Journal of Applied Ecology volume 48 tahun 2011

Journal of Applied Ecology, 48, 1–2 doi: 10.1111/j.1365-2664.2010.01938.x EDITORIAL Practitioner’s perspectives: intr...

Author: Edited by: Jane Memmott | EJ Milner-Gulland | Phil Hulme | Marc Cadotte and Mark Whittingham with Gill Kerby

9 downloads 537 Views 16MB Size Report

This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!

Report copyright / DMCA form

DOWNLOAD PDF

Journal of Applied Ecology, 48, 1–2

doi: 10.1111/j.1365-2664.2010.01938.x

EDITORIAL

Practitioner’s perspectives: introducing a different voice in applied ecology Philip E. Hulme* The Bio-Protection Research Centre, Lincoln University, PO Box 84, Canterbury, New Zealand

Most researchers working in applied ecology are aware that much of what is published in leading ecological and environmental science journals makes little diﬀerence to the day-today management of species and ecosystems (Nature 2007). In recent years, several editorials have made this point and attempted to identify a way forward to bridge ‘The Great Divide’ (Born, Boreux & Lawes 2009; Milner-Gulland et al. 2010; Memmott et al. 2010). The onus has largely been on the scientiﬁc community to communicate the value of its science more clearly, become increasingly involved in extension activities and, heaven forbid, step down from their ivory towers and get their hands dirty. Yet the reality is that many scientists are doing this already, several very successfully (Possingham 2009). A more serious concern is that academic journals are simply not the best medium to communicate practical messages to a wide audience who need speciﬁc solutions to particular problems that have to be delivered on a tight budget. We should not be surprised; much academic research aims to be innovative, internationally competitive and globally relevant – aims which are not always congruent with ﬁnding practical solutions. It is against these criteria that many editorial decisions regarding whether or not to accept a paper for publication are made. Irrespective of how much hand-wringing might take place among editors this situation is unlikely to change as publishers judge the viability of journals using bibliometric indices and numbers of institutional subscriptions. Yet, potentially there is another way. Communication is a two-way street, even if much of the academic traﬃc is heading in one direction with no clear destination. So how do scientiﬁc researchers hear about the concerns and needs of those tackling problems in the ﬁeld? Surveys of stakeholders are certainly one way to conﬁrm that they feel the scientiﬁc community is not listening to them (e.g. Andreu, Vila` & Hulme 2009) but questionnaires do not address the problem. An alternative is to provide an opportunity within the pages of academic journals for non-standard pieces to be written by individuals who have a diﬀerent perspective on what is needed in applied ecology research and whether the papers published in academic journals get anywhere near it. With this aim, the Journal of Applied Ecology launches its ﬁrst ‘Practitioner’s Perspective’. These ‘prick our conscience’ pieces can be contributed by anyone who has a strong opinion on the current state of applied ecology research, whether academic or not, as long as they can provide an original perspective and a constructive

way forward. Although practitioners have been identiﬁed as a distinct group of actors in applied ecology that ‘buy land, put up fences, set ﬁres, put out ﬁres, lobby politicians, negotiate with farmers, spray invasive weeds, poison rats and guard against poachers’ (Nature 2007), we are not placing restrictions on who is or is not a ‘practitioner’. Thus, we welcome pieces from academics (at least those with a bit of dirt under their ﬁngernails) as well as civil servants, environmental consultants, park managers and environmental lobbyists. The truth is we are unsure what to expect in terms of submissions under this new feature, hopefully provocative pieces from writers whose voices are rarely heard in our journal. To kick-start this initiative we have commissioned a few articles that might give a ﬂavour of the pieces we would like to see published in the future. Our greatest challenge to date has been to prevent these pieces from becoming advertorials for the activities of NGOs, conservation groups or consultancies. This is certainly not what we want, but we do welcome examples of best practice that may not have made it into the wider academic literature. The ﬁrst Practitioner’s Perspective appears in this issue (Goulson et al. 2011) and illustrates the viewpoint of the Bumblebee Conservation Trust, although the lead author is a senior academic at Stirling University, UK. Hopefully, in addition to highlighting how science informs the conservation of bumblebees it will challenge readers to consider what more needs to be done. We encourage future submissions under Practitioner’s Perspectives but please be sure to contact the Editors to discuss your piece beforehand. There is no prescribed structure to Practitioner’s Perspectives apart from our hope that they will be thought-provoking and challenge the science community to consider the perspectives of those individuals addressing applied ecological issues. However, authors may wish to consider covering the activities of the individual or organization with regard to ecological management, the key issues they are addressing (see Sutherland et al. 2006, 2009 for a range of key questions), the extent to which applied ecological research has supported their activities (if at all), how future research might assist them to address ecological problems more eﬀectively and how this might best be achieved (e.g. through greater dialogue, joint projects, new research techniques etc.).

References Andreu, J., Vila`, M. & Hulme, P.E. (2009) An assessment of stakeholder perceptions and management of alien plants in Spain. Environmental Management, 43, 1244–1255.

*Correspondence author. E-mail: [email protected] Ó 2011 The Author. Journal of Applied Ecology Ó 2011 British Ecological Society

2 Editorial Born, J., Boreux, V. & Lawes, M.J. (2009) Synthesis: sharing ecological knowledge – the way forward. Biotropica, 41, 586–588. Goulson, D., Rayner, P., Dawson, R. & Darvill, B. (2011) Translating research into action; bumblebee conservation as a case study. Journal of Applied Ecology, 48, 3–8. Memmott, J., Cadotte, M., Hulme, P.E., Kerby, G., Milner-Gulland, E.J. & Whittingham, M.J. (2010) Editorial: putting applied ecology into practice. Journal of Applied Ecology, 47, 1–4. Milner-Gulland, E.J., Fisher, M., Browne, S., Redford, K.H., Spencer, M. & Sutherland, W.J. (2010) Do we need to develop a more relevant conservation literature? Oryx, 44, 1–2. Nature (2007) The great divide. Nature, 450, 135–136. Possingham, H. (2009) Dealing with ‘The great divide’. Decision Point, 28, 2. Sutherland, W.J., Armstrong-Brown, S., Armsworth, P. R., Brereton, T., Brickland, J., Campbell, C.D., Chamberlain, D. E., Cooke, A.I., Dulvy, N.K., Dusic, N.R., Fitton, M., Freckleton, R.P., Godfray, H.C., Grout, N., Harvey, H.J., Hedley, C., Hopkins, J.J., Kift, N.B., Kirby, J., Kunin, W.E.,

MacDonald, D.W., Markee, B., Naura, M., Neale, A.R., Oliver, T., Osborn, D., Pullin, A.S., Shardlow, M.E.A., Showler, D.A., Smith, P.L., Smithers, R.J., Solandt, J.-L., Spencer, J., Spray, C.J., Thomas, C.D., Thompson, J., Webb, S.E., Yalden, D.W. & Watkinson, A.R. (2006) The identiﬁcation of 100 ecological questions of high policy relevance in the UK. Journal of Applied Ecology, 43, 617–627. Sutherland, W.J., Adams, W.M., Aronson, R.B., Aveling, R., Blackburn, T.M., Broad, S., Ceballos, G., Coˆte´, I.M., Cowling, R.M., da Fonseca, G.A.B., Dinerstein, E., Ferraro, P.J., Fleishman, E., Gascon, C., Hunter Jr, M., Hutton, J., Kareiva, P., Kuria, A., Macdonald, D.W., MacKinnon, K., Madgwick, F.J., Mascia, M.B., McNeely, J., Milner-Gulland, E.J., Moon, S., Morley, C.G., Nelson, S., Osborn, D., Pai, M., Parsons, E.C.M., Peck, L.S., Possingham, H., Prior, S.V., Pullin, A.S., Rands, M.R.W., Ranganathan, J., Redford, K.H., Rodriguez, J.P., Seymour, F., Sobel, F., Sodhi, N.S., Stott, A., Vance-Borland, K. & Watkinson, A.R. (2009) One hundred questions of importance to the conservation of global biological diversity. Conservation Biology, 23, 557–567.

Ó 2011 The Author. Journal of Applied Ecology Ó 2011 British Ecological Society, Journal of Applied Ecology, 48, 1–2

Journal of Applied Ecology 2011, 48, 163–173

doi: 10.1111/j.1365-2664.2010.01890.x

Assessing spatial patterns of disease risk to biodiversity: implications for the management of the amphibian pathogen, Batrachochytrium dendrobatidis Kris A. Murray1*, Richard W. R. Retallick2, Robert Puschendorf3, Lee F. Skerratt4, Dan Rosauer5,6, Hamish I. McCallum7, Lee Berger4, Rick Speare4 and Jeremy VanDerWal3 1

The Ecology Centre, School of Biological Sciences, University of Queensland, Brisbane, Queensland 4072, Australia; GHD Pty Ltd, 8 ⁄ 180 Lonsdale Street, Melbourne, Victoria 3000, Australia; 3Centre for Tropical Biodiversity and Climate Change Research, School of Marine and Tropical Biology, James Cook University, Townsville, Queensland 4811, Australia; 4Amphibian Disease Ecology Group, School of Public Health, Tropical Medicine and Rehabilitation Sciences, James Cook University, Townsville, Queensland 4811, Australia; 5School of Biological, Earth and Environmental Sciences, University of New South Wales, Sydney, New South Wales 2052, Australia; 6Centre for Plant Biodiversity Research, GPO Box 1600, Canberra, Australian Capital Territory 2601, Australia; and 7School of Environment, Griﬃth University, Nathan Campus, Queensland 4111, Australia 2

Summary 1. Emerging infectious diseases can have serious consequences for wildlife populations, ecosystem structure and biodiversity. Predicting the spatial patterns and potential impacts of diseases in freeranging wildlife are therefore important for planning, prioritizing and implementing research and management actions. 2. We developed spatial models of environmental suitability (ES) for infection with the pathogen Batrachochytrium dendrobatidis, which causes the most signiﬁcant disease aﬀecting vertebrate biodiversity on record, amphibian chytridiomycosis. We applied relatively newly developed methods for modelling ES (Maxent) to the ﬁrst comprehensive, continent-wide data base (comprising >10000 observations) on the occurrence of infection with this pathogen and employed novel methodologies to deal with common but rarely addressed sources of model uncertainty. 3. We used ES to (i) predict the minimum potential geographic distribution of infection with B. dendrobatidis in Australia and (ii) test the hypothesis that ES for B. dendrobatidis should help explain patterns of amphibian decline given its theoretical and empirical link with organism abundance (intensity of infection), a known determinant of disease severity. 4. We show that (i) infection with B. dendrobatidis has probably reached its broad geographic limits in Australia under current climatic conditions but that smaller areas of invasion potential remain, (ii) areas of high predicted ES for B. dendrobatidis accurately reﬂect areas where population declines due to severe chytridiomycosis have occurred and (iii) that a host-speciﬁc metric of ES for B. dendrobatidis (ES for Bdspecies) is the strongest predictor of decline in Australian amphibians at a continental scale yet discovered. 5. Synthesis and applications. Our results provide quantitative information that helps to explain both the spatial distribution and potential eﬀects (risk) of amphibian infection with B. dendrobatidis at the population level. Given scarce conservation resources, our results can be used immediately in Australia and our methods applied elsewhere to prioritize species, regions and actions in the struggle to limit further biodiversity loss. Key-words: amphibian declines, bioclimatic modelling, chytrid fungus, chytridiomycosis, infectious disease, Maxent, species distribution model

*Correspondence author. E-mail: [email protected] 2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society

164 K. A. Murray et al.

Introduction Emerging infectious diseases can have serious consequences for wildlife populations, ecosystem structure and biodiversity (Crowl et al. 2008). Alarmingly, their incidence appears to be rising as a result of anthropogenic inﬂuences that favour the growth, dispersal and transmission of pathogens (Daszak, Cunningham & Hyatt 2000; Jones et al. 2008). Assessing the extent, eﬀects and dynamics of diseases in host populations are therefore important for predicting disease emergence and its consequences, and to plan, prioritize and implement research and management actions. Arguably the most serious wildlife disease impacting vertebrate biodiversity at this time is chytridiomycosis. This disease, caused by infection with the fungal pathogen Batrachochytrium dendrobatidis Longcore, Pessier & Nichols (1999), has been implicated in many rapid and recent amphibian declines and extinctions (Stuart et al. 2004; Skerratt et al. 2007; Bielby et al. 2008; Wake & Vredenburg 2008). Batrachochytrium dendrobatidis (hereafter Bd) appears to have undergone recent global expansion after the outbreak of a single clonal lineage, the origin of which remains uncertain (Morehouse et al. 2003; Rachowicz et al. 2005; but see Goka et al. 2009; James et al. 2009). As an international notiﬁable disease, reporting of Bd detection to the World Organisation for Animal Health (OIE) is now obligatory for member countries (World Organisation for Animal Health 2008). Bd is currently known from hundreds of amphibian species and from all continents where amphibians occur (Speare & Berger 2000; Kusrini et al. 2008; Olson & Ronnenberg 2008). Given its broad host range, many more species are likely to be suitable hosts and this number will rise as search eﬀort and reporting increases. The potential distribution of Bd is, however, still relatively poorly understood; the native range has not been delineated and Bd may still be expanding its range worldwide (Lips et al. 2008; Rohr et al. 2008; James et al. 2009). In Australia, it is now widely accepted that the invasion and spread of Bd is the probable cause of many frog declines (Skerratt et al. 2007). Despite this, little quantitative data on risk of disease have been available to researchers and managers at broad spatial scales, hampering eﬀorts to pinpoint areas and species warranting immediate management attention. Tools for predicting the spread or establishment of Bd and for identifying areas of high disease risk are therefore critical for policy makers, researchers and managers charged with detecting this pathogen, developing management actions and prioritizing resource expenditure (Gascon et al. 2007; Skerratt et al. 2008). Predicting the dispersal and potential range of organisms is commonly approached by characterizing environmental suitability (ES) with correlative species distribution models (SDMs) (Guisan & Thuiller 2005; Kearney & Porter 2009). ‘Presence-only’ SDMs are being used increasingly for their application to species occurrence data sets for which no reliable absence records may be available (e.g. museum ⁄ herbarium collections, atlases, non-targeted surveys etc.) (Pearce & Boyce 2006). Rarely used in studies of infectious disease, presence-only SDMs appear well suited to investigating the

distribution of infection with some pathogens because, analogous to verifying true absence of rare or endangered species (Gibson, Barrett & Burbidge 2007), it is a statistical and sampling challenge to assert ‘freedom from disease’ (Digiacomo & Koepsell 1986; Ziller et al. 2002; Skerratt et al. 2008). This challenge is rarely met for wildlife pathogens because the cost of suﬃcient sampling (including diagnostics, personnel, logistics, etc.) at broad spatial scales is typically prohibitive. Furthermore, pathogen prevalence may be low in a host population or may ﬂuctuate temporally, and the host itself may be diﬃcult to detect, particularly if the pathogen has resulted in host declines as has been the case with Bd (e.g. Lips et al. 2006). Correlative SDMs will only be appropriate where the distribution of infection with a pathogen is expected to be regulated by spatially quantiﬁable predictors that capture ES, such as climate or habitat type. For many pathogens, this may be inappropriate if hosts provide a highly regulated ‘habitat’ in which to grow and no stage of the life cycle is exposed to external environmental conditions (e.g. for internal, directly transmitted pathogens of endotherms). In the case of Bd, however, infections occur on ectothermic amphibian hosts and there is a direct eﬀect of the environment (particularly temperature and moisture) on growth and survival of both free-living and parasitic life stages (Johnson & Speare 2003; Berger et al. 2004; Piotrowski, Annis & Longcore 2004; Woodhams et al. 2008). SDMs should thus be highly suited to characterizing ES for infection with Bd to provide important insights into its potential distribution, shed light on the probability of pathogen establishment following invasion into previously naı¨ ve areas [as has been demonstrated for other invasive species (Ficetola, Thuiller & Miaud 2007), and to help improve detection probability while reducing cost and eﬀort of surveying for the pathogen in the future (Guisan et al. 2006)]. In an adaptive management context, such models are ideally suited to tailoring future data collection, which can in turn be used to iteratively improve the model (Wintle, Elith & Potts 2005). We hypothesized that modelling ES for infection with Bd may also provide useful information about the risk to amphibian populations posed by chytridiomycosis. Recently, VanDerWal et al. (2009b) demonstrated that modelling ES broadly predicts an organism’s abundance. For chytridiomycosis, Bd abundance (infection intensity) on the host is a direct determinant of disease development, severity and population eﬀects (Carey et al. 2006; Voyles et al. 2007; Briggs, Knapp & Vredenburg 2010). Indeed, seasonal and elevational variation in the prevalence, intensity and virulence of Bd infections has long implicated climatic suitability as a major factor governing its eﬀects in the wild (Berger 2001; Berger et al. 2004; Woodhams & Alford 2005; Kriger & Hero 2007), and this has been consistently supported by laboratory infection experiments (Woodhams, Alford & Marantelli 2003; Berger et al. 2004; Carey et al. 2006). We would thus expect that our ES results not only reﬂect proliferation of Bd on the host at the time scale used in model training (average annual) but also the risk of severe chytridiomycosis to populations as a whole, a link we

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 163–173

Spatial patterns of disease risk 165 test herein by examining patterns of disease-induced amphibian population declines. A published SDM already exists for Bd (Ron 2005), in which a correlative, presence-only SDM (GARP) with relatively few (n = 44) presence records in the New World was used to predict the global potential range of this pathogen. While this was of great use at the time of publication, the model appears to exaggerate suitable area in Australia (Fig. S1, Supporting Information), is at a spatial scale too coarse to be useful for regional management or for predicting population declines, and is likely to suﬀer from several sources of uncertainty inherent to correlative SDMs (e.g. extrapolation beyond the training region, limited sample size, algorithm nuances, inappropriate pseudo-absence selection; see e.g. Araujo & Guisan 2006; Pearson et al. 2007; Peterson, Papes & Eaton 2007; VanDerWal et al. 2009a). These issues together necessitate the development of independent, regionally speciﬁc predictions for planning future research and management actions on Bd at ﬁner spatial scales. To this end, we applied a relatively novel SDM method (Maxent) (Phillips, Anderson & Schapire 2006) to the most comprehensive, continent-wide data base available to date on the occurrence of infection with Bd (Murray et al. 2010) to model ES for this pathogen. We employed novel methodologies to deal with common but rarely addressed sources of SDM uncertainty to provide maximum robustness in our predictions of ES in Australia given the available data. The predictions were used to estimate the minimum potential geographic distribution of infection with Bd in Australia and to test the hypothesis linking ES to disease risk as indicated by patterns of disease-induced population declines. We used our results to identify where chytridiomycosis may pose the greatest risk to endangered species, allowing prioritization of species, regions and actions when considering research and management options given scarce conservation funds (Wilson et al. 2007).

Materials and methods MODEL DESCRIPTION

The SDM software used was Maxent (ver. 3.3.0), for which the underlying theory and assumptions have been described in detail elsewhere

(Phillips, Anderson & Schapire 2006; Dudik, Phillips & Schapire 2007). Brieﬂy, Maxent has been shown to generally outperform other correlative (both presence-only and presence-absence) SDM algorithms (Elith et al. 2006; Peterson, Papes & Eaton 2007; Graham et al. 2008; Wisz et al. 2008). It requires presence records only (but uses random background points to sample available environmental space), accounts for interactions among variables and identiﬁes areas that fall beyond the range of environmental conditions used during training when making projections (identiﬁed as ‘clamped’ areas). The output of Maxent corresponds with an index of ES for the organism, where higher values correspond to a prediction of better conditions (Phillips, Anderson & Schapire 2006). We used Bd occurrence records from Murray et al. (2010). Full details of these data and their collection methods are described in the Metadata provided therein. Brieﬂy, this newly compiled data set represents the ﬁrst comprehensive, continent-wide data base describing occurrence patterns of Bd on wild amphibian hosts. The data base comprises 821 sites in Australia at which frogs or tadpoles have been tested for Bd and includes 10 183 records from >80 contributors spanning collection dates from 1956 to 2007. Bd was detected on 63 (55%) of the 115 species in the data set (c. 28% of Australia’s 223 species) (Table S1, Supporting Information). Two hundred and eighty-four Bd-positive sites had suﬃcient geographic accuracy for inclusion in the model (Table 1, Fig. S2, Supporting Information). Few localities in the data base comprise statistically defensible absence records given the diﬃculty of asserting freedom from chytridiomycosis. The data base represents records of clinical and aclinical infection with Bd, which by deﬁnition is considered synonymous with chytridiomycosis (ranging from severe and clinical to benign and aclinical) by disease authorities (sensu Berger et al. 1998 and as per the ‘Deﬁnitions’ of the OIE’s Aquatic Animal Health Code; see http://www.oie.int/eng/normes/fcode/en_chapitre_1.1.1.htm) but distinct from including records of the free-living stage which may also be detected oﬀ-host (Kirshtein et al. 2007; Walker et al. 2007). Bd’s current occurrence pattern in Australia is highly consistent with the hypothesis that environmental characteristics, such as climate or habitat type, place direct limits on its distribution. Its extensive distribution nation-wide (Fig. S2; Murray et al. 2010) demonstrates that it has had suﬃcient opportunity to spread great distances and into new geographic areas from its hypothesized point(s) of introduction (major ports) (Murray et al. 2010). The large number of known hosts and the spectrum of potentially susceptible amphibian hosts nationally (e.g. Litoria spp.) in currently uninfected regions strongly suggest that Bd is not limited in Australia by the unavailability of susceptible amphibian host species. Similarly, its presence in some remote and sparsely populated regions of the

Table 1. Summary of Batrachochytrium dendrobatidis (Bd) data base records. Geo-referenced Bd+ sites are those where the pathogen was detected and an accurate geographic coordinate was obtained for input to the distribution model. Individuals tested is a minimum estimate; many site records in the database did not include total number of individuals tested (see Fig. S2 for map and key to State names)

State

Database records

Individuals tested

Sites with records

Bd+ sites

Georeferenced Bd+ sites

ACT NSW NT QLD SA TAS VIC WA Australia

77 494 14 6660 42 146 26 2647 10 106

77 887 14 8789 42 574 32 2446 12 861

7 79 2 359 16 122 11 225 821

1 39 0 165 8 45 6 76 340

1 39 0 165 8 43 6 22 284

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 163–173

166 K. A. Murray et al. country and its absence in some populated regions suggest that it is not simply dependent on humans for its establishment and persistence, although in some cases human aided spread seems likely (Morgan et al. 2007; Skerratt et al. 2007). In contrast, Adams et al. (2010) report that Bd occurrence in Oregon and California, USA, does not correlate with any hypothesized environmental factors, but that Bd detectability increases with human inﬂuence on the landscape. We thus also evaluated the predictive power of a human-inﬂuence hypothesis for predicting Bd’s current occurrence pattern in Australia and compared it with our ES model (Fig. S8, Supporting Information). We used 19 bioclimatic variables (all continuous), one geo-physical variable (distance to water; continuous) and one vegetation type variable (categorical) at a resolution of c. 250 m (9 arc-seconds) for our models (Table S2, Supporting Information). We knew a priori that many of the variables were correlated and potentially meaningful contributors to the model; to avoid over-sized models (including variables with no predictive value) or over-ﬁtted models (too many parameters for the data set) (Parolo, Rossi & Ferrarini 2008), we ﬁrst selected the top ranking variables that together contributed c. 90% of the information to a full model run. We then re-ran a ‘pruned’ model with the most important variables (Table S2). Model accuracy was assessed with the area under the curve (AUC) of the receiver operator characteristic (ROC), which is a single measure of discrimination ability (presence from random background, where a value of 1 = perfect prediction, 0Æ5 = prediction no better than random) of the models (Fielding & Bell 1997). To incorporate uncertainty into our predictions, we used bootstrapping (N = 100) with unique sets of training and testing data (70 : 30% respectively). Many presence-only SDMs require background points (or pseudo-absences), the selection of which can inﬂuence the outcome of the models (Phillips et al. 2009; VanDerWal et al. 2009a). We provide an extended discussion of our background point selection in Fig. S2 which we used in order to limit as far as possible the eﬀects of unquantiﬁable sampling bias and modelling an organism with considerable invasion potential.

DISEASE RISK

To investigate the hypothesized relationship between ES for Bd and the risk of chytridiomycosis to susceptible amphibian populations, we assessed whether our results were consistent with descriptions of population decline attributed to severe chytridiomycosis (Berger et al. 1998, 2004). We anticipated that decline sites would be strongly skewed towards higher values of ES for Bd. Declines attributed to chytridiomycosis have been best described from uplands in the Australian Alps and from montane rainforest areas in Queensland, where ill and dead frogs have been rigorously diagnosed as dying from chytridiomycosis at the time of declines (Berger et al. 1998, 2004; Hines, Mahony & McDonald 1999; McDonald & Alford 1999; Osborne, Hunter & Hollis 1999) (Fig. S3, Supporting Information). We next averaged our ES predictions across amphibian occurrence records for each species in the data set described by Slatyer, Rosauer & Lemckert (2007 updated 2009, D. Rosauer unpubl. data) to derive a species-speciﬁc metric of ES for Bd that we termed ‘ES for Bdspecies’. Slatyer et al.’s extensive data set comprises 291 942 occurrence records for all of Australia’s amphibian species. We removed duplicate records from the same locality (leaving 140 897 records; mean per species = 640) for calculations. Further details of the metric are provided in Fig. S4, Supporting Information. Species range size has previously been identiﬁed as the major risk factor for decline and extinction in Australian amphibians after controlling for other lifehistory and ecological factors (Murray & Hose 2005). We thus tested

for an eﬀect of ES for Bdspecies, controlling for the range size eﬀect, in contributing to whether amphibians have experienced declines or not. Amphibian trend classiﬁcations were sourced from the IUCN (2008). Range sizes were calculated from extent of occurrence polygons developed for the Global Amphibian Assessment (GAA) (Stuart et al. 2004). Finally, we calculated mean ES values across Australia’s biogeographic regions to identify those most suitable for infection with the pathogen (Fig. S5, Supporting Information). We related these results to amphibian species richness and endemism statistics from the study of Slatyer, Rosauer & Lemckert (2007) to indicate where infection with Bd most threatens anuran biodiversity in Australia (Fig. S6, Supporting Information).

Results MODEL SELECTION, VALIDATION AND VARIABLE CONTRIBUTIONS

After the pruning step, mean test AUC was 0Æ900 (range 0Æ874–0Æ925) and the model contained eight variables. The jack-knife procedure, which examines the eﬀect of individual variables, indicated that mean diurnal temperature range and annual precipitation had the most useful information as single variables on training data (highest gain scores in isolation) as well as the highest predictive power (highest AUC in isolation) (Fig. 1). Response curves characterizing the relationships between ES and each of the two most inﬂuential predictor variables are shown in Fig. S7, Supporting Information. In the comparative analysis incorporating human population density (HPD), predictive performance of the full model was unchanged (0Æ903, range = 0Æ852–0Æ936) and HPD had inferior predictive power in isolation (AUC = 0Æ763, range = 0Æ716–0Æ811) relative to many of the environmental predictors. For subsequent analyses, we thus used the model incorporating environmental variables only (see Fig. S8, Supporting Information for results and further discussion).

PREDICTED DISTRIBUTION

The model suggested that infection with Bd should be largely restricted to the eastern and southern seaboards of Australia, with nearly all of inland and northern Australia unsuitable. Figure 2 represents the average Maxent predictions of ES (available for download in Appendix S2, Supporting Information). Clamping indicated that all of Australia fell within the environmental limits used to train the model (data not shown).

DISEASE RISK

Decline sites (mean ESdecline = 0Æ758; 95% CI = 0Æ714–0Æ802, n = 39) were highly skewed towards higher ES values compared to all sites used for model training and testing (mean ESall = 0Æ577, 95% CI = 0Æ550–0Æ604, n = 284) (Fig. 4a,b,d). Mean ES for Bdspecies varied between population trend categories (Fig. 3a); three extinct species had the highest value, 42 declining species had an intermediate value and 151 stable

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 163–173

Spatial patterns of disease risk 167 Only variable

(a)

Without variable

Full Vegetation type Precip cold quart Precip warm quart Precip dry quart Annual precip Mean temp dry quart Temp ann range Mean diurnal range 0

0·2

0·6

Only variable

(b)

Fig. 1. Variable contributions to (a) training gain and (b) AUC of the ﬁnal ‘pruned’ model for Batrachochytrium dendrobatidis in Australia. ‘Only variable’ indicates the results of the model when a single variable is run in isolation; ‘without variable’ indicates the eﬀect of removing a single variable from the full model (jack-knife). Values are means from 100 replicates. See Table S2 for full variable names and descriptions.

0·4

0·8 1 Training gain

1·2

1·4

1·6

Without variable

Full Vegetation type Precip cold quart Precip warm quart Precip dry quart Annual precip Mean temp dry quart Temp ann range Mean diurnal range

species exhibited a comparatively low value. In a logistic model in which species were grouped by whether they had declined or not (unknown trend species omitted), ES for Bdspecies was a highly signiﬁcant predictor of decline (Ddev = 20Æ932, d.f. = 1, P < 0Æ001), even after controlling for a signiﬁcant inﬂuence of narrow species range size (Ddev = 22Æ831, d.f. = 1, P < 0Æ001). The best model in terms of AIC contained ES for Bdspecies as a highly signiﬁcant term (P < 0Æ001), range size as a marginally signiﬁcant term (P = 0Æ098) and no interaction term. Table S3, Supporting Information presents a list of priority species for research and management indicating where investigation of Bd as a potential threatening process is critical. Table S4, Supporting Information presents the full list of Australian species. Mean ES varied considerably across biogeographic regions (Fig. 3b); the Wet Tropics (see also Fig. 4a) was predicted to have the highest mean suitability for Bd, followed by the Central Mackay Coast (Fig. 4b), Tasmania’s southern ranges, northern slopes, north-east (Ben Lomond) and King Island (Fig. 4e) and the NSW north coast (Fig. 4c). South-east Queensland (Fig. 4c), the Australian Alps (Fig. 4d), the Swan Coastal Plain (around Perth) (Fig. 4f) and the Tasmanian south-east also showed high mean ES values. Many regions with low mean ES nevertheless showed limited areas of very high ES as indicated by their maximum values (e.g. Brigalow Belts, Einasleigh Uplands, NSW south-western slopes) (Fig. 3b).

Discussion Infection with B. dendrobatidis occurs across a broad range of climates in Australia, in areas that are at times very hot, cold,

0·65

0·7

0·75

0·8 AUC

0·85

0·9

0·95

dry or wet. Those locations range from the hot, humid coastal lowlands of north-eastern Australia to the highest peaks of the Australian Alps, where winter snow occurs. Despite its broad tolerance of conditions, the model suggested that speciﬁc environmental conditions will restrict infection with Bd to the generally cooler and wetter areas of Australia (Figs 2 and 4). In this respect, our model was highly consistent with that of Ron (2005) (Fig. S1; Fig. 2); however, our results suggested that Bd should be more restricted, with the majority of central (arid) Australia being broadly unsuitable for Bd persistence (see also Fig. S8). The model indicated that ES increased with annual precipitation (with a minimum extreme of c. 500 mm) (Fig. S7). This is not surprising since desiccation is known to rapidly kill Bd in vitro (Berger 2001; Johnson et al. 2003) and the presence of permanent water is known to be an important feature for sustaining Bd, probably because the transmission stage for Bd is an aquatic zoospore (Berger et al. 1998). The model also suggested that mean diurnal temperature range was an important variable; the response curve indicated that ES declined rapidly in highly variable temperature regimes, where the diﬀerence in daily maxima and minima is greater than c. 11 C. Variation in temperature of itself has not previously been shown to aﬀect chytridiomycosis (Woodhams, Alford & Marantelli 2003). However, high temperatures are known to be lethal to Bd and the eﬀect of temperature variability may be explained by the observation that areas with higher temperature variability (e.g. the arid ⁄ semi-arid interior of the country) also typically exhibit very high maximum temperatures. This suggestion is supported by the response of Bd to maximum temperature of the warmest month, which showed maximum ES in the range of

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 163–173

168 K. A. Murray et al.

Fig. 2. Model predictions of environmental suitability (after bootstrapping N = 100) for Batrachochytrium dendrobatidis in Australia.

maximum temperatures 18–30 C beyond which there is a precipitous decrease (data not shown). Our results are thus highly consistent with those of previous studies indicating that high temperatures are detrimental to Bd (Kriger & Hero 2007; Muths, Pilliod & Livo 2008; Puschendorf et al. 2009).

(termed ES for Bdspecies) was a very strong predictor of amphibian decline at a national level. These ﬁndings support the hypothesis that ES for infection with the pathogen as modelled here is broadly predictive of suitability for, and severity of, the disease chytridiomycosis via a theoretical and empirical link with intensity of infection (VanDerWal et al. 2009b). We thus interpret our ES for Bd results as being a highly useful source of quantitative information relevant to explaining the potential eﬀects of infection with B. dendrobatidis (disease risk). This association should be interpreted cautiously, however, as in order for ES for Bdspecies to translate to risk of decline a number of other conditions relevant to the epidemiology of chytridiomycosis must be fulﬁlled, most importantly transmission. Species with more aquatic life-histories and an association with permanent water are most susceptible and at greatest risk of severe disease (Berger et al. 1998, 2004). Further, species inhabiting diﬀerent micro-habitats can vary in their relative risk of infection within a single location (Woodhams & Alford 2005; Skerratt et al. 2008). As such, actual disease risk will be a product of the ES for the pathogen, the susceptibility of the species and the factors that make it susceptible to decline (e.g. see Bielby et al. 2008) given the former. This is an important consideration as it will be necessary to stratify host life-history traits for prioritization purposes (Table S3). Bielby et al. (2008) found that small range size, altitude and an aquatic life stage are risk factors for rapid decline in Bd-positive species. However, applying these risk factors to all species as they do is a considerable extrapolation because not all species are equally susceptible to infection. For example, very high-risk values in that study were assigned to many species with largely terrestrial life-histories, including many of Australia’s microhylid frogs (e.g. Cophixalus sp). While some of these also exhibit high ES for Bdspecies values as described herein, neither index identiﬁes actual risk from Bd as species in this group appear far less susceptible than stream-dwelling and permanent water-associated species from the same region (N = 557 negative results in areas that are Bd-positive; K. Hauselberger & D. Mendez et al. unpubl. data; Skerratt et al. 2008). A more sophisticated risk analysis can be performed when more information is available about the innate susceptibilities of diﬀerent amphibian species to chytridiomycosis and to decline. Integrating host-life history and ecological traits with the pathogen’s environmental requirements (as modelled here) to predict infection and decline is the focus of our current research eﬀorts (Murray et al. in press).

DISEASE RISK

Two key results from this study are that (i) our predictions of ES are strikingly consistent with known associations between Bd and amphibian population declines in Queensland ⁄ New South Wales (Fig. S3 and Fig. 4a–c) (Hines, Mahony & McDonald 1999; McDonald & Alford 1999) and in the Australian Alps (Osborne, Hunter & Hollis 1999; Berger et al. 2004) (Fig. 4d) and (ii) the species-speciﬁc metric of ES for Bd

MANAGEMENT IMPLICATIONS

We have shown that ES for Bdspecies is a strong predictor of decline at a continental scale. This result was independent of a previously reported, dominant eﬀect of narrow geographic range size. Our study provides a species-speciﬁc metric, representing the environmental requirements of the pathogen, with which to begin to assess this risk and calls for targeted vigilance

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 163–173

(a)

Environmental suitability for Bd (95%CI)

Spatial patterns of disease risk 169 1 0·9 0·8 0·7 0·6 0·5 0·4 0·3 0·2 0·1 0 Extinct (3)

Declining (42)

Stable (151)

Unknown (17)

IUCN trend category Environmental suitability for Bd (±1SD)

(b) 1·0 MEAN Max

0·9 0·8 0·7 0·6 0·5 0·4 0·3 0·2 0·1 0·0

IBRA Ecoregion

Fig. 3. Mean environmental suitability for Batrachochytrium dendrobatidis across: (a) species with diﬀerent IUCN trend classiﬁcations, (b) IBRA ecoregions (averaged across entire region; only regions where the maximum or mean value >0Æ2 are shown). See Fig. S5 for map.

in sampling for this disease and monitoring for its potentially insidious eﬀects (Murray et al. 2009; Pilliod et al. 2010). Bd records exist from most regions that were deemed suitable by the model, indicating that it has probably reached its broad geographic limits on this continent. There are, however, at least two areas that show marginal suitability for Bd beyond the known range of the pathogen where testing has failed to detect it: Cape York (Skerratt et al. 2008) (Fig. 4a) and Tasmania’s World Heritage south-west (Pauza & Driessen 2008) (Fig. 4e). It is possible that Bd has simply not yet dispersed to these regions, as they are at the extreme limits of the distribution in northern and southern Australia. However, the results of this study may also suggest that establishment or disease risk could be relatively low in these regions. Our results provide a testable hypothesis and surveys should continue in these areas where suitability is predicted to be highest (Fig. 4a,e). Prevention of spread nevertheless remains the best management strategy and these areas should not be regarded as areas in which Bd could not establish and cause mortalities (Fig. 2). Hygiene protocols should therefore be enforced for people entering these areas (Phillott et al. 2010). Several other marginal to highly suitable regions exist where no sampling has occurred. These regions represent important

areas for future sampling to establish the actual geographic limit of Bd in Australia and to establish amphibian population health. Identiﬁcation of naive populations at high disease risk is a particular priority. Examples include uplands in the far north of Queensland (Cape York; Fig. 4a), upland areas in the Brigalow Belt (Fig. 4c), a large expanse of the western slopes of the Great Dividing Range in NSW (Fig. 4c), the south-west central tablelands of NSW and the regions surrounding Mount Gambier and the Mt Lofty Ranges in South Australia (Fig. 2). Conversely, our results suggest that several declining species for which chytridiomycosis is a suspected threatening process may have relatively low ES for Bdspecies (e.g. Litoria piperata and Litoria castanea), which provides a way forward when considering management and research activities. Similarly, some declining species may be at high risk of disease only in a subset of populations (e.g. Litoria spenceri, as indicated by maximum vs. mean ES values – Table S3). We are not asserting that Bd will be absent from these species or populations, but that other factors may also be involved in their decline, an example of where our results provide some useful and testable hypotheses that should be pursued. Regions of high disease risk in association with high host endemism should be the highest priority for population moni-

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 163–173

170 K. A. Murray et al.

(a)

(b)

(c)

(d)

(e)

(f)

toring and ⁄ or management activity. Of the 11 centres of exceptional anuran endemism identiﬁed by Slatyer, Rosauer & Lemckert (2007), six occur in regions predicted to be highly suitable for Bd, including the Wet Tropics (Fig. 4a), Central Mackay Coast (Mackay ⁄ Eungella; Fig. 4b), Gladstone (Kroombit Tops), South-east Queensland (Gympie-Coﬀs Harbour; Fig. 4c) and south-west Western Australia (Walpole and Bunbury-Augusta; Fig. 4f) (see Fig. S5 for ecoregion names and Fig. S6 for endemism ⁄ richness). Records of Bd exist from all of these areas. An additional two areas (Townsville and Cape York) are predicted to have more restricted regions that are marginally or highly suitable for Bd (Fig. 4a). Three endemism hotspots are predicted to be at negligible risk from Bd (Kakadu and the Arnold River region in the NT and the Mitchell Plateau in WA). Establishing and maintaining a disease-free status should be their regional priority. The methods and results from this study can be used as a tool for establishing cost-sharing arrangements, prioritizing future eﬀorts to detect and manage this pathogen (e.g. disease

Fig. 4. Selected regions in Australia predicted to have high average environmental suitability for Batrachochytrium dendrobatidis (see Fig. 2 for key to colours). Stars = ill and dead frogs positive for Bd in association with population declines (Qld ⁄ Aust. Alps). Grey lines = IBRA ecoregion boundaries (see Fig. S5 for map and key).

surveys, preventing further spread to naı¨ ve areas), for prioritizing monitoring programmes for Bd and Australia’s anuran fauna (e.g. Skerratt et al. 2008) and for identifying priority species for potential emergency captive-breeding programmes (Gascon et al. 2007) (Table S3). We envisage this to be an iterative process, with models such as ours regularly updated and scrutinized as new systematically collected data accrue (Wintle, Elith & Potts 2005). Critically, our methods can be directly and rapidly applied to other regions of the world experiencing amphibian declines; such results will aid in the task of developing informed management and surveillance decisions for Bd (Skerratt et al. 2008) and will help to make the most of limited conservation funds for prioritizing species, regions and actions for biodiversity conservation outcomes (Wilson et al. 2007).

LIMITATIONS AND FUTURE DIRECTIONS

While our model had high predictive performance and clamping indicated a well sampled environmental space, relatively lit-

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 163–173

Spatial patterns of disease risk 171 tle sampling has occurred on the western margins of the Great Dividing Range and in inland Australia, and Queensland and Western Australia were over represented compared with other regions. In addition, frogs may take refuge in environments that are not captured by interpolated bioclimatic or vegetation mapping data (see also Fig. S8) and we have limited ability to incorporate microclimatic features into our models given the enormous diversity of amphibian hosts and their habitats in this country. Similarly, beyond considerations of HPD (Fig. S8) we were unable to incorporate models of pathogen dispersal given very limited knowledge regarding how this pathogen is spread. We model the realized niche of this invasive species in an invaded range; there is thus the possibility that Bd’s distribution in Australia has not approached an equilibrium state, potentially resulting in an underprediction of its potential range. We consider this an unlikely source of major bias in our results given Bd’s extensive distribution nation-wide and the spectrum of potentially susceptible amphibian hosts (e.g. Litoria spp.) and hypothesized vectors (e.g. humans) in currently uninfected regions. Nevertheless, our model represents a baseline, minimum potential distribution rather than a ﬁnite prediction of this organism’s fundamental niche; we encourage scrutiny and ongoing iteration (e.g. integrated use of new systematically collected data), particularly to increase representation of apparently disease-free areas into future models. Dispersal models should also be a future priority, particularly in areas that are newly invaded. Finally, genetic diﬀerentiation has been noted geographically (Morgan et al. 2007; James et al. 2009), and strains may undergo local adaptation (Fisher et al. 2009) and ⁄ or show strain speciﬁc diﬀerences in adaptive plasticity so distribution in Australia with respect to available environmental space may not necessarily correspond exactly to other regions or to the results of other predictive models. Comparison of these and future studies will thus identify important areas and avenues for further research and it is imperative that the predictions of any SDM be independently compared with other SDM methods and data sources (see Elith et al. 2006), other methods (e.g. mechanistic models; K.A.M. unpubl. data) (Morin & Thuiller 2009) and by comprehensive ﬁeld surveys during sampling periods that maximize detection probability (Skerratt et al. 2008).

Acknowledgements We are indebted to the many authors and contributors named in Murray et al. (2010) for the production of the Bd occurrence data base. In particular, we thank K. McDonald, K. Aplin, H. Hines, D. Mendez, A. Felton, P. Kirkpatrick, D. Hunter, R. Campbell, M. Pauza, M. Driessen, S. Richards, M. Mahony, A. Freeman, A. Phillott, J-M. Hero, K. Kriger and D. Driscoll. KAM thanks D. Segan, M. Watts and C. Klein for GIS wisdom and spatial data, R. Wilson and H. Possingham for lab space, B. Sutherst, M. Zalucki and D. Kriticos for fruitful discussions and M. Araujo and R. Pearson for running a timely SDM workshop at the University of Queensland. We also thank Dr Marc Cadotte, Professor Christl Donnelly and three anonymous reviewers for excellent comments and discussion on earlier versions of the manuscript. KAM was supported by an Australian Postgraduate Award, an Australian Biosecurity CRC professional development award and a Wildlife Preservation Society of Australia student research award. Part of this work was conducted when RWRR was supported by the Australian Research Council, the School of Public Health and Tropical Medicine, James Cook University and a National Science Foundation Integrated Research Challenges in Environmen-

tal Biology grant awarded to J. Collins at Arizona State University, USA. RWRR thanks J. Collins and C. Carey.

References Adams, M.J., Chelgren, N.D., Reinitz, D., Cole, R.A., Rachowicz, L.J., Galvan, S., McCreary, B., Pearl, C.A., Bailey, L.L., Bettaso, J., Bull, E.L. & Leu, M. (2010) Using occupancy models to understand the distribution of an amphibian pathogen, Batrachochytrium dendrobatidis. Ecological Applications, 20, 289–302. Araujo, M.B. & Guisan, A. (2006) Five (or so) challenges for species distribution modelling. Journal of Biogeography, 33, 1677–1688. Berger, L. (2001) Diseases in Australian frogs. PhD, James Cook University, Townsville. Berger, L., Speare, R., Daszak, P., Green, D.E., Cunningham, A.A., Goggin, C.L., Slocombe, R., Ragan, M.A., Hyatt, A.D., McDonald, K.R., Hines, H.B., Lips, K.R., Marantelli, G. & Parkes, H. (1998) Chytridiomycosis causes amphibian mortality associated with population declines in the rain forests of Australia and Central America. Proceedings of the National Academy of Sciences of the United States of America, 95, 9031–9036. Berger, L., Speare, R., Hines, H.B., Marantelli, G., Hyatt, A.D., McDonald, K.R., Skerratt, L.F., Olsen, V., Clarke, J.M., Gillespie, G., Mahony, M., Sheppard, N., Williams, C. & Tyler, M.J. (2004) Eﬀect of season and temperature on mortality in amphibians due to chytridiomycosis. Australian Veterinary Journal, 82, 434–439. Bielby, J., Cooper, N., Cunningham, A., Garner, T. & Purvis, A. (2008) Predicting susceptibility to future declines in the world’s frogs. Conservation Letters, 1, 82–90. Briggs, C., Knapp, R.A. & Vredenburg, V.T. (2010) Enzootic and epizootic dynamics of the chytrid fungal pathogen of amphibians. Proceedings of the National Academy of Sciences of the United States of America, 107, 9695– 9700. Carey, C., Bruzgul, J.E., Livo, L.J., Walling, M.L., Kuehl, K.A., Dixon, B.F., Pessier, A.P., Alford, R.A. & Rogers, K.B. (2006) Experimental exposures of boreal toads (Bufo boreas) to a pathogenic chytrid fungus (Batrachochytrium dendrobatidis). EcoHealth, 3, 5–21. Crowl, T.A., Crist, T.O., Parmenter, R.R., Belovsky, G. & Lugo, A.E. (2008) The spread of invasive species and infectious disease as drivers of ecosystem change. Frontiers in Ecology and the Environment, 6, 238–246. Daszak, P., Cunningham, A. & Hyatt, A. (2000) Emerging infectious diseases of wildlife – threats to biodiversity and human health. Science, 287, 443–449. Digiacomo, R.F. & Koepsell, T.D. (1986) Sampling for detection of infection or disease in animal populations. Journal of the American Veterinary Medical Association, 189, 22–23. Dudik, M., Phillips, S.J. & Schapire, R.E. (2007) Maximum entropy density estimation with generalized regularization and an application to species distribution modeling. Journal of Machine Learning Research, 8, 1217–1260. Elith, J., Graham, C.H., Anderson, R.P., Dudik, M., Ferrier, S., Guisan, A., Hijmans, R.J., Huettmann, F., Leathwick, J.R., Lehmann, A., Li, J., Lohmann, L.G., Loiselle, B.A., Manion, G., Moritz, C., Nakamura, M., Nakazawa, Y., Overton, J.M., Peterson, A.T., Phillips, S.J., Richardson, K., Scachetti-Pereira, R., Schapire, R.E., Soberon, J., Williams, S., Wisz, M.S. & Zimmermann, N.E. (2006) Novel methods improve prediction of species’ distributions from occurrence data. Ecography, 29, 129–151. Ficetola, G.F., Thuiller, W. & Miaud, C. (2007) Prediction and validation of the potential global distribution of a problematic alien invasive species – the American bullfrog. Diversity and Distributions, 13, 476–485. Fielding, A.H. & Bell, J.F. (1997) A review of methods for the assessment of prediction errors in conservation presence ⁄ absence models. Environmental Conservation, 24, 38–49. Fisher, M.C., Bosch, J., Yin, Z., Stead, D.A., Walker, J., Selway, L., Brown, A.J.P., Walker, L.A., Gow, N.A.R., Stajich, J.E. & Garner, T.W.J. (2009) Proteomic and phenotypic proﬁling of the amphibian pathogen Batrachochytrium dendrobatidis shows that genotype is linked to virulence. Molecular Ecology, 18, 415–429. Gascon, C., Collins, J.P., Moore, R.D., Church, D.R., McKay, J.E. & Mendelson, J.R. III (eds) (2007) Amphibian Conservation Action Plan: Proceedings: IUCN ⁄ SSC Amphibian Conservation Summit 2005. The World Conservation Union (IUCN), Gland, Switzerland. Gibson, L., Barrett, B. & Burbidge, A. (2007) Dealing with uncertain absences in habitat modelling: a case study of a rare ground-dwelling parrot. Diversity and Distributions, 13, 704–713. Goka, K., Yokoyama, J., Une, Y., Kuroki, T., Suzuki, K., Nakahara, M., Kobayashi, A., Inaba, S., Mizutani, T. & Hyatt, A.D. (2009) Amphibian chy-

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 163–173

172 K. A. Murray et al. tridiomycosis in Japan: distribution, haplotypes and possible route of entry into Japan. Molecular Ecology, 18, 4757–4774. Graham, C.H., Elith, J., Hijmans, R.J., Guisan, A., Peterson, A.T., Loiselle, B.A. & Group, N.P.S.W. (2008) The inﬂuence of spatial errors in species occurrence data used in distribution models. Journal of Applied Ecology, 45, 239–247. Guisan, A. & Thuiller, W. (2005) Predicting species distribution: oﬀering more than simple habitat models. Ecology Letters, 8, 993–1009. Guisan, A., Broennimann, O., Engler, R., Vust, M., Yoccoz, N.G., Lehmann, A. & Zimmermann, N.E. (2006) Using niche-based models to improve the sampling of rare species. Conservation Biology, 20, 501–511. Hines, H., Mahony, M. & McDonald, K. (1999) An assessment of frog declines in wet subtropical Australia. Declines and Disappearances of Australian Frogs (eds A. Campbell), pp. 44–63. Environment Australia, Canberra. IUCN (2008) The IUCN red list of threatened species. IUCN. http:// www.iucnredlist.org, accessed July 2008. James, T.Y., Litvintseva, A.P., Vilgalys, R., Morgan, J.A.T., Taylor, J.W., Fisher, M.C., Berger, L., Weldon, C., du Preez, L. & Longcore, J.E. (2009) Rapid global expansion of the fungal disease chytridiomycosis into declining and healthy amphibian populations. PLoS Pathogens, 5, e1000458. Johnson, M.L. & Speare, R. (2003) Survival of Batrachochytrium dendrobatidis in water: quarantine and disease control implications. Emerging Infectious Diseases, 9, 922–925. Johnson, M.L., Berger, L., Philips, L. & Speare, R. (2003) Fungicidal eﬀects of chemical disinfectants, UV light, desiccation and heat on the amphibian chytrid Batrachochytrium dendrobatidis. Diseases of Aquatic Organisms, 57, 255–260. Jones, K.E., Patel, N.G., Levy, M.A., Storeygard, A., Balk, D., Gittleman, J.L. & Daszak, P. (2008) Global trends in emerging infectious diseases. Nature, 451, 990–994. Kearney, M. & Porter, W. (2009) Mechanistic niche modelling: combining physiological and spatial data to predict species’ ranges. Ecology Letters, 12, 334–350. Kirshtein, J.D., Anderson, C.W., Wood, J.S., Longcore, J.E. & Voytek, M.A. (2007) Quantitative PCR detection of Batrachochytrium dendrobatidis DNA from sediments and water. Diseases of Aquatic Organisms, 77, 11–15. Kriger, K.M. & Hero, J.M. (2007) Large-scale seasonal variation in the prevalence and severity of chytridiomycosis. Journal of Zoology, 271, 352–359. Kusrini, M.D., Skerratt, L.F., Garland, S., Berger, L. & Endarwin, W. (2008) Chytridiomycosis in frogs of Mount Gede Pangrango, Indonesia. Diseases of Aquatic Organisms, 82, 187–194. Lips, K.R., Brem, F., Brenes, R., Reeve, J.D., Alford, R.A., Voyles, J., Carey, C., Livo, L., Pessier, A.P. & Collins, J.P. (2006) Emerging infectious disease and the loss of biodiversity in a neotropical amphibian community. Proceedings of the National Academy of Sciences of the United States of America, 103, 3165–3170. Lips, K.R., Diﬀendorfer, J., Mendelson, J.R. & Sears, M.W. (2008) Riding the wave: reconciling the roles of disease and climate change in amphibian declines. PLoS Biology, 6, e72. Longcore, J.E., Pessier, A.P. & Nichols, D.K. (1999) Batrachochytrium dendrobatidis gen et sp nov, a chytrid pathogenic to amphibians. Mycologia, 91, 219–227. McDonald, K. & Alford, R. (1999) A review of declining frogs in Northern Queensland. Declines and Disappearances of Australian Frogs (eds A. Campbell), pp. 9–22. Environment Australia, Canberra. Morehouse, E.A., James, T.Y., Ganley, A.R.D., Vilgalys, R., Berger, L., Murphy, P.J. & Longcore, J.E. (2003) Multilocus sequence typing suggests the chytrid pathogen of amphibians is a recently emerged clone. Molecular Ecology, 12, 395–403. Morgan, J.A.T., Vredenburg, V.T., Rachowicz, L.J., Knapp, R.A., Stice, M.J., Tunstall, T., Bingham, R.E., Parker, J.M., Longcore, J.E., Moritz, C., Briggs, C.J. & Taylor, J.W. (2007) Population genetics of the frog-killing fungus Batrachochytrium dendrobatidis. Proceedings of the National Academy of Sciences of the United States of America, 104, 13845–13850. Morin, X. & Thuiller, W. (2009) Comparing niche- and process-based models to reduce prediction uncertainty in species range shifts under climate change. Ecology, 90, 1301–1313. Murray, B.R. & Hose, G.C. (2005) Life-history and ecological correlates of decline and extinction in the endemic Australian frog fauna. Austral Ecology, 30, 564–571. Murray, K.A., Skerratt, L.F., Speare, R. & McCallum, H. (2009) Impact and dynamics of disease in species threatened by the amphibian chytrid fungus, Batrachochytrium dendrobatidis. Conservation Biology, 23, 1242–1252. Murray, K.A., Retallick, R., McDonald, K., Mendez, D., Aplin, K., Kirkpatrick, P. et al. (2010) The distribution and host range of the pandemic disease

chytridiomycosis in Australia spanning surveys from 1956 to 2007. Ecology, 91, 1557. Murray, K.A., Rosauer, D., McCallum, H. & Skerratt, L.F. (in press) Integrating species traits with extrinsic threats: closing the gap between predicting and preventing species declines. Proceedings of the Royal Society B-Biological Sciences. Muths, E., Pilliod, D.S. & Livo, L.J. (2008) Distribution and environmental limitations of an amphibian pathogen in the Rocky Mountains, USA. Biological Conservation, 141, 1484–1492. Olson, D. & Ronnenberg, K. (2008) Global Bd Mapping Project: Update – 17 January 2008. Available from http://www.parcplace.org/bdmap2008update. html (accessed January 2008). USDA Forest Service, Paciﬁc Northwest Research Station, Corvallis, Oregon. Osborne, W., Hunter, D. & Hollis, G. (1999) Population declines and range contraction in Australian alpine frogs. Declines and Disappearances of Australian Frogs (eds A. Campbell), pp. 145–157. Environment Australia, Canberra. Parolo, G., Rossi, G. & Ferrarini, A. (2008) Toward improved species niche modelling: Arnica montana in the Alps as a case study. Journal of Applied Ecology, 45, 1410–1418. Pauza, M., Driessen, M. & Skerratt, L.F. (2010) Distribution and risk factors for spread of amphibian chytrid fungus Batrachochytrium dendrobatidis in the Tasmanian Wilderness World Heritage Area, Australia. Diseases of Aquatic Organisms, published online September 23 2010, 1–7. Pearce, J.L. & Boyce, M.S. (2006) Modelling distribution and abundance with presence-only data. Journal of Applied Ecology, 43, 405–412. Pearson, R.G., Raxworthy, C.J., Nakamura, M. & Peterson, A.T. (2007) Predicting species distributions from small numbers of occurrence records: a test case using cryptic geckos in Madagascar. Journal of Biogeography, 34, 102– 117. Peterson, A.T., Papes, M. & Eaton, M. (2007) Transferability and model evaluation in ecological niche modeling: a comparison of GARP and Maxent. Ecography, 30, 550–560. Phillips, S.J., Anderson, R.P. & Schapire, R.E. (2006) Maximum entropy modeling of species geographic distributions. Ecological Modelling, 190, 231–259. Phillips, S.J., Dudik, M., Elith, J., Graham, C.H., Lehmann, A., Leathwick, J. & Ferrier, S. (2009) Sample selection bias and presence-only distribution models: implications for background and pseudo-absence data. Ecological Applications, 19, 181–197. Phillott, A.D., Speare, R., Hines, H.B., Meyer, E., Skerratt, L.F., McDonald, K.R., Cashins, S.D., Mendez, D. & Berger, L. (2010) Minimising exposure of amphibians to pathogens during ﬁeld studies. Diseases of Aquatic Organisms, in press. doi: 10.3354/dao02162. Pilliod, D.S., Muths, E., Scherer, R.D., Bartelt, P.E., Corn, P.S., Hossack, B.R., Lambert, B.A., McCaﬀery, R. & Gaughan, C. (2010) Eﬀects of amphibian chytrid fungus on individual survival probability in wild boreal toads. Conservation Biology, 24, 1259–1267. Piotrowski, J.S., Annis, S.L. & Longcore, J.E. (2004) Physiology of Batrachochytrium dendrobatidis, a chytrid pathogen of amphibians. Mycologia, 96, 9– 15. Puschendorf, R., Carnaval, A.C., VanDerWal, J., Zumbado-Ulate, H., Chaves, G., Bolanos, F. & Alford, R.A. (2009) Distribution models for the amphibian chytrid Batrachochytrium dendrobatidis in Costa Rica: proposing climatic refuges as a conservation tool. Diversity and Distributions, 15, 401–408. Rachowicz, L.J., Hero, J.M., Alford, R.A., Taylor, J.W., Morgan, J.A.T., Vredenburg, V.T., Collins, J.P. & Briggs, C.J. (2005) The novel and endemic pathogen hypotheses: competing explanations for the origin of emerging infectious diseases of wildlife. Conservation Biology, 19, 1441–1448. Rohr, J.R., Raﬀel, T.R., Romansic, J.M., McCallum, H. & Hudson, P.J. (2008) Evaluating the links between climate, disease spread, and amphibian declines. Proceedings of the National Academy of Sciences of the United States of America, 105, 17436–17441. Ron, S.R. (2005) Predicting the distribution of the amphibian pathogen Batrachochytrium dendrobatidis in the New World. Biotropica, 37, 209–221. Skerratt, L.F., Berger, L., Speare, R., Cashins, S., McDonald, K.R., Phillott, A.D., Hines, H.B. & Kenyon, N. (2007) Spread of chytridiomycosis has caused the rapid global decline and extinction of frogs. EcoHealth, 4, 125– 134. Skerratt, L.F., Berger, L., Hines, H.B., McDonald, K.R., Mendez, D. & Speare, R. (2008) Survey protocol for detecting chytridiomycosis in all Australian frog populations. Diseases of Aquatic Organisms, 80, 85–94. Slatyer, C., Rosauer, D. & Lemckert, F. (2007) An assessment of endemism and species richness patterns in the Australian Anura. Journal of Biogeography, 34, 583–596.

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 163–173

Spatial patterns of disease risk 173 Speare, R. & Berger, L. (2000) Global Distribution of Chytridiomycosis in Amphibians. Amphibian Diseases Research Group, Townsville. http:// www.jcu.edu.au/shool/phtm/PHTM/frogs/chyglob.htm, accessed July 2007. Stuart, S.N., Chanson, J.S., Cox, N.A., Young, B.E., Rodrigues, A.S.L., Fischman, D.L. & Waller, R.W. (2004) Status and trends of amphibian declines and extinctions worldwide. Science, 306, 1783–1786. VanDerWal, J., Shoo, L.P., Graham, C. & William, S.E. (2009a) Selecting pseudo-absence data for presence-only distribution modeling: how far should you stray from what you know? Ecological Modelling, 220, 589– 594. VanDerWal, J., Shoo, L.P., Johnson, C.N. & Williams, S.E. (2009b) Abundance and the environmental niche: environmental suitability estimated from niche models predicts the upper limit of local abundance. American Naturalist, 174, 282–291. Voyles, J., Berger, L., Young, S., Speare, R., Webb, R., Warner, J., Rudd, D., Campbell, R. & Skerratt, L.F. (2007) Electrolyte depletion and osmotic imbalance in amphibians with chytridiomycosis. Diseases of Aquatic Organisms, 77, 113–118. Wake, D.B. & Vredenburg, V.T. (2008) Are we in the midst of the sixth mass extinction? A view from the world of amphibians. Proceedings of the National Academy of Sciences of the United States of America, 105, 11466– 11473. Walker, S.F., Salas, M.B., Jenkins, D., Garner, T.W.J., Cunningham, A.A., Hyatt, A.D., Bosch, J. & Fisher, M.C. (2007) Environmental detection of Batrachochytrium dendrobatidis in a temperate climate. Diseases of Aquatic Organisms, 77, 105–112. Wilson, K.A., Underwood, E.C., Morrison, S.A., Klausmeyer, K.R., Murdoch, W.W., Reyers, B., Wardell-Johnson, G., Marquet, P.A., Rundel, P.W., McBride, M.F., Pressey, R.L., Bode, M., Hoekstra, J.M., Andelman, S., Looker, M., Rondinini, C., Kareiva, P., Shaw, M.R. & Possingham, H.P. (2007) Conserving biodiversity eﬃciently: what to do, where, and when. PLoS Biology, 5, 1850–1861. Wintle, B.A., Elith, J. & Potts, J.M. (2005) Fauna habitat modelling and mapping: a review and case study in the Lower Hunter Central Coast region of NSW. Austral Ecology, 30, 719–738. Wisz, M.S., Hijmans, R.J., Li, J., Peterson, A.T., Graham, C.H., Guisan, A. & NCEAS (2008) Eﬀects of sample size on the performance of species distribution models. Diversity and Distributions, 14, 763–773.

Woodhams, D.C. & Alford, R.A. (2005) Ecology of chytridiomycosis in rainforest stream frog assemblages of tropical Queensland. Conservation Biology, 19, 1449–1459. Woodhams, D.C., Alford, R.A. & Marantelli, G. (2003) Emerging disease of amphibians cured by elevated body temperature. Diseases of Aquatic Organisms, 55, 65–67. Woodhams, D.C., Alford, R.A., Briggs, C.J., Johnson, M. & Rollins-Smith, L.A. (2008) Life-history trade-oﬀs inﬂuence disease in changing climates: strategies of an amphibian pathogen. Ecology, 89, 1627–1639. World Organisation for Animal Health (2008) Aquatic Animal Code 2008. World Organisation for Animal Health, Paris. Available from http:// www.oie.int/eng/normes/fcode/en_sommaire.htm (accessed July 2008). Ziller, M., Selhorst, T., Teuﬀert, J., Kramer, M. & Schluter, H. (2002) Analysis of sampling strategies to substantiate freedom from disease in large areas. Preventive Veterinary Medicine, 52, 333–343. Received 31 May 2010; accepted 28 September 2010 Handling Editor: Marc Cadotte

Supporting Information Additional Supporting Information may be found in the online version of this article: Appendix S1. Supporting Information (Tables S1–S4, Figs S1–S8). Appendix S2. Results of the model (Bd_in_Australia.asc). As a service to our authors and readers, this journal provides supporting information supplied by the authors. Such materials may be re-organized for online delivery, but are not copy-edited or typeset. Technical support issues arising from supporting information (other than missing ﬁles) should be addressed to the authors.

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 163–173

Journal of Applied Ecology 2011, 48, 174–176

doi: 10.1111/j.1365-2664.2010.01891.x

FORUM

Modelling the future distribution of the amphibian chytrid fungus: the inﬂuence of climate and human-associated factors Jason R. Rohr*, Neal T. Halstead and Thomas R. Raffel University of South Florida, Department of Integrative Biology, Tampa, FL 33620, USA

Summary 1. Many of the global losses of amphibians are believed to be caused by the chytrid fungus, Batrachochytrium dendrobatidis (Bd). Hence, determining its present and future environmental suitability should help to inform management and surveillance of this pathogen and curtail the amphibian biodiversity crisis. 2. In this issue of Journal of Applied Ecology, Murray et al. (2011) oﬀer an important step in this direction by providing a species distribution model that projects the environmental suitability of Bd across Australia and predicts locations of chytridiomycosis and amphibian declines. Batrachochytrium dendrobatidis presence was predicted by diurnal temperature range (a measure of temperature variability) and mean precipitation. Human population density, a positive predictor of Bd, accounted for the most variation when removed from the statistical model. 3. This work represents an invaluable case study and has great potential for managing chytridiomycosis and associated amphibian declines, but its value in practice will depend on how well managers understand the limitations of species distribution models. 4. Synthesis and applications. To improve the management of chytridiomycosis, amphibian-chytrid research should attempt to understand how humans may aﬀect the distribution of Bd, how climatic means and variances aﬀect Bd transmission, how much variation in the distribution of Bd is unique to and shared among climate, human, and other factors, whether human-related factors and climate statistically interact, and how these potentially correlated factors and any interactions aﬀect the predictability of species distribution models. In response to the swift spread of Bd and our rapidly changing planet, we encourage the application of Bd distribution models to other regions of the globe and predictions of Bd’s distribution under future climate change scenarios. Key-words: Batrachochytrium dendrobatidis, bioclimatic envelope models, biotic interactions, chytridiomycosis, climate change, disease, dispersal, diurnal temperature range, management, species distribution models

Introduction Alarmingly, almost a third of amphibian species are considered threatened and more than 43% are experiencing some form of population decline (Stuart et al. 2004). Many of these amphibian losses are believed to be driven by the disease chytridiomycosis, caused by the pathogenic chytrid fungus Batrachochytrium dendrobatidis (Bd; Wake & Vredenburg 2008). Given that Bd may be spreading and ⁄ or emerging (Rohr et al. 2008; James et al. 2009) and that climate is changing, one of *Correspondence author. E-mail: [email protected] Conﬂicts of interest: No conﬂicts declared

the priorities for managing this devastating pathogen is to determine its present and future environmental limitations at global and local scales. If this can be done accurately, it might help to predict the future distribution of Bd, identify where Bd poses the greatest future threats, and facilitate prioritization of species and locations for monitoring and management given scarce conservation funds. In this issue of Journal of Applied Ecology, Murray et al. (2011) oﬀer an important step in this direction by providing a species distribution model that projects the environmental suitability for Bd across the entire continent of Australia. Murray et al. (2011) base their models on an impressive spatiotemporal database consisting of 821 sites in Australia where

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society

Future distribution of the amphibian chytrid 175 115 amphibian species and 10 183 individuals were tested for Bd from 1956–2007. Previous large-scale databases associated with Bd-related declines have proven to be useful in understanding the biology of this pathogen (Lips et al. 2008; Rohr et al. 2008; Rohr & Raﬀel 2010), and there is little question that this new database will also be valuable for addressing the amphibian biodiversity crisis and questions of disease ecology in general. From 19 climatic variables, as well as information on distance to water, vegetation type, and human population density, Murray et al. (2011) identiﬁed annual precipitation and diurnal temperature range (a measure of temperature variation) as important predictors of Bd presence. Moisture has been well established as a crucial factor for Bd persistence (Johnson et al. 2003; Berger et al. 2004). Evidence for the importance of diurnal temperature range is interesting in light of a recent study that revealed that temperature variability, in general, and diurnal temperature range speciﬁcally, might drive amphibian declines putatively caused by chytridiomycosis (Rohr & Raﬀel 2010). Using recent advances in bioclimatic envelope modelling (Phillips, Anderson & Schapire 2006), Murray et al. (2011) estimated the range of environments suitable for Bd (its bioclimatic envelope) and then predicted its minimum potential geographic distribution across Australia. Importantly, sites with documented amphibian declines associated with severe chytridiomycosis had high Bd environmental suitability values, and environmental suitability values speciﬁc to each of Australia’s 196 amphibian species were signiﬁcant positive predictors of whether species experienced declines (even after controlling for species’ range sizes). Hence, the developed species distribution model appears to be capable of predicting locations of high risk for both chytridiomycosis and amphibian losses. Some caution should be used interpreting these correlations, however, because sampling for Bd was likely biased towards locations where amphibians were in decline and ⁄ or showing signs of chytridiomycosis, so environmental suitability estimates might also be biased toward these locations. Finally, in an eﬀort to facilitate and target management and monitoring, Murray et al. (2011) importantly identiﬁed amphibian species that have high environmental suitability scores for Bd and regions of Australia that have both high suitability scores and high amphibian species richness and endemism. Although this work clearly has great potential for managing chytridiomycosis and associated amphibian declines, its value in practice will depend on how well managers understand the limitations of bioclimatic envelope models (BEMs) and how well the assumptions of these models were met by this study. BEMs are strictly correlational and do not directly model abiotic and biotic interactions, dispersal, or evolution, all of which can be important for predicting the eﬀects of climate change on biodiversity (Davis et al. 1998; Rohr & Madison 2003; Araujo & Luoto 2007; Harmon, Moran & Ives 2009). Further, BEMs often have considerable uncertainty despite systematically overestimating model ﬁts during model validation (Hampe 2004). The over-ﬁtting occurs because BEMs generally do not account for spatial autocorrelation

among their data (a pseudoreplication issue; Segurado, Araujo & Kunin 2006). Finally, the results of BEMs can be quite sensitive to the model parameterization and model selection procedures that were implemented (Araujo & Guisan 2006). Ideally, several parameterization and selection procedures should be used to evaluate the robustness of the BEM results. Nevertheless, BEMs provide a useful ﬁrst approximation and working hypothesis when identifying a species’ future distribution (Pearson & Dawson 2003). These approximations should be improved upon with additional data and adaptive management approaches, as advocated by Murray et al. (2011). Murray et al. (2011) also oﬀer some important insights into how their model might be improved upon. Intriguingly, Murray et al. (2011) discovered that human population density (HPD) was a positive predictor of the presence of Bd, with Bd almost exclusively being found near port cities and the highways connecting them (their Fig. S1). Furthermore, the removal of HPD from their statistical model resulted in the greatest change in variation relative to all other predictors (their Fig. S8), indicating that HPD accounted for the greatest unique variation in the distribution of Bd. While it is possible that the correlation is driven by humans and Bd simply preferring the same climate, this seems unlikely given that the eﬀect of humans was still evident after accounting for variation due to the climatic factors. Hence, the relationship between HPD and Bd presence suggests that, in addition to climate, human-associated factors might aﬀect the distribution of this pathogen. This is not surprising given that humans are believed to be a major dispersal agent for Bd (Skerratt et al. 2007), a hypothesis that received support from two molecular studies concluding that the distribution of Bd was consistent with human-assisted migration (Morgan et al. 2007; James et al. 2009), and from a survey in Oregon and Northern California revealing that detectability of Bd increased markedly with human inﬂuence on the landscape (Adams et al. 2010). However, if humans are regularly introducing Bd into areas of low environmental suitability, this could violate the underlying assumption of the BEM, that the climatic conditions where Bd is presently found will either match the conditions of its future range or at least be adequate surrogates for the factors that dictate its future distribution (Pearson & Dawson 2003). Research on source-sink dynamics and metapopulations, and more recently neutral theory, has shown that organisms can often appear in suboptimal habitats when immigration is high (Davis et al. 1998; Hanski 1998; Hubbell 2005). The fact that Bd is predominantly found in and around coastal cities might further exacerbate this concern because the coastline essentially functions as a giant drift fence, forcing dispersal along and away from the coast. If there are multiple introductions at nearby port cities and crossing waves of dispersal along the coast, then BEMs will provide greater weight to these coastal regions, even if they do not represent optimal Bd conditions. It remains to be seen how much of a concern this is to the reliability of Murray et al.’s (2011) predictions, but the potential inﬂuence of dispersal and human-associated factors point to some major challenges for future BEMs and amphibianchytrid work.

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 174–176

176 J. R. Rohr, N. T. Halstead & T. R. Raﬀel Future research on BEMs should evaluate whether accounting for spatial autocorrelation using traditional approaches adequately accounts for dispersal barriers and limitations and spread from known or presumed introduction sites. This is particularly important given that the goal of BEMs, in many cases, is to model invasive species, which are often introduced at locations of high human population density, such as at cities with ports and airports. Amphibian-chytrid research should attempt to understand how humans aﬀect the distribution of Bd; how much variation in the distribution of Bd is unique to and shared among climate, human, and other factors; whether human-related factors and climate interact statistically; and how these correlated factors and any interactions aﬀect the predictability of BEMs. Finally, amphibian-chytrid research should better validate BEMs by determining how accurately they predict the spread of Bd. Despite their limitations, BEMs represent an important tool for predicting the future distributions of species and Murray et al. (2011) provide an invaluable case study that should guide others in applying these tools worldwide. Undoubtedly, the reliability of these models for predicting the distribution of Bd will improve with a better understanding of the factors that dictate the persistence and transmission of this pathogen. In response to the swift spread of Bd (Skerratt et al. 2007; Rohr et al. 2008) and our rapidly changing planet, we encourage the application of BEMs to other regions of the globe and the use of BEMs and ensemble climate models to predict the distribution of Bd under future climate change scenarios (e.g. Lawler et al. 2009). Global climate change is creating a climatically more variable world (Raisanen 2002) and thus research must consider how changes to both the mean and variance of climatic variables aﬀect Bd-amphibian interactions and species interactions, in general (Raﬀel et al. 2006; Rohr & Raﬀel 2010). These proposed eﬀorts should help inform management and surveillance, and will hopefully curtail our amphibian biodiversity crisis.

Acknowledgements We thank Marc Cadotte for providing us with the opportunity to write this commentary and for discussing with us the reviewers’ praise and comments on this paper. Funds were provided by a National Science Foundation (DEB 0516227) grant to J.R.R., US Department of Agriculture (NRI 2006-01370, 2009-35102-05043) grants to J.R.R., and a US Environmental Protection Agency STAR grant to J.R.R. and T.R.R. (R833835).

References Adams, M.J., Chelgren, N.D., Reinitz, D., Cole, R.A., Rachowicz, L.J., Galvan, S., McCreary, B., Pearl, C.A., Bailey, L.L., Bettaso, J., Bull, E.L. & Leu, M. (2010) Using occupancy models to understand the distribution of an amphibian pathogen, Batrachochytrium dendrobatidis. Ecological Applications, 20, 289–302. Araujo, M.B. & Guisan, A. (2006) Five (or so) challenges for species distribution modelling. Journal of Biogeography, 33, 1677–1688. Araujo, M.B. & Luoto, M. (2007) The importance of biotic interactions for modelling species distributions under climate change. Global Ecology and Biogeography, 16, 743–753. Berger, L., Speare, R., Hines, H.B., Marantelli, G., Hyatt, A.D., McDonald, K.R., Skerratt, L.F., Olsen, V., Clarke, J.M., Gillespie, G., Mahony, M., Sheppard, N., Williams, C. & Tyler, M.J. (2004) Eﬀect of season and

temperature on mortality in amphibians due to chytridiomycosis. Australian Veterinary Journal, 82, 434–439. Davis, A.J., Jenkinson, L.S., Lawton, J.H., Shorrocks, B. & Wood, S. (1998) Making mistakes when predicting shifts in species range in response to global warming. Nature, 391, 783–786. Hampe, A. (2004) Bioclimate envelope models: what they detect and what they hide. Global Ecology and Biogeography, 13, 469–471. Hanski, I. (1998) Metapopulation dynamics. Nature, 396, 41–49. Harmon, J.P., Moran, N.A. & Ives, A.R. (2009) Species response to environmental change: impacts of food web interactions and evolution. Science, 323, 1347–1350. Hubbell, S.P. (2005) Neutral theory in community ecology and the hypothesis of functional equivalence. Functional Ecology, 19, 166–172. James, T.Y., Litvintseva, A.P., Vilgalys, R., Morgan, J.A.T., Taylor, J.W., Fisher, M.C., Berger, L., Weldon, C., du Preez, L. & Longcore, J.E. (2009) Rapid global expansion of the fungal disease chytridiomycosis into declining and healthy amphibian populations. PLoS Pathogens, 5, e1000458. Johnson, M.L., Berger, L., Philips, L. & Speare, R. (2003) Fungicidal eﬀects of chemical disinfectants, UV light, desiccation and heat on the amphibian chytrid Batrachochytrium dendrobatidis. Diseases of Aquatic Organisms, 57, 255–260. Lawler, J.J., Shafer, S.L., White, D., Kareiva, P., Maurer, E.P., Blaustein, A.R. & Bartlein, P.J. (2009) Projected climate-induced faunal change in the Western Hemisphere. Ecology, 90, 588–597. Lips, K.R., Diﬀendorfer, J.E., Mendelson, J.R. & Sears, M.W. (2008) Riding the wave: reconciling the roles of disease and climate change in amphibian declines. PLoS Biology, 6, 441–454. Morgan, J.A.T., Vredenburg, V.T., Rachowicz, L.J., Knapp, R.A., Stice, M.J., Tunstall, T., Bingham, R.E., Parker, J.M., Longcore, J.E., Moritz, C., Briggs, C.J. & Taylor, J.W. (2007) Population genetics of the frog-killing fungus Batrachochytrium dendrobatidis. Proceedings of the National Academy of Sciences of the United States of America, 104, 13845–13850. Murray, K.A., Retallick, R.W.R., Puschendorf, R., Skerratt, L.F., Rosauer, D., McCallum, H.I., Berger, L., Speare, R. & VanDerWal, J. (2011) Assessing spatial patterns of disease risk to biodiversity: implications for the management of the amphibian pathogen, Batrachochytrium dendrobatidis. Journal of Applied Ecology, 48, 163–173. Pearson, R.G. & Dawson, T.P. (2003) Predicting the impacts of climate change on the distribution of species: are bioclimate envelope models useful? Global Ecology and Biogeography, 12, 361–371. Phillips, S.J., Anderson, R.P. & Schapire, R.E. (2006) Maximum entropy modeling of species geographic distributions. Ecological Modelling, 190, 231–259. Raﬀel, T.R., Rohr, J.R., Kiesecker, J.M. & Hudson, P.J. (2006) Negative eﬀects of changing temperature on amphibian immunity under ﬁeld conditions. Functional Ecology, 20, 819–828. Raisanen, J. (2002) CO2-induced changes in interannual temperature and precipitation variability in 19 CMIP2 experiments. Journal of Climate, 15, 2395–2411. Rohr, J.R. & Madison, D.M. (2003) Dryness increases predation risk in efts: support for an amphibian decline hypothesis. Oecologia, 135, 657–664. Rohr, J.R. & Raﬀel, T.R. (2010) Linking global climate and temperature variability to widespread amphibian declines putatively caused by disease. Proceedings of the National Academy of Sciences of the United States of America, 107, 8269–8274. Rohr, J.R., Raﬀel, T.R., Romansic, J.M., McCallum, H. & Hudson, P.J. (2008) Evaluating the links between climate, disease spread, and amphibian declines. Proceedings of the National Academy of Sciences of the United States of America, 105, 17436–17441. Segurado, P., Araujo, M.B. & Kunin, W.E. (2006) Consequences of spatial autocorrelation for niche-based models. Journal of Applied Ecology, 43, 433– 444. Skerratt, L.F., Berger, L., Speare, R., Cashins, S., McDonald, K.R., Phillott, A.D., Hines, H.B. & Kenyon, N. (2007) Spread of chytridiomycosis has caused the rapid global decline and extinction of frogs. EcoHealth, 4, 125–134. Stuart, S.N., Chanson, J.S., Cox, N.A., Young, B.E., Rodrigues, A.S.L., Fischman, D.L. & Waller, R.W. (2004) Status and trends of amphibian declines and extinctions worldwide. Science, 306, 1783–1786. Wake, D.B. & Vredenburg, V.T. (2008) Are we in the midst of the sixth mass extinction? A view from the world of amphibians. Proceedings of the National Academy of Sciences of the United States of America, 105, 11466– 11473. Received 10 September 2010; accepted 29 September 2010 Handling Editor: Marc Cadotte

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 174–176

Journal of Applied Ecology 2011, 48, 86–95

doi: 10.1111/j.1365-2664.2010.01893.x

Optimizing search strategies for invasive pests: learn before you leap Peter W.J. Baxter* and Hugh P. Possingham The University of Queensland, The Ecology Centre and Centre for Applied Environmental Decision Analysis, School of Biological Sciences, St. Lucia, Qld 4072, Australia

Summary 1. Strategic searching for invasive pests presents a formidable challenge for conservation managers. Limited funding can necessitate choosing between surveying many sites cursorily, or focussing intensively on fewer sites. While existing knowledge may help to target more likely sites, e.g. with species distribution models (maps), this knowledge is not ﬂawless and improving it also requires management investment. 2. In a rare example of trading-oﬀ action against knowledge gain, we combine search coverage and accuracy, and its future improvement, within a single optimization framework. More speciﬁcally we examine under which circumstances managers should adopt one of two search-and-control strategies (cursory or focussed), and when they should divert funding to improving knowledge, making better predictive maps that beneﬁt future searches. 3. We use a family of Receiver Operating Characteristic curves to reﬂect the quality of maps that direct search eﬀorts. We demonstrate our framework by linking these to a logistic model of invasive spread such as that for the red imported ﬁre ant Solenopsis invicta in south-east Queensland, Australia. 4. Cursory widespread searching is only optimal if the pest is already widespread or knowledge is poor, otherwise focussed searching exploiting the map is preferable. For longer management timeframes, eradication is more likely if funds are initially devoted to improving knowledge, even if this results in a short-term explosion of the pest population. 5. Synthesis and applications. By combining trade-oﬀs between knowledge acquisition and utilization, managers can better focus – and justify – their spending to achieve optimal results in invasive control eﬀorts. This framework can improve the eﬃciency of any ecological management that relies on predicting occurrence. Key-words: adaptive management, containment, eradication, invasive species, optimal management, receiver operating characteristic (ROC) curve, species distribution models, statedependent management, stochastic dynamic programming (SDP), value of information

Introduction Invasive species comprise one of the main threats to global biodiversity (Sala et al. 2000) and their annual economic impact is substantial (Pimentel et al. 2001). While considerable economic resources can be allocated to invasive species management, it is important to strategise spending in a coherent decision-making framework, to maintain cost-eﬃciency as well as increase the likelihood of programme success (Regan et al. 2006; Bogich & Shea 2008). Such a framework should ideally take account of all economic factors in the programme, includ-

*Correspondence author. E-mail: [email protected]

ing investing in knowledge acquisition to improve future management. This notion of learning now, to make better decisions later, underpins adaptive management (Walters 1986) and theories of learning in animal behaviour (Stephens & Krebs 1986). In this paper we investigate how best to allocate a restricted budget among options for research and control of an invasive pest when we have some information about its distribution, as well as the ability to improve that information. For any detection-and-control programme constrained by time, budget and human resources, trade-oﬀs exist between the diﬀerent search strategies and the acquisition of information to inform future searches. Therefore managers are confronted with the following questions: How many sites should we search, which ones, and at what intensity? Should we invest

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society

Optimizing search strategies for invasives 87 resources in reﬁning our methods for choosing sites, to improve future search success? The ﬁrst questions have been addressed to some extent in the literature. For example, controlling new satellite populations may be preferable to reducing the density of the core pest population (Moody & Mack 1988), although this may not be optimal when costs are taken into account (Whittle, Lenhart & Gross 2007). For pest metapopulations, it may be optimal to attempt eradication of mediumdensity sub-populations, while still balancing colonization from, and containment of, high-density patches (Bogich & Shea 2008). With enough money, however, prioritizing highdensity patches can become optimal (Taylor & Hastings 2004). Within a patch, detection probability depends on both the search method and search intensity or coverage, which in turn depend on budget (Cacho, Hester & Spring 2007). The tradeoﬀ between cursory and focussed searching has not yet been examined in the context of (mis-) information about the distribution of the species. Predicting a pest’s distribution and spread often involves a species distribution model of its likely occurrence (e.g. Baret et al. 2006; Hauser & McCarthy 2009), allowing some degree of strategic searching within a region. The predictive accuracy of species distribution models is commonly assessed using Receiver Operating Characteristic (ROC) curves (Pearce & Ferrier 2000; Wintle, Elith & Potts 2005; Latimer et al. 2009), which graph the rates of occurrence of true vs. false positives. In species distribution models (henceforth, ‘maps’) this translates as rates of classifying occupied sites as being occupied, vs. rates of misclassifying unoccupied sites as occupied, implying when searches would be worthwhile or futile, respectively. As maps can be improved at some cost, namely redirecting funds from active control (Murray et al. 2009), optimal investment in research can be determined if we know the future beneﬁt of having better predictive maps that will result from the short term reduction in on-ground eﬀort. The trade-oﬀ of knowledge gain vs. immediate action is implicit in every area of applied ecology. Despite the practical beneﬁts and broader implications of exploring such a trade-oﬀ, however, this has not yet been done in a theoretical or practical framework. Although some cost-beneﬁt analysis has been applied to the use of ROC curves in clinical settings (Swets & Pickett 1982; Metz 1986), this has focussed on the direct costs of treating false-positive vs. true-positive diagnoses rather than deferring treatments while diagnostic tests are improved. As no map can predict a pest’s distribution with 100% precision, some unoccupied sites will inevitably be searched (Va´clavı´ k & Meentemeyer 2009). As the overall search area increases we expect to reach more of the occupied sites, but we also experience a concomitant increase in the proportion of empty sites that are searched. In the extreme case, exhaustive searching entails looking for the pest throughout the entire area of unsuitable habitat as well as in all the more likely sites. Therefore it may be better to select fewer sites and increase the search eﬀort at each site to increase the probability of detecting the pest, similar to intensive vs. extensive search modes of animal foraging behaviour (Fortin 2002). The search strategy for an invasive species will ideally incorporate some idea of how

accurate the distribution predictions are, and therefore what proportion of searches are likely to be futile due to an incorrect choice of site. Consequently it may even be beneﬁcial to redirect resources to reﬁning our knowledge of the organism’s expected distribution, to better identify candidate sites for future searches. We address these issues here by optimizing the trade-oﬀs among widespread and more focussed search areas (allowing low and high search intensity per site, respectively), against knowledge acquisition to improve future searches. We use stochastic dynamic programming (SDP), a procedure that identiﬁes optimal strategies by considering the possible changes in the states of a system over time (Bellman 1957). We also compare, by simulation under diﬀerent assumptions, the relative performances of the SDP recommendations and alternative management strategies. SDP is commonly applied in behavioural ecology (Mangel & Clark 1988; McNamara & Houston 1996), including examining when foraging organisms should learn about resource distribution by moving between patches (Eliassen et al. 2009). It is being used increasingly to solve state-dependent management problems in ecology: choosing between ﬁre management options (Richards, Possingham & Tizard 1999; McCarthy, Possingham & Gill 2001), how to allocate management eﬀort within or among sites (Baxter et al. 2007; McDonald-Madden, Baxter & Possingham 2008), and when to cease management or monitoring altogether (Regan et al. 2006; Chade`s et al. 2008). We frame our analysis around the invasion of red imported ﬁre ants Solenopsis invicta Buren in the Brisbane, Queensland, region, dating from February 2001 (Jennings & McCubbin 2004). Native to South America, their establishment as an invasive alien species is greatly facilitated by anthropogenic disturbance (King & Tschinkel 2008), and they are capable of considerable environmental, social and economic damage (Williams 1994; Callcott & Collins 1996). The ﬁre ant incursion into Australia is therefore potentially very serious, given extensive suitable habitat (Moloney & Vanderwoude 2002; Sutherst & Maywald 2005), and the invasion has been listed as a Key Threatening Process under the 1999 Commonwealth Environment Protection and Biodiversity Conservation Act. The Queensland invasion dynamics have been modelled previously (Scanlan & Vanderwoude 2006), allowing reasonable biological parameterisation for our purposes. In the absence of detailed management cost data, however, we keep our approach general and demonstrate the method in a form broadly applicable to invasive species management (and indeed to applied ecology in general), rather than presenting a speciﬁc case analysis of the ﬁre-ant invasion. This novel approach shows how current and future prediction capability should aﬀect current and future search strategies to optimize invasive species control and planning.

Materials and methods In order to model the maps’ predictive quality, we use a family of ROC curves. In practice, ROC curves can take any shape between (0,0) and (1,1), with one measure of map quality being the area under

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 86–95

88 P. W. J. Baxter & H. P. Possingham the curve (AUC); the closer AUC gets to one, the better the map (Swets 1988; but see Lobo, Jime´nez-Valverde & Real 2008). We assume a family of ROC curves given by y ¼ x1=a

eqn 1

(Fig. 1a; after Swets 1986), and bounded at (0,0) and (1,1) as required. For this equation, the area under the curve takes a simple form, AUC = a ⁄ (a + 1), which asymptotically approaches one as a increases. Therefore a higher value of a implies a more reliable map (a = 1 essentially implies no knowledge and random searches).

SEARCH STRATEGY AND SUCCESS

We assume that a pest is present at some density / within a region of area A (we list symbols and parameters in Table 1). In a map, A can be measured as the number of cells in a grid covering the region (i.e.

number of sites in which the pest may potentially occur), and / as the proportion of those cells that are infested. The map directing our searches will produce sites labelled as occupied, either correctly (in eqn 1, proportion y of occupied sites) or incorrectly (proportion x of empty sites). When we use the map to choose sA sites to search (a proportion s of the region), eqn 1 allows us to express this in terms of the sites searched that are either occupied (/Ay) or unoccupied ([1–/]Ax): sA ¼ /Ay þ ð1 /ÞAx ) s ¼ /y þ ð1 /Þya :

The value of s increases with y, A and / (Fig. 1b). We can use eqn 2 to ﬁnd the proportion of a region, s1, needed to be searched in order to visit some proportion y1 of the occupied sites. Alternatively we could search fewer sites (proportion s2 < s1), spending longer in each, giving:

0·8

800

Number of sites to search

(b) 1000

Proportion of true positives

(a) 1·0

0·6

0·4 a = 25 a = 10 a= 5 a= 2 a= 1

0·2

0

0

0·2

0·4

0·6

0·8

600

400

200

0

1·0

φ = 0·1; a = 2 φ = 0·1; a = 5 φ = 0·1; a = 8 φ = 0·3; a = 2 φ = 0·3; a = 5 φ = 0·3; a = 8 φ = 0·5; a = 2 φ = 0·5; a = 5 φ = 0·5; a = 8

0

0·2

Search success; a = 2

Expected no· colonies found

Expected no· colonies found

200

150

100

50

0

0

0·2

0·4

0·6

0·6

0·8

1·0

Search success; a = 8 (d) 300

$50 000; cursory $100 000; cursory $200 000; cursory $50 000; focussed $100 000; focussed $200 000; focussed

250

0·4

Proportion of true positives required

Proportion of false positives

(c) 300

eqn 2

0·8

Regional pest density (φ)

1·0

$50 000; cursory $100 000; cursory $200 000; cursory $50 000; focussed $100 000; focussed $200 000; focussed

250

200

150

100

50

0

0

0·2

0·4

0·6

0·8

1·0

Regional pest density (φ)

Fig. 1. Relationships underpinning the framework of active searching vs. knowledge acquisition. (a) Family of theoretical Receiver Operating Characteristic curves, described by y = x1 ⁄ a, where higher values of a reﬂect better predictive capability. Two example search strategies are indicated for a habitat map quality of a = 5. The cursory-widespread strategy (‘+’) attempts to ﬁnd 95% of the infestation (380 sites) and the focussed strategy (‘o’), visits half that number, allowing double the search time per site. (b) The total number of sites (empty and occupied) required to search, to visit given proportions of occupied sites, depending on regional infestation density, /, and quality of habitat map, a. (c, d) Expected number of colonies detected in one time step as a function of regional pest density, for three diﬀerent budgets and the two search strategies (cursory and focussed). Two levels of map quality are shown, (c) a = 2 and (d) a = 8. 2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 86–95

Optimizing search strategies for invasives 89 Table 1. Parameters used, with their symbols and values (where applicable; otherwise indicated as a variable or a function f(…) of other parameters)

Symbol

Description

Value

/ /0 kmax kd ku A B Di S95 a a0 d k s v x

proportion of region infested initial proportion of region infested in simulations maximum rate of invasion spread rate of spread due to detected colonies rate of spread due to uncontrolled colonies area of region (#grid cells in species distribution model) budget per time-step expected proportion of colonies detected with action i value of s giving 95% worthwhile searches ROC curve exponent initial ROC curve exponent in simulations probability of detecting a colony if present search eﬀort at which d = 0Æ5 proportion of region to be searched search eﬀort per site ($) proportion of all futile sites searched (‘false positive proportion’) proportion of all worthwhile sites searched (‘true positive proportion’)

variable 0Æ01 1Æ19 f(/), £ 1Æ09 f(/), £ 1Æ19 1000 $100,000 f(/,A,B,k,si,yi) f(/,a) variable 2 f(v,k) $500 f(y,/,a) f(A,B,s) 0–1

y

0–1, f(x,a)

ROC, Receiver Operating Characteristic.

s1 ¼ /y1 þ ð1 /Þy1 a and s2 ¼ /y2 þ ð1 /Þy2 a ;

d ¼ v=ðk þ vÞ; eqn 3

with y1 > y2 when s1 > s2 (Fig. 1b). For example we can set s1 = S95, which we deﬁne as the proportion of sites needed to include 95% of all occupied sites (y1 = 0Æ95): s1 ¼ S95 ¼ 095/ þ ð1 /Þ095a

eqn 4

(Fig. 1a, ‘+’). If we want to ﬁnd the ‘hit’ rate y2 for searching half those sites we can use the equation s2 ¼ 05S95 ¼ /y2 þ ð1 /Þy2 a ;

eqn 5

and (knowing S95 and estimating a and /) we can ﬁnd y2 numerically (Fig. 1a, ‘o’). For example, for a regional infestation density of / = 0Æ4, and map quality a = 5, we get S95 = 0Æ844, and y2 = 0Æ735 (giving s2 = 0Æ422 = 0Æ5S95). This means that, for an area of 1000 sites, in order to search in 380 sites that are occupied (95% of the 400 infestations), we need to search in 844 sites in total, as our map will misdirect us to 464 sites (so 55% of searches are futile – still better than random, which would lead to 60% of searches being futile). By halving the number of sites searched (422), our map would send us to 0Æ4(0Æ735) = 294 occupied sites, and leave us with 128 futile searches (30% of all searches). Therefore when we employ the map, reducing the total search coverage also reduces the proportion of searches that are futile. Furthermore, searching half the sites allows us to double our per-site search eﬀort, increasing the detection probability at occupied sites. Nonetheless, the increase in proportion of worthwhile searches and detection probability must be traded oﬀ against the reduction in overall search coverage.

PROBABILITY OF DETECTION

Detecting a pest at an occupied site is more likely if we expend more eﬀort searching the site. Assume the probability of pest detection (conditional on its presence), d, is a saturating function of search eﬀort (v):

eqn 6

where we have a 50% chance of ﬁnding an infestation if we search with eﬀort v = k. Search eﬀort could reﬂect, for example, time spent at a site, number of ﬁeldworkers and diﬀerent search methods used. It is convenient therefore to measure eﬀort in terms of its total cost, which gives the eﬀort per site searched as v ¼ B=sA;

eqn 7

where the budget for the management period is $B. Combining eqns 6 and 7 gives the detection probability in an occupied site as a function of budget, proportion of areas searched, and the area of the region: d ¼ B=ðksA þ BÞ:

eqn 8

IMPROVING THE SPECIES DISTRIBUTION MODEL

Another possible strategy is to defer searching to concentrate resources on improving the predictive accuracy of the map (increasing a in eqn 1). These improvements could come from updating habitatpredictive algorithms, or acquiring more or better environmental data relating to pest habitat, including development of novel techniques to do so.

DECISION TRADE-OFFS

In this example, therefore, we choose from three actions (i) at each time step: i = 1: search proportion s1 = S95 of the region, visiting 95% of all occupied sites, 0Æ95/A. In each we have a probability B ⁄ (ks1A + B) of detecting the pest, but we also search A(1 – /)0Æ95a sites in vain, and leave (1 – S95)A sites unsearched; i = 2: search proportion s2 = 0Æ5S95 of the region, which gives us /y2A occupied sites in which we have a probability B ⁄ (ks2A + B) of detecting the pest. We will also search ð1 /Þya2 A empty sites and leave (1 – 0Æ5S95)A sites unsearched; or i = 3: postpone searching (s3 = 0) and develop a better map (increase a).

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 86–95

90 P. W. J. Baxter & H. P. Possingham For the ﬁrst two options, the expected proportion of colonies detected and destroyed is Di ¼

/yi AB ksi A þ B

i ¼ 1; 2

eqn 9 PERFORMANCE EVALUATION

(Fig. 1 c,d). For the third option (i = 3) we assume that all /A colonies remain undetected (D3 = 0).

PARAMETERIZATION AND OPTIMIZATION

Cost parameters For purposes of illustration, we assume that the map covers A = 1000 sites, or grid-cells, each of which may or may not contain the pest species; that the budget per time step is B = $100 000; and that the cost of the eﬀort required to have 50% chance of detecting a pest present on an infested site is k = $500. These values give a detection probability per occupied site of d = 0Æ17, if all sites are searched, increasing to d = 0Æ67 if only 10% of the region is searched.

The spread of the infestation will depend on both the organism’s biology and search-and-control success. In this example we use simpliﬁed ﬁre ant population dynamics to demonstrate our approach. Scanlan & Vanderwoude (2006) modelled the spread of ﬁre ants in Australia at two spatial scales, and assumed that invasion extent doubled every 2–4 years when measured at the broader scale (10 000-km2 blocks), with faster dynamics at local scales. We compromise between these two scales and assume a maximum doubling period, in the absence of control or density-dependence, of 24 months. Thus, in a 6-month management period, the maximum rate of spread is kmax = 2(6 ⁄ 24) 1Æ19. We assume that the increase in regional density follows logistic growth (Scanlan & Vanderwoude 2006; see also Shryock et al. 2008), giving a rate of increase of ku ð/Þ ¼ 1 þ ðkmax 1Þð/ 1Þ=1

To test the performance of employing the optimization (SDP) results vs. three simpler management regimes, we simulated 20-year management of a pest invasion with dynamics as above (Table 1), beginning at 1% regional infestation; this level would in practice be dependent on both timeliness of detection and the spatial resolution of the map. We assumed an initial knowledge level of a = 2, reﬂecting a reasonable lower-end AUC value (Latimer et al. 2009; Va´clavı´ k & Meentemeyer 2009). The alternative management regimes were based on those used for the SDP formulation and comprise: always search S95 sites; always search S95 ⁄ 2 sites (doubling eﬀort-per-site); or rotate between search and learning modes. The ‘rotating strategy’ iteratively followed the sequence: widespread control (S95) - upgrade map (increase a) - focussed control (S95 ⁄ 2).

Results

Invasion parameters

eqn 10

if colonies are uncontrolled (the denominator of 1 indicates the ‘carrying capacity’ of / = 1 in a fully colonized area). Even if a colony is detected and destroyed, it may already have reproduced. Assuming that colonies are discovered on average halfway through their reproductive cycle, the rate of increase of detected-and-removed colonies is kd ð/Þ ¼ ku ð/Þ1=2 :

(Fraser et al. 2006), avoided costs of ongoing management, or even societal values).

eqn 11

For ku = 1Æ19 at its maximum value, this implies that at most 9% of detected ant colonies will have spread prior to their destruction (1Æ19½ 1Æ09). For the optimization (Appendix S1, Supporting Information), we describe the system state by the combination of map quality (a in eqn 1) and regional pest density /. The system undergoes transitions between states with probabilities governed by the outcome of each management option. Expected future pest density depends on the rates of spread from controlled and uncontrolled sites, and we assume that learning improves the map quality a by one (with probability 0Æ2) or two (probability 0Æ8) units, giving diminishing increases in AUC. We accord a ‘utility’ value to each state depending on the pest density only (map quality has no utility apart from improving future searches): utility increases linearly as pest density decreases, with a 100-fold bonus if the pest is eradicated (this high bonus could reﬂect renewed access to export markets in the case of agricultural pests

STATE-DEPENDENT OPTIMIZATION

The optimal strategies for learning about and controlling an invasion (i.e. the SDP solution; Fig. 2) depend on the system state (pest infestation density / and map quality a), and on the management time horizon T. We ﬁrst consider long-term management recommendations (e.g. more than a decade; Fig. 2a–c). For the longest management horizon considered, 20 years, improving the predictive quality of the map initially takes precedence over either search method for most of the system states. Exceptions are when our map is already excellent (a > 15, or AUC > 0Æ9375; not unrealistically high values; Zurell et al. 2009) or when the infestation is at moderate densities. If the infestation is at very low densities we can aﬀord to delay searching until we have a better map to improve targeting of future searches. On the other hand, if the infestation is widespread then searching with a restricted budget will have little eﬀect on pest density and so we again delay searching until we have a better map. If we strive for shorter-term success (Fig. 2d–f) the value of improving the map diminishes and the optimal strategy is usually immediate search-and-control. Generally, if the pest is already widespread, then cursory widespread searching is optimal (i = 1 above), as even undirected searches will have high success. If the pest is at lower densities then we should use targeted site-intensive searching (i = 2). This strategy depends on having a reasonable-quality map to exploit: if the map is very poor, we should still use cursory searches (close in eﬀect to random searches: a 1). Nonetheless, at all but the shortest management timeframes, it is always recommended to improve a very poor map if the infestation is still at low levels (e.g. / = 0Æ01; bottom-left of Fig. 2a–e).

PERFORMANCE EVALUATION

As expected, the state-dependent optimization performed better in our simulations than the alternative strategies (Fig. 3a,

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 86–95

Optimizing search strategies for invasives 91 (c)

15-year horizon

25

25

20

20

20

15

15

15

10

10

10

5

5

5

0

0·5

1

(d)

5-year horizon

0

0·5

(e)

2-year horizon

1

0·5

(f)

6-month horizon

25

25

20

20

20

15

15

15

10

10

10

5

5

5

0

0·5

1

0

0·5

Pest density (φ)

10-year horizon

0

25

(a)

1

0

Pest density (φ)

0·5

1

1

Pest density (φ)

(b)

0·035

Cursory Focussed Optimal Rotating

0·030 0·025 0·020 0·015 0·010 0·005 0·000

(b)

20-year horizon

0

5

10

15

20

Time (years)

Mean model quality (a)

Mean infestation density (φ)

Fig. 2. Optimal strategies depending on regional infestation density (/) and current quality of species distribution model (a), over a selection of management time horizons. The optimal strategies for each (/, a) state are indicated as: white = cursory widespread searching (s1 = S95); grey = fewer, more focussed and intensive searches (s2 = S95 ⁄ 2; v2 = 2v1); black = re-direct funding towards improving the species distribution model.

Map quality (a)

Map quality (a)

(a) 25

30

Optimal Rotating

25 20 15 10 5 0

0

5

10

15

20

Time (years)

Fig. 3. Simulated performance of pest management over 20 years under default model assumptions. (a) Comparison of four management strategies: cursory widespread searches; intensive focussed searches; optimal state-dependent strategy recommended by stochastic dynamic programming; and continual rotating between cursory search, model-improvement and focussed search. (b) Acquisition of knowledge when the optimal and rotational strategies are implemented, showing mean (±SD) values of the Receiver Operating Characteristic curve exponent a. The two non-learning strategies (cursory and focussed searching) remain at the initial level of a = 2 (dotted line).

mean trajectories shown), usually achieving eradication by year 20. This eventual success is dependent on tolerating an initial increase in pest density as funds are initially devoted to improving the map (Fig. 3b). Other strategies avoid the initial ‘spike’ in pest density but fail to achieve eradication over the long-term. The cursory-widespread search strategy performs worst, with the site occupancy steadily increasing. The ‘focussed’ strategy results in a steady, but slow, decrease in density. The ‘rotating’ strategy allows some map improvement as well as search-and-control eﬀorts, and thus performs comparatively well. Nonetheless, in terms of achieving eradication the optimal strategy performed considerably better, eradicating the pest in 97% of simulations (compared to 59% for the rotating strategy, and never for the two non-learning strategies). We also investigated departures from our default assumptions and parameter values (Fig. 4). Increasing the budget by 50% (Fig. 4a) improved the performance of all strategies (unsurprisingly), with most strategies achieving high levels of

suppression. Starting with a better map (a = 5; Fig. 4b) also led to improvements in all strategies, particularly the ‘focussed’ strategy, which particularly depends on reliable site selection for its success. We also examined our assumption that research increases the value of a by 2 units with probability pa = 0Æ8, however, this appeared to have little overall eﬀect on results (Figs 4c,d). The spike in pest density when the optimization solution is applied decreases with pa, because managers will be less willing to allow temporary population explosions if the scope for improving the map is reduced, and so switch to search ⁄ control operations sooner. Similar eﬀects resulted from eliminating the eradication bonus (Fig. 4e) or incorporating a 3% discount rate (Fig. 4f): these scenarios reduce the emphasis on eradication vs. containment (Odom et al. 2003; Fraser et al. 2006) either in the longer-term (discounting) or permanently (no bonus), leading to decreased incentive to improve knowledge. All strategies had more success controlling slower invasions (Fig. 4g), while rapidly-spreading invasions (Fig. 4h) were only able to be suppressed by the optimal and rotating

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 86–95

Larger management budget (a)

0·035

Cursory Focussed Optimal Rotating

0·030 0·025 0·020 0·015 0·010 0·005 0·000 0

5

10

15

20

Mean infestation density (φ)

Mean infestation density (φ)

92 P. W. J. Baxter & H. P. Possingham Better initial knowledge (b)

0·035 0·030 0·025 0·020 0·015 0·010 0·005

0·000 0

5

0·030 0·025 0.020 0·015 0·010 0·005 5

10

15

20

Mean infestation density (φ)

Mean infestation density (φ)

Moderate knowledge gain (c)

0·035

0·000 0

0·030 0·025 0·020 0·015 0·010 0·005 10

15

20

Mean infestation density (φ)

Mean infestation density (φ)

No eradication bonus (e)

5

0·025 0·020 0·015 0·010 0·005 5

10

15

20

Mean infestation density (φ)

Mean infestation density (φ)

Slower lnvasion speed (g)

0

20

15

20

0·030 0·025 0·020 0·015 0·010 0·005 0·000 0

5

10

Discounting of benefits (f)

0·035 0·030 0·025 0·020 0·015 0·010 0·005 0·000

0

5

10

Time (years)

0·030

0·000

15

Slower knowledge gain (d)

Time (years)

0·035

20

Time (years)

0·035

0

15

0·035

Time (years)

0·000

10

Time (years)

Time (years)

0·09

Faster lnvasion speed (h)

0·08 0·07 0·06 0·05 0·04 0·03 0·02 0·01 0·00 0

5

Time (years)

strategies. Overall, the ranking of performances was robust to changes in assumptions (including others not shown; e.g. higher initial pest density). The tendency of the SDP solution to allocate approximately 3–4 years’ initial funding to map improvement was similarly consistent.

Discussion We have introduced and demonstrated an approach for trading-oﬀ actions that search for and remove a pest, against an action that only gains knowledge. Our optimizations indicate

10

Time (years)

15

20

Fig. 4. Mean performances of four management strategies (as Fig. 3a) under alternative parameter values and assumptions. (a) Budget per time-step of $150 000; (b) initial good quality species distribution model (a = 5); (c) 50% and (d) 20% probability of increasing a by 2 units (otherwise a increases by 1 unit); (e) no eradication bonus given; (f) applying a 3% discount rate to the performance beneﬁts; and maximum doubling time of invasion set to (g) 30 months (kmax = 0Æ15) and (h) 18 months (kmax = 1Æ26; note diﬀerent vertical scale).

that spending time improving knowledge about the pest’s habitat preferences, before searching for it, is optimal. We were surprised to discover this, as deciding to improve knowledge while a pest incursion grows exponentially seems like ﬁddling while Rome burns. Nevertheless this highlights the value of learning even when at the expense of control operations under seemingly urgent conditions. With a long-term perspective, it is optimal to learn rather than take direct action at the start of an invasion (Fig. 2), but exactly for how long we should delay action and learn depends on many factors. These include the cost of learning relative to

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 86–95

Optimizing search strategies for invasives 93 on-ground search and control operations, the desire for eradication vs. containment, the likely improvement in our knowledge of habitat preferences and the biological characteristics of the pest species (Fig. 4). Once the initial learning phase is over, searching should be focussed and intensive, rather than widespread and cursory, when we have less widespread infestation or intermediate predictive capabilities (Fig. 2d–f). Simulation of alternative plausible management strategies conﬁrmed the expected superiority of the optimization approach (Fig. 4). The overall similarity of trajectories under diﬀerent assumptions also highlighted that at the initial stages of pest incursion (when / is still small), allowing some spread of the infestation may be acceptable so long as we can improve our map to better predict and control the pest in future. This is partly because our management objective gives more weight to eventual eradication than to merely suppressing the population (cf. Fig. 4e), and increased knowledge is ultimately required to improve searching for and removing the invader. This result raises the interesting question of whether the best emergency response to a new incursion is actually to ‘do nothing’ – take no immediate direct action but concentrate funds into developing high-quality predictive maps to maximize eﬃcacy of future management. By taking a long-term focus (and deferring control), it could be argued that managers are being more pragmatic, trading oﬀ the apparent urgency of a new incursion against the strategic allocation of resources to knowledge acquisition and better chance of success further in the future. This however must also be evaluated in the light of the structure and assumptions presented here. For example, the framework can be expanded to include many diﬀerent strategies (e.g. more search areas si and their corresponding search intensities). More reﬁned budgetary options in the optimization could allow both searching and habitat modelling simultaneously by selecting the proportion of funds allocated to learning vs. control, instead of the all-or-nothing choice presented here. Alternatively some funding could be allocated to improving detection probability at occupied sites by developing enhanced on-site search techniques. Pursuing this research will provide further insights into knowledge ⁄ action trade-oﬀs. The optimization (Fig. 2) indicates cases where we should search and control straight away rather than improving a map. The most obvious case is when we already have a reasonable quality map. The simulation results indicate that the switch from knowledge acquisition to active control takes place after about 3–4 years’ research (average map quality of a = 16Æ4, AUC = 0Æ94), but of course this depends on other factors such as the severity of the incursion and the management time horizon. Nonetheless, the shorter the management timeframe, the less the relative merits of acquiring additional information and the more likely control action will be taken. This diminishing return on investment in data acquisition has recently been shown for conservation of South African fynbos ﬂora (Grantham et al. 2009), in which case optimal decisions on choosing patches to reserve could be made after a relatively small initial data-gathering eﬀort. Other cases demanding quicker action are when the pest population spreads quickly, or the management timeframe imposes too tight a deadline to be able to act

on the research results. These two aspects reﬂect increased urgency in countering the invasion, either in terms of spatial spread or the wish to produce positive outcomes quickly. Another factor to favour immediate action is having a large budget, which allows greater search eﬀort and better success even with poor predictive ability. The size of the budget may itself reﬂect an urgent desire to control an invasion. This was the case in the Queensland ﬁre-ant incursion, for which a large budget was available from the outset (AU$123 million; George 2007). Such well-funded programmes allow us the luxury of immediate action as well as simultaneous production of a predictive map; however many conservation eﬀorts operate on much tighter budgets, making trade-oﬀs between knowledge gain and control eﬀorts unavoidable. Even in the ﬁre-ant case success has proven elusive despite more than AU$200 million invested (Williams 2010). While we deliberately used simple models to demonstrate the learn-or-act trade-oﬀ, the assumptions made in developing our framework should be noted, particularly if being applied to a real-world situation. For example, we have assumed that diversion of funds into research will deﬁnitely have a (measureable) positive outcome, that a will increase precisely by either 1 or 2 (with probability 0Æ2 and 0Æ8 respectively); while altering the relative values of these probabilities had little eﬀect on simulated performance (Fig. 4c,d), the possibilities of no map improvement, or even perverse disimprovement, were not considered. We have assumed that this improvement costs the same per time-step as searching for the pest; the actual costs and beneﬁts of map improvement would need to be estimated based on acquiring suitable personnel and infrastructure (software, data layers etc.), and anticipating the projected map improvements (reduction in level of false positives with an improved map) – these estimates, while uncertain, may provide suﬃcient insight into whether greater weight should be given to searching or knowledge improvement. The current value of a could be estimated from the map’s performance in predicting the species’ native range (with caution), or from previous search results if available. We have disregarded the spatio-temporal dynamics of the infestation. In reality dispersing ants may not ﬁnd all suitable available habitat in which to found a new colony, so our assumptions also (conservatively) overestimate doubling speed. Just as a searched site may be ideal habitat, but not yet colonized by a dispersing pest, a site deemed to be of marginal suitability may become colonized if the surrounding area is already saturated with colonies. We have also assumed that the map extends to all possible areas of spread – judgement may be required to trade its spatial extent oﬀ against suﬃcient grain to provide meaningful information on ﬁner scales. We could use more sophisticated optimization methods to address some of our assumptions. D’Evelyn et al. (2008) demonstrate the value of incorporating search results directly into estimates of pest population density, in order to choose the optimal eﬀort of control in later years, emphasizing, as here, the value of early learning (in their case via control). Our optimization is dependent on both infestation density and map accuracy, both of which may be imprecisely known (Va´clavı´ k

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 86–95

94 P. W. J. Baxter & H. P. Possingham & Meentemeyer 2009). We could therefore re-express the problem in terms of our belief of what the current values of / and a are, in a partially-observable Markov decision process (POMDP), to optimize our future ‘belief state’ rather than the actual, but unknowable, state of the system (for ecological examples see Lane 1989; White 2005; Chade`s et al. 2008). In terms of economic simpliﬁcations, we have ignored the costs of travelling between sites; so that the selection of many sites to search may incur extra costs if extensive travel is involved. McDonald-Madden, Baxter & Possingham (2008) demonstrate a succinct approach to this problem when the locations of the populations are known. We have also assumed that the cost of pest removal, once discovered in a site, is negligible, or integrated into the search costs. Similarly we ignore the possibility of multiple pest occurrences on one site which would inﬂuence both the search success and removal costs; or other aspects of spatial contagion that also aﬀect costs and strategy performance. We have made these assumptions and simpliﬁcations so that we could get to the heart of the act-or-learn problem in invasion management; some of these assumptions will be relaxed in future work. Nonetheless, striking results emerge, particularly the consistent recommendation that learning ﬁrst, and looking (more successfully) later on, is the long-term optimal approach to new pest incursions.

Acknowledgements We thank Iadine Chade`s, Hedley Grantham, Cindy Hauser, Dane Panetta, the editor and two anonymous reviewers for helpful discussions and comments. Financial support was provided by the Australian Centre of Excellence for Risk Analysis (ACERA) and the Applied Environmental Decision Analysis (AEDA) research hub, a Commonwealth Environmental Research Facility.

References Baret, S., Rouget, M., Richardson, D.M., Lavergne, C., Egoh, B., Dupont, J. & Strasberg, D. (2006) Current distribution and potential extent of the most invasive alien plant species on La Re´union (Indian Ocean, Mascarene islands). Austral Ecology, 31, 747–758. Baxter, P.W.J., Wilcox, C., McCarthy, M.A. & Possingham, H.P. (2007) Optimal management of an annual weed: a stochastic dynamic programming approach. MODSIM 2007 International Congress on Modelling and Simulation (eds L. Oxley & D. Kulasiri), pp. 2223–2229. Modelling and Simulation Society of Australia and New Zealand, http://www.mssanz.org.au/MOD SIM07/papers/41_s34/OptimalManagement_s34_Baxter_.pdf. Bellman, R.E. (1957) Dynamic Programming. Princeton University Press, Princeton, NJ. Bogich, T. & Shea, K. (2008) A state-dependent model for the optimal management of an invasive metapopulation. Ecological Applications, 18, 748–761. Cacho, O.J., Hester, S. & Spring, D. (2007) Applying search theory to determine the feasibility of eradicating an invasive population in natural environments. Australian Journal of Agricultural and Resource Economics, 51, 425–443. Callcott, A.M.A. & Collins, H.L. (1996) Invasion and range expansion of imported ﬁre ants (Hymenoptera: Formicidae) in North America from 19181995. Florida Entomologist, 79, 240–251. Chade`s, I., McDonald-Madden, E., McCarthy, M.A., Wintle, B., Linkie, M. & Possingham, H.P. (2008) When to stop managing or surveying cryptic threatened species. Proceedings of the National Academy of Sciences, 105, 13936– 13940. D’Evelyn, S.T., Tarui, N., Burnett, K. & Roumasset, J.A. (2008) Learningby-catching: Uncertain invasive-species populations and the value of information. Journal of Environmental Management, 89, 284–292.

Eliassen, S., Jørgensen, C., Mangel, M. & Giske, J. (2009) Quantifying the adaptive value of learning in foraging behavior. The American Naturalist, 174, 478–489. Fortin, D. (2002) Optimal searching behaviour: the value of sampling information. Ecological Modelling, 153, 279–290. Fraser, R.W., Cook, D.C., Mumford, J.D., Wilby, A. & Waage, J.K. (2006) Managing outbreaks of invasive species: eradication versus suppression. International Journal of Pest Management, 52, 261–268. George, C.R.R. (2007) Establishment risk modelling of invasive organisms: Solenopsis invicta as a case study. PhD Thesis, University of Queensland, Brisbane. Grantham, H.S., Wilson, K.A., Moilanen, A., Rebelo, T. & Possingham, H.P. (2009) Delaying conservation actions for improved knowledge: how long should we wait? Ecology Letters, 12, 293–301. Hauser, C.E. & McCarthy, M.A. (2009) Streamlining ‘search and destroy’: cost-eﬀective surveillance for invasive species management. Ecology Letters, 12, 683–692. Jennings, C. & McCubbin, K. (2004) The National Red Imported Fire Ant Eradication Program Overview – October 2004. Fire Ant Control Centre, Queensland Department of Primary Industries, Brisbane, Australia. King, J.R. & Tschinkel, W.R. (2008) Experimental evidence that human impacts drive ﬁre ant invasions and ecological change. Proceedings of the National Academy of Sciences, 105, 20339–20343. Lane, D.E. (1989) A partially observable model of decision-making by ﬁshermen. Operations Research, 37, 240–254. Latimer, A.M., Banerjee, S., Sang Jr, H., Mosher, E.S. & Silander Jr, J.A. (2009) Hierarchical models facilitate spatial analysis of large data sets: a case study on invasive plant species in the northeastern United States. Ecology Letters, 12, 144–154. Lobo, J.M., Jime´nez-Valverde, A. & Real, R. (2008) AUC: a misleading measure of the performance of predictive distribution models. Global Ecology and Biogeography, 17, 145–151. Mangel, M. & Clark, C.W. (1988) Dynamic Modeling in Behavioral Ecology. Princeton University Press, Princeton, NJ. McCarthy, M.A., Possingham, H.P. & Gill, A.M. (2001) Using stochastic dynamic programming to determine optimal ﬁre management for Banksia ornata. Journal of Applied Ecology, 38, 585–592. McDonald-Madden, E., Baxter, P.W.J. & Possingham, H.P. (2008) Subpopulation triage: how to allocate conservation eﬀort among populations. Conservation Biology, 22, 656–665. McNamara, J.M. & Houston, A.I. (1996) State-dependent life histories. Nature, 380, 215–221. Metz, C.E. (1986) ROC methodology in radiologic imaging. Investigative Radiology, 21, 720–733. Moloney, S. & Vanderwoude, C. (2002) Red imported ﬁre ants: a threat to eastern Australia’s threatened wildlife? Ecological Management and Restoration, 3, 167–175. Moody, M.E. & Mack, R.N. (1988) Controlling the spread of plant invasions the importance of nascent foci. Journal of Applied Ecology, 25, 1009–1021. Murray, J.V., Goldizen, A.W., O’Leary, R.A., McAlpine, C.A., Possingham, H.P. & Choy, S.L. (2009) How useful is expert opinion for predicting the distribution of a species within and beyond the region of expertise? A case study using brush-tailed rock-wallabies Petrogale penicillata. Journal of Applied Ecology, 46, 842–851. Odom, D.I.S., Cacho, O.J., Sinden, J.A. & Griﬃth, G.R. (2003) Policies for the management of weeds in natural ecosystems: the case of scotch broom (Cytisus scoparius, L.) in an Australian national park. Ecological Economics, 44, 119–135. Pearce, J. & Ferrier, S. (2000) Evaluating the predictive performance of habitat models developed using logistic regression. Ecological Modelling, 133, 225– 245. Pimentel, D., McNair, S., Janecka, J., Wightman, J., Simmonds, C., O’Connell, C., Wong, E., Russel, L., Zern, J., Aquino, T. & Tsomondo, T. (2001) Economic and environmental threats of alien plant, animal, and microbe invasions. Agriculture Ecosystems & Environment, 84, 1–20. Regan, T.J., McCarthy, M.A., Baxter, P.W.J., Panetta, D.F. & Possingham, H.P. (2006) Optimal eradication: when to stop looking for an invasive plant. Ecology Letters, 9, 759–766. Richards, S.A., Possingham, H.P. & Tizard, J. (1999) Optimal ﬁre management for maintaining community diversity. Ecological Applications, 9, 880–892. Sala, O.E., Chapin, F.S., Armesto, J.J. et al. (2000) Global biodiversity scenarios for the year 2100. Science, 287, 1770–1774. Scanlan, J.C. & Vanderwoude, C. (2006) Modelling the potential spread of Solenopsis invicta Buren (Hymenoptera: Formicidae) (red imported ﬁre ant) in Australia. Australian Journal of Entomology, 45, 1–9.

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 86–95

Optimizing search strategies for invasives 95 Shryock, K.A., Brown, S.L., Sanders, N.J. & Burroughs, E. (2008) A reactiondiﬀusion equation modeling the invasion of the argentine ant population, Linepithema humile, at Jasper Ridge Biological Preserve. Natural Resource Modeling, 21, 330–342. Stephens, D.W. & Krebs, J.R. (1986) Foraging Theory. Princeton University Press, Princeton, N.J. Sutherst, R.W. & Maywald, G. (2005) A climate model of the red imported ﬁre ant, Solenopsis invicta Buren (Hymenoptera: Formicidae): Implications for invasion of new regions, particularly Oceania. Environmental Entomology, 34, 317–335. Swets, J.A. (1986) Indexes of discrimination or diagnostic-accuracy – their ROCs and implied models. Psychological Bulletin, 99, 100–117. Swets, J.A. (1988) Measuring the accuracy of diagnostic systems. Science, 240, 1285–1293. Swets, J.A. & Pickett, R.M. (1982) Evaluation of Diagnostic Systems: Methods From Signal Detection Theory. Academic Press, New York. Taylor, C.M. & Hastings, A. (2004) Finding optimal control strategies for invasive species: a density-structured model for Spartina alterniﬂora. Journal of Applied Ecology, 41, 1049–1057. Va´clavı´ k, T. & Meentemeyer, R.K. (2009) Invasive species distribution modeling (iSDM): Are absence data and dispersal constraints needed to predict actual distributions? Ecological Modelling, 220, 3248–3258. Walters, C.J. (1986) Adaptive Management of Renewable Resources. Macmillan, New York. White, B. (2005) An economic analysis of ecological monitoring. Ecological Modelling, 189, 241–250. Whittle, A.J., Lenhart, S. & Gross, L.J. (2007) Optimal control for management of an invasive plant species. Mathematical Biosciences and Engineering, 4, 101–112. Williams, D.F. (1994) Control of the introduced pest Solenopsis invicta in the United States. Exotic Ants: Biology, Impact, and Control of Intro-

duced Species (ed. D.F. Williams), pp. 282–292. Westview Press, Boulder, CO. Williams, B. (2010) Fire ants advance. Courier Mail 31 March 2010. News Limited, Brisbane, pp 1–2. Wintle, B.A., Elith, J. & Potts, J.M. (2005) Fauna habitat modelling and mapping: a review and case study in the Lower Hunter Central Coast region of NSW. Austral Ecology, 30, 719–738. Zurell, D., Jeltsch, F., Dormann, C.F. & Schro¨der, B. (2009) Static species distribution models in dynamically changing systems: how good can predictions really be? Ecography, 32, 733–744. Received 27 January 2010; accepted 30 September 2010 Handling Editor: Andy Sheppard

Supporting Information Additional Supporting Information may be found in the online version of this article. Appendix S1. Details of the optimization. As a service to our authors and readers, this journal provides supporting information supplied by the authors. Such materials may be re-organized for online delivery, but are not copy-edited or typeset. Technical support issues arising from supporting information (other than missing ﬁles) should be addressed to the authors.

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 86–95

Journal of Applied Ecology 2011, 48, 96–101

doi: 10.1111/j.1365-2664.2010.01894.x

REVIEW

The frequency and magnitude of non-additive responses to multiple nutrient enrichment Jacob E. Allgeier1*, Amy D. Rosemond1 and Craig A. Layman2 1

Odum School of Ecology, University of Georgia, Athens, GA 30602, USA; and 2Marine Sciences Program, Department of Biological Sciences, Florida International University, 3000 NE 151st Street, North Miami, FL 33181, USA

Summary 1. Anthropogenic eutrophication is among the greatest threats to ecosystem functioning globally, often occurring via enrichment of both nitrogen (N) and phosphorus (P). As such, recent attention has focused on the implications of non-additive responses to dual nutrient enrichment and the inherent diﬃculty associated with predicting their combined eﬀects. 2. We used a simple metric to quantify the frequency and magnitude of non-additive responses to enrichment by N, P and N + P in 653 experiments conducted across multiple ecosystem types and locations. 3. Non-additive responses were found to be common in all systems. Freshwater ecosystems and temperate latitudes tended to have frequent synergistic responses to dual nutrient enrichment, i.e. the response was greater than predicted by an additive model. Terrestrial and arctic systems were dominated by antagonistic responses (responses to N + P that were less than additive). 4. The mean of all experiments was synergistic because despite being less common, synergistic responses were generally of greater magnitude than antagonistic ones. 5. Synthesis and applications. Our study highlights the ubiquity of non-additive eﬀects in response to dual nutrient enrichment and further elucidates the complex ways in which ecosystems respond to human impacts. Our results suggest how alternative nutrient limitation scenarios can be used to guide approaches to conservation and management of nutrient loading to ecosystems. This review provides the ﬁrst published summary of non-additive responses by primary producers. Key-words: Antagonism, co-limitation, eutrophication, interaction, nitrogen, nutrient loading, phosphorus, primary production, synergy

Introduction The ecological impacts of excessive nutrient loading are substantial, driving losses of ecosystem services world-wide (Vitousek et al. 1997; Smith & Schindler 2009) and stimulating debate over how to most eﬀectively regulate anthropogenic nutrient inputs (Conley et al. 2009). At the crux of the debate is whether controlling nitrogen (N), phosphorus (P) or both, should frame conservation initiatives (Carpenter 2008; Conley et al. 2009). The underpinning research that has informed this debate is generally based on quantifying the primary producer response to enrichment by these key nutrients. Most notably, measuring the production response to multiple nutrients, i.e. both N and P, has received much attention because many anthropogenic stressors tend to alter concentrations of both *Correspondence author. E-mail: [email protected]

nutrients simultaneously (Sala & Knowlton 2006; Halpern et al. 2008). A recent study by Elser et al. (2007) demonstrated the prevalence of nutrient co-limitation across ecosystems. Here we deﬁne nutrient co-limitation as a greater response to simultaneous enrichment by both nutrients than enrichment by either nutrient individually. Some interpretations of these ﬁndings have suggested that they likewise imply a dominance of synergy in ecosystems, assuming that co-limitation is necessarily synergistic (Davidson & Howarth 2007; Elser et al. 2007). However, a synergism only occurs when the response is greater than additive, whereas co-limitation can also be an equal to or less than additive response. Understanding these diﬀerent outcomes forms the basis of our ability to predict how an ecosystem will respond to nutrient enrichment and, therefore, our ability to develop eﬀective management strategies.

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society

Non-additive responses to nutrient enrichment 97 We developed a simple metric to quantify the relative response to additions of both N and P compared to predicted additive eﬀects in plant production to: (i) quantitatively assess the generalities of non-additive responses to nutrient enrichment and (ii) distinguish diﬀerent types of co-limitation across ecosystem types and latitudinal gradients. We also test the hypothesis that the distribution of these data is consistent with null distributions based on random values. Our results suggest how alternative nutrient limitation scenarios can be used to guide approaches to conservation and management of nutrient loading to ecosystems.

Materials and methods We developed the interaction eﬀect index (IEI) to quantify the response of primary producers to N and P additions: IEI ¼ ln½response NP=ðresponse N + response PÞ:

eqn 1

Where response NP is the primary producer biomass (and in some cases the change in mass) reported for N + P treatments (hereafter NP) and response N and response P are primary producer biomass responses in those treatments. Taking the natural log of the quotient proportionally centers the IEI values around zero. For example, an IEI value generated from an experiment, where response NP is two times greater than response N + response P (i.e., ln(2)) is equal to the absolute value of an experiment, where response N + response P is two times greater than response NP (i.e., ln(0Æ5)). We applied the IEI to 653 experiments from marine, freshwater and terrestrial ecosystems that tested for primary producer responses to enrichment in all three treatments: N, P and NP (compiled in Elser et al. 2007; obtained via the National Center for Ecological Analysis and Synthesis). Experiments that used the metric of biomass per unit area or volume were included, but proxy variables for biomass were also allowed (e.g. chlorophyll a concentration, ash-free dry mass, carbon mass, biovolume, per cent cover; Elser et al. 2007). We included only studies that reported mean community-level biomass responses to nutrient enrichment. Thus, the only single species responses that

(a)

(b)

were included were drawn from communities dominated by single species. One hundred and twenty-nine studies were conducted in laboratory settings; the rest of the experiments were conducted in situ. A total of 39 of the 653 experiments included additional manipulations (e.g. grazer exclusion), but only data from unmanipulated controls (e.g. grazers at natural densities) were included. Because of the nature of our categories, all experiments were classiﬁed simultaneously in two categories (based on ecosystem type and latitudinal zone) A simple prediction regarding dual nutrient enrichment is that NP response would be equal to the sum of individual N and P responses (i.e. an additive response; Fig. 1b). Our metric provides a continuous measure to assess the relative departure from additivity. IEI values close to zero, either positive or negative, can be characterized by additive co-limitation (AD; Fig. 1b). As IEI increases or decreases, the non-additive eﬀect becomes more pronounced and can be classiﬁed into one of three response categories: synergistic co-limitation (SC), antagonistic co-limitation (AC) and absolute antagonism (AA; Fig. 1a,c,d). Co-limitation implies that the producer is limited by both nutrients (Arrigo 2005; Davidson & Howarth 2007), and is demonstrated when the response to both nutrients is greater than either nutrient individually. Synergistic co-limitation results when there is a positive nonadditive response, whereby NP response is greater than the sum of N and P responses (Fig. 1a). Antagonistic co-limitation is a less than additive response that occurs when NP response is less than the sum of N and P responses, but is still greater than response to either single nutrient. Absolute antagonisms occur where NP response is less than at least one of the single nutrient enrichments. The relative strength of the non-additive eﬀect (i.e. SC, AC, AA) increases as the IEI value deviates from zero, either positively (SC) or negatively (AC, AA). The term nutrient co-limitation has been subject to various interpretations and requires speciﬁc clariﬁcation (Arrigo 2005; Lewis & Wurtsbaugh 2008). According to Liebig’s law of the minimum, only one nutrient can functionally limit primary production at a given point in time. However, with dual nutrient enrichment, an individual (or producer assemblage with similar physiological nutrient demands) may oscillate between single nutrient limitation of two nutrients (here N and P). In this case, the supply of one nutrient is

(c)

(d)

Fig. 1. A conceptual diagram of possible responses from enrichment by N, P and NP. An additive response is indicated in each panel by summing the individual N (yellow) and P (blue) responses. (a) Synergistic co-limitation (SC) such that the biomass or production response to dual enrichment (NP) is greater than the additive response of both single nutrient treatments (N and P alone). (b) Additive co-limitation (AD), whereby the response to NP is equal to that of the sum of N alone and P alone. (c) Antagonistic co-limitation (AC), whereby the response to NP is greater than that of either N or P alone, but not their sum. (d) Absolute antagonism (AA), whereby NP results in less biomass or production than either N or P alone. 2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 96–101

98 J. E. Allgeier, A. D. Rosemond & C. A. Layman suﬃcient to shift demand towards that of the other, next most limiting nutrient. This interplay continues until either another factor becomes limiting or a saturation state is reached (Davidson and Howarth 2007). As such, over the course of time, e.g. an experimental time period, an individual producer (or producer assemblage) may be considered functionally co-limited, even though a single factor may always be limiting at any instantaneous point. We test the hypotheses that the distribution of the data from each category (e.g. freshwater) was consistent with null distributions based on random numbers. To do this we compared the distribution of a given category (e.g., marine; n = 105) with the distribution of a randomly sampled data set of the same size, using Kolmogorov–Smirnov tests for 1000 permutations. Then we pooled the P-values from these permutations to determine the proportion of the model runs that showed statistical diﬀerence between the randomly generated and the observed distributions (a = 0Æ05). The data set of artiﬁcial IEI values from which the null distributions for each category was sampled, was generated by randomizing each response variables (N, P and NP) from the original data set and recalculating IEI values based on these numbers. The null distribution for each permutation was then sampled from this data set.

Results Synergistic co-limitation, AC and AA occurred in all ecosystem types and latitudinal zones (Fig. 2). When comparing the frequency of each response for all experiments combined, 37% were SC, 40% were AC and 23% were AA (Fig. 2). Across all six subcategories (marine, freshwater, terrestrial, arctic, temperate and tropical), SC occurred more frequently in all but terrestrial and arctic ecosystems, in which AC occurred 64% and 71% of the time, respectively (Fig. 2). AA occurred more frequently than SC in arctic (8% SC, 21% AA) and terrestrial

systems (18% SC and 18% AA), but never occurred more frequently than AC (Fig. 2). Across all categories, SC occurred substantially less frequently than antagonistic responses (i.e. AC and AA combined). A study that incorporates multiple experimental units can be considered additive if the mean of all experiments does not signiﬁcantly diﬀer from zero (i.e. the 95% conﬁdence intervals overlap zero). Because of the complex nature of our data set, applying such conﬁdence intervals to individual studies was inappropriate. Thus to provide perspective as to the number of studies that were characterized by values close to additive (i.e. zero), we chose an arbitrary positive and negative interval of 10% from perfect additivity (0Æ095 > IEI > )0Æ095). Under these conditions, we found only 5% of experiments yielded additive responses (AD). Extending the interval to 15% (0Æ139 > IEI > )0Æ139), the frequency of such responses increased to only 11%. All experiments combined reﬂect a mean SC response (IEI = 0Æ12, P < 0Æ001 for t-test of IEI = 0). Freshwater, temperate and tropical subcategories had mean net SC IEI values [P < 0Æ005 for t-test of IEI = 0 for freshwater and temperate; tropical did not diﬀer from zero, P = 0Æ43 (see Appendix S1 in Supporting Information)]. Marine, terrestrial and arctic subcategories had mean AC IEI values [P < 0Æ001 for t-test of IEI = 0 for terrestrial and arctic, marine did not diﬀer from zero P = 0Æ83 (Appendix S1)] (red lines; Fig. 3). SC values were on average of greater magnitude than AC or AA values in most subcategories (coloured bars; Fig. 3). Freshwater ecosystems had the greatest mean SC value (IEI = 1Æ23 ± 0Æ07, NP responses 3Æ4· greater than additivity). Tropical and marine systems demonstrated the lowest IEI

Fig. 2. Frequency of IEI values within each subcategory. In each plot, the white background bars indicate the frequency of IEI values for all experiments combined. A positive value represents synergistic co-limitation, a negative value indicates either antagonistic co-limitation or absolute antagonism and zero represents additive co-limitation. Categories are not orthogonal, thus experiments can be within multiple categories (i.e., temperate and marine). 2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 96–101

Non-additive responses to nutrient enrichment 99 of random organization of data, or some underlying pattern driving these trends. Over 99% of random permutations of the data set diﬀered from the observed distribution of values from all the experiments combined. With the exception of marine and tropical categories, >95% of the random permutations of the data set diﬀered from the observed distribution of values in every category. These ﬁndings provide evidence that the distribution of these data is a product of underlying patterns that emerge from each subcategory.

Discussion

Fig. 3. Full range of all values (grey bars) and mean values for each response type (as indicated by the height of coloured bars; e.g. SC) for diﬀerent ecosystem types and latitudinal zones. Positive and negative values as in Fig. 2. The red line indicates the net mean IEI value for the respective category. For context, an absolute IEI value of 0Æ69 or 1Æ09 indicate a 100% or 200% increase or decrease from additivity, respectively. The coloured bars indicate mean values for each category: yellow bars for synergistic co-limitation (SC), green for antagonistic co-limitation (AC) and blue for absolute antagonism (AA). Categories include fundamental ecosystem types (Mar = marine, Fresh = freshwater and Terr = terrestrial) and well as categories based on latitudinal zones [Arct = arctic (latitudes >66Æ5), Temp = temperate (latitudes 23Æ5–66Æ5) and Trop (latitudes 23Æ5N to 23Æ5S)].

values (IEI = )0Æ88 ± 0Æ5, )0Æ92 ± 0Æ12; NP responses 2Æ4· and 2Æ5· less than additivity, respectively). Terrestrial ecosystems and arctic latitudes were the categories that had greater absolute mean AA than SC values. The highest IEI value (IEI = 5Æ01; NP response 150· greater than additivity) was from a benthic freshwater stream (Chessman, Hutton & Burch 1992). However, of the top 50 highest IEI values, all but two (both in benthic freshwater environments) experiments were conducted in pelagic freshwater and marine environments. The lowest IEI value (IEI = )2Æ81; NP response 16· less than additivity) was conducted on the benthos of a temperate marine estuary (Taylor et al. 1995). Unlike the positive IEI values, the lower IEI values were not dominated by experiments from any category. A bimodal trend is apparent in freshwater, marine, temperate and tropical categories, whereby there is a secondary mode centred around IEI 2 (Fig. 2). Examination of the data showed that this trend was strongly driven by a single set of experiments in temperate lakes (62 of the 82 studies) (Maberly et al. 2002). Of the 82 experiments that fall within the range of 1Æ5 < IEI < 2Æ5, we found that all but two were conducted in the pelagic zone of freshwater or marine environments, emphasizing that pelagic environments may tend towards relatively strong synergistic response to dual nutrient enrichment. Comparing the distribution of the data within each category with that of a randomly generated null distribution allows inference as to the probability that these data were the product

Synergies have garnered much attention in the ecological literature, often under the assumption that they occur frequently and with great magnitude (Myers 1995; Sala & Knowlton 2006; Halpern et al. 2008). Our ﬁndings provide more detail to this broad generalization. Though synergistic responses (SC) were often demonstrated, they occurred less frequently than antagonistic responses (the combination of AC and AA). However, where they occurred, SC tended to be of greater magnitude than antagonisms, as is supported by the bimodal distribution of the data with the second mode occurring approximately around 2. Thus, although the distribution of experiments is skewed towards negative IEI values (Fig. 2), the overall mean IEI is positive. The presumed mechanism for synergisms results from primary production that is limited by both nutrients to such a degree that little production occurs under enrichment by a single nutrient. SC is generally a result of oscillating nutrient limitation, whereby ambient availability of nutrients is minimal, and given supply of one nutrient, limitation shifts towards limitation by the other (Davidson & Howarth 2007). Thus, limitation oscillates between nutrients (if supply rate of both nutrients is constant relative to demand) until either production is maximized or another factor becomes limiting. These conditions are often prevalent in extremely nutrient poor ecosystems (Arrigo 2005). Antagonistic co-limitation, the most common response type, can be explained by a third (or additional) limiting factor. Other micronutrients (e.g. iron, magnesium, molybdate, sylica), as well as physical factors (e.g. light, water), can limit production (Howarth, Marino & Cole 1988; Arrigo 2005; Davidson & Howarth 2007). Thus, stimulating production beyond a certain level may incur limitation by a resource(s) besides N or P. Another mechanism may derive from physiological and ⁄ or environmentally related limitations (e.g. maximum physical size, disturbance or grazing), whereby the upper bound of community or individual primary production is constrained in mass or size irrespective of nutrient resources (Rosemond 1993). An additional plausible mechanism for AC may occur if increased supply of one nutrient concomitantly decreases the need for another. An example is the requirement of N for the anabolism of phosphatase enzymes which can be used to process organic P at low availability of inorganic or bioavailable P (Chrost 1991). In this case, enrichment of N can enhance net primary production (via increased production of phosphatase,

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 96–101

100 J. E. Allgeier, A. D. Rosemond & C. A. Layman and thus increased access to inorganic P). However, under conditions of enrichment with N and P, the availability of inorganic P can simultaneously inhibit production of phosphatase resulting in potentially similar or only slightly higher production than with N additions alone. The net response to NP would then be less than additive, but still greater than the response to N or P alone (Ivancic et al. 2009; Rees et al. 2009; Scott et al. 2009). Absolute antagonisms, perhaps the most counterintuitive response, were the most infrequent response category. The eﬀect of grazing could lead to AA, whereby the grazer could selectively feed on the resource with the highest production rate, or potentially with the highest nutrient content from enrichment (DeMott, Gulati & Siewertsen 1998; Heck et al. 2006). However, there are multiple examples that suggest that antagonisms could simply be experimental artefacts. For example, Taylor et al. (1995) reports a strong AA response (IEI = )2Æ81) by eelgrass to enrichment by NP. The enrichment study was conducted in mesocosms where, under enriched conditions, phytoplankton, which was growing simultaneously with eelgrass, responded synergistically to nutrient addition (Taylor et al. 1995). This experiment was characterized by a large algal bloom, causing light limitation and thus reducing seagrass biomass. These ﬁndings are consistent with the widely predicted response of seagrass to nutrient enrichment at an ecosystem scale (Deegan et al. 2002), and arose due to complex interactions involving two diﬀerent producer assemblages. The experiments compiled in this study measured the biomass response to enrichment by monocultures (e.g. a stand of a single tree species) as well as entire assemblages of diﬀerent producer species (e.g. a phytoplankton assemblage). The differences in response to nutrient enrichment between an individual species and a community of diﬀerent species can be substantial. For example, a diverse assemblage of producers probably consists of organisms with varying physiological requirements (e.g. N limited or P limited) and growth potential (e.g. greater size ⁄ growth rate). As such, under various magnitudes and time duration of nutrient enrichment, diﬀerential non-additive responses may be expected, and knowledge of the existing community is required to fully understand the mechanisms behind these responses. These diﬀerences may help explain the disparity in ﬁndings between terrestrial and aquatic (freshwater and marine together) systems, whereby aquatic systems are characterized by a greater range in IEI values with notably greater frequency and magnitude of SC. Many aquatic studies were conducted on assemblages of producers, whereas the majority of studies conducted on monocultures were from terrestrial ecosystems. These ﬁndings are consistent with the fact that pelagic environments with mixed species assemblages (e.g. phytoplankton) tend to be particularly susceptible to large production responses (e.g. algal blooms) from multiple nutrient enrichment (Conley et al. 2009). Distributional trends that emerge from these data appear to be the product of underlying ecological patterns as opposed to randomness within the data. Yet, isolating speciﬁc factors that determine the frequency of the type of non-additive eﬀects are

diﬃcult given the biological complexity (i.e. species life history, physical conditions, etc.) associated with interaction of multiple nutrients. A notable ﬁnding from our study was the dominance of antagonistic responses (AC and AA combined) in terrestrial and arctic subcategories. One explanation for terrestrial ecosystems may be that the growth rate is typically slower and generation time of terrestrial producers is typically greater than for producers in aquatic systems due to the greater requirement of structural and supporting tissue (Cebrian 1999; Chapin 2002). Thus, even given adequate experimental time frames, physiological constraints may hinder synergistic responses. Consistent with this observation, the strongest synergistic eﬀects tended to occur in aquatic ecosystems, particularly in the pelagic zone, occurring among more speciose assemblages with relatively minimal structural demands (see Appendix S2). As for arctic regions, a less than additive response to nutrient enrichment may reﬂect the fact that producer growth rates are positively correlated with temperature and thus temperature could be a physical factor limiting synergistic responses (Chapin 2002). However, despite the similarity in frequency of response types between terrestrial and arctic subcategories, arctic experiments were primarily conducted in freshwater ecosystems (S2). Our ﬁndings have important implications for management of nutrient loading to aquatic ecosystems. The prevalence of non-additive eﬀects across all systems suggests that when possible, both nutrients should be controlled in conservation and management because the ecological repercussion of simultaneous nutrient enrichment is relatively unpredictable. This is particularly relevant in ecosystems where IEI is close to zero, as they are often characterized by a relatively large response to at least one, but more often both, nutrients individually (Fig. 1b,c). As the IEI value deviates from zero, positively or negatively, it may indicate the potential for eﬀective control of nutrient loading by focusing on the single most limiting nutrient. For example, a large IEI value (i.e. a synergistic response) generally indicates that both nutrients are critical for enhancement of production, thus controlling the single most limiting nutrient (in the case of Fig. 1a; P is most important to control) may be an eﬀective way to mitigate unwanted ecosystem responses. Likewise, an extremely negative IEI value (i.e. AA) generally indicates that only one nutrient is signiﬁcantly limiting and thus suggests that controlling the loading rate of this most limiting nutrient may provide a signiﬁcant reduction in ecosystem-scale responses. In a perfect world, all stressors that negatively aﬀect ecosystems would be carefully managed. Yet, conservation eﬀorts are constrained by cost, time and societal will to manage ecosystems. Our ﬁndings show frequent and strong non-additive responses to nutrient enrichment across ecosystem types and locations. We emphasize that a single conservation model for mitigating nutrients is not appropriate and stress that future eﬀorts need to account for the complex nature of dual nutrient limitation. We further highlight the importance of incorporating all treatments (N, P and NP) into enrichment experiments in conjunction with quantitatively assessing the nature of the interaction on a system-speciﬁc basis. These data are critical

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 96–101

Non-additive responses to nutrient enrichment 101 for building eﬀective predictive models needed to inform conservation and management decision-making regarding nutrient control.

Acknowledgements We would like to thank D.S. Gruner, J.J. Elser, W.S. Harpole and their colleagues for allowing access to their data and Cynthia Tant, Ashley Helton, Andrew Mehring and William Lewis for comments that improved the manuscript. We also thank an anonymous reviewer and the editor for helpful comments. Funding was provided by a University-wide Graduate Student Fellowship, University of Georgia and National Science Foundation Grant OCE#0746164.

References Arrigo, K.R. (2005) Marine microorganisms and global nutrient cycles. Nature, 437, 349–355. Carpenter, S.R. (2008) Phosphorus control is critical to mitigating eutrophication. Proceedings of the National Academy of Sciences of the United States of America, 105, 11039–11040. Cebrian, J. (1999) Patterns in the fate of production in plant communities. American Naturalist, 154, 449–468. Chapin, F.S.I. (2002) Principles of Terrestrial Ecosystem Ecology. Springer Science. Chessman, B.C., Hutton, P.E. & Burch, J.M. (1992) Limiting nutrients for periphyton growth in sub-alpine, forest, agricultural and urban streams. Freshwater Biology, 28, 349–361. Chrost, R.J. (1991) Environmental Control of Synthesis and Activity of Aquatic Microbial Ectoenzymes. Springer-Verlag, New York. Conley, D.J., Paerl, H.W., Howarth, R.W., Boesch, D.F., Seitzinger, S.P., Havens, K.E., Lancelot, C. & Likens, G.E. (2009) Controlling eutrophication: nitrogen and phosphorus. Science, 323, 1014–1015. Davidson, E.A. & Howarth, R.W. (2007) Environmental science – nutrients in synergy. Nature, 449, 1000–1001. Deegan, L.A., Wright, A., Ayvazian, S.G., Finn, J.T., Golden, H., Merson, R.R. & Harrison, J. (2002) Nitrogen loading alters seagrass ecosystem structure and support of higher trophic levels. Aquatic Conservation-Marine and Freshwater Ecosystems, 12, 193–212. DeMott, W.R., Gulati, R.D. & Siewertsen, K. (1998) Eﬀects of phosphorusdeﬁcient diets on the carbon and phosphorus balance of Daphnia magna. Limnology and Oceanography, 43, 1147–1161. Elser, J.J., Bracken, M.E.S., Cleland, E.E., Gruner, D.S., Harpole, W.S., Hillebrand, H., Ngai, J.T., Seabloom, E.W., Shurin, J.B. & Smith, J.E. (2007) Global analysis of nitrogen and phosphorus limitation of primary producers in freshwater, marine and terrestrial ecosystems. Ecology Letters, 10, 1135–1142. Halpern, B.S., Walbridge, S., Selkoe, K.A., Kappel, C.V., Micheli, F., D’Agrosa, C., Bruno, J.F., Casey, K.S., Ebert, C., Fox, H.E., Fujita, R., Heinemann, D., Lenihan, H.S., Madin, E.M.P., Perry, M.T., Selig, E.R., Spalding, M., Steneck, R. & Watson, R. (2008) A global map of human impact on marine ecosystems. Science, 319, 948–952. Heck, K.L., Valentine, J.F., Pennock, J.R., Chaplin, G. & Spitzer, P.M. (2006) Eﬀects of nutrient enrichment and grazing on shoalgrass Halodule wrightii and its epiphytes: results of a ﬁeld experiment. Marine Ecology-Progress Series, 326, 145–156. Howarth, R.W., Marino, R. & Cole, J.J. (1988) Nitrogen ﬁxation in freshwater, estuarine and marine ecosystems: biogeochemical controls. Limnology and Oceanography, 33, 688–701.

Ivancic, I., Radic, T., Lyons, D.M., Fuks, D., Precali, R. & Kraus, R. (2009) Alkaline phosphatase activity in relation to nutrient status in the northern Adriatic Sea. Marine Ecology-Progress Series, 378, 27–35. Lewis, W.M. & Wurtsbaugh, W.A. (2008) Control of lacustrine phytoplankton by nutrients: erosion of the phosphorus paradigm. International Review of Hydrobiology, 93, 446–465. Maberly, S.C., King, L., Dent, M.M., Jones, R.I. & Gibson, C.E. (2002) Nutrient limitation of phytoplankton and periphyton growth in upland lakes. Freshwater Biology, 47, 2136–2152. Myers, N. (1995) Environmental unknowns. Science, 269, 358–360. Rees, A.P., Hope, S.B., Widdicombe, C.E., Dixon, J.L., Woodward, E.M.S. & Fitzsimons, M.F. (2009) Alkaline phosphatase activity in the western English Channel: elevations induced by high summertime rainfall. Estuarine Coastal and Shelf Science, 81, 569–574. Rosemond, A.D. (1993) Interactions among irradiance, nutrients, and herbivores constrain a stream algal community. Oecologia, 94, 585–594. Sala, E. & Knowlton, N. (2006) Global marine biodiversity trends. Annual Review of Environment and Resources, 31, 93–122. Scott, J.T., Lang, D.A., King, R.S. & Doyle, R.D. (2009) Nitrogen ﬁxation and phosphatase activity in periphyton growing on nutrient diﬀusing substrata: evidence for diﬀerential nutrient limitation in stream periphyton. Journal of the North American Benthological Society, 28, 57–68. Smith, V.H. & Schindler, D.W. (2009) Eutrophication science: where do we go from here? Trends in Ecology & Evolution, 24, 201–207. Taylor, D., Nixon, S., Granger, S. & Buckley, B. (1995) Nutrient limitation and the eutrophication of coastal lagoons. Marine Ecology-Progress Series, 127, 235–244. Vitousek, P.M., Mooney, H.A., Lubchenco, J. & Melillo, J.M. (1997) Human domination of Earth’s ecosystems. Science, 277, 494–499. Received 8 April 2010; accepted 6 October 2010 Handling Editor: Marc Cadotte

Supporting Information Additional Supporting Information may be found in the online version of this article. Appendix S1. T-test results and conﬁdence intervals for all designated categories (ecosystem type and latitudinal zone). Appendix S2. Frequency of IEI values for each latitudinal zone (arctic, temperate, tropical) within each ecosystem type (freshwater, marine, terrestrial). In each plot, the white background bars indicate the frequency of IEI values for all experiments within that given ecosystem type (e.g. the ﬁrst row the white bars indicate the IEI values for all experiments in freshwater ecosystems). As a service to our authors and readers, this journal provides supporting information supplied by the authors. Such materials may be re-organized for online delivery, but are not copy-edited or typeset. Technical support issues arising from supporting information (other than missing ﬁles) should be addressed to the authors.

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 96–101

Journal of Applied Ecology 2011, 48, 220–227

doi: 10.1111/j.1365-2664.2010.01896.x

Long-term impact of changes in sheep Ovis aries densities on the breeding output of the hen harrier Circus cyaneus Arjun Amar1*, Jacob Davies2, Eric Meek3, Jim Williams4, Andy Knight3 and Steve Redpath5 1

RSPB–Scotland, Dunedin House, 25 Ravelston Terrace, Edinburgh, EH4 3TP, UK; 2Banks, Northside, Birsay, Orkney KW17 2LU, UK; 3RSPB–Scotland, 12 ⁄ 14 North End Rd., Stromness, Orkney KW16 3AG, UK; 4Fairholm, Finstown, Orkney KW17 2EQ, UK; and 5Aberdeen Centre for Environmental Sustainability, Aberdeen University & Macaulay Institute, Tillydrone Avenue, Aberdeen AB24 2TZ, UK

Summary 1. Livestock grazing is an important form of land use across the globe and changes in grazing pressure can have profound eﬀects on vertebrate populations. 2. In Scotland, over the last decade sheep numbers in many areas have declined from historically high levels, providing an opportunity to explore the implications of these declines for biodiversity. 3. The hen harrier Circus cyaneus is a bird of high conservation importance in the UK, and a species that may be heavily inﬂuenced by the indirect eﬀects of sheep on habitat and prey. The hen harrier population on the Orkney Islands in Scotland has been monitored since 1975 and represents an ideal case study for considering the impact of sheep de-stocking on a key predator. 4. Declines in the harrier population were associated with a doubling in sheep numbers between the early 1980s and the late 1990s. Subsequently, as sheep numbers have fallen the harrier population has recovered. These changes indicate an association but no clear mechanism, so we tested whether reductions in sheep numbers have led to increases in harrier prey or preferred foraging habitat. We then tested whether breeding output over the last 33 years correlates with sheep stocking levels or variation in weather conditions (rainfall and temperature). 5. Orkney sheep numbers declined by about 20% between 1998 and 2008. Surveys in 1999 ⁄ 2000 and repeated in 2008 showed increases in rough grassland, the preferred harrier foraging habitat, and increases in a key prey species, the Orkney vole Microtus arvalis orcadensis. 6. Overall, hen harrier breeding output over the last 33 years was signiﬁcantly negatively correlated to both sheep abundance and spring rainfall. 7. Synthesis and application. This study provides strong evidence for the consequences of changes in sheep numbers on a top predator. Our results indicate that reductions in sheep numbers are likely to prove beneﬁcial for some upland species, particularly small mammals and their predators. Key-words: agriculture, grasslands, grazing, grouse moors, Orkney, predation, rainfall, voles

Introduction The impacts of grazing on vegetation structure and composition, and on ecosystem processes have received considerable attention (Milchunas & Lauenroth 1993; Augustine & McNaughton 1998; Perevolotsky & Seligman 1998; Knapp et al. 1999; Cote et al. 2004; Hanley et al. 2008). In Britain, sheep Ovis aries are the principal domestic grazing species in the uplands, and between 1950 and 1990 their numbers rose *Correspondence author. E-mail: [email protected]

from 19Æ7 million to 41Æ2 million with particularly dramatic increases apparent during the 1980s and the early 1990s (Fuller & Gough 1999; Amar & Redpath 2005; Condliﬀe 2009). These increases stem from changes to the subsidy and support systems operated through the Common Agricultural Policy (CAP) (Fuller & Gough 1999; Hanley et al. 2008). Increased sheep abundance dramatically aﬀected some bird species, particularly in the uplands, and almost certainly reduced the habitat quality for some ground nesting bird species (Fuller & Gough 1999; Thirgood et al. 2000). In contrast,

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society

Grazing impacts on hen harriers 221 increased grazing may have beneﬁted those species preferring shorter or more grass-dominated vegetation (Smith et al. 2001; Pearce-Higgins & Grant 2006) and their predators, as well as carrion feeders which would have beneﬁted from the increase in sheep carcasses (Watson, Rae & Stillman 1992; Ratcliﬀe 1997; Fuller & Gough 1999; Redpath & Thirgood 1999). However, since the late 1990s sheep numbers in Scotland have declined quite dramatically (SAC 2008). The outcome for biodiversity of these declines will inevitably vary between vegetation types and between bird species (Pearce-Higgins et al. 2009). There is an urgent need to understand the biodiversity responses to spatial and temporal changes in grazing patterns, indeed understanding the consequences of changes in upland grazing regimes for biodiversity is currently viewed as one of the most important ecological questions of high policy relevance for the UK (Sutherland et al. 2006). Changes to vegetation structure or communities caused by high levels of grazing can have a direct impact on vertebrate and invertebrate species that rely on ungrazed or lightly grazed habitats (Jepson-Innes & Bock 1989; Steen, Mysterud & Austrheim 2005; Evans et al. 2006). Reductions in the abundance of these species can in turn inﬂuence the abundance of the predators that feed on them, although links between these diﬀerent trophic guilds have been poorly studied due to their inherent complexities (Duﬀ 1979; Douglass & Frisina 1993; Steenhof et al. 1999; Johnson & Horn 2008). For example, small mammal populations are known to be aﬀected by grazing levels (Steen, Mysterud & Austrheim 2005; Evans et al. 2006) and avian predators of small mammals, such as hen harriers Circus cyaneus whose populations can be strongly inﬂuenced by vole abundance (Redpath, Thirgood & Clarke 2002a), may therefore be inﬂuenced by changes in grazing densities. The Orkney Islands in North Scotland are an important breeding area for hen harriers in the UK (Sim et al. 2007). Most hen harriers on Orkney breed in the west of the island of Mainland (hereafter referred to as West Mainland) and this population has been monitored on the same area annually from 1975 (Amar et al. 2005). A doubling in sheep numbers is believed to have been responsible for a decline in hen harriers between the 1970s and 1990s when the numbers of chicks produced each year declined by 73% (Amar & Redpath 2005; Amar et al. 2005). The mechanism for this was thought to have been a reduction in rough grassland (the preferred habitat for foraging harriers) and Orkney voles, reducing the amount of prey available to harriers during the critical pre-laying period (Amar & Redpath 2002, 2005; Amar, Redpath & Thirgood 2003a). However, since the end of the 1990s the hen harrier population on Orkney has recovered. Between 1998 and 2004, the numbers of breeding females on Orkney increased by 118% from 34 to 74 (Sim et al. 2001, 2007), which contributed to an overall population increase for the UK, the Orkney population representing 12% of the Scottish population in 2004 (Sim et al. 2007). In this paper, we describe this population recovery on Orkney to 2008, and quantify the changes in sheep abundance

within their breeding and foraging areas. Secondly, we repeat vegetation and prey surveys ﬁrst undertaken in the late 1990s and test the hypothesis that a reduction in sheep numbers has allowed the amount of rough grassland and the abundance of key prey species to recover. We also test if hen harrier breeding output on Orkney correlates with weather variables because previous work has shown that hen harrier breeding success is inﬂuenced by the eﬀect of weather on prey delivery and nestling mortality (Picozzi 1984; Redpath et al. 2002b). Lastly, we explore if changes in sheep abundance and ⁄ or weather can account for the changes in breeding output of this harrier population over the last 33 years.

Materials and methods HARRIER DATA

Hen harriers have been monitored on Orkney (5910’ N, 312’ W) to a varying degree since 1953 (Amar et al. 2005). Since 1975 the same areas of moorland on West Mainland have been systematically monitored, with the total number of broods and total number of young produced being the minimum data recorded each year. This population decline dramatically between the 1970s and 1990s and intensive monitoring revealed that the key demographic change during this decline was a reduction in the proportion of breeding females (linked to lower levels of polygny) and a reduction in breeding success of secondary females, with little variation apparent in the brood size of successful nests (Amar et al. 2005). Both these factors led to a reduction in the number of broods ﬂedging and a lower number of young produced. Due to the labour intensive methods, data on the proportion of breeding females and levels of polygny are unavailable throughout the whole study period. For this study, we instead used the total number of young produced each year between 1975 and 2008. This measure combines several variables into a single productivity estimate, including numbers of breeding females and breeding success rate, together with the small variation in brood size at ﬂedging. During 1999 and 2000, a supplementary feeding experiment was undertaken and improved the population’s breeding performance (Amar & Redpath 2002). Therefore, data from these 2 years are excluded from our analyses examining productivity.

SHEEP AND WEATHER DATA

We obtained the total annual number of sheep from the seven regional areas or parishes (Birsay, Harray, Evie, Rendall, Firth, Orphir and Stenness) with breeding harriers on West Mainland from 1975 to 2008 from the June Agricultural Census data. Weather data for the same period was extracted from the Met Oﬃce MIDAS Land Surface Observation Stations dataset, held by the British Atmospheric Data Centre. All data came from the Kirkwall weather station (58Æ9N, 2Æ9W), situated about 20 km from the main breeding areas for harriers on West Mainland. Using the same divisions as Redpath et al.’s (2002b) analysis of weather on harrier breeding success, we summarized data from March and April as ‘spring’ and June and July as ‘summer’. Rainfall data were the sum of rainfall (mm) in both months in each season. Mean spring and summer maximum and minimum temperatures ( C) were derived from the average minimum and maximum temperatures from both months in each season. Temperature data were missing for a small number of seasons (max. spring = 4; max. summer = 2; min. spring = 4; min. summer = 2), so we used a fuller dataset from Lerwick, Shetland

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 220–227

222 A. Amar et al. (60Æ1N, 1Æ1W), about 100 miles north of West Mainland, to complete the Kirkwall data using predictive linear regressions (Whitﬁeld, Fielding & Whitehead 2008) (R2 for all correlations>0Æ81).

study), along two 1-km transects, giving 100 quadrats per square. Data were collected in July or August, and although this was later than our surveys, previous work in Orkney suggests that vole indices change little between these two time periods (Amar 2001).

PREY AND HABITAT SURVEYS

Surveys of Orkney voles, lagomorphs (rabbit Oryctolagus cuniculus and brown hare Lepus capensis), and meadow pipits Anthus pratensis, and the area of rough grassland were carried out using line transects within 18 1-km squares. Squares were selected (non-randomly) to allow unobstructed observation of hunting harriers (as part of another study), and were all sited within 5 km of active harrier territories, the locations of which remained similar from year to year (Amar & Redpath 2005; Amar et al. 2008). Within each square, surveys were undertaken along two parallel transects, located at 250 m and 750 m from a randomly selected side of each square. Surveys of voles, lagomorphs and rough grassland were carried out in spring, when voles formed the largest component of the hen harrier’s diet; surveys of meadow pipits took place in summer, when meadow pipits become an important part of the diet (Amar 2001). Rough grassland measures, and vole and lagomorph sign indices were recorded simultaneously. Quadrats (25 · 25 cm) were placed every 40 m along the two transect lines within each square, giving 50 quadrats per square. Vole and lagomorph abundance was indexed using the presence or absence of fresh vole or lagomorph droppings in each quadrat, and we recorded the number of quadrats dominated by unmanaged grass. Fresh vole droppings provided the most reliable index of relative vole abundance on Orkney as estimated by simultaneous snap trapping (Oates 1996). Rough grassland was deﬁned by a build up of dead vegetation forming a litter mat under the living vegetation, adequate to conceal a moving vole (Hewson 1982). Original surveys of voles, lagomorphs and rough grassland were undertaken in nine of the 18 squares between 23 and 31 March 1999, with the remaining nine surveyed between 2 and 24 March 2000. Repeat surveys were undertaken between 1 and 10 April 2008. Meadow pipits were surveyed using standard passerine transects along the same transect routes as the vole and vegetation surveys, between 06:00 and 09:00 h, as this period provided the highest repeatability in estimating passerine numbers (Thirgood, Leckie & Redpath 1995). Numbers of individuals seen within 200 m on either side of the transect line were recorded. Counts were undertaken between 2 July and 6 August in either 1998 or 1999 (all but two counts occurring by 20 July), with the repeat surveys undertaken slightly earlier between 24 June and 9 July 2008. This discrepancy in survey date (average of 10 days between surveys) is unlikely to have inﬂuenced our comparison of meadow pipit counts between surveys because we found no correlation between meadow pipit abundance and date (F1,34 = 0Æ29 P = 0Æ59).

LONGER-TERM VOLE MONITORING

We used longer-term vole abundance data from a separate study (Royal Society of the Protection of Birds unpublished data) to examine whether any changes found between our original and repeat surveys were likely to reﬂect real changes or larger-scale temporal ﬂuctuations. Vole abundance data collected annually from 1999 to 2008 were available from three 1-km squares in West Mainland. Habitat within these squares was principally heather moorland with some rough grassland with little or no grazing. These surveys used comparable methods to our study, recording the presence of fresh vole droppings within 25 · 25 cm quadrats placed every 20 m (cf. 40 m in our

STATISTICAL ANALYSIS

Trends in the numbers of young ﬂedged and the numbers of sheep were analysed using linear regression with a normal error structure. Changes in the abundance of prey (count of vole signs or meadow pipits) and in the number of rough grassland dominated quadrats were analysed using a Generalised Linear Mixed Model (GLMM), with a unique identiﬁer for ‘square’ as a random factor and survey period (original or resurvey) as a categorical ﬁxed eﬀect. For meadow pipit abundance, the GLMM would not converge, so for this analysis we used a Generalised Linear Model with ‘square’ and survey period as ﬁxed eﬀects. Models were ﬁtted with a Poisson error structures and a log link function and were corrected for over-dispersion. Denominator degrees of freedom for the GLMM were estimated using Satterthwaite’s formula (Littell et al. 1996). To examine the inﬂuence of weather variables and sheep abundance on the number of young ﬂedged, we used a General Linear Model, with a normal error structure and an identify link function. All analyses were carried out in sas version 9.1 (SAS Institute Inc. 2004).

Results TRENDS IN HARRIER BREEDING OUTPUT AND SHEEP ABUNDANCE

The number of young hen harriers ﬂedged declined from the end of the 1970s to a low during the 1990s before rising again at the start of the 2000s (Fig. 1). The number of young ﬂedged declined by 79% between 1975 and 1997 (F1,.21 = 32Æ73, P < 0Æ001) and then increased by 92% between 1998 and 2008 (F1,.9 = 6Æ53, P = 0Æ03). In contrast, sheep numbers increased by 140% between 1975 and 1997 (1975 – c. 20 000, 1997 – c. 48 000; F1,.21 = 418Æ96, P < 0Æ001); and declined by about 20% between 1998 and 2008 (1998 – c. 50 000, 2008 – c. 40 000; F1,.9 = 59Æ70, P < 0Æ001) (Fig. 1). CHANGES IN HABITAT AND PREY ABUNDANCE

We found an increase in the number of quadrats dominated by rough grassland (F1,15 = 4Æ97, P = 0Æ04) and an increase in quadrats with vole signs (F1,13 = 7Æ25, P = 0Æ01) between spring 1999 ⁄ 2000 and spring 2008 (Fig. 2). However, no diﬀerences in the number of lagomorph signs were found between the two surveys (F1,20 = 0Æ60; P = 0Æ44). We also found no diﬀerence in the number of meadow pipits counted between summer 1998 ⁄ 1999 and summer 2008 (F1,17 = 21Æ25, P = 0Æ26). Longer-term data on vole abundance from three 1-km squares subject to relatively constant low levels of grazing suggested that a wider pattern of temporal ﬂuctuations in vole abundance was unlikely to explain the increase in voles detected from the 18 squares surveyed in 2008. Indeed, these longer-term data displaying the annual ﬂuctuations, suggested that vole abundance was actually higher in 1999 and 2000 than it was in 2008 (Fig. 3).

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 220–227

Grazing impacts on hen harriers 223 60 000

90 80

50 000

50 30 000 40 20 000

30

Total sheep numbers

40 000

60

20 10 000 10 0 2007

2005

2003

2001

1999

1997

1995

1993

1991

1989

1985

1987

1983

1981

1979

1977

1975

0

Year

8

12

7

10

6

Vole signs

Number of quadrats

Fig. 1. Graph showing the total number of young hen harriers ﬂedged from West Mainland Orkney (open circles–the two close circles show the years when a diversionary feeding experiment took place) between 1975 and 2008 together with the 3-year running mean (dashed line). Also shown are the total numbers of sheep (closed squares) recorded between 1975 and 2008 from the June Agricultural Census in the seven parishes with breeding harriers.

Total young fledged

70

5 4 3

8 6 4

2

2

1

0 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 Year

0 Unmanaged grass

Voles

Fig. 2. Changes between 1999 ⁄ 2000 (shaded bars) and 2008 (clear bars) in rough grass abundance and fresh vole signs (per 50 quadrats) within eighteen 1-km squares distributed throughout West Mainland, Orkney. Data are mean (±1 SE) model estimates from the GLMM ﬁtted with survey period as a ﬁxed eﬀect and square as a random term in the model.

Fig. 3. Graph showing the inter-annual changes in vole signs (per 50 quadrats) between 1999 and 2008 on three 1 km squares. Squares were additional to the main study and were largely un-grazed throughout the period. Data shown are the mean counts taken in July ⁄ August across the three squares ± 1 SE. Data indicate that changes in vole abundance between 1999 and 2008 in the main study were not explained by broad-scale between year ﬂuctuations in vole abundance.

RELATIONSHIP BETWEEN CLIMATE, SHEEP DENSITIES AND HARRIER BREEDING SUCCESS

No relationship was found between total young ﬂedged and either spring or summer temperature or summer rainfall (Table 1a, Fig. 4). However, there was a signiﬁcant negative association between spring rainfall and numbers of young ﬂedged (Table 1a, Fig. 4). Between 1975 and 1997, spring rainfall increased (F1,.21 = 5Æ03, P = 0Æ04), although there was no trend between 1998 and 2008 (F1,.9 = 0Æ05, P = 0Æ83). We also found a highly signiﬁcant relationship between the abundance of sheep and the numbers of young ﬂedged between 1975 and 2008 (Table 1a). Combining these signiﬁcant terms in a full model, both the abundance of sheep (Fig. 5) and the spring rainfall (Fig. 6) remained signiﬁcant (Table 1b), with no signiﬁcant interaction (F1,.28 = 1Æ03, P = 0Æ31). These two terms in the ﬁnal model accounted for nearly 40% of the variation in the numbers of young ﬂedged between years (Table 1b).

Discussion This study provides convincing evidence that the decline and subsequent recovery of the hen harrier population on Orkney was due to changes in sheep abundance. It is important to note that this conclusion is not based on just a one way relationship auto-correlated with time, but with numbers of sheep both increasing and decreasing and harriers following the converse trend. Increases in sheep are thought to have reduced the amount of rough grassland, the habitat preferred by foraging male harriers (Amar et al. 2003b; Amar & Redpath 2005) which in turn led to a decrease in the abundance of voles, an important prey species which is heavily dependent on this habitat type (Amar & Redpath 2005). During the period of harrier recovery, sheep numbers declined in the main hen harrier breeding areas on Orkney by over 20%, with the loss of around

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 220–227

224 A. Amar et al. Table 1. Relationships between the number of young hen harriers ﬂedged on West Mainland, Orkney and the spring and summer climate variables (rainfall and min. and max. temperature) and sheep abundance recorded on Orkney from 1975 to 2008 (omitting 1999 and 2000), (a) univariate relationships for each term, (b) ﬁnal full GLM model including only terms signiﬁcant at the univariate stage. Signiﬁcant terms are presented in bold Variable

Intercept

Estimate

d.f.

(a) Spring rainfall Summer rainfall Spring max temp Summer max temp Spring min temp Summer min temp Total sheep

67Æ05 53Æ31 36Æ83 10Æ05 30Æ76 )2Æ40 77Æ28

)0Æ200 )0Æ160 )0Æ123 1Æ88 1Æ469 3Æ901 )0Æ001

1,30 1,30 1,30 1,30 1,30 1,30 1,30

(b) Spring rainfall Total sheep

)0Æ149 )0Æ001

96Æ16

Residual young fledged per year

Rainfall (mm)

0Æ01 0Æ08 0Æ98 0Æ67 0Æ74 0Æ47 <0Æ001

11Æ15

14Æ3% 6Æ6% 0Æ0% 0Æ0% 0Æ0% 0Æ0% 32Æ1%

0Æ04 0Æ001

39Æ6%

30

250 200 150 100 50

20 10 0 –10 –20 –30 –40 –50 80

1980

1985

1990 1995 Year

2000

2005

2010

Fig. 4. Annual rainfall data from Kirkwall weather station between 1975 and 2008, together with 3 year running means for spring (sum of March and April daily rainfall) (solid line and circles) and summer (sum of June and July daily rainfall) (dashed line and open circles).

50 Residual young fledged per year

R2

P

6Æ19 3Æ18 0Æ00 0Æ19 0Æ11 0Æ53 15Æ66

2,29

300

0 1975

F

40 30 20 10 0 –10 –20 –30 –40 20 000 25 000 30 000 35 000 40 000 45 000 50 000 55 000 Total sheep numbers

Fig. 5. Relationship between the total number of sheep on the seven parishes on West Mainland containing hen harriers (from the June agricultural records) and the residuals of the numbers of young hen harriers ﬂedged on West Mainland Orkney between 1975 and 2008 after controlling for the signiﬁcant relationship between of the total number of hen harriers ﬂedged and spring rainfall. Data from 1999 and 2000 when a supplementary feeding experiment took place are excluded.

130

180 Spring rainfall

230

280

Fig. 6. Relationship between the spring rainfall (mm), and the residuals of the numbers of young hen harriers ﬂedged on West Mainland Orkney between 1975 and 2008 after controlling for the signiﬁcant relationship between of the total number of hen harriers ﬂedged and the total number of sheep. Data from 1999 and 2000 when a supplementary feeding experiment took place are excluded.

10 000 sheep. Scottish Natural Heritage (SNH) launched the Orkney Hen Harrier Recovery Scheme in 2002 (http://www.snh.org.uk/pdfs/about/orkneyHH.pdf), and this, together with other agri-environment schemes in Orkney, is likely to have improved habitat conditions for harriers, through sheep removal and exclusion to increase vole abundance and rough grassland. However, the bulk of these sheep declines are probably unrelated to this speciﬁc scheme, given that they are typical of wider declines seen across northern and western Scotland, without such schemes. Rather, the decline in sheep numbers is more likely to be due to changes in wider subsidy reforms of the CAP (SAC 2008). Our study provides support for the hypothesis that the decrease in sheep numbers on Orkney has led to an increase in rough grassland habitat with signiﬁcant increases detected between the late 1990s and 2008. Modelling suggests that only relatively small increases in this habitat type are necessary to improve harrier breeding success (Amar et al. 2008). Previous research has identiﬁed that food shortages were most critical for this hen harrier population during the pre-lay

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 220–227

Grazing impacts on hen harriers 225 and incubation period (Amar & Redpath 2002; Amar, Redpath & Thirgood 2003a), when voles were the most important prey species. The selection of rough grassland habitats by foraging harriers (Amar & Redpath 2005) and the positive relationship between rough grass and breeding success (Amar et al. 2008) are thought to be a direct consequence of the higher vole abundance in habitats with more unmanaged and ungrazed grassland. Several non-experimental studies have suggested beneﬁts of reduced grazing on vole densities (Hewson 1982; Hill, Evans & Bell 1992) and these are now supported by experimental studies (Steen, Mysterud & Austrheim 2005; Evans et al. 2006). Our study provides further support for these ﬁndings, with signiﬁcant increases in vole abundance also apparent on Orkney over the time that sheep numbers declined and rough grass increased. Furthermore, this study advances our knowledge, by demonstrating how changes in grazing pressure may aﬀect predatory species which depend on voles as a key prey species. The inﬂuence of grazing levels on the vegetation type and structure, and therefore on vole abundance, will vary depending on the original vegetation. For example, some studies on mainland Scotland have demonstrated that higher levels of historical grazing on moorlands can increase the ratio of grass to heather, thus improving the quality of habitat for meadow pipits and ﬁeld voles Microtus agrestis, and therefore favouring higher hen harrier settling densities (Redpath & Thirgood 1997, 1999; Smith et al. 2001). The eﬀects of changes in sheep numbers are therefore likely to vary both spatially and temporally. Thus, decreases in sheep numbers might cause improved conditions in the short term, but in some areas, particularly areas where hen harrier home ranges are dominated by heather, these changes could reduce habitat quality for hen harriers in the longer term. Unfortunately, the June Agricultural Census does not attribute sheep numbers to habitat types and therefore we are unable to determine whether sheep reductions varied between habitats on Orkney. Anecdotal observations suggest that sheep abundance was never very high in most of the moorland nesting areas, especially as most are under conservation management by the Royal Society of the Protection of Birds (RSPB) or are designated Sites of Special Scientiﬁc Interest (SSSIs) on which management agreements with landowners have kept grazing levels down. Declines are more likely on marginal habitats, and given that these habitats already contain a higher ratio of grass-to-heather (Amar 2001) than moorland habitats, any reduction in sheep numbers may provide disproportionately positive beneﬁts. Previous grazing in these habitats may have increased the plant species that voles prefer, although until the grazing pressure is relieved the vegetation structure may not have been suitable for voles to colonize. Indeed improvement in vole habitat by limited grazing was considered a possibility from a grazing experiment in Norway where voles appeared to beneﬁt from light grazing over controls with no grazing (Steen, Mysterud & Austrheim 2005) and in Iceland, grazing by sheep increased the cover of Carex bigelowii, the main forage plant of both sheep and voles (Jo´nsdo´ttir 1991). This might also explain why the recovery of harriers on Orkney

was so rapid following the start of the decline in sheep numbers, and we know that voles can respond very quickly to reduced grazing, with eﬀects becoming apparent after only 1 year (Evans et al. 2006). Caution must be applied when comparing two points from a time series since change between points may be due to variation around a long-term trend, rather than being due to the trend itself. The longer-term vole data available from the study area suggested that inter-annual variation was unlikely to account for the contrast in vole numbers between the ﬁrst and second vole surveys in this study. Data from 2008 did not suggest that this year represented a particular peak in vole abundance in Orkney; indeed this longterm data suggested that 2008 was a year of relatively low vole abundance compared with 1999 ⁄ 2000. Thus, our ﬁndings on vole abundance in 2008 may actually underestimate the real level of increase. These data also suggest that although Orkney vole populations appear to show large ﬂuctuations between years, there is no evidence that these ﬂuctuations are cyclic in nature, unlike many other vole populations on mainland Britain (Lambin, Petty & MacKinnon 2000; Redpath, Thirgood & Clarke 2002a) or continental Europe (Oksanen & Oksanen 1992; Lambin, Bretagnolle & Yoccoz 2006). Although accounting for less variation than sheep numbers, spring rainfall also had an negative inﬂuence on hen harrier breeding output, most probably through reducing the amount of time harriers could spend hunting (Newton 1986; Sergio 2003). Harriers rarely hunt in rainy conditions (A. Amar, pers, obs.) and Redpath et al. (2002b) found that prey provisioning to nests by male harriers was reduced by rainfall. High rainfall in spring may therefore reduce the amount of food males can supply to their females during the critical pre-lay period in spring. Steenhof, Kochert & MacDonald (1997) found that prey and weather (number of days above 32 C) interacted to limit golden eagle Aquila chrysaetos productivity; in our study however, spring rainfall appeared to act independently of sheep numbers. Thus, even when sheep numbers were low and prey were therefore considered to be more abundant, rainfall had a similar negative eﬀect as when sheep numbers were high and prey was therefore assumed to be less abundant. Should climate change result in increases in spring rainfall our results suggest that this may have negative consequences for the harrier population on Orkney. National surveys of hen harriers between 1998 and 2004 showed considerable increases on other Scottish islands and in the west of Scotland, with declines only apparent in areas where grouse moor management was most prevalent, presumably due to illegal killing which is heavily associated with management for recreational shooting of red grouse (Etheridge, Summers & Green 1997; Sim et al. 2007; Thompson et al. 2009; Redpath et al. 2010). Many of these western areas have also seen considerable declines in sheep abundance (SAC 2008), and may have also contributed to the increases in these harrier populations. Declines in sheep numbers may have implications for other vole predators of conservation importance. For example, voles are an important component in the

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 220–227

226 A. Amar et al. diet of kestrels Falco tinnunculus (Village 1990), a rapidly declining species in Scotland (Risely, Noble & Baillie 2008). However, voles are also important prey for foxes Vulpes vulpes and pine martins Martes martes, and these species can be important predators of species of conservation concern, such as curlew (Grant et al. 1999) and capercaillie Tetrao urogallus (Summers, Willi & Selvidge 2009). Therefore, in theory, changes in grazing pressure may have far reaching indirect eﬀects on other bird species through changing the abundance of their main predators. Changes to agricultural support mechanisms can inﬂuence stocking densities of grazing animals and changes to grazing levels can have profound ecological impacts on vegetation and, as this study demonstrates, these changes can have considerable inﬂuence on other grazing species, such as small mammals and their predators. Thus, the large reductions in sheep numbers recorded over the last decade in the Scottish uplands are likely to beneﬁt some species but to have negative eﬀects on others. ‘Winning species’ may include small mammals and their predators, or other species that beneﬁt from nesting or feeding opportunities created by the changes in sward structure described in this study, such as ground nesting passerines. However, ‘losing species’ may include species, which prefer shorter vegetation such as golden plover Pluvialis apricaria or skylarks Alauda arvensis, which beneﬁt from the maintenance of more extensive areas of short, open swards (Pearce-Higgins & Grant 2006). Similar reductions in the abundance of open country bird species have occurred in the Alps (Laiolo et al. 2004) and Argentina (Garcia et al. 2008) following the loss of livestock. Some have speculated that reduced grazing as a consequence of lower sheep numbers may be compensated by increases in the abundance of deer (Clutton-Brock, Coulson & Milner 2004); Orkney has no wild deer, and this should be considered when transferring these ﬁndings to habitats on the mainland where deer are present. The potential implications of changes in sheep grazing for biodiversity are of broad policy interest. Our results demonstrate that policy-driven changes in grazing intensity, which although not designed with clear biodiversity objectives, nonetheless, can have a large eﬀect on a species of high conservation concern through impacts on the food chain.

Acknowledgements We are very grateful to the many professional researchers and volunteers that have surveyed the harrier populations over the years. JD is grateful for support from Ben Sheldon. In particular, we are grateful to many RSPB staﬀ and Orkney members of the Scottish Raptor Study Group. Hen harrier surveys on Orkney have beneﬁted from funding provided by both RSPB and SNH. Lastly, we thank Jeremy Wilson, Vicki Swales, Helen Riley, the editors and two anonymous referees for comments that greatly improved the manuscript.

References Amar, A. (2001) Determining the cause of the hen harrier decline on Orkney. PhD thesis, University of Aberdeen. Amar, A. & Redpath, S. (2002) Determining the cause of the hen harrier decline on the Orkney Islands: an experimental test of two hypotheses. Animal Conservation, 5, 21–28.

Amar, A. & Redpath, S. (2005) Habitat use by hen harriers Circus cyaneus on Orkney: implications of land use change on this declining population. Ibis, 147, 37–47. Amar, A., Redpath, S. & Thirgood, S. (2003a) Evidence for food limitation in the declining hen harrier population on the Orkney Island, Scotland. Biological Conservation, 111, 374–388. Amar, A., Redpath, S., Lambin, X. & Meek, E. (2003b) Could the hen harrier decline on Orkney be due to a shortage of food? Birds of Prey in a Changing Environment (eds D.B. Thompson, S.M. Redpath, A. Fielding, M. Marquiss & C.A. Galbraith), pp. 377–391. The Stationery Oﬃce, Edinburgh. Amar, A., Picozzi, N., Meek, E.R., Lambin, X. & Redpath, S.M. (2005) Decline of the Orkney Hen Harrier Circus cyaneus population: do changes to demographic parameters and mating system ﬁt a declining food hypothesis? Bird Study, 52, 18–24. Amar, A., Arroyo, B.E., Meek, E., Redpath, S. & Riley, H. (2008) Inﬂuence of habitat on occupancy and breeding performance of hen harriers in a declining population. Ibis, 150, 400–404. Augustine, D.J. & McNaughton, S.J. (1998) Ungulate eﬀects on the functional species composition of plant communities: herbivore selectivity and plant tolerance. Journal of Wildlife Management, 4, 1165–1183. Clutton-Brock, T.H., Coulson, T. & Milner, J.M. (2004) Red deer stocks in the highlands of Scotland. Nature, 429, 261–262. Condliﬀe, I. (2009) Policy change in the uplands. Drivers of Environmental Change in Uplands (eds A. Bonn, T. Allot, K. Hubacek & J. Stewart), pp. 59–90. Routledge, London ⁄ New York. Cote, S.D., Rooney, T.P., Tremblay, J.P., Dussault, C. & Waller, D.M. (2004) Ecological impacts of deer overabundance. Annual Review of Ecology, Evolution and Systematics, 35, 113–147. Douglass, R.J. & Frisina, M.R. (1993) Mice and management on Mount Haggin Wildlife Management Area. Rangelands, 15, 8–12. Duﬀ, D.A. (1979) Ripearian habitat recovery on Big Creek, Rich County, Utah. Proceedings of the Forum – Grazing and Riparian ⁄ Stream Ecosystems (ed. O.B. Cope), pp. 91–92. Trout Unlimited, Denver, CO. Etheridge, B., Summers, R.W. & Green, R.E. (1997) The eﬀects of illegal killing and destruction of nests by humans on the population dynamics of the hen harrier Circus cyaneus in Scotland. Journal of Applied Ecology, 34, 1081– 1105. Evans, D.M., Redpath, S.M., Elston, D.A., Evans, S.A., Mitchell, R.J. & Dennis, P. (2006) To graze or not to graze? Sheep, voles, forestry and nature conservation in the British uplands Journal of Applied Ecology, 43, 499–505. Fuller, R.J. & Gough, S.J. (1999) Changes in sheep numbers in Britain: implications for bird populations. Biological Conservation, 91, 73–89. Garcia, C., Renison, D., Cingolani, A.M. & Fernandez-Juricic, E. (2008) Avifaunal changes as a consequence of large scale livestock exclusion in the mountains of Central Argentina. Journal of Applied Ecology, 45, 351–360. Grant, M.C., Orsman, C., Easton, J., Lodge, C., Smith, M., Thompson, G., Rodwell, S. & Moore, N. (1999) Breeding success and causes of breeding failure of curlew Numenius arquata in Northern Ireland. Journal of Applied Ecology, 36, 59–74. Hanley, N., Davies, A., Angelopoulos, K., Hamilton, A., Ross, A., Tinch, D. & Watson, F. (2008) Economics determinants of biodiversity change over a 400-year period in the Scottish uplands. Journal of Applied Ecology, 45, 1557–1565. Hewson, R. (1982) The eﬀect upon Field vole (Microtus agrestis) habitat on removing sheep from moorland in west Scotland. 1982. Journal of Zoology, 197, 304–307. Hill, M.O., Evans, D.F. & Bell, S.A. (1992) Long-term eﬀects of excluding sheep from hill pastures in North Wales. Journal of Ecology, 80, 1–13. Jepson-Innes, K. & Bock, C.E. (1989) Response of grasshoppers (Orthoptera: Acrididae) to livestock grazing in southeastern Arizona: diﬀerences between seasons and subfamilies. Oecologia, 78, 430–431. Johnson, M.D. & Horn, C.M. (2008) Eﬀects of rotational grazing on rodents and raptors in a coastal grassland. Western North American Naturalist, 68, 444–452. Jo´nsdo´ttir, I.S. (1991) Eﬀects of grazing on tiller size and population dynamics in a clonal sedge (Carex bigelowii). Oikos, 62, 177–188. Knapp, A.K., Blair, J.M., Briggs, J.M., Collins, S.L., Hartnett, D.C., Johnson, L.C. & Towne, E.G. (1999) The keystone role of Bison in North American Tallgrass Prairie. BioScience, 49, 39–50. Laiolo, P., Dondero, F., Ciliento, E. & Rolando, A. (2004) Consequences of pastoral abandonment for the structure and diversity of the alpine avifauna. Journal of Applied Ecology, 41, 294–304. Lambin, X., Bretagnolle, V. & Yoccoz, N.G. (2006) Vole population cycles in northern and southern Europe: is there a need for diﬀerent explanations for single pattern? Journal of Animal Ecology, 75, 340–349.

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 220–227

Grazing impacts on hen harriers 227 Lambin, X., Petty, S.J. & MacKinnon, J.L. (2000) Cyclic dynamics in ﬁeld vole populations and generalist predation. Journal of Animal Ecology, 69, 106– 118. Littell, R.C., Milliken, G.A., Stroup, W.W. & Wolﬁnger, R.D. (1996) SAS System for Mixed Models. SAS Institute Inc., Cary, NC. Milchunas, D.G. & Lauenroth, W.K. (1993) Quantitative eﬀects of grazing on vegetation and soils over a global range of environments. Ecological Monographs, 63, 327–366. Newton, I. (1986) The Sparrowhawk. Carlton, Poyser. Oates, G. (1996) Designing and validating Orkney vole abundance indices based on live-trapping and counts of signs. Unpublished BSc. Hons. thesis. University of Aberdeen. Oksanen, L. & Oksanen, T. (1992) Long-term microtine dynamics in north Fennoscandian tundra: the vole cycle and the lemming chaos. Ecography, 15, 226–236. Pearce-Higgins, J.W. & Grant, M.C. (2006) Relationships between bird abundance and the composition and structure of moorland vegetation. Bird Study, 53, 112–125. Pearce-Higgins, J.W., Grant, M.C., Beale, C.M., Buchanan, G.M. & Sim, I.M.W. (2009) International importance and drivers of change of upland bird populations. Drivers of Change in Upland Environments (eds A. Bonn, K. Hubacek, J. Stewart & T. Allott), pp. 209–227. Routledge, Oxford, UK. Perevolotsky, A. & Seligman, N.G. (1998) Degradation of Mediterranean rangeland ecosystems by grazing: inversion of a paradigm. BioScience, 48, 1007–1017. Picozzi, N. (1984) Breeding biology of polygynous hen harriers Circus c. cyaneus in Orkney. Ornis Scandinavica, 15, 1–10. Ratcliﬀe, D. (1997) The Raven: A Natural History of Britain and Ireland. T & A D Poyser, London. Redpath, S.M. & Thirgood, S.J. (1997) Birds of Prey and Red Grouse. Stationery Oﬃce, London. Redpath, S.M. & Thirgood, S.J. (1999) Numerical and functional responses in generalist predators: hen harriers and peregrines on Scottish grouse moors. Journal of Animal Ecology, 68, 879–892. Redpath, S.M., Thirgood, S.J. & Clarke, R. (2002a) Field Vole Microtus agrestis abundance and Hen Harrier Circus cyaneus diet and breeding in Scotland. Ibis, 144, E33–E38. Redpath, S.M., Arroyo, B.E., Etheridge, B., Leckie, F., Bouwman, K. & Thirgood, S.J. (2002b) Temperature and hen harrier productivity: from local mechanisms to geographical patterns. Ecography, 25, 533–540. Redpath, S., Amar, A., Smith, A., Thompson, D. & Thirgood, S. (2010). People and nature in conﬂict: can we reconcile raptor conservation and game management? Species Management: Challenges and Solution for the 21st Century (eds J. Baxter & C.A. Galbraith), pp. 335–350. The Stationary Oﬃce, Edinburgh. Risely, K., Noble, D.G. & Baillie, S.R. (2008) The Breeding Bird Survey 2007. BTO Research Report 508, British Trust for Ornithology, Thetford. SAC. (2008) Farming’s Retreat from the Hills. SAC, Edinburgh. SAS Institute Inc. (2004) SAS ⁄ STAT 9.1 User’s Guide. SAS Institute Inc., Cary, NC. Sergio, F. (2003) From individual behaviour to population pattern: weatherdependent foraging and breeding performance in black kites. Animal Behaviour, 66, 1109–1117.

Sim, I.M.W., Gibbons, D.W., Bainbridge, I.P. & Mattingley, W.A. (2001) Status of the Hen Harrier Circus cyaneus in the UK and the Isle of Man in 1998. Bird Study, 48, 341–353. Sim, I.M.W., Dillon, I.A., Eaton, M.A., Etheridge, B., Lindley, P., Riley, H., Saunders, R., Sharpe, C. & Tickner, M. (2007) Status of the Hen Harrier Circus cyaneus in the UK and the Isle of Man in 2004, and a comparison with the 1988 ⁄ 89 and 1998 surveys. Bird Study, 54, 256–267. Smith, A., Redpath, S.M., Campbell, S.T. & Thirgood, S.J. (2001) Relationships between habitat characteristics of managed grouse moors and the abundance of meadow pipits and red grouse. Journal of Applied Ecology, 38, 390–400. Steen, H., Mysterud, A. & Austrheim, G. (2005) Sheep grazing and rodent populations: evidence of negative interactions from a landscape scale experiment. Oecologia, 143, 357–364. Steenhof, K., Kochert, M.N. & MacDonald, T.L. (1997) Interactive eﬀects of prey and weather on golden eagle reproduction. Journal of Animal Ecology., 66, 350–362. Steenhof, K., Kochert, M.N., Carpenter, L.B. & Lehman, R.N. (1999) Longterm prairie falcon population changes in relation to prey abundance, weather, land uses, and habitat conditions. The Condor, 101, 28–41. Summers, R.W., Willi, J. & Selvidge, J. (2009) Capercaillie Tetrao urogallus nest loss and attendance at Abernethy Forest, Scotland. Wildlife Biology, 15, 319–327. Sutherland, W.J., Armstrong-Brown, S., Armsworth, P.R., Tom, B., Brickland, J., Campbell, C.D., Chamberlain, D.E., Cooke, A.I., Dulvy, N.K., Dusic, N.R., Fitton, M., Freckleton, R.P., Godfray, H.C.J., Grout, N., Harvey, H.J., Hedley, C., Hopkins, J.J., Kift, N.B., Kirby, J., Kunin, W.E., Macdonald, D.W., Marker, B., Naura, M., Neale, A.R., Oliver, T., Osborn, D., Pullin, A.S., Shardlow, M.E.A., Showler, D.A., Smith, P.L., Smithers, R.J., Solandt, J.-L., Spencer, J., Spray, C.J., Thomas, C.D. & Thompson, J. (2006) The identiﬁcation of 100 ecological questions of high policy relevance in the UK. Journal of Applied Ecology, 43, 617–627. Thirgood, S.J., Leckie, F.M. & Redpath, S.M. (1995) Diurnal and seasonal variation in line transect counts of moorland passerines. Bird Study, 42, 257– 259. Thirgood, S.J., Redpath, S.M., Haydon, D.T., Rothery, P., Newton, I. & Hudson, P.J. (2000) Habitat loss and raptor predation: disentangling long- and short-term causes of red grouse declines. Proceedings of the Royal Society London Series B, 267, 651–656. Thompson, P., Amar, A., Hoccom, D.G., Knott, J. & Wilson, J.D. (2009) Resolving the conﬂict between driven-grouse shooting and conservation of hen harriers. Journal of Applied Ecology, 46, 950–954. Village, A. (1990) The Kestrel. T & A D Poyser, London, UK. Watson, J., Rae, S.R. & Stillman, R. (1992) Nesting density and breeding success of Golden Eagle Aquila chrysaetos in relation to food supply in Scotland. Journal of Animal Ecology, 61, 543–550. Whitﬁeld, D.P., Fielding, A.H. & Whitehead, S. (2008) Long-term increase in the fecundity of hen harriers in Wales is explained by reduced human interference and warmer weather. Animal Conservation, 11, 144–152. Received 13 May 2010; accepted 20 October 2010 Handling Editor: Des Thompson

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 220–227

Journal of Applied Ecology 2011, 48, 133–142

doi: 10.1111/j.1365-2664.2010.01897.x

Cost-effective age structure and geographical distribution of boreal forest reserves Johanna Lundstro¨m1*, Karin O¨hman2, Karin Perhans1,4, Mikael Ro¨nnqvist3 and Lena Gustafsson1 1

Department of Ecology, Swedish University of Agricultural Sciences, PO Box 7044, SE-750 07 Uppsala, Sweden; Department of Forest Resource Management, Swedish University of Agricultural Sciences, SE-901 83 Umea˚, Sweden, 3Department of Finance and Management Science, Norwegian School of Economics and Business Administration, NO-5045 Bergen, Norway; and 4The Ecology Centre, University of Queensland, QLD 4072, Australia 2

Summary 1. Forest reserves are established to preserve biodiversity, and to maintain natural functions and processes. Today there is heightened focus on old-growth stages, with less attention given to early successional stages. The biodiversity potential of younger forests has been overlooked, and the costeﬀectiveness of incorporating diﬀerent age classes in reserve networks has not yet been studied. 2. We performed a reserve selection analysis in boreal Sweden using the Swedish National Forest Inventory plots. Seventeen structural variables were used as biodiversity indicators, and the cost of protecting each plot as a reserve was assessed using the Heureka system. A goal programming approach was applied, which allowed inclusion of several objectives and avoided a situation in which common indicators aﬀected the result more than rare ones. The model was limited either by budget or area. 3. All biodiversity indicators were found in all age classes, with more than half having the highest values in ages ‡ 100 years. Several large-tree indicators and all deadwood indicators had higher values in forests 0–14 years than in forests 15–69 years. 4. It was most cost-eﬀective to protect a large proportion of young forests since they generally have a lower net present value compared to older forests, but still contain structures of importance for biodiversity. However, it was more area-eﬀective to protect a large proportion of old forests since they have a higher biodiversity potential per area. 5. The geographical distribution of reserves selected with the budget-constrained model was strongly biassed towards the north-western section of boreal Sweden, with a large proportion of young forest, whereas the area-constrained model focussed on the south-eastern section, with dominance by the oldest age class. 6. Synthesis and applications. We show that young forests with large amounts of structures important to biodiversity such as dead wood and remnant trees are cheap and cost-eﬃcient to protect. This suggests that reserve networks should incorporate sites with high habitat quality of diﬀerent forest ages. Since young forests are generally neglected in conservation, our approach is of interest also to other forest biomes where biodiversity is adapted to disturbance regimes resulting in open, early successional stages. Key-words: biodiversity, conservation planning, early succession, goal programming, indicator, old-growth, reserve selection, Swedish National Forest Inventory, young forest

Introduction The boreal forest belt runs circumpolar in the Northern Hemisphere and comprises more than 30% of all global forest area, *Correspondence author. E-mail: [email protected]

with Russia and Canada having by far the largest area covered in forest (FAO 2006). Although northern Europe has a comparatively small forested area, forests are the dominant ecosystem type, and the forest industry is a major export industry in several northern European countries. For example, intensive forestry has been conducted since the mid 20th

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society

134 J. Lundstro¨m et al. century in Sweden on a large proportion of the productive forest land, and this has led to even-aged and fragmented forest landscapes with small amounts of important features for biodiversity such as dead wood and old trees (O¨stlund, Zackrisson & Axelsson 1997; Lo¨fman & Kouki 2001). In the past boreal forests were mainly shaped by ﬁre (Zackrisson 1977), leading to a dynamic forest structure consisting of stands with diﬀerent ages (Angelstam 1998). As a consequence of ﬁre exclusion and suppression, today few natural forests remain and hundreds of species are threatened (Esseen et al. 1997; Ga¨rdenfors 2005). To mitigate this situation, reserves have been established with the aim of maintaining natural functions and processes, and preserving indigenous species in viable populations (Swedish Government 2005). Today, 1% of the productive boreal forests in Sweden below the mountain region are protected in reserves or in national parks (Anonymous 2007a). To rationalize the process of ﬁnding and designing reserves, a systematic approach to conservation planning has been developed in which complementarity and representativeness are key aspects (Margules & Pressey 2000). One way for forests in a reserve network to represent a natural range of structures and composition is to include diﬀerent successional stages (Junninen et al. 2006). This composition approach has been largely overlooked in forest conservation strategies. For instance, in boreal Sweden, 76% of the protected forests are over 100 years and only 1Æ5% are under 15 years (NFI data). Since humans have aﬀected the forests in boreal Fennoscandia for centuries, knowledge of the natural age composition in the forest landscape is lacking. Most probably, the amount of open areas and remnant structures varied substantially in space and time depending on disturbance pattern, indicating a complex forest composition (Kuuluvainen 2009). A post-disturbance deciduous forest with large amounts of dead wood is one of the most species-rich forest types in the boreal zone because of the possibility of existence for species adapted to both early and late successional stages (Esseen et al. 1997). Young natural forests have a unique species composition not found in any other successional stage (Spies & Franklin 1991; Swanson et al. 2010) and early successional stages are considered important for the protection of some red-listed species (Tikkanen et al. 2006). Many of the species associated with old-growth forests might not be dependent on old forests per se, but more on structures occurring there, e.g. dead wood (Kouki et al. 2001). To our knowledge, research has not been conducted on the value of including forests of diﬀerent ages in boreal forest reserve networks. The annual National Forest Inventory (NFI) in Sweden oﬀers an opportunity for such studies because of the large number of plots surveyed and the data collected on a number of structural variables such as dead wood, tree species, tree sizes and ages (Ranneby et al. 1987). Since the economic value of forests varies greatly with forest age, economic aspects are also important to consider, and can be calculated with high precision for the NFI plots. The main aim of our study was to analyse the cost-eﬀectiveness and biodiversity potential of protecting forests in diﬀerent age classes. To investigate this, optimal combinations of forest

age classes with diﬀerent constraints regarding area and budget were identiﬁed. A reserve selection framework building on optimization models was applied with the potential for biodiversity assessed from structural forest characteristics. In addition, geographical diﬀerences in the distribution of reserves were studied using models constrained by area and budget.

Materials and methods STUDY AREA AND NFI

The study area covered the counties of Va¨rmland, O¨rebro, Dalarna, Ga¨vleborg, Ja¨mtland, Va¨sternorrland, Va¨sterbotten and Norrbotten (about 14 million ha of productive forest land), an area roughly coinciding with the boreal zone in Sweden (Ahti, Ha¨met-Ahti & Jalas 1968). The boreal forest is characterized by a relatively homogeneous structure dominated by Scots pine Pinus sylvestris L. and Norway spruce Picea abies (L.) Karst. The main broad-leaved trees are birches Betula pendula Roth. and B. pubescens Ehrh. and aspen Populus tremula L. (Gustafsson & Ahle´n 1996). NFI is an annual survey of all land in Sweden that started in 1923 (Anonymous 1932), with a present systematic cluster design that was established in 1983 (Ranneby et al. 1987). Circular sample plots with a diameter of 7 or 10 m are clustered along the borders of square tracts, with a total of approximately 11 000 plots surveyed each year (Anonymous 2007b). Two-thirds of the tracts are permanent and are surveyed every ﬁfth year, whereas the remaining one-third of the tracts are surveyed only once. Plot numbers and tract sizes diﬀer between regions and between permanent and temporary tracts (Anonymous 2005). In this study, temporary and permanent NFI-plots within the study area were used from productive forest land outside of reserves and surveyed between the years 2003 and 2007 (17 599 plots in total). For the analysis, the plots were aggregated into 112 larger 50 · 50 km plots; 292 NFI-plots were excluded since a 50 · 50 km plot had to contain a minimum of 30 NFI-plots. The excluded plots were located along the outer border of the study area. The 50 · 50 km plots were in turn grouped into 6 geographical regions (Fig. 1) matching administrative county borders. Four smaller counties were grouped together (two and two) in order to make the regions more equally sized. The forest within each aggregated 50 · 50 km plot was divided into ﬁve age classes: 0–14, 15–39, 40–69, 70–99 and ‡ 100 years, with the total area in each age class comprising 2Æ3, 3Æ4, 2Æ9, 2Æ0 and 3Æ6 million ha respectively. Age class division was based on the introduction of tree retention practices (Lindenmayer & Franklin 2003) about 15 years ago, and on a normal rotation time of about 100 years.

BIODIVERSITY INDICATORS

Structure-based indicators registered in the NFI were used as proxies for biodiversity potential. We chose indicators based on what are commonly thought to be the important substrates and environmental conditions for a majority of forest species (Ferris & Humphrey 1999; Lindenmayer, Margules & Botkin 2000; Spanos & Feest 2007): structural heterogeneity in the form of gaps and water dynamics (Kuuluvainen 2002); uneven age and multi-layered tree canopy (Esseen et al. 1997); deciduous trees (both old trees in conifer forests and younger ones in early successional deciduous forests) (Hagar 2007); and large-diameter trees and dead wood (Nilsson, Hedin & Niklasson 2001; Siitonen 2001). The underlying assumption was that

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 133–142

Cost-eﬀective boreal forest reserves 135 OPPORTUNITY COST

The opportunity cost, i.e. the economic value of the plots, was estimated by using the net present value (NPV) from future forestry activities. The NPV was based on the sum of income and cost of future activities from timber harvest from period one to inﬁnity discounted back to today with a 3% interest rate. This is consistent with the value of an investment of moderate risk in northern Europe (Ibbotson & Sinqueﬁeld 1986). For each plot, up to 100 diﬀerent treatment schedules were ﬁrst simulated using the Heureka system and then the most proﬁtable schedule was selected to calculate the economic value. The Heureka system is a newly developed planning system for multiple-use forestry (La¨ma˚s et al. 2006). Each treatment schedule consisted of variable timing for the associated silvicultural activities from period one to inﬁnity (i.e. thinning and clear cutting with appropriate regeneration following a harvest). Economic data (timber, regeneration and harvesting costs) used for calculating the NPV for each schedule were based on a timber price list for 2008 retrieved from the forest owners’ association in northern Sweden, and was representative for the region.

DECISION MODELS

Fig. 1. Map of the study area. The analysed NFI-plots were grouped into 50 · 50 km large plots which were further aggregated into 6 geographical regions, broadly following the borders of administrative counties: 1. Norrbotten, 2. Va¨sterbotten, 3. Ja¨mtland, 4. Va¨sternorrland and Ga¨vleborg, 5. Dalarna, 6. Va¨rmland and O¨rebro.

the potential for high richness of indigenous species, including rare species, would increase if the presence of structural indicators was maximized. Forests with large trees, plentiful dead wood and dying trees, and a high proportion of broad-leaved trees should currently be prioritized in forest conservation according to political guidelines (Swedish Government. 2005). The registration of each indicator on the NFI plots was translated into a point between 0 and 100 (Table 1), and each aggregated 50 · 50 km plot was given a biodiversity indicator point ha)1 for each indicator and age class. The biodiversity indicator point ha)1 was calculated as the sum of points (for each indicator) from all NFI plots in the 50 · 50 km plot (in each age class) divided by the total area that all NFI plots in the 50 · 50 km plot (in each age class) represent. The indicator ‘volume of dead wood’ was calculated as the total volume of dead wood (for each age class) in the 50 · 50 km plot divided by the total area (for each age class). This point was normalized from 0 to 100 for all volumes less than or equal to 20 m3 ha)1. Volumes greater than 20 m3 ha)1 were given 100 points to avoid disproportional inﬂuence of outliers. The presence of the exotic Pinus contorta Dougl. var. latifolia Engelm and the stand character ‘plantation character’ (no structures from previous stands, > 90% of the trees of the same species, even-aged and consisting of one layer) were considered negative for biodiversity, and plots with these registrations were excluded from the selection (1005 plots) since it would not be realistic to establish a reserve in an area containing a plantation.

There were large diﬀerences in points between indicators; therefore a method that could accommodate these variations was essential. When using ordinary linear programming (LP) it is necessary to ﬁnd weights for each indicator to include into one single objective function. We used a goal programming approach to allow impartial treatment of all indicators and avoid manual weight determination. In a two-phase approach, we ﬁrst found the best possible outcome for each indicator, which became a goal. In the second phase, we searched for a solution that was as close as possible to each individual indicator goal but that considered all indicators at the same time. Each of the models used are described below (see Table 2 for parameters and decision variables). The ﬁrst LP problem can be formulated as follows: XXX ½P1 max z ¼ we pite xit eqn 1 i2I t2T e2E

subject to XX cit xit b

eqn 2

i2I t2T

XX i2I t2T

xit q

XX

ait

eqn 3

i2I t2T

xit ait ; 8i 2 I; t 2 T

eqn 4

xit 0; 8i 2 I; t 2 T

eqn 5

The objective function, eqn 1, maximizes the sum of the points from the biodiversity indicators in the selected areas (hereafter referred to as the biodiversity indicator score). Constraint set in eqn 2 is the budget constraint preventing the total cost of the selected areas to exceed the available budget (b). Constraint set in eqn 3 is the area constraint, preventing the total selected area from exceeding a certain proportion (q) of the total area. Constraint set in eqn 4 ensures that the area selected from each 50 · 50 km plot and age class is smaller than its existing area, and set in eqn 5 is the non-negativity constraints on the decision variables. It is diﬃcult to establish weights we that can be considered to be appropriate (Polasky, Camm & Garber-Yonts 2001). Typically, an indicator with large value will dominate and the solution tends to

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 133–142

136 J. Lundstro¨m et al. Table 1. List of biodiversity indicators and criteria for points Indicator

100 points

50 points

0 points

Uneven age1 Gaps2 Stand character3 Tree layer4 Ground structure5 Large pine Large spruce Large birch Large aspen Large deciduous tree (other than aspen or birch) Dead conifer tree lying Dead deciduous tree lying Dead conifer tree standing Dead deciduous tree standing Presence of rowan Aﬀected by water (moving water ⁄ spring ⁄ temporarily ﬂooded) Volume of dead wood

Not even-aged Several gaps Pristine Fully layered ⁄ several layers Very uneven ⁄ fairly uneven >40 cm dbh >40 cm dbh >40 cm dbh >40 cm dbh >40 cm dbh Tree > 20 cm dbh Tree > 20 cm dbh Tree > 20 cm dbh Tree > 20 cm dbh Present Yes >20 m3 ha)1

Fairly even-aged Some gaps

Completely even-aged No gaps Normal One layer ⁄ no layer Very even Not present Not present Not present Not present Not present Not present Not present Not present Not present Not present No

Two layers Fairly even >30 cm dbh >30 cm dbh >30 cm dbh >30 cm dbh >30 cm dbh

£ 20 m3 ha)1*

*Normalized point according to the volume of dead wood ha)1, from 0 to 100. 1 Totally even-aged: > 95% of the volume within an age interval of 5 years, fairly even-aged: > 80% of the volume within an age interval of 20 years; remaining stands classed as uneven aged. 2 Gap: an area without main crop seedlings ⁄ main trees larger than a square with a length of 2Æ5 times the average distance between main crop seedlings ⁄ main trees, but at least 5 m. Several gaps: at least 4 gaps within a 20 m radius from the centre of the plot, some gaps: 2–3 gaps; remaining stands are classed as no gaps. 3 Pristine character: presence of coarse (> 25 cm diameter) dead wood and no trace of management actions during the last 25 years. 4 Tree layer: group of trees amongst which the height is approximately the same, but their mean height diﬀers from other layers. Fully layered: all diameter classes represented, the biggest tree > 20 cm in diameter, the number of stems increasing with increasing diameter class, and the volume density (relationship between the actual volume in the stand and the potential volume) > 0Æ5. 5 Ground structure: Classiﬁcation based on height and frequency of irregularities (rocks, small hills and holes) on the ground.

Table 2. Parameters and decision variables for the model Notation

Description

Parameters I T E pite ait cit we q b

Set of 50 · 50 km plots (i = 1,...,n) Set of age classes (t = 1,...,m) Set of biodiversity indicators (e = 1,...,o) Point of biodiversity indicator e in plot i and age class t Area (ha) of plot i in age class t Cost ha)1 of plot i and age class t Weight of biodiversity indicator e Maximum proportion that can be selected Available budget (SEK)

Decision variables xit

Area (ha) selected in plot i and age class t

select areas with high values for one (or maybe a few) indicators. Goal programming is an approach that includes several objectives (expressed as goals) in the same objective function and still allows a trade-oﬀ that is considered impartial. In goal programming, we establish goals in phase 1. In our case, we simply solved the problem [P1] as many times as we had diﬀerent indicators. When we solved [P1] for indicator e, we set we to 1 for that indicator and 0 for all other indicator weights. This means that we independently searched for the best possible value for each indicator. We let those 17 values be denoted as ze. In the second phase, we wanted to ﬁnd a solution in which all indicators were as close as possible to their goal. Since it would not be

possible to reach the goal of each indicator, a quadratic deviation from these goals was minimized. The goal programming model in phase 2 can be formulated as !2 X XX ½P2 min w ¼ ðze pite xit Þ=ze eqn 1b e2E

i2I t2T

subject to eqns 2–5 In this objective function, eqn 1b, the squared diﬀerence between the goal and the actual biodiversity indicator score for each indicator is minimized. The diﬀerence is scaled with the goal value and hence we measure the deviation as a percentage deviation. Problem [P2] is a quadratic programming problem. The problem is convex (Lundgren, Ro¨nnqvist & Va¨rbrand 2010) so a global optimal solution is guaranteed. The models were formulated in the modelling language AMPL and solved using the software CPLEX 11Æ2 (ILOG 2006). All tests were conducted on a standard PC with 2Æ99 GHz and 3Æ25 GB of internal memory. The number of variables (in both models) was 9520 (112*5*17) and the number of constraints was 562 (1 + 1 + 112*5). The solution time for each problem was within a fraction of a second.

Results BIODIVERSITY INDICATORS

In the original NFI data used for analysis, the biodiversity indicators were distributed unevenly over the ﬁve age classes, but with all indicators represented in all age classes (Table 3). The

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 133–142

Cost-eﬀective boreal forest reserves 137 Table 3. Biodiversity indicator data from NFI (based on individual plots) with mean points ± standard deviation, as well as total area and total cost for the ﬁve age classes Age class 0–14 Biodiversity indicator Uneven age Gaps Stand character Tree layer Ground structure Large pine Large spruce Large birch Large aspen Large deciduous tree (other than aspen or birch) Dead conifer tree lying Dead deciduous tree lying Dead conifer tree standing Dead deciduous tree standing Presence of rowan Aﬀected by water (moving water ⁄ spring ⁄ temporarily ﬂooded) Volume of dead wood Total area (1000 ha) Total cost (billion SEK)

7Æ8 15Æ7 0Æ03 24Æ5 35Æ3 5Æ8 0Æ2 0Æ4 0Æ3 0Æ1 13Æ2 3Æ6 4Æ9 1Æ0 34Æ4 0Æ8

15–39

± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ±

18Æ8 31Æ6 1Æ8 30Æ5 40Æ4 19Æ5 2Æ9 5Æ4 4Æ6 2Æ3 33Æ8 18Æ5 21Æ6 9Æ7 47Æ5 9Æ0

23Æ2 ± 34Æ1 2346 21Æ6

28Æ4 21Æ3 0Æ13 42Æ5 36Æ6 1Æ4 0Æ6 0Æ2 0Æ2 0Æ2 7Æ3 1Æ5 1Æ1 0Æ3 32Æ0 1Æ1

± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ±

40–69

27Æ4 34Æ0 3Æ6 27Æ3 41Æ0 10Æ1 5Æ7 3Æ3 3Æ6 3Æ9 25Æ9 12Æ3 10Æ5 5Æ3 46Æ7 10Æ4

10Æ3 ± 23Æ1 3396 67Æ8

45Æ1 27Æ7 0Æ14 43Æ6 31Æ4 6Æ7 5Æ5 1Æ7 0Æ6 0Æ4 6Æ2 1Æ7 2Æ4 0Æ8 29Æ1 1Æ9

± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ±

‡ 100

70–99

28Æ0 37Æ0 3Æ8 29Æ8 39Æ4 19Æ8 18Æ0 9Æ9 6Æ1 5Æ1 24Æ1 12Æ8 15Æ4 8Æ9 45Æ4 13Æ7

17Æ1 ± 29Æ3 2975 87Æ8

63Æ3 24Æ8 0Æ61 48Æ1 29Æ1 15Æ5 13Æ3 1Æ9 0Æ7 0Æ3 10Æ5 2Æ2 7Æ5 2Æ4 21Æ7 1Æ6

± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ±

27Æ0 35Æ1 7Æ8 30Æ6 39Æ4 27Æ9 27Æ0 10Æ1 6Æ8 4Æ6 30Æ6 14Æ5 26Æ3 15Æ2 41Æ3 12Æ5

29Æ6 ± 37Æ5 2021 74Æ7

73Æ8 29Æ0 3Æ65 45Æ3 35Æ3 22Æ5 13Æ9 0Æ9 0Æ6 0Æ3 15Æ8 2Æ5 12Æ0 2Æ4 15Æ3 1Æ9

± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ±

27Æ2 36Æ7 18Æ8 30Æ8 41Æ9 31Æ1 27Æ8 7Æ2 6Æ4 4Æ6 36Æ5 15Æ7 32Æ5 15Æ3 36Æ0 13Æ6

37Æ1 ± 40Æ1 3550 141Æ6

The points for the indicator ‘Volume dead wood’ were given proportionally to the volume ha)1, with volumes > 20 m3 ha)1 given 100 points. The actual volumes per 1000 ha are shown.

relative magnitude of the mean points should not be interpreted as a sign of importance since the optimization models neutralized the advantage of common indicators. Instead diﬀerences between age classes are of interest. More than half of the 17 indicators peaked at ages > 100 years. Several of the large-tree indicators and all deadwood indicators had a higher mean point in the youngest forests (0–14 years) than in the subsequent age class. All deadwood indicators had the lowest values at intermediate age classes. ‘Uneven age’ increased over time, whereas ‘rowan’ decreased.

OPTIMAL AGE DISTRIBUTIONS

To investigate the question of whether the optimal combination of forest ages diﬀered when a budget constraint or an area constraint were used, two versions of the stated model were solved. In the ﬁrst version, budget was limiting (i.e. the area constraint, eqn 3, was omitted). This model was solved 100 times with an incremental increase in budget, starting at 1% of the total cost of all forest, up to 100%, with automatic registration of age distribution in each stage, resulting in 1800 ((17 + 1)*100) optimizations. In the second version, area was limiting and the budget constraint, eqn 2, was omitted. This model was also solved 100 times, with an incremental increase in area limit, starting at 1% of the total area, up to 100%, also with automatic registration of age distribution in each stage. The optimal age distributions when cost was limiting differed markedly from the optimal age distributions when area was limiting. With a budget-constrained approach, a large proportion of young forest was chosen at small budgets (Fig. 2a)

whereas forests in the 40- to 99-year age class were selected to a lower extent. However, when the selection was made with an area-constrained approach, the proportion of old forest was clearly more dominant (Fig. 2b). Forests in the 15- to 39-year age class were selected the least when costs were not considered, but younger forests were also selected in small proportions at low area limits. In general, the area-constrained approach covered less area but with higher biodiversity indicator scores, whereas the budget-constrained approach covered more area, but with lower biodiversity indicator scores (Fig. 3). The budget-constrained approach achieved a higher biodiversity indicator score compared to the area-constrained approach at any given cost (Fig. 3a).

GEOGRAPHICAL DISTRIBUTION

Two scenarios were used to analyse the diﬀerences in geographical distribution of selected forests between a budget-constrained model and an area-constrained model. In the ﬁrst scenario, a budget was set to 10 billion SEK (2Æ5% of the cost for the total area). The limit was chosen based on the current political targets in Sweden for nature reserve establishment, with 6 billion SEK allocated to forest protection for the years 1998–2008 (Swedish Government 2009). The area limit was set to 714 000 ha (5% of the total area) since this scenario gave approximately the same biodiversity indicator score. The 10 billion SEK budget scenario led to a reserve area of 1Æ2 million ha with a strong bias for selection of areas in the north-western section of boreal Sweden (Fig. 4a). The scenario with an area limit of 714 000 ha corresponded to a cost

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 133–142

138 J. Lundstro¨m et al. (a)

(a)

(b)

(b)

Fig. 3. The biodiversity indicator score plotted as a function of (a) cost and (b) area for the budget-constrained and area-constrained model.

Discussion Fig. 2. Optimal age distributions of forest reserves in boreal Sweden plotted as a function of (a) cost and (b) area. The age distributions toward the left in the graphs are most relevant for the actual situation in Sweden, with about 6 billion SEK allocated to forest protection during the last 10-year period (Swedish Government. 2009), and with an environmental target of protecting an additional 900 000 ha. When the limits increase and approach the total area or total cost (the right hand side of the graphs) the age distribution equals the original distribution in the data set.

of 41 billion SEK, and was strikingly diﬀerent with strong representation in the south-eastern part of the boreal region (Fig. 4b). As in the analysis on optimal age-distributions, the forests chosen in the budget-constrained scenario were mostly young, whereas the forests chosen in the area-constrained scenario were mostly old (Table 4). The biodiversity indicator score in the scenarios with a budget constraint and an area constraint were both based on contributions from all indicators (Table 5). Indicators on dead wood were more represented in the budget-constrained scenario whereas large trees were much more represented in the area-constrained scenario. The goal programming approach led to a biodiversity indicator score in which the contributions of all indicators were higher, in some cases substantially, than the mean of their contribution when each indicator was maximized separately (Table 5).

The results clearly show that it is more cost-eﬀective to protect young forests and more area-eﬀective to protect old forests, but that a combination of age classes always gives the highest biodiversity indicator score. This indicates that all age classes have a value to biodiversity, and that a reserve network ideally should consist of forests of diﬀerent ages regardless of whether the selection is limited by budget or by area. This is also challenging since it demonstrates that there is a need to reorient current boreal forest conservation strategies, which almost exclusively target the oldest forests. It was evident that it is more cost-eﬀective to use a budgetconstrained model compared to an area-constrained model when selecting reserves. Previous studies have shown that when land prices vary and area is used as a limitation, more money than necessary is spent, which is unfortunate since conservation is always restricted by scarce resources (Ando et al. 1998; Polasky, Camm & Garber-Yonts 2001). When using an area-constrained model the same biodiversity indicator score can be obtained in a smaller total area. Decision makers, therefore, have to integrate ecological and economic data and balance short- and long-term constraints in terms of cost and area in order to design cost-eﬀective conservation strategies (Polasky, Camm & Garber-Yonts 2001; Juutinen et al. 2004; Messer 2006; Naidoo et al. 2006).

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 133–142

Cost-eﬀective boreal forest reserves 139 (a)

(b)

Fig. 4. The proportion of the total area selected in each 50 · 50 km plot with (a) a budget limit of 10 billion SEK (9% of total area) or (b) an area limit of 715, 000 ha (5% of total area). The biodiversity indicator score in both scenarios was approximately the same (51 million in the budget-constrained scenario and 50 million in the areaconstrained scenario). For names of geographical regions see Fig. 1.

The analyses show that if costs are considered, large areas of young forests in north-western Sweden are selected. There are already numerous large nature reserves in this section of the boreal region, and more than 75% of all protected forests in the country are found here. However, the present reserves are overwhelmingly old, and by setting aside young, structurally rich forests nearby, dispersal and colonization of some species might be facilitated. One risk of concentrating reserves in the northwest region is that rare species conﬁned to more eastern regions would not be protected When costs are not considered, as in the area-constrained model, old forests are mostly selected and these are primarily located in the south-eastern section of the study area. One reason for this southern dominance could be a higher productivity, leading to higher representation of large trees, tentatively a subject to be further scrutinized in subsequent studies along with complementarity analyses between our model selections and existing reserves. The NFI data showed that a majority of the structure-based variables of importance to biodiversity are most common in the oldest age classes, but that there are substantial amounts also found in younger forests. For the youngest age class, 0–14 years, this mainly reﬂects the practice of tree retention, (leaving trees for biodiversity at clear cutting) introduced in Sweden and other countries a few decades ago. However, at least in terms of dead wood the same pattern is also found following natural disturbances such as forest ﬁres or storms, with plenty of dead wood present in the early successional stages, whilst amounts are lower at intermediate ages and then higher again in old-growth stages (Siitonen 2001). Much of the high conservation values of young forests reported here are thus likely to prevail both in young forests originating from management practices and in those originating from natural disturbances, although the latter would probably host higher amounts of dead wood and large living trees (Uotila et al. 2001). The biodiversity indicators chosen were to some extent biassed towards features most common in old-growth forests (e.g. uneven age, multi-layered canopy and presence of large trees). A more unbiassed list of indicators would also include those that are important for rare species that prefer young age classes, such as sun-exposed dead trees (Kaila, Martikainen & Punttila 1997; Jonsell, Weslien & Ehnstro¨m 1998). Including a larger proportion of such indicators would further strengthen the result that young forests host important biodiversity potential. There are some weaknesses in the structure-based approach. The same structures can be present in young and old forests, but support diﬀerent species compositions, mainly due to

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 133–142

140 J. Lundstro¨m et al. Table 4. Area distribution (proportion of selected area, %) in the six geographical regions and ﬁve age classes selected under a budget constraint of 10 billion SEK (9% of the total area) and an area constraint of 714 000 ha (5% of the total area)

15–39 years

40–69 years

70–99 years

‡ 100 years

All age classes

Budget constraint 10 billion SEK (area 1Æ2 million ha) Norrbotten 22Æ6 Va¨sterbotten 10Æ4 Ja¨mtland 14Æ8 Va¨sternorrland and Ga¨vleborg 14Æ6 Dalarna 6Æ3 Va¨rmland and O¨rebro 2Æ3 All regions 70Æ8

9Æ8 3Æ7 1Æ2 0 1Æ2 0 15Æ8

0 0Æ8 0Æ2 0 0Æ8 0 1Æ8

1Æ2 0 0Æ5 0 0 0 1Æ8

5Æ6 2Æ4 1Æ8 0 0 0 9Æ8

39Æ2 17Æ2 18Æ4 14Æ6 8Æ3 2Æ3 100

Area constraint 714 000 ha (cost 41 billion SEK) Norrbotten 0 Va¨sterbotten 0 Ja¨mtland 0 Va¨sternorrland and Ga¨vleborg 1Æ5 Dalarna 0 Va¨rmland and O¨rebro 0 All regions 1Æ5

0 0 0 0 0 0 0

0 0 0Æ3 0 0 16Æ7 17Æ0

0 0 0Æ7 15Æ8 3Æ2 13Æ4 33Æ0

0Æ6 4Æ2 10Æ6 9Æ0 7Æ8 16Æ3 48Æ5

0Æ6 4Æ2 11Æ6 26Æ2 11Æ0 46Æ4 100

0–14 years

Table 5. The contribution to the biodiversity indicator score of each indicator from both phases in the goal programming under a budget constraint and an area constraint Budget scenario

Area scenario

Biodiversity indicator

Max1

Mean2

Min3

% of mean4

Max

Mean

Min

% of mean

Uneven age Gaps Stand character Tree layer Ground structure Large pine Large spruce Large birch Large aspen Large deciduous tree Dead conifer tree lying Dead deciduous tree lying Dead conifer tree standing Dead deciduous standing Presence of rowan Aﬀected by water Volume of dead wood5

56986 48442 7306 59801 69991 14910 9682 2944 1953 1574 32103 11012 12124 5090 62552 5789 66677

26488 24171 1265 32115 32317 6005 2258 542 313 234 14262 3907 4746 1150 27191 1106 33714

15502 8801 0 14258 9415 2559 139 149 36 49 4814 1199 2351 199 7429 301 13509

145 143 131 182 211 109 138 138 176 166 149 123 140 127 261 125 150

79266 43714 9832 58258 61775 38861 30494 6176 3780 2498 24202 10475 19116 7990 57944 8756 70850

52006 24332 2122 39889 35602 16751 13372 1873 860 510 11366 3240 8924 2429 24975 1934 39003

25247 18587 100 28864 21916 8570 4019 398 150 149 4995 1804 4108 700 10306 700 18945

112 110 133 164 201 120 115 106 224 202 179 141 173 103 179 111 127

1 The biodiversity indicator score when maximizing each indicator separately (the goal) gives a maximal sum of points that each indicator can obtain (when the optimization is made only considering that speciﬁc indicator). 2 A mean sum of points from all 17 goal optimizations. 3 The lowest sum of points that the indicator gets from another indicator’s goal optimization. 4 The minimization of all indicator’s quadratic percentage deviation from their goal (phase 2 in the goal programming) gives a contribution from each indicator to the biodiversity indicator score shown here as the percentage of its mean value. 5 Shown in units of 1000 points.

diﬀerences in microclimatic conditions and colonization opportunities. Further, the presence of structures is no guarantee of the presence of associated species since other aspects, such as forest history and connectivity might be decisive for species occurrences (Nilsson, Hedin & Niklasson 2001). Therefore, a further development of our study would be to repeat the analyses for species distribution data and compare the results with those based on structural data. There are detailed data on

occurrences of red-listed species from organism groups such as vascular plants, birds, bryophytes, lichens and fungi in boreal Sweden which could be used for such a comparison. The proposed goal programming approach provides impartial and objective weighting. We used the same weighting for analysis of the deviation from the goals, but this can be easily modiﬁed if desired. If weights are decided manually, care must be taken since the results can greatly disadvantage some

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 133–142

Cost-eﬀective boreal forest reserves 141 variables (Table 5, ‘Min’ column). We note that the importance of diﬀerent indicators are indirectly weighted when deciding criteria for points, but those decisions are based on existing knowledge of which features are important for biodiversity. A further development of the model could potentially be to add speciﬁc requirements regarding geographical distribution or a minimum amount of diﬀerent indicators. This, however, needs to be substantiated by high-quality ecological studies on the critical requirements of diﬀerent species. In northern Europe, with its long history of relatively intense forest use, there are so few old-growth forests left that areas strongly impacted by humans need to be included when new forests are protected. Consequently, it is vital to prioritize the few high-quality old-growth remnants that still exist, although our analyses indicate that it is more cost-eﬀective to include young forests in reserve networks. The young forests that we propose for protection have a decidedly diﬀerent character than those normally regenerating after clear-cutting, even with tree retention. A careful selection will be needed for sites especially rich in dead wood, remnant live trees and other qualities of importance to biodiversity. Protection of young forest allows much more land to be set aside than protection of oldgrowth forest, due to lower net present values. A shift towards more protection of young forest might therefore eventually cause a reduction in timber volumes available for forest industry. A novel conservation strategy, and a future research challenge, is to analyse if some reserves with old forests could be systematically replaced by younger forests without causing biodiversity decline at the landscape level. Possibly, such a dynamic reserve scheme could beneﬁt both timber production and biodiversity protection. In general, early-successional stages are overlooked in forest conservation. Thus, our approach with protection of diﬀerent age classes has a general interest also for other biomes where biodiversity is adapted to frequent disturbances and where early successional stages are common in natural forest landscapes.

Acknowledgements We thank Per Nilsson for help with the NFI-data and Peder Wikstro¨m and Torgny Lind for assistance during the Heureka calculations. We also thank Viktor Johansson and three anonymous reviewers for constructive comments on the manuscript. The study was ﬁnancially supported by FORMAS.

References Ahti, T., Ha¨met-Ahti, L. & Jalas, J. (1968) Vegetation zones and their sections in northwestern Europe. Annales Botanici Fennici, 5, 169–211. Ando, A., Camm, J., Polasky, S. & Solow, A. (1998) Species distributions, land values, and eﬃcient conservation. Science, 279, 2126–2128. Angelstam, P. (1998) Maintaining and restoring biodiversity in European boreal forests by developing natural disturbance regimes. Journal of Vegetation Science, 9, 593–602. Anonymous. (1932) Uppskattning av Sveriges skogstillga˚ngar. Jordbruksdepartementet, Stockholm, Sweden (in Swedish). Anonymous. (2005) RIS Riksinventeringen av skog. Fa¨ltinstruktion 2005. Institutionen fo¨r skoglig resurshusha˚llning och geomatik SLU, Umea˚, Sweden (in Swedish). Anonymous. (2007a) Skogsstatistisk a˚rsbok 2007. Sveriges oﬃciella statistik, Skogsstyrelsen, Jo¨nko¨ping, Sweden (in Swedish).

Anonymous. (2007b) Skogsdata 2007. Sveriges oﬃciella statistik, Institutionen fo¨r skoglig resurshusha˚llning, SLU, Umea˚ (in Swedish). Esseen, P.A., Ehnstro¨m, B., Ericson, L. & Sjo¨berg, K. (1997) Boreal forests. Ecological Bulletins, 46, 16–47. FAO (Food and Agriculture Organization of the United Nations). (2006) Global Forest Resources Assessment 2005: Progress towards sustainable forest management. FAO, Rome, Italy. Ferris, R. & Humphrey, J.W. (1999) A review of potential biodiversity indicators for application in British forests. Forestry, 72, 313–328. Ga¨rdenfors, U. (Ed.). (2005) The 2005 Red List of Swedish Species. Swedish Species Information Centre, Swedish University of Agricultural Sciences, Uppsala, Sweden. Gustafsson, L. & Ahle´n, I. (1996) The National Atlas of Sweden. Geography of Plants and Animals. SNA, Stockholm, Sweden. Hagar, J.C. (2007) Wildlife species associated with non-coniferous vegetation in Paciﬁc Northwest conifer forests: a review. Forest Ecology and Management, 246, 108–122. Ibbotson, R.G. & Sinqueﬁeld, R.A. (1986) Stocks, Bonds, Bills and Inﬂation: 1986 Yearbook. Ibbotson Associates, Chicago. ILOG (2008) CPLEX system 11.2. Gentilly, France. Jonsell, M., Weslien, J. & Ehnstro¨m, B. (1998) Substrate requirements of redlisted saproxylic invertebrates in Sweden. Biodiversity and Conservation, 7, 749–764. Junninen, K., Simila, M., Kouki, J. & Kotiranta, H. (2006) Assemblages of wood-inhabiting fungi along the gradients of succession and naturalness in boreal pine-dominated forests in Fennoscandia. Ecography, 29, 75–83. Juutinen, A., Mantymaa, E., Mo¨nkko¨nen, M. & Salmi, J. (2004) A cost-eﬃcient approach to selecting forest stands for conserving species: a case study from northern Fennoscandia. Forest Science, 50, 527–539. Kaila, L., Martikainen, P. & Punttila, P. (1997) Dead trees left in clear-cuts beneﬁt saproxylic Coleoptera adapted to natural disturbances in boreal forest. Biodiversity and Conservation, 6, 1–18. Kouki, J., Lo¨fman, S., Martikainen, P., Rouvinen, S. & Uotila, A. (2001) Forest fragmentation in Fennoscandia: linking habitat requirements of woodassociated threatened species to landscape and habitat changes. Scandinavian Journal of Forest Research, 16, 27–37. Kuuluvainen, T. (2002) Natural variability of forests as a reference for restoring and managing biological diversity in boreal Fennoscandia. Silva Fennica, 36, 97–125. Kuuluvainen, T. (2009) Forest management and biodiversity conservation based on natural ecosystem dynamics in Northern Europe: the complexity challenge. AMBIO: A Journal of the Human Environment, 38, 309–315. La¨ma˚s, T., Dahlin, B., Olsson, H., Sallna¨s, O., Stenlid, J. & Sta˚hl, G. (2006) Preface. Scandinavian Journal of Forest Research, 21, 3–4. Lindenmayer, D.B. & Franklin, J.F. (2003) Towards Forest Sustainability. CSIRO Publishing, Melbourne, Australia. Lindenmayer, D.B., Margules, C.R. & Botkin, D.B. (2000) Indicators of biodiversity for ecologically sustainable forest management. Conservation Biology, 14, 941–950. Lo¨fman, S. & Kouki, J. (2001) Fifty years of landscape transformation in managed forests of Southern Finland. Scandinavian Journal of Forest Research, 16, 44–53. Lundgren, J., Ro¨nnqvist, M. & Va¨rbrand, P. (2010) Optimization. Studentlitteratur, Lund, Sweden. Margules, C.R. & Pressey, R.L. (2000) Systematic conservation planning. Nature, 405, 243–253. Messer, K.D. (2006) The conservation beneﬁts of cost-eﬀective land acquisition: a case study in Maryland. Journal of Environmental Management, 79, 305–315. Naidoo, R., Balmford, A., Ferraro, P.J., Polasky, S., Ricketts, T.H. & Rouget, M. (2006) Integrating economic costs into conservation planning. Trends in Ecology & Evolution, 21, 681–687. Nilsson, S.G., Hedin, J. & Niklasson, M. (2001) Biodiversity and its assessment in boreal and nemoral forests. Scandinavian Journal of Forest Research, 16, 10–26. O¨stlund, L., Zackrisson, O. & Axelsson, A.L. (1997) The history and transformation of a Scandinavian boreal forest landscape since the 19th century. Canadian Journal of Forest Research, 27, 1198–1206. Polasky, S., Camm, J.D. & Garber-Yonts, B. (2001) Selecting biological reserves cost-eﬀectively: an application to terrestrial vertebrate conservation in Oregon. Land Economics, 77, 68–78. Ranneby, B., Cruse, T., Ha¨gglund, B., Jonasson, H. & Swa¨rd, J. (1987) Designing a New National Forest Survey for Sweden. Swedish University of Agricultural Sciences, Uppsala, Sweden.

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 133–142

142 J. Lundstro¨m et al. Siitonen, J. (2001) Forest management, coarse woody debris and saproxylic organisms: Fennoscandian boreal forests as an example. Ecological Bulletins, 49, 11–41. Spanos, K. & Feest, A. (2007) A review of the assessment of biodiversity in forest ecosystems. Management of Environmental Quality: An International Journal, 18, 475–486. Spies, T.A. & Franklin, J.F. (1991) The Structure of Natural Young, Mature, and Old-growth Douglas-ﬁr Forests in Oregon and Washington. U.S. Department of Agriculture, Forest Service, Paciﬁc Northwest Research Station, Portland, OR, USA. Swanson, M.E., Franklin, J.F., Beschta, R.L., Crisafulli, C.M., DellaSala, D.A., Hutto, R.L., Lindenmayer, D.B. & Swanson, F.J. (2010) The forgotten stage of forest succession: early-successional ecosystems on forest sites. Frontiers in Ecology and the Environment, in press. doi: 10.1890/090157. Swedish Government. (2005) Svenska miljo¨ma˚l – ett gemensamt uppdrag. Proposition 2004 ⁄ 2005:150. Swedish Government, Stockholm, Sweden (in Swedish).

Swedish Government. (2009) Ha˚llbart skydd av naturomra˚den. Proposition 2008 ⁄ 09:214. Swedish Government, Stockholm, Sweden (in Swedish). Tikkanen, O., Martikainen, P., Hyvarinen, E., Junninen, K. & Kouki, J. (2006) Red-listed boreal forest species of Finland: associations with forest structure, tree species, and decaying wood. Annales Zoologici Fennici, 43, 373–383. Uotila, A., Maltamo, M., Uuttera, J. & Isoma¨ki, A. (2001) Stand structure in semi-natural and managed forests in eastern Finland and Russian Karelia. Ecological Bulletins, 49, 149–158. Zackrisson, O. (1977) Inﬂuence of forest ﬁres on the North Swedish boreal forest. Oikos, 29, 22–32. Received 9 March 2010; accepted 20 October 2010 Handling Editor: Harald Bugman

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 133–142

Journal of Applied Ecology 2011, 48, 265–273

doi: 10.1111/j.1365-2664.2010.01898.x

Variation partitioning in canonical ordination reveals no effect of soil but an effect of co-occurring species on translocation success in Iris atrofusca Sergei Volis1*, Michael Dorman1, Michael Blecher2, Yuval Sapir3 and Lev Burdeniy1 1

Life Sciences Department, Ben-Gurion University of the Negev, Beer Sheva, 84105 Israel; 2Ein Gedi Nature Reserve, Israel Nature and Parks Authority, Dead Sea, 86980 Israel; and 3The Botanical Garden, Department of Plant Sciences, Tel Aviv University, Tel Aviv, 69978 Israel

Summary 1. Despite being expensive, complicated and less successful than the conservation of primary habitat, translocation is rapidly gaining importance as a conservation approach due to accelerated loss of natural environment. Finding the optimal abiotic and biotic conditions needed for successful translocation of plants can be diﬃcult for species with limited information on prior distribution. Unfortunately, this is often the case with endangered plant species, including those urgently needing action. 2. We present a method of evaluating the relative importance of multiple environmental parameters in translocation success. This method is based on the application of variation partitioning in canonical ordination and it allows usage of not only multiple independent biotic and abiotic variables, but also multiple dependent variables for ﬁtness estimates. 3. In this study, six soil parameters together with the abundance of 61 plant species and their total biomass were used to explain the variation in translocation success of Iris atrofusca plants among 22 microsites. The relative importance of each of the three factors was estimated using ordination techniques. 4. Soil characteristics and total biomass of other plants did not signiﬁcantly aﬀect the performance of translocated irises, but the species composition of the surrounding vegetation did have a signiﬁcant eﬀect. The abundance of relatively rare species was closely correlated with iris performance. It is likely that these species do not aﬀect the irises directly but instead represent environmental conditions not measured in this study, which are necessary for the survival of irises. 5. Synthesis and applications. Variation partitioning appears to be a highly promising method for planning the translocation of plants and evaluating success due to its ability to estimate the unique contribution of each of two or more sets of environmental factors. It can be used to monitor success, and to identify the key contributory factors, in experimental translocations preceding actual introduction of plants in conservation programmes. Key-words: canonical correspondence analysis, endangered plant species, habitat-suitability, niche space, redundancy analysis, soil characteristics, species abundance, variation partitioning

Introduction Translocation of plants refers to their accidental or deliberate movement, within or beyond their natural range, by humans. The goal of translocations for conservation purposes is to establish new populations of rare and endangered species in order to increase the survival of the species as a whole (Hey-

*Correspondence author. E-mail: [email protected]

wood & Iriondo 2003). This practice may become more prevalent as human impact on the natural environment increases, even though it is more expensive, complicated and less successful than the conservation of primary habitat (e.g. Gordon 1996; Milton et al. 1999). Currently, the evidence for the long-term success of translocations is limited (Maunder 1992; Seddon, Armstrong & Maloney 2007) and the reasons for success or failure can be diﬃcult to determine (e.g. Morgan 1999). In the past, many translocation projects were performed without scientiﬁc rigour

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society

266 S. Volis et al. or hypothesis testing. Today, it is recognized that the factors determining success or failure of a translocation have to be studied using a scientiﬁcally based approach combining ecological theory and empirical tests (Sarrazin & Barbault 1996; Seddon, Armstrong & Maloney 2007; Menges 2008). When plants are moved to a site outside the known native range (termed ‘introduction’; IUCN 1987) one of the most important decisions is selecting the relocation site, and then the best microsites within the relocation site (e.g. Adamec & Lev 1999; Jusaitis 2005; Maschinski & Duquesnel 2006; Colas et al. 2008). By deﬁnition, the location where a population can be established has to be within the ecological niche of the species, i.e. the set of abiotic and biotic conditions under which the species can maintain populations without immigration (Grinnell 1917). However, determination of the species niche is not an easy task because the actual distribution of the species can be limited due to interspeciﬁc interactions (the diﬀerence between fundamental and realized niches, MacArthur 1972), as well as limited colonization ability and local extinction (Burkey 1995; Peterson, Sobero´n & Sa´nchez-Cordero 1999). The method currently used for identifying the most suitable habitats for species is to search for a quantitative relationship between ecological and environmental features in the landscape and either (i) species occurrence in space prior to relocation or (ii) population establishment after relocation. Much progress has been made in the last two decades in the development of techniques predicting species distribution and then estimating potential site suitability for establishment (reviewed in Guisan & Zimmermann 2000; Stauﬀer 2002; Guisan & Thuiller 2005; Richards, Carstens & Lacey Knowlrs 2007; Elith & Leathwick 2009). These techniques generally involve the use of spatially explicit data through geographic information systems (GIS) and modelling a species’ ecological niche. Particular examples include BIOCLIM (Busby 1991), HABITAT (Walker & Cocks 1991), DOMAIN (Carpenter, Gillison & Winter 1993), Genetic Algorithm for Rule-set Prediction (GARP) (Stockwell & Peters 1999) and Ecological-Niche Factor Analysis (ENFA) (Hirzel et al. 2002; Engler, Guisan & Rechsteiner 2004; Basille et al. 2008). These methods require detailed maps of species presence and environmental parameters, which limit the application of species distribution modelling to (i) large geographic scale of kilometres and (ii) species with documented and relatively wide distributions. Species that occupy only a limited number of locations including a patchy distribution at the ﬁne geographic scale (tens or hundreds of metres) or where there is little information on prior distribution, require another approach. In these cases, an alternative approach is to analyse the success of an experimentally translocated population. As the environmental variables aﬀecting performance of relocated organisms can be complex, they need to be identiﬁed and diﬀerentiated using multivariate statistical methods. One approach is to use stepwise multiple regression to develop a predictive equation for success of translocation in which independent variables will be ranked by their importance (Griﬃth et al. 1989). However, this approach requires a single dependent variable for translocation success, which can be binary

(success vs. failure), ordinal (varying degrees of success) or continuous (e.g. population growth rate, percentage of survived plants or populations that became self-sustained). A requirement for a single variable summarizing the existing information on relocated plants ⁄ populations limits the application of this approach because many estimates of ﬁtness are stage speciﬁc and vary in time. A potentially more eﬃcient approach is one using not only multiple independent variables representing diﬀerent environmental biotic and abiotic eﬀects, but also multiple dependent variables for ﬁtness estimates. Canonical correspondence analysis (CCA) and redundancy analysis (RDA) are constrained multivariate ordination techniques widely used in ecology and vegetation science to extract the major gradients in response (dependent, usually biotic) variables attributed to the explanatory (independent, usually environmental) variables (e.g. Pivello, Shida & Meirelles 1999; Clarke, Latz & Albrecht 2005; Svenning & Skov 2005). A particular strength of CCA and RDA is their ability to remove the eﬀect of undesirable variables (covariables) by regressionbased covariance analysis prior to the analysis itself. This procedure is called partial canonical correspondence analysis (pCCA) or partial redundancy analysis (pRDA; ter Braak 1986). It is possible to measure the fraction of the variation in the dependent variables explained by each set of environmental variables alone as well as the fraction of the variation shared by the sets of variables by using CCA and pCCA (or RDA and pRDA) (Borcard, Legendre & Drapeau 1992; Noe & Zedler 2001; Volis et al. 2004). High analytical power and the ability to eﬃciently reduce large data sets of both independent and dependent variables into only a few canonical axes make CCA and RDA attractive for ﬁnding a range of suitable environmental conditions for successful relocation. Speciﬁcally, CCA ⁄ RDA can help to ﬁnd which environmental factors have the highest eﬀect on individual performance and also the unique contribution of each of two or more sets of environmental factors to explaining the variation in individual performance. For example, species composition may signiﬁcantly aﬀect the performance of introduced plants (Elmendorf & Moore 2007). However, a hypothetical situation is possible when the vegetation eﬀect may be indirect and reﬂecting other eﬀects, such as diﬀerences in soil properties. Thus, both vegetation and soil will appear important for successful relocation, while in fact only the soil is important. Iris atrofusca Baker is a highly endangered species in Israel (Shmida & Pollak 2007), with habitats in the Northern Negev being the most vulnerable throughout its distribution. Rapid destruction of the natural habitat of I. atrofusca due to land clearing and a lack of nature reserves containing populations of I. atrofusca in the Negev leave very limited conservation options for this species. Thus there is no alternative to translocation, i.e. introduction of the species into seemingly suitable protected areas with no record of prior occupancy. The habitat characteristics needed for success are not known for this species or similar endangered irises in Israel. For example, translocation of Iris hermona Dinsmore in the Golan Heights was unsuccessful, even though translocation was into a very

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 265–273

Variation partitioning in translocation 267 similar habitat to the one that had been destroyed (Y. Sapir, unpublished data). In order to investigate species habitat preferences for I. atrofusca, we set up a translocation experiment, using rhizomes rescued from a site under threat of destruction. We applied CCA ⁄ RDA for (i) partitioning the variation in performance of I. atrofusca into components due to vegetation composition, total plant biomass and soil properties and their shared eﬀects and (ii) for determining the environmental factor(s) directly aﬀecting species introduction success.

70

Produced fruits

60

Produced flowers

50

Non-reproducing

40 30 20 10 0 70

Materials and methods The study area was Lahav North Nature Reserve which is 1Æ15 km in size and located in the semi-arid climatic zone of Israel (ca. 300 mm annual rainfall; Shachak et al. 2008). The reserve is typiﬁed by low hills less than 500 m above sea level, with plant formation typical for the transitional zone between Mediterranean and desert vegetation (known as batha), dominated by Sarcopoterium spinosum (L.) Spach, Phlomis brachyodon (Boiss.) Zohary, Asphodelus ramosus L. and Gundelia tournefortii L. (Tsoar & Ramon 2002; Fig. 1). There are no records of I. atrofusca ever occupying this site. However, it is within the discontinuous distribution range of I. atrofusca, with the nearest population found 9 km south-west of the Reserve at the Dudaim forest. Rhizomes rescued in spring 2006 in the nearby Goral Hills region (road-building strip for new railroad tracks) were planted in autumn 2006 in sets of 62 rhizomes in each of 22 microhabitats in Lahav North Nature Reserve (Fig. 1). Each set comprised the following size classes (number in parentheses): <5 g (14), 5–10 g (10), 10–20 g (23), 20– 30 g (10), 30–40 g (3) and >40 g (2). In spring 2007, 2008 and 2009 we counted the number of plants that emerged, ﬂowered and set fruit at each site, making a total of nine performance variables (Fig. 2). In addition, in the second year (spring 2008) the plant community at each site was sampled using random quadrats of either 1 m2 (one per site) or 0Æ125 m2 (two per site). Quadrat size was chosen according to the vegetation density and homogeneity of the site. All plants from a quadrat were harvested, brought to the laboratory, identiﬁed to species level, counted and dried to constant weight to determine the total plant biomass at each site. The vegetation data set had, therefore, two parts: the number of individuals per species (Table S1, Supporting information) and total plant biomass (Table S2, Supporting

Number of plants

60 2

50 40 30 20 10 0 70 60 50 40 30 20 10 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22

Site Fig. 2. Survival and reproduction of I. atrofusca (number of plants) 1, 2 and 3 years after experimental introduction at Lahav North Reserve in the 22 sites (from top to bottom). 62 rhizomes of Goral origin with equal representation of diﬀerent size classes were introduced at each site in fall of 2006, counting was done in spring of 2007, 2008 and 2009. information) per 1 m2 per site (when 0Æ125 m2 quadrats were used, the data were standardized per 1 m2). In total, 75 plant species were found across the sites. During the same year (spring 2008), soil samples were taken at each site and analysed for six soil characteristics (Tables 1 and S3, Supporting information). Therefore, we had four multivariate data sets: species abundance (75 variables), biomass (one variable), soil (six variables) and iris performance (nine variables) data for each of 22 sites.

DATA ANALYSIS

Fig. 1. Example of ﬁve relocation sites within the Lahav North Nature Reserve.

The eﬀects of the three sets of environmental variables (species abundance, biomass and soil) on variation in plant performance were examined with ordination techniques, using canoco (ver. 4.02; ter Braak & Smilauer 2002). Since a relationship between each of the three factors and performance of relocated iris plants may be caused by partial redundancy with the other factors, we applied the method of variation partitioning (Borcard, Legendre & Drapeau 1992). We estimated the following components of variation: ‘pure’ eﬀects of abundance, biomass and soil (i.e. variation that can be explained by

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 265–273

268 S. Volis et al. Table 1. Qualitative soil characteristics, their description and categorization Soil characteristics

Description

Categories

Depth

Depth of soil proﬁle to parent or subjacent material

<20 cm; 20–40 cm; 40–60 cm; >60 cm

Humus

Humus of root-inhabited horizon, according to Munsell Soil Colour Charts

Low (10 YR 4 ⁄ 4, 5 ⁄ 3, 5 ⁄ 4); intermediate (10 YR 4 ⁄ 3, 5 ⁄ 2); high (10YR 3 ⁄ 3, 4 ⁄ 2; 7Æ5YR 4 ⁄ 2)

Presence of Bk horizon

The subsoil layer (horizon B) with large accumulation of carbonates (k)

Absent; low (Bk fragmentary, weak or unstable sub-angular structure); intermediate (Bk visible, moderate sub-angular structure); high (Bk clear, sub-angular to cubic or angular structure)

Alluvium

Depth of crumb or granular structure

Absent; low (<40 cm); high (<60 cm)

Stoniness

Rock fragments by volume (%)

Low (<15%); intermediate (15–30%); high (30–60%)

Parent material

Type of parent material

El – eluvium of lime rocks or limestone; L – mainly loessial sediments; D – diluvium, talus sediments

each of the three factors and is not shared by the other factors); variation that is shared by two factors, either ‘abundance & biomass’, ‘abundance & soil’ or ‘biomass & soil’; variation that is shared by three factors, ‘abundance & biomass & soil’; and variation explained by no factor, which is the unexplained variation in iris performance. Performance data was represented by nine variables: number of plants that survived, ﬂowered and produced fruit in 2007, 2008 and 2009 (Fig. 2). No plants set fruit in 2008 because all ﬂowers were consumed by insects that year. To justify usage of more than one performance variable, Pearson product–moment correlation coeﬃcients, and their signiﬁcance after sequential Bonferroni adjustment, were used to assess the correlation among those variables. Soil data comprised a single matrix with six independent variables, including one categorical (parent material) and ﬁve ordinal variables (Table S3, Supporting information). Plant abundance data included abundance per 1 m2 of 75 plant species (Table S1, Supporting information), belonging to the following life forms: annual forbs (39), annual grasses (14), perennial forbs (9), geophytes (8), perennial grasses (3) and annual ⁄ perennial forbs (not identiﬁed to the species level) (2). However, prior to the analysis 14 species which occurred only once in the vegetation dataset were excluded (Boeken & Shachak 1994), leaving 61 variables of species abundance for the analysis. Plant abundance data was log-transformed and biomass data was represented as the total plant biomass per 1 m2 at each site (Table S2, Supporting information). In order to reduce the number of independent variables, both species abundance and soil data sets were reduced using either Principal Components Analysis (PCA) or Correspondence Analysis (CA), and the values of the ﬁrst several axes were used as the derived variables (Fig. 3). The choice between PCA and CA was made according to ter Braak & Smilauer (2002), assuming either a unimodal model (CA) or linear model (PCA) for the relationships of independent variables with the ordination axes. The number of derived axes for the sites scores, to be used in the variation partitioning analyses, was chosen to represent as much variation as possible in the original data, but not to exceed the number of dependent variables (which is equal to nine).

Variation partitioning Technically, variation partitioning is done by CCA, pCCA, and CA (Borcard, Legendre & Drapeau 1992) or by RDA, pRDA, and PCA. The RDA ⁄ pRDA ⁄ PCA was chosen in our case, since we assumed a linear, and not a unimodal, relationship between the environmental gradients and iris performance. The PCA estimated the total amount

Independent variables

Dependent variables

Species data Biomass data Soil data (60*22) (1*22) (6*22) (DCA) (DCA) CA PCA Species data CA scores (5*22)

Iris data (9*22)

Soil data PCA scores (3*22)

1 PCA 6 RDA + permutation tests 6 pRDA

Fig. 3. Conceptual diagram of the multivariate statistical procedure used to estimate the proportions of variation in performance of translocated iris due to environmental variation in vegetation composition (abundance), total plant biomass and soil characters. DCA was conducted to determine the gradient length in species and soil data, in order to choose between PCA and CA for these variables. Then the three independent sets of variables were used in 12 combinations (see Table 2), as constraining variables and ⁄ or covariables in RDA, while the dependent variable was always the iris performance data. A permutation test was done in each analysis to assess the signiﬁcance of the eﬀects.

of variation (sum of all eigenvalues) in the dependent variables. Several RDA and pRDA were conducted (Fig. 3), with diﬀerent combinations of one, two or three variables as the constraining variables and ⁄ or covariables. The sum of all canonical eigenvalues in these analyses was used to partition the variation into components. When variables from diﬀerent data sets were used together either as constraining variables or covariables, both datasets were simply combined. For example, the dataset ‘abundance + biomass’ (A + B; see Table 2) is a dataset of six variables (5 + 1). Calculation of variation components is described below.

Signiﬁcance of effects Statistical signiﬁcance of each environmental eﬀect was determined by a Monte Carlo permutation test with 999

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 265–273

Variation partitioning in translocation 269 Table 2. Pearson product–moment correlations among nine performance traits T 2007

T 2008

T 2009

FL 2007

FL 2008

FL 2009

FR 2007

T 2008 0Æ10 T 2009 0Æ09 0Æ65* FL 2007 0Æ59 0Æ05 )0Æ34 FL 2008 )0Æ37 0Æ26 0Æ25 )0Æ33 FL 2009 )0Æ21 0Æ55 0Æ80* )0Æ45 0Æ65* FR 2007 0Æ70* )0Æ14 )0Æ31 0Æ85* )0Æ38 )0Æ44 FR 2009 )0Æ12 0Æ61 0Æ80* )0Æ37 0Æ56 0Æ95* )0Æ32 T – total; FL – ﬂowered; FR – set fruits. *Signiﬁcance (P < 0Æ05) after sequential Bonferroni adjustment.

permutations in the relevant ordination analysis. The environmental variables explaining most of the variation in dependent variables were selected using the forward selection procedure in canoco, with a cut oﬀ point of P = 0Æ10 (default).

Results PERFORMANCE DATA

There were no strong and consistent correlations among the three performance traits in the same year and across years (Table 3). Only seven out of 28 trait correlations were signiﬁcant.

VEGETATION AND SOIL DATA

Species abundance and soil data sets were reduced using CA and PCA, respectively (Fig. 3). Based on the gradient length in DCA, PCA was chosen for the soil data and CA for the (logtransformed) species data (1Æ6 and 4Æ4 SD, respectively). For plant abundance data, the ﬁrst ﬁve CA axes were used, summarizing 61Æ2% of the original data matrix variation. For soil data, the eigenvalues of the ﬁrst three principal components

were 0Æ54, 0Æ17 and 0Æ16. The eigenvalues of the remaining axes did not exceed 0Æ06 and therefore only the ﬁrst three PCs were retained, summarizing 86Æ8% of the variation. Overall, after data reduction through CA and PCA, we had four data sets: abundance (ﬁve variables), biomass (one variable), soil (three variables) and iris performance (nine variables; Fig. 2). The derived soil variables (principal components) can be interpreted based on the contribution of original soil variables (Table 2). The ﬁrst principal component (PC1) is positively associated with soil permeability and water accumulation as indicated by its positive correlation with soil depth and diluvium, and negative correlation with eluvium. Thus the sites having high scores for PC1 are those allowing accumulation of water runoﬀ, such as terrace slopes and valleys. The second principal component (PC2) is negatively correlated with soil depth indicating conditions of high runoﬀ and shallow soil. The third principal component (PC3) is positively correlated with loess soil and high humus content indicating high soil fertility. Among the 22 relocation sites of the Lahav Nature Reserve, ﬁve grasses out of 61 species were the most abundant (per 1 m2): Stipa capensis Thunb., Carex pachystylis J. Gay, Brachypodium distachyum (L.) P. Beauv., Avena sterilis L. and Hordeum spontaneum K. Koch., comprising 53, 23, 12, 5 and 2%, respectively, of the total number of plants in the 22 sites. These ﬁve species together comprised 94% of plant abundance, while the abundance of each of the remaining 56 species was less than 1%. S. capensis was the most abundant species (53%), with over 15 000 plants, and an average density of 997 plants m)2 where it was recorded. However, the most common species across sites was A. sterilis, which was found at 17 of the 22 sites. It is followed by S. capensis and H. spontaneum, found in 16 and 11 sites, respectively; while the remaining 58 species were found in 8 sites or fewer. The other two abundant species mentioned above, C. pachystylis and B. distachyum, were found at just 3 and 6 sites, respectively.

VARIATION PARTITIONING

Table 3. Contribution of soil characteristics to the ﬁrst three axes derived by PCA on soil data (total variation explained: 86Æ8%) Principal components Soil characteristics

1

2

3

Depth of soil Humus Presence of Bk horizon Alluvium Stoniness EL D L Variation explained (%)

0Æ73 )0Æ03 0Æ82

)0Æ61 0Æ07 0Æ54

0Æ18 0Æ85 0Æ12

0Æ64 0Æ85 )0Æ89 0Æ83 0Æ14 54Æ1

0Æ02 )0Æ27 0Æ07 )0Æ13 0Æ14 16Æ8

)0Æ66 0Æ02 )0Æ18 0Æ01 0Æ43 15Æ9

El – eluvium of lime rocks or lime stones; L – mainly loessial sediments; D – diluvium, talus sediments.

We used 12 ordination analyses (Table 4, lines 2–13) to calculate the explained variation (%). Variation in the iris data alone, in the unconstrained analysis (PCA), is scaled to equal unity in canoco (line 1); therefore the sum of eigenvalues in the constrained analyses (RDA and pRDA) is a measure of variation in the response data (performance of relocated plants) accounted for by the constraining data (lines 2–13). From the sums of eigenvalues in the 12 ordination analyses (Table 4) we calculated the proportions of variation in iris performance data explained by each respective factor, shown in a Venn diagram (Fig. 4). Each rectangle area represents the proportion of variation explained by species abundance, biomass and soil data (a, b and s, respectively). The area of overlap between rectangles represents the two- or three-factor shared eﬀects (ab, as, bs and abs). Species abundance explained the largest proportion of the iris performance data (47Æ4%), and after removal of the component shared with biomass and soil it still explained 40Æ0%.

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 265–273

270 S. Volis et al. Table 4. Results of ordination analyses, including the multivariate method, variable(s) and covariable(s) used, the component of variation explained, its value and signiﬁcance

1 2 3 4 5 6 7 8 9 10 11 12 13

Statistical analysis

Constraining variable*

Covariable*

Component of variation

(PCA) RDA RDA RDA RDA RDA RDA pRDA pRDA pRDA pRDA pRDA pRDA

– A B S A A B A B S A A B

– – – – – – – B + S A + S A + B S B A

a b s a a b a b s a a b

+ B + S + S

+ B + S + S

+ + + + + +

as + ab + abs ab + bs + abs as + bs + abs b + ab + as + bs + abs s + as + ab +bs + abs s + bs + ab +as + abs

+ b + ab + s + as + s + bs

Sum of all canonical eigenvalues

Signiﬁcance of all canonical axes (P-value)

(1) 0Æ474 0Æ070 0Æ194 0Æ509 0Æ598 0Æ232 0Æ400 0Æ035 0Æ124 0Æ439 0Æ563 0Æ159

0Æ013 0Æ221 0Æ219 0Æ035 0Æ015 0Æ285 0Æ029 0Æ325 0Æ250 0Æ036 0Æ028 0Æ281

*Species abundance data were the ﬁrst ﬁve axes of CA on the log-transformed species abundance matrix. Biomass data were the original total biomass values. Soil data were the ﬁrst three axes from the PCA on the soil characters matrix. Iris data (response variable) were the original values in the nine-parameter matrix. Explanatory variable ⁄ component of variation; A ⁄ a – species abundance; B ⁄ b – biomass; S ⁄ s – soil.

a 40·0% as 3·9%

ab 0·4%

s 12·4%

abs 3·1%

b 3·5%

bs 0%

Unexplained = 36·7% Fig. 4. Venn diagram of variation partitioning of iris performance. Each area represents variation in iris performance data explained by the following independent variables: a – species abundance; b – biomass; s – soil; ab, as, bs, abs – two- and three-factor shared eﬀects.

Soil data explained 19Æ4% of the variation and 12Æ4% after accounting for abundance and biomass. Biomass of vegetation in each site explained 7Æ0% of the variation and 3Æ5% when other eﬀects were removed. The shared eﬀects were small in magnitude. Together species abundance and soil, species abundance and biomass, and biomass and soil explained 3Æ9, 0Æ4 and 0% of the variation, respectively. The shared eﬀect of all three factors together (species abundance, biomass and soil) comprised 3Æ1%. The remaining 36Æ7% of iris performance variation were not explained by any of the variables.

The RDA and pRDA revealed, using Monte Carlo permutation tests, that only the eﬀect of species abundance on iris performance was signiﬁcant (Table 4). This eﬀect was signiﬁcant when confounding eﬀects of other variables were not removed (lines 2, 5, 6, 11, 12; Table 2), as well as when these eﬀects were fractioned out (line 8). The eﬀects of total plant biomass and soil on iris performance were not signiﬁcant (Table 4). A relationship between iris performance variables, experimental relocation sites and species composition ⁄ abundance (summarized in two CA axes) can be seen on ordination (RDA) triplot (Fig. 5). The triplot shows a substantial variation among diﬀerent performance measures ⁄ years of observation, and that population at site 7 had consistently higher performance than other populations across performance measures and years of observation. Vegetation (species composition ⁄ abundance) was associated with iris performance two and three years after introduction, but not in the following introduction year. Forward selection of signiﬁcance testing has further shown that only one of the abundance variables had a signiﬁcant eﬀect – the second CA axis summarizing species abundance data (P = 0Æ001; P > 0Æ05 for the other four axes). This means that abundance of species with the highest contribution to Axis 2 of the CA is most correlated with iris performance. Thus, the species with the highest absolute values on Axis 2 are the most important (Fig. 6). Among these species, two abundant species B. distachyum and C. pachystylis were associated with poor iris performance. Other species with high absolute scores were relatively rare, and they did not belong to particular life forms (Fig. 5). For example, the ﬁve species with the highest scores (two negative and three positive) were: an annual grass, Avena barbata Pott ex Link; a geophyte, Ornithogalum narbonense L.; and three annual forbs, Crupina crupinastrum (Moris) Vis., Hedypnois rhagadioloides (L.)

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 265–273

Variation partitioning in translocation 271 1·0 22 15 17 16

18

19

0

12

FL 8 FL 9

2 20

FR 9

9

CA axis 2

10

S = Survived FL = Flowered FR = Fruited 7 = 2007 8 = 2008 9 = 2009

—1·0

21

3

13 8 5 4 11 6 1

S8

14

S9

FR 7 FL 7

CA axis 1

7

S7

0

—1·0

1·0

Fig. 5. Ordination (RDA) triplot showing variation in iris performance (arrows) among the experimental relocation sites (dots), and eﬀect of species composition ⁄ abundance summarized in two CA axes. Note that only CA Axis 2 had a signiﬁcant eﬀect on iris performance. 2·5 2

L om pro C ru cru

1·5

B ra dis

Hed rha

Score on CA axis 2

1

C ar pac

0·5 0

S ti cap Ave s te

0

0·5

1

1·5

2

2·5

3

3·5

4

4·5

Hor s po

–0·5

Annual forb

–1

Perennial forb

–1·5

O rn nar Ave bar

–2 –2·5

Annual grass Perennial grass Geophyte

10 × Abundance per 22 m²

Fig. 6. Species scores on CA Axis 2 as a function of log-transformed abundance per 22 m2 (in all sites combined). Species are marked according to their life form. The CA Axis 2 (Y axis) was the only CA axis having signiﬁcant eﬀect on iris performance. This means that abundance of species with the highest absolute scores on this axis correlates with iris success.

F. W. Schmidt and Lomelosia prolifera (L.) Greuter & Burdet. Conversely, the importance of three of the ﬁve most common species was low, as suggested by their low scores on Axis 2 of the CA (Fig. 5).

Discussion Analysis of performance variables suggests that (i) survival, probability of ﬂowering and fruit set, are largely independent

from each other in relocated I. atrofusca plants and (ii) interannual climatic variation diﬀerentially aﬀects plants across sites. This means that no single performance trait measured in a given year can be used for comparisons between sites. Instead, multi-year data on a suite of performance traits are needed for reliable inferences about relocation success across sites. The RDA performed on a set of environmental (explanatory) and plant performance (response) variables allowed precise partitioning of eﬀects of species abundance, total plant biomass and soil properties. Species abundance and soil eﬀects on iris performance were largely independent and not correlated among sites, and only the eﬀect of the former was signiﬁcant. The eﬀect of total plant biomass was also non-signiﬁcant, and explained only a small portion of variation in iris performance. Therefore, relocation success for this species could be inﬂuenced either directly by interaction with plant community composition; or indirectly by some unmeasured environmental (biotic or abiotic) factors of which plant abundance is an indicator. Although the ﬁrst explanation may apply to a few species (e.g. B. distachyum and C. pachystylis), the second explanation seems more likely for two reasons. First, the eﬀect of total biomass was very small, suggesting that the competitive eﬀect of vegetation on the relocated irises is low. Secondly, the abundance of several rare plant species was most strongly correlated with iris performance while abundance of the common species, except for B. distachyum and C. pachystylis, did not correlate with iris performance (Fig. 6). It is likely that these rare species do not aﬀect the irises directly but are indicators of environmental conditions, other than the soil characters measured, which are important for iris success. It is known that environmental conditions vary greatly on a small scale in water-limited systems, due to resources redistribution creating landscape heterogeneity (Boeken & Shachak 1994; Boeken et al. 1995; Shachak, Sachs & Moshe 1998; Shachak et al. 2008). This pattern was studied in several sites in Israel, including an area (Lehavim) only 4 km southwest of our study area (Shachak et al. 2008). In general, both physical environment (rock-soil ratio, topography) and biological landscape modulation (shrub mounds, biological soil crust, animal diggings) were found to aﬀect runoﬀ redistribution, thus creating water and nutrient enriched or impoverished patches, which in turn aﬀects vegetation community composition. If the performance of iris plants is aﬀected by the patch characteristics as well, plant species indicative of the preferred patches can be identiﬁed and used for selection of the introduction sites. Species of the section Oncocyclus in the genus Iris are well adapted to aridity and are mostly found in the semi-arid habitats of the Middle East (Avishai 1977). In Israel, these species occupy open low-herbaceous or shrub communities (Sapir et al. 2002). There is evidence that they suﬀer from light competition with shrubs, such as Sarcopoterium spinosum, as succession progresses in herbaceous plant communities (Segal 2006). However, the relocation sites in our study area were always chosen to be herbaceous vegetation patches, even where the overall Sarcopoterium spinosum cover was high (Fig. 1). Therefore, competition with shrubs as a factor negatively

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 265–273

272 S. Volis et al. aﬀecting relocation success can be ruled out. Competition with herbaceous vegetation is also unlikely, as indicated by the relatively small eﬀect of the most abundant plant species on iris performance and by the very low amount of variation in iris performance explained by variation in plant biomass. As for the negative eﬀects of B. distachyum and C. pachystylis on iris plants, this may be due to allelopathy. For example, litter of Brachypodium retusum negatively aﬀected seedling growth of an endangered shrub Cistus heterophyllus (Navarro-Cano 2008). The results of this study have demonstrated the usefulness of CCA ⁄ RDA for understanding the causes of relocation success or failure. The many biotic and abiotic factors aﬀecting individual performance are often complex and ⁄ or highly inter-correlated, therefore elucidating the primary environmental eﬀect(s) on relocated plants is challenging. Counter intuitively, the eﬀect of soil was less important than plant abundance ⁄ composition. The eﬀect of the total biomass, indicative of intensity of competition for resources, was negligible; only the eﬀect of composition ⁄ abundance of surrounding relocated plants vegetation was signiﬁcant and explained more than 40% of variation in performance of iris plants. Counter intuitively again, abundances of rare rather than common species, correlated with iris performance. These ﬁndings show that CCA ⁄ RDA can help with the choice of future translocation microsites, the most important decisions in translocation, according to the knowledge gained about the ecological niche of the species. On the other hand, we have seen that even in relocations of perennial plants, it is diﬃcult to ascertain what environmental factors determine success. To conclude, variation partitioning can be used to monitor success and to identify the key factors in experimental translocations preceding actual introduction of plants in conservation programmes. The (partial) ordination techniques, due to their ability to integrate multiple datasets and reduce the list of possible eﬀects, can be a useful tool for improving our knowledge of the ecological requirements of endangered species.

Acknowledgments This project was supported by a grant from the Israel Ministry of Sciences and a grant from Israel Nature and Parks Authority. We would like to thank Israel Nature and Parks Authority for assistance in completion of this project and David Saltz for helpful comments on an early version of the manuscript.

References Adamec, L. & Lev, J. (1999) The introduction of the aquatic carnivorous plant Aldrovanda vesiculosa to the new potential sites in the Czech Republic: a ﬁveyear investigation. Folia Geobotanica, 34, 299–305. Avishai, M. (1977) Species relationships and cytogenetic aﬃnities in section Oncocyclus of the genus Iris. PhD thesis, The Hebrew University of Jerusalem. Basille, M., Calenge, C., Marboutin, E., Andersen, R. & Gaillard, J.-M. (2008) Assessing habitat selection using multivariate statistics: some reﬁnements of the ecological-niche factor analysis. Ecological Modelling, 211, 233–40. Boeken, B. & Shachak, M. (1994) Desert plant communities in human-made patches – implications for management. Ecological Applications, 4, 702–716.

Boeken, B., Shachak, M., Gutterman, Y. & Brand, S. (1995) Patchiness and disturbance: plant community responses to porcupine diggings in the central Negev. Ecography, 18, 410–422. Borcard, D., Legendre, P. & Drapeau, P. (1992) Partialling out the spatial component of ecological variation. Ecology, 73, 1045–1055. ter Braak, C.J.F. (1986) Canonical correspondence analysis: a new eigenvector technique for multivariate direct gradient analysis. Ecology, 67, 1167– 1179. ter Braak, C.J.F. & Smilauer, P. (2002) CANOCO Reference Manual and CanoDraw for Windows User’s Guide: Software for Canonical Community Ordination (version 4Æ5). Microcomputer Power, Ithaca, NY, USA, 500pp. Burkey, T.V. (1995) Extinction rates in archipelagoes: implications for populations in fragmented habitats. Conservation Biology, 9, 527–541. Busby, J.R. (1991) BIOCLIM – a bioclimate analysis and prediction system. Nature Conservation: Cost Eﬀective Biological Surveys and Data Analysis (eds C.R. Margules & M.P. Austin), pp. 64–68. CSIRO, Melbourne, Australia. Carpenter, G., Gillison, A.N. & Winter, J. (1993) DOMAIN – a ﬂexible modeling procedure for mapping potential distributions of plants and animals. Biodiversity and Conservation, 2, 667–680. Clarke, P.J., Latz, P.K. & Albrecht, D.E. (2005) Long-term changes in semi-arid vegetation: invasion of an exotic perennial grass has larger eﬀects than rainfall variability. Journal of Vegetation Science, 16, 237– 248. Colas, B., Kirchner, F., Riba, M., Olivieri, I., Mignot, A., Imbert, E., Beltrame, C., Carbonell, D. & Freville, H. (2008) Restoration demography: a 10-year demographic comparison between introduced and natural populations of endemic Centaurea corymbosa (Asteraceae). Journal of Applied Ecology, 45, 1468–1476. Elith, J. & Leathwick, J.R. (2009) Species distribution models: ecological explanation and prediction across space and time. Annual Review of Ecology, Evolution and Systematics, 40, 677–697. Elmendorf, S.C. & Moore, K.A. (2007) Plant competition varies with community composition in an edaphically complex landscape. Ecology, 88, 2640– 2650. Engler, R., Guisan, A. & Rechsteiner, L. (2004) An improved approach for predicting the distribution of rare and endangered species from occurrence and pseudo-absence data. Journal of Applied Ecology, 41, 263–274. Gordon, D.R. (1996) Experimental translocation of the endangered shrub Apalachicola rosemary Conradina glabra to the Apalachicola Bluﬀs and Ravines Preserve, Florida. Biological Conservation, 77, 19–26. Griﬃth, B., Scott, J.M., Carpenter, J.W. & Reed, C. (1989) Translocation as a species conservation tool: status and strategy. Science, 245, 477–480. Grinnell, J. (1917) Field tests of theories concerning distributional control. American Naturalist, 51, 115–128. Guisan, A. & Thuiller, W. (2005) Predicting species distribution: oﬀering more than simple habitat models. Ecological Letters, 8, 993–1009. Guisan, A. & Zimmermann, N.E. (2000) Predictive habitat distribution models in ecology. Ecological Modelling, 135, 147–186. Heywood, V.H. & Iriondo, J.M. (2003) Plant conservation: old problems, new perspectives. Biological Conservation, 113, 321–335. Hirzel, A.H., Hausser, J., Chessel, D. & Perrin, N. (2002) Ecological-niche factor analysis: how to compute habitat-suitability maps without absence data? Ecology, 83, 2027–2036. IUCN (the International Union for Conservation of Nature). (1987) The IUCN Position Statement on Translocation of Living Organisms: Introductions, Re-introductions and Restocking. IUCN, Gland. Jusaitis, M. (2005) Translocation trials conﬁrm speciﬁc factors aﬀecting the establishment of three endangered plant species. Ecological Management & Restoration, 6, 61–67. MacArthur, R.H. (1972) Geographical Ecology. Princeton University Press, Princeton. Maschinski, J. & Duquesnel, J. (2006) Successful reintroductions of the endangered long-lived Sargent’s cherry palm, Pseudophoenix sargentii, in the Florida Keys. Biological Conservation, 134, 122–129. Maunder, M. (1992) Plant reintroduction: an overview. Biodiversity and Conservation, 1, 51–61. Menges, E.S. (2008) Restoration demography and genetics of plants: when is a translocation successful? Australian Journal of Botany, 56, 187–196. Milton, S.J., Bond, W.J., Du Plessis, M.A., Gibbs, D., Hilton-Taylor, C., Linder, H.P., Raitt, L., Wood, J. & Donaldson, J.S. (1999) A protocol for plant conservation by translocation in threatened lowland Fynbos. Conservation Biology, 13, 735–743.

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 265–273

Variation partitioning in translocation 273 Morgan, J.W. (1999) Have tubestock planting successfully established populations of rare grassland species into reintroduced sites in western Victoria? Biological Conservation, 89, 235–243. Navarro-Cano, J.A. (2008) Eﬀect of grass litter on seedling recruitment of the critically endangered Cistus heterophyllus in Spain. Flora - Morphology, Distribution, Functional Ecology of Plants, 203, 663–668. Noe, G.B. & Zedler, J.B. (2001) Spatio-temporal variation of salt marsh seedling establishment in relation to the abiotic and biotic environment. Journal of Vegetation Science, 12, 61–74. Peterson, A.T., Sobero´n, J. & Sa´nchez-Cordero, V. (1999) Conservatism of ecological niches in evolutionary time. Science, 285, 1265–1267. Pivello, V.G., Shida, C.N. & Meirelles, S.T. (1999) Alien grasses in Brazilian savannas: a threat to biodiversity. Biodiversity and Conservation, 8, 1281– 1294. Richards, C.L., Carstens, B.C. & Lacey Knowlrs, L. (2007) Distribution modelling and statistical phylogeography: an integrative framework for generating and testing alternative biogeographical hypotheses. Journal of Biogeography, 34, 1833–1845. Sapir, Y., Shmida, A., Fragman, O. & Comes, H.P. (2002) Morphological variation of the Oncocyclus irises (Iris: Iridaceae) in the southern Levant. Botanical Journal of the Linnean Society, 139, 369–382. Sarrazin, F. & Barbault, R. (1996) Reintroduction: challenges and lessons for basic ecology. Trends in Ecology & Evolution, 11, 474–478. Seddon, P.J., Armstrong, D.P. & Maloney, R.F. (2007) Developing the science of reintroduction biology. Conservation Biology, 21, 303–312. Segal, B. (2006) Ecological aspects of the Iris bismarckiana population in the iris reserve for development of reservation management. MSc thesis, Technion, Israel Institute of Technology (in Hebrew). Shachak, M., Sachs, M. & Moshe, I. (1998) Ecosystem management of desertiﬁed shrublands in Israel. Ecosystems, 1, 475–483. Shachak, M., Boeken, B., Groner, E., Kadmon, R., Lubin, Y., Meron, E., Ne’eman, G., Perevolotsky, A., Shkedy, Y. & Ungar, E.D. (2008) Woody species as landscape modulators and their eﬀect on biodiversity patterns. BioScience, 58, 209–221. Shmida, A. & Pollak, G. (2007) Red Data Book: Endangered Plants of Israel, Vol. 1. Israel Nature and Parks Authority, Jerusalem, 495 pp [In Hebrew]. Stauﬀer, D.E. (2002) Linking populations and habitats: Where have we been? Where are we going? Predicting Species Occurrences: Issues of Accuracy and Scale (eds J.M. Scott, P.J. Heglund, M.L. Morrison, M.G. Raphael & W.A. Wall), pp. 53–61. Island Press, Covelo, CA. Stockwell, D.R.B. & Peters, D. (1999) The GARP modeling system: problems and solutions to automated spatial prediction. International Journal of Geographic Information Science, 13, 143–58.

Svenning, J.-C. & Skov, F. (2005) The relative roles of environment and history as controls of tree species composition and richness in Europe. Journal of Biogeography, 32, 1019–1033. Tsoar, A. & Ramon, U. (2002) Lahav Survey – Survey, Analysis and Evaluation of Nature, Landscape and Heritage Resources. The Society for the Protection of Nature in Israel, Tel-Aviv, 127 pp. (in Hebrew). Volis, S., Anikster, Y., Olsvig-Whittaker, L. & Mendlinger, S. (2004) The inﬂuence of space in genetic-environment relationships when environmental heterogeneity and seed dispersal occur at similar scales. The American Naturalist, 163, 312–327. Walker, P.A. & Cocks, K.D. (1991) HABITAT: a procedure for modeling a disjoint environmental envelope for a plant or animal species. Global Ecology and Biogeography Letters, 1, 108–118. Received 29 April 2010; accepted 20 October 2010 Handling Editor: Jennifer Firn

Supporting information Additional Supporting Information may be found in the online version of this article. Table S1. Abundance and life form of 75 plant species per 1 m2 for 22 sites Table S2. Total plant biomass (dry weight) per 1 m2 for 22 sites Table S3. Soil characteristics measured at the 22 sites. See ‘Materials and methods’ section for explanation of sampling methods and abbreviations As a service to our authors and readers, this journal provides supporting information supplied by the authors. Such materials may be re-organized for online delivery, but are not copy-edited or typeset. Technical support issues arising from supporting information (other than missing ﬁles) should be addressed to the authors.

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 265–273

Journal of Applied Ecology 2011, 48, 143–147

doi: 10.1111/j.1365-2664.2010.01899.x

FORUM

Connectivity, dispersal behaviour and conservation under climate change: a response to Hodgson et al. Veronica A. J. Doerr1,2*, Tom Barrett3 and Erik D. Doerr1,2 1

CSIRO Ecosystem Sciences, GPO Box 284, Canberra ACT 2601, Australia; 2Research School of Biology, Australian National University, Canberra ACT 0200, Australia; and 3New South Wales Department of Environment, Climate Change & Water, PO Box 494, Armidale NSW 2350, Australia

Summary 1. Hodgson et al. [Journal of Applied Ecology 46 (2009) 964] argue that connectivity is complex and uncertain, that it can be improved incidentally by increasing habitat extent, and that connectivity conservation is unlikely to be eﬀective under climate change. 2. We believe that they have overlooked recent research on dispersal behaviour and structural connectivity, which has improved our understanding of functional connectivity and revealed that it will not necessarily increase with habitat extent. 3. New modelling techniques including least-cost path models incorporate this more detailed understanding of connectivity into conservation planning, facilitating the true aim of connectivity conservation – to ensure appropriate interactions between habitat extent, quality and connectivity. 4. Synthesis and applications. Advances in behavioural research and modelling techniques allow us to manage structural connectivity with as much certainty as we manage extent and quality of habitat. Successful landscape conservation to address both current threats and future climate change must manage these three elements in concert. Key-words: aggregation, behavioural ecology, connectivity conservation, corridor, fragmentation, gap-crossing, metapopulation, population viability, range shift, stepping stone

Introduction For most of the world’s ecosystems, human-induced habitat loss, degradation and fragmentation are primary causes of declines in biodiversity (Fahrig 2003; Lindenmayer & Fischer 2006). Furthermore, climate change is predicted to interact with and intensify the eﬀects of these problems. ‘Connectivity conservation’ has emerged as an overarching solution with considerable political and popular support (Crooks & Sanjayan 2006). However, Hodgson et al. (2009) highlight the dangers of investing in connectivity per se, and argue that other strategies may provide better protection for species in a changing climate. We wholeheartedly agree with Hodgson et al. that connectivity should not be the sole focus of conservation actions, and that conservation investments should be based on analysis of their likely beneﬁts. Yet Hodgson et al. suggest that connectivity conservation is never likely to be a robust strategy, and here we disagree. Speciﬁcally, Hodgson et al. argue that there is too

*Correspondence author. E-mail: [email protected]

much uncertainty surrounding connectivity, that connectivity is primarily a result of habitat aggregation, and it can coincidentally be improved by increasing habitat extent. They also suggest that connectivity conservation sacriﬁces long-term conservation success under a changing climate in favour of short-term gains. In this response, we suggest that Hodgson et al. have overlooked recent advances in our understanding of connectivity, particularly arising from research on dispersal behaviour. These advances provide a clearer distinction between structural and functional connectivity and greater certainty regarding the eﬀects of structural connectivity. They also bring a new awareness that increases in habitat extent alone will not necessarily increase functional connectivity. In addition, we believe that Hodgson et al. have misinterpreted connectivity conservation, which carries a speciﬁc meaning that involves more than just conserving structural connectivity, and can provide long-term solutions to many of the threats associated with climate change including those highlighted by Hodgson et al. Finally, we suggest that the diﬀerences in our perspectives may partly result from diﬀerences in the scales at which empirical research and conservation planning are conducted. Fortunately, new modelling techniques are allowing us

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society

144 V. A. J. Doerr, T. Barrett & E. D. Doerr to move beyond simple measures of aggregation to incorporate a more detailed behavioural understanding of connectivity into conservation planning, despite diﬀerences in scale.

Dispersal behaviour and structural connectivity The intent of connectivity is to facilitate dispersal of individuals. Thus, an empirical understanding of connectivity depends on understanding animal behaviour, particularly movement and dispersal behaviour, to reveal what parts of landscapes individuals are willing to move through and why (Lima & Zollner 1996; Chetkiewicz, Clair & Boyce 2006). The value of behavioural research for conservation has been debated (Blumstein & Fernandez-Juricic 2004; Caro 2007), and thus it is unsurprising if conservation biologists are not familiar with the movement behaviour literature, very little of which existed during early discussions about connectivity. Yet movement behaviour is a rapidly growing ﬁeld (Nathan 2008), with extensive empirical analyses and emerging theories that can provide a strong foundation for modelling and conserving connectivity. When connectivity began to be viewed from a behaviourbased perspective, it became a characteristic of the matrix between subpopulations (Taylor, Fahrig & With 2006), rather than a characteristic of patches or landscapes. Movement could be dependent not just on distances between subpopulations (i.e. aggregation) but on the physical characteristics of the matrix itself, particularly the presence of habitat elements too small for settlement but which might nonetheless facilitate movement. As a result, a much clearer distinction emerged between structural and functional connectivity. ‘Structural connectivity’ refers to physical characteristics of the landscape between patches of occupied habitat. ‘Functional connectivity’ refers to the degree to which movement of individuals and ⁄ or their genetic material actually occurs, and is inﬂuenced by both movement potential due to structural connectivity and by local subpopulation dynamics (Hilty, Lidicker & Merenlender 2006). Empirical research on movement behaviour has concentrated on revealing what types of structural connectivity provide the potential for dispersal movements and thus contribute to functional connectivity. A number of studies have shown that various species use corridors to move through fragmented landscapes (Haddad et al. 2003; Haddad & Tewksbury 2005), and some have demonstrated the use of simpler landscape elements such as scattered trees (Fischer & Lindenmayer 2002; Doerr, Doerr & Davies 2010). Research on gap-crossing behaviour has been particularly critical, identifying gap distances either within corridors or among scattered trees that may prevent movements, thus revealing details of structural connectivity that contribute to movement potential (St. Clair et al. 1998; Grubb & Doherty 1999; Robertson & Radford 2009). In addition, research on overall movement strategies such as foray search is illustrating that distances between subpopulations (i.e. aggregation) may have a threshold eﬀect rather than a linear eﬀect on movement potential. When

individuals use a foray-based search strategy, they may have a maximum search distance beyond which they will not travel, regardless of how much structural connectivity is present in the landscape (Conradt, Roper & Thomas 2001; Doerr & Doerr 2005; Doerr, Doerr & Davies 2010). The species-speciﬁc nature of behavioural research may be viewed as an impediment to its usefulness for ecosystem conservation, but patterns are emerging which suggest that responses to structural connectivity may not be as species-speciﬁc as was once thought (Haddad et al. 2003; Doerr, Doerr & Davies 2010; Gilbert-Norton et al. 2010). Instead, movement behaviour may be shaped by the structure of environments experienced over evolutionary time, and species in any given ecological community with broadly similar life-histories may have evolved similar movement behaviours as responses to their shared environments (Fahrig 2007). For example, Doerr, Doerr & Davies (in press) found that use of scattered trees, foray distances, and gap distances crossed were similar among ﬁve Australian woodland birds despite substantial diﬀerences in their ecology. Belisle (2005) proposed that travel costs may provide one mechanism through which landscapes can exert similar evolutionary pressures across species. Thus, theories from behavioural ecology such as the marginal value theorem could provide the basis for general theories of connectivity, allowing us to predict the eﬀects of diﬀerent types of structural connectivity for large suites of species at once (Belisle 2005).

Conservation certainty All of these advances are making structural connectivity a much more measurable and manageable concept than Hodgson et al. suggest. Functional connectivity remains complex because it integrates movement potential with the dynamics of subpopulations (which is why Hodgson et al. deem it too intractable for conservation planning). Yet structural connectivity contributes signiﬁcantly to functional connectivity by determining movement potential. The resulting eﬀects on population persistence are also increasingly predictable thanks to controlled research in experimental landscapes which is demonstrating that connected patches experience fewer local extinctions than isolated patches (Damschen et al. 2006; Brudvig et al. 2009). Thus, structural connectivity can be directly quantiﬁed in the landscape, has predictable eﬀects on movement potential, and is known to contribute to population persistence, making it a worthwhile focus for management. Unfortunately, Hodgson et al. omit structural connectivity from their schematic of the place of connectivity in conservation (Hodgson et al., Fig. 2). We have revised their diagram to distinguish between structural and functional connectivity and depict the relationships between them, as well as relationships with the area and quality of habitat suitable for settlement (Fig. 1). Structural connectivity is independent of habitat area and quality and is what deﬁnes habitat for dispersal, just as area and quality are what deﬁne habitat for settlement. Structural connectivity, habitat area and quality interact to determine functional connectivity, but they also interact to determine subpopulation dynamics and thus the eﬀective

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 143–147

Connectivity and dispersal behaviour 145 What the organism needs for survival and reproduction

What the organism needs for dispersal

…Defines…

…Defines…

Habitat for settlement

Habitat for dispersal …Is quantified in terms of…

…Is quantified in terms of… Area

Structural connectivity

Quality

…Which is the interaction between… Landscape characteristics between areas of habitat for settlement

Distance between areas of habitat for settlement

…Which determine…

Potential rates of dispersal between subpopulations

…Which interact to determine… Functional connectivity (actual rates of dispersal) and subpopulation size and dynamics …Which are integrated over multiple subpopulations to determine… Effective population size, spatial distribution and persistence probability

Fig. 1. A schematic illustrating the role of both structural and functional connectivity in spatial ecology and conservation (revised from Fig. 2 in Hodgson et al. (2009)). Functional connectivity results from interactions between the amount and quality of habitat suitable for settlement as well as the inﬂuence of the rest of the landscape (i.e. structural connectivity) on the potential for dispersal. Structural connectivity is therefore a vital component of functional connectivity that is tractable to model and manage.

population size of the population as a whole. None has a direct inﬂuence on populations completely independent of the others – all provide the same degree of conservation certainty because their beneﬁts depend on the interactions between them. Hodgson et al. also argue we can be relatively certain about the positive eﬀects of increasing habitat extent and habitat quality – often assumed to be accomplished through increasing the size and ⁄ or number of protected areas. Yet in highly disturbed ecosystems, there may be little habitat left outside of already existing protected areas. Thus, increasing habitat extent and improving habitat quality involves restoring habitat in areas where it has been lost to other land uses. Unfortunately, there are substantial limitations and uncertainties in our ability to restore ecosystems. For example, nitrogen enrichment via fertilisation reduces plant diversity as well as the stability of ecosystems worldwide (McIntyre 2008; Bai et al. 2010). These eﬀects can last long after fertilisation has ceased, inhibiting full recovery of the ecosystem despite restoration attempts (Munro et al. 2009). It is also reasonable to argue that habitat quality will often be more species-speciﬁc than structural connectivity, as habitat for settlement must

provide for many more of a species’ needs than habitat for dispersal (Haddad & Tewksbury 2005; Doerr, Doerr & Davies 2010). Restoring habitat for settlement, either for a single species or particularly for an entire community, may thus be more complex and uncertain than restoring habitat for dispersal (i.e. increasing structural connectivity). Finally, Hodgson et al. argue that increasing habitat extent will coincidentally improve connectivity by increasing aggregation and thus reducing distances between patches. However, behavioural research suggests that reducing inter-patch distances without providing structural connectivity will only be beneﬁcial once patches become close enough to allow gap-crossing between them. That distance may be as little as 60–100 m (unlikely to be achieved by most eﬀorts to increase habitat extent), as many species are unwilling to cross gaps any larger (Desrochers & Hannon 1997; Robertson & Radford 2009; Doerr, Doerr & Davies 2010). Individuals can traverse much greater distances between habitat patches if structural connectivity is present, but the existence of foray-based search means that increases in aggregation may only be beneﬁcial if distances between patches can be reduced below a critical foray distance threshold, which may only be 1–2 km (Doerr, Doerr & Davies 2010, in press). Thus, the beneﬁts of reducing aggregation per se (as opposed to managing it in concert with structural connectivity) are risky because they are not commensurate with eﬀort.

‘Connectivity conservation’ is more than just conserving connectivity Hodgson et al. interpret connectivity conservation as the eﬀort to increase structural connectivity with the primary purpose of enabling species’ range shifts due to climate change. However, as highlighted above, structural connectivity interacts with other aspects of the landscape and thus is not necessarily the sole or most important aspect to improve in every landscape. Connectivity conservation acknowledges this, and has a very speciﬁc meaning in the literature (IUCN WCPA 2006; Worboys 2010), much like ‘systematic conservation planning’ has a speciﬁc meaning and doesn’t merely refer to taking a systematic approach to planning conservation actions (Margules & Pressey 2000). As a result, connectivity conservation is broader than Hodgson et al.’s interpretation. Connectivity conservation can be deﬁned as coordinated eﬀorts to achieve metapopulation viability across a range of spatial scales, which involves evaluating and improving the interactions between habitat area, habitat quality and structural connectivity (Crooks & Sanjayan 2006; Worboys 2010). There is no overarching rule about which action is always more eﬀective – this depends on the existing conditions in a given landscape. Connectivity conservation thus aims to develop ﬂexible solutions, tailored to the diﬀerent needs of diﬀerent landscapes. This may involve protecting large continuous areas of existing habitat, but may also involve protecting or increasing connections between multiple small discontinuous areas of habitat where that is all that remains. The preference of Hodgson et al. to focus on habitat area and habitat quality can thus be

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 143–147

146 V. A. J. Doerr, T. Barrett & E. D. Doerr encompassed by connectivity conservation wherever these actions are deemed to provide the greatest beneﬁts. Finally, the ultimate purpose of connectivity conservation is not simply to facilitate range shifts, but to increase the resilience of populations to the variety of threats caused by or intensiﬁed by climate change. Under connectivity conservation, structural connectivity is desired where it links multiple subpopulations via dispersal, allowing subpopulations to function collectively as one larger, more resilient population. These principles can be applied at any scale, not just scales that might be relevant for possible range shifts under climate change (Opdam & Wascher 2004). Thus, connectivity conservation can be used to reduce pressures other than climate change, can be applied to increase viability of populations in centres of endemism, and can concentrate on areas of high environmental heterogeneity – all of which are principles that Hodgson et al. suggest will underlie robust conservation strategies under climate change.

Moving beyond aggregation in large-scale conservation planning models Conservation modellers may still be unsure how to incorporate advances in our understanding of connectivity and connectivity conservation due to the diﬀerent scales at which behavioural research and conservation planning are usually conducted. Conservation planning often occurs at very large scales – regions to global scales. Yet a behavioural understanding of connectivity is shaped at scales relevant to movement of individuals – local to landscape scales. The diﬃculty is that incorporating small-scale detail in large-scale models is often deemed computationally intractable. Fortunately, there are promising new advances that can model connectivity over large spatial scales in ways that align more closely with a behaviourbased view of connectivity. First, we have already noted that the types of structural connectivity that facilitate dispersal movements are not necessarily species-speciﬁc. Thus, models may only need to incorporate general principles (such as threshold distances between habitat patches) rather than behavioural detail speciﬁc to many diﬀerent species. Another way in which conservation planning models can incorporate a tractable amount of behavioural detail is through the use of least-cost path modelling and state-space modelling. These new types of models simultaneously explore behavioural and landscape parameters to identify which landscape details most need to be incorporated into large-scale models (Chetkiewicz, Clair & Boyce 2006; Kadoya 2009), which can then be kept relatively simple by modelling only the few most relevant small-scale parameters. Remaining computational challenges can often be overcome by decreasing the sizes of grid cells only to a relevant scale. Further behavioural detail can then be incorporated by modelling resistance of grid cells that have diﬀerent compositions (McRae & Beier 2007). One example of the success of these new approaches comes from our own work, in which data on gap-crossing distances (Doerr, Doerr & Davies 2010) were used to deﬁne the maximum distance individuals will move through non-habitat, and

data on foray-based search behaviour and foray distances (Doerr & Doerr 2005; Doerr, Doerr & Davies 2010) were used to deﬁne the maximum distance individuals will move through structural connectivity, modelled as suboptimal habitat. Using modern satellite imagery, suboptimal habitat could be mapped at a ﬁne scale of resolution to detect very small elements of structural connectivity known to support dispersal movements, such as single trees. Fine-scale habitat mapping and simpliﬁed behavioural rules were then modelled together using the least-cost path approach (Drielsma, Manion & Ferrier 2007), with habitat quality as a surrogate for movement cost. This modelling approach simultaneously evaluates habitat area, quality and structural connectivity, identifying where these elements currently exist in concert in the landscape versus where they are unable to interact due to deﬁciencies in one or more elements (Barrett et al. 2010). This gives conservation planners the ability to make practical recommendations that maximise the likelihood that actions at a local scale will contribute to population viability and resilience at large scales (Barrett et al. 2010). These models are currently being used to guide conservation planning decisions in several regions of New South Wales, Australia.

Reasons to be cheerful – indeed! As Hodgson et al. suggest, it is easy to be overwhelmed by the challenges of conservation under climate change. It is worthwhile returning to basic principles, focusing on actions that will be cost-eﬀective and that will address current threats as well as those anticipated due to climate change. Fortunately, connectivity conservation provides such an approach by focusing on habitat area, quality, and structural connectivity as independent attributes that must all work together to support viable, resilient populations. Thanks to a growing body of behavioural research, we can reliably identify situations in which fostering structural connectivity in the matrix is likely to yield positive beneﬁts for populations. We can also use behavioural information to model connectivity alongside habitat extent and quality and thus tailor management to speciﬁc landscapes. Ultimately, these new techniques ensure that large-scale conservation planning can take advantage of up-to-date connectivity knowledge to truly provide evidence-based conservation guidance.

Acknowledgements Thanks to members of the Great Eastern Ranges Initiative, David Westcott, Paul Sunnucks, Sasha Pavlova, and Colleen Cassady St. Clair for discussions that shaped these ideas. The manuscript was greatly improved by the comments of Richard Fuller, Sue McIntyre, Dan Lunney, Vicki Logan, and ﬁve anonymous reviewers.

References Bai, Y.F., Wu, J.G., Clark, C.M., Naeem, S., Pan, Q.M., Huang, J.H., Zhang, L.X. & Han, X.G. (2010) Tradeoﬀs and thresholds in the eﬀects of nitrogen addition on biodiversity and ecosystem functioning: evidence from inner Mongolia Grasslands. Global Change Biology, 16, 358–372. Barrett, T., Drielsma, M., Love, J., Howling, G. & Anderson, G. (2010) Modelling of Fauna Habitat and Connectivity Values to Support Conservation Planning Decisions in the Hunter Region of NSW as Part of the Great Eastern

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 143–147

Connectivity and dispersal behaviour 147 Ranges Initiative. New South Wales Department of Environment, Climate Change and Water, New South Wales. Belisle, M. (2005) Measuring landscape connectivity: the challenge of behavioral landscape ecology. Ecology, 86, 1988–1995. Blumstein, D.T. & Fernandez-Juricic, E. (2004) The emergence of conservation behavior. Conservation Biology, 18, 1175–1177. Brudvig, L.A., Damschen, E.I., Tewksbury, J.J., Haddad, N.M. & Levey, D.J. (2009) Landscape connectivity promotes plant biodiversity spillover into non-target habitats. Proceedings of the National Academy of Sciences of the United States of America, 106, 9328–9332. Caro, T. (2007) Behavior and conservation: a bridge too far? Trends in Ecology & Evolution, 22, 394–400. Chetkiewicz, C.L.B., Clair, C.C.S. & Boyce, M.S. (2006) Corridors for conservation: integrating pattern and process. Annual Review of Ecology Evolution and Systematics, 37, 317–342. Conradt, L., Roper, T.J. & Thomas, C.D. (2001) Dispersal behaviour of individuals in metapopulations of two British butterﬂies. Oikos, 95, 416–424. Crooks, K.R. & Sanjayan, M., ed. (2006) Connectivity Conservation. Cambridge University Press, Cambridge. Damschen, E.I., Haddad, N.M., Orrock, J.L., Tewksbury, J.J. & Levey, D.J. (2006) Corridors increase plant species richness at large scales. Science, 313, 1284–1286. Desrochers, A. & Hannon, S.J. (1997) Gap crossing decisions by forest songbirds during the post-ﬂedging period. Conservation Biology, 11, 1204–1210. Doerr, E.D. & Doerr, V.A.J. (2005) Dispersal range analysis: quantifying individual variation in dispersal behaviour. Oecologia, 142, 1–10. Doerr, V.A.J., Doerr, E.D. & Davies, M.J. (2010) Systematic Review #44: Does Structural Connectivity Facilitate Dispersal of Native Species in Australia’s Fragmented Terrestrial Landscapes? Collaboration for Environmental Evidence, Bangor. Doerr, V.A.J., Doerr, E.D. & Davies, M.J. (in press) Dispersal behaviour of Brown Treecreepers predicts functional connectivity for several other woodland birds. Emu. Drielsma, M., Manion, G. & Ferrier, S. (2007) The spatial links tool: automated mapping of habitat linkages in variegated landscapes. Ecological Modelling, 200, 403–411. Fahrig, L. (2003) Eﬀects of habitat fragmentation on biodiversity. Annual Review of Ecology Evolution and Systematics, 34, 487–515. Fahrig, L. (2007) Non-optimal animal movement in human-altered landscapes. Functional Ecology, 21, 1003–1015. Fischer, J. & Lindenmayer, D.B. (2002) The conservation value of paddock trees for birds in a variegated landscape in southern New South Wales. 2. Paddock trees as stepping stones. Biodiversity and Conservation, 11, 833–849. Gilbert-Norton, L., Wilson, R., Stevens, J.R. & Beard, K.H. (2010) A metaanalytic review of corridor eﬀectiveness. Conservation Biology, 24, 660–668. Grubb, T.C. & Doherty, P.F. (1999) On home-range gap-crossing. The Auk, 116, 618–628. Haddad, N.M. & Tewksbury, J.J. (2005) Low-quality habitat corridors as movement conduits for two butterﬂy species. Ecological Applications, 15, 250–257.

Haddad, N.M., Bowne, D.R., Cunningham, A., Danielson, B.J., Levey, D.J., Sargent, S. & Spira, T. (2003) Corridor use by diverse taxa. Ecology, 84, 609– 615. Hilty, J.A., Lidicker, W.Z. & Merenlender, A.M. (2006) Corridor Ecology: The Science and Practice of Linking Landscapes for Biodiversity Conservation. Island Press, Washington, D.C. Hodgson, J.A., Thomas, C.D., Wintle, B.A. & Moilanen, A. (2009) Climate change, connectivity and conservation decision making: back to basics. Journal of Applied Ecology, 46, 964–969. IUCN WCPA (2006) Papallacta Declaration. Mountain Forum Bulletin, 7, 46. Kadoya, T. (2009) Assessing functional connectivity using empirical data. Population Ecology, 51, 5–15. Lima, S.L. & Zollner, P.A. (1996) Towards a behavioral ecology of ecological landscapes. Trends in Ecology and Evolution, 11, 131–135. Lindenmayer, D.B. & Fischer, J. (2006) Habitat Fragmentation and Landscape Change: An Ecological and Conservation Synthesis. Island Press, Washington, D.C.. Margules, C.R. & Pressey, R.L. (2000) Systematic conservation planning. Nature, 405, 243–253. McIntyre, S. (2008) The role of plant leaf attributes in linking land use to ecosystem function in temperate grassy vegetation. Agriculture Ecosystems & Environment, 128, 251–258. McRae, B.H. & Beier, P. (2007) Circuit theory predicts gene ﬂow in plant and animal populations. Proceedings of the National Academy of Sciences of the United States of America, 104, 19885–19890. Munro, N.T., Fischer, J., Wood, J. & Lindenmayer, D.B. (2009) Revegetation in agricultural areas: the development of structural complexity and ﬂoristic diversity. Ecological Applications, 19, 1197–1210. Nathan, R. (2008) An emerging movement ecology paradigm. Proceedings of the National Academy of Sciences of the United States of America, 105, 19050–19051. Opdam, P. & Wascher, D. (2004) Climate change meets habitat fragmentation: linking landscape and biogeographical scale levels in research and conservation. Biological Conservation, 117, 285–297. Robertson, O.J. & Radford, J.Q. (2009) Gap-crossing decisions of forest birds in a fragmented landscape. Austral Ecology, 34, 435–446. St. Clair, C.C., Belisle, M., Desrochers, A. & Hannon, S. (1998) Winter responses of forest birds to habitat corridors and gaps. Conservation Ecology [online], 2, 13. Taylor, P.D., Fahrig, L. & With, K.A. (2006) Landscape connectivity: a return to the basics. Connectivity Conservation (eds K.R. Crooks & M. Sanjayan), pp. 29–43. Cambridge University Press, Cambridge. Worboys, G.L. (2010) The connectivity conservation imperative. Connectivity Conservation Management – A Global Guide (eds G.L. Worboys, W.L. Francis & M. Lockwood), pp. 3–21. Earthscan, London. Received 16 June 2010; accepted 20 October 2010 Handling Editor: Morten Frederiksen

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 143–147

Journal of Applied Ecology 2011, 48, 47–55

doi: 10.1111/j.1365-2664.2010.01900.x

Making better sense of monitoring data from low density species using a spatially explicit modelling approach Max Post van der Burg1*†, Bartholomew Bly2, Tammy VerCauteren3 and Andrew J. Tyre1 1

School of Natural Resources, 3310 Holdrege Street, University of Nebraska-Lincoln, Lincoln, NE 68583, USA; Nebraska Prairie Partners, PO Box 489, Scottsbluﬀ, NE 69363, USA; and 3Rocky Mountain Bird Observatory, PO Box 1232, Brighton, CO 80601, USA 2

Summary 1. Wildlife managers are limited in the inferences they can draw about low density populations. These limits are imposed by biases in monitoring data not regularly accounted for. 2. We developed a Bayesian hierarchical model to correct biases arising from imperfect detection and spatial autocorrelation. Our analysis incorporated model selection uncertainty by treating model probabilities as parameters to be estimated in the context of model ﬁtting. We ﬁtted our model to count data from a monitoring programme for the mountain plover Charadrius montanus, a low density bird species in Nebraska, USA. 3. Our results demonstrated that previous accounts of the abundance and distribution of plovers in Nebraska were impacted by low detection probabilities (5–20%). Uncorrected relative abundance estimates showed that the average number of birds per agricultural section increased over time, whereas corrected estimates showed that average abundance was stable. 4. Our method spatially interpolated relative abundance to produce distribution maps. These predictions suggested that birds were selecting some sites more frequently than others based on some habitat feature not explored in our study. Variation in mountain plover abundance appeared more heavily inﬂuenced by changes in the number of individuals occupying a few high quality sites, rather than from changes in abundance across many sites. Thus, conservation eﬀorts may not be as eﬃcient when focusing on low to moderate quality sites. 5. Synthesis and applications. Managers who must make decisions based on data-poor systems should adopt rigorous statistical approaches for drawing inferences. Spatial predictions provide information for deciding where to implement management, which is just as important as knowing what kind of management to apply. Our approach provides a step in the direction of making the biological signal in data-poor monitoring programmes more informative for conservation and management. Key-words: Bayesian hierarchical models, detection error, modelling uncertainty, spatial statistics

Introduction Wildlife managers are limited in the inferences they can draw from surveys of low density populations. These limits are partially caused by biases in monitoring data induced by detection errors and errors arising from spatial autocorrelation. Detection errors arise in surveys as a result of factors such as species-

*Correspondence author. E-mail: [email protected] †Present address: School of Forestry and Wildlife Sciences, Auburn University, 3203 Forestry and Wildlife Sciences Building, Auburn, AL, 36849, USA.

speciﬁc behaviour or diﬀering observer abilities (Royle 2004; Field, Tyre & Possingham 2005a; Johnson 2008). Accounting for variation in detectability has garnered much attention in the ecological literature because it erodes the power to estimate population parameters (Royle & Nichols 2003; Tyre et al. 2003; Field et al. 2005b). Failure to account for detectability also aﬀects our understanding of species’ distributions. This is important for agencies tasked with targeting management. Often, distributional data is derived from the location of positive detections in the context of monitoring (e.g. McConnell et al. 2009). However, methods such as point counts or presence–absence

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society

48 M. Post van der Burg et al. surveys do not cover all occupied sites which could make observed distributions biased. Rigorous distributional predictions can be derived from capturing spatial autocorrelation between observations and using estimates of spatial dependence to map predicted abundance or occupancy (e.g. Latimer et al. 2006; Thogmartin, Knutson & Sauer 2006). Few studies have dealt with both detection and spatial coverage biases in the context of estimating abundance (but see Royle et al. 2007). These combined errors could be of concern for low density populations for two reasons. First, low density populations are diﬃcult to ﬁnd and it would be helpful to know whether this low density is because of a biological process that warrants management action. Secondly, the non-random distribution of habitat is a likely source of spatial autocorrelation in count data (Legendre 1993). Non-random sampling of habitat patches will increase detections, but will also compromise the estimation of variability of abundance within a particular region. Therefore, accounting for the spatial non-independence of sampling sites and detection error is necessary to accurately predict a species’ abundance and distribution. We developed and applied a Bayesian hierarchical modelling approach that accounts for detection and spatial errors in count data. Our approach also deals with model selection uncertainty by treating model selection as a part of the ﬁtting procedure. We ﬁtted our model to data from a 3 year monitoring programme for the mountain plover Charadrius montanus, a low density bird species in Nebraska, USA. We had four main goals with this study. First, to compare detectioncorrected and uncorrected relative abundance estimates for mountain plovers. Secondly, to compare mountain plover abundance between two types of agricultural habitats: grazed lands and arable ﬁelds. Thirdly, to compare abundance between sites that were actively managed to improve the reproductive success of plovers versus those that were not. Lastly, to generate predictions of mountain plover distributions to inform management.

Materials and methods CASE STUDY: THE MOUNTAIN PLOVER

Mountain plovers are of conservation concern within North America (Knopf & Wunder 2006). The eastern edge of this species’ range extends into a small portion of Nebraska, USA, where previous surveys indicated that plovers rarely occupied the state (Clausen 1990; Dinsmore 1997). Bly, Snyder & VerCauteren (2008) conducted patch surveys in the southwestern corner of the Nebraska panhandle and concluded that mountain plovers were more numerous than previously believed. Thus, it is likely that mountain plovers in Nebraska had been under-sampled. This is problematic for the conservation of this population for at least two reasons. First, the population in Nebraska is at the edge of the species’ range and might be expected to experience wide ﬂuctuations. Therefore, distinguishing between measurement and process error is crucial for conservation. Secondly, identifying locations that have high conservation value will be compromised by biased abundance estimates. This could lead to ineﬃcient or misdirected allocation of conservation eﬀort.

Numerous studies have recognized that mountain plovers are diﬃcult to detect because they are cryptically coloured and often encountered in relatively low densities (Knopf & Wunder 2006). Past studies have accounted for detectability by using distance sampling (Wunder, Knopf & Pague 2003; Plumb, Knopf & Anderson 2005), markrecapture methods (Dinsmore, White & Knopf 2003), repeated visits (Dreitz, Lukacs & Knopf 2006; Tipton, Dreitz & Doherty 2008) and removal methods (McConnell et al. 2009; Tipton, Doherty & Dreitz 2009).

STUDY AREA

Data used in this study came from surveys described in Bly, Snyder & VerCauteren (2008). These surveys were conducted in the southwestern portion of the Nebraska panhandle (Kimball, Banner and Cheyenne counties) and covered approximately 4500 km2. Roughly 90% of the land in this region was privately owned and used for agricultural production (Hiller et al. 2009). Approximately 59% of the landscape was used for grazing livestock and the remaining 41% was used for growing crops. Both intensively grazed lands with bare ground and wheat ﬁelds provide mountain plovers with breeding habitat in Nebraska: relatively ﬂat areas with greater than 30% bare ground (Knopf & Miller 1994). Mountain plovers nest in agricultural ﬁelds, where their nests are exposed to tillage operations (Knopf & Rupert 1999; Shackford, Leslie & Harden 1999). In Nebraska, the Rocky Mountain Bird Observatory and Nebraska Game and Parks Commission had been applying a nest marking management programme in some of the same areas that the count data was recorded within. The nest marking programme focused on reducing nest loss due to agricultural tillage by marking nests so that soil tillage operations could avoid them.

SURVEY METHODOLOGY

Data were collected in the 2005, 2006 and 2007 breeding seasons. Counts were conducted within the primary land division unit, 2Æ56 km2 sections. Our analysis was based on 102 randomly selected sections in 2005 (43 previously occupied sections, 59 randomly selected sections), 111 sections in 2006 (73 previously occupied sections, 38 randomly selected sections), and 150 sections in 2007 (88 previously occupied sections, 62 randomly selected sections). In each section, surveyors aimed to maximize the number of detections by selecting a 4 ha patch that contained suitable nesting habitat. Scaling from the level of the patch to the agricultural section has the potential to inﬂuence our ability to accurately predict mountain plover abundance. For instance, if the patch contained a portion of the plovers in a section, scaling could underpredict average abundance. However, we were only interested in predicting relative abundance at the scale of the section so scaling issues were not as important for our analysis as they might be for ﬁner scale habitat selection analyses. In all three years, roughly 25% of the patches were located in rangeland sections and about 75% of the patches were located in arable sections. Surveyors visited each patch three times in 2005, four times in 2006 and three times in 2007. Surveys were conducted from mid-April to the beginning of June each year. Diﬀerent observers conducted the surveys in 2005 and 2006, but were the same in 2006 and 2007. Each visit consisted of two three-minute point counts: one prior to playing a territorial or alarm call and one post call playback. Call playback was standardized so that calls were played in the same direction and location during each visit. Surveys were conducted between sunrise and 10:00 or between 17:00 and sunset when it was not raining and winds were <32 km h)1.

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 47–55

Making better sense of monitoring data 49 STATISTICAL ANALYSIS

We performed all analyses using the statistical computing language R (R Development Core Team 2008). We used a hierarchical Bayesian approach because it accommodated multi-level processes and included prior information. Our approach was based on the work of Royle, Link & Sauer (2002) and Royle et al. (2007). Our model of the count process followed Royle (2004): ! ! I J Y Y pðyij jki ; pÞ ¼ Binðyij ; Ni ; pÞ PoisðNi ; ki Þ eqn 1 i¼1

j¼1

where yij were observed counts arising from a binomial process at patch i and visit j, Ni was the unobserved patch-speciﬁc abundance and p was the rate at which individuals were detected. Notice that including N in the binomial model says that there was some true number of individuals, but that observers could only ﬁnd some proportion of them. We assumed that N was distributed as a Poisson process where ki was the patch speciﬁc Poisson mean (i.e. detection-corrected average relative abundance). We assumed that N had a constant uniform prior distribution. We believe our method corrected our observations for detection error based on simulations (not presented). Others have found that similar methods may not always yield fully corrected estimates of ‘true’ abundance (Eﬀord & Dawson 2009). Thus, to be conservative, we refer to our estimates as corrected relative abundance. We modelled detection rate p as a function of temporal and observer-speciﬁc covariates using a logistic model. We assumed that p did not vary across sites. This assumption was reasonable because our sites did not contain factors, such as vegetation, that would make detection vary. We modelled the patch-speciﬁc mean abundance u(s) using an overdispersed log linear model: uðsÞ ¼ loge ðkðsÞÞ ¼ lðsÞ þ zðsÞ þ eðsÞ P

eqn 2

where lðsÞ ¼ b0 þ b1 x1 þ . . . þbm xm was the sum of spatially indexed covariates, z(s) was a random eﬀect representing spatially autocorrelated error, and e(s) was a random eﬀect representing uncorrelated residual error. We assumed a normal prior with a mean of zero and variance of 10 on the detection covariates and a constant prior (i.e. equal to 1) on the abundance covariates. Each of the sets of covariate parameters was drawn from a set of candidate models with a uniform prior (1 ⁄ number of models). By updating the model parameters conditionally on the chosen model, we model averaged our posterior estimates and simultaneously estimated the probability of the model. We used the term ‘model averaging’ (sensu Burnham & Anderson 2002) to refer to the process of estimating model parameters conditional on model probabilities. The frequency with which each model was drawn can then be used to calculate Bayes’ factors, which we used to calculate posterior model probabilities (Link & Barker 2006). Our model structure allowed us to include the eﬀects of spatial autocorrelation in terms of a departure from the systematic mean l. The z(s) term was modelled as a multivariate normal distribution: z MVNð0; r2z KÞ, where r2z was the spatial variance and K was a correlation function that speciﬁed how correlated the error terms were. We used an exponential correlation model: K ¼ ejdi dx j=h , where h represents the degree of spatial dependence in metres (i.e. range parameter in geostatistics). We assumed a uniform prior on h (U(0,10 000)). We chose this correlation function because there is little theoretical justiﬁcation for choosing complicated multiparameter correlation functions (Royle, Link & Sauer 2002). The uncorre-

lated error term represents the small scale variation in the data and was included to help account for overdispersion in our observations. We modelled uncorrelated error as e Normalð0; r2e Þ, where r2e was the residual variance. In our model, both r2z and r2e were parameterized as sz ¼ 1=r2z andse ¼ 1=r2e , known as Bayesian precision parameters. We assumed inverse gamma priors with a mean of one and a variance of 10 on these parameters. We used Markov chain Monte Carlo (MCMC) simulation to ﬁt our models (Gilks, Richardson & Spiegelhalter 1995). Our algorithm was a Metropolized Gibbs sampler because it combined Gibbs sampling when the full conditional distribution was available and employed Metropolis–Hastings (M–H) when the full conditional was not available, as when updating based on the mixture likelihood (1) (Gilks, Richardson & Spiegelhalter 1995). We hierarchically centred our continuous covariates by subtracting the mean of each covariate from the data and dividing by the standard deviation in order to promote better mixing. We made additional improvements in mixing by reparameterizing the systematic mean and spatial random eﬀect as: z MVNðl; r2z KÞand u MVNðz; r2z KÞ(Gelfand, Sahu & Carlin 1995; Royle et al. 2007). For each iteration, we estimated the model parameters and then made spatial predictions by making draws from the full conditional distribution (Appendix S1–S4, Supporting information). We analysed data for each year separately, with and without prior information on detectability. We assumed uninformative priors on all detection parameters for the 2005 data. We used the posteriors estimated from 2005 as priors on the 2006 analysis, and we used 2006 posteriors as priors on the 2007 analysis. Prior information did slightly inﬂuence our posterior detection parameter estimates and also reduced the posterior variance of those estimates, but did not inﬂuence our posterior abundance estimates. Our results were based on models including prior information. We ran the models for 110 000 iterations, using three chains per model. We used more than one chain to ensure that we converged on the same answer each time we ran the algorithm. We discarded the ﬁrst 10 000 iterations as a burn-in period. We made our inferences from the last 100 000 iterations, and to further reduce serial autocorrelation we thinned the chains using every 100th iteration. One chain took approximately 5–7 h to run on a computer with a 2Æ66 GHz processor. This long running time was largely due to the multivariate normal formulation of the spatial process, which is generally regarded as being more ﬂexible than faster running conditional autoregressive models. Due to the long periods of time needed to run each model we adopted a parallel processing approach.

CANDIDATE MODELS

We considered four abundance models. Our ﬁrst model assumed that average abundance was similar across patches (Null). Our second model contained an eﬀect of the linear distance between the centre of the patch and the nearest road (Distance). We considered this because mountain plovers may select or avoid sites that experience more anthropogenic disturbance. Our third model included a binary variable (Grazed) for whether the site was in an arable ﬁeld (0) or in grazed land (1). We considered these eﬀects because mountain plover abundance could be higher in patches that were more similar to native grasslands (i.e. pastures and grazed lands) compared to arable ﬁelds. Our fourth model included a binary variable that described whether nest protection measures were applied in the section (1) or not (0). We considered this because mountain plovers might prefer sites with ongoing management. We did not consider models without the spatial random eﬀects because we

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 47–55

We detected a total of 18 birds in 2005, 26 birds in 2006 and 86 birds in 2007. Thus, our surveys yielded low observed abundances of mountain plovers across years, but there was an increase in mean naı¨ ve abundances over time (Fig. 1a). Our results suggested that this increase was probably a result of increasing detectability across years (Fig. 1a). In our model selection exercise, we found that the detection error model with the highest posterior probability in 2005 included an eﬀect of observer and call, whereas the model with the highest posterior in 2006 and 2007 contained call and time eﬀects (Table 1). Our posterior parameter estimates suggested that playing a call had the strongest eﬀect on detection error (Table 2). In general, playing a call increased detectability within each year and appeared to have caused the most dramatic increase in 2007 (Fig. 2). We found only weak eﬀects of observer and time on detection error (Table 2). Model selection results showed a high degree of uncertainty for models of abundance (Table 1). We found that average relative abundance seemed to have remained fairly constant through time (Fig. 1a). In terms of parameter eﬀects, we found no eﬀect of distance from road on mountain plover relative abundance (Table 2). We did, however, ﬁnd a slight increase in relative abundance for patches located in grazed land compared to arable ﬁelds, and this eﬀect was most pronounced in 2006 (Fig. 1b). We found virtually no correlation between nest management and observed abundance (Table 2). Our estimates of spatial autocorrelation in abundance among patches varied between years with the range parameter

1·0 0·8 0·6 0·4

Mean relative abundance

0·2 0·0

Arable

1·5

Grazed

1·0

Mean relative abundance

2007

(b)

2·0

2·5

2006

0·5

Results

(a)

2005

0·0

wanted to make spatially explicit predictions of mountain plover abundance. We considered three detection models. All three models contained an intercept and an eﬀect of call playback (Call). We included this parameter in all the models because visual inspection of the data showed higher maximum counts after the call was played. Thus, our ﬁrst model contained only an eﬀect of call. Our second model added an observer eﬀect (Observer). For these surveys there were only two observers in each year. We built our third model to include the eﬀects of call, a quadratic eﬀect of the time of day (Time + Time2) and a quadratic eﬀect of the ordinal day in the breeding season (Day + Day2). We presented all of our parameter estimates as posterior means and standard errors. These values were calculated by computing the mean and standard error of each parameter from values in the Markov chains. We then made predictions using our posterior parameter estimates. We presented these predictions as means with 95% Bayesian conﬁdence intervals (BCI). Our spatial mountain plover predictions were made using the centre coordinates of 1720 sections within the survey area. For each iteration of the algorithm we drew 1720 predicted spatial random eﬀects for each of the sections from the posterior predictive distribution (Royle, Link & Sauer 2002). We made spatial abundance predictions by combining these spatial random eﬀects with the estimated model parameters for landuse. Within each section we speciﬁed whether the dominant form of landuse was arable land or grazed land using 2005 landcover data (obtained from Nebraska Department of Natural Resources) in ArcGIS 9.2 (ESRI 2008).

1·2

50 M. Post van der Burg et al.

2005

2006

2007

Year Fig. 1. (a) Mountain plover relative abundance estimates from western Nebraska during the breeding seasons of 2005–2007. Closed triangles represent mean naı¨ ve abundance estimates (with 95% BCI) based on the maximum number of observed individuals. Closed circles represent posterior detection corrected model estimates of mean relative abundance (with 95% BCI). (b) Bars represent posterior mean model abundance estimates (with 95% BCI) for arable and grazed patches over the course of 3 years.

(h) being similar in 2005 and 2007, but lower in 2006 (Table 2). This parameter was diﬃcult to interpret, but generally it is thought of as representing the distance at which the correlation between points weakens by a factor of 0Æ37 (i.e. e)1 = 0Æ37; Isaaks & Srivastava 1989). Across all 3 years, we found variation in mountain plover abundance could be attributed to variability between locations (ss) compared to variation within each location (sg). Recall, that these were precision parameters and must be inverse transformed. Our spatially explicit predictions of abundance showed a patchy distribution of mountain plovers across our study area (Fig. 3). This distributional pattern also changed between years with 2006 showing a weaker pattern compared with 2005 and 2007. Between years the spatial distribution shifted from a western to a southern

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 47–55

Making better sense of monitoring data 51

0Æ02 0Æ02 0Æ96

0Æ00 0Æ00 1Æ00

The abundance models included a null eﬀect (Null), an eﬀect of distance from road (Distance), the presence (Nests = 1) or absence (Nests = 0) of marked nests, and whether abundance varied between arable crop ﬁelds (Grazed = 0) and grazed land (Grazed = 1). The detection model set included a model with the eﬀect of call (Call), an eﬀect of call and observer (Observer) and quadratic eﬀects of time of day (Time2) and ordinal day of the breeding season (Day2). Model probabilities were calculated using Bayes’ factors.

2007

log(k) )0Æ03 (0Æ07) )0Æ06 (0Æ08) 0Æ00 (0Æ07) Distance 0Æ00 (0Æ03) 0Æ00 (0Æ03) 0Æ00 (0Æ03) Grazing 0Æ03 (0Æ13) 0Æ15 (0Æ30) 0Æ04 (0Æ10) Nests 0Æ01 (0Æ07) )0Æ01 (0Æ07) )0Æ06 (0Æ13) logit(p) )4Æ65 (0Æ56) )4Æ25 (0Æ31) )3Æ14 (0Æ34) Call 1Æ35 (0Æ41) 1Æ37 (0Æ27) 1Æ38 (0Æ15) Observer 0Æ64 (0Æ58) 0Æ00 (0Æ03) 0Æ00 (0Æ01) Time )0Æ30 (0Æ59) 0Æ01 (0Æ36) )0Æ42 (0Æ35) Time2 0Æ99 (1Æ77) 1Æ61 (1Æ11) 2Æ09 (1Æ63) Day )0Æ08 (0Æ15) )0Æ29 (0Æ12) )0Æ32 (0Æ23) Day2 0Æ01 (0Æ10) 0Æ06 (0Æ08) 0Æ09 (0Æ08) h 5360Æ86 (2790Æ35) 2893Æ75 (2586Æ74) 4196Æ95 (2277Æ51) ss 4Æ03 (3Æ03) 3Æ78 (2Æ40) 1Æ86 (0Æ91) sg 7Æ35 (4Æ52) 6Æ31 (4Æ20) 7Æ01 (4Æ61) The covariates for abundance were the eﬀect of distance from road (Distance), landuse: arable ﬁeld (Grazed = 0), grazed land (Grazed = 1) and whether managed nests were absent (Nests = 0) or present (Nests = 1). The detection covariates were the eﬀect of call (Call), eﬀect of observer (Observer), as well as a main and quadratic eﬀect of both time and day. Additional model parameters included the range of spatial autocorrelation (h), spatial precision (ss) and residual precision (sg).

distribution. Assuming we made predictions within the region of Nebraska where mountain plovers are likely to be found, we estimated the total abundance of plovers as 1650 (95% BCI: 400–6681); 1617 (95% BCI: 367–6966); and 1568 (95% BCI: 277–8681) for the 3 years, respectively.

Discussion The primary uses of monitoring data in species conservation should be to answer scientiﬁc questions or to assess the eﬃcacy

0·00

2006

of

0·30

2005

2006

2007

0·20

Table 2. Posterior parameter estimates (standard errors) abundance (k) and detection probability (p) on the linear scale

0·10

0Æ07 0Æ63 0Æ30

0·00

Detection models pCall pCall + Observer pCall + Time2 + Day2

0·30

0Æ24 0Æ27 0Æ27 0Æ22

0·20

0Æ25 0Æ27 0Æ25 0Æ23

Call

0·10

0Æ26 0Æ26 0Æ24 0Æ25

Detection probability

Abundance models kNull kDistance kNests kGrazed

No call

0·20

2007

2005

0·10

2006

0·00

2005

0·30

Table 1. Model selection table for linear models explaining variation in abundance and detection probability of mountain plovers in Western Nebraska for the 2005–2007 breeding seasons

1

2

Observer Fig. 2. Detection probability estimates for mountain plover surveys during the 2005–2007 breeding seasons. Bars represent the posterior mean estimates of detection probability between two observers and before and after alarm calls were played. Error bars represent 95% BCI.

of certain management practices (Nichols & Williams 2006). Inferences based on naı¨ ve estimates of abundance are likely to impact which steps are taken in terms of interpreting results and applying management action. For example, had we implemented a management practice to improve breeding habitat within agricultural sections in 2005, we would have drawn the wrong conclusion about the impact of management on the average number of birds expected in each section. Likewise, had we not accounted for spatial error we might not have detected changes in sections under management or identiﬁed portions of the landscape that warranted further attention.

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 47–55

52 M. Post van der Burg et al.

5HODWLYHDEXQGDQFH HVWLPDWHV

Fig. 3. Spatially explicit posterior mean predictions of mountain plover relative abundance in the panhandle of Nebraska for the 2005–2007 breeding seasons. Open circles represent count locations where no birds were detected. Triangles represent locations where one or more birds were detected. Grid cells represent 2Æ56 km2 agricultural sections. Map scale is in metres.

Rather, our habitat eﬀects would have told us that abundance was largely similar across patches and that the expected number of birds within a section was fairly uniform across the landscape. Therefore, in order to make progress in terms of creating new management plans for this species (or for any low density species), a rigorous modelling approach is necessary to make better sense of the data.

MOUNTAIN PLOVER ABUNDANCE

Our estimates of detection rate were, in general, lower than those found elsewhere: 0Æ94–0Æ65 in Oklahoma, USA (McConnell et al. 2009), 0Æ38 in Colorado, USA (Wunder, Knopf & Pague 2003). The overall pattern in our results mirrored those of other surveys designed to increase the number of detections within sampling locations (e.g. McConnell et al. 2009). Other studies of this species have shown that increasing the number of visits to the same site does reduce detection error (Dreitz, Lukacs & Knopf 2006). While such strategies can reduce variation in detection error, this reduction usually comes at the cost of increasing variation in abundance estimates (based on simulations, not shown). The amount of eﬀort invested in revisits could also explain why we found a very weak eﬀect of time on detection error. Dreitz, Lukacs & Knopf (2006) sampled some

sites as many as 12 times. In our study, eﬀort remained ﬁxed at three to four visits because of personnel and ﬁnancial constraints. Tipton, Dreitz & Doherty (2008) found similar results for this species using ﬁxed and comparatively low amounts of sampling eﬀort. Thus, studies designed to monitor low density species should weigh the statistical trade-oﬀ between the number of sites and visits, as well as the ﬁnancial trade-oﬀs between costs associated with increased eﬀort and costs of making decisions based on incorrect inferences (Field et al. 2004; Field, Tyre & Possingham 2005a). Within the limits of our data, our model-based estimates suggested that mountain plover relative abundance was higher in Nebraska than previously suggested in Bly, Snyder & VerCauteren (2008). Furthermore, past claims about the size of this population in Nebraska were probably confounded by chronic undersampling. It would appear that mountain plover abundance did not respond to the types of agricultural landuse we considered. Our estimated eﬀect of agricultural landuse was similar to that found elsewhere (Dreitz, Lukacs & Knopf 2006; Tipton, Dreitz & Doherty 2008; Tipton, Doherty & Dreitz 2009). There was also a great deal of variability in terms of mean relative abundance across the mountain plover’s range. Our estimates of relative abundance on a per hectare basis were 0Æ20–0Æ30 birds ha–1. Others have found abundances as high as 1Æ56 birds ha–1 (assuming 1Æ6 ha patches) in Colorado (Dreitz,

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 47–55

Making better sense of monitoring data 53 Lukacs & Knopf 2006) and as low as 0Æ003 birds ha–1 in Oklahoma (McConnell et al. 2009). One possible explanation for this heterogeneity could be related to the methods of sampling and detection correction applied across the various surveys. Our study and that of Dreitz, Lukacs & Knopf (2006) both used a multiple revisit method of sampling and similar statistical methods. Plumb, Knopf & Anderson (2005) and Wunder, Knopf & Pague (2003) used distance sampling which requires a diﬀerent sampling strategy and diﬀerent statistical approach. Finally, McConnell et al. (2009) used a point count sampling approach and post hoc detection correction method, and Tipton, Doherty & Dreitz (2009) used a probability based sampling approach and a design based method of deriving abundance. Therefore, comparing multiple studies suggests that we may not clearly know whether abundance diﬀerences are attributable to a biological signal or whether these diﬀerences are attributable to variation in ﬁeld and statistical methods.

SPATIAL PREDICTIONS

Without spatially explicit predictions we would have to estimate total mountain plover abundance in Nebraska by assuming that our model could be used to predict abundance at unobserved locations (i.e. extrapolation), which could lead to unreliable estimates. Instead, we estimated the spatial structure of abundance using the data and constructed predictions at unobserved locations between observed locations (i.e. interpolation). In many of the studies previously mentioned, total population estimates were arrived at using extrapolated estimates of density. However, our total population estimates for Nebraska, based on interpolated values, are less than those for Colorado (Wunder, Knopf & Pague 2003; Tipton, Doherty & Dreitz 2009) and Montana (Dinsmore, White & Knopf 2003), and are generally more uncertain. This was to be expected because ignoring spatial autocorrelation can result in biased parameter estimates and misleading inferences (Legendre 1993; Dormann 2007; Beale et al. 2010). Our study was unable to tease apart the broader reasons for the spatial structuring in our data. Spatial structuring could be driven by either extrinsic (e.g. environmental or geomorphic) factors or intrinsic (e.g. behavioural or phenotypic) factors. Because habitat variables did not adequately account for the variability in our data it is likely that much of the intrinsic and extrinsic structuring was swept into the spatial autocorrelation term. Therefore, it could be that the temporal variation in our spatial predictions was caused by interacting static and dynamic spatial processes. Additionally, including a spatial error term, as we did, does not necessarily mean that we accounted for all of the spatial variation in our data. For instance, Wintle & Bardos (2006) show that including a spatial autocorrelation term in a model reduces residual spatial error in data with intrinsic structuring, but not completely. Thus, it is possible that there may be lingering spatial variation in our data that we did not account for. The way to solve this would be to include additional linear and non-linear spatially indexed covariates in our abundance models and utilize the Bayesian

model averaging approach to compare the performance of those additional predictors. Despite this, we can view our relative spatial predictions as measures of habitat quality if we assume that the number of birds in a section is proportional to the number and quality of habitat patches in a section (Fretwell & Lucas 1970). Treating our corrected mean estimate of relative abundance as a Poisson random variable representing birds per section (our unit of prediction), we could expect to ﬁnd one to two mountain plovers in each section, but could occasionally ﬁnd a maximum of ﬁve. When we consider the spatial heterogeneity in mountain plover habitat use, as measured by our map, we might expect to see as many as ten birds in some sections and almost none in others. This is interesting biologically, because our mean abundance estimate suggests that, in the absence of a spatial process, mountain plovers should be spacing themselves so that their densities are fairly low (around 1–2 birds per section). This would only work if there were one or two high quality patches in each section. If we drew inferences from only this value we might have expected that an increase in the abundance of mountain plovers would lead to more sections being occupied by individuals. In terms of management, this would be an important aspect of mountain plover biology to understand because it should inform the scale at which management is applied. In the case of mountain plovers, this would suggest that a larger area was necessary to increase the population. However, this is not what our results indicate. Our spatial predictions show that the number of mountain plovers in some sections increased. We might expect this result if something were improving the number of high quality patches within each section. One potential explanation for this within-section increase could be that mountain plovers were selecting nesting sites in response to nest success (Greenwood 1980). Our modelling results indicated that nest management had little eﬀect on our estimates of relative abundance. This is likely to be due to the fact that few of the sites where nests were marked corresponded to survey locations. However, the survey locations with the highest predicted numbers of mountain plovers in our 2007 map overlap the regions where the highest densities of nests were found between 2004 and 2007 (B. Bly, unpublished data). If we compare the average predicted abundances between sections that contained marked nests (1Æ51) and those that did not (0Æ89) we see that plover abundance is slightly higher in the managed sections. Bly, Snyder & VerCauteren (2008) also found that the observed number of mountain plover nests in these sections had increased from 49 nests in 2005 to 112 nests by 2007. Average nest success rates in this region are high (75%) and similar between years (B. Bly, M. Post van der Burg, A. Tyre, L. Snyder, J. Jorgensen & T. Vercauteren, unpublished data). However, we are still unsure whether these sections contain some as yet unobserved habitat characteristic that could be attracting breeding plovers, thus giving the impression that the increase in abundance is due to management eﬀorts. Knowing this information would be particularly useful in a largely homogeneous landscape like our study area. Our

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 47–55

54 M. Post van der Burg et al. analysis suggested that small-scale conservation programmes, which tend to be more expensive in terms of money and labour, might be more beneﬁcial for some subsets of the continental population of mountain plovers. This might be particularly true when preferred habitats, such as prairie dog Cynomys ludovicianus colonies, are nearly absent from the landscape. The population level beneﬁt of management could also vary with regional shifts in preferred tillage techniques. Variation in the value of management strategies across a species’ range raises questions of where to put conservation eﬀort, rather than how to implement it across the landscape. Conservation practices certainly accrue greater beneﬁt when they are coordinated at larger scales (Kark et al. 2009). The types of conservation practices for migratory species like mountain plovers appear fragmented from one region to another. Federal and regional agencies can make recommendations, but have no regulatory authority to compel state agencies or funding bodies to direct funds to species where the greatest local beneﬁt would accrue. Our study does not provide ‘rules-of-thumb’ for low density population management, but it does provide a modelling framework to study spatial variation that can be used to target management across the landscape. Considering the variability in approaches and methods for surveys and analysis of population trends, we suggest that a more synthetic study of these methodologies is warranted.

Acknowledgements We would like to thank the Nebraska Game and Parks Commission (NGPC) for ﬁnancial support. Our research was funded with State Wildlife Grant (T-47), Nebraska Environmental Trust Grant (05-182), and NGPC Nongame and Endangered Species Conservation Funds. The UNL Research Computing Facility provided the computational resources for ﬁtting the models. We especially would like to thank Joel Jorgenson, Matthew Giovanni, Wayne Thogmartin, Carsten Dormann, the editor and one anonymous reviewer for helpful comments on this manuscript. This research would not have been possible without the help of Larry Snyder and his knowledge of local landowners. Likewise, without those landowners granting us access to their property, our observations would have been more limited. We would also like to acknowledge the previous work that Cris Carnine and Courtney Kerns did to establish mountain plover surveys in Nebraska. Finally, we thank the technicians for their work in the ﬁeld: Travis Wooten, Cameron Shelton and Brian Monser.

References Beale, C.M., Lennon, J.J., Yearsley, J.M., Brewer, M.J. & Elston, D.A. (2010) Regression analysis of spatial data. Ecology Letters, 13, 246–264. Bly, B.L., Snyder, L. & VerCauteren, T. (2008) Migration chronology, nesting ecology, and breeding distribution of Mountain Plover (Charadrius montanus) in Nebraska. Nebraska Bird Review, 76, 120–128. Burnham, K.P. & Anderson, D.R. (2002) Model Selection and Multimodel Inference: A Practical Information-Theoretic Approach. Springer, New York, USA. Clausen, M.K. (1990) Mountain Plover sighted in Kimball County. Nebraska Bird Review, 58, 98–99. Dinsmore, S.J. (1997) Mountain Plover nest in Kimball County. Nebraska Bird Review, 65, 131. Dinsmore, S.J., White, G.C. & Knopf, F.L. (2003) Annual survival and population estimates of mountain plovers in southern Philips County, Montana. Ecological Applications, 13, 1013–1026. Dormann, C.F. (2007) Eﬀects of incorporating spatial autocorrelation into the analysis of species distribution data. Global Ecology and Biogeography, 16, 129–138.

Dreitz, V.J., Lukacs, P.M. & Knopf, F.L. (2006) Monitoring low density avian populations: an example using Mountain Plovers. Condor, 108, 700–706. Eﬀord, M.G. & Dawson, D.K. (2009) Eﬀect of distance-related heterogeneity on population size estimates from point counts. Auk, 126, 100–111. ESRI (2008) ArcGIS: Release 9.3. Environmental Systems Research Institute, Redlands, CA. Field, S.A., Tyre, A.J. & Possingham, H.P. (2005a) Optimizing allocation of monitoring eﬀort under economic and observational constraints. Journal of Wildlife Management, 69, 473–482. Field, S.A., Tyre, A.J., Jonzen, N., Rhodes, J.R. & Possingham, H.P. (2004) Minimizing the cost of environmental management decisions by optimizing statistical thresholds. Ecology Letters, 7, 669–675. Field, S.A., Tyre, A.J., Thorn, K.H., O’Connor, P.J. & Possingham, H.P. (2005b) Improving the eﬃciency of wildlife monitoring by estimating detectability: a case study of foxes (Vulpes vulpes) on the Eyre Peninsula, South Australia. Wildlife Research, 32, 253–258. Fretwell, S.D. & Lucas, H.L. (1970) On territorial behavior and other factors inﬂuencing habitat distribution in birds. I. Theoretical development. Acta Biotheoretica, 19, 16–36. Gelfand, A.E., Sahu, S.K. & Carlin, B.P. (1995) Eﬃcient parametrizations for normal linear mixed models. Biometrika, 82, 479–488. Gilks, W., Richardson, S. & Spiegelhalter, D. (1995) Markov Chain Monte Carlo in Practice. Chapman and Hall, London. Greenwood, P.J. (1980) Mating systems, philopatry and dispersal in birds and mammals. Animal Behavior, 28, 1140–1162. Hiller, T.L., Powell, L.A., McCoy, T.D. & Lusk, J.J. (2009) Long-term agricultural land-use trends in Nebraska, 1866–2007. Great Plains Research, 19, 225–237. Isaaks, E.H. & Srivastava, R.M. (1989) An Introduction to Applied Geostatistics. Oxford University Press, New York, USA. Johnson, D.H. (2008) In defense of indices: the case of bird surveys. Journal of Wildlife Management, 72, 857–868. Kark, S., Levin, N., Grantham, H.S. & Possingham, H.P. (2009) Betweencountry collaboration and consideration of costs increase conservation planning eﬃciency in the Mediterranean Basin. Proceedings of the National Academy of Sciences of the United States of America, 106, 15360–15365. Knopf, F.L. & Miller, B.J. (1994) Charadrius montanus: montane, grassland, or bare-ground plover? Auk, 111, 504–506. Knopf, F.L. & Rupert, J.R. (1999) Use of cultivated ﬁelds by breeding mountain plovers in Colorado. Studies in Avian Biology, 19, 81–86. Knopf, F.L. & Wunder, M.B. (2006) Mountain Plover (Charadrius montanus). The Birds of North America Online (ed. A. Poole). Cornell Lab of Ornithology, Ithaca; Retrieved from the Birds of North America Online: http:// bna.birds.cornell.edu/bna/species/211 Latimer, A.M., Wu, S.S., Gelfand, A.E. & Silander, J.A. (2006) Building statistical models to analyze species distributions. Ecological Applications, 16, 33–50. Legendre, P. (1993) Spatial autocorrelation – trouble or new paradigm. Ecology, 74, 1659–1673. Link, W.A. & Barker, R.J. (2006) Model weights and the foundations of multimodel inference. Ecology, 87, 2626–2635. McConnell, S., O’Connell, T.J., Leslie, D.M. & Shackford, J.S. (2009) Mountain Plovers in Oklahoma: distribution, abundance, and habitat use. Journal of Field Ornithology, 80, 27–34. Nichols, J.D. & Williams, B.K. (2006) Monitoring for conservation. Trends in Ecology and Evolution, 21, 668–673. Plumb, R.E., Knopf, F.L. & Anderson, S.H. (2005) Minimum population size of mountain plovers breeding in Wyoming. Wilson Bulletin, 117, 15–22. R Development Core Team (2008) R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing, Vienna, Austria. Royle, J.A. (2004) N-mixture models for estimating population size from spatially replicated counts. Biometrics, 60, 108–115. Royle, J.A., Link, W.A. & Sauer, J.R. (2002) Statistical Mapping of Count Survey Data. Predicting Species Occurrences: Issues of Accuracy and Scale. Island Press, Washington, D.C. Royle, J.A. & Nichols, J.D. (2003) Estimating abundance from repeated presence–absence data or point counts. Ecology, 84, 777–790. Royle, J.A., Kery, M., Gautier, R. & Schmid, H. (2007) Hierarchical spatial models of abundance and occurrence from imperfect survey data. Ecological Monographs, 77, 465–481. Shackford, J.S., Leslie, D.M. & Harden, W.D. (1999) Range-wide use of cultivated ﬁelds by Mountain Plovers during the breeding season. Journal of Field Ornithology, 70, 114–120.

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 47–55

Making better sense of monitoring data 55 Thogmartin, W.E., Knutson, M.G. & Sauer, J.R. (2006) Predicting regional abundance of rare grassland birds with a hierarchical spatial count model. Condor, 108, 25–46. Tipton, H.C., Doherty, P.F. & Dreitz, V.J. (2009) Abundance and density of mountain plover (Charadrius montanus) and burrowing owl (Athene cunicularia) in eastern Colorado. Auk, 126, 493–499. Tipton, H.C., Dreitz, V.J. & Doherty, P.F. (2008) Occupancy of mountain plover and burrowing owl in Colorado. Journal of Wildlife Management, 72, 1001–1006. Tyre, A.J., Tenhumberg, B., Field, S.A., Niejalke, D., Parris, K. & Possingham, H.P. (2003) Improving precision and reducing bias in biological surveys: estimating false-negative error rates. Ecological Applications, 13, 1790–1801. Wintle, B.A. & Bardos, D.C. (2006) Modeling species–habitat relationships with spatially autocorrelated observation data. Ecological Applications, 16, 1945–1958. Wunder, M.B., Knopf, F.L. & Pague, C.A. (2003) The high-elevation population of Mountain Plovers in Colorado. Condor, 105, 654–662. Received 1 February 2010; accepted 20 October 2010 Handling Editor: Brendan Wintle

Appendix S1. R code for MCMC algorithm for analysis of spatial data. Appendix S2. R script ﬁle for proposal and likelihood functions used in Appendix S1. Appendix S3. Comma delimited Microsoft Excel ﬁle containing data for analysis. Appendix S4. Comma delimited Microsoft Excel ﬁle with data used in spatial interpolation. As a service to our authors and readers, this journal provides supporting information supplied by the authors. Such materials may be re-organized for online delivery, but are not copy-edited or typeset. Technical support issues arising from supporting information (other than missing ﬁles) should be addressed to the authors.

Supporting Information Additional Supporting Information may be found in the online version of this article.

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 47–55

Journal of Applied Ecology 2011, 48, 14–24

doi: 10.1111/j.1365-2664.2010.01901.x

Can the abundance of tigers be assessed from their signs? Yadvendradev Jhala1*, Qamar Qureshi1 and Rajesh Gopal2 1

Wildlife Institute of India, PO Box 18, Dehradun 248001, India; and 2National Tiger Conservation Authority, Bikeneer House, New Delhi, India

Summary 1. Indices of abundance oﬀer cost eﬀective and rapid methods for estimating abundance of endangered species across large landscapes, yet their wide usage is controversial due to their potential of being biased. Here, we assess the utility of indices for the daunting task of estimating the abundance of the endangered tiger at landscape scales. 2. We use double sampling to estimate two indices of tiger abundance (encounters of pugmarks and scats per km searched) and calibrate those indices against contemporaneous estimates of tiger densities obtained using camera-trap mark–recapture (CTMR) at 21 sites (5185 km2) in Central and North India. We use simple and multiple weighted regressions to evaluate relationships between tiger density and indices. A model for estimating tiger density from indices was validated by Jackknife analysis and precision was assessed by correlating predicted tiger density with CTMR density. We conduct power analysis to estimate the ability of CTMR and of indices to detect changes in tiger density. 3. Tiger densities ranged between 0Æ25 and 19 tigers 100 km)2 were estimated with an average coefﬁcient of variation of 13Æ2(SE 2Æ5)%. Tiger pugmark encounter rates explained 84% of the observed variability in tiger densities. After removal of an outlier (Corbett), square root transformed scat encounter rates explained 82% of the variation in tiger densities. 4. A model including pugmark and scat encounters explained 95% of the variation in tiger densities with good predictive ability (PRESS R2 = 0Æ99). Overall, CTMR could detect tiger density changes of >12% with 80% power at a = 0Æ3, while the index based model had 50% to 85% power to detect >30% declines. The power of indices to detect declines increased at high tiger densities. 5. Synthesis and applications. Indices of tiger abundance obtained from across varied habitats and a range of tiger densities could reliably estimate tiger abundance. Financial and temporal costs of estimating indices were 7% and 34% respectively, of those for CTMR. The models and methods presented herein have application in evaluation of the abundance of cryptic carnivores at landscape scales and form part of the protocol used by the Indian Government for evaluating the status of tigers. Key-words: camera trap, double sampling, indices of abundance, mark–recapture, Panthera tigris, power analysis, regression models

Introduction Information on abundance and change in abundance is important for the eﬀective management of endangered species (Gibbs, Snell & Causton 1999). Assessing the abundance of low density, wide ranging and cryptic species is extremely demanding in terms of time and resources (Garshelis 1992). In the absence of abundance information, conservation *Correspondence author. E-mail: [email protected]

management decisions are often based on crude estimates, expert opinion or educated guesses, which may result in erroneous decisions that can be counterproductive for conservation (Blake & Hedges 2004). Predictive models based on indices of abundance oﬀer an economical, practical and timely solution to this problem (Hutto & Young 2003; Conn, Bailey & Saeur 2004; Johnson 2008). An index of abundance is deﬁned as any measurable correlative of density (Caughley 1977) typically estimated without a measure of detection rate (Conroy & Carroll 2009). Use of indices as surrogates of abundance has been

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society

Tiger abundance from signs 15 criticized as most indices are rarely calibrated with density (Pollock et al. 2002; Williams, Nichols & Conroy 2002; Skalski, Ryding & Millspaugh 2005), or tested for precision in detecting population change (MacKenzie & Kendall 2002; Conn, Bailey & Saeur 2004). This latter aspect of population estimates, i.e. ability to detect change in abundance is vital for monitoring trends, essential information for adaptive management and for evaluating success of conservation programmes (Williams, Nichols & Conroy 2002; Barlow et al. 2008). The key in making an index useful is to link the observed numbers in the index to true abundance or density (Conroy & Carroll 2009). Probably the best and most cost eﬀective approach is to use double sampling (Cochran 1977) where a subgroup of the sample sites is subject to both the index and quantitative estimator and then the relationship between them determined. The added advantage of double sampling is that it can directly address the issue of incomplete detection in an index (a potentially biased estimator) since it is calibrated against an unbiased accurate estimate of abundance (Conroy & Carroll 2009). The world is witnessing the highest concern society has ever shown towards conservation of large carnivores and their ecosystems (Mech 1996). Yet, the numbers and range of most large carnivores continue to decline (Check 2006; Dinerstein et al. 2007). Due to the resource intensive nature of the techniques used for estimating large carnivore abundances, those techniques are rarely applied to large landscapes (but see Hayward et al. 2002; Barlow et al. 2008). In the case of the tiger Panthera tigris (Linnaeus 1758) that occupies wide inaccessible landscapes, obtaining reliable abundance estimates over much of its range is a daunting task (Karanth et al. 2003; Sanderson et al. 2006). Examples of tiger density estimates obtained using resource intensive camera trap mark–recapture (CTMR) in tiger occupied landscapes of India, Nepal, Bhutan and South East Asia include Karanth & Nichols (1998); O’Brien, Wibisono & Kinnaird (2003); Karanth et al. (2004); Kawanishi & Sunquist (2004); Wegge, Pokheral & Jnawali (2004); Karanth et al. (2006); Linkie et al. (2006); Jhala, Gopal & Qureshi (2008); Wang & Macdonald (2009); and Lynam et al. (2009). Most camera trapped areas were ‘small’ subsets of larger tiger occupied landscapes and often cameras were placed in areas that have relatively high tiger density within this landscape (e.g. Karanth et al. 2004; Jhala, Gopal & Qureshi 2008). Therefore, density estimates obtained from camera trapped areas cannot be extrapolated to occupied landscapes (Garshelis 1992; but see Linkie et al. 2006), and have limited application in estimating population size or evaluating the status of tigers at landscape, state or country scale. Occurrence of tigers in a forest patch can be ascertained by detection of their sign in the form of pugmark trails, scat, rake marks, scrape marks and vocalization (Karanth & Nichols 2002; Jhala, Qureshi & Gopal 2005a). Quanta of signs in an area are likely to be related to abundance (Smallwood & Fitzhugh 1995; Stander 1998). An attempt to quantify relationships between tiger densities and abundance of tiger signs is needed for developing models that would help in evaluating the status of tigers and conservation potential of large landscapes from indices in a timely and cost eﬀective manner (Lynam et al. 2009).

The country wide total count of tigers using experts to individually identify each individual from their pugmark impressions has been severely criticized (Karanth et al. 2003). The grave status of tigers in India gained global attention when the oﬃcial census continued to report good numbers even when the species became locally extinct from Sariska Tiger Reserve in 2004 and later in Panna Tiger Reserve (2009) due to poaching (Check 2006; Rajesh et al. 2010). Subsequently, the Prime Minister established a Tiger Task Force in 2005 to investigate and resolve the tiger crisis in the country. The Tiger Task Force identiﬁed, amongst others, the lack of a credible status assessment system as a major problem (Narain et al. 2005). In this article we evaluate relationships between indices of tiger abundance and tiger density using a double sampling approach (Cochran 1977; Eberhardt & Simmons 1987). We estimate absolute tiger densities by camera trap-based mark– recapture simultaneously with estimates of quanta of tiger sign from 21 diﬀerent sites (5185 km2) from amongst 53 787 km2 of tiger occupied forests in Central and North India (Jhala, Gopal & Qureshi 2008). We conduct a power analysis to determine the ability of CTMR and our index-based models to detect change in tiger abundance. The methods and concept presented herein form an important component of a country-wide tiger status evaluation protocol that was assessed and recommended by the Tiger Task Force (Jhala, Qureshi & Gopal 2005b).

Materials and methods STUDY AREAS

We sampled 18 tiger populations in central and north India (Jhala, Gopal & Qureshi 2008) with 21 independent sampling units. Sampled sites were located in the states of Uttarakhand, Uttar Pradesh, Bihar, Rajasthan, Madhya Pradesh, Orissa, Andhra Pradesh and Maharashtra (Fig. 1) and were sampled between 2006 and 2007. Based on a priori knowledge and pilot surveys we stratiﬁed potential sampling areas into high, medium, low and very low tiger abundance categories. Based on the conservation importance of the tiger populations and logistical constraints we sampled ﬁve units with high, seven units with medium, six units with low and three units with very low tiger abundance. Sampled areas ranged between 125 and 300 km2. At a few sites (e.g. Dhaulkhand) the size of tiger occupied forest patches were small thereby restricting the area coverage of our samples. In Corbett the sample coverage was large (545 km2) due to availability of additional resources and contiguous tiger occurrence over a large area. All sampled areas were larger than the average home-range of tigers in these habitats (Smith, McDougal & Sunquist 1987; Chundawat, Gogate & Johnsingh 1999; Sharma et al. 2009). Sampled sites covered all types of tiger habitats found in Central and North India ranging from the Terai grasslands and moist Sal Shorea robusta forests of Corbett, Dudhwa, Similipal and Valmiki; arid thorn and Aneogeissus forests of Ranthambore and Kuno; dry deciduous mixed forests of Panna; mesic Sal forests of Kanha and Bandhavgarh; teak Tectona grandis dominated forests in Pench, Tadoba and Melghat and deciduous forests of the Eastern Ghats in Sri Sailam Tiger Reserve (Fig. 1). The topography and rainfall also varied greatly between the study sites.

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 14–24

16 Y. Jhala, Q. Qureshi & R. Gopal

Fig. 1. Study sites for estimating tiger density overlaid on the forest cover map of India showing tiger occupied forests. Site codes are referenced in Table 1.

ESTIMATING TIGER DENSITY

Absolute tiger densities were estimated by closed capture–recapture models (Pollock et al. 1990) using camera traps (Karanth & Nichols 1998). We placed one double-sided remote camera unit (TrailmasterTM and ⁄ or DeerCamTM) in each 2 · 2 km grid cell overlaid on the study area at each site. Within each grid cell a team of wildlife biologists with the assistance of local forest guards searched for the best location to deploy the camera unit to maximize photo-capture of tigers. This approach ensured that no gaps were left and that there was a reasonable density of

camera trap units within the range of all tigers (Sharma et al. 2009). Sampling durations lasted from 20 to 96 days with 20 to 120 double-sided camera units deployed at each site. Cameras were deployed simultaneously to cover the entire study area of a site, except in Corbett where we sampled the area in two blocks with equal sessions (days) in each block (Karanth et al. 2004). Individual tigers were identiﬁed from their unique stripe patterns and a capture matrix ‘‘X’’ generated for each site (Karanth & Nichols 1998). Despite substantial camera trap eﬀort we could not photo capture tigers at two sites (Kuno and Phen) and captured less than ﬁve individuals at three sites (Table 1).

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 14–24

680 450 11495 1643 462 2840 1080 750 960 825 1290

800 1000 750 1860 1716 600 2146 1080 1600 1140

210 196 545 221 108 243 280 400 267 125 272

185 152 400 75 275 136 323 227 250 215

Camera trap nights

7 13 NA 1 26 4 5 5 10 3

22 5 102 13 12 12 17 NA 11 8 3 Mh Mh NA NA Mb NA Mth Mh Mh NA

Mh Mo Mth Mh Mh Mo Mh NA Mh Mb NA

Best site model

0Æ36 0Æ82 NA NA 0Æ003 NA 0Æ09 0Æ98 0Æ99 NA

0Æ3 0Æ45 0Æ43 0Æ07 0Æ22 0Æ5 0Æ47 NA 0Æ96 0Æ14 NA

Closure test P value

(1Æ0) (0Æ1) NA

(4Æ2) (0Æ5) (9Æ4) (1Æ5) (1Æ8) (2Æ73) (4Æ2)

9 (1Æ7) 17 (2Æ6) 2 NA 28 (2Æ8) NA 7 (1Æ1) 6 (2Æ0) 10 (0Æ3) NA

26 5 127 14 13 14 18 1 11 8

Tiger numbers site estimate (SE)

9 (1Æ7) 15 (1Æ1) 2 1 (0Æ001) 27 (1Æ35) 5 (1Æ55) 5 (0Æ01) 5 (0Æ02) 10 (0Æ1) 3 (1Æ1)

29 (3Æ2) NA 104 (1Æ6) 14 (1Æ2) 13 (2Æ1) 12 (1Æ0) 17 (1Æ3) 1 11 (0Æ01) 10 (1Æ8) 3 (0Æ003)

Tiger numbers pooled estimate (Mh Two Mixture Model) (SE)

0Æ09 0Æ72 NA 1Æ00 0Æ42 0Æ76 0Æ64 0Æ79 0Æ27 1Æ00

0Æ92 NA 0Æ90 0Æ64 0Æ22 0Æ91 0Æ23 NA 0Æ18 0Æ77 0Æ51

(0) (0Æ11) (0Æ23) (0Æ23) (0Æ19) (0Æ15) (0)

(0Æ2) (0Æ12)

(0Æ12) (0Æ16) (0Æ36)

(0Æ03) (0Æ15) (0Æ05) (0Æ09) (0Æ19)

(0Æ08)

Capture probability Mh Mixture Model (SE)

4Æ9 9Æ5 0Æ50 0Æ55 9Æ81 1Æ64 1Æ87 2Æ59 4Æ00 1Æ49

13Æ6 2Æ54 19 6Æ15 12Æ00 5Æ01 6Æ23 0Æ25 4Æ12 7Æ25 1Æ49 (1Æ5) (0Æ84) (0Æ10) (0Æ02) (0Æ75) (0Æ55) (0Æ06) (0Æ95) (0Æ21) (0Æ52)

(1Æ7) (0Æ70) (0Æ54) (0Æ87) (2Æ0) (0Æ52) (0Æ76) (0Æ10) (0Æ15) (1Æ57) (0Æ05)

Tiger density 100 km)2 (SE)

2Æ3 12Æ9 1Æ45 1Æ32 7Æ19 0Æ79 2Æ2 2Æ36 5Æ06 2Æ18

(0Æ35) (0Æ77) (0Æ39) (0Æ54) (0Æ63) (0Æ45) (0Æ39) (0Æ38) (0Æ36) (0Æ41)

12Æ8 (1Æ15) 1Æ15 (0Æ47) 12Æ8 (5Æ9) 4Æ91 (0Æ31) 12Æ23 (1Æ13) 4Æ31 (0Æ32) 4Æ97 (0Æ32) 0Æ05 (0Æ51) 2Æ5 (0Æ38) 4Æ1 (0Æ33) 3Æ87 (0Æ29)

Model predicted tiger density (SE) §

169Æ0 101Æ0 145Æ0 111Æ0 113Æ0 221Æ0 159Æ9 261Æ0 188Æ0 112Æ5

213Æ0 75Æ2 214Æ7 138Æ5 90Æ0 87Æ1 91Æ3 151Æ0 197Æ0 144Æ0 101Æ4

Sign survey eﬀort (km)

0Æ16 1Æ48 0Æ02 0Æ24 0Æ60 0Æ08 0Æ10 0Æ13 0Æ51 0Æ36

1Æ31 0Æ23 4Æ24 0Æ65 1Æ10 0Æ56 0Æ60 0Æ01 0Æ46 0Æ37 0Æ50

(0Æ04) (0Æ10) (0Æ00) (0Æ09) (0Æ09) (0Æ02) (0Æ03) (0Æ10) (0Æ10) (0Æ08)

(0Æ24) (0Æ11) (0Æ47) (0Æ10) (0Æ22) (0Æ15) (0Æ16) (0Æ00) (0Æ06) (0Æ07) (0Æ80)

Pugmark Track set encounters km)1 (SE)

0Æ25 2Æ83 0Æ17 0Æ02 1Æ95 0Æ04 0Æ28 0Æ29 0Æ67 0Æ07

4Æ15 0Æ03 0Æ52 0Æ47 4Æ13 0Æ37 0Æ56 0Æ01 0Æ11 0Æ55 0Æ24

(0Æ07) (0Æ10) (0Æ06) (0Æ01) (0Æ41) (0Æ02) (0Æ07) (0Æ17) (0Æ11) (0Æ03)

(1Æ90) (0Æ02) (0Æ15) (0Æ13) (0Æ53) (0Æ11) (0Æ29) (0Æ00) (0Æ03) (0Æ10) (0Æ05)

Scat encounters km)1 (SE)

*Tiger densities not obtained by camera trap mark–recapture. †Tiger density estimate from Harihar, Pandav & Goyal (2009). ‡Tiger density estimate from Sharma et al. (2010), all other estimates obtained as part of this study. §Tiger densities were predicted by Jackknife procedure using the best multiple regression model that had tiger pugmark and scat encounter rates as the independent variables, see text for details. Site codes in parenthesis (#) denote the location of the site in Fig. 1.

Bandhavgarh (1) Chilla† (2) Corbett (3) Dudhwa (4) Kanha‡ (5) Katarniaghat (6) Kishenpur (7) Kuno* (8) Melghat (9) Mukki (10) Nagarjun Sagar (11) Panna (12) Pench (13) Phen* (14) Dhaulkhand (15) Ranthambore (16) Satpura (17) Simlipal (18) Suphkar (19) Tadoba (20) Valmiki (21)

Location (site codes)

Eﬀectively sampled area (km2)

No of tigers photocaptured

Table 1. Abundance, density and sign encounter rates of tigers from 21 sites in Central and Northern India using photographic mark–recapture and search paths. Numbers are estimates with standard errors in parenthesis

Tiger abundance from signs 17

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 14–24

18 Y. Jhala, Q. Qureshi & R. Gopal Due to the small number of tiger captures at these sites, it was not possible to obtain reliable estimates of population size, evaluate model selection or test for population closure using standard mark– recapture analysis (Kendall 1999; Stanley & Burnham 1999; Karanth & Nichols 2002). We therefore, used two separate approaches for obtaining density estimates. For sites where ﬁve or more individual tigers were captured, population size was estimated using program CAPTURE (Rexstad & Burnham 1991), whilst demographic and geographic closure population was tested using CloseTest (Stanley & Burnham 1999). For comparability with earlier estimates (Karanth et al. 2004), density estimates were obtained following Karanth & Nichols (1998, 2002); this method required that a buﬀer strip of half the mean maximum distance moved (1 ⁄ 2 MMDM) by recaptured tigers was added to the camera trap area to determine the eﬀectively trapped area. To obtain abundance estimates for the low-sample size populations, we carried out a combined analysis of the data sets where models were ﬁtted to the pooled encounter histories from all populations (MacKenzie et al. 2005). Thus, information on capture probability parameters was combined across populations. The combined analysis used program MARK (White & Burnham 1999) and model Mh (with a two-point ﬁnite mixture). Since population estimates did not diﬀer between the two analyses (individual site capture matrix and pooled matrix for all sites), the more precise population estimates obtained by the combined site analysis were used for computing tiger densities and developing our models. The average cost of obtaining absolute density of tigers at each site considering wages of biologists, and ﬁeld personnel, vehicle rental, discounted equipment and material costs, was estimated to be Rs. 850 000 (about US$ 17 000); the process took, on average, 720 man-days to accomplish.

ESTIMATING TIGER SIGN

Though this article focuses on small areas within large tiger occupied landscapes where CTMR and sign-based indices of abundance were simultaneously estimated, the methodology for sign survey was developed for application throughout all potential forested areas likely to have tigers (Jhala, Qureshi & Gopal 2005a; b). Thus, sampling for tiger sign needed to be systematically distributed throughout the forested areas of the landscape. Grid designs are statistically well suited to distribute sampling units; however, locating a grid in a large forested landscape is not an easy task for ﬁeld sampling. Most of India’s forests are delineated into hierarchical administrative units of forest divisions, ranges, beats and compartments. Boundaries of beats and compartments are based on natural features like ridges, waterways and dirt tracks. Each division is administered by Divisional forest oﬃcer; a Range is administered by a Range Forest Oﬃcer and a Beat by a Beat Guard who has intimate knowledge of these areas. The Beat Guard participated as a team member for sign surveys conducted in his beat. The average beat size in our study areas was 16Æ5 (SE 4Æ16) km2. We estimated the length of the search path that had to be sampled to minimize the eﬀect of habitat and substrate on encounter rates of tiger sign (Hayward et al. 2002). During a pilot study conducted in 2002–2004 in c. 50 000 km2 of the Satpura-Maikal landscape we incrementally increased the length of the search path by 1 km, from 1 to 12 km and estimated encounter rates of tiger sign, we found that encounter rates of pugmark and scats stabilized after a 4 to 5 km search (Jhala & Qureshi, unpubl. data). Subsequently for this study, each survey consisted of a 5 km search for tiger signs. Surveys were not random, but instead conducted along features that were likely to have tiger sign (e.g. dirt roads, dry water courses and animal trails).

Three spatially diﬀerent surveys were conducted within each beat; this served to distribute our survey eﬀorts throughout the study area. A team of 30 biologists was trained for a period of 15 days in identifying and searching, and for consistency in classiﬁcation of tiger sign between observers. This team was then deployed at various study sites and each search path was walked by two observers. All encounters of tiger pugmark track sets and scats were recorded. These were distinguished from those of other carnivores based on criteria described by Jhala, Qureshi & Gopal (2005b) and Karanth & Nichols (2002), and by ﬁeld training. An accurate measure of each search path was recorded using a hand held GPS unit (Garmin 72TM). Encounter rates of each sign category were computed as the number of signs per km of search path. The cost of obtaining data on tiger sign indices was on average Rs. 62 000 (about US$ 1240) for each sampled site, which consisted of 22 beats (SE 2Æ44) on average. The time taken to sample each site for tiger signs was, on average, 220 man-days.

CALIBRATING TIGER SIGN INDICES WITH TIGER DENSITIES

Pearson’s correlation coeﬃcients were computed from 21 values obtained from the sampled sites between tiger densities and encounter rates of tiger pugmark trails, and tiger scats per km. Scatter plots of pugmark encounters and scat encounters versus tiger density were examined (see Appendix S1, Figs S1 and S2, Supporting information). A square root transformation of tiger scat encounters linearized relationship with tiger density (Sokal & Rohlf 1995). Simple and multiple linear regressions were used to investigate relationships and calibrate quanta of tiger sign with tiger density (Draper & Smith 1981; Eberhardt & Simmons 1987). Ideally, tiger densities should be known with certainty for developing relationships with indices (Engeman 2003). As this is practically impossible to achieve in free ranging populations, we used estimates of tiger densities obtained by mark–recapture methods (Pollock et al. 1990). To account for variability in precision of tiger density estimates we used a weighted regression approach for all the models (Sokal & Rohlf 1995). Each tiger density estimate was weighed by the reciprocal of the coeﬃcient ^ divided by the median tiger of variation of tiger density (CV [D]) density (Wiewel, Clark & Sovada 2007). Thus, precise density esti^ made a greater contribution to the regression mates (smaller CV [D]) model. Since one objective was to develop predictive models to estimate tiger densities from tiger abundance indices, we used least squares regression to assess the relationship between (i) pugmark trail encounter rates and tiger density, (ii) square root transformed encounter rates of tiger scat and tiger densities and (iii) multiple regression analysis with pugmark trail encounters and square root transformed scat encounters as independent variables and tiger densities as the dependent variable (Draper & Smith 1981). We assessed model ﬁt and performance by coeﬃcient of determination (R2), root mean square error (RMS) and Akaike Information Criteria (Draper & Smith 1981; Sokal & Rohlf 1995; Burnham & Anderson 2002). The ability of the indices to predict tiger density was assessed using a Jackknife analysis wherein we dropped each site, re-computed the best regression model, and used it to predict the tiger density of the excluded site (Krebs 1989). The predictive performance was summarized by the predicted sum of squares R2 (PRESS R2), and correlation of Jackknife model estimated tiger density with CTMR tiger density (Draper & Smith 1981). No photographic captures of tigers were obtained after over 750 trap nights in the best potential tiger habitat of 150 km2 of Phen Sanctuary and Kuno Sanctuary. However, we obtained reliable evidence

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 14–24

Tiger abundance from signs 19 of tiger occurrence through sign when a larger area of about 400 km2 was searched in and around each camera trap site. Based on pugmark track sets (Sharma, Jhala & Sawarkar 2005) and tiger scat obtained during the extensive search within the landscape, we conservatively estimated that two tigers operated in and around Phen Sanctuary and one tiger occurred in the Kuno landscape. Conservative Tiger densities estimated for these two sites were 0Æ25 and 0Æ5 tigers per 100 km2 for Kuno and Phen, respectively. Since tiger densities for these two sites were not estimated by mark–recapture, we construct regression models with, as well as without, data from these two sites.

POWER OF INDICES TO DETECT CHANGE IN TIGER DENSITIES

We carried out analyses to evaluate the power of CTMR estimates of tiger density, indices, and our models based on indices to detect changes of various magnitudes at ﬁxed type I error rates (a levels) of 0Æ3. Since the type I error is strictly a value judgment, and the consequence of rejecting a true null hypothesis of ‘no change’ is of lesser importance (to the management objective of conserving the endangered tiger) than failing to detect population declines (Beier & Cunningham 1996) we use a ‘larger than usual’ type I error rate. Tiger populations show natural ﬂuctuations of substantial magnitude as a consequence of recruitment, dispersal and immigration (Karanth et al. 2006; Barlow et al. 2009). Yet, it is pertinent for wildlife managers to be able to detect changes of at least 30% size eﬀect (especially declines), to react in a timely manner with appropriate actions. Thus, we set minimum management standards of achieving 80% power to detect 30% changes between two subsequent survey eﬀorts (Hayward et al. 2002; Barlow et al. 2008). We used program monitor (Gibbs & Eduard Ene 2010) to conduct Monte Carlo simulations at a desired power of 0Æ8 for evaluating the precision of CTMR, pugmark and scat encounter rates in detecting changes between two subsequent surveys. We follow Barlow et al. (2008) and Hayward et al. (2002) and use exponential response and lognormal measures to model changes in CTMR tiger density and indices of tiger abundance. Other options selected in the software were paired plot comparisons, which compare the same sites between two sampling intervals and test the hypothesis that the diﬀerence between the ﬁrst and second survey averaged across all sites is greater or less than zero (two tailed tests). To evaluate the power of using observed changes in sign indices to detect tiger population declines we ﬁtted linear regression models in which the indices (i) were treated as the dependent variable, and CTMR tiger density (D) as the independent variable. We account for imperfect tiger density estimates by ﬁtting a measurement error model using the estimated variance of density estimates as the measurement error variance for each observation. In the context of our problem, the null hypothesis is of the form: H0: D(t+1) – D(t) = 0 and we reject based on the diﬀerence in observed index values i(t+1) – i(t) based on an index i. That is, we reject the hypothesis of ‘no change in density’ if the observed change in index values is large. The power of this test is the probability that the hypothesis is rejected given a certain prescribed change in density. For a ﬁxed type I error rate (a) this is a standard power calculation (Casella & Berger 1990). Combining both indices into the calculation of power assumed that the indices are independent, conditional on tiger density. Further details of the power analysis are provided in Appendix S2 (Supporting Information). All statistical analyses was done using spss 11 (SPSS 2001), NCSS (Hintze 2006) and R Development Core Team (2004) software packages.

Results TIGER DENSITY

The highest tiger densities were estimated for Corbett Tiger Reserve at 19 tigers per 100 km2 where we photographed 102 individual tigers. The lowest estimate obtained by CTMR was for Rajaji Dhaulkhand, where a single tigress was photographed twice, and density was estimated at 0Æ55 tigers per 100 km2 (Table 1). Due to very low tiger densities we could not obtain camera trap photographs of tigers at Kuno and Phen. The best model selected by CAPTURE for most sites with more than ﬁve tiger captures had some form of heterogeneity in capture probabilities. CloseTest supported population closure for all sites except for Ranthambore (Table 1). Model Mh [two-point ﬁnite mixture model, where p1 = 0Æ038 (SE 0Æ002), p2 = 0Æ177 (SE 0Æ008)], using the pooled capture matrix for all sites provided more precise estimates of tiger numbers than site speciﬁc analyses in CAPTURE (Table 1).

TIGER ABUNDANCE INDICES

On average 147 (SE 12, range 75–221) km of search eﬀort was invested at each site. The maximum number of tiger pugmark sets was recorded in Corbett and the minimum number in Kuno (Table 1). The maximum number of tiger scats was obtained from Bandhavgarh and the minimum number from Kuno (Table 1).

CALIBRATING TIGER SIGNS WITH TIGER DENSITY

Tiger pugmark set encounters had the best linear correlation with tiger densities (R = 0Æ92, P < 0Æ0001, n = 21) across all sites. Tiger scat encounter rates had a quadratic relationship with tiger densities probably due to greater persistence of scats than pugmarks within the environment. Thus, at equilibrium (where scat deposition equals decomposition), scat density would be disproportionately higher for high tiger density probably resulting in a curvilinear relationship (see Fig. S2, Supporting information). Scat encounters were found to be low in Corbett in comparison to tiger density. Transformed tiger scat encounter rates, after excluding Corbett, had a high linear correlation with tiger density (R = 0Æ91, P < 0Æ0001, n = 20). Tiger pugmark and scat encounter rates explained 84% and 30%, respectively, of the observed variation in tiger densities (Table 2). Tiger pugmarks and tiger scats together explained 94% of the variation in tiger densities. All the three models had good predictive ability for tiger densities (PRESS R2 = 0Æ997, Table 2). The multiple regression model with pugmarks and tiger scats had the lowest AIC and RMS values and was therefore selected as the best model (Table 2). Jackknife predicted tiger densities correlated well with mark–recapture-based estimates of tiger densities (R = 0Æ91, P = 0Æ0001, Table 1). The regression coeﬃcients presented in Table 2 were developed using data from 21 sites including the two sites where we could

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 14–24

20 Y. Jhala, Q. Qureshi & R. Gopal Table 2. Least square regression models for estimating tiger densities from tiger sign indices (n = 21 sites)

Model 1 2 3

Independent variables

Slope

Slope P value

Pug mark SqRt (Scat) Pug mark SqRt (Scat)

4Æ24 (0Æ42) 6Æ69 (2Æ32) 3Æ84(0Æ26) 4Æ07(0Æ69)

<0Æ0001 0Æ0097 <0Æ0001 <0Æ0001

Intercept

Intercept P value

2Æ02 (0Æ61) 1Æ107 (1Æ84) )0Æ31(0Æ53)

<0Æ004 0Æ55 0Æ57

R2 0Æ84 *0Æ303 0Æ947

Adj. R2 0Æ834 *0Æ267 0Æ94

PRESS R2

RMS

AIC

0Æ989 0Æ99 0Æ997

17Æ23 36Æ2 10Æ29

105Æ95 137Æ1 85Æ12

R2 – coeﬃcient of determination, Adj. R2 – R2 adjusted for degrees of freedom, PRESS R2 – R2 value computed from prediction sum of squares, RMS – root mean square error, AIC – Akaike information criteria. *R2 and Adj R2 values after removal of outlier Corbett data were 0Æ824 and 0Æ814 respectively. Table 3. Least square regression models for estimating tiger densities from tiger sign indices, two of the sites where tiger densities were not estimated by camera trap mark–recapture models have been omitted (n = 19 sites)

Model

Independent variables

Slope

Slope P value

1 2 3

Pugmark SqRt(Scat) Pugmark SqRt(Scat)

4Æ21 6Æ58 3Æ83 4Æ06

0Æ0001 0Æ016 0Æ0001 0Æ0001

(0Æ44) (2Æ45) (0Æ28) (0Æ73)

Intercept

Intercept P value

R2

Adj. R2

PRESS R2

RMS

2Æ09 (0Æ64) 1Æ25 (1Æ98) 0Æ28 (0Æ57)

0Æ005 0Æ53 0Æ63

0Æ88 0Æ3 0Æ94

0Æ83 0Æ25 0Æ94

0Æ989 0Æ99 0Æ997

18Æ1 38Æ1 10Æ8

R2 – coeﬃcient of determination, Adj. R2 – R2 adjusted for degrees of freedom, PRESS R2 – R2 value computed from prediction sum of squares, RMS – root mean square error.

not photo-capture tigers. By excluding these two sites from our regression models the regression coeﬃcients, ﬁt, or performance were not altered (Table 3). We therefore retain the two sites in our ﬁnal analysis, to have a wide range of tiger density coverage (0Æ25 to 19 tigers per 100 km2).

POWER TO DETECT CHANGE IN TIGER DENSITY

Considering all sites, CTMR could detect 12% change in tiger density with a power of 0Æ8 between two subsequent surveys. At the desired power of 0Æ8, pugmark encounter rates could detect a decline of 25% and an increase of 23% over all sites, while scat encounter rates could only detect declines of 35% and increases of 37%. The desired power of 0Æ8 to detect minimum density declines of 30% at each site by subsequent surveys could be achieved only if the CTMR density estimate had a coeﬃcient of variation less than 20%. The power of the best index-based model to detect a 30% decline in tiger density ranged between 50% and 85% at a type 1 error rate of 0Æ3. The power to detect a decline in tiger density was higher for pugmark encounter rates in comparison to scat encounter rates. The power of the models increased at higher tiger densities (Fig. 2 and Appendix S2, Supporting information ).

Discussion ESTIMATING TIGER DENSITIES

Tiger densities estimated by camera trap data in a mark– recapture framework ranged between 0Æ55 and 19 tigers per 100 km2, and were estimated with an average precision of 13

Fig. 2. The power of index based models to detect tiger density declines of various magnitudes at a type 1 error rate of 0Æ3.

(SE 3Æ2)% CV (Table 1). In the cases of Phen and Kuno, tiger density was so low that population estimation through CTMR was impractical and would have required much greater eﬀort than we invested (Kawanishi & Sunquist 2004). Such low density areas rarely harbour breeding populations of tigers in Indian forests and, thus, contribute little to tiger abundance (Jhala, Gopal & Qureshi 2008). However, their importance to serve as dispersal corridor habitats or their potential to harbour breeding populations in the future should not be disregarded. The duration of sampling lasted for a maximum of 96 days (for Corbett) while all others were less than 60 days. Most of our study areas abutted hard boundaries on some sides. Considering the longevity of tigers

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 14–24

Tiger abundance from signs 21 and by considering tigers >1Æ5 years for population estimation, we believe that we were justiﬁed in assuming demographic closure. The statistical test for closure (Stanley & Burnham 1999) supported population closure in the majority of our sampled sites for which the number of tiger captures was suﬃcient to perform a meaningful test (Kendall 1999). Population closure was not supported for Ranthambore (Table 1). However, from our additional intensive and extensive camera trapping eﬀorts and ongoing telemetry study of tigers, we were reasonably certain that the Ranthambore population was geographically and demographically closed during our sampling period.

INDICES OF ABUNDANCE

Use of indices for evaluating abundance and population trends of endangered species has been a matter of serious debate (Ellington & Lukacs 2003; Hutto & Young 2003; Conn, Bailey & Sauer 2004; Jhonson 2008). Most proponents of indices advocate the practicality of using indices as surrogates for abundance (Hutto & Young 2003). It would be extremely diﬃcult and resource intensive to attempt to estimate populations of tigers throughout their range using robust approaches like CTMR (Lynam et al. 2009). Simple indices of tiger sign oﬀer a cost eﬀective alternative to the evaluation of tiger status over larger landscapes (Pollock et al. 2002; Linkie et al. 2006). The cost and time required for estimating indices of tiger abundance was only 7% and 33% respectively of the cost and time required to estimate tiger abundance with camera traps. The main arguments against the use of indices are that they are rarely calibrated with absolute abundance estimates (Williams, Nichols & Conroy 2002; Conroy & Carroll 2009). Herein, we address this concern using the double sampling approach (Cochran 1977; Pollock et al. 2002) and collect index and density data from the same areas simultaneously (Skalski, Ryding & Millspaugh 2005; Conroy & Carroll 2009). Amongst the 19 sites sampled successfully with camera trap, tiger densities ranged from 0Æ55 to 19 tigers per hundred km2 (Table 1), giving a wide spectrum of density for calibrating indices. Skalski, Ryding & Millspaugh (2005) recommend using a double sampling approach with a minimum data set of n = 5 when correlations between sign and abundance are above 0Æ85. Our data adequately satisfy these recommendations. The regression model included both tiger pugmark and scat as independent variables was selected as the best model by AIC (Table 2). This model had exceptionally good predictive ability (PRESS R2 = 0Æ99) across a wide spectrum of naturally occurring tiger densities (0Æ25 to 19 tigers per 100 km2) and did not suﬀer from problems of collinearity (variance inﬂation factor <1Æ1, tolerance = 0Æ97, condition number <14). The high signiﬁcance level of both predictor variables (P < 0Æ0001), and substantial increment in the predictive power of the model by inclusion of scat encounter rates, justiﬁes the use of a full, two predictor model for estimating

tiger densities (Whittingham et al. 2006). The model should be used to estimate tiger densities when tiger sign data are collected in the manner described here to generate similar observational data within the data range used to develop the models.

POWER TO DETECT TIGER DENSITY DECLINES

It is important for population estimates to be able to detect changes in abundance (Gibbs, Snell & Causton 1999), especially declines in the case of the endangered tigers (Beier & Cunningham 1996). The average precision (CV) of our tiger densities using individual site population estimates was 23(SE 3)% and 13Æ2 (SE 2Æ5)% using the combined site analysis approach. This level of precision seems to be typical for CTMR density estimates for tigers, as the average CV observed from 29 tiger density estimates was 28Æ8 (SE 4Æ1)%. To detect a 30% decline at each site by two consecutive surveys the required precision was <20% CV. Our CTMR site estimates met this criterion in 68% of cases. The low power of the index-based models to detect declines in tiger abundance between two consecutive surveys was expected (Fig. 2), given that our power analysis took into account variability of CTMR density, pugmark encounter rates and scat encounter rates. Type I error (or the probability of rejecting a true null hypothesis i.e. no change in tiger density) was set quite high at 0Æ3 in comparison to traditional statistical norms. This error does not have serious consequences with regard to tiger conservation, in comparison to our inability to detect a decline in tiger density. With a = 0Æ3, the index-based model could detect 30% of declines in tiger density with power ranging between 50% and 85%. The management threshold of 80% power to detect 30% of declines between two subsequent surveys was achieved for sites with a tiger density >10 tigers per 100 km2 (Fig. 2). The current estimate of power and eﬀect size is computed for two subsequent samples. The power for detecting tiger density declines could be increased by improving the precision of CTMR density through use of likelihood based spatially explicit estimator models (Eﬀord, Brochers & Byrom 2009; Royle et al. 2009), improving precision of indices by increasing sampling eﬀort (Eberhardt & Simmons1987), and by using time series data for trend analysis in place of two sample comparisons (Gibbs, Snell & Causton 1999). The relationships between indices of tiger abundance and tiger density developed here are applicable across Central and North India encompassing about 300 000 km2 of forested habitat (Fig. 1). Similar relationships could be investigated for other regions and species. The initial costs and eﬀort of double sampling vast landscapes are a major deterrent to this eﬀort, but once undertaken, they lead to rapid and cost eﬀective assessments of the status of the target species. The precision of model predictions will increase as more double sampling data accumulate with further time and eﬀort (Eberhardt & Simmons 1987; Pollock et al. 2002).

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 14–24

22 Y. Jhala, Q. Qureshi & R. Gopal Tiger populations in India are characterized by source–sink dynamics (Pulliam 1988). Most reserves harbouring a breeding population of tigers (currently about 13 000 km2) serve as sources to populate and maintain tiger occupancy of sink habitats (currently about another 80 000 km2) (Jhala, Gopal & Qureshi 2008). Source populations across the tiger’s range are under threat from commercial poaching (Check 2006). Poaching can deplete a source and cause extinctions within a short period of time (Chapron et al. 2008; Rajesh et al. 2010). Due to the small size of most source populations in India (Jhala, Gopal & Qureshi 2008), habitat linkages between sources that permit exchange of individuals are important elements for long-term survival of tigers (Wikramanayaka et al. 2004; Dinerstein et al. 2007). Tiger habitats throughout their range are threatened by development projects and human pressures (Sanderson et al. 2006). Considering the precarious status of tigers (Sanderson et al. 2006), it is essential that wildlife managers and policy makers are able to protect tiger populations eﬀectively wherever they are declining, by timely deployment of remedial measures (Gibbs, Snell & Causton 1999). Source populations of tigers are of paramount importance and should be monitored annually by resource intensive CTMR. Population estimates made using CMTR can detect population changes over short time periods and provide additional information on population dynamics (e.g. survival rates and recruitment, see Karanth et al. 2006). In addition, all areas occupied by tigers should be surveyed every 2 years to provide up-todate indices of abundance. Index-based surveys can provide assessments of spatial occupancy, population extent and the viability of connecting corridor habitats (Linkie et al. 2006). Now, with the calibration of tiger sign indices, they can also provide reliable estimates of tiger abundance. Implementation of a continuous monitoring programme (CTMR and index surveys) will substantially increase our ability to detect trends in tiger density. As with all large carnivores, the conservation of tigers is dependent on the appropriate management of large areas of landscape (Woodroﬀe & Ginsberg 1998; Wikramanayaka et al. 2004). Vast areas of tiger habitat are rapidly vanishing (Dinerstein et al. 2007) and major investment is required to monitor, manage and safeguard these habitats to ensure their long-term survival. The approach and models developed herein permit rapid and cost eﬀective assessments of abundance to monitor the status of tigers at landscape scales. This information is vital for conservation investment, habitat management, planning development projects, formulation of policy and for law enforcement.

Acknowledgements We thank the National Tiger Conservation Authority, Government of India, for funding support. The State Forest Departments of Madhya Pradesh, Rajasthan, Uttar Pradesh, Uttrakhand, Orissa, Bihar, Andhra Pradesh and Maharashtra are thanked for logistical support. P. Ghosh, P.R. Sinha, V.B. Mathur and K. Sankar are acknowledged for their support and facilitation. We thank the team of research biologists who worked hard to collect data on tiger density and tiger signs across India. S. Dutta is specially thanked for assistance with MARK. Comments by two anonymous reviewers and the editor greatly

improved the manuscript. We thank J. Andrew Royle for reviewing the manuscript, providing valuable advice on the analyses, and for doing part of the power analyses presented in this article and in Appendix S2.

References Barlow, A.C.D., Ahmed, M.I.U., Rahman, M.M., Howlader, A., Smith, A.C. & Smith, J.L.D. (2008) Linking monitoring and intervention for improved management of tigers in the Sundarbans of Bangladesh. Biological Conservation, 141, 2031–2040. Barlow, A.C.D., McDougal, C., Smith, J.L.D., Gurung, B., Bhatta, S.R., Kumal, S., Mahato, B. & Tamang, D.B. (2009) Temporal variation in tiger (Panthera tigris) populations and its implications to monitoring. Journal of Mammalogy, 90, 472–478. Beier, P. & Cunningham, S.C. (1996) Power of track surveys to detect changes in cougar populations. Wildlife Society Bulletin, 24, 540–546. Blake, S. & Hedges, S. (2004) Sinking the ﬂagship: the case of the forest elephants in Asia and Africa. Conservation Biology, 18, 1191–1202. Burnham, K.P. & Anderson, D.R. (2002) Model Selection and Multimodel Inference: A Practical Information-Theoretic Approach, 2nd edn. SpringerVerlag, New York. Casella, G. & Berger, R.L. (1990) Statistical Inference. Duxbury Press, Belmount, CA. USA. Caughley, G. (1977) Analysis of Vertebrate Populations. John Wiley & Sons, New York. Chapron, G., Miquelle, D.G., Lambert, A., Goodrich, J.M., Legender, S. & Clobert, J. (2008) The impact on tigers of poaching versus prey depletion. Journal of Applied Ecology, 45, 1667–1674. Check, E. (2006) The tigers retreat. Nature, 441, 927–930. Chundawat, R.S., Gogate, N. & Johnsingh, A.J.G. (1999) Tiger in Panna: preliminary results from an Indian tropical dry forest. Riding the Tiger (eds J. Seidensticker, S. Christie & P. Jackson), pp. 123–129. Cambridge University Press, Cambridge, UK. Cochran, W.G. (1977) Sampling Techniques, 3rd edn. John Wiley & Sons, Inc., New York. Conn, P.B., Bailey, L.L. & Sauer, J.R. (2004) Indices as surrogates to abundance for low-abundance species. Sampling Rare or Elusive Species (ed. W.L. Thompson), pp. 59–74. Island Press, Washington. Conroy, M.J. & Carroll, J.P. (2009) Quantitative Conservation of Vertebrates. Wiley-Blackwell, West Sussex, UK. Dinerstein, E., Loucks, C., Wikramanayake, E., Ginsburg, J., Sanderson, E., Seidensticker, J., Forrest, J., Bryja, G., Heydlauﬀ, A., Klenzendorf, S., Leimgruberg, P., Mills, J., O’Brien, G.T., Shrestha, M., Simons, R. & Songer, M. (2007) The fate of wild tigers. BioScience, 57, 508–514. Draper, N. & Smith, H. (1981) Applied Regression Analysis, 2nd edn. John Wiley & Sons, New York, NY. Eberhardt, L.L. & Simmons, M.A. (1987) Calibrating population indices by double sampling. Journal of Wildlife Management, 51, 665–675. Eﬀord, G.M., Brochers, D.L. & Byrom, A.E. (2009) Density estimation by spatially explict capture-recapture: likelihood-based methods. Modeling Demographic Process in Marked Populations (eds D.L. Thomson, E.G. Cooch & M.J. Conroy), pp. 255–259. Springer, New York, USA. Ellington, A.R. & Lukacs, P.M. (2003) Improving methods for regional land bird monitoring: a reply to Hutto and Young. Wildlife Society Bulletin, 31, 896–902. Engeman, R.M. (2003) More on the need to get the basics right: population indices. Wildlife Society Bulletin, 31, 286–287. Garshelis, D.L. (1992) Mark-recapture density estimation for animals with large home ranges. Wildlife 2001: Populations (eds D.R. McCullough & R.H. Barrett), pp. 1098–1111. Elsevier Applied Science, New York. Gibbs, J.P. & Eduard, E. (2010) Program Monitor: Estimating the statistical power of ecological monitoring programs. Version 11.0.0., URL: http:// www.esf.edu/efb/gibbs/monitor/. Gibbs, J.P., Snell, H.L. & Causton, C.E. (1999) Eﬀective monitoring for adaptive wildlife management: lessons from the Galapagos Islands. Journal of Wildlife Management, 63, 1055–1065. Harihar, A., Pandav, B. & Goyal, S.P. (2009) Responses of tiger (Panthera tigris) and their prey to removal of anthropogenic inﬂuences in Rajaji National Park, India. European Journal of Wildlife Research, 55, 97–105. Hayward, G.D., Miquelle, D.G., Smirnov, E.N. & Nations, C. (2002) Monitoring Amur tiger populations: characteristics of track surveys in snow. Wildlife Society Bulletin, 4, 1150–1159.

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 14–24

Tiger abundance from signs 23 Hintze, J. (2006) NCSS, PASS, and GESS. NCSS Kaysville, Utah. http:// www.ncss.com. Hutto, R.L. & Young, J.S. (2003) On the design of monitoring programs and the use of population indices: a reply to Ellington and Lukacs. Wildlife Society Bulletin, 31, 903–910. Jhala, Y.V., Gopal, R. & Qureshi, Q. (2008) Status of Tigers, Co-Predators and Prey in India. National Tiger Conservation Authority and Wildlife Institute of India, Dehradun. TR 08 ⁄ 001 pp. 164. Jhala, Y.V., Qureshi, Q. & Gopal, R. (2005a) Monitoring Tigers, Co-Predators, Prey and their Habitat: Field Guide. Technical Publication of the Project Tiger Directorate, New Delhi and the Wildlife Institute of India, Dehradun, India. Jhala, Y.V., Qureshi, Q. & Gopal, R. (2005b) Methodology for estimating and monitoring tigers, prey, and habitat: technical note. Indian Forester, 131, 1393–1398. Johnson, D.H. (2008) In defense of indices: the case of bird surveys. Journal of Wildlife Management, 72, 857–868. Karanth, K.U. & Nichols, J.D. (1998) Estimating tiger densities in India from camera trap data using photographic captures and recaptures. Ecology, 79, 2852–2862. Karanth, K.U. & Nichols, J.D. (Eds) (2002) Monitoring Tigers and their Prey: A Manual for Researchers, Managers and Conservationists in Tropical Asia. Centre for Wildlife Studies, Bangalore, India. Karanth, K.U., Nichols, J.D., Scidensticker, J., Dinerstein, E., Smith, J.L.D., McDougal, C., Johnsingh, A.J.T., Chundawat, R.S. & Thapar, V. (2003) Science deﬁciency in conservation practice: the monitoring of tiger populations in India. Animal Conservation, 6, 141–146. Karanth, K.U., Nichols, J.D., Kumar, N.S., Link, W.A. & Hines, J.E. (2004) Tigers and their prey: predicting carnivore densities from prey abundance. Proceedings of Natural Academy of Science, 101, 4854– 4858. Karanth, K.U., Nichols, J.D., Kumar, S. & Hines, J.E. (2006) Assessing tiger population dynamics using photographic capture-recapture sampling. Ecology, 87, 2925–2937. namegroup type="author">Kawanishi, K. & Sunquist, M.E. (2004) Conservation status of tigers in a primary rainforest of Peninsular Malaysia. Biological Conservation, 120, 329–344. Kendall, W.L. (1999) Robustness of closed capture-recapture methods to violations of closure assumption. Ecology, 80, 2517–2525. Krebs, C.J. (1989) Ecological Methodology. Harper & Row. Pub., New York. Linkie, M., Chapron, G., Martyr, D.J., Holden, J. & Leader-Williams, N. (2006) Assessing the viability of tiger populations in a fragmented landscape. Journal of Applied Ecology, 43, 576–586. Lynam, A.J., Rabinowitz, A., Myint, T., Maung, M., Latt, K.T. & Po, S.H.T. (2009) Estimating abundance with sparse data: tigers in northern Myanmar. Population Ecology, 51, 115–121. MacKenzie, D.I. & Kendall, W.L. (2002) How should detection probability be incorporated into estimates of relative abundance? Ecology, 83, 2387– 2393. MacKenzie, D.I., Nichols, J.D., Sutton, N., Kawanishi, K. & Bailey, L.L. (2005) Improving inferences in population studies of rare species that are detected imperfectly. Ecology, 86, 1101–1113. Mech, L.D. (1996) A new era for carnivore conservation. Wildlife Society Bulletin, 24, 397–401. Narain, S., Panwar, H.S., Gadgil, M., Thapar, V. & Singh, S. (2005) Joining the dots: The report of the Tiger Task Force. Project Tiger Directorate, Union Ministry of Environment, Government of India, New Delhi. O’Brien, T.G., Wibisono, H.T. & Kinnaird, M.F. (2003) Crouching tigers hidden prey: status of Sumatran tigers in the Bukit Barisan Slatan National Park, Sumatra, Indonesia. Animal Conservation, 6, 131–139. Pollock, K.H., Nichols, J.D., Brownie, C. & Hines, J.E. (1990) Statistical inference for capture-recapture experiments. Wildlife Monographs, 107, 97. Pollock, K.H., Nichols, J.D., Simons, T.R., Fransworth, G.L., Bailey, L.L. & Sauer, J.R. (2002) Large scale wildlife monitoring studies: statistical methods for design and analysis. Environmetrics, 13, 105–119. Pulliam, H.R. (1988) Source, sinks, and population regulation. The American Naturalist, 132, 652–661. R Development Core Team (2004) R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing Vienna, Austria. Rajesh, G., Qureshi, Q., Bharadwaj, M., Singh, R.K.J. & Jhala, Y.V. (2010) Evaluating the status of the endangered tiger Panthera tigris and its prey in Panna Tiger Reserve, Madhya Pradesh, India. Oryx, 44, 383–398.

Rexstad, E. & Burnham, K.P. (1991) User’s Guide for Interactive Program CAPTURE. Colorado Cooperative Fish & Wildlife Research Unit, Colorado State University, Fort Collins, Colorado. Royle, J.A., Nichols, J.D., Karanth, K.U. & Gopalaswamy, A.M. (2009) A hierarchical model for estimating density in camera-trap studies. Journal of Applied Ecology, 46, 118–127. Sanderson, E., Forrest, J., Loucka, C., Ginsberg, J., Dinerstein, E., Seidensticker, J., Leimgruber, P., Songer, M., Heydlauﬀ, A., O’Brian, T., Bryja, G., Klenzendorf, S. & Wikramanayake, E. (2006) Setting Priorities for the Conservation and Recovery of Wild Tigers: 2005–2015. The Technical Assessment. Wildlife Conservation Society, World Wildlife Fund, Smithsonian, and Save the Tiger Fund, Washington, DC. Sharma, S., Jhala, Y. & Sawarkar, V.B. (2005) Identifying individual tigers from their pugmarks. Journal of Zoology, 267, 9–18. Sharma, R.K., Jhala, Y.V., Qureshi, Q., Vattakaven, J., Gopal, R. & Nayak, K. (2010) Evaluating capture-recapture population and density estimation of tigers in a population with known parameters. Animal Conservation, 13, 94–103. Skalski, J.R., Ryding, K.E. & Millspaugh, J.J. (2005) Wildlife Demography: Analysis of Sex, Age, and Count Data. Elsevier Academic Press, New York, USA. Smallwood, K.S. & Fitzhugh, E.L. (1995) A track count for estimating mountain lion Felis concolor californica population trend. Biological Conservation, 71, 251–259. Smith, J.L.D., McDougal, C.W. & Sunquist, M.E. (1987) Female land tenur system in tigers. Tigers of the World (eds R.L. Tilson & U.S. Seal), pp. 97– 109. Noyes Publication, New Jersey. Sokal, R.B. & Rohlf, F.J. (1995) Biometry, the Principle and Practice of Statistics in Biological Research. W. H. Freeman, New York, NY, USA. SPSS (2001) SPSS Inc. Chicago, IL. http://www.spss.com Stander, P.E. (1998) Spoor counts as indices of large carnivore populations: the relationship between spoor frequency, sampling eﬀort and true density. Journal of Applied Ecology, 35, 378–385. Stanley, T.R. & Burnham, K.P. (1999) A closure test for time-speciﬁc capture-recapture data. Environmental Ecological Statistics, 6, 197–209. Wang, S.W. & Macdonald, D.W. (2009) The use of camera traps for estimating tiger and leopard populations in the high altitude mountains of Bhutan. Biological Conservation, 142, 606–613. Wegge, P., Pokheral, C.P. & Jnawali, S.R. (2004) Eﬀects of trapping eﬀort and trap shyness on estimates of tiger abundance from camera trap studies. Animal Conservation, 7, 251–256. White, G.C. & Burnham, K.P. (1999) Program MARK: survival estimation from populations of marked animals. Bird Study, 46(Supplement), 120– 138. Whittingham, M.J., Stephens, P.A., Bradbury, R.B. & Freckleton, R. (2006) Why do we still use stepwise modeling in ecology and behavior? Journal of Animal Ecology, 75, 1182–1189. Wiewel, A.S., Clark, W.R. & Sovada, M.A. (2007) Assessing small mammal abundance with track-tube indices and mark-recapture population estimates. Journal of Mammalogy, 88, 250–260. Wikramanayaka, E.D., McKnight, M., Dinerstein, E., Joshi, A., Gurung, B. & Smith, D. (2004) Designing a conservation landscape for tigers in human-dominated environments. Conservation Biology, 18, 839–844. Williams, B.K., Nichols, J.D. & Conroy, M.J. (2002) Analysis and Management of Animal Populations. Academic Press, San Diego, CA, USA. Woodroﬀe, R. & Ginsberg, J.R. (1998) Edge eﬀects and the extinction of populations inside protected areas. Science, 280, 2126–2128. Received 21 March 2010; accepted 20 October 2010 Handling Editor: Phil Stephens

Supporting Information Additional Supporting Information may be found in the online version of this article. Appendix S1. Scatter plots of indices of tiger abundance versus Camera Trap Mark–Recapture Tiger Density. Appendix S2. Power analysis from Abundance Index Data.

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 14–24

24 Y. Jhala, Q. Qureshi & R. Gopal Fig. S1. Tiger pugmark sets encountered per kilometre walk plotted against tiger density (tigers 100 km)2) estimates obtained by camera traps using mark–recapture closed population estimators. Fig. S2. Tiger scats encountered per kilometre walk plotted against tiger density (tigers 100 km)2) estimates obtained by camera trap mark–recapture.

As a service to our authors and readers, this journal provides supporting information supplied by the authors. Such materials may be reorganized for online delivery, but are not copy-edited or typeset. Technical support issues arising from supporting information (other than missing ﬁles) should be addressed to the authors.

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 14–24

Journal of Applied Ecology 2011, 48, 228–237

doi: 10.1111/j.1365-2664.2010.01902.x

REVIEW

Preserving frugivorous birds in agro-ecosystems: lessons from Spanish olive orchards Pedro J. Rey* Department Biologı´a Animal, Vegetal y Ecologı´a, Universidad de Jae´n, E-23071 Jae´n, Spain

Summary 1. Frugivorous birds are a priority for conservation. They are experiencing the transformation of natural habitats to agro-ecosystems worldwide and some are taking advantage of agricultural production of ﬂeshy-fruited plants. However, the mechanisms through which some birds are able to thrive in agricultural landscapes while others become extinct are poorly known. 2. This manuscript provides an overview of avian frugivory in olive orchards, one of the principal agro-ecosystems in the Mediterranean region and an important winter refuge for birds. The hypothesis that Mediterranean avian frugivores are pre-adapted to olive orchards is used to consider potential constraints to bird occupation of wider agro-ecosystems. 3. Agricultural practices and artiﬁcial selection of fruit cause habitat and landscape simpliﬁcation and reduction of bird food resources in orchards, with resulting negative eﬀects on bird diet and body condition, as well as on bird abundance and diversity. Some of these constraints can be partially overcome by the presence of small hedgerows and copses in the agricultural landscape. 4. Five pre-adaptive features determine the successful occurrence of a bird species in olive orchards: (1) frugivorism intensity; (2) ability to track variation in fruit availability; (3) diet plasticity to cope with low fruit diversity and unbalanced food; (4) fruit handling plasticity to cope with oversized fruits; and (5) ecomorphology and foraging niche conditions which increase the ability to respond to habitat simpliﬁcation. 5. Synthesis and applications. Management practices have the potential to alleviate the constraints that habitat simpliﬁcation puts on bird movement and diet. Two inter-related approaches to management are proposed: increasing landscape and habitat diversity by the occurrence of hedgerows and forest remnants; and increasing food availability through reducing the use of pesticides and promoting fruit diversity in hedges. Key-words: agricultural landscape, bird pre-adaptation, blackcap, frugivorous bird conservation, fruit tracking, habitat structure simpliﬁcation, hedges, thrushes

Introduction The way in which animals and plants cope with new environments in a human-impacted world is of great interest for biodiversity management under global change scenarios (Pimm & Gittleman 1992; Benton, Vickery & Wilson 2003). The successful settlement of wildlife in croplands is of particular interest because major changes in these landscapes have led to the disappearance of many animal and plant species (Mas & Dietsch 2004). However, the mechanisms that permit some species to thrive in croplands remain largely unexplored. Many plants depend on a mutualistic association with frugivorous animals, hence frugivores are a conservation priority *Correspondence author. E-mail: [email protected]

(Cordeiro & Howe 2001; Tellerı´ a, Ramı´ rez & Pe´rez-Tris 2005). The use of ﬂeshy-fruited plants (Vitis, Rubus, Olea, Vaccinium, Prunus, Citrus, Coﬀea, etc.) by agriculture occurs worldwide. Vertebrates consume the fruits of the wild relatives of these crops and many of the agricultural croplands are used by frugivores. The successful establishment of frugivores in croplands can be interpreted in terms of pre-adaptation to these cropping conditions. Understanding these pre-adaptive processes is fundamental to understanding the requirements of these animals in new environments. The use of olive orchards as wintering quarters by frugivorous birds in the Mediterranean region (Rey 1993) provides an example of such pre-adaptive processes and their implications for frugivore conservation. The cultivated olive tree Olea europaea var. europaea was developed over millennia by human

2010 The Author. Journal of Applied Ecology 2010 British Ecological Society

Frugivorous birds in agro-ecosystems 229 selection of the wild olive Olea europaea var. sylvestris (Breton et al. 2006). Olive orchards occupy large areas of the Mediterranean Basin, with approximately 6 million hectares in Spain, Italy, Greece, Tunisia, France, Turkey, Israel, Morocco and Cyprus (Loussert & Brousse 1980; Junta de Andalucı´ a 2003). There are also 0Æ5 millions hectares of olive orchards in Portugal in areas under Mediterranean climate. The cultivated olive tree exists in the same ecological environment as the wild plant. The occurrence of Mediterranean avian frugivores in olive orchards is thus expected to mirror the relationships between frugivores and native ﬂeshy-fruited plants in natural Mediterranean habitats. Population regulation during winter may be especially important for many temperate migrant birds that, although widespread during the breeding season, assemble in restricted geographic areas during the winter. The loss of natural habitats and food supply that aﬀects local populations in wintering grounds can have disproportionate, large-scale eﬀects on breeding population dynamics (Webster et al. 2002; Bibby 2003). The Mediterranean region has a long history of natural habitat loss, transformation to agricultural landscapes and reduction of fruit supplies. Olive orchards therefore have the potential to act as alternative habitats for the conservation of frugivores in the region. This study reviews frugivory in olive orchards in southern Spain and makes comparison with frugivory in natural habitats in the same region, where numerous studies have been conducted in recent years (reviewed in Herrera 1995). Southern Spain forms the focus because (i) it contains the largest amount of olive cultivation in the world; and (ii) it is one of the major wintering grounds and migratory pathways for frugivorous–insectivorous birds of the Palearctic (Tellerı´ a, Asensio & Dı´ az 1999). In addition, major insights into frugivory in the Mediterranean Basin have come from studies in southern Spain. First, habitat structure and landscape heterogeneity is considered together with comparative information on abundance and diversity of frugivorous birds and fruits to formulate the hypothesis of Mediterranean avian frugivore pre-adaptation to olive orchards (Herrera 1983; Mun˜oz-Cobo 1987). Subsequently, the major pre-adaptive features which constrain the use of orchards by birds are identiﬁed and used to highlight mechanisms that limit the occupation of this cropland by different frugivores. Finally, the ﬁndings are used to suggest management practices to enhance frugivore diversity in this and other agro-ecosystems.

Physical characteristics and agricultural landscape structure of southern Spain Although distributed throughout the Mediterranean, olives groves occupy more than 2 400 000 ha in Spain and around 1 500 000 ha in Andalusia (southern Spain), where they are mainly located in the centre and northeast of the Guadalquivir River Valley and on the Betic Mountains hillsides (frequently up to 800 m.a.s.l.). The Guadalquivir Valley is a triangular area of 35 000 km2 located between the natural boundaries of Sierra

Morena in the north and the Betic Mountains in the south (Fig. S1, Supporting information). The valley has historically been used for intensive agriculture of olives, cereals, vineyards and sunﬂowers. The area is generally ﬂat, rising from sea level to 700 m near Sierras de Cazorla, Segura y Las Villas Natural Park where the Guadalquirvir River surfaces. The climate is Mediterranean with hot and dry summers and cool, humid winters. Mean annual temperatures range from 15 to 18Æ5 C, and annual precipitation is between 400 and 1020 mm. Although virtually eliminated from the area, the Mediterranean maquis associated with Quercus ilex (accompanied by Quercus suber in the southwest of the Valley and Q. faginea in wetter conditions) is the natural climax vegetation across the entire region (Aparicio 2008). The area under olive grove area retracted and expanded repeatedly during the twentieth century, but since 1995 it has expanded in southern Spain, with a total of 120 000 ha previously devoted to cereal and other crops now dedicated to olive plantation (Junta de Andalucı´ a 2003).

AGRICULTURAL LANDSCAPE STRUCTURE

In the central and upper Guadalquivir Valley (Co´rdoba and Jae´n provinces), olive groves form a monoculture that results in relatively homogeneous landscapes. Native vegetation represents <1% of the area of the Gudalquivir Valley (Aparicio 2008). Intensiﬁcation practices eliminated hedgerows and other live fences between properties, and the rare hedges still present are small, and often undetectable in the landscape. While woodlots and copses are extremely rare in lowlands, they are better represented in the olive groves on mountain hillsides. Topography is thus one of the rare sources of heterogeneity in the landscape and contributes to the diﬀerentiation between olive orchards from the valley and mountains. Other sources of heterogeneity in this agricultural landscape include diﬀerences between groves in herbaceous cover treatment (Valera et al. 1997) and number of trees per hectare (between 70 in old plantations and 400 in some young plantations; Mun˜ozCobo & Purroy 1980). These variations seem to more greatly aﬀect the breeding than the wintering communities, which are more inﬂuenced by food availability (Rey, Alca´ntara & Sa´nchez-Lafuente 1996; Valera et al. 1997; Mun˜oz-Cobo & Moreno-Montesino 2003). Finally, another source of variation is the diﬀerence between cultivars. Many olive cultivars are planted in Andalusia (Junta de Andalucı´ a 2003) but the most abundant are regionally separated (Fig. S1). Cultivars diﬀer in olive ripening phenology, harvesting timing, fruit crop and the extent of their inter-annual variation (Fig. S2, Supporting information). These diﬀerences generate a regional mosaic in olive availability with consequences for avian frugivore occurrence in the agricultural landscape.

Olive orchards as wintering grounds for European frugivores The Mediterranean environments in southern Europe and northern Africa are important wintering grounds for many

2010 The Author. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 228–237

230 P. J. Rey European birds, including frugivores (e.g. Tellerı´ a, Asensio & Dı´ az 1999; Tellerı´ a, Ramı´ rez & Pe´rez-Tris 2005). Frugivores are more abundant in wild olive forests (58–90 birds ⁄ 10 ha) and other lowland habitats dominated by ﬂeshy-fruited plants (Fig. 1a). However, loss of natural habitats and the decrease in fruit resources caused by unselective clearing of forest undergrowth (a major threat for biodiversity in the region; Andre´s & Ojeda 2002) has displaced frugivores to olive orchards during winter (Mun˜oz-Cobo & Purroy 1980; Sua´rez & Mun˜oz-Cobo 1984; Rey 1993). Frugivore abundance in orchards of Andalusia ranges between 23 and 55 birds ⁄ 10 ha (Rey 1993; Mun˜ozCobo et al. 2001). Thus, 3Æ5–8 million frugivores overwinter in olive orchards in southern Spain. In Andalusia, between three and ﬁve species of frugivorous birds commonly occur in lowlands olive orchards. Sylvia atricapilla Linnaeus, S. melanocephala Gmelin and Turdus philomelos Brehm regularly overwinter in this habitat, whereas

T. iliacus Linnaeus and Erithacus rubecula Linnaeus mostly appear in low densities early in winter. T. merula Linnaeus does not occur in lowland olive orchards but frequently occurs in orchards in mountain areas containing small remnants of native vegetation (Mun˜oz-Cobo et al. 2001; P.J. Rey, pers. obs.). Up to 11 frugivorous species can occur in wild olive scrublands and other lowland habitats (Fig. 1a). Locally, ﬂocks of Sturnus unicolor Temminck and S. vulgaris Linnaeus and the crows Corvus monedula Linnaeus, C. corone Linnaeus, C. corax Linnaeus, Pyrrhocorax pyrrhocorax Linnaeus and Cyanopica cyanus Pallas may also consume fruits in the three habitats (Jordano 1987a; Blanco, Fargallo & Cuevas 1994; Alca´ntara et al. 1997). Based on their inter-habitat variation (Fig. 1b,c; Table S1, Supporting Information), T. philomelos, T. iliacus and S. atricapilla are the least aﬀected by habitat changes, whereas T. merula, T. viscivorus Linnaeus, E. rubecula, Phoenicurus ochruros Gmelin, S. melanocephala, S. undata

(a) Total frugivores 200 180 160

Scrublands

12

Wild olive scrublands

10 8

Olive orchards 140

6

120

4 2

100

0

Num. frugivores species

80 60 40 20 0

Frugivore abundance

% Frugivores

(c)

60

6

50

5

40

4

30

3

20

2

10

1

un S.

cu

la

a be

er ul

E. ru

os

m

m

el

ha

ilo ph

ep

T.

T.

la

a ill ap S.

m

el

an

oc

ric at

da ta ili ac T. us vi sc iv or us C .c ya nu P. oc s hr St ur ur o nu s s sp p

0

0

S.

Minor frugivores

T.

Birds per 10 ha

(b) Major frugivores

Fig. 1. Frugivore assemblages in three lowland habitats of southern Spain. Data from sclerophyllous scrublands are from Jordano (1985, 1987a), Arroyo & Tellerı´ a (1983), Tellerı´ a, Ramı´ rez & Pe´rez-Tris (2005) and Alca´ntara et al. (1997). Data from wild olive scrublands are from Sua´rez & Mun˜oz-Cobo (1984) and Rey (1992, 1993). Data from olive orchards are from Sua´rez & Mun˜oz-Cobo (1984) and Rey (1992, 1993). Means and standard error bars are shown for: (a) data on frugivore assemblage abundance (birds ⁄ 10 ha), percentage of frugivores in the bird community, and number of species; (b) abundance of the most abundant frugivores; (c) abundance of the minor frugivores. 2010 The Author. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 228–237

Frugivorous birds in agro-ecosystems 231 Boddaert, C. cyanus and Garrulus glandarius Linnaeus are seriously aﬀected (Rey 1993). Some of these birds, like T. philomelos, are declining in Northern Europe (Peach, Robinson & Murray 2004), and olive orchards appear to represent an important winter surrogate habitat for turdids.

can support 6–10 diﬀerent species of fruiting plants. Fruit diversity increases further in other Mediterranean scrublands in southern Spain (Herrera 1984a; Jordano 1984) with 16–21 fruit species per site (Table S2).

AVAILABILITY OF ALTERNATIVE FOODS FOR BIRDS

Disparities and similarities between orchards and natural habitats HABITAT STRUCTURE

Olive orchards are mono-speciﬁc stands with olive trees uniformly separated (6–12 m between trees). Tree cover ranges between 5% in young plantations and 30% in old plantations. Scrub layer is predominantly lacking and the herbaceous layer is ephemeral in most groves due to the use of herbicides and ploughing (Valera et al. 1997). Their spatial heterogeneity and habitat structure are extremely simple compared to native scrublands and forests (Rey, Valera & Sa´nchez-Lafuente 1997).

Arthropod availability over the winter is higher in natural habitats than in olive orchards (Fig. 2), probably as a consequence of agricultural practices (pesticide, ploughing) and habitat simpliﬁcation.

NUTRIENT COMPOSITION OF THE FRUITS

Cultivated and wild olives have very high lipid contents that range from 40 to 66% of dry mass (Jordano 1987a; Rey 1992). Among the Mediterranean fruits, only Pistacia lentiscus L. (58Æ8%), Pistacia terebithus L. (55Æ6%) and Laurus nobilis L. (54Æ3%) have such a high lipid content (Herrera 1987). Thus, olives are among the most energy-rich fruits in the Mediterranean Basin.

FRUIT SIZE FRUIT ABUNDANCE

Cultivated olives are larger than the winter fruits of the Mediterranean region. The average width of the fruits in the scrublands of southern Spain is 6Æ9 mm for ﬂeshy fruits, 7Æ6 mm for berries with one to three seeds and 10Æ7 mm for berries with more than three seeds (Jordano 1984). The mean width of wild olives is 8Æ4 mm, whereas cultivated olives range from 12Æ7 to 17Æ8 mm (Rey & Gutie´rrez 1996; Rey et al. 1997). Therefore, the oversized fruits of cultivated trees do not ﬁt the gape of Mediterranean avian frugivores and suitable fruits are scarce compared to wild olives or other fruits (Table 1).

Production of fruit in the scrublands (Herrera 1984a; Jordano 1985, 1992) approaches the levels of orchards in terms of fruits ha–1, reaching a maximum of 3 · 106 (Mun˜oz-Cobo 1987). However, there is extreme inter-annual and among zones variation in fruit crop yield (Rey 1995; Fig. S2, Supporting information), which is related to supra-annual cycles of fruit production (Loussert & Brousse 1980). This is common in many scrublands in the Mediterranean region (Herrera 1984a, 1998), as well as in the wild olive fruit production (Jordano1987a; Alca´ntara et al. 1997).

DIVERSITY OF FLESHY FRUITS

FRUITING AND RIPENING PHENOLOGY

The diversity of fruit is poor in the olive orchard landscape because there are fewer hedges and copses next to streams supporting other fruit-bearing plants (Table S2, Supporting information).Wild olive-dominated scrublands are diverse and

Cultivated and wild olives have winter ripening phenologies (Fig. S2 Supporting Information), common to many Mediterranean fruits (Herrera 1984a; Jordano 1984; Alca´ntara et al. 1997). Their maximum ripe fruit availability (December to

Table 1. Body metrics of avian frugivores and associated fruit consumption and feeding behaviour. Predominant diet and % fruit in diet refer to frugivory in natural habitats. Fruit suitable to gape is stated in terms of percentage of fruits ﬁtting the bird’s gape. Frequency of pecking is the percentage of total olive consumption obtained through pecking. Swallowing and pecking energetic is the energy rendered by each feeding behaviour. Data from Rey & Gutie´rrez (1996, 1997); Rey et al. (1997)

Body mass (g) Gape width (mm) Predominant diet % Fruit in diet Olives suitable to gape Wild olives suitable to gape Frequency of olive pecking Pecking energetic (kJ s–1) Swallowing energetic (kJ s–1)

Sylvia atricapilla

Sylvia melanocephala

Sylvia undata

Turdus philomelos

Erithacus rubecula

18Æ5 8Æ5 Heavily frugivorous >70% 0–2% 60–88% >95% 0Æ03 ± 0Æ02 0Æ12 ± 0Æ04

11Æ5 7Æ1 Omnivorous 50–70% 0 13–58% >95% No data No data

9Æ1 5Æ6 Insectivorous <25% 0 <10% No data No data No data

70Æ0 13Æ7 Heavily frugivorous >70% 2–75% 100% <60% 0Æ05 ± 0Æ02 0Æ63 ± 0Æ08

17Æ0 8Æ0 Heavily frugivorous >70% 0 35–79% >95% No data No data

2010 The Author. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 228–237

Arthropods per dm2

232 P. J. Rey account for the majority of the frugivore diet (Tejero, Camacho & Soler 1983; Tejero, Soler & Camacho 1984; Soler, Tejero & Camacho 1988; Soler et al. 1988; Rey & Valera 1999). Some frugivorous birds thus appear to be pre-adapted to olive culture due to structural and fruit-related similarities between olive orchards and the native scrublands that frugivores naturally inhabit (Mun˜oz-Cobo 1987).

Ground 2 χ = 5·2, df = 2, P = 0·07

4

3

2 1

0

Arthropods per 5 min census

Nov

Dec

Jan

Feb

Mar

Principal branches

15

Z = 2·02, P = 0·04

Five pre-adaptive features inﬂuence the successful occupation of olive orchards by a bird species.

10

5

1. HIGH WINTER FRUGIVORISM

0

Nov

Dec

Jan

Feb

Mar

Secondary branches

Arthropods per 5 min census

Pre-adaptive features that favour the settlement of birds in olive orchards

15

Z = 2·02, P = 0·04

10

5

0

Nov

Dec

Jan

Feb

Mar

Fig. 2. Monthly abundance of arthropods in lowland habitats of southern Spain: olive orchards (open circle), wild olive scrublands (black squares) and other scrublands (black triangles). To compare monthly abundances, the results of Friedman anova (ground) and Wilcoxon test (tree branches) are shown. Data from olive orchards and wild olive scrubland are from Rey (1992) and Rey & Valera (1999). Data from other scrublands are from Jordano (1989). Only arthropod abundance on the ground is available from other scrublands. Arthropods on the ground were sampled with pitfalls (0Æ38 dm2) in orchards and wild olive scrublands, and with sticky plastic sheets in other scrublands. Arthropods on branches were sampled by ﬁve-minute censuses.

Frugivorism intensity (sensu Herrera 1984a,b; Jordano 1987c) of a bird species aﬀects the successful occupation of olive orchards. The relative decrease in abundance between orchards and natural habitats (Table S1) is signiﬁcantly related to percentage of fruit in the diet (r = )0Æ84, P = 0Æ005) and the frugivorism index (r = )0Æ69, P = 0Æ038). Hence, birds with greater frugivorism in their natural habitat will experience fewer diﬀerences in their population abundance between natural habitats and orchards. In contrast, insectivorous birds will be likely to experience greater decreases in abundance (r = 0Æ80, P = 0Æ01). This pattern is found even within the same genus. Among the three overwintering Sylvia species, there is a size gradation directly related to the frugivorism intensity (Jordano 1987b) and to the successful occupation of the orchards. Sylvia atricapilla (18Æ5 g of body weight) is heavily frugivorous, S. melanocephala (11Æ5 g) is omnivorous and S. undata (9Æ1 g) is mostly insectivorous (Table 1). Thus, S. atricapilla is abundant in olive orchards and in scrublands, whereas S. melanocephala is less abundant and S. undata is extremely rare in orchards (Fig. 1; Table S1).

2. FRUIT HANDLING PLASTICITY TO COPE WITH OVERSIZED FRUITS

February) matches the winter peak of frugivore abundance in the Mediterranean Basin (Jordano 1985; Rey 1995; Herrera 1998).

The pre-adaptation hypothesis The olive cultivation zones in the Mediterranean Basin were originally occupied by Mediterranean scrublands or forests. Autumn and winter lipid-rich fruits, such as Pistacia lentiscus, P. terebinthus, Olea europaea L., Viburnun tinus L., Phillyrea angustifolia L, Jasminum fruticans L. and Hedera helix L., are well represented in these plant communities (Herrera 1984a, 1995; Jordano 1984). Because of their high reward, these fruits are major components in the winter diet of the Mediterranean avian frugivores (Herrera 1981, 1984a; Jordano & Herrera 1981; Jordano 1987a,b, 1988, 1989). In the orchards, olives

The scarcity of fruits of suitable size for frugivores in olive orchards may limit their successful occupation of this habitat. Among those species overwintering in olive orchards, the largest species (T. philomelos, T. iliacus and S. atricapilla) survive better in orchards than smaller species (E. rubecula, S. melanocephala and S. undata) (Fig. 1; Table S1). Within the Sylvia genus, the larger species have greater gape width and are consistently more abundant in olive orchards. Olive orchards are clearly unsuitable winter habitats for S. undata and S. melanocephala, because of the limitations imposed by the size of the available fruit, but they are more appropriate for S. atricapilla. Some birds develop an opportunistic feeding behaviour (pecking olive pulp) to cope with the low availability of suitable fruits. Pecking is particularly frequent for S. atricapilla, but larger birds such as T. philomelos, and smaller birds such as E. rubecula and S. melanocephala, also peck fruit. The feeding

2010 The Author. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 228–237

Frugivorous birds in agro-ecosystems 233 Sylvia atricapilla

eﬃciency of this habit is less proﬁtable than swallowing whole fruit (Table 1) but it is still proﬁtable enough to defray the metabolic requirements of these birds (Rey & Gutie´rrez 1996, 1997; Rey et al. 1997). These birds have also been reported to peck fruit in cherry orchards (Herna´ndez 2008) and large wild olives and other fruits in natural habitats (Jordano 1987a; Alca´ntara et al. 1997). The capacity of some birds to peck as an alternative to swallowing the fruits is a pre-adaptive feature that allows frugivores to survive in croplands where fruit size has been increased by artiﬁcial selection.

100

1·30

1·37

2·14

NFPF Animal

80

Other fruits Olive

60 40 20 0

Sylvia melanocephala 100

3. OPPORTUNISM AND DIET PLASTICITY TO COPE WITH

0·92

1·25

1·83

1·60

80

LOW FRUIT DIVERSITY AND UNBALANCED FOODS

60 40

Percentage volume in diet

Birds have very diﬀerent diets in olive orchards and natural habitats (Fig. 3). In orchards, birds feed on fewer fruit species and the amount of fruit in the diet decreases in comparison to natural scrublands (Rey & Valera 1999). Consumption of arthropods is no lower in orchards relative to scrublands despite the decrease in arthropods in orchards (Fig. 3). It seems that frugivores actively search for arthropods to diversify their diet and to balance the intake of nutrients and energy. Most surprisingly, frugivores in olive orchards systematically include large amounts of leaves, ﬂowers and seeds of weeds in their winter diets (Fig. 3). Consumption of this suboptimal food suggests feeding plasticity to balance nutritional requirements, especially the need for minerals and micronutrients (Rey & Valera 1999).

1·02

20 0

Erithacus rubecula 100

0·99

1·33

1·42

1·36

80 60 40 20 0

Turdus philomelos 4. BIRD ECOMORPHOLOGY, FORAGING NICHE AND THE

100

ABILITY TO COPE WITH THE STRUCTURAL SIMPLICITY

80

OF THE ORCHARD HABITAT

60

Birds feed on olives from branches or on the ground (Rey & Gutie´rrez 1997) although this is limited by the birds’ external morphology. Thrushes and starlings are ground foragers (Smith 1974; Snow & Snow 1988) and consume olives on the ground; in contrast, Sylvia species are perching birds (Leisler 1980) and consume olives in the trees (Rey & Gutie´rrez 1997). Diﬀerences in perching ability associated with hind limb morphology have been described for Sylvia (Leisler 1980; Leisler & Winkler 1985; Jordano 1987c). S. atricapilla forages frequently in trees and is better adapted to perch than its congeners S. melanocephala and S. undata, which forage in medium and small shrubs (Cuadrado 1987; Jordano 1987b). Erithacus rubecula (less adapted than the Sylvia species to forage using perches) forages on the ground among the vegetation or consumes fruits by hovering (Snow & Snow 1988). Olive orchards thus attract fruit gleaning-perching birds and ground foragers but cannot support strict scrub foragers. Compared to scrublands, the probability of successful occupation of orchards is 0Æ75 for perching frugivores, 0 for strict scrub foragers and 0Æ33 for other frugivores (log-likelihood = )8Æ24, d.f. = 2, P = 0Æ03; N = 12 species; bird categorisation in Table S1). It seems that lack of shrubs in orchards directly constrains bird occurrence. The lack (T. merula), rarity (S. undata) or pronounced decline (E. rubecula and S. melanocepahla) of some Mediterranean

40

0·86

1·6

1·47

20 0

Olive Hedges Wildolive Other orchards in olive scrublands scrublands orchards

Fig. 3. Variation in winter diet of four avian frugivores in four lowland habitats of southern Spain: olive orchards, hedges in orchards, wild olive scrublands and other natural sclerophyllous scrublands. Numbers on each bar indicate number of fruit species per diet sample. NFPF refers to plant matter other than fruit. Data for olive orchards and wild olive scrublands are from Rey (1992), Rey, Alca´ntara & Sa´nchez-Lafuente (1996) and Rey & Valera (1999), complemented for some species with Soler et al. (1988), Soler, Tejero & Camacho (1988), Tejero, Camacho & Soler (1983) and Tejero, Soler & Camacho (1984). Data for other scrublands are from Jordano & Herrera (1981), Herrera (1984a), Jordano (1987a) for S. atricapilla and S. melanocephala, Herrera (1981) for E. rubecula, and Herrera (1984a) for T. philomelos.

frugivores in orchards is probably related to structural simpliﬁcation and lack of a shrub understorey in the orchards. Ecomorphological and niche constraints to bird occupation of fruit orchards are not exclusive of olive landscapes.

2010 The Author. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 228–237

234 P. J. Rey A paucity of understorey birds in fruit orchards, compared to remnants of native vegetation, has been reported in other regions and linked to insuﬃcient shrub understorey (Little & Crowe 1994; Round, Gale & Brockelman 2006).

5. THE ABILITY OF BIRDS TO TRACK SPATIOTEMPORAL VARIATION IN FRUIT AVAILABILITY

Olive groves oﬀer a continuum of olive supply across spatial and temporal scales. This continuum of ripe fruit availability is a consequence of the variation between diﬀerent regional zones in olive crops, ripening phenology and harvesting (Rey 1995), which are all linked to the peculiarities of the cultivars planted at each locality (Figs S1 and S2). By virtue of such a continuum, the olives become a predictable resource at a regional and landscape level each winter from November to March. In order to survive, birds must track the availability of ripe olives in space and time. The intrinsic variation (in phenology and location) between cultivars and the landscape continuity of the orchards are crucial for frugivore populations. Many studies have shown that frugivores are able to track spatio-temporal variations of fruit availability (Levey 1988; Mogenburg & Levey 2003) and that such variation becomes fundamental to survival in a region (see Tellerı´ a, Ramı´ rez & Pe´rez-Tris 2005; for implications on avian frugivore conservation in the Mediterranean Basin). Tracking is possible because some birds exhibit nomadism and regional migration (Levey & Stiles 1992). Sylvia atricapilla and some Turdus species (especially T. philomelos) have winter nomadic behaviour (Simms 1978; Debussche & Isenmann 1984; Tellerı´ a & Pe´rez-Tris 2003) that allows them to respond to spatial and temporal changes in fruit availability, and they have been found tracking olive availability at both small and regional scales (Rey 1995). By contrast, S. melanocephala, E. rubecula and T. merula do not display winter nomadism (Debussche & Isenmann 1984; for E. rubecula; Lundberg 1985; for T. merula; and Cantos 1992, for S. melanocephala) and do not track olive abundance in orchards. Thus, olive orchards are suitable habitats for frugivores with high nomadic movement and regional migration and unsuitable for frugivores with low nomadic capacity.

Managing frugivore biodiversity in Spanish olive orchards Pre-adaptive features are fundamental to the successful occurrence of some birds in olive orchards during their winter stay in the Mediterranean region. Although olive orchards diﬀer from natural habitats within the region, they still maintain structural and functional similarities. Interestingly, most of these pre-adaptive features are properties of the most abundant frugivores (S. atricapilla and T. philomelos) but are not as pronounced in less frequent frugivores. Orchard management for increasing frugivore diversity should encourage habitat diversity to facilitate bird movement, diet diversiﬁcation, fruit foraging and niche requirements. It is suggested that this can be achieved by two inter-related approaches to management:

1. MANAGEMENT OF THE OLIVE ORCHARDS STRUCTURE: TOWARDS AN INCREASED HETEROGENEITY

Heterogeneity should be managed at several spatial scales (Benton, Vickery & Wilson 2003). At small scales, the vertical and horizontal structural heterogeneity may be enhanced by extending the practice of separating olive orchards with hedges. This would provide a shrub layer for those frugivores for which this is a fundamental niche requirement (providing protection, foraging sites and movement facilities). Mist-netting in hedges demonstrates that 2–5 m-width hedges are enough to increase the local occurrence of some species (Fig. S3 Supporting information). At the landscape scale, hedgerows, rocky outcrops, copses, and stream vegetation belts should be promoted. All these elements diversify the olive orchard landscape, create heterogeneity and complexity, and connect the landscape for nonnomadic birds. At the regional scale, the range of distinct, locally predominant olive cultivars must be maintained, since diﬀerences in ripening of diﬀerent cultivars increases the timeframe during which olives are available to birds. The homogenisation of orchards to only a few cultivars would compromise frugivore overwintering in the region.

2. MANAGEMENT OF THE FOOD AVAILABILITY

Hedgerows should be planted with native species that produce ﬂeshy fruits to diversify the fruit resources. The occurrence of wild fruits in remnant hedges around olive orchards makes these orchards more similar to natural habitats in fruit diversity and phenology. Hedgerows and other sources of fruit (e.g. groups of isolated fruiting trees) have been found to maintain frugivory in agricultural landscapes worldwide (Hinsley & Bellamy 2000; Croxton & Sparks 2004). Hedges in the agricultural landscape of southern Spain may contain not only winter but also summer and autumn fruits that would help support other migrant birds that display summer and autumn frugivory and use native habitats, but not olive orchards, as stopover sites. Hedges also increase the abundance of arthropods and other food resources, as demonstrated in many agricultural landscapes (Thomas & Marshall 1999; Pollard & Holland 2006). The increase of organic agricultural practices in the region, especially reducing the use of pesticides, should be highly beneﬁcial for winter frugivores by increasing arthropod availability as a food resource. A reduction in the use of pesticides has been shown to enhance bird diversity in fruit farmlands in the Mediterranean (Genghini, Gellini & Gustin 2006) and other regions (Chamberlain, Wilson & Fuller 1999; Freemark & Kirk 2001).

Generalizations to olive landscapes in the Mediterranean Basin Much of the patterns of habitat simpliﬁcation and homogenization described for olive landscapes in southern Spain are common to other olive landscapes in the Mediterranean. Agricultural intensiﬁcation has been common in Mediterranean Europe since the 1950s (Kizos & Koulouri 2006),with

2010 The Author. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 228–237

Frugivorous birds in agro-ecosystems 235 increasing mechanization, removal of natural vegetation, fertilization and the use of pesticides, all resulting in an overall decrease in fruit supplies and other resources for birds (Genghini, Gellini & Gustin 2006). Other olive cultivation zones are also experiencing similar phenomena. For example, Santos & Cabral (2003) showed detrimental eﬀects of olive cultivation intensiﬁcation for several passerine guilds in Portugal. Thus, the recommendations to promote landscape heterogeneity and diversiﬁcation of food resources are also applicable to most olive cultivation zones of the Mediterranean Basin. However, substantial diﬀerences exist in the extent and continuity of olive groves. In France, Italy and Greece, olive groves are frequently interspersed with other crops, producing agricultural mosaics. Live fences (hedgerows and vegetated stonewalls) and woodlots are better conserved in many of the agricultural landscapes in these countries compared to Mediterranean Spain (Genghini, Gellini & Gustin 2006; Kizos & Koulouri 2006) and, consequently, there is a greater diversity of food sources, including fruits. Ancient olive plantations in some regions of Italy and Greece have been maintained and provide refuge for biodiversity. There are action plans for the protection of this habitat (LIFE07 NAT ⁄ IT ⁄ 000450), but no similar conservation strategies are being developed in Spain. The habitat simpliﬁcation scenario is to some extent diﬀerent for many olive cultivation zones of North Africa, where mechanization and use of pesticides is less common and the olive trees are frequently interspersed with other cultivated trees and some natural vegetation (P. J. Rey personal observation). No detailed information on bird abundance and frugivory in olive landscapes exists outwith Spain, but it would be expected that such scenarios would be favourable to frugivore conservation. However, the expansion and intensiﬁcation of olive cultivation that has happened in North Africa in the last two decades is putting pressure on these heterogeneous landscapes and it would be expected that this would have adverse impacts on wintering populations of Mediterranean frugivores.

Lessons from olive orchards. Comparison with other fruit croplands A fundamental lesson from studies of frugivory in Spanish olive orchards is that the cultivation of fruit crops derived from native instead of exotic plant species will better preserve the original animal biodiversity of the region. Such agricultural landscapes maintain some of the structural and functional (the plant–animal interactions) properties of the natural habitats to which animals are adapted. On the other hand, it is important to acknowledge that diﬀerent bird species have diﬀerent pre-adaptive features that will enable them to thrive in agroecosystems. Most fruit croplands of the world are aﬀected by intensiﬁcation, landscape and habitat structural simpliﬁcation and human selection of fruit size. As a result there are often food shortages for frugivores (e.g. Little & Crowe 1994; Nelson et al. 2000) similar to those described in olive orchards. It would therefore be expected that the pre-adaptive features inﬂuencing bird diversity in olive orchards will also be relevant in other fruit production systems.

This review demonstrates that modiﬁcations of the agricultural practices in a region can easily make agro-ecosystems more suitable to a greater number of bird species by incorporating features that will favour birds poorly adapted to croplands (Benton, Vickery & Wilson 2003). In particular, hedgerows are known to have positive eﬀects on local bird diversity in other agro-ecosystems (e.g. Hinsley & Bellamy 2000; Herzon & O’Hara 2007). Likewise, the retention of natural elements in the agricultural landscape has been shown to increase bird diversity in many other regions (e.g. Haslem & Bennett 2008a,b). These natural elements connect the landscape (Hinsley & Bellamy 2000; Donald & Evans 2006) by acting as natural corridors and also provide additional food resources, which are fundamental to a diverse diet. Finally, we can make tentative generalizations from the comparison of olive orchards with other fruit croplands claimed as important reservoirs for biodiversity. Rustic (shade) coﬀee plantations in Central America have repeatedly been proposed as functional surrogates of the tropical forest for biodiversity (reviewed in Philpott et al. 2008). Coﬀee plantations are exotic in these areas, but their function for biodiversity is achieved from the structural and taxonomical similarities with tropical forests due to the species that provide shade for coﬀee production in rustic plantations. Coﬀee plantations have become fundamental as winter refuges and stopover sites for Neotropical migrant birds because their structural complexity and taxonomical diversity provide suitable food sources and niche requirements. However, modern sun plantations are structurally and taxonomically simpliﬁed, mirroring to some extent some phenomena occurring in olive orchards. Most native plant species are removed leading to habitat homogenization, reductions of insects and fruits (food supplies for birds), and a concomitant reduction of bird biodiversity. Unlike olive cultivation zones, however, there is increasing awareness of the importance of bird conservation in agricultural landscapes of the Neotropics. The repercussions for biodiversity of diﬀerent management regimes in coﬀee plantations are being thoroughly investigated in these systems (Philpott et al. 2008). There are rigorous certiﬁcation protocols for deﬁning biodiversity-friendly practices as standbrand for the coﬀee market (Mas & Dietsch 2004). Similar certiﬁcation programmes for olive production should be encouraged to conserve frugivorous ⁄ insectivorous European migrant birds in their winter Mediterranean quarters.

Future directions Research programmes should be implemented to improve our knowledge of the role of diﬀerent agricultural practices (organic vs. integrated and conventional farming) in maintaining functional biodiversity as well as landscape properties in olive orchards. A detailed knowledge is needed of landscape structure, the occurrence of natural elements in the landscape, the connectivity of these elements, their function and their relationship to biodiversity. Future studies should involve experimental trials combining diﬀerent hedgerows settings and diﬀerent amounts of herbaceous cover. These investigations should consider the eﬀects of biodiversity-friendly practices on

2010 The Author. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 228–237

236 P. J. Rey crops and the economics of crop reduction and subsidy inputs after achieving the eco-conditionality criteria of European Commission regulations. Avian studies in orchards need to be extended to olive cultivation zones in North Africa and the Eastern Mediterranean to allow a complete picture to be formed of the role of olive orchards as reservoirs for frugivorous ⁄ insectivorous European birds.

Acknowledgements I thank comments from F. Valera, J. E. Gutie´rrez and three anonymous referees. R. Zamora encouraged me to write this review. While writing this paper I was funded by project CGL2006-02848 ⁄ BOS of the Spanish MEC and FEDER.

References Alca´ntara, J.M., Rey, P.J, Valera, F., Sa´nchez-Lafuente, A.M. & Gutie´rrez, J.E. (1997) Habitat alteration and plant intra-speciﬁc competition for seed dispersers. An example with Olea europaea var. sylvestris. Oikos, 79, 291– 300. Andre´s, C. & Ojeda, F. (2002) Eﬀects of aﬀorestation with pines on woody plant diversity of Mediterranean heathlands in southern Spain. Biodiversity and Conservation, 11, 1511–1520. Aparicio, A. (2008) Descriptive analysis of the ‘relictual’ Mediterranean landscape in the Guadalquivir River valley (southern Spain): a baseline for scientiﬁc research and the development of conservation action plans. Biodiversity and Conservation, 17, 2219–2232. Arroyo, B. & Tellerı´ a, J.L. (1983) La invernada de las aves en el a´rea de Gibraltar. Ardeola, 30, 23–31. Benton, T.G, Vickery, J.A. & Wilson, J.D. (2003) Farmland biodiversity: is habitat heterogeneity the key? Trends in Ecology & Evolution, 18, 182–188. Bibby, C.J. (2003) Conservation of migratory birds. Avian Migration (eds) P. Berthold, E. Gwinner & E. Sonnenschein), pp. 407–420. Springer, Berlin. Blanco, G., Fargallo, J.A. & Cuevas, J.A. (1994) Consumption rates of olives by choughs in central Spain. Variations and importance. Journal of Field Ornithology, 65, 482–489. Breton, C., Medail, F., Pinatel, C. & Berville, A. (2006) From olive to oleaster: origin and domestication of Olea europaea L. in the Mediterranean basin. Cahiers Agricultures, 15, 329–336. Cantos, F. (1992). Migracio´n e invernada de la Familia Sylvidae (Orden Paseriformes, Clase Aves) en la Penı´nsula Ibe´rica. PhD thesis, Universidad Complutense. Chamberlain, D.E., Wilson, J.D. & Fuller, R.J. (1999) A comparison of bird populations on organic and conventional farm systems in southern Britain. Biological Conservation, 88, 307–320. Cordeiro, N.J. & Howe, H.F. (2001) Low recruitment of trees dispersed by animals in African forest fragments. Conservation Biology, 15, 1733–1741. Croxton, P.J. & Sparks, T.H. (2004) Timing of berry depletion rates of three common hedgerow shrubs. Agriculture, Ecosystems & Environment, 104, 663–666. Cuadrado, M. (1987) Winter foraging behaviour of Blackcap and Sardinian warbler in a Mediterranean scrubland. Ardea, 76, 107–110. Debussche, M. & Isenmann, P. (1984) Origine et nomadisme des fauvettes a tete´ noire (Sylvia atricapilla) hyvernant et zone mediterrane´ene francaise. L’Oiseau et R.F.O., 54, 101–107. Donald, P.F. & Evans, A.D. (2006) Habitat connectivity and matrix restoration: the wider implications of agri-environment schemes. Journal of Applied Ecology, 43, 209–218. Freemark, K.E. & Kirk, D.A. (2001) Birds on organic and conventional farms in Ontario: partitioning eﬀects of habitat and practices on species composition and abundance. Biological Conservation, 101, 337–350. Genghini, M., Gellini, S. & Gustin, M. (2006) Organic and integrated agriculture: the eﬀects on bird communities in orchard farms in northern Italy. Biodiversity and Conservation, 15, 3077–3094. Haslem, A. & Bennett, A.F. (2008a) Birds in agricultural mosaics: the inﬂuence of landscape pattern and countryside heterogeneity. Ecological Applications, 18, 185–196. Haslem, A. & Bennett, A.F. (2008b) Countryside elements and the conservation of birds in agricultural environments. Agricultural, Ecosystem & Environment, 125, 191–203.

Herna´ndez, A. (2008) Cherry removal by seed-dispersing mammals: mutualism through commensal association with frugivorous birds. Polish Journal of Ecology, 56, 127–138. Herrera, C.M. (1981) Fruit food of Robins wintering in southern Spanish mediterranean scrubland. Bird Study, 28, 115–122. Herrera, C.M. (1983) Coevolucio´n de plantas y frugı´ voros: la invernada mediterra´nea de algunos paseriformes. Alytes, 1, 177–189. Herrera, C.M. (1984a) A study of avian frugivores, bird-dispersed plants, and their interaction in Mediterranean scrublands. Ecological Monographs, 54, 1–23. Herrera, C.M. (1984b) Adaptation to frugivory of mediterranean avian seed dispersers. Ecology, 65, 609–617. Herrera, C.M. (1987) Vertebrate-dispersed plants of the Iberian Peninsula: a study of fruits characteristics. Ecological Monographs, 57, 305–331. Herrera, C.M. (1995) Plant-vertebrate dispersal systems in the Mediterranean: ecological, evolutionary, and historical determinants. Annual Review of Ecology and Systematics, 26, 705–727. Herrera, C.M. (1998) Long-term dynamics of Mediterranean frugivorous birds and ﬂeshy-fruits: a 12-year study. Ecological Monographs, 68, 511– 538. Herzon, I. & O’Hara, R.B. (2007) Eﬀects of landscape complexity on farmland birds in the Baltic States. Agriculture, Ecosystems & Environment, 118, 297– 306. Hinsley, S.A. & Bellamy, P.E. (2000) The inﬂuence of hedge structure, management and landscape context on the value of hedgerows to birds: a review. Journal of Environmental Management, 60, 33–49. Jordano, P. (1984) Relaciones entre Plantas y Aves Frugı´ voras en el Matorral Mediterra´neo del A´rea de Don˜ana. PhD thesis, Universidad de Sevilla. Jordano, P. (1985) El ciclo anual de los paseriformes frugı´ voros en el matorral mediterra´neo, del sur de Espan˜a: Importancia de su invernada y variaciones interanuales. Ardeola, 32, 69–94. Jordano, P. (1987a) Avian fruit removal: eﬀects of fruit variation, crop size, and insect damage. Ecology, 68, 1711–1723. Jordano, P. (1987b) Frugivory, external morphology and digestive system in mediterranean sylviid warblers Sylvia spp. Ibis, 129, 175–189. Jordano, P. (1987c) Nota sobre la dieta no insectı` vora de algunos Muscicapidae. Ardeola, 34, 89–98. Jordano, P. (1988) Diet, fruit choice and variation in body condition of frugivorous warblers in mediterranean scrubland. Ardea, 76, 193–209. Jordano, P. (1989) Variacio´n de la dieta frugı´ vora oton˜o-inviernal del petirrojo (Erithacus rubecula): Efectos sobre la condicio´n corporal. Ardeola, 36, 161– 184. Jordano, P. (1992) Fruits and frugivory. Seeds. The Ecology of Regeneration in Plant Communities (ed. M. Fenner), pp. 125–166. C.A.B. International, Wallingford. Jordano, P. & Herrera, C.M. (1981) The frugivorous diet of Blackcap populations Sylvia atricapilla wintering in southern Spain. Ibis, 123, 502–507. Junta de Andalucı´ a (2003) El olivar andaluz. Consejerı´ a de Agricultura y Pesca, Sevilla. Kizos, T. & Koulouri, M. (2006) Agricultural landscape dynamics in the Mediterranean: Lesvos (Greece) case study using evidence from the last three centuries. Environmental Science & Policy, 9, 330–342. Leisler, B. (1980) Morphological aspects of ecological organization in bird genera. Okol Vogel, 2, 199–220. Leisler, B. & Winkler, H. (1985) Ecomorphology. Current Ornithology, Vol. 2 (ed. R.E. Johnston), pp. 155–186. Kluwer, Dordrecht. Levey, D.J. (1988) Spatial and temporal variation in Costa Rica fruit and fruiteating bird abundance. Ecological Monographs, 58, 251–269. Levey, D.J. & Stiles, F.G. (1992) Evolutionary precursors of long-distance migration: resource availability and movement patterns in Neotropical landbirds. American Naturalist, 140, 447–476. Little, R.M. & Crowe, T.M. (1994) Conservation implications of deciduous fruit farming on birds in the elgin district, Western Cape province, SouthAfrica. Transactions of the Royal Society of South Africa, 49, 185–198. Loussert, R. & Brousse, G. (1980) El olivo. Editorial Mundiprensa, Madrid. Lundberg, P. (1985) Dominance, behaviour, body weight and fat variations, a partial migration in European blackbirds Turdus merula. Behaviour Ecology and Sociobiology, 17, 185–189. Mas, A.H. & Dietsch, T.V. (2004) Linking shade coﬀee certiﬁcation to biodiversity conservation: butterﬂies and birds in Chiapas, Mexico. Ecological Applications, 14, 642–654. Mogenburg, S.M. & Levey, D.J. (2003) Do frugivores respond to fruit harvest? An experimental study of short-term responses. Ecology, 84, 2600–2612. Mun˜oz-Cobo, J. (1987) Las comunidades de aves de los olivares de Jae´n. PhD thesis, Universidad Complutense.

2010 The Author. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 228–237

Frugivorous birds in agro-ecosystems 237 Mun˜oz-Cobo, J. & Moreno-Montesino, J. (2003) Uso del agroecosistema olivar por las aves. (II) Variables estructurales en la estacio´n oton˜o-invernal. Boletı´n de Sanidad Vegetal: Plagas, 29, 171–183. Mun˜oz-Cobo, J. & Purroy, F.J. (1980) Wintering bird communities in olive tree plantations of Spain. Proceedings of VI International Conference on Bird Census Work, Gottingen, pp. 185–199. Mun˜oz-Cobo, J., Moreno-Montesino, J., Romero, C. & Ruiz, M. (2001) Ana´lisis cualitativo y cuantitativo de las comunidades de aves en cuatro tipos de olivares de Jae´n. (II) Comunidades oton˜ales e invernantes. Boletı´n de Sanidad Vegetal: Plagas, 27, 127–141. Nelson, S.L., Miller, M.A., Heske, E.J. & Fahey, G.C. (2000) Nutritional consequences of a change in diet from native to agricultural fruits for the Samoan fruit bat. Ecography, 23, 393–401. Peach, W.J, Robinson, R.A. & Murray, K.A. (2004) Demographic and environmental causes of the decline of rural Song Thrushes Turdus philomelos in lowland Britain. Ibis, 146, 50–59. Philpott, S.M., Arendt, W.J., Armbrecht, I., Bichier, P., Diestch, T.V. & Gordon, C. (2008) Biodiversity Loss in Latin American Coﬀee Landscapes: Review of the Evidence on Ants, Birds, and Trees. Conservation Biology, 22, 1093–1105. Pimm, S.L. & Gittleman, J.L. (1992) Biological diversity: where is it? Science, 225, 940. Pollard, K.A. & Holland, J.M. (2006) Arthropods within the woody element of hedgerows and their distribution pattern. Agricultural and Forest Entomology, 8, 203–211. Rey, P.J. (1992) Preadaptacio´n de la avifauna frugı´vora invernante al cultivo del olivar. PhD thesis, Universidad de Granada. Rey, P.J. (1993) The role of olive orchards in the wintering of frugivorous birds in Spain. Ardea, 81, 151–160. Rey, P.J. (1995) Spatio-temporal variation in fruit and frugivorous birds abundance in olive plantations. Ecology, 76, 1625–1635. Rey, P.J., Alca´ntara, J.M & Sa´nchez-Lafuente, A.M. (1996) Temporal variation in food availability and diet of Blackcaps in olive orchards. Journal of Field Ornithology, 67, 592–603. Rey, P.J. & Gutie´rrez, J.E. (1996) Pecking of olives by frugivorous birds: a shift in feeding behaviour to overcome gape limitation. Journal of Avian Biology, 27, 327–333. Rey, P.J. & Gutie´rrez, J.E. (1997) Eleccio´n de fruto y conducta de alimentacio´n de aves frugı´ voras en olivares y acebuchares: una estrategia o´ptima basada en la razo´n beneﬁcio ⁄ tiempo de manipulacio´n. Ardeola, 44, 27–39. Rey, P.J. & Valera, F. (1999) Diet plasticity in blackcap (Sylvia atricapilla): the ability to overcome nutritional constraints imposed by agricultural intensiﬁcation. Ecoscience, 6, 429–438. Rey, P.J., Valera, F. & Sa´nchez-Lafuente, M. (1997) Avifauna reproductora y estructura del ha´bitat en la campin˜a y sierras Subbeticas de Jae´n. Don˜ana Acta Vertebrata, 24, 115–142. Rey, P.J., Gutie´rrez, J.E., Alca´ntara, J.M. & Valera, F. (1997) Fruit size in wild olives: Implications for avian seed dispersal. Functional Ecology, 11, 611– 618. Round, P.D., Gale, G.A. & Brockelman, W.Y. (2006) A comparison of bird communities in mixed fruit orchards and natural forest at Khao Luang, southern Thailand. Biodiversity and Conservation, 15, 2873–2891. Santos, M. & Cabral, J.A. (2003) Development of a stochastic dynamic model for ecological indicators’ prediction in changed Mediterranean agroecosystems of north-eastern Portugal. Ecological Indicators, 3, 285–383. Simms, E. (1978) British Thrushes. Collins, London. Smith, J.N.M. (1974) The food searching behaviour of two European thrushes I. Description and analyses of search paths. Behaviour, 48, 276– 302. Snow,B.K.&Snow,D.W.(1988)BirdsandBerries.T.andA.D.Poyser,Calton. Soler, M., Tejero, E. & Camacho, I. (1988) Alimentacio´n de la Curruca capirotada (Sylvia atricapilla) en olivares de la provincia de Jae´n durante el periodo de oton˜o-invierno. Alytes, 4, 93–104.

Soler, M., Pe´rez-Gonza´lez, J.A., Tejero, E. & Camacho, I. (1988) Alimentacio´n del Zorzal alirrojo (Turdus iliacus) durante su invernada en olivares de Jae´n (Sur de Espan˜a). Ardeola, 35, 183–196. Sua´rez, F. & Mun˜oz-Cobo, J. (1984) Comunidades de aves invernantes en cuatro medios diferentes de la provincia de Co´rdoba. Don˜ana Acta Vertebrata, 11, 45–63. Tejero, E., Camacho, I. & Soler, M. (1983) La alimentacio´n de la curruca cabecinegra (Sylvia melanocephala) en olivares de la provincia de Jae´n (oton˜o-invierno). Don˜ana Acta Vertebrata, 10, 133–153. Tejero, E., Soler, M. & Camacho, I. (1984) Alimentacio´n del zorzal comu´n (Turdus philomelos) en olivares de la provincia de Jae´n (oton˜o invierno). Anales I.N.I.A. Servicio Forestal, 8, 9–32. Tellerı´ a, J.L., Asensio, B. & Dı´ az, M. (1999) Aves Ibe´ricas. Vol. 2. Passeriformes. J. M. Reyero Editor, Madrid. Tellerı´ a, J.L. & Pe´rez-Tris, J. (2003) Seasonal distribution of a migratory bird: eﬀects of local and regional resource tracking. Journal of Biogeography, 36, 1583–1591. Tellerı´ a, J.L., Ramı´ rez, A. & Pe´rez-Tris, J. (2005) Conservation of seed-dispersing migrant birds in Mediterranean habitats: shedding light on patterns to preserve processes. Biological Conservation, 124, 493–502. Thomas, C.F.G. & Marshall, E.J.P. (1999) Arthropod abundance and diversity in diﬀerently vegetated margins of arable ﬁelds. Agriculture, Ecosystems & Environment, 72, 131–144. Valera, F., Rey, P.J., Sa´nchez-Lafuente, A.M. & Alca´ntara, J.M. (1997) Efecto de los sistemas de laboreo sobre las aves. Agricultura de Conservacio´n: Fundamentos agrono´micos, Medioambientales y Econo´micos (eds L. Garcı´ a-Torres & P. Gonza´lez-Ferna´ndez), pp. 225–242. AELC ⁄ SV, Co´rdoba. Webster, M.S., Marra, P.P., Haig, S.M., Bensch, S. & Holmes, R.T. (2002) Links between worlds: unraveling migratory connectivity. Trends in Ecology & Evolution, 17, 76–83. Received 5 March 2010; accepted 20 October 2010 Handling Editor: Davy McCracken

Supporting Information Additional Supporting Information may be found in the online version of this article. Table S1. Foraging niche and habitat requirement of fruit-eating birds of this study. Table S2. Fruit species appearing in diﬀerent lowland habitats from southern Spain.

Fig. S1. Distribution of main olive cultivars in Andalusia. Fig. S2. Crop, fruiting and harvesting phenology in olive orchards. Fig. S3. Mist-netting capture rates of birds in olive orchards. As a service to our authors and readers, this journal provides supporting information supplied by the authors. Such materials may be reorganized for online delivery, but are not copy-edited or typeset. Technical support issues arising from supporting information (other than missing ﬁles) should be addressed to the authors.

2010 The Author. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 228–237

Journal of Applied Ecology 2011, 48, 257–264

doi: 10.1111/j.1365-2664.2010.01903.x

Lucerne-dominated ﬁelds recover native grass diversity without intensive management actions Pe´ter To¨ro¨k1*, Andra´s Kelemen1, Orsolya Valko´1, Bala´zs Dea´k2, Bala´zs Luka´cs2 and Be´la To´thme´re´sz1 1

Department of Ecology, University of Debrecen, PO Box 71, H-4010 Debrecen, Hungary; and 2Hortoba´gy National Park Directorate, Sumen u´t 2, H-4024 Debrecen, Hungary

Summary 1. Spontaneous succession is often underappreciated in restoration after the cessation of intensive agricultural management. Spontaneous succession could improve the success of restoration programmes, and oﬀers a cost-eﬀective option with little active intervention. 2. We studied the spontaneous recovery of loess grasslands in extensively managed lucerne Medicago sativa ﬁelds mown twice a year using space for time substitutions to highlight the importance of spontaneous processes in grassland restoration. 3. With increasing ﬁeld age a gradual replacement of lucerne by perennial native grasses and forbs and increase of mean species richness was detected. As the age of ﬁelds increased, lucerne decreased from 75% to 2% of total vegetation cover, whereas perennial graminoids increased from 0Æ5 to 50% cover. Mean total cover showed no signiﬁcant diﬀerences between the age groups; weed cover was less than 10%. 4. The phytomass of lucerne was negatively correlated with graminoid phytomass. As the age of the ﬁelds increased, lucerne phytomass decreased and grass phytomass increased. We found a negative correlation between litter and forb phytomass but there was no relationship with the age of the ﬁeld. There was no litter accumulation and no increase of mean total phytomass as the age of ﬁelds increased. 5. Synthesis and applications. Native grasses within loess grasslands recovered within 10 years, but characteristic native forbs remained rare. The advantages of spontaneous succession in lucerne ﬁelds compared to technical reclamation include: (i) no early stages dominated by weeds, (ii) minimal litter accumulation, (iii) a spontaneous decrease in lucerne over time, and (iv) negligible cost. In addition, the requirement for twice yearly mowing in the early years will guarantee farmer involvement because of the high forage value of lucerne. The complete restoration of species rich grasslands will require more active management such as propagule transfer by hay and ⁄ or moderate grazing to encourage the return of native forbs. Key-words: alfalfa, Medicago sativa, old ﬁeld, phytomass, space for time substitution, succession, weed control

Introduction The aim of grassland restoration is to recover and ⁄ or improve grassland biodiversity and ecosystem functions (Firn 2007; Reid et al. 2009). Two contrasting approaches are used most often: technical reclamation or spontaneous succession (Prach & Hobbs 2008). Both methods are generally followed up by site management for weed suppression using techniques such as mowing and ⁄ or grazing (Warren, Christal & Wilson 2002; Lepsˇ et al. 2007; Kiehl et al. 2010). Recovery *Correspondence author. E-mail: [email protected]

can be accelerated and directed by technical reclamation methods. In most cases this means adding seeds of desirable species using hay transfer or seed sowing (Pywell et al. 2002; Ho¨lzel & Otte 2003). An alternative approach is spontaneous succession, where seeds are not added and the system is left to recover naturally (Prach & Pysˇ ek 2001). Technical reclamation is preferred worldwide despite several promising examples of spontaneous recovery of grasslands (e.g. Ruprecht 2006; Prach & Rehounkova´ 2008). This is especially true when there is an urgent need to heal landscape scars, prevent erosion or suppress weeds (To¨ro¨k et al. 2010; Tropek et al. 2010).

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society

258 P. To¨ro¨k et al. Recently, there have been attempts to link theories of spontaneous succession with direct restoration eﬀorts to mitigate costs and improve the success of restoration (del Moral, Walker & Bakker 2007; Walker, Walker & del Moral 2007). For example, patterns in vegetation dynamics could be used to judge whether or not invasive weed cover will develop rapidly after agriculture ceases or to judge whether active intervention is necessary to eliminate former crops. Spontaneous succession has several advantages over technical reclamation. (i) The natural value of spontaneously regenerated sites is often higher than that of reclaimed ones (Hodacˇova´ & Prach 2003). (ii) Spontaneously colonising species are expected to be better adapted to local conditions than species originating from commercial sources or non-local sites (Mijnsbrugge, Bischoﬀ & Smith 2010). (iii) Increased vegetation patchiness at spontaneously regenerated sites provides improved refugees for animals compared to technical reclamation sites (Tropek et al. 2010). Finally, (iv) spontaneous succession oﬀers cost-eﬀective restoration with a low rate of active intervention (Prach & Hobbs 2008). Spontaneous succession also has some drawbacks compared to technical reclamation, concerning (i) the low level of predictability and control of initial vegetation composition, density and pattern, and (ii) the relatively slow development of vegetation towards to the target state, especially where proper donor sites for colonisation are missing (Ruprecht 2006; Prach & Hobbs 2008). However, the value of spontaneous succession in restoration programmes is becoming more widely appreciated, which underlines the importance of reporting relevant case (Prach & Pysˇ ek 2001; Prach, Pysˇ ek & Bastl 2001). There is large scale abandonment in rural areas where productivity is low in Central- and Eastern Europe (Jongepierova´, Mitchley & Tzanopoulos 2007; To¨ro¨k et al. 2010). After the collapse of state owned agricultural cooperatives, the socioeconomical changes resulted in large scale abandonment of croplands (Prach, Lepsˇ & Rejma´nek 2007; Pullin et al. 2009). Between 1990 and 2004, 600 000 ha of croplands have been abandoned in Hungary (Hobbs & Cramer 2007). This has provided an opportunity to use these areas to restore grasslands and improve their continuity for nature conservation (Stevenson, Bullock & Ward 1995; Simmering, Waldhardt & Otte 2006; Lindborg et al. 2008). Most studies reporting spontaneous succession have focused on abandoned ﬁelds formerly cultivated with annual crops or the previous history of the site (e.g. last crop) has been ignored (Csecserits & Re´dei 2001; Ruprecht 2006). Generally in these studies, weedy short-lived species are found to dominate in the ﬁrst years after abandonment (Blumenthal, Jordan & Svenson 2005; Prach, Lepsˇ & Rejma´nek 2007). Weed dominance is generally associated with high levels of soil nutrients, which can be diﬃcult and costly to control (Blumenthal, Jordan & Svenson 2003). The dominance of early colonising weedy species can also slow down the regeneration of native vegetation for many years (Collins, Wein & Philippi 2001; Prach & Pysˇ ek 2001). Secondary succession after intensive cultivation of perennial crops has not previously been studied.

One of the most important perennial crops worldwide is lucerne Medicago sativa L. Lucerne is often used as silage or hay for cattle forage (Horrocks & Valentine 1999; Li, Xu & Wang 2008). In Hungary more than 130 000 ha of croplands were sown with lucerne although intensity of use has decreased in recent years (2004–2008; K.S.H. 2008). We studied the regeneration of loess grasslands in extensively managed (mown twice a year) lucerne ﬁelds using space for time substitutions. We addressed the following questions: (i) How eﬀective is lucerne in weed control? (ii) How quickly does lucerne disappear? (iii) How fast does grassland recover in extensively managed lucerne ﬁelds? The overall aim of this study was to examine the value of spontaneous succession in the restoration of grasslands in former lucerne ﬁelds as a cost-eﬀective strategy for grassland conservation.

Materials and methods STUDY AREA

The study area is located in the Hortoba´gy Puszta (Hortoba´gy National Park), in East-Hungary. Hortoba´gy Puszta with an area of 85 000 ha is one of the largest grassland ecosystems in Europe, with vegetation characteristic of alkali and loess grasslands. The climate is moderately continental with a mean annual temperature of 9Æ5 C. Mean annual precipitation is about 550 mm. The yearly maximum precipitation falls in June (mean 80 mm) with high year-to-year ﬂuctuations (Pe´csi 1989). Historically, loess grassland vegetation (Festucion rupicolae) covered the highest elevations in the region (Borhidi 2003). At the lower elevations, loess grasslands were surrounded by dry alkali short grasslands (Festucion pseudovinae), alkali wet meadow (Alopecurion pratensis) and alkali marsh vegetation (Bolboschoenetalia maritimi) (for more details see Molna´r et al. 2008; Molna´r & Borhidi 2003). The loess grasslands have been ploughed up in the last centuries and many of the remaining fragments are degraded by moderate or heavy grazing by cattle and ⁄ or sheep. The most degraded loess pastures (Cynodonti-Poe¨tum angustifoliae) are characterised by a high cover of grazing tolerant graminoids [Cynodon dactylon (L.) Pers., Poa angustifolia L., Festuca pseudovina Hack. ex Wiesb., Festuca rupicola Heuﬀ. and Carex stenophylla Wahlbg.] and forbs [Galium verum L., Euphorbia cyparissias L., Cruciata pedemontana (Bell) Ehrend., Myosotis stricta Link, Achillea collina L., and Convolvulus arvensis L.]. At heavily grazed sites, thistles dominate (Ononis spinosa L., Eryngium campestre L.). Only small patches of less degraded loess steppe grasslands (Salvio nemorosae-Festucetum rupicolae) have remained. The characteristic graminoids for these grasslands are Festuca rupicola, Bromus inermis Leyss, Koeleria cristata (L.) Pers., Stipa capillata L., Alopecurus pratensis L., and Poa angustifolia. They are rich in perennial forb species, and harbour several characteristic loess specialist species (Salvia nemorosa L., Salvia austriaca Jacq., Phlomis tuberosa L., Thalictrum minus L., Thymus glabrescens Willd.). In the study region lucerne or alfalfa Medicago sativa L. is sown after deep ploughing at the high elevations formerly covered by loess grasslands. Seed sowing density is typically 30 kg ha–1. There are intensively and extensively managed lucerne ﬁelds. Intensive management means regular mowing associated with the application of fertilisers and pesticides. After 3 years intensively managed ﬁeld are re-sown or shallow disked. Extensive management means only regular mowing twice a year. Every year 10–50 ha intensively managed

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 257–264

Grassland recovery in lucerne ﬁelds 259 lucerne ﬁelds were replaced by extensively managed ones in the Hortoba´gy National Park.

SAMPLING

The vegetation of 1-, 3-, 5- and 10-year-old extensively managed lucerne ﬁelds (three ﬁelds in each age group) was monitored in 2009. The study ﬁelds were situated on loess plateaux between 87 and 94 m a.s.l., within a 50 km radius, in the vicinity of the villages of Egyek, Tiszacsege, Karcag and Na´dudvar (N47 26¢; E21 01¢). None of the study ﬁelds were directly connected to loess grasslands, which was the most common vegetation at this elevation in the region (To¨ro¨k et al. 2010). The ﬁelds were mown twice a year but no further management was applied. Small patches of loess grasslands and, at lower elevations, alkali marshes, alkali wet meadows and alkali short grasslands were present in close proximity to most of the ﬁelds. In each ﬁeld three 25-m2 sample blocks were chosen randomly. Within each block, the cover of vascular plants was recorded in four 1 m2 plots in early June, before the ﬁrst mowing. In addition, within each block and near to the plots (<1 m), 10 aboveground phytomass samples were collected (in total 30 per ﬁeld, 20 · 20 cm, total aboveground green phytomass and litter). We recorded the vegetation of three variously degraded stands of loess grasslands (Festucion rupicolae) for base-line vegetation reference: (i) a formerly heavy grazed Cynodonti-Poe¨tum stand, (ii) a species rich loess balk stand with Bromus inermis dominance, and (iii) a regularly mown species rich stand of Salvio nemorosaeFestucetum rupicolae grassland (for detailed species lists see Appendix S1 Supporting Information). We used the same sampling design as described above. Phytomass samples were dried (65 C, 24 h), then sorted to litter, graminoids (Poaceae and Cyperaceae), lucerne and forbs. Dry weights were measured in a laboratory with an accuracy of 0Æ01 g.

DATA ANALYSIS

We classiﬁed the species into four functional groups using life-form (based on Raunkiaer’s life form system, Raunkiaer 1934) and morphological categories (grasses and forbs). These were perennial graminoids, perennial forbs, short-lived graminoids, and short-lived forbs. Annuals and biennials are short-lived, and geophytes, hemikryptophytes, and chamaephytes are perennials. The functional groups of the weed species were classiﬁed using Grime C-S-R strategy types

(Grime 1979) which was modiﬁed and adapted to local conditions by Borhidi (1995). The cover, species richness and phytomass data of the diﬀerently aged ﬁelds were compared using General Linear MixedEﬀect Models (GLMM) and Tukey test (Zuur et al. 2009). Field age (time) was included as a ﬁxed eﬀect and ﬁeld ⁄ block structure as a random eﬀect. To analyse correlations between the diﬀerent phytomass groups and sites we used DCA ordination, with square root transformed datasets. DCA was calculated by CANOCO 4.5 (ter Braak & Sˇmilauer 2002). We used cover based Shannon diversity to characterise vegetation diversity, and Sørensen dissimilarity for vegetation changes. Characteristic species of diﬀerently aged lucerne ﬁelds and reference grasslands were identiﬁed by the IndVal procedure (Dufreˆne & Legendre 1997); during the calculations 10 000 random permutations were used. The IndVal procedure was executed by a revised version of the R code published as the electronic appendix of Bakker (2008). To explore similarities between restored and reference sites, we used NMDS ordination with Bray–Curtis similarity (Legendre & Legendre 1998). Other statistical analyses were performed using the R statistical environment (version 2.11.1, R Development Core Team 2010). Nomenclature follows Borhidi (2003) for syntaxa, and Simon (2000) for taxa.

Results VEGETATION AND PHYTOMASS

The vegetation of 1- and 3-year-old lucerne ﬁelds was characterized by the high cover of lucerne. Several weed species were present; their mean cover was less than 5% (e.g. Conyza canadensis (L.) Cronq., Lamium amplexicaule L., Polygonum aviculare L., Stellaria media (L.) Vill., see Appendix S1 in Supporting Information). The mean cover of lucerne decreased from 75Æ2 to 2Æ2% with increasing ﬁeld age. In the vegetation of 5-year-old ﬁelds the cover of lucerne was lower than 50% in all studied plots; moreover in one of the 10-year-old ﬁelds no lucerne cover was detected. Conversely, the mean cover of perennial graminoids increased from 0Æ5 to 50Æ2% parallel with increasing ﬁeld age (GLMM, P < 0Æ001, d.f. = 134, t = 14Æ30; Table 1). The mean total cover of diﬀerently aged lucerne ﬁelds ﬂuctuated between 77Æ6 and 86Æ1% (Table 1). Altogether 104 vascular plant species were recorded in the

Table 1. Cover, species richness and Shannon diversity scores of functional species groups Age of lucerne ﬁelds

Cover (%, mean ± SE) Total Medicago sativa Perennial forbs (excl. M. sativa) Perennial graminoids Short-lived forbs Short-lived graminoids Species richness (mean ± SE) Perennial species Short-lived species Shannon diversity

1-year-old

3-year-old

5-year-old

10-year-old

85Æ4 75Æ2 0Æ7 0Æ5 8Æ9 0Æ1

85Æ8 72Æ8 6Æ5 0Æ9 5Æ4 0Æ2

86Æ1 24Æ1 10Æ7 29Æ8 10Æ6 11Æ0

77Æ6 2Æ3 16Æ3 50Æ2 6Æ2 2Æ6

± ± ± ± ± ±

0Æ4 1Æ1a 0Æ2a 0Æ2a 1Æ6 0Æ1a

2Æ4 ± 0Æ2a 6Æ1 ± 0Æ7a 0Æ5 ± 0Æ1a

± ± ± ± ± ±

4Æ7 11Æ0a 4Æ5b 0Æ1a 2Æ2 0Æ1a

3Æ3 ± 0Æ4a 5Æ2 ± 1Æ6a 0Æ6 ± 0Æ3a

± ± ± ± ± ±

12Æ9 4Æ9b 2Æ7b 14Æ1b 7Æ6 3Æ9b

6Æ0 ± 1Æ1b 8Æ7 ± 2Æ1b 1Æ6 ± 0Æ2b

± ± ± ± ± ±

12Æ6 2Æ3c 2Æ2c 15Æ0c 0Æ5 1Æ5a

5Æ8 ± 0Æ4b 8Æ1 ± 1Æ0b 1Æ5 ± 0Æ2b

Diﬀerent superscripted letters indicate signiﬁcant diﬀerences tested with General Linear Mixed-Eﬀect Models and Tukey test (P < 0Æ05) 2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 257–264

260 P. To¨ro¨k et al. (a)

(b)

(c)

(d)

Fig. 1. Phytomass scores of Medicago sativa and three functional groups in diﬀerent aged lucerne ﬁelds. Notations: a = Medicago sativa, b = graminoids, c = litter, d = other forbs. Diﬀerent letters indicate signiﬁcant diﬀerences within a phytomass group between years (General Linear Mixed-Eﬀect Models and Tukey test, P < 0Æ05; tests were executed on 20 · 20 cm samples).

vegetation of the studied lucerne ﬁelds. The mean total species richness (from 8Æ5 to 13Æ9–14Æ7), the mean species richness of perennials (from 2Æ4 to 5Æ8–6Æ0), and the mean Shannon diversity scores (from 0Æ5 to 1Æ5–1Æ6) were increased with ﬁeld age (GLMM, P < 0Æ001, d.f. = 134, t = 11Æ04 and 11Æ17, respectively; Table 1). No signiﬁcant diﬀerences were found between the total phytomass of diﬀerently aged lucerne ﬁelds (means ranged between 286 and 689 g m–2). As for cover, the phytomass of lucerne decreased with increasing ﬁeld age (GLMM, P < 0Æ001, d.f. = 350, t = 17Æ17). The phytomass of graminoids was highest in the 5- and 10-year-old ﬁelds (Fig. 1). A negative correlation was detected between the phytomass of lucerne and that of graminoids. Litter and forb phytomass were also negatively correlated, but no clear temporal trend was detected. A decreasing lucerne phytomass and an increasing grass phytomass were detected with increasing ﬁeld age (Fig. 2.)

LUCERNE FIELDS AND REFERENCE GRASSLANDS

Characteristic grass species for reference grasslands (e.g. Festuca rupicola and Bromus inermis) were found at low levels of cover in 5- and 10-year-old lucerne ﬁelds. Conversely, some common grasses were dominant (e.g. Festuca pseudovina, Poa angustifolia, Agropyron intermedium (Host) P.B., Alopecurus pratensis; see Appendix S1). Decreasing mean dissimilarity of species composition was detected with increasing ﬁeld age (from a mean of 0Æ96 in 1-year-old ﬁelds to a mean of 0Æ76 in 10-year-old ﬁelds). Characteristic forb species of native loess grasslands were only present in 5- and 10-year-old lucerne ﬁelds (e.g. Vicia hirsuta (L.) S.F., V. angustifolia L., Galium verum, Medicago minima (L.) Grufbg., Trifolium angulatum W. et Kit., T. retusum Ho¨jer, Lathyrus tuberosus L.). Several other characteristic perennial forbs were not detected even in the

Fig. 2. The relationship between the various phytomass fractions and time using DCA. The points (main data) were based on mean species percentage cover. All data were pooled at the ﬁeld’s level. Notations for the lucerne ﬁelds: 1-year-old – ; 3-year-old – ; 5-year-old – ; 10-year-old – . Notations for the background variables (arrows): Lucerne = phytomass of alfalfa; Forbs = forb phytomass; Grasses = graminoid phytomass, Time = ﬁeld age; Litter = litter phytomass. Eigenvalues are 0Æ52 and 0Æ08 for axis 1 and 2, respectively.

vegetation of 10-year-old lucerne ﬁelds (e.g. Ajuga genevensis L., Salvia nemorosa, S. austriaca, Pimpinella saxifraga L., Thymus degenianus Lyka, Euphorbia cyparissias, Veronica prostrata L.; see Appendix S1). Several disturbance tolerant and weedy perennial forbs were more frequent in the lucerne ﬁelds than in reference grasslands (e.g. Cirsium arvense (L.) Scop., Convolvulus arvensis, Taraxacum oﬃcinale Weber ex Wiggers). Species composition in the lucerne ﬁelds showed a clear shift along the ﬁrst axis in the NMDS ordination (Fig. 3). Time is represented by the ﬁrst axis, and the age groups are separated along it. The vegetation of the 1 and 3-year-old ﬁelds showed low variability, while the variability of plots of the older ﬁelds was much higher (Fig. 3). The vegetation of the 10-year-old ﬁelds showed the most similarity with the vegetation of reference grasslands.

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 257–264

Grassland recovery in lucerne ﬁelds 261 & Mahn 2005). An allelopathic eﬀect of lucerne may be responsible for low weed cover: Ells & McSay (1991) showed that lucerne leaf extract (containing phenolic allelochemicals) was detrimental to germination and diﬀerentiation of susceptible plants.

COVER AND PHYTOMASS OF LUCERNE

Fig. 3. Vegetation composition of diﬀerent aged lucerne ﬁelds and reference grasslands. Ordination plot was based on percentage cover data of the sample plots using NMDS ordination and Bray–Curtis similarity (Stress = 15Æ91). Lucerne ﬁelds: 1-year-old – ; 3-year-old – ; 5-year-old – ; 10-year-old – ; Species poor Cynodonti-Poe¨tum loess grassland – ; Loess balk with high cover of Bromus inermis – ; Species rich Salvio-Festucetum loess grassland – .

Discussion WEED CONTROL

Previous studies have reported high weed cover after abandonment of intensively managed crop ﬁelds, e.g. weed cover of 5–40% for sandy ﬁelds abandoned for 1–10 years (CentralHungary; Csecserits & Re´dei 2001; Csecserits et al. 2007), and 10–60% for 1- to 12-year-old abandoned loess ﬁelds (Ruprecht 2005, 2006). Low weed cover was found after abandonment only where crop production lasted just a few years, and no mineral fertilizers had been applied (e.g. Jongepierova´, Jongepier & Klimes 2004). It has been suggested that the rapid development of weed cover can be avoided by sowing mixtures of seeds of characteristic late successional species (Prach & Pysˇ ek 2001; Pywell et al. 2002; Warren, Christal & Wilson 2002) or cover crop grasses (Hansson & Fogelfors 1998). In our study weedy species did not dominate in the early years. The total cover of weeds was low at less than 5% cover, regardless of the age of the ﬁelds. Our results support the ﬁndings of Li, Xu & Wang (2008), where lucerne and other legume species were found to aid in suppressing weeds. It is well known that seeds of weed species are present in the soils of croplands in high density (Hutchings & Booth 1996; Manchester et al. 1999). To¨ro¨k et al. (2010) detected a high cover of short-lived weeds after ploughing and sowing of perennial graminoids in former lucerne ﬁelds (1–3 years old), which suggests a high amount of weed seeds in the soil of lucerne ﬁelds. The low cover of weeds detected in the present study is most likely to be explained by the presence of lucerne, than by the absence of weed seeds in the soil. The high cover and phytomass of lucerne in the ﬁrst years caused weed suppression by increased shading of the soil surface (Gu¨sewell & Edwards 1999), and ⁄ or the competitive exclusion of short-lived weeds (Bischoﬀ, Auge

In our study the cover of lucerne was over 70% in 1- and 3-year-old lucerne ﬁelds. A sharp decline was detected after the third year. This is in accordance with the common agricultural practice in this region, where the lucerne is re-sown after 3–4 years of cultivation. In a sowing experiment conducted by Li, Xu & Wang (2008) in loess plateaux in China, the mean cover of lucerne decreased after the ﬁrst year of sowing (about 50% of cover in the ﬁrst, and 29% in the third year after sowing, respectively). The more rapid decrease in lucerne cover can be explained by the lower sowing density than in our study (22Æ5 kg ha–1, in our region 30 kg ha–1 is typical). Our results suggest that lucerne could disappear within a decade from grasslands under extensive management by mowing. The disappearance of lucerne could also be facilitated by low intensity grazing, which would select for leguminous species (Stroh et al. 2002). In previous studies a signiﬁcant increase in total vegetation cover (Ruprecht 2005; Li, Xu & Wang 2008) or an increase of cover and ⁄ or phytomass of perennials (Sˇtolcova´ 2002; Feng et al. 2007a,b; To¨ro¨k et al. 2008) has been found during secondary succession. In our study, no such trend was detected. The total cover and also the total phytomass scores remained stable during secondary succession. This was caused by the gradual replacement of lucerne by perennial grasses. To¨ro¨k et al. (2010) found litter accumulation of one order of magnitude higher between the ﬁrst and second years after restoration of grasslands with low diversity mixtures in former lucerne ﬁelds (ﬁrst year litter: 28–37 g m–2; second year litter: 280– 289 g m–2). The litter scores in the second and the third year of this study were about two to three times higher than that detected in the present study. Accumulated plant litter was identiﬁed as negatively aﬀecting vascular plant species richness in several studies (Huhta et al. 2001; Enyedi, Ruprecht & Dea´k 2007). Therefore, high amounts of litter with high perennial cover are especially eﬀective in weed suppression (To¨ro¨k et al. 2010). Litter accumulation can also be negative as litter can reduce the micro-topographical heterogeneity (Tropek et al. 2010), and decrease the availability of colonisation sites (Jensen & Gutekunst 2003), which can stabilise the community in an undesirable state (Hobbs et al. 2006). High amounts of litter could also hamper the immigration and establishment of several target species by limiting microsite availability (Foster & Gross 1998; Bissels et al. 2006). In this study, there was no litter accumulation detected and, as a result, germination and colonisation was not hampered and species richness increased with ﬁeld age. Other studies reporting spontaneous grassland succession have found similar links with litter accumulation and reduction in germination and colonisation (Jongepierova´, Jongepier & Klimes 2004; Ruprecht 2006; Feng et al. 2007a).

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 257–264

262 P. To¨ro¨k et al.

RECOVERY OF GRASSLANDS

We found that the recovery of species poor loess grasslands dominated by perennial native species in former lucerne ﬁelds was possible within 10 years. Other old-ﬁeld studies found 6– 23 years after abandonment was suﬃcient time for the spontaneous succession of loess grasslands (Molna´r & Botta-Duka´t 1998; Ruprecht 2005; Csecserits et al. 2007; Feng et al. 2007a,b). The dissimilarity in species composition between lucerne ﬁelds and reference grasslands has continuously decreased with increasing ﬁeld age. Dissimilarity scores were, however, high even between 5 and 10-year-old ﬁelds and reference grasslands. Several perennial forbs found at high frequency in loess grasslands were not detected in lucerne ﬁelds; and several short-lived weeds detected with low cover but high frequency in lucerne ﬁelds were missing from reference grasslands (see Appendix S1). Previous studies have reported that the spontaneous immigration of desirable target species is a diaspore limited process (Donath et al. 2007; Kiehl et al. 2010). There are two reasons for diaspore limitation: (i) limited spatial dispersal (e.g. missing dispersal agents and heavy seeds) reduces the movement of seeds into target sites (Simmering, Waldhardt & Otte 2006); (ii) long-term agricultural use often depletes the local seed bank, and also increases the amount of weed seeds in the soil (Coulson et al. 2001). Therefore, spontaneous recovery will be most eﬀective where native grassland sites are located nearby (O¨ster et al. 2009). A further explanation for the persistent diﬀerences in species composition between the old ﬁelds and reference grasslands is that the perennial forbs may require more time to establish in extensively managed ﬁelds (e.g. Prach, Lepsˇ & Rejma´nek 2007).

PRACTICAL IMPLICATIONS FOR POLICY

Our results suggest that the recovery of initial loess grasslands may not require technical reclamation methods (i.e. sowing competitor grasses and ⁄ or forbs) in lucerne ﬁelds where nearby grasslands are present as a seed source. We found that after a decade of regular mowing, lucerne ﬁelds were transformed into loess grasslands dominated by native perennial grasses. However, most of the characteristic loess grasslands forbs are missing. Similar results were found under the more common technical reclamation method of sowing low diversity seed mixtures (Hansson & Fogelfors 1998; Lepsˇ et al. 2007; To¨ro¨k et al. 2010). The full recovery of loess grasslands requires more time and ⁄ or should be facilitated by technical introduction of some of the target species (Kirmer et al. 2008; Kiehl et al. 2010). The transfer of hay and ⁄ or low intensity grazing combined with continued mowing can be another option to facilitate the establishment of desirable species. Our results suggest that sowing lucerne in abandoned ﬁelds and following this with extensive management can combine the advantages of both spontaneous succession and technical reclamation in grassland restoration. It oﬀers a cost eﬀective solution from the economic (agricultural) and conservation management point of view. The method has several advantages over technical reclamation. In particular, there is no weed

dominated stage and no intensive litter accumulation. Lucerne gradually decreases in abundance once re-sowing and ⁄ or fertilizing stop so we there will be a lower microsite limitation rate compared to technical reclamation sites where competitor grasses are sown. Finally, spontaneous succession is cheaper than technical reclamation, and provides a high value hay harvest in the ﬁrst few years in lucerne ﬁelds.

Acknowledgments We thank I. Kapocsi, L. Ga´l, S. U´jfalusi, S. To´th from the Hortoba´gy National Park for their help. We are indebted to T. Migle´cz, K. To´th, Sz. Tasna´dy graduate students for their help in ﬁeld and laboratory works. We are grateful to J. Memmott, and J. Firn for improving the former draft of the paper.

References Bakker, J.D. (2008) Increasing the utility of indicator species analysis. Journal of Applied Ecology, 45, 1829–1835. Bischoﬀ, A., Auge, H. & Mahn, E.-G. (2005) Seasonal changes in the relationship between plant species richness and community phytomass in early succession. Basic and Applied Ecology, 6, 385–394. Bissels, S., Donath, T.W., Ho¨lzel, N. & Otte, A. (2006) Eﬀects of diﬀerent mowing regimes on seedling recruitment in alluvial grasslands. Basic and Applied Ecology, 7, 433–442. Blumenthal, D.M., Jordan, N.R. & Svenson, E.L. (2003) Weed control as a rationale for restoration: the example of tallgrass prairie. Conservation Ecology, 7, [online] URL: http://www.consecol.org/vol7/iss1/art6/ Blumenthal, D.M., Jordan, N.R. & Svenson, E.L. (2005) Eﬀects of prairie restoration on weed invasions. Agriculture, Ecosystems & Environment, 107, 221–230. Borhidi, A. (1995) Social behaviour types, the naturalness and relative indicator values of the higher plants in the Hungarian Flora. Acta Botanica Hungarica, 39, 97–181. Borhidi, A. (2003) Magyarorsza´g No¨ve´nyta´rsula´sai (Plant associations of Hungary). Akade´miai Kiado´, Budapest, Hungary. (in Hungarian). ter Braak, C.J.F. & Sˇmilauer, P. (2002) CANOCO Reference Manual and CanoDraw for Windows User’s Guide: Software for Canonical Community Ordination (version 4.5). Microcomputer Power Ithaca, NY, USA. Collins, B., Wein, G. & Philippi, T. (2001) Eﬀects of disturbance intensity and frequency on early old-ﬁeld succession. Journal of Vegetation Science, 12, 721–728. Coulson, S.J., Bullock, J.M., Stevenson, M.J. & Pywell, R.F. (2001) Colonization of grassland by sown species: dispersal versus microsite limitation in responses to management. Journal of Applied Ecology, 38, 204– 216. Csecserits, A. & Re´dei, T. (2001) Secondary succession on sandy old-ﬁelds in Hungary. Applied Vegetation Science, 4, 63–74. Csecserits, A., Szabo´, R., Halassy, M. & Re´dei, T. (2007) Testing the validity of successional predictions on an old-ﬁeld chronosequence in Hungary. Community Ecology, 8, 195–207. Donath, T.W., Bissels, S., Ho¨lzel, N. & Otte, A. (2007) Large scale application of diaspore transfer with plant material in restoration practice – impact of seed and microsite limitation. Biological Conservation, 138, 224–234. Dufreˆne, M. & Legendre, P. (1997) Species assemblages and indicator species: the need for a ﬂexible asymmetrical approach. Ecological Monographs, 67, 345–366. Ells, J.E. & McSay, A.E. (1991) Allelopathic eﬀects of alfalfa plant residues on emergence and growth of cucumber seedlings. HortScience, 26, 368– 370. Enyedi, M.Z., Ruprecht, E. & Dea´k, M. (2007) Long-term eﬀects of the abandonment of grazing on steppe-like grasslands. Applied Vegetation Science, 11, 55–62. Feng, D., Hong-Bo, S., Lun, S., Zong-Suo, L. & Ming-Ana, S. (2007a) Secondary succession and its eﬀects on soil moisture and nutrition in abandoned old-ﬁelds of hilly region of Loess Plateau, China. Colloids and Surfaces B: Biointerfaces, 58, 278–285. Feng, D., Zongsuo, L., Xuexuan, X., Lun, S. & Xingchang, Z. (2007b) Community biomass of abandoned farmland and its eﬀects on soil nutrition in

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 257–264

Grassland recovery in lucerne ﬁelds 263 the Loess hilly region of Northern Shaanxi, China. Acta Ecologica Sinica, 27, 1673–1683. Firn, J. (2007) Developing strategies and methods for rehabilitating degraded pastures using native grasses. Ecological Management & Restoration, 8, 183– 187. Foster, B.L. & Gross, K.L. (1998) Species richness in a successional grassland: eﬀects of nitrogen enrichment and plant litter. Ecology, 79, 2593–2602. Grime, J.P. (1979) Plant strategies and Vegetation Processes. Wiley, Chichester. Gu¨sewell, S. & Edwards, P. (1999) Shading by Phragmites australis: a threat for species-rich fen meadows? Applied Vegetation Science, 2, 61–70. Hansson, M. & Fogelfors, H. (1998) Management of permanent set-aside on arable land in Sweden Journal of Applied Ecology, 35, 758–771. Hobbs, R.J. & Cramer, V.A. (2007) Why Old Fields? Socioeconomic and ecological causes and consequences of land abandonment. Old Fields: Dynamics and Restoration of Abandoned Farmland (eds V.A. Cramer & R.J. Hobbs), pp. 1–15. Island Press, Washington. Hobbs, R.J., Arico, S., Aronson, J., Baron, J.S., Bridgewater, P., Cramer, V.A., Epstein, P.R., Ewel, J.J., Klink, C.A., Lugo, A.E., Norton, D., Ojima, D., Richardson, D.M., Sanderson, E.W., Valladares, F., Vila`, M., Zamora, R. & Zobel, M. (2006) Novel ecosystems: theoretical and management aspects of the new ecological world order. Global Ecology and Biogeography, 15, 1–7. Hodacˇova´, D. & Prach, K. (2003) Spoil heaps from brown coal mining: technical reclamation vs. spontaneous re-vegetation. Restoration Ecology, 11, 385– 391. Ho¨lzel, N. & Otte, A. (2003) Restoration of a species-rich ﬂood meadow by topsoil removal and diaspore transfer with plant material. Applied Vegetation Science, 6, 131–140. Horrocks, R.D. & Valentine, J.F. (1999) Harvested Forages. Academic Press, San Diego, CA, USA. Huhta, A.-P., Rautio, P., Tuomi, J. & Laine, K. (2001) Restorative mowing on an abandoned semi-natural meadow: short-term and predicted long-term eﬀects. Journal of Vegetation Science, 12, 677–686. Hutchings, M.J. & Booth, K.D. (1996) Studies on the feasibility of re-creating chalk grassland vegetation on ex-arable land. I. The potential roles of the seed bank and the seed rain. Journal of Applied Ecology, 33, 1171– 1181. Jensen, K. & Gutekunst, K. (2003) Eﬀects of litter on establishment of grassland plant species: the role of successional status. Basic and Applied Ecology, 4, 579–587. Jongepierova´, I., Jongepier, J.W. & Klimes, L. (2004) Restoring grassland on arable land: an example of a fast spontaneous succession without weed-dominated stages. Preslia, 76, 361–369. Jongepierova´, I., Mitchley, J. & Tzanopoulos, J. (2007) A ﬁeld experiment to recreate species rich hay meadows using regional seed mixtures. Biological Conservation, 139, 297–305. Kiehl, K., Kirmer, A., Donath, T.W., Rasran, L. & Ho¨lzel, N. (2010) Species introduction in restoration projects – evaluation of diﬀerent techniques for the establishment of semi-natural grasslands in Central and Northwestern Europe. Basic and Applied Ecology, 11, 285–299. Kirmer, A., Tischew, S., Ozinga, W.A., von Lampe, M., Baasch, A. & Groenendael, J.M. (2008) Importance of regional species pools and functional traits in colonisation processes: predicting re-colonization after large-scale destruction of ecosystems. Journal of Applied Ecology, 45, 1523–1530. K.S.H. (Hungarian Central Statistical Oﬃce) (2008) Economic Accounts for Agriculture in Hungary, 2008. Statistical Reﬂections, 26, [online] http://portal.ksh.hu/portal/ Legendre, P. & Legendre, L. (1998) Numerical Ecology. Elsevier Science, Amsterdam, The Netherlands. Lepsˇ , J., Dolezˇal, J., Bezemer, T.M., Brown, V.K., Hedlund, K., Igual Arroyo, M., Jo¨rgensen, H.B., Lawson, C.S., Mortimer, S.R., Peix Geldart, A., Rodrı´ guez Barrueco, C., Santa Regina, I., Sˇmilauer, P. & van der Putten, W.H. (2007) Long-term eﬀectiveness of sowing high and low diversity seed mixtures to enhance plant community development on ex-arable ﬁelds. Applied Vegetation Science, 10, 97–110. Li, J.-H., Xu, D.-H. & Wang, G. (2008) Weed inhibition by sowing legume species in early succession of abandoned ﬁelds on Loess Plateau, China. Acta Oecologica, 33, 10–14. Lindborg, R., Bengtsson, J., Berg, A., Cousins, S.A.O., Eriksson, O., Gustafsson, T., Per Hasund, K., Lenoir, L., Pihlgren, A., Sjo¨din, E. & Stenseke, M. (2008) A landscape perspective on conservation of semi-natural grasslands. Agriculture, Ecosystems & Environment, 125, 213–222. Manchester, S.J., McNally, S., Treweek, J.R., Sparks, T.H. & Mountford, J.O. (1999) The cost and practicality of techniques for the reversion of arable land

to lowland wet grassland - an experimental study and review. Journal of Environmental Management, 55, 91–109. Mijnsbrugge, K.V., Bischoﬀ, A. & Smith, B. (2010) A question of origin: where and how to collect seed for ecological restoration. Basic and Applied Ecology, 11, 300–311. Molna´r, Zs. & Borhidi, A. (2003) Hungarian alkali vegetation: origins, landscape history syntaxonomy, conservation. Phytocoenologia, 33, 377–408. Molna´r, Zs. & Botta-Duka´t, Z. (1998) Improved space-for-time substitution for hypothesis generation: secondary grasslands with documented site history in SE-Hungary. Phytocoenologia, 28, 1–29. Molna´r, Zs., Bı´ ro´, M., Bo¨lo¨ni, J. & Horva´th, F. (2008) Distribution of the (semi-)natural habitats in Hungary I. Marshes and grasslands. Acta Botanica Hungarica, 50, Suppl. 1, 59–106. del Moral, R., Walker, L.R. & Bakker, J.P. (2007) Insights gained from succession fort he restoration of landscape structure and function. Linking Restoration and Ecological Succession (eds L.R. Walker, J. Walker & R.J. Hobbs), pp. 19–45. Springer-Verlag, New York, USA. O¨ster, M., Ask, K., Ro¨mermann, C., Tackenberg, O. & Eriksson, O. (2009) Plant colonization of ex-arable ﬁelds from adjacent species-rich grasslands: the importance of dispersal vs. recruitment ability. Agriculture, Ecosystems & Environment, 130, 93–99. Pe´csi, M. (eds) (1989) Magyarorsza´g Nemzeti Atlasza (National Atlas of Hungary). Kartogra´ﬁai va´llalat, Budapest (in Hungarian). Prach, K. & Hobbs, R.J. (2008) Spontaneous succession versus technical reclamation in the restoration of disturbed sites. Restoration Ecology, 16, 363– 366. Prach, K., Lepsˇ , J. & Rejma´nek, M. (2007) Old ﬁeld succession in Central Europe: local and regional patterns. Old Fields: Dynamics and Restoration of Abandoned Farmland (eds V.A. Cramer & R.J. Hobbs), pp. 180–201. Island Press, Washington. Prach, K. & Pysˇ ek, P. (2001) Using spontaneous succession for restoration of human-disturbed habitats: experience from Central Europe. Ecological Engineering, 17, 55–62. Prach, K., Pysˇ ek, P. & Bastl, M. (2001) Spontaneous vegetation succession in human-disturbed habitats: a pattern across seres. Applied Vegetation Science, 4, 83–88. Prach, K. & Rˇehounkova, K. (2008) Spontaneous vegetation succession in gravel-sand pits: a potential for restoration. Restoration Ecology, 16, 305–312. Pullin, A.S., Ba´ldi, A., Can, O.E., Dieterich, M., Kati, V., Livoreil, B., Lo¨vei, G., Miho´k, B., Nevin, O., Selva, N. & Sousa-Pinto, I. (2009) Conservation focus on Europe: major conservation policy issues that need to be informed by Conservation Science. Conservation Biology, 23, 818–824. Pywell, R.F., Bullock, J.M., Hopkins, A., Walker, K.J., Sparks, T.H., Burke, M.J.W. & Peel, S. (2002) Restoration of species-rich grassland on arable land: assessing the limiting processes using a multi-site experiment. Journal of Applied Ecology, 39, 294–309. R Development Core Team (2010) R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing, Vienna, Austria. ISBN 3-900051-07-0, URL http://www.R-project.org. Raunkiaer, C. (1934) The Life Forms of Plants and Statistical Plant Geography, Being the Collected Papers of C. Raunkiaer. Oxford University Press, Oxford. Reid, A.M., Morin, L., Downey, P.O., French, K. & Virtue, J.G. (2008) Does invasive plant management aid restoration of natural ecosystems? Biological Conservation, 142, 2342–2349. Ruprecht, E. (2005) Secondary succession in old-ﬁelds in the Transylvanian Lowland (Romania). Preslia, 77, 145–157. Ruprecht, E. (2006) Successfully recovered grassland: a promising example from Romanian old-ﬁelds. Restoration Ecology, 14, 473–480. Simmering, D., Waldhardt, R. & Otte, A. (2006) Quantifying determinants contributing to plant species richness in mosaic landscapes: a single- and multi-patch perspective. Landscape Ecology, 21, 1233–1251. Simon, T. (2000) A magyarorsza´gi ede´nyes ﬂo´ra hata´rozo´ja (Vascular Flora of Hungary). Nemzeti Tanko¨nyvkiado´, Budapest, Hungary (In Hungarian). Stevenson, M.J., Bullock, J.M. & Ward, L.K. (1995) Re-creating semi-natural communities: eﬀect of sowing rate on establishment of calcareous grassland. Restoration Ecology, 3, 279–289. Sˇtolcova´, J. (2002) Secondary succession on an early abandoned ﬁeld: vegetation composition and production of biomass. Plant Protection Science, 38, 149–154. Stroh, M., Storm, C., Zehm, A. & Schwabe, A. (2002) Restorative grazing as a tool for directed succession with diaspore inoculation: the model of sand ecosystems. Phytocoenologia, 32, 595–625. To¨ro¨k, P., Matus, G., Papp, M. & To´thme´re´sz, B. (2008) Secondary succession in overgrazed Pannonian sandy grasslands. Preslia, 80, 73–85.

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 257–264

264 P. To¨ro¨k et al. To¨ro¨k, P., Dea´k, B., Vida, E., Valko´, O., Lengyel, Sz. & To´thme´re´sz, B. (2010) Restoring grassland biodiversity: sowing low diversity seed mixtures can lead to rapid favourable changes. Biological Conservation, 143, 806–812. Tropek, R., Kadlec, T., Karesova, P., Spitzer, L., Kocarek, P., Malenovsky, I., Banar, P., Tuf, I.H., Hejda, M. & Konvicka, M. (2010) Spontaneous succession in limestone quarries as an eﬀective restoration tool for endangered arthropods and plants. Journal of Applied Ecology, 47, 139–147. Walker, L.R., Walker, J. & del Moral, R. (2007) Forging a new alliance between succession and restoration. Linking Restoration and Ecological Succession (eds L.R. Walker, J. Walker & R.J. Hobbs), pp. 1–19. SpringerVerlag, New York, USA. Warren, J., Christal, A. & Wilson, F. (2002) Eﬀects of sowing and management on vegetation succession during grassland habitat restoration. Agriculture, Ecosystems & Environment, 93, 393–402. Zuur, A., Ieno, E.N., Walker, N., Saveiliev, A.A. & Smith, G.M. (2009) Mixed Eﬀects Models and Extensions in Ecology with R. Springer, New York, USA.

Supporting Information Additional Supporting Information may be found in the online version of this article. Appendix S1. Characteristic species of target loess grasslands and extensively managed lucerne ﬁelds identiﬁed by an IndVal procedure; 10,000 random permutations were used. As a service to our authors and readers, this journal provides supporting information supplied by the authors. Such materials may be reorganized for online delivery, but are not copy-edited or typeset. Technical support issues arising from supporting information (other than missing ﬁles) should be addressed to the authors.

Received 9 June 2010; accepted 20 October 2010 Handling Editor: Jennifer Firn

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 257–264

Journal of Applied Ecology 2011, 48, 35–46

doi: 10.1111/j.1365-2664.2010.01905.x

Conserving a moving target: planning protection for a migratory species as its distribution changes Navinder J. Singh* and Eleanor J. Milner-Gulland Department of Life Sciences, Imperial College London, Silwood Park Campus, Buckhurst Road, Ascot, Berkshire SL57PY, UK

Summary 1. Conservation of declining migratory species is a challenging task, as the factors that may have determined their past distribution may not determine their current and future distribution. Saiga antelope Saiga tatarica populations have massively declined due to poaching. The species is now beginning to recover in Kazakhstan and protected areas are being implemented. Using 25 years of aerial monitoring data, we identiﬁed changes in the spring distribution and predicted densities of saiga to prioritize areas for protection under scenarios of climate change together with changes in disturbance and population size. Conserving the spring distribution is critical as spring calving aggregations are of particular importance to population viability. 2. The current distribution is strongly inﬂuenced by disturbance, whereas climate had a stronger inﬂuence in the past. The area of highly suitable habitat has halved and become fragmented in the last decade. The existing and proposed protected areas are relatively complementary and perform well under most scenarios of future change. However there is a need to widen the geographical scope of protected area planning if potential future high suitability areas are to be eﬀectively protected. 3. Climate change interacts with other factors to determine the distribution of suitable habitat within and outside protected areas. Scenarios in which conservation has increased saiga population size and density tend to show limited impacts of climate change, while scenarios in which the saiga population fails to recover and disturbance continues show, worsening patchiness and reduced suitable habitat. 4. Synthesis and applications. We provide evidence for changing distribution and density of a migratory species over a large spatio-temporal scale, and suggest that future distribution may be more constrained and spatially heterogeneous. These results have important implications for designing future conservation measures for migratory species, such that areas that robustly show high suitability under a range of potential scenarios of change can be included in protected area expansion plans. Protected area placement based only on current, rather than projected distribution risks wasting opportunities for proactive conservation, particularly for a highly disturbed, recovering species likely to be aﬀected by climate change. Key-words: climate change, conservation planning, GAMM, HSM, long distance migration, NDVI, poaching, protected areas, Saiga tatarica tatarica, ungulates Introduction Many ungulate migrations worldwide have been disrupted in the last two centuries due to human impacts (Bolger et al. 2008; Harris et al. 2009). Several studies have raised concern over the plight of migratory species and urged the need for proper monitoring and establishment of protected areas (PAs) (Bolger et al. 2008; Harris et al. 2009). However, protection of such species is challenging because the animals

*Correspondence author. E-mail: [email protected]

occur in large numbers, are frequently on the move and are distributed over vast areas (Mueller et al. 2008). Often, migratory routes vary by season and year, which makes establishment of static PAs challenging. Many migratory ungulates are also subjected to poaching and habitat modiﬁcation, which cause changes in habitat use (Shuter et al. 2010). The prioritization of survey areas and PA planning can become diﬃcult in such situations, especially for species whose populations have declined substantially and so are potentially not currently occupying typical habitat. The issue of climate change has received signiﬁcant attention recently, as it is expected to induce major changes in migratory

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society

36 N. J. Singh & E. J. Milner-Gulland systems and hence raise further challenges to the management of migratory species (Lundberg & Moberg 2003; Hodgson et al. 2009). Especially in the case of migratory ungulates for which the timing of spring green up is tightly coupled with the onset of the spring migration and dense calving aggregations, climatic changes in temperature and productivity are likely to induce major range shifts during spring (Lundberg & Moberg 2003; Wilcove & Wikelski 2008; Singh et al. 2010a; b). This implies that an understanding of the drivers of the spring distribution of these species is particularly important for conservation planning. Pressey et al. (2007) suggest that eﬀective conservation planning must recognise both the inherent dynamism of ecological processes and the eﬀects of anthropogenic alterations on habitat use and availability. Following on from this, there is a need to integrate uncertainty into spatial predictions of species presence and to make precautionary decisions that avoid underestimating threats. Decisions also need to be revisited regularly in the light of novel and potential threats (Wilson et al. 2007). It is also important to evaluate the eﬀects of conservation on future species distributions, in the light of ongoing environmental change. When planning responses to climate change scenarios, Wiens & Bachelet (2010) urge conservationists to align the scale of conservation action to the scales of climate change projections. An additional challenge is the uncertainty in identiﬁcation of important habitats for threatened or declining species. Freckleton, Noble & Webb (2006) explored the likely relationship between population density and occupancy and found a generally positive relationship, but departures from this relationship were potentially informative as to population status. Especially for endangered or declining species, there may not always be a linear relationship between occurrence and abundance. Predicting the spatial relationship between density and occupancy is hence potentially important in evaluating the proportion of the population likely to beneﬁt from particular PAs. Eﬀective stratiﬁcation of population monitoring also depends upon an understanding of the relationship between the density and distribution of a population. An excellent example of a species facing these issues is the migratory saiga antelope Saiga tatarica (Linnaeus 1766) of Central Asia, which has experienced a 95% reduction in population size over the last two decades (Milner-Gulland et al. 2001, 2003). Simultaneously, group sizes have decreased (McConville et al. 2009), density has decreased and sex ratio and conception rates (Milner-Gulland et al. 2003) and the location and density of calving aggregations have all been aﬀected by human disturbance (Fry 2004; Singh et al. 2010a). For the last four decades, annual aerial surveys have been conducted over the spring range of saiga in Kazakhstan to estimate population size, density and identify aggregations. Considerable conservation eﬀorts are now being undertaken for saiga on a landscape-scale, mostly within the spring range. For example the Altyn Dala Conservation Initiative (ADCI) is working in Central Kazakhstan, carrying out activities such as research, community engagement, anti-poaching patrols, improvement of monitoring and the designation of new PAs (Klebelsberg 2008). As a result of investment in initiatives such

as these by the Government of Kazakhstan and international NGOs, saigas are currently recovering rapidly in this region, and new PAs are being planned and implemented (CMS 2010). Under these circumstances, it is essential that the designation of PAs, which consumes substantial time, resources, and legislative activity, is based on best estimates of the future needs of the species. The regional predictions for Kazakhstan from global climate change models predict precipitation changes and an increase in spring temperature in the region (IPCC 2007). Recent studies based on analysis of time series satellite data also predict a decline in productivity across the saiga range (de Beurs, Wright & Henebry 2009; Zhao & Running 2010). Kazakhstan is currently undergoing rapid development, including new infrastructure, urban growth and human population increases, which are likely to change patterns of human disturbance, and together with increasing wealth, may aﬀect poaching pressure (Ku¨hl et al. 2009). These changes will combine to aﬀect saiga densities and distribution in future, which may render PAs based upon current saiga presence less well targeted than they could be. Long term aerial survey data provide a unique opportunity to identify changes in spring saiga distribution patterns and to assess the adequacy of diﬀerent PA plans (Singh et al. 2010a; b). We used the last 25 years of aerial survey data to identify the changing drivers of saiga distribution in Central Kazakhstan. We analysed trends in distribution patterns, explored factors aﬀecting the spring distribution, and then identiﬁed areas of high potential saiga density, where monitoring and protection could be focused. We compared density and occupancy predictions to identify areas of mismatch between the two. Since future changes in climate and human activities may cause changes in saiga distribution, we projected the current distribution under a range of scenarios, based on IPCC projections of spring temperature rise and vegetation productivity changes in the region (IPCC – Intergovernmental Panel for Climate Change 2007), in combination with assumptions about the success or failure of current conservation interventions in reducing human disturbance. We then evaluated the location of current and planned PAs with respect to potential future spring saiga distributions. This approach provides a robust basis for the continuing conservation planning process in Central Kazakhstan, as well as providing guidance for similar studies on other migratory species. It ﬁlls an important gap in the existing literature on modelling the eﬀects of climate and anthropogenic factors to understand their synergistic impact on species density and distribution.

Materials and methods STUDY AREA

We consider the Betpak Dala saiga population, located in central Kazakhstan (Fig. S1, Supporting information). This was historically both the largest and the widest ranging of the four populations of sub-species S.t. tatarica, covering an estimated area of about 1Æ08 million km2. The area is generally ﬂat, covered by treeless steppe, semidesert and desert vegetation. Vegetation zones occur in a latitudinal gradient. The northernmost steppe zone is followed by semi-desert

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 35–46

Landscape-scale planning for a migratory species 37 and desert zones with annual precipitation of around 300 mm, 200– 250 mm and <200 mm, respectively. Saigas inhabit all three zones on a seasonal basis (Fadeev & Sludskii 1982; Bekenov, Grachev & Milner-Gulland 1998). Their annual migrations, 600–1200 km in length, are driven by the need for new green pastures in the spring (March–May) and by the presence of deep snow in the steppe zone starting in autumn (September–October). The area in which spring calving was historically observed in this population is 400 km in length north–south and 700 km east–west (Bekenov, Grachev & Milner-Gulland 1998). The period, route, distance and speed of migration diﬀer from year to year depending on climatic conditions, the condition of pastures, the number of watering places and the degree of disturbance experienced by the animals (Bekenov, Grachev & Milner-Gulland 1998; Singh et al. 2010a). This population was severely and rapidly aﬀected by poaching due to its relative accessibility from the capital Almaty and from the main saiga horn markets in China. The population declined from an estimated 375 000 animals in the 1980s to an estimated 1800 at its nadir in 2003 (although this is likely to be a substantial underestimate; McConville et al. 2009). The 2010 population estimate is 53 400, around 14% of the 1980s size, but substantially larger than only a few years previously (CMS 2010; Fig. S2, Supporting information). This population has been the recipient of signiﬁcant conservation action, which is likely to be one reason for its rapid recovery.

AERIAL SURVEY DATA

We acquired aerial survey reports from the Institute of Zoology of Kazakhstan for the period 1983–2008, with some years missing (1984, 1990, 1998 and 2000). These reports contain information on the dates and timing of aerial surveys, the areas surveyed, locations of the groups with 500 m resolution, group sizes observed and the population estimates made (see Fig. S3, Supporting information for an example of the maps generated). These observations were transferred into a Geographic Information System. The survey technique was developed during Soviet times and remained unchanged throughout the study period. The aerial survey team ﬂew transects over the areas of highest saiga density, determined through local and expert knowledge. Transects were, where possible, 10 km apart and the assumed strip width was 1 km each side of the aircraft, such that 20% of the area was covered. The estimated size and approximate location of each group were marked on a topographic map of scale 1:1 000 000. The team then extrapolated the population estimate by dividing the number of saiga seen by the coverage (usually 0Æ2) and multiplying by the size of the area of saiga concentration, calculated by drawing a line around the observations and counting grid squares within that line. There was no measure of error. The population crash towards the end of the 1990s led to a change in the group size distribution, which aﬀected the population estimates. Formerly, saiga aggregated in large numbers during the spring migration, with approximately 40% of the groups containing over 500 individuals (Fadeev & Sludskii 1982). This has changed considerably in recent years, with maximum herd sizes in the low hundreds, and the majority of herds numbering less than 50 (Institute of Zoology, unpublished data). The extent to which these changes have aﬀected the bias in the population estimates is unclear but potentially signiﬁcant (McConville et al. 2009). Signiﬁcant issues with the sampling procedures have also recently been identiﬁed (Norton-Griﬃths & McConville 2007), which are now being addressed (Zuther 2009).

VARIABLES

The spring distribution of saiga is determined by temperature, availability of water and green forage (Bekenov, Grachev & Milner-Gulland 1998; Singh et al. 2010b). We selected mean diurnal range of temperature during the survey period (March–May), vegetation productivity and distance to water and settlements as predictor variables. Singh et al. (2010b) showed for the whole of Kazakhstan that all these variables were important determinants of saiga presence ⁄ absence in spring. We used temperature range as it made sense biologically and was a better ﬁt to the data than other temperature metrics. In early spring the maximum temperature may be inﬂuential in the onset of spring migration as a result of snow melt, but later in the period, during calving, survival of new born saiga calves is limited by minimum temperature (Milner-Gulland 1994; Bekenov, Grachev & MilnerGulland 1998). Mean diurnal range of temperature, henceforth ‘temperature’ was estimated from monthly maximum and minimum temperature data for the survey period, obtained from the National Climate Data Center (ftp://ftp.ncdc.noaa.gov/pub/data/ghcn/v2). These data were obtained from various ground stations and were gridded to be used as raster layers with 1 km2 resolution (http:// www.esrl.noaa.gov/psd/data/gridded/). We rescaled all the variables to a pixel size of 25 km2, since the scale of the aerial survey is large and groups are spread over vast areas. We used vegetation productivity instead of precipitation as these two variables are confounded (Singh et al. 2010a). We used the Integrated Normalized Diﬀerence Vegetation Index (INDVI) as an index of vegetation productivity (Pettorelli et al. 2006). We acquired NDVI data from 15-day GIMMS (Global Inventory Modelling and Mapping Studies) dataset (8 km2 spatial resolution – ftp://ftp.glcf. umiacs.umd.edu/glcf/GIMMS/Geographic/; Tucker et al. 2005) and from the bimonthly MODIS dataset (for years 2007–08, 1 km resolution; Global Land Cover Facility, https://wist.echo.nasa.-gov/ wist-bin/api/ims.cgi?mode=MAINSRCH&JS=1) resampled to the scale of GIMMS data as GIMMS was only available up to 2006. We estimated INDVI as the sum of NDVI from March to mid-May, because the aerial surveys were conducted towards the end of April and ﬁrst half of May. The straight line distance to the nearest natural water and settlements were estimated for each calving location based on maps acquired from ‘Biogeomencer project’ (http://www.biogeo mancer.org/) for water and ‘Global database on Administrative areas’ (http://biogeo.berkeley.edu/gadm/) for settlements. Both temporary and permanent water sources were included, because temporary water sources are full in spring. The straight-line distance to the nearest settlement in the database was regarded as an indicator of the level of disturbance due to poaching pressure. Settlement location has remained consistent in the region over the period, despite changes in settlement size, and poaching pressure is not likely to be straightforwardly frequency dependent (Ku¨hl et al. 2009), hence the size of the settlement was not included.

SAIGA DISTRIBUTION MODELLING

Habitat suitability models (HSMs) apply ecological niche principles, using environmental variables to predict the presence ⁄ absence or abundance of a species throughout a study area, with the primary aim of identifying key variables that determine the niche (Hirzel & Le Lay 2008). However, the strength of the distribution-niche linkage depends on the ecology of the species, local constraints and historical events (Hirzel & Le Lay 2008). We hence tested ﬁve diﬀerent modelling approaches, consisting of both presence-pseudo-absence and presence only methods, to select the best approach for modelling

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 35–46

38 N. J. Singh & E. J. Milner-Gulland annual saiga distributions based on climatic and anthropogenic variables (Table S1, Supporting information; Thuiller et al. 2009). A presence-pseudo-absence approach using logistic Generalized Additive Models (GAM) with a binary response variable and continuous explanatory variables consistently performed better than other methods based on the area under the relative operating characteristic curve (AUC) values (Hastie & Tibshirani 1990; Table S1). We therefore used this approach instead of a presence-only approach such as Ecological Niche Factor Analysis (ENFA) or Maxent as they overpredicted suitable habitat, as has been observed in other studies (Chefaoui & Lobo 2008; Matthiopoulos & Aarts 2010). We generated 2000 pseudo-absences randomly within the annual range area of the Betpak Dala population, of which 1890 were left after removing the points from water bodies. After estimating the yearly models, we tested whether pooling the data into three periods would be more representative of the changes over the study period. This was done to overcome the high yearly variability in saiga locations, probably a result of the confounded eﬀects of weather and observer bias. The yearly observations were pooled into three periods – Period I: 1986–1996, II: 1997–2002 and III: 2003– 2008, representing three diﬀerent saiga population states: I – high and relatively stable, II – declining and III – low ⁄ recovering (Fig. S1). We ran these models using generalized additive mixed models with a cubic spline smoother value of 3 (GAMMS; Wood 2004), year as the random eﬀect and other explanatory variables as ﬁxed eﬀects (Table S2, Supporting information). We accounted for spatial autocorrelation by incorporating ‘X’ and ‘Y’ coordinates of the centroids of the saiga group locations as ﬁxed eﬀects. Spatial predictions were made for all three periods and models were validated using k-folds cross validation (Boyce et al. 2002). We used <0Æ2, 0Æ2–0Æ5 and >0Æ5 suitability thresholds to classify the predictions into low, medium and high suitability areas. To estimate variable contributions we ﬁrst estimated the correlation score between the standard predictions vs. the predictions of the model from a randomly permutated variable (Thuiller et al. 2009). We then estimated 1 minus the mean correlation for each variable where a high score meant a high importance. The results were then converted to give a relative importance (summing to one; Thuiller et al. 2009).

CHANGE SCENARIOS

The best model for period III as the current baseline was used to test the eﬀects of potential future changes on saiga distribution. Two dimensions of change were explored; climate change leading to a rise in spring temperature and ⁄ or reduced vegetation productivity, and anthropogenic change leading to increased or reduced disturbance and ⁄ or increases in population and group size. We manipulated the dataset and environmental variables keeping the model structure

constant, to get a general idea of the consequent distributional changes. The main scenarios tested were (Table 1): T+: Temperature + 2Æ5 C, modelled by creating a future temperature layer using the predicted rise in maximum spring temperature from the 2007 IPCC report. N): NDVI )13Æ5% due to increased droughts, based on predictions for the area in de Beurs, Wright & Henebry (2009) and Zhao & Running (2010), modelled similarly to the T+ scenario. D+: Increased disturbance; in this scenario saigas may move further away from towns, hence each saiga observation was moved 4 km further away from the nearest town than its observed location. The value of 4 km was chosen based on the observed shift of saiga groups away from settlements between periods I and III (Table S2, Supporting information). D): Disturbance decreases; we decreased the distance of observations from the nearest settlement to Period I levels (i.e. by 4 km). P+: Population recovers to 150 000 (based upon the Government of Kazakhstan’s goal to increase saiga numbers in Kazakhstan to 200 000 in the next 10 years, and because the Betpak Dala population is the only one that is currently increasing rapidly). We increased the frequency of all group sizes in the database equivalently, to mimic an increased population size but the same group size distribution as currently. G+: Population recovers and also becomes less fragmented (based upon the fact that group sizes were very much larger in previous periods than in period III): We increased the population size to 150 000 by increasing the number of saigas in the top 10% of group sizes. Finally, we combined some of the scenarios to give composite scenarios that included the most likely potential combinations of eﬀects. These composites included one that mimicked conservation success, in which disturbance was reduced and the population increased, with a concomitant increase in mean group size, but in the context of climate change (T+P)D)G+), and one mimicking conservation failure, in which disturbance increased and the population stayed at the current level (T+P)D+; Table 1).

FUTURE PA EVALUATION

To assess the relevance of existing and proposed PAs to potential future distributions, we mapped the two existing and three proposed PAs in the Betpak Dala population’s spring range, based on information from the Association for the Conservation of Biodiversity in Kazakhstan (ACBK; http://www.acbk.kz; Fig. S3) and the United Nations Development Programme’s Steppe Conservation Project (http://www.undp.kz). We overlaid this map over our model predictions for period I (representing healthy saiga populations in the past), period III (representing the current situation) and for the future change scenarios, then identiﬁed the suitable habitat falling under each PA type. We considered ﬁve PAs of which two already exist, and three are proposed (Fig. S4, Supporting information). The total area covered by these PAs is 20 250 km2, made up of 14 681 km2 of existing

Table 1. Percentage area under three habitat suitability categories of predicted saiga presence for nine diﬀerent change scenarios

High Medium Low

Period III

T+

N)

D+

D)

P+

G+

T+ N)

T+N)D+

T+N)D)G+

16Æ84 42Æ72 40Æ43

6Æ96 6Æ73 86Æ31

13Æ15 56Æ28 30Æ58

13Æ50 39Æ26 47Æ24

30Æ82 20Æ14 49Æ04

42Æ43 40Æ67 16Æ90

51Æ12 33Æ27 15Æ61

10Æ09 21Æ60 68Æ31

14Æ91 35Æ21 49Æ88

27Æ26 36Æ34 36Æ40

Period III is the current situation. T+: Temperature increases by 2Æ5 C; N): NDVI decreases by 13Æ5%; D+: Disturbance increases by 4 km (i.e. more people or more hunting); D): Disturbance to 1980s levels (conservation programmes succeed); P+: Population increases to 150 000; G+: Population increases and groups become less fragmented. Combined scenarios include climate change only (T+N)), climate change with increased disturbance (‘conservation failure’; T+N)D+), and climate change with decreased disturbance and a larger, less fragmented population (‘conservation success’, T+N)D)G+). 2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 35–46

Landscape-scale planning for a migratory species 39 PAs and 5568 km2 of proposed PAs, compared to the identiﬁed current spring saiga range of about 650 000 km2. We then estimated the proportion of the PA with high (>0Æ5), medium (0Æ2–0Æ5) and low (<0Æ2) habitat suitability. We also estimated the proportion of each of these suitability types within the PAs vs. in the entire spring range.

Year X,Y

Dist2town Dist2water

INDVI Temperature

100% 90% 80%

PREDICTING FUTURE DENSITY

70%

We predicted the spatial distribution of saiga density, to compare the distribution of density and occupancy, understand model limitations and identify which model is a better metric for future scenarios. Predicted relationships for period III and the future scenarios could also be used to guide the allocation of monitoring and protection eﬀorts to areas of high saiga density. The aerial survey data included information about group size and extent from which density could be inferred. However, observed group size distribution from Kazakhstan’s aerial surveys are likely to be severely biased by observation error, artiﬁcially reducing the number of small groups in the dataset (McConville et al. 2009). Hence we reconstructed putative original group size distributions using the bias correction algorithm developed by McConville et al. (2009), obtaining an estimate and range of corrected group sizes for each year. We assumed the total population size and then back-calculated the actual group size distribution using lognormal curves. This distribution was then corrected using the detectibility function used by McConville et al. (2009) and was simulated 100 times for each year to be used in the models of density distribution. Once the bias-corrected group size distribution had been obtained from the original survey data, we ﬁtted a kernel density estimator to our observed point pattern with a smoothing parameter estimated using cross validation (Baddeley 2008; Hengl et al. 2009). We ﬁtted the relative density as a ratio between the local density and maximum density from all locations (Hengl et al. 2009; Hengl 2009). The advantage of using the relative density is that the values are in the range [0, 1], regardless of the bandwidth and sample size. We then used the logistic regression kriging technique from Hengl et al. (2009) to predict saiga group density by: (i) converting the group sizes to logits (if the input values are equal to 0 or 1, replacing them with the second smallest ⁄ highest value); (ii) ﬁtting a regression model (GAMM in our case) with temperature, NDVI, distance to towns and water as explanatory variables, based on the variables found important in previous modelling exercises (Singh et al. 2010a,b and the distribution model in this paper); (iii) ﬁtting a variogram for the residuals (logits); (iv) producing predictions by ﬁrst predicting the regression model part, then interpolating the residuals using ordinary kriging, and ﬁnally adding the predicted trend and residuals together; (v) backtransforming the interpolated logits to the original [0, 1] scale. We performed these analyses for period III. The density predictions were then stratiﬁed into three strata based on the predicted density of the groups (<0Æ02, 0Æ02–0Æ05, >0Æ05 animals km-2). To identify the relationship between predicted density and occupancy of saiga, we used mean density and mean habitat suitability for each pixel and estimated the product–moment correlations (Freckleton, Noble & Webb 2006).

60%

Results SAIGA DISTRIBUTION MODELLING

The spring saiga distribution was determined by an intermediate range of temperature, intermediate productivity, areas at

50% 40% 30% 20% 10% 0%

Period I

Period II

Period III

Fig. 1. Percentage variable contribution of explanatory variables in each periodic Generalized Additive Mixed Model for the distribution of saiga antelopes in Betapak Dala. Temperature–Mean diurnal range of temperature during the survey period; INDVI – Integrated Normalized Diﬀerence Vegetation Index; Dist2water – Distance to nearest water source; Dist2town – Distance to nearest settlement; X,Y – Geographic coordinates of the groups; Year – Year of survey. Period I – 1986–1996; II – 1997–2002, III – 2003–2008.

intermediate distance from water and away from settlements. The yearly models demonstrate a high variability in the percentage contribution of these variables, with temperature consistently explaining most of the variability, while other variables varied in time (Fig. S5, Supporting information). The periodic models captured the same consistency (Table S2), but were more robust to interannual variation, and so were used for the scenario analyses in preference to the yearly models. In period I, temperature explained most variation in group location (52Æ8%) followed by NDVI (15Æ8%; Fig. 1, Table S2). For period II, the trends were similar, with temperature and NDVI still explaining most of the variability (Temperature – 51Æ3% NDVI – 11Æ0%), but the contribution of distance to settlements slightly increased (from 2Æ35 to 6Æ7%). However during period III, the contribution of temperature and NDVI declined (29Æ5 and 13Æ9%, respectively) while distance to settlements increased substantially (22Æ6%, Fig. 1). Percentage variation explained by spatial autocorrelation of groups was 9Æ9% and by year, about 15Æ1%. There was no signiﬁcant spatial autocorrelation observed in spatial variograms of the residuals. The high Spearman rank correlation coeﬃcients from cross validations for each model showed that the models had good predictive ability (Table S3 Supporting information). There was a 50% decline in the amount of high suitability area in period III compared to period I and a slight decline compared to period II. The extent of medium suitability areas increased by about 25% in period III compared to period I (Fig. 2).

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 35–46

40 N. J. Singh & E. J. Milner-Gulland 50°0'0"E

60°0'0"E

70°0'0"E

80°0'0"E

90°0'0"E

50°0'0"E

60°0'0"E

(a)

70°0'0"E

80°0'0"E

90°0'0"E

(b)

50°0'0"N

50°0'0"N

Russia

Kyrgyzstan

Uzbekistan

Caspian Sea

China

Turkmenistan

50°0'0"E

60°0'0"E

40°0'0"N

40°0'0"N

Kazakhstan

Tajikistan

70°0'0"E

80°0'0"E

90°0'0"E

40°0'0"N

50°0'0"N

(c)

0

200

400

800 km

Fig. 2. Spring distribution of the Betpak Dala population of Kazakhstan in three diﬀerent time-periods, based on the GAMM models. (a) Period I 1986–1996; (b) II 1997–2002 and (c) III 2003–2008. Black cells represent areas with high suitability (>0Æ5), dark grey cells – areas with medium suitability (0Æ2–0Æ5), and white with low suitability (<0Æ2). Light grey polygons represent water bodies.

CHANGE SCENARIOS

The predictions from the diﬀerent scenarios were highly variable both in terms of the quantity and spatial distribution of suitable habitat. A 60% loss in high and 80% loss in medium suitability areas was predicted with rising temperatures, with patchier spatial occurrence of high suitability areas (Table 1, Fig. S6, Supporting information). A decline in productivity had less eﬀect on the total amount of each habitat type, but increased the patchiness of the distribution (Table 1, Fig. S6). Increasing disturbance had a particularly strong eﬀect in the northwest, whereas decreased disturbance lead to consolidation of the distribution, with both high and low probability habitat gaining at the expense of the medium probability habitat. Similar, and more extreme, consolidation within high probability areas was seen in the population increase scenarios, with a reduction in group fragmentation in particular leading to more than half of the occupied saiga range having a high suitability. The climate-only combined scenario, T+N) led to a shift in the distribution to the south and east, with large areas of the northern part of the spring range becoming low probability. Including the anthropogenic factors into either a ‘best’ or ‘worst’ conservation scenario illustrated that population

fragmentation and the location of the areas of high probability of presence were strongly dependent on the outcome of conservation, although this eﬀect was modulated by the predicted changes in climate. ‘Conservation success’ led to a homogenous distribution similar to period I, whereas ‘conservation failure’ led to a widely distributed, patchy population. In all three combined scenarios, there was more high probability habitat in the southeast part of the range than currently, and less in the north-west.

FUTURE PA EVALUATION

The current PA system was more eﬀective at protecting current, rather than past, high suitability areas, with the proportion of current high suitability habitat within the PA system being 23%, compared to 8% under the period I saiga distribution (Table 2, Fig. 3a,b). However it was not substantially better than random placement within the spring range, as 17% of the current spring range was rated as high suitability (Table 2). If the proposed PAs were added, the proportion of current high suitability habitat in PAs would be much improved, to 43%. The current PA system performed much better than random under increased disturbance (62% high suitability

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 35–46

Landscape-scale planning for a migratory species 41 Table 2. Percentage of high, medium and low probability habitat occurring within current PAs, proposed PAs, and current and proposed PAs combined (‘Total’), compared to the Predicted Total Area of each habitat type in the saiga spring range for Periods I and III and nine future scenarios (see Table 1 for scenario explanations) Current

Period I Period III T+ N) D+ D) P+ G+ T+N) T+N)D+ T+N)D)G+

Proposed

Total

Predicted total area

High

Medium

Low

High

Medium

Low

High

Medium

Low

High

Medium

Low

8 23 8 38 62 38 23 8 38 31 8

46 54 0 46 23 15 62 38 31 23 46

46 23 92 15 15 46 15 54 31 46 46

75 75 13 11 38 88 88 75 38 13 75

13 25 13 44 0 13 13 25 0 13 25

13 0 75 44 63 0 0 0 63 75 0

33 43 10 27 52 57 48 33 38 24 33

33 43 5 45 14 14 43 33 19 19 38

33 14 86 27 33 29 10 33 43 57 29

29 17 7 13 13 16 17 17 16 10 17

36 43 7 31 39 15 41 27 33 22 29

35 40 86 56 47 69 42 56 51 68 54

Low Medium High

80°0'0"E

The values are rounded to the nearest whole number.

60°0'0"E

70°0'0"E

Kazakhstan boundary

80°0'0"E

60°0'0"E

70°0'0"E

(a)

Betpak Dala population range Water bodies

Planned PAs

(b)

60°0'0"E

50°0'0"N

50°0'0"N

ADCI_boundary

70°0'0"E

80°0'0"E

60°0'0"E

70°0'0"E

50°0'0"N

50°0'0"N

80°0'0"E

(d)

(c)

0

100

200

400 km

Fig. 3. Maps showing the ADCI Boundary and PAs overlaid on modelled spring distribution of saiga for the (a) past and (b) current situation as well as the (c) conservation success and (d) failure scenarios (see Table 1 for details on scenarios and Fig. S6 for all other scenarios). Black cells show the predicted high suitability, grey – medium and white cells the low suitability areas for saiga.

habitat), whereas the proposed PAs performed best under reduced disturbance and increased population size. The proposed PAs performed very diﬀerently to the existing PAs in

some scenarios, which suggested a degree of complementarity which may enhance robustness to uncertainty. For example under reduced productivity, the proposed PAs performed very

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 35–46

42 N. J. Singh & E. J. Milner-Gulland badly (11%) and existing PAs improved their performance (to 27%), while if the population became larger and less fragmented, proposed PAs remained eﬀective (75%) while existing ones performed very badly (8%). This was due to changing distributions under these diﬀerent scenarios (Figs. 3c, 3d and S6, Supporting information). The PA system performed relatively poorly under all three of the combined scenarios; in all cases the combined performance dropped below current levels, substantially so in the case of conservation failure (43% high probability habitat currently, 24% projected). The actual coverage of the existing and proposed PA system was low, which was expected given the very large areas involved; the percentage of the total high quality habitat currently covered by the existing PAs was only 1%, with an additional 2% in the proposed PAs (Table S4, Supporting information).

PREDICTING FUTURE DENSITY

The density models from periods I and III produced similar results to the occupancy models, where distance to towns explained most of the variation in the model in Period III. The regression-kriging model explained 91% of the original variation and predicted densities well (Table S5 Supporting infor-

High density

DENSITY–DISTRIBUTION RELATIONSHIPS

The distribution of habitat suitability was bimodal for period I, whereas for period III and the two conservation scenarios it

80°0'0"E

60°0'0"E

70°0'0"E

80°0'0"E

(b)

(a)

50°0'0"N

70°0'0"E

50°0'0"N

60°0'0"E

mation). The maximum observed densities were 0Æ08 animals km-2 for Period I, with the groups spread in a relatively homogeneous line at a constant latitude (Fig. 4a). The stratiﬁed map of current (period III) predicted highest density showed at least three major areas of high density for saiga in the spring range, near the towns of Turgay, Irgiz and Zhezkazgan (Fig. 4b). The high density areas were shifted north and west compared to Period I. For the conservation failure scenario (T+N–D+; Fig. 4c), the predicted density was much lower on average, at 0Æ02 animals km-2 and the high density areas were more centrally located. In the conservation success scenario (T+N–D–G+; Fig. 4d), with a density of 0Æ09 animals km-2, the population was similar in distribution to Period III, but with high density areas reappearing in the east. However the results for the conservation success scenario suﬀer from a degree of circularity, given the assumption of population increase being concentrated in the largest herd sizes.

² 60°0'0"E

70°0'0"E

80°0'0"E

60°0'0"E

70°0'0"E

50°0'0"N

50°0'0"N

80°0'0"E

(d)

(c)

0

100

200

400 km

Fig. 4. Predictions of saiga spring density in Betpak Dala based on (a) Period I model (healthy past population); (b) Period III (current situation); (c) a scenario representing conservation success, in which the population recovers and group sizes get larger, with reduced disturbance but the climatic changes still occur (T+N)D)G+); and (d) a scenario representing conservation failure, in which disturbance increases and the population does not grow (T+N)D+). Refer to Table 1 for details on scenarios. 2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 35–46

Landscape-scale planning for a migratory species 43 was left skewed (Fig. S7, Supporting information). There was a signiﬁcant positive relationship between mean density and occupancy for all the models (Period I – rs = 0Æ458, n = 1998, P < 0Æ001; Period III – rs = 0Æ876, n = 1998, P < 0Æ001; Conservation success – rs = 0Æ785, n = 1998, P < 0Æ001; Conservation failure – rs = 0. 434, n = 1998, P < 0Æ001). In period III and under the conservation success scenario, the strong correlations were matched by a strong spatial overlap between high density areas and areas with a high probability of occupancy (Fig. S8, Supporting information). However in period I, when the population was large and healthy, there was a substantial swathe of the south of the range which was highly likely to be occupied, but where density was not predicted to be high. In the conservation failure scenario, with more disturbed and fragmented populations, areas of predicted high occupancy were also spatially mismatched with areas of predicted high density, except in core areas in the centre of the range (Fig. S8).

Discussion Saigas show a similar pattern to other long distance migratory ungulates, where the spring migration is driven by a rise in temperatures and an increase in productivity (Bolger et al. 2008; Singh et al. 2010a). Suitable habitat in the spring range is characterized by intermediate temperatures and productivity, is an intermediate distance from water and away from settlements. Temperature is a limiting factor for the survival of new born calves in spring, which may drive the role of temperature in deﬁning suitable habitat (Bekenov, Grachev & Milner-Gulland 1998). With suitable habitat characterized by intermediate productivity, saiga may trade-oﬀ forage quantity for quality, as observed in several other migratory ungulates (Hebblewhite, Merrill & McDermid 2008; Mueller et al. 2008; Holdo, Holt & Fryxell 2009). Sites at an intermediate distance from water may indicate avoidance of disturbance from other animals, predators and humans, which also use water bodies. Moreover, availability of water may not be limiting in spring, when precipitation is relatively high. Finally, sites away from settlements and an increase in the role of this variable with time clearly indicates avoidance of areas with high disturbance (Singh et al. 2010b), although it is not possible to determine whether this is because at low population sizes there is no need to approach settlements in order to access high quality habitat, or whether increased disturbance in recent years has driven saigas away from otherwise high quality areas. As expected, the scenarios of future change lead to signiﬁcant predicted changes in saiga spring distribution. Climate change was predicted to reduce the extent of highly suitable habitat, and produce a more heterogeneous pattern of habitat. It was also predicted to lead to shifts in highly suitable habitat with increases in some areas (especially the south east) and decreases in others (the northwest). This may have implications for migratory routes, foraging behaviour and selection of favourable areas for calving. An indirect eﬀect on population productivity might be expected. These likely eﬀects of climate

change are similar to those predicted for other migratory species (Lundberg & Moberg 2003; Shuter et al. 2010), but in our study we have been able to characterise potential changes spatially and in more detail than is commonly possible. It has also been instructive to consider how conservation outcomes may interact with climate change; successful conservation tends to mitigate the eﬀects of climate change, while conservation failure exacerbates the patchiness and loss of suitable habitat. For the saiga, as is likely for many similar species (Harris et al. 2009), disturbance currently overrides the contribution of other variables. This suggests that a conservation priority should be to reduce the eﬀects of human disturbance in otherwise highly suitable areas, for example by reducing poaching levels, limiting human use of natural watering places or restricting entry to areas of high saiga density such as calving aggregations. Currently, only a few PAs exist in the vast spring range of the Betpak Dala saiga population, covering only 1Æ05% of the current high suitability areas (Fig. S3). This is a typical picture for migratory species which cover large areas during their long-range movements (Harris et al. 2009). A particular concern for saigas is that the current and planned PAs are concentrated in the north-west of the spring range, where saiga is currently at high density. However, the eastern parts of the range were important in the 1980s and could be again, as populations increase. The eastern areas were particularly prominent in our more realistic combined scenarios. It is important to aim for geographic representativeness when planning PAs (Hodgson et al. 2009), and our analyses demonstrate that if making PAs robust to plausible future scenarios is also an aim (Pressey et al. 2007; Hannah 2008), a priority should be to improve PA coverage in areas of past and predicted high density in the east. Our study only considers the spring range, prior to calving, because this is when the available data were collected. The winter range of the saiga is under-represented in the PA system at present with only one PA, and no new PAs proposed for this area. However, the winter period is also critical for the saiga, as the winter range is more limited in extent than the other seasonal ranges and is shared with livestock, with the potential for food limitation as well as direct mortality from harsh weather (Sludskii 1963; ; Bekenov, Grachev & Milner-Gulland 1998; Coulson, Milner-Gulland & CluttonBrock 2000). There is an urgent need for similar analyses to be conducted within the winter range, but this requires data on saiga distribution at this time of year, which are currently lacking. GPS collars could provide an eﬀective way to gather this data, given the diﬃculties of aerial and ground surveys in the winter months. The recent and potential increases in patchiness of saiga distributions have important implications for monitoring, since surveying patchy distributions entails increased cost and lower reliability. In such a situation, our predicted density maps may be useful in guiding monitoring eﬀorts towards high density areas. Our results showed a signiﬁcantly positive relationship between predicted density and occupancy for a migratory species in a variety of circumstances ranging from a healthy

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 35–46

44 N. J. Singh & E. J. Milner-Gulland and abundant population phase to a critically endangered low density phase. However, the spatial mismatches observed in the healthy population phase and conservation failure scenario showed that such relationships may not follow a consistent pattern at large spatial and temporal scales, and as population status changes. Freckleton, Noble & Webb (2006) showed that when habitat suitability was left-skewed, density– occupancy relationships were positive and linear, whereas for bimodal distributions, the relationships were weak; this ﬁnding was echoed in our study. Models of habitat suitability based on data collected from disturbed populations may be misleading if density and occupancy are assumed to be proxies for each other. Overall, however, the positive relationships between density and occupancy which we found suggested that either may be useful when designing PAs or monitoring strategies.

MANAGEMENT IMPLICATIONS

Work of the type presented here is crucial in providing baselines for landscape-scale conservation initiatives. An approach such as ours may ensure that monitoring remains cost eﬀective and that PA networks are robust to future changes in distribution and densities of key target species. Here we have used the available data to model the likelihood of saiga presence and density in a particular season in a spatially explicit manner. This is a fundamental requirement for embarking on a systematic conservation planning process. We have also made some extrapolations to the future, based upon realistic scenarios. These are crude, and assume that the structure of the models governing current distribution remains the same as conditions change and as we extrapolate away from the data. However they are the best we have currently available and can give general indications of the eﬀectiveness of current and proposed PAs for saiga conservation at the landscape and decadal scales. This approach also has broad resonance for the conservation of similar long distance migratory species. A dynamic predictive approach that both assesses past and current drivers of distribution and predicts future distribution under plausible scenarios is more robust than relying on current status. It can provide insights into the contrasting drivers of current, potentially highly disrupted and fragmented distributions, and of healthier populations in the past. One potential application of the scenario-based approach is evaluation of conservation interventions against a spatially explicit set of objectives. For example, a PA network might be planned to ensure that it contains more than a threshold level of predicted high quality habitat under a range of likely scenarios of environmental and anthropogenic change.

Acknowledgements We would like to thank the Institute of Zoology, Almaty, Kazakhstan for providing access to their long term datasets and especially Dr Iu. A. Grachev and Dr A.B. Bekenov for hosting NJS in Almaty. This work was supported by the Leverhulme Trust and by a Royal Society Wolfson Research Merit Award to EJMG. Thanks to Joaquin Hortal, Miguel A´ Olalla-Ta´rraga, Steﬀen Zuther and Nils Bunnefeld for helpful comments. The editors and one anonymous

reviewer also provided valuable comments that greatly improved the manuscript.

References Baddeley, A. (2008) Analysing Spatial Point Patterns in R. CSIRO, Canberra, Australia. Bekenov, A.B., Grachev, I.A. & Milner-Gulland, E.J. (1998) The ecology and management of the saiga antelope in Kazakhstan. Mammal Review, 28, 1–52. de Beurs, K.M., Wright, C.K. & Henebry, G.M. (2009) Dual scale trend analysis for evaluating climatic and anthropogenic eﬀects on the vegetated land surface in Russia and Kazakhstan. Environmental Research Letters, 4, 045012. doi:10.1088/1748-9326 ⁄ 4 ⁄ 4 ⁄ 045012. Bolger, D.T., Newmark, W.D., Morrison, T.A. & Doak, D.F. (2008) The need for integrative approaches to understand and conserve migratory ungulates. Ecology Letters, 11, 63–77. Boyce, M., Vernier, P., Nielsen, S. & Schmiegelow, F. (2002) Evaluating resource selection functions. Ecological Modelling, 157, 281–300. Chefaoui, R.M. & Lobo, J.M. (2008) Assessing the eﬀects of pseudo-absences on predictive distribution model performance. Ecological Modelling, 210, 478–486. CMS (2010) Overview Report for the Second Meeting of the Signatories to the CMS MOU on Saiga Conservation. Convention on Migratory Species, http://www.cms.int. Bonn, Germany. Coulson, T.N., Milner-Gulland, E.J. & Clutton-Brock, T. (2000) A comparison of the relative roles of density and climatic variation on fecundity rates in three contrasting ungulate species. Proceeding of the Royal Society of London, B, 267, 1771–1779. Fadeev, V.A. & Sludskii, A.A. (1982) Saigak v Kazakhstan, Alma-Ata, Kazakhstan: Academy of Sciences, USSR, pp. 160. [The Saiga in Kazakhstan]. Freckleton, R.P., Noble, D. & Webb, T.J. (2006) Distribution of habitat suitability and the abundance-occupancy relationship. The American Naturalist, 167, 260–275. Fry, M. (2004) The status of the Saiga antelope in the Ustiurt region of western Kazakhstan. MSc Thesis, Imperial College, London. Available online at http://www.iccs.org.uk. Hannah, L. (2008) Protected areas and climate change. Annals of the New York Academy of Sciences, 1134, 201–212. Harris, G., Thirgood, S., Hopcraft, G., Cromsigt, J.P.G.M. & Berger, J. (2009) Global decline in aggregated migrations of large terrestrial mammals. Endangered Species Research, 7, 55–76. Hastie, T.J. & Tibshirani, R. (1990) Generalized Additive Models. Chapman and Hall, London. Hebblewhite, M., Merrill, E.H. & McDermid, G. (2008) A mutli-scale test of the forage maturation hypothesis for a partially migratory montane elk population. Ecological Monographs, 78, 141–166. Hengl, T. (2009) A Practical Guide to Geostatistical Mapping, 2nd edn. University of Amsterdam, Amsterdam. pp. 291, http://www.lulu.com. Hengl, T., Sierdsema, H., Radovic, A. & Dilo, A. (2009) Spatial prediction of species’ distribution from occurrence-only records: combining point pattern analysis, ENFA and regression-kriging. Ecological Modelling, 220, 3499– 3511. Hirzel, A. & Le Lay, G. (2008) Habitat suitability modelling and niche theory. Journal of Applied Ecology, 45, 1372–1381. Hodgson, J.A., Thomas, C.D., Wintle, B.A. & Moilanen, A. (2009) Climate change, connectivity and conservation decision making: back to basics. Journal of Applied Ecology, 46, 964–969. Holdo, R.M., Holt, R.D. & Fryxell, J.M. (2009) Opposing rainfall and nutrients gradients best explain the wildebeest migration in the Serengeti. The American Naturalist, 173, 431–445. IPCC – Intergovernmental Panel for Climate Change (2007) Fourth Assessment Report on Climate Change. http://www.ipcc.ch/ipccreports/assessments-reports.htm. Klebelsberg, E. (2008) The Altyn Dala Conservation Initiative: a long term commitment to save the steppe and its saigas, an endangered couple. Saiga News, 8, 1–2. Ku¨hl, A., Mysterud, A., Grachev, I.A., Bekenov, A.B., Ubushaev, B.S., Lushchekina, A.A. & Milner-Gulland, E.J. (2009) Monitoring population productivity in the saiga antelope. Animal Conservation, 12, 355–363. Lundberg, J. & Moberg, F. (2003) Mobile link organism and ecosystem functioning – implications for ecosystem resilience and management. Ecosystems, 6, 87–98. Matthiopoulos, J. & Aarts, G. (2010) The spatial analysis of marine mammal abundance. Marine Mammal Ecology and Conservation: A Handbook of

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 35–46

Landscape-scale planning for a migratory species 45 Techniques (eds I.L. Boyd, D. Bowen & S. Iverson), pp. 68–97. Oxford University Press, Oxford. McConville, A.J., Grachev, Iu.A., Keane, A., Coulson, T., Bekenov, A.B. & Milner-Gulland, E.J. (2009) Reconstructing the observation process to correct for changing detection probability of a critically endangered species. Endangered Species Research, 6, 231–237. Milner-Gulland, E.J. (1994) A population-model for the management of the saiga antelope. Journal of Applied Ecology, 31, 25–39. Milner-Gulland, E.J., Kholodova, M.V., Bekenov, A.B., Bukreeva, O.M., Grachev, Iu.A., Amgalan, L. & Lushchekina, A.A. (2001) Dramatic declines in saiga antelope populations. Oryx, 35, 340–345. Milner-Gulland, E.J., Bukreevea, O.M., Coulson, T., Lushchekina, A.A., Kholodova, M.V., Bekenov, A.B. & Grachev, I.A. (2003) Reproductive collapse in saiga antelope harems. Nature, 422, 135. Mueller, T., Olson, K.A., Fuller, T.K., Schaller, G.B., Murray, M.G. & Leimgruber, P. (2008) In search of forage: predicting dynamic habitats of Mongolian gazelles using satellite based estimates of vegetation productivity. Journal of Applied Ecology, 45, 649–658. Norton-Griﬃths, M. & McConville, A.J. (2007) Technical evaluation of the aerial survey programme for saiga antelope (Saiga tatarica) with recommendations for improvements. Final Report – ZGF-Project No. 1293 ⁄ 02 Saiga Schutz Kasachstan ⁄ Evaluation der Tierzahlungen. Frankfurt Zoological Society, Frankfurt, Germany. Pettorelli, N., Gaillard, J.M., Mysterud, A., Duncan, P., Stenseth, N.C., Delorme, D., van Laere, G., Toigo, C. & Klein, F. (2006) Using a proxy of plant productivity (NDVI) to ﬁnd key periods for animal performance. Oikos, 112, 565–572. Pressey, R.L., Cabeza, M., Watts, M.E., Cowling, R.M. & Wilson, K.E. (2007) Conservation planning in a changing world. Trends in Ecology & Evolution, 22, 583–592. Shuter, J.L., Broderick, A.C., Agnew, D.J., Jonzen, N., Godley, B.J., MilnerGulland, E.J. & Thirgood, S. (2010) Conservation and management of migratory species. Animal Migration: A Synthesis (eds E.J. Milner-Gulland, J. Fryxell & A.R.E. Sinclair), pp. 172–206. Oxford University Press, Oxford. Singh, N.J., Grachev, I.A., Bekenov, A.B. & Milner-Gulland, E.J. (2010a) Tracking greenery in central Asia – the migration of the saiga antelope. Diversity and Distributions, 16, 663–675. Singh, N.J., Grachev, I.A., Bekenov, A.B. & Milner-Gulland, E.J. (2010b) Saiga antelope calving site selection is increasingly driven by human disturbance. Biological Conservation, 143, 1770–1779. Sludskii, A.A. (1963) Dzhuts in Eurasian steppes and deserts. Trudi Instituta Zoologii AN KazSSR, [Proceedings of the Zoological Institute of the Kazakhstan Academy of Sciences], 20, 5–88. Thuiller, W. (2003) BIOMOD: Optimising predictions of species distribution and projecting potential future shifts under global change. Global Change Biology, 9, 1353–1362. Thuiller, W., Lafourcade, B., Engler, R. & Arau´jo, M.B. (2009) BIOMOD – a platform for ensemble forecasting of species distribution. Ecography, 32, 369–373. Tucker, C.J., Pinzon, J.E., Brown, M.E., Slayback, D., Pak, E.W., Mahoney, R., Vermote, E. & ElSaleous, N. (2005) An extended AVHRR 8-km NDVI data set compatible with MODIS and SPOT vegetation NDVI data. International Journal of Remote Sensing, 26, 4485–5598. Wiens, J.A. & Bachelet, D. (2010) Matching the multiple scales of conservation with the multiple scales of climate change. Conservation Biology, 24, 51–62. Wilcove, D.S. & Wikelski, M. (2008) Going, going, gone: is animal migration disappearing? PLoS Biology, 6, 1361–1364. Wilson, K.A., Underwood, E.C., Morrison, S.A., Klausmeyer, K.R., Murdoch, W.W., Reyers, B., Wardell-Johnson, G., Marquet, P.A., Rundell, P.W., McBride, M.F., Pressey, R.L., Bode, M., Hoekstra, J.M., Andelman, S., Looker, M., Rondinni, C., Kareiva, P., Shaw, M.R. & Possingham, H. (2007) Conserving biodiversity eﬃciently: what to do, where, and when. PLoS Biology, 5, 1860–1861. Wood, S.N. (2004) Stable and eﬃcient multiple smoothing parameter estimation for generalized additive models. Journal of the American Statistical Association, 99, 673–686. Zhao, M. & Running, S.W. (2010) Drought-induced reduction in global terrestrial net primary production from 2000 through 2009. Science, 329, 940–943. Zuther, S. (2009) Aerial census of the Betpak Dala population of saiga antelope. Technical Report, Altyn Dala Conservation Initiative. Association of Conservation of Biodiversity of Kazakhstan, Kazakhstan. Received 11 June 2010; accepted 20 October 2010 Handling Editor: Mark Hebblewhite

Supporting Information Additional Supporting Information may be found in the online version of this article. Fig. S1. Map of Kazakhstan showing the annual ranges of three Saiga populations: Betpak Dala (71 24¢ E, 48 7¢ N), Ustiurt and Ural. Fig. S2. Estimates of the size of the Betpak Dala population of Kazakhstan from the spring aerial surveys for the three periods I, II and III. Y axis represents saiga numbers. Data from the Institute of Zoology, Kazakhstan. Fig. S3. Past aerial survey maps of saiga population surveys from Institute of Zoology, Almaty, Kazakhstan, showing ﬂight paths and the location and size of saiga aggregations. The years shown here are (a) 1995 and (b) 2001. Fig. S4. Map showing the key PAs of diﬀerent categories and status in the study area. Map adapted from the UNDP project on steppe conservation (2010). The Green areas show the existing PAs and the blue Proposed PAs. The thick black line is the boundary of the ADCI. The three blue areas are proposed to be combined into the ‘Altyn Dala’ State Nature Reservat. Fig. S5. Percent contribution of explanatory variables in each yearly GAM for the distribution of saiga antelopes in the Betapak Dala population. NDVI – Normalized Diﬀerence Vegetation Index; Dist2water – Distance to nearest water source; Dist2towns – Distance to nearest settlement; Temperature – Mean diurnal range of temperature during the survey period. Fig. S6. Maps showing the ADCI Boundary and PAs (red lines) overlaid on modelled spring distribution of saigas under nine diﬀerent scenarios. Black cells show the predicted high suitability, grey the medium and white cells the low suitability areas for saigas. See Table 2 for the details on the scenarios. Fig. S7. Distribution of habitat suitability for saiga antelope under four diﬀerent scenarios, where habitat suitability is deﬁned as the probability of occupancy. Period I, III and Conservation Success and Failure scenarios. Fig. S8. Spatial overlap between occupancy (grey; probability >0.5) and predicted density (black; >0.05 animals km-2) of saiga antelopes in Kazakhstan. Table S1. Cross validation score for the best model for each survey year from BIOMOD (Thuiller 2003, Thuiller et al. 2009). Table S2. Estimates and standard errors for the variables used in GAMM models of saiga antelope distribution in Betpak Dala, Kazakhstan for each period. Temperature = Mean diurnal range of temperature. Period I – 1986–1996; II – 1997–2002, III – 2003–2008. Table S3. Cross-validated Spearman rank correlations (rs) between ﬁnal GAMM model bin ranks and area-adjusted frequencies for individual and average model sets for each decadal model. Table S4. Cross-validation score for the period III regression kriging model.

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 35–46

46 N. J. Singh & E. J. Milner-Gulland Table S5. Proportion of the total amount of High, Medium and Low suitability habitat in the spring saiga range that occurs within the ‘Current’ and ‘Proposed’ PAs, and in the PA system as a whole (‘Total’), for diﬀerent scenarios of change (see Table 1 for scenario details and Table 2 for the percentage of each habitat type within the PAs).

As a service to our authors and readers, this journal provides supporting information supplied by the authors. Such materials may be reorganized for online delivery, but are not copy-edited or typeset. Technical support issues arising from supporting information (other than missing ﬁles) should be addressed to the authors.

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 35–46

Journal of Applied Ecology 2011, 48, 247–256

doi: 10.1111/j.1365-2664.2010.01906.x

Habitat or fuel? Implications of long-term, post-ﬁre dynamics for the development of key resources for fauna and ﬁre Angie Haslem1,2*, Luke T. Kelly2, Dale G. Nimmo2, Simon J. Watson2, Sally A. Kenny3, Rick S. Taylor1, Sarah C. Avitabile1, Kate E. Callister1, Lisa M. Spence-Bailey1, Michael F. Clarke1 and Andrew F. Bennett2 1

Department of Zoology, La Trobe University, Bundoora Vic. 3086, Australia; 2School of Life and Environmental Sciences, Deakin University, Burwood Vic. 3125, Australia; and 3Department of Botany, La Trobe University, Bundoora Vic. 3086, Australia

Summary 1. Managing ﬁre to achieve hazard reduction while providing for biodiversity conservation is complex in ﬁre-prone regions. This challenge is exacerbated by limited understanding of post-ﬁre changes in habitat and fuel attributes over time-scales commensurate with their development, and a paucity of empirical research integrating the eﬀects of ﬁre on these attributes. 2. We used a 110-year post-ﬁre chronosequence to investigate temporal development in habitat resources used by fauna, and fuels for ﬁre in semi-arid Mallee vegetation, south-eastern Australia. Fire-history mapping previously limited investigation to 35 years post-ﬁre. The patterns of temporal change over 110 years for 13 variables, representing key attributes of habitat and fuel, were explored using nonlinear mixed models and data from 549 sites. 3. Most habitat and fuel attributes exhibited changes in abundance and rate of development over extended periods, emphasizing the importance of documenting post-ﬁre dynamics over long timeframes. Further, developmental patterns were mostly nonlinear, indicating that a shorter temporal perspective (e.g. 20–30 years post-ﬁre) may obscure, or provide an inaccurate understanding of, long-term changes. 4. There were striking diﬀerences in the post-ﬁre dynamics of some habitat and fuel attributes. Leaf litter and spinifex grass Triodia scariosa, which function as both habitat and fuel, increased rapidly after ﬁre followed by a plateau or slow decline after 20–30 years. In contrast, live tree stems were not predicted to develop hollows until 40 years, after which time the density of live hollow-bearing stems, an important habitat feature, increased steadily. 5. Synthesis and applications. Fire aﬀects the development and abundance of resources over substantially longer periods than can be examined using ﬁre-mapping based on satellite imagery. Our results demonstrate that post-ﬁre changes in mallee vegetation inﬂuence ﬁre hazard and faunal habitat in diﬀerent ways. Critically, the cover ⁄ abundance of most primary fuel sources did not increase substantially beyond around 30 years post-ﬁre; whereas important habitat attributes changed in ways that aﬀect faunal occurrence for over a century. Fire management must explicitly acknowledge the potential for ﬁre to aﬀect fauna and fuel diﬀerently, and for these eﬀects to operate over time-frames that may extend well beyond current understanding. Key-words: Australia, ﬁre chronosequence, generalized additive mixed model, mallee vegetation, prescribed ﬁre, succession, wildﬁre

*Correspondence author. E-mail: [email protected] 2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society

248 A. Haslem et al.

Introduction Fire is a natural process that shapes the structure and function of ecosystems across the globe (Bowman et al. 2009). Wildﬁre can also threaten human life and assets in ﬁre-prone regions (Bradstock 2008). For both reasons, considerable attention and resources are invested in ﬁre management, with prescribed ﬁre commonly being used to reduce the risk of uncontrollable wildﬁres (Fernandes & Botelho 2003). Fire management for hazard reduction, however, may not be compatible with conservation objectives (Morrison et al. 1996). The potential for prescribed ﬁre to negatively aﬀect biodiversity is increased by inadequate understanding of biotic responses to ﬁre (Clarke 2008; Driscoll et al. 2010) and strong public pressure to minimize ﬁre hazard (Stephens & Ruth 2005). Successfully balancing fuel reduction and biodiversity considerations is a challenge faced by managers in ﬁre-prone systems throughout the world (James & M’Closkey 2003; Ucitel, Christian & Graham 2003). A key issue in addressing the potentially competing demands of fuel reduction and biodiversity conservation is the length of time over which successional processes operate, compared with the short time-period over which changes in fuels or biota have often been documented. Techniques such as dendrochronology and radiocarbon dating have allowed examination of post-ﬁre dynamics across extended time-frames (e.g. 230 years: Pare & Bergeron 1995; 2355 years: Lecomte et al. 2006), but more commonly research is constrained to periods of known ﬁre-history (e.g. <40 years: Sah et al. 2006; Driscoll & Henderson 2008). Temporal mismatches between decades or centuries of successional change (Lecomte et al. 2006; Mack et al. 2008), and the much shorter period over which changes in biota frequently have been studied, hamper eﬀective ﬁre management. A second issue concerns understanding of post-ﬁre dynamics in fuel sources and habitat attributes used by fauna for foraging, refuge or breeding. Empirical research into ﬁre-fauna relationships has largely been undertaken in isolation from studies investigating the association between ﬁre and fuel (Ucitel, Christian & Graham 2003). Thus, there are few explicit guidelines for managers attempting to address both considerations. Post-ﬁre responses of fauna are often interpreted using a framework of secondary succession (Friend 1993; Fox, Taylor & Thompson 2003; Torre & Diaz 2004). This approach is based on the understanding that vegetation succession following ﬁre drives faunal occurrence via the diﬀerential availability of resources along post-ﬁre temporal gradients (Fox 1982). Other factors may also inﬂuence faunal responses to ﬁre, such as altered biotic interactions (Higgs & Fox 1993), abiotic conditions (Letnic et al. 2004), and ﬁre frequency ⁄ scale (Bradstock et al. 2005). However, associations between faunal occurrence and habitat attributes, and the strength of post-ﬁre succession in many vegetation types (Hanes 1971; Vandvik et al. 2005), means that understanding ﬁre–fauna relationships will beneﬁt from knowledge of changes in habitat resources over time-frames commensurate with successional processes.

Management of fuel–ﬁre relationships has been informed by the positive associations between fuel availability, wildﬁre severity, and time-since-ﬁre (Hanes 1971; Sah et al. 2006). Thus, the use of prescribed ﬁre is often guided by fuel characteristics, and understanding temporal patterns in fuel accumulation (Fernandes & Botelho 2003). For example, estimates of fuel loads are widely used to measure ﬁre hazard in parts of southern Australia (McCarthy, Tolhurst & Chatto 1999). Such estimates can identify maximum tolerable fuel loads (Burrows 2008) and, when combined with predictions of fuel accumulation, suggest optimum inter-ﬁre intervals for hazard reduction (Department of Sustainability and Environment 2008). Documenting temporal patterns in fuel development is critical if such management approaches are to eﬀectively reduce wildﬁre hazard (Sah et al. 2006). Our study system, the Murray Mallee region in south-eastern Australia, is strongly inﬂuenced by ﬁre. Large wildﬁres (c. 100 000 ha) typically occur somewhere in the region on a bi-decadal basis (Noble & Vines 1993), and prescribed ﬁre is used to minimize the risk of wildﬁres burning extensive areas (Sandell et al. 2006). Fire also aﬀects the status of native fauna (Brown, Clarke & Clarke 2009; Kelly et al. 2010). Current ﬁre-history mapping based on satellite imagery restricts investigation of post-ﬁre responses in habitat and fuel to a 35-year period. We use empirical data and age predictions for sites of previously unknown ﬁre-age (Clarke et al. 2010) to investigate post-ﬁre dynamics over a chronosequence extending to 110 years. We address two objectives: (i) to document post-ﬁre resource dynamics across a 110-year chronosequence, and establish whether extending the time-scale under examination alters the resultant understanding of post-ﬁre change; and (ii) to explicitly compare post-ﬁre trajectories of habitat attributes and fuel sources.

Materials and methods STUDY AREA

The study area encompasses 100 000 km2 of the Murray Mallee, a low-lying region covering parts of Victoria, South Australia and New South Wales, Australia (see Fig. S1, Supporting Information). The climate is semi-arid (220–330 mm rainfall ⁄ year) with high summer temperatures (mean daily maxima ‡32 C) and mild winters (mean daily maxima 16 C) (data supplied by the Australian Bureau of Meteorology). The region is characterized by extensive dune ⁄ swale systems that reﬂect underlying variation in soil characteristics and moisture availability (Land Conservation Council 1987). The current distribution of native vegetation has been strongly inﬂuenced by agriculture, primarily cereal cropping and stock grazing: remnant native vegetation occurs predominantly on less fertile soils of the duneﬁelds (Land Conservation Council 1987). The most common vegetation, tree mallee, comprises short (typically <5 m), multi-stemmed (‘mallee’) Eucalyptus trees above an understorey of shrubs and perennial and ephemeral grasses. An important feature of this vegetation is the ability of mallee eucalypts to regenerate from underground lignotubers after ﬁre by coppicing multiple new stems (Gill 1981). Fire is a dominant process in the study area. Tree mallee vegetation is highly ﬂammable and the reproductive strategies of many plant

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 247–256

Habitat or fuel? Post-ﬁre resource dynamics 249 species are strongly tied to ﬁre events (Bradstock & Cohn 2002). Key components of perennial fuels include the hummock-forming grass Triodia scariosa N.T. Burbidge, leaf litter accumulated beneath trees, and decorticating ribbons of bark (Bradstock & Cohn 2002). After large rainfall events, ephemeral grasses also provide a critical fuel source that connects the otherwise patchily-distributed perennial fuels (Noble & Vines 1993). Wildﬁres are actively suppressed in much of the region and prescribed ﬁre is used predominantly to reduce ﬁre hazard (Sandell et al. 2006).

STUDY DESIGN

This study is part of a project examining landscape-level responses of multiple taxa to the properties of ﬁre mosaics. Consequently, the 549 sites included here were grouped in 26 clusters, each representing an individual landscape unit of 12Æ6 km2 (Fig. S1). The mean distance between study landscapes was 130 km (range: 6–218 km) and all were located within conservation reserves. Previous work identiﬁed and mapped three broad types of tree mallee vegetation across the region (Haslem et al. 2010). Here, we focus on the two most extensive types: Triodia Mallee (TM) and Chenopod ⁄ Shrubby Mallee (hereafter Chenopod Mallee, CM). These communities diﬀer on the basis of canopy dominants, understorey composition, soil and topography. Triodia Mallee is characterized by an increased abundance of T. scariosa, an important habitat component for fauna (Bennett, Lumsden & Menkhorst 2006) and a key source of fuel (Noble, Smith & Leslie 1980). The understorey of Chenopod Mallee comprises a diversity of low chenopod shrubs, many of which have low ﬂammability (Pausas & Bradstock 2007). We employed a space-for-time approach to investigate post-ﬁre temporal dynamics, with sites of diﬀering ﬁre-ages representing a post-ﬁre chronosequence. Fire in tree mallee vegetation removes both the canopy and understorey (Caughley 1985), resetting the system to ‘year zero’. Importantly, for this reason both wildﬁres and prescribed ﬁres in this system are commonly ‘stand-replacing’ and have similar eﬀects on post-ﬁre succession. To ensure results were not inﬂuenced by ﬁre intensity, sites considered to have been burnt only partially ⁄ patchily in the most recent ﬁre, based on the occurrence of multiple cohorts of eucalypt stems, were excluded from analyses.

DATA COLLECTION

A range of vegetation characteristics was assessed at each site. The percentage cover of diﬀerent types of ground cover (bare ground, cryptogamic crust, leaf litter, plant matter, logs), and the depth of leaf litter, were recorded at 1-m intervals along a 50-m transect (n = 50 sample points). Vegetation structure was assessed at each sample point by recording the number of vegetation contacts with a vertical ranging pole in four height strata (<0Æ5 m, 0Æ5–1 m, 1–2 m, >2 m). Quadrats were established to record: (i) the characteristics of Eucalyptus trees and stems (50 · 4-m quadrat) including canopy height, number of trees, number stems ⁄ tree, stems with hollows, amount of decorticating bark ⁄ tree; and (ii) the number of logs (50 · 10-m quadrat). Vegetation data were collected between June and August 2007. We selected 13 variables to represent habitat attributes for fauna, and fuel sources for ﬁre (Table 1). These variables include measures of the availability of speciﬁc habitat or fuel resources (e.g. T. scariosa, leaf litter, tree hollows), and characteristics of habitat structure and fuel distribution (e.g. vegetation cover in diﬀerent height strata). They are referred to as ‘habitat’ when viewed from the perspective of fauna, and ‘fuel’ when considered in relation to ﬁre; many constitute both. For example, Triodia hummocks and leaf litter are used for foraging

and refuge by many of the 55 reptile species occurring in the region; but also are a primary source of surface fuels that sustain ﬁres. Ephemeral grasses, associated with large ﬁres (Noble, Smith & Leslie 1980), were not investigated as their occurrence is more strongly inﬂuenced by rainfall than time-since-ﬁre. The ﬁre-age of study sites was determined using two methods. Landsat satellite imagery from 15 individual years (1972–2007), combined with local knowledge, was used to identify the exact year of the most recent ﬁre for sites burnt since 1972. For sites burnt before 1972, the lack of historical records and satellite imagery necessitated an alternative approach. We used linear regression models to quantify the relationship between tree age (indicated by ﬁre-year) and mean stem diameter for each eucalypt species at sites of known ﬁre-year (Clarke et al. 2010). These models were then used to predict tree age (thus infer ﬁre-year) for sites where ﬁre-year was unknown but stem diameter data were available. Validation of these models with independent data from new sites revealed a highly signiﬁcant correlation between known and predicted tree ages (r = 0Æ71, P < 0Æ001, n = 88) (Clarke et al. 2010) conﬁrming the reliability of this approach. The ﬁre-age of sites was calculated as the diﬀerence between actual ⁄ predicted ﬁre-year and 2007 (when vegetation data were collected). Sites with a predicted ﬁre-age >110 years were excluded from analyses due to low sample sizes. Table 2 shows the distribution of sites across the 110-year ﬁre chronosequence, and two vegetation types, investigated. The lack of sites aged between 11 and 20 years reﬂects reduced ﬁre activity between 1987 and 1996.

STATISTICAL ANALYSES

We used generalized additive mixed models (GAMMs: Wood 2006; Zuur et al. 2009) to investigate patterns in the development of each habitat ⁄ fuel attribute across the post-ﬁre chronosequence. Three factors contributed to the selection of this approach. First, inspection of raw data showed that nonlinear models were appropriate. Generalized additive models (GAMs) are nonparametric regression models that use smoothing functions to ﬁt nonlinear response curves (Wood 2006). Secondly, the clustered distribution of sites (grouped in landscape units) suggested a mixed model approach. Mixed models are recommended when data are structured by some factor (here, landscape unit) that introduces systematic variation of potential inﬂuence over the relationship between predictor and response variables (Zuur et al. 2009). Lastly, diﬀerences in the structure, ﬂoristic composition and abiotic associations of Triodia Mallee and Chenopod Mallee indicated that time-since-ﬁre responses might diﬀer between vegetation types. To account for this, we used ‘variable coeﬃcient’ GAMMs which produce diﬀerent smoothed terms for each level of a categorical variable (Wood 2006; Zuur et al. 2009). Models were ﬁtted with ‘landscape’ as a random eﬀect and a separate smoothed term for time-since-ﬁre in each vegetation type. The amount of smoothing used to model time-since-ﬁre was selected internally during the model-ﬁtting process (Wood 2004). Outliers, as identiﬁed by residual plots, were removed from ﬁnal models (see Table 3). Models were evaluated using a measure of model ﬁt (deviance explained) and cross-validation. Sevenfold cross-validations were used to assess the stability and predictive accuracy of models (Pearce & Ferrier 2000). This involved randomly dividing study landscapes into seven groups (‘folds’), ﬁtting a model to data from six folds, then using it to predict to data from the seventh fold. This process continued until all sites had predictions derived from independent data. The mean correlation (and associated

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 247–256

250 A. Haslem et al. Table 1. Variables used to represent habitat and fuel attributes in tree mallee vegetation, and their roles in providing these resources Role of resourcea Description

Habitatb

Fuelc

Abbreviation

Percent cover of Triodia scariosa (<0Æ5 m high)

F⁄R⁄B

S,f

Triodia

Leaf litter

Mean depth (cm) of leaf litter ‡1 cm deep

F⁄R⁄B

S,f

Litter

Overall ground fuel

Combined percent cover of leaf litter (‡1 cm deep), plant matter and logs

F⁄R

S,v

GroundFuel

Logs

Density of logs >3 cm diameter and >50 cm length (no. ⁄ ha)

F⁄R⁄B

S,c

Logs

Decorticating bark

Mean ordinal measure across trees, where: (i) no hanging bark (>30 cm in length); (ii) hanging bark present <50% stem surface area; (iii) hanging bark present >50% stem surface area

F⁄R

L,c

Bark

Tree hollows – live stems

Estimated density of live hollow-bearing tree stems (no. ⁄ ha)

R⁄B

n⁄a

Hollows(Live)

Tree hollows – dead stems

Estimated density of dead hollow-bearing tree stems (no. ⁄ ha)

R⁄B

n⁄a

Hollows(Dead)

Tree hollows – proportion stems

Overall proportion of tree stems containing a tree hollow

R⁄B

n⁄a

HollowProportion

Collectively represent habitat structure rather than speciﬁc habitat resources

Represents coarse standing fuel density

StemDensity

Variable Speciﬁc resource types Triodia scariosa

Habitat structure ⁄ fuel distribution Tree stems Estimated density of tree stems (no. ⁄ ha)

Low vegetation cover

Percent cover of plant matter (dead or alive) <0Æ5 m high

S,v

LowCover

Mid vegetation cover

Percent cover of plant matter (dead or alive) 0Æ5–2 m high

U,v

MidCover

Canopy vegetation cover

Percent cover of plant matter (dead or alive) >2 m high

C,v

CanopyCover

Canopy heightd

Canopy height (m) determined using a range ﬁnder

Represents canopy fuel height

CanopyHeight

a

Information collated from cited references. F = foraging habitat, R = refuge ⁄ shelter habitat, B = breeding habitat. c S = surface fuel, U = understorey fuel, L = ladder fuel, C = canopy fuel, f = ﬁne fuel (<3 cm), c = coarse fuel (>3 cm), v = variable fuel size. d Data available for a reduced number of sites (see Table 3). b

standard error) between observed and predicted values, averaged across folds, was used to evaluate models. The median of three crossvalidation trials is reported. Regression modelling and cross-validation were undertaken using the mgcv package v.1.4–1 (Wood 2004) and source scripts (also used to calculate model deviance: Elith, Leathwick & Hastie 2008) in R v.2.8.0 (R Development Core Team 2008).

Results All habitat ⁄ fuel attributes except the density of dead hollowbearing stems, Hollows(Dead), showed a signiﬁcant relationship with time-since-ﬁre (Table 3). Two attributes, Triodia and MidCover, exhibited signiﬁcant post-ﬁre responses in only one

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 247–256

Habitat or fuel? Post-ﬁre resource dynamics 251 Table 2. Distribution of 549 study sites across the 110-year post-ﬁre chronosequence. Years-since-ﬁre has been split into 10-year intervals for Triodia Mallee (TM) and Chenopod Mallee (CM) vegetation Number of sites Years-since-ﬁre

TM

CM

1–10 11–20 21–30 31–40 41–50 51–60 61–70 71–80 81–90 91–100 101–110 Total

89 0 50 124 15 28 22 13 15 8 5 369

12 0 6 51 22 30 25 12 10 7 5 180

Table 3. Results of models describing the relationship between habitat ⁄ fuel attributes and time-since-ﬁre. Details of the smoothed terms for time-since-ﬁre in Triodia Mallee (TM) and Chenopod Mallee (CM) are shown for each attribute, together with the number of sites sampled in each Smoothed term for time-since-ﬁre Habitat ⁄ fuel attribute

Vegetation type

Sites (no.)

edf a

F

P

Triodia

TM CM TMb CMb TM CM TMb CM TM CM TM CM TM CM TM CM TM CMb TM CM TM CM TM CM TM CM

369 180 367 178 369 180 368 180 368 180 369 180 369 180 369 180 369 179 369 180 369 180 369 180 241 130

5Æ33 1Æ00 5Æ26 3Æ27 6Æ06 3Æ63 6Æ71 1Æ00 1Æ83 1Æ00 3Æ78 2Æ70 1Æ00 1Æ00 3Æ03 1Æ00 4Æ39 4Æ74 5Æ66 1Æ00 6Æ68 1Æ00 5Æ59 3Æ08 4Æ48 3Æ84

15Æ55 0Æ12 17Æ11 4Æ85 26Æ07 6Æ00 3Æ72 4Æ94 54Æ04 29Æ06 12Æ53 11Æ22 0Æ27 2Æ22 16Æ85 24Æ62 21Æ26 19Æ37 4Æ54 4Æ18 7Æ07 0Æ10 32Æ19 5Æ87 98Æ91 33Æ61

<0Æ001 0Æ824 <0Æ001 0Æ001 <0Æ001 <0Æ001 0Æ001 0Æ014 <0Æ001 <0Æ001 <0Æ001 <0Æ001 0Æ696 0Æ124 <0Æ001 <0Æ001 <0Æ001 <0Æ001 <0Æ001 0Æ026 <0Æ001 0Æ853 <0Æ001 <0Æ001 <0Æ001 <0Æ001

Litter GroundFuel Logs Bark Hollows(Live) Hollows(Dead) HollowProportion StemDensity LowCover MidCover CanopyCover CanopyHeight

a

Estimated degrees of freedom. One outlier removed (large residual value).

b

vegetation type: Triodia Mallee. Post-ﬁre dynamics did not diﬀer markedly between vegetation types for most attributes, as shown by overlapping conﬁdence intervals of predictions (Fig. 1).

Most attributes exhibited nonlinear patterns of change following ﬁre (Fig. 1); the trend and rate of their development were not consistent across the 110-year post-ﬁre chronosequence. For example, tree stem density increased relatively rapidly for the ﬁrst 20 years following ﬁre and then declined; rapidly at ﬁrst and then at a slower rate after around 50 years post-ﬁre (Fig. 1). Linear responses to time-since-ﬁre were observed only in Chenopod Mallee: the variables HollowProportion and Bark increased, while Logs and LowCover decreased, as time-since-ﬁre increased (Fig. 1). No attribute had reached a stable state by 35 years-sinceﬁre, the current extent of ﬁre-history records (Fig. 1). The rate of development in some attributes did decrease after 35 years (e.g. GroundFuel [CM], CanopyCover [TM]) but few showed no further change after this time. In contrast, many attributes still exhibited change at 110-years post-ﬁre (e.g. HollowProportion, Bark, Triodia [TM], Litter [TM]). All attributes, except Bark, showed a marked change in the rate of development at around 20–30 years post-ﬁre (Fig. 1). Developmental shifts took one of three forms: (i) rapid increase followed by slower increase or plateau (Litter, GroundFuel, LowCover, CanopyCover, CanopyHeight); (ii) rapid increase followed by decline (Triodia, Logs, StemDensity, MidCover), or (iii) minimal change followed by increase (Hollows(Live), HollowProportion). Both methods used to evaluate models, cross-validation and model ﬁt, returned similar results (Fig. 2). CanopyHeight, Triodia and CanopyCover showed the strongest relationship with time-since-ﬁre, and these models had relatively high predictive accuracy and stability. Time-since-ﬁre explained comparatively less variation (<15%) in MidCover and Logs, and the mean cross-validation correlation for these attributes was low (<0Æ3). Between 24% and 37% of variation in the remaining eight attributes was related to time-since-ﬁre, with cross-validation indicating a moderate performance of models explaining these relationships (Fig. 2). To further examine temporal development in tree hollows, the mean proportion of hollow-bearing stems that were alive and dead was compared for each 10-year post-ﬁre interval (Fig. 3). On average, over 80% of all hollow-bearing stems were dead in ﬁre-age periods £40 years-since-ﬁre. Parity in the proportion of live and dead stems containing hollows was not reached until 60 years since last burn.

Discussion Most habitat and fuel attributes investigated here showed a signiﬁcant response to time-since-ﬁre, emphasizing the inﬂuence of ﬁre on the temporal availability of resources used by fauna, and fuel accumulation patterns. Time-since-ﬁre had a stronger eﬀect on some attributes than others: over half the variation in the cover of the hummock grass Triodia scariosa was attributed to the inﬂuence of time-since-ﬁre; whereas it explained only 10% of variation in log density. Many factors other than timesince-ﬁre aﬀect the abundance and development of these resources. For example, other potential inﬂuences on log

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 247–256

252 A. Haslem et al. 25

Triodia

2·5

50

Litter

20

2·0

40

15

1·5

30

10

1·0

20

5

0·5

10

0

0

0

2·0

Bark

300

Hollows(Live)

250

1·5

200 150

1·0

100

0·5

50 0

0 50

50

LowCover

MidCover

50

CanopyCover

8

30

30

30

20

20

20

10

10

10

20

40 60 80 Years since fire

100

20

40 60 80 Years since fire

100

StemDensity

7000 6000 5000 4000 3000 2000 1000 0

40

0

Logs

HollowsProportion

40

0 0

700 600 500 400 300 200 100 0

0·35 0·30 0·25 0·20 0·15 0·10 0·05 0

40

0

GroundFuel

0

6 4 2 0

20

40 60 80 Years since fire

100

0

CanopyHeight 0

20

40 60 80 Years since fire

100

90

0·9

80

0·8

70

0·7

60

0·6

50

0·5

40

0·4

30

0·3

20

0·2

10

Cross-validation results

Deviance explained (%)

Fig. 1. Predicted post-ﬁre dynamics in habitat ⁄ fuel attributes across a 110-year time-frame. Predicted trends and their 95% conﬁdence intervals are shown for Triodia Mallee (black) and Chenopod Mallee (grey). Vertical lines indicate the extent of temporal understanding based on ﬁre-history records. See Table 1 for Y-axis measurement unit.

0·1

0 Logs

MidCover

HollowProportion

Bark

Litter

StemDensity

LowCover

GroundFuel

Hollows(live)

CanopyCover

Triodia

CanopyHeight

0

Fig. 2. Measures used in model evaluation: (a) percentage deviance explained (crosses); (b) mean cross-validation correlation (squares, including standard error bars).

density include inter-ﬁre interval and termite activity (Whitford, Ludwig & Noble 1992).

LONG-TERM RESOURCE DYNAMICS FOLLOWING FIRE

Fire-age predictions for sites of previously unknown age (Clarke et al. 2010) allowed us to investigate post-ﬁre resource dynamics over a chronosequence three times that provided by ﬁre-history mapping. Extending the chronosequence provided a greater understanding of post-ﬁre changes. First, it highlighted that diﬀerent resources develop and change in abundance over diﬀerent time-scales. For example, leaf litter depth and the cover of Triodia hummocks increased rapidly post-ﬁre, reaching a maximum within 20–30 years (Fig. 1). Other

attributes displayed a much longer process of development: canopy height reached an asymptote at around 60 years, while the proportion of tree stems with hollows, and the abundance of decorticating bark, continued to increase for over a century (Fig. 1). Such variation in the temporal scale of resource development makes identifying inter-ﬁre intervals that are appropriate for all management objectives diﬃcult. Secondly, it revealed how attributes change in abundance over long time-frames. The density of fallen timber, for example, varies through time. In Triodia Mallee vegetation, it peaks in the ﬁrst 20 years post-ﬁre as dead stems collapse, declines over the subsequent 20 years as these stems decay, and gradually increases from 50 to 100 years post-ﬁre as the next cohort of maturing trees shed limbs. The nonlinearity of long-term trends identiﬁed here, and in other systems (e.g. Pare & Bergeron 1995; Hall, Burke & Hobbs 2006), has important implications. It suggests that reduced temporal understanding may lead to an inaccurate interpretation of post-ﬁre dynamics. Trends observed over the ﬁrst 35 years of the chronosequence (the limit of known ﬁre-history) were not indicative of patterns observed in following decades for most attributes. This may have repercussions if feedback mechanisms govern the relationship between vegetation characteristics and ﬁre (e.g. Bradstock 1989a), as management based on shorter-term understanding may favour some species and communities at the expense of others.

TEMPORAL DEVELOPMENT OF HABITAT AND FUEL ATTRIBUTES

The relatively rapid increase in the cover of Triodia hummocks and the depth of litter layers following ﬁre (Fig. 1) has implications for fauna, as well as fuel accumulation. Mallee vegetation

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 247–256

Habitat or fuel? Post-ﬁre resource dynamics 253 1·0

Fig. 3. Proportion of hollow-bearing stems that were dead (black) and alive (grey), averaged across sites in time-since-ﬁre intervals of 10 years. Data for Triodia Mallee and Chenopod Mallee have been pooled; mean numbers of hollow-bearing stems ⁄ tree are shown for sites in each interval.

Proportion stems with hollows

0·15

0·37

0·26

0·45

0·53

0·46

0·52

0·67

21–30

31–40

41–50

51–60

61–70

71–80

81–90

0·71

0·84

0·8

0·6

0·4

0·2

0·0 1–10

11–20

may be suitable for Triodia-associated reptiles, such as southern mallee ctenotus Ctenotus atlas, within six years post-ﬁre (Caughley 1985), while at least 15 years is required before vegetation is suitable for the Triodia-dependent mallee emu-wren Stipiturus mallee (Brown, Clarke & Clarke 2009). In terms of its development as a fuel source, it takes around 15 years before Triodia hummocks carry ﬁre (Noble & Vines 1993). In contrast, leaf litter often accumulates rapidly to ﬂammable levels, sometimes within ﬁve years (Raison, Woods & Khanna 1983). It takes longer for litter layers to provide adequate habitat for the litter-nesting malleefowl Leiopoa ocellata, which exhibits highest breeding-density 60 years post-ﬁre (Benshemesh 1989); and litter-dependent reptiles, including Boulenger’s skink Morethia boulengeri Ogilby, which are most likely to occur 100 years post-ﬁre (D. Nimmo unpublished data). However, strong fauna–habitat associations do not always translate into predictable fauna–ﬁre relationships (Driscoll & Henderson 2008). Nonetheless, variation in faunal responses to time-since-ﬁre (see also Briani et al. 2004) highlights the complexity of managing ﬁre for biodiversity conservation, even before hazard reduction imperatives are considered. The slow development of some resources over long periods also has implications for the occurrence of fauna, and the provision of fuel sources. Recently burnt areas are unlikely to provide suﬃcient decorticating bark for species that use this resource, such as the endangered black-eared miner Manorina melanotis which forages amongst bark ribbons (Woinarski 1999). As a fuel resource, hanging bark will likewise have greater inﬂuence on ﬁre behaviour in older tree mallee vegetation. Dead trees present similar challenges in the USA, where park management often involves the removal of this fuel type, yet dead trees provide important microhabitats used by many lizard species for shelter, foraging and courting (James & M’Closkey 2003). Species that depend on tree hollows, such as the hollow-nesting striated pardalote Pardalotus striatus (Woinarski 1999), are unlikely to ﬁnd tree mallee suitable for all their requirements until it is at least 40 years old (Fig. 1). It will take many more decades (>100 years) before live eucalypt stems are of a diameter suitable for large hollow-nesting species

91–100 101–110

Years since fire

like Major Mitchell’s cockatoo Cacatua leadbeateri (Clarke 2005). Recurrent ﬁres within 40 years will result in hollows being provided predominately by dead stems (Fig. 3) which are more susceptible to ﬁre. While ﬁre may create hollows or increase their rate of development (Inions, Tanton & Davey 1989), our results suggest it will not markedly increase hollow availability in mallee vegetation, especially in the long-term. The occurrence of hollow-bearing stems was proportionally lowest in early post-ﬁre years, and the density of dead hollowbearing stems was not related to time-since-ﬁre. The identiﬁcation of habitat resources that develop over many decades, and the documentation of associated developmental patterns, can inform ﬁre management by identifying minimum and maximum ﬁre intervals for fauna, as undertaken using plant species attributes (e.g. age at ﬁrst seed set, longevity: Noble & Slatyer 1980). Management based primarily on ﬁre intervals derived from plant attributes will not be adequate for the provision of all faunal requirements, due to the extended time-frames over which some habitat resources develop (Clarke 2008). Fuel continuity plays an important role in inﬂuencing wildﬁre behaviour (Van Wilgen, Lemaitre & Kruger 1985). The continuity of most primary fuels in tree mallee vegetation does not increase substantially beyond around 30 years since ﬁre. After this time, there is a greater distance between surface and canopy fuels, fewer eucalypt stems (coarse standing fuels), and a reduced cover of understorey fuel. The ‘opening-up’ of vegetation as time-since-ﬁre increases has been observed in this and other systems (Hanes 1971; Clarke, Boulton & Clarke 2005). Nonetheless, there is strong potential for bark to contribute to spotting behaviour in ﬁres (Bradstock & Cohn 2002). Bark continues to accumulate on trees after 30 years post-ﬁre, and may increase fuel continuity and ﬁre-spread in older mallee vegetation. Post-ﬁre changes in vegetation structure and complexity also have implications for faunal habitats and assemblages (Catling, Coops & Burt 2001; Fox, Taylor & Thompson 2003). For example, the composition of bird communities in tree mallee vegetation aged between 10–30 years diﬀers from

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 247–256

254 A. Haslem et al. communities occupying older vegetation with a sparser understorey and taller canopy (Woinarski 1999; S. Watson, unpublished data). Patterns of vegetation succession represent a sequence of habitats which beneﬁt diﬀerent species; hence, descriptions of post-ﬁre change in habitat structure do not provide clear guidance for ecological ﬁre management unless ﬁre plans target particular species (e.g. MacHunter, Menkhorst & Loyn 2009).

FURTHER IMPLICATIONS FOR FIRE MANAGEMENT

Our results demonstrate that the temporal scale of investigation inﬂuences the perception of post-ﬁre resource dynamics. Critically, extension of the chronosequence alters understanding of ﬁre–fauna and fuel–ﬁre relationships in diﬀerent ways, such that management recommendations derived from shorter (<30 years) and longer (110 years) temporal scales may diﬀer. Results from the longer, 110-year chronosequence indicate greater scope for integrating management for conservation and hazard reduction objectives. Fuel quantity and continuity increased rapidly in early post-ﬁre years, as previously documented in this (Bradstock 1989b; Noble & Vines 1993) and other systems (Raison, Woods & Khanna 1983; Sah et al. 2006). However, the longer perspective revealed that most primary sources of fuel did not increase substantially after around 30 years-since-last burn, suggesting a potential plateau in fuel-related ﬁre hazard. In contrast, a short-term perspective precludes appreciation of the importance of the long-term development of some habitat resources (e.g. mature canopy layer, tree hollows) and ongoing changes in vegetation structure and complexity. In combination, these insights suggest a reduced urgency to burn ‘long-unburnt’ mallee vegetation for the purpose of hazard reduction, and a corresponding opportunity for ﬁre planners to focus greater attention on conservation objectives. Attributes such as Triodia hummocks, leaf litter and decorticating bark provide both important resources for fauna and fuel for ﬁre (Bradstock & Cohn 2002). At local scales, management to reduce these fuels is incompatible with the requirements of animals using these resources for foraging or shelter (e.g. James & M’Closkey 2003; Ucitel, Christian & Graham 2003). Strategic use of prescribed ﬁre to reduce the risk of wildﬁre involves a trade-oﬀ. Maintaining vegetation in early postﬁre conditions as a ‘ﬁre-break’ may prevent extensive wildﬁre (e.g. >100 000 ha) and facilitate the persistence of a greater proportion of the landscape at more advanced stages along the post-ﬁre chronosequence (Sandell et al. 2006). However, the use of prescribed ﬁre over large areas (e.g. landscape burning) demands careful consideration of the overall amount and spatial distribution of vegetation of all ﬁre-ages (Clarke 2008), and recognition that some habitat components continue to develop for at least a century after ﬁre. In contrast with tree mallee vegetation, where prescribed ﬁres and wildﬁres result in similar post-ﬁre changes, in many other ecosystems (such as forests) diﬀering ﬁre severity combined with taller canopy vegetation mean that prescribed ﬁre and wildﬁre may initiate diﬀerent post-ﬁre trajectories

(e.g. Sah et al. 2006). Consequently, the resources available in prescribed ﬁre and wildﬁre scars of identical ﬁre-ages will also diﬀer. Further research is needed before explicit recommendations can be made about the spatial properties of ﬁre plans, and the eﬀect of diﬀerential post-ﬁre trajectories (e.g. prescribed ﬁre cf. wildﬁre) on resource availability at broad spatial extents. Nevertheless, the impacts of ﬁre on the dynamics of fuel and habitat attributes must be considered in ﬁre planning if management is to successfully contribute to both hazard reduction and conservation objectives. Furthermore, the potential for post-ﬁre resource dynamics to operate over temporal scales that exceed current understanding must be factored into management plans.

Acknowledgements Funding and support were received from Parks Victoria, Department of Sustainability and Environment (Vic), Mallee Catchment Management Authority, NSW National Parks and Wildlife Service, Department of Environment and Climate Change (NSW), Lower Murray Darling Catchment Management Authority, Department for Environment and Heritage (SA), Land and Water Australia, Natural Heritage Trust, Birds Australia (Gluepot Reserve), Australian Wildlife Conservancy (Scotia Sanctuary), and the Murray Mallee Partnership. We are grateful to the Doyle and Barnes families for access to Petro and Lethero Stations, respectively. Thanks to Lauren Brown and many volunteers who assisted with vegetation surveys. Jane Elith provided helpful advice on statistical analyses. We thank editorial staﬀ and two referees for constructive feedback on this manuscript.

References Bennett, A.F., Lumsden, L.F. & Menkhorst, P.W. (2006) Mammals of the mallee region, Victoria: past, present and future. Proceedings of the Royal Society of Victoria, 118, 259–280. Benshemesh, J. (1989) Management of the malleefowl with regard to ﬁre. The Mallee Lands: A Conservation Perspective (eds P.J. Joss & G.K. Jones), pp. 206–211. CSIRO, Adelaide. Bowman, D., Balch, J.K., Artaxo, P., Bond, W.J., Carlson, J.M., Cochrane, M.A., D’Antonio, C.M., DeFries, R.S., Doyle, J.C., Harrison, S.P., Johnston, F.H., Keeley, J.E., Krawchuk, M.A., Kull, C.A., Marston, J.B., Moritz, M.A., Prentice, I.C., Roos, C.I., Scott, A.C., Swetnam, T.W., van der Werf, G.R. & Pyne, S.J. (2009) Fire in the earth system. Science, 324, 481–484. Bradstock, R.A. (1989a) Dynamics of a perennial understorey. Mediterranean Landscapes in Australia, Mallee Ecosystems and Their Management (eds J.C. Noble & R.A. Bradstock), pp. 141–154. CSIRO, Australia. Bradstock, R.A. (1989b) Relationships between ﬁre regimes, plant species and fuels in mallee communities. The Mallee Lands: A Conservation Perspective (eds P.J. Joss & G.K. Jones), pp. 218–223. CSIRO, Adelaide. Bradstock, R.A. (2008) Eﬀects of large ﬁres on biodiversity in south-eastern Australia: disaster or template for diversity? International Journal of Wildland Fire, 17, 809–822. Bradstock, R.A. & Cohn, J.S. (2002) Fire regimes and biodiversity in semi-arid mallee ecosystems. Flammable Australia: The Fire Regimes and Biodiversity of a Continent (eds R.A. Bradstock, J.E. Williams & A.M. Gill), pp. 238– 258. Cambridge University Press, Cambridge. Bradstock, R.A., Bedward, M., Gill, A.M. & Cohn, J.S. (2005) Which mosaic? A landscape ecological approach for evaluating interactions between ﬁre regimes, habitat and animals. Wildlife Research, 32, 409–423. Briani, D.C., Palma, A.R.T., Vieira, E.M. & Henriques, R.P.B. (2004) Post-ﬁre succession of small mammals in the Cerrado of central Brazil. Biodiversity and Conservation, 13, 1023–1037. Brown, S., Clarke, M. & Clarke, R. (2009) Fire is a key element in the landscape-scale habitat requirements and global population status of a threatened bird: the Mallee Emu-wren (Stipiturus mallee). Biological Conservation, 142, 432–445. Burrows, N.D. (2008) Linking ﬁre ecology and ﬁre management in south-west Australian forest landscapes. Forest Ecology and Management, 255, 2394– 2406.

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 247–256

Habitat or fuel? Post-ﬁre resource dynamics 255 Catling, P.C., Coops, N.C. & Burt, R.J. (2001) The distribution and abundance of ground-dwelling mammals in relation to time since wildﬁre and vegetation structure in south-eastern Australia. Wildlife Research, 28, 555–564. Caughley, J. (1985) Eﬀect of ﬁre on the reptile fauna of Mallee. Biology of Australasian Frogs and Reptiles (eds G. Grigg, R. Shine & H. Ehmann), pp. 31–34. Surrey Beatty and Sons, Sydney. Clarke, R.H. (2005) Ecological Requirements of Birds Specialising in Mallee Habitats: Modelling the Habitat Suitability for Threatened mallee Birds. Department of Zoology, La Trobe University, Melbourne. Clarke, M.F. (2008) Catering for the needs of fauna in ﬁre management: science or just wishful thinking? Wildlife Research, 35, 385–394. Clarke, R.H., Boulton, R.L. & Clarke, M.F. (2005) Estimating population size of the Black-eared Miner, with an assessment of landscape-scale habitat requirements. Paciﬁc Conservation Biology, 11, 174–188. Clarke, M.F., Avitabile, S.C., Brown, L., Callister, K.E., Haslem, A., Holland, G.J., Kelly, L.T., Kenny, S.A., Nimmo, D.G., Spence-Bailey, L.M., Taylor, R.S., Watson, S.J. & Bennett, A.F. (2010) Ageing mallee eucalypt vegetation after ﬁre: insights for successional trajectories in semiarid mallee ecosystems. Australian Journal of Botany, 58, 363–372. Department of Sustainability and Environment (2008) Fire Protection Plan: North West Region – Mildura Fire District. Department of Sustainability and Environment, East Melbourne. Driscoll, D.A. & Henderson, M.K. (2008) How many common reptile species are ﬁre specialists? A replicated natural experiment highlights the predictive weakness of a ﬁre succession model. Biological Conservation, 141, 460–471. Driscoll, D.A., Lindenmayer, D.B., Bennett, A.F., Bode, M., Bradstock, R.A., Cary, G.J., Clarke, M.F., Dexter, N., Fensham, R., Friend, G., Gill, M., James, S., Kay, G., Keith, D.A., MacGregor, C., Russell-Smith, J., Salt, D., Watson, J.E.M., Williams, R.J. & York, A. (2010) Fire management for biodiversity conservation: key research questions and our capacity to answer them. Biological Conservation, 143, 1928–1939. Elith, J., Leathwick, J.R. & Hastie, T. (2008) A working guide to boosted regression trees. Journal of Animal Ecology, 77, 802–813. Fernandes, P.M. & Botelho, H.S. (2003) A review of prescribed burning eﬀectiveness in ﬁre hazard reduction. International Journal of Wildland Fire, 12, 117–128. Fox, B.J. (1982) Fire and mammalian secondary succession in an Australian coastal heath. Ecology, 63, 1332–1341. Fox, B.J., Taylor, E. & Thompson, P.T. (2003) Experimental manipulation of habitat structure: a retrogression of the small mammal succession. Journal of Animal Ecology, 72, 927–940. Friend, G.R. (1993) Impact of ﬁre on small vertebrates in mallee woodlands and heathlands of temperate Australia – a review. Biological Conservation, 65, 99–114. Gill, A.M. (1981) Adaptive responses of Australian vascular plant species to ﬁres. Fire and the Australian Biota (eds R.H. Groves & I.R. Noble), pp. 243– 272. Australian Academy of Science, Canberra. Hall, S.A., Burke, I.C. & Hobbs, N.T. (2006) Litter and dead wood dynamics in ponderosa pine forests along a 160-year chronosequence. Ecological Applications, 16, 2344–2355. Hanes, T.L. (1971) Succession after ﬁre in chaparral of southern California. Ecological Monographs, 41, 27–52. Haslem, A., Callister, K.E., Avitabile, S.C., Griﬃoen, P.A., Kelly, L.T., Nimmo, D.G., Spence-Bailey, L.M., Taylor, R.S., Watson, S.J., Brown, L., Bennett, A.F. & Clarke, M.F. (2010) A framework for mapping vegetation over broad spatial extents: a technique to aid land management across jurisdictional boundaries. Landscape and Urban Planning, 97, 296–305. Higgs, P. & Fox, B.J. (1993) Interspeciﬁc competition – a mechanism for rodent succession after ﬁre in wet heathland. Australian Journal of Ecology, 18, 193– 201. Inions, G.B., Tanton, M.T. & Davey, S.M. (1989) Eﬀect of ﬁre on the availability of hollows in trees used by the common brushtail possum, Trichosurus vulpecula Kerr, 1792, and the ringtail possum, Pseudocheirus peregrinus Boddaerts, 1785. Australian Wildlife Research, 16, 449–458. James, S.E. & M’Closkey, R.T.M. (2003) Lizard microhabitat and ﬁre fuel management. Biological Conservation, 114, 293–297. Kelly, L.T., Nimmo, D.G., Spence-Bailey, L.M., Clarke, M.F. & Bennett, A.F. (2010) The short-term responses of small mammals to wildﬁre in semiarid mallee shrubland, Australia. Wildlife Research, 37, 292–300. Land Conservation Council (1987) Report on the Mallee Area Review. Land Conservation Council, Melbourne. Lecomte, N., Simard, M., Fenton, N. & Bergeron, Y. (2006) Fire severity and long-term ecosystem biomass dynamics in coniferous boreal forests of eastern Canada. Ecosystems, 9, 1215–1230.

Letnic, M., Dickman, C.R., Tischler, M.K., Tamayo, B. & Beh, C.L. (2004) The responses of small mammals and lizards to post-ﬁre succession and rainfall in arid Australia. Journal of Arid Environments, 59, 85–114. MacHunter, J., Menkhorst, P. & Loyn, R. (2009) Towards a Process for Integrating Vertebrate Fauna into Fire Management Planning. Arthur Rylah Institute for Environmental Research, Melbourne. Mack, M.C., Treseder, K.K., Manies, K.L., Harden, J.W., Schuur, E.A.G., Vogel, J.G., Randerson, J.T. & Chapin, F.S. (2008) Recovery of aboveground plant biomass and productivity after ﬁre in mesic and dry black spruce forests of interior Alaska. Ecosystems, 11, 209–225. McCarthy, G.J., Tolhurst, K.G. & Chatto, K. (1999) Overall Fuel Hazard Guide. Department of Natural Resources and Environment, East Melbourne. Morrison, D.A., Buckney, R.T., Bewick, B.J. & Cary, G.J. (1996) Conservation conﬂicts over burning bush in south-eastern Australia. Biological Conservation, 76, 167–175. Noble, I.R. & Slatyer, R.O. (1980) The use of vital attributes to predict successional changes in plant communities subject to recurrent disturbances. Vegetatio, 43, 5–21. Noble, J.C., Smith, A.W. & Leslie, H.W. (1980) Fire in the mallee shrublands of western New South Wales. Australian Rangeland Journal, 2, 104–114. Noble, J.C. & Vines, R.G. (1993) Fire studies in mallee (Eucalyptus spp.) communities of western New South Wales: grass fuel dynamics and associated weather patterns. Rangeland Journal, 15, 270–297. Pare, D. & Bergeron, Y. (1995) Above-ground biomass accumulation along a 230-year chronosequence in the southern portion of the Canadian boreal forest. Journal of Ecology, 83, 1001–1007. Pausas, J.G. & Bradstock, R.A. (2007) Fire persistence traits of plants along a productivity and disturbance gradient in Mediterranean shrublands of south-east Australia. Global Ecology and Biogeography, 16, 330–340. Pearce, J. & Ferrier, S. (2000) Evaluating the predictive performance of habitat models developed using logistic regression. Ecological Modelling, 133, 225– 245. R Development Core Team (2008) R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing, Vienna. Raison, R.J., Woods, P.V. & Khanna, P.K. (1983) Dynamics of ﬁne fuels in recurrently burnt eucalypt forests. Australian Forestry, 46, 294–302. Sah, J.P., Ross, M.S., Snyder, J.R., Koptur, S. & Cooley, H.C. (2006) Fuel loads, ﬁre regimes, and post-ﬁre fuel dynamics in Florida Keys pine forests. International Journal of Wildland Fire, 15, 463–478. Sandell, P., Tolhurst, K., Dalton, J., Scott, B. & Smith, M. (2006) Fire management prescriptions for the Victorian mallee parks. Proceedings of the Royal Society of Victoria, 118, 395–412. Stephens, S.L. & Ruth, L.W. (2005) Federal forest-ﬁre policy in the United States. Ecological Applications, 15, 532–542. Torre, I. & Diaz, M. (2004) Small mammal abundance in Mediterranean postﬁre habitats: a role for predators? Acta Oecologica, 25, 137–142. Ucitel, D., Christian, D.P. & Graham, J.M. (2003) Vole use of coarse woody debris and implications for habitat and fuel management. The Journal of Wildlife Management, 67, 65–72. Van Wilgen, B.W., Lemaitre, D.C. & Kruger, F.J. (1985) Fire behaviour in South-African fynbos (macchia) vegetation and predictions from Rothermel Fire Model. Journal of Applied Ecology, 22, 207–216. Vandvik, V., Heegaard, E., Maren, I.E. & Aarrestad, P.A. (2005) Managing heterogeneity: the importance of grazing and environmental variation on post-ﬁre succession in heathlands. Journal of Applied Ecology, 42, 139– 149. Whitford, W.G., Ludwig, J.A. & Noble, J.C. (1992) The importance of subterranean termites in semi-arid ecosystems in south-eastern Australia. Journal of Arid Environments, 22, 87–91. Woinarski, J.C.Z. (1999) Fire and Australian birds: a review. Australia’s Biodiversity: Responses to Fire (eds A.M. Gill, J.C.Z. Woinarski & A. York), pp. 56–101. Department of Environment and Heritage, Canberra. Wood, S.N. (2004) Stable and eﬃcient multiple smoothing parameter estimation for generalized additive models. American Statistical Association, 99, 673–686. Wood, S.N. (2006) Generalized Additive Models: An Introduction with R. Chapman and Hall ⁄ CRC Press, Boca Raton. Zuur, A.F., Ieno, E.N., Walker, N.J., Saveliev, A.A. & Smith, G.M. (2009) Mixed Eﬀects Models and Extensions in Ecology with R. Springer, New York. Received 12 July 2010; accepted 21 October 2010 Handling Editor: Brian Wilsey

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 247–256

256 A. Haslem et al.

Supporting Information Additional Supporting Information may be found in the online version of this article:

As a service to our authors and readers, this journal provides supporting information supplied by the authors. Such materials may be re-organized for online delivery, but are not copy-edited or typeset. Technical support issues arising from supporting information (other than missing ﬁles) should be addressed to the authors.

Figure S1. Location of 26 study landscapes in Victoria, South Australia and New South Wales, Australia. All study sites (n = 549) were situated within these landscapes.

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 247–256

Journal of Applied Ecology 2011, 48, 102–111

doi: 10.1111/j.1365-2664.2010.01907.x

The impact of alternative harvesting strategies in a resource–consumer metapopulation Chloe¨ M. J. Strevens1 and Michael B. Bonsall*1,2 1

Mathematical Ecology Research Group, Department of Zoology, University of Oxford, South Parks Road, Oxford, OX1 3PS, UK; and 2St. Peters College, New Inn Hall Street, Oxford, OX1 2DL, UK

Summary 1. The majority of our knowledge of harvested populations has been drawn from studies on singlespecies occupying continuous landscapes. Population dynamics in spatially structured landscapes are markedly diﬀerent from those in continuous habitats. We investigate the eﬀects of several harvesting strategies on the metapopulation dynamics of the bruchid beetle Callosobruchus maculatus and its parasitoid Anisopteromalus calandrae, when locally extinction prone populations of bruchids were harvested. By harvesting the resource in this closed resource–consumer interaction, we could examine the scenario where humans and natural predators share a common resource. 2. Using population-level models in single and coupled-patch systems, simulations were run to estimate sustainable harvesting levels for each harvesting strategy. These harvesting levels were then implemented in experimental metapopulation microcosms. Experiments were run for multiple generations and long-term time series of population size and harvest yield were collected. 3. Controversially, harvesting resulted in larger regional population sizes in harvested metapopulations than in unharvested metapopulations. Similarly, conservative harvesting strategies such as ﬁxed escapement harvesting and harvesting with refuges resulted in smaller population sizes than ﬁxed quota or ﬁxed proportion harvesting strategies. Fixed proportion harvesting gave rise to the largest population sizes, fewest local extinctions and largest yields. Assuming both species are of ecological importance, ﬁxed proportion is therefore the optimal harvesting strategy for this model system. 4. Synthesis and applications. We demonstrate that, under certain conditions, increasing local mortality can increase population sizes. This ‘hydra eﬀect’ may be caused by the advantage of higher rates of local population turnover for dispersive species in patchy landscapes, or due to the interruption of overcompensatory density-dependence in the host populations. These results are of particular relevance both in the development of sustainable harvesting policies in multi-species communities or, conversely, the control of pest species inhabiting spatially structured landscapes. We have shown that predictive models can be a useful tool in the estimation of sustainable harvesting limits where no a priori knowledge of the system exists. We also demonstrate that spatial structure, the eﬀects of interspeciﬁc interactions and knowledge of density-dependent population dynamics must be included in the generation of sustainable harvesting theory. Key-words: Anisopteromalus calandrae, Callosobruchus maculatus, escapement, host–parasitoid, proportion, quota, refuges

Introduction Most of our knowledge about harvesting animals assumes that populations are distributed continuously across space (McCullough 1996). However, due to the prevalence of global habitat

*Correspondence author. E-mail: [email protected]

loss and fragmentation this theory is no longer adequate. Rather, a synthesis of metapopulation ecology and current harvesting theory is likely to prove much more relevant to the generation of sustainable harvesting strategies. Metapopulation ecology has developed over the last four decades from the original homogeneous model (Levins 1969) to systems containing various degrees of spatial and temporal heterogeneity (Fahrig 2007; Strevens & Bonsall in press). In all

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society

Harvesting a resource–consumer metapopulation 103 cases the persistence of a metapopulation relies upon the balance between local extinction and recolonization rates (Ovaskainen & Hanski 2004). Harvesting local populations negatively aﬀects both of these critical rates (McCullough 1996). By reducing local population sizes, harvesting increases the likelihood of local extinction. Harvesting the prey in a predator–prey interaction generally results in a decline in both predator and prey population sizes, while harvesting the predator decreases predator populations size but increases prey population sizes (Beddington & Cooke 1982; Hoekstra & van den Bergh 2005). By reducing local population sizes, harvesting also reduces dispersal, particularly when dispersal is densitydependent (Hansson 1991; Matthysen 2005). Exceptions to the negative inﬂuence of harvesting may occur where high population density causes major habitat destruction (McCullough 1996) or when harvesting mortality is compensated for by a relaxation in some other form of density-dependent mortality (Murton, Westwood & Isaacson 1974; Hilborn, Walters & Ludwig 1995). The term ‘hydra eﬀect’ has been coined to describe the paradox of increased population sizes as a result of increased mortality and is predominantly due to the eﬀect of harvesting on overcompensatory density-dependence in populations (Abrams & Matsuda 2005; Abrams 2009). However, assuming that harvesting reduces local population sizes and dispersal it would suggest that metapopulations could only support limited and highly controlled levels of harvesting. Furthermore, in heterogeneous metapopulations only certain patches may be able to support harvesting (Tuck & Possingham 1994; Supriatna & Possingham 1998; Robinson et al. 2008). This is highly likely in metapopulations which are composed of source and sink patches, due to diﬀerences in either local growth rates (Pulliam 1988) or net dispersal directions (Cronin 2007). Despite the likely ecological problems surrounding the harvesting of spatially explicit environments, there is often a social and ⁄ or economic imperative to do so. As a result, studies need to consider the most sustainable harvesting strategies to employ. However, replicating ﬁeld experiments across time and space is often diﬃcult for practical reasons because species are often long-living, cryptic and widely distributed. Furthermore, the cost of failure of manipulative experiments in natural populations is high (Hilborn, Walters & Ludwig 1995). This has meant that information on the eﬀects of diﬀerent harvesting strategies must be drawn from other studies on similar species ⁄ systems, from historical harvesting records, from theoretical studies and from microcosm experiments. The latter provides an excellent opportunity to replicate experimental systems in controlled environments and allows for the comparison of several diﬀerent treatments (Bonsall & Hassell 2005; Fryxell, Smith & Lynn 2005). Here, we used the host–parasitoid interaction between Callosobruchus maculatus and Anisopteromalus calandrae. This is an excellent model system because the interaction is tightly coupled and well understood. Only females search for hosts and oviposition generally results in the development of one parastoid (Godfray & Shimada 1999). This means that the interaction can be well described by simple population models (Kareiva 1989; Hassell 2000). It is ideal for

use in metapopulation studies such as ours as high parasitoid eﬃciency results in overexploitation and extinction at the local scale which is mitigated by dispersal (by adults) between patches (Bonsall, French & Hassell 2002; Bonsall et al. 2005; Bull et al. 2006). In this study we examined the eﬀect of harvesting the resource species (C. maculatus). Humans often compete with natural predators for a shared resource, as a result understanding the eﬀect of harvesting on the persistence and population sizes of both the resource and its consumer is essential in order to maintain a functioning ecosystem. Examples of this include the negative eﬀect that harvesting Southern Ocean krill stocks has had on populations of seals, birds and ﬁsh (May et al. 1979), and the eﬀect of harvesting shellﬁsh on shore bird populations (Stillman et al. 2001; Goss-Custard et al. 2004; Hoekstra & van den Bergh 2005). In general, the objective of most sustainable harvesting strategies is to maximize yield while maintaining a persistent population of the exploited resource. In another context, harvesting theory may also be used in the ﬁeld of pest control where the objective is to reduce population sizes and persistence (Murton, Westwood & Isaacson 1974; Brooks & Lebreton 2001; Matsuoka & Seno 2008). There are several commonly used harvesting strategies. Classic maximum sustainable yield is, theoretically, the largest yield which can be taken from a population while maximizing population growth continuously across time. However, it fails to account for either demographic or environmental stochasticity and as a result it is ineﬀective in dealing with natural populations, generally resulting in overexploitation of resources (Quinn & Deriso 1999). Alternative harvesting policies include ﬁxed quota, ﬁxed proportion and ﬁxed escapement strategies. Of these, ﬁxed quota has been shown to be the least sustainable, resulting in large ﬂuctuations in population density and an increase in extinction risk (Beddington & May 1977; Pascual & Hilborn 1995; Fryxell, Smith & Lynn 2005). Fixed proportion and ﬁxed escapement strategies both reduce the risk of population collapse, the former by providing a compensatory mechanism to accommodate population ﬂuctuations and the latter by ﬁxing the minimum population size. In particular, ﬁxed escapement harvesting minimizes the probability of population extinction in highly ﬂuctuating populations (Lande, Engen & Saether 1995; Lande, Saether & Engen 1997; Aanes et al. 2002). Unfortunately both of these approaches require exact knowledge about population size (which may not be possible to measure) and result in a more variable yield leading to negative economic consequences (Fryxell, Smith & Lynn 2005). Finally, an alternative way to promote population persistence is through the creation of harvesting refuges within the landscape (McCullough 1996; Joshi et al. 2009). However, the closure of patches to harvesting can result in high opportunity costs which must be thoroughly considered (Sanchirico & Wilen 2001). In this study we use mathematical models to estimate sustainable harvesting levels for four common harvesting strategies; ﬁxed quota, ﬁxed proportion, ﬁxed escapement and the inclusion of refuges. Assuming that both species are ecologically valuable, our objective was to ﬁnd a strategy that

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 102–111

104 C. M. J. Strevens & M. B. Bonsall promoted population persistence while maintaining the largest harvest yield possible. Harvesting strategies were implemented in replicated Callosobruchus maculatus–Anisopteromalus calandrae metapopulation microcosms. Our results indicated that the optimum harvesting strategy for this resource–consumer system was ﬁxed proportion harvesting which maximized both population size and yield. Contrary to expectation, unharvested systems had the smallest population sizes. We suggest that harvesting in these spatially explicit systems increases population sizes by disrupting the ecological dynamics of the resource–consumer interaction.

Table 1. Functions for harvest levels for each strategy Harvest strategy

hN

No harvest

hN = 0

Fixed quota

hN = q where q is constant

Fixed proportion

hN = p%(Nt) where p is some constant percentage of Nt

Fixed proportion with refuges

hN = p%(Nt) where p is some constant percentage of Nt in peripheral patches hN = 0 in core patches

Fixed escapement

hN = p%(Nt) for Nt > e where e is some constant threshold population size hN = 0 for Nt £ e where e is some constant threshold population size

Materials and methods THEORETICAL PREDICTIONS

Our approach to understanding the ecological implications of harvesting in metapopulations is twofold. First, we use an existing data set to establish appropriate harvesting levels for three focal harvesting strategies; ﬁxed quota, ﬁxed proportion and ﬁxed escapement. A fourth strategy of harvesting with refuges was studied experimentally (see below) however we excluded it from the preliminary theoretical analyses as we constrained the maximum number of patches modelled to two (which is too few to sensibly include a refuge eﬀect). All harvesting was conducted at the local scale and harvesting eﬀort was equal in all patches. Model scenarios were either composed of single patches or two coupled patches. We then designed a set of metapopulation microcosm experiments to test the sustainability and productivity of these harvesting policies. In order to deﬁne the most suitable harvesting levels within each of the strategies a set of spatially explicit host–parasitoid time series from an existing study were used (Bonsall et al. 2005; Bull et al. 2006, 2007; Bull & Bonsall 2008). These data were considered appropriate proxies for experiments in this study as the metapopulation structure, dispersal and environmental conditions were identical. Data consisted of four replicate time series for the interaction between the bruchid beetle Callosobruchus maculatus and its parasitoid Anisopteromalus calandrae within a 16 patch metapopulation. The impact of diﬀerent harvesting levels on population dynamics were explored using a model which best ﬁt the pre-existing time series data. The deterministic model framework used was based on a basic continuous-time birth–death model (eqn 1): dN ¼ NðrfðNÞ dlðNÞÞ: dt

eqn 1

Here, r and d are the intrinsic rates of increase and decrease in a population of size N, respectively. The functions f(N) and l(N) describe density-dependent births and deaths, respectively and were based on a set of density-dependent functions (Bellows 1981) (Table S1, Supporting information). By ﬁtting the models (through a numerical optimization algorithm implemented in C – further details are given in Appendix S1, Supporting information) to the patch-level time series for each replicate separately it was possible to identify the best model describing the local population dynamics. The two models which best described these data were then ﬁtted to all of the replicates simultaneously so that a set of parameter values for r and d could be estimated which best described the local population dynamics across the replicate metapopulations. The best ﬁt of these two models was selected based on their AIC (Akaike Information Criterion) scores (Table S1, Supporting information). In this case, the deterministic model that

Simulations of population dynamics following a birth-death model (eqn 3) were run with varying values of q (ﬁxed quota), p (ﬁxed proportion) and e (ﬁxed escapement). The values of q, p and e chosen for the experimental microcosms were two individuals, 25% and three individuals, respectively. best described the local population dynamics was a single-parameter linear density-dependence function (MacFadyen 1963), acting on death rate with a density independent growth rate (eqn 2): dN ¼ r N d Nða NÞ dt

eqn 2

where a is the strength of density-dependence. Simulations were then run using the selected birth-death model (eqn 2) combined with a harvesting function, h(N) (eqn 3, Table 1): dN ¼ r N d Nða NÞ hðNÞ: dt

eqn 3

Both deterministic and stochastic versions of this model acting in single and coupled patches were explored. Rather than trying to predict the precise population dynamics that would occur in the experimental (16 patch) systems, our simple single-species, single and coupled-patch models allowed us to describe the baseline response of populations inhabiting continuous and structured landscapes to diﬀerent forms of harvesting. Being larger, our experimental systems were likely to be more persistent than the small model systems studied (Bonsall, French & Hassell 2002). Stochasticity was introduced into these models by deﬁning the likelihood of population increase and decrease (and dispersal in the two patch systems) at any time step as a probabilistic process (see Appendix S2, Supporting information). In the stochastic coupled-patch models we examined the eﬀect of harvesting when dispersal was symmetrical and asymmetrical between patches. As a result of density-dependent dispersal the net direction of dispersal at any point in time is often asymmetrical (Bowler & Benton 2005; Matthysen 2005). Although resource distribution in our systems was homogeneous, patches supported diﬀering host population sizes as a consequence of the inﬂuence of parasitoids. While we did not incorporate density-dependent dispersal or heterogeneous patch quality explicitly in our models, we studied the eﬀect of asymmetrical dispersal by altering the relative dispersal rates in and out of patches. In symmetrical systems 50% of the population dispersed out of the patch each time step and in asymmetrical systems

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 102–111

Harvesting a resource–consumer metapopulation 105 (a)

(b)

(c)

Fig. 1. Time series from the deterministic single-patch models with; (a) ﬁxed quota (based on 25% (1Æ319), 33% (1Æ758) or 50% (2Æ637) of the equilibrium population size), (b) ﬁxed proportion (50% or 75% of the population size) and (c) ﬁxed escapement (50% or 75% over a threshold population size of three individuals). 75% of one patch dispersed while only 25% of the second patch dispersed. These relatively high dispersal rates were necessary in order to explore an obvious eﬀect of dispersal on population dynamics in these simple population models. Deterministic coupled-patch models used asymmetrical dispersal only. Deterministic models were run until they reached a stable equilibrium. Stochastic models were run for 1000 time units and replicated 100 times after which the probability of population persistence was calculated. A range of values of h(N) for each treatment were simulated and the highest harvesting levels which also maximized persistence were used in the microcosm experiments.

able. Fixed quota harvesting at any level in the stochastic model was unsustainable such that even the minimum quota of one individual caused 95% of populations to go extinct. In single patch systems Table 2. Extinction probabilities from stochastic models in single and coupled-patch populations (determined from 100 simulations) at the regional and local (bracketed) scale Population structure

Harvest strategy

Extinction probability

No harvest Fixed quota (1) Fixed proportion (10%) Fixed proportion (25%) Fixed escapement (2, 10%) Fixed escapement (2, 25%) Fixed escapement (3, 10%) Fixed escapement (3, 25%)

0Æ24 0Æ95 0Æ45 0Æ78 0Æ38 0Æ63 0Æ36 0Æ50

Coupled patches, No harvest symmetrical Fixed quota (1) dispersal Fixed quota (2) Fixed proportion (10%) Fixed proportion (25%) Fixed escapement (2, 10%) Fixed escapement (2, 25%) Fixed escapement (3, 10%) Fixed escapement (3, 25%)

0Æ04 0Æ18 0Æ20 0Æ12 0Æ30 0Æ09 0Æ16 0Æ08 0Æ10

(0Æ18, (0Æ30, (0Æ33, (0Æ26, (0Æ44, (0Æ22, (0Æ29, (0Æ23, (0Æ24,

0Æ14) 0Æ41) 0Æ56) 0Æ29) 0Æ45) 0Æ25) 0Æ30) 0Æ25) 0Æ25)

Coupled patches, No harvest asymmetrical Fixed quota (1) dispersal Fixed quota (2) Fixed proportion (10%) Fixed proportion (25%) Fixed escapement (2, 10%) Fixed escapement (2, 25%) Fixed escapement (3, 10%) Fixed p escapement (3, 25%)

0Æ04 0Æ12 0Æ16 0Æ11 0Æ26 0Æ10 0Æ16 0Æ08 0Æ09

(0Æ05, (0Æ16, (0Æ23, (0Æ12, (0Æ30, (0Æ11, (0Æ17, (0Æ10, (0Æ09,

0Æ41) 0Æ27) 0Æ31) 0Æ63) 0Æ73) 0Æ57) 0Æ64) 0Æ55) 0Æ57)

Single patch MODEL RESULTS

Single-patch models In the absence of harvesting, deterministic populations settled at an equilibrium population size of 5Æ274 individuals. Harvesting was initiated in populations at equilibrium and the eﬀect with respect to persistence ⁄ extinction was determined and plotted as time series (Fig. 1). In the stochastic single-patch models without harvesting there was a 0Æ24 chance of population extinction. Representative time series for this model are shown in Fig. 2. Given the large number of replicates of these models, results were gathered in terms of the probability of extinction due to the speciﬁc harvesting levels (Table 2). Fixed quotas of 50%, 33% and 25% of the equilibrium value (2Æ637, 1Æ758 and 1Æ319, respectively) were explored (Fig. 1a). This showed that a ﬁxed quota of only 1Æ319 (or 1 individual) was sustain-

Fig. 2. 100 replicate time series from a stochastic single-patch model without harvesting. Stochastic models were replicated 100 times and the proportion of times the system went extinct was taken to represent the persistence of the system.

Coupled patches were connected by either symmetrical dispersal where 50% of each patch dispersed each week or asymmetrical dispersal where 25% of one patch and 75% of the other patch dispersed each week. Local populations were harvested at a ﬁxed quota of either one or two individuals per week or a ﬁxed proportion of 10% or 25% of the population size per week. For systems undergoing ﬁxed escapement harvesting, a threshold population size of either two or three individuals was imposed below which there was no dispersal and above which populations were harvested at a ﬁxed proportion of 10% or 25%.

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 102–111

106 C. M. J. Strevens & M. B. Bonsall stochasticity frequently resulted in extremely small population sizes so that harvesting any ﬁxed quota forced the populations to collapse. Fixed proportion harvesting in deterministic models was sustainable up to 75% of the population size (Fig. 1b) but was much less sustainable in stochastic population models. In the latter case, the percentage of populations that went extinct reduced as the proportion of the population harvested decreased from 25% to 10%. Fixed escapement strategies in deterministic populations were also highly sustainable with the populations reaching a steady state at the value of the threshold escapement size below which no harvesting took place (Fig. 1c). In the stochastic populations, the use of a threshold population size to limit harvesting reduced the extinction risk compared to the ﬁxed proportion strategy. When harvested at 10%, the extinction risk was similar regardless of the escapement size (36–38% extinction risk). In contrast, when 25% of the population was harvested a ﬁxed escapement of two individuals resulted in 63% of populations to go extinct as opposed to 50% when the escapement was three. Threshold sizes above three did not improve population persistence relative to ﬁxed proportion harvesting. While this threshold is relatively low, bruchids are extremely persistent even at low population sizes so this was considered a reasonable value.

regardless of the dispersal pattern between patches. This suggests that stochasticity acting in coupled-patch systems increases the size of the sustainable quota compared to that in deterministic systems. Fixed proportion harvesting 25% of the population size in coupled deterministic populations was sustainable for both patches. Above that only the patch which had a net increase in population size due to dispersal persisted. As in the single patch models, extinction risk in stochastic models was positively related to the proportion removed. The extinction risk was almost twice as high when harvesting 25% of the population than when 10% was harvested and this held true regardless of dispersal regime. Fixed escapement harvesting in coupled deterministic populations was highly sustainable, with populations reaching a steady state at or above the threshold population size depending on the direction of net dispersal. In the stochastic simulations this strategy resulted in a lower extinction rate than in ﬁxed proportion harvesting. When the escapement was two individuals, harvesting 25% of the population increased the percentage of extinct populations from 10% (at 10% harvesting) to 16%. However, when the escapement was increased to three individuals, harvesting at 25% and 10% resulted in similar extinction risks.

Coupled-patch models

HARVESTING IN EXPERIMENTAL MICROCOSMS

In deterministic coupled-patch models one patch had a lower dispersal rate than the other (75% and 25%) and therefore maintained a higher equilibrium population size (Fig. 3). This disparity between local population sizes was also seen in stochastic populations with asymmetrical dispersal. Symmetrical dispersal in stochastic populations resulted in slightly larger population sizes than in systems with asymmetrical dispersal although the overall regional persistence was the same in both coupled-patch model systems. At a local level, patches which lost the most individuals through dispersal had higher probabilities of extinction. In general however, coupling patches increased population sizes and reduced extinction risks at both the regional and local scales. Symmetrical dispersal was not modelled in coupled deterministic populations as dispersal in and out of each identical patch would be balanced out and dynamics would be the same as in single patch models. Fixed quota harvesting in deterministic models had a similar eﬀect to that in single-patch populations where a maximum harvest of a single individual was sustainable. In the stochastic models quotas of one and two individuals were similarly sustainable

(a)

(b)

Study species Laboratory metapopulations of the interaction between the bruchid beetle Callosobruchus maculatus and it parasitoid Anisopteromalus calandrae were used in this study. This otherwise non-persistent interaction has been shown to persist in spatially structured landscapes, giving rise to locally extinction prone and regionally persistent populations (Bonsall, French & Hassell 2002; Bonsall & Hastings 2004). It is an ideal model system due to the self-contained, closely coupled and easily managed life-cycles of both species (Bonsall & Hassell 2005). The bruchid beetle Callosobruchus maculatus is a pest species of substantial economic importance as it parasitises leguminous bean plants and stores of dried beans (Ngamo et al. 2007). It exists as a mobile adult for approximately 1 week during which time it copulates and oviposits on the surface of the host bean. The rest of the development is within the bean where the bruchid larvae develop over the course of approximately three weeks before hatching. For a more detailed description of the bruchid life-cycle see Bellows (1982).

(c)

Fig. 3. Harvesting in deterministic, coupledpatch populations connected by asymmetrical dispersal. Patches either lost 25% (black lines) or 75% (blue lines) of the population through dispersal. Each patch was harvested under; (a) ﬁxed quota (1 individual or 25% (1Æ319), 33% (1Æ758) or 50% (2Æ637) of the equilibrium population size), (b) ﬁxed proportion (25%, 33% or 50% of the population size) or (c) ﬁxed escapement (50% or 75% over a threshold population size of three individuals). 2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 102–111

Harvesting a resource–consumer metapopulation 107 Anisopteromalus calandrae is an ectoparasitoid on bruchid larvae. It has a mobile adult phase lasting approximately 1 week when it can disperse, copulate and oviposit onto third and fourth instar bruchid larvae (Heong 1981). The larva takes about 10 days to develop to adulthood, sustained by a single host larva. This species tends to overexploit its host driving both species to (local) extinction and as a result has been proposed as a natural means of pest control (Ngamo et al. 2007).

Experimental design Metapopulation microcosms were constructed following a well-studied design (Bonsall, French & Hassell 2002; Bonsall & Hastings 2004; Bull et al. 2006). Clear plastic boxes (73 · 73 · 30 cm) were arranged in a 4 · 4 lattice. Each of these boxes represented a patch supporting local host–parasitoid populations. Dispersal between these patches was facilitated through 5 cm long tubes (4Æ5 mm diameter) linking the connecting faces of the boxes (Fig. 4). In order to ensure asynchrony in local population dynamics dispersal was limited to two hours each week (Bonsall, French & Hassell 2002; Hunt & Bonsall 2009). While explicit measures of dispersal were not recorded, observations of dispersal events in these species have shown that Callosobruchus maculatus have high density-dependent dispersal rates and movement in this limited two hour period is generally restricted to neighbouring patches only. In contrast, parasitoids exhibited much lower dispersal rates. Initially, all patches were seeded with three beans and two pairs of bruchids each week for three weeks. This homogeneous resource distribution ensured that patches were all able to support breeding populations of Callosobruchus maculatus. By the fourth week beetles from the oldest beans in the patches were emerging and thenceforth only new beans were added each week. The oldest beans were removed from the patches every week and stored for an additional four weeks to allow for late emergents which were reintroduced to their natal patches. At the time of resource renewal, censuses of both live and

dead adult beetles were conducted and all dead individuals were removed. After 8 weeks, parasitoids were introduced to the systems. Two pairs of wasps were introduced to four randomly selected patches in each of the metapopulations for two consecutive weeks. Adult wasps were counted, and the dead individuals removed, each week at the same time as the beetle census. These data provide local scale time series for beetle and wasp abundance. Due to overlapping generations resulting from variable longevity and larval development periods these time series describe continuous generations of both species. Harvesting strategies (Table 1) were commenced in week 26 after the populations had settled to a host-parasitoid steady state. Harvesting was conducted on adult bruchids at a local level. Individuals within a patch were harvested at random using a grid system. Grids were approximately the size of one adult beetle and cells for harvesting were chosen at random. All experiments were run for a total of 48 weeks, replicated four times and carried out under constant environmental conditions of 30 C, 70% humidity and 16 : 8 h of light : dark.

STATISTICAL ANALYSIS

The time series abundance data was divided into three sections; host only, host : parasitoid and host : parasitoid : harvest. The eﬀect of harvesting regime on population size and the variation in population size were analysed using linear mixed eﬀects models in which replicate was designated as a random variable. This enabled us to determine the level of variation arising between replicates. Even though replicates were identical, this is an important consideration in statistically modelling the eﬀects of harvesting as each time series is a single realization of an underlying stochastic process (Chatﬁeld 2004). The number of local host extinction events was compared between harvesting treatments using a generalized linear model with Poisson errors (Crawley 2007). Finally, the yield from diﬀerent harvesting regimes was compared using analysis of variance. All analyses were conducted using R version 2.5.1 (R Foundation for Statistical Computing 2009).

Results

Fig. 4. Schematic of the patch lattice constructed in these experiments. Harvesting was carried out at the patch scale. All patches were harvested in all treatments except one in which core patches (grey) were allocated as refuges.

Time series for the abundances of C. maculatus and A. calandrae are shown in Fig. 5. These graphs illustrate the three phases of the experiment; host only, host : parasitoid and host : parasitoid : harvest. Across all treatments and replicates the weekly bruchid population size decreased as the experiment progressed through each of the successive stages (F2,950 = 353Æ186, P < 0Æ001). Conversely wasp population size increased from one phase to the next (F1,763 = 335Æ770, P < 0Æ001). There was a signiﬁcant diﬀerence between the size of the weekly regional bruchid populations depending on the harvesting treatment (F4,452 = 6Æ574, P < 0Æ001) with the largest population sizes occurring in populations undergoing ﬁxed proportion harvest and the smallest population sizes, unexpectedly, in the control (no harvest) populations. All other population sizes were similar. At the regional scale, parasitoid populations showed no diﬀerences in abundance between harvesting treatments (F4,453 = 1Æ488, P = 0Æ205). The amount of variation in the regional time series did not diﬀer between treatments for either species (F4,30 = 0Æ988, P < 0Æ429).

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 102–111

108 C. M. J. Strevens & M. B. Bonsall

Fig. 5. Time series for each of the four replicates (rows) of the unharvested control and the four harvesting strategies tested; ﬁxed quota, ﬁxed proportion, ﬁxed escapement, ﬁxed proportion with refuges. Depicted on each graph is the point at which parasitoids were introduced (week 9) and when harvesting commenced (week 26).

(a)

(b)

Fig. 6. Box and whisker plots illustrating; (a) the median and spread of the regional yield and (b) the variation in regional yield between treatments; ﬁxed quota (FQ), ﬁxed proportion (FP), ﬁxed escapement (FE), harvesting with refuges (R).

There was a signiﬁcant diﬀerence between the number of local population extinctions depending on the harvesting regime (Z4,315 = )5Æ924, P < 0Æ05). In particular, the highest number of local extinctions was in metapopulations without harvesting (73 ± 10). The lowest number was observed in replicates under ﬁxed proportion (41 ± 11) and ﬁxed escapement (48 ± 8) harvesting. The weekly yield size varied signiﬁcantly between treatments (F3,364 = 20Æ377, P < 0Æ001), see Fig. 6a. The largest yield was provided by ﬁxed proportion harvesting and the lowest yield by refuge harvesting. The variation in the yield also differed depending on the treatment (F3,12 = 5Æ414, P < 0Æ05). The greatest amount of variation occurred in harvest refuge

systems whilst the least occurred in ﬁxed quota metapopulations (Fig. 6b). Based on these results the optimal harvesting strategy for this resource–consumer system is ﬁxed proportion harvesting as it maximizes both yield and population size. On the other hand, harvesting with a core refuge was the poorest performing strategy giving rise to the smallest population sizes (with the exception of the control) and lowest yields.

Discussion In this study we have used theoretical predictions to estimate sustainable harvesting levels for several harvesting strategies

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 102–111

Harvesting a resource–consumer metapopulation 109 and applied these to a well-studied resource–consumer metapopulation system (Callosobruchus maculatus–Anisopteromalus calandrae). This allowed us to consider the eﬀects of harvesting in a spatially explicit system and to identify the harvesting strategy which resulted in the largest population sizes and greatest yields simultaneously. Theoretical predictions of the impacts of harvesting were drawn from mathematical simulations of population dynamics in both single and coupled populations. Providing there is adequate and precise information about population size and dynamics, using population models in this way can avoid time consuming and expensive monitoring surveys (Aanes et al. 2002; Hauser, Pople & Possingham 2006). Furthermore, the use of laboratory systems such as our metapopulation microcosms allows the collection of long-term, statistically replicated data on population dynamics which can be used to understand the processes driving other ecological systems (Lawton 1995; Bonsall & Hassell 2005). To date, the majority of studies on harvesting still focus on single, continuous populations. One example of particular relevance here is the comprehensive research comparing harvesting strategies in microcosms of the ciliate Tetrahymena thermophila by Fryxell, Smith & Lynn (2005). In their study they use population models to estimate the eﬀect of harvesting under a number of alternative strategies and compare the model results with those from experimental trials. They found corroboration between theoretical predictions and empirical results in the identiﬁcation of the optimal harvesting strategy for their system. There are also a number of purely theoretical studies comparing harvesting strategies in both continuous and spatially structured populations (Tuck & Possingham 1994; Lande, Saether & Engen 1997; Supriatna & Possingham 1998; Aanes et al. 2002). However, there is a dearth of information on the eﬀect of harvesting in experimental metapopulations and even fewer studies on harvesting in resource–consumer metapopulations. The most common harvesting strategies described in the literature are: ﬁxed quota, ﬁxed proportion, ﬁxed escapement and the use of harvesting refuges within a landscape and these, along with an unharvested control, were the treatments which were used in this study. Another strategy to note, but which was not studied here, is the diﬀerential harvest of source and sink patches in heterogeneous metapopulations (Tuck & Possingham 1994; Robinson et al. 2008; Wilberg et al. 2008). Furthermore, work by Supriatna & Possingham (1998) suggest diﬀerent harvesting approaches should be used in predator– prey metapopulations depending on the spatial vulnerability of the prey, eﬃciency of the predator and the existence of prey refuges. Speciﬁcally, they ﬁnd that more conservative harvesting of prey should be implemented in patches where the prey is less vulnerable to predation or where predators are not present. If predator eﬃciency varies spatially then they recommend more conservative harvesting in patches where predator eﬃciency is higher. In our Callosobruchus maculatus–Anisopteromalus calandrae metapopulations prey vulnerability and predator eﬃciency are considered to be homogeneous across the landscape and both species had equal access to all parts of the patch lattice.

Our results demonstrate that host population sizes decreased and parasitoid populations grew after harvesting was commenced. This is likely to be a consequence of the length of time the parasitoid population took to establish itself across the entire patch lattice. Anisopteromalus calandrae was introduced into a limited number of patches within the metapopulation and, due to a low dispersal rate, took several weeks to spread throughout the entire metapopulation. With respect to Callosobruchus maculatus there was a signiﬁcant diﬀerence between population sizes arising from diﬀerent treatments. Surprisingly, smaller population sizes and higher numbers of local extinctions were seen in unharvested metapopulations compared to harvested metapopulations. Increased population sizes as a result of increased mortality has been termed the ‘hydra eﬀect’ (Abrams & Matsuda 2005; Abrams 2009). It is similar to the concepts of ‘pest resurgence’ (Matsuoka & Seno 2008) and ‘press perturbation’ (Abrams 1987) where increased mortality relaxes the eﬀect of strong overcompensatory density-dependence, changes the nature of population ﬂuctuations and reduces parasitoid ⁄ predator eﬃciency (Abrams 2009). As far as we are aware, our ﬁndings are one of the ﬁrst empirical examples of this phenomenon (Cameron & Benton 2004; Zipkin et al. 2008), and this will be investigated more elsewhere. Of the harvesting strategies tested, the largest population sizes were in the ﬁxed proportion treatments. Harvesting a ﬁxed proportion accommodates the natural ﬂuctuations in population size by scaling the harvesting amounts in relation to population size and has been demonstrated to be less prone to population collapse than ﬁxed quota harvesting (Pascual & Hilborn 1995; Milner-Gulland et al. 2001; Fryxell, Smith & Lynn 2005). Fixed proportion harvesting in our experiments also resulted in the highest yields whilst the more conservative ﬁxed escapement policy resulted in moderate population sizes and was unable to capitalize on the abundance of beetles to the same extent as ﬁxed proportion harvesting resulting in lower yields. Fixed escapement harvesting is most important in systems with abundant environmental stochasticity, where parameter estimates are uncertain or the population is on the threshold of collapse (Lande 1987). The smallest yield was gathered from harvesting in refuge systems, where population sizes were also small. In our experimental 4 · 4 systems, core patches were designated as unharvested refuges. While our systems were homogeneous in terms of the distribution of resources for the host species, patches diﬀered in their connectivity index. In general higher levels of connectivity increase population persistence in patchy landscapes (Hanski, Kuussaari & Nieminen 1994; Holyoak 2000; Goodwin & Fahrig 2002). Here, the core refuge patches had the highest connectivity indices in order to maximize the beneﬁt of the refuges. Whether patches are situated at the edge or centre of a patch system can be very important, particularly where species have low edge tolerance (Anderson & Danielson 1997; With & King 2001) or if the habitat outside the bounds of the metapopulation is hostile (Cronin & Reeve 2005). Given the reﬂective boundary and the homogeneous environmental conditions across the patch lattice these factors are not relevant in this

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 102–111

110 C. M. J. Strevens & M. B. Bonsall study. We assume that the small population sizes in this treatment were due to the strong host–parasitoid interaction occurring in the unharvested refuges. This strongly coupled interaction suppressed host population size and thereby reduced the yield. Contrary to this result, reserves ⁄ refuges are frequently used to increase population persistence by providing source populations to feed into other patches in the metapopulation (McCullough 1996; Fryxell, Lynn & Chris 2006; Joshi et al. 2009). Another important facet of yield is its reliability and consistency across time as this will have considerable economic impacts (Hilborn, Walters & Ludwig 1995; Bulte & van Kooten 1999). From this perspective the least variable, and therefore the most beneﬁcial harvesting approach, was ﬁxed quota harvesting, while the largest amount of variation occurred in refuge treatments. All of the harvesting strategies discussed here carry some practical disadvantages; for instance ﬁxed quota, proportion and escapement harvesting all require in depth knowledge of the population sizes in order to estimate sustainable harvesting levels and this may be diﬃcult to gather (Fryxell, Smith & Lynn 2005). Environmental stochasticity further inﬂates this diﬃculty by changing the nature of population dynamics across time and demands longer term data in order to avoid overexploitation of harvested resources (Lande, Saether & Engen 1997; Aanes et al. 2002). Reserves or refuges where harvesting is not permitted can be costly both to create and in terms of lost opportunity costs (Sanchirico & Wilen 2001). Finally the control of harvesting is diﬃcult for both practical and socio-economic reasons (Hilborn, Walters & Ludwig 1995). We express caution at the direct extrapolation of the results from this resource–consumer study to natural and inherently more complex systems. Instead we propose that our ﬁndings may enhance the current understanding of sustainable harvesting in spatially structured resource–consumer systems.

Acknowledgements We thank the Natural Environment Research Council, The Royal Society, Magdalen College and the Department of Zoology for ﬁnancial support. We are also grateful to Lucas Cunningham for his assistance with data collection and Nina Alphey, Inigo Montes, Brian and Jennifer Strevens and two anonymous reviewers for their comments and advice.

References Aanes, S., Engen, S., Saether, B.E., Willebrand, T. & Marcstrom, V. (2002) Sustainable harvesting strategies of Willow Ptarmigan in a ﬂuctuating environment. Ecological Applications, 12, 281–290. Abrams, P.A. (1987) On classifying interactions between populations. Oecologia, 73, 272–281. Abrams, P.A. (2009) When does greater mortality increase population size? The long history and diverse mechanisms underlying the hydra eﬀect. Ecology Letters, 12, 462–474. Abrams, P.A. & Matsuda, H. (2005) The eﬀect of adaptive change in the prey on the dynamics of an exploited predator population. Canadian Journal of Fisheries and Aquatic Sciences, 62, 758–766. Anderson, G.S. & Danielson, B.J. (1997) The eﬀects of landscape composition and physiognomy on metapopulation size: the role of corridors. Landscape Ecology, 12, 261–271. Beddington, J.R. & Cooke, J.G. (1982) Harvesting from a prey–predator complex. Ecological Modelling, 14, 155–177.

Beddington, J.R. & May, R.M. (1977) Harvesting natural-populations in a randomly ﬂuctuating environment. Science, 197, 463–465. Bellows, T.S. (1981) The descriptive properties of some models for density dependence. Journal of Animal Ecology, 50, 139–156. Bellows, T.S. (1982) Analytical models for laboratory populations of Callosobruchus-chinensis and Callosobruchus-maculatus (coleoptera, bruchidae). Journal of Animal Ecology, 51, 263–287. Bonsall, M.B., French, D.R. & Hassell, M.P. (2002) Metapopulation structures aﬀect persistence of predator–prey interactions. Journal of Animal Ecology, 71, 1075–1084. Bonsall, M.B. & Hassell, M.P. (2005) Understanding ecological concepts: the role of laboratory systems. Advances in Ecological Research, Population Dynamics and Laboratory Ecology, 37, 1–36. Bonsall, M.B. & Hastings, A. (2004) Demographic and environmental stochasticity in predator–prey metapopulation dynamics. Journal of Animal Ecology, 73, 1043–1055. Bonsall, M.B., Bull, J.C., Pickup, N.J. & Hassell, M.P. (2005) Indirect eﬀects and spatial scaling aﬀect the persistence of multispecies metapopulations. Proceedings of the Royal Society B-Biological Sciences, 272, 1465–1471. Bowler, D.E. & Benton, T.G. (2005) Causes and consequences of animal dispersal strategies: relating individual behaviour to spatial dynamics. Biological Reviews, 80, 205–225. Brooks, E.N. & Lebreton, J.D. (2001) Optimizing removals to control a metapopulation: application to the yellow legged herring gull (Larus cachinnans). Ecological Modelling, 136, 269–284. Bull, J.C. & Bonsall, M.B. (2008) Overcompensatory population dynamic responses to environmental stochasticity. Journal of Animal Ecology, 77, 1296–1305. Bull, J.C., Pickup, N.J., Hassell, M.P. & Bonsall, M.B. (2006) Habitat shape, metapopulation processes and the dynamics of multispecies predator–prey interactions. Journal of Animal Ecology, 75, 899–907. Bull, J.C., Pickup, N.J., Pickett, B., Hassell, M.P. & Bonsall, M.B. (2007) Metapopulation extinction risk is increased by environmental stochasticity and assemblage complexity. Proceedings of the Royal Society B-Biological Sciences, 274, 87–96. Bulte, E.H. & van Kooten, G.C. (1999) Metapopulation dynamics and stochastic bioeconomic modeling. Ecological Economics, 30, 293–299. Cameron, T.C. & Benton, T.G. (2004) Stage-structured harvesting and its eﬀects: an empirical investigation using soil mites. Journal of Animal Ecology, 73, 996–1006. Chatﬁeld, C. (2004) The Analysis of Time Series: An Introduction, 6th edn. Chapman & Hall, CRC. Crawley, M.J. (2007) The R Book. Wiley-Blackwell, Oxford. Cronin, J.T. (2007) From population sources to sieves: the matrix alters host– parasitoid source-sink structure. Ecology, 88, 2966–2976. Cronin, J.T. & Reeve, J.D. (2005) Host–parasitoid spatial ecology: a plea for a landscape-level synthesis. Proceedings of the Royal Society B-Biological Sciences, 272, 2225–2235. Fahrig, L. (2007) Landscape heterogeneity and metapopulation dynamics. Key Topics in Landscape Ecology (eds J. Wu & R.J. Hobbs), pp. 78–91. Cambridge University Press, New York. Fryxell, J.M., Lynn, D.H. & Chris, P.J. (2006) Harvest reserves reduce extinction risk in an experimental microcosm. Ecology Letters, 9, 1025–1031. Fryxell, J.M., Smith, I.M. & Lynn, D.H. (2005) Evaluation of alternate harvesting strategies using experimental microcosms. Oikos, 111, 143–149. Godfray, H.C.J. & Shimada, M. (1999) Parasitoids as model organisms for ecologists. Researches on Population Ecology, 41, 3–10. Goodwin, B.J. & Fahrig, L. (2002) How does landscape structure inﬂuence landscape connectivity? Oikos, 99, 552–570. Goss-Custard, J.D., Stillman, R.A., West, A.D., Caldow, R.W.G., Triplet, P., Durell, S. & McGrorty, S. (2004) When enough is not enough: shorebirds and shellﬁshing. Proceedings of the Royal Society of London Series B-Biological Sciences, 271, 233–237. Hanski, I., Kuussaari, M. & Nieminen, M. (1994) Metapopulation structure and migration in the butterﬂy Melitaea cinxia. Ecology, 75, 747–762. Hansson, L. (1991) Dispersal and connectivity in metapopulations. Metapopulation Dynamics: Empirical and Theoretical Investigations (eds M.E. Gilpin & I. Hanski), pp. 89–103. Academic Press, London. Hassell, M.P. (2000) Host–parasitoid population dynamics. Journal of Animal Ecology, 69, 543–566. Hauser, C.E., Pople, A.R. & Possingham, H.P. (2006) Should managed populations be monitored every year? Ecological Applications, 16, 807– 819.

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 102–111

Harvesting a resource–consumer metapopulation 111 Heong, K.L. (1981) Searching preference of the parasitoid, Anisopteromalus calandrae (Howard) for diﬀerent stages of the host, Callosobruchus maculatus (F.) in the laboratory. Researches on Population Ecology, 23, 177–191. Hilborn, R., Walters, C.J. & Ludwig, D. (1995) Sustainable exploitation of renewable resources. Annual Review of Ecology and Systematics, 26, 45–67. Hoekstra, J. & van den Bergh, J. (2005) Harvesting and conservation in a predator–prey system. Journal of Economic Dynamics & Control, 29, 1097– 1120. Holyoak, M. (2000) Habitat patch arrangement and metapopulation persistence of predators and prey. American Naturalist, 156, 378–389. Hunt, J.J.F.G. & Bonsall, M.B. (2009) The eﬀects of colonization, extinction and competition on co-existence in metacommunities. Journal of Animal Ecology, 78, 866–879. Joshi, H.R., Herrera, G.E., Lenhart, S. & Neubert, M.G. (2009) Optimal dynamic harvest of a mobile renewable resource. Natural Resource Modeling, 22, 322–343. Kareiva, P.M. (1989) Renewing the dialogue between theory and experiments in population ecology. Perspectives in Ecological Theory (eds J. Roughgarden, R.M. May & S.A. Levin), pp. 68–88. Princeton University Press, Princeton. Lande, R. (1987) Extinction thresholds in demographic-models of territorial populations. American Naturalist, 130, 624–635. Lande, R., Engen, S. & Saether, B.E. (1995) Optimal harvesting of ﬂuctuating populations with a risk of extinction. American Naturalist, 145, 728–745. Lande, R., Saether, B.E. & Engen, S. (1997) Threshold harvesting for sustainability of ﬂuctuating resources. Ecology, 78, 1341–1350. Lawton, J.H. (1995) Ecological experiments with model systems. Science, 269, 328–331. Levins, R. (1969) Some demographic and genetic consequences of environmental heterogeneity for biological control. Bulletin of the Entomological Society of America, 15, 237–240. MacFadyen, A. (1963) Animal Ecology: Aims and Methods, 2nd edn. Pitman, London. Matsuoka, T. & Seno, H. (2008) Ecological balance in the native population dynamics may cause the paradox of pest control with harvesting. Journal of Theoretical Biology, 252, 87–97. Matthysen, E. (2005) Density-dependent dispersal in birds and mammals. Ecography, 28, 403–416. May, R.M., Beddington, J.R., Clark, C.W., Holt, S.J. & Laws, R.M. (1979) Management of multispecies ﬁsheries. Science, 205, 267–277. McCullough, D.R. (1996) Spatially structured populations and harvest theory. Journal of Wildlife Management, 60, 1–9. Milner-Gulland, E.J., Shea, K., Possingham, H., Coulson, T. & Wilcox, C. (2001) Competing harvesting strategies in a simulated population under uncertainty. Animal Conservation, 4, 157–167. Murton, R.K., Westwood, N.J. & Isaacson, A.J. (1974) Study of wood-pigeon shooting – exploitation of a natural animal population. Journal of Applied Ecology, 11, 61–81. Ngamo, T.S.L., Kouninki, H., Ladang, Y.D., Ngassoum, M.B., Mapongmestsem, P.M. & Hance, T. (2007) Potential of Anisopteromalus calandrae (Hymenoptera: Pteromalidae) as biocontrol agent of Callosobruchus maculatus (F.) (Coleopetera: Bruchidae). African Journal of Agricultural Research, 2, 168–172. Ovaskainen, O. & Hanski, I. (2004) Metapopulation dynamics in highly fragmented landscapes. Ecology, Genetics and Evolution of Metapopulations (eds I. Hanski & O.E. Gaggioti), pp. 73–104. Elsevier Academic Press, Oxford. Pascual, M.A. & Hilborn, R. (1995) Conservation of harvested populations in ﬂuctuating environments – the case of the serengeti wildebeest. Journal of Applied Ecology, 32, 468–480. Pulliam, H.R. (1988) Sources, sinks, and population regulation. American Naturalist, 132, 652–661. Quinn, T.J. & Deriso, R.B. (1999) Quantitative Fish Dynamics. Oxford University Press, New York. R Foundation for Statistical Computing (2009) R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing, Vienna, http://www.R-project.org. Robinson, H.S., Wielgus, R.B., Cooley, H.S. & Cooley, S.W. (2008) Sink populations in carnivore management: Cougar demography and immigration in a hunted population. Ecological Applications, 18, 1028–1037.

Sanchirico, J.N. & Wilen, J.E. (2001) A bioeconomic model of marine reserve creation. Journal of Environmental Economics and Management, 42, 257– 276. Stillman, R.A., Goss-Custard, J.D., West, A.D., Durell, S., McGrorty, S., Caldow, R.W.G., Norris, K.J., Johnstone, I.G., Ens, B.J., Van der Meer, J. & Triplet, P. (2001) Predicting shorebird mortality and population size under diﬀerent regimes of shellﬁshery management. Journal of Applied Ecology, 38, 857–868. Supriatna, A.K. & Possingham, H.P. (1998) Optimal harvesting for a predatorprey metapopulation. Bulletin of Mathematical Biology, 60, 49–65. Tuck, G.N. & Possingham, H.P. (1994) Optimal harvesting strategies for a metapopulation. Bulletin of Mathematical Biology, 56, 107–127. Wilberg, M.J., Irwin, B.J., Jones, M.L. & Bence, J.R. (2008) Eﬀects of source– sink dynamics on harvest policy performance for yellow perch in southern Lake Michigan. Fisheries Research, 94, 282–289. With, K.A. & King, A.W. (2001) Analysis of landscape sources and sinks: the eﬀect of spatial pattern on avian demography. Biological Conservation, 100, 75–88. Zipkin, E.F., Sullivan, P.J., Cooch, E.G., Kraft, C.E., Shuter, B.J. & Weidel, B.C. (2008) Overcompensatory response of a smallmouth bass (Micropterus dolomieu) population to harvest: release from competition? Canadian Journal of Fisheries and Aquatic Sciences, 65, 2279–2292. Received 17 March 2010; accepted 21 October 2010 Handling Editor: Marc Cadotte

Supporting Information Additional Supporting Information may be found in the online version of this article: Table S1. Population-level models with density-dependent functions describing births (fN) and deaths (lN) in the bruchid metapopulations [taken from Bellows (1981)] were ﬁtted to data from an independent study (Bonsall et al. 2005; Bull et al. 2006, 2007; Bull & Bonsall 2008) in order to estimate parameter values describing population dynamics in these systems [growth rate r, death rate d, density-dependence a (the function from Hassell et al. (1976) has a second parameter for density-dependence b) and likelihood variance v]. Models were ﬁt to replicates separately so that the models which best described the systems could be identiﬁed. Models estimated diﬀerent parameter values depending on the nature of density-dependence acting on births or deaths. The estimated parameter values for each model ﬁtted to each replicate are listed. The ﬁt of the models were measured using AIC values calculated from the model likelihood values. The two best ﬁtting models (MacFadyen birth and death) were then ﬁtted to all of the replicates combined to estimate the parameter values which best describe the population dynamics across all the time series. The AIC value and parameter estimates with 95% conﬁdence limits (Morgan 2000) for the model which best describes the dynamics are listed in bold print. Appendix S1. Model Fitting. Appendix S2. Stochastic models. As a service to our authors and readers, this journal provides supporting information supplied by the authors. Such materials may be re-organized for online delivery, but are not copy-edited or typeset. Technical support issues arising from supporting information (other than missing ﬁles) should be addressed to the authors.

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 102–111

Journal of Applied Ecology 2011, 48, 192–198

doi: 10.1111/j.1365-2664.2010.01909.x

The potential effect of exotic Paciﬁc rats Rattus exulans on vectors of scrub typhus Chi-Chien Kuo1,2, Hsi-Chieh Wang2* and Ching-Lun Huang2 1

Department of Wildlife, Fish, and Conservation Biology, University of California, Davis, CA 95616, USA; and Department of Health, Research and Diagnostic Center, Centers for Disease Control, No. 6, Linsen S. Rd., Taipei, Taiwan, ROC

2

Summary 1. The role and inﬂuence of exotic species in indigenous vector-borne diseases are important but remain understudied. Ascertaining whether disease vectors prefer exotic vs. native hosts has important implications for human health. Moreover, evaluating whether exotic hosts are intrinsically less susceptible to vectors, or whether vector loads can vary with environment, is instructive for possible disease dynamics in the face of range expansion of exotic species. Rattus exulans has recently been recorded for the ﬁrst time in eastern Taiwan, where scrub typhus is prevalent. We assessed the role of R. exulans as a host for chiggers (larval trombiculid mites), and the degree to which vector loads of R. exulans exhibited spatial variation. 2. We deployed live traps in two villages in Taiwan that diﬀered in human population density and in the incidence of scrub typhus. We recovered and tallied chiggers from small mammals, and identiﬁed over one-ﬁfth of chiggers from each host to species. Chiggers were assayed for Orientia tsutsugamushi (OT) infection with nested-PCR. 3. R. exulans was the most common small mammal species captured (31Æ4% total captures), and supported about one-ﬁfth of total chiggers recovered. Leptotrombidium imphalum dominated the chigger assemblage of most native species (>90%), but Gahrliepia spp. was commonly found in R. exulans (39Æ1%). We detected OT in the genus Leptotrombidium (39%) but not in the Gahrliepia. Prevalence and loads of chiggers in R. exulans were about 5· and 17· higher, respectively, in the less densely populated village; a similar trend also occurred with native R. losea. 4. Synthesis and applications. Currently, R. exulans appears to play a relatively minor role in supporting chiggers. However, the fact that both prevalence and loads of chiggers in R. exulans vary greatly with environment, along with the abundance of most exotic species and the ecological ﬂexibility of R. exulans, implies a potential health risk as this species expands to areas with more chiggers. Our study suggests that a clearer understanding of interactions among native and exotic hosts and native parasite species can facilitate prediction of the impact of exotic hosts on the dynamics of vector-borne diseases. Key-words: chigger, exotic species, host-parasite interactions, human health, Orientia tsutsugamushi, Taiwan

Introduction It has been well documented that exotic species can introduce wildlife diseases. For example, the signiﬁcant decline of the endemic Hawaiian avifauna is partially attributable to the introduction of avian malaria and avian pox by exotic birds and their transmission by introduced mosquitoes (Warner *Correspondence author. Centers for Disease Control, Department of Health, No. 6, Linsen S. Rd., Taipei, Taiwan. E-mail: [email protected]

1969). Similarly, rinderpest virus in imported cattle killed a huge proportion of native artiodactyls in Africa (Plowright 1982). Reduction of native red squirrels Sciurus vulgaris followed by the introduction of eastern gray squirrels S. carolinensis in Britain and Europe likely is related to the more pathogenic eﬀects of parapox virus on red squirrels (Tompkins, White & Boots 2003). Dobson & Foufopoulos (2001) reviewed recent emerging and re-emerging wildlife diseases in North America and reported that the majority of diseases are exotic or likely of exotic origin. Examples include West Nile virus, bovine tuberculosis, and avian pox, but many others exist.

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society

Exotic Paciﬁc rats and scrub typhus 193 Exotic species may also inﬂuence the dynamics of indigenous diseases via their interaction with native hosts, vectors, or pathogens. For instance, the introduction of bank voles Clethrionomys glareolus, a less competent host for the indigenous pathogen Bartonella, reduced disease prevalence in native wood mice Apodemus sylvaticus (Telfer et al. 2005); in contrast, an exotic snail Pomacea canaliculata in China is very susceptible to an indigenous nematode Angiostrongylus cantonensis and has led to outbreaks of a human brain disease caused by the nematode (Wang, Chen & Lun 2007). Although rarely studied, the inﬂuence of introduced species on indigenous vector-borne diseases, especially those impacting humans, is critical in the face of ongoing expansion of exotic species, because human diseases may increase if native vectors can exploit exotic hosts. The further impact of expanding exotic species on disease vectors can be appraised by determining whether exotic hosts are intrinsically less susceptible to vectors, or if vector loads are determined by the environment. Under the former scenario, vectors may not increase with introduction of exotic hosts. Alternatively, if the impact of introduced species is contextdependent, then exotics could be a major concern for human health if they were able to expand into areas with proliﬁc vectors. This is especially risky due to the abundance of most exotic species. The Paciﬁc rat Rattus exulans Peale is widely distributed in Southeast Asia and Paciﬁc islands, and this species was ﬁrst recorded in Taiwan in 1999 (Motokawa et al. 2001), although only in Ji-an village of Hua-lien County in eastern Taiwan (Chu et al. 2007). The distribution of this exotic species is expanding southward, and by 2005 it had reached Shou-feng village, located across the Mu-gua River (Fig. 1; C.-C. Kuo, unpublished data). Elsewhere, R. exulans is a major threat to seabirds (Jones et al. 2008), and has had devastating eﬀects in introduced regions, especially on islands (Courchamp, Cha-

puis & Pascal 2003; Towns, Atkinson and Daugherty 2006; Athens 2009). The introduction of R. exulans may also pose human health concerns because they can host plague as well as murine typhus and scrub typhus (Audy & Harrison 1951; Walton et al. 1980; Wodzicki & Taylor 1984). In Taiwan, the seropositivity rates of Hantavirus in R. exulans were higher than other native rodents (Wang 2004). Scrub typhus, vectored by larval trombiculid mites (i.e. chiggers) harbouring the rickettsia Orientia tsutsugamushi (OT), is an acute, febrile human infectious disease prevalent in the West Paciﬁc, South Asia, and northeast Australia, where about one million cases occur annually, and one billion people are at risk. The symptoms of scrub typhus include fever, headache, and rash and fatality rate can be as high as 50% if not treated appropriately (Kawamura, Tanaka & Takamura 1995; Coleman et al. 2003; Lu et al. 2010). The life cycle of trombiculid mites includes seven stages: egg, deutovum, larva (chigger), protonymph, deutonymph, tritonymph and adult. Only chiggers are parasitic. Nymphs and adults are free living in the soil, feeding mainly on the eggs and larvae of arthropods (Kawamura, Tanaka & Takamura 1995). Leptotrombidium chiggers are the primary vectors, and murine rodents, especially Rattus species, are the predominant hosts of chiggers in regions with endemic scrub typhus (Traub & Wisseman Jr. 1974; Kawamura, Tanaka & Takamura 1995). Humans are accidental hosts, and get infected with scrub typhus when bitten by chiggers infected with OT. Trombiculid mites are the only reservoirs of OT, which can be transmitted transstadially (from larva to nymph to adult) and transovarially (from female to next generations), while vertebrate hosts provide chiggers with food resources but play little roles in transmitting OT (Kawamura, Tanaka & Takamura 1995). Although one of Taiwan’s least populated counties, Hualien County had the country’s highest number of human cases of scrub typhus between 1998 and 2007, except for one oﬀshore

Fig. 1. Location of study plots in Ji-an and Shou-feng villages in eastern Taiwan between 2007 and 2008. Plots where Paciﬁc rats Rattus exulans were absent or present are indicated by circle symbol shading. Also shown are plots beyond the Paciﬁc rat invasion that should be re-sampled once R. exulans expands to this area. 2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 192–198

194 C.-C. Kuo, H.-C. Wang & C.-L. Huang islet (Taiwan Centers for Disease Control 2008), and scrub typhus has been endemic there for at least 95 years (Hatori 1919). The introduction of R. exulans to Hua-lien County provides a good opportunity to examine its possible impact on this indigenous human disease. We evaluated the relative role of several small mammal species as hosts of chiggers in Hua-lien, where apart from R. exulans and the rare Rattus argentiventer, and possibly Bandicota indica, all small mammals are native to this region (Lin 1980; Motokawa et al. 2001; Chen 2008). In the study, hosts were deﬁned solely as providers of food resource, without any implication of being reservoirs of disease. We also evaluated whether prevalence and loads of chiggers in R. exulans increased signiﬁcantly when this exotic species expanded into areas with more abundant chiggers.

detached from these skin samples were transferred to 70% ethanol after 2 days, and subsequently tallied. Blood was collected directly by heart puncture for R. exulans and R. argentiventer, whereas for native species we sampled blood either from the submandibular area (small native rodents) or the saphenous vein. Sera was retrieved after centrifugation and stored at )70 C for later determination. Exotic rodent species were then humanely euthanized. Other rodents were fur clipped and released ‡5 km away from the study areas. Shrews were screened for ectoparasite infestation. Those infested with chiggers were euthanized with an overdose of Zoletil 50, and blood collected from their hearts. Those without ectoparasites were marked with fur clips and released outside the study areas without collecting blood. All procedures were approved by the University of California, Davis Animal Use and Care Administrative Advisory Committee, and met guidelines recommended by the American Society of Mammalogists (Gannon, Sikes and the Animal Care and Use Committee of the American Society of Mammalogists 2007).

Materials and methods CHIGGER IDENTIFICATION STUDY AREA

The study was implemented in the lowlands of Ji-an and Shou-feng villages (Fig. 1) in Hua-lien County of eastern Taiwan, where agriculture dominates land use and villages are interspersed among ﬁelds. Ji-an is adjacent to the largest city (Hua-lien City) in Hua-lien County, and is much more densely populated (1211 people km)2) than Shoufeng (88 people km)2) (http://web.hl.gov.tw/static/). Between 1998 and 2006, the incidence of scrub typhus in Shou-feng was over twice that in Ji-an (1Æ45 vs. 0Æ67 cases ⁄ 1000 people per year) (Kuo 2010).

SMALL MAMMAL TRAPPING AND COLLECTION OF CHIGGERS

We surveyed abandoned agricultural ﬁelds from January 2007 to March 2007 (as a preliminary study), and from August 2007 to March 2008 [covering periods with the lowest and the highest number of cases of scrub typhus in humans (Lee et al. 2006)]. In the former period, each ﬁeld was sampled with two parallel transect lines containing 10 Sherman traps (26Æ5 · 10 · 8Æ5 cm) at 10-m intervals, and two hand-made live traps (27 · 16 · 13 cm) at 50-m intervals. Hand-made traps were used to target the larger but less abundant B. indica. Adjacent transect lines were separated by 10 m. After August 2007, we added a third hand-made live trap to each transect line and placed these at 30-m intervals. We used between one and three parallel transect lines in each ﬁeld in accordance with ﬁeld size. Because R. exulans remains limited in distribution, some sampling ﬁelds were only 50 m apart and some ﬁelds were surveyed two to three times to increase R. exulans captures (to avoid concerns over independence, data were pooled across ﬁelds for each village; see ‘Data analysis’). Traps were opened and baited in the evening and checked for captures early in the morning. Fields were surveyed for three consecutive nights. Ji-an and Shou-feng were both surveyed in each trapping bout. Trapped small mammals (rodents and shrews) were transferred to a clean nylon mesh bag which was carefully examined beforehand to ensure that no arthropod vectors remained from earlier captures. Rodents were anesthetized with Zoletil 50 (Fa. Virbac. Carros, France) and examined for gender and reproductive status. We also recorded body weight (g) and length of body, tail, ear and hind-foot (mm). Rodents were thoroughly examined for ectoparasites by carefully combing their fur. Skin with attached chiggers was detached with minimal injury to the animals and preserved in vials; chiggers

For each individual host, at least one-ﬁfth of parasitized chiggers were randomly selected for species identiﬁcation. Chiggers were soaked for 2–3 30-min periods in deionized water and then slidemounted in Berlese ﬂuids (Asco Laboratories, Manchester, UK). Chiggers were examined under a light microscope and identiﬁed with published keys (Wang & Yu 1992; Li, Wang & Chen 1997).

ORIENTIA TSUTSUGAMUSHI DETECTION IN CHIGGERS

We detected OT in chiggers of the genera Leptotrombidium and Gahrliepia (see Results) with nested PCR. To retrieve enough DNA, PCR was performed on pools of 30 chiggers each of the same genus. Chiggers were grouped into genera because further classiﬁcation (i.e. to species) would require the use of Berlese ﬂuids, which destroys DNA material. Chiggers of the same pool were recovered from the same host, except from R. exulans or Apodemus agrarius, where some individuals were infested with <30 chiggers. In those cases, chiggers were pooled from diﬀerent rodent hosts of the same species. This should not pose any problem because rodents do not play any role in the transmission of OT (Kawamura, Tanaka & Takamura 1995), and our main purpose was to examine whether Gahrliepia, commonly found in R. exulans, was also infective of OT (see below). We modiﬁed the method of Kawamori et al. (1993) in the detection of OT in chiggers. This method targeted a well conserved DNA corresponding to 56-kDa type-speciﬁc antigen located on the OT outer membrane. Primers 5¢-AGAATCTGCTCGCTTGGATCCA-3¢ and 5¢-ACCCT ATAGTCAATACCAGCACAA-3¢ were applied for the ﬁrst step PCR, while primers 5¢-GAGCAG AGATAGGTGTT ATGTA-3¢ and 5¢-TATTCATTATAGTAGGCTGA-3¢ were used in the second stage PCR. The positive PCR products were separated by electrophoresis in 1Æ5% agarose gels, stained with ethidium bromide, and identiﬁed under UV ﬂuorescence. Laboratory Karp and Gilliam strains and deionized water were used as positive and negative controls respectively.

DATA ANALYSIS

We tested diﬀerences in trapping success and prevalence of chiggers for each host species between the two villages using Pearson’s chisquare test with Yates’ correction for continuity. Because some ﬁelds could not be considered valid replicates due to their proximity,

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 192–198

Exotic Paciﬁc rats and scrub typhus 195 we pooled data across ﬁelds within each village (e.g. analytical sample size was n = 2 villages). Data are presented as percentages, but chisquare tests were based on raw data. When comparing chigger loads for each host species between villages, host individuals were treated as independent units. We conﬁrmed normality and homogeneity of variance with Shapiro–Wilk and Levene tests, respectively; data were transformed if necessary. We used a t-test when both assumptions were met and a non-parametric Mann–Whitney U-test otherwise. All procedures were implemented in spss 16.0 (SPSS Inc., Chicago, IL, USA).

Results

(13Æ4) and R. exulans (9Æ5) had intermediate chigger loads, whereas the other three common species (M. caroli, Mus musculus and S. murinus) supported very few chiggers (Table 1). We calculated total chiggers, the product of mean chigger loads and total captures for a given host species, to represent the number of chiggers supported by each host species. Rattus losea sustained the most chiggers (62Æ2% of all chiggers recovered). In spite of its moderate chigger loads, the abundance of R. exulans resulted in moderately high chigger populations (21Æ1%), followed by A. agrarius (8Æ9%) and B. indica (7Æ2%). Other species altogether sustained <1% of total chiggers (Table 1).

SMALL MAMMAL TRAPPING COMPOSITION OF CHIGGERS AMONG SMALL MAMMAL

We trapped small mammals in 79 ﬁelds (6249 trap-nights): 31 ﬁelds were surveyed from January to March 2007 (2232 trapnights), and 69 ﬁelds from August 2007 to March 2008 (4017 trap-nights). Nine ﬁelds were sampled in both periods, and 12 ﬁelds were surveyed twice within the second period. We captured 1597 individual small mammals (706 in the ﬁrst time period, 891 in the second) belonging to 10 species (three shrew and seven rodent species). The most commonly captured species was the exotic R. exulans (31Æ4% of total captures), followed by the natives Suncus murinus (23Æ9%) and Mus caroli (21Æ2%). Rattus argentiventer, the other exotic species, constituted only 0Æ2% of total captures (Table 1).

PREVALENCE AND LOADS OF CHIGGERS AMONG SMALL MAMMAL HOSTS

A total of 22 483 chiggers were recovered. Prevalence was greatest in Rattus losea (68%), followed by A. agrarius (61%) and B. indica (43%). Rattus exulans had moderate prevalence (31%) (Table 1). Mean chigger loads were much higher in R. losea (189Æ1) and B. indica (70Æ5) than other species. Apodemus agrarius

Table 1. Prevalence and loads of chiggers (larval trombiculid mites) among small mammal hosts trapped from January to March 2007, and from August 2007 to March 2008 in Hua-lien, eastern Taiwan

Host species

Total No. of Prevalence chiggers captures (%) Chigger recovered (% of all) of chiggers loads (% of all)

HOSTS

We identiﬁed approximately one-third of the chiggers recovered in this study (7111 chiggers = 31Æ6%). Of these, 839 (11Æ8% of identiﬁed) were excluded because they could not be keyed to species (but were identiﬁed to be Leptotrombidium spp.) due to unsuitable position for keying when mounted on the slides. Of the remaining 6272 chiggers, Leptotrombidium imphalum dominated the assemblage (83Æ9%), followed by Gahrliepia spp. (10Æ8%). The remaining 5Æ3% comprised small proportions of L. deliense, Walchia spp. and other species of Leptotrombidium (Table 2). Most host species were infested mainly by L. imphalum (>90Æ0%), except R. exulans, M. musculus and Crocidura attenuata (Table 2). The latter two species, as well as Crocidura suaveolens and S. murinus warrant further study as we were unable to obtain suﬃcient chiggers from these species for further assessment. Chiggers recovered from R. exulans comprised mostly L. imphalum (52Æ3%), followed by Gahrliepia spp. (39Æ1%) (Table 2).

PREVALENCE OF OT AMONG CHIGGERS

Due to the low chigger loads in the most commonly found host species (Table 1), only 80 pools of chiggers were examined for OT infection. We detected OT in the Leptotrombidium spp. (39%, N = 65) recovered from B. indica (80%, N = 5), A. agrarius (61%, N = 18), R. losea (28%, N = 25) and R. exulans (18%, N = 17). OT was not found in the genus Gahrliepia (0%, N = 15) recovered from R. exulans.

A COMPARISON OF PREVALENCE AND LOADS OF

Apodemus agrarius Bandicota indica Crocidura attenuata Crocidura suaveolens Mus caroli Mus musculus Rattus argentiventer Rattus exulans Rattus losea Suncus murinus Total

149 (9Æ3) 23 (1Æ4) 1 (0Æ1) 1 (0Æ1)

61 43 100 100

13Æ4 70Æ5 5 16

339 (21Æ2) 124 (7Æ8) 3 (0Æ2) 502 (31Æ4) 74 (4Æ6) 381 (23Æ9) 1597

0 3 100 31 68 1 20

0 0Æ1 26 9Æ5 189Æ1 0Æ1 14Æ1

1990 1621 5 16

(8Æ9) (7Æ2) (0Æ02) (0Æ1)

0 12 78 4747 13 994 20 22 483

(0) (0Æ1) (0Æ4) (21Æ1) (62Æ2) (0Æ1)

CHIGGERS AMONG SMALL MAMMAL HOSTS IN JI-AN VS. SHOU-FENG

We surveyed 23 plots (1932 trap-nights) in Ji-an, and 56 plots (4317 trap-nights) in Shou-feng. R. exulans was the most common species in Ji-an (41Æ8%), followed by S. murinus (24Æ4%) and M. musculus (22Æ2%) (Table 3). R. exulans was also the most abundant species in Shou-feng (28Æ5%), followed by M. caroli (24Æ9%) and S. murinus (23Æ7%) (Table 3). Trapping success (captures ⁄ trap-nights) was approximately 50% greater in Shou-feng (28Æ8%) than in Ji-an (18Æ2%)(chi-square test

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 192–198

196 C.-C. Kuo, H.-C. Wang & C.-L. Huang Table 2. Species composition of chiggers (%) among small mammal hosts trapped from January to March 2007, and from August 2007 to March 2008 in Hua-lien, eastern Taiwan Chigger species composition (%) within each host species

Host species

Leptotrombidium imphalum

Leptotrombidium deliense

Leptotrombidium others

Walchia sp.

Gahrliepia sp.

No. chiggers identiﬁed

Apodemus agrarius Bandicota indica Crocidura attenuata Crocidura suaveolens Mus musculus Rattus argentiventer Rattus exulans Rattus losea Suncus murinus Total

94Æ6 96Æ6 33Æ3 100 0 97Æ2 52Æ3 94Æ5 91Æ7 83Æ9

3Æ6 1Æ7 66Æ7 0 0 2Æ8 1Æ3 1Æ9 8Æ3 2Æ0

1Æ6 0Æ9 0 0 100 0 1Æ3 1Æ0 0 1Æ3

0 0Æ9 0 0 0 0 6Æ0 0Æ9 0 2Æ0

0 0 0 0 0 0 39Æ1 1Æ8 0 10Æ8

799 351 3 9 10 36 1573 3479 12 6272

Table 3. Total captures and relative abundance of each small mammal host species in Ji-an and Shou-feng villages from January to March 2007, and from August 2007 to March 2008 in Hua-lien, eastern Taiwan Ji-an

Shou-feng

Host species

Relative Relative abundance No. of abundance No. of captures (%) captures (%)

Apodemus agrarius Bandicota indica Crocidura attenuata Crocidura suaveolens Mus caroli Mus musculus Rattus argentiventer Rattus exulans Rattus losea Suncus murinus Total Total trap-nights

0 3 0 0 29 78 0 147 9 86 352 1932

0 0Æ9 0 0 8Æ2 22Æ2 0 41Æ8 2Æ6 24Æ4 –

149 20 1 1 310 46 3 355 65 295 1245 4317

12Æ0 1Æ6 0Æ1 0Æ1 24Æ9 3Æ7 0Æ2 28Æ5 5Æ2 23Æ7 –

with Yates’ correction, v2 = 78Æ56, P < 0Æ001), but trapping success of R. exulans was similar in both villages (Ji-an: 7Æ6%; Shou-feng: 8Æ2%; v2 = 0Æ60, P = 0Æ44). Six small mammal species occurred in both villages. Chigger prevalence was greater in Shou-feng for all host species except M. musculus (Ji-an: 4%; Shou-feng: 2%) and M. caroli (0% for both villages) (Fig. 2a). Prevalence of chiggers on R. exulans in Shou-feng was nearly 5· that in Ji-an (v2 = 50Æ15, P < 0Æ001), and that on R. losea was >2· greater in Shou-feng (v2 = 3Æ85, P = 0Æ05). The other four species did not diﬀer in prevalence between these villages (Fig. 2a). Chigger loads also were higher in Shou-feng, with the exception of M. caroli (0 for both villages). Chigger loads in R. exulans were >17· higher in Shou-feng than Ji-an (Mann– Whitney U-test, U = 17303, Z = )7Æ23, P < 0Æ001), and those in R. losea were 11· higher in Shou-feng (U = 132Æ5, Z = )2Æ69, P = 0Æ007). Chigger loads in the other four species did not diﬀer between the two villages (Fig. 2b). Total chiggers recovered from all small mammals were much higher

Fig. 2. Comparison of prevalence (a) and loads (b) of chiggers in small mammal host species between Ji-an and Shou-feng villages. Statistical results are from Chi-square contingency tests (a) and Mann–Whitney U-test (b). Notice that the y-axis in (b) is a logarithmic scale.

in Shou-feng (22 191) than in Ji-an (292); after correcting for diﬀerential trapping eﬀort, total chiggers recovered in Shoufeng were nearly 34 times those in Ji-an.

Discussion The exotic R. exulans was the most abundant small mammal in our study sites, but it played a relatively minor role in

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 192–198

Exotic Paciﬁc rats and scrub typhus 197 supporting chigger populations, contributing to about oneﬁfth of total chiggers recovered. Moreover, about 40% of chiggers recovered from this species belonged to Gahrliepia spp. in which OT was not detected, indicating that R. exulans may support a lower proportion of chiggers that could potentially transmit OT to humans. Rattus argentiventer, the other recently introduced species, sustained very few chiggers (0Æ35% of total chiggers) due to its rare occurrence. In contrast, the native R. losea was not common (<5% of total captures), but was by far the dominant host for chiggers (62Æ24% total chiggers). It is still not clear whether the introduction of R. exulans has replaced or even facilitated other native species which also host chiggers. These data suggest that, currently, the presence of R. exulans is not likely to increase greatly the susceptibility of local residents to scrub typhus. Some rodent species in our study appear intrinsically less susceptible to chiggers (M. caroli and M. musculus), exhibiting similarly low prevalence and loads in both study regions. This may be due to their smaller body size (both <15 g) and less amount of energy reserve, thus less tolerant of vector infestation (Hart et al. 1992; Olubayo et al. 1993). In contrast, the prevalence and loads of chiggers in R. exulans (mean body weight >30 g) were very diﬀerent in Shou-Feng and Ji-an (5· and 17· higher, respectively, in the former), supporting the hypothesis that prevalence and loads can greatly increase in response to local conditions. Such signiﬁcant variation despite similar population density (trapping success) of R. exulans at both villages suggests that the degree of chigger infestation could be determined by the environment (or other native mammals, or both), although the ineﬃciency of sampling free-living chiggers in Taiwan (Wang et al. 2005) precluded us from comparing populations of host-seeking chiggers at these two areas. However, even after correcting for diﬀerential trapping eﬀort, the total number of chiggers recovered from small mammals in Shou-feng was nearly 34· than in Ji-an. Rodents in abandoned ﬁelds harboured many more chiggers than those in frequently disturbed ﬁelds (Kuo 2010), so one explanation for the diﬀerence in chigger abundance between Shou-feng and Ji-an could be the much lower human density and associated disturbance at the former, although this is confounded by the higher small mammal abundances there. Because both prevalence and loads of chiggers in R. exulans can vary with local conditions, monitoring the expansion of R. exulans and its subsequent inﬂuence on chigger abundance and human incidence of scrub typhus is imperative. For comparative purposes, we trapped small mammals and collected chiggers in regions ahead of the R. exulans invasion front (Fig. 1), using identical sampling and during the same time period as this study (Kuo 2010). In areas where R. exulans had not yet invaded, A. agrarius was the dominant host of chiggers (47Æ8% of total chiggers recovered), followed by R. losea (40Æ8%). Prevalence and loads of chiggers in A. agrarius were much lower in Shou-feng than in the area beyond the invasion front (prevalence: 61% vs. 97%; loads: 13Æ4 vs. 105Æ3). Similar trends were observed for R. losea (prevalence: 72% vs. 100%; loads: 212Æ6 vs. 544Æ1) and B. indica (prevalence: 50% vs. 90%; loads: 81Æ1 vs. 288Æ1)

(Kuo 2010). Together with the variable susceptibility of R. exulans to chiggers, we expect that if this rodent can expand into and maintain high population density in this region, its presence may signiﬁcantly increase the total number of chiggers. In southeast Asia and Paciﬁc islands, R. exulans occurs in houses, plantations, grasslands, and forest, but rarely in virgin rainforest (Audy & Harrison 1951; Williams 1973; Roberts & Craig 1990). Almost all of lowland Hua-lien has been converted for human use. It is thus likely that R. exulans will expand further within this region. However, the outcome of expansion of R. exulans on chigger populations will also depend on the responses of native species, especially the main hosts of chiggers (A. agrarius and R. losea). Our study provides valuable baseline data for investigating the response of chiggers and native small mammal species to the ongoing expansion of R. exulans. The prevalence and loads of chiggers were lower in R. exulans than in some native small mammal species (R. losea, B. indica and A. agrarius), but higher than others (M. caroli, M. musculus and S. murinus). Given these observations, it seems that inferences about the susceptibility of exotic hosts that have been extrapolated from species-poor systems (normally 1–2 species; e.g. Malmstrom et al. 2005; Telfer et al. 2005; Georgiev et al. 2007) may yield an overly simplistic view of the role of exotics within more diverse assemblages. This further suggests that the inﬂuence of exotic species on indigenous disease dynamics is context-dependent. When exotics invade areas with highly susceptible native host species, their impact on disease dynamics may be negligible, or even negative if they suppress native host populations. On the contrary, vectors and diseases may be ampliﬁed if the invasive species is a viable host for the vectors and the indigenous community is mostly less susceptible. Consequently, predicting whether a vector-borne disease will emerge after the introduction of a new species requires an understanding of host–parasite interactions in the native communities, as well as of the extent to which native parasites can exploit exotic hosts. Many studies have documented the health risks that exotic species can pose to humans. Introduced Norway rats Rattus norvegicus and black rats (Rattus rattus) are reservoirs of plague in Madagascar (Chanteau et al. 1998). Hantavirus has been isolated from introduced R. norvegicus in Korea and Brazil (Lee, Baek & Johnson 1982; LeDuc et al. 1985). Rickettsia felis, which is pathogenic to humans, has been retrieved from ﬂeas on exotic house mice Mus musculus in Hawaii (Eremeeva et al. 2008). Our study adds R. exulans to the list of exotic species potentially impacting human health. More importantly, similar studies have been implemented primarily from the perspectives of public health, leaving underlying ecological processes poorly evaluated and possible disease dynamics underappreciated. Our study demonstrates that the impact of exotic species on human diseases may be contingent on local conditions. A better understanding of the interactions between exotic species and disease vectors will help to predict the emergence of vector-borne disease. Given the pace of species introductions globally, this knowledge will be crucial for successful management and control.

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 192–198

198 C.-C. Kuo, H.-C. Wang & C.-L. Huang

Acknowledgements We thank HY Wu and her students for logistical support during the ﬁeld work. DA Kelt, JE Foley, and DH Van Vuren provided constructive comments on this manuscript. This study would not have been possible without the ﬁnancial support by Taiwan Centers for Disease Control (DOH96-DC-2019). C.-C. Kuo was also ﬁnancially supported by the University of California, Davis, the Paciﬁc Rim Research Program and the American Society of Mammalogists.

References Athens, J.S. (2009) Rattus exulans and the catastrophic disappearance of Hawai’i’s native lowland forest. Biological Invasions, 11, 1489–1501. Audy, J.R. & Harrison, J.L. (1951) A review of investigations of mite typhus in Burma and Malaya, 1945–1950. Transactions of the Royal Society of Tropical Medicine and Hygiene, 44, 371–404. Chanteau, S., Ratsifasoamanana, L., Rasoamanana, B., Rahalison, L., Randriambelosoa, J., Roux, J. & Rabeson, D. (1998) Plague, a reemerging disease in Madagascar. Emerging Infectious Diseases, 4, 101–104. Chen, Y.W. (2008) Genetic Variation of the Greater Bandicoot Rat (Bandicota indica) in Taiwan Based on Molecular Data. Master Thesis. Department of Life Science, Tunghai University, Tai-chung, Taiwan. Chu, J.H., Tsao, Y.J., Chuang, Y.J., Chen, H.C., Lin, Y.S. & Wu, H.Y. (2007) Genetic patterns and biogeographic origins of the exotic rodent (Rattus exulans) in Taiwan. Endemic Species Research, 9, 1–11. Coleman, R.E., Monkanna, T., Linthicum, K.J., Strickman, D.A., Frances, S.P., Tanskul, P., Kollars, T.M., Inlao, I., Watcharapichat, P., Khlaimanee, N., Phulsuksombati, D., Sangjun, N. & Lerdthusnee, K. (2003) Occurrence of Orientia tsutsugamushi in small mammals from Thailand. American Journal of Tropical Medicine and Hygiene, 69, 519–524. Courchamp, F., Chapuis, J.L. & Pascal, M. (2003) Mammal invaders on islands: impact, control and control impact. Biological Review, 78, 347–383. Dobson, A.P. & Foufopoulos, J. (2001) Emerging infectious pathogens in wildlife. Philosophical Transactions of the Royal Society of London-B, 356, 1001–1012. Eremeeva, M.E., Warashina, W.R., Sturgeon, M.M., Buchholz, A.E., Olmsted, G.K., Park, S.Y., Eﬄer, P.V. & Karpathy, S.E. (2008) Rickettsia typhi and R. felis in rat ﬂeas (Xenopsylla cheopis), Oahu, Hawaii. Emerging Infectious Diseases, 14, 1613–1615. Gannon, W.L., Sikes, R.S. & The Animal Care and Use Committee of the American Society of Mammalogists (2007) Guidelines of the American Society of Mammalogists for the use of wild mammals in research. Journal of Mammalogy, 88, 809–823. Georgiev, B.B., Sa´nchez, M.I., Vasileva, G.P., Nikolov, P.N. & Green, A.J. (2007) Cestode parasitism in invasive and native brine shrimps (Artemia spp.) as a possible factor promoting the rapid invasion of A. franciscana in the Mediterranean region. Parasitology Research, 101, 1647–1655. Hart, B.L., Hart, L.A., Mooring, M.S. & Olubayo, R. (1992) Biological basis of grooming behaviour in antelope: the body-size, vigilance and habitat principles. Animal Behaviour, 44, 615–631. Hatori, J. (1919) On the endemic tsutsugamushi disease of Formosa. Annals of Tropical Medicine and Parasitology, 13, 233–258. Jones, H.P., Tershy, B.R., Zavaleta, E.S., Croll, D.A., Keitt, B.S., Finkelstein, M.E. & Howald, G.R. (2008) Severity of the eﬀects of invasive rats on seabirds: a global review. Conservation Biology, 22, 16–26. Kawamori, F., Akiyama, M., Sugieda, M., Kanda, T., Akahane, S., Yamamoto, S., Ohashi, N. & Tamura, A. (1993) Two-step polymerase chain reaction for diagnosis of scrub typhus and identiﬁcation of antigenic variants of Rickettsia tsutsugamushi. Journal of Veterinary Medical Science, 55, 749–755. Kawamura, A., Tanaka, H. & Takamura, A. (1995) Tsutsugamushi Disease: An Overview. University of Tokyo Press, Tokyo, Japan. Kuo, C.C. (2010) Ecology of scrub typhus in Taiwan: the impact of environment, rodent community, and socio-economy. PhD dissertation. University of California, Davis. LeDuc, J.W., Smith, G.A., Pinheiro, F.P., Vasconcelos, P.F., Rosa, E.S. & Maiztegui, J.I. (1985) Isolation of a Hantaan-related virus from Brazilian

rats and serologic evidence of its widespread distribution in South America. American Journal of Tropical Medicine and Hygiene, 34, 810–815. Lee, H.W., Baek, L.J. & Johnson, K.M. (1982) Isolation of Hantaan virus, the etiologic agent of Korean hemorrhagic fever, from wild urban rats. Journal of Infectious Diseases, 146, 638–644. Lee, Y.S., Wang, P.H., Tseng, S.J., Ko, C.F. & Teng, H.J. (2006) Epidemiology of scrub typhus in eastern Taiwan, 2000–2004. Japanese Journal of Infectious Diseases, 59, 235–238. Li, J., Wang, D. & Chen, X. (1997) Trombiculid Mites of China: Studies on Vector and Pathogen of Tsutsugamushi Disease. Guangdong Science and Technology Publishing, Guangzhou, China. Lin, L.G. (1980) Mammal research in Taiwan. Master Thesis. Department of Biology, Tunghai University, Tai-chung, Taiwan. Lu, H.Y., Tsai, K.H., Yu, S.K., Cheng, C.H., Yang, J.S., Su, C.L., Hu, H.C., Wang, H.C., Huang, J.H. & Shu, P.Y. (2010) Phylogenetic analysis of 56kDa type-speciﬁc antigen gene of Orientia tsutsugamushi isolates in Taiwan. American Journal of Tropical Medicine and Hygiene, 83, 658–663. Malmstrom, C.M., McCullough, A.J., Johnson, H.A., Newton, L.A. & Borer, E.T. (2005) Invasive annual grasses indirectly increase virus incidence in California native perennial bunchgrasses. Oecologia, 145, 153–164. Motokawa, M., Lu, K.H., Harada, M. & Lin, L.K. (2001) New records of the Polynesian rat Rattus exulans (Mammalia: Rodentia) from Taiwan and the Ryukyus. Zoological Studies, 40, 299–304. Olubayo, R.O., Jono, J., Orinda, G., Groothenhuis, J.G. & Hart, B.L. (1993) Comparative diﬀerences in densities of adult ticks as a function of body size on some East African antelopes. African Journal of Ecology, 31, 26–34. Plowright, W. (1982) The eﬀects of rinderpest and rinderpest control on wildlife in Africa. Symposia of the Zoological Society of London, 50, 1–28. Roberts, M. & Craig, J.L. (1990) The demography of kiore, Rattus exulans in three habitats. New Zealand Journal of Zoology, 17, 43–53. Taiwan Centers for Disease Control (2008) Notiﬁable Infectious Disease Statistics System. Available at: http://nidss.cdc.gov.tw/ (accessed 6 December 2008). Telfer, S., Brown, K.J., Sekules, R., Begon, M., Hayden, T. & Birtles, R. (2005) Disruption of a host-parasite system following the introduction of an exotic host species. Parasitology, 130, 661–668. Tompkins, D.M., White, A.R. & Boots, M. (2003) Ecological replacement of native red squirrels by invasive greys driven by diseases. Ecology Letters, 6, 189–196. Towns, D.R., Atkinson, I.A.E. & Daugherty, C.H. (2006) Have the harmful eﬀects of introduced rats on islands been exaggerated? Biological Invasions, 8, 863–891. Traub, R. & . Wisseman Jr, C.L. (1974) The ecology of chigger-borne rickettsiosis (scrub typhus). Journal of Medical Entomology, 11, 237–303. Walton, D.W., Brooks, J.E., Thinn, K.K. & Tun, U.M.M. (1980) Reproduction in Rattus exulans in Rangoon, Burma. Mammalia, 44, 349–360. Wang, H.C. (2004) Epidemiology of tsutsugamushi disease and hantavirus infection in Taiwan, pp. 93–111. Proceedings of the symposium on rodent damage and control strategy, Taichung County, Taiwan. September 16. Wang, Q.P., Chen, X.G. & Lun, Z.R. (2007) Invasive freshwater snail, China. Emerging Infectious Disease, 13, 1119–1120. Wang, D.Q. & Yu, Z.Z. (1992) Chigger mites of the genus Leptotrombidium: key to species and their distribution in China. Medical and Veterinary Entomology, 6, 389–395. Wang, H.C., Teng, H.J., Huang, S.C., Chen, S.H., Lin, T.H. & Yuan, S.Y. (2005) A report on ﬁeld control of trombiculid mites, vectors of scrub typhus, in Kinmen. Taiwan Epidemiology Bulletin, 21, 36–45. Warner, R.E. (1969) The role of introduced diseases in the extinction of the endemic Hawaiian avifauna. Condor, 70, 101–120. Williams, J.M. (1973) The ecology of Rattus exulans (Peale) reviewed. Paciﬁc Science, 27, 120–127. Wodzicki, K. & Taylor, R.H. (1984) Distribution and status of the Polynesian rat Rattus exulans. Acta Zoologica Fennica, 172, 99–101.

Received 9 November 2009; accepted 27 October 2010 Handling Editor: Christl Donnelly

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 192–198

Journal of Applied Ecology 2011, 48, 121–132

doi: 10.1111/j.1365-2664.2010.01910.x

Conserving pelagic habitats: seascape modelling of an oceanic top predator Maite Louzao1,2*, David Pinaud1, Clara Pe´ron1, Karine Delord1, Thorsten Wiegand2 and Henri Weimerskirch1 1

Centre d’Etudes Biologiques de Chize´, CNRS UPR 1934, 79369 Villiers en Bois, France; and 2UFZ–Helmholtz Centre for Environmental Research, Permoserstrasse 15, 04318 Leipzig, Germany

Summary 1. Currently pelagic ecosystems are changing signiﬁcantly due to multiple threats. An important management policy is to establish marine protected areas, until now overlooked due to the diﬃculty of declaring ‘high seas’ protected areas, obtaining long-term distribution data on indicator species and the dynamic nature of these ecosystems. 2. Within this framework, we developed predictive habitat suitability models of an oceanic predator, the vulnerable wandering albatross Diomedea exulans, in the highly dynamic Southern Ocean. Based on a long-term tracking database (1998–2008), we estimated three quantitative ecological indices that complementarily describe the hierarchical habitat use of the species at multiple spatial scales: where the species (i) spent more time (the seascape, based on the time spent per area), (ii) searched for prey (the foraging habitat, based on zones of increased foraging intensity using ﬁrst passage time), and (iii) fed (the feeding habitat, based on prey capture data). 3. Predictive habitat models reasonably matched the observed distribution patterns and described albatross multi-scale habitat use as a hierarchical arrangement: albatrosses foraged over topographic features in subtropical waters, nested within the wider seascape due to the constraint imposed by the colony eﬀect, whereas feeding occurred nested over the continental shelf and seamounts in areas of low oceanographic variability within the Polar Frontal Zone. 4. Within the current oceanographic conditions, the location of key pelagic habitats for albatrosses breeding in the southern Indian Ocean encompassed certain topographic features such as pelagic areas surrounding main breeding sites, seamounts and submarine mountain ranges. The placement of these pelagic hotspots depends on the current sea surface temperature conditions. 5. Synthesis and applications. The present study provides two key conservation and management tools. First, we provide the ﬁrst map to support the development of a prospective network of priority conservation zones across the southern Indian Ocean based on habitat predictions of an oceanic indicator species. This could be used not only to support conservation of top predators but also the underlying biodiversity associated with pelagic key habitats. Secondly, the developed habitat modelling procedure is widely applicable and could be used to track changes in species distribution in both marine and terrestrial environments within the current global change scenario. Key-words: ﬁrst passage time, habitat modelling, indicator species, network of marine protected areas, prey capture data, Southern Ocean, time spent per unit area, wide-ranging predators, wandering albatross

Introduction The structure and functioning of pelagic ecosystems are changing signiﬁcantly due to multiple threats (e.g. climate change, overﬁshing and pollution; Game et al. 2009). It is increasingly *Correspondence author. E-mail: [email protected]

reported that overexploitation of natural resources following new technological development has caused adverse impacts on the oceanic environment (Game et al. 2009). For instance, deep ocean habitats in remote areas are being exploited as traditional ﬁshing grounds have been depleted. The need for an ecosystem-based approach for marine conservation applies to the protection of all trophic levels, including top predators

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society

122 M. Louzao et al. (Hooker & Gerber 2004). Species belonging to the higher trophic levels play a key role in ecosystem functioning (Heithaus et al. 2008), but they are declining worldwide at a rapid rate (Myers & Worm 2003), driven to severely low levels, and consequently aﬀecting other species and habitats (Norse & Crowder 2005). Similarly, the loss of these taxa will have important eﬀects in pelagic ecosystems, but being diﬃcult to detect, further research is needed in order to elucidate their ecological eﬀects (Heithaus et al. 2008). A major conservation initiative for the protection of marine ecosystems is the establishment of marine protected areas (MPAs), specially focused on coastal areas for protecting sessile and sedentary taxa (Hooker & Gerber 2004; Game et al. 2009). The protection of pelagic ecosystems has been largely overlooked until now due to the diﬃculty of (i) declaring ‘high seas’ protected areas, (ii) obtaining long-term distribution data on pelagic species and (iii) the dynamic character of pelagic habitats (Game et al. 2009). Some of the challenges of identifying key pelagic habitats could be overcome by using ecological indicator species characteristic of a particular habitat or biological community. These species could be used to assess biodiversity hotspots for candidate protected areas (Zacharias & Roﬀ 2001). In the case of marine conservation and management, indicator species should be distributed over wide distribution ranges, easy to observe and not be continuously harvested (Zacharias & Roﬀ 2001). Although selecting indicator species is controversial, seabirds have been identiﬁed as potentially good indicators (Zacharias & Roﬀ 2001). They are easy to monitor thanks to their land-based breeding which facilitates the study of their at-sea distribution via tracking devices. Albatrosses and petrels are the most pelagic of seabirds, occurring in all oceans and are therefore especially suited as indicator species (Furness & Camphuysen 1997). Moreover, they are highly sensitive components of the marine ecosystem since major system shifts will be reﬂected in their population sizes (Weimerskirch et al. 2003). Within this framework, we studied the distribution patterns of the wandering albatross Diomedea exulans (Linnaeus, 1758) in a highly dynamic pelagic ecosystem: the Southern Ocean, where the species breeds in several sub-Antarctic islands. The need for urgent conservation measures is highlighted by analysis of long-term demographic data that has revealed a rapid population decline over three generations and has classiﬁed the wandering albatross as Vulnerable (BirdLife International 2009). Fishing bycatch is suspected to be the main factor aﬀecting this decline, through a reduction in adult survival and juvenile recruitment (Weimerskirch, Brothers & Jouventin 1997). Understanding the features that determine the distribution of oceanic predators is a prerequisite for identifying key areas for their conservation, and hence guides the establishment of MPAs. We developed habitat suitability models to deﬁne the pelagic habitats of this threatened albatross breeding in French Southern Territories, which comprises 40% of the global breeding population (Delord et al. 2008). Our study is based on a long-term tracking database (1998–2008). This provides a unique opportunity for considering diﬀerent oceanographic scenarios, which allows us to consider the dynamic character

of pelagic ecosystems when delineating MPAs. Based on an integrative habitat modelling approach, we estimated three different quantitative ecological indices that complementarily describe the hierarchical habitat use of the species at multiple spatial scales: where the species (i) spent more time (the seascape, based on the time spent per area), (ii) searched for prey (the foraging habitat, based on zones of increased foraging intensity using ﬁrst passage time, FPT) and (iii) fed (the feeding habitat, based on prey capture data). Within the distribution range of the species, the identiﬁcation of the seascape provides a global and wider perspective of the pelagic habitat use, integrating information on diﬀerent behaviours such as foraging and resting. The delineation of the foraging habitat gives insights into a more speciﬁc behaviour: prey searching. Ultimately, the deﬁnition of the feeding habitat aﬀords the most speciﬁc activity: feeding, without consideration of other habitat uses. Since albatrosses are central place foragers and could change their foraging strategies and habitats depending on the breeding stage (Weimerskirch et al. 1993), we studied in more detail habitat use during the incubation and brooding periods. Finally, we interpreted the complementarity of all three habitat modelling outputs in relation to the ecology of the species, within the oceanographic context of the southern Indian Ocean. We discuss the implication of our results in the current conservation scenario, which involves diﬀerent Regional Fisheries Management Organisations (RFMOs). To our knowledge, this is the ﬁrst time that diﬀerent ecological indices quantifying habitat use at diﬀerent spatial scales have been applied to characterize the pelagic habitat of a wide-ranging top predator.

Materials and methods BIRD TRACKING

Birds of known age and sex were tracked at Crozet and Kerguelen Islands (southern Indian Ocean), during both incubation (December–March) and brooding (April) periods over 8 years from 1998 to 2008 (Table S1, Supporting information). Albatrosses were equipped with three diﬀerent tracking devices: Argos PTT Satellite Transmitters powered with battery and working in continuous mode (134 birds; 1998–2003), global positioning system (GPS; 18 birds during the 2002–2005 period) and duty-cycle GPS ⁄ Argos satellite transmitters solar panel (22 birds; 2008). Proportionally, 13% of the equipped birds were from Kerguelen. Additionally, some of the birds tracked with Argos PTT (1998–2001) and GPS (2002–2005) were also equipped with stomach temperature transmitters (STT; Weimerskirch, Gault & Cherel 2005; Weimerskirch et al. 2007). Since some of the tracked individuals performed more than one foraging trip, we randomly selected one per bird (to avoid pseudo-replication) totalling 149 foraging trips from Crozet and 18 from Kerguelen. The total mass of devices was below the recommended 3% threshold (Phillips, Xavier & Croxall 2003) and the same procedure has been used over the last 20 years (Weimerskirch et al. 2007). Analyses were performed on complete foraging trips (93% of trips); although incomplete trips were also included when prey capture data were available (STT were regurgitated before birds returned to the colony). We used all Argos locations (classes A, B, 0, 1–3),

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 121–132

Seascape modelling of a marine predator 123 after ﬁltering positions above 90 km h)1 (McConnell, Chambers & Fedak 1992). Speed ﬁltering led to the removal (in average) of the 5Æ1% of the positions of one foraging trip (range: 0–37Æ1, Table S1, Supporting information).

which animals alter their movements (i.e. concentrates its foraging eﬀort) in response to landscape heterogeneity based on FPT analysis (Pinaud & Weimerskirch 2005). Also, we identiﬁed nested spatial scales which might have an important biological meaning (Fauchald & Tveraa 2003; Pinaud & Weimerskirch 2005). Then, we identiﬁed zones of high FPT values (i.e. higher foraging eﬀort) along the tracks for both maximum and nested spatial scales above a FPT threshold value determined from its multimodal distribution (see Fig. S1, Supporting information; Pinaud & Weimerskirch 2007). Finally, we assigned a categorical binomial variable to each position of the interpolated track, indicating whether the albatross was ‘foraging’ or ‘not foraging’ within a given cell. Note that the entire foraging trip was analysed following Fauchald & Tveraa (2003) and Pinaud & Weimerskirch (2005) and consequently we did not remove locations with high FPT values occurring at night (Pinaud & Weimerskirch 2007).

HABITAT USE ECOLOGICAL INDICES: A THREE-LEVEL HIERARCHICAL APPROACH

We applied diﬀerent methodological approaches to identify diﬀerent marine habitat use of the wandering albatross at multiple spatial scales, within the R environment (R Development Core Team 2008). We built a standard spatial grid based on the geographic limits of the tracking data (from 1W to 124E and from 30S to 68S) where tracking locations and environmental data were overlaid. The 0Æ25º cell size (152 · 500 cells) was chosen according to the available oceanographic data (Table 1) and the accuracy of the tracking devices.

Prey capture Time spent per unit area

A total of 34 breeders were tracked and equipped with STT to locate prey capture events (Weimerskirch, Gault and Cherel 2005, 2007). Stomach temperature data were analysed in order to determine positions of feeding events along each track, and then re-coded into a binary ‘feeding’ ⁄ ’not feeding’ variable, indicative of whether at least one feeding event occurred within a given cell.

We used the tripGrid function (trip package) which resamples each individual track at a higher temporal resolution by linear interpolation (every 60 s) in order to approximate the time spent in each 0Æ25size cell (more details in http://staﬀ.acecrc.org.au/~mdsumner/Rutas/ trip-demo.pdf). Then, we estimated the percentage of time spent in each cell relative to the total duration of the trip.

HABITAT MODELLING PROCEDURE: IDENTIFYING

First passage time

PELAGIC HABITATS

Within the marine environment, resources are distributed heterogeneously, thus animals often alter their movement rates and ⁄ or frequencies of turns in response to local resource abundance by adopting an area-restricted search behaviour, resulting in slow speed and sinuous trajectories (Benhamou 1992). We detected the scale at

We used a hierarchical modelling approach to identify those environmental variables (see details in Table 1) that most accurately reﬂected the seascape and both foraging and feeding habitats of wandering albatross within the information theoretic approach (Fig. S1, Supporting information; Louzao et al. 2009).

Table 1. Biologically relevant explanatory variables used for habitat modelling and associated oceanographic processes. Dynamic variables were downloaded on a monthly basis. Since they diﬀered in spatial resolutions, they were aggregated to match the standard grid of 0Æ25 cell size. Static variables were extracted once and aggregated. BAT, SST, CHL, SLA, and WIND were not normally distributed and we used the median as it is less strongly inﬂuenced by outliers (Zuur, Ieno & Smith 2007) Range2 (min–max)

Oceanographic process

Explanatory variables

Satellite

Spatial resolution

Dynamic Chlorophyll a (CHL, mg m)3) CHL gradient (CHLG)3 Sea surface temperature (SST, C) SST gradient (SSTG)3 Sea level anomaly (SLA, cm) SLA gradient (SLAG)3 Wind speed (WIND, m s)1)

SEAWIFS SEAWIFS PATHFINDER PATHFINDER AVISO AVISO BLENDED

0Æ1 0Æ1 0Æ04 0Æ04 0Æ25 0Æ25 0Æ25

0Æ051–1Æ657 0Æ000–99Æ487 0Æ45–24Æ60 2Æ00–82Æ33 )0Æ619–1Æ052 0Æ214–58Æ912 5Æ822–12Æ746

Ocean productivity domains Frontal systems Water mass distribution Frontal systems Presence of eddies Frontal systems Wind speed

Static Bathymetry (BAT, m) BAT gradient (BATG)3

ETOPO ETOPO

0Æ03 0Æ03

80Æ735–5847Æ816 0Æ187–96Æ522

–

–

12Æ256–3354Æ582

Coastal vs. pelagic domains Presence of topographic features (shelf-break, seamount) Breeding colony inﬂuence on central-place-foragers

1

Distance to colony (COLONY, km)

1 Extracted from the Environmental Research Division, Southwest Fisheries Science Center and US National Marine Fisheries Service (http://coastwatch.pfel.noaa.gov/coastwatch/CWBrowserWW360.jsp). 2 Oceanographic data ranges are based on the time spent data, the most extensive training dataset (n = 23 021 observations). 3 Spatial gradients were estimated as their proportional change (PC) within a surrounding 3 · 3 cell (0Æ75º · 0Æ75º) grid using a moving window as follows: PC = [(maximum value–minimum value) · 100] ⁄ (maximum value).

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 121–132

124 M. Louzao et al. Selecting predictors Prior to modelling, all environmental variables were standardized (Zuur, Ieno & Smith 2007). Strongly ‘correlated’ (|rs| > 0Æ5) predictors were identiﬁed by estimating all pair-wise Spearman rank correlation coeﬃcients (Table S2, Supporting information). Then, we removed those explaining less deviance by comparing Akaike Information Criteria values (AICs) of generalized linear mixed models (GLMMs) with only one predictor to avoid colinearity and related problems with parameter estimations (Zuur et al. 2007). This approach led to the removal of diﬀerent predictors depending on the habitat index and breeding stage considered (indicated in Table 2).

Habitat models Once ‘non-correlated’ environmental variables were identiﬁed, GLMMs were ﬁtted for all possible linear combinations of predictors based on the lmer function (lme4 package; Pinheiro & Bates 2000). For each breeding stage, the (log-transformed) percentage of timespent per unit area was ﬁtted with a Gaussian error distribution (identity link), whereas the two binomial dependent variables (‘foraging ⁄ not foraging’ and ‘feeding ⁄ not feeding’) were ﬁtted with a binomial error distribution (logit link). We only included the ‘individual identity’ as a random term in order to account for individual eﬀects, although ‘year’ and ⁄ or ‘sex’ eﬀects were also tested (to account for inter-annual variability in sampling eﬀort and sex-related foraging ground location) but AIC values did not improve (decrease).

cation: >0Æ9 excellent, 0Æ9–0Æ8 good, 0Æ8–0Æ7 reasonable, 0Æ7–0Æ6 poor and 0Æ6–0Æ5 unsuccessful (Swets 1988). We applied a cross-validation procedure using two diﬀerent approaches: (1) an independent dataset for time spent and foraging patterns in order to assess the predictive performance of averaged models (built with data from Crozet) in predicting distribution patterns of birds from Kerguelen and (2) bootstrapping the original data for feeding patterns (no independent dataset) which provides an alternative approach for evaluating the model with the original data (Guisan & Zimmermann 2000; McAlpine et al. 2008). Working on two spatially distinct groups (Crozet and Kerguelen) allowed us to assess the model performance to predict in diﬀerent conditions ⁄ areas. Although both populations diﬀered slightly in their habitat availability, a previous study showed no evidence of diﬀerence in habitat selection (Pinaud & Weimerskirch 2007). During 1000 simulations, models within the 95% conﬁdence set were ﬁtted to 70% of the test dataset and the modelling output was then used to predict distribution patterns of the remaining 30%. Then, the C-index was estimated for each simulation (up to 1000) and the mean, upper and lower 95% conﬁdence interval (CI) of the C-index were used as a cross-validation measure of the predictive performance of the models (McAlpine et al. 2008). If the lower 95% CI limit did not include the 0Æ5 value, there was evidence that averaged habitat models were able to accurately predict beyond training dataset.

Mapping predictions Model selection and inference Within the Information Theoretic Approach, we evaluated competing models by assessing their relative support (based on AIC and Akaike weight) in relation to observed data, rather than using the best single model approach (Burnham & Anderson 2002). When the model with lowest AIC value has an Akaike weight value lower than 0Æ9, a model averaging procedure might be more appropriate to account for parameter uncertainty (Burnham & Anderson 2002). Therefore, we constructed a 95% conﬁdence set of models where the sum of Akaike weights was >0Æ95 (Louzao et al. 2009). Accordingly, averaged coeﬃcients were estimated from the 95% conﬁdence set of models containing that variable, as well as variance estimator in order to assess the precision of the estimates (Burnham & Anderson 2002).

Model checking In parallel, we checked the distribution and spatial autocorrelation of the residuals, but no signiﬁcant evidence was found (results not presented) and we did not consider any spatial autocorrelation structure in GLMMs.

Model evaluation To assess the predictive performance of habitat models, we estimated the concordance index (C-index) of the averaged models estimated with the Hmisc package (Harrell 2001). This index is equivalent to the area under the Receiver Operating Characteristics curve (AVC) and probably the most useful measurement for distribution modelling (Vaughan & Ormerod 2005), since it allowed the comparison of the predictive performance of all three models (time spent: continuous, foraging and feeding: binomial; Harrell 2001). The C-index varies from 0Æ5 to 1 with the following model predictive performance classiﬁ-

We mapped the predicted spatial distribution of the three habitat use ecological indices. Predictors were extracted yearly for each month (January–March: incubation and April: brooding) from 1998 to 2008 and we applied the 95% conﬁdence set of models to predict the seascape and both foraging and feeding habitats. The 11-years predictions were averaged for each month and the standard deviation (SD) was used as a measure of predicted habitat stability (low and high SD representing stable and unstable habitats, respectively). Habitat predictions for albatrosses were represented as continuous surface probabilities. Once time spent, foraging, and feeding predictions were mapped, we further analysed their relationship with water mass distribution. We ﬁrst extracted the mean predictions of habitat models for January and April (as representative of the incubation and brooding, respectively) over the 11-years study period (1998–2008) around the mean distribution ranges of both breeding periods (see Results). Secondly, mean sea surface temperature (SST) values averaged over the 11-years study period corresponding to January and April were extracted as previously and matched to habitat predictions cell by cell. Finally, SST values were aggregated in relation to main water masses described in the southern Indian Ocean (Park et al. 2002): Subtropical Zone (SST > 13 C), sub-Antarctic zone (9º < SST < 13 C), Polar Frontal Zone (4º < SST < 9 C) and Antarctic zone (SST < 4 C).

Results Wandering albatrosses travelled up to thousands of kilometres from the colony during incubation (mean: 1176 km, range: 61–3381) and brooding (mean 450 km, range: 88–1800) (Fig. 1). During the brooding period, adults made shorter trips (mean: 72 h, range: 5–286) compared to the incubation period

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 121–132

Models within 95% conﬁdence set

Mean C-index (CI 95%) 70% test dataset 30% test dataset

C-index averaged model Mean (± SD) Cross-validation

Averaged model Intercept Chlorophyll a (CHL)* CHL gradient (CHLG)* Sea Surface Temperature (SST)* SST gradient (SSTG)* Sea Level Anomaly (SLA)* SLA gradient (SLAG)* Wind Speed (WIND)* Bathymetry (BAT) BAT gradient (BATG) Distance to colony (COLONY) 0Æ078 0Æ009 0Æ009 0Æ031

± 0Æ032 ± 0Æ01

± 0Æ009 ± 0Æ01

± ± ± ±

3

0Æ651 (0Æ644–0Æ657) 0Æ647 (0Æ632–0Æ662)

0Æ680 ± 0Æ004

)1Æ209 )0Æ037 0Æ042 0Æ156 NI )0Æ048 )0Æ028 NI NI 0Æ166 0Æ029 0Æ006 0Æ027 0Æ007 0Æ006 0Æ006

± 0Æ035

± ± ± ± ±

± 0Æ104

13

0Æ638 (0Æ621–0Æ654) 0Æ618 (0Æ580–0Æ658)

0Æ701 ± 0Æ01

0Æ201 NI 0Æ023 0Æ155 )0Æ01 )0Æ009 )0Æ01 NI NI 0Æ181 NI

0Æ043 0Æ043 0Æ009 0Æ087 0Æ012 0Æ007 0Æ008

± 0Æ358 ± 0Æ235

± ± ± ± ± ± ±

14

0Æ590 (0Æ565–0Æ614) 0Æ554 (0Æ503–0Æ605)

0Æ693 ± 0Æ015

0Æ013 )0Æ197 0Æ051 0Æ292 )0Æ061 )0Æ026 )0Æ017 NI NI 0Æ599 0Æ485

Incubation

Incubation

Brooding

Foraging Habitat

Seascape

0Æ296 0Æ043 0Æ064 0Æ088

± 0Æ061

± ± ± ±

± 0Æ063 ± 0Æ065

10

0Æ777 (0Æ717–0Æ852) 0Æ717 (0Æ600–0Æ828)

0Æ601 ± 0Æ034

0Æ011 )0Æ201 NI 0Æ545 0Æ131 )0Æ131 )0Æ217 NI NI 0Æ225 NI

Brooding

± 0Æ052

± 0Æ051 ± 0Æ101 ± 0Æ017

± 0Æ143 ± 0Æ051

27

0Æ620 (0Æ563–0Æ710) 0Æ559 (0Æ453–0Æ656)

0Æ600 ± 0Æ077

)0Æ777 0Æ207 NI )0Æ161 )0Æ282 0Æ071 NI NI NI 0Æ21 NI

Incubation

Feeding Habitat

0Æ165 0Æ01 0Æ011 0Æ014 0Æ118 0Æ018 0Æ187

± 0Æ076

± ± ± ± ± ± ±

59

0Æ736 (0Æ682–0Æ783) 0Æ656 (0Æ590–0Æ725)

0Æ731 ± 0Æ049

)1Æ112 )0Æ004 0Æ039 0Æ033 )0Æ324 )0Æ051 )0Æ407 NI NI 0Æ269 NI

Brooding

Table 2. Averaged (± SE) coeﬃcients of habitat models (seascape, foraging and feeding habitats) developed for the incubation and brooding stages for wandering albatross. Also, the Concordance index (C-index) is shown for averaged models and cross-validation, as well as the number of models within the 95% conﬁdence set. The most important predictors for each habitat model are denoted in bold. NI: indicate non included predictors (i.e. ‘Correlated’ environmental variables). Dynamic variables are marked with an asterisk

Seascape modelling of a marine predator 125

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 121–132

126 M. Louzao et al. (mean: 251 h, range: 63–559). The tracking data revealed that a core marine area surrounding Crozet that extended to the west (1400 km from west to east and 300 km from north to south) was exploited by 6–113 wandering albatrosses between the South Subtropical Front and the Subantarctic Front (green area surrounding Crozet, Fig. S2, Supporting information). In addition, at least two birds visited a large buﬀer area surrounding Crozet and Kerguelen (yellow area, Fig. S2, Supporting information).

MODELLING THE SEASCAPE–TIME SPENT PER AREA

A total of 149 foraging trips from Crozet were analysed corresponding to 23 021 observations (86% to incubation, Fig. 1a). For incubation, the model with the lowest AIC included all ‘non-correlated’ environmental variables, whereas during brooding only SST and bathymetric gradient (BATG) were included. Since these two models yielded an Akaike weight of 0Æ85 and 0Æ240, respectively, and some model uncertainty was present (3 and 13 models in the 95% conﬁdence set for incubation and brooding, respectively), a model averaging approach was applied. For both breeding stages, averaged models showed a reasonable model performance (C-index values) and SST and BATG had the strongest positive eﬀect on time spent (based on the sign of averaged coeﬃcients, Table 2). These results indicated that wandering albatrosses from Crozet spent more time in areas of high bathymetric variability such as shelf-breaks and seamounts during both breeding stages, which corresponded to subtropical waters during incubation and to sub-Antarctic waters during brooding (Kruskall–Wallis test for incubation H3,8227 = 2408Æ36, P < 0Æ001; Kruskall– Wallis test for brooding H2,1216 = 192Æ08, P < 0Æ001, Fig. 2). Overall, our model predictions showed an increasing trend of time spent from Antarctic to subtropical waters. Model predictions matched observed patterns within the range of wandering albatrosses and identiﬁed pelagic areas beyond the training dataset where albatrosses might spend more time in the southern Indian Ocean: areas surrounding sub-Antarctic oceanic breeding colonies, the Southwest Indian Ridge (N-NW sector of Crozet), and seamounts such as Del Cano Rise (E of Prince Edward Islands) and the seamount complex of Ob and Llena south of Crozet and Kerguelen (bluer areas, Figs 1a and 3a). Those areas were consistently identiﬁed as important during incubation (low SD in predictions), whereas the marine area associated with the retroﬂection of the Agulhas Current was especially variable, in response to the highly dynamic nature of this frontal system (which was not detected during brooding; Fig. 3a). The C-index values of the cross-validation indicated that averaged habitat models from Crozet had the ability to predict time spent patterns from Kerguelen (Table 2).

MODELLING THE FORAGING HABITAT–ZONES OF INCREASED FORAGING INTENSITY

Foraging behaviour was recorded for 122 birds from Crozet totalling 5716 observations (82% corresponding to incuba-

tion; Fig. 1b). Chlorophylla (CHL), SST and BATG were common to both models with the lowest AIC, in addition to distance to the colony (COLONY) during incubation and SSTG, sea level anomaly (SLA) and SLA gradient (SLAG) during brooding. Since these two models yielded an Akaike weight of 0Æ136 and 0Æ395 and model uncertainty was present (14 and 10 models in the 95% conﬁdence set for incubation and brooding, respectively; Table 2), models were averaged. During incubation, averaged models showed a reasonable model performance (but poorer predictions for brooding) and BATG and SST (in addition to COLONY during incubation) had the stronger positive eﬀect on albatrosses foraging probability (Table 2). Thus, foraging might occur within the same oceanographic context where albatrosses spent more time, but was more constrained by the distance to the colony (compared Fig. 1a,b). Also, all wandering albatrosses searched for prey more intensively in subtropical waters during incubation, but with no preference between sub-Antarctic or Polar Frontal Zone waters (shorter trips) during brooding (Kruskall–Wallis test for incubation H3,8227 = 1274Æ70, P < 0Æ001; Kruskall–Wallis test for brooding H2,1216 = 186Æ95, P < 0Æ001, Fig. 2). Predicted foraging habitat not only matched the spatial location of predicted seascapes, but also the stability of pelagic habitats: stable around main breeding colonies, seamounts and mountain ranges, in contrast to the dynamic habitat related to the Agulhas retroﬂection current (compared Fig. 3a,b). The cross-validation output indicated that foraging patterns from Kerguelen were better predicted during brooding than during the incubation period (Table 2).

MODELLING THE FEH–PREY CAPTURE

A total of 34 independent breeders were equipped with STT yielding 754 observations (31% during incubation; Table S1, Supporting information) within the shelf area of Crozet (green areas, Fig. 1c). Models with the lowest AIC values included SSTG for both breeding stages, in addition to SLAG and BATG during brooding. Since these models had an Akaike weight of only 0Æ068 and 0Æ148 for incubation and brooding, models were averaged and the 95% conﬁdence set was comprised by 27 and 59 models, respectively (Table 2). Averaged feeding habitat models yielded poor predictions for incubation, independently of the very reasonable predictions for brooding. SSTG and BATG (in addition to SLAG during brooding) had the strongest negative and positive eﬀect on albatross feeding probability during both breeding stages, respectively. Therefore, feeding occurred in areas of high bathymetric variability characterized by low SST variability within the Polar Frontal Zone during both breeding stages (Kruskall–Wallis test for incubation H3,8227 = 462Æ10, P < 0Æ001; Kruskall–Wallis test for brooding H2,1216 = 336Æ37, P < 0Æ001, Figs 2 and S3d, Supporting information). Again, averaged predictions highlighted similar key feeding areas of the wandering albatross for both breeding stages (Fig. 3c). Marine areas of high mesoscale activity yielded the strongest variability in predictions at 40ºS related to the retroﬂection of the Agulhas Front to the west

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 121–132

Seascape modelling of a marine predator 127 (a)

Incubation 10° E

20° E

30° E

40° E

50° E

Brooding 60° E

70° E

80° E

90° E

Africa

–30° s

10° E

20° E

30° E

40° E

50° E

60° E

70° E

80° E

90° E

Africa

–30° s

–30° s

–30° s

–40° s

–40° s

–40° s

–40° s

–50° s

–50° s

–50° s

–50° s

–60° s

–60° s

–60° s Mean time per cent/cell

–60° s Mean time per cent/cell

11·0

–70° s

(b)

36·6

0

Antarctica

0·0

10° E

20° E

30° E

40° E

10° E

20° E

30° E

40° E

50° E

500 km

–70° s

–70° s

70° E

80° E

90° E

10° E

20° E

30° E

40° E

60° E

70° E

80° E

90° E

10° E

20° E

30° E

40° E

Incubation 50° E

0

Antarctica

0·0

60° E

50° E

500 km

60° E

70° E

80° E

90° E

60° E

70° E

80° E

90° E

–70° s

Brooding

Africa

50° E

Africa

–30° s

–30° s

–30° s

–30° s

–40° s

–40° s

–40° s

–40° s

–50° s

–50° s

–50° s

–50° s

–60° s

–60° s

–60° s

Foraging birds/cell 0

–70° s

0

Antarctica

7–9

10° E

–60° s Foraging birds/cell 0

1 2–3 4–6

20° E

30° E

40° E

50° E

60° E

70° E

500 km

80° E

–70° s

1 2–3 4–6

–70° s

90° E

20° E

30° E

Incubation

(c)

30° E

40° E

0

Antarctica

10° E

40° E

50° E

60° E

70° E

80° E

500 km

–70° s

90° E

Brooding 50° E

30° E

60° E

40° E

50° E

60° E

–40° s

–40° s

–40° s

–40° s

–50° s

–50° s

–50° s

–50° s

–60° s

–60° s

Feeding birds/cell 0

Feeding birds/cell

–60° s

0

0

30° E

1

500 km

0–1

0

2

40° E

50° E

60° E

30° E

–60° s

2–3

500 km

4–7

40° E

50° E

60° E

Fig. 1. Observed patterns of (a) time spent (mean percentage of time spent by all birds visiting each cell), (b) foraging (number of foraging birds in each cell) and (c) feeding patterns (number of feeding birds in each cell) of Crozet during incubation and brooding periods (1998–2008). Mean position of the main frontal systems (Agulhas Front: AF in blue; the North and South Subtropical fronts: NSTF in green and SSTF in dark blue, respectively; Subantarctic Polar front: SAF in fuchsia and Polar Front: PF in rose, respectively) are identiﬁed (Belkin & Gordon 1996). Studied breeding colonies are represented by red triangles.

and the South Subtropical Front to the east (see Figs 3c and S3). Predictions yielded poor values for incubation, but reasonable predictions for brooding (see cross-validation results in Table 2).

Discussion Given their extraordinary movement capacities (Weimerskirch et al. 2000), the habitat modelling of this wide-ranging animal presents an exceptional and challenging opportunity to consider the dynamic nature of pelagic ecosystems. By combining three diﬀerent quantitative ecological indices, this study oﬀers a comprehensive ecological picture of marine habitat use of a

top predator and the ﬁrst integrative spatially explicit ecological study of the wandering albatross with important conservation implications. Previous studies have focused on one of the applied methodologies (e.g. time spent–Hyrenbach, Ferna´ndez & Anderson 2002; FPT – Pinaud & Weimerskirch 2005), but none of them have developed an integrative modelling procedure.

LINKING ALBATROSS ECOLOGY AND AN INTEGRATIVE HABITAT MODELLING PROCEDURE

Wandering albatrosses have a distinct foraging strategy based on extensive movement at low cost, by travelling constantly

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 121–132

128 M. Louzao et al. Incubation

Brooding 2·6

Predicted percentage of time spent

Predicted percentage of time spent

0·6

0·5

0·4

0·3

2·2 2·0 1·8 1·6 1·4 1·2 1·0

Foraging probability

0·8

0·7 0·6 0·5 0·4 0·3

0·7 0·6 0·5 0·4 0·3

0·2

0·2

0·6

0·6

Feeding probability

Feeding probability

Foraging probability

0·2 0·8

2·4

0·5 0·4 0·3 0·2

0·5 0·4 0·3

0·2 Antarctic

Polar

Sub-Antarctic Subtropical

Antarctic

Water masses

and quickly to maximize their probability of encountering isolated prey patches (Weimerskirch et al. 2000, 2005). Breeding albatrosses are typical central-place foragers that adjust their movements at various scales, from ocean basin (thousand of kilometres) to ﬁne scales (100 m) in response to the oceanographic context (Fritz, Said & Weimerskirch 2003), i.e. the patchy and dispersed distribution of their main prey, squids (Weimerskirch, Gault and Cherel 2005, 2007). Moreover, the species uses two diﬀerent foraging tactics: ‘foraging in ﬂight’ and ‘sit and wait’, the former being the main and more eﬃcient tactic (Weimerskirch et al. 2007). Taken together, this evidence underscores the fact that the species occupies an unique niche in the marine environment (Weimerskirch, Gault and Cherel 2005). Our predictive habitat models reasonably matched the observed distribution patterns and described albatross multiscale habitat use with the expected hierarchical arrangement of marine resource distribution: small scale feeding habitat nested within larger scale habitats (Fauchald & Tveraa 2003). Results for time spent and foraging patterns were similar, indicating that albatrosses foraged over topographic features in subtropical waters, and nested within the wider seascape due to the constraint imposed by the colony eﬀect. Prey searching behaviour along the shelf-break (Weimerskirch et al. 2007) conﬁrms the importance of this topographic feature as foraging ground of the species. Since foraging behaviour spatially overlaps with areas where albatrosses spent more time, one could hypothe-

Polar

Sub-Antarctic

Water masses

Fig. 2. Predictions (median, 25–75% interquartile range, non-outlier range, and outliers) of all three predicted habitats in relation to water masses during both breeding stages (subtropical: Subtropical Zone, sub-Antarctic: sub-Antarctic zone, polar: Polar Frontal Zone, and Antarctic: Antarctic zone; Park et al. 2002). Note the diﬀerent y-axis scale of predicted percentage of time spent for both periods.

size that they invest time in areas where they search for prey. The hierarchical system is usually used to describe prey patch arrangement, which supports the use of the wandering albatross as an ecological indicator species (Cherel & Weimerskirch 1999). One of the main contributions of this study to the ecology and conservation of pelagic top predators is the capacity to predict key pelagic habitats in the near future or when data are not available for speciﬁc years (Guisan & Zimmermann 2000). Kernel analysis is the traditional method for the identiﬁcation of key habitats, drawing probability contours of equal density from the tracking locations (Worton 1989), but it does not allow the prediction of potential habitats. Thus, developing habitat suitability models can overcome these limitations and improve our current knowledge on species distribution. However, this approach must be considered cautiously because it assumes that the habitat associations deﬁned for speciﬁc conditions can be extrapolated to non-sampled areas. Our cross-validation exercise showed the general ability of averaged models to predict the distribution patterns of wandering albatrosses in two closely related populations, but also revealed the wide distribution range of the species and high inter-individual diﬀerences (i.e. relative low C-index in some cases). Species with less restricted ecological requirements and ⁄ or distribution ranges can be modelled less accurately than species with more restricted requirements ⁄ ranges (Segurado & Arau´jo 2004).

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 121–132

Seascape modelling of a marine predator 129

PELAGIC HABITATS OF WIDE-RANGING ANIMALS

Understanding the movement patterns and habitat associations of these highly mobile organisms is critical to the eﬀective monitoring and implementation of conservation measures. Our tracking study highlighted that the core distribution area of breeding albatrosses was restricted on average to ca. 1200 and 450 km during incubation and brooding, respectively. This core area was limited by both the South Subtropical Front and the Sub-Antarctic Front in the north and south, respectively (Weimerskirch et al. 2005). Although the pelagic habitats of wandering albatrosses changed between incubation and brooding, the foraging strategies and total mass of prey captured did not, which indicated that birds used similar foraging strategies, and that prey availability was probably similar in both stages (Weimerskirch, Gault and Cherel 2005). Overall, both static and dynamics variables were involved in explaining multi-scale habitat use of the species, with important implications for delineation of MPAs. Within the breeding range, three topographical features were identiﬁed as key pelagic habitats: marine areas surrounding sub-Antarctic oceanic breeding islands (Prince Edward Islands, Crozet, Kerguelen and Heard), seamounts (Ob and Llena south Crozet; Del Cano Rise between Crozet and Prince Edward) and submarine mountain ranges (Southwest Indian Ridge). Two oceanographic variables (SST and BATG) directly drive key habitats of wandering albatrosses, which were consistently identiﬁed across breeding stages and years. This could be explained by the fact that these topographical features promote the conﬂuence of the main frontal systems not only in the Crozet and Kerguelen Basins, but also in the Southwest Indian Ridge (Park et al. 2002). These convergence zones are areas of strong mesoscale activity, where primary productivity is higher and intense upwelling ⁄ downwelling processes occur (Nel et al. 2001; Park et al. 2002). Within these convergence zones, high aggregations of prey occur, and they have been identiﬁed previously as high conservation areas for seabirds (Harris et al. 2007). In the Southern Ocean, the association of top predators with frontal systems inﬂuenced by bathymetric features has been well documented and seems to be a major feature driving top predator distribution within this biogeographic area (Nel et al. 2001). Finally, the position of the retroﬂection of the Agulhas current showed the strongest variability in predictions, suggesting substantial inter-annual variability of the habitat preference in this area. All these results underline the importance of considering the dynamic nature of pelagic habitats when planning conservation initiatives to protect highly mobile animals.

IMPLICATIONS FOR CONSERVATION AND MANAGEMENT

The identiﬁcation of key marine areas (e.g. foraging areas and migration corridors) might be a priority action for the conservation of a given species or community (Hooker & Gerber 2004). However, a precondition for this is to establish standard guidelines in order to similarly collect and analyse distribution

data that allows comparison of key habitats of diﬀerent oceanic species on a global scale. Based on our integrative habitat modelling procedure (especially model evaluation) and the ecological context of each habitat use, the wider time spent per unit area might be the most useful ecological index for conservation purposes, since it integrates diﬀerent habitat uses. Once the location and extent of key top predator habitat areas are identiﬁed, it is necessary to ensure their legal protection (Louzao et al. 2009). Currently, diﬀerent international conservation agreements force governments to protect 20–30% of all marine habitats under their jurisdiction (i.e. Economic Exclusive Zones – EEZs) by 2012 (Lombard et al. 2007). However, seabirds are not subjected to any speciﬁc spatially explicit conservation initiative in the southern Indian Ocean and most of the current regulations are associated with the management of ﬁsheries (Game et al. 2009). The distribution of wandering albatrosses breeding in Crozet and Kerguelen overlaps with three diﬀerent RFMOs: the Indian Ocean Tuna Commission (IOTC), the Commission for the Conservation of Southern Blueﬁn Tuna and the Commission for the Conservation of Antarctic Marine Living Resources (CCAMLR), as well as the Illegal, Unregulated and Unreported ﬁshing (IUU, Fig. 3). Wandering albatrosses spend most of their life travelling across the Southern Ocean encountering diﬀerent RFMOs and IUU ﬂeets (Weimerskirch et al. 1997), the latter representing the major portion of the annual seabird bycatch (BirdLife International 2009). Our results provide two key conservation and management tools: (i) the ﬁrst map to support development of a prospective network of priority conservation zones (marine Important Bird Areas) across the southern Indian Ocean; and (ii) habitat suitability models for tracking changes in the distribution of a given species. By plotting habitat predictions, a prospective network of pelagic sites can be planned that could encompass marine areas surrounding sub-Antarctic oceanic breeding islands (Prince Edward Islands, Crozet, Kerguelen and Heard), seamounts (Ob and Llena south Crozet; Del Cano Rise between Crozet and Prince Edward) and submarine mountain ranges (Southwest Indian Ridge). Our results match well with the recently created Prince Edward Islands MPA (the ﬁrst MPA within the distribution of the species; Lombard et al. 2007). This protection could be expanded to other key pelagic habitats. A network of important protected sites could be used not only to support conservation of top predators (with important implications for ﬁshery and ecosystem management), but also the underlying biodiversity associated with key habitats of top predators in the pelagic realm (e.g. Louzao et al. 2006, 2009). These pelagic habitats are important within certain sea surface temperature conditions and their boundaries might be tracked based on water mass distribution (i.e. SST), which is already an essential tool of the spatial management of certain ﬁsheries (e.g. Hobday & Hartmann 2006). The association of wandering albatrosses with dynamic oceanographic variables reﬂects the importance of dynamic, spatially explicit conservation initiatives for oceanic top predators. Finally, present habitat

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 121–132

130 M. Louzao et al. Incubation

(a) 40° E

50° E

60° E

Brooding 70° E

80° E

40° E

50° E

60° E

70° E

80° E

–40° S

–40° S –40° S

–40° S

–50° S

–50° S –50° S

–50° S

TS mean Jan 1·12

TS mean Apr 4·66

0·11

–60° S

40° E

50° E

60° E

70° E

–60° S –60° S 80° E

40° E

50° E

60° E

70° E

80° E

0·8

40° E

50° E

60° E

70° E

–60° S 80° E

40° E

50° E

60° E

70° E

80° E

–40° S

–40° S –40° S

–40° S

–50° S

–50° S –50° S

–50° S

TS sd Apr 0·15

TS sd Jan 0·2 0

–60° S

–60° S –60° S 40° E

50° E

40° E

50° E

60° E

70° E

0

80° E

–60° S 40° E

50° E

40° E

50° E

Incubation

(b)

60° E

60° E

70° E

80° E

Brooding 70° E

80° E

60° E

70° E

80° E

–40° S

–40° S –40° S

–40° S

–50° S

–50° S –50° S

–50° S

FoH mean Apr 1

FoH mean Jan 1

–60° S

0·04

–60° S –60° S 40° E 40° E

50° E 50° E

60° E 60° E

70° E 70° E

0

80° E 80° E

–60° S 40° E

50° E

60° E

70° E

80° E

40° E

50° E

60° E

70° E

80° E

–40° S

–40° S –40° S

–40° S

–50° S

–50° S –50° S

–50° S

FoH sd Apr 0·3

FoH sd Jan 0·35

–60° S

0

–60° S –60° S 40° E

50° E

40° E

50° E

60° E

0

–60° S

70° E

80° E

40° E

50° E

70° E

80° E

40° E

50° E

Incubation

(c)

60° E

60° E

70° E

80° E

70° E

80° E

Brooding 60° E

–40° S

–40° S –40° S

–40° S

–50° S

–50° S –50° S

–50° S

FeH mean Jan 1

–60° S

FeH mean Apr 0·7

0·16

–60° S –60° S

0·08

–60° S

40° E

50° E

60° E

70° E

80° E

40° E

50° E

60° E

70° E

80° E

40° E

50° E

60° E

70° E

80° E

40° E

50° E

60° E

70° E

80° E

–40° S

–40° S –40° S

–40° S

–50° S

–50° S –50° S

–50° S

FeH sd Apr 0·18

FeH sd Jan 0·22

–60° S

0

–60° S –60° S 40° E

50° E

60° E

70° E

80° E

0

–60° S 40° E

50° E

60° E

70° E

80° E

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 121–132

Seascape modelling of a marine predator 131

Fig. 3. Mean (± SD) predictions of habitat models (upper and lower panel, respectively) for (a) time spent (TS), (b) foraging habitat (FoH) and (c) feeding habitat (FeH) during incubation (January) and brooding (April) stages (left and right panel, respectively) over the 11-years study period (1998–2008). We also highlighted the main management units: EEZs (lines of black dots) are shown surrounding main oceanic islands (Prince Edward Islands, Crozet, Kerguelen, Heard Island and Amsterdam, from west to east) and Regional Fisheries Management Organizations (IOTC: Indian Ocean Tuna Commission, CCAMLR: Commission for the Conservation of Antarctic Marine Living Resources). Mean position of the main frontal systems (Agulhas Front: AF in orange; the North and South Subtropical fronts: NSTF in rose and SSTF in red, respectively; Subantarctic Polar front: SAF in yellow and Polar Front: PF in white, respectively) are identiﬁed (Belkin & Gordon 1996). Studied breeding colonies are represented by red triangles.

suitability models are useful to predict trends in key pelagic areas of wandering albatrosses during the next decades in a global change scenario. In a wider context, our integrative modelling approach is applicable to a wide range of species for habitat conservation in both marine and terrestrial environments.

Acknowledgements We wish to thank all participants of the ﬁeldwork, especially S.A. Shaﬀer, T. Guionnet, J. Martin, G. Mabille, F. Bailleul, V. Lecomte, A. Jaeger and M. Berlincourt. Statistical valuable advice was provided by A. Goarant, P. Inchausti and V. Rolland. C. Barbraud, F. Bartumeus, B. Martin, J. Rodrı´ guez and J.D. Anado´n provided useful comments to the earlier version of the manuscript. The Ethic Committee of the Institut Polaire-Paul Emile Victor (IPEV, programme no. 109) approved the ﬁeld procedure. The oceanographic data were extracted thanks to the Environmental Research Division, Southwest Fisheries Science Center and US National Marine Fisheries Service. The study was ﬁnanced by IPEV (programme no. 109) and the Prince Albert II de Monaco Foundation, and is part of the Program ANR Biodiversite´ 2005-REMIGE and ANR Biodiversite´ 2007-GLIDES. M.L. was funded by a postdoctoral contract of the Spanish Ministry of Education and Science (Ref. EX-2007-1148) and Marie Curie Individual Fellowship (PIEF-GA-2008-220063). We acknowledge C. Freitas, the Associate Editor and Editor for their help with bringing the manuscript to its ﬁnal state.

References Belkin, I.M. & Gordon, A.L. (1996) Southern Ocean fronts from the Greenwich meridian to Tasmania. Journal of Geophysical Research, 101, 3675– 3696. Benhamou, S. (1992) Eﬃciency of area-concentrated searching behavior in a continuous patchy environment. Journal of Theoretical Biology, 159, 67– 81. BirdLife International (2009) Species factsheet: Diomedea exulans. Downloaded from http://www.birdlife.org on 4 ⁄ 6 ⁄ 2009. Burnham, K.P. & Anderson, D.R. (2002) Model Selection and Multimodel Inference: A Practical Information-Theoretic Approach. Springer-Verlag, New York, NY. Cherel, Y. & Weimerskirch, H. (1999) Spawning cycle of onychoteuthid squids in the southern Indian Ocean: new information from seabird predators. Marine Ecology Progress Series, 188, 93–104. Delord, K., Besson, D., Barbraud, C. & Weimerskirch, H. (2008) Population trends in a community of large Procellariiforms of Indian Ocean: potential eﬀects of environment and ﬁsheries interactions. Biological Conservation, 141, 1840–1856. Fauchald, P. & Tveraa, T. (2003) Using ﬁrst-passage time in the analysis of area-restricted search and habitat selection. Ecology, 84, 282–288. Fritz, H., Said, S. & Weimerskirch, H. (2003) Scale-dependent hierarchical adjustments of movement patterns in a long-range foraging seabird. Proceedings of the Royal Society of London, Series B: Biological Sciences, 270, 1143–1148. Furness, R.W. & Camphuysen, C.J. (1997) Seabirds as monitors of the marine environment. ICES Journal of Marine Science, 54, 726–737. Game, E.T., Grantham, H.S., Hobday, A.J., Pressey, R.L., Lombard, A.T., Beckley, L.E., Gjerde, K., Bustamante, R., Possingham, H.P. & Richardson, A.J. (2009) Pelagic protected areas: the missing dimension in ocean conservation. Trends in Ecology and Evolution, 24, 360–369.

Guisan, A. & Zimmermann, N.E. (2000) Predictive habitat distribution models in ecology. Ecological Modelling, 135, 147–186. Harrell, F.E. (2001) Regression Modeling Strategies: with Applications to Linear Models, Logistic Regression, and Survival Analysis. Springer, New York, NY. Harris, J., Haward, M., Jabour, J. & Woehler, E.J. (2007) A new approach to selecting marine protected areas (MPAs) in the Southern Ocean. Antarctic Science, 19, 189–194. Heithaus, M.R., Frid, A., Wirsing, A.J. & Worm, B. (2008) Predicting ecological consequences of marine top predator declines. Trends in Ecology and Evolution, 23, 202–210. Hobday, A.J. & Hartmann, K. (2006) Near real-time spatial management based on habitat predictions for a longline bycatch species. Fisheries Management and Ecology, 13, 365–380. Hooker, S.K. & Gerber, L.R. (2004) Marine reserves as a tool for ecosystembased management: the potential importance of megafauna. BioScience, 54, 27–39. Hyrenbach, K.D., Ferna´ndez, P. & Anderson, D.J. (2002) Oceanographic habitats of two sympatric North Paciﬁc albatrosses during the breeding season. Marine Ecology Progress Series, 233, 283–301. Lombard, A.T., Reyers, B., Schonegevel, L.Y., Cooper, J., Smith-Adao, L.B., Nel, D.C., Froneman, P.W., Ansorge, I.J., Bester, M.N. & Tosh, C.A. (2007) Conserving pattern and process in the Southern Ocean: designing a marine protected area for the Prince Edward Islands. Antarctic Science, 19, 39–54. Louzao, M., Be´cares, J., Rodrı´ guez, B., Hyrenbach, K.D., Ruiz, A. & Arcos, J.M. (2009) Combining vessel-based surveys and tracking data to identify key marine areas for seabirds. Marine Ecology Progress Series, 391, 183–197. Louzao, M., Hyrenbach, K.D., Arcos, J.M., Abello´, P., Gil de Sola, L. & Oro, D. (2006) Oceanographic habitat of a critically endangered mediterranean Procellariiform: implications for the design of marine protected areas. Ecological Applications, 16, 1683–1695. McAlpine, C.A., Rhodes, J.R., Bowen, M.E., Lunney, D., Callaghan, J.G., Mitchell, D.L. & Possingham, H.P. (2008) Can multiscale models of species’ distribution be generalized from region to region? A case study of the koala. Journal of Applied Ecology, 45, 558–567. McConnell, B.J., Chambers, C. & Fedak, M.A. (1992) Foraging ecology of the southern elephant seals in relation to bathymetry and productivity of the Southern Ocean. Antarctic Science, 4, 393–398. Myers, R.A. & Worm, B. (2003) Rapid worldwide depletion of predatory ﬁsh communities. Nature, 423, 280–283. Nel, D.C., Lutjeharms, J.R.E., Pakhomov, E.A., Ansorge, I.J., Ryan, P.G. & Klages, N.T.W. (2001) Exploitation of mesoscale oceanographic features by grey-headed albatross Thalassarche chrysostoma in the southern Indian Ocean. Marine Ecology Progress Series, 217, 15–26. Norse, E.A. & Crowder, L.B. (2005) Marine Conservation Biology: the Science of Maintaining the Sea’s Biodiversity. Island Press, Washington, DC. Park, Y.H., Pollard, R.T., Read, J.F. & Leboucher, V. (2002) A quasi-synoptic view of the frontal circulation in the Crozet Basin during the Antares-4 cruise. Deep-Sea Research Part II, 49, 1823–1842. Phillips, R.A., Xavier, J.C. & Croxall, J.P. (2003) Eﬀects of satellite transmitters on albatrosses and petrels. The Auk, 120, 1082–1090. Pinaud, D. & Weimerskirch, H. (2005) Scale-dependent habitat use in a long-ranging central place predator. Journal of Animal Ecology, 74, 852– 863. Pinaud, D. & Weimerskirch, H. (2007) At-sea distribution and scale-dependent foraging behaviour of petrels and albatrosses: a comparative study. Ecology, 76, 9–19. Pinheiro, J.C. & Bates, D.M. (2000) Mixed-Eﬀects Models in S and S-PLUS. Springer, New York, NY.

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 121–132

132 M. Louzao et al. R Development Core Team (2008) R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing, Vienna, Austria. ISBN 3-900051-07-0, URL http://www.R-project.org Segurado, P. & Arau´jo, M.B. (2004) An evaluation of methods for modelling species distributions. Journal of Biogeography, 31, 1555–1568. Swets, J.A. (1988) Measuring the accuracy of diagnostic systems. Science, 240, 1285–1293. Vaughan, I.P. & Ormerod, S.J. (2005) The continuing challenges of testing species distribution models. Journal of Applied Ecology, 42, 720–730. Weimerskirch, H., Brothers, N. & Jouventin, P. (1997) Population dynamics of wandering albatross Diomedea exulans and Amsterdam albatross D. amsterdamensis in the Indian Ocean and their relationships with long-line ﬁsheries: conservation implications. Biological Conservation, 79, 257–270. Weimerskirch, H., Gault, A. & Cherel, Y. (2005) Prey distribution and patchiness: factors in foraging success and eﬃciency of Wandering Albatrosses. Ecology, 86, 2611–2622. Weimerskirch, H., Guionnet, T., Martin, J., Shaﬀer, S.A. & Costa, D.P. (2000) Fast and fuel eﬃcient? Optimal use of wind by ﬂying albatrosses. Proceedings of the Royal Society B: Biological Sciences, 267, 1869–1874. Weimerskirch, H., Inchausti, P., Guinet, C. & Barbraud, C. (2003) Trends in bird and seal populations as indicators of a system shift in the Southern Ocean. Antarctic Science, 15, 249–256. Weimerskirch, H., Pinaud, D., Pawlowski, F. & Bost, C.-A. (2007) Does prey capture induce area-restricted search? A ﬁne-scale study using GPS in a marine top predator, the Wandering Albatross. The American Naturalist, 170, 734–743. Weimerskirch, H., Salamolard, M., Sarrazin, F. & Jouventin, P. (1993) Foraging strategy of wandering albatrosses through the breeding season: a study using satellite telemetry. The Auk, 110, 325–342. Worton, B.J. (1989) Kernel methods for estimating the utilization distribution in home-range studies. Ecology, 70, 164–168. Zacharias, M.A. & Roﬀ, J.C. (2001) Use of focal species in marine conservation and management: a review and critique. Aquatic Conservation: Marine and Freshwater Ecosystems, 11, 59–76.

Zuur, A.F., Ieno, E.N. & Smith, G.M. (2007) Analysing Ecological Data. Springer, New York, NY. Received 3 May 2010; accepted 27 October 2010 Handling Editor: Melanie Austen

Supporting Information Additional Supporting Information may be found in the online version of this article: Table S1. Summary of the tracking procedures. Table S2. Summary of pair-wise correlation analysis of environmental variables. Figure S1. Workﬂow of the habitat modelling procedure. Figure S2. Number of birds visiting each 0Æ25- cell from Crozet and Kerguelen Islands. Figure S3. Mean (± SD) of diﬀerent oceanographic variables averaged over the 11-year study period (1998–2008). As a service to our authors and readers, this journal provides supporting information supplied by the authors. Such materials may be re-organized for online delivery, but are not copy-edited or typeset. Technical support issues arising from supporting information (other than missing ﬁles) should be addressed to the authors.

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 121–132

Journal of Applied Ecology 2011, 48, 25–34

doi: 10.1111/j.1365-2664.2010.01911.x

Using presence-only and presence–absence data to estimate the current and potential distributions of established invasive species Andrew M. Gormley1, David M. Forsyth1*, Peter Grifﬁoen2, Michael Lindeman1, David S.L. Ramsey1, Michael P. Scroggie1 and Luke Woodford1 1

Arthur Rylah Institute for Environmental Research, Department of Sustainability and Environment, 123 Brown Street, Heidelberg 3084, Australia; and 2Peter Griﬃoen Consulting, 3 ⁄ 391 Upper Heidelberg Road, Ivanhoe 3079, Australia

Summary 1. Predicting the current and potential distributions of established invasive species is critical for evaluating management options, but methods for diﬀerentiating these distributions have received little attention. In particular, there is uncertainty among invasive species managers about the value of information from incidental sightings compared to data from designed ﬁeld surveys. This study compares the two approaches, and develops a unifying framework, using the case of invasive sambar deer Cervus unicolor in Victoria, Australia. 2. We ﬁrst used 391 incidental sightings of sambar deer and 12 biophysical variables to construct a presence-only habitat suitability model using Maxent. We then used that model to stratify ﬁeld sampling, with proportionately greater sampling of cells with high predicted habitat suitability. Field sampling, consisting of faecal pellet surveys, sign surveys and camera trapping, was conducted in 80 4-km2 grid cells. A Bayesian state-space occupancy model was used to predict probability of suitable habitat from the ﬁeld data. 3. The Maxent and occupancy models predicted similar spatial distributions of habitat suitability for sambar deer in Victoria and there was a strong positive correlation between the rankings of cells by the two approaches. The congruence of the two models suggests that any spatial and detection biases in the presence-only data were relatively unimportant in our study. 4. We predicted the extent of suitable habitat from the occupancy model using a threshold that gave a false negative error rate of 0Æ05. The current distribution was the suitable habitat within a kernel that had a 99Æ5% chance of including the presence locations pooled from incidental sightings and ﬁeld surveys: the potential distribution was suitable habitat outside that kernel. Several discrete areas of potential distribution were identiﬁed as priorities for surveillance monitoring with the aim of detecting and managing incursions of sambar deer. 5. Synthesis and applications. Our framework enables managers to robustly estimate the current and potential distributions of established invasive species using either presence-only and ⁄ or presence– absence data. Managers can then focus control and ⁄ or containment actions within the current distribution and establish surveillance monitoring to detect incursions within the potential distribution. Key-words: camera trap, Cervus unicolor, detection probability, habitat suitability models, kernel smoothing, Maxent, occupancy, sambar deer, state-space modelling, Victoria Introduction Invasive species can have important detrimental environmental, economic and social impacts (Mack et al. 2000; Pimentel

*Correspondence author. E-mail: [email protected] Re-use of this article is permitted in accordance with the Terms and Conditions set out at http://wileyonlinelibrary.com/onlineopen# OnlineOpen_Terms

et al. 2005; Lodge et al. 2006) and there is much interest in managing these populations (Myers et al. 2000; Hulme 2006; Lodge et al. 2006). Predicting and quantifying the current and potential distributions of established invasive species is a critical step in evaluating management options: for example, control and eradication eﬀorts should focus on the current distribution, containment should focus on the interface between the current and potential distributions, and incursion monitoring should focus on the potential distribution (Myers

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society

26 A. M. Gormley et al. et al. 2000; Leung et al. 2005; Lodge et al. 2006). However, methods for diﬀerentiating the current and potential distributions of established invasive species have received little attention. Since the distributions of many established invasive plants and animals may be much smaller than their maximum distributions (e.g. for recent and ⁄ or slow invaders; Ward 2007; Phillips, Chipperﬁeld & Kearney 2008), methods are required for discriminating suitable habitat that is occupied from that which is unoccupied. The ﬁrst step is to distinguish ‘suitable’ from ‘unsuitable’ habitat, and two general approaches have been used to do this. Presence-only data (e.g. from atlas records) and biophysical variables can be used to ﬁt predictive ‘niche-based models’ of distribution using numerous methods (Elith et al. 2006). Models of presence-only data produce spatially explicit suitability surfaces that represent habitat suitability (Elith et al. 2006). However, presence-only modelling based on incidental sightings may be subject to major spatial and detection biases (Gu & Swihart 2004; Wintle, Elith & Potts 2005; Arau´jo & Guisan 2006). An alternative approach is to conduct ﬁeld surveys in a way that accounts for potential spatial biases (by using a known sampling design; Thompson, White & Gowan 1998) and imperfect detection of the species of interest (MacKenzie et al. 2002): modelling such data estimates the probability of occupancy (MacKenzie et al. 2006). Occupancy models constructed from observed presence– absence data also predict habitat suitability when projected across the landscape. A threshold is needed to distinguish the output of habitat suitability models (from presence-only and presence–absence models) into ‘suitable’ and ‘unsuitable’ habitat (Liu et al. 2005). The second step is to estimate which areas of predicted suitable habitat are ‘occupied’ (‘current distribution’) and ‘unoccupied’ (‘potential distribution’). Point pattern analysis (‘kernel smoothing’; Diggle 2003; Hengl et al. 2009) is a particularly promising method for estimating the current distributions of established invasive species because it can use presences pooled from presence-only and presence–absence data. The aim of this study is to estimate the current and potential distributions of invasive sambar deer C. unicolor Kerr in the state of Victoria, Australia. We ﬁrst construct habitat suitability models for sambar deer using presence-only data from incidental sightings and presence–absence data from a designed ﬁeld survey. After comparing the predictions of the two methods we then use threshold occupancy and kernel smoothing methods to delineate the current and potential distributions of sambar deer in Victoria.

Fig. 1. Sambar deer photographed at a camera trap during our presence–absence ﬁeld survey.

(Menkhorst 1995; Bentley 1998). There is concern about the continued range expansion of sambar deer in Victoria because of their potential negative impacts on native biodiversity (Department of Sustainability and Environment 2009a) and agriculture (Lindeman & Forsyth 2008). We subdivided Victoria into 56 764 cells of 2 · 2 km. A cell size of 4 km2 was chosen because it approximated estimates of sambar deer home range size in invasive populations (Lewis et al. 1990; Fraser & Nugent 2005) and was a practical unit size for conducting ﬁeld surveys (sensu Karanth et al. 2009).

PREDICTOR VARIABLES

Thirty biophysical variables were identiﬁed from the literature as potentially important predictors of sambar deer distribution and abundance in Victoria (review in Forsyth et al. 2009; see Appendix S1, Supporting information). The variables were generated, for each of the 4 km2 cells, from GIS layers supplied by the Victorian State Government’s Corporate Geospatial Data Library (O’Brien 2004). Prior to model building we assessed the strength of Pearson’s correlation coeﬃcients between pairs of variables: if variables were highly correlated (rp > 0Æ7) then one of the variables was removed from the set. A ﬁnal set of 12 candidate variables remained for model building (Table 1).

HABITAT SUITABILITY MODEL FROM INCIDENTAL SIGHTINGS

Materials and methods STUDY AREA AND SPECIES

The state of Victoria (237 629 km2), south-eastern mainland Australia, was our study area. Sambar deer (Fig. 1), sourced from Sri Lanka, India and the Philippines, were introduced at four sites in Victoria during the 1860s and have subsequently expanded their distribution to the north, north-east and south-east of Victoria

Incidental sightings Presence-only data for sambar deer were obtained from the Atlas of Victorian Wildlife Database (AVWD) containing data from 1974 to 2007 (Department of Sustainability and Environment 2009b). The AVWD is a geographically registered relational database of incidental sightings of fauna by government agency staﬀ and the public. Sambar deer observations consisted of a date, latitude ⁄ longitude and a measure of locational precision. We only used records (n = 391)

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 25–34

Modelling invasive species distributions 27 Table 1. Biophysical covariates used in our models of sambar deer distribution in Victoria Covariate

Description

Units

Grass Gullies Homogeneity NativeGrassShrub

Amount of grassland Number of gullies Similarity of land use Amount of native grassland ⁄ shrubland Annual precipitation Seasonal diﬀerence in precipitation Distance from nearest road Annual mean temperature Minimum annual temperature Distance from water Amount of wet sclerophyll forest Average slope

% (0–100) Count 0–100 % (0–100)

AnnualPrecip SeasonalPrecip RoadDistance MeanTemp MinimumTemp WaterDistance WetForestCover Slope

mm mm m C · 10 C · 10 m % (0–100) (0–90)

with a location precision of less than 1 km in our analyses, and binned these observations into the 4 km2 cells.

Maxent model Incidental sightings of sambar deer were modelled using Maxent 3Æ2Æ19 (Phillips, Anderson & Schapire 2006), a machine learning approach based on maximum entropy. Maxent has been shown to perform as well as, or better than, other methods for modelling presenceonly data (Elith et al. 2006). Maxent uses the presence-only data and a user-deﬁned number (in our case, 10 000) of randomly selected points (‘pseudo-absences’) and combines these with the biophysical covariates to construct an index of habitat suitability for each cell ranging from 0 (least suitable habitat) to 1 (most suitable habitat). We allowed linear and ⁄ or quadratic relationships between the index of habitat suitability and each covariate (Phillips & Dudı´ k 2008). The relative contribution of each covariate to the Maxent distribution, and the relationship between each variable and the predicted index of habitat suitability, was also calculated (Phillips, Anderson & Schapire 2006). Model performance was assessed by determining how well the model discriminates between unsuitable and suitable habitat over a range of thresholds (Fielding & Bell 1997). For any threshold of habitat suitability index, presence locations are either correctly classiﬁed as being in suitable habitat (‘true positives’) or misclassiﬁed as being in unsuitable habitat (‘false negatives’). Similarly, absence data are either correctly classiﬁed as being in unsuitable habitat (‘true negatives’) or misclassiﬁed as being in suitable habitat (‘false positives’). Because false positives cannot be estimated for presence-only data, Maxent estimates the fractional predicted area (FPA), which is the proportion of cells predicted to have suitable habitat for the species (Phillips, Anderson & Schapire 2006). To assess performance of the Maxent model we plotted a receiver operating characteristic curve, which compares the model sensitivity (true positives) against 1 – speciﬁcity (false positives) over the entire range of thresholds (Fielding & Bell 1997). For presence-only modelling, the area under this curve (AUC) represents the probability that a randomly chosen presence site will be ranked as more suitable than a randomly chosen pseudoabsence site. A model that performs no better than random will have an AUC of 0Æ5 whereas a model with perfect discrimination would have an AUC of 1. An additional measure of model performance is the regularized training gain (‘Gain’), which describes how much

better the Maxent distribution ﬁts the presence data compared to a uniform distribution. The exponential of the Gain is a measure of how many times higher the sample likelihood is compared to a random cell (Yost et al. 2008).

OCCUPANCY MODEL FROM FIELD SURVEYS

Sampling methodology Our aim here was to develop a model of potential distribution of sambar deer based on the relationship between presence ⁄ absence data and biophysical variables. Since resources were available to conduct ﬁeld surveys in only 80 cells, it was desirable to spend more eﬀort sampling areas of high-habitat suitability (sensu McDonald 2004). We therefore allocated a greater proportion of sites to areas of higher habitat suitability estimated by our Maxent model. Sixty cells were randomly selected and retained with probability equal to the corresponding habitat suitability index of that cell. The other 20 cells were selected entirely at random.

Field surveys We used three survey methods to estimate occupancy rates of sambar deer between July 2008 and April 2009. First, we assessed presence ⁄ absence of sambar deer faecal pellets along three randomly located transects in each of the 80 cells using the method described in Forsyth et al. (2007). Brieﬂy, we navigated to the start of each 150-m transect using a hand-held GPS and counted the number of intact pellets in circular plots of 1 m radius spaced at 5 m intervals (i.e. 30 plots per transect). The presence and absence of pellets in cell i and transect j was indicated by Yij = 1 and 0, respectively, for j = 1–3. Secondly, we searched for signs of sambar deer along a 400 m transect in each of the 80 cells. The sign transect was subjectively located by ﬁeld staﬀ to maximize the detection of deer (e.g. along a trail or watercourse likely to be used by sambar deer; Bentley 1998). Any of the following signs of sambar deer seen along the survey route were recorded: sightings of live or dead deer, tree-rubbings, tracks, cast antlers, wallows and faecal pellets. The presence ⁄ absence of sambar deer sign on transects was denoted as Yi4 = 1 and 0, respectively. Thirdly, in a randomly selected 40 of the 80 cells we set two heat-inmotion remote cameras along the sign survey route. Cameras [TrailMAC Digital (Trail Sense Engineering, Middletown, DE, USA) and PixController DigitalEyeTM (PixController Inc., Export, PA, USA)] were set, unbaited, for 21 days. The presence ⁄ absence of images of sambar deer on the cameras was indicated by Yi5 = 1 and 0, respectively.

Statistical model The presence–absence data were modelled using a Bayesian statespace occupancy model consisting of a process model and an observation model (Royle & Ke´ry 2007). The process model describes the true occupancy at each site and the observation model described the observation process conditional on the true occupancy state of each site. For each site i, the true occupancy state zi was modelled as a random variate from a Bernoulli distribution with probability wi equal to the probability of occupancy at site i: zi Bernðwi Þ:

eqn 1

The probability of occupancy at site i was modelled as a function of one or more biophysical covariates, denoted in general as:

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 25–34

28 A. M. Gormley et al. logitðwi Þ ¼ bXi :

eqn 2

For each survey method there is a probability of detection given that the site is occupied. The observed presences ⁄ absences were modelled as: Yij Bernðzi pj Þ;

eqn 3

where Yij is the observed presence ⁄ absence at site i for survey j, and pj is the detection probability for that survey (recall j = 1–3 denotes faecal pellet transects, j = 4 sign surveys and j = 5 camera surveys). If a site is unoccupied then zi = 0 and Yij = 0 is observed with probability 1. If a site is occupied then zi = 1 and Yij = 1 is observed with probability pj, and Yij = 0 with probability 1–pj. Assuming independence of the surveys, the overall probability of detection, conditional on presence, p*, from k surveys is: p ¼ 1

k Y

ð1 pj Þ:

Spearman’s correlation coeﬃcient (rs). We also compared the spatial output from each of the two models following rescaling as deciles.

DEFINING ‘SUITABLE HABITAT’

The probabilities of suitable habitat for each cell from the occupancy model were delineated into suitable and unsuitable habitat using a threshold. The choice of a threshold depends on whether one wishes to minimize false negative or false positive errors, or balance them in some other way (Liu et al. 2005). A threshold that is too high will result in a high number of false negative errors and low number of false positives, leading to a higher proportion of the study area being classiﬁed as unsuitable when it is suitable. Conversely, a threshold that is too low will result in lower false negative and higher false positive error rates, leading to a relatively high proportion of the study area being classiﬁed as suitable when it is not (Ward 2007). We selected a threshold by setting the false negative error rate at 0Æ05.

eqn 4

j¼1

ESTIMATING CURRENT DISTRIBUTION

The same twelve biophysical variables used in the Maxent model (Table 1) were used as potential covariates in the occupancy model.

Parameter estimation Models were ﬁtted using WinBUGS 1Æ4Æ3 (Lunn et al. 2000). Prior distributions of Normal(0, 100) were used for the covariate coeﬃcient parameters b. All covariates were standardized to a mean of 0 and standard deviation of 1. Prior distributions of Beta(1, 1) were used for the detection probabilities pj for each of the three survey methods. Three replicate Markov-chains were constructed using diﬀerent initial values to check for convergence. The chains were run for 1000 iterations to tune the algorithm and ensure convergence. The ‘burnin’ samples were discarded and the algorithm run for a further 20 000 samples before the three chains were combined to provide a sample of 60 000 values from the joint posterior distribution of each parameter. Our WinBUGS code is provided in Appendix S2 (Supporting information).

Model selection and averaging

We delimited the current distribution of sambar deer, conditional on areas of suitable habitat, by two-dimensional kernel smoothing the pooled sambar deer presence data (i.e. using both incidental sightings and ﬁeld survey data). The function ‘kde2d’ in R package ‘MASS’ version 7Æ2 (Venables & Ripley 2002) with a bivariate Gaussian kernel was used to estimate the density surface. This method has been widely used to estimate the utilization distribution of individual animals based on location data. The resulting density surface can be thought of as indicating the relative intensity (i.e. points per unit area) of species presence records for any location within the study area. The bandwidth for smoothing was calculated using the ‘solve-the-equation’ method of Sheather & Jones (1991) and we deﬁned a percentage level that ensured 99Æ5% of the presence records were included in the current distribution. Kernel smoothing was applied conditional on the cell being classiﬁed as suitable habitat (see above).

Results HABITAT SUITABILITY MODEL USING INCIDENTAL SIGHTINGS

We calculated the deviance information criterion (DIC) value for each model following Spiegelhalter et al. (2002). We ﬁrst evaluated models containing the 12 biophysical variables individually and in pairs. This was followed by models with combinations of three and four variables, using variables that had consistently lower DIC values as individuals and pairs. Rather than selecting a single ‘best’ model we used model averaging (Burnham & Anderson 2002; McCarthy 2007) to predict sambar deer occupancy. Model weights (w) were summed from largest to the smallest, and the models with a cumulative sum of 0Æ9 used as the model averaging set (Burnham & Anderson 2002). The resulting model-averaged predictive equation was applied to each 4 km2 grid cell in our study area to produce a map of predicted probability of suitable habitat for sambar deer.

COMPARING PREDICTIONS OF THE MAXENT AND OCCUPANCY MODELS

Although Maxent and occupancy models both give results on the unit scale, these are not directly comparable. We therefore compared the predictions (i.e. cell rankings from lowest to highest) of the presenceonly Maxent model and the presence–absence occupancy model using

The 391 sightings of sambar deer occurred in 322 cells (Fig. 2a). The AUC (0Æ942) and Gain (1Æ61) values indicate that the Maxent model of the incidental sightings (Table 2) had a high discriminatory ability (Fig. 2b). The plot of false negative errors and FPA (Fig. 3a) showed little overlap, further conﬁrming the usefulness of the Maxent model. Three variables (WetForestCover, AnnualPrecip and Gullies) had a relative contribution of 83% to the Maxent model and when used on their own showed a reasonable ﬁt to the data in terms of Gain (Fig. 3b). Conversely, the variables SeasonalPrecip and RoadDistance achieved little Gain when used alone (Fig. 3b). Results from omitting each variable whilst including all others showed that no one variable contained a substantial amount of information that was not contained in the other variables. Three other variables (MeanTemp, MinimumTemp and Slope) showed a reasonable to ﬁt to the data in terms of Gain when used alone despite having small relative contributions to the model built using all variables. The probability

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 25–34

Modelling invasive species distributions 29 of presence increased with increasing WetForestCover and AnnualPrecip, but had a concave-up relationship with Gullies and MeanTemp (Fig. S1, Supporting information).

DETECTION PROBABILITIES AND OCCUPANCY MODEL

Sambar deer were detected in 40 of the 80 sampled cells (Fig. 2c). They were detected on one or more faecal pellet transects in 26 cells, on sign transects in 35 cells and at camera traps (Fig. 1) in 10 of the 40 cells sampled with that method. The highest probability of detection, conditional on presence, was associated with sign surveys, followed by transects and cameras (Table 3 and Fig. 4). The overall probability of detection from three faecal pellet transects was 0Æ736 (95% CI = 0Æ628–0Æ832), and from two cameras was 0Æ507 (95% CI = 0Æ288–0Æ722). The site-level detection probability, combining all methods (eqn 4), was 0Æ932 (95% CI = 0Æ851–0Æ974) at sites where only faecal pellet transects and sign surveys were used and 0Æ967 (95% CI = 0Æ896–0Æ992) at sites where all three methods were used. The variables Gullies, Homogeneity, AnnualPrecip, AnnualTemp, MinimumTemp, WetForestCover and Slope had consistently lower DIC values relative to the other covariates when used alone and when included in pairs. Subsequently, all three-way and four-way combinations of these seven covariates were modelled. A total of 148 models with various

(a)

combinations of covariates were evaluated. The best model (i.e. lowest DIC) included the variables Gullies, AnnualPrecip, AnnualTemp, and MinimumTemp (Table S1, Supporting information). However, there were many models with similar DIC values: the 17 highest ranked models had a cumulative model selection weight of 0Æ906. The variables MinimumTemp and AnnualPrecip were included in 17 and 16 of the reduced set of 17 models used for model averaging, respectively (Tables S1, Supporting information and 3). There was a strong negative eﬀect of MinimumTemp, and a strong positive eﬀect of AnnualPrecip, on probability of occupancy (Table 3). The eﬀects of the other variables included in the model-averaged occupancy model were more equivocal (Table 3).

COMPARISON OF THE MAXENT AND OCCUPANCY MODELS

There was a strong positive correlation (rs = 0Æ89) between the rankings of cells by the two methods (Fig. 5): cells with a higher habitat suitability index from Maxent had higher probabilities of suitable habitat from the occupancy model. Both models indicated that areas of highest habitat suitability for sambar deer were in eastern Victoria and that the northern, western and southern areas of the state were of lowest suitability (Fig. 2b,d). There were several large patches of moderate habitat suitability in central and southern Victoria.

(b) Urban areas Incidental sightings

0

Habitat suitability index

Urban areas

1 0·8 0·6 0·4 0·2 0

50 100

200

0

km

50 100

200

km

(c)

(d) Urban areas Present Absent

Probability of suitable habitat

Urban areas

1 0·8 0·6 0·4 0·2 0

0

50 100 km

200

0

50 100

200

km

Fig. 2. Habitat suitability and occupancy models for sambar deer in Victoria estimated from incidental sighting (presence-only) and ﬁeld survey (presence–absence) data, respectively. (a) Incidental sightings used in the Maxent habitat suitability model. (b) Predictions of the Maxent habitat suitability model. (c) The 804-km2 cells in which presence–absence ﬁeld surveys were undertaken. (d) Predictions of the occupancy model. 2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 25–34

30 A. M. Gormley et al. Table 2. Relative contribution of variables to the Maxent model of incidental sightings of sambar deer in Victoria

Variable

Relative contribution

WetForestCover AnnualPrecip Gullies WaterDistance MinimumTemp RoadDistance SeasonalPrecip NativeGrassShrub Homogeneity Grass MeanTemp Slope

65Æ4 9Æ4 8Æ4 6Æ7 5Æ5 1Æ7 1Æ2 0Æ6 0Æ5 0Æ3 0Æ2 0Æ1

Comparison of each cell’s deciles showed several coastal areas were ranked higher in the Maxent model than the occupancy model, although they were still ranked relatively low overall (Fig. S2, Supporting information).

(a)

(b)

CURRENT AND POTENTIAL DISTRIBUTIONS OF SAMBAR DEER IN VICTORIA

We used all presence-only data (i.e. including all incidental sighting records from 1974–2007 and our ﬁeld survey presences) to estimate current distribution. The target false negative rate of 0Æ05 was achieved at a threshold level of 0Æ40, which had a corresponding commission error rate of 0Æ225 (Fig. 6). The threshold value of 0Æ40 was therefore used to delineate between unsuitable and suitable sambar deer habitat. Using this threshold there are an estimated 58 340 km2 of suitable sambar deer habitat in Victoria. The 99Æ5% utilization distribution gave a current estimated distribution of 42 888 km2 (Fig. 7). Major areas of apparently suitable but unoccupied range outside the current distribution include the Great Otway National Park and Grampians National Park, both in western Victoria (Fig. 7).

Discussion We used presence-only (incidental sightings) and presence– absence (ﬁeld surveys) data to diﬀerentiate the current and potential distributions of invasive sambar deer in Victoria such that potentially important spatial and detection biases were minimized. We ﬁrst used incidental sightings to estimate a habitat suitability index. We then used the habitat suitability index to stratify our ﬁeld survey eﬀort and our ﬁeld surveys used methods that enabled imperfect detection to be accounted for in the estimated probability of suitable habitat. We then used a threshold to delineate the predictions of the occupancy model into suitable (i.e. potential distribution) and unsuitable habitat. Finally, we applied kernel smoothing to the pooled presence data (i.e. from both incidental sightings and ﬁeld surveys) to further delineate the suitable habitat into estimates of current and potential range. Our analyses indicated that

Fig. 3. Performance of the Maxent habitat suitability model of incidental sightings of sambar deer in Victoria. (a) False negative error rate (solid line) and FPA (dashed line) for all threshold values. (b) Regularized Training Gain, with variables ranked depending on the Gain from a model with only that variable.

sambar deer occupied c. 74% of suitable habitat (42 888 km2) in Victoria in 2008–2009 but that several large, discrete areas of potential range exist in western Victoria.

CONGRUENCE OF HABITAT SUITABILITY AND OCCUPANCY MODELS

Although the units of Maxent and occupancy models diﬀer, there was strong agreement between the relative rankings of the predictions of the two approaches for sambar deer in Victoria (Fig. 5). To our knowledge, this is the ﬁrst study to use independently collected presence–absence data to test the

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 25–34

Modelling invasive species distributions 31 Table 3. Model-averaged parameter estimates from occupancy models of sambar deer in Victoria. SD is the square root of the unconditional variance estimator. Importance is calculated for b coeﬃcients as the sum of the model weights for models containing that parameter Parameter

Mean

SD

2Æ5%

97Æ5%

Importance

a [Intercept] 1Æ062 1Æ172 )0Æ809 3Æ370 NA b [Gullies] 0Æ389 0Æ638 )0Æ081 1Æ119 0Æ37 b [Homogeneity] 0Æ197 0Æ409 )0Æ123 0Æ648 0Æ21 b [AnnualPrecip] 2Æ355 1Æ100 0Æ595 4Æ659 0Æ99 b [MeanTemp] 0Æ723 1Æ299 )0Æ549 1Æ970 0Æ40 b [MinimumTemp] )4Æ263 1Æ573 )7Æ699 )1Æ664 1Æ00 b [WetForestCover] )0Æ025 0Æ194 )0Æ336 0Æ296 0Æ21 b [Slope] 0Æ145 0Æ306 )0Æ051 0Æ586 0Æ26 p [Faecal Pellet 0Æ362 0Æ043 0Æ281 0Æ449 NA Transect] p [Sign] 0Æ746 0Æ068 0Æ605 0Æ870 NA p [Camera] 0Æ302 0Æ815 0Æ156 0Æ479 NA

Fig. 5. Scatter plot of cell ranks from the predictions of Maxent and occupancy models.

Fig. 4. Conditional probabilities of detection for our three ﬁeld survey methods. Cumulative probabilities are shown for one, two or three faecal pellet transects, and one and two camera traps. Vertical bars are 95% credible intervals.

predictions of a habitat suitability model constructed from presence-only data: previous comparisons have used pseudoabsences (e.g. Elith et al. 2006; Pearson et al. 2007). The congruence of the two models suggests that any spatial and detection biases in the presence-only data (Gu & Swihart 2004; Wintle, Elith & Potts 2005; Arau´jo & Guisan 2006) were unimportant in our case study. The combination of repeated surveys and multiple ﬁeld methods when collecting the presence– absence data resulted in a high cumulative detection probability and thus a very small probability of false negatives. However, such biases might be more important for a recently established invader with a small current range and ⁄ or few sightings, or when unmodelled processes constrain range expansion (Pearson et al. 2007). Furthermore, issues related to detectability are likely to be greater for rare and ⁄ or elusive

Fig. 6. False negative and false positive error rates from the occupancy model for all threshold values. The black dashed line indicates a 5% false negative error rate and the red lines indicate the resulting threshold between suitable and unsuitable habitat and the corresponding false positive error rate for that threshold.

species. MacKenzie et al. (2006) give an excellent summary of the logic for using occupancy estimated from designed ﬁeld surveys, rather than habitat suitability derived from incidental sightings, to estimate species distributions. We chose to use Maxent to model incidental sightings of sambar deer in Victoria, but many other methods are available for modelling presence-only data (Elith et al. 2006). In the absence of any presence-only data, expert opinion could be used to develop a habitat suitability model (e.g. Yamada et al.

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 25–34

32 A. M. Gormley et al. If multiple methods were not used, the rate of false negatives would have been much greater. These results highlight the need to carefully consider detection probability in the design of presence–absence surveys (MacKenzie et al. 2002, 2006; Royle & Ke´ry 2007).

Unoccupied range Occupied range Urban areas

ESTIMATING CURRENT AND POTENTIAL DISTRIBUTIONS

0

Grampians National Park

Great Otway National Park

50 100

200

km

Fig. 7. Occupied (orange) and unoccupied (green) ranges of sambar deer in Victoria from an occupancy model and kernel smoothing of presence locations (circles) from incidental sightings and ﬁeld surveys.

2003; Ray & Burgman 2006) for stratifying ﬁeld sampling. However, the ability of experts to extrapolate beyond their geographic area of expertise may be poor (Murray et al. 2009). An alternative to choosing among diﬀerent modelling approaches when estimating species distributions is to combine inferences using the ensemble model framework (Arau´jo & New 2007). Future studies could use that framework to combine the outputs of presence-only (e.g. Maxent) and presence–absence models.

STRATIFYING FIELD SAMPLING USING HABITAT SUITABILITY MODELS

Field surveys are expensive and occupancy rate will be estimated more precisely if proportionately more sampling is conducted in areas where a species is known or predicted to occur relative to areas where they have not previously been observed. We chose to randomly allocate 75% of our ﬁeld surveys to cells using an unequal probability sampling scheme according to the habitat suitability index and the remainder randomly to all cells. The choice of how to stratify ﬁeld sampling is determined by the goal of the study. The aim of our study was to estimate current and potential distribution. If the aim was to detect new incursions ⁄ range expansions then relatively more eﬀort should be placed in areas of lower habitat suitability. Further work is required to generate rules of thumb for the allocation of survey eﬀort based on habitat suitability maps, and adaptive sampling may be a useful approach (Thompson, White & Gowan 1998).

DETECTION PROBABILITIES

Although all three ﬁeld survey methods had detection probabilities <1 (Fig. 4), the use of multiple methods and spatial replication of two of those methods (faecal pellet transects and camera traps) reduced the overall probability of false negatives in sampled cells. Sambar deer are cryptic, being largely nocturnal and spending daylight hours in dense forest (Bentley 1998).

A key decision in estimating suitable and unsuitable habitat is the choice of threshold (Liu et al. 2005). We chose to use a threshold that gave a false negative error rate of 0Æ05 because we had reliable information on presences but due to imperfect detection there may have been some sites where deer were present but unobserved. For invasive species it may often be desirable to minimize the false negative error. In some cases the spatial predictions of probability of suitable habitat may be more useful to managers than the demarcation into suitable and unsuitable habitat. We used kernel smoothing to deﬁne the current range of sambar deer in Victoria. Kernel smoothing has previously been applied to the modelling of presence-only species distribution data (Hengl et al. 2009) but our innovation was to use the resulting distribution to delimit suitable habitat estimated from the occupancy model into occupied habitat (current range) and unoccupied habitat (potential range), a critical parameter in decision-making for invasive species (Hulme 2006). The kernel density estimator, and hence estimates of current distribution, can be particularly sensitive to the choice of smoothing parameter (Diggle 2003). As well as calculating the smoothing parameter using the robust method developed by Sheather & Jones (1991), we also used historical information on the range expansion of sambar deer in Victoria (Menkhorst 1995; Bentley 1998) to help us determine the ‘best’ model of occupied and unoccupied range.

MANAGEMENT APPLICATIONS

The draft management policy focuses on containing invasive sambar deer within their current distribution in Victoria (Department of Sustainability and Environment 2009a). Although we have shown that sambar deer occupy c. 74% of their potential range in Victoria, our analysis has identiﬁed several discrete areas of suitable habitat that sambar deer do not currently occupy (Fig. 7). The Great Otway National Park and Grampians National Park are both separated from occupied range by agricultural land that sambar deer are unlikely to disperse across (Downes 1983; Bentley 1998). However, illegal translocation to establish new populations of deer has commonly occurred in Australia (Moriarty 2004). Rapid eradication of new populations has been proposed as a priority management action for sambar deer in Victoria (Department of Sustainability and Environment 2009a) and establishing surveillance monitoring in areas of suitable but unoccupied habitat using our detection methods (Fig. 4) would enable such populations to be quickly detected and dealt with.

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 25–34

Modelling invasive species distributions 33

Conclusion Our framework enables managers to robustly estimate the current and potential distributions of established invasive species using either presence-only and ⁄ or presence–absence data, and could be applied to any plant or animal taxa. Invasive species managers can use this information to better target control and ⁄ or containment actions within the current distribution and establish surveillance monitoring to detect incursions within the potential distribution.

Acknowledgements This study was funded by the Department of Sustainability and Environment (Land Management Branch), the Department of Primary Industries (Invasive Plants and Animals Branch) and Parks Victoria. We thank D. MacKenzie and C. Liu for statistical discussions. R. Chick, K. Howard, M. Main, J. Reside, S. Coulson, A. Douglas, H. Snow, and N. Trikojus assisted with ﬁeld work. D. MacKenzie, C. Liu, G. Paroz, M. White, F. Huettmann, the editors and an anonymous reviewer made valuable comments on previous manuscript drafts.

References Arau´jo, M.B. & Guisan, A. (2006) Five (or so) challenges for species distribution modelling. Journal of Biogeography, 33, 1677–1688. Arau´jo, M.B. & New, M. (2007) Ensemble forecasting of species distributions. Trends in Ecology and Evolution, 22, 42–47. Bentley, A. (1998) An Introduction to the Deer of Australia. Australian Deer Research Foundation Ltd., Melbourne. Burnham, K.P. & Anderson, D.R. (2002) Model Selection and Multimodal Inference: A Practical Information-Theoretic Approach, 2nd edn. SpringerVerlag, New York. Department of Sustainability and Environment (2009a) Draft Flora and Fauna Guarantee Action Statement: Reduction in Biodiversity of Native Vegetation by Sambar Deer (Cervus unicolor). Department of Sustainability and Environment, Melbourne. Department of Sustainability and Environment. (2009b) Victorian Fauna Database. http://www.viridans.com/FISVFD/VFD1.HTM Accessed 21 March, 2009. Data current to May 2007. Diggle, P.J. (2003) Statistical Analysis of Spatial Point Patterns, 2nd edn. Arnold, London. Downes, M. (1983) The Forest Deer Project 1982. Volume II. Ecology and Hunting. Forests Commission, Melbourne. Elith, J., Graham, C., Anderson, R., Dudik, M., Ferrier, S., Guisan, A., Hijmans, R., Huettmann, F., Leathwick, J., Lehmann, A., Li, J., Lohmann, L., Loiselle, B., Manion, G., Moritz, C., Nakamura, M., Nakazawa, Y., Overton, J., Peterson, A.T., Phillips, S., Richardson, K., Scachetti-Pereia, R., Schapire, R., Soberon, J., Williams, S., Wisz, M. & Zimmermann, N. (2006) Novel methods improve prediction of species distributions from occurrence data. Ecography, 29, 129–151. Fielding, A.H. & Bell, J.F. (1997) A review of methods for the assessment of prediction errors in conservation presence ⁄ absence models. Environmental Conservation, 24, 38–49. Forsyth, D.M., Barker, R.J., Morriss, G. & Scroggie, M.P. (2007) Modeling the relationship between fecal pellet indices and deer density. Journal of Wildlife Management, 71, 964–970. Forsyth, D.M., McLeod, S.R., Scroggie, M.P. & White, M. (2009) Modelling the abundance of wildlife using ﬁeld surveys and GIS: non-native sambar deer (Cervus unicolor) in the Yarra Ranges, south-eastern Australia. Wildlife Research, 36, 231–241. Fraser, K.W. & Nugent, G.N. (2005) Sambar Deer. The Handbook of New Zealand Mammals (ed C.M. King), pp. 436–442. Oxford University Press, Melbourne. Gu, W.D. & Swihart, R.K. (2004) Absent or undetected? Eﬀects of non-detection of species occurrence on wildlife habitat models. Biological Conservation, 116, 195–203. Hengl, T., Sierdsema, H., Radovic´, A. & Dilo, A. (2009) Spatial prediction of species’ distributions from occurrence-only records: combining point pattern analysis, ENFA and regression-kriging. Ecological Modelling, 220, 3499– 3511.

Hulme, P.E. (2006) Beyond control: wider implications for the management of biological invasions. Journal of Applied Ecology, 43, 835–847. Karanth, K.K., Nichols, J.D., Hines, J.E., Karanth, K.U. & Christensen, N.L. (2009) Patterns and determinants of mammal species occurrence in India. Journal of Applied Ecology, 46, 1189–1200. Leung, B., Finnoﬀ, D., Shogren, J.F. & Lodge, D. (2005) Managing invasive species: rules of thumb for rapid assessment. Ecological Economics, 55, 24– 36. Lewis, J.C., Flynn, L.B., Marchinton, R.L., Shea, S.M. & Marchinton, E.M. (1990) Biology of sambar deer on St. Vincent National Wildlife Refuge, Florida. Tall Timbers Research Station Bulletin, 25, 1–107. Lindeman, M.J. & Forsyth, D.M. (2008) Agricultural Impacts of Wild Deer in Victoria. Arthur Rylah Institute for Environmental Research Technical Report Series No. 182. Department of Sustainability and Environment, Heidelberg, Victoria. Liu, C., Berry, P.M., Dawson, T.P. & Pearson, R.G. (2005) Selecting thresholds of occurrence in the prediction of species distributions. Ecography, 28, 385– 393. Lodge, D.M., Williams, S., MacIsaac, H.J., Hayes, K.R., Leung, B., Reichard, S., Mack, R.N., Moyle, P.B., Smith, M., Andow, D.A., Carlton, J.T. & McMichael, A. (2006) Biological invasions: recommendations for U.S. policy and management. Ecological Applications, 16, 2035–2054. Lunn, D.J., Thomas, A., Best, N. & Spiegelhalter, D. (2000) WinBUGS – a Bayesian modelling framework: concepts, structure, and extensibility. Statistics and Computing, 10, 325–337. Mack, R.N., Simberloﬀ, D., Lonsdale, W.M., Evans, H., Clout, M. & Bazzaz, F. (2000) Biotic invasions: causes, epidemiology, global consequences and control. Ecological Applications, 10, 689–710. MacKenzie, D.I., Lachman, G.B., Droege, S., Royle, J.A. & Langtimm, C.A. (2002) Estimating site occupancy rates when detection probabilities are less than one. Ecology, 83, 2248–2255. MacKenzie, D.I., Nichols, J.D., Royle, J.A., Pollock, K.H., Bailey, L.L. & Hines, J.E. (2006) Occupancy Estimation and Modeling: Inferring Patterns and Dynamics of Species Occurrence. Elsevier Academic Press, Burlington, MA. McCarthy, M.A. (2007) Bayesian Methods for Ecology. Cambridge University Press, Cambridge. McDonald, L.L. (2004) Sampling rare populations. Sampling Rare or Elusive Species (ed W.L. Thompson), pp. 11–43. Island Press, Washington. Menkhorst, P.W. (1995) Mammals of Victoria: Distribution, Ecology and Conservation. Oxford University Press, Melbourne. Moriarty, A. (2004) The liberation, distribution, abundance and management of wild deer in Australia. Wildlife Research, 31, 291–299. Murray, V.V., Goldizen, A.W., O’Leary, R.A., McAlpine, C.A., Possingham, H.P. & Low Choy, S. (2009) How useful is expert opinion for predicting the distribution of a species within and beyond the region of expertise? A case study using brush-tailed rock wallabies Petrogale penicillata. Journal of Applied Ecology, 46, 842–851. Myers, J.H., Simberloﬀ, D., Kuris, A.M. & Carey, J.R. (2000) Eradication revisited: dealing with exotic species. Trends in Ecology and Evolution, 15, 316–320. O’Brien, M. (2004) Corporate Geospatial Data Library Catalogue, 17th edn, January 2004. Department of Primary Industries and Department of Sustainability and Environment, Melbourne. Pearson, R.G., Raxworthy, C.J., Nakamura, M. & Peterson, A.T. (2007) Predicting species distributions from small numbers of occurrence records: a test case using cryptic geckos in Madagascar. Journal of Biogeography, 34, 102–117. Phillips, B.L., Chipperﬁeld, J.D. & Kearney, M.R. (2008) The toad ahead: challenges of modelling the range and spread of an invasive species. Wildlife Research, 35, 222–234. Phillips, S.J., Anderson, R.P. & Schapire, R.E. (2006) Maximum entropy modeling of species geographic distributions. Ecological Modelling, 190, 231–259. Phillips, S.J. & Dudı´ k, M. (2008) Modeling of species distributions with Maxent: new extensions and a comprehensive evaluation. Ecography, 31, 161–175. Pimentel, D., Zuniga, R. & Morrison, D. (2005) Update on the environmental and economic costs associated with alien-invasive species in the United States. Ecological Economics, 52, 273–288. Ray, N. & Burgman, M.A. (2006) Subjective uncertainties in habitat suitability maps. Ecological Modelling, 195, 172–186. Royle, J.A. & Ke´ry, M. (2007) A Bayesian state-space formulation of dynamic occupancy models. Ecology, 88, 1813–1823.

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 25–34

34 A. M. Gormley et al. Sheather, S.J. & Jones, M.C. (1991) A reliable data-based bandwidth selection method for kernel density estimation. Journal of the Royal Statistical Society Series B, 53, 683–690. Spiegelhalter, D.J., Best, N.G., Carlin, B.P. & Van der Linde, A. (2002) Bayesian measures of model complexity and ﬁt (with discussion). Journal of the Royal Statistical Society Series B, 64, 583–616. Thompson, W.L., White, G.C. & Gowan, C. (1998) Monitoring Vertebrate Populations. Academic Press, San Diego. Venables, W.N. & Ripley, B.D. (2002) Modern Applied Statistics with S, 4th edn. Springer, New York. Ward, D.F. (2007) Modelling the potential geographic distribution of invasive ant species in New Zealand. Biological Invasions, 9, 723–735. Wintle, B.A., Elith, J. & Potts, J. (2005) Fauna habitat modelling and mapping: a review and case study in the Lower Hunter Central Coast region of NSW. Austral Ecology, 30, 719–738. Yamada, K., Elith, J., McCarthy, M. & Zerger, A. (2003) Eliciting and integrating expert knowledge for wildlife habitat modelling. Ecological Modelling, 165, 251–264. Yost, A.C., Peterson, S.L., Gregg, M. & Miller, R. (2008) Predictive modeling and mapping of sage grouse (Centrocercus urophasianus) nesting habitat using Maximum Entropy and a long-term dataset from Southern Oregon. Ecological Informatics, 3, 375–386. Received 5 May 2010; accepted 2 November 2010 Handling Editor: Paul Lukacs

Appendix S1. The 30 biophysical variables considered as predictor variables of sambar deer habitat suitability and occupancy. Appendix S2. WinBUGS code used to ﬁt the occupancy model to presence–absence data from ﬁeld surveys. Table S1. Model selection summary for the 17 models of sambar deer occupancy used for model averaging. Fig. S1. Maxent output of the response of sambar deer to the four covariates that explained the most variation in the habitat suitability model of incidental sightings. Fig. S2. Output of the (a) Maxent habitat suitability and (b) occupancy models rescaled as deciles. As a service to our authors and readers, this journal provides supporting information supplied by the authors. Such materials may be re-organized for online delivery, but are not copy-edited or typeset. Technical support issues arising from supporting information (other than missing ﬁles) should be addressed to the authors.

Supporting Information Additional Supporting Information may be found in the online version of this article.

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 25–34

Journal of Applied Ecology 2011, 48, 199–209

doi: 10.1111/j.1365-2664.2010.01912.x

Response of raptors to a windfarm Julia C. Garvin*, Christopher S. Jennelle, David Drake and Steven M. Grodsky Department of Forest and Wildlife Ecology, University of Wisconsin – Madison, 1630 Linden Drive, Madison, WI 53706, USA

Summary 1. The global growth of wind energy has outpaced our assessment of possible impacts on wildlife. There is a pressing need for studies with pre- and post-construction data to determine whether wind facilities will have detrimental eﬀects on susceptible avian groups such as raptors. 2. A pre- and post-construction study was conducted to determine the impact of a windfarm on the abundance and behaviour of raptors in Wisconsin, USA. Variation in abundance and behaviour was examined both within and among years and relative to selected spatial, temporal and weather covariates. Raptor avoidance rates and indices of collision risk were calculated. 3. Raptor abundance post-construction was reduced by 47% compared to pre-construction levels. Flight behaviour varied by species, but most individuals remained at a distance of at least 100 m from turbines and above the height of the rotor zone. 4. Turkey vultures Cathartes aura and red-tailed hawks Buteo jamaicensis displayed high-risk ﬂight behaviours more often than all other raptor species, but also showed signs of avoidance. Red-tailed hawks were the only raptor species found dead beneath turbines during mortality searches. There were few observed mortalities and corrected mortality estimates were comparable to those from other windfarm studies. 5. Synthesis and applications. The decline in raptor abundance post-construction together with other lines of evidence suggests some displacement from the windfarm project area. While certain species may be at risk, ﬂight behaviour data and mortality estimates indicate that the majority of raptors may not be directly aﬀected by the presence of turbines. The avoidance rates recorded in this study should be used to improve collision risk models, and both current and future windfarms should investigate avoidance behaviour post-construction. Key-words: avoidance rates, birds, collision risk, displacement, ﬂight behaviour, wildlife, wind turbines

Introduction The growth of the global wind energy industry has outpaced our understanding of the possible impacts on wildlife, speciﬁcally birds and bats which may be most aﬀected. Of those studies which have been completed, most lack pre-construction data, thereby providing no context in which to place postconstruction ﬁndings. Additionally, current eﬀorts to model collision risk suﬀer from a dearth of information about avian avoidance rates that can bias estimates (Chamberlain et al. 2006). Lastly, much of the research on wildlife-impacts is restricted to ‘grey’ literature (de Lucas et al. 2008), and the availability of current information is often restricted by developers and utilities in order to protect their interests within a competitive industry. *Correspondence author. E-mail: [email protected]

Studies indicate that raptors are especially susceptible to negative impacts by windfarms (Rugge 2001; Howe, Evans & Wolf 2002; Barrios & Rodriguez 2004; Hoover & Morrison 2005; Percival 2005; Stewart, Pullin & Coles 2007; Kikuchi 2008; de Lucas et al. 2008; Smallwood, Rugge & Morrison 2009). Raptors are more likely to collide with turbine blades than many other avian species due to their morphology and foraging behaviour (e.g. heavy wing loading, focus on distant prey; Janss 2000; Kikuchi 2008). Furthermore, research has shown that raptors forage, perch and ﬂy within 50 m of wind turbines disproportionately more often than expected by chance alone, with individuals often perching on turbine towers (Orloﬀ & Flannery 1992; Barrios & Rodriguez 2004; Smallwood & Thelander 2004). Compounding the problem, raptors occur at relatively low densities and most are long-lived with low reproductive output, making them especially susceptible to additive mortality (Kikuchi 2008).

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society

200 J. C. Garvin et al. There are three main threats to birds posed by windfarms: risk of collision, disturbance and habitat loss (reviewed in Langston & Pullan 2003; Percival 2005). While most research has focused on collision risk, disturbance due to windfarm construction, and habitat loss and fragmentation caused by turbine access roads may lead to displacement of resident raptors, as well as other species (Madders & Whitﬁeld 2006). Such displacement may be driven by reductions in the availability of nest-sites or areas for foraging and other activities (Madders & Whitﬁeld 2006). While there is equivocal evidence of windfarm-induced displacement of raptors [e.g. golden eagle Aquila chrysaetos and northern harrier Circus cyaneus (also known as hen harrier), see Pearce-Higgins et al. 2009], a lack of standardized protocols makes comparison between species and studies diﬃcult (reviewed in both Drewitt & Langston 2006; Madders & Whitﬁeld 2006). We conducted a pre- and post-construction study to evaluate potential impacts of a 129-MW windfarm in southeast Wisconsin, USA on the abundance (used as an index of raptor activity) and behaviour of raptors within the project area. While each wind facility is somewhat unique, the agricultural setting of this windfarm made it widely applicable to many current and planned windfarms in the USA and around the world, especially as developers are pressured to avoid building in areas with high densities of sensitive species. We recorded behavioural observations to improve collective knowledge about how raptors in ﬂight respond to turbines (e.g. avoidance or attraction), as well as to generate observed avoidance rates. A concurrent mortality study provided information on collision rates, and, subsequently, estimation of avoidance rates (extrapolated from rates of mortality or non-avoidance). We accounted for important covariates that are often neglected, such as weather conditions which modulate collision risk (reviewed in Drewitt & Langston 2008), detectability (Robbins 1981) and ﬂight height (Shamoun-Baranes et al. 2006). This study is uniquely capable of assessing displacement, avoidance and mortality rates. As such, the speciﬁc aims of our study were to determine (i) whether raptors were being displaced from within the windfarm, and if so, which species were most vulnerable; (ii) the proportion of raptors displaying avoidance behaviours as they approached a turbine; (iii) the relative risk of collision for all raptors as a group and for individual raptor species; and (iv) potential correlations with observed mortality and estimated avoidance rates. The location of the windfarm within a predominantly agricultural area with little suitable habitat made it unlikely to have the high densities of raptor activity seen at windfarms in California, USA and Tarifa, Spain, and served as an ideal site with minimal threat to raptors. Furthermore, the results of similar studies in the vicinity (Howe & Atwater 1999; Howe, Evans & Wolf 2002) and pre-construction assessments at this windfarm suggest there will be no diﬀerence in abundance between years or between reference and windfarm project areas. Based on previous studies of avoidance behaviour and collision risk (Howe, Evans & Wolf 2002; reviewed in Langston & Pullan 2003), we expected to see species-speciﬁc diﬀerences, with the particular ﬂight behaviour and hunting ecology of American

kestrels Falco sparverius, red-tailed hawks Buteo jamaicensis and turkey vultures Cathartes aura causing them to ﬂy more often than other species within 100 m of turbines and to have the highest risk of collision, and northern harriers, which typically course low over the ground and rarely collide with turbines (Whitﬁeld & Madders 2006), having the highest rates of avoidance. Although all these species have been observed to ﬂy within the rotor zone near turbines at windfarms in Wisconsin, we expected our results to be similar to comparable windfarm studies in the region and around the world which found few raptor mortalities and high avoidance rates, probably due to eﬀective siting measures and risk-recognition by raptors in ﬂight (Howe, Evans & Wolf 2002; reviewed in Madders & Whitﬁeld 2006; Gruver et al. 2009). We present one year of pre-construction and two years of post-construction data, their management implications, and suggestions for future avenues of research.

Materials and methods STUDY AREA

The study windfarm, the Forward Wind Energy Centre, encompasses approximately 13 110 hectares in southeastern Wisconsin (8827– 34¢N, 4332–39¢W; Fig. 1 inset). Approximately 97% of the project area is agricultural land, and 2% is deciduous woodland. The landscape within the project area is mostly ﬂat with an elevational gradient of less than 90 m. The windfarm consists of 86 General Electric 1Æ5sle wind turbines for a combined maximum capacity of 129 MW of energy annually. These turbines have a single tubular tower conﬁguration that measures 80 m high at the hub, and reaches 118 m at the rotor-tip. Adopting terminology from Smallwood, Rugge & Morrison (2009), the rotor plane (area swept by the rotor blades) measures 77 m across, covers 4657 m2 and is characterized by a rotor zone spanning 41–118 m aboveground. Turbines are clustered and typically spaced at least 500 m apart. The windfarm became commercially operational on 14 May, 2008.

FIELD METHODS AND ANALYTICAL APPROACH

As part of a larger study, we compared pre- and post-construction measures of raptor abundance and ﬂight behaviour. Pre-construction data collection by the consulting ﬁrm Curry and Kerlinger, LLC, was not done within reference areas, preventing the use of the standard Before-After-Control-Impact design. However, data were collected from reference areas post-construction, allowing evaluation of windfarm eﬀects on avian use metrics in both spatial and temporal dimensions. Thus our study was more robust than a pre- or post-construction-only study, or a simple comparison of impact and reference areas. Four ﬂight transects oriented in a north-south direction were established parallel to the geographical boundaries of the Horicon National Wildlife Refuge at distances extending 1, 3, 6 and 10 km east from the refuge (Fig. 1). Raptor survey stations were established at the intersections of these ﬂight transects with three east–west transects for a total of 12 survey stations (Fig. 1). The use of this grid system, combined with high visibility at the selected sites, allowed for nearly full visual coverage of the entire project area. Eight survey stations were established in a similar manner exterior to the windfarm project area in June of 2009 to serve as reference stations for

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 199–209

Raptor behaviour within a windfarm 201

Fig. 1. Windfarm project area, Wisconsin, USA. Heavy grey lines indicate east–west study transects. Inset depicts location of the windfarm within the state of Wisconsin. comparison with project (interior) stations (Fig. 1). Data analysis using reference stations was restricted to the summer of 2009 in order to keep comparisons valid among years. Therefore, unless otherwise noted, all results exclude data from reference stations. Raptor survey methodology was adapted from standardized count methodologies (Fuller & Mosher 1981; Bibby et al. 2000), and was designed in conjunction with state and federal agencies. The circular plot survey method was used with no distance cut-oﬀ. We conducted surveys between 08:30 and 15:00, with each survey lasting 60 min, and the order of sites determined via randomization without replacement. Observations were primarily restricted to resident raptor species (accipiters, buteos, eagles, falcons, harriers, osprey, owls and vultures) that were detectable using binoculars of 10· magniﬁcation. A ﬁeld scope (25–75 · 82) mounted on a tripod was used to supplement identiﬁcation with binoculars. For each raptor observed, we recorded date, time, species, number of individuals in same-species groups, behaviour (e.g. ﬂying or perched), ﬂight height with respect to the rotor zone (below = 0–40 m, within = 41–118 m, above = 119 m or higher above ground level) and direction. Initial ﬂight height category was recorded, as well as any subsequent changes in ﬂight height during the survey. Flight paths were not mapped because there were often too

many individuals to monitor within the visual area. Behavioural response when within 100 m of turbines was also recorded, categorized as avoidance, no response or high-risk. Avoidance was deﬁned as changes in ﬂight height category or ﬂight direction that deviated away from turbines or turbine blades, regardless of distance to turbines (i.e. small-scale and last-second avoidance, Blew et al. 2008). High-risk behaviours were deﬁned as ﬂights directly toward a turbine without signs of avoidance, circling around a turbine and within the rotor plane. We used the number of birds ﬂying through the rotor zone at any time during the survey while within 500 m of a turbine array as an additional index of collision risk. To reduce double-counting within a single observation period, we excluded subsequent counts of individuals suspected to have been previously recorded (e.g. same species in the same general area), but included any changes in ﬂight height, behaviour and response to turbines with the ﬁrst observation for that individual. Pre-construction raptor surveys were conducted by a biologist working for Curry and Kerlinger, LLC, and occurred year-round from 4 April 2005 to 31 March 2006. Post-construction raptor surveys were conducted by J.C.G. from 12 June 2008 to 31 August 2008, and by J.C.G. and her ﬁeld assistant C. Kowalchuk (C.K.) from 15 April 2009 to 31 August 2009. Stations were visited approximately

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 199–209

202 J. C. Garvin et al. ﬁve times each between April and May and four times between June and August. We restricted pre-construction data to 15 April 2005–31 August 2005 for comparison with post-construction data. Bird mortality data were obtained from a complementary study conducted by S.M.G. as part of a larger study of this windfarm (Drake et al. 2010; see Appendix S1 in Supporting Information for methodology). Data were also used from US Breeding Bird Surveys, both state-wide and restricted to the region near the windfarm (USGS 2010), to investigate annual trends in raptor abundance. Scientiﬁc and common names for birds are derived from the 7th edition of the Check-list of North American Birds, produced by the American Ornithologists’ Union. Individuals could not always be identiﬁed to the species level, and thus taxonomic groupings such as buteo and accipiter were used when appropriate. The research described in this paper was approved by the University of WisconsinMadison’s animal care and use committee and assigned protocol number A01354-0-06-08.

HABITAT DATA

We quantiﬁed the habitat occurring within a 3-km radius around each survey station using GIS data derived from the US 2001 National Land Cover Database, which is the most current land cover database available for this region. We calculated the percentage of each circular plot that was covered in each land cover category. In order to determine the amount of natural habitat, deﬁned here as areas largely unmodiﬁed by humans, we excluded categories that included pasture, crop, barren or developed land. We then combined the remaining cover categories (e.g. wetland, forest, grassland) into the broader category of natural habitat.

WEATHER DATA

During pre-construction surveys cloud cover, temperature and average wind speed data were collected on-site at the time of each raptor survey. We collected post-construction data on-site for cloud cover, while 10-min incremental wind speed data were obtained from a turbine anemometer in the centre of the study area. Post-construction hourly temperature data were obtained from the National Weather Service station at the Fond Du Lac, Wisconsin airport, located 17 km from the centre of the project area. Variation among years was investigated using daily averages, but there was no systematic variation in any weather variables that might inﬂuence the outcome of the results based on our modelling.

STATISTICAL ANALYSES

The abundance of raptors within the windfarm (number of birds per survey) was analysed with respect to selected temporal, spatial and environmental covariates. All statistical analyses were conducted using sas software (Version 9.2, SAS Institute Inc., Cary, NC, USA) to evaluate generalized linear mixed models (GLMM) which allow the incorporation of both ﬁxed and random eﬀects. We used proc glimmix to construct a separate predictive model for seven raptor count response variables. The unit of analysis was the count of raptors within a 60-min survey on a given visit. We modelled turkey vultures with a Poisson distribution, while all raptors combined, red-tailed hawks and all raptors from summer 2009 only were modelled with a negative binomial distribution. Due to the small numbers of accipiter, American kestrel and northern harrier (average < 1 per survey), we performed a logistic regression with proc glimmix (logit link) to analyse diﬀerences in the odds of observing

these species as a function of selected covariates. Other raptor species were observed too infrequently to analyse. Degrees of freedom were calculated using the approximation of Kenward & Roger (1997). Parameter estimation was performed using Restricted Maximum Subject-Speciﬁc Pseudo-Likelihood (RSPL), which eﬀectively accounts for random eﬀects (Molenberghs & Verbeke 2006). We conducted post hoc analyses on least-square means using a Bonferroni adjustment to evaluate between-year diﬀerences. We used a single predictive model which included the primary sources of variation for each response variable to evaluate the eﬀect of windfarm construction on raptor abundance. This avoided multiple model comparisons with pseudo-AIC criteria. We considered year as a ﬁxed eﬀect to distinguish between pre- and post-construction. Environmental and temporal ﬁxed eﬀects included percentage cloud cover, temperature, wind speed, percentage of natural habitat and time of day modelled as a quadratic eﬀect. This latter eﬀect accounted for declining activity of raptors as the survey day progressed. Inferences regarding environmental variables were limited to the ranges observed during the study (range: cloud cover = 0–100%, temperature = 38–90 F, wind speed = 0–33 mph, natural habitat = 3–35%). We used Pearson’s correlation coeﬃcients (r) calculated with the SAS procedure proc corr to evaluate and limit collinear environmental variables before inclusion in the model. Given that there were unequal time intervals between raptor survey visits, we accounted for non-independence of repeat counts at the same site by using a one-dimensional spatial power covariance structure appropriate for accounting for temporal covariance among visits (with the exception of accipiters and American kestrels, in which limited data precluded its use). Both survey station and visit nested within survey station and year were modelled as random eﬀects. Avoidance rates, deﬁned here as the probability of a bird taking avoidance action when encountering a turbine (Chamberlain et al. 2006), were estimated as: 1 – [(corrected estimate of actual mortality per carcass search period) ⁄ (total number of birds at risk during carcass search period)] (Madders & Whitﬁeld 2006). We calculated the number of birds at risk (per species) as the number of birds ﬂying through the rotor zone (at any time during a survey) within 500 m of turbine arrays. Observations were restricted to this focal area to improve estimate accuracy by (i) limiting them to birds that actually ﬂew within turbine arrays, and (ii) decreasing the error caused by estimating ﬂight heights long distances away from turbines. We then divided by the hours of observation to generate the mean passage rate, and subsequently multiplied by the average day length during the study period (14Æ5 h). This daily rate was then multiplied by the number of days of mortality searches to generate the number of birds passing through the windfarm within the rotor zone during carcass searches. Observed small-scale avoidance rates were derived from the proportion of individuals ﬂying within 100 m of turbines that showed avoidance behaviours. Estimates of raptor mortality were corrected for searcher eﬃciency and scavenger removal, and calculated using the Huso estimator (Huso 2010). Full details of the calculations of mortality estimates can be found in Appendix S1 Supporting Information.

Results ABUNDANCE – VARIATION AMONG YEARS

A total of 93, 48 and 108 surveys were conducted in 2005, 2008 and 2009, respectively, with abundance of species and groups

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 199–209

Raptor behaviour within a windfarm 203 varying among years (Table 1). The most abundant species were the same in all years: turkey vulture, red-tailed hawk, northern harrier and American kestrel (Table 1). Overall raptor abundance was on average 47% lower post-construction (Tables 2a and 3a). The abundances of all ﬁve species ⁄ groups examined were aﬀected by year, and were lower post-construction compared to pre-construction (Tables 2–4). Abundance of red-tailed hawks was on average 51% lower post-construction (Tables 2b and 3b), while turkey vulture abundance did not diﬀer between 2005 and 2008, but decreased by 50% in 2009, and was marginally lower in 2009 compared to 2008 (Table 3c). American kestrels, northern harriers and accipiters all had slightly lower chances of being observed in 2008 compared to 2005 (although conﬁdence intervals included 1Æ0), but signiﬁcantly lower chances of being observed in 2009 compared to 2005 (Table 4). Environmental and temporal variables aﬀected the abundances of species ⁄ groups diﬀerently, except for the presence of natural habitat which universally had no eﬀect (Table 2). Of the species ⁄ groups examined, only the northern harrier was a species of special concern in Wisconsin. US Breeding Bird Survey data collected state-wide from 2005, 2008 and 2009 showed no diﬀerence among years in raptor abundance using matched-pairs t-tests on the number of birds per count for each species (2005–2008: t9 = 1Æ35, P = 0Æ210, 2005–2009: t9 = 1Æ16, P = 0Æ275, 2008–2009: t9 = 1Æ07, P = 0Æ311). Survey data collected within the region of the windfarm (Wisconsin routes 59, 60 and 61) from 2005 to 2009 yielded similar results. Because data were far sparser, a one-way ANOVA of the eﬀect of year on ln-transformed number per count by species was used (F4,15 = 1Æ66, P = 0Æ211).

ABUNDANCE – VARIATION BETWEEN REFERENCE AND PROJECT STATIONS

We examined the overall abundance of raptors restricted to June–August 2009 in a GLMM which included station type

(reference or project) as an additional ﬁxed eﬀect. Overall raptor abundance was on average 61% greater at reference stations (mean raptors per survey: Reference = 9Æ75 ± 1Æ70; Project = 6Æ07 ± 0Æ67) and varied with environmental eﬀects (Table 2g).

FLIGHT HEIGHT

Despite training to ensure continuity between observers, observations by C.K. were signiﬁcantly diﬀerent to observations by other observers (ﬂight height frequency by observer Pearson test: v2 = 60Æ8, P < 0Æ001). Thus ﬂight height results from 2009 exclude observations by C.K. unless noted diﬀerently. Greater numbers of raptors ﬂew above the rotor zone relative to below or within it during both pre- and post-construction (initial ﬂight heights used; Fig. 2). We compared the relative frequencies of birds ﬂying within each ﬂight height category (frequencies of initial ﬂight heights per survey) between pre- and post-construction and found no diﬀerence (Wilcoxon test, chi-square approximation below RZ: v21 = 0Æ960, P = 0Æ327; within RZ: v21 = 0Æ041, P = 0Æ839; above RZ: v21 = 0Æ384, P = 0Æ535). We did not attempt to examine whether ﬂight heights exterior from the windfarm were diﬀerent because the relevant surveys had biased ﬂight heights (see above). Species-speciﬁc ﬂight behaviours did not appear to change greatly between pre- and post-construction for those species with enough numbers to compare.

RISK OF COLLISION

Of the 1480 raptors observed post-construction (includes observations by C.K.), 1455 (98%) were in ﬂight at some point during the survey. Of those in ﬂight, 913 (63%) ﬂew within 500 m of turbine arrays, and nearly half of these individuals (49%, N = 445) ﬂew within the rotor zone at some point during the survey. This is a conservative estimate because

Table 1. Annual abundance (number of raptors per survey) observed by species ⁄ group

Species Accipiter American kestrel Bald eagle Broad-winged hawk Buteo Great horned owl Northern harrier Osprey* Peregrine falcon Red-shouldered hawk Red-tailed hawk Turkey vulture Unidentiﬁed raptor Total

Status

SC

SC ST SE ST

2005 abundance

2008 abundance

2009 abundance

0Æ333 0Æ376 0Æ022 0Æ086 0Æ011 0 0Æ688 0Æ022 0Æ022 0Æ011 4Æ591 6Æ075 0Æ602 12Æ839

0Æ083 0Æ104 0 0 0 0 0Æ229 0 0Æ021 0 2Æ063 6Æ354 0Æ042 8Æ896

0Æ065 0Æ028 0 0Æ019 0Æ123 0Æ009 0Æ296 0 0 0Æ019 2Æ324 3Æ500 0Æ250 6Æ639

(31) (35) (2) (8) (1) (64) (2) (2) (1) (427) (565) (56) (1194)

(4) (5)

(11) (1) (99) (305) (2) (427)

(7) (3) (2) (14) (1) (32)

(2) (251) (378) (27) (717)

Numbers of birds are given in brackets. Conservation status of each species ⁄ group is denoted. SE = State Endangered, ST = State Threatened, SC = Species of Concern in Wisconsin, *currently in the process of being de-listed. 2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 199–209

204 J. C. Garvin et al. Table 2. Evaluation of ﬁxed eﬀects from generalized linear mixed model of abundance (a) all raptors, (b) red-tailed hawk, (c) turkey vulture, (d) American kestrel, (e) northern harrier, (f) accipiter and (g) all raptors for summer 2009 only Model eﬀects

Estimate (SE)

df

t

P

Table 2. (Continued) Model eﬀects Time of day Time of day2 Natural habitat

Estimate (SE)

df

t

0Æ008 (0Æ041) )0Æ000 (0Æ000) 3Æ796 (2Æ871)

237 237 8

0Æ20 )0Æ15 1Æ32

0Æ838 0Æ882 0Æ222

125 21 126 112 132 126 126 16

)1Æ54 )2Æ21 )1Æ37 0Æ00 )2Æ54 1Æ96 )1Æ90 )0Æ68

0Æ127 0Æ038 0Æ175 0Æ999 0Æ012 0Æ052 0Æ060 0Æ509

P

(a) All raptors combined Int )8Æ552 Year 2005 0Æ711 Year 2008 0Æ241 Wind speed )0Æ023 Temperature )0Æ000 Cloud cover )0Æ005 Time of day 0Æ029 Time of day2 )0Æ000 Natural habitat 0Æ681

(3Æ059) (0Æ127) (0Æ150) (0Æ007) (0Æ005) (0Æ001) (0Æ009) (0Æ000) (0Æ602)

201 34 53 229 176 223 201 200 35

)2Æ80 5Æ61 1Æ61 )3Æ04 )0Æ03 )4Æ66 3Æ38 )3Æ19 1Æ13

0Æ006 <0Æ001 0Æ114 0Æ003 0Æ977 <0Æ001 0Æ001 0Æ002 0Æ266

(g) All raptors summer 2009 only Int )9Æ354 (6Æ081) Station type )0Æ474 (0Æ214) Wind speed )0Æ014 (0Æ010) Temperature 0Æ000 (0Æ007) Cloud cover )0Æ005 (0Æ002) Time of day 0Æ033 (0Æ017) Time of day2 )0Æ000 (0Æ000) Natural habitat )0Æ690 (1Æ020)

(b) Red-tailed hawk Int Year 2005 Year 2008 Wind speed Temperature Cloud cover Time of day Time of day2 Natural habitat

)4Æ959 0Æ687 )0Æ038 )0Æ032 )0Æ021 )0Æ008 0Æ021 )0Æ000 0Æ505

(3Æ772) (0Æ142) (0Æ191) (0Æ009) (0Æ006) (0Æ001) (0Æ011) (0Æ000) (0Æ694)

191 27 59 221 191 219 195 197 27

)1Æ31 4Æ84 )0Æ20 )3Æ43 )3Æ61 )5Æ44 1Æ96 )1Æ83 0Æ73

0Æ190 <0Æ001 0Æ844 <0Æ001 <0Æ001 <0Æ001 0Æ051 0Æ069 0Æ473

(c) Turkey vulture Int Year 2005 Year 2008 Wind speed Temperature Cloud cover Time of day Time of day2 Natural habitat

)15Æ23 0Æ702 0Æ466 )0Æ032 0Æ012 )0Æ005 0Æ041 )0Æ000 0Æ517

(5Æ250) (0Æ160) (0Æ192) (0Æ013) (0Æ008) (0Æ002) (0Æ015) (0Æ000) (0Æ737)

207 153 121 222 209 208 207 204 127

)2Æ90 4Æ38 2Æ43 )2Æ56 2Æ64 )2Æ68 2Æ79 )2Æ64 0Æ70

0Æ004 <0Æ001 0Æ017 0Æ011 0Æ009 0Æ008 0Æ006 0Æ009 0Æ484

(d) American kestrel Int )2Æ330 (15Æ109) Year 2005 2Æ781 (0Æ689) Year 2008 1Æ464 (0Æ814) Wind speed 0Æ048 (0Æ043) Temperature 0Æ018 (0Æ023) Cloud cover 0Æ010 (0Æ006) Time of day )0Æ012 (0Æ043) Time of day2 0Æ000 (0Æ000) Natural habitat 0Æ732 (3Æ716)

237 237 237 237 237 237 237 237 12

)0Æ15 4Æ03 1Æ80 1Æ11 0Æ76 1Æ66 )0Æ27 0Æ28 0Æ20

0Æ878 <0Æ001 0Æ074 0Æ269 0Æ446 0Æ099 0Æ787 0Æ782 0Æ847

observations by C.K. tended to be biased low (i.e. relatively more observations recorded within RZ and fewer recorded above RZ compared to other observers). Based on the proportion of a species ﬂying through the rotor zone within 500 m of turbines, American kestrel (57%, 4 of 7), red-tailed hawk (56%, 151 of 270) and turkey vulture (48%, 269 of 564) had the highest collision risk, compared to accipiter (33%, 3 of 9) and northern harrier (10%, 3 of 29). Only 11% of raptors ﬂying within 500 m of turbines were observed ﬂying within 100 m of a turbine. Most demonstrated especially high-risk behaviours, while the remainder displayed signs of small-scale avoidance (Table 5). The raptors with no response to turbines were typically individuals on a straight ﬂight path passing through the windfarm. Red-tailed hawks and turkey vultures made up the majority of the birds that ﬂew within 100 m of turbines (Table 5). Over half (57%) of the observations of raptors ﬂying within 100 m of a turbine were seen at three of the survey stations, indicating a non-random distribution in space.

(e) Northern harrier Int )9Æ646 (11Æ849) Year 2005 1Æ356 (0Æ357) Year 2008 0Æ292 (0Æ493) Wind speed 0Æ017 (0Æ029) Temperature )0Æ014 (0Æ017) Cloud cover 0Æ005 (0Æ004) Time of day 0Æ024 (0Æ034) )0Æ000 (0Æ000) Time of day2 Natural habitat 1Æ804 (1Æ623)

237 129 177 237 213 237 237 237 103

)0Æ81 3Æ79 0Æ59 0Æ57 )0Æ82 1Æ07 0Æ72 )0Æ73 1Æ11

0Æ416 <0Æ001 0Æ554 0Æ566 0Æ412 0Æ284 0Æ470 0Æ464 0Æ269

(f) Accipiter Int Year 2005 Year 2008 Wind speed Temperature Cloud cover

237 237 237 237 237 237

)0Æ37 3Æ54 0Æ37 )1Æ13 )0Æ62 )0Æ76

0Æ708 <0Æ001 0Æ712 0Æ259 0Æ533 0Æ449

)5Æ383 1Æ740 0Æ261 )0Æ044 )0Æ013 )0Æ004

(14Æ362) (0Æ492) (0Æ706) (0Æ039) (0Æ021) (0Æ005)

RATES OF MORTALITY

Two red-tailed hawks were found during searches beneath a subset of turbines (N = 29), one in August 2008 and the other in May 2009. Three red-tailed hawks were found near turbines outside of search transects or search schedules (incidental ﬁnds; see Table S1, Supporting Information). No other raptor mortalities were reported. Carcasses appeared to be randomly distributed throughout the windfarm. All carcasses showed injuries typical of collision with turbine blades as revealed by X-ray and necropsy (e.g. wing injuries ⁄ amputations, neck injuries, decapitation, Barrios & Rodriguez 2007). The corrected estimates of mortality were: autumn 2008 = 0Æ003 red-tailed hawks turbine)1 day)1 (95% CI = 0, 0Æ009), spring 2009 = 0Æ005 red-tailed hawks turbine)1day)1 (95% CI = 0, 0Æ017), and autumn 2009 = 0Æ000 (no raptors found). Assuming one raptor carcass had been found in autumn 2009, the estimate of mortality would have been at most 0Æ003 raptors turbine)1 day)1 (95% CI = 0, 0Æ009). Conservative annual rates of mortality based on the total number of days within the search period(s) per year were 0Æ363

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 199–209

Raptor behaviour within a windfarm 205 Table 3. Post hoc comparison of year eﬀects (a) total raptors, (b) red-tailed hawk and (c) turkey vulture Year comparison

Estimate (SE)

(a) All raptors 2005 vs. 2008 2005 vs. 2009 2008 vs. 2009 (b) Red-tailed hawk 2005 vs. 2008 2005 vs. 2009 2008 vs. 2009 (c) Turkey vulture 2005 vs. 2008 2005 vs. 2009 2008 vs. 2009

df

t

Adj P

Year

Mean Count (SE)

0Æ467 (0Æ155) 0Æ711 (0Æ127) 0Æ241 (0Æ145)

56 34 53

3Æ03 5Æ61 1Æ61

0Æ012 <0Æ001 0Æ345

2005 2008 2009

12Æ64 (1Æ14) 7Æ90 (0Æ97) 6Æ21 (0Æ54)

0Æ725 (0Æ195) 0Æ687 (0Æ142) )0Æ038 (0Æ191)

58 27 59

3Æ72 4Æ84 )0Æ20

0Æ002 <0Æ001 1Æ000

2005 2008 2009

4Æ14 (0Æ41) 2Æ01 (0Æ33) 2Æ08 (0Æ21)

0Æ236 (0Æ203) 0Æ702 (0Æ160) 0Æ466 (0Æ192)

116 153 121

1Æ16 4Æ38 2Æ43

0Æ740 <0Æ001 0Æ049

2005 2008 2009

4Æ97 (0Æ55) 3Æ93 (0Æ63) 2Æ47 (0Æ27)

Table 4. Post hoc comparison of year eﬀects (a) American kestrel, (b) northern harrier, and (c) accipiter Year comparison

Estimate (SE)

df

t

Adj P

OR

95% CI

(a) American kestrel 2005 vs. 2008 2005 vs. 2009 2008 vs. 2009

1Æ317 (0Æ651) 2Æ781 (0Æ689) 1Æ464 (0Æ814)

237 237 237

2Æ02 4Æ03 1Æ80

0Æ133 <0Æ001 0Æ220

3Æ73 16Æ14 4Æ32

0Æ78, 17Æ94 3Æ06, 85Æ08 0Æ61, 30Æ81

(b) Northern harrier 2005 vs. 2008 2005 vs. 2009 2008 vs. 2009

1Æ064 (0Æ497) 1Æ356 (0Æ357) 0Æ292 (0Æ493)

154 129 177

2Æ14 3Æ79 0Æ59

0Æ102 0Æ001 1Æ000

2Æ90 3Æ88 1Æ34

0Æ87, 9Æ66 1Æ63, 9Æ22 0Æ41, 4Æ41

(c) Accipiter 2005 vs. 2008 2005 vs. 2009 2008 vs. 2009

1Æ479 (0Æ654) 1Æ740 (0Æ492) 0Æ261 (0Æ706)

237 237 237

2Æ26 3Æ54 0Æ37

0Æ074 0Æ002 1Æ000

4Æ39 5Æ70 1Æ30

0Æ91, 21Æ21 1Æ74, 18Æ65 0Æ24, 7Æ12

Odds ratios (OR) with adjusted 95% conﬁdence intervals are presented.

10

Table 5. Behavioural response of all raptors that approached within 100 m of wind turbines

Mean # raptors per survey

9 8

Response

7

Accipiter American kestrel Buteo Northern harrier Red-tailed hawk Turkey vulture Total

6 5 4 3

N

R

1 1 2 3 12 17 32

11 6 18

21 30 54

Total 1 1 2 3 44 53 104

Type of response to turbines is categorized as avoidance (A), no response (N) or high-risk (R).

2 1 0

A

1

2

3

Flight height category Fig. 2. Flight height frequency distributions pre- (open) and postconstruction (ﬁlled). Flight height category 1 = below rotor zone, 2 = within rotor zone, 3 = above rotor zone.

red-tailed hawks turbine)1 year)1 (95% CI = 0, 1Æ128) for 2008, and 0Æ275 red-tailed hawks turbine)1 year)1 (95% CI = 0, 0Æ835) for 2009 (actual estimates from spring and autumn combined). Incidental carcasses were excluded from mortality calculations. Because no carcasses were found for other raptor species, one can assume that mortality estimates for these species would be less than the values provided above.

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 199–209

206 J. C. Garvin et al.

AVOIDANCE RATES

Avoidance rates were estimated for the ﬁve most abundant raptor species ⁄ groups (accipiter, American kestrel, northern harrier, red-tailed hawk and turkey vulture; Table 6). Observed small-scale avoidance was signiﬁcantly lower compared to estimates of avoidance for nearly all species, with the exception of northern harrier (Table 6).

Discussion ABUNDANCE

Although we predicted abundance would remain relatively constant, raptor abundance was lower post-construction compared to pre-construction levels. This index of raptor activity was also aﬀected by spatial, temporal and weather covariates (similar to previous studies, e.g. Bunn, Klein & Bildstein 1995). Additionally, American kestrel, red-tailed hawk and turkey vulture appeared to decline more than other species. The decline in raptor abundance post-construction may have been a result of general avoidance (displacement) of the windfarm. Raptors may have relocated to areas outside the windfarm, which was supported by our ﬁnding that abundance was higher at reference stations outside of the windfarm project area, although pre-construction baseline data were not collected at reference stations. Another study in Wisconsin (Howe, Evans & Wolf 2002) found that open-country raptors were more abundant in the reference area surrounding their windfarm than within the windfarm itself. Moreover, all ﬁve species ⁄ groups analysed at the windfarm were less abundant post-construction, providing further evidence of a possible displacement eﬀect. Displacement could have been caused by the disturbance of windfarm construction and the ongoing presence of turbines and maintenance machinery (reviewed in both Langston & Pullan 2003; Madders & Whitﬁeld 2006). Habitat fragmentation or loss is unlikely to have caused this apparent displacement because the windfarm is located in a primarily agricultural setting with very little apparent, suitable raptor habitat (habitat eﬀects reviewed in Langston & Pullan 2003), and the land use has been consistent since before 2005. On average, only 11% of the habitat assessed was considered nat-

ural habitat, and this variable did not inﬂuence abundance for any raptor species or group. Whether this possible displacement eﬀect will remain constant over time, become more pronounced (see Stewart, Pullin & Coles 2007) or decrease through gradual habituation as with pink-footed geese Anser brachyrhynchus (Madsen & Boertmann 2008), will require additional years of study. The abundance of accipiter, American kestrel, northern harrier and turkey vulture decreased signiﬁcantly from 2005 to 2009, but not between 2005 and 2008, suggesting that for some species there may be a temporal lag in the possible displacement eﬀect in response to windfarm construction. No raptor groups analysed in this study changed signiﬁcantly in abundance from 2008 to 2009 (although it was marginally lower in 2009 for turkey vulture), suggesting that raptor activity did not rebound over the temporal scale considered, minimizing the possibility that post-construction declines were an artefact caused by annual variation or observer diﬀerences. This is further supported by analyses of annual state and regional breeding bird survey data which indicated that raptor numbers were not abnormally high in 2005, and thus our single year of pre-construction data should provide an unbiased comparison. While there may be slight diﬀerences between observers in technique, visual acuity and skill in species identiﬁcation, using identical study protocols should aid in controlling for any inter-observer diﬀerences. Furthermore, observer identity (as a random variable) was investigated within statistical models, and only signiﬁcantly impacted observations of ﬂight height which were then adjusted accordingly. Therefore, the decrease in raptor abundance seems most likely to be caused by displacement of raptors in the vicinity of the windfarm.

COLLISION RISK, AVOIDANCE, AND MORTALITY RATES

Overall, birds ﬂew least often within the rotor zone, indicating that risk of collision with turbines was minimal, similar to our predictions. This was further supported by the fact that although half of the individuals observed within turbine arrays ﬂew within the rotor zone, few of these approached within 100 m of turbines. While observer bias may inﬂuence estimation of ﬂight heights, especially during surveys lacking turbine height references, observations of birds within 100 m of turbines were likely to be accurate and of greatest relevance

Table 6. Estimated and observed avoidance rates 2008

2009

Species

E. Mortality

No. at Risk

E. Avoid.

O. Avoid.

E. Mortality

No. at Risk

E. Avoid.

O. Avoid.

Accipiter American kestrel Northern harrier Red-tailed hawk Turkey vulture

0 0 0 31Æ22 0

42Æ81 128Æ43 42Æ81 1412Æ71 2397Æ33

100Æ0% 100Æ0% 100Æ0% 97Æ8% 100Æ0%

0% 0% – 20% 20%

0 0 0 23Æ65 0

36Æ58 18Æ29 36Æ58 2158Æ02 3895Æ41

100Æ0% 100Æ0% 100Æ0% 98Æ9% 100Æ0%

– – 100% 29% 43%

The corrected estimate of mortality per year within the entire windfarm, number of birds at risk, estimates of avoidance and observed small-scale avoidance are presented for the most abundant species for each year post-construction. 2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 199–209

Raptor behaviour within a windfarm 207 when observing small-scale avoidance and predicting collision risk. There were species-speciﬁc diﬀerences in both collision risk and avoidance rates. While American kestrels, red-tailed hawks and turkey vultures followed predictions by having the highest proportions of individuals at risk of collision, and similarly high proportions of birds displaying high-risk behaviours near turbines, these indices of collision risk did not correlate with observed mortalities. Observed small-scale avoidance rates were highest for northern harrier (all individuals avoided turbines), with the next highest values for turkey vulture, followed by red-tailed hawk and American kestrel. In comparison, estimated avoidance rates were 100% for all but red-tailed hawks. Thus, avoidance behaviour, at any scale, does appear to strongly aﬀect collision risk, and through it, mortality. According to Orloﬀ & Flannery (1992), turkey vultures have low risk of collision, while red-tailed hawks are at high-risk. In contrast, we found these species had similar ﬂight behaviours within 100 m of turbines, similar to ﬁndings by Howe, Evans & Wolf (2002). Nonetheless, of all raptor species, we only recorded ﬁve red-tailed hawk mortality events (three as incidentals), suggesting that foraging strategies and other speciesspeciﬁc diﬀerences may also aﬀect the level of collision risk (Orloﬀ & Flannery 1992; de Lucas et al. 2008). As expected, few mortalities were observed, and uncorrected numbers were similar to studies at comparable windfarms conducted within the same region and with similar methodology to our windfarm [one incidental raptor per year, Gruver et al. (2009); ﬁve raptors including three incidentals per year, BHE Environmental, Inc. 2010]. Our results were also similar to other studies in the Upper Midwestern USA (e.g. Howe, Evans & Wolf 2002; Johnson et al. 2002), as well as elsewhere in the USA and Europe, with the exceptions of Altamont Pass Wind Resource Area, California, USA and Tarifa, Spain which documented very high raptor mortality rates (reviewed in both Erickson et al. 2002; Drewitt & Langston 2006). While a long-term study by de Lucas et al. (2008) found that mortality is not correlated with abundance (a measure of avian use), Smallwood, Rugge & Morrison (2009) found that red-tailed hawk fatalities increased with both utilization rates and frequency of ﬂights through turbine rows. Our results support the former ﬁndings as no carcasses were found of the most abundant species (turkey vulture), suggesting that mortality is inﬂuenced by more than abundance alone (see above). Mortality rates did not directly correlate with our index of collision risk or observations of small-scale avoidance. Although red-tailed hawks, the only species for which carcasses were found, ranked among the highest in terms of collision risk and high-risk behaviours, no carcasses were found for the other high-risk species (e.g. American kestrel, turkey vulture). Avoidance behaviour varied by location within the windfarm, similar to studies where raptor mortalities were unequally distributed in space (Barrios & Rodriguez 2004). However, mortalities in our study appeared to be distributed randomly, and did not occur in the same areas as the majority of observations of birds within 100 m of turbines.

MANAGEMENT IMPLICATIONS

The northern (hen) harrier was the only species of concern (at the state level) which declined post-construction. Our study supports the notion that northern harriers appear to be at low risk of collision (reviewed in Whitﬁeld & Madders 2006; Smallwood, Rugge & Morrison 2009). A recent study in the UK showed that northern harriers avoided ﬂying within 250 m of turbines (Pearce-Higgins et al. 2009). However, Pearce-Higgins et al. predicted that this avoidance could result in a 53% reduction in ﬂight activity within 500 m of turbine arrays. Our ﬁndings also support the presence of such indirect negative eﬀects, and we encourage continued monitoring of harrier population levels near windfarms both in the USA and globally. Our observations of avoidance ﬂight behaviours and estimation of avoidance rates, although technically simple, based on small sample sizes and uncorrected for several factors (Madders & Whitﬁeld 2006), still provide important information because avoidance rates of operational windfarms are extremely rare and have a strong inﬂuence on estimated collision risk. Indeed, collision risk modelling, an important tool used in windfarm development, has been limited by the lack of speciesand state-speciﬁc avoidance data (Chamberlain et al. 2006). Our reported avoidance rates will provide a reference useful for collision risk modelling of future windfarms. However, site-speciﬁc assessments of avoidance behaviour are strongly recommended, and we caution against calculating avoidance rates using estimates of mortality that are based on hypothetical data, since this may involve invalid assumptions. Lastly, our results suggest that construction of the windfarm poses a minimal mortality risk for raptors, and may not be sufﬁcient to cause population-wide eﬀects. However, our windfarm is only one of four large (>50 MW) windfarms within the region. While impacts (e.g. displacement) from each windfarm separately may be negligible to raptor populations, the cumulative eﬀects may be biologically signiﬁcant, as has been found for populations of the Egyptian vulture Neophron percnopterus in Spain by Carrete et al. (2009) (but see Smales & Muir 2005). Future research should consider raptor population dynamics across multiple windfarms in a broader spatial context in order to predict impacts at the population level.

Conclusions Our study provides evidence for possible displacement and an increased collision risk for raptors (particularly turkey vultures and red-tailed hawks) near the windfarm. However, our observed and estimated avoidance rates, coupled with low mortality events, may indicate eﬀective avoidance behaviours by individuals that remain within the windfarm project area. Determining whether resident species will habituate to the presence of the windfarm and return to their pre-construction levels will require additional years of study. Additionally, cumulative eﬀects from the multiple windfarms in the area on local populations may be biologically signiﬁcant and should be estimated. Lastly, providing empirically determined avoidance rates for several raptor species may aid in the advancement of

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 199–209

208 J. C. Garvin et al. collision risk modelling in general, and in the planning of future windfarms. Taken together, these ﬁndings may aid in reducing the negative impacts of windfarms on susceptible groups such as raptors, both in the USA and elsewhere.

Acknowledgements We would like to thank our research assistants, especially C. Kowalchuk and T. Prestby, for their assistance with logistics, data collection and data entry. Additional thanks to staﬀ from Forward Energy, LLC and Invenergy, LLC, especially M. Chang and L. Miner for assistance with maps and project management, respectively. Lastly, we would like to thank M. Carrete, N. Cutright, P. Dunn and two anonymous reviewers for their insightful comments which improved this manuscript. This study was funded by Invenergy, LLC as part of the permitting requirements determined by the Public Service Commission of Wisconsin

References Barrios, L. & Rodriguez, A. (2004) Behavioural and environmental correlates of soaring-bird mortality at on-shore wind turbines. Journal of Applied Ecology, 41, 72–81. Barrios, L. & Rodriguez, A. (2007) Spatiotemporal patterns of bird mortality at two wind farms of southern Spain. Birds and Wind Farms: Risk Assessment and Mitigation (eds M. de Lucas, G.F.E. Janss & M. Ferrer), pp. 229– 239. Quercus, Madrid. BHE Environmental, Inc. (2010) Post-construction bird and bat mortality study: Cedar Ridge wind farm, Fond Du Lac County, Wisconsin - Interim Report. BHE Environmental, Inc., Cincinnati, OH. Bibby, C.J., Burgess, N.D., Hill, D.A. & Mustoe, S. (2000) Bird Census Techniques. Academic Press, San Diego, CA. Blew, J., Hoﬀmann, M., Nehls, G. & Hennig, V. (2008) Investigations of the bird collision risk and the responses of harbour porpoises in the oﬀshore wind farms Horns Rev, North Sea, and Nysted, Baltic Sea, in Denmark Part I: Birds. Created by Universita¨t Hamburg and BioConsult SH. Bunn, A.G., Klein, W. & Bildstein, K.L. (1995) Time-of-day eﬀects on the numbers and behavior of non-breeding raptors seen on roadside surveys in eastern Pennsylvania. Journal of Field Ornithology, 66, 544–552. Carrete, M., Sanchez-Zapata, J.A., Benitez, J.R., Lobon, M. & Donazar, J.A. (2009) Large scale risk-assessment of wind-farms on population viability of a globally endangered long-lived raptor. Biological Conservation, 142, 2954– 2961. Chamberlain, D.E., Rehﬁsch, M.R., Fox, A.D., Desholm, M. & Anthony, S.J. (2006) The eﬀect of avoidance rates on bird mortality predictions made by wind turbine collision risk models. Ibis, 148, 198–202. Drake, D., Garvin, J.C., Grodsky, S.M. & Watt, M.A. (2010) Post-construction Bird and Bat Monitoring at the Forward Energy Center – Final Report. University of Wisconsin – Madison, Madison, WI. Drewitt, A.L. & Langston, R.H.W. (2006) Assessing the impacts of wind farms on birds. Ibis, 148, 29–42. Drewitt, A.L. & Langston, R.H.W. (2008) Collision eﬀects of wind-power generators and other obstacles on birds. Year in Ecology and Conservation Biology 2008 (eds R.S. Ostfeld & W.H. Schlesinger), Vol. 1134, pp. 233–266. Blackwell Publishing, Oxford. Erickson, W.P., Johnson, B.G., Young, D., Strickland, D., Good, R.E., Bourassa, M., Bay, K. & Sernka, K.J. (2002) Synthesis and Comparison of Baseline Avian and Bat Use, Raptor Nesting and Mortality Information from Proposed and Existing Wind Developments. Report prepared for Bonnerville Power Administration by WEST Inc., Cheyenne, WY. Fuller, M.R. & Mosher, J.A. (1981) Methods of detecting and counting raptors: a review. Estimating numbers of terrestrial birds (eds C.J. Ralph & J.M. Scott), pp. 235–246. Blackburn Press, Caldwell, NJ. Gruver, J.C., Sonnenberg, M., Bay, K. & Erickson, W.P. (2009) Results of a Post-Construction Bat and Bird Fatality Study at Blue Sky Green Field Wind Energy Center, Fond du Lac County, Wisconsin, July 2008-May 2009. Final report prepared for We Energies, Milwaukee, WI by WEST Inc., Cheyenne, Wyoming. Hoover, S.L. & Morrison, M.L. (2005) Behavior of red-tailed hawks in a wind turbine development. Journal of Wildlife Management, 69, 150–159. Howe, R.W. & Atwater, R. (1999) The Potential Eﬀects of Wind Power Facilities on Resident and Migratory Birds in Eastern Wisconsin. University of Wisconsin - Green Bay, Green Bay, WI.

Howe, R.W., Evans, W. & Wolf, A.T. (2002) Eﬀects of Wind Turbines on Birds and Bats in Northeastern Wisconsin. University of Wisconsin – Green Bay, Green Bay, WI. Huso, M.M.P. (in press) An estimator of wildlife fatality from observed carcasses. Environmetrics, 21, DOI: 10.1002/env.52. Janss, G.F.E. (2000) Avian mortality from power lines: a morphological approach of a species-speciﬁc mortality. Biological Conservation, 95, 353–359. Johnson, G.D., Erickson, W.P., Strickland, M.D., Shepherd, M.F., Shepherd, D.A. & Sarappo, S.A. (2002) Collision mortality of local and migrant birds at a large-scale wind-power development on Buﬀalo Ridge, Minnesota. Wildlife Society Bulletin, 30, 879–887. Kenward, M.G. & Roger, J.H. (1997) Small sample inference for ﬁxed eﬀects from restricted maximum likelihood. Biometrics, 53, 983– 997. Kikuchi, R. (2008) Adverse impacts of wind power generation on collision behaviour of birds and anti-predator behaviour of squirrels. Journal for Nature Conservation (Jena), 16, 44–55. Langston, R. & Pullan, J. (2003) Windfarms and Birds: An Analysis of the Eﬀects of Windfarms on Birds, and Guidance on Environmental Assessment Criteria and Site Selection Issues. pp. 1–58. Birdlife International & Bern Convention, Sandy, UK. de Lucas, M., Janss, G.F.E., Whitﬁeld, D.P. & Ferrer, M. (2008) Collision fatality of raptors in wind farms does not depend on raptor abundance. Journal of Applied Ecology, 45, 1695–1703. Madders, M. & Whitﬁeld, D.P. (2006) Upland raptors and the assessment of wind farm impacts. Ibis, 148, 43–56. Madsen, J. & Boertmann, D. (2008) Animal behavioral adaptation to changing landscapes: spring-staging geese habituate to wind farms. Landscape Ecology, 23, 1007–1011. Molenberghs, G. & Verbeke, G. (2006) Models for Discrete Longitudinal Data. Springer, Berlin. Orloﬀ, S. & Flannery, A. (1992) Wind Turbine Eﬀects on Avian Activity, Habitat Use and Mortality in Altamont Pass and Solano County Wind Resource Areas. Report prepared for California Energy Commission by Biosystems Analysis, Tiburon, CA. Pearce-Higgins, J.W., Stephen, L., Langston, R.H.W., Bainbridge, I.P. & Bullman, R. (2009) The distribution of breeding birds around upland wind farms. Journal of Applied Ecology, 46, 1323–1331. Percival, S. (2005) Birds and windfarms: what are the real issues? British Birds, 98, 194–204. Robbins, C.S. (1981) Bird activity levels related to weather. Estimating Numbers of Terrestrial Birds (eds C.J. Ralph & J.M. Scott). pp. 301–310. Allen Press, Inc., Lawrence, KS. Rugge, L.M. (2001) An avian risk behavior and mortality assessment at the Altamont Wind Resource Area in Livermore, California. MS Thesis, California State University, Sacramento. Shamoun-Baranes, J., van Loon, E., van Gasteren, H., van Belle, J., Bouten, W. & Buurma, L. (2006) A comparitive analysis of the inﬂuence of weather on the ﬂight altitudes of birds. Bulletin of American Meteorological Society, 87, 47–61. Smales, I. & Muir, S. (2005) Modelled Cumulative Impacts on the Tasmanian Wedge-tailed Eagle of Wind Farms Across the Species’ Range. Report for Australian Department of Environment and Heritage by Biosis Research Pty. Ltd. Smallwood, K.S., Rugge, L. & Morrison, M.L. (2009) Inﬂuence of behavior on bird mortality in wind energy developments. Journal of Wildlife Management, 73, 1082–1098. Smallwood, K.S. & Thelander, C.G. (2004) Developing Methods to Reduce Bird Mortality in the Altamont Pass Wind Resource Area, p. 363. BioResource Consultants & California Energy Commission, Sacramento, CA. Stewart, G.B., Pullin, A.S. & Coles, C.F. (2007) Poor evidence-base for assessment of windfarm impacts on birds. Environmental Conservation, 34, 1–11. USGS (2010) North American Breeding Bird Survey. US Geological Survey Patuxent Wildlife Research Center. http://www.pwrc.usgs.gov/bbs/ March 12, 2010. Whitﬁeld, D.P. & Madders, M. (2006) A Review of the Impacts of Wind Farms on Hen Harrier Circus cyaneus and an Estimation of Collision Avoidance Rates. Natural Research Information Note 1 (revised). Natural Research Ltd, Banchory, UK. Received 10 June 2010; accepted 2 November 2010 Handling Editor: Esteban Fernandez-Juricic

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 199–209

Raptor behaviour within a windfarm 209

Supporting Information Additional Supporting Information may be found in the online version of this article:

As a service to our authors and readers, this journal provides supporting information supplied by the authors. Such materials may be reorganized for online delivery, but are not copy-edited or typeset. Technical support issues arising from supporting information (other than missing ﬁles) should be addressed to the authors.

Appendix S1. Raptor mortality methodology. Table S1. Mortality search results.

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 199–209

Journal of Applied Ecology 2011, 48, 210–219

doi: 10.1111/j.1365-2664.2010.01914.x

Negative impact of trafﬁc noise on avian reproductive success Wouter Halfwerk1*, Leonard J. M. Holleman2, C(Kate). M. Lessells2 and Hans Slabbekoorn1 1

Behavioural Biology, Institute of Biology, Leiden University, 2333 BE Leiden, The Netherlands; and 2Netherlands Institute of Ecology (NIOO-KNAW), 6666 ZG Heteren, The Netherlands

Summary 1. Traﬃc aﬀects large areas of natural habitat worldwide. As a result, the acoustic signals used by birds and other animals are increasingly masked by traﬃc noise. Masking of signals important to territory defence and mate attraction may have a negative impact on reproductive success. Depending on the overlap in space, time and frequency between noise and vocalizations, such impact may ultimately exclude species from suitable breeding habitat. However a direct impact of traﬃc noise on reproductive success has not previously been reported. 2. We monitored traﬃc noise and avian vocal activity during the breeding season alongside a busy Dutch motorway. We measured variation in space, time and spectrum of noise and tested for negative eﬀects on avian reproductive success using long-term breeding data on great tits Parus major. 3. Noise levels decreased with distance from the motorway, but we also found substantial spatial variation independent of distance. Noise also varied temporally with March being noisier than April, and the daytime being noisier than night-time. Furthermore, weekdays were clearly noisier than weekends. Importantly, traﬃc noise overlapped in time as well as acoustic frequency with avian vocalization behaviour over a large area. 4. Traﬃc noise had a negative eﬀect on reproductive success with females laying smaller clutches in noisier areas. Variation in traﬃc noise in the frequency band that overlaps most with the lower frequency part of great tit song best explained the observed variation. 5. Additionally, noise levels recorded in April had a negative eﬀect on the number of ﬂedglings, independent of clutch size, and explained the observed variation better than noise levels recorded in March. 6. Synthesis and applications. We found that breeding under noisy conditions can carry a cost, even for species common in urban areas. Such costs should be taken into account when protecting threatened species, and we argue that knowledge of the spatial, temporal and spectral overlap between noise and species-speciﬁc acoustic behaviour will be important for eﬀective noise management. We provide some cost-eﬀective mitigation measures such as traﬃc speed reduction or closing of roads during the breeding season. Key-words: anthropogenic noise, clutch size, great tit, Parus major, reproductive success, trafﬁc noise ﬂuctuations

Introduction Anthropogenic noise currently aﬀects large areas of natural habitat worldwide (Forman 2000; Barber, Crooks & Fristrup 2009). Masking by noise interferes with the use of the acoustic signals critical to many animal species (Bradbury & Vehren-

*Correspondence author. E-mail: [email protected]

camp 1998; Brumm & Slabbekoorn 2005). As a consequence, animals living in areas exposed to anthropogenic noise may suﬀer reduced reproductive success, which may ultimately lead to the exclusion of species from otherwise suitable habitat (Slabbekoorn & Ripmeester 2008). The majority of areas aﬀected by noise are situated along major transport links, such as motorways and railways (Forman 2000; Barber, Crooks & Fristrup 2009). The impact of traﬃc noise has been explored in a diverse range of taxa (frogs;

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society

Traﬃc noise and avian reproductive success 211 Bee & Swanson 2007; bats; Schaub, Ostwald & Siemers 2008), but has been studied most intensively in birds (e.g. Reijnen et al. 1995; Stone 2000). Many studies have shown a reduction in breeding numbers in the vicinity of motorways (e.g. van der Zande, ter Keurs & van der Weijden 1980; Reijnen & Foppen 1991), but no study to date has been able to exclude confounding factors associated with roads and thus identify traﬃc noise as the key threat to birds (Warren et al. 2006). An impact of anthropogenic noise on breeding numbers (Bayne, Habib & Boutin 2008) and species richness (Francis, Ortega & Cruz 2009) without confounding factors has been demonstrated in the vicinity of noisy gas compressor stations. However, extrapolating these ﬁndings to motorway noise is far from straightforward. For instance, noise at gas compressor stations is constant in amplitude throughout the day and year (Francis, Ortega & Cruz 2009), whereas most anthropogenic noise levels show strong daily, weekday versus weekend, and seasonal variation (Bautista et al. 2004; Warren et al. 2006). The negative eﬀect of traﬃc noise on birds depends on the temporal and spectral overlap with relevant acoustic sounds (Brumm & Slabbekoorn 2005). Birds use a variety of vocalizations throughout the day but many species restrict the use of song, which is important in both territorial defence and female attraction, to the period around dawn (Catchpole & Slater 2008). The overlap between dawn song and peaks in traﬃc activity (e.g. the rush hour) may be an important factor in determining negative eﬀects, and depends primarily on the time of year in combination with longitude and latitude (Warren et al. 2006). Assessing temporal variation in noise levels is therefore an important step in understanding when noise overlaps most with the vocal activity of birds (Slabbekoorn & Ripmeester 2008; Barber, Crooks & Fristrup 2009). Spectral overlap is most dramatic for birds vocalizing at low frequencies (e.g. cuckoos, owls, woodpeckers and grouse) as traﬃc noise is typically loudest at lower frequencies (Pohl et al. 2009) and low sounds attenuate less with distance and vegetation density (Wiley & Richards 1978; Padgham 2004). Furthermore, ﬂuctuations in low frequency transmission can change dramatically with weather conditions (Ovenden, Shaffer & Fernando 2009) resulting in unpredictable overlap levels. Even when there is clear temporal and spectral overlap between traﬃc noise and birdsong, assessing whether there is a negative impact on reproductive success in the ﬁeld is not straightforward. The eﬀect on breeding numbers may underestimate the impact and provides little insight into the mechanisms by which birds are aﬀected. For example, breeding success and welfare may be impaired, but breeding densities remain high because of compensating eﬀects of noise on predation rates (Francis, Ortega & Cruz 2009) or competition for food (Slabbekoorn & Halfwerk 2009). Furthermore, inexperienced or low quality birds may be more likely to occupy noisy areas (Reijnen & Foppen 1991; Habib, Bayne & Boutin 2007). Therefore, understanding the mechanisms underlying the negative eﬀects of noise is best achieved by focusing on individual life history traits that are components of reproductive success.

The great tit Parus major (Linneaus 1758) is a common species that is currently not under threat, but the availability of long-term data from a population bordering a major motorway provides a rare opportunity to investigate whether noise has more subtle eﬀects than simply excluding birds from otherwise suitable habitat. This species prefers artiﬁcial nest-boxes to natural cavities (Kluyver 1951) even when they are situated in suboptimal habitat. This is probably one reason why great tits breed in substantial numbers in areas adjacent to motorways (Junker-Bornholdt et al. 1998), allowing collection of breeding data in noisy areas. Great tit singing behaviour has been repeatedly related to noise at both the population (Slabbekoorn & den Boer-Visser 2006; Mockford & Marshall 2009) and individual level (Slabbekoorn & Peet 2003). We know that relatively low frequency songs are detected less well when there is traﬃc-like noise (Pohl et al. 2009), and great tits can switch between song types when exposed to experimental noise (Halfwerk & Slabbekoorn 2009). However, it is unknown whether such behavioural ﬂexibility prevents any negative eﬀects of anthropogenic noise. We studied spatial, temporal and spectral variation in the loudness of traﬃc noise and bird acoustic behaviour in a nestbox population of great tits adjacent to a Dutch motorway with a heavy traﬃc load. Traﬃc noise and bird song were recorded during two important breeding stages: March, when territories are formed, and April, when eggs are laid and incubated. We used these data, together with habitat and long-term breeding data to explore the following questions: How does traﬃc noise in habitat adjacent to a motorway vary in space? To what extent do traﬃc noise and bird vocal activity overlap in time and frequency, and does the amount of overlap diﬀer between breeding stages? Is there an impact of traﬃc noise levels on breeding success? Does seasonal variation in traﬃc noise aﬀect particular breeding stages? And does spectral overlap between great tit song and traﬃc noise play a role in the eﬀect on reproductive success? Answers to these questions will be valuable in identifying conservation measures and applying eﬀective noise management in natural areas polluted by traﬃc noise.

Materials and methods STUDY SITE AND SPECIES

We collected data from a nest-box population of great tits Parus major breeding at the Buunderkamp (0545¢ E; 5201¢ N) in the Netherlands (Fig. 1a). The area is bounded in the north by a four-lane motorway and in the south by a railway line (about 20 trains h)1). The habitat is mixed woodland consisting of plots of varying sizes, and age and species of trees, with Pinus sylvestris and Quercus rubra dominant (see Drent 1987 for further description of the area). The great tit is a hole-nesting passerine that sings in the frequency range of 2–9 kHz (Halfwerk & Slabbekoorn 2009). Territory defence starts in mid-January and peaks towards the end of March (Kluyver 1951). Egg-laying in the study population starts in April and is accompanied by a strong increase in dawn singing activity. We used long-term breeding data on great tits for the period 1995– 2009 during which no major changes have been made in the area

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 210–219

212 W. Halfwerk et al. (a)

(b) "

"

"

"

"

"

"

"

"

"

"

" "

"

"

"

"

"

"

"

"

"

"

"

"

"

"

"

"

"

"

!

!

!

!

!

!

!

!

!

!

!

!

!

!

1

!

!

!

!

!

!

!

!

! !

!

!

!

!

2!

" "

"

"

"

"

"

"

"

"

"

"

"

"

"

"

"

"

!

"

"

"

"

"

"

" "

" "

! !

N

"

"

"

" "

! !

" "

"

"

"

"

!

"

"

" "

" "

"

"

"

"

"

"

"

! !

! !

"

" "

" "

! "

"

"

"

!

!

"

"

" "

! !

" "

"

"

"

!

"

" "

"

"

" " "

"

"

"

" "

"

" "

"

!

!

" "

" "

"

"

"

!

!

"

"

"

"

"

"

! !

"

" "

"

"

"

" "

!

"

"

"

" " "

"

"

!

"

"

" " "

" "

"

"

" " " "

"

"

!

"

"

"

" "

"

" " "

" "

"

"

" "

!

!

!

"

"

"

"

!

!

!

" "

"

"

" "

"

" "

"

"

3

" " "

" "

" "

"

!

!

"

"

"

" "

"

!

!

"

"

"

"

"

"

"

" "

" " "

" " "

"

"

" "

"

"

" "

"

"

" "

" "

" "

"

"

"

" "

" " "

! "

"

"

"

"

" "

"

"

" "

"

"

"

" "

"

"

"

" "

"

"

"

"

"

"

"

"

"

"

" "

"

"

"

"

"

" "

"

"

"

"

"

"

"

"

"

" "

"

" " "

"

"

"

"

"

"

" "

" "

"

"

"

"

"

"

" "

" "

"

"

"

"

!

!

4

!

!

!

! !

"

" " "

100 m

"

!

! ! !

(c) > 65 dB > 62 dB > 59 dB > 56 dB > 53 dB > 50 dB > 47 dB > 44 dB

1 km

Fig. 1. Maps of the Buunderkamp area showing nest-boxes, sampling locations and noise levels. Motorway (triple line) and railway (dashed line) are shown. a) nest-box distribution (small dots). Only breeding data from nest-boxes within the rectangle was used. b) sampling locations (ﬁlled rectangles) along 10 transects (open rectangles, 2 of them shown). Numbers refer to locations of example recordings used in Fig. 2. c) GIS-map showing spatial variation in sound levels. Traﬃc noise shows a strong decrease with distance from the motorway (absolute range at sampling locations 46Æ5–67Æ8 dB SPL, A-weighted), but there is substantial spatial variation in this decline. that would have aﬀected the spatial spread of noise coming from the motorway.

NOISE DATA ACQUISITION

We made sound recordings between March and May 2008, before major leaﬁng of the deciduous trees. We sampled sound levels along ten transects perpendicular to the motorway (Fig. 1b), with automatic SongMeter recorders (16 bit, 24 kHz sample rate; Wildlife Acoustics Inc., Concorde, MA, USA). Exact sampling locations were determined with a GPS (Garmin 60CSx, Olathe, KS, USA). The sampling transects started 100 m from the mid-line of the motorway and six sampling locations at approximately 100 m intervals were chosen within each transect. The transects were spaced 80–100 m apart and two transects were sampled simultaneously for 3–5 consecutive days. Transects were each sampled twice in a random order, once between 8th and 30th of March, and once between 31st of March and 1st of May. The sampling grid encompassed most

of the area, but we used two additional SongMeters to monitor the remaining area. Recorders were attached to large trees (>40 cm in diameter) at 2 m above the ground with the recording microphone directed towards the motorway. Recording levels for the microphones were adjusted to a sensitivity ranging from 0Æ0 to 1Æ5 dBV pa)1 (reaching full scale between 92Æ5 and 94Æ0 dB SPL) and amplitude levels were adjusted according to the eﬀective sensitivity of each individual Song Meter recorder. Recorders were randomly swapped between sampling locations to control for any remaining variation in recording levels. Recorders were scheduled to record for 30 s at 30 min intervals, day and night. We analysed sound recordings in the computer program Matlab (Mathworks Inc., Natick, MA, USA). We measured overall sound levels (using an A-weighted ﬁlter), and also sound levels in four adjacent octave-bands, centred at frequencies of 0Æ5, 1Æ0, 2Æ0, and 4Æ0 kHz. Sound measurements were averaged over either 30-min or 24-h intervals, and ⁄ or sampling locations, depending on the type of analysis.

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 210–219

Traﬃc noise and avian reproductive success 213 We used 76 sampling locations to visualize spatial variation in noise levels for the Buunderkamp in the computer program ArcGis (version 9Æ0, ESRI). Sixty locations from the sampling transects and 16 additional sampling locations were plotted onto a geo-annotated reference map from which noise maps were derived with the Spatial Analyst toolbox. Spatial resolution was set at 5 m and raster values between sampling locations were calculated with a weighted distance interpolation tool (IDW). Additionally we calculated distances for all nest boxes and sampling locations to the nearest mid-point on the motorway. We assessed the temporal overlap between traﬃc noise and vocal bird activity throughout the season and at diﬀerent times of day. At our study site most of the non-anthropogenic sound comes from vocalizing birds with the majority of acoustic energy in the range of 2–8 kHz. We selected a subset of sampling locations at distances over 400 m from the motorway where there is little traﬃc noise present in the 4 kHz octave band so temporal variation in sound levels was mainly related to the vocal activity by birds. For these locations, we compared sound levels, averaged over 1 or 24 h intervals, in the 1 kHz band (mainly due to traﬃc noise) with those in the 4 kHz band (mainly due to bird activity, including great tits).

LONG-TERM BREEDING DATA

Great tit breeding data were collected between 1995 and 2009 by the Netherlands Institute of Ecology (NIOO-KNAW). We used data from both large and small nest-boxes within the sampling grid (Fig 1a) on laying date, clutch size, number of hatchlings, number of ﬂedglings and ﬂedging mass (average weight of chicks for the brood when chicks are 15 days old) for all ﬁrst great tit clutches over this period, except for 2007 and 2008 when data were excluded because of an unrelated experiment. Additional data on female identity, female age and ﬂedging mass were only available for 1995–1999, 2001 and 2009. For analysis of breeding performance we used only ﬁrst clutches (categorised using female identity or because laying date was within 30 days of the ﬁrst laying date for a given year). For analyses of laying date we used only clutches for which this could be reliably calculated. We were interested in the mechanisms underlying breeding success and therefore focused on life history traits that reﬂected decisions made by the birds. For the analysis of clutch size we therefore excluded clutches that were not incubated, because including nests that were abandoned (either through a decision by the parents, or predation of the parents) would introduce unwanted heterogeneity in the data. Similarly, we excluded nests where no chicks hatched or ﬂedged from the analyses of the number of hatchlings and ﬂedglings, respectively, because it was usually unknown whether failure was caused by death of all the embryos or chicks, abandonment by the parents or predation of the parents (away from the nest).

WEATHER DATA AND HABITAT MEASUREMENTS

We assessed habitat characteristics, including tree density, tree diameter and species composition, at the level of woodland plots (0Æ2– 1Æ0 ha). We measured tree density and diameter and noted tree species at each of the 60 sampling locations, and at the 2 nest boxes nearest to these locations. We calculated the percentage of deciduous trees per plot and averaged tree density and diameter over all locations within a plot. We used weather data on daily wind direction and speed, and temperature, recorded by the Royal Netherlands Meteorological

Institute (KNMI) at de Bilt (situated ±50 km to the west of the Buunderkamp).

STATISTICAL ANALYSIS

We analysed all data using SPSS (version 17Æ0) and log-transformed variables when necessary to meet model assumptions. Temporal variation in daily and seasonal sound levels were explored using repeated measures anovas with sound level grouped by sampling location as the dependent variable and time of day or date as an explanatory variable. Additionally, we compared recordings made on weekdays with recordings from weekends with type of day as a ﬁxed factor. We examined the eﬀect of daily weather conditions on the propagation of noise with full factorial linear mixed models. To test for the eﬀect of wind direction we discriminated between days with northerly (coming from the direction of the motorway) and southerly winds (going towards the direction of the motorway). Wind direction was included as a ﬁxed factor, and sample location as a random factor. Distance to the motorway, wind speed, and daily temperature were included as covariates. We constructed a set of linear mixed models for each life history trait and compared them using a model selection approach based on Akaike’s information criterion (Burnham & Anderson 2002). Models always included nest-box type (large or small), sampling location and breeding year as random factors. Depending on the model, we also included other reproductive traits as explanatory variables (cf. Wilkin et al. 2006). For instance, clutch size can correlate with laying date and an eﬀect of noise on clutch size could be indirectly caused by an eﬀect of noise on laying date. Including laying date in the clutch size model therefore allows us to test for a direct eﬀect of noise. For the number of hatchlings we included clutch size and for the number of ﬂedglings we included number of hatchlings in the models. For the ﬂedging mass model we included both clutch size and laying date as these factors are known to have a large eﬀect on ﬂedging mass (e.g. Wilkin et al. 2006). In a ﬁrst analysis we compared models that included overall noise levels, distance to the motorway, tree density, tree diameter and percentage deciduous trees as explanatory factors only. Models contained single factors or in combination with other factors as main eﬀects as we had no a priori knowledge that interactions among factors would be of importance. The total set contained 32 models to be compared for each trait, including the Null model. We calculated for each explanatory factor the probability that it would be in the best approximating model using Akaike weights (see e.g. Whittingham et al. 2005; Garamszegi et al. 2009). We used the subset of models with a delta-AIC < 4Æ0 from the top model to get model-averaged estimates and standard errors for each factor (cf. Burnham & Anderson 2002). In a second analysis we focused on temporal overlap between noise sampling period and breeding stage. We used the models with delta-AIC < 4Æ0 from the previous analysis and only exchanged the overall noise with noise levels sampled either in March or in April. In a third analysis, we repeated this procedure, but focused on the spectral overlap with song and explored whether noise in a certain frequency range (0Æ5, 1Æ0, 2Æ0, or 4Æ0 octave band, or overall noise) better explained variation of the data. Breeding performance is known to be age-dependent (Kluyver 1951; Wilkin et al. 2006) and we therefore re-ran analyses for which we found strong support using the subset of data for which female age was known. Female identity was added as a random factor and female age (ﬁrst year or older) as a ﬁxed factor.

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 210–219

214 W. Halfwerk et al. distances from the motorway. For instance, at 700 m from the motorway, sound levels below 1 kHz could increase by over 10 dB SPL on cold days or days with northerly winds (Fig. 2c,d).

Results SPATIAL PATTERNS IN NOISE LEVELS

Overall sound levels gradually decreased with distance from the motorway (F5,54 = 200Æ5, P < 0Æ001) with an average drop of 20 dB SPL (A-weighted) over less than 500 m (Fig. 1c). Furthermore, high frequencies attenuated faster than low frequencies (F3,59 = 12Æ03, P < 0Æ001; Fig. 2a). There was substantial spatial variation in traﬃc noise, independent of distance to the motorway (Fig. 1c): diﬀerent locations at medium (>300 m) to large (>700 m) distances from the motorway diﬀered by more than 9 dB SPL (A-weighted) in noise level (Fig. 1c). Train noise can be very loud (see e.g. Fig. 2b) but, in contrast to motorway noise, is transient and average daily noise levels near the railway line were among the lowest (Fig. 1c).

TEMPORAL FLUCTUATIONS IN TRAFFIC NOISE LEVELS AND THE OVERLAP WITH BIRD ACTIVITY

Traﬃc noise levels changed throughout the season (F1,59 = 7Æ57 P = 0Æ008) with March being noisier and more variable than April (Fig. 3a). Additionally, noise levels on weekdays were signiﬁcantly higher than at the weekend (F1,59 = 4Æ87 P = 0Æ032; Fig. 3). Noise levels showed a strong daily pattern (F1,59 = 8Æ776 P = 0Æ005), with a clear drop between 0:00 and 4:00 AM, but no distinct rush-hour peaks (Fig. 3b). Screening of recordings revealed that, at distances over 400 m from the motorway, variation in sound levels in the 4 kHz band was indeed mainly inﬂuenced by bird vocal activity, and we therefore used recordings at these distances to assess seasonal and daily overlap of traﬃc noise and bird vocal behaviour. Bird vocal activity as measured at the peak of the dawn chorus increased throughout the season (4 kHz-band; F1,59 = 7Æ88, P < 0Æ001) whereas traﬃc noise during this time period decreased (1 kHz-band; F1,59 = 5Æ13, P < 0Æ001; Fig. 3a). Bird vocal behaviour showed a temporal shift between early March and late April due to changes in the time of sunrise, but despite this, the temporal overlap with traﬃc noise remained remarkably high on weekdays (Fig. 3b), probably due to the change from winter to summer time (i.e. clock

WEATHER-DEPENDENT NOISE LEVELS

Wind direction, wind speed and daily temperature all had an eﬀect on overall sound levels (see Table 1). Furthermore, wind direction and temperature interacted with distance to the motorway (Table 1). We reanalysed a subset of recordings made at distances of 400–700 m from the motorway to explore the eﬀect of weather conditions on sounds in diﬀerent octave bands. Both temperature (F1,59 = 27Æ78; P < 0Æ0001) and wind direction (F1,59 = 5Æ27; P = 0Æ001) interacted with frequency, with the strongest eﬀect at lower frequencies and large

70

70 (b)

50

40

40

30

30

20

20

10

60

1

2

3

4

5

6

Cold

(c)

1

April 16, 2008 16:00;6·4 °C

50

April 25, 2008 16:00;12·6 °C

60

20

20

10

10

0 2

3

4

5

2

3

6

0

4

5

Northern

(d)

April 18, 2008 16:00

6 2 C

Southern

40

E

30

1

1

50

30

0

0

1

Warm

40

B

10

4

4 D April 21, 2008 19:30

60

50

0

Train No train

1

April 21, 2008 D 20:00

60

Noise level (dB SPL)

3

Near Far

(a)

March 31, 2008 16:00

0

Frequency (kHz)

1

2

3

4

5

2 A

6

Fig. 2. Variation in sound proﬁles across different environmental conditions. a) powerspectrographic example comparing sound proﬁles near to (±100 m), and far from (±700 m), the motorway. At larger distances, the high-frequency components of traﬃc noise are more attenuated and even disappear above ±3 kHz. b) recordings made near the railway (±100 m from the track and ±1 km from the motorway) shortly before and during the passage of a train. c) comparison of sound proﬁles on days with diﬀerent temperatures, but similar wind conditions illustrates large eﬀect of weather conditions on noise levels. d) comparison of sound proﬁles on days with opposite wind directions, but similar temperature and wind speed. Small numbers refer to locations illustrated in Fig. 1. Capital letters refer to recording days illustrated in Fig. 3.

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 210–219

Traﬃc noise and avian reproductive success 215 and 3) and virtually no support in the remaining life history models. Overall noise levels had an independent negative eﬀect on clutch size, with females laying on average about 10% fewer eggs across a noise gradient of 20 dB SPL (A-weighted) (Table 3). Reanalysing the top clutch size model to include female identity and age conﬁrmed the eﬀect of noise (F1,268 = 7Æ82, P = 0Æ007), but failed to show an eﬀect of female age on clutch size (F1,268 = 0Æ20, P = 0Æ82). Noise levels had a negative eﬀect on ﬂedging mass (Table 3), but in none of the top models was the eﬀect signiﬁcant (all P > 0Æ2).

Table 1. Results from mixed model showing eﬀect of weather condition on overall noise levels. Sampling location (N = 60) was added as random factor. Only ﬁrst order interactions are reported Source

d.f.

F

P

Distance Wind direction (N vs. S) Daily temperature Wind speed* Distance · Wind direction Distance · Daily temperature Distance · Wind speed* Wind direction · Daily temperature Wind direction · Wind speed* Daily temperature · Wind speed*

5 1 1 1 5 5 5 1 1 1

6Æ61 10Æ92 9Æ65 29Æ30 3Æ81 2Æ73 1Æ75 1Æ32 10Æ38 11Æ26

<0Æ001 0Æ001 0Æ002 <0Æ001 0Æ002 0Æ019 0Æ12 0Æ25 0Æ001 0Æ001

TEMPORAL AND SPECTRAL VARIATION IN NOISE PREDICTS SMALLER CLUTCHES AND FEWER

*log-transformed.

FLEDGLINGS

Reﬁning the models with noise sampled either in March or in April did not change the level of support, except for the number of ﬂedglings model (Tables 4 and 5). Noise sampled in April was about seven times more likely to explain variation of the data compared to noise sampled in March (Table 5). Higher noise levels in April correlated with lower numbers of ﬂedglings (Table 5). We re-ran the top model to include clutch size instead of the number of ﬂedglings as ﬁxed factor. Clutch size had a large eﬀect on the number of ﬂedglings (B = 0Æ57 ± 0Æ070; F1,364 = 65Æ51, P < 0Æ0001), but we found noise sampled in April to have an additional negative eﬀect (B = )0Æ061 ± 0Æ027; F1,364 = 5Æ09, P = 0Æ028) as well. Finally, we found that variation in noise levels in the 2 kHz octave band best explained variation in clutch size, although

time advancing by 1 h on 30 March). Peak activity of avian vocal behaviour showed the least overlap with traﬃc noise during the weekends, especially in late April (Fig. 3b).

NEGATIVE EFFECT OF TRAFFIC NOISE ON BREEDING PERFORMANCE

Overall noise levels received strong support in the model selection procedure for clutch size and ﬂedging mass models and moderate support for the number of ﬂedglings model (Tables 2 and 3). Tree diameter and tree density received strong support in all life history models (Tables 2 and 3), but the eﬀect was not consistent across models and the variance was high (Table 3). Distance to the highway and percentage deciduous trees received weak support in the ﬂedging mass model (Tables 2 60

(a)

Traffic noise Bird sounds Weekend

Noise level (dB SPL)

50

40

A

30

B

March 60

(b)

Week

60

C

D

Weekend

60

Week

60

50

50

50

50

40

40

40

40

30

30

30

30

20

0 2 4 6 8 10 12 14 16 18 20 22 24

20

E

April

0 2 4 6 8 10 12 14 16 18 20 22 24

20

0 2 4 6 8 10 12 14 16 18 20 22 24

20

Weekend

0 2 4 6 8 10 12 14 16 18 20 22 24

Time of day (h)

Fig. 3. Temporal patterns in traﬃc noise levels and bird vocal activity. Recordings made at distances of 400–700 m from the motorway, averaged over 1 or 24 h intervals. We compared amplitude ﬂuctuations in the 1 and 4 kHz band, which are mainly inﬂuenced by traﬃc and bird vocal activity respectively. a) seasonal pattern in sound levels between March and May 2008. Recordings made at dawn during peak singing activity. Traﬃc noise levels decrease throughout the season, but show substantial variation due to changes in traﬃc activity (week days being noisier than weekend days) and changes in weather conditions. Bird vocal activity increases throughout the season. b) daily pattern in traﬃc noise and bird sound levels during either the period of great tit territory defense (8–16 of March, left two graphs) or egg-laying (19–27 of April, right two graphs), on weekdays or at the weekend. 2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 210–219

216 W. Halfwerk et al.

Dependent trait

Model

AIC

D AIC

Akaike weight

Laying date (N = 542)

d+t d Noise+d+t

3523Æ81 3525Æ61 3526Æ86

0Æ00 1Æ80 3Æ06

0Æ52 0Æ21 0Æ11

Clutch size (N = 505)

Noise+d Noise+d+t Noise d d+t Noise+t

1727Æ51 1727Æ92 1729Æ43 1730Æ05 1730Æ41 1730Æ41

0Æ00 0Æ41 1Æ92 2Æ54 2Æ90 2Æ90

0Æ32 0Æ26 0Æ12 0Æ09 0Æ07 0Æ07

Number of hatchlings (N = 470)

d d+t Null t Noise+d Noise+d+t

917Æ53 917Æ71 920Æ18 921Æ05 921Æ19 921Æ38

0Æ00 0Æ18 2Æ65 3Æ53 3Æ66 3Æ85

0Æ36 0Æ33 0Æ10 0Æ06 0Æ06 0Æ05

Number of ﬂedglings (N = 387)

d d+t Noise+d Noise+d+t Null t

1956Æ24 1956Æ65 1957Æ61 1958Æ08 1958Æ98 1959Æ22

0Æ00 0Æ42 1Æ37 1Æ85 2Æ74 2Æ99

0Æ29 0Æ24 0Æ15 0Æ12 0Æ07 0Æ07

Fledging mass (N = 215)

Noise+d+t+%d Noise+d+t Noise+d+t+dH+%d

2070Æ29 2072Æ26 2072Æ86

0Æ00 1Æ96 2Æ57

0Æ52 0Æ19 0Æ14

Table 2. Life history model selection procedure based on traﬃc noise, distance and ⁄ or habitat features using Akaike’s information criterion. Overall noise level (Noise), distance to the highway (dH), tree diameter (d), tree density (t) and percentage of deciduous trees (%d) were entered as main eﬀects in mixed models. Only models with a D AIC < 4Æ0 are shown for each life history trait

Table 3. Results from model selection procedure showing selection probabilities (calculated across the whole model set) and parameter estimates (using a subset of the models with D AIC < 4Æ0 and model averaging procedures; see text and Table 2). Only factors that were used for model averaging are shown

overall noise and noise in the 0Æ5 and 1Æ0 kHz band also received moderate support (Table 6a). Noise in the 2 kHz band frequency range overlaps the lower part of great tit song in our study population and had a negative eﬀect on the number of eggs laid by females (Table 6b).

Dependent trait ⁄ independent parameter

Selection probability

B

SE

Discussion

Laying date Tree diameter Tree density Noise

0Æ92 0Æ70 0Æ18

)1 07 0Æ34 0Æ044

3 92 0Æ93 0Æ075

Clutch size Noise Tree diameter Tree density

0Æ80 0Æ75 0Æ42

)0Æ053 0Æ18 0Æ17

0Æ021 1Æ23 0Æ25

Number of hatchlings Tree diameter Tree density Noise

0Æ81 0Æ46 0Æ14

0Æ72 )0Æ05 )0Æ039

1Æ64 0Æ35 0Æ030

Number of ﬂedglings Tree diameter Tree density Noise

0Æ80 0Æ45 0Æ33

)0Æ75 )0Æ18 )0Æ044

0Æ83 0Æ15 0Æ020

Fledging mass Tree diameter Tree density Noise Distance to highway Percentage deciduous

0Æ99 0Æ99 0Æ93 0Æ22 0Æ16

145Æ1 )10Æ87 )3Æ14 0Æ005 0Æ56

151Æ7 26Æ96 2Æ67 0Æ11 0Æ39

We recorded high traﬃc noise levels in forest bird breeding habitat related to the proximity of a motorway. However, we also found spatial variation in noise levels independent of distance to the motorway that allowed us to demonstrate a negative relationship between noise levels and the reproductive success of great tits. Furthermore, noise levels varied substantially with the time of day, season and weather conditions, and both temporal and spectral overlap with vocalizing birds is high under a wide range of conditions. Finally, we found noise levels in April to have a negative eﬀect on the number of ﬂedglings, while noise variation in the frequency with most spectral overlap with great tit song best predicted a negative eﬀect on clutch size.

EXPLAINING NOISE IMPACT ON REPRODUCTIVE SUCCESS

We found an impact of traﬃc noise on avian reproductive success manifest by smaller clutches and fewer ﬂedged chicks in the noisier areas. We also explored relationships between breeding traits and temporal and spectral overlap of noise,

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 210–219

Traﬃc noise and avian reproductive success 217 Table 4. Results from model selection procedure focusing on temporal variation in noise. Models as used in Table 2 were adjusted to include noise levels recorded either in March or April Dependent trait

Model

AIC

D AIC

Akaike weight

Laying date (N = 542)

d+t d Noise Noise Noise Noise

April+d+t March+d+t April+d March+d

3523Æ81 3525Æ61 3527Æ03 3527Æ30 3528Æ85 3529Æ14

0 1Æ80 3Æ23 3Æ49 5Æ05 5Æ34

0Æ52 0Æ21 0Æ10 0Æ09 0Æ04 0Æ04

Clutch size (N = 505)

Noise Noise Noise Noise Noise Noise

March+d April+d April+d+t March+d+t March April

1725Æ97 1726Æ20 1726Æ56 1726Æ79 1728Æ05 1728Æ11

0 0Æ23 0Æ59 0Æ82 2Æ08 2Æ14

0Æ25 0Æ22 0Æ19 0Æ17 0Æ09 0Æ09

Number of hatchlings (N = 470)

d d+t Null t

917Æ53 917Æ71 920Æ18 921Æ05

0 0Æ18 2Æ65 3Æ53

0Æ36 0Æ33 0Æ09 0Æ06

Number of ﬂedglings (N = 387)

Noise April+d

1955Æ34

0

0Æ32

Noise April+d+t d d+t

1956Æ08 1956Æ24 1956Æ65

0Æ74 0Æ89 1Æ31

0Æ22 0Æ21 0Æ17

Noise April+d+t+%d

2070Æ16

0

0Æ39

Noise Noise Noise Noise Noise

2071Æ12 2072Æ30 2073Æ01 2073Æ13 2073Æ84

0Æ96 2Æ14 2Æ85 2Æ97 3Æ68

0Æ24 0Æ13 0Æ09 0Æ09 0Æ06

Fledging mass (N = 215)

Marchl+d+t+%d April+d+t March+d+t April+d+t+dH+%d March+d+t+dH+%d

Table 5. Temporal variation in noise is related to breeding performance. Selection probabilities and parameter estimates of noise recorded either in March or April from model selection procedures are shown (see text and Table 4) Noise parameter (sampling period)

Selection probability

B

SE

Laying date

March April

0Æ13 0Æ14

0,007 0,024

0,050 0,059

Clutch size

March April

0Æ50 0Æ50

)0,038 )0Æ040

0,020 0Æ020

Number of hatchlings

March April

0Æ07 0Æ09

)0Æ027 )0Æ032

0Æ027 0Æ029

Number of ﬂedglings

March April

0Æ08 0Æ55

)0Æ033 )0Æ051

0Æ019 0Æ019

Fledging mass

March April

0Æ39 0Æ61

)1Æ97 )3Æ16

1Æ80 2Æ48

Dependent trait

which could provide some insight into the mechanisms by which birds are aﬀected. We believe there are at least four possible mechanisms, all related to signal masking to some degree, which could explain how anthropogenic noise has a negative impact on avian reproductive success.

The ﬁrst explanation is related to interference with acoustic assessment of mate quality. Female birds are known to rely on song in assessment of male quality and subsequent investment decisions (Holveck & Riebel 2009). High noise levels could reduce perceived song quality and cause females to breed later, allocate less energy to the eggs or provide less maternal care to the chicks. Our data show that spectral overlap between noise and great tit song best predicts patterns in clutch size, suggesting that noise may indeed interfere with song-based assessment of male quality and subsequently lower female investment. The second explanation for the eﬀect of traﬃc noise on reproductive success could be related to the non-random distribution of individuals across the habitat. Birds may perceive a noisy territory as being of lesser quality (Slabbekoorn & Ripmeester 2008) and therefore try to avoid these areas. For instance, both Reijnen & Foppen (1991) and Habib, Bayne & Boutin (2007) found less experienced birds breeding in more noisy territories. We did not ﬁnd traﬃc noise or clutch size to covary with female age and we have no insight into distribution and performance of lower quality individuals (e.g. immigrants, who are known to produce smaller clutches; Kluyver 1951), but it is likely that noise may play an important role at the time that individuals are settling and defending territories. The third explanation is that increased noise levels could also cause physiological stress due to reduced foraging

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 210–219

218 W. Halfwerk et al. Table 6. Spectral overlap between noise and song predicts clutch size. a) model selection using clutch size models with strong support in previous analysis (see text and Table 2). Only models with a D AIC < 4Æ0 are shown. b) Selection probabilities for noise in diﬀerent frequency ranges and parameter estimates after model averaging. Only results for noise variables are shown a)

Model (Noise frequency range)

AIC

D AIC

Akaike weight

2 kHz band+d+t overall (A-weighted)+d+t 0Æ5 kHz band+d+t 1 kHz band+d+t 4 kHz band+d+t 2 kHz band+d b)

1726Æ59 1727Æ92 1728Æ14 1728Æ62 1729Æ57 1730Æ27

0 1Æ33 1Æ54 2Æ02 2Æ98 3Æ67

0Æ28 0Æ15 0Æ13 0Æ10 0Æ06 0Æ05

Noise frequency range

Selection probability

B

SE

2 kHz band overall (A-weighted) 0Æ5 kHz band 1 kHz band 4 kHz band

0Æ38 0Æ20 0Æ18 0Æ15 0Æ09

)0Æ058 )0Æ053 )0Æ070 )0Æ064 )0Æ069

0Æ016 0Æ021 0Æ022 0Æ018 0Æ027

opportunities, because prey are less easy to detect (Schaub, Ostwald & Siemers 2008), or because more time has to be spent scanning for predators (Quinn et al. 2006). Individuals living in noisy areas may therefore have less energy to invest in their eggs and oﬀspring. And ﬁnally, the fourth explanation could be that noise can have an impact on parent-oﬀspring communication and adults may therefore not be able to meet their chicks’ demands (Leonard & Horn 2008). We did not ﬁnd a signiﬁcant eﬀect on ﬂedging mass, but we did ﬁnd that high noise levels in April have a negative eﬀect on the number of ﬂedglings, independent of clutch size. Whether this is related to higher stress levels, reduced foraging or decreased communication is diﬃcult to disentangle, but it does suggest that noise interference could aﬀect food provisioning to the chicks.

EXPLAINING TRAFFIC NOISE HETEROGENEITY

The opportunity to test for an impact of traﬃc noise on avian reproductive success relied on the heterogeneity of noise levels independent of distance to the motorway. Many earlier studies have designed ways to predict spatial and temporal variation of traﬃc noise, using a combination of ﬁeld data and theoretical modelling (Steele 2001). However, these models have tended to focus either on noise data at the source (taking traﬃc and road variables into account; e.g. Li et al. 2002; Parris & Schneider 2009) or on transmission data (e.g. Ovenden, Shaﬀer & Fernando 2009). The few models that have integrated these aspects have assumed that the areas adjacent to motorways are environmentally homogeneous (Steele 2001). In contrast, our

study reveals a high level of heterogeneity at a local scale that should be taken into account when trying to understand the impact of noise on bird breeding populations. In addition to revealing the pattern of noise heterogeneity, we were able to provide some insight into the causal explanations for the noise variation in space, time, and frequency. We found substantial spatial variation throughout our study area that was not related to the distance to the motorway. The eﬀect was most pronounced at a few hundred metres from the motorway, with nearby areas diﬀering by over 9 dB in mean noise levels. Transmission of traﬃc noise is known to depend on motorway architecture, and ground and vegetation structure (Bucur 2006). However, the architecture of the motorway does not vary over the length adjacent to our study area and the spatial noise heterogeneity that we found is most likely to be caused by variation in tree densities in the areas close to the motorway (Padgham 2004). Noise levels close to the motorway source are known to depend on traﬃc load (Parris & Schneider 2009) which can vary between day and night, and between weekdays and the weekend (Bautista et al. 2004). Noise amplitude is also strongly related to traﬃc speed (Makarewicz & Kokowski 2007), which is probably why we did not detect a clear rush-hour peak in noise, because traﬃc during the rush-hour is often much slower or even stationary. Finally, we not only conﬁrmed that lower frequency sounds were transmitted over a larger area than higher frequency sounds but that relatively low frequencies were also more inﬂuenced by changing weather conditions.

Conclusions We have shown that traﬃc noise levels in roadside forest vary substantially in space, time and frequency, which allowed us to reveal a negative relationship with reproductive success in a common species. Great tit females laid fewer eggs and pairs ﬂedged fewer young in noisier areas. As the impact of noise is potentially even higher for species vocalizing at lower frequencies than great tits our data could have signiﬁcance for the conservation of species that are less abundant or under threat. Consequently, we believe that integration of data on speciesspeciﬁc acoustic behaviour with noise prediction models and actual ﬁeld measurements could be a useful approach in exploring ways to protect threatened birds in noise-polluted wildlife sanctuaries. Mitigation measures to reduce the negative impact of noise on breeding birds could include sound barriers (Slabbekoorn & Ripmeester 2008), alternative, more sound-eﬃcient transport by buses through nature reserves (Laube & Stout 2000) or closing roads during acoustically critical phases in the breeding cycle (Groot Bruinderink et al. 2002). Traﬃc noise could also be reduced by introducing a ‘noise tax’ for a given time of day or season based on the type of car or tyres and the average vehicle speed – factors that are known to aﬀect noise levels (Makarewicz & Kokowski 2007). It is clear that the trade-oﬀ between ecological and economic values will play a crucial role in the implementation of these kinds of applications. Furthermore, suﬃcient insight into species-speciﬁc acoustic behaviour and

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 210–219

Traﬃc noise and avian reproductive success 219 noise distribution data is typically still lacking. Nevertheless, we hope our results help to raise awareness of the potentially negative impact of anthropogenic noise on breeding birds in general.

Acknowledgements We thank C. ten Cate, H. Kunc and three anonymous referees for critically reviewing previous versions of the manuscript. This study was funded through a NWO grant (no. 817Æ01Æ003) to H.S and material support from the ‘Dobberke stichting’ to W.H.

References Barber, J.R., Crooks, K.R. & Fristrup, K.M. (2009) The costs of chronic noise exposure for terrestrial organisms. Trends in Ecology & Evolution, 25, 180– 189. Bautista, L.M., Garcia, J.T., Calmaestra, R.G., Palacin, C., Martin, C.A., Morales, M.B., Bonal, R. & Vinuela, J. (2004) Eﬀect of weekend road traﬃc on the use of space by raptors. Conservation Biology, 18, 726–732. Bayne, E.M., Habib, L. & Boutin, S. (2008) Impacts of Chronic Anthropogenic Noise from Energy-Sector Activity on Abundance of Songbirds in the Boreal Forest. Conservation Biology, 22, 1186–1193. Bee, M.A. & Swanson, E.M. (2007) Auditory masking of anuran advertisement calls by road traﬃc noise. Animal Behaviour, 74, 1765–1776. Bradbury, J.W. & Vehrencamp, S.L. (1998) Principles of Animal Communication. Sinauer Associates, Sunderland, MA. Brumm, H. & Slabbekoorn, H. (2005) Acoustic communication in noise. Advances in the Study of Behavior, Vol 35 (eds P.J.B. Slater, C.T. Snowdown, H.J. Brockmann & M. Naguib), pp. 151–209. Elsevier Academic Press, San Diego, USA. Bucur, V. (2006) Urban Forest Acoustics. Springer-Verlag, Berlin, Heidelberg. Burnham, K.P. & Anderson, D.R. (2002) Model Selection and Multimodel Inference: A Practical-Theoretic Approach, 2nd edn. Springer-Verlag, Berlin. Catchpole, C.K. & Slater, P.J.B. (2008) Bird Song: Biological Themes and Variations. Cambridge University Press, Cambridge. Drent, P.J. (1987) The importance of nestboxes for territory settlement, survival and density of the great tit. ARDEA, 75, 59–71. Forman, R.T.T. (2000) Estimate of the area aﬀected ecologically by the road system in the United States. Conservation Biology, 14, 31–35. Francis, C.D., Ortega, C.P. & Cruz, A. (2009) Cumulative consequences of noise pollution: noise changes avian communities and species interactions. Current Biology, 19, 1415–1419. Garamszegi, L.Z., Calhim, S., Dochtermann, N., Hegyi, G., Hurd, P.L., Jorgensen, C., Kutsukake, N., Lajeunesse, M.J., Pollard, K.A., Schielzeth, H., Symonds, M.R.E. & Nakagawa, S. (2009) Changing philosophies and tools for statistical inferences in behavioral ecology. Behavioral Ecology, 20, 1363– 1375. Groot Bruinderink, G.W.T.A., Brandjes, G.J., van Eekelen, R., Niewold, F.J.J., ten Den, P.G.A. & Waardenburg, H.W. (2002) Faunabeheerplan Nationaal Park Sallandse Heuvelrug i.o. Alterra, Research Instituut voor de Groene Ruimte, Wageningen. Habib, L., Bayne, E.M. & Boutin, S. (2007) Chronic industrial noise aﬀects pairing success and age structure of ovenbirds Seiurus aurocapilla. Journal of Applied Ecology, 44, 176–184. Halfwerk, W. & Slabbekoorn, H. (2009) A behavioural mechanism explaining noise-dependent frequency use in urban birdsong. Animal Behaviour, 78, 1301–1307. Holveck, M.J. & Riebel, K. (2009) Low-quality females prefer low-quality males when choosing a mate. Proceedings of the Royal Society B-Biological Sciences, 277, 153–160. Junker-Bornholdt, R., Wagner, M., Zimmermann, M., Simonis, S., Schmidt, K.H. & Wiltschko, W. (1998) The impact of a motorway in construction and after opening to traﬃc on the breeding biology of Great Tit (Parus major) and Blue Tit (P. caeruleus). Journal Fur Ornithologie, 139, 131–139. Kluyver, H.N. (1951) The population ecology of the great tit, Parus major L. ARDEA, 39, 1–135.

Laube, M.M. & Stout, R.W. (2000) Grand Canyon National Park – assessment of transportation alternatives. Transit: Planning, Intermodal Facilities, Management, and Marketing, 1735, 59–69. Leonard, M.L. & Horn, A.G. (2008) Does ambient noise aﬀect growth and begging call structure in nestling birds? Behavioral Ecology, 19, 502–507. Li, B.G., Tao, S., Dawson, R.W., Cao, J. & Lam, K. (2002) A GIS based road traﬃc noise prediction model. Applied Acoustics, 63, 679–691. Makarewicz, R. & Kokowski, P. (2007) Prediction of noise changes due to trafﬁc speed control. Journal of the Acoustical Society of America, 122, 2074– 2081. Mockford, E.J. & Marshall, R.C. (2009) Eﬀects of urban noise on song and response behaviour in great tits. Proceedings of the Royal Society B-Biological Sciences, 276, 2979–2985. Ovenden, N.C., Shaﬀer, S.R. & Fernando, H.J.S. (2009) Impact of meteorological conditions on noise propagation from freeway corridors. Journal of the Acoustical Society of America, 126, 25–35. Padgham, M. (2004) Reverberation and frequency attenuation in forests–implications for acoustic communication in animals. Journal of the Acoustical Society of America, 115, 402–410. Parris, K.M. & Schneider, A. (2009) Impacts of traﬃc noise and traﬃc volume on birds of roadside habitats. Ecology and Society, 14, 29. Pohl, N.U., Slabbekoorn, H., Klump, G.M. & Langemann, U. (2009) Eﬀects of signal features and environmental noise on signal detection in the great tit, Parus major. Animal Behaviour, 78, 1293–1300. Quinn, J.L., Whittingham, M.J., Butler, S.J. & Cresswell, W. (2006) Noise, predation risk compensation and vigilance in the chaﬃnch Fringilla coelebs. Journal of Avian Biology, 37, 601–608. Reijnen, R. & Foppen, R. (1991) Eﬀect of road traﬃc on the breeding site tenacity of male willow warblers (Phylloscopus trochilus). Journal Fur Ornithologie, 132, 291–295. Reijnen, R., Foppen, R., Terbraak, C. & Thissen, J. (1995) The eﬀects of car trafﬁc on breeding bird populations in woodland. 3. Reduction of density in relation to the proximity of main roads. Journal of Applied Ecology, 32, 187–202. Schaub, A., Ostwald, J. & Siemers, B.M. (2008) Foraging bats avoid noise. Journal of Experimental Biology, 211, 3174–3180. Slabbekoorn, H. & den Boer-Visser, A. (2006) Cities change the songs of birds. Current Biology, 16, 2326–2331. Slabbekoorn, H. & Halfwerk, W. (2009) Behavioural ecology: noise annoys at community level. Current Biology, 19, R693–R695. Slabbekoorn, H. & Peet, M. (2003) Ecology: birds sing at a higher pitch in urban noise. Nature, 424, 267–267. Slabbekoorn, H. & Ripmeester, E.A.P. (2008) Birdsong and anthropogenic noise: implications and applications for conservation. Molecular Ecology, 17, 72–83. Steele, C. (2001) A critical review of some traﬃc noise prediction models. Applied Acoustics, 62, 271–287. Stone, E. (2000) Separating the noise from the noise: a ﬁnding in support of the ‘‘Niche Hypothesis,’’ that birds are inﬂuenced by human-induced noise in natural habitats. Anthrozoos, 13, 225–231. Warren, P.S., Katti, M., Ermann, M. & Brazel, A. (2006) Urban bioacoustics: it’s not just noise. Animal Behaviour, 71, 491–502. Whittingham, M.J., Swetnam, R.D., Wilson, J.D., Chamberlain, D.E. & Freckleton, R.P. (2005) Habitat selection by yellowhammers Emberiza citrinella on lowland farmland at two spatial scales: implications for conservation management. Journal of Applied Ecology, 42, 270–280. Wiley, R.H. & Richards, D.G. (1978) Physical constraints on acoustic communication in atmosphere – implications for evolution of animal vocalizations. Behavioral Ecology and Sociobiology, 3, 69–94. Wilkin, T.A., Garant, D., Gosler, A.G. & Sheldon, B.C. (2006) Density eﬀects on life-history traits in a wild population of the great tit Parus major: analyses of long-term data with GIS techniques. Journal of Animal Ecology, 75, 604–615. van der Zande, A.N., ter Keurs, W.J. & van der Weijden, W.J. (1980) The impact of roads on the densities of four bird species in an open ﬁeld habitat – evidence of a long-distance eﬀect. Biological Conservation, 18, 299–321. Received 8 July 2010; accepted 2 November 2010 Handling Editor: Mark Whittingham

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 210–219

Journal of Applied Ecology 2011, 48, 76–85

doi: 10.1111/j.1365-2664.2010.01915.x

Optimally managing under imperfect detection: a method for plant invasions Tracey J. Regan1*†, Iadine Chade`s2,3† and Hugh P. Possingham4 1

The School of Botany, The University of Melbourne, Parkville, Vic., 3010, Australia; 2CSIRO Ecosystem Sciences, GPO Box 2583, Brisbane, QLD 4001, Australia, 3Unite´ de Biome´trie et Intelligence Artiﬁcielle, Institut National de la Recherche Agronomique, BP 27 F-31326 Castanet-Tolosan, France; and 4The Ecology Centre, The University of Queensland, St. Lucia, Qld 4072, Australia

Summary 1. Failing to account for uncertainty in the detection of invasive plants may lead to ineﬃcient management strategies and wasted resources. Smart strategies to manage plant invasions requires consideration of the economic costs and beneﬁts, and plant life-history characteristics as well as imperfect detection. 2. We develop a partially observable Markov decision process (POMDP) to provide optimal management actions when we are uncertain about the presence of invasive plants. The optimal strategy depends on the probability of being in a particular state. We ask the question, ‘When is it preferable to use a less eﬃcient, less costly action to a more eﬃcient, more costly action?’ We apply the POMDP to branched broomrape Orobanche ramosa, a parasitic plant species at the centre of a national eradication campaign in South Australia. 3. The optimal strategy depends on the ability to detect the invasive species and the location of the infested site. For high detection rates, if the site is a satellite infestation, management should employ the more eﬃcient, more costly action (i.e. soil fumigation) the year the weed is detected followed by monitoring. When the detection probability is low, then it is optimal to employ the less eﬃcient, low cost action (i.e. host denial) in the years the species is not detected. For sites in the centre of the infestation, management should employ the less costly, less eﬃcient action. While the optimal strategy is insensitive to colonization, the likelihood of local eradication diminishes as colonization probability increases, highlighting the importance of limiting colonization if eradication is to be achieved. 4. Synthesis and applications. Providing decision support for managing ecological systems is a key role of applied research. Formulating this support within a decision theory context provides a framework for good decision-making. The POMDP model is a novel decision support tool for optimal sequential decision making when invasive plants are diﬃcult to detect. The model can determine the best management action to employ based on the location of the infestation and can inform when to switch to alternative management actions that buﬀer against imperfect detection. Key-words: branched broomrape, decision making, detectability, eradication, invasive plants, Orobanche ramosa, partially observable Markov decision process

Introduction Invasive plants are a major threat to natural and managed systems. They are notoriously diﬃcult to control or eradicate and require large amounts of eﬀort and resources to manage eﬀectively (Pimental 2002; Panetta & Timmins 2004). With the *Correspondence author. E-mail: [email protected] †The ﬁrst two authors have contributed equally to this paper.

world wide economic impact of invasive plants estimated as US$560 billion annually (Pimental et al. 2001; Pimental, Zuniga & Morrison 2005), it is crucial that resources are directed towards the most eﬀective management activities and not wasted on management actions that are ineﬃcient. Yet deciding the best course of action for managing invasive plants can be excruciatingly diﬃcult due to the complex interaction of factors such as the extent of the invasion, the ecology of the species, the dynamics of the system, and how the species

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society

Optimally managing plant invasions 77 responds to diﬀerent management actions (Taylor & Hastings 2004). The decision process is exacerbated further by our inability to observe the invasion perfectly. Imprecise survey techniques, the cryptic nature of some species, and persistent seed banks make it diﬃcult to verify whether the imposed management actions are successful or not. Formal decision theory tools provide a useful avenue to investigate these complex interactions by systematically incorporating them in a transparent and consistent manner. These tools can calculate the optimal management actions given a speciﬁc objective and any constraints imposed on the system (i.e. cost) (Possingham 2001). Methods exist for determining optimal strategies and several have been applied to ecological systems (Richards, Possingham & Tizard 1999; McCarthy, Possingham & Gill 2001; Westphal et al. 2003; Field et al. 2004; Tenhumberg et al. 2004), and speciﬁcally in invasive species management (Shea & Possingham 2000; Taylor & Hastings 2004; Regan et al. 2006; Hauser & McCarthy 2009; Rout, Salomon & McCarthy 2009). These optimizing methods generally describe the system of interest as a Markov chain; a set of mutually exclusive states and transitions between states determined by the diﬀerent processes governing the system. In the case of an invasive plant the states are often either some measure of abundance, number of sites, or occupancy, and the processes are generally colonization, germination, and the persistence of the adult population and the seed bank. In addition, any management imposed on the system can alter the rate of one or more of these processes and hence their inﬂuence on the state. The method uses a reward system that incorporates the cost of the action and the beneﬁt of being in a particular state. Optimization algorithms, such as stochastic dynamic programming (SDP), a backward iterative procedure that ﬁnds the optimal solution for a stochastic system, can be applied to the Markov chain to determine the optimal state-dependent strategy based on the objective of maximizing the net expected beneﬁt (Mangel & Clark 1988). The combination of the Markov chain (the model representing the system) and the decision process (objective, management options, costs and rewards) are commonly known as a Markov decision process (MDP) (Bellman 1957). While Markov decision processes are very useful for representing dynamic systems, choosing the best action depends on the assumption that the state of the system is known precisely. This may not always be the case when managing ecological systems. While there is some underlying truth about the state of the system at any particular point in time, it is often not known precisely because of imperfect detection. Individuals may go undetected during a survey, and for plants, seeds may be viable in the seed bank and go unnoticed for a number of years. Despite the fact we rarely know the state of any ecological system perfectly, examples of using optimization methods that deal with imperfect detectability are limited in the ecological literature (but see Regan et al. 2006; Chades et al. 2008; Rout, Salomon & McCarthy 2009). However in Operations Research and Artiﬁcial Intelligence there is a thriving literature on developing optimization methods that speciﬁcally address the issue of incomplete knowl-

edge of the state of the system (Monahan 1982; Cassandra 1998; Kaelbling, Littman & Cassandra 1998). This general methodology is a special case of Markov decision processes, aptly named partially observable Markov decision process (POMDP). In addition to the MDP formulation, POMDP deﬁnes observed states of the system and uses information about the probability of the true state given what is known. For example, if an adult invasive plant has not been observed in a site for 3 years, what is the chance a seed bank remains? Rather than the optimal solution being state dependent, as in the MDP case, for the POMDP case, the optimal solution is dependent on the probability of being in a particular state at a given time (Cassandra 1998). In this research we develop a POMDP model to determine optimal management strategies for an invasive plant species where the states of the system are not known perfectly. We investigate how the optimal solution changes depending on the value of eradicating the species, the likelihood the site is re-colonized, and our ability to detect adults or seeds. We illustrate this approach using a case study of branched broomrape, Orobanche ramosa L., a parasitic species that is currently targeted by a national eradication program in South Australia (Jupp, Warren & Secomb 2002). While we tailor the problem to the case study, the approach is general enough to be applicable to a wide range of agricultural and environmental invasive plants as well as threatened plant species that are cryptic in nature and ⁄ or have persistent soil seed banks.

Materials and methods Branched broomrape is an annual obligate parasitic plant that attaches itself to the root system of its host plant. It originates from Mediterranean Europe, the Middle East, and northern Africa. It was ﬁrst discovered in Australia in 1911 but disappeared and was rediscovered in 1992 in southern Australia (Jupp, Warren & Secomb 2002). Branched broomrape is a threat to production crops with reported losses in biomass of up to 75%–90% (Linke, Sauerborn & Saxena 1989). Its biggest threat, however, is to export markets that prohibit the import of goods that are likely to carry broomrape seed, potentially costing the Australian industry $1Æ7 billion (Wilson & Bowran 2002). Much of the life cycle of branched broomrape occurs underground. Seeds germinate and attach to hosts in response to chemical triggers from susceptible host roots. Dust-like seeds set within 3–4 weeks after plant emergence and have the potential to survive up to 12 years in the soil (Panetta & Lawes 2007). Plants are short lived and only visible for a very short time, making detection diﬃcult (Kebreab & Murdoch 2001). A single plant can produce over 50,000 seeds that can disperse by sticking to animals and equipment that come in contact with the plant. Seeds can also be dispersed by wind and water, soil, fodder, on footwear and clothing, and can remain viable after passage through digestive tracts. A national campaign to eradicate branched broomrape in Australia commenced in 2000. Currently the species is contained within a quarantine zone approximately 70 · 70 km, which restricts the movement of fodder, crops, livestock, machinery and soil out of the area. The quarantine area is a large continuous area that encompasses the earliest known infestations. In addition there are several satellite infestations surrounding the main quarantine area where the species has been detected subsequently.

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 76–85

78 T. J. Regan, I. Chade`s & H. P. Possingham MANAGEMENT ACTIONS

Determining the most suitable course of action requires consideration of the trade-oﬀs between the cost and eﬃciency of alternative actions. Here we investigate when employing a less expensive, less eﬃcient action is preferable to a more expensive, more eﬃcient action. In the case of branched broomrape, it is thought that the key to eradication is to attack the seed bank (Cooke 2002). Soil fumigation attacks the seed bank directly, killing all seeds on contact. However it is extremely expensive and can be highly toxic, thus it is only applied to areas where branched broomrape is known to be present. Consequently there is the potential to miss areas that have broomrape seed. The soil fumigation option involves the use of methyl bromide. Surveys suggest that the application of methyl bromide is 79Æ6% eﬀective. That is, adult plants were detected the following year in 20Æ4% of the fumigated sites (N. Secomb, pers. comm.). The second management action is to promote depletion of the seed bank through host denial. For this method, a non-host crop is planted and any plants that might serve as hosts are controlled so that the seed bank decays naturally through time. The host denial action is much less expensive than soil fumigation, is much less ecologically damaging and has no human health risks, but it is also much less eﬃcient than soil fumigation. Lastly we include a ‘do nothing’ action that has minimal cost and eﬃciency. This action is included to highlight the situations where the other direct management actions are suboptimal. Since ‘do nothing’ still requires ‘doing something’, we represent the ‘do nothing’ action as pasture but with no other management imposed to eradicate the species, so there may be other plants within the site that can serve as hosts. All three management actions assume some monitoring is implemented during the ﬂowering season to determine presence or absence of the species.

THE POMDP MODEL

We model branched broomrape within a single management unit. In this case the management unit is a paddock, a fenced area used to grow crops, pasture or graze livestock. For other situations it may be a patch or any other unit where small scale direct management is focused. The system is represented by a Markov chain with three mutually exclusive states; S ¼ fse ; ss ; sw g, where se represents an empty paddock, free of broomrape, ss is a paddock with broomrape seeds, and sw is when the paddock has both seeds and adult plants. In any particular year, the paddock can only be in one of these states. The ability of the system to move from one state to another is governed by processes such as colonization, c, seed bank decay, q, and germination, g. These rates may alter depending on the action imposed on the system. The set of management actions is A ¼ fFumigation; HostDenial; DoNothingg. Transitions between states are calculated as the conditional probability of being in state s’ in the next time step given state s and action a in the previous time step. This is represented mathematically as: Pðs0 j a; sÞ ¼ Pr ðstþ1 ¼ s0 jst ¼ s; at ¼ aÞ . To determine the optimal solution requires deﬁning a reward function for the states and actions of the system. The reward function is a combination of the relative beneﬁt for being in a particular state of the system and the relative cost of the management action: Rðs; aÞ ¼ fr1 ; ; rN g To take into account the imperfect detection of the species we also deﬁne a ﬁnite set of observations: Z ¼ fza ; zp gwhere za is an observa-

tion where broomrape was absent and zp represents an observation where broomrape was present. An observation function, also called the detection rate (d), relates the observations to the real states of the system. The probability of observation z, given the system is in state s can be represented mathematically as OðzjsÞ ¼ Pr ðztþ1 ¼ z j stþ1 ¼ sÞ

A schematic representation of the POMDP with three real states and two observed states is shown in Fig. 1. The processes governing the transitions between states are shown in Table 1.

OPTIMIZATION ALGORITHM

We use stochastic dynamic programming to ﬁnd optimal solutions. Since the real state of the system is not known precisely, belief states are used to summarize and overcome the diﬃculties of incomplete detection. Astro¨m (1965) has shown that belief states are suﬃcient statistical tools to summarize all the observable history of a POMDP without loss of optimality. A POMDP can be cast into a framework of a fully observable Markov decision process where belief states represent the continuous but fully observable state space. Here, a belief state b is deﬁned as a probability distribution over the three real states of the system (empty, seeds, and adults and seeds). In our case, solving a POMDP is ﬁnding a strategy p:B · s ﬁ A such that the strategy p maps a current belief state (b 2 B) and a time-step (t 2 s) on to a management action, a2A. An optimal strategy maximizes the expected sum of rewards over a ﬁnite time horizon, T. This expected summation is also referred to as the ‘value function’ and essentially ranks strategies by assigning a real value to each belief state, b. Using the Bellman principle of optimality (i.e. subdividing the problem into simpler manageable parts) (Bellman 1957) and the POMDP parameters, we can calculate the optimal t-step value function from the (t - 1)-step value function: " V0 ðbÞ

¼ max a2A

Vt ðbÞ ¼ max a2A

þ

X

# Rðs; aÞ PrðsjbÞ ;

eqn 1

s2S

X

Rðs; aÞ PrðsjbÞ

s2S

XXX

# 0

0

PrðsjbÞPðs js; aÞOðzjs

ÞVt1 ðbaz Þ

eqn 2

;

s2S s0 2S z2Z

where Pr(s|b) represents the probability of being in state s given a belief state b, and baz is the belief state assuming action a and observation z. Equation 1 maximizes the expected sum of instantaneous rewards when there is no time left to manage the species. Similarly when there are t-steps to go, Equation 2 maximizes the instantaneous rewards and the future expected rewards for the remaining t - 1 steps.

PARAMETER ESTIMATION

Three main processes govern the system dynamics; germination (g), seed bank decay [q1 and q2 (see below for description)] and colonization (c). Since most of the life-cycle for branched broomrape occurs underground, germination in this situation is calculated as an

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 76–85

Optimally managing plant invasions 79

Absent, (za)

Pr(za| se)

Present (zp)

Pr(za |sw)

Pr(za |ss)

Pr(zp |sw)

Pr(sw|se,ai) Pr(ss|se,ai)

Pr(sw|ss,ai) Pr(sw|sw,ai)

Empty, (Se)

Pr(se |se,ai)

Seeds, (Ss)

Pr(se |ss,ai)

Pr(ss|ss,ai)

Weeds and seeds, (Sw)

Pr(ss|sw,ai) Pr(se |sw,ai)

Fig. 1. Schematic representation of the POMDP model including the real states of the system represented by the circles with solid lines (Empty, Seeds and Weeds and seeds) and the observed states dashed circles (Present, Absent). Solid arrows represent the transitions between the real states of the system while dashed arrows represent the relationship between the observed state and the real states of the system.

Table 1. State transition probabilities for the three alternative management strategies

Transitions

State to State

Processes

Action1 (Do nothing)

Action2 (Soil fumigation)

Action3 (Host denial)

Pr(se|se,a) Pr(ss|se,a) Pr(sw|se,a) Pr(se|ss,a) Pr(ss|ss,a) Pr(sw|ss,a) Pr(se|sw,a) Pr(ss|sw,a) Pr(sw|sw,a)

Empty ﬁ Empty Empty ﬁ Seeds Empty ﬁ Weeds Seeds ﬁ Empty Seeds ﬁ Seeds Seeds ﬁ Weeds Weeds ﬁ Empty Weeds ﬁ Seeds Weeds ﬁ Weeds

(1 - c) c(1 - g) c*g q1 (1 - q1)*(1 - g) (1 - q1)*g q2 (1 - q2)*(1 - g) (1 - q2)*g

1 0 0 0Æ08 0Æ57 0Æ35 0 0Æ62 0Æ38

1 0 0 – – – 0Æ23 0Æ48 0Æ29

1 0 0 0Æ08 0Æ75 0Æ17 0 0Æ81 0Æ19

emergence event and incorporates several processes such as conditioning, seed stimulation, germination of seeds, host penetration and emergence. We estimate germination rates from land use data from years 2001 to 2004 and knowledge of the known infestations (N. Secomb, unpublished data). Since broomrape seed can remain viable for many years, we assume that if a paddock was infested any time in the previous 4 years (i.e. the length of the available survey data for this study), it is still infested in the current time step, unless some additional management has been imposed on the area. The probability of germination is calculated as the proportion of paddocks in a particular year where the land use was pasture and where broomrape was found, given that it had been found there any time in the past. Pasture in this case was identiﬁed based on the known management in each paddock. This resulted in an average germination rate of 0Æ27 per paddock. Since detection of an infestation is not perfect, the germination rates have been scaled by the detection rate. Thus the emergence probability used in the model is 0Æ38 per year. Little information is available to estimate the eﬀectiveness of host denial other than it is thought to be much less eﬀective than soil fumigation. To investigate the trade-oﬀs between the diﬀerent management strategies we set the eﬀectiveness of host denial as a 50% reduction in the background germination rate (i.e. 0Æ19). Given the uncertainty associated with this

action we perform sensitivity analysis to investigate the importance of this estimate. There are two types of seed bank decay for branched broomrape. The background decay probability (q1) is the probability that the seed bank decays given that no new seeds have entered the seed bank in the current time step. It is thought that the seed bank will persist for approximately 12 years if no new seeds are added (Panetta & Lawes 2007). This equates to a decay probability of q1 = 0Æ08. The second decay probability (q2) is post seed set decay and implies that new seeds have entered the seed bank in the current year. We take a precautionary approach and assume that this decay rate is negligible and set it to zero. There are no estimates available for the colonization rate even though there is management imposed to prevent the dispersal of seed between sites. The speciﬁc rates will depend on the vector of seed dispersal and the distance from the infested site. Since the POMDP model is not spatially explicit, estimating this parameter becomes difﬁcult. Rather than ignoring this process, we look at the likely impact of colonization through investigating a range of values that could be representative of broomrape dispersal vectors that are not addressed through management provisions (i.e. wind). We initially assume colonization rate is zero and then increase it incrementally to 0Æ15. Data indicate that if soil fumigation is implemented then the probability that a paddock in the adult state will remain in the adult state is

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 76–85

80 T. J. Regan, I. Chade`s & H. P. Possingham 0Æ204 (N. Secomb, unpublished data). We use this value as the eﬀectiveness of soil fumigation but scale it by the detection factor. Once the fumigant is applied it kills all seeds on site. In general, however, only sites known to be infested with broomrape are fumigated. It is not economically feasible to fumigate areas where the plant has not been detected in the past, thus we constrain the model so that the soil fumigant can only be applied in the year broomrape is observed. Transition probabilities for each action are given in Table 1. We assume that monitoring occurs every year around ﬂowering time. The probability of detecting adult branched broomrape is calculated based on double blind surveys within the quarantine area (Kuhnert & Possingham 2003). We use an average detection probability of 0Æ7 (the probability an adult plant is detected given it is present). As this parameter is highly uncertain we investigate scenarios where the detection rate ranges from 0Æ1 to 1Æ0.

REWARDS

Management time horizon (years)

The reward function is expressed as a ratio between the beneﬁt (or reward) for being in a particular state and the cost of the management option imposed. In this case, soil fumigation amounted to an economic cost on average of $18 000 per paddock (N. Secomb, unpublished data). This cost includes basic application but does not include environmental and human health costs thus perhaps underestimating the true cost of soil fumigation. For the management action ‘host denial’ we assume that there is some value in the crop itself in that it

can be sold (usually within the quarantine area) or used for feed for livestock. For the purposes of illustration we assume the beneﬁt from selling the non-host crop is equal to the cost of planting, maintaining and harvesting the crop. For the ‘do nothing’ action, there is a cost of managing the paddock as pasture but we assume there is no value in the pasture itself. We set the cost for managing the paddock as pasture at $2000. The monitoring costs are assumed to be equal across the three alternative management strategies. In this case, the beneﬁts of being in a particular state are independent of the action employed. Given the objective is to eradicate branched broomrape from the site, there is no beneﬁt for the system to be in a seed or adult and seed state. However for circumstances where control is the objective, there may be some beneﬁt for being in one of the other two stages. This can easily be incorporated into the POMDP formulation by adjusting the rewards for the alternative states based on the objective. The reward for being in an empty state may depend on the number of remaining infested sites. If the site of interest is the last known infested area then the beneﬁt for being in an empty state will be very large as it will constitute complete eradication of the species. Alternatively, the reward for a site being in an empty state will be less if there are other infested sites in the area. It is plausible that the reward could also depend on the proximity of the site to other infestations within in the area. Satellite infestations or sites on the edge of the main infested area may have a high reward if they become empty as they are likely to be deemed locally eradicated. On the contrary, empty sites in the

(a)

(b)

(c)

(d)

Time since detection (years) Fig. 2. The optimal management strategy for diﬀerent reward ⁄ cost ratios (R ⁄ cost) under diﬀerent management time horizons from 20 years down to 1 year where colonization is c = 0. Detection probability is d = 0Æ7. Black bars are soil fumigation, dark grey bars are host denial and light grey bars are ‘do nothing’ (pasture). 2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 76–85

Optimally managing plant invasions 81

Probability of eradication under optimal management

1 0·9 0·8 0·7 0·6 Do nothing

0·5 0·4 0·3

c=0 c = 0·05 c = 0·1 c = 0·15

0·2 0·1

Soil fumigation

0 0

5

10

15

20

Time since weed detection (years)

Fig. 3. The probability of eradication given the optimal management strategy for diﬀerent levels of colonization. Reward cost ratio is 100 and detection probability is d = 0Æ7. Thick black line indicates when the optimal management action changes.

interior of the infestation may not be deemed eradicated until the neighboring sites are also clear. Thus there may not be a very large associated beneﬁt with this outcome. We apply a range of rewards to the empty state to assess how the location and circumstances of the paddock changes the best management strategy. The rewards function includes a discount rate of 4% (Sumaila & Walters 2005).

Results The optimal strategy graphs in Fig. 2 map the optimal action for each time step and observation, present or absent, over the entire management time horizon (20 years). For example, in Fig. 2b where the cost ratio (i.e. reward of being in an empty state ⁄ cost of the action) is 1Æ0, colonization rate = 0 and detection probability = 0Æ7, if there are 20 years remaining and branched broomrape is detected (i.e. time since detection, t = 0), the optimal action to employ is soil fumigation. In subsequent years if we continue to not detect any plants (i.e. time since detection, t ‡ 1), it is optimal to switch to host denial until year 4 after which the ‘do nothing’ action becomes optimal. Once broomrape is detected the optimal action depends on how many years remain in the management time horizon. For instance if broomrape is again detected in the 12th year the optimal action with eight years remaining is to employ the host denial action. The optimal course of action for managing branched broomrape depends on the reward for being in an empty state (Fig. 2). For areas where the cost ratio is ‡10, such as a satellite infestation or a site on the edge of the main infested area, when the species is detected, the optimal action is to perform the most eﬃcient and costly action (i.e. soil fumigation). In subsequent years, the optimal strategy when adult plants are not detected is to ‘do nothing’ unless there are only 4 years left to manage, then host denial is optimal (Fig. 2c,d). For areas of low relative reward (i.e. R ⁄ cost = 0Æ5), such as sites in the centre of the infestation, the less eﬃcient and less costly action (i.e. host denial) is the best strategy (Fig. 2a). For areas where the cost ratio is equal to 1, the best strategy depends on the man-

agement time horizon: when the species is detected at the early stage of the management period, the more eﬃcient action is optimal. The less eﬃcient action becomes optimal only when the management time horizon is less than nine years. Under other conditions ‘do nothing’ is optimal (Fig. 2b). When colonization is a factor, the optimal strategies do not change. For a reward ⁄ cost ratio of 100 the optimal management involves soil fumigation in the year plants are detected followed by the ‘do nothing’ action in subsequent years when plants are not detected (Fig. 3). While the optimal set of management actions are stable as the colonization rate increases, the probability of eradicating the weed from a site diminishes. When colonization is zero (c = 0), the probability of eradication approaches 1Æ0 in approximately 15 years. However if the colonization rate is c = 0Æ05 the best that can be achieved in the management time horizon is a probability of eradication of 0Æ86. When the colonization rate increases then the likelihood of eradication becomes less. At a colonization rate of 0Æ15 the best possible outcome is a probability of eradication of 0Æ62. Under these colonization scenarios even if the management time horizon were extended (>20 years), it would not increase the probability of eradication by a substantial amount as all the curves reach an asymptote by year 20. The diﬀerence between the MDP and the POMDP formulation is the incorporation of imperfect detection. This has an eﬀect on the optimal course of action as well as the probability of eradication within the management time horizon (Figs 4 and 5). For the MDP formulation where detection is perfect (d = 1Æ0), for a reward ⁄ cost ratio = 100 and colonization = 0, the optimal course of action is to fumigate the year the weed is detected and then ‘do nothing’ when it is not detected. Under this scenario the probability of eradication approaches 1Æ0 in approximately 12 years (Fig. 5). This is consistent with the current philosophy of branched broomrape management. However if the ability to detect the weed is imperfect, not only does the probability of eradication change but so does the optimal course of action. For high detection probabilities (0Æ5–0Æ7) the optimal course of action is the same as the perfect detection scenario (Fig. 4c,d) but the probability of eradication diminishes. The probability of eradication approaches 1Æ0 within the 20-year management time horizon but at a much slower rate (Fig. 5). For a detection probability of 0Æ7, the probability of eradication approaches 1Æ0 around 15 years after detection, while for a detection probability of 0Æ5, the probability of eradication approaches 1Æ0 around 20 years after initial detection. When dealing with low probability of detection (0Æ1–0Æ3), the optimal course of action remains, i.e. to fumigate in the year the weed is detected but it is now followed by host denial in the years when the weed is absent (Fig. 4a,b). This results in a probability of eradication of 0Æ82–0Æ90 at the end of 20 years for the two detection rates respectively. The ‘do nothing’ action at the end of the management time horizon in these scenarios is an artefact of the time horizon constraint. Since there are only a couple of years left to manage, the gain from doing a more costly action (host denial in this case) will not improve the overall situation, thus a less costly option becomes optimal in the ﬁnal stages of management.

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 76–85

Management time horizon (years)

82 T. J. Regan, I. Chade`s & H. P. Possingham (a)

(b)

(c)

(d)

Time since detection (years) Fig. 4. The optimal management strategy for diﬀerent detection probabilities under diﬀerent management time horizons from 20 years down to 1 year when there is no colonization (c = 0Æ0). Reward cost ratio is 100. Black bars are soil fumigation, dark grey bars are host denial and light grey bars are ‘do nothing’ (pasture).

Data for the eﬀectiveness of host denial were unavailable for this study. Sensitivity analysis was performed for two situations, high cost ratio and low detection (Fig. 6a) versus low cost ratio, high detection (Fig. 6b). The eﬀectiveness of host denial was set at 50% the background germination rate. In the sensitivity analysis the germination rate was decreased from 50% to 100% of the background rate. Results revealed the model to be relatively insensitive to changes in the eﬀectiveness

of host denial (EHD) with only small changes in the switching time from host denial to the ‘do nothing’ strategy at the end of the management time horizon in Fig. 6a. The largest diﬀerence in the probability of eradication occurred at around nine years where it ranged between 0Æ45 and 0Æ7 for decreases in the germination rate of 50% and 100% respectively. There were also small changes in the optimal time to switch from host denial to the ‘do nothing’ action.

1

Probability of eradication under optimal management

0·9

Do nothing

Do nothing

0·8 0·7 Host denial

0·6 0·5 0·4 0·3

d = 0·1 d = 0·3 d = 0·5 d = 0·7 d=1

0·2 0·1 Soil fumigation

0 0

2

4

6

8

10

12

14

Time since weed detection (years)

16

18

20

Fig. 5. The probability of eradication under optimal management for diﬀerent levels of detection probability. This scenario assumes no colonization between paddocks and a reward cost ratio of 100. Thick black lines indicate when the optimal management action changes.

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 76–85

Optimally managing plant invasions 83 (a)

Do nothing

Host denial d = 0·3 R/cost = 100 c = 0·0 EHD = 0·5 EHD = 0·8 EHD = 1·0

Soil fumigation

(b)

Do nothing

Fig. 6. Sensitivity analysis of the eﬀectiveness of host denial (EHD). (a) Detection probability is 0Æ3, R ⁄ cost ratio = 100 and colonization is 0. (b) Detection probability is 0Æ7, R ⁄ cost ratio is 0Æ5 and colonization is 0. EHD values are the proportional decreases in the background germination rate. Thick black lines indicate when the optimal action changes.

Discussion One of the main roles of applied research is to provide decision support for managing ecological systems (Buckley 2008). Formulating this support within a decision theory context provides a framework for good decision making (Possingham 2001; Ayaz, Leung & Miao 2008; Franklin et al. 2008). The adoption of a POMDP model as a decision support tool provides many beneﬁts that a more informal approach to decision making may lack. One of the major challenges in managing ecological systems eﬀectively is appropriately accounting for multiple processes and decision factors. The POMDP model explicitly incorporates imperfect detection of an invasive plant, as well as the interacting processes at play (i.e. dispersal, germination, seed bank decay, colonization). It also includes important management factors such as the probability of successful management, beneﬁts, costs and management time horizons. The POMDP model informs on the best management action to employ under imperfect detection. It also informs on where, when, and for how long diﬀerent actions should be imple-

Host denial

d = 0·7 R/cost = 0·5 c = 0·0 EHD = 0·5 EHD = 0·8

EHD = 1·0

mented. The POMDP framework can quantify the likelihood of eradication which is useful when deciding when to stop managing the invasive species. It is unlikely and diﬃcult to verify whether a more informal decision process takes into account of all these factors appropriately and consistently. For our case study, branched broomrape, current management prescriptions include some direct management via host denial or soil fumigation followed by monitoring in subsequent years. If the weed is not detected at the site for 12 consecutive years (i.e. the longevity of the seed bank) then the weed is deemed eradicated from the site (Panetta & Lawes 2005). Our model concurs with this line of management only in the situation where the colonization rate is zero and the detection probability is very high. While managers acknowledge that colonization and imperfect detection are important factors in managing invasive species, they are not necessarily accounted for adequately in the informal decision process. The POMDP model demonstrates that ignoring imperfect detection of invasive plants may result in inappropriate management. In circumstances when detection is imperfect, if an

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 76–85

84 T. J. Regan, I. Chade`s & H. P. Possingham MDP formulation of the model is adopted instead of the POMDP model, it may lead to poor management decisions in the years where the species is not detected. In the MDP a nondetection assumes an empty state, so the optimal action will always be ‘do nothing’. In the POMDP model, a non-detection is used to update the probability of being in an empty state. If this translates to a low probability of being in an empty state the ‘do nothing’ action may not be optimal and adopting alternative actions to buﬀer the uncertainty in non-detection years may be needed. In our case, we constrained soil fumigation, the most costly action, to instances when the species was detected, thus the beneﬁt from using a POMDP is when the host denial action is optimal in the years the plant is not observed. Constraining management actions to times of detection is common in invasive species management (i.e. spraying or removal of adults). If, however, there are multiple alternative actions that could be performed when the species is not detected, such as placing barriers to hinder dispersal, or other methods for treating the seed bank, the usefulness of the POMDP will be even greater. The POMDP model resulted in several useful general management recommendations that are dependent on the location of the infestation relative to other infected sites. For areas with low relative beneﬁt when empty, such as sites within the interior of the infestation that are unlikely to be declared eradicated until sites around it are also clear, the use of a less eﬃcient and inexpensive action (i.e. host denial in our case study) is preferred. However once the beneﬁts of being in an empty state outweigh the cost of the most eﬃcient action (e.g. soil fumigation in this case), the most eﬃcient, and more costly action should be employed the year the species is detected. This would be the case for satellite infestations or newly infested sites on the edge of the main infestation. This result is robust to increases in the value of the site so exact estimates for these values are not important. In our case study, the cost of applying the soil fumigant did not incorporate human health costs explicitly. An optimal solution incorporating human health costs can be inferred from the results by using relative measures of human health costs versus the value of an empty site. When management areas are at risk of being re-colonized by surrounding infected areas, the likelihood of eradication diminishes as the colonization rate increases, even though the optimal set of actions does not change signiﬁcantly. This highlights the importance of limiting colonization from surrounding areas if eradication is ever to be achieved, otherwise, management can only endeavour to achieve a level of control rather than eradication. These results emphasize the challenge of managing invasive plants with high colonization rates under imperfect detection. Understanding the magnitude of additional potential seed dispersal vectors and how they can be eﬀectively limited should be considered a research priority. The POMDP model is based on the data available, in this case presence, absence data on the species. For situations where alternative actions based on diﬀerent abundances are warranted, then increasing the number of states may be advan-

tageous. However the complexity of solving the POMDP prevents us from using a detailed state space when looking for exact solutions. Thus POMDP is appropriate when the system can be summarized into a small number of states. We present several graphs of the probability of eradication under optimal management illustrating that while the optimal management actions may not change with changes in speciﬁc parameter values, the eﬀectiveness of eradicating the weed can change. This pattern was particularly evident when the colonization rate increased. The probability of eradication at the end of the time horizon ranged from 1Æ0 to 0Æ62 across the colonization rates explored (0-0Æ15). While it is tempting to use these graphs for deciding when to declare eradication of a site, we do not recommend it. Instead the decision to declare eradication involves consideration not only of the probability of eradication but also the trade-oﬀ between the cost of continued monitoring and the consequences should the invasive species escape, colonize new areas and causes more damage if eradication is declared too soon. For sites where the cost of escape is relatively low, such as species where the potential damage is likely to be small, it may not be worthwhile to continue monitoring even if the probability of eradication is relatively low. Alternatively, for sites that have a high consequence of escape, managers may be advised to continue monitoring to a point where the probability of eradication approaches 1Æ0 to avoid the high cost of escape and damage. We recommend that when managers are deciding when to declare eradication of a particular site they should apply the method outlined in Regan et al. (2006) that makes the trade-oﬀs between potential errors explicit. In summary the POMDP formulation is a convenient and novel model for optimal sequential decision making when invasive plants are partially detectable. While the management strategies are fairly robust to uncertainty, if the invasive plant has a low probability of detection, then an alternative course of action should be adopted. The model also allows an explicit investigation into how management may change when there are alternative rewards for being in an empty state. This led to several useful recommendations on how best to manage an invasive species depending on the location of the infestation. These types of recommendations are diﬃcult to deduce without a formal decision support tool such as the POMDP model. The model presented is not spatially explicit even though we address the potential threat of colonization from neighbouring areas. It is plausible that the optimal course of action will also depend on the state of neighbouring sites. This is an active area of research in the utility of the POMDP formulation.

Acknowledgements We thank Dane Panetta, Yvonne Buckley and two anonymous reviewers for providing useful comments on earlier versions of this paper. We are grateful to Anthony Cassandra for insightful discussions on POMDP and the branched broomrape eradication team for support and provision of data for the model. TJR and HPP are ﬁnancially supported by AEDA and the ARC. IC is ﬁnancially supported by INRA, CSIRO, MASCOS, AEDA and ACERA. A grant from the GRDC and support from the Weeds CRC enabled this work to be completed.

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 76–85

Optimally managing plant invasions 85

References Astro¨m, K. J. (1965) Optimal control of Markov decision processes with incomplete state estimation. Journal of Mathematical Analysis and Applications, 10, 174–205. Ayaz, H., Leung, B. & Miao, Z. (2008) Integrating data, biology, and decision models for invasive species management: application to leafy spurge (Euphorbia esula). Ecology and Society, 13, online URL: http://www.ecology andsociety.org/vol13/iss12/art12/. Bellman, R. (1957) A Markovian decision process. Journal of Mathematics and Mechanics, 6, 679–684. Buckley, Y. M. (2008) The role of research for integrated management of invasive species, invaded landscapes and communities. Journal of Applied Ecology, 45, 397–402. Cassandra, A. R. (1998) Exact and approximate algorithms for partially observable Markov decision processes. PhD Thesis, Brown University. Chades, I., McDonald-Madden, E., McCarthy, M. A., Wintle, B., Linkie, M. & Possingham, H. P. (2008) When to stop managing or surveying cryptic threatened species. Proceedings of the National Academy of Sciences of the United States of America, 105, 13936–13940. Cooke, D. (2002) Control of Branched Broomrape. Department of Water, Land and Biodiversity Conservation, Adelaide. Field, S. A., Tyre, A. J., Jonzen, N., Rhodes, J. R. & Possingham, H. P. (2004) Minimizing the cost of environmental management decisions by optimizing statistical thresholds. Ecology Letters, 7, 669–675. Franklin, J., Sisson, S. A., Burgman, M. A. & Martin, J. K. (2008) Evaluating extreme risks in invasion ecology: learning from banking compliance. Diversity and Distributions, 14, 581–591. Hauser, C. E. & McCarthy, M. A. (2009) Streamlining ‘search and destroy’: cost-eﬀective surveillance for invasive species management. Ecology Letters, 12, 683–692. Jupp, P., Warren, P. & Secomb, N. (2002) The branched broomrape eradication program: methodologies, problems encountered and lessons learnt. Proceedings of the 13th Australian Weeds Conference (eds H. Spaﬀord Jacob, J. Dodd & J. H. Moore), pp. 270–273. Plant Protection Society of Western Australia, Perth. Kaelbling, L. P., Littman, M. L. & Cassandra, A. R. (1998) Planning and acting in partially observable stochastic domains. Artiﬁcial Intelligence, 101, 99– 134. Kebreab, E. & Murdoch, A. J. (2001) Simulation of integrated control strategies for Orobanche spp. based on a life cycle model. Experimental Agriculture, 37, 37–51. Kuhnert, P. & Possingham, H. (2003) Development of Decision Support Models for Weed Containment and Eradication. Report V: Detectability of Branched Broomrape using Double Count Method with Independent Observers. CRC for Weed Management and the Animal and Plant Control Commission, Brisbane, Queensland. Linke, K. H., Sauerborn, J. & Saxena, M. C. (1989) Orobanche Field Guide. Parasitic Weeds Collaborative Research Program. Institute of Plant Production in the Tropics and Subtropics. Hohenheim, Germany. Mangel, M. & Clark, T. W. (1988) Dynamic Modeling in Behavioural Ecology. Princeton University Press, Princeton, New Jersey. McCarthy, M. A., Possingham, H. P. & Gill, A. M. (2001) Using stochastic dynamic programming to determine optimal ﬁre management for Banksia ornata. Journal of Applied Ecology, 38, 585–592.

Monahan, G. E. (1982) A Survey of Partially Observable Markov Decision Processes: Theory, Models, and Algorithms. Management Science, 28, 1– 16. Panetta, F. D. & Lawes, R. (2005) Evaluation of the performance of weed eradication programs: the delimitation of extent. Diversity and Distributions, 11, 435–442. Panetta, F. D. & Lawes, R. (2007) Evaluation of the Australian branched broomrape (Orobanche ramosa) eradication program. Weed Science, 55, 644–651. Panetta, F. D. & Timmins, S. M. (2004) Evaluating the feasibility of eradication for terrestrial weed incursions. Plant Protection Quarterly, 19, 5–11. Pimental, D. (2002) Biological Invasions: Economic and Environmental Cost of Alien Plant, Animal, and Microbe Species. CRC Press, Boca Raton, Florida. Pimental, D., Zuniga, R. & Morrison, D. (2005) Update on the environmental and economic costs associated with alien-invasive species in the United States. Ecological Economics, 52, 273–288. Pimental, D., McNair, S., Janecka, J., Wightman, J., Simmonds, C., O’Connell, C., Wong, E., Russel, L., Zern, J., Aquino, T. & Tsomondo, T. (2001) Economic and environmental threats of alien plant, animal, and microbe invasions. Agriculture, Ecosystems and Environment, 84, 1–20. Possingham, H. P. (2001) The Business of Biodiversity: Applying Decision Theory Principles to Nature Conservation. TELA Series No. 9, Australian Conservation Foundation, Fitzroy, Victoria. Regan, T. J., McCarthy, M. A., Baxter, P. W. J., Panetta, F. D. & Possingham, H. P. (2006) Optimal eradication: when to stop looking for an invasive plant. Ecology Letters, 9, 759–766. Richards, S. A., Possingham, H. P. & Tizard, J. (1999) Optimal ﬁre management for maintaining community diversity. Ecological Applications, 9, 880–892. Rout, T. M., Salomon, Y. & McCarthy, M. A. (2009) Using sighting records to declare eradication of an invasive species. Journal of Applied Ecology, 46, 110–117. Shea, K. & Possingham, H. P. (2000) Optimal release strategies for biological control agents: an application of stochastic dynamic programming to population management. Journal of Applied Ecology, 37, 77–86. Sumaila, U. R. & Walters, C. (2005) Intergenerational discounting: a new intuitive approach. Ecological Economics, 52, 135–142. Taylor, C. M. & Hastings, A. (2004) Finding optimal control strategies for invasive species: a density-structured model for Spartina alterniﬂora. Journal of Applied Ecology, 41, 1049–1057. Tenhumberg, B., Tyre, A. J., Shea, K. & Possingham, H. P. (2004) Linking wild and captive populations to maximize species persistence: optimal translocation strategies. Conservation Biology, 18, 1–11. Westphal, M. I., Pickett, M., Getz, W. M. & Possingham, H. P. (2003) The use of stochastic dynamic programming in optimal landscape reconstruction for metapopulations. Ecological Applications, 13, 543–555. Wilson, B. J. & Bowran, D. G. (2002) Report on the Review of the Branched Broomrape Eradication Program. Land, Water and Biodiversity Committee, Canberra. Received 13 July 2010; accepted 2 November 2010 Handling Editor : Jason Matthiopoulos

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 76–85

Journal of Applied Ecology 2011, 48, 181–191

doi: 10.1111/j.1365-2664.2010.01916.x

Qualitative modelling of invasive species eradication on subantarctic Macquarie Island Ben Raymond1*, Julie McInnes2, Jeffrey M. Dambacher3, Sarah Way2 and Dana M. Bergstrom1 1

Australian Antarctic Division, Department of Sustainability, Environment, Water, Population and Communities, 203 Channel Highway, Kingston 7050, Tas., Australia; 2Parks and Wildlife Service, Department of Primary Industries, Parks, Water and Environment, GPO Box 1751, Hobart 7001, Tas., Australia; and 3CSIRO Mathematics, Informatics and Statistics, GPO Box 1538, Hobart 7001, Tas., Australia

Summary 1. Invaded ecosystems present complex management issues. This problem is exacerbated in many situations by a lack of knowledge about the ecosystem. However, delaying conservation action to collect further data and so reduce such uncertainty is often either impractical or inadvisable. 2. The Macquarie Island Pest Eradication Project, currently underway, is attempting to eradicate rabbits, rats, and mice from the island. We undertook qualitative modelling of this project, examining a range of likely outcomes and their possible ecological consequences. The results were aggregated across a large number of possible models, in order to account for uncertainty concerning interactions within the ecosystem. 3. The results strongly support the current actions of simultaneous eradication of all three pest species, as simulated eradications of only one or two generally led to continued impacts on the island’s native biota. The results also provided support for the anticipated positive outcomes of the project, with predicted recoveries of tall tussock vegetation, and burrow- and surface-nesting seabirds. 4. However, the model predictions also highlighted potential risks: the eradication of mice from the island may not succeed, due in part to the structural position of this species within the ecosystem. Successful eradication of all three target species could potentially release the self-introduced, nonnative redpolls and starlings, allowing expansion of their populations, with possible impacts on macro-invertebrates and vegetation. 5. Synthesis and applications. These results demonstrate that qualitative modelling approaches can in some cases deliver consistent results, despite high levels of uncertainty regarding interactions within the underlying ecosystem. Such outcomes can provide assistance in the development of strategic contingency plans and ongoing future management action. Key-words: cats, conservation, invasive species, mice, rabbits, rats, starlings, subantarctic island, uncertainty

Introduction Invaded ecosystems present complex management issues, with multiple interacting elements and indirect eﬀects, giving sometimes unexpected eﬀects of management action. Even in relatively simple island ecosystems, the dynamics can be poorly understood and outcomes counter-intuitive (e.g. Rayner et al. 2009). There is a need for applied research in order to provide

*Correspondence author. E-mail: [email protected]

decision support for management actions (Buckley 2008; Bergstrom et al. 2009), but uncertainty about outcomes can be a diﬃcult issue to overcome (Maguire 2004). Delaying conservation action in order to collect additional data can reduce such uncertainty, but this may not always be the best strategy (Grantham et al. 2009). Subantarctic Macquarie Island is currently being subjected to one of the largest island eradication actions attempted (Parks & Wildlife Service 2008) with the aim to eradicate European rabbits Oryctolagus cuniculus L., ship rats Rattus rattus L., and house mice Mus musculus L. Brodifacoum-laced cereal

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society

182 B. Raymond et al. bait is being aerially spread, to be followed up by hunting teams and rabbit detector dogs. Action to eradicate rabbits and rodents was long planned as part of the island’s integrated pest management framework (Copson 2002). Before this action could be implemented, however, the island suﬀered substantial destruction of vegetation because of an increase in the rabbit population, which followed the eradication of cats and cessation of Myxoma virus releases (see Scott & Kirkpatrick 2008; Bergstrom et al. 2009; Dowding et al. 2009). Signiﬁcant gaps exist in our knowledge of the Macquarie Island ecosystem, and the ecological outcomes, should one or more of the target species survive the eradication attempt, are acknowledged as being largely unknown (Parks & Wildlife Service 2009b). However, the immediate priority of removing the pests from the island (particularly the rabbits, because of the extensive vegetation damage and onward impacts) far outweighed the advantages of further research. The island’s ecosystem involves a large number of potential interactions between species. Not only are the functional forms of these interactions often poorly understood, in some cases it is not even apparent whether or not an interaction between a pair of species is of suﬃcient importance to warrant inclusion in an ecosystem model. Such structural uncertainty can be an important source of indeterminacy in ecological predictions and associated decision making (Punt & Hilborn 1997; Hosack, Hayes & Dambacher 2008). Qualitative modelling (Levins 1974; Puccia & Levins 1985), which focuses on model structure rather than the quantitative details of a model’s

components, can be a valuable tool in such circumstances. Such modelling is also well suited to the exploration of the relative balance of direct and indirect eﬀects in an ecosystem. Indirect eﬀects (that is, an eﬀect from one member of the community upon another, mediated by one or more intervening members) can in some cases oppose and outweigh direct eﬀects, causing counter-intuitive outcomes such as a population increasing despite reduced food availability (Sih et al. 1985; Yodzis 1988). Qualitative modelling has previously been applied to the analyses and simulation of pest management actions (e.g. Ramsay & Veltman 2005; Ramsay & Norbury 2009). This study explores the possible consequences of eradication activities under various success scenarios, and investigates whether informative results can be obtained, despite the relatively high levels of uncertainty.

Materials and methods STUDY SITE AND HISTORY

Subantarctic Macquarie Island (5430¢S, 15857¢E) is a tundra-covered, World Heritage listed sliver of land in the Southern Ocean. Oiling gangs plundered the seal and penguin populations on the island throughout the 19th and early 20th centuries. This activity was accompanied by a litany of non-native species arrivals at various times (Fig. 1): dogs Canis lupus familiaris (1815), cats (1820), wekas Gallirallus australis scotti (1870s), rabbits (1879), mice (1890), and rats (1900) (see Cumpston 1968; Copson & Whinam 2001). Several nonnative bird species also arrived, probably through self-introduction

Fig. 1. Timeline of pest species introductions, management actions, and other events on Macquarie Island. 2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 181–191

Modelling of invasive species eradication 183 from New Zealand and Australia (Copson & Brothers 2008): starlings Sturnus vulgaris L. in 1930 (Falla 1937), common redpolls Carduelis ﬂammea L. in 1912 (Falla 1937) and mallards Anas platyrhynchos L. in the 1950s (Cumpston 1968). Dogs died out after 1820, but the other species established and began to impact native biota. Two native ground-dwelling bird species, the Macquarie Island parakeet Cyanoramphus erythrotis Wagler and the buﬀ-banded rail Gallirallus philippensis macquariensis Hutton, became extinct by the 1890s, probably through a hyperpredation interaction with cats and rabbits (see Vestjens 1963; Taylor 1979). Extensive rabbit grazing of the native vegetation was noted in the early 1900s (Taylor 1955; Costin & Moore 1960). Rabbit control eﬀorts began in 1968 with the release of the European rabbit ﬂea Spilopsyllus cuniculi Dale followed by successful introduction of the Myxoma virus in 1978 (Brothers et al. 1982), resulting in substantial reductions in the rabbit population. The Myxoma virus was spread until 2006, but eﬃcacy was possibly compromised after about 2002 (Copson 2002; Dowding et al. 2009). The reduction in rabbit numbers probably resulted in increased cat predation on burrow-nesting seabirds (Brothers 1984; Copson & Whinam 2001) and the cat management programme was refocused in 1985, intensiﬁed in 1998, and the last cat was shot in 2000 (Copson & Whinam 2001; Copson 2002). From 2002, rabbit numbers on the island increased to 40 000–130 000 in 2005 (Fig. 1; Terauds 2009), with the upper estimate close to the estimated maxima of 150 000 prior to Myxoma release (Sobey et al. 1973; Copson, Brothers & Skira 1981). The recent increase in rabbits has again caused substantial vegetation destruction, with particular impact on coastal slopes (Scott & Kirkpatrick 2008; Bergstrom et al. 2009), and the impacts to some areas of seabird breeding habitat are reaching critical levels (McInnes & Way 2010). The Macquarie Island Pest Eradication Project (Parks & Wildlife Service 2008) began in May 2010, with the objectives of the recoveries of vegetation, burrowing seabird habitat, and populations of invertebrates and burrowing seabirds (Parks & Wildlife Service 2009b).

MODEL CONSTRUCTION

The dynamics of n interacting species is often represented as a LotkaVolterra system of diﬀerential equations of the form: 1 dNi ¼ f1 ðN1 ; N2 ; . . . ; Nn ; c1 ; c2 ; . . . ; cm Þ Ni dt

eqn 1

where Ni is the density of population i, fi is a function describing the per-capita growth rate of that population, and ch represents growth parameters that account for processes of birth, death and migration. The interaction coeﬃcient aij measures the eﬀect of a change in the density of species j on the per-capita growth rate of species i, and is deﬁned as the partial derivative of fi with respect to Nj (Levins 1968; Berlow et al. 2004): aij ¼ @fi =@Nj :

eqn 2

At equilibrium, the negative inverse of the matrix A (which has elements aij) can be used to estimate the long-term eﬀects of a press perturbation (Bender, Case & Gilpin 1984; Nakajima 1992), which is deﬁned as a sustained shift in the magnitude of a growth parameter ch of a species (Puccia & Levins 1985). Precise quantitative speciﬁcation

of A is rarely practicable (Levins 1998), and not possible for the Macquarie Island ecosystem. One solution is to instead specify A only in terms of the signs of its interaction coeﬃcients (Levins 1974). This qualitative approach permits inclusion of important variables and interactions in a model, despite an inability to precisely measure them, and can give qualitative predictions of press perturbation impacts (Levins 1974; Puccia & Levins 1985). However, for even moderately complex models, there can be multiple interaction pathways that connect species, and the qualitative predictions from the inverse matrix can be ambiguous because of the propagation of both negative and positive eﬀects between species (Dambacher, Li & Rossignol 2002). Our particular situation is further complicated by another form of ambiguity: model structure uncertainty (there are a number of interactions that could potentially be included or excluded from the model). These issues, and previously described approaches to the problem, are discussed further in the Appendix S1 (Supporting information). In this work, we adopt a hybrid solution, which simultaneously addresses uncertainty within and between qualitative models. For a given model structure (that is, a set of community members and their signed interactions), a realisation of this structure was obtained by assigning randomly selected interaction strengths for the non-zero elements of A. Strengths between 0Æ01 and 1Æ0 were used, with the sign of each interaction remaining unchanged. Self-eﬀects (aii) were assigned a magnitude between )1 and )0Æ25. The Lyapunov stability of this realisation was checked, and an unstable realisation discarded. Predictions were obtained from the negative inverse of A, and compared against validation information. If the signs of the predictions matched the signs of the known (validation) responses, then this realisation was considered plausible and added to the pool of accepted realisations. This process was repeated for each model structure until 1000 stable, valid realisations were obtained. Structural uncertainty was addressed by considering a large number of model structures, encompassing all possible combinations of unknown interactions (sensu Montan˜o-Moctezuma, Li & Rossignol 2007). Results were aggregated across this set of model structures. A summary of the method is shown in ﬂowchart form in Fig. 2. The use of random interaction strengths (a similar approach to that used by e.g. Yodzis 1988; Dambacher et al. 2003; Hosack, Hayes & Dambacher 2008) shares the advantage of purely qualitative methods, in that it requires knowledge only of the signs of the species interactions. Our method provides an additional advantage by constraining the allowable parameter values to those which are consistent with both system stability and the validation data. Increasing constraints can potentially be achieved by increasing the number of responses in the validation set through experiments or expert knowledge. Example R code is available in the Supporting information, or from http://data.aad.gov.au/analysis/qualitative/.

THE MACQUARIE ISLAND MODEL

The model comprised three categories of vegetation, macro-invertebrates, seven categories of native seabird, rabbits, and their control agent the Myxoma virus, the introduced meso- and apex predators cats, rats, mice, and redpolls and starlings. A description of these elements, including the species within each group and their Latin names, is given in Table 1, and their interactions are shown in Fig. 3 (details of the interactions can be found in Table S1, Supporting information). All community members in the model are self-limited, although these self-interactions are not shown on Fig. 3 for clarity of presentation. Self-limitation represents processes such as competition

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 181–191

184 B. Raymond et al. models that passed the stability and validation criteria in the previous simulations, and also passed a further stability test (i.e. with the cats and Myxoma removed). The eradications of various species from this system were then simulated.

(a)

Results CAT AND MYXOMA VIRUS SUPPRESSION SIMULATIONS

The model has 17 unknown interactions, giving 217 (c. 131 000) possible model structures. We thus obtained a pool of 131 million (1000 · 217) model realisations that passed the stability and validation tests. Figure 4 summarizes the predicted responses of the community members to the suppression of cats and the Myxoma virus. Rabbits increased and tall tussock vegetation decreased in all cases by deﬁnition, because these were the validation criteria. The results provided moderate support for decreases in albatrosses (79% of cases) and burrow-nesting seabirds (70%), stronger support for decreases in redpolls and starlings (91%) and giant petrels (94%), and moderate support for an increase in skuas (76%) and Antarctic prions (70%). Grassland showed no change, as the model included no interactions that aﬀected it (other than its own self-limitation). The predictions for the remaining community members did not give clear indication of increases or decreases, with majorities of 52–65%.

(b)

COMBINED ERADICATION SIMULATIONS

Fig. 2. (a) Summary ﬂowchart of the modelling method used. (b) Details of step 3 (evaluation of each of the possible model structures).

for food and breeding territory, as well as dependence on elements not explicitly included in model, such as marine resources for seabirds. The dashed lines in Fig. 3 represent interactions that are poorly understood, and the importances of such interactions (to either of the interacting elements, or to other elements in the system) are unknown. We consider that such interactions are of a known type (e.g. predator–prey) and that their signs are therefore known, but which of those interactions should be included in the model is not clear. Simulations were run in two broad phases. First, the suppression of cats and the Myxoma virus within the island ecosystem was simulated. A pool of candidate model structures was assembled, using all possible combinations of presence or absence of the unknown interactions. For each combination, 1000 stable, valid realisations were generated, each with random interaction weights, as described above. Following the eradication of cats and cessation of Myxoma virus releases, the rabbit population increased and the extent of tall tussock vegetation was reduced (Bergstrom et al. 2009; Dowding et al. 2009). These two responses were used as validation data for the ﬁrst phase of simulations. The eradication project was then simulated. Cats and Myxoma were completely removed from the models. We used only those

The target species in a given scenario did not necessarily decrease in response to simulated suppression (Table 2), with non-negative responses of target species generally more prevalent in simulations where multiple species were being targeted simultaneously. For example, simulations suppressing mice alone were almost always successful. In contrast, mice decreased in 65% of cases with simultaneous suppression of mice and rats, and in only 51% of cases with simultaneous suppression of rabbits, rats, and mice. Indeed, for the simultaneous suppression of rabbits, rats, and mice (i.e. the target species of the eradication project), the most common outcome was decreases in rabbits and rats, but not in mice (occurring in 45% of cases). The successful suppression of all three target species occurred in 33% of cases (Table 3). Figure 5 shows the predicted responses for the suppression of rabbits, rats, and mice. These results are presented using only those cases in which the target species actually decreased (that is, eﬀectively assuming that the eradication actions were successful in causing a sustained decline in the target species). Successful suppression of all three target species predicted increases in all native biota except skuas and penguins, with levels of support ranging from 72% to 99% (Fig. 5). Redpolls and starling were universally predicted to increase. The response of skuas was ambiguous, and there was moderate support (80%) for decreases in penguins. The results of other simulated eradication scenarios (i.e. targeting only one of the three target species, as well as combinations of those species and also with redpolls and starlings) are

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 181–191

Modelling of invasive species eradication 185 Table 1. Model community members. Indicators of abundance (abundant, common, uncommon, rare) from Copson (2002) Community member

Description

Penguins

King penguin Aptenodytes patagonicus (abundant) Gentoo penguin Pygoscelis papua (common) Royal penguin Eudyptes schlegeli (abundant) Rockhopper penguin E. chrysocome ﬁlholi (abundant) Subantarctic skua Catharacta lonnbergi (common) Southern giant petrel Macronectes giganteus (common) Northern giant petrel M. halli (common) White-headed petrel Pterodroma lessonii (uncommon) Sooty shearwater Puﬃnus griseus (uncommon) Blue petrel Halobaena caerulea (uncommon) Grey petrel Procellaria cinerea (uncommon) Fairy prion Pachyptila turtur (rare) Soft-plumaged petrel Pterodroma mollis (rare) Diving petrels Pelecanoides georgicus and Pelecanoides urinatrix (rare) Antarctic prion Pachyptila desolata (abundant) Light-mantled sooty albatross Phoebetria palpebrata (common) Black-browed albatross Thalassarche melanophrys (uncommon) Grey-headed albatross Thalassarche chrysostoma (uncommon) Wandering albatross Diomedea exulans (rare) Macquarie Island shag Phalacrocorax purpurascens Antarctic tern Sterna vittata bethunei Cape petrel Daption capense Paciﬁc black duck Anas superciliosa Grey teal Anas gracilis Mallard* Anas platyrhynchos Kelp gull Larus dominicanus Terrestrial and aquatic macro-invertebrates, including moths and moth larvae, spiders, worms, and ﬂies The tall tussock vegetation class (Selkirk et al. 1990). Dominated by the tall tussock grass Poa foliosa and the megaherb Stilbocarpa polaris, with other tall species including Polystichum vestitum The herbﬁeld vegetation class (Selkirk et al. 1990). Dominated by Pleurophyllum hookeri and ⁄ or S. polaris, with Blechnum penna-marina and Acaena spp. The short grassland vegetation class (Selkirk et al. 1990). Meadow-like vegetation dominated by species of Agrostis, Luzula, Uncinia, Deschampsia, or Festuca Feral cat* Felis catus European rabbit* Oryctolagus cuniculus Ship rat* Rattus rattus House mouse* Mus musculus Redpoll* Acanthis ﬂammea (common) Starling* Sturnus vulgaris (common)

Skuas Giant petrels Burrow-nesting seabirds

Antarctic prions Albatrosses

Small surface-nesting seabirds

Macro-invertebrates Tall tussock vegetation system

Herbﬁeld vegetation system

Grassland vegetation system

Cats Rabbits Rats Mice Redpolls and starlings

Asterisk denotes introduced species. Naming authorities can be found in Selkirk, Seppelt & Selkirk (1990).

presented in Fig. S1 (Supporting information). Successful suppression of rabbits and rats, but not targeting mice, yielded broadly similar but much less deﬁnitive results compared with the successful suppression of all three species. Burrow-nesting seabirds, tall tussock vegetation, redpolls and starlings, and giant petrels were all predicted to increase, albeit with lower support (76–86%). However, the responses of albatrosses, macro-invertebrates, small surface-nesting seabirds, skuas, and Antarctic prions were ambiguous. Interestingly, simulations targeting all three pest species, but only successfully suppressing rabbits and rats, gave intermediate results, that is, predictions lying between those obtained with the successful suppression of all three, and those targeting only rabbits and rats (results not shown).

The sensitivities of the predictions to structural uncertainties are presented in Fig. 6. These results apply to the simultaneous suppression of rabbits, rats, and mice. The values indicate the dissimilarities between predicted responses with the link present, and with the link absent, aggregated over all model realisations. The predictions were most sensitive to those structural uncertainties involving small surface-nesting seabirds, and redpolls and starlings. The uncertainties involving small surface-nesting seabirds generally had little eﬀect beyond the seabirds themselves. However, the redpoll and starling uncertainties had implications for the majority of the community members, including members without direct interactions with redpolls and starlings. The community members with the

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 181–191

186 B. Raymond et al.

Fig. 3. Interaction network of subantarctic Macquarie Island. A line terminated with an arrowhead indicates a positive inﬂuence; one terminated with a dot indicates a negative inﬂuence: a link with both an arrow and a dot therefore indicates a predator–prey relationship. Dashed lines indicate interactions that are not well understood. All members have a limiting (negative) self-interaction, but for clarity these are not shown. Red indicates introduced pest species; blue indicates native species; orange indicates self-introduced alien species; green is vegetation.

Discussion MODELLING APPROACH

Fig. 4. Responses to simulated suppression of cats and Myxoma virus on Macquarie Island. Bars indicate the proportion of models in which the response was of a given direction. Light grey, negative response; dark grey, no change; mid-grey, positive response.

largest overall sensitivity to structural uncertainties were small surface-nesting seabirds, the herbﬁeld vegetation complex, Antarctic prions, and skuas.

The approach used here examines outcomes across a large suite of plausible model structures. Results are consistent across all possible model structures are therefore robust, despite the structural uncertainty of the models examined, and are thus potentially informative in a management and planning context. There are a number of potential issues with the modelling methodology used here. Eradication actions involve high death rates of target species, and modelled perturbations are intended to simulate the depression of target populations to near-zero or zero abundances. Model predictions are therefore unlikely to be valid if such actions cause the remaining members of the ecosystem to change the way they interact (Dambacher & Ramos-Jiliberto 2007). This could occur if the target species aﬀects the way two other species interact (Wootton 1994), or if a predator responds to eradication actions by consuming a species it previously did not consume. Such issues did not aﬀect the current study, but may be of concern for more general application to pest eradication modelling. The qualitative predictions made in these analyses represent long-term responses, as the system arrives at a new equilibrium. Field observations following the eradication of cats, spanning only a decade or so, probably represent responses that can be considered as short-term with respect to the longlived species, or mid-term with respect to vegetation and shorter-lived species. These observations should therefore be

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 181–191

Modelling of invasive species eradication 187 Table 2. The percentage of simulations in which each target species was successfully suppressed, for various eradication scenarios Target species in scenario

Species response

Rabbits

Rabbits Rats Mice Redpolls and starlings

98Æ6

Rats

Mice

Rabbits, mice

Rabbits, rats

89Æ0

96Æ2 94Æ6

99Æ4 99Æ5

94Æ7

Mice, rats

Rabbits, rats, mice

Rabbits, rats, mice, redpolls and starlings

91Æ9 65Æ2

96Æ2 80Æ2 51Æ4

94Æ5 60Æ7 41Æ1 51Æ9

Table 3. The percentage of simulations by various outcome, for simulated suppression of rabbits, rats, and mice Outcome: taxa successfully suppressed

Taxa remaining

Percentage of simulations

Rabbits Rats

Mice Redpolls and starlings

45%

Highly likely increase in mice and redpolls and starlings Possible impact on macro-invertebrates, small surface-nesting seabirds, albatrosses, Antarctic prions, skuas, and penguins

Rabbits Rats Mice

Redpolls and starlings

33%

Highly likely increase in redpolls and starlings Possible impact on macro-invertebrates and vegetation

Mice

Rabbits Rats Redpolls and starlings

16%

Highly likely increase in rabbits and rats Likely increase in redpolls and starlings Possible impacts to all native biota

Predicted ecological consequences

Outcomes not listed here (e.g. suppression of rats but not mice or rabbits) occurred in less than 3% of simulations.

compared with model predictions with caution. Short-term responses are known to be poor indicators of long-term responses (Yodzis 1988). In particular, short-term responses are likely to be dominated by direct eﬀects (i.e. those directly related to cat predation in the case of the cat eradication). Rabbit numbers appear to be declining following their initial increase after cat eradication in 2000. The indirect eﬀects of the increased rabbit numbers (e.g. vegetation loss and potential impacts on the integrity of seabird breeding habitat) are still becoming apparent, particularly with albatrosses and other seabird populations that have relatively slow response times. The simulations do not include impacts on non-target species through primary or secondary poisoning, or human disturbance from eradication activities (e.g. trampling of seabird burrows). These risks, along with the actions being undertaken in order to minimize their impacts, are discussed in detail in the project’s environmental impact statement (Parks & Wildlife Service 2009a). Some additional limitations of the modelling approach are outlined in Appendix S2, Supporting information.

CAT AND MYXOMA VIRUS SUPPRESSION SIMULATIONS

Fig. 5. Responses to the simulated suppression of rabbits, rats, and mice on Macquarie Island. Bars indicate the proportion of models in which the response was of a given direction. Light grey, negative response; dark grey, no change; mid-grey, positive response.

The predictions made by the ﬁrst phase of simulations were generally in agreement with observations made on the island, and thus provide some conﬁdence in the predictions for the rabbit, rat, and mouse eradication project.

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 181–191

188 B. Raymond et al.

Fig. 6. Sensitivities of the responses of community members to uncertainties in the model structure, for simulated suppression of rabbits, rats, and mice. SNS, small surface-nesting seabirds; BNS, burrow-nesting seabirds; HF, herbﬁeld vegetation; TTV, tall tussock vegetation; Albs, albatrosses; RS, redpolls and starlings; ANPR, Antarctic prions.

The simulations predicted that burrow-nesting seabirds would likely have decreased following cat and Myxoma virus suppression. This prediction was due to impacts from increased rabbit numbers, combined with the loss of tall tussock vegetation, that together oﬀset the positive beneﬁt of reduced predation from cats. On the island, the decline in breeding success of white-headed petrels and sooty shearwaters from 2006 to 2009 was thought to be due primarily to rabbit grazing impacts (Way, McInnes & Derry 2009). Some burrownesting seabird populations showed favourable responses to the eradication of cats from the island (Schulz, Robinson & Gales 2005; Brothers & Bone 2008; Dowding et al. 2009). However, with the decrease in vegetation, these species are now mainly found around the northern tip of the island and on rock stacks, where rabbits are absent or in low numbers. In some cases, these recoveries may have been driven by shortterm transient eﬀects: the breeding success of grey petrels peaked in 2005 but then declined again in all sites except the northern tip, again thought to be due primarily to decreases in tall tussock vegetation and associated ground instability and erosion (Way & McInnes 2010). Model results matched island skua surveys, which showed an increase in the population on the plateau from 1997 to 2004, whereas the coastal population remained constant (Carmichael 2008). Populations remained relatively stable to 2009 (McInnes, Way & Achurch 2010). Albatrosses were predicted to decrease due to the reduction in tall tussock vegetation. Albatross breeding success remained roughly constant from 2001 to 2007. However, the majority of

albatross breeding areas have now lost substantial areas of tall tussock vegetation. The breeding successes of black-browed, grey-headed, and light-mantled sooty albatrosses have declined over the last two seasons. For black-browed and greyheaded albatrosses, the breeding successes recorded in the 2009 ⁄ 10 season were the lowest for 16 years (McInnes & Way 2010). It is diﬃcult to quantify the extent to which this habitat degradation contributed to these declines, as other inﬂuences (e.g. prey availability) also have signiﬁcant eﬀects on breeding success. The model prediction of decreases in giant petrels was not matched by observations on the island: both giant petrel populations are increasing. These increases are thought to be due to increased adult survival through improved mitigation of ﬁsheries impacts and increased breeding success through reduced disturbance around colonies following the introduction of special management areas and expeditioner education (R. Alderman, unpublished data). These factors were not included in the model.

SUCCESSFUL SUPPRESSION OF RABBITS, RATS, AND MICE

The simulations provided good support for the anticipated positive outcomes of the eradication project, with predicted recoveries of tall tussock vegetation, burrow- and small surface-nesting seabirds, and albatrosses. Redpolls and starlings were, however, also predicted to increase. The eradication project anticipates this as a possible

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 181–191

Modelling of invasive species eradication 189 outcome, with unknown consequences (Parks & Wildlife Service 2009a). Our results suggest that an increase in redpolls and starlings might inhibit the recovery of macro-invertebrates and herbﬁeld vegetation: support for their recoveries with the suppression of rabbits, rats, and mice were moderate (79% and 72%). The additional simulated suppression of redpolls and starlings raised these levels of support to 100%. While starlings have been shown to be capable of exerting signiﬁcant pressures on soil invertebrates and vegetation in other parts of the world (Whitehead, Wright & Cotton 1996; Linz et al. 2007), an improved understanding of the pressures that redpolls and starlings might bring to bear on a post-eradication Macquarie Island ecosystem would allow more rigorous assessment of the risk that these species might present. Redpolls, for example, are known to feed on seed heads of Pleurophyllum hookeri and Leptinella plumosa (Warham 1969; Parks & Wildlife Service 2009a), but it is not known whether they will forage in recovering and increasingly dense tussock and herbﬁeld. The eradication project monitoring plan does not explicitly include any ongoing action regarding redpolls or starlings. Our results suggest that such monitoring and ⁄ or additional action is warranted. Giant petrels were predicted to increase with the recovery of tall tussock vegetation, which could provide expanded northern giant petrel breeding habitat. However, we note that vegetation changes have rarely been implicated in investigations of giant petrel population changes (e.g. Patterson et al. 2008; but see also McInnes & Way 2010), and that the model does not take into account at-sea factors that might be aﬀecting giant petrel populations. This prediction should be therefore be assessed with caution, and similarly the predicted decrease in penguins, which was driven by increased giant petrel predation.

on the mice (Caut et al. 2007). This phenomenon, whereby the decline of a top predator can allow smaller predators to increase, is known as mesopredator release (Rayner et al. 2009). Here, the eﬀect may be exacerbated by the diﬃculty in targeting all individuals of the mouse population (thus reducing the magnitude of the direct negative impact of control actions on the mouse population). Consistent with these previous analyses, our results showed increased failure to suppress mice when targeted as part of a multi-species eradication scenario (Table 2). While mice on Macquarie Island have not yet been observed to have major impacts on seabird or other native species, recent publications have provided graphic documentation of the potential impacts that mice may have if left as the only introduced mammals on the island (e.g. Jones & Ryan 2009). Even relatively simple island ecosystems can have poorly understood ecosystem dynamics, and the eradication of multiple invasive species from such ecosystems is complex because of interdependencies and indirect eﬀects. Our study has shown that quite robust conclusions can be derived about some management actions by using qualitative analyses, without requiring detailed knowledge of the interaction strengths. Our results not only raise the prospect of many successful outcomes for the current eradication actions but also highlight the risks of failure to eradicate mice and the competitive release of redpolls and starlings. Reﬁnements to this model could be made as new data is collected and as the eradication project unfolds.

OTHER ERADICATION SCENARIOS

References

The results support the current actions of simultaneous eradication of all three pest species, with generally sub-optimal outcomes with suppression of only one or two pest species. Brothers & Bone (2008) suggested that if the eradication of all three species was not possible, the eradication of rabbits (but not rodents) would be adequate for the recovery and increased breeding success of the majority of burrow-nesting petrels. Our results showed high support (89%) for this prediction; however, rabbit-only suppression had unsatisfactory outcomes for other taxa, particularly macro-invertebrates. A common outcome of the simulations was the failure to suppress mice. This has been identiﬁed as a risk for the eradication project (Parks & Wildlife Service 2009a). Indeed, in other island eradications, failure to eradicate mice has been more common than failure to eradicate rats (Howald et al. 2007). Mouse eradication is increasingly problematic when other pest species are present (Caut et al. 2007; Harper & Cabrera 2009). This is, at least in part, a result of the structural position of mice within the ecosystem. The simultaneous control of rats and mice can, on balance, be beneﬁcial to mice, if the indirect positive eﬀect to mice (i.e. the suppression or removal of rats) outweighs the direct negative eﬀect of the control programme

Acknowledgements We are grateful to two anonymous reviewers, and to Keith Springer of the Tasmanian Parks and Wildlife Service, whose advice and comments greatly improved this manuscript.

Bender, E., Case, T.J. & Gilpin, M.E. (1984) Perturbation experiments in community ecology: theory and practice. Ecology, 65, 1–13. Bergstrom, D.M., Lucieer, A., Kiefer, K., Wasley, J., Belbin, L., Pedersen, T.K. & Chown, S.L. (2009) Indirect eﬀects of invasive species removal devastate World Heritage Island. Journal of Applied Ecology, 46, 73–81. Berlow, E.L., Neutel, A.-M., Cohen, J.E., De Ruiter, P.C., Ebenman, B., Emmerson, M., Fox, J.W., Jansen, V.A.A., Iwan Jones, J., Kokkoris, G.D., Logofet, D.O., McKane, A.J., Montoya, J.M. & Petchey, O. (2004) Interaction strengths in food webs: issues and opportunities. Journal of Animal Ecology, 73, 585–598. Brothers, N.P. (1984) Breeding, distribution and status of burrow-nesting petrels at Macquarie Island. Australian Wildlife Research, 11, 113–131. Brothers, N. & Bone, C. (2008) The response of burrow-nesting petrels and other vulnerable bird species to vertebrate pest management and climate change on sub-Antarctic Macquarie Island. Papers and Proceedings of the Royal Society of Tasmania, 142, 123–148. Brothers, N.P., Eberhard, I.E., Copson, G.R. & Skira, I.J. (1982) Control of rabbits on Macquarie Island by myxomatosis. Australian Wildlife Research, 9, 477–485. Buckley, Y.M. (2008) The role of research for integrated management of invasive species, invaded landscapes and communities. Journal of Applied Ecology, 45, 397–402. Carmichael, N. (2008) Some Island-Scale Indicators of Rabbit Impacts and Distribution on Macquarie Island. Master’s Thesis, University of New England, Armidale, NSW. Caut, S., Casanovas, J.G., Virgos, E., Lozano, J., Witmer, G.W. & Courchamp, F. (2007) Rats dying for mice: modelling the competitor release eﬀect. Austral Ecology, 32, 858–868.

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 181–191

190 B. Raymond et al. Copson, G.R. (2002) Integrated vertebrate pest management on subantarctic Macquarie Island 1997–2002, Final Report. Department of Primary Industries, Water and Environment, Hobart, Tasmania. Copson, G.R. & Brothers, N.P. (2008) Notes on rare, vagrant and exotic avifauna at Macquarie Island, 1901-2000. Papers and Proceedings of the Royal Society of Tasmania, 142, 105–116. Copson, G.R., Brothers, N.P. & Skira, I.J. (1981) Distribution and abundance of the rabbit, Oryctolagus cuniculus (L.), at subantarctic Macquarie Island. Australian Wildlife Research, 8, 597–611. Copson, G. & Whinam, J. (2001) Review of ecological restoration programme on subantarctic Macquarie Island: pest management progress and future directions. Ecological Management & Restoration, 2, 129–138. Costin, A.B. & Moore, D.M. (1960) The eﬀects of rabbit grazing on the grasslands of Macquarie Island. Journal of Ecology, 48, 729–732. Cumpston, J.S. (1968) Macquarie Island. Australian National Antarctic Research Expedition A.N.A.R.E. Reports, Series A. Commonwealth of Australia, Department of External Aﬀairs, Melbourne. Dambacher, J.M., Li, H.W. & Rossignol, P.A. (2002) Relevance of community structure in assessing indeterminacy of ecological predictions. Ecology, 83, 1372–1385. Dambacher, J.M. & Ramos-Jiliberto, R. (2007) Understanding and predicting eﬀects of modiﬁed interactions through a qualitative analysis of community structure. Quarterly Review of Biology, 82, 227–250. Dambacher, J.M., Luh, H.-K., Li, H.W. & Rossignol, P.A. (2003) Qualitative stability and ambiguity in model ecosystems. The American Naturalist, 161, 876–888. Dowding, J.E., Murphy, E.C., Springer, K., Peacock, A.J. & Krebs, C.J. (2009) Cats, rabbits, Myxoma virus, and vegetation on Macquarie Island: a comment on Bergstrom et al. (2009). Journal of Applied Ecology, 46, 1129– 1132. Falla, R.A. (1937) Birds. British, Australian and New Zealand Antarctic Research Expedition (BANZARE) Reports, Series B. Hassell Press, Adelaide. Grantham, H.S., Wilson, K.A., Moilanen, A., Rebelo, T. & Possingham, H.P. (2009) Delaying conservation actions for improved knowledge: how long should we wait? Ecology Letters, 12, 293–301. Harper, G.A. & Cabrera, L.F. (2009) Response of mice (Mus musculus) to the removal of black rats (Rattus rattus) in arid forest on Santa Cruz Island, Gala´pagos. Biological Invasions, 12, 1449–1452. Hosack, G.R., Hayes, K.R. & Dambacher, J.M. (2008) Assessing model structure uncertainty through an analysis of system feedback and Bayesian networks. Ecological Applications, 18, 1070–1082. Howald, G., Donlan, C.J., Galva´n, J.P., Russell, J.C., Parkes, J., Samaniego, A., Wang, Y., Veitch, D., Genovesi, P., Pascal, M., Saunders, A. & Tershy, B. (2007) Invasive rodent eradication on islands. Conservation Biology, 21, 1258–1268. Jones, M.G.W. & Ryan, P.G. (2009) Evidence of mouse attacks on albatross chicks on sub-Antarctic Marion Island. Antarctic Science, 22, 39–42. Levins, R. (1968) Evolution in Changing Environments: Some Theoretical Explorations. Princeton University, Princeton. Levins, R. (1974) Qualitative analysis of partially speciﬁed systems. Annals of the New York Academy of Sciences, 231, 123–138. Levins, R. (1998) Qualitative mathematics for understanding, prediction and intervention in complex ecosystems. Ecosystem Health: Principles and Practices (eds D. Rapport, R. Costanza, P.R. Epstein, C. Gaudet & R. Levins), pp. 178–204. Blackwell Science, Malden, MA. Linz, G.M., Homan, H.J., Gaukler, S.M., Penry, L.B. & Bleier, W.J. (2007) European starlings: a review of an invasive species with far-reaching impacts. Managing Vertebrate Invasive Species: Proceedings of an International Symposium (eds G.W. Witmer, W.C. Pitt & K.A. Fagerstone). USDA ⁄ APHIS ⁄ WS, National Wildlife Research Center, Fort Collins, CO. Maguire, L.A. (2004) What can decision analysis do for invasive species management? Risk Analysis, 24, 859–868. McInnes, J. & Way, S. (2010) The Conservation and Status of Albatrosses and Giant-Petrels on Macquarie Island: Report on the 2009 ⁄ 2010 Field Season. Unpublished report produced for the Department of Primary Industries, Parks, Water and Environment, Tasmania, Hobart. McInnes, J., Way, S. & Achurch, H. (2010) Macquarie Island Skua Report 2009 ⁄ 10, Census and Dietary Analysis. Unpublished report produced for the Department of Primary Industries, Parks, Water and Environment, Tasmania, Hobart. Montan˜o-Moctezuma, G., Li, H.W. & Rossignol, P.A. (2007) Alternative community structures in a kelp-urchin community: a qualitative modeling approach. Ecological Modelling, 205, 343–354.

Nakajima, H. (1992) Sensitivity and stability of ﬂow networks. Ecological Modelling, 62, 123–133. Parks & Wildlife Service (2008) Macquarie Island Pest Eradication Plan. Part A: Overview. Department of the Environment, Parks, Heritage and the Arts, Hobart, Tasmania. Available from http://www.parks.tas.gov.au/index.aspx?base = 6186. Parks & Wildlife Service (2009a) Macquarie Island Pest Eradication Plan. Part C: Environmental Impact Statement. Department of Primary Industries, Parks, Water and the Environment, Hobart, Tasmania. Available from http://www.parks.tas.gov.au/ﬁle.aspx?id = 16721. Parks & Wildlife Service (2009b) Macquarie Island Pest Eradication Plan. Part F: Monitoring Plan. Department of Primary Industries, Parks, Water and the Environment, Hobart, Tasmania. Patterson, D.L., Woehler, E.J., Croxall, J.P., Cooper, J., Poncet, S., Peter, H.U., Hunter, S. & Fraser, W.R. (2008) Breeding distribution and population status of the Northern Giant Petrel Macronectes halli and Southern Giant Petrel M. giganteus. Marine Ornithology, 36, 115–124. Puccia, C.J. & Levins, R. (1985) Qualitative Modelling of Complex Systems: An Introduction to Loop Analysis and Time Averaging. Harvard University Press, Cambridge, MA. Punt, A.E. & Hilborn, R. (1997) Fisheries stock assessment and decision analysis: the Bayesian approach. Reviews in Fish Biology and Fisheries, 7, 35–63. Ramsay, D.S.L. & Norbury, G.L. (2009) Predicting the unexpected: using a qualitative model of a New Zealand dryland ecosystem to anticipate pest management outcomes. Austral Ecology, 34, 409–421. Ramsay, D. & Veltman, C. (2005) Predicting the eﬀects of perturbations on ecological communities: what can qualitative models oﬀer? Journal of Animal Ecology, 74, 905–916. Rayner, M.J., Hauber, M.E., Imber, M.J., Stamp, R.K. & Clout, M.N. (2009) Spatial heterogeneity of mesopredator release within an oceanic island system. Proceedings of the National Academy of Sciences, 104, 20862–20865. Schulz, M., Robinson, S. & Gales, R. (2005) Breeding of the Grey Petrel (Procellaria cinerea) on Macquarie Island: population size and nesting habit. Emu, 105, 323–329. Scott, J.J. & Kirkpatrick, J.B. (2008) Rabbits, landslips and vegetation change on the coastal slopes of subantarctic Macquarie Island, 1980–2007: implications for management. Polar Biology, 31, 409–419. Selkirk, P.M., Seppelt, R.D. & Selkirk, D.R. (1990) Subantarctic Macquarie Island: Environment and Biology. Cambridge University Press, Cambridge. Sih, A., Crowley, P., McPeek, M., Petranka, J. & Strohmeier, K. (1985) Predation, competition, and prey communities: a review of ﬁeld experiments. Annual Review of Ecology and Systematics, 16, 269–311. Sobey, W.R., Adams, K.M., Johnston, G.C., Gould, L.R., Simpson, K.N.G. & Keith, K. (1973) The introduction of the European rabbit ﬂea Spilopsyllus cuniculi (Dale) as a possible vector for Myxomatosis. The Journal of Hygiene, 71, 299–308. Taylor, B.W. (1955) The Flora, Vegetation and Soils of Macquarie Island. Australian National Antarctic Research Expeditions A.N.A.R.E. Reports, Series B. Commonwealth of Australia, Department of External Aﬀairs, Melbourne. Taylor, R.H. (1979) How the Macquarie Island Parakeet became extinct. New Zealand Journal of Ecology, 2, 42–45. Terauds, A. (2009) Changes in Rabbit Numbers on Macquarie Island 1974–2008. Tasmanian Parks and Wildlife Service, Hobart, Tasmania. Vestjens, W.J.M. (1963) Remains of the extinct banded rail at Macquarie Island. The Emu, 62, 249–250. Warham, J. (1969) Notes on some Macquarie Island birds. Notornis, 16, 190– 197. Way, S. & McInnes, J. (2010) Grey Petrel Procellaria cinerea Monitoring Report, Macquarie Island Marine Park 2009. Unpublished report produced for the Tasmanian Parks and Wildlife Service, Hobart. Way, S., McInnes, J. & Derry, A. (2009) Baseline Population Monitoring of Key Seabird Species, Macquarie Island Marine Park: Report on the 2008 ⁄ 09 Season. Unpublished report produced for the Tasmanian Parks and Wildlife Service, Hobart. Whitehead, S.C., Wright, J. & Cotton, P.A. (1996) Measuring the impact of parental foraging by starlings (Sturnus vulgaris) on soil invertebrate prey availability: an exclosure experiment. Oikos, 76, 511–521. Wootton, J.T. (1994) Putting the pieces together: testing the independence of interactions among organisms. Ecology, 75, 1544–1551. Yodzis, P. (1988) The indeterminacy of ecological interactions as perceived through perturbation experiments. Ecology, 69, 508–515. Received 23 July 2010; accepted 2 November 2010 Handling Editor: Marc Cadotte

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 181–191

Modelling of invasive species eradication 191

Supporting Information Additional Supporting Information may be found in the online version of this article: Appendix S1. Model construction: further details. Appendix S2. Additional limitations of the model. Appendix S3. Example R code.

Table S1. Details of the interactions within the Macquarie Island ecosystem. As a service to our authors and readers, this journal provides supporting information supplied by the authors. Such materials may be re-organized for online delivery, but are not copyedited or typeset. Technical support issues arising from supporting information (other than missing ﬁles) should be addressed to the authors.

Figure S1. Responses to simulated eradication scenarios of rabbits, rats, mice, and redpolls and starlings on Macquarie Island.

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 181–191

Journal of Applied Ecology 2011, 48, 9–13

doi: 10.1111/j.1365-2664.2010.01917.x

Monitoring species abundance and distribution at the landscape scale Julia P. G. Jones* School of Environment, Natural Resources and Geography, Thoday Building, Deniol Road, Bangor University, Bangor, Gwynedd LL57 2UW, UK

Summary 1. The abundance and distribution of a species are aﬀected by processes which operate at multiple scales. Large-scale dynamics are increasingly recognized in conservation responses such as metapopulation management, transfrontier protected areas and softening the agricultural matrix. Landscape-scale monitoring is needed both to inform and judge their eﬃcacy. In this Special Proﬁle we address some of the challenges presented by monitoring at the landscape scale, how models of species distribution can be used to inform policy, and we discuss how monitoring at the global-scale could be approached. 2. Collecting data over a large area is inherently costly, so methods which can provide robust information at low-cost are particularly valuable. We present two papers which test low-cost approaches against more data-hungry methods (indices of abundance vs. direct density estimates, and species distribution models built from presence-only vs. presence ⁄ absence data). 3. Occupancy modelling is a useful approach for landscape-scale monitoring due to the relatively low-cost of collecting detection ⁄ non-detection data. We discuss challenges, such as non-random sampling locations and periodical unavailability for detection, in using detection ⁄ non-detection data for monitoring species distribution. Such data can also provide estimates of abundance and we show how existing models have been modiﬁed to allow the abundance of multiple species to be estimated simultaneously. 4. Models of species distribution can be used to project likely future scenarios and thus inform conservation planning where distributions are likely to change because of climate change or changing disturbance patterns. We also discuss how an optimization framework can be used to make eﬃcient management decisions for invasive species management in the light of imperfect information. 5. Synthesis and applications. Monitoring is needed for many purposes including auditing past management decisions and informing future choices. Much monitoring data are collected at the site scale, although management authorities increasingly recognize landscape-scale dynamics. Recent global targets for conservation require monitoring which can report trends at the global-scale. Integrating data collected at a variety of scales to draw robust inference at the scale required is a challenge which deserves more attention from applied ecologists. Key-words: camera-trap, cost-effective, density, detectability, habitat suitability model, invasive species, migratory, occupancy model, tiger

Introduction Over the past two decades there has been a shift in emphasis among conservation biologists from managing populations of threatened species at a single site, to considering larger scale dynamics (Baillie et al. 2000; Brown, Spector & Wu 2008; Lindenmayer et al. 2008; Levi et al. 2009; Bailey et al. 2010). Such landscape (or seascape) scale approaches to conservation *Correspondence author: E-mail: [email protected]

make sense because the drivers of biodiversity loss (such as habitat conversion and fragmentation, overexploitation and climate change) tend to operate at large scales, and a substantial body of evidence has demonstrated the importance of dispersal between subpopulations and source-sink dynamics in the persistence of a species in a landscape. Consideration of these large-scale dynamics is behind such conservation responses as metapopulation management (Esler 2000; Rouquette & Thompson 2007), transfrontier protected areas (Smith et al. 2008) and drives to soften the agricultural matrix

2011 The Author. Journal of Applied Ecology 2011 British Ecological Society

10 J. P. G. Jones around conservation areas (Donald & Evans 2006; Perfecto & Vandermeer 2010; Koh et al. 2010). Monitoring at the appropriate scale is essential both to judge, and to improve, the eﬃcacy of such approaches (Radford & Bennett 2007). The term ‘monitoring’ has many diﬀerent deﬁnitions and usages. Here we follow Yoccoz, Nichols & Boulinier (2001) in deﬁning monitoring as the process of gathering information about a state variable (such as the abundance or distribution of a species) to assess the state of the system and draw inferences about changes over time. Monitoring species abundance or distribution at the landscape scale presents particular challenges. (i) Data collection over a large area is inherently costly so methods for minimizing costs, always important in any monitoring study, will be particularly signiﬁcant. (ii) Species detectability tends to vary over space or time. Accounting for variable detectability will be particularly important in landscape-scale studies as surveys will inevitably cover a variety of habitats, and may have to be carried out in diﬀerent seasons; both of which will inﬂuence detectability. In this Special Proﬁle we bring together ﬁve papers which address these challenges of monitoring species abundance and distribution at the landscape scale, and three which demonstrate novel ways in which models based on species distribution data can be used to inform policy.

Challenges in monitoring species abundance and distribution at the landscape scale MINIMIZING THE COSTS, WHILE ENSURING DATA QUALITY

Monitoring is costly and in a resource-limited world conservationists will seek to maximize the cost-eﬀectiveness of monitoring, freeing up resources to spend on other activities (Murray et al. 2009). There will often be a trade-oﬀ between the cost of a monitoring method and the quality of the information it provides and if a low-cost method provides noisy data that does not allow trends to be detected robustly, it will be of little use for decision making and the resources invested will be wasted (Legg & Nagy 2005). Validation of low-cost methods against more data-intensive methods is therefore an essential step but is seldom done (Jones et al. 2008; De Barba et al. 2010). In this issue, two papers explicitly validate the use of lower-cost methods against more data-hungry and therefore costly approaches. The ﬁrst (Jhala, Qureshi & Gopal 2011), looks at the cost-eﬀectiveness of indices of abundance relative to obtaining direct estimates of density for a threatened species across a landscape. The second (Gormley et al. 2011), compares the value of presence-only to presence ⁄ absence data for modelling the distribution of an invasive species. Counts of some easily measurable sign that is assumed to correlate with population density are often used in place of density estimates because the data collection is less costly. However, the use of such indices of abundance has been widely criticized as they are seldom calibrated with direct estimates of density, or tested for precision in detecting change in population size (MacKenzie & Kendall 2002). The tiger Panthera tigris is a highly charismatic and threatened species with popu-

lations scattered across 13 countries. The importance of robust monitoring was highlighted when it emerged that tigers had gone extinct from the Sariska Tiger Reserve, India, during a period when the oﬃcial census continued to report a healthy tiger population in the area (Narain et al. 2005). Appropriate methods for monitoring the tiger have been vigorously debated. Work by Karanth et al. (2003) showed the weakness of the widely used pugmark census technique. However, although camera-trap based mark–recapture can provide precise estimates of tiger densities (Karanth et al. 2006), it is expensive and more suited to use at relatively small scales (Linkie et al. 2006). There is therefore a need for a quantitative assessment of cheaper and easier methods that can be deployed at the landscape scale. In this Special Proﬁle we present the work by scientists from the Wildlife Institute of India and the National Tiger Conservation authority calibrating indices of abundance (pugmarks and scats encountered per kilometre searched) against estimates of tiger density made at the same time using state-of-the-art camera-trap mark–recapture methods. Their results (Jhala, Qureshi & Gopal 2011) are encouraging and suggest that indices of tiger abundance can reliably indicate tiger density, across a range of habitats and densities, and at much lower cost than camera-trap mark– recapture. Of course these results will not negate the need for detailed and precise studies in targeted areas. However, the work will be valuable for developing a cost-eﬀective landscapescale tiger monitoring programme. Cost-eﬀective monitoring is equally important in the study of invasive species. Knowing the current distribution of an invasive species, and being able to predict its potential distribution, is important information for those seeking to eradicate or control it (Hulme 2006). Opportunistic sightings (presence-only data) are often available at relatively low-cost but due to inherent biases may be expected to give less valuable information than data from carefully designed ﬁeld surveys. In this Special Proﬁle Gormley et al. (2011) compare estimates of the current and potential distribution of the introduced sambar deer Cervus unicolor in Australia, from models built using presence-only data, and presence ⁄ absence data. They were able to robustly estimate the current and potential distribution of sambar deer from either data set and could identify priority areas for surveillance monitoring to detect an expansion of the species range even from relatively low-cost and easily obtainable presence-only data.

ACCOUNTING FOR VARIABLE DETECTABILITY (IN SPACE AND TIME)

In 2004 the Journal of Applied Ecology published a Special Proﬁle on new paradigms in species distribution modelling. Since then approaches such as information-theoretic-based model selection and incorporating remotely sensed data into habitat suitability models have, as predicted in the accompanying editorial (Rushton, Ormerod & Kerby 2004), become mainstream. The methods used to model species distributions continue to advance, providing ever more powerful

2011 The Author. Journal of Applied Ecology 2011 British Ecological Society, Journal of Applied Ecology, 48, 9–13

Monitoring at the landscape scale 11 approaches for targeting monitoring and management of threatened or invasive species (see also Gormley et al. 2011; Singh & Milner-Gulland 2011). Habitat suitability models are often constructed using data on occupancy (the fraction of sampling units in a landscape where a species is present). However a species that is present in an area may go undetected, whatever method is used for surveying. Unless such false absences can be accounted for, apparent changes in occupancy and estimates of habitat preference will only be valid if detectability of the species remains stable over time and across diﬀerent habitats (Mackenzie & Royle 2005); an unrealistic assumption. Occupancy modelling makes use of repeat surveys of sampling units to explicitly model and estimate detection probability. The approach was developed more than 20 years ago but was not well used until advances in computing made it possible to fully exploit its potential. Occupancy models, and extensions such as dynamic occupancy models (MacKenzie et al. 2003) and multi-scale occupancy models (Mordecai 2007; Nichols et al. 2008), are now used with diﬀerent types of data including that provided by cameratraps (Rowcliﬀe et al. 2008) and from multiple detection methods in one study (Nichols et al. 2008), to allow for temporary emigration or immigration (Rota et al. 2009) and to identify predictors of detectability as a guide for focusing future monitoring eﬀort (Guillera-Arroita et al. 2010). Occupancy modelling is a particularly useful approach for landscape-scale monitoring due to the relatively lower cost of collecting detection ⁄ non-detection data relative to surveys estimating abundance (Joseph et al. 2006) and because larger-scale studies may cover a greater range of habitats resulting in more variation in detectability than in smaller-scale studies. Two papers included in this Special Proﬁle present further advances in the use of this ﬂexible and useful modelling approach. In all surveys, the location of sampling units would ideally be fully independent. However in many situations it is more practical to locate sampling units (point counts, transects, etc.) along an existing access route or transects than to site them randomly, and in studies of low-density and rare species nonrandom sampling may be done to increase detection. Both these reasons for non-random sampling may introduce biases. Van der Burg et al. (2010) use random eﬀects in a Bayesian hierarchical model to account for spatial dependence of sampling units, allowing them to account for such biases. They use their resulting models of the distribution of a rare bird (the mountain plover Charadrius montanus) to investigate the eﬃcacy of management of this species. Conservationists and environmental managers often have to deal with relatively sparse or poor-quality data (because a study has been poorly designed or because the species is rare). The work by Van der Burg et al. (2011) shows how careful modelling can maximize the value of such data to conservation and management. Mordecai et al. (2011) focus on the challenges posed when studying mobile or episodic species (which are therefore only sometimes available for detection) in a landscape where survey locations need to be spatially clustered for logistical reasons. Although these challenges have been addressed in previous studies, they extend the occupancy modelling framework to

account for both problems simultaneously. The approach they use builds on previous multi-scale occupancy models (Mordecai 2007; Nichols et al. 2008) and allows simultaneous estimation of occupancy (which they deﬁne as the probability that a site is occupied by a species at least once in the survey period), use (the probability that the species is available to be detected given that the site is occupied), and detection probability (the probability that the species is detected on a given visit given that the site is being used). In traditional occupancy models, use and detection are confounded into a single parameter. Mordecai et al. make a convincing case that there are many ecological problems where separating these parameters, as their model allows, is important. For example, where an animal may be absent from much of its home range at any given time and a researcher wants to investigate patterns of occupancy. Like Van der Burg et al. (2011), they also deal with issues of spatial dependency of sampling locations in their model using random eﬀects. They argue that their approach has a wide application for studying clustered detection–nondetection data for elusive species across a range of spatial and temporal scales. Ecologists often not only want to know a species’ distribution but how many individuals there are. Royle & Nichols (2003) showed how simple detection–non-detection data can provide information on abundance. In this issue, Yamaura et al. (2011) extend the Royle & Nichols model to estimate the abundance of a number of species simultaneously. They apply their model to multiple repeat bird surveys in a forest recovering after a ﬁre. This model has enormous potential for application to landscape-scale monitoring problems as it can be used to extract information on community structure as well as the dynamics of individual species from relatively easily obtained data.

Using landscape-scale species data to inform policy A number of papers included in this Special Proﬁle (Van der Burg et al. 2011; Gormley et al. 2011; Mordecai et al. 2011) suggest ways that models of species distributions can be improved or can make more eﬃcient use of available data. Three further papers consider how such models can be used to make the best use of available information to inform policy. Singh & Milner-Gulland (2011) use a combination of long-term aerial survey data, remote-sensed habitat data, and projections of likely future scenarios of climate change and disturbance to inform landscape-scale conservation for a migratory ungulate. Regan, Chade´s & Possingham (2011), and Baxter & Possingham (2011) show how optimization frameworks can be used to make eﬃcient decisions for invasive species management in the light of imperfect information on species distributions. The large-scale movements of migratory species, mean that monitoring and management at the landscape-scale is necessary (Sutherland 1998). Although many migrations follow relatively established routes, their precise location and timing may vary and may be inﬂuenced by climate change and human disturbance (Wilcove & Wikelski 2008). Such variability makes

2011 The Author. Journal of Applied Ecology 2011 British Ecological Society, Journal of Applied Ecology, 48, 9–13

12 J. P. G. Jones it diﬃcult to monitor trends in abundance or distribution, or chose where to locate static conservation interventions. The saiga antelope Saiga tatarica is considered Critically Endangered following a 95% decline in population size over the last two decades (Milner-Gulland et al. 2003) but the understanding of trends is hampered by biases in the monitoring techniques which have been used (McConville et al. 2009). Saiga are currently the focus of considerable conservation attention in parts of its range and, given the level of threat facing the species, it is essential that this eﬀort is eﬀectively deployed. Information on current, and likely future, distributions is required to inform conservation planning, and monitoring that can robustly detect trends given the challenges posed by the species’ ecology is needed. A remarkable data set from 25 years of aerial surveys allowed Singh & Milner-Gulland (2011) to identify factors predicting the spring distributions of saiga in Kazakhstan. They used their model to predict the distribution under likely scenarios of climate change and human disturbance. They found that the distribution and density of saiga has changed over time and is likely to continue to change into the future. Their models of likely future distributions can be used to improve the placement of planned protected areas. The approach taken by Singh & Milner-Gulland (2011) – using species distribution models in conjunction with projections of future scenarios – is likely to be useful in targeting monitoring and conservation action at the landscape scale in a range of circumstances. Another paper in this Special Proﬁle (Regan, Chade´s & Possingham 2011) considers how the detectability of a species (and therefore how costly it is to get reliable information on a species’ distribution) inﬂuences the optimal strategy for its management. Using the case study of broomrape Orobanche ramosa, an invasive parasitic plant in Australia, they show how the optimal strategy for eradicating the species depends, in part, on the species’ detectability. Using a partially observable Markov decision process (POMDP) they show that costly, eﬀective actions for controlling the species (such as soil fumigation), should be used in preference to less eﬀective but lower cost methods (such as reducing the availability of hosts) if detection rates are high. However when detection rates are low, it is optimal to continue managing the species using the lower cost methods even when the species is not detected, in order to buﬀer against this low detectability. Despite the great advances in using occupancy models to estimate detectability, this novel study may be the ﬁrst to explicitly consider how variable detection might change the optimal management strategy for a species. Staying with invasive species management in Australia, Baxter & Possingham (2011) consider the trade-oﬀ between investing in action to control the invasive ant Solenopsis invicta and further surveys to allow better predictive maps beneﬁting future searches. Their work shows the importance of investing in knowledge, as long as that knowledge acquisition has a clear purpose for informing management. Their work has relevance wherever ecologists seek to predict species distribution for the purpose of making management decisions; for example identifying priority areas for new protected areas. Given the diﬃ-

culty and cost of gathering information over the vast range of the saiga antelope, this framework could perhaps be applied to the case study highlighted by Singh & Milner-Gulland to help identify the optimal investment in further monitoring to improve understanding of the species’ likely distribution vs. investment in conservation interventions on the ground.

Scaling up from the landscape scale? Natural ecosystems interact at all spatial scales. For example, the persistence and abundance of a species at a site may be inﬂuenced by the level of exploitation it is exposed to at that location, by the pattern of habitat fragmentation at the landscape scale, and by the spread of a damaging invasive species around the globe. Therefore, although monitoring and management at the landscape scale is important and justiﬁably attracting increased attention, it will not always be suﬃcient for understanding and combating biodiversity loss. Targets for conserving biodiversity are increasingly set globally (Perrings et al. 2011) and the failure to meet the ﬁrst global biodiversity target was blamed partially on at the lack of measurable targets and appropriate monitoring (Butchart et al. 2010). Monitoring which can provide robust inference at the global scale is therefore needed to properly audit progress against these global targets (Jones et al. 2011). However this is not necessarily a call for a prescriptive, top-down global monitoring programme as that would be prohibitively expensive (Scholes et al. 2008). Most monitoring is carried out to meet local or regional management objectives and cost-eﬀective global-scale biodiversity monitoring will have to make use of these data. As the papers in this Special Proﬁle demonstrate, applied ecologists are grappling with the challenges of monitoring eﬀectively and eﬃciently at the landscape scale. Interesting and important challenges remain in tackling how monitoring data, collected at a variety of scales, can be integrated to monitor biodiversity change globally.

Acknowledgements The author thanks E.J. Milner-Gulland, Aidan Keane, Neal Hockley, Brady Mattsson, Tracey Regan and Andy Royle for helpful discussion and comments.

References Bailey, D., Schmidt-Entling, M. H., Eberhart, P., Herrmann, J. D., Hofer, G., Kormann, U. & Herzog, F. (2010) Eﬀects of habitat amount and isolation on biodiversity in fragmented traditional orchards. Journal of Applied Ecology, 47, 1003–1013. Baillie, S. R., Sutherland, W. J., Freeman, S. N., Gregory, R. D. & Paradis, E. (2000) Consequences of large-scale processes for the conservation of bird populations. Journal of Applied Ecology, 37, 88–102. Baxter, P. & Possingham, H. P. (2011) Optimizing search strategies for invasive pests: learn before you leap. Journal of Applied Ecology, 48, 86–95. Brown, K. A., Spector, S. & Wu, W. (2008) Multi-scale analysis of species introductions: combining landscape and demographic models to improve management decisions about non-native species. Journal of Applied Ecology, 45, 1639–1648. Butchart, S. H. M., Walpole, M., Collen, B., van Strien, A., Scharlemann, J. P. W., Almond, R. E. A., Baillie, J. E. M., Bomhard, B., Brown, C., Bruno, J., Carpenter, K. E., Carr, G. M., Chanson, J., Chenery, A. M., Csirke, J.,

2011 The Author. Journal of Applied Ecology 2011 British Ecological Society, Journal of Applied Ecology, 48, 9–13

Monitoring at the landscape scale 13 Davidson, N. C., Dentener, F., Foster, M., Galli, A., Galloway, J. N., Genovesi, P., Gregory, R. D., Hockings, M., Kapos, V., Lamarque, J. F., Leverington, F., Loh, J., McGeoch, M. A., McRae, L., Minasyan, A., Morcillo, M. H., Oldﬁeld, T. E. E., Pauly, D., Quader, S., Revenga, C., Sauer, J. R., Skolnik, B., Spear, D., Stanwell-Smith, D., Stuart, S. N., Symes, A., Tierney, M., Tyrrell, T. D., Vie, J. C. & Watson, R. (2010) Global biodiversity: indicators of recent declines. Science, 328, 1164–1168. De Barba, M., Waits, L. P., Genovesi, P., Randi, E., Chirichella, R. & Cetto, E. (2010) Comparing opportunistic and systematic sampling methods for noninvasive genetic monitoring of a small translocated brown bear population. Journal of Applied Ecology, 47, 172–181. Donald, P. F. & Evans, A. D. (2006) Habitat connectivity and matrix restoration: the wider implications of agri-environment schemes. Journal of Applied Ecology, 43, 209–218. Esler, D. (2000) Applying metapopulation theory to conservation of migratory birds. Conservation Biology, 14, 366–372. Gormley, A., Forsyth, D., Griﬃoen, P., Lindeman, M., Ramsey, D., Scroggie, M. & Woodford, L. (2011) Using presence-only and presence-absence data to estimate the current and potential distributions of established invasive species. Journal of Applied Ecology, 48, 25–34. Guillera-Arroita, G., Lahoz-Monfort, J. J., Milner-Gulland, E. J., Young, R. P. & Nicholson, E. (2010) Using occupancy as a state variable for monitoring the critically endangered Alaotran gentle lemur Hapalemur alaotrensis. Endangered Species Research, 11, 157–166. Hulme, P. E. (2006) Beyond control: wider implications for the management of biological invasions. Journal of Applied Ecology, 43, 835–847. Jhala, Y., Qureshi, Q. & Gopal, R. (2011) Can the abundance of tigers be assessed from their signs? Journal of Applied Ecology, 48, 14–24. Jones, J. P. G., Andriamarovolona, M. M., Hockley, N., Gibbons, J. M. & Milner-Gulland, E. J. (2008) Testing the use of interviews as a tool for monitoring trends in the harvesting of wild species. Journal of Applied Ecology, 45, 1205–1212. Jones, J. P. G., Collen, B., Atkinson, G., Baxter, P. W. J., Bubb, P., Illian, J. B., Katzner, T. E., Keane, A., Loh, J., McDonald-Madden, E., Nicholson, E., Pereira, H. M., Possingham, H. P., Andrew, S. P., Rodrigues, A. S. L., RuizGutierrez, V., Sommerville, M. & Milner-Gulland, E. J. (2011) The why, what and how of global biodiversity indicators beyond the 2010 target. Conservation Biology., DOI: 10.1111/j.1523-1739.2010.01605.x. Joseph, L. N., Field, S. A., Wilcox, C. & Possingham, H. P. (2006) Presenceabsence versus abundance data for monitoring threatened species. Conservation Biology, 20, 1679–1687. Karanth, K. U., Nichols, J. D., Seidensticker, J., Dinerstein, E., Smith, J. L. D., McDougal, C., Johnsingh, A. J. T., Chundawat, R. S. & Thapar, V. (2003) Science deﬁciency in conservation practice: the monitoring of tiger populations in India. Animal Conservation, 6, 141–146. Karanth, K. U., Nichols, J. D., Kumar, N. S. & Hines, J. E. (2006) Assessing tiger population dynamics using photographic capture-recapture sampling. Ecology, 87, 2925–2937. Koh, L. P., Lee, T. M., Sodhi, N. S. & Ghazoul, J. (2010) An overhaul of the species-area approach for predicting biodiversity loss: incorporating matrix and edge eﬀects. Journal of Applied Ecology, 47, 1063–1070. Legg, C. J. & Nagy, L. (2005) Why most conservation monitoring is, but need not be, a waste of time. Journal of Environmental Management, 78, 194–199. Levi, T., Shepard, G. H., Ohl-Schacherer, J., Peres, C. A. & Yu, D. W. (2009) Modelling the long-term sustainability of indigenous hunting in Manu National Park, Peru: landscape-scale management implications for Amazonia. Journal of Applied Ecology, 46, 804–814. Lindenmayer, D., Hobbs, R. J., Montague-Drake, R., Alexandra, J., Bennett, A., Burgman, M., Cale, P., Calhoun, A., Cramer, V., Cullen, P., Driscoll, D., Fahrig, L., Fischer, J., Franklin, J., Haila, Y., Hunter, M., Gibbons, P., Lake, S., Luck, G., MacGregor, C., McIntyre, S., Mac Nally, R., Manning, A., Miller, J., Mooney, H., Noss, R., Possingham, H., Saunders, D., Schmiegelow, F., Scott, M., Simberloﬀ, D., Sisk, T., Tabor, G., Walker, B., Wiens, J., Woinarski, J. & Zavaleta, E. (2008) A checklist for ecological management of landscapes for conservation. Ecology Letters, 11, 78–91. Linkie, M., Chapron, G., Martyr, D. J., Holden, J. & Leader-Williams, N. (2006) Assessing the viability of tiger subpopulations in a fragmented landscape. Journal of Applied Ecology, 43, 576–586. MacKenzie, D. I. & Kendall, W. L. (2002) How should detection probability be incorporated into estimates of relative abundance? Ecology, 83, 2387–2393. Mackenzie, D. I. & Royle, J. A. (2005) Designing occupancy studies: general advice and allocating survey eﬀort. Journal of Applied Ecology, 42, 1105–1114. MacKenzie, D. I., Nichols, J. D., Hines, J. E., Knutson, M. G. & Franklin, A. B. (2003) Estimating site occupancy, colonization, and local extinction when a species is detected imperfectly. Ecology, 84, 2200–2207.

McConville, A. J., Grachev, I. A., Keane, A., Coulson, T., Bekenov, A. B. & Milner-Gulland, E. J. (2009) Reconstructing the observation process to correct for changing detection probability of a critically endangered species. Endangered Species Research, 6, 231–237. Milner-Gulland, E. J., Bukreevea, O. M., Coulson, T., Lushchekina, A. A., Kholodova, M. V., Bekenov, A. B. & Grachev, I. A. (2003) Conservation – reproductive collapse in saiga antelope harems. Nature, 422, 135–135. Mordecai, R. S. (2007) Ecological patterns of occupancy and use: new methods for analysis and applications to species and communities. PhD thesis, University of Georgia, Georgia. Mordecai, R., Mattsson, B., Tzilkowski, C. J. & Cooper, R. (2011) Addressing challenges when studying mobile or episodic species: Hierarchical Bayes estimation of occupancy and use. Journal of Applied Ecology, 48, 56–66. Murray, J. V., Goldizen, A. W., O’Leary, R. A., McAlpine, C. A., Possingham, H. P. & Choy, S. L. (2009) How useful is expert opinion for predicting the distribution of a species within and beyond the region of expertise? A case study using brush-tailed rock-wallabies Petrogale penicillata. Journal of Applied Ecology, 46, 842–851. Narain, S., Panwar, H. S., Gadgil, M., Thapar, V. & Singh, S. P. (2005) Joining the dots: The report of the Tiger Task Force. Project Tiger Directorate, Union Ministry of Environment, Government of India, New Delhi. Nichols, J. D., Bailey, L. L., O’Connell, A. F., Talancy, N. W., Grant, E. H. C., Gilbert, A. T., Annand, E. M., Husband, T. P. & Hines, J. E. (2008) Multiscale occupancy estimation and modelling using multiple detection methods. Journal of Applied Ecology, 45, 1321–1329. Perfecto, I. & Vandermeer, J. (2010) The agroecological matrix as alternative to the land-sparing ⁄ agriculture intensiﬁcation model. Proceedings of the National Academy of Sciences of the United States of America, 107, 5786–5791. Perrings, C., Naeem, S., Ahrestani, F., Bunker, D. E., Burkill, P., Canziani, G., Elmqvist, T., Ferrati, R., Fuhrman, J., Jaksic, F., Kawabata, Z., Kinzig, A., Mace, G. M., Milano, F., Mooney, H., Prieur-Richard, A. H., Tschirhart, J. & Weisser, W. (2010) Ecosystem Services for 2020. Science, 330, 323–324. Radford, J. Q. & Bennett, A. F. (2007) The relative importance of landscape properties for woodland birds in agricultural environments. Journal of Applied Ecology, 44, 737–747. Regan, T., Chade´s, I. & Possingham, H. (2011) Optimally managing under imperfect detection: a method for plant invasions. Journal of Applied Ecology, 48, 76–85. Rota, C. T., Fletcher, R. J., Dorazio, R. M. & Betts, M. G. (2009) Occupancy estimation and the closure assumption. Journal of Applied Ecology, 46, 1173–1181. Rouquette, J. R. & Thompson, D. J. (2007) Patterns of movement and dispersal in an endangered damselﬂy and the consequences for its management. Journal of Applied Ecology, 44, 692–701. Rowcliﬀe, J. M., Field, J., Turvey, S. T. & Carbone, C. (2008) Estimating animal density using camera traps without the need for individual recognition. Journal of Applied Ecology, 45, 1228–1236. Royle, J. A. & Nichols, J. D. (2003) Estimating abundance from repeated presence-absence data or point counts. Ecology, 84, 777–790. Rushton, S. P., Ormerod, S. J. & Kerby, G. (2004) New paradigms for modelling species distributions? Journal of Applied Ecology, 41, 193–200. Scholes, R. J., Mace, G. M., Turner, W., Geller, G. N., Jurgens, N., Larigauderie, A., Muchoney, D., Walther, B. A. & Mooney, H. A. (2008) Ecology – toward a global biodiversity observing system. Science, 321, 1044–1045. Singh, N. J. & Milner-Gulland, E. J. (2011) Conserving a moving target: planning protection for a migratory species as its distribution changes. Journal of Applied Ecology, 48, 35–46. Smith, R. J., Easton, J., Nhancale, B. A., Armstrong, A. J., Culverwell, J., Dlamini, S. D., Goodman, P. S., Loﬄer, L., Matthews, W. S., Monadjem, A., Mulqueeny, C. M., Ngwenya, P., Ntumi, C. P., Soto, B. & LeaderWilliams, N. (2008) Designing a transfrontier conservation landscape for the Maputaland centre of endemism using biodiversity, economic and threat data. Biological Conservation, 141, 2127–2138. Sutherland, W. J. (1998) The eﬀect of local change in habitat quality on populations of migratory species. Journal of Applied Ecology, 35, 418–421. Van der Burg, M. P, Bly, B., VerCauteren, T. & Tyre, A. J. (2011) Making better sense of monitoring data from low density species using a spatially explicit modeling approach. Journal of Applied Ecology, 48, 47–55. Wilcove, D. S. & Wikelski, M. (2008) Going, going, gone: Is animal migration disappearing? PLoS Biology, 6, 1361–1364. Yamaura, Y., Royle, J. A., Kuboi, K., Tada, T., Ikeno, S. & Makino, S. (2011) Modeling community dynamics based on species-level abundance models from detection ⁄ nondetection data. Journal of Applied Ecology, 48, 67–75. Yoccoz, N. G., Nichols, J. D. & Boulinier, T. (2001) Monitoring of biological diversity in space and time. Trends in Ecology & Evolution, 16, 446–453.

2011 The Author. Journal of Applied Ecology 2011 British Ecological Society, Journal of Applied Ecology, 48, 9–13

Journal of Applied Ecology 2011, 48, 153–162

doi: 10.1111/j.1365-2664.2010.01918.x

Independent effects of habitat loss, habitat fragmentation and structural connectivity on the distribution of two arboreal rodents Alessio Mortelliti1,2*, Giovanni Amori1, Dario Capizzi3, Cristina Cervone2, Stefano Fagiani2, Barbara Pollini2 and Luigi Boitani2 1

CNR-Institute for Ecosystem Studies, c ⁄ o Department of Animal and Human Biology, Sapienza University of Rome, Viale dell’Universita` 32, 00185, Rome, Italy; 2Department of Biology and Biotechnology ‘‘Charles Darwin’’, Sapienza University of Rome, Viale dell’Universita` 32, 00185, Rome, Italy; and 3Arp, Regional Park Agency, via del Pescaccio 96, 00166 Rome, Italy

Summary 1. Habitat loss must be distinguished from habitat fragmentation so that appropriate conservation management can be applied. Few studies have evaluated the independent eﬀects of habitat loss and habitat fragmentation on the distribution of vertebrates, and none has evaluated the independent eﬀect of changes in structural connectivity. We carried out a landscape-scale experiment to assess the independent contribution of these three processes and to examine what landscape scale factors aﬀect the distribution of two forest-dependent arboreal rodents: the hazel dormouse Muscardinus avellanarius and the red squirrel Sciurus vulgaris. 2. Habitat loss, rather than habitat fragmentation per se, was the major driver of distribution patterns for both species. As predicted, structural connectivity (hedgerow networks) played an important role in determining the distribution of the hazel dormouse, but not of the red squirrel. 3. Our models predict that long lengths of hedgerows (>30 km) are unlikely to increase the probability of occurrence of hazel dormouse in landscapes where there are low levels of forest cover (<5%–10%). 4. Synthesis and applications. Our empirical ﬁndings indicate that structural connectivity and habitat loss may have additive eﬀects on vertebrate distribution. For the hazel dormouse, improving structural connectivity will be ineﬀective if the amount of forest cover in the landscape is less than 5–10%. The key message from this study is that resources should not be invested in landscape linkages until their eﬃcacy for the given level of suitable habitat has been assessed. Key-words: corridors, forest cover, fragmentation, hedgerows, Italy, landscape, Muscardinus avellanarius, Sciurus vulgaris, spatial conﬁguration

Introduction Habitat loss and fragmentation are commonly regarded as being among the greatest threats to global biodiversity and are major processes contributing to landscape change. Habitat loss and habitat fragmentation are two distinct processes, however the term ‘habitat fragmentation’ is often used ambiguously for both processes. Habitat loss is the loss of habitat for a given species from an area whilst habitat fragmentation per se is the breaking apart of formerly contiguous habitat (Fahrig 2003).

*Correspondence author. E-mail: [email protected]

Disentangling the diﬀerent processes often included in the term ‘habitat fragmentation’ is crucial to identify and quantify the underlying mechanisms that threaten species and ecosystems. Although these processes are strongly interrelated in real landscapes, their distinction is important because they require diﬀerent conservation measures (Lindenmayer & Fischer 2007). The extent of habitat and the spatial conﬁguration of elements (key-elements of the fragmentation paradigm) are only two of several properties of landscapes that may aﬀect the distribution of species. A holistic view of landscapes has been proposed by Bennett, Radford & Haslem (2006) who framed four properties of land mosaics (agricultural landscapes with

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society

154 A. Mortelliti et al. diﬀerent land uses): (i) total extent of habitat (e.g. forest cover); (ii) composition of the mosaic (proportions of diﬀerent landuse types); (iii) spatial conﬁguration of elements; (iv) geographic position and physical environment. The spatial conﬁguration of elements is an emergent property of land mosaics and habitat fragmentation per se is the most important process determining it. Bennett, Radford & Haslem (2006) identiﬁed four aspects of the conﬁguration of habitat patches: the ﬁrst three – subdivision (number of patches), aggregation (clumping of the patches) and symmetry (referring to the skew in distribution of elements) – are strictly related to the intuitive concept of ‘fragmentation per se’. According to Bennett, Radford & Haslem (2006) the fourth aspect, structural connectivity, refers to the physical continuity of elements in the landscape. In agricultural mosaics such continuity is often provided by hedgerows (Bennett, Henein & Merriam 1994; Davies & Pullin 2007). When structural connectivity is provided by hedgerows, it is clearly distinct from other aspects of spatial conﬁguration in that for a given level of aggregation and subdivision of patches in a landscape the complexity of the hedgerow networks may be diﬀerent. Here we consider structural connectivity (in the speciﬁc form of hedgerow networks) as an independent (albeit interrelated and co-occurring) process. The eﬀectiveness of hedgerows as corridors has been reviewed by Davies & Pullin (2007) who concluded that there is no empirical evidence to substantiate their eﬀectiveness at the landscape-scale (but see Tewksbury et al. 2002 for a broader perspective on the eﬀectiveness of corridors in increasing plant and animal movements as well community interactions). The amount of hedgerows in the landscape has been considered by previous studies (Goheen et al. 2003; Moore & Swihart 2005; Pereira & Rodrı` guez 2010); nevertheless, very few landscape-scale studies speciﬁcally designed to separate the role of habitat loss and habitat fragmentation, have included a quantiﬁcation of linear elements (e.g. Haslem & Bennett 2008). A research priority is therefore to assess the contribution of structural connectivity at the landscape scale in conjunction with testing the independent role of habitat loss and habitat fragmentation. Most fragmentation studies are carried out at the patch or patch-landscape scale (sensu McGarigal & Cushman 2002) that do not allow either mosaic-level inference or the disentanglement of the roles of the diﬀerent processes such as habitat loss and habitat fragmentation (Fahrig 2003; Bennett, Radford & Haslem 2006), and often overlook the role of the surrounding matrix (Prugh et al. 2008). Landscape scale studies – where both the independent and dependent variables are included at the landscape scale – are not uncommon (e.g. Trzcinski, Fahrig & Merriam 1999), but few are based on the deliberate selection of independent landscapes for systematic sampling and analysis (McGarigal & McComb 1995; Trzcinski, Fahrig & Merriam1999; Villard, Trzcinski & Merriam 1999; Radford, Bennett & Cheers 2005; Radford & Bennett 2007). These studies generally report that the extent of habitat is more important than the conﬁguration (McGarigal & McComb 1995; Trzcinski, Fahrig & Merriam 1999; Radford

& Bennett 2007). Caution is warranted, however, since generalizations are hampered by the fact that studies are biased towards birds (Fahrig 2003; Bennett, Radford & Haslem 2006). In order to disentangle the diﬀerential contribution of habitat loss, habitat fragmentation per se and disruption of structural connectivity, we carried out a landscape scale mensurative experiment (McGarigal & McComb 1995). Our study design allowed us to separate the contribution of the different processes: for a given amount of habitat we selected landscapes with contrasting spatial conﬁguration of habitat patches and with a gradient of structural connectivity. Since no other research with a similar study design has focused on mammals, we chose two forest-dependent arboreal rodents as model species: the hazel dormouse Muscardinus avellanarius (L.) and the red squirrel Sciurus vulgaris (L.). These two species are known to be sensitive to the loss and fragmentation of forest habitat (Wauters et al. 1994; Rodriguez & Andren 1999; Verbeylen, De Bruyn & Matthysen 2003; Koprowski 2005). We predicted that habitat loss would be the main determinant of distribution patterns and that habitat fragmentation per se would play a secondary role (Fahrig 2003). Following Bright (1998), we also predicted that connectivity would be more important for the hazel dormouse than for the red squirrel, which is able to cross non-forested lands more easily (Mortelliti, Santulli Sanzo & Boitani 2009; Wauters et al. 2010). In order to establish the value of hedgerows as a conservation measure, we explored whether the inﬂuence of structural connectivity varies depending on the amount of forest habitat in the landscape.

Materials and methods STUDY AREA

The study area was located in central Italy (Fig. S1, Supporting Information) encompassing an area of 18 000 km2 (4320¢N 1120¢E to 4127¢N 1351¢E). All the landscape units (41 squares measuring 4 · 4 km) were located within the climax of deciduous oak woodland (Quercus cerris and Quercus pubescens). Land use patterns varied across the study area, with extensive cereal cultivations dominating in plain and coastal areas, and orchards (olives, hazelnuts and vineyards) on the hills. The central, south-east and south western areas are characterized by large urban settlements. There is a system of relatively well connected hedgerows, quite narrow (often width is <5 m), and composed of thick shrub vegetation (main species are Rubus sp., Crataegus monogyna, Rosa canina) and oaks (Quercus sp.).

STUDY DESIGN AND LANDSCAPE SELECTION

The ‘experimental’ units were 4 · 4 km ‘landscape’ squares that were large enough to contain populations of the target species. Data for the dependent variable (presence ⁄ absence of the two target species) and the independent variables were gathered at the landscape scale. Landscape units were selected strategically to represent a gradient in: (i) forest cover from <5% to approximately 80%; (ii) the level of subdivision and aggregation of forest patches; and (iii) the amount of hedgerows in the landscape and level of

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 153–162

Rodent distribution in fragmented landscapes 155 connectedness of forest patches in the landscape. For each of the following amounts of forest cover in the landscape: <5%; 5–10%; 10–15%; 15–20%; 20–40%; 40–80%, we chose pairs of landscape squares with contrasting conﬁguration and contrasting level of connectedness (Table S1, Supporting Information). Landscapes with extremely low levels of connectedness (no hedgerows) were selected only for low levels of forest cover (<15%) as we were unable to ﬁnd them for higher levels. Logistical constraints reduced the total sample to 41 landscapes. Explanatory variables (Table 1; Table S2, Supporting Information) were calculated with Arcview 3Æ3 using Corine Land Cover (with a resolution of 0Æ1 ha) and digitized aerial photographs as main layers.

FIELD SURVEYS

The presence of hazel dormouse in the woodland patches was assessed using 745 wooden nest-boxes spaced at least 70 m apart and inspected at regular intervals four times in spring–summer and three times the following autumn–winter. The survey was carried out during spring 2006 to spring 2007 (10 landscapes) and spring 2008 to spring 2009 (31 landscapes). The distribution of red squirrels was assessed using plastic hairtubes with a strip of adhesive material on the upper side. A mixture of chocolate, maize and sunﬂower seeds was used as bait; hazelnuts were glued to the inner part of the tube to prevent mice or birds from immediately consuming the bait. Tubes were spaced at least 70 m apart and were inspected every 10 days, for one month (3 inspections). A total of 591 hair-tubes were left in the ﬁeld for one month giving 18,321 tube-days of activation. The survey was carried out during spring-summer (March–July) 2007 (10 landscapes) and spring-summer 2008 (31 landscapes). We had to exclude four landscapes from the analysis because hair tubes were damaged.

SITE SELECTION AND SAMPLING DESIGN

We followed MacKenzie & Royle (2005) and the sampling protocol developed by Mortelliti & Boitani (2008) to determine the number of nest-boxes and hair-tubes per patch required to ensure detection with a high degree of certainty. First, we estimated detection probability and presence probability as a function of patch size (Mortelliti & Boitani 2008) by ﬁtting occupancy models on capture history data gathered in the ﬁrst 10 landscapes sampled (sampled in 2007). Secondly, we determined total survey eﬀort (total number of nest-boxes ⁄ hairtubes required) for each patch in the study area with a ﬁxed number of three visits per patch (MacKenzie et al. 2006) and a desired standard error of 0Æ2 in the estimate of presence probability (obtained as a compromise of available nest-boxes ⁄ tubes and logistical constraints). The number of nest-boxes and tubes varied with the size of the patch (approximately 1 per hectare) with a maximum of 15 in patches over 15 hectares. The number of patches sampled in each 4 · 4 km landscape increased with the number of patches present, ranging from one – in landscapes with one single patch (e.g. non fragmented landscapes) – to a maximum of six. In total we sampled 110 patches (median number of sampled patches per landscape = 2Æ5). We always sampled the two largest patches where a higher probability of presence was expected (Mortelliti, Santulli Sanzo & Boitani 2009); where applicable (e.g. if more than 2 patches were present in the landscape), the other sampled patches were selected in order to spread the sampling throughout the square. In summary, our design involved sampling several sites per patch (up to a maximum of 15 nest-boxes ⁄ tubes per patch) and several patches per landscape (up to a maximum of 6 patches), and always included the one or two largest patches in the landscape. Preliminary analysis of data from the ﬁrst 10 landscapes revealed that sampling the two largest patches was suﬃcient to reduce the risk of false absences in the data at the landscape level (Mortelliti, unpublished data).

Table 1. Results of the Principal Component Analysis (Varimax rotation) of the landscape characteristics measured in 41 landscapes in central Italy. All variables were Log10 transformed prior to analyses; values in bold indicate variables more strongly correlated with the principal components

PCA on extent and conﬁguration of habitat Variance explained Forest Cover (ha) Number of patches Mean patch size Mean of the Proximity index (1000 m threshold) Sum of the Proximity Indices (1000 m threshold) Mean EdgeDistance (1000 m threshold) Total number of hedgerows Total length of hedgerows (km) Mean number of hedgerows ⁄ patch PCA on land use variables Variance explained Arable land (ha) Olive groves Small scale cultivations Urban settlements Vineyards Orchards Longitude (X-UTM33) Latitude (Y-UTM 33)

Principal component HA

Principal component C

Principal component HF

50% 0Æ862 )0Æ017 0Æ686 0Æ91 0Æ907 )0Æ191 0Æ392 0Æ182 0Æ147

23% 0Æ278 )0Æ040 0Æ241 0Æ103 0Æ277 0Æ070 0Æ859 0Æ918 0Æ843

14% )0Æ044 0Æ939 )0Æ574 )0Æ253 0Æ008 0Æ877 0Æ091 0Æ139 )0Æ306

Luse 1

Luse 2

41% )0Æ530 0Æ572 0Æ476 0Æ758 0Æ002 0Æ213 0Æ841 )0Æ832

16% )0Æ066 0Æ295 0Æ043 )0Æ083 0Æ857 0Æ854 0Æ138 )0Æ415

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 153–162

156 A. Mortelliti et al. STATISTICAL ANALYSES

We carried out two series of principal components analysis (PCA): the ﬁrst with variables related to the extent and conﬁguration of forest habitat and the length of hedgerows in the landscape, and the second with land-use variables related to the matrix. As expected, the ﬁrst PCA reﬂected the study design. The ﬁrst three components explain, altogether, 87% of the variance in the dataset. The ﬁrst component (HA in Table 1) was interpreted as a gradient reﬂecting primarily habitat amount since it is strongly correlated with forest cover (0Æ862), the two proximity indices (0Æ91 and 0Æ907) and mean patch size (0Æ686). The second component (C in Table 1) was interpreted as a gradient of structural connectivity (hedgerow network) since it is strongly correlated with total length (km) of hedgerows in the landscape (0Æ918), total number of hedgerows in the landscape (0Æ859) and mean number of hedgerows per patch (0Æ843). The third component (HF in Table 1) is interpreted as a gradient of subdivision and aggregation of forest habitat, since it is strongly correlated with the number of patches in the landscape (0Æ939), the mean edge-to-edge distance between patches in the landscape (0Æ877) and negatively correlated with mean patch size ()0Æ574). Factor score values for the three components are reported in Table S3 and Fig. S2 (Supporting Information). The PCA of land-use variables and geographic coordinates reﬂect the north–south and east–west gradients in land use patterns of the Lazio and southern Tuscany regions of Italy. The ﬁrst component (Luse 1 in Table 1) was interpreted as a gradient reﬂecting the north– south and west–east variations in land use patterns: arable land is predominant along the coast (west) and in the north, whilst olive groves increase in the south and on the east towards the Apennine mountains. The second component depicts a gradient from areas with high abundance of vineyards and orchards to areas with low abundance of these land use types (Luse 2 in Table 1). Since component HA was correlated with the component Luse 2 (Pearson correlation coeﬃcient: z = )0Æ415, P < 0Æ05) and the component C was correlated with the component Luse 1 (Pearson correlation coeﬃcient z = 0Æ326, P < 0Æ05) these were not introduced simultaneously in models, with the exception of the global model (Burnham & Anderson 2002). Binary response data (presence ⁄ absence of the two target species in a landscape unit) was modelled as a function of explanatory variables using logistic regression models and the software spss (SPSS Inc., Chicago, IL, USA). Residuals were tested for spatial autocorrelation using the Moran I test.

MODEL SELECTION AND HYPOTHESIS TESTING

We followed an information-theoretic approach for model selection. Models were ﬁrst ranked according to AICc (second order Akaike Information Criteria) values (Burnham & Anderson 2002). We calculated Nagelgerke R2 as a goodness-of-ﬁt measure; departure from the logistic model was assessed through the Hosmer–Lemeshow test on the global model (Tabachnick & Fidell 2001). Relative variable importance was assessed by summing Akaike weights of the models including the target variable (Burnham & Anderson 2002). Within the information-theoretic approach, each model corresponds to a speciﬁc hypothesis (Burnham & Anderson 2002). Using the factor scores of principal components (HL, HF, C, Luse 1, Luse 2) we tested 13 hypotheses including multiple predictor variables (e.g. HA+C) and interaction eﬀects between the predictor variables (e.g. HL*C); a complete list of hypotheses tested is provided in Table 2.

Table 2. Logistic regression models (corresponding to each hypothesis tested) predicting the probability of occurrence of the hazel dormouse M. avellanarius and red squirrel S. vulgaris in fragmented landscapes. Models are ranked according to AICc. Goodness-of-ﬁt was assessed with Nagelkerke R2. HA = principal component interpreted as a gradient of habitat amount; HF = principal component interpreted as a gradient of habitat fragmentation per se; C = principal component interpreted as a gradient of structural connectivity (in the form of hedgerows); Luse 1 and Luse 2 = principal components interpreted as geographical gradients in land use patterns. See Methods for a description of the components Model form

DAICc

W

R2

Muscardinus avellanarius HA+C HA HA+HF+C HA+HF Global Luse 1 Luse 2 HAxHF HF C HF+C HFxC HAxC

0Æ00 0Æ97 2Æ07 2Æ38 9Æ92 12Æ10 17Æ75 18Æ13 20Æ23 20Æ39 21Æ11 21Æ73 21Æ75

0Æ44 0Æ27 0Æ16 0Æ13 0Æ00 0Æ00 0Æ00 0Æ00 0Æ00 0Æ00 0Æ00 0Æ00 0Æ00

0Æ634 0Æ568 0Æ642 0Æ587 0Æ762 0Æ300 0Æ134 0Æ122 0Æ054 0Æ048 0Æ101 0Æ002 0Æ002

Sciurus vulgaris HA HA+C HA+HF HA+HF+C Luse 2 Global HAxHF HFxC Luse 1 HAxC C HF HF+C

0Æ00 1Æ80 2Æ29 4Æ28 5Æ42 5Æ85 6Æ77 7Æ99 10Æ31 10Æ90 11Æ52 11Æ98 13Æ88

0Æ49 0Æ20 0Æ16 0Æ06 0Æ03 0Æ03 0Æ02 0Æ01 0Æ00 0Æ00 0Æ00 0Æ00 0Æ00

0Æ377 0Æ392 0Æ379 0Æ394 0Æ222 0Æ695 0Æ180 0Æ140 0Æ061 0Æ040 0Æ017 0Æ001 0Æ018

Results The hazel dormouse was found in 29 landscapes (N = 41) whilst the red squirrel was found in 14 landscapes (N = 37). No signiﬁcant spatial autocorrelation was found for the residuals of the global models for both species (Moran I test on residuals: P > 0Æ05, 999 permutations, lag distance 2Æ5–75 km). The Hosmer–Lemeshow test on the global model for both species showed that data did not depart from a logistic regression model (hazel dormouse: v2 = 6Æ287, d.f. = 8, P > 0Æ05; red squirrel: v2 = 6Æ68, d.f. = 7; P > 0Æ05). The top ranked model for the hazel dormouse found that probability of presence was related to habitat amount (principal component HL: b = 2Æ735; SE = 0Æ894; constant b = 1Æ769; SE = 0Æ638) and structural connectivity (principal component C: b = 0Æ892; SE = 0Æ548; Table 2). The probability of hazel dormouse presence increased with the amount of forest cover and increasing structural connectivity (hedgerows) in the landscape (Fig. 1). The second model expresses presence

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 153–162

Rodent distribution in fragmented landscapes 157

Fig. 1. Probability of presence of the hazel dormouse expressed as a function of structural connectivity (principal component C) controlling for habitat amount (principal component HA). Each line represents a ﬁxed value of the HA component. An increase in structural connectivity can increase the probability of presence in landscapes with moderate levels of habitat, but for low levels of habitat amount even high values of structural connectivity will have little eﬀect.

probability as a function of habitat amount (principal component HL: b = 2Æ443; SE = 0Æ819; constant: b = 1Æ531; SE = 0Æ563). The other candidate models (DAICc > 2; Table 2) received less support in the model selection procedure (Burnham & Anderson 2002). The two variables with highest relative importance were HA (w = 1) followed by C (w = 0Æ6); HF had considerably less importance (w = 0Æ29; Fig. S3). Model averaged parameter estimates are shown in Table 3. Modelling probability of presence of hazel dormouse as a function of connectivity, while controlling for habitat amount, predicted that increasing connectivity would increase presence probability in landscapes with moderate amount of habitat (Fig. 1). For low levels of habitat amount (values of HA roughly corresponding to 5–10% of forest cover in the landscape), however, even extremely high values of structural connectivity were predicted to have little eﬀect on probability of presence (Fig. 1). The top ranked regression model for the red squirrel found that probability of presence was related to habitat amount (principal component HL: b = 1Æ568; SE = 0Æ573; constant:

b = )0Æ624; SE = 0Æ416; Table 2). Probability of red squirrel presence increased with the amount of forest cover in the landscape (Fig. 2). The second ranked model expresses presence probability as a function of habitat amount (principal component HL: b = 1Æ571; SE = 0Æ581; constant: b = )0Æ582; SE = 0Æ417; Table 2) and structural connectivity (principal component C: b = 0Æ36; SE = 0Æ491). The variable with the highest relative importance is HA (w = 0Æ93) followed by HF and C with less importance (C: w = 0Æ28; HF: w = 0Æ24; Fig. S3). Modelling red squirrel probability of presence expressed as a function of the principal component HA (ﬁrst ranked model; Fig. 2) predicted that relatively high values of presence probability (e.g. presence probability >0Æ8) would be reached only for landscapes with relative high amounts of forest habitat (values of HA roughly corresponding to more than 25% forest cover). Modelling red squirrel probability of presence expressed as a function of connectivity and habitat fragmentation per se, while controlling for habitat amount, predicted that increasing connectivity or increasing the subdivision of habitat would increase

Table 3. Logistic regression model averaged parameter estimates and standard errors. See Methods for a description of the components b constant Hazel dormouse Red squirrel

1Æ653 )0Æ621

SE

b HA

SE

bC

SE

b HF

SE

0Æ608 0Æ424

2Æ585 1Æ508

0Æ861 0Æ557

0Æ521 0Æ137

0Æ325 0Æ159

0Æ115 0Æ026

0Æ148 0Æ109

b Luse 1

)0Æ043

SE

0Æ020

b Luse 2

)0Æ083

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 153–162

SE

0Æ045

158 A. Mortelliti et al.

Fig. 2. Results of the ﬁrst ranked model for the red squirrel: probability of presence is expressed as a function of the principal component HA interpreted as a gradient in habitat cover: increased forest habitat in the landscape increases the probability of occurrence of the red squirrel.

presence probability in landscapes with a relatively high amount of habitat but not where the amount of habitat is relatively low (Fig S4ab Supporting Information). However, the impact of increasing fragmentation per se is particularly small and leads to a small variation in probability of presence; the impact of increasing structural connectivity is slightly stronger but still weaker than that observed for the hazel dormouse.

Discussion Our results show that habitat amount, rather than habitat fragmentation per se, is a main driver of distribution patterns for both the hazel dormouse and the red squirrel. As predicted, structural connectivity (hedgerow networks) played an important role in determining the distribution of the hazel dormouse, but less so for the red squirrel, where it was no more important than habitat fragmentation per se. The logistic model for the hazel dormouse predicted that increasing connectivity would increase presence probability only for moderate levels of residual forest cover. For lower levels of forest cover, even extensive hedgerow networks will not increase presence probability to any great extent.

RELIABILITY OF THE SURVEY METHODS

Nest-boxes and hair-tubes are often used to survey dormice and squirrels (Bright, Morris & Mitchell-Jones 2006; Mortelliti

et al. 2010a). The issue of false absences is particularly important when studying elusive animals such as arboreal rodents, since false absences are likely to bias model parameter estimates (Gu & Swihart 2004). Our sampling design was structured to reduce the chances of including false absences in the data and has been tested elsewhere (Mortelliti & Boitani 2008; Mortelliti et al. 2010a). In the case of the red squirrel we have shown previously that one month of sampling is suﬃcient to be 95% sure of detecting the species, if present (Mortelliti & Boitani 2008). We are therefore conﬁdent that the likelihood of having included false absences in our data is low.

EFFECTIVE SEPARATION OF THE DIFFERENT LANDSCAPE SCALE PROCESSES

Optimal separation of the role of the diﬀerent processes included in the term ‘habitat fragmentation’ can be carried out only with true landscape scale experiments, which would be impractical for forest mammals [but see Lindenmayer (2009)]. Carefully designed observational studies are, therefore, the best available option (Mortelliti et al. 2010b) but adequate separation of these processes requires two steps. First, the study design must include diﬀerent amounts of habitat and contrasting levels of habitat fragmentation per se and structural connectivity. The second step (PCA) allows reduction of collinearity (Tabachnick & Fidell 2001). We opted for the

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 153–162

Rodent distribution in fragmented landscapes 159 PCA approach (Fahrig 2003) rather than regression of residuals (Villard, Trzcinski & Merriam 1999) since the latter approach has been criticized because it does not allow separation of the joint variance of the two correlated variables (Koper, Schmiegelow & Merrill 2007).

HAZEL DORMOUSE IN FRAGMENTED LANDSCAPES

The hazel dormouse is an arboreal species that avoids moving across open ground (Bright 1998). Habitat loss and fragmentation per se are thought to be important factors in their decline (Bright, Morris & Mitchell-Jones 2006). Although more studies are needed to test the applicability of our ﬁndings to other areas, our results have two important implications for the conservation and management of this species: 1. The amount of habitat in the landscape is a crucial factor aﬀecting its distribution: therefore preservation of existing habitat and habitat restoration are a conservation priority. The second most important factor is the structural connectivity represented by the network of hedges. Although restoring habitat is more expensive than implementing a networks of hedgerows, with low amounts of residual forest (e.g. less than 5–10% of forest cover in the landscape), restoration of habitat is the most eﬀective strategy. However, increasing the hedgerows network is an eﬀective measure for landscapes with moderate amounts of habitat. 2. The subdivision of habitat has a minor role in determining distribution patterns. We recorded the presence of the species even in highly subdivided habitats with highly scattered small patches, as long as these were connected by hedgerows. It should be emphasised that hedgerows also provide habitat for the hazel dormouse (Wolton 2009), therefore any increase in the length of hedgerow will not only increase connectivity in the landscape but will also increase the available habitat for this species.

RED SQUIRREL IN FRAGMENTED LANDSCAPES

There have been a number of studies of red squirrels in fragmented landscapes (Wauters, Casale & Dhondt 1994; Rodriguez & Andren 1999; Verbeylen et al. 2009; Wauters et al. 2010) but this study extends previous ﬁndings by showing the landscape scale response of the species. We have shown that habitat amount is an important driver of distribution patterns, whilst habitat subdivision and structural connectivity are less important (Fig. S3, Supporting Information). Importantly, the weight of these variables comes from the models with HA as additional covariate (HA+C and HA+HF; Table 2, Fig. S3). These results are consistent with our prediction that the red squirrel more readily crosses open ﬁelds (Mortelliti, Santulli Sanzo & Boitani 2009). The most important objective for the conservation of the red squirrel is the preservation and restoration of existing forest habitat. Increasing structural connectivity by increasing the hedgerow network will be beneﬁcial but increasing the amount of habitat will be more eﬀective. Con-

servation management should be prioritized accordingly. Note that the grey squirrel does not inhabit our study area and its presence would have a signiﬁcant eﬀect. Clearly, dormice and squirrels diﬀer in body size (adult hazel dormouse weight range 15–30 g; adult red squirrel weight range 250–350 g) and movement capabilities (squirrels have larger home-range and are more mobile). These diﬀerences explain their respective requirements at the landscape scale (Figs 1 and 2).

IMPLICATIONS FOR THE MANAGEMENT OF LANDSCAPES

We examined the response of arboreal mammals to landscape elements. We add to existing knowledge by showing that arboreal mammals show diﬀerential sensitivity to landscape processes. In particular the subdivision of habitat plays a minor role. Eﬀective conservation strategies must acknowledge the independent contribution of structural connectivity. A species may persist in a subdivided landscape if these are connected, for example by hedgerows. However, while increasing the amount of hedgerows may be eﬀective in connecting moderate amounts of habitat in the landscape, it may prove ineﬀective if the total amount of habitat is too low (e.g. less than 5–10% of residual forest cover). The importance of habitat amount in determining species’ distribution is consistent with observations of birds where it is known to be an important driver of the occurrence of species in agricultural landscapes (Fahrig 2003; Bennett, Radford & Haslem 2006). Trzcinski, Fahrig & Merriam (1999) concluded that the role of habitat fragmentation per se was marginal and did not increase with decreasing forest cover as suggested by Andren (1994) and Fahrig (1998). Subsequently Fahrig (2003) concluded that fragmentation per se plays a minor role that may be either positive or negative. On the other hand Radford & Bennett (2007) concluded that, when inﬂuential, the role is essentially negative. In general, there is agreement that the role of habitat conﬁguration is secondary to the predominant role of habitat loss. In our study of the red squirrel, the eﬀect of fragmentation per se was weakly positive (as found for some species by Villard, Trzcinski & Merriam 1999). This is unsurprising because the red squirrel will cross open ﬁelds and can incorporate several patches in its home-range (Wauters, Casale & Dhondt 1994). Our values of 5–10% forest cover are clearly arbitrary and our results are context-speciﬁc, depending on the biology of the species in the area. However, the key-message is that an increase in the length of hedgerows may be eﬀective only within certain amounts of residual habitat. We suggest that planting new hedgerows should not be considered a panacea for areas where the amount of residual habitat is particularly low. In some circumstances the restoration of habitat may be more cost-eﬀective than planting a network of hedgerows, especially given that the increase in presence probability following an increase in the amount of habitat is higher than that from increasing connectivity (Table 3: a higher value of b corresponds to a steeper slope).

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 153–162

160 A. Mortelliti et al. We emphasize four key-issues relevant for future studies: 1. We did not consider the quality or vegetation structure of forest patches, which is an important predictor in fragmented landscapes for many species including the hazel dormouse and the red squirrel (Van Apeldoorn, Celada & Nieuwenhuizen 1994; Bright & Morris 1996; Rima et al. 2010). Part of the unexplained variance found in the ﬁrst ranked model of both species could be explained by the quality of patches in the landscapes. Future studies should gather data on the quality of patches. Moreover spatial correlation in patch quality should be evaluated since, although the quality of a patch is a patch scale quantity, it may have landscape scale eﬀects in the presence of spatial autocorrelation (Schooley & Branch 2007). 2. Future studies should include quantitative information on the structure and the internal characteristics of hedgerows. We have focused on the spatial properties measurable from digitized aerial photos; however, the internal features of hedgerows are likely to inﬂuence their utilization by species (Wolton 2009). 3. Landscape scale studies based on snapshots of species’ distribution tell us nothing about dynamic patterns of colonization, extinction and the general trend of populations. As far as we know, for instance, landscapes with low habitat amount could be large scale sinks, therefore increasing the hedgerow network might facilitate individuals ﬂowing into this ‘landscape trap’. Regardless of the kind of spatially structured population that both species may present in certain fragmented landscapes, such as patchy populations, isolated populations or true metapopulations (Driscoll 2007), future studies should follow a metapopulation approach (sensu Hanski & Gaggiotti 2004) to determine the actual dynamics behind the observed eﬀect of a presence ⁄ absence at the landscape scale (see Verbeylen, De Bruyn & Matthysen 2003 for red squirrels). 4. We stress that the approach followed in Fig. 1 and Fig. S4ab is inferential. Therefore, it involves predictions for combinations of variables that may have not been sampled; however, this is expected with any kind of regression analysis. Nevertheless, we stress that our conclusions depend heavily on the stability and robustness of the models: our models were relatively stable as can be seen from standard deviation of parameter estimates, therefore we believe they allowed us relatively unbiased inference.

Conclusions Our results have broader implications than the conservation of the two target species examined. Ongoing analyses on the distribution of forest dependent birds in the same study area, for instance, show similar patterns to those reported here (Mortelliti et al. 2010c). Our results provide strong empirical evidence that halting habitat loss and carrying out habitat restoration should be conservation priorities. However, as shown in the case of the dormouse, the situation may be more complex: restoration of structural connectivity should not be carried out regardless of the amount of habitat in the landscapes. As a rule

of thumb our results suggest that with less than 5–10% of forest cover in the landscape, establishment of a hedgerow network, even if extensive (e.g. 30 km of hedgerows in the landscape), may prove ineﬀective for species such as the hazel dormouse. Landscape scale studies should be conducted to assess the amount of remnant habitat and the dependence of vulnerable species upon it, before investing resources in the establishment of landscape linkages.

Acknowledgements Thanks to the park wardens and naturalists of the Regione Lazio for help in the ﬁeld surveys; special thanks to Sergio Muratore, Pietro Politi and Fabiola Iannarilli. Thanks to Joyce Keep for language revision and to Andrew Bennett, the editors and ﬁve anonymous reviewers for extremely useful comments on an earlier version of the manuscript.

References Andren, H. (1994) Eﬀects of habitat fragmentation on birds and mammals in landscapes with diﬀerent proportions of suitable habitat – a review. Oikos, 71, 355–366. Bennett, A.F., Henein, K. & Merriam, G. (1994) Corridor use and the elements of corridor quality: Chipmunks and fencerows in a farmland mosaic. Biological Conservation, 68, 155–165. Bennett, A.F., Radford, J.Q. & Haslem, A. (2006) Properties of land mosaics: Implications for nature conservation in agricultural environments. Biological Conservation, 133, 250–264. Bright, P.W. (1998) Behaviour of specialist species in habitat corridors: arboreal dormice avoid corridor gaps. Animal Behaviour, 56, 1485–1490. Bright, P.W. & Morris, P.A. (1996) Why are dormice rare? A case study in conservation biology. Mammal Review, 26, 157–187. Bright, P.W., Morris, P. A. & Mitchell-Jones, A. J. (2006) The Dormouse Conservation Handbook. English Nature, Peterborough. Burnham, K.P. & Anderson, D. R. (2002) Model Selection and Multimodel Inference: A Practical Information-theoretic Approach, 2nd edn. Springer, New York. Davies, Z.G. & Pullin, A.S. (2007) Are hedgerows eﬀective corridors between fragments of woodland habitat? An evidence-based approach. Landscape Ecology, 22, 333–351. Driscoll, D.A. (2007) How to ﬁnd a metapopulation. Canadian Journal of Zoology-Revue Canadienne De Zoologie, 85, 1031–1048. Fahrig, L. (1998) When does fragmentation of breeding habitat aﬀect population survival? Ecological Modelling, 105, 273–292. Fahrig, L. (2003) Eﬀects of habitat fragmentation on biodiversity. Annual Review of Ecology, Evolution, and Systematics, 34, 487–515. Goheen, J.R., Swihart, R.K., Gehring, T.M. & Miller, M.S. (2003) Forces structuring tree squirrel communities in landscapes fragmented by agriculture: species diﬀerences in perceptions of forest connectivity and carrying capacity. Oikos, 102, 95–103. Gu, W. & Swihart, R.K. (2004) Absent or undetected? Eﬀects of non-detection of species occurrence on wildlife-habitat models. Biological Conservation, 116, 195–203. Hanski, I. & Gaggiotti, O. (2004) Ecology, Genetics, and Evolution of Metapopulations. Elsevier, Burlington. Haslem, A. & Bennett, A.F. (2008) Birds in agricultural mosaics: the inﬂuence of landscape patterns and countryside heterogeneity. Ecological Applications, 18, 185–196. Koper, N., Schmiegelow, F.K.A. & Merrill, E.H. (2007) Residuals cannot distinguish between ecological eﬀects of habitat amount and fragmentation: implications for the debate. Landscape Ecology, 22, 811–820. Koprowski, J.L. (2005) The response of tree squirrels to fragmentation: a review and synthesis. Animal Conservation, 8, 369–376. Lindenmayer, D.B. (2009) Large-scale Landscape Experiments: Lessons from Tumut. Cambridge University Press, Cambridge. Lindenmayer, D.B. & Fischer, J. (2007) Tackling the habitat fragmentation panchreston. Trends in Ecology and Evolution, 22, 127–132. MacKenzie, D.I. & Royle, J.A. (2005) Designing occupancy studies: General advice and allocating survey eﬀort. Journal of Applied Ecology, 42, 1105– 1114.

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 153–162

Rodent distribution in fragmented landscapes 161 MacKenzie, D.I., Nichols, J.D., Royle, J.A., Pollock, K.H., Bailey, L.L. & Hines, J.E. (2006) Occupancy Estimation and Modeling – Inferring Patterns and Dynamics of Species Occurrence. Elsevier, San Diego, USA. McGarigal, K. & Cushman, S.A. (2002) Comparative evaluation of experimental approaches to the study of habitat fragmentation eﬀects. Ecological Applications, 12, 335–345. McGarigal, K. & McComb, W.C. (1995) Relationships between landscape structure and breeding birds in the Oregon coast range. Ecological Monographs, 65, 235–260. Moore, J.E. & Swihart, R.K. (2005) Modeling patch occupancy by forest rodents: Incorporating detectability and spatial autocorrelation with hierarchically structured data. Journal of Wildlife Management, 69, 933– 949. Mortelliti, A. & Boitani, L. (2008) Inferring red squirrel (Sciurus vulgaris) absence with hair tubes surveys: a sampling protocol. European Journal of Wildlife Research, 54, 353–356. Mortelliti, A., Santulli Sanzo, G. & Boitani, L. (2009) Species’ surrogacy for conservation planning: caveats from comparing the response of three arboreal rodents to habitat loss and fragmentation. Biodiversity and Conservation, 18, 1131–1145. Mortelliti, A., Cervone, C., Amori, G. & Boitani, L. (2010a) The eﬀect of nontarget species in presence-absence distribution surveys: a case study with hair-tubes. Italian Journal of Zoology, 77, 211–215. Mortelliti, A., Amori, G., Capizzi, D., Rondinini, C. & Boitani, L. (2010b) Experimental design and taxonomic scope of fragmentation studies of European mammals: current status and future priorities. Mammal Review, 40, 125–154. Mortelliti, A., Fagiani, S., Battisti, C., Capizzi, D. & Boitani, L. (2010c) Independent eﬀects of habitat loss and structural connectivity on landscape occupancy of forest dependent birds. Diversity and Distributions, 16, 941–951. Pereira, M. & Rodrı` guez, A. (2010) Conservation value of linear woody remnants for two forest carnivores in a Mediterranean agricultural landscape. Journal of Applied Ecology, 47, 611–620. Prugh, L.R., Hodges, K.E., Sinclair, A.R.E. & Brashares, J.S. (2008) Eﬀect of habitat area and isolation on fragmented animal populations. Proceedings of the National Academy of Sciences of the United States of America, 105, 20770–20775. Radford, J.Q. & Bennett, A.F. (2007) The relative importance of landscape properties for woodland birds in agricultural environments. Journal of Applied Ecology, 44, 737–747. Radford, J.Q., Bennett, A.F. & Cheers, G.J. (2005) Landscape-level thresholds of habitat cover for woodland-dependent birds. Biological Conservation, 124, 317–337. Rima, P.C., Cagnin, M., Aloise, G., Preatoni, D. & Wauters, L.A. (2010) Scaledependent environmental variables aﬀecting red squirrel (Sciurus vulgaris meridionalis) distribution. Italian Journal of Zoology, 77, 92–101. Rodriguez, A. & Andren, H. (1999) A comparison of Eurasian red squirrel distribution in diﬀerent fragmented landscapes. Journal of Applied Ecology, 36, 649–662. Schooley, R.L. & Branch, L.C. (2007) Spatial heterogeneity in habitat quality and cross-scale interactions in metapopulations. Ecosystems, 10, 846–853. Tabachnick, B.G. & Fidell, L. S. (2001) Using Multivariate Statistics. Allyn and Bacon, Boston. Tewksbury, J.J., Levey, D.J., Haddad, N.M., Sargent, S., Orrock, J.L., Weldon, A., Danielson, B.J., Brinkerhoﬀ, J., Damschen, E.I. & Townsend, P. (2002) Corridors aﬀect plants, animals, and their interactions in fragmented landscapes. Proceedings of the National Academy of Sciences of the United States of America, 99, 12923–12926. Trzcinski, M.K., Fahrig, L. & Merriam, G. (1999) Independent eﬀects of forest cover and fragmentation on the distribution of forest breeding birds. Ecological Applications, 9, 586–593. Van Apeldoorn, R.C., Celada, C. & Nieuwenhuizen, W. (1994) Distribution and dynamics of the red squirrel (Sciurus vulgaris L.) in a landscape with fragmented habitat. Landscape Ecology, 9, 227–235. Verbeylen, G., De Bruyn, L. & Matthysen, E. (2003) Patch occupancy, population density and dynamics in a fragmented red squirrel Sciurus vulgaris population. Ecography, 26, 118–128. Verbeylen, G., Wauters, L.A., De Bruyn, L. & Matthysen, E. (2009) Woodland fragmentation aﬀects space use of Eurasian red squirrels. Acta OecologicaInternational Journal of Ecology, 35, 94–103. Villard, M.A., Trzcinski, M.K. & Merriam, G. (1999) Fragmentation eﬀects on forest birds: Relative inﬂuence of woodland cover and conﬁguration on landscape occupancy. Conservation Biology, 13, 774–783.

Wauters, L., Casale, P. & Dhondt, A.A. (1994) Space use and dispersal of red squirrels in fragmented habitats. Oikos, 69, 140–146. Wauters, L.A., Hutchinson, Y., Parkin, D.T. & Dhondt, A.A. (1994) The eﬀects of habitat fragmentation on demography and on the loss of genetic variation in the red squirrel. Proceedings of the Royal Society B: Biological Sciences, 255, 107–111. Wauters, L.A., Verbeylen, G., Preatoni, D., Martinoli, A. & Matthysen, E. (2010). Dispersal and habitat cuing of Eurasian red squirrels in fragmented habitats. Population Ecology, 52, 527–536. Wolton, R. (2009) Hazel dormouse Muscardinus avellanarius (L.) nest site selection in hedgerows. Mammalia, 73, 7–12. Received 26 April 2010; accepted 3 November 2010 Handling Editor: Chris Dickman

Supporting Information Additional Supporting Information may be found in the online version of this article. Fig. S1. Map of the study area; studied landscapes (4x4 Km squares) were distributed throughout the Lazio region and the Province of Siena. Fig. S2. Values of percent forest cover (within the 4x4km squares) and PCA components interpreted as gradients in habitat amount (HA), habitat fragmentation per se (HF), and structural connectivity (C) for each of the 41 studied landscape in central Italy. Fig. S3. Relative importance of variables in models for the dormouse (stippled) and red squirrel (clear), obtained by summing the Akaike weights for all models in the set including the target variable. HA = principal component interpreted as a gradient of habitat loss; HF = principal component interpreted as a gradient of habitat fragmentation per se; C = principal component interpreted as a gradient of structural connectivity (in the form of hedgerows); Luse 1 and Luse 2 = principal components interpreted as north-south-east-west gradients in land use patterns. See materials and methods for a detailed description of the components. Fig. S4. a) Results of the second ranked model for the red squirrel (Sciurus vulgaris): red squirrel probability of presence is expressed as a function of habitat fragmentation per se (principal component HF) controlling for habitat amount (principal component HA). Each line represents a ﬁxed value of the HA component: e.g. a value of HA= )2.8 roughly corresponds to landscapes with less than 5% residual forest cover. An increase in habitat subdivision can increase the probability of presence only in landscape with relatively high amounts. b) Results of the third ranked model: red squirrel probability of presence is expressed as a function of structural connectivity (principal component C) controlling for habitat amount (principal component HA). Each line represents a ﬁxed value of the HA component. An increase in structural connectivity can increase probability of presence in landscapes with moderate levels of habitat, but below a certain threshold of habitat even high values of structural connectivity will not increase probability of presence to relatively high values. Table S1. Study design showing the combination of landscapes variables for each category; numbers correspond to landscapes in Fig. S1. See Fig. S2 for a graphical representation of the PCA axis values.

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 153–162

162 A. Mortelliti et al. Table S2. Descriptive statistics of the landscape characteristics measured in 41 4x4Km landscapes in central Italy. Table S3. Values of the principal components and presence (=1) ⁄ absence (=0) of the target species (Sciurus vulgaris and Muscardinus avellanarius) in each of the 41 sampled landscapes in central Italy.

As a service to our authors and readers, this journal provides supporting information supplied by the authors. Such materials may be re-organized for online delivery, but are not copy-edited or typeset. Technical support issues arising from supporting information (other than missing ﬁles) should be addressed to the authors.

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 153–162

Journal of Applied Ecology 2011, 48, 177–180

doi: 10.1111/j.1365-2664.2010.01920.x

FORUM

Issues with modelling the current and future distribution of invasive pathogens Kris A. Murray1*, Richard W. R. Retallick2, Robert Puschendorf3, Lee F. Skerratt4, Dan Rosauer5,6, Hamish I. McCallum7, Lee Berger4, Rick Speare4 and Jeremy VanDerWal3 1

The Ecology Centre, School of Biological Sciences, University of Queensland, Brisbane, Queensland 4072, Australia; GHD Pty Ltd, 8 ⁄ 180 Lonsdale Street, Melbourne, Victoria 3000, Australia; 3School of Marine and Tropical Biology, Centre for Tropical Biodiversity and Climate Change Research, James Cook University, Townsville, Queensland 4811, Australia; 4School of Public Health, Tropical Medicine and Rehabilitation Sciences and the Amphibian Disease Ecology Group, James Cook University, Townsville, Queensland 4811, Australia; 5School of Biological, Earth and Environmental Sciences, University of New South Wales, Sydney, New South Wales 2052, Australia; 6Centre for Plant Biodiversity Research, GPO Box 1600, Canberra, Australian Capital Territory 2601, Australia; and 7School of Environment, Griﬃth University, Nathan Campus, Queensland 4111, Australia 2

Summary 1. Correlative species distribution models can be used to produce spatially explicit estimates of environmental suitability for organisms. This process can provide meaningful information for a range of purposes (e.g. estimating a species’ current or future distribution, estimating dispersal limits, predicting occupancy for conservation planning) but, like all statistical exercises, is subject to numerous assumptions and can be inﬂuenced by several sources of potential bias. 2. In this issue of Journal of Applied Ecology, we (Murray et al. 2011) employ a correlative species distribution model for infection with the pathogen Batrachochytrium dendrobatidis (Bd), cause of amphibian chytridiomycosis, to derive useful information for the immediate management and research of this pathogen in Australia. Also in this issue, Rohr, Halstead & Raﬀel (2011) comment on some of the potential limitations of our approach and the value of our results in practice. 3. Synthesis and applications. Here we show that while a focus on mechanisms of dispersal and transmission among hosts, as advocated in both studies, is an important objective for modelling Bd distribution under climate change or at invasion fronts, correlative models can be of immediate value for their ability to generate a baseline hypothesis about the current potential distribution of this lethal pathogen and for eﬃciently identifying gaps in current knowledge. As demonstrated in our paper, this should help improve the immediate allocation of limited research and management resources for future surveillance eﬀorts and proactive species conservation. Key-words: amphibian declines, bioclimatic modelling, chytrid fungus, chytridiomycosis, infectious disease, Maxent, species distribution model

Introduction In this issue of Journal of Applied Ecology (Murray et al. 2011), we employ a species distribution model (SDM) to spatially estimate environmental suitability for infection with Batrachochytrium dendrobatidis (Bd), cause of the pandemic amphibian disease chytridiomycosis. This new source of information proved to be a signiﬁcant predictor of amphibian declines in Australia. The implication for researchers and

*Correspondence author. E-mail: [email protected]

managers is that actions, geographic regions, species and even populations can potentially be targeted for further attention as the most at risk regions and species can be rapidly evaluated. Rohr, Halstead & Raﬀel’s (2011) commentary is generally encouraging and raises some interesting questions relevant to our study that both scrutinize the details of our models and echo our recommendations for further work. Rohr, Halstead & Raﬀel’s (2011) key contribution is to review some of the questions regarding the validity of SDMs for the prediction of chytridiomycosis distribution and impacts in the future; for example, under climate change or at invasion fronts. Rohr, Halstead & Raﬀel (2011) conclude, as we

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society

178 K. A. Murray et al. do, that a focus on mechanisms, such as dispersal of the pathogen over long distances and transmission among hosts, will be important. However, since these mechanisms are still poorly understood, in many cases correlative SDMs will be of immediate value for their ability to generate a baseline hypothesis about the current potential distribution of this lethal pathogen and for eﬃciently identifying gaps in current knowledge. As demonstrated in our paper, this should help improve the allocation of limited research and management resources for future surveillance eﬀorts and proactive species conservation.

Key issues for modelling species’ distributions The issues raised by Rohr, Halstead & Raﬀel (2011) include some of the general principles of correlation and assumptions for SDMs. Some of the issues, such as dealing with sampling bias and spatial autocorrelation, remain highly active areas of research (e.g. Phillips et al. 2009; Veloz 2009) and under some circumstances may bias the results of correlative SDMs. Over-ﬁtting, model parameterization and model selection are also mentioned by Rohr, Halstead & Raﬀel (2011) as potential limitations of our models. These are important issues; however, as detailed in our paper, we took great care in building our models to maximize predictive performance and minimize the eﬀects of bias due to these limitations as far as was practically possible. Our hypotheses are strongly underpinned by biological reasoning and our choice of modelling tools and our methods together reﬂect signiﬁcant recent advances in correlative species distribution modelling from incomplete information (Elith et al. 2006; VanDerWal et al. 2009). The key strengths and limitations of the Maxent approach have been described extensively elsewhere (Phillips, Anderson & Schapire 2006; Phillips & Dudik 2008; Phillips et al. 2009). Brieﬂy, the Maxent approach avoids or minimizes many of the limitations of earlier bioclimatic ‘envelope’ model methods, such as BIOCLIM and DOMAIN, that are referred to by Rohr, Halstead & Raﬀel (2011). Our study demonstrates how presence-only datasets of poorly studied wildlife pathogens may be eﬀectively employed in this relatively new framework to identify new avenues for research and management. With greater traction are Rohr, Halstead & Raﬀel’s (2011) comments relating to the validity and underlying assumptions of using correlative SDMs for modelling the future distribution of chytridiomycosis; for example, under climate change or at invasion fronts. In our study, we did not attempt to incorporate climate change predictions. We did, however, make projections geographically across Australia from our well sampled ‘training’ region (ﬁg. S2; Murray et al. 2011). Two potential issues arise in this case that could lead to spurious projections: (1) invalid statistical extrapolation may occur and (2) models may be biased if occurrence records are not suﬃciently representative of the environmental space inhabitable by the organism. Extrapolation (point 1 above) may occur in geographic or environmental space (these are often linked but are not

synonymous) and can be a source of uncertainty in some cases. Species interactions, for example, may impose limits on a species’ observed distribution through competition. Such interactions are unavoidably captured in the occurrence records used in correlative models but, as Rohr, Halstead & Raﬀel (2011) point out, they are not explicitly modelled. Predictive uncertainty thus arises where the interactions (or other factors) are not transferable to the new geographic area (e.g. the competing species may be absent). In terms of species interactions, however, our model was not unreasonably extrapolative because Bd infects a very broad range of hosts that were available in both training and projection spaces. Extrapolation in environmental space is arguably more important than extrapolation in geographic space. With its ‘clamping’ function, Maxent allows the user to evaluate whether signiﬁcant extrapolation in environmental space has occurred when making projections (see methods; Murray et al. 2011). Our projections were not extrapolative in this sense either (see results; Murray et al. 2011) and as such this issue is unlikely to be a major source of bias in our study. Several of Rohr, Halstead & Raﬀel’s (2011) remaining concerns arise from whether our models were built upon occurrence records that fairly represent the environmental space inhabitable by Bd in Australia (point 2 above). This would not be the case for a species that is not at or near equilibrium in the region in which the model is trained (i.e. other factors, such as dispersal routes, may be more important for prediction, as is often the case when modelling invasive species undergoing range expansion). Similarly, this would not be the case if sampling was strongly biased in environmental space. We show that these are also unlikely to be major sources of bias in our study. On the basis of our results regarding the predictive ability of human population density (HPD) as a variable in the models (ﬁg. S8; Murray et al. 2011), Rohr, Halstead & Raﬀel (2011) correctly suggest that human associated factors might be aﬀecting the observed distribution of Bd in Australia. We agree and discuss with reference to relevant literature several reasons how this might be the case in relation to our study system and results. In their commentary, Rohr, Halstead & Raﬀel (2011) speculate further on mechanisms to explain this result and further emphasize the eﬀects that this could have on our results. However, while it is true that, as a single predictor, HPD contained the most unique predictive information that was not together present in the other variables in the model, the loss of this unique information scarcely reduced overall predictive ability (omission equated to c. 1Æ5% decrease in accuracy). In addition, as a single explanatory factor, HPD was only the sixth best predictor in the model and was considerably worse than the best environmental predictors. The ability of HPD to predict Bd distribution over and above the environmental variables is thus likely to be restricted to a small fraction of the occurrence records. In contrast, the ability of the environmental variables to explain Bd distribution over and above HPD is clearly more signiﬁcant, particularly as it relates to the aims, hypotheses and methods of our study (Murray et al. 2011).

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 177–180

Reply to Rohr et al. (2011) 179 This is not to say that we think that humans have not played a role in the shaping of Bd’s observed distribution in Australia. Our more conservative and biologically plausible interpretations of these results given the uncertainties ⁄ lack of supporting evidence are discussed in ﬁg. S8 (Murray et al. 2011) with reference to the relevant literature. Rohr, Halstead & Raﬀel (2011) also make two arguments about the eﬀect that humans will have on sampling bias. First they suggest, as we do in ﬁg. S3 (Murray et al. 2011), that sampling is likely to have been biased towards areas of higher environmental suitability for Bd because these are the areas where frog declines and mortalities have occurred due to severe chytridiomycosis. Later, and in contrast to their ﬁrst suggestion, Rohr, Halstead & Raﬀel (2011) propose that frequent human introductions and spread should bias the occurrence records towards generally unsuitable environments around ports, cities, highways and the coast (which would act as a giant drift-fence of sorts), where they state that Bd is almost exclusively found. Bd is not almost exclusively found near ports, cities and the coast in Australia and repeated introductions and humanaided spread within Australia remains a hypothesis in need of substantive evidence. As described in Murray et al. (2010a), Skerratt et al. (in press) and Berger et al. (2004), the majority of frog decline sites in Australia are located in remote, relatively pristine upland wilderness areas away from cities and sampling for Bd in Australia has occurred across a wide range of environments that are likely to be representative of both suitable and unsuitable conditions. Our database consists of opportunistic, systematic and retrospective (museum) sources. Some of these are likely biased towards areas inhabited by humans (e.g. public submission of sick or dead frogs; Berger 2001), while others are biased towards areas in remote wilderness generally far removed from human inﬂuences (e.g. McDonald et al. 2005). Quantifying and reducing the overall sampling bias is nevertheless identiﬁed as a future priority in our study (Murray et al. 2011). Hence, while we agree with Rohr, Halstead & Raﬀel (2011) that methods and routes of dispersal will be important for predicting the future distribution of chytridiomycosis, it is less important for characterizing current distribution in Australia where the majority of potential dispersal appears to have already occurred. This hypothesis is described in detail in the manuscript and largely supported by our results (Murray et al. 2011). Possible exceptions at the northern and southern extremes of Bd’s distribution are discussed in the paper, as are the main sources of bias and the weak signal that HPD retains some explanatory value over and above environmental factors. The value of our results in practice are demonstrated and discussed as are our recommendations for management given the uncertainties (e.g. Phillott et al. 2010). Finally, transparency is one of our study’s key features: all data have been made publicly available for scrutiny and for others to use (Murray et al. 2010a). We hope this fosters researcher coordination and expedites the development of improved models for amphibian conservation.

Key issues for modelling the future distribution of invasive pathogens Many of Rohr, Halstead & Raﬀel’s (2011) concerns nevertheless remain relevant to predicting the distribution and impacts of chytridiomycosis and other pathogens under climate change and at invasion fronts. In these cases, models need to avoid or be robust to signiﬁcant extrapolation by being transferable. One way of doing this is to shift the focus from correlative SDMs (which require only occurrence records) to mechanistic or process-based SDMs (which require more complete knowledge of a species’ responses to its biotic and abiotic environment) (see Kearney & Porter 2009 for a review). Combinations of these methods may also oﬀer advantages, as each can inform the other. Our approach to date for these more extrapolative situations has thus focussed on the mechanisms of Bd’s growth, dispersal and transmission. Simple mechanistic models that link Bd proliferation with environment (using, for example, biophysical performance curves denoting optimal conditions for growth and lethal limits) have already proved useful for describing seasonal and inter-annual infection patterns in wild frogs in subtropical Australia, where environmental suitability for Bd is generally very high but varies temporally (K.A. Murray, unpublished data). These models also show considerable promise for characterising Bd’s fundamental niche and hence potential distribution globally (K.A. Murray, unpublished data). We anticipate that this class of models will be particularly useful for estimating the inﬂuence that climate change may have on chytridiomycosis distribution, dynamics and impacts. In the meantime, correlative SDMs can provide a wealth of useful information that can be immediately adopted. We have demonstrated this in our paper and have subsequently explored how our metric of environmental suitability for infection with Bd may be useful among multiple threats and life-history and ecological traits in more detailed studies of decline risk in amphibians (Murray et al. 2010b). We have also subsequently employed the results to short-list unknown wild hosts for Bd (Murray & Skerratt in press). We therefore encourage the use of correlative SDMs in other regions of the world confronting similar conservation challenges to Australia in the face of this global threat, particularly where data and resources are sparse but biodiversity values are high.

References Berger, L. (2001) Diseases in Australian frogs. PhD, James Cook University, Townsville. Berger, L., Speare, R., Hines, H.B., Marantelli, G., Hyatt, A.D., McDonald, K.R., Skerratt, L.F., Olsen, V., Clarke, J.M., Gillespie, G., Mahony, M., Sheppard, N., Williams, C. & Tyler, M.J. (2004) Eﬀect of season and temperature on mortality in amphibians due to chytridiomycosis. Australian Veterinary Journal, 82, 434–439. Elith, J., Graham, C.H., Anderson, R.P., Dudik, M., Ferrier, S., Guisan, A., Hijmans, R.J., Huettmann, F., Leathwick, J.R., Lehmann, A., Li, J., Lohmann, L.G., Loiselle, B.A., Manion, G., Moritz, C., Nakamura, M., Nakazawa, Y., Overton, J.M., Peterson, A.T., Phillips, S.J., Richardson, K., Scachetti-Pereira, R., Schapire, R.E., Soberon, J., Williams, S., Wisz, M.S.

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 177–180

180 K. A. Murray et al. & Zimmermann, N.E. (2006) Novel methods improve prediction of species’ distributions from occurrence data. Ecography, 29, 129–151. Kearney, M. & Porter, W. (2009) Mechanistic niche modelling: combining physiological and spatial data to predict species’ ranges. Ecology Letters, 12, 334–350. McDonald, K., Mendez, D., Muller, R., Freeman, A. & Speare, R. (2005) Decline in the prevalence of chytridiomycosis in frog populations in North Queensland, Australia. Paciﬁc Conservation Biology, 11, 114–120. Murray, K.A. & Skerratt, L.F. (in press) Predicting wild hosts for amphibian chytridiomycosis: integrating host life-history traits with pathogen environmental requirements. Human and Ecological Risk Assessment, in press. Murray, K.A., Retallick, R., McDonald, K., Mendez, D., Aplin, K., Kirkpatrick, P., Berger, L., Hunter, D., Hines, H.B., Campbell, R., Pauza, M., Driessen, M., Speare, R., Richards, S.J., Mahony, M., Freeman, A., Phillott, A.D., Hero, J.-M., Kriger, K., Driscoll, D., Felton, A., Puschendorf, R. & Skerratt, L.F. (2010a) The distribution and host range of the pandemic disease chytridiomycosis in Australia spanning surveys from 1956 to 2007. Ecology, 91, 1557. Murray, K.A., Rosauer, D., McCallum, H. & Skerratt, L.F. (2010b) Integrating species traits with extrinsic threats: closing the gap between predicting and preventing species declines. Proceedings of the Royal Society B-Biological Sciences, Published online October 27, 2010. Murray, K.A., Retallick, R.W.R., Puschendorf, R., Skerratt, L.F., Rosauer, D., McCallum, H.I., Berger, L., Speare, R. & VanDerWal, J. (2011) Assessing spatial patterns of disease risk to biodiversity: implications for the management of the amphibian pathogen, Batrachochytrium dendrobatidis. Journal of Applied Ecology, 48, 163–173. Phillips, S.J., Anderson, R.P. & Schapire, R.E. (2006) Maximum entropy modeling of species geographic distributions. Ecological Modelling, 190, 231–259.

Phillips, S.J. & Dudik, M. (2008) Modeling of species distributions with Maxent: new extensions and a comprehensive evaluation. Ecography, 31, 161– 175. Phillips, S.J., Dudik, M., Elith, J., Graham, C.H., Lehmann, A., Leathwick, J. & Ferrier, S. (2009) Sample selection bias and presence-only distribution models: implications for background and pseudo-absence data. Ecological Applications, 19, 181–197. Phillott, A.D., Speare, R., Hines, H.B., Meyer, E., Skerratt, L.F., McDonald, K.R., Cashins, S.D., Mendez, D. & Berger, L. (2010) Minimising exposure of amphibians to pathogens during ﬁeld studies. Diseases of Aquatic Organisms, 92, 175–185. Rohr, J.R., Halstead, N.T. & Raﬀel, T.R. (2011) Modelling the future distribution of the amphibian chytrid fungus: the inﬂuence of climate means and variances and human-associated factors. Journal of Applied Ecology, 48, 174–176. Skerratt, L.F., McDonald, K.R., Hines, H.B., Berger, L., Mendez, D., Phillott, A.D., Cashins, S., Murray, K.A. & Speare, R. (2010) Application of the survey protocol for chytridiomycosis to Queensland, Australia. Diseases of Aquatic Organisms, 92, 117–129. VanDerWal, J., Shoo, L.P., Graham, C. & William, S.E. (2009) Selecting pseudo-absence data for presence-only distribution modeling: how far should you stray from what you know? Ecological Modelling, 220, 589– 594. Veloz, S.D. (2009) Spatially autocorrelated sampling falsely inﬂates measures of accuracy for presence-only niche models. Journal of Biogeography, 36, 2290–2299. Received 19 October 2010; accepted 9 November 2010 Handling Editor: Marc Cadotte

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 177–180

Journal of Applied Ecology 2011, 48, 56–66

doi: 10.1111/j.1365-2664.2010.01921.x

Addressing challenges when studying mobile or episodic species: hierarchical Bayes estimation of occupancy and use Rua S. Mordecai1†, Brady J. Mattsson1*‡, Caleb J. Tzilkowski2 and Robert J. Cooper1 1

Warnell School of Forestry and Natural Resources, University of Georgia, Athens, GA 30602, USA; and 2National Park Service, Eastern Rivers and Mountains Network, Forest Resources Building, University Park, PA 16802, USA

Summary 1. Understanding the distribution and ecology of episodic or mobile species requires us to address multiple potential biases, including spatial clustering of survey locations, imperfect detectability and partial availability for detection. These challenges have been addressed individually by previous modelling approaches, but there is currently no extension of the occupancy modelling framework that accounts for all three problems while estimating occupancy (w), availability for detection (i.e. use; h) and detectability (P). 2. We describe a hierarchical Bayes multi-scale occupancy model that simultaneously estimates site occupancy, use, and detectability, while accounting for spatial dependence through a state-space approach based on repeated samples at multiple spatial or temporal scales. As an example application, we analyse the spatiotemporal distribution of the Louisiana waterthrush Seiurus motacilla with respect to catchment size and availability of potential prey based on data collected along Appalachian streams of southern West Virginia, USA. In spring 2009, single observers recorded detections of Louisiana waterthrush (henceforth, waterthrush) within 75 m of point-count stations (i.e. sites) during four 5-min surveys per site, with each survey broken into 1-min intervals. 3. Waterthrushes were widely distributed (w range: 0Æ6–1Æ0) and were regularly using (h range: 0Æ4– 0Æ6) count circles along forested mountain streams. While accounting for detection biases and spatial dependence among nearby sampling sites, waterthrushes became more common as catchment area increased, and they became more available for detection as the per cent of the benthic macroinvertebrates that were of the orders Ephemeroptera, Plecoptera or Trichoptera (EPT) increased. These results lend some support to the hypothesis that waterthrushes are inﬂuenced by instream conditions as mediated by watershed size and benthic macroinvertebrate community composition. 4. Synthesis and applications. Although several available modelling techniques provide estimates of occupancy at one scale, hierarchical Bayes multi-scale occupancy modelling provides estimates of distribution at two scales simultaneously while accounting for detection biases and spatial dependencies. Hierarchical Bayes multi-scale occupancy models therefore hold signiﬁcant potential for addressing complex conservation threats that operate at a landscape scale (e.g. climate change) and probably inﬂuence species distributions over multiple scales. Key-words: detection probability, Louisiana waterthrush, Markov chain Monte Carlo, monitoring, multi-scale occupancy, Seiurus motacilla, spatiotemporal distribution, state-space modelling, WinBUGS

*Correspondence author. E-mail: [email protected] Present addresses: †U.S. Fish & Wildlife Service, South Atlantic Landscape Conservation Cooperative, Raleigh, NC 27699–1701. ‡U.S. Geological Survey, Patuxent Wildlife Research Center, Laurel, MD 20708, USA

Introduction Much ecological research seeks to understand drivers of species distributions across space and time. Examples include studies of metapopulation ecology (Hanski 1994), population viability (Beissinger & Westphal 1998), community

2011 The Authors. Journal of Applied Ecology 2011 British Ecological Society

Hierarchical Bayes multi-scale occupancy 57 composition and dynamics (Mordecai, Cooper & Justicia 2009; Zipkin, Dewan & Royle 2009), resource selection (MacKenzie 2006) and disease spread (Thompson 2007). Modelling distribution of species based on presence–absence data using occupancy models oﬀers ﬂexibility in addressing such diverse questions with relatively simple sampling designs that account for detectability (Mackenzie & Royle 2005). Understanding the distribution and ecology of episodic or mobile species, however, requires us to address multiple challenges related to sampling biases (Pollock et al. 2004; Ke´ry & Schmidt 2008; Ke´ry et al. 2009). In particular when studying vocal species, challenges include (i) individuals may be more detectable in acoustically favourable environments (Paciﬁci et al. 2008; Mattsson & Marshall 2009), (ii) individuals periodically become unavailable for detection within a sample unit (Farnsworth et al. 2002; Diefenbach et al. 2007; Rota et al. 2009) and (iii) spatial clustering of survey locations may induce spatial dependence among nearby points (for review see Campomizzi et al. 2008). The ﬁrst challenge (imperfect detection) can be addressed by simultaneously estimating occupancy and detection probabilities based on repeated detection ⁄ non-detection data (MacKenzie et al. 2002; Mattsson & Marshall 2009). If unaddressed, variation in detectability can produce misleading inferences regarding species distribution (Williams, Nichols & Conroy 2002; Gu & Swihart 2004). Detection bias has been recognized and addressed in several applications that investigate distributions of species (Wintle et al. 2005; O’Connell et al. 2006; Bailey et al. 2007; Ke´ry & Schmidt 2008).

A second challenge is that periodic unavailability for detection due to species movement or phenology may violate the closure assumption of occupancy models and may generate biased estimates of patch occupancy (Pollock et al. 2004; Ke´ry & Schmidt 2008). The robust design is a sampling design comprised of nested primary and secondary surveys and allows application of models that account for (i) variation in detectability during each secondary survey and (ii) violation of the closure assumption among primary surveys (Pollock 1982). In addition to providing a means to account for potential biases, the robust design oﬀers an opportunity to distinguish occupancy at two nested scales (Fig. 2). At a coarser scale, we can estimate the probability that a site is usable (i.e. that a species may use the site), which we deﬁne here as occupancy (w). Given species occupancy at the coarser scale, we can then estimate the probability that a species uses a site during each primary survey, which we deﬁne here as use (h). Taken together, this multi-scale modelling approach allows examination of species distribution at two scales simultaneously. Multi-scale occupancy models may be ﬁt to detection data collected using the robust design, and they simultaneously provide estimates of occupancy, use and detection (Mordecai 2007; Nichols et al. 2008). As such, these models may be particularly useful for investigators that are interested in examining patch occupancy across multiple primary surveys and during each of >2 primary surveys (e.g. days or weeks). Use (i.e. availability for detection) and detection are often separated when estimating local species abundance (for review see

GARI

Point count station New River Federal boundary Federal or State land

NERI

0 2·5 5

10

15

20

km

Fig. 1. Louisiana waterthrush survey locations in New River Gorge National River (NERI) and Gauley River National Recreation Area (GARI), WV. As illustrated in the transect map at Arbuckle Creek, NERI (inset), each transect contained ﬁve point-count stations and could have included as many as four Louisiana waterthrush territories. 2011 The Authors. Journal of Applied Ecology 2011 British Ecological Society, Journal of Applied Ecology, 48, 56–66

58 R. S. Mordecai et al. Johnson 2008), and such estimates can be provided by generalized Horvitz–Thompson estimators (Pollock et al. 2004; Diefenbach et al. 2007). In contrast, distinguishing patterns in species distribution (i.e. occupancy) from use and detection has received little attention (Mordecai 2007; Nichols et al. 2008). The third, and perhaps the least addressed, challenge is that survey locations are often clustered and conspeciﬁcs may aggregate or occupy areas covering multiple sampling locations, which induces spatial dependence and therefore underestimation of variation among nearby sample units (Sauer, Link & Royle 2005). A solution to this dependence is to apply a random eﬀect that references a coarser, aggregate sampling unit when predicting distribution at ﬁner spatial scales (Royle & Dorazio 2006; Royle et al. 2007). Although this may be accomplished through maximum-likelihood estimation and linear mixed modelling, hierarchical Bayes models oﬀer ﬂexible and robust approaches to modelling distributions of species based on sparse detections while accounting for spatiotemporal dependencies and detectability (Royle & Dorazio 2008, pp. 106–124). Applying a hierarchical Bayes approach to multi-scale occupancy models oﬀers a robust and extensible solution for dealing with multiple challenges of studying nested patterns of distribution or resource use by mobile or episodic species.

(a)

Here, we describe a multi-scale site occupancy model that integrates existing occupancy modelling approaches by simultaneously estimating site occupancy (w) and use (h) while accounting for detectability (P) and spatial dependence through the use of random eﬀects. In particular, this model addresses challenges to studying episodic or mobile species by employing a Bayesian state-space modelling approach and is an extension of existing multi-scale occupancy models that assume no spatial dependence among sample units (Mordecai 2007; Nichols et al. 2008). We ﬁrst demonstrate that multiscale occupancy models are a generalization of single-scale occupancy models, and then we describe sampling designs necessary to simultaneously estimate occupancy and temporal or spatial patterns of use while accounting for detectability. We then present an analysis based on bird data collected in southern West Virginia as part of a long-term monitoring programme administered by the National Park Service (NPS). In particular, we examine occupancy and temporal patterns of use by the Louisiana waterthrush Seiurus motacilla Vieillot, a riparian obligate passerine, based on catchment area and a measure of benthic macroinvertebrate community composition. Finally, we discuss the importance and potential extensions of hierarchical Bayes multi-scale occupancy modelling for addressing many questions in ecology, management and conservation biology.

Temporal replication of subsamples

Spatial replication of subsamples

(ψi)

Minute 1 (Pi,1)

...

Site

...

Point 1 (Pi,1)

Point j (Pi,j)

Minute j (Pi,j)

(b)

(θi,1)

Minute 1 (Pi,1,1)

...

Survey 1

...

Point 1 (Pi,1,1)

Point k (Pi,1,k)

Site (ψi)

...

Minute k (Pi,1,k)

(θi,j)

Minute 1 (Pi,j,1)

...

Survey j

...

Point 1 (Pi,j,1)

Point k (Pi,j,k)

Minute k (Pi,1,k)

Fig. 2. Comparison of example single-scale (a) and multi-scale (b) occupancy designs with either temporal or spatial replication of subsamples. A square represents a site, surveys take place at sites (here, point count circles), and wi is the probability that site i is occupied by a species. In single-scale occupancy design (a), P is the probability of detecting that species during a subsample (e.g. minute of point count) given the site is occupied. In multi-scale occupancy design (b), h is the probability that the species uses the site on a speciﬁc survey given the site is occupied, and P is the probability of detecting that species during a subample given the site is used during a speciﬁc survey. 2011 The Authors. Journal of Applied Ecology 2011 British Ecological Society, Journal of Applied Ecology, 48, 56–66

Hierarchical Bayes multi-scale occupancy 59

Materials and methods STUDY AREA AND FIELD PROTOCOL

As part of the NPS Inventory and Monitoring Programme in the Eastern Rivers and Mountains Network, a long-term streamside bird monitoring programme was developed, in part, to monitor the distribution of the Louisiana waterthrush (henceforth, waterthrush; Mattsson & Marshall 2010) which has been demonstrated to be an indicator of biotic integrity in headwater streams (Mattsson & Cooper 2006; Mulvihill, Newell & Latta 2008). This riparian obligate warbler consumes primarily benthic macroinvertebrates along stream margins, and their mostly linear territories typically extend 250– 300 m along stream networks in this region (Mattsson et al. 2009). Watershed conditions such as catchment area, topography and land cover can aﬀect the community composition of potential prey for waterthrushes (Klemm et al. 2002; Roy et al. 2003; King et al. 2005), which in turn may aﬀect waterthrush distribution. Waterthrush monitoring is one element of a larger ‘Vital Signs’ monitoring eﬀort in the network that includes monitoring of water quality and benthic macroinvertebrates (Marshall & Piekielek 2007). From a total of 80 candidates, 28 tributary watersheds (2nd–3rd Strahler stream order; Strahler 1952) were selected for waterthrush monitoring in two National Parks (i.e. New River Gorge National River and Gauley River National Recreation Area) of southern West Virginia (3757¢ N, 814¢ W‘; Fig. 1; Mattsson & Marshall 2010). These parks are characterized by steep, forested 1st–3rd order drainages that ﬂow into larger rivers that bisect each park. Watersheds within these parks were selected using a stratiﬁed randomization based on underlying features, including watershed size, geology and land ownership. Watersheds were delineated and catchment area (i.e. the area of land that drains to a focal point in the landscape, also known as watershed area) within each watershed was estimated based on a 10-m digital elevation model (US Geological Survey 2004) using ArcGIS 9Æ1 (ESRI 2005). Once delineated, the number of cells that ﬂow into any focal point within a watershed were converted into a measure of catchment area for that point. Some candidate watersheds were excluded from monitoring due to logistical limitations (e.g. safe access). Within each selected watershed, a 1-km streamside transect was established within a predetermined range of catchment areas (i.e. 1– 99Æ9 km2). As such, transects were established along reaches that were perennial and wadeable. If >1 km of stream was available, then a series of four adjacent 250-m segments were selected at random. A 1km streamside transect was expected to contain up to four mostly linear waterthrush territories (Fig. 1), based on territory mapping of colour-banded waterthrushes in this region (Mulvihill, Newell & Latta 2008). Along each 1-km transect, a point-count station was established every 250 m, totalling ﬁve stations per transect and 140 points throughout the two parks. Detectability of waterthrush pairs is particularly high during the ﬁrst month following ﬂedging of nests (Mattsson & Cooper 2006). As such, each transect was visited twice from 23 May to 19 June 2009 to coincide with the peak period of waterthrush ﬂedgling care. Transect visits were 4–20 days apart for any given transect. During each visit day, one of four observers traversed a transect twice (i.e. upstream and downstream), conducted 5-min point counts at each station during both passes, and recorded per-minute detections (aural or visual) of waterthrush adults or young within an estimated 75 m of the point-count station. This resulted in two levels of temporal replication (Fig. 2). Before conducting any transect surveys, observers underwent ‡5 days of training that focused on improving

accuracy in estimating distances to waterthrushes within 75 m using both aural and visual cues. Each point-count station was therefore sampled four times (four passes over two days), and there were ﬁve subsamples (ﬁve 1-min intervals) per sample (i.e. 4 · 5 = 20 subsamples per point throughout the season). Due to travel time between sites and the limited daily period of waterthrush vocal activity, it was more reasonable for an observer to conduct two passes per transect visit than it would have been for that observer to conduct surveys along two transects per day. To account for observer variability with respect to detection of waterthrushes while minimizing the number of days between visits, one observer conducted counts on both passes along a transect on a given day, and another observer conducted counts on both passes along a transect on the second day. Early spring is the season when benthic macroinvertebrate communities are typically most diverse (Huryn, Wallace & Anderson 2008); consequently, benthic macroinvertebrates were sampled from 26 of 28 transects during March 2009. This period also coincides with waterthrush territory establishment in the region (Mattsson et al. 2009). The benthic macroinvertebrate sampling protocol was based on methods developed for the US Geological Survey (Moulton et al. 2000, 2002). For more details on sampling methods, see Tzilkowski, Weber & Ferreri (2009). Substrate disturbance sampling, with a 0Æ25 m2 template and Slack sampler (500 lm mesh), was used to collect subsamples from ﬁve riﬄes throughout each transect. Stream conditions (i.e. substrate, water velocity and depth) were measured and kept consistent among riﬄe subsamples. These subsamples were composited into one sample for each stream transect, preserved in 95% ethanol and transported to the laboratory. Fixed-count subsamples of 240–360 individuals were identiﬁed to genus for all taxa, except for chironomid midges and oligochaete worms, using standard dichotomous keys (Peckarsky et al. 1990; Merritt, Cummins & Berg 2008). For the analysis, the percentage of individuals belonging to the insect orders Ephemeroptera, Plecoptera or Trichoptera (henceforth, % EPT) was calculated for each sample, as this metric is related to waterthrush distribution in other parts of its range (Mattsson & Cooper 2006).

HIERARCHICAL BAYES MULTI-SCALE OCCUPANCY MODEL

We ﬁrst illustrate how a single-season, single-scale occupancy model (MacKenzie et al. 2002) is generalized to a single-season, multi-scale occupancy model (Mordecai 2007; Nichols et al. 2008) based on the sampling design for examining waterthrush distribution patterns. In doing so, we largely follow the theory and notation of MacKenzie et al. (2002). Suppose that each transect were visited during only a single day, and a 5-min survey was repeated twice per day at each point-count station (i.e. site) following a temporal replication sampling design (Fig. 2). A single-season occupancy model could then be applied to estimate the probability that a waterthrush occupied a site that day (w) and the minute-by-minute probability of detecting the waterthrush (P), given the site is occupied. For a given site-visit day, using 1 to denote a detection and 0 a non-detection to create a detection history, if we only detected a waterthrush during the third minute of the ﬁrst pass (i.e. a detection history of 00100 00000), then we could conclude that the species occupied the site. Alternatively, if we never detected a waterthrush at the site (i.e. 00000 00000), then either (i) the site was occupied but the species was not detected or (ii) the site was not occupied. As such, minute-by-minute detection ⁄ non-detection data provides information to estimate both w and P. In reality, however, transects were visited on two diﬀerent days (Fig. 3a). A multi-scale occupancy model can therefore be used to

2011 The Authors. Journal of Applied Ecology 2011 British Ecological Society, Journal of Applied Ecology, 48, 56–66

60 R. S. Mordecai et al. use and occupancy, and an observation model for the repeated detections themselves. The state process model is composed of two equations, starting with the binary site occupancy state:

Occupancy probability

(a)

Zi Bernoulliðwi Þ for i ¼ 1; 2; . . . N followed by the binary use state, which is conditional on the respective site occupancy state: uij jZi Bernoulliðhij Þ for j ¼ 1; 2; . . . V

Catchment area (km2)

Daily use probability

(b)

where, under a temporal replication sampling design (Fig. 3a), i indexes the N sites and j indexes the V surveys. Therefore, a species occupies a site according to a Bernoulli trial with parameter w, and the species uses (i.e. is available for detection at) the site during a survey according to another Bernoulli trial with parameter h. The observation model, which is conditional on the state of use is denoted as follows: yijk juij BernoulliðPijk Þ for k ¼ 1; 2; . . . S

Fig. 3. Eﬀect of catchment area and % EPT (in benthic macorinvertebrate community) on probability of occupancy (a) and of daily use (b) by the Louisiana waterthrush within 75-m radius count circles along streams in two national parks of southern West Virginia during spring of 2009. Per cent EPT was held at 70% for graph A, and catchment area was held at 20 km2 in graph B. Dashed lines represent 95% BCIs. estimate (i) the probability that a waterthrush occupied a site (w) at least once from the start of the ﬁrst survey to the end of the ﬁnal survey of that site, (ii) the probability of use (h) by a waterthrush during an individual point-count survey given the site is occupied and (iii) the probability of detecting a waterthrush (P) during an individual survey given the site was used. For example, if we only detect a waterthrush on the third minute of the ﬁrst survey (00100 00000 00000 00000), then we could assert that the species: (i) occupied the site, (ii) used the site during the ﬁrst survey and (iii) either used the site during the subsequent surveys and was not detected or did not use the site during these subsequent surveys. Alternatively, suppose no individuals were detected during any survey of the site (i.e. 00000 00000 00000 00000). In this case, a waterthrush either: (i) did not occupy the site; (ii) occupied but did not use the site (i.e. species was unavailable for detection) during any survey or (iii) was not detected despite occupying and using the site during either survey. Minute-by-minute detection ⁄ non-detection data during surveys that are repeated on multiple days therefore provide information to estimate not only w and P, but also h. Note that a single-scale occupancy model may also be ﬁt to such a data set, and its performance could be directly compared with that of a multi-scale occupancy model. This is analogous to the case where single-season and dynamic occupancy models can be ﬁt to the same data set (MacKenzie et al. 2003). We formulated the multi-scale occupancy model as a state-space model (Royle & Ke´ry 2007) that comprises two submodels, including a state process model for the latent or partially observed processes of

where k indexes S subsamples, y is a three-dimensional array of 1’s or 0’s representing detections or non-detections of a species for each site-survey-subsample combination, and P is the corresponding three-dimensional array of detection probabilities for each site-survey-subsample combination. Thus, if the species uses an occupied site during a survey, then the species is detected during that survey according to a Bernoulli trial with parameter P. To demonstrate how sampling design dictates interpretations of occupancy, use, and detectability, we refer again to the waterthrush sampling design where N = 140 point-count stations are surveyed during V = 4 surveys, and detection ⁄ non-detection data are collected during S = 5 successive 1-min counts during each survey. In this case, parameters could be interpreted as follows: (i) occupancy (w) is the probability that a site is usable, i.e. that a waterthrush may use the site; (ii) use (h) is the probability that a waterthrush uses the site by vocalizing during a survey given a site is occupied; and (iii) detection (P) is the probability of detecting a waterthrush during a survey given that a waterthrush uses the site that day. It is therefore possible for a waterthrush to occupy a site but not use that site during the four surveys. This design, where the replication is temporal (Fig. 3a), focuses on the frequency that a waterthrush uses a site (or is available for detection). It is important to note that the model structure is easily adaptable for questions focused on spatial patterns in addition to temporal patterns of use among plots within sites (Fig. 3b). In particular, V would instead represent the number of plots per site, and S would represent the number of temporally replicated surveys per plot. Interpretations of occupancy, use, and detection are therefore contingent on the sampling design (Mackenzie & Royle 2005). Spatial replication, however, may introduce Markovian dependence due to animal movements and require approaches that accommodate such dependence (Hines et al. 2010). Covariates and missing data can be easily incorporated into the state-space multi-scale occupancy model as they have for other statespace occupancy models (Royle & Dorazio 2008). Eﬀects of site-level covariates on w, h, and P, patch or survey-level covariates on h and P, and subsample-level covariates on P can be modelled using the logit transformation, where Y is the response parameter of interest (i.e. either w, h or P), X is the covariate information and B is the vector of logistic model coeﬃcients for estimation: Y¼

eXB 1 þ eXB

2011 The Authors. Journal of Applied Ecology 2011 British Ecological Society, Journal of Applied Ecology, 48, 56–66

Hierarchical Bayes multi-scale occupancy 61 Under some standard sample designs, sites are nested within larger spatial units (henceforth, aggregates) to improve sampling eﬃciency or accommodate logistical constraints (Bibby & Burgess 2000; Sauer, Fallon & Johnson 2003; Newson et al. 2008). Spatial dependence due to nested or clustered distribution of species among nearby sites, unless taken into account, may yield biased estimates of distribution (for review see Dormann et al. 2007). In addition to covariates, a random intercept for aggregates (e.g. transects each comprised of multiple survey sites; b0i) can be incorporated into the model to account for this dependence: logitðwi Þ ¼ a0a þ a1 x1 þ a2 x2 þ þ ar xr where a indexes aggregates, i indexes sites and the model may contain any number of ﬁxed eﬀects, indexed by r. In the hierarchical Bayes analysis, prior distributions are deﬁned such that aggregate intercept values share a common mean and variance (Royle & Dorazio 2006; Howell, Peterson & Conroy 2008): b0a Normallðl; r2 Þ

This variance then represents variation among aggregates or the level of spatial dependence.

MODEL ASSUMPTIONS

Obtaining accurate estimates via the multi-scale occupancy model presented here requires important assumptions. Unlike single-scale, single-season occupancy models (MacKenzie et al. 2002), multi-scale occupancy models allow for the possibility that species become occasionally unavailable for detection at a site. For example, a waterthrush may move into or out of a count circle as it passes along its streamside territory. Similar to dynamic occupancy models (MacKenzie et al. 2003), multi-scale occupancy models assume that sites are closed (i.e. availability for detection remains constant) during each primary survey. For example, waterthrushes do not move into or out of a count circle throughout a 5-min point-count survey. Secondly, species are identiﬁed correctly upon detection, or no species are misidentiﬁed. Again, this assumption may be relaxed to account for false positives, as it has been for single-scale occupancy modelling (Royle & Link 2006). Thirdly, covariates must be included to account for any detection biases, such as diﬀerences among observers, temporal variation in species perceptibility (e.g. singing rates), sampling eﬀort and environmental conditions (Mattsson & Marshall 2009). Finally, parameters must be incorporated to account for any dependencies of detections among sites (e.g. spatially clustered or large territories), surveys (e.g. temporally clustered availability for detection) or subsamples (e.g. observer expectation bias or temporally clustered availability for detection).

WATERTHRUSH ANALYSIS

We investigated patterns of waterthrush occupancy and use along tributaries by ﬁtting a hierarchical Bayes multi-scale occupancy model to temporally replicated detection ⁄ non-detection data (Fig. 2). We assume that use (h) represents availability for detection during a point count, as waterthrushes move along their ca. 250-m territories throughout the day. Waterthruses are thus only available for detection when they are present within the 75-m detection radius (henceforth, count circle) during a 5-min count. Therefore, occupancy (w) is the probability that a point is occupied by ‡1 waterthrushes at least once during the study period, h is the probability that ‡1 waterthrushes use the count circle during a particular pass given the point

is occupied, and P is the probability of detecting a waterthrush during one of the 5 min given they use that count circle on a particular pass. A count circle, therefore, is a site in a traditional, single-season occupancy model. Occupancy, use, and detectability of waterthrushes in count circles may depend on watershed-scale attributes, including local waterthrush territory density, and this dependence would manifest throughout a transect (Fig. 1). With this in mind, we applied the hierarchical Bayes multi-scale occupancy model that included a random intercept for variation among transects with respect to count-circle w and h. We also assumed that sources of variation in detectability of a waterthrush would be accounted for when considering observer-speciﬁc attributes (Mattsson & Marshall 2009) and previous detection of a waterthrush. Detections of waterthrushes may not be independent during a 5-min count due to an expectation bias of observers and ⁄ or temporally clustered singing activity of a waterthrush (Riddle et al. 2010). We therefore included in the model ﬁxed categorical covariates for observer (Obs) and detection during the previous minute (Prev). We also included as predictors of w, h and P ﬁxed eﬀects for catchment area and for % EPT. We predicted that w and h would increase with increasing catchment area and % EPT, as these are both expected to correspond directly with waterthrush food availability (Klemm et al. 2002; Mattsson et al. 2009). Catchment area and % EPT had weak, if any correlation (–0Æ19). As such, we developed a series of logistic regression equations to estimate the level of spatial dependence and relationships between covariates and response parameters of the multi-scale occupancy model: logitðwi Þ ¼ a0a þ a1 catchmenti þ a2 EPTi logitðhij Þ ¼ b0a þ b1 catchmenti þ b2 EPTi logitðPijk Þ ¼ d0 þ d1 catchmenti þ d2 EPTi þ d3;Obs Obsij þ d4 Previjk where i indexes sites, j indexes surveys within sites, k indexes subsamples within surveys, and Obs indexes observers, and Prev is a binary variable for previous detection. Parameters a0a and b0a represent random intercepts that account for transect-level spatial dependence when modelling occupancy and use, respectively, and d0 represents the ﬁxed intercept for detectability. Parameters a1, b1, and r1 represent slopes for the catchment area eﬀect, whereas a2, b2, and r2 represent slopes for the eﬀect of % EPT on occupancy and use, respectively. For modelling detectability, r3,Obs and r4 represent the slopes for the eﬀect of observer and previous detection, respectively. We used WinBUGS version 1Æ4 (Spiegelhalter et al. 2003) to ﬁt the model to the waterthrush data, which uses an Markov chain Monte Carlo (MCMC) algorithm. R and WinBUGS code for ﬁtting the model is provided in Appendix S1 (Supporting information), and we provided simpliﬁed code (i.e. without covariates) for running the hierarchical Bayes multi-scale occupancy model in WinBUGS in Appendix S2 (Supporting information). We chose a relatively uninformative normal prior with a logit-scale mean of 0 (i.e. 0Æ5 on the probability scale) and standard deviation of 1Æ58 (i.e. 0Æ83 on probability scale and precision of 0Æ4 on logit scale) for the mean of the random intercepts for w and h. Unlike a uniform prior on the logit scale, this normal prior results in an approximately uniform distribution from 0 to 1 on the probability scale. We also used this uninformative normal prior for the remaining parameters, except we used a uniform prior of 0Æ001–10 on the logit scale for the standard deviation of random intercepts (SDRs) for w and h. For the latter, the lower

2011 The Authors. Journal of Applied Ecology 2011 British Ecological Society, Journal of Applied Ecology, 48, 56–66

62 R. S. Mordecai et al. and upper bounds represent no spatial dependence and strong spatial dependence, respectively. To improve mixing of the MCMC algorithm, we truncated the normal priors from –10 to 10, disallowed values below 1E)5 or above 1–1E)5 for probabilities, and disallowed values below 1E-3 for standard deviation. We exported model output from WinBUGS to programme R (R Development Core Team 2006) and assessed convergence using the default values for the Raftery– Lewis test implemented in the R package BOA, which is based on a single chain of MCMC iterations (Raftery & Lewis 1992a,b; Smith 2007). Inferences regarding eﬀect sizes and direction were based on posterior means and 95% Bayesian Credible Intervals (BCIs; 2Æ5th– 97Æ5th percentile of the distribution), and parameter estimates are reported as mean with BCI in square brackets. In particular, if the BCI surrounding a slope estimate did not include zero, then we interpreted this as a statistically signiﬁcant (henceforth, signiﬁcant) eﬀect.

Occupancy Use No dependence

Results Waterthrushes were detected in 70 of 140 point-count circles for a naı¨ ve occupancy estimate of 0Æ500, and waterthrushes were detected during 99 of 280 surveys in these 70 count circles for a naı¨ ve use estimate of 0Æ353. Catchment areas ranged from 1Æ08 to 74 km2, benthic macroinvertebrate densities ranged from 436 to 14 469 individuals m)2, and % EPT ranged from 0Æ7 to 90Æ0%. The hierarchical model, when ﬁt to these data, converged after 2Æ5 million MCMC iterations following 100 000 discarded (i.e. burn-in) iterations. Based on estimates

Fig. 5. Spatial dependence among count circles within transects with respect to count-circle occupancy and survey-speciﬁc use by the Louisiana waterthrush along streams of two National Parks in southern West Virginia during spring 2009. The histogram shows the prior probability density, and the curves show the posterior probability distributions for the logit-scale standard deviation of random intercepts (SDR) for transect. Vertical dotted line at 0 standard deviation indicates the expectation if there were no spatial dependence of occupancy or use.

from this model, waterthrush occupancy of count circles increased on average from 59 to 100% across the range of catchment areas while holding % EPT at the mean value across sites (i.e. 70%), and this eﬀect was signiﬁcant (a1 = 1Æ775 [0Æ445, 3Æ745]; Fig. 3). Likewise, use increased on average from 35 to 66% across the range of % EPT while holding catchment area at the mean value across sites (i.e. 20 km2), and the credible interval was almost entirely above zero (b2 = 1Æ620 [)0Æ106, 3Æ527]; Fig. 3). Minute-speciﬁc detectability of waterthrushes increased as catchment size increased by 1 ha, and the credible interval was almost entirely above zero (r1 = 0Æ169 [)0Æ004, 0Æ330]; Fig. 4). Detectability more than doubled when a waterthrush was detected during the previous minute compared to when no waterthrushes were detected previously (Fig. 4). Catchment area had weak, if any, eﬀect on waterthrush daily use, and % EPT had weak, if any, eﬀects on waterthrush occupancy or detection. Based on posterior distributions for SDRs in the model, spatial dependence among count circles within transects was evident with respect to occupancy and use (Fig. 5). The mode of SDR for occupancy (0Æ2) was less than that for use (0Æ8).

Probability of detection

(a)

Probability of detection

(b)

Discussion Catchment area (km2) Fig. 4. Per-minute detectability with (a) and without (b) prior detection of the Louisiana waterthrush within 75-m radius count circles along streams in two National Parks of southern West Virginia during spring of 2009. Whiskers represent 95% BCIs.

Questions about species distribution and resource use are central to many ecological studies and their application for management and conservation. Hierarchical Bayes multi-scale occupancy modelling is an extension of existing approaches that model occupancy at a single scale while accounting for detectability and ⁄ or spatial dependence (MacKenzie et al.

2011 The Authors. Journal of Applied Ecology 2011 British Ecological Society, Journal of Applied Ecology, 48, 56–66

Hierarchical Bayes multi-scale occupancy 63 2002; Royle & Dorazio 2006; Mordecai 2007; Nichols et al. 2008). Speciﬁcally, hierarchical Bayes multi-scale occupancy modelling allows inference regarding species occurrence at two diﬀerent spatial or temporal scales while enabling incorporation of random eﬀects that account for nested sampling designs (e.g. count circles along transects). At two diﬀerent temporal scales, investigators can model both the probability of species occurrence at a site during the study period and the frequency of species occurrence (i.e. use of or availability for detection) at that site while accounting for detectability and spatial dependence among nearby sites. Alternatively, at two diﬀerent spatial scales, investigators can model both the probability of species occurrence in an area (e.g. management unit) and, if the species occurs in that area, the probability of species occurrence in smaller regions nested within that area (e.g. stands within the management unit) while accounting for spatial dependence among adjacent management units. An alternative, common approach to study species distribution at multiple scales is radiotelemetry (e.g. Michalski et al. 2006; Matson et al. 2007; Rittenhouse & Semlitsch 2007). Although radiotelemetry can provide data on both spatial and temporal patterns in use by individual animals, it is often logistically challenging and expensive to obtain suﬃcient sample size to detect diﬀerences in use among habitats (Murray 2006). In contrast, detection–nondetection data for highly visible or audible species tend to be inexpensive and easy to collect; thus, conducting repeated surveys of unmarked animals may be more eﬃcient than conducting telemetry for studying patterns of resource use by many species. Based on our analysis and rather simple sampling design, waterthrushes were not only widely distributed (w > 0Æ6) but were also often using (h > 0Æ4) riparian areas along forested mountain tributaries. While accounting for detection biases and spatial dependence among nearby sampling sites, waterthrushes became more common as catchment area increased. Community composition of instream prey varies with watershed size (Klemm et al. 2002; Roy et al. 2003; King et al. 2005), and waterthrushes may be attracted to assemblages of benthic macroinvertebrates found in larger watersheds. Daily use probability increased with increasing %EPT, and this result does not refute the hypothesis that waterthrushes are indicators of benthic macroinvertebrate community composition (Mattsson & Cooper 2006; Mulvihill, Newell & Latta 2008). Until additional macroinvertebrate metrics, water quality metrics (e.g. pH) or sampling methods are explored, catchment area appears to be a more important driver of waterthrush distribution; whereas % EPT is associated with waterthrush availability for detection. With respect to patterns of detection, waterthrushes that were using an area during a count were easier to detect if they were also detected earlier in that count. While the speciﬁc reason for the increased detectability is unclear, observer knowledge of the previous location of individual waterthrushes may have increased the rate of redetection. Alternatively, waterthrushes may have multi-minute bouts of vocalization, yielding clumped minute-by-minute detections. Whatever the mechanism, increased detection probability after a prior detection

may be common in bird surveys, and not accounting for this dependence can result in a downward bias in occupancy estimators (Riddle et al. 2010).

MULTI-SCALE PERSPECTIVE OF DISTRIBUTION ESTIMATES

In traditional occupancy modelling, use and detection probabilities are combined into a single parameter, P. There are many ecological problems, however, where separating use and detection would be particularly important. One common situation, as illustrated by the waterthrush analysis, involves an animal that is absent from large portions of its home range at any given time. In this case, an investigator can apply multi-scale occupancy models to investigate patterns of occupancy within patches of home ranges in addition to frequency of use for those patches. Furthermore, modelling w, h and P in a hierarchical framework becomes particularly informative when, as indicated by the waterthrush example, P < 1 and variation exists among coarser-scale sampling units (i.e. streamside transects) with respect to both occupancy and use. Modelling h and P could also be useful for revealing contrasting patterns in use and detection in relation to a factor of interest. For example, many bird species occasionally use nonforested habitat but spend a majority of their time in forested habitat (Lent & Capen 1995; Annand & Thompson 1997; Mordecai, Cooper & Justicia 2009). When these facultative species occupy non-forested habitat, they may be easier to detect due to the reduced visual and auditory obstructions in an open area. However, the probability that these species use non-forest habitat at a given time is lower, because they spend less time in that habitat. While multi-scale and single-scale occupancy models both predict occupancy rates for each habitat, multi-scale occupancy models could also estimate the negative trend in use and positive trend in detectability associated with non-forested habitat. Therefore, multi-scale occupancy models distinguish between parameters that are typically of ecological interest, occupancy (i.e. w) and use (i.e. h) and a parameter that is generally estimated only to account for perceptibility by observers for detecting that species (i.e. P; Johnson 2008). Multi-scale occupancy models provide estimates of distribution at two scales (i.e. w and h), and these estimates are subject to an important assumption about spatial independence. Using the state-space framework, spatial dependence may be addressed by adding to the process model a random eﬀect that indexes coarser spatial units. As shown in the waterthrush analysis, species distributions may be clustered such that nested sampling designs (e.g. point-transects) warrant inclusion of this random eﬀect to account for spatial dependence.

COMPARING MULTI-SCALE AND DYNAMIC OCCUPANCY MODELS

Use may be estimated based on the robust design under two alternative modelling approaches. First, dynamic occupancy models explicitly account for transitions in patch occupancy

2011 The Authors. Journal of Applied Ecology 2011 British Ecological Society, Journal of Applied Ecology, 48, 56–66

64 R. S. Mordecai et al. between successive primary surveys such as months, seasons or years (MacKenzie et al. 2003; Rota et al. 2009). Secondly, multi-scale occupancy models account for the possibility that an occupied patch may be periodically unused by providing estimates of (i) patch occupancy across primary surveys, (ii) use of occupied patches (i.e. availability for detection of at least one individual) during each primary survey and (iii) detectability of species within used patches during secondary surveys (Mordecai 2007; Nichols et al. 2008). Patch occupancy during each primary survey under the dynamic-occupancy modelling approach is analogous to the use parameter in multi-scale occupancy models. Therefore, initial occupancy in a dynamic occupancy model is the likelihood of use during the ﬁrst season. The two approaches, in fact, provide identical estimates for use of occupied patches under a random immigration model where immigration and emigration are equal. In contrast with dynamic occupancy models that allow estimation of use for a single scale of sampling units, multi-scale occupancy modelling allows estimation of species distribution at two nested temporal and ⁄ or spatial scales. As such, dynamic occupancy models provide a parameter for ‘seasonal’ use (h) but have no parameter for ‘cross-seasonal’ occupancy (w) as deﬁned in multi-scale occupancy models. Researchers interested in examining immigration or emigration between consecutive surveys (e.g. monthly, seasonal or annual) may be better served by the dynamic occupancy model (MacKenzie et al. 2003), whereas investigators that are interested in examining patterns of species distribution at multiple nested scales would be better served by the multi-scale occupancy model described here and elsewhere (Mordecai 2007; Nichols et al. 2008).

EXTENDING HIERARCHICAL MULTI-SCALE OCCUPANCY MODELS

Hierarchical Bayes multi-scale occupancy models can be expanded in numerous ways. Current extensions to singleseason occupancy models such as species interactions (MacKenzie, Bailey & Nichols 2004), community-level metrics (Dorazio & Royle 2005; Dorazio et al. 2006), dynamic occupancy models (MacKenzie et al. 2003), and false positives (Royle & Link 2006) could all be applied to multi-scale occupancy models. Additionally, probabilities of occupancy, use and detection at any scale could be estimated with doubleobserver sampling (Cook & Jacobson 1979), removal models (Moran 1951; Seber 1982) or distance sampling (Reynolds, Scott & Nussbaum 1980; Buckland, Burnham & Laake 1993). In conclusion, hierarchical Bayes multi-scale occupancy models have many potential applications and extensions for studying the distribution and resource use patterns of mobile or episodic species that exhibit spatial heterogeneity.

IMPLICATIONS FOR CONSERVATION PLANNING

Conservation organizations are increasingly challenged by complex threats, such as climate change, which may aﬀect species distributions at multiple scales (Elith & Leathwick 2009; Galatowitsch, Frelich & Phillips-Mao 2009). Evaluating

conservation policies to address these threats will probably require analysis of clustered detection–nondetection data for elusive species across a wide range of spatial and temporal scales. The proposed hierarchical Bayes extension to multiscale occupancy models will allow conservation organizations to evaluate alternative management options while accounting for challenges associated with clustered sampling designs for species that are highly mobile or episodic.

References Annand, E.M. & Thompson, F.R. (1997) Forest bird response to regeneration practices in central hardwood forests. Journal of Wildlife Management, 61(1), 159–171. Bailey, L.L., Hines, J.E., Nichols, J.D. & MacKenzie, D.I. (2007) Sampling design trade-oﬀs in occupancy studies with imperfect detection: examples and software. Ecological Applications, 17(1), 281–290. Beissinger, S.R. & Westphal, M.I. (1998) On the use of demographic models of population viability in endangered species management. Journal of Wildlife Management, 62(3), 821–841. Bibby, C.J. & Burgess, N.D. (2000) Bird Census Techniques. Academic Press, London, UK. Buckland, S.T., Burnham, D.R. & Laake, J.L. (1993) Density Estimation Using Distance Sampling. Chapman and Hall, London. Campomizzi, A.J., Butcher, J.A., Farrell, S.L., Snelgrove, A.G., Collier, B.A., Gutzwiller, K.J., Morrison, M.L. & Wilkins, R.N. (2008) Conspeciﬁc attraction is a missing component in wildlife habitat modeling. Journal of Wildlife Management, 72(1), 331–336. Cook, R.D. & Jacobson, J.O. (1979) Design for estimating visibility bias in aerial surveys. Biometrics, 35(4), 735–742. Diefenbach, D.R., Marshall, M.R., Mattice, J.A. & Brauning, D.W. (2007) Incorporating availability for detection in estimates of bird abundance. Auk, 124(1), 96–106. Dorazio, R.M. & Royle, J.A. (2005) Estimating size and composition of biological communities by modeling the occurrence of species. Journal of the American Statistical Association, 100(470), 389–398. Dorazio, R.M., Royle, J.A., Soderstrom, B. & Glimskar, A. (2006) Estimating species richness and accumulation by modeling species occurrence and detectability. Ecology, 87(4), 842–854. Dormann, C.F., McPherson, J.M., Araujo, M.B., Bivand, R., Bolliger, J., Carl, G., Davies, R.G., Hirzel, A., Jetz, W., Kissling, W.D., Kuhn, I., Ohlemuller, R., Peres-Neto, P.R., Reineking, B., Schroder, B., Schurr, F.M. & Wilson, R. (2007) Methods to account for spatial autocorrelation in the analysis of species distributional data: a review. Ecography, 30(5), 609–628. Elith, J. & Leathwick, J.R. (2009) Species distribution models: ecological explanation and prediction across space and time. Annual Review of Ecology, Evolution and Systematics, 40, 677–697. ESRI (2005) ArcGIS 9.1. ESRI Inc., Redlands, CA. Farnsworth, G.L., Pollock, K.H., Nichols, J.D., Simons, T.R., Hines, J.E. & Sauer, J.R. (2002) A removal model for estimating detection probabilities from point-count surveys. Auk, 119(2), 414–425. Galatowitsch, S., Frelich, L. & Phillips-Mao, L. (2009) Regional climate change adaptation strategies for biodiversity conservation in a midcontinental region of North America. Biological Conservation, 142(10), 2012–2022. Gu, W.D. & Swihart, R.K. (2004) Absent or undetected? Eﬀects of non-detection of species occurrence on wildlife-habitat models. Biological Conservation, 116(2), 195–203. Hanski, I. (1994) A practical model of metapopulation dynamics. Journal of Animal Ecology, 63(1), 151–162. Hines, J.E., Nichols, J.D., Royle, J.A., MacKenzie, D.I., Gopalaswamy, A.M., Kumar, S. & Karanth, K.U. (2010) Tigers on trails: occupancy modeling for cluster sampling. Ecological Applications, 20, 1456–1466. Howell, J.E., Peterson, J.T. & Conroy, M.J. (2008) Building hierarchical models of avian distributions for the state of Georgia. Journal of Wildlife Management, 72(1), 168–178. Huryn, A.D., Wallace, J.B. & Anderson, N.H. (2008). Habitat, life history, secondary production, and behavioral adaptations of aquatic insects. An Introduction to the Aquatic Insects of North America, 4th edn (eds R.W. Merritt, K.W. Cummins & M.B. Berg), pp. 55–103. Kendall ⁄ Hunt Publishing Company, Dubuque, Iowa. Johnson, D.H. (2008) In defense of indices: the case of bird surveys. Journal of Wildlife Management, 72(4), 857–868.

2011 The Authors. Journal of Applied Ecology 2011 British Ecological Society, Journal of Applied Ecology, 48, 56–66

Hierarchical Bayes multi-scale occupancy 65 Ke´ry, M. & Schmidt, B.R. (2008) Imperfect detection and its consequences for monitoring for conservation. Community Ecology, 9(2), 207–216. Ke´ry, M., Royle, J.A., Plattner, M. & Dorazio, R.M. (2009) Species richness and occupancy estimation in communities subject to temporary emigration. Ecology, 90(5), 1279–1290. King, R.S., Baker, M.E., Whigham, D.F., Weller, D.E., Jordan, T.E., Kazyak, P.F. & Hurd, M.K. (2005) Spatial considerations for linking watershed land cover to ecological indicators in streams. Ecological Applications, 15(1), 137– 153. Klemm, D.J., Blocksom, K.A., Thoeny, W.T., Fulk, F.A., Herlihy, A.T., Kaufmann, P.R. & Cormier, S.M. (2002) Methods development and use of macroinvertebrates as indicators of ecological conditions for streams in the Mid-Atlantic Highlands Region. Environmental Monitoring and Assessment, 78(2), 169–212. Lent, R.A. & Capen, D.E. (1995) Eﬀects of small-scale habitat disturbance on the ecology of breeding birds in a Vermont (USA) Hardwood Forest. Ecography, 18(2), 97–108. MacKenzie, D.I. (2006) Modeling the probability of resource use: the eﬀect of, and dealing with, detecting a species imperfectly. Journal of Wildlife Management, 70(2), 367–374. MacKenzie, D.I., Bailey, L.L. & Nichols, J.D. (2004) Investigating species co-occurrence patterns when species are detected imperfectly. Journal of Animal Ecology, 73(3), 546–555. Mackenzie, D.I. & Royle, J.A. (2005) Designing occupancy studies: general advice and allocating survey eﬀort. Journal of Applied Ecology, 42(6), 1105– 1114. MacKenzie, D.I., Nichols, J.D., Lachman, G.B., Droege, S., Royle, J.A. & Langtimm, C.A. (2002) Estimating site occupancy rates when detection probabilities are less than one. Ecology, 83(8), 2248–2255. MacKenzie, D.I., Nichols, J.D., Hines, J.E., Knutson, M.G. & Franklin, A.B. (2003) Estimating site occupancy, colonization, and local extinction when a species is detected imperfectly. Ecology, 84(8), 2200–2207. Marshall, M.R. & Piekielek, N.B. (2007) Eastern Rivers and Mountains Network Ecological Monitoring Plan. National Park Service, Fort Collins, CO. Matson, T.K., Putland, D.A., Jarman, P.J., Le Roux, J. & Goldizen, A.W. (2007) Inﬂuences of parturition on home range and microhabitat use of female black-faced impalas. Journal of Zoology, 271(3), 318–327. Mattsson, B.J. & Cooper, R.J. (2006) Louisiana Waterthrushes (Seiurus motacilla) and habitat assessments as cost-eﬀective indicators of instream biotic integrity. Freshwater Biology, 51, 1941–1958. Mattsson, B.J. & Marshall, M.R. (2009) Occupancy modeling as a framework for designing avian monitoring programs: a case study along Appalachian streams in southern West Virginia. Proceedings of the Fourth International Partners in Flight Conference (eds T.D. Rich, C.D. Thompson, D. Demarest & C. Arizmendi), pp. 617–632. McAllen, TX. Mattsson, B.J. & Marshall, M.R. (2010) Streamside Bird Monitoring Protocol for the Eastern Rivers and Mountains Network. Technical Report NPS ⁄ NER ⁄ NRTR ⁄ DRAFT. National Park Service, Philadelphia, PA. Mattsson, B.J., Master, R., Mulvihill, S. & Robinson, W.D. (2009) Louisiana Waterthrush (Seiurus motacilla). The Birds of North America. (ed. A. Poole), Cornell Lab of Ornithology, Ithaca, NY. http://bna.birds.cornell. edu/bna/species/ISI. Merritt, R.W., Cummins, K.W. & Berg, M.B. (2008) An Introduction to the Aquatic Insects of North America. 3rd edn. Kendall ⁄ Hunt, Dubuque, Iowa. Michalski, F., Crawshaw, P.G., de Oliveira, T.G. & Fabian, M.E. (2006) Notes on home range and habitat use of three small carnivore species in a disturbed vegetation mosaic of southeastern Brazil. Mammalia, 70(1–2), 52–57. Moran, P.A.P. (1951) A mathematical theory of animal trapping. Biometrika, 38(3–4), 307–311. Mordecai, R.S. (2007) Ecological patterns of occupancy and use: new methods for analysis and applications to species and communities. PhD dissertationUniversity of Georgia, Athens. Mordecai, R.S., Cooper, R.J. & Justicia, R. (2009) A threshold response to habitat disturbance by forest birds in the Choco Andean corridor, Northwest Ecuador. Biodiversity and Conservation, 18, 2421–2431. Moulton, S.R., Carter, J.L., Grotheer, S.A., Cuﬀney, T.F. & Short, T.M. (2000) Methods of analysis by the U.S. Geological Survey National Water Quality Laboratory-processing, taxonomy, and quality control of benthic macroinvertebrate samples. US Geological Survey Open-File Report 00-212, Denver, CO. Moulton, S.R., Kennen, J.G., Goldstein, R.M. & Hancock, J. (2002) Revised protocols for sampling algal, invertebrate, and ﬁsh communities as part of the National Water-Quality Assessment Program. US Geological Survey Open-File Report 02-150, Denver, CO.

Mulvihill, R.S., Newell, F.L. & Latta, S.C. (2008) Eﬀects of acidiﬁcation on the breeding ecology of a stream-dependent songbird, the Louisiana waterthrush (Seiurus motacilla). Freshwater Biology, 53, 2158–2169. Murray, D.L. (2006) On improving telemetry-based survival estimation. Journal of Wildlife Management, 70(6), 1530–1543. Newson, S.E., Evans, K.L., Noble, D.G., Greenwood, J.J.D. & Gaston, K.J. (2008) Use of distance sampling to improve estimates of national population sizes for common and widespread breeding birds in the UK. Journal of Applied Ecology, 45(5), 1330–1338. Nichols, J.D., Bailey, L.L., O’Connell, A.F., Talancy, N.W., Grant, E.H.C., Gilbert, A.T., Annand, E.M., Husband, T.P. & Hines, J.E. (2008) Multiscale occupancy estimation and modelling using multiple detection methods. Journal of Applied Ecology, 45(5), 1321–1329. O’Connell, A.F., Talancy, N.W., Bailey, L.L., Sauer, J.R., Cook, R. & Gilbert, A.T. (2006) Estimating site occupancy and detection probability parameters for meso- and large mammals in a coastal ecosystem. Journal of Wildlife Management, 70(6), 1625–1633. Paciﬁei, K., Simons, T.R. & Pollock, K.H. (2008) Eﬀects of vegetation and background noise on the detection process in auditory avian point-count surveys. Auk, 125, 600–607. Peckarsky, B.L., Fraissinet, P.R., Penton, M.A. & Conklin Jr, D.J. (1990) Freshwater Macroinvertebrates of Northeastern North America. Cornell University Press Ithaca, NY. Pollock, K.H. (1982) A capture-recapture design robust to unequal probability of capture. Journal of Wildlife Management, 46(3), 752–757. Pollock, K.H., Marsh, H., Bailey, L.L., Farnsworth, G.L., Simons, T.R. & Alldredge, M.W. (2004) Separating components of detection probability in abundance estimation: an overview with diverse examples. Sampling Rare and Elusive Species: Concepts, Designs and Techniques for Estimating Population Parameters (ed W.L. Thompson), pp. 43–58. Island Press, Washington DC. R Development Core Team, R. (2006) R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria, ISB, 07-0. Raftery, A.E. & Lewis, S. (1992a) How many iterations in the Gibbs sampler. Bayesian Statistics, 4, 763–773. Raftery, A.E. & Lewis, S.M. (1992b) One long run with diagnostics: Implementation strategies for Markov chain Monte Carlo. Statistical Science, 7, 493– 497. Reynolds, R.T., Scott, J.M. & Nussbaum, R.A. (1980) A variable circular-plot method for estimating bird numbers. Condor, 82(3), 309–313. Riddle, J.D., Mordecai, R.S., Pollock, K.H. & Simons, T.R. (2010) Eﬀects of prior detections on estimates of detection probability, abundance, and occupancy. The Auk, 127, 94–99. Rittenhouse, T.A.G. & Semlitsch, R.D. (2007) Distribution of amphibians in terrestrial habitat surrounding wetlands. Wetlands, 27(1), 153– 161. Rota, C.T., Fletcher, R.J., Dorazio, R.M. & Betts, M.G. (2009) Occupancy estimation and the closure assumption. Journal of Applied Ecology, 46(6), 1173–1181. Roy, A.H., Rosemond, A.D., Leigh, D.S., Paul, M.J. & Wallace, J.B. (2003) Habitat-speciﬁc responses of stream insects to land cover disturbance: biological consequences and monitoring implications. Journal of the North American Benthological Society, 22(2), 292–307. Royle, J.A. & Dorazio, R.M. (2006) Hierarchical models of animal abundance and occurrence. Journal of Agricultural Biological and Environmental Statistics, 11(3), 249–263. Royle, J.A. & Dorazio, R.M. (2008) Hierarchical Modeling and Inference in Ecology: The Analysis of Data from Populations, Metapopulations and Communities. Academic, London, UK. Royle, J.A. & Ke´ry, M. (2007) A Bayesian state-space formulation of dynamic occupancy models. Ecology, 88(7), 1813–1823. Royle, J.A. & Link, W.A. (2006) Generalized site occupancy models allowing for false positive and false negative errors. Ecology, 87(4), 835–841. Royle, J.A., Ke´ry, M., Gautier, R. & Schmid, H. (2007) Hierarchical spatial models of abundance and occurrence from imperfect survey data. Ecological Monographs, 77(3), 465–481. Sauer, J.R., Fallon, J.E. & Johnson, R. (2003) Use of North American breeding bird survey data to estimate population change for bird conservation regions. Journal of Wildlife Management, 67(2), 372–389. Sauer, J.R., Link, W.A. & Royle, J.A. (2005). Hierarchical models and bayesian analysis of bird survey information. Bird Conservation Implementation and Integration in the Americas: Proceedings of the Third International Partners in Flight Conference (eds C.J. Ralph & T.D. Rich), Vol. 2, pp. 762–770. Asilomar, California.

2011 The Authors. Journal of Applied Ecology 2011 British Ecological Society, Journal of Applied Ecology, 48, 56–66

66 R. S. Mordecai et al. Seber, G.A.F. (1982) Estimation of Animal Abundance and Related Parameters. Macmillan, New York. Smith, B.J. (2007) boa: An R package for MCMC output convergence assessment and posterior inference. Journal of Statistical Software, 21(11), 1548– 7660. Spiegelhalter, D.A., Thomas, A., Best, N. & Lunn, D. (2003) WinBUGS User Manual, version 1.4. MRC Biostatistics Unit, Cambridge. Strahler, A.N. (1952) Hypsometric (area-altitude) analysis of erosional topography. Geological Society of America Bulletin, 63(11), 1117. Thompson, K.G. (2007) Use of site occupancy models to estimate prevalence of Myxobolus cerebralis infection in trout. Journal of Aquatic Animal Health, 19(1), 8–13. Tzilkowski, C.J., Weber, A.S. & Ferreri, C.P. (2009) Benthic Macroinvertebrate Monitoring Protocol for Wadeable Streams in the Eastern Rivers and Mountains Network. Natural Resource Report NPS ⁄ ERMN ⁄ NRR-2009 ⁄ XXX. National Park Service, Fort Collins, Colorado. US Geological Survey (2004) National Elevation Dataset. US Geological Survey, Sioux Falls, South Dakota. Williams, B.K., Nichols, J.D. & Conroy, M.J. (2002) Analysis and Management of Animal Populations. Academic Press, Sand Diego, California. Wintle, B.A., Kavanagh, R.P., McCarthy, M.A. & Burgman, M.A. (2005) Estimating and dealing with detectability in occupancy surveys for forest owls and arboreal marsupials. Journal of Wildlife Management, 69(3), 905–917. Zipkin, E.F., Dewan, A. & Royle, J.A. (2009) Impacts of forest fragmentation on species richness: a hierarchical approach to community modelling. Journal of Applied Ecology, 46(4), 815–822.

Received 1 February 2010; accepted 10 November 2010 Handling Editor: Andy Royle

Supporting Information Additional Supporting Information may be found in the online version of this article. Appendix S1. R2WinBUGS code for a simulating data and applying a state-space hierarchical Bayes multi-scale occupancy model for the waterthrush analysis. Appendix S2. WinBUGS speciﬁcation, initial values and data for running the hierarchical Bayes multi-scale occupancy model without covariates. As a service to our authors and readers, this journal provides supporting information supplied by the authors. Such materials may be re-organized for online delivery, but are not copy-edited or typeset. Technical support issues arising from supporting information (other than missing ﬁles) should be addressed to the authors.

2011 The Authors. Journal of Applied Ecology 2011 British Ecological Society, Journal of Applied Ecology, 48, 56–66

Journal of Applied Ecology 2011, 48, 67–75

doi: 10.1111/j.1365-2664.2010.01922.x

Modelling community dynamics based on species-level abundance models from detection ⁄ nondetection data Yuichi Yamaura1*‡, J. Andrew Royle2, Kouji Kuboi3†, Tsuneo Tada3, Susumu Ikeno4 and Shun’ichi Makino1 1

Department of Forest Entomology, Forestry and Forest Products Research Institute, 1 Matsunosato, Tsukuba, Ibaraki 305-8687, Japan; 2U.S. Geological Survey, Patuxent Wildlife Research Center, Laurel, MD 20708, USA; 3 NPO School for Enjoying Nature in Forests, Mt. Sukegawa Preservatin Club, 4-12-18 Nishinarusawa-cho, Hitachi, Ibaraki 316-0032, Japan; and 4Ibaraki Branch, Wild Bird Society of Japan, 925-6 Nakagachichou, Mito, Ibaraki 310-0002, Japan

Summary 1. In large-scale ﬁeld surveys, a binary recording of each species’ detection or nondetection has been increasingly adopted for its simplicity and low cost. Because of the importance of abundance in many studies, it is desirable to obtain inferences about abundance at species-, functional group-, and community-levels from such binary data. 2. We developed a novel hierarchical multi-species abundance model based on species-level detection ⁄ nondetection data. The model accounts for the existence of undetected species, and variability in abundance and detectability among species. Species-level detection ⁄ nondetection is linked to species-level abundance via a detection model that accommodates the expectation that probability of detection (at least one individuals is detected) increases with local abundance of the species. We applied this model to a 9-year dataset composed of the detection ⁄ nondetection of forest birds, at a single post-ﬁre site (from 7 to 15 years after ﬁre) in a montane area of central Japan. The model allocated undetected species into one of the predeﬁned functional groups by assuming a prior distribution on individual group membership. 3. The results suggest that 15–20 species were missed in each year, and that species richness of communities and functional groups did not change with post-ﬁre forest succession. Overall abundance of birds and abundance of functional groups tended to increase over time, although only in the winter, while decreases in detectabilities were observed in several species. 4. Synthesis and applications. Understanding and prediction of large-scale biodiversity dynamics partly hinge on how we can use data eﬀectively. Our hierarchical model for detection ⁄ nondetection data estimates abundance in space ⁄ time at species-, functional group-, and community-levels while accounting for undetected individuals and species. It also permits comparison of multiple communities by many types of abundance-based diversity and similarity measures under imperfect detection. Key-words: data augmentation, detection ⁄ nondetection data, detectability, functional group, hierarchical Bayesian modelling, monitoring, presence ⁄ absence data

Introduction The estimation of abundance is fundamental to ecology (Williams, Nichols & Conroy 2001; Krebs 2009). For example, *Correspondence author. E-mail: [email protected] †Deceased. ‡Present address: Division of Environmental Resources, Graduate School of Agriculture, Hokkaido University, Nishi 9 Kita 9, Kitaku, Sapporo, Hokkaido 060-8589, Japan.

abundance is directly linked to persistence of populations (Hanski 1999) and ecosystem function ⁄ services (Ellison et al. 2005; Gaston & Fuller 2008). However, large-scale surveys are very costly so abundance can be expensive to count. Binary recording of species (species-level detection ⁄ nondetection) could reduce sampling eﬀort at each site (MacKenzie et al. 2006), and yet retain some ability to study temporal population dynamics (Joseph et al. 2006; Pollock 2006). Occupancy estimated from binary data is also useful in metapopulation

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society

68 Y. Yamaura et al. and landscape ecology because theoretical studies use occupancy as a state variable of populations (e.g. Hanski 1999). Collecting binary detection ⁄ nondetection data is thus increasingly adopted in large-scale ﬁeld surveys (Marsh & Trenham 2008). Nevertheless, because of the relevance of abundance in many situations, it is desirable to obtain inferences about abundance from binary data (He & Gaston 2000; Royle & Nichols 2003). Recently, Royle & Nichols (2003) developed a hierarchical model for estimating abundance of single species based on detection ⁄ nondetection data, i.e. the Royle–Nichols (RN) model. In the RN model, species-level detection ⁄ nondetection is linked to abundance via a detection model that accommodates the expectation that detection probability increases with local abundance of the species. On the contrary, Dorazio & Royle (2005) and Dorazio et al. (2006) developed a multi-species occupancy model also based on species-level detection ⁄ nondetection data, i.e. the Dorazio–Royle (DR) model. By accounting for the existence of undetected species and variability in occupancy and detectability among species, the DR model allows for estimation of species richness and community composition at a given site. Here we developed a novel hierarchical multi-species abundance model by combining these two diﬀerent modelling frameworks, i.e. the Royle–Nichols and Dorazio–Royle models. Based on species-level detection ⁄ nondetection data (also called presence ⁄ absence data), our model provides estimates of abundance of each species within communities while accounting for undetected species and variability in abundance and detectability among species. The concept of a functional group or guild, which is a group of species with similar ecological characteristics, is a basic tool in community ecology (Simberloﬀ & Dayan 1991). It is increasingly important in understanding and predicting how biological communities drive ecosystem function ⁄ services (Violle et al. 2007; Hillebrand & Matthiessen 2009), and should be incorporated into multi-species abundance models. In our model, we do this by prescribing a prior distribution for species-level group membership in which group identity is a latent categorical covariate. The development of this model made use of 9 years of bird monitoring data from a single site following a forest ﬁre. Forest ﬁres play a major role in forest ecosystems where they maintain habitat heterogeneity and species diversity at a landscape level (Turner, Gardner & O’Neill 2001). They are widely used in conservation management plans (Lindenmayer & Franklin 2002). Although the eﬀect of ﬁre on birds has been widely studied, most have used a chronosequence approach (but see Smucker, Hutto & Steele 2005; Haney, Apfelbaum & Burris 2009; Jacquet & Prodon 2009) where bird communities in ﬁre and control sites are examined concurrently to identify diﬀerences due to ﬁre. However, the assumption of a space-for-time substitution is not necessarily met due to the existence of confounding factors such as diﬀerences in initial conditions and site history among sites (Johnson & Miyanishi 2008). Therefore, long-term monitoring studies examining the eﬀects of ﬁre are very valuable (Brawn, Robinson & Thompson 2001). Species detection was recorded over a 9-year period following a major (>200 ha) forest ﬁre in Hitachi-city, central Japan

(i.e. years 7–15, post-ﬁre). Sampling included both breeding and wintering seasons, allowing comparison of bird responses between two seasons. Long-term monitoring data after forest ﬁre are rare especially in East Asia, so these data provide a unique opportunity to understand the responses of birds to such events. The data have only a single spatial replicate (n = 1) but have temporal replications (n = 9 years), and possess two characteristics typical in long-term (or broadspatial) data: variable sampling eﬀort and possible changes in detectability among the samples. These factors complicate estimation of the eﬀects of the covariates on populations and communities because, ideally, sampling should be constant across space and time (i.e. the proportionality assumption: Thompson 2002). We develop a hierarchical model that explicitly incorporates detection processes and accommodates variable sampling eﬀort and changes in detectability, and allows inferences to be made about the post-ﬁre changes in bird populations and communities. Several long-term studies from temperate zones (North America: Haney, Apfelbaum & Burris 2009; Europe: Jacquet & Prodon 2009) have shown that bird communities change signiﬁcantly in the ﬁrst 10 years following a forest ﬁre, after which they become relatively stable and approach pre-ﬁre communities. Canopy trees return over this period so the abundance of a functional group feeding in the canopy layer (canopy gleaners) is also expected to increase (Haney, Apfelbaum & Burris 2009). However, the species richness and abundance of communities and other functional groups may not change markedly (Haney, Apfelbaum & Burris 2009; Jacquet & Prodon 2009). Succession could cause not only changes in bird communities but also a decline in the detectability of birds due to the development of habitat structure (Bibby & Buckland 1987; Schieck 1997), in which case we predicted that detectability of many species would decline. We tested these predictions by applying our multi-species abundance model to the 9-year monitoring data set.

Materials and methods THE MODEL

We describe our model using a spatially replicated sampling design as an example, although the model can be easily applied to temporally replicated sampling design by substituting space (e.g. sites) by time (e.g. years). Indeed, we apply this model to temporal dynamics in a bird community in a single site.

Detection process model We assume a sampling design in which a number of spatial sample units – ‘sites’ – are visited multiple times, allowing for inferences to be made about species- and community-level state variables (e.g. occupancy, abundance and species richness) while accounting for imperfect detection (Royle & Dorazio 2008). We utilized binary recording in the ﬁeld surveys to obtain encounter histories for each species. This gave a series of detection ⁄ nondetection in the visitations to each site, e.g. [1 0 1 0 1] for a certain species during ﬁve visits to a site. We compiled these encounter histories into a dataset of the number of visits in

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 67–75

Multi-species abundance model 69 which species i are detected (yij) out of total visits (vj) for site j. Here vj needs to be larger than 1, but could vary among sites. A key element of our model is that we link these binary detection ⁄ nondetection recordings of species to species-level abundance, following Royle & Nichols (2003). If individuals of species i in site j are detected independently of one another with probability rij, then site- and species-speciﬁc detection probability is pij ¼ 1 ð1 rij ÞZij :

eqn 1

where pij is the ‘net’ probability of detection of species i, i.e. Pr(at least 1 individual is detected), and Zij is abundance of species i in site j that may be the object of inference either directly or indirectly. We assume that yij follows a binomial distribution in the form yij Binomial(vj, pij). Detectability could also diﬀer among sites, depending on site-speciﬁc relevant covariates, which can also aﬀect the abundance of each species (Bibby & Buckland 1987; Schieck 1997). In such cases, we must consider the dependency of the detection probability on the covariates (Ke´ry 2008): logitðrij Þ ¼ a0i þ xj ai

eqn 2

where xj are measured covariates for site j, rij is individual-level detection probability of species i in site j, and a0i (intercept) and ai (slope) are the parameters to be estimated for species i. Because detection processes are modelled separately from ecological (i.e. abundance) processes, changes in abundance would not be confounded with changes in detectability in our model (Ke´ry 2008).

Ecological process model The abundance of species i, Zij, could vary in space depending on covariates in the form Zij Poisson (kij) with logðkij Þ ¼ b0i þ xj bi

eqn 3

where b0i and bi are the parameters to be estimated for species i. Here b0i (intercept) is a random species eﬀect. If there are no covariates, i.e. if the right hand side of eqn 3 has only b0i, the model allows for heterogeneity only in base-line abundance among species. Deviations from such base-line abundance, which depend on covariates of each site, are modelled by xhj bhi (h = 1, 2, …, H). Note that variation in abundance might not be fully captured by measured covariates and the Poisson assumption (Royle & Dorazio 2008). To overcome this potential overdispersion, we can include (spatially structured) random site eﬀects into the right hand side of the eqn 3 (Royle et al. 2007; Ke´ry et al. 2009). If this model is applied to a temporally replicated design, it can be extended to include dynamic eﬀects, i.e. to incorporate recruitment and mortality rates, which can be approximated by linear or quadratic time trends in this equation (Ke´ry et al. 2009; Russell et al. 2009).

MODELLING VARIATIONS IN PARAMETERS AMONG SPECIES

Although parameters of well-detected species could be independently estimated, there are far too many parameters to be precisely estimated for rare species (and some species will not be detected at all). Hence, we add to the model an additional hierarchical layer and assume that parameters (e.g. ai, and bi) are independent normal random eﬀects, each governed by community-level hyper-parameters (Ke´ry & Royle 2009). For example, we assume that bi Normal(lb, rb2) where lb is

the community mean response (across species) to a covariate and rb is the standard deviation (among species). In sum, the hyper-parameters are simply the mean and variance for each covariate as measured across species. Using such hierarchical models we can estimate species- and community-level state variables more precisely, especially for rarely detected species (Zipkin, DeWan & Royle 2009; Zipkin et al. 2010), which is often referred to colloquially as ‘borrowing strength’, but species-level estimates (random eﬀects) are pulled in (‘shrunk’) towards the community-level means (Sauer & Link 2002; Ke´ry 2010). Parameters that describe variation among species – i.e. community-level hyperparameters – are relatively well-estimated because the model eﬀectively aggregates data from among species (Link 1999; Sauer & Link 2002). On the contrary, the precision of species-level estimates are highly variable depending on the sample sizes and are relatively more inﬂuenced by the hierarchical model structure.

BAYESIAN ANALYSIS BY DATA AUGMENTATION

Let N denote the unknown number of distinct species in the sampled community and let n denote the total number of species that are detected through the sampling across sites (i.e. n £ N). We cannot know how many species are undetected by sampling, i.e. number of undetected species (N – n). As such, we prescribe a prior distribution for N and it is estimated along with the other model parameters by Markov chain Monte Carlo (MCMC) using a method based on data augmentation (Royle, Dorazio & Link 2007). As described above, let yi = (yi1, yi2, …, ynj) denote a vector of the j site-speciﬁc binary observations of species i (number of visits in which species i are detected in site j). We create a supercommunity of species, one that comprises the n detected species and an arbitrarily large, but known, number of undetected species (potential species) for which yi = 0 (i = n + 1, n + 2, …, N, N + 1, …, M). The supercommunity size M is ﬁxed. Formally, data augmentation arises under a uniform prior distribution for the community size N where M is the upper bound for the uniform prior. A useful fact associated with the uniform prior is that it can be constructed ‘hierarchically’ by introducing a set of latent indicator variables, say wi, which takes the value 1 if a species in the supercommunity is a member of the community exposed to the sampling and 0 otherwise. The uniform prior for N arises by assuming that wi are independent, Bernoulli-distributed random variables indexed by parameter X: wi Bernoulli[X]; X Uniform(0,1). We P can estimate N as a derived parameter, by calculating M i¼1 wi . If the assigned value of M is too low, the posterior distribution of X will be concentrated near the upper limit of its support, and N would be underestimated. Conversely, if M is too large, the computational burden increases (Royle & Dorazio 2008). Under data augmentation in our multi-species abundance model, the observations for which wi = 0 are ‘structural’ or ﬁxed 0s. This model admits that the number of visits in which species i is detected (yij) of the species exposed to the sampling (i.e. wi = 1) can be more than 0 depending on covariates, whereas those of the species unexposed to the sampling (i.e. wi = 0) is always 0, that is, Pr(yi = 0) = 1. This is a form of zero-inﬂation model for the augmented data set, and the model is a mixture of stochastic and structural zeros. To formulate this in the model, we modify eqn 1 so that Zijwi rather than Zij appears in the model for encounter probability. We can estimate site-speciﬁc species richness and total abundance by P counting species with Zijwi > 0, and calculating M i¼1 Zij wi , respectively. We calculated the posterior distributions of these derived parameters by MCMC.

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 67–75

70 Y. Yamaura et al. We extended the model to include a species-level functional group attribute, as a discrete, latent variable, so that undetected species could be allocated to prescribed functional groups. To formalize this in the model, we introduce a species-level covariate, Gi, the group membership of species i, which is unobserved for the undetected species. We assume that the group membership probabilities have a conventional non-informative Dirichlet prior distribution (Spiegelhalter et al. 2003; McCarthy 2007) on functional group membership. Speciﬁcally, we assume that Gi Categorical(propk) where propk is the proportion of species in group k (k = 1, 2, …, K). We used the gamma distribution to construct the Dirichlet prior (Spiegelhalter et al. 2003): gk PK Gamma(1,1); propk ¼ gk k¼1 gk . The parameters gk are also estimated by MCMC along with the other parameters of the model. We can estimate species richness of each functional group P by calculating M i¼1 wi only for members of each group. We can also estimate site-speciﬁc species richness and abundance of each functional group by counting members with Zijwi > 0, and calP culating M i¼1 Zij wi only for the constituent species, respectively. MODEL APPLICATION

We applied our model to 9-years of bird monitoring data after a forest ﬁre at a single site (N36 35¢ 32¢¢, E140 37¢ 20¢¢) by substituting time (e.g. years) for space (e.g. sites). The monitoring site was located in a montane area in Hitachi-city, Ibaraki prefecture, central Japan. The montane area dominates the western area of the city, whereas industrial and residential area dominates eastern area. The ﬁre burnt 218 ha on 7–8 March 1991. The ﬁre was a stand-replacing ﬁre because many canopy trees were burned and became snags, and shrubs and forest ﬂoor vegetation was burnt out (Appendix S1, Supporting information). The dominant species before the ﬁre were planted Japanese cedar Cryptomeria japonica and red pine Pinus densiﬂora, and natural broadleaved species. A single observer (Kouji Kuboi) walked similar routes (c. 3Æ5 km) in the area, and recorded species detection from January 1998 to November 2006 (1176 sampling visits; 7–15 years post-ﬁre). Routes were composed of three sections. The second section (c. 600 m) was a single segment, whereas ﬁrst and third sections were composed of 2–3 segments within 500 m each other. Although the observer arbitrarily selected segments in the ﬁrst and third sections (and selected routes were not recorded), the total lengths were similar (diﬀerences were <200 m). The three sections of the routes were sub-samples aggregated into single encounter histories in a single route (site) to form species-level encounter histories. That is, a species was recorded as encountered if it was observed on any section of the route. Data were pooled within breeding (May–July) and wintering (December–February) seasons separately in each year so that the data are composed of total detections out of number of visits for each species in each season (Appendix S2, Supporting information). Stand level covariates such as stand height were not recorded. In each season, we allocated each species exclusively to one of the seven functional groups (or guilds) based on their feeding ecology and habitat associations following published literature (Amano & Yamaura 2007; Yamaura et al. 2009) and expert knowledge. For the wintering season birds were classiﬁed as bush user, canopy gleaner, edge species, ﬂoor user, open land species, seed eater, and stem prober; in the breeding season we classiﬁed birds as air searcher, bush user, canopy gleaner, edge species, ﬂycatcher, open land species, and stem prober. We excluded water birds in both seasons, and in the breeding season we excluded species that did not breed in the study area (i.e. transients) from the analyses. The data set used for the anal-

ysis contained 47 and 34 bird species in the wintering and breeding seasons, respectively (Appendix S2). Using the model described above, we estimated temporal dynamics of species richness and abundance within communities and functional groups, and abundance of each species allowing for undetected species. Here we assumed that our data were independent between years because our primary objective was the development and application of the simple model. We augmented the data set with 100 potential species in both seasons. Median estimates of inclusion rate (X) were lower than 1 in both seasons (<0Æ45), indicating that the number of augmented species was suﬃciently large (Royle & Dorazio 2008). Because Jacquet & Prodon (2009) showed that bird abundances could change nonlinearly during the 7–15 post-ﬁre years, we modelled abundance response of each species to year since burning (Yearj; covariate) quadratically: logðkij Þ ¼ b0i þ b1i Yearj þ b2i Year2j :

eqn 4

As succession progresses, the vegetation structure will become more complex and bird detectability will decrease (Bibby & Buckland 1987; Schieck 1997). Therefore, we also assumed that detectability of each species could change quadratically among years: logitðrij Þ ¼ a0i þ a1i Yearj þ a2i Year2j :

eqn 5

Year, which took values 1–9, was standardized to a mean 0 and a SD of 1Æ0. We allocated each of the undetected species into one of the seven functional groups using a categorical distribution with uninformative Dirichlet prior. We used customary vague priors that reﬂect ignorance about the parameter values: X Uniform(0,1); lb0 ; lb1 ; lb2 Normal(0,1000); la0 ; la1 ; la2 Normal(0,1000). We used uniform distribution for the standard deviations (Gelman 2006): rb0 ; rb1 ; rb2 Uniform(0,10); ra0 ; ra1 ; ra2 Uniform(0,10). We used WinBUGS Ver. 1Æ4Æ3 (Lunn et al. 2000) and R2WinBUGS R package Ver. 2Æ1–16 (Sturtz, Ligges & Gelman 2005) to ﬁt the model using MCMC. We ran three chains of 88 000 iterations with diﬀerent initial values, discarded the ﬁrst 8000 and thinned by 20, which resulted in 12 000 iterations used for inference. Model convergence was assessed with R^ values (the Gelman–Rubin statistic), and R^ of all above described parameters were less than 1Æ12, indicating that our model convergence was good (Gelman & Hill 2007).

Results We recorded 47 and 34 bird species post-ﬁre in the wintering and breeding seasons, respectively. The posterior median speP cies richness ( M i¼1 wi ) was 57 (95% credible interval [CI]: 48– 109) and 55 (36–127) in each season over the whole nine years. This suggests that c. 10 (95% CI: 1–62) and 21 (2–93) species were not detected in each season throughout the survey. Consistent with previous studies (Haney, Apfelbaum & Burris 2009; Jacquet & Prodon 2009), there were no distinct post-ﬁre annual changes in estimated species richness of communities or functional groups in both seasons (Fig. 1, Appendix S3, Supporting Information). The estimated species richness of communities suggests that 15–20 species were missed in each year. However, the estimated number of missed species varied among functional groups. While most species would have been detected in the bush user functional group in the breeding

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 67–75

Wintering season

Breeding season

All species

All species 120

120 100 80 60 40 20 0

28 54 42 46 30 49 35 53 40 8

10

12

14

Est. species rich.

Est. species rich.

Multi-species abundance model 71

100 80 60 40 20 0

16

37 27 39 32 24 41 44 14 17 8

Year after fire

Bush user

14

16

Bush user 30

Est. species rich.

Est. species rich.

12

Year after fire

30 25 20 15 10 5

25 20 15 10 5 0

0 8

10

12

14

8

16

Year after fire

10

12

14

16

Year after fire

Canopy gleaner

Canopy gleaner 30

Est. species rich.

30

Est. species rich.

10

25 20 15 10 5

25 20 15 10 5 0

0 8

10

12

14

16

Year after fire

8

10

12

14

16

Year after fire

Fig. 1. Estimated annual changes in species richness of bird communities and two functional groups. Left and right columns show the results of wintering and breeding seasons, respectively. Grey lines show observed species richness in each year. Black solid and dotted lines show predicted values of median and 95% credible intervals (CIs), respectively. Numbers above the x axis show the number of visits in each year. Forest ﬁre occurred in 1991, which is here treated as year 1. Because upper limits of CI in canopy gleaners in the breeding season had large values (42–43), this line was not included in the ﬁgure. We did not directly estimate community-level state variables (e.g. species richness ⁄ abundance of communities ⁄ functional groups); rather, these are derived parameters that are functions of species-level abundance.

season, 5–15 species would have been consistently missed in canopy gleaners and edge species in both seasons (see also Appendix S3). In the winter, the abundance of canopy gleaners showed a weak increasing trend, which was consistent with our prediction (Fig. 2, Appendix S3). The abundance of communities (all species) and many functional groups also showed increasing trends. The abundances of constituent species also showed an increasing trend in the winter (e.g. Japanese bush warbler Cettia diphone for bush user; great tit Parus major for canopy gleaner), indicating that changes in their abundance contributed to the increases in the abundance of communities and functional groups (Fig. 3, Appendix S3). In contrast, few annual changes in the abundance of communities and functional groups were observed in the breed-

ing season, and few species showed annual changes in their abundance (Fig. 3, Appendix S3). Annual changes in detectability varied among species. The detectabilities of many species were quite low and unchanged across years, and these species were rarely detected in the survey (Appendices S1 and S3). Brown-eared bulbul Hypsipetes amaurotis was the only species whose detectability increased with years in both seasons (Appendix S3). However, as predicted, the detectability of many well-detected species decreased with year since ﬁre, particularly in the later years (Fig. 3, Appendix S3). For example, detection probabilities of the black-faced bunting Emberiza spodocephala, Japanese bush warbler, and the Siberian meadow bunting Emberiza cioides decreased as a function of time in the wintering, breeding, and both seasons, respectively.

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 67–75

72 Y. Yamaura et al.

Breeding season

Wintering season All species

All species 1000

Est. abundance

Est. abundance

1000 800 600 400 200 0

800 600 400 200 0

8

10

12

14

16

8

Bush user

14

16

200

Est. abundance

Est. abundance

12

Bush user

200 150 100 50 0

150 100 50 0

8

10

12

14

16

8

10

12

14

16

Year after fire

Year after fire

Canopy gleaner

Canopy gleaner 200

Est. abundance

200

Est. abundance

10

Year after fire

Year after fire

150 100 50 0

150 100 50 0

8

10

12

14

16

Year after fire

8

10

12

14

16

Year after fire

Fig. 2. Estimated annual changes in abundance of bird communities and two functional groups. Because upper limits of credible interval (CI) in canopy gleaners in the breeding season had large values (>250), this line was not included.

Discussion It is important to combine ecological and detection processes in statistical models because ignoring imperfect detection (e.g. false absence) could lead not only to underestimation of occupancy ⁄ abundance (MacKenzie et al. 2006; McCarthy 2007; Royle & Dorazio 2008), but also to a misunderstanding of the eﬀects of covariates (Tyre et al. 2003; Mazerolle, Desrochers & Rochefort 2005) and undesirable management outcomes (Yoccoz, Nichols & Boulinier 2001). Here we develop a method to examine the eﬀects of covariates on population and community structure while accounting for imperfect detection of individuals. We build on earlier development of such models (e.g. Ke´ry & Royle 2008; Russell et al. 2009; Zipkin, DeWan & Royle 2009); speciﬁcally, our model is novel in that it is the ﬁrst community-level abundance model based on species-level binary detection ⁄ nondetection data that accommodates the existence of undetected species (Appendix S4, Supporting Information). Consideration of abun-

dance and undetected species in community models is important because abundance is a fundamental state variable in basic and applied ecology, and undetected species commonly occur in most practical community sampling scenarios. The model also accommodates functional group structure by assuming a prior distribution of functional group membership and estimating the parameters of that distribution. We applied this model to bird monitoring data collected after a forest ﬁre. The results suggested that 10 (95% CI: 1–62) and 21 (2–93) species remained undetected throughout the survey in the wintering and breeding seasons, respectively, and 15–20 species were also missed in each year. Estimates of the number of undetected species throughout the survey were similar to those derived from the regional species pool. Systematic survey of birds in Ibaraki prefecture suggests that 5–10 and 21–29 undetected species could inhabit the area in each season (Ibaraki Branch 2008; S. I. person. comm.). This suggests that our abundance model may yield good inferences about the sampled community. The estimated number of undetected

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 67–75

Multi-species abundance model 73

Wintering season

Breeding season

Japanese bush warbler

Japanese bush warbler

0·6 10

0·4

5

0·2

0

0·0 12

14

0·6 10

0·4

5

0·2

0

0·0 8

16

Year after fires

10

Great tit

15

0·6

10

0·4

5

0·2 0·0

0 12

14

1·0 0·8

15

0·6 10

0·4

5

0·2 0·0

0

16

8

Year after fires

15

0·6

10

0·4

5

0·2

0

0·0 16

Year after fires

Est. abundance

0·8

14

14

16

1·0

20

Est. detectability

Est. abundance

20

12

12

Siberian meadow bunting 1·0

10

10

Year after fires

Siberian meadow bunting

8

16

20

Est. abundance

0·8

Est. detectability

Est. abundance

20

10

14

Great tit 1·0

8

12

Year after fires

Est. detectability

10

0·8

15

0·8

15

0·6 10

0·4

5

0·2

Est. detectability

8

Est. abundance

15

1·0

20

Est. detectability

Est. abundance

0·8

Est. detectability

1·0

20

0·0

0 8

10

12

14

16

Year after fires

Fig. 3. Estimated annual changes in abundance of three bird species. Japanese bush warbler, great tit, and Siberian meadow bunting were representative species of bush user, canopy gleaner, and edge species, respectively. Black lines show predicted abundances indexed by left-side ordinate. Thin grey lines shows expected abundances (kij) obtained by eqn 4 and median estimated parameters (bi0, bi1, and bi2). Thick grey lines show expected individual-level detectabilities (rij) obtained by eqn 5 and median estimated parameters (ai0, ai1, and ai2) indexed by right-side ordinate. Diﬀerences in predicted and expected values are due to the fact that predicted values represent a compromise between expected values and observations, which were not strictly consistent with the quadratic models.

species diﬀered among functional groups. We found few annual changes in the estimated species richness of communities and functional groups in both seasons, which is consistent with previous studies (Haney, Apfelbaum & Burris 2009; Jacquet & Prodon 2009). As predicted, the abundance of canopy gleaners showed a trend towards a slight increase across years. The abundance of communities, many functional groups and species also showed consistent increasing trends, but only in the wintering season. Foraging resources, including those in canopy layers, would have increased as succession progressed. In contrast, we found few annual changes in species abundances during the breeding season. Under conditions of severe weather and limited food resources, habitat and landscape structure in winter could have large eﬀects on bird populations and communities (e.g. Doherty & Grubb 2002; Johnson et al. 2006) with further eﬀects in the subsequent breeding season (Greenberg & Marra 2005). Bird–habitat relationships in post-ﬁre sites may also be important in winter. Previous studies have shown that bird communities are relatively stable after the ﬁrst 10 years post-ﬁre in the United

States and Europe (Haney, Apfelbaum & Burris 2009; Jacquet & Prodon 2009). Although we found that bird responses to the year since burning diﬀered between seasons, our study in East Asia conﬁrmed this by accounting for detectabilities over this period. The ﬁrst 5–10 years after a forest ﬁre are the most dynamic, during which open-land species (granivores), ﬂycatchers, and woodpeckers colonize, and some of them desert the site (Brawn, Robinson & Thompson 2001; Jacquet & Prodon 2009). It is thought that disturbed ⁄ early successional habitats and the species inhabiting them are now declining in many countries (Askins 2001; Yamaura et al. 2009), therefore future studies should examine the responses of birds to such short-lived disturbed habitats at multiple scales, preferably using complementary chronosequence and long-term (revisiting) approaches (e.g. Donner, Ribic & Probst 2010). In our model, we assumed the unit of detection was a single individual of a species (Zij = number of individuals). However, many bird species form ﬂocks in winter, for which the observers detect groups of birds rather than single individuals (Zij = number of ‘functional individuals’). When estimating abundance of communities and ⁄ or comparing abundance

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 67–75

74 Y. Yamaura et al. among species, group size should be considered whenever possible. However, obtaining explicit information about group size entails additional sampling protocol modiﬁcations and also additional model assumptions. The development of methods that can accommodate group size would be worthwhile. Rarely detected species had low detectabilities, and the credible intervals (CIs) of community-level state variables (i.e. species richness and abundance of communities and functional groups) were very wide. These results are attributable to the low densities of many species in the community, factors that cause low detectability (e.g. topographic complexity along the census route, poor census ability of the observer), short length of the census route, and the limited amount of information provided by binary detection ⁄ nondetection data. In addition, our model is highly parameterized relative to the sparse data available; however, this is partly a consequence of biological considerations. In order to accommodate temporal variability in the parameters in the model, it is necessary to consider heterogeneity in model parameters among species. In our case, we considered a quadratic trend containing species-level linear and quadratic parameters. For many species, these specieslevel parameters are poorly informed by the sparseness of available data. Indeed, narrower CIs could have been achieved if we assumed constant detectability across years, i.e. by using ri rather than rij (results not shown). Such a model might be reasonable in a standard design where spatial units with similar habitat structure are used as replicates, and sampled within the same year. However, in our case, such a model could not account for the decreases in detectabilities in the later years, and estimated bird abundances showed slight decreases in that period especially in the winter. If we use count data rather than binary detection ⁄ nondetection, sparse data availability would be partially resolved. For example, in each site, we may obtain count data such as counts of individuals in the series of the visits, e.g. [2 1 3 0 2] (Ke´ry, Royle & Schmid 2005). Let yijt (t = 1, 2, …, T) be independent counts of individuals of species i made at sites j = 1, 2, …, J so that yijt BinomialðZij ; rij Þ

eqn 6

where Zij and rij is abundance and individual-level detectability of species i in site j, respectively. Development of multi-species models for this binomial observation model is ongoing.

Conclusion Because our model estimates the abundance of each species, we can compare multiple communities by many types of abundance-based diversity and similarity measures under imperfect detection (e.g. Shannon diversity index, Euclidean distance and v2 metric: Legendre & Legendre 1998). A further extension is the estimation of a, b, and c diversity. Using the additive partitioning framework (Lande 1996; Veech et al. 2002), the estimated species richness of the sampled community (N) could be deﬁned as c diversity (regional diversity), while average site-speciﬁc species rich-

ness could be deﬁned as a diversity (local diversity). Diﬀerences between these two diversity measures could be an estimate of b diversity (among sites diversity). Collection of detection ⁄ nondetection data for species is increasingly adopted in large-scale ﬁeld surveys. Development, use, and improvement of the statistical models for such simple data should be encouraged. Our hierarchical multi-species model allows explicit modelling and estimation of abundance in space ⁄ time at species-, functional group- and communitylevels accounting for undetected individuals and species.

Acknowledgements Suggestions from two anonymous reviewers and the editor greatly improved our manuscript. We are grateful to M. Ke´ry, R. Russell, and E. F. Zipkin for their helpful reviews of a draft of this manuscript. We also thanks for T. Amano, K. Matsumoto, Y. Mitsuda, H. Taki, K. Okabe, and S. Sugiura for useful suggestions for this study. Y. Yamaura was supported by Research grant No. 200802 of the Forestry and Forest Products Research Institute, and JSPS KAKENHI (Grand-in-Aid for JSPS Fellows No. 21-7033).

References Amano, T. & Yamaura, Y. (2007) Ecological and life-history traits related to range contractions among breeding birds in Japan. Biological Conservation, 137, 271–282. Askins, R.A. (2001) Sustaining biological diversity in early successional communities: the challenge of managing unpopular habitats. Wildlife Society Bulletin, 29, 407–412. Bibby, C.J. & Buckland, S.T. (1987) Bias of bird census results due to detectability varying with habitat. Acta Oecologica, 8, 103–112. Brawn, J.D., Robinson, S.K. & Thompson, F.R. III. (2001) The role of disturbance in the ecology and conservation of birds. Annual Review of Ecology and Systematics, 32, 251–276. Doherty, P.F.J. & Grubb, T.C.J. (2002) Survivorship of permanent-resident birds in a fragmented forested landscape. Ecology, 83, 844–857. Donner, D.M., Ribic, C.A. & Probst, J.R. (2010) Patch dynamics and the timing of colonization-abandonment events by male Kirtland’s Warblers in an early succession habitat. Biological Conservation, 143, 1159–1167. Dorazio, R.M. & Royle, J.A. (2005) Estimating size and composition of biological communities by modeling the occurrence of species. Journal of the American Statistical Association, 100, 389–398. Dorazio, R.M., Royle, J.A., So¨derstro¨m, B. & Glimska¨r, A. (2006) Estimating species richness and accumulation by modeling species occurrence and detectability. Ecology, 87, 842–854. Ellison, A.M., Bank, M.S., Clinton, B.D., Colburn, E.A., Elliott, K., Fort, C.R. & Foster, D.R. (2005) Loss of foundation species: consequences for the structure and dynamics of forested ecosystems. Frontiers in Ecology and Environment, 3, 479–486. Gaston, K.J. & Fuller, R.A. (2008) Commonness, population depletion and conservation biology. Trends in Ecology and Evolution, 23, 14–19. Gelman, A. (2006) Prior distributions for variance parameters in hierarchical models. Bayesian Analysis, 1, 513–534. Gelman, A. & Hill, J. (2007) Data Analysis Using Regression and Multilevel ⁄ hierarchical Models. Cambridge University Press, Cambridge. Greenberg, R. & Marra, P.P. (2005) Birds of Two Worlds: The Ecology and Evolution of Migration. Johns Hopkins University Press, Baltimore. Haney, A., Apfelbaum, S. & Burris, J.M. (2009) Thirty years of post-ﬁre succession in a southern boreal forest bird community. American Midland Naturalist, 159, 421–433. Hanski, I. (1999) Metapopulation Ecology. Oxford University Press, Oxford. He, F. & Gaston, K.J. (2000) Estimating species abundance from occurrence. American Naturalist, 156, 553–559. Hillebrand, H. & Matthiessen, B. (2009) Biodiversity in a complex world: consolidation and progress in functional biodiversity resarch. Ecology Letters, 12, 1405–1419. Ibaraki Branch (2008) Report on the Distribution of Wild Birds and Mammals in Ibaraki Prefecture (2006, 2007). Ibaraki Branch, Wild Bird Society of Japan, Mito.

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 67–75

Multi-species abundance model 75 Jacquet, K. & Prodon, R. (2009) Measuring the postﬁre resilience of a bird–vegetation system: a 28-year study in a Mediterranean oak woodland. Oecologia, 161, 801–811. Johnson, E.A. & Miyanishi, K. (2008) Testing the assumptions of chronosequences in succession. Ecology Letters, 11, 419–431. Johnson, M.D., Sherry, T.W., Holmes, R.T. & Marra, P.P. (2006) Assessing habitat quality for a migratory songbird wintering in natural and agricultural habitats. Conservation Biology, 20, 1433–1444. Joseph, L.N., Field, S.A., Wilcox, C. & Possingham, H.P. (2006) Presenceabsence versus abundance data for monitoring threatened species. Conservation Biology, 20, 1679–1687. Ke´ry, M. (2008) Estimating abundance from bird counts: binomial mixture models uncover complex covariate relationships. Auk, 125, 336–345. Ke´ry, M. (2010) Introduction to WinBUGS for Ecologists: A Bayesian Approach to Regression, ANOVA, Mixed Models and Related Analysis. Academic Press, San Diego. Ke´ry, M. & Royle, J.A. (2008) Hierarchical Bayes estimation of species richness and occupancy in spatially replicated surveys. Journal of Applied Ecology, 45, 589–598. Ke´ry, M. & Royle, J.A. (2009) Inference about species richness and community structure using species-speciﬁc occupancy models in the national Swiss breeding bird survey MHB. Modeling Demographic Processes in Marked Populations (eds D.L. Thomson, E.G. Cooch & M.J. Conroy), pp. 639–656. Springer, New York. Ke´ry, M., Royle, J.A. & Schmid, H. (2005) Modeling avian abundance from replicated counts using binomial mixture models. Ecological Applications, 15, 1450–1461. Ke´ry, M., Dorazio, R.M., Soldaat, L., van Strien, A., Zuiderwijk, A. & Royle, J.A. (2009) Trend estimation in populations with imperfect detection. Journal of Applied Ecology, 46, 1163–1172. Krebs, C.J. (2009) Ecology: The Experimental Analysis of Distribution and Abundance. Pearson Education, Chollerstrasse. Lande, R. (1996) Statistics and partitioning of species diversity, and similarity among multiple communities. Oikos, 76, 5–13. Legendre, P. & Legendre, L. (1998) Numerical Ecology, 2nd edn. Elsevier, Amsterdam. Lindenmayer, D.B. & Franklin, J.F. (2002) Conserving Forest Biodiversity: A Comprehensive Multiscaled Approach. Island Press, Washington, DC. Link, W.A. (1999) Modeling pattern in collections of parameters. Journal of Wildlife Management, 63, 1017–1027. Lunn, D.J., Thomas, A., Best, N. & Spiegelhalter, D. (2000) WinBUGS – a Bayesian modelling framework: concepts, structure, and extensibility. Statistics and Computing, 10, 325–337. MacKenzie, D.I., Nichols, J.D., Royle, J.A., Pollock, K.H., Bailey, L.L. & Hines, J.E. (2006) Occupancy Estimation and Modeling: Inferring Patterns and Dynamics of Species Occurrence. Academic Press, Amsterdam. Marsh, D.M. & Trenham, P.C. (2008) Current trends in plant and animal population monitoring. Conservation Biology, 22, 647–655. Mazerolle, M.J., Desrochers, A. & Rochefort, L. (2005) Landscape characteristics inﬂuence pond occupancy by frogs after accounting for detectability. Ecological Applications, 15, 824–834. McCarthy, M.A. (2007) Bayesian Methods for Ecology. Cambridge University Press, Cambridge. Pollock, J.F. (2006) Detecting population declines over large areas with presence-absence, time-to-encounter, and count survey methods. Conservation Biology, 20, 882–892. Royle, J.A. & Dorazio, R.M. (2008) Hierarchical Modeling and Inference in Ecology: The Analysis of Data From Populations, Metapopulations and Communities. Academic Press, Amsterdam. Royle, J.A., Dorazio, R.M. & Link, W.A. (2007) Analysis of multinomial models with unknown index using data augmentation. Journal of Computational and Graphical Statistics, 16, 67–85. Royle, J.A. & Nichols, J.D. (2003) Estimating abundance from repeated presence-absence data or point counts. Ecology, 84, 777–790. Royle, J.A., Ke´ry, M., Gautier, R. & Schmid, H. (2007) Hierarchical spatial models of abundance and occurrence from imperfect survey data. Ecological Monographs, 77, 465–481. Russell, R.E., Royle, J.A., Saab, V.A., Lehmkuhl, J.F., Block, W.M. & Sauer, J.R. (2009) Modeling the eﬀects of environmental disturbance on wildlife communities: avian responses to prescribed ﬁre. Ecological Applications, 19, 1253–1263. Sauer, J.R. & Link, W.A. (2002) Hierarchical modeling of population stability and species group attributes from survey data. Ecology, 86, 1743–1751.

Schieck, J. (1997) Biased detection of bird vocalizations aﬀects comparisons of bird abundance among forested habitats. Condor, 99, 179–190. Simberloﬀ, D. & Dayan, T. (1991) The guild concept and the structure of ecological communities. Annual Review of Ecology and Systematics, 22, 115– 143. Smucker, K.M., Hutto, R.L. & Steele, B.M. (2005) Changes in bird abundance after wildﬁre: importance of ﬁre severity and time since ﬁre. Ecological Applications, 15, 1535–1549. Spiegelhalter, D., Thomas, A., Best, N. & Lunn, D. (2003) WinBUGS User Manual. Ver. 1.4. MRC Biostatistics Unit, Institute of Public Health, Cambridge. Sturtz, S., Ligges, U. & Gelman, A. (2005) R2WinBUGS: a package for running WinBUGS from R. Journal of Statistical Software, 12, 1–16. Thompson, W.L. (2002) Towards reliable bird surveys: accounting for individuals present but not detected. Auk, 119, 18–25. Turner, M.G., Gardner, R.H. & O’Neill, R.V. (2001) Landscape Ecology in Theory and Practice: Pattern and Process. Springer, New York. Tyre, A.J., Tenhumberg, B., Field, S.A., Niejalke, D., Parris, K. & Possingham, H.P. (2003) Improving precision and reducing bias in biological surveys: estimating false-negative error rates. Ecological Applications, 13, 1790–1801. Veech, J.A., Summerville, K.S., Crist, T.O. & Gering, J.C. (2002) The additive partitioning of species diversity: recent revival of an old idea. Oikos, 99, 3–9. Violle, C., Navas, M.-L., Vile, D., Kazakou, E., Fortunel, C., Hummel, I. & Garnier, E. (2007) Let the concept of trait be functional! Oikos, 116, 882– 892. Williams, B.K., Nichols, J.D. & Conroy, M.J. (2001) Analysis and Management of Animal Populations. Academic Press, San Diego. Yamaura, Y., Amano, T., Koizumi, T., Mitsuda, Y., Taki, H. & Okabe, K. (2009) Does land-use change aﬀect biodiversity dynamics at a macroecological scale? A case study of birds over the past 20 years in Japan. Animal Conservation, 12, 110–119. Yoccoz, N.G., Nichols, J.D. & Boulinier, T. (2001) Monitoring of biological diversity in space and time. Trends in Ecology and Evolution, 16, 446–453. Zipkin, E.F., DeWan, A. & Royle, J.A. (2009) Impacts of forest fragmentation on species richness: a hierarchical approach to community modelling. Journal of Applied Ecology, 46, 815–822. Zipkin, E.F., Royle, J.A., Dawson, D.K. & Bates, S. (2010) Multi-species occurrence models to evaluate the eﬀects of conservation and management actions. Biological Conservation, 143, 479–484. Received 5 June 2010; accepted 10 November 2010 Handling Editor: Marc Cadotte

Supporting information Additional Supporting Information may be found in the online version of this article. Appendix S1. Some pictures of the study area. Appendix S2. Encounter histories of forest birds and guild classiﬁcation. Appendix S3. Annual changes in birds. Appendix S4. Model categorization by inference targets and state variables. Appendix S5. R code and required data ﬁles to conduct our multi-species abundance model. As a service to our authors and readers, this journal provides supporting information supplied by the authors. Such materials may be re-organized for online delivery, but are not copy-edited or typeset. Technical support issues arising from supporting information (other than missing ﬁles) should be addressed to the authors.

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 67–75

Journal of Applied Ecology 2011, 48, 3–8

doi: 10.1111/j.1365-2664.2010.01929.x

FORUM

Translating research into action; bumblebee conservation as a case study Dave Goulson1,2*, Pippa Rayner1, Bob Dawson1 and Ben Darvill1 1

Bumblebee Conservation Trust, University of Stirling, Stirling, FK9 4LA, UK; and 2School of Biological and Environmental Sciences, University of Stirling, Stirling, FK9 4LA, UK

Key-words: agri-environment sheme, Bombus, Fabaceae, forage use, habitat restoration, pollination, population decline, raising awareness, species-rich grasslands

Introduction Bumblebees belong to the genus Bombus, which comprises about 250 species, largely conﬁned to the temperate Northern Hemisphere. They are wholly dependent on ﬂowers for their energetic and developmental requirements. Most are social species, with nest sizes varying from 50 to 400 workers. As such, they have attracted considerable attention regarding their role as pollinators. There is a growing body of evidence that bumblebees have declined in Europe, North America and Asia in recent decades because of multiple causes probably including habitat loss, impacts of pesticides, competition from non-native species and the introduction of non-native diseases (Goulson, Lye & Darvill 2008a; Williams & Osborne 2009). Recent health problems aﬀecting honeybees and a perception that other pollinators may be declining has led to serious concern that we might be facing a global ‘pollination crisis’ aﬀecting pollination of crops and wildﬂowers (e.g. Aizen & Harder 2009). The global value of crop pollination by bumblebees is unknown; Gallai et al. (2009) estimate that for the EU25 countries in 2005, the value of insect pollination of agricultural crops was €14Æ2 billion, with a large (but unquantiﬁed) proportion of this coming from bumblebees. Most crop pollination delivered by bumblebees is because of a handful of common species, so that from an economic viewpoint there may be no need to conserve a diversity of bumblebee species. However, bumblebees also provide pollination services to natural ecosystems, with numerous wild plant species largely or entirely dependent on bumblebees for pollination. As bumblebee species each occupy distinct (albeit often overlapping) niches with regard to their patterns of ﬂoral visitation (e.g. Goulson, Lye & Darvill 2008b), it is probable that many bumblebee species are needed to maintain functionality of natural ecosystems (Williams & Osborne 2009). Bumblebee nests also support a diversity of parasitic and commensal organisms. For these reasons, it can be argued that bumblebees are ‘keystone species’, *Correspondence author. E-mail: [email protected]

upon which the survival of many other organisms depend (Goulson, Lye & Darvill 2008a). Perhaps as a result of perceived declines, academic interest in bumblebees has risen markedly in recent decades. This can be simply illustrated by plotting the number of papers in Thomson’s ISI Web of Knowledge which have Bombus in the abstract or key words (Fig. 1). Over the last 20 years the number of papers published per year has grown steadily from 12 to 144, a 12-fold increase (for comparison, studies concerning two other pollinator groups, Lepidoptera or Syrphidae, have each increased by a factors of c. 2Æ5 over the same period, Fig. 1). The studies of bumblebees encompass diverse topics from ‘pure’ research of, for example, social structure, foraging behaviour, population genetics, pheromones and navigation, to applied studies addressing how particular land management methods inﬂuence bumblebee numbers. There is no doubt that we understand far more about the biology of bumblebees than we once did, although there remains much more to learn (for example mating behaviour of many species has rarely been seen, and because natural nests are hard to ﬁnd we know little about the factors aﬀecting their survival and success). Recent papers on bumblebees (and many grant applications) often start by summarizing evidence for bumblebee declines, the implication being that the research may contribute to our understanding of the causes of decline and so help us to reverse them. However, publishing a paper, no matter how good the science may be, does not in itself improve the fortunes of a single bumblebee. It is only when the research reaches the right audiences and is translated into practical action that it makes any diﬀerence. Very few farmers, gardeners, politicians or nature reserve wardens sit down of an evening to read a scientiﬁc journal, nor should we expect them to. If they did, they might struggle to make sense of most of it. Academics must take some of the blame for this situation; many researchers make little eﬀort to communicate their work beyond the traditional use of scientiﬁc journals, publications which are all but incomprehensible to the layman. This in turn is largely because the traditional criteria used for judging academic success (publications and grant income) pay little attention to the

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society

4 D. Goulson et al. 180 Bombus 160

Lepidoptera/10 Syrphidae

Publications per year

140 120 100 80 60 40 20 0 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010

Year

Fig. 1. The number of publications in Thomson’s ISI Web of Knowledge which have Bombus, Lepidoptera or Syrphidae in the abstract, key words or title, plotted against year of publication. Numbers for Lepidoptera are divided by 10 for ease of comparison.

impact of the research. In some areas of science the breakdown of communication between scientists and the public may not be too disastrous, but in conservation this matters profoundly, because if conservation research is not communicated to those who might implement it, then the research eﬀort (and funds) were wasted. Yet there remains a yawning gulf between the research consensus and practical on-the-ground habitat management, and it is not clear whose job it is to bridge this gap. This applies both at the level of rare species conservation (arguably the territory of conservation NGOs) and at the more basic level of maintaining healthy pollinator populations. The Bumblebee Conservation Trust (BBCT, http:// www.bumblebeeconservation.org) was founded in the UK in 2006 by academics (Goulson & Darvill), with the speciﬁc aim of linking science and practice by translating our growing understanding of bumblebee ecology into well-informed practical conservation measures, which would halt and hopefully reverse declines in bumblebee populations. It provides an example of one possible route by which conservation research can usefully feed directly into on-the-ground conservation. We discuss, in turn: What recent research has revealed about the conservation needs of bumblebees; how this has fed into the strategy of BBCT; current knowledge gaps with regard to bumblebee ecology of relevance to conservation; and barriers facing the more widespread implementation of research in conservation.

What recent research has revealed about the conservation needs of bumblebees? Six key points have emerged in recent years: 1. Population size. Recent studies have demonstrated that the social nature of bumblebees renders them particularly sensitive to habitat fragmentation. Their eﬀective population size is c. 1Æ5 times the number of successful nests, for each nest contains just one breeding female and the sperm she has

stored from a (single) haploid male. Although some species remain widespread and hence have large populations, others thrive only in areas containing high densities of their favourite forage plants. Because of habitat loss, mainly attributable to agricultural intensiﬁcation, these ﬂower-rich areas are now often small and fragmented. Most nature reserves in the UK might only support a handful of nests of a rare bumblebee species, and are thus far too small to support viable populations. For example, Ellis et al. (2006) estimated that surviving UK populations of the rare Bombus sylvarum contained between 26 and 48 nests, and that the remaining populations are isolated from one another. Such tiny populations are unlikely to be viable in the long term, and it seems likely that a breakdown of metapopulation structure has already led to the UK extinction of B. subterraneus (Goulson, Lye & Darvill 2008a). Genetic studies have demonstrated that the rarer species such as B. sylvarum, B. muscorum and B. distinguendus are genetically depauperate compared to more common species (e.g. Ellis et al. 2006). There is a real risk that surviving populations of rare species will disappear in the near future because of stochastic eﬀects, inbreeding, or both. 2. Dispersal abilities. Recent research has revealed marked diﬀerences in the dispersal ranges of the sexual stages of bumblebees, suggesting that some bumblebee groups such as the subgenus Pyrobombus (which includes B. pratorum, B. jonellus, B. hypnorum and B. monticola) may have relatively high dispersal abilities. In contrast the Thoracobombus group seem to be relatively sedentary (e.g. Darvill et al. 2010); this group includes B. pascuorum, B. sylvarum, B. humilis, B. rudararius and B. muscorum, of which all but B. pascuorum have undergone marked declines. Diﬀerences in dispersal ability dictate the scale of habitat fragmentation which an individual species can withstand. 3. Foraging range. The foraging range of worker bees determines the area which a nest can exploit. It is hard to quantify, and there have been many attempts, but it seems highly likely that there are important diﬀerences between species, with foraging ranges varying from c. 400 m to 1Æ5 km (e.g. Osborne et al. 2008a). Species with long foraging ranges such as B. terrestris will be able to cope with more patchy availability of ﬂoral resources than those with shorter foraging ranges (thought to include B. pascuorum, B. muscorum and B. sylvarum). Bumblebees do not store large quantities of nectar or pollen which means temporal and spatial patchiness in local forage availability is more diﬃcult for species with short foraging ranges to withstand. 4. Forage use. Studies suggest that bumblebees do not require a high ﬂoral diversity to survive (e.g. Carvell et al. 2007). There is high dietary overlap between species, but most bumblebee species could be catered for throughout most of the season by providing a plentiful supply of 10 or fewer suitable plants. Fabaceae appear to be very important in providing protein-rich pollen for bumblebees, and it is likely that the large scale loss of species-rich grasslands and clover leys (habitats characterized by high densities of Fabaceae) are primary drivers of bumblebee declines in Europe (Goulson, Lye & Darvill 2008a). Trifolium pratense and its close relatives

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 3–8

Bumblebee conservation as a case study 5 appear to be particularly important sources of both nectar and pollen for many long-tongued bumblebee species such as B. distinguendus, the UK’s rarest species and a UK Biodiversity Action Plan priority species. Nevertheless, where available, bumblebees are known to forage from (and pollinate) a great diversity of ﬂowering plants. Hence, bumblebees beneﬁt from a diverse ﬂora, and vice versa. 5. The value of agri-environment schemes. Simple agri-environment scheme options are available in many EU countries and the USA; for example pollen and nectar strips in ﬁeld margins can be very eﬀective in providing forage for bumblebees in the UK (Carvell et al. 2007). Schemes promoting the restoration and creation of species-rich grasslands also have the potential to greatly beneﬁt bumblebees. There is now a wealth of information on how best to establish and manage both short and long-term pollen and nectar strips or speciesrich grasslands (e.g. Pywell et al. 2002; Carvell et al. 2007). However, even in non-competitive, entry-level schemes uptake of these options can be poor. In total only 6000 ha of pollen and nectar mix has been sown in England, which has a total area of over 13 million ha (0Æ05%). Farmland areas under agri-environment schemes but not incorporating these targeted options are often no better for bumblebees than conventional farmland (e.g. Lye et al. 2009), indicating the need for interventionist measures where these are not currently available. If food prices rise in the future, as seems probable, agri-environment schemes may become less attractive to farmers, and in Europe, future Rural Development Programme funding could have a signiﬁcant impact on the menu of fundable measures. 6. Urban areas. It has become clear that both gardens and brownﬁeld sites in urban areas support higher densities of bumblebees and bumblebee nests than do typical farmed areas (Osborne et al. 2008b). In some cases these urban areas can also support rare and declining species (e.g. B. sylvarum and B. humilis in the Thames Estuary). Brownﬁeld sites are often under threat as legislation prioritizes them for development.

How this research has fed into the strategy of BBCT A number of practical messages have emerged from bumblebee research. From our growing knowledge of bumblebee population structure, it is clear that conservation measures need to focus on enhancing the size and connectivity of extant populations of the rare species, which otherwise are likely to go extinct one by one. Nature reserves are too small to support viable populations, so conservation measures need to target the wider countryside, i.e. farmland. Stepping-stone habitat is needed to link existing populations, and this is likely to be particularly critical for the more sedentary species. Studies of foraging range suggest that some species forage much further aﬁeld than others. For the species with shorter foraging ranges, nests will only survive if there are patches of suitable ﬂowers available through the season within a c. 400 m radius of the nest. However, even for the less mobile species,

patches of ﬂoral resources clearly need not be contiguous and could readily be incorporated into most farming systems as patches of ﬂowers interspersed among much larger areas of crops. Suitable conservation measures include: (i) maintenance of ﬂower-rich sites; (ii) restoration of species-rich grasslands; (iii) sowing pollen and nectar mixes; (iv) encouraging clover ley crops and a return to crop rotations as an alternative to the use of fertilizers; (v) promoting wildlife-friendly gardening. As bumblebees are found throughout the UK, these activities have some value wherever they take place; as a minimum they will help to boost populations of the common species. However, to conserve rare species and prevent further bumblebee extinctions, activities need to be targeted at appropriate sites close to or within areas where rare species persist. It is clearly not possible for a small NGO to buy and manage suﬃcient land to make any signiﬁcant impact on bumblebee populations at a national scale, so the challenge is to persuade land owners and managers to change their practices. The key stakeholders here are farmers, local councils, gardeners, and those involved in the management of nature reserves, national parks and other protected areas. Land management can also be improved indirectly by inﬂuencing government policy. To reach this diverse and substantial group of people is a considerable task for a small organization. The Trust has adopted a range of strategies to achieve its aims: 1. Raising awareness. Where stakeholders are numerous and diverse the simplest way to reach them is through popular media. The Trust seeks to engage with and educate the general public as to the importance of bumblebees and how they can be helped. If even a small proportion of farmers, gardeners, and other land managers can be inﬂuenced then diﬀuse eﬀects might be achieved across large areas. Awareness-raising has been achieved so far through: articles in numerous popular media, including national newspapers, radio and television; development of a primary-school education pack; dissemination of information through our newsletter which goes to the 7000 trust members; and setting up ‘citizen science’ recording schemes which encourage members of the public to photograph or identify the bees in their local area and send in records. 2. Targeted habitat management to support species recovery. The Trust has dedicated conservation oﬃcers for the UK’s two most threatened bumblebee species, B. sylvarum and B. distinguendus. Their role is to promote favourable management and habitat restoration in areas within a 10 km radius of populations of these species. This ranges from fundraising for speciﬁc grassland restoration projects, to encouraging farmers to enter appropriate agri-environment schemes. An important component of their work is carried out in partnership with other NGOs. 3. A reintroduction programme for the extinct B. subterraneus. This species was last recorded in the UK in 1988, but stock of UK origin persists in New Zealand, to which they were introduced over 120 years ago. Reintroduction attempts are often rightly criticized as requiring substantial resources with limited likelihood of success and rather narrow biodiver-

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 3–8

6 D. Goulson et al. sity beneﬁts, when the organism to be reintroduced could often be conserved much more cheaply in the places where it survives. However, in this instance the species is a non-native alien in New Zealand so it will not receive aid by conservationists. More importantly, the habitat creation and restoration work being carried out in South East England in preparation for the reintroduction is beneﬁtting several other endangered bumblebee species which still persist in the area, and much else besides. In a collaborative project with other stakeholders and numerous local landowners and farmers, several hundred hectares of species-rich grassland have been created and managed for biodiversity. 4. Collaboration. Many other organizations are involved in conservation of bumblebees and their habitats, and it makes sense wherever possible to work together and avoid duplication of eﬀort. For example in the UK a number of organizations have overlapping remits: in addition to BBCT, there is Buglife (a charity devoted to invertebrate conservation with an extensive record of successful policy level engagement, http://www.buglife.org.uk), Hymettus Ltd (an organization providing advice and expertise relating to the conservation of bees, wasps and ants, http://www.hymettus.org.uk), and the Bees, Wasps & Ants Recording Society (who monitoring the changing distributions of species, http://www.bwars.com). A key driver of bumblebee declines is loss of species-rich grasslands, and creation and restoration of this habitat are clear mechanisms for reversing declines. Species-rich grasslands are important habitats for numerous other organisms, so this aspect of bumblebee conservation is eminently suitable for collaborative projects with other conservation NGOs. It is notable that, at present, most conservation NGOs work primarily at a national level, and there is scope for more collaboration between similar organizations in diﬀerent countries [for example between the Xerces Society in the United States (a not-for-proﬁt organization devoted to the conservation of invertebrates, http://www.xerces.org) and invertebrate conservation organizations in Europe]. It seems probable that these organizations could learn much from one another’s successes and failures. 5. Promoting wildlife gardening. Gardens cover c. 1 million ha in the UK. Members of the public can get directly involved in bumblebee conservation by planting appropriate ﬂowers in their garden. Thus far the Trust has distributed >20 000 packets of wildﬂower seeds, has produced and sold 8000 copies of a booklet Gardening for Bumblebees, has run stands at various national ﬂower shows, and is currently collaborating with a large garden centre chain to develop and promote a range of bumblebee-friendly plants for the garden. It must be noted that the approaches described here for conserving bumblebees would not be possible with many less endearing organisms. Bumblebees are large, colourful, and furry; they have media appeal. In contrast, most invertebrates and lower plants would be much harder to sell to the general public. Even in the UK where interest in natural history is high, and there appears to be an expert for any taxon, however humble, it would probably be impossible to attract suﬃcient members to provide adequate core income for a charity for the

conservation of, say, nematodes or true bugs. However, as bumblebee conservation requires conservation of highly biodiverse habitats such as species-rich grasslands, they can usefully act as umbrella species for large numbers of less charismatic organisms, including a diversity of other pollinators and economically beneﬁcial species.

Current knowledge gaps with regard to bumblebee ecology of relevance to conservation Some aspects of the ecology and conservation of bumblebees remain poorly understood, and urgently require research. In particular, we need information on the following areas if we are to design appropriate mitigation ⁄ conservation strategies: 1. We currently have no data on population trajectories of either common or rare bumblebee species. BBCT are in the process of setting up a UK-wide transect recording scheme, ‘Beewalks’, modelled along the lines of the very successful butterﬂy monitoring scheme, which will begin to address this problem for the UK, but similar schemes are needed elsewhere. 2. At present there is little knowledge as to the impacts of pesticides on bumblebees, although among the non-scientiﬁc community this is a topic of great interest and much speculation. In particular, the possible role of neonicotinoids in causing bee mortality has received considerable media attention but few hard data are available. Sublethal eﬀects of pesticides, such as impairment of learning ability which might lead to drastic eﬀects at the colony level, have rarely been investigated. 3. The possible impacts of the global trade in commercial bumblebee nests include competition with native species, hybridization with native species, and accidental spread of pathogens, but these subjects remain poorly researched (reviewed by Goulson 2003; see also Ings, Ward & Chittka 2006). Non-native bumblebees are now established in the wild in many parts of the world (e.g. Chile, Japan, Tasmania) but their likely long-term impacts are not yet known. The relative importance of pathogens as causes of mortality in wild bumblebee populations, and the role of commercial bumblebees in spreading pathogens is poorly understood, although there is evidence that the accidental introduction of a non-native pathogen to North America with commercial bees may have caused catastrophic declines in some native bumblebee species (Winter et al. 2006). We know very little about which viruses infect bumblebees, although evidence suggests that honeybees and bumblebees share some viruses. 4. There is growing concern amongst practitioners that entry-level agri-environment schemes oﬀer little concrete beneﬁt to biodiversity, even where a diverse menu of measures is available. Because of limited funding, farmers have to compete for entry into higher level schemes, so uptake of these schemes is inevitably low. In addition, there is often inadequate targeting with respect to the biodiversity which might be present in a particular locality. Also, measures are not always successfully implemented. Some schemes might bene-

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 3–8

Bumblebee conservation as a case study 7 ﬁt a small number of species or particular taxonomic groups but not others. There is a clear need to establish which of the existing schemes ⁄ options are eﬀective and for what, to establish why implementation is often poor and to balance the conservation requirements of diﬀerent organisms (Kleijn et al. 2006). Given the substantial level of public subsidy for agriculture in Europe, there is an urgent need, but limited capacity, to achieve progress towards eﬀective, balanced delivery of public goods from a productive agricultural sector, before the next 7-year funding programme of rural development in the EU is ﬁnalized. 5. We do not as yet know if wildﬂowers or crops routinely suﬀer from pollination shortages, or conversely whether pollinator populations remain largely adequate. This remains a fundamental question of huge ecological and economic signiﬁcance. Recent and high proﬁle health problems with honeybees highlight the danger inherent in relying heavily on a single species for pollination of many crops, and suggest that a less risky approach would be to maintain a diverse natural pollinator population in addition to domesticated bees. If clear evidence could be found that pollinator populations were now so low in agricultural landscapes that crop yields were beginning to be depressed then this would provide a powerful lever to gain funding for pollinator conservation. It would likewise provide an important indicator that conservation of biodiversity requires implementation of measures to support pollinators. 6. Bumblebees are comparatively well-studied in the UK, Western Europe, Japan and to some extent North America but little is known about the distribution, ecology or conservation status of the majority of species which live elsewhere. Citizen Science schemes, such as those operated by BBCT in the UK and the ‘Beespotter’ programme run by the University of Illinois can provide eﬀective means of gathering these data.

Barriers facing the more widespread implementation of research in conservation For bumblebees, considerable progress has been made in transferring scientiﬁc knowledge into practical conservation, but the gulf between evidence and practice remains in some areas, particularly with regard to policy. A major problem in the UK and elsewhere is that no clear mechanism exists for translating scientiﬁc evidence into governmental policy. There is little discourse between governmental organizations responsible for conservation and academics carrying out conservation-related research. Decision-making with regard to policies aﬀecting conservation (including agri-environment schemes) is not transparent. Any academic wishing to have an input into conservation policy would be hard put to identify a mechanism by which to do so. Similarly, small conservation bodies such as BBCT struggle to have their voice heard. Conservation policy tends to reﬂect the popularity of the respective taxa and the resultant lobbying power of attendant NGOs, but also those taxa that are relatively simple to monitor, such as plants, birds and butterﬂies, because these provide

indicators of long-term change that oﬀer powerful reporting and lobbying tools. That policy decisions are weighted towards certain taxonomic groups, such as birds, is therefore no surprise given their popular appeal and the capacity for dedicated, research, policy, advocacy and advisory skills of associated individual NGOs and the BirdLife partner network. Bees, and other pollinators, have yet to make a similar impact, although that remains a clear aspiration, given their signiﬁcant economic and ecological importance. It is our view that national governments should do much more to ensure not only that that conservation policy is based on scientiﬁc evidence, but that policy is not unduly biased towards conservation of a small number of vertebrate species and instead reﬂects a balanced approach to conservation of biodiversity and ecosystem function. There are also issues with the targeting of conservation action. Particular agri-environment schemes such as pollen and nectar strips might be highly beneﬁcial for boosting populations of rare bumblebee species, but only if they are implemented in locations where these rare species are likely to occur. Many farmers have no idea whether they have rare bumblebee species in their locality. Of course this applies to all taxonomic groups. There is considerable scope for improving the value obtained from agri-environment schemes by better targeting of schemes to appropriate areas according to the rare species present, but this requires coordination of knowledge of species distributions, decisions over which species or taxonomic groups to prioritize in each area, and then communication of this information to farmers. At present, large sums of taxpayers’ money are spent in the EU on both ecological research and on conservation (through agri-environment schemes and funds for governmental agencies), yet biodiversity continues to decline according to most measures and our environment is in a parlous state. Signatories to the International Convention on Biodiversity pledged ‘to achieve by 2010 a signiﬁcant reduction of the current rate of biodiversity loss at the global, regional and national level’. It is notable that very few of the speciﬁc targets agreed in this convention have been met. One might reasonably argue that taxpayers are getting poor value for money at present, and that this could be greatly improved by involving researchers in discussions over environmental priorities and policy (e.g. see Sutherland et al. 2010). With a little more joined-up thinking and appropriate use of existing scientiﬁc evidence when designing conservation strategies, public money could be spent more wisely and result in much greater beneﬁts to biodiversity and the environment.

References Aizen, M.A. & Harder, L.D. (2009) The global stock of domesticated honey bees is growing slower than agricultural demand for pollination. Current Biology, 19, 915–918. Carvell, C., Meek, W.R., Pywell, R.F., Goulson, D. & Nowakowski, M. (2007) Comparing the eﬃciency of agri-environment schemes to enhance bumble bee abundance and diversity on arable ﬁeld margins. Journal of Applied Ecology, 44, 29–40.

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 3–8

8 D. Goulson et al. Darvill, B., O’Connor, S., Lye, G.C., Lepais, O. & Goulson, D. (2010) Cryptic diﬀerences in dispersal lead to diﬀerential sensitivity to habitat fragmentation in two bumblebee species. Molecular Ecology, 19, 53–63. Ellis, J.S., Knight, M.E., Darvill, B. & Goulson, D. (2006) Extremely low eﬀective population sizes, genetic structuring and reduced genetic diversity in a threatened bumblebee species, Bombus sylvarum (Hymenoptera: Apidae). Molecular Ecology, 15, 4375–4386. Gallai, N., Salles, J., Settele, J. & Vaissie`re, B.E. (2009) Economic valuation of the vulnerability of world agriculture confronted with pollinator decline. Ecological Economics, 68, 810–821. Goulson, D. (2003) Eﬀects of introduced bees on native ecosystems. Annual Review of Ecology and Systematics, 34, 1–26. Goulson, D., Lye, G.C. & Darvill, B. (2008a) Decline and conservation of bumblebees. Annual Review of Entomology, 53, 191–208. Goulson, D., Lye, G.C. & Darvill, B. (2008b) Diet breadth, coexistence and rarity in bumblebees. Biodiversity and Conservation, 17, 3269–3288. Ings, T.C., Ward, N.L. & Chittka, L. (2006) Can commercially imported bumble bees out-compete their native conspeciﬁcs? Journal of Applied Ecology, 43, 940–948. Kleijn, D., Baquero, R.A., Clough, Y., Diaz, M., De Esteban, J., Fernandez, F., Gabriel, D., Herzog, F., Holzschuh, A., Johl, R., Knop, E., Kruess, A., Marshall, E.J.P., Steﬀan-Dewenter, I., Tscharntke, T., Verhulst, J., West, T.M. & Yela, J.L. (2006) Mixed biodiversity beneﬁts of agri-environment schemes in ﬁve European countries. Ecology Letters, 9, 243–254. Lye, G.C., Park, K., Osborne, J.L., Holland, J. & Goulson, D. (2009) Assessing the value of Rural Stewardship schemes for providing foraging resources and nesting habitat for bumblebee queens (Hymenoptera: Apidae). Biological Conservation, 142, 2023–2032. Osborne, J.L., Martin, A.P., Carreck, N.L., Swain, J.L., Knight, M.E., Goulson, D., Hale, R.J. & Sanderson, R.A. (2008a) Bumblebee ﬂight distances in relation to the forage landscape. Journal of Animal Ecology, 77, 401–415. Osborne, J.L., Martin, A.P., Shortall, C.R., Todd, A.D., Goulson, D., Knight, M.E., Hale, R.J. & Sanderson, R.A. (2008b) Quantifying and comparing bumblebee nest densities in gardens and countryside habitats. Journal of Applied Ecology, 45, 784–792. Pywell, R.F., Bullock, J.M., Hopkins, A., Walker, K.J., Sparks, T.H., Burke, M.J.W. & Peel, S. (2002) Restoration of species-rich grassland on arable land: assessing the limiting processes using a multi-site experiment. Journal of Applied Ecology, 39, 294–309. Sutherland, W.J., Albon, S.D., Allison, H., Armstrong-Brown, S., Bailey, M.J., Brereton, T., Boyd, I.L., Carey, P., Edwards, J., Gill, M., Hill, D., Hodge, I., Hunt, A.J., Le Quesne, W.J.F., Macdonald, D.W., Mee, L.D., Mitchell, R., Norman, T., Owen, R.P., Parker, D., Prior, S.V., Pullin, A.S., Rands, M.R.W., Redpath, S., Spencer, J., Spray, C.J., Thomas, C.D., Tucker, G.M., Watkinson, A.R. & Clements, A. (2010) The identiﬁcation of

priority policy options for UK nature conservation. Journal of Applied Ecology, 47, 955–965. Williams, P.H. & Osborne, J.L. (2009) Bumblebee vulnerability and conservation world-wide. Apidologie, 40, 367–387. Winter, K., Adams, L., Thorp, R., Inouye, D., Day, L., Ascher, J. & Buchmann, S. (2006) Importation of non-native bumble bees into North America: potential consequences of using Bombus terrestris and other non-native bumble bees for greenhouse crop pollination in Canada, Mexico and the United States. A white paper of the North American Pollinator Protection Campaign (NAPPC). Received 20 September 2010; accepted 16 November 2010 Handling Editor: Jane Memmott

Biosketch Dave Goulson is an academic specializing in studies of insect ecology and conservation, with a particular focus on bumblebees. Ben Darvill did his PhD with Goulson on the population genetics of rare bumblebees, followed by a postdoc on the factors aﬀecting populations of common bumblebees in arable ecosystems. Together, Goulson and Darvill founded the Bumblebee Conservation Trust in 2006 at the University of Stirling, where its Head Oﬃce remains. Darvill now works for the BBCT as Development Manager and Ecologist. Bob Dawson is Conservation Oﬃcer for Scotland for BBCT. His previous research concerned bird systematics and genetics before working on a captive breeding project in Morocco. He then worked for RSPB, carrying out ecological research and co-ordinating the Volunteer and Farmer Alliance in South & West Scotland. Pippa Rayner is Conservation Oﬃcer for England & Wales with the Bumblebee Conservation Trust. She did her PhD on creation and management of species-rich grasslands and then worked on restoring upland hay meadows in the Yorkshire Dales before bringing her expertise to BBCT in 2009.

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 3–8