Developments in Atmospheric Science 24 A P P R O A C H E S TO SCALING OF TRACE GAS FLUXES IN ECOSYSTEMS
Further titles in this series
1. 2. 3. 4. 5. 6.
7. 8. 9. 10. 11. 12. 13. 14. 15.
16. 17. 18. 19. 20. 21.
22. 23.
F. VERNIANI (Editor) - Structure and Dynamics of the Upper Atmosphere E.E. GOSSARD and W.H. HOOKE- Waves in the Atmosphere L.P. SMITH - Methods in Agricultural Meteorology O. ESSENWANGER- Applied Statistics in Atmospheric Science G.W. PALTRIDGE and C.M.R. PLATT - Radiative Processes in Meteorology and Climatology P. S C H W E R D T F E G E R - Physical Principles of Micro-Meteorological Measurements S. TWOMEY- Atmospheric Aerosols E. FUKUI (Editor) - The Climate of Japan A.L. FYMAT and V.E. ZUEV (Editors) - Remote Sensing of the Atmosphere: Inversion Methods and Applications W. BACH, J. PANKRATH and W. KELLOGG (Editors) - Man's Impact on Climate A. LONGHETTO (Editor) - Atmospheric Planetry Boundary Layer Physics C. MAGONO- Thunderstorms S. IKEDA, E. SUZUKI, E. UCHIDA and M.M. YOSHINO (Editors) - Statistical Climatology E.E. GOSSARD and R.G. STRAUCH - Radar Observation of Clear Air and Clouds W. S C H W E R D T F E G E R - Weather and Climate of the Antartic A.L. BERGER and C. NICOLIS (Editors) - New Perspectives in Climate Modelling R.W. PREISENDORFER and C.D. MOBLEY - Principal Component Analysis in Meteorology and Oceanography D. MARTYN - Climates of the World M.E. SCHLESINGER (Editor) - Greenhouse-Gas-Induced Climatic Change: A Critical Appraisal of Simulations and Observations R. REITER- Phenomena in Atmospheric and Environmental Electricity S.G. D O B R O L O V S K I - Global Climatic Changes in Water and Heat TransferAccumulation Processes B.BURMAN and L.O. POCHOP- Evaporation, Evapotransportation and Climatic Data D. CAMUFFO - Microclimate for Cultural Heritage
Developments in Atmospheric Science 24
A P P R O A C H E S TO SCALING OF TRACE GAS FLUXES IN ECOSYSTEMS Edited by
A.F. B O U W M A N National Institute of Public Health and Environment Protection, P.O. Box 1, 3720 BA Bilthoven, The Netherlands
Report of the workshop
"Scaling of trace gas fluxes between terrestrial and aquatic ecosystems and the atmosphere" Organized
by
International Soil Reference and Information Centre (ISRIC)
ELSEVIER AMSTERDAM
- LAUSANNE
- NEW YORK - OXFORD - SHANNON - SINGAPORE - TOKYO
ELSEVIER SCIENCE B.V. Sara Burgerhartstraat 25 P.O. Box 211, 1000 AE Amsterdam, The Netherlands
9 1998 Elsevier Science B.V. All rights reserved.
This work is protected under copyright by Elsevier Science, and the following terms and conditions apply to its use: Photocopying Single photocopies of single chapters may be made for personal use as allowed by national copyright laws. Permission of the publisher and payment of a fee is required for all other photocopying, including multiple or systematic copying, copying for advertising or promotional purposes, resale, and all forms of document delivery. Special rates are available for educational institutions that wish to make photocopies for non-profit educational classroom use. Permissions may be sought directly from Elsevier Science Rights & Permissions Department, PO Box 800, Oxford OX5 1DX, UK; phone: (+44) 1865 843830, fax: (+44) 1865 853333, e-mail:
[email protected]. You may also contact Rights & Permissions directly through Elsevier's home page (http://www.elsevier.nl), selecting first 'Customer Support', then 'General Information', then 'Permissions Query Form'. In the USA, users may clear permissions and make payments through the Copyright Clearance Center, Inc., 222 Rosewood Drive, Danvers, MA 01923, USA; phone: (978) 7508400, fax: (978) 7504744, and in the UK through the Copyright Licensing Agency Rapid Clearance Service (CLARCS), 90 Tottenham Court Road, London W1P 0LP, UK; phone: (+44) 171 436 5931; fax: (+44) 171 436 3986. Other countries may have a local reprographic rights agency for payments. Derivative Works Tables of contents may be reproduced for internal circulation, but permission of Elsevier Science is required for external resale or distribution of such material. Permission of the publisher is required for all other derivative works, including compilations and translations. Electronic Storage or Usage Permission of the publisher is required to store or use electronically any material contained in this work, including any chapter or part of a chapter. Contact the publisher at the address indicated. Except as outlined above, no part of this work may be reproduced, stored in a retrieval system or transmitted in any form or by any means, electronic, mechanical, photocopying, recording or otherwise, without prior written permission of the publisher. Address permissions requests to: Elsevier Science Rights & Permissions Department, at the mail, fax and e-mail addresses noted above. Notice No responsibility is assumed by the Publisher for any injury and/or damage to persons or property as a matter of products liability, negligence or otherwise, or from any use or operation of any methods, products, instructions or ideas contained in the material herein. Because of rapid advances in the medical sciences, in particular, independent verification of diagnoses and drug dosages should be made.
First edition 1999 Library of Congress Cataloging in Publication Data A catalog record from the Library of Congress has been applied for.
ISBN: 0-444-82934-2
@ The paper used in this publication meets the requirements of ANSI/NISO Z39.48-1992 (Permanence of Paper). Printed in The Netherlands.
TABLE OF CONTENTS
Foreword
vii
Acknowledgements
ix
Towards reliable global bottom-up estimates of temporal and spatial patterns of emissions of trace gases and aerosols from land-use related and natural sources A.F. Bouwman, R.G. Derwent and F.J. Dentener
Methods for stable gas flux determination in aquatic and terrestrial systems
27
R.L. Lapitan, R. Wanninkhof and A.R. Mosier
Some recent developments in trace gas flux measurement techniques
67
O.T. Denmead, R. Leuning, D. W.T. Griffith and C.P. Meyer
Working group report: How can fluxes of trace gases be validated between different scales?
85
W.A.H. Asman, M.O. Andreae, R. Conr:ld, O.T. Denmead, L.N. Ganzeveld, W. Helder, T. Kaminski, M.A. Sofiev and S.E. Trurnbore
Experimental designs appropriate for flux determination in terrestrial and aquatic ecosystems
99
D. Fowler .
Toward the use of remote sensing and other data to delineate functional types in terrestrial and aquatic systems
123
J.E. Estes and T.R. Loveland ,
Working group report: How can we best define functional types and integrate state variables and properties in time and space ?
151
S. Seitzinger, J.-P. Malingreau, N.H. Batjes, A.F. Bouwman, J.P. Burrows, J.E. Estes, J. Fowler, M. Frankignoulle and R.L. Lapitan
8.
Modelling carbon dioxide in the ocean: A review
169
D. Archer
Simulation models of terrestrial trace gas fluxes at soil microsites to global scales
185
D.S. Schimel and N.S. Panikov
10.
The application of compensation point concepts in scaling of fluxes R. Conrad and F.J. Dentener
203
vi Working group report: Relations between scale, model approach and model parameters
11.
217
s Middelburg, P.S. Liss, F.J. Dentener, T. Kaminski, C. Kroeze, s Malingreau, M. Nov6k, N.S. Panikov, R. Plant, M. Starink and R. Wanninkhof
12.
Validation of model results on different scales M.A. Sofiev
233
13.
Role of isotopes and tracers in scaling trace gas fluxes S.E. Trumbore
257
14.
Inverse modelling approaches to infer surface trace gas fluxes from observed atmospheric mixing ratios
275
M. Heimann and T. Kaminski
15.
Working group report: How should the uncertainties in the results of scaling be investigated and decreased ? R.G. Derwent, A.R. Mosier, S. Bogdanov, s Dzo,zer, V. Garcon, S. Houweling, M.A. Softer, H.A.C. Denier van der Gon, F. Wania and R. Wanninkhof
16.
Current and future passive remote sensing techniques used to determine atmospheric constituents J.P. Burrows
297
315
Participants and contributing authors
349
Index
353
vii
FOREWORD
The world's terrestrial and aquatic ecosystems are important sources of a number of greenhouse gases and aerosols which cause atmospheric pollution and disturb the energy balance of the Earth-atmosphere system. In recent decades the measurement techniques and instrumentation for quantifying gas fluxes have been improved considerably. Yet, the uncertainties in the regional and global budgets for a number of atmospheric compounds have not been reduced due to the large spatial heterogeneity and temporal variability of the factors that control gaseous fluxes in ecosystems. Techniques used for extrapolating measurements or properties and constraining results between different temporal and spatial scales are nowadays referred to as "scaling". All scaling methods are embedded in the data. Apart from uncertainties associated with the data used, errors may be caused by generalization of the basic data (e.g. in soil maps, ocean maps). Moreover, much of the spatial and temporal variation at a detailed level is obscured as a result of aggregation. Possible errors caused by the use of aggregated or generalized data in models are generally not explicitly analyzed. An important step in scaling of gas exchanges between ecosystems and the atmosphere is the delineation of functional types where distinct differences in structure, composition or properties of landscapes or water bodies coincide with functions or processes relevant for gas fluxes. Delineation reduces the variability of state variables, and therefore functional types form a good basis for measurement strategies and model development. Models are widely used tools in bottom-up scaling approaches. Models can also be used to calculate flux values for regions where less intensive or no measurement data are available. One of the challenges in model development is the integration of properties or variables in space and time, accounting for the spatial and temporal variability of processes involved in gas production, consumption and exchange. Scaling not only comprises bottom-up approaches, but also top-down methods, such as inverse modelling to calculate from the atmospheric concentrations back to the sources. Topdown scaling in general involves the validation of estimates obtained at a lower scale level against constraints given at a higher level of scale. Hence, scaling requires uncertainty analysis at all levels considered. The present book is a collective effort of a diverse group of scientists to review the stateof-the art in the field of scaling of fluxes of greenhouse gases and ozone and aerosol precursors. It focuses on identification of gaps in knowledge, and on finding solutions and determining future research directions. The book is the result of an international workshop on "Scaling of trace gas fluxes between terrestrial and aquatic ecosystems and the atmosphere", held from 18-22 January 1998 at kasteel "Hoekelum", Bennekom, the Netherlands. The workshop was organized by the International Soil Reference and Information Centre (ISRIC) as a follow-up to the international conference on "Soils and the Greenhouse Effect" which ISRIC organized in 1989. The overall goal of the workshop was to investigate approaches to reduce uncertainties in estimates of fluxes of trace gases and aerosols between terrestrial and aquatic ecosystems and the atmosphere at the landscape, regional and global scale. To achieve that goal, the participants concentrated on: (i) Identification of data gaps in scaling approaches between
viii field, landscape, regional and global scales; (ii) Development of procedures to bridge process level information between different scales; (iii) Assessment of methods for integration, aggregation and other data operations; and, (iv) Assessment of approaches to uncertainty analysis in bottom-up and top-down scaling. The workshop was one of researchers with many different backgrounds, including soil science, microbiology, oceanography, rec.ote sensing and atmospheric sciences. The group included experts in the determination of gas fluxes, modellers, specialists in the use of isotopes and tracers, and researchers working on the compilation of regional and global inventories and maps of soils, vegetation, land use and emissions. Twelve invited background papers, providing a review of the field, were distributed prior to the workshop, but were not presented at the meeting. Instead, the scientific programme of the workshop consisted of five days of discussions according to the well-known Dahlem workshop model. The participants were divided in four interdisciplinary working groups which met to address the workshop aims and give concise and practical recommendations, concentrating on the following questions: (i) How can fluxes of trace gas species be validated between different scales ?; (ii) How can we best define functional types and integrate state variables and properties in time and space ?; (iii) What is the relation between scale, the model approach and the model parameters selected ?; (iv) How should the uncertainties in the results of scaling be investigated ? The four group reports are included in this volume as separate chapters together with the peer-reviewed background papers. The organizing committee for the workshop, which started discussions Jn 1996, included the following members: A.F. Bouwman (National Institute of Public Health and the Environment, Bilthoven), N.H. Batjes (International Soil Reference and Information Centre, Wageningen), H.A.C. Denier van der Gon (Soil Science and Geology Department, Wageningen Agricultural University), F.J. Dentener (Institute for Marine and Atmospheric Research, Utrecht University), J. Duyzer (TNO Institute of Environmental Sciences, Energy Research and Process Innovation, Apeldoorn), W. Helder (Netherlands Institute for Sea Research, Den Burg), J. Middelburg (Netherlands Institute of Ecology, Centre for Estuarine and Coastal Ecology, Yerseke). The organization of the workshop was made possible through funds of the Commission of the European Communities (CEC-DG XII), European IGAC Office (EIPO), International Fertilizer Industry Association (IFA), Kemira Agro Oy, National Institute of Public Health and the Environment (RIVM), Norsk Hydro, Netherlands Royal Academy of Arts and Sciences (KNAW), Shell Nederland b.v., and the Netherlands Organization for Applied Scientific Research (TNO). Cooperating organizations were the Intemational Society of Soil Science (ISSS), International Geosphere-Biosphere Programme (IGBP), International Global Atmospheric Chemistry Programme (IGAC), Global Emission Inventories Activity (GEIA), Centre for Climate Research (CKO), and the Climate Change and Biosphere Programme of the Wageningen Agricultural University (CCB) Dr. L.R. Oldeman Director, Intemational Soil Reference and Information Centre (ISRIC) October 1998
ix
ACKNOWLEDGEMENTS
This volume is the result of an international workshop on "Scaling of trace gas fluxes between terrestrial and aquatic ecosystems and the atmosphere", held from 18-22 January 1998 at kasteel "Hoekelum", Bennekom, the Netherlands. It is a collective effort of a diverse group of scientists. The choice of topics, identification of authors for the invited background papers, and the scientific programme around four key questions is the result of discussions in the organizing committee. Thanks are due to the members of this committee, Niels Batjes, Hugo Denier van der Gon, Frank Dentener, Jan Duyzer, Wim Helder and Jack Middelburg. Thanks to the enthusiastic involvement of the committee members the workshop became a very successful one. I wish to thank the chairmen and rapporteurs of the four working groups for leading the discussions and summarizing the various contributions of the working group members in four reports which are included in this book: Andi Andreae and Willem Asman (group I), JeanPaul Malingreau and Sybil Seitzinger (group 2), Peter Liss and Jack Middelburg (group 3), Arvin Mosier and Dick Derwent (group 4). I am indebted to all participants for reviewing the invited background papers. Carl Brenninkmeier, who could not attend the meeting, was so kind to provide a review of one of the papers. I am also thankful to Niels Batjes for critically reading two papers, and to Ruth de Wijs-Christensen of RIVM for editing one of the background papers. I very much appreciated the support and ideas of Roel Oldeman Hans van Baren of ISRIC during the preparation of the workshop. Special thanks are due to Jan Brussen, Yolanda Karpes-Liem and Hans Berendsen for their enthusiasm and input during the preparations of the workshop and the workshop. Finally, I am grateful to Wouter Bomer for designing the workshop logo (also presented on the cover of this book) and for preparing some of the figures in the book. Finally, I wish to thank Fred Langeweg of RIVM for giving me the opportunity to work on this projectl Lex Bouwman, editor October 1998
This Page Intentionally Left Blank
Chapter 1
TOWARDS RELIABLE GLOBAL B O T T O M - U P ESTIMATES OF T E M P O R A L AND SPATIAL PATTERNS OF EMISSIONS OF TRACE GASES AND AEROSOLS F R O M LAND-USE RELATED AND NATURAL SOURCES
A.F. Bouwman, R.G. Derwent and F.J. Dentener
This Page Intentionally Left Blank
Approaches to scaling a trace gas fluxes in ecosystems A.F. Bouwman, editor 9 Elsevier Science B.V. All rights reserved
TOWARDS RELIABLE GLOBAL BOTTOM-UP ESTIMATES OF TEMPORAL AND SPATIAL PATTERNS OF EMISSIONS OF TRACE GASES AND AEROSOLS FROM LAND-USE RELATED AND NATURAL SOURCES
A.F. Bouwman l, R.G. Derwent 2 and F.J. Dentener 3 ~National Institute of Public Health and the Environment, P.O. Box 1, 3720 BA Bilthoven, The Netherlands 2 Meteorological Office, London Road, Bracknell, RG 12 2SZ Berkshire, UK 3 University of Utrecht, Institute for Marine and Atmospheric Research, Princetonplein 5, 3584 CC Utrecht, The Netherlands
I. I n t r o d u c t i o n
Emission inventories play a dual role in global air pollution issues. Firstly, they can be used directly to establish the more important source categories, to identify trends in emissions and to examine the impact of different policy approaches. Secondly, emission inventories are used to drive atmospheric models applied to assess the environmental consequences of changing trace gas emissions and concentrations and to provide advice to policy makers. This second role contributes to the atmospheric modelling community being an important user of emission inventories. The assessment process for global air pollution problems has a number of identifiable steps: (i) it quantifies the changes in trace gas composition of the atmosphere; (ii) it quantifies changes in atmospheric chemistry, transport, deposition and radiative forcing; (iii) it identifies the climate responses to the changes in atmospheric composition of the radiatively active trace gases; and, (iv) it quantifies the biological and ecological responses to the predicted changes in climate. The atmospheric modelling community will need a hierarchy of emission inventories to complete an assessment of global air pollution problems based on these steps over the next decade or so. In their simplest form, atmospheric models merely require no more than fixed global emission fields of each relevant species. However, in their most complex form, future atmospheric models will require emission fields whose spatial patterns and magnitudes will respond in a wholly self-consistent manner to changes in economic prosperity, demography, land use, climate change and policy. The requirements placed on the emission inventories will change from the provision of fixed fields to the implementation of emission algorithms within the modelling system. Gridded emission fields may slowly change from being the essential input to being the output of the modelling system. Alongside this anticipated increase in complexity in moving towards a process-based approach to emission inventories, there is a growing interest in the emissions of a wider range of species. For example, in climate change research at the start of the Intergovernmental Panel on Climate Change (IPCC) process, assessment work was performed with present-day and doubled atmospheric carbon dioxide (CO2) concentrations. This "steady state" or "equilibrium" approach has now been replaced by the transient scenario approach in which CO2 concentrations increase with time in response to emission projections and carbon cycle modelling. Further scenarios have been added to deal with the other major radiatively active trace gases: methane (CH4), nitrous oxide (N20), tropospheric ozone (03), stratospheric ozone
4
A.F. B o u w m a n , R.G. D e r w e n t a n d F.J. D e n t e n e r
Table 1. Some recent global-scale chemistry-transport r Model
Type
GETTM GEOS-I DAS GFDL/GCTM GISS/CTM GRANTOUR IMAGES IMARU
.tels (types, spatial and time scales and che :'ical mechanisms),
Spatial resolution of meteorological data
Time resolution of meteorological data
Chemical mechanism
Reference a
Eulerian Eulerian Eulerian Eulerian Lagrangian Eulerian Eulerian
2.8~215176215 18 levels 2.5%2~215 levels 265 km• kmxl I levels 4~215 levels 7.5~215 12 levels 5%5~215 levels 3.75%3.75% 19 levels
3-hourly 6-hourly 6-hourly 4-hourly to 5-day 12-hourly monthly 1/2-hourly
1 2 3 4 5 6 7
KFA/GISS KNMI/CTMK MCT/UiB MOGUNTIA
Eulerian Eulerian Eulerian Eulerian
10%8% 15 levels 2.5%2.5~ 15 levels 150 kmx 150 kmx 10 levels 10~ 10% 10 levels
8-hourly 12-hourly hourly monthly
MOZART RGLK
Eulerian Eulerian
2.8%2.8% 18 levels 10~ 10% 0 levels
6-hourly Monthly
STOCHEM UiO/GISS
Lagrangian Eulerian
3.75~215 8~ 10~
radioactive decay Simplified Simplified DMS chemistry 76 species 47 species CH4 and NO• chemistry Simplified 13 species 51 species CH4 and NOx chemistry 45 species SO2, NOx and NH3 chemistry 70 seecies 50 species
19 '. ~els levels
6-hourly 8-hourly to 5-day
8 9 10 11 12 13 14 15
a 1, Li and Chang (1996); 2, Allen et al. (1996); 3, Moxim et al. (1996); 4, Chin et al. (1996); 5, Penner et al. (1994); 6, Mt~ller and Brasseur (1995); 7, Roelofs and Lelieveld (1995); 8, Kraus et al. (1996); 9, Wauben et al. (1997); 10, Flatoy and Hov (1996); 11, Dentener and Crutzen (1993); 12, Brasseur et al. (1997); 13, Rodhe et al. (1995); 14, Collins et al. (1997); 15, Berntsen et al. (1996).
and chlorofluorocarbons (CFCs). More recently, sulphur dioxide (SO2), dimethylsulphide (DMS), ammonia (NH3) and other aerosol species have been incorporated into the scenario approach to take into account the climate consequences of man-made fuel combustion and biomass burning. There has therefore been an increasing interest in the details of the emission inventories of a wider range of trace gases and aerosols. Emission inventories are implemented in current atmospheric models to represent the processes by which trace gases and aerosols are discharged into the model atmosphere. The models, commonly three-dimensional chemistry transport models (CTMs), then simulate the dispersion, diffusion and advection of t h i material away from its source r~'gions in response to a continuously varying turbulent and chaotic atmospheric flow. At some point, the material may be removed from the atmospheric circulation by dry or wet deposition or uptake in the oceans or it may undergo chemical transformation. An immense amount of meteorological, chemical and process information is required to drive current CTMs. This information can be made available from archived meteorological analyses or from the global climate models (GCM) used to predict future climate change. The CTM may be incorporated within the GCM, in which case the atmospheric model is "on-line"; alternatively, the model is referred to as "off-line" if the GCM and CTM are separated. By way of example, details of the time and spatial resolution of 15 of the current CTMs are provided in Table 1. At present some 20 CTMs are being used to assess global air pollution problems. Currently, CTMs use emission inventories for the trace gases and aerosol species listed in Table 2. Each emission is usually subdivided into up to about 10 major source categories. Most source categories have different spatial grids applied and work with different seasonal and sometimes diurnal variations. We will focus here on the issues of "scaling" in the implementation of emission inventories in current and future CTMs. Scaling comprises all techniques used for extrapolating measurements or properties and constraining results between different temporal and spatial scales. Very similar problems of scaling occur across various disciplines, such as ecology (Ehleringer and Field, 1993), soil science (Wagenet and Hutson, 1995; Hoosbeek and Bryant,
Towards reliable global estimates of emissions of trace gases and aerosols
5
Table 2. Some of the trace gases and aerosol species handled by current chemistry transport models (CTMs) for the
assessment of global air pollution problems. Radiatively active gases
Aerosols
CO2 N20 CFCs: 11,12,113 HCFCs: 22, 141b, 142b HFCs: 134a, 152a Perfluoro molecules: SF6, CF4, C2F6,C4F8
Black carbon Organic particles Wind-blown dust Sea-salt Resuspended material Volcanic emissions Biomass burning
Ozone precursor and depleting gases
Aerosol precursor gases
CO NOx H2 Synthetic hydrocarbons: light C2 - C~o hydrocarbons, oxygenates Biogenic hydrocarbons: isoprene, terpenes CH3CC13,C2C14,C2HC13, CH2C12 CH3Br CH3CI Synthetic bromine compounds: 12B 1, 13B 1
SO2 DMS H2S NH3
CH4
1992) and global change research in general (O'Neill, 1988). Two approaches are used for scaling gas fluxes: bottom-up and top-down scaling. Bottomup approaches, calculated from smaller to larger scales, involve extending calculations from an easily measured and reasonably well understood unit to more encompassing processes. In bottom-up scaling, various problems occur, such as how to aggregate the spatial and temporal variation of properties or fluxes. Other problems are the various data uncertainties involved, and translating mechanisms and processes between different scales. Top-down approaches can mean using the measurements at a higher scale level which set the boundary conditions for problem identification, and stimulate the testing of general relationships for specific cases (see Heimann and Kaminski, 1999). Examples of observations at a higher level of scale that are used to constrain flux estimates include atmospheric concentrations and mixing-ratios of stable isotopes (see Trumbore, 1999). Comparison of the concentrations or deposition velocities simulated by transport models with observations can result in an expression of scientific confidence or a warning that crucial !r,formation is still missing. Between these two extremes, top-down scaling can be very useful for testing hypotheses and identifying missing information. A number of methods exist to scale information, the most important being aggregation, generalization, stratification and modelling. Aggregation involves the collection or uniting of information into an aggregate unit, generally by counting and addition. Aggregated results can be presented as the mean or median coupled with statistical information such as standard deviations. Generalization is the description of a group on the basis of properties of a sub-unit or member of the group considered to be representative, commonly based on expert knowledge. This method is generally used when observational or statistical data on individual members of the group are scarce. The reverse action of aggregation, whereby the aggregate is subdivided into different components, may be the classification of a system into functional units with similar properties and environmental and management conditions that regulate trace gas fluxes (see Seitzinger et al. (1999). This is sometimes referred to as "stratification" (Matson et al. (1989).
6
A.F. Bouwman, R.G. Derwent and F.J. Dentener
Models break down a system into its main components and describe the behaviour of the system through the interaction of those components. A discussion of the different types of models used can be found in Archer (1999) for aquatic systems and Schimel and Panikov (1999) for terrestrial systems. We will focus here on bottom-up scaling approaches for trace gas fluxes between aquatic and terrestrial ecosystems (including agroecosystems) and the atmosphere used in the development of global gridded emission inventories. The discussion will be primarily on emission inventories prepared for scientific purposes such as atmospheric modelling. Although our findings may also hold for other types of inventories, we will not discuss these inventories explicitly on the country or provincial (sub-national) scale. Such inventories are now being prepared for non-scientific purposes (e.g. national communications in the United Nations Framework Convention on Climate Change). The first, and major, part of this paper discusses the uncertainties and problems of aggregation, generalization, stratification and modelling in the compilation of inventories. Next, the available global emission inventories for land-use related and natural sources of trace gases will be discussed on the basis of their spatial and temporal resolution. Finally, the spatial and temporal resolution of current CTMs will be confronted with the available emission inventories.
2. Uncertainties in emission estimates Among the various approaches to estimating fluxes, the major ones in use are the emission factor approach and modelling. In emission factor approaches emission estimates are derived by combining measurement data with geographic and statistical information on the ecosystem processes and economic activity. This can be represented as: E = A • Eu
(1)
where E is the emission, A the activity level (e.g. area of a functional unit, animal population, fertilizer use, burning of biomass) and EU the emission factor (e.g. the emission per unit of area, animal, unit of fertilizer applied or biomass burnt). When using the emission factor approach, both the stratification scheme for delineating functional types (e.g. management systems, ecosystems, environmental provinces or entities) as a basis fol scaling, and the reliability of the emission factor determine the accuracy of the flux estimates. The firmest basis for scaling is developing an understanding of the mechanisms that regulate spatial and temporal patterns of processes, and describing these mechanisms in models. Models are used to break down a system into its component parts and describe the behaviour of the system through their interaction. In general, trace gas flux models include descriptions of the processes responsible for the cycling of carbon or nitrogen and the fluxes associated with these processes. Various types of models exist, including regression models, empirical and process (or mechanistic) models. In the following sections the various sources of uncertainty in the estimates of emission rates for the emission factor approach, trace gas flux models and farm-scale models will be discussed, followed by uncertainties associated with the spatial and temporal distribution of the data underlying flux estimates. We will not discuss uncertainties in the measurement data. This problem will be examined in more detail by Lapitan et al. (1999), Fowler (1999), Denmead et al. (1999) and Sofiev (1999).
Towards reliable global estimates of emissions of trace gases and aerosols
7
2.1. U n c e r t a i n t i e s in the e m i s s i o n rates
2.1.1. Emission factor approach
Uncertainty ranges for emission inventories are usually presented for the global total emission only, and not on a regional or grid-by-grid basis. Uncertainties in emission inventories may be caused by uncertainties in the environmental and economic activity data used and in the measurement data themselves. Uncertainties can also result from the lack of representative measurem'ents to resolve the full range of ,mvironmental conditions occurrhag in the systems considered and in the models used. These sources of uncertainty will be discussed for the different approaches to flux estimation on the basis of a number of examples for different scales. - M e a s u r e m e n t data. In a review of measurement data for biomass burning, Andreae (1991) proposed emission factors for several gas species. Although the ranges in measured fluxes in field and laboratory experiments varied by more than a factor of 2 for most species as a result of differences in fuel and burning conditions, one single emission factor was proposed for each gas species, representing the aggregated flux for smouldering and flaming fires for all fuel types (grass, wood, crop residues, etc.). For biomass burning it is difficult to delineate the types of fires and the different techniques used may introduce systematic differences, especially where reactive and difficult-to-measure species (such as NOx and NH3) are involved. Clearly, one emission factor cannot describe all the burning conditions and fuel
types. Another example illustrating the lack of measurement data concerns the emission coefficients used for animal housings in Europe. In housings with mechan!cal ventilation the gas flux can be easily determined from the gas concentration in the ventilation air and the flow rate. The trace gas emissions from naturally ventilated housings can only be determined indirectly and with greater uncertainty. In such "open" housings the emission depends on the opening and closing of doors. In large parts of Europe, housings for cattle - the most important category- are naturally ventilated (Asman, 1992). Besides being scant, the available measurement data need not be representative. For example, the NH3 ammonia emission per animal may vary by a factor of 4 within the same type of housing (Pedersen et al., 1996). This may be caused by differences in the ventilation over the slurry between housings and by differences in waste management practices such as cleaning. Guenther et al. (1995) were also confronted by a lack of measurement data in their global invemory of fluxes of volatile organic compounds (VOC) from vegetation. Measurements represented only 26 of the 59 global land-cover types considered; the remaining land-cover types, including tropical seasonal forests and savannas, were assigned an emission on the basis of expert knowledge. In this database, most of the simulated VOC emissions come from systems where very few or no measurements are available. - Functional types. Guenther et al. (1994) proposed emission factors for VOC for 91 woodland landscapes in the USA by combining emissions from 49 genera of plants. In their global modelling ofVOC, Guenther et al. (1995) used emission factors on the basis of the 59 land-cover types defined by Olson et al. (1985). This aggregation causes considerable loss of information, as the detailed estimates for the USA vary by as much as a factor of 5 for various aggregated landscapes on the global scale. Yienger and Levy II (1995) used a combination of emission factors and modelling approaches to estimate global emissions of NO from soils. They first calculated "biome factors" based on NO flux estimates from the literature. These biome-dependent average fluxes were modified by an algorithm to account for pulse events of NO production following
8
A.F. Bouwman, R.G. Derwent and F.J. Dentener
wetting of dry soil and another algorithm to account for the effect of varying temperature. Yienger and Levy (1995) also made an attempt to model the effects of NOx uptake by plants on net NOx emission to the atmosphere. They calculated absorption factors based on leaf area indices, and then multiplied these absorption factors by the estimated soil emissions to calculate net ecosystem emissions. The model of Yienger and Levy has some mechanistic components, such as the wetting and temperature functions, but is primarily based on averaged biome factors that are not substantially different from an emission factor approach. Davidson and Kingerlee (1997) also derived emission factors based on data for biomes from the literature. Although in their study more soil NOx measurement data were used in comparison to Yienger and Levy's study (1995), the major differences between the two studies are the stratification scheme and t;L,~coupling of environmental con.~ition descriptions at the measurement sites with the functional types distinguished. Davidson and Kingerlee (1997) presented a global annual emission which exceeds the estimate of Yienger and Levy (1995) by a factor of 2. It is clear that the differences between the two studies described will have an enormous impact on the results of atmospheric models. 2.1.2. Regression approaches
Bouwman et al. (1993) calculated the N20 emission from soils under natural vegetation using a simple global model describing the spatial and temporal variability of the major controlling factors of N20 production in soils. The basis for the model is the strong relationship between N20 fluxes and the amount of nitrogen (N) being cycled through the soil-plant-microbial biomass system. The model calculates the monthly N20 production potential from five indices representing major regulators of N20 production (soil fertility, organic matter input, soil moisture status, temperature and soil oxygen status). These five indices were combined in the final N20 index (Figure 1). Comparison of the N20 index with reported measurements for about 30 locations in six ecosystems correlated with an r2 o f - 0.6 (Figure l a). The resulting regression equation was used to calculate emissions on a l ~ 1~ resolution. However, the correlation coefficient is not a robust statistical method (see Sofiev, 1999), and minor differences in only one of the measurement sites can cause major shifts in the correlation coefficient (Bouwman et al., 1993). A major problem causing unreliability of the regression equation is the lack of measurement data, particularly for a number of important ecosystems and world regions that have not been sampled at all. It is not known how the model performs in these areas (Figure 1b). 2.1.3. Process models
Reliable regional or global estimates of trace gas emissions depend on an examination of methodologies to reduce the current high uncertainty in the estimates. One potential way to do this is to develop predictive flux models. Such models have been developed for different processes and gas species on different scales. Examples will be given of the magnitude of the uncertainties in global models, the value 9f models developed for speci~c ecosystems for extrapolation and the problem of selecting the appropriate scale of process descriptions in models. Finally, the advantages of using a range of models on different scales will be discussed. Uncertainties in global flux models. Here, examples for oceanic flux models will be given, although very similar examples also exist for terrestrial models. In aquatic systems, fluxes can generally not be directly determined. Models commonly used describe fluxes on the basis of wind speed and anomalies of concentrations between surface water and air. Nevison et al. (1995) calculated the air-sea exchange using three different relationships for the NzO-air -
Towards reliable global estimates of emissions of trace gases and aerosols
a. R e l a t i o n b e t w e e n
550
o
measured
NzO fluxes and modelled
9
N20
index
Measured flux Regression equation
450
-rr
350 o
0
E z
25O
o
6 z ~
o
150 o 0
50
oO 0
-50
o
o.........oi" s
0
i
o
+ N20
b. L o c a t i o n of m e a s u r e m e n t
;
index
sites
90
90
"""
.,~. ".. ~,.~-,,:,~:..,
"
;~
.9
-30
,
:'~.',]:-
-
-
' ~-~,~,,,
tc-'..-~S'.:.----~
:-~.-.~~ ....,:.....~ . ~I " ~ . - 1,~).. - ......
-.,..'
: ,-. ..,,':
.:..,~...,...
!
"... ~:..(., :... _.
"~ :
:':
, ..~='--"
"-~ .... ~
"
-
O M e a s u r e m e n t sites .................... , ............ ,::':G..........
-90
-
,-.(~: . ~~.~ -:.:-
~. ',%~: . . . -
.-,~.-
"
.
...-
- . . , . -,,"
-,..
~
::
-60
:
,,
6
......................................
..9 . . . . .
-30
" .,.:.-
- .............................................
-60
-. . . . . . . .
-1 ,o
6
6'o
-90
Figure 1. a) Relationship between measured N20 flux and simulated N20 index; and b) the location of the measurement sites. Figure 1a was modified from Bouwman et al. (1993) with kind permission from the American Geophysical Union.
gas transfer velocity from global N 2 0 surface anomalies. The highest N20 fluxes were obtained using a quadratic function of wind speed for the transfer velocity, while linear functions yielded much lower values. An intermediate relationship was the stability-dependent method based on the occurrence of whitecaps, also used by Guenther et al. (1995) for estimating the VOC emission inventory for oceans. The uncertainty in the global emission is illustrated by the difference of more than a factor of 4 between the lowest and the highest global emission estimate. - Limitations of e c o s y s t e m m o d e l s . Although models developed for specific ecosystems may show fewer uncertainties than global models, their value for extrapolation may be
10
A.F. Bouwman, R.G. Derwent and F.J. Dentener
limited. Mosier and Parton (1985) developed a model for the estimation of N20 fluxes over large areas of semi-arid grassland soils, accounting for spatial and interannual variability. Model parameters were developed by relating N20 flux to soil moisture and temperature for two sites representing much of the variability in the Colorado shortgrass ecosystem. Because no time-series data of NO3 and NH4+ are available on the target scale of the study, the model was simplified with an empirical multiplier representing N availability. It is especially empirical multipliers like these that cause problems when models are applied to other ecosystems with different environmental and climatic conditions. Scale of process descriptions. Some models seem to include an imbalance in the detail and the particular scale on which different processes are described. For example, Li et al. (1992) developed a model to simulate N20 fluxes from decomposition and denitrification in soils on the field scale. The model can also describe NO• fluxes by using soil, climate and data on management to drive three submodels (i.e. thermal-hydraulic, denitrification and decomposition submodels). The management practices considered include tillage timing and intensity, fertilizer and manure application, irrigation (amount and timing), and crop type and rotation. One of the processes simulated by the model is microbial growth. Since model results appear to be dominated by the effect of temperature and soil moisture, which operate at nearly all levels in the model, the question arising is whether there is an imbalance in the scales according to which processes are described. The similarity of the results obtained for shortgrass ecosystems by Mosier and Parton (1985) with their simple approach to those of Li
Simulated (g m"2) 75-
(a)
60-
O O
model correspondence (rA2=0.905, n=36)
45-
1"1 correspondence
~
30-
~ o (:)
o
oo
15-
0 - . / model correspondence u (rA2=0.893, n=36)
(b)
45
o
~
j
1"1 correspondence
~176
0
9
9e
15
30
45
60
75
Measured (g m2) Figure 2. Comparison of simulated and measuredtotal seasonal methane emissions from Texas flooded rice paddy soils during the 1991-95 growing seasons employingconsecutively (a) the simulation model and (b) the simplified model. Model correspondence is the regression line of simulated vs. measured methane emissions. Reprinted from Huang et al. (1998) with kind permission of Blackwell Science Ltd.
Towards reliable global estimates of emissions of trace gases and aerosols
11
et al. (1992) illustrates the need to match the scale of process description with that of the scale
at which the model is applied. Comparisons of different models to predict N20 fluxes from fields (Frolking et al., 1997) reveal major differences in the simulated N gas fluxes from soils. Apparently, the major problem in developing trace gas flux models is the description of soil processes that operate in "hot spots" in field models. - M o d e l s at different scales. Apart from the above-mentioned problem of the scale on which processes operate, a very practical problem is formed by the available model input data. To overcome this problem, sometimes summary models are developed on the basis of the detailed process model. These summary models can be used to predict fluxes in regions with limited data availability. Progress with the use of models on different scales for flooded rice paddy fields was made by Huang et al. (1998). Understanding the processes of methane production, oxidation and emission in flooded rice fields enabled them to develop a semiempirical model. They also derived a simplified (summary) version of the model for application to a wider range of conditions but with limited data sets. Huang et al. (1998) hypothesized methanogenic substrates as being primarily derived from rice plants and added organic matter. Rates of methane production in flooded rice soils are determined by the availability of methanogenic substrates a,~d the influence of environmemal factors. Model validation against observations from single-rice growing seasons in Texas (USA) demonstrated that the seasonal variation of methane emission is regulated by rice growth and development. A further validation of the model against measurements from irrigated rice paddy soils in various regions of the world, including Italy, China, Indonesia, Philippines and the United States, suggested that methane emission could be predicted from rice net productivity, cultivar character, soil texture and temperature, and organic matter amendments. The detailed model and the summary model gave similar results (Figure 2), illustrating the advantage of using simplified models. 2.1.4. F a r m nutrient balance models
On the farm scale, trace gas fluxes occur in the stable, during grazing or during and after spreading of animal manure. A model is therefore required to describe farm-scale processes and cycles. For example, the model of Hutchings et al. (1996) describes NH3 losses from animal housings, stored slurry, application of slurry and urine patches. The model builds on knowledge acquired from various experiments and model studies of animal housing, waste storage and farming practices. The model tracks the N input as animal feed until it is lost as NH3. The problem of applying farm-scale models is the variety in management styles occurring within groups of farms. Representative farms or averages for a group of farms have to be used to obtain aggregated data. Differences in fluxes as a result of differences in management may disappear due to this aggregation.
2.2. U n c e r t a i n t i e s in the spatial distributions
The environmental spatial data used as a basis for stratification schemes for delineation of functional types underpins the emission factor approach and, if sufficient attribute data are available, drives flux models. When no spatial data are available to distribute activities or emissions, a proxy or surrogate distribution has to be used. Clearly, this introduces an unknown uncertainty in the spatial distribution. We will give a number of examples of databases that describe environmental conditions in aquatic and terrestrial ecosystems, emphasizing their uncertainties. A comprehensive review of the data required for global terrestrial modelling can be found in Cramer and Fischer (1996). The list of examples given
12
A.F. Bouwman, R.G. Derwent and F.J. Dentener
here is not intended to be complete, but does illustrate data limitations and aggregation problems. The weaknesses and strong points in the databases discussed may serve to improve future database development. The examples considered include databases for climate, oceans, soils and vegetation/land cover, as well as the problem of surrogate spatial distributions. 2.2.1. Climate
The example of a database on current climate for a global terrestrial 0.5 ~ x 0.50 grid given by Leemans and Cramer (1991; update in preparation) includes average monthly, average minimum and maximum air temperature, precipitation and cloudiness values. Data limitations. The weather records were usually limited to at least five observational years from the period of 1931-1960. Not all stations considered have complete coverage. Based on selection criteria, the final number of stations worldwide was found to be 6280 for temperature and 6090 for precipitation. The cloudiness data set, defined as the number of recorded bright sunshine hours as a percentage of potential number, was based on fewer stations and often derived from estimated rather than recorded data. Aggregation. To aggregate the point data to a spatial grid an interpolation onto 0.5 ~ grid boxes was done using a triangulation network followed by smooth surface fitting. For regions with no primary data, the temperature val.aes were corrected for altitude using an estimated moist adiabatic lapse rate and a global topography data set, while precipitation was not corrected; this was due to the more complex relationships between precipitation and altitude. - Uncertainty. The major problem is the inappropriate data coverage for large areas of the world. The uncertainty of temperatures is particularly high in mountainous areas because there are only a few weather stations in these regions and none of them are located on a clear altitudinal gradient. The average moist adiabatical lapse rate for mountainous areas may result in underestimation of temperatures for these areas. The spatial precipitation patterns resulting from straight interpolation of measured values causes great uncertainty in areas with sparse data coverage. Although the major annual cloud dynamics are represented, the regional reliability of the cloudiness data is low. -
-
2.2.2. Oceans
The best known chemo-physical global ocean data sets are included in the World Ocean Atlas (Conkright et al., 1994; Levitus and Boyer, 1994a, b; Levitus et al., 1994). This database includes spatial information on a l~ 1~ grid at various depths between 0 and 5500 m below the surface for ocean temperature, salinity, dissolved oxygen, apparent oxygen utilization, oxygen saturation, phosphate, and nitrate and silicate. Data for temperature and salinity have a monthly time resolution and apply to depths between 0 and 1000 m below the surface; those for dissolved oxygen, apparent oxygen utilization and oxygen saturation are on a seasonal temporal scale and phosphate; nitrate and silicate concentrations taken on an annual basis. Data limitations. The World Ocean Atlas is based on many observations. For example, the temperature data set is based on 4.5 million profiles. Although the number of observations is much higher than that used to produce the soil, vegetation/land cover and climate databases, there is a problem of areas with a low density or absence of observations; furthermore, the timing of the measurements may differ between profiles. - Aggregation. The data at the observed depth were interpolated to standard depths. The accuracy of the observed and standard level data was checked and flagged using a number of procedures. The point data for depth profiles were interpolated onto a 1o grid. There are many regions where measurements are scant or even absent. To describe the density of observations, there are accompanying mask files for all the data listed -
-
U n c e r t a i n t y .
Towards reliable global estimates of emissions of trace gases and aerosols
13
above, containing the number of grid points with data within the radius of influence surrounding each grid box. If a grid box contains three or fewer observations within its radius of influence, the mask value for that 1~ grid box will be zero. This file is used in plotting routines to "mask" or cover up areas with three or fewer observations. 2.2.3. Soils
Soil fertility, and soil chemical and physical parameters, play an important role in the production and exchange of trace gases. Recently, a 0.5 ~ • 0.5 ~ global soil database was developed on the basis of an edited version of the 1:5 million scale FAO Soil Map of the World (FAO, 1991), combining geographic information on soil types with a set of representative soil profiles held in a profile-attribute database (Batjes and Bridges, 1994). Data limitations. The density of available soil profile data varies from one region to the other. Important geographic gaps are in China, the New Independent States and the Northwest Territories of Canada. Similarly, a number of soil units are underrepresented in the profile database; these units account for about 28% of the terrestrial globe of which total Lithosols (shallow soils) account for about 40%. - Aggregation. The FAO Soil Map of the World is a compilation of many national and regional soil maps. Therefore coverage is not spatially constant. The soil profile information for each soil unit was coupled to the soil units distinguished region-wise. Based on the number of profiles available, statistical analysis was performed by Batjes (1997), allowing refinement of ratings for soil quality in global environmental studies. Uncertainty. The variability of the reliability of the spatial information has already been mentioned. The attribute files containing soil profile data in Batjes and Bridges (1994) represent a major improvement on the FAO soil map as such. However, this aggregation may not realistically describe the variability actually occurring within a soil unit in regions where the density of observations is low. -
-
2.2.4. Vegetation~Land cover
Similar to the soil information, land-use and land-cover information is required to scale up information from the field to landscapes or ecosystems. Two examples of widely used vegetation/land-cover maps are those compiled by Matthews (1983) and Olson et al. (1985) with 1~ and 0.5 ~ spatial resolution, respectively. A recent development is the creation of a global 1-km resolution global land-cover characterization (Loveland et al., 1997) based on remotely sensed data. For the pan-European region (from Gibraltar to the Ural and from the North Cape to Athens) a land-cover database with a 10% 10 minutes resolution was developed (Veldkamp et al., 1996). Data limitations. Matthews (1983) used the Unesco (1973) vegetation classification scheme, while the database by Olson et al. (1985) is based on a land systems grouping. Estimates of the extent of vegetation/land-cover types excluding cultivated land show a considerable difference between the two databases. The global area of cultivated land is similar in all the maps and corresponds well with FAO statistics, although regional discrepancies may exist. The Olson and Loveland et al. databases include estimates for carbon stocks in each land-cover type. Apart from definitional problems, there is generally a major lack of observational data describing the properties of the vegetation/land-cover types distinguished. As in the soil database of Batjes and Bridges (1994), the map unit characteristics will be included in attribute files, allowing use of the data for different purposes in a variety of models. Aggregation. The Matthews and Olson databases were compiled from maps, atlases and -
-
14
A.F. Bouwman, R.G. Derwent and F.J. Dentener
other information available. For spatial aggregation satellite observations may form a considerable improvement. The 10~ x 10~ resolution for the pan-European region (Veldkamp et al., 1996) includes eight classes produced from a combination of spatial data in vector format (based on various sources, including satellite data) and tabular statistical data. A calibration routine was used to ensure that no land-use class deviated more than 5% from the statistical information. The Loveland et al. database is derived from 1-km Advanced Very High Resolution Radiometer (AVHRR) d:,,a, spanning a 12-month period (April 1992-March 1993). It is based on seasonal land-cover region concepts, which provide a framework for presenting the temporal and spatial patterns of vegetation in the database. Uncertainty. Major uncertainties in the traditional databases, such as Matthews (1983) and Olson et al. (1985), are seen in the classification scheme used, the underlying data and the aggregation method, which is illustrated by the disagreement in the spatial distributions between these two databases. The database of Veldkamp et al. (1996) may suffer from the small number of types distinguished; this may not allow a proper description of the observed variability necessary for ecosystem and trace gas studies. However, the combination with soil and climate data may form an improvement here. The database also lacks data on the characteristics of the vegetation type itself in the form of attribute data. Since the Loveland et al. database is still in development, its uncertainty is as yet unknown. A review of the use of remote sensing and other data in vegetation mapping is given by Estes and Loveland (1999) -
2.2.5. Surrogate distributions
When the exact location or distribution of an activity or process is not known, surrogate distributions are used to distribute activities, volumes or emissions over the grids. For example, the grassland distribution is generally used to distribute cattle populations, while for other animal categories the rural human population distribution or the distribution of arable land is used as a surrogate distribution. However, the human population distribution is generally not well known in rural areas, as statistics and atlases give data on populations in major towns only. Using surrogate distributions may be realistic in some regions. However, in others with specific stratifications of management, environmental or demographic conditions, surrogate distributions may cause major errors (see, for example, the dairy cattle discussed in 2.4). 2.2. 6. General remarks
The major uncertainties in databases are generally related to the scarcity of data, and variable density of data coverage and quality. With reference to the data problem, the mask files (containing the number of grid points for data within the radius of influence surrounding each grid box) provided in the ocean database form a good tool for describing the data density and the point-by-point accuracy or reliability in other databases as well. Compared to the classification schemes for vegetation and land cover in the traditional maps and databases, satellite observations may provide a more flexible way of describing ecosystem characteristics. Attribute files with descriptive data of the map units distinguished (e.g. in the soil database of Batjes and Bridges, 1994) are very useful for modellers. These attribute data also enable performance of statistical analysis of the data by unit. Furthermore, correction of the satellite data with actual statistical information is a good way to improve the accuracy of the spatial data. Finally, a combination of vegetation/land-cover data with climate and soil information may provide a basis for classification into functions.
Towards reliable global estimates of emissions of trace gases and aerosols
|5
2.3. Uncertainties in the economic data on land use
The major forms of economic land use activities generating emissions of trace gases include livestock production, crop production and forestry. Livestock production is the most complex system. In livestock production systems, trace gas fluxes can be determined in a stable fi~r either individual animals or a group. The comp.ete production system, from feeu to excretion and emission in the stable and during grazing, has to be known for extrapolation of these measurements. For example, to estimate NH3 emissions from animal manure during storage and during and after application as a fertilizer, we need to know the number of animals in each animal category (e.g. dairy cattle) according to age class, live weight; N content and relative share of the various amino acids, N use efficiency (feed conversion to milk and meat); housing system and period of confinement, and form, mode and period of storage of manure. Further, we need to know weather conditions during spreading (turbulence, air temperature, air humidity and rainfall), properties of the soil to which the manure is applied, amount of manure per unit area, mode of manure application and the period between application and cultivation. Outside Europe and North America all these data are scant. Data on animal populations by category, and within a category (according to age and weight class) are almost non-existent. For many countries only the total number of animals within a category is available for a specific year. Data are not available on some animal categories, such as house pets, horses, buffalo, donkeys, camels, or on housing, and the type and form of manure. Estimates for regions within countries may be availai~:e, but do not always correspo,d to the official statistics or are outdated. Data on the coverage of stored manure, which may highly vary in effectiveness, are lacking. Geographic data on the application rate and timing of manure application, soil conditions, and weather conditions during application are not available. In addition to spatial variability, manure application rates, and mode and timing of application, show a strong interannual variability, which is not easy to include in scaling exercises. Storage and spreading of manure are regulated by law to reduce emissions in a number of countries. It is difficult to obtain information on the actual observance of these laws and the emission reductions achieved. Data on crop production systems that are essential for estimating trace gas fluxes envelop fertilizer use (including animal manure) by type and by crop, timing and mode of fertilizer application, amount and timing of field-residue burning, animal waste management, number of rice crops per year combined with soil and water management practices and fertilizer application rates. Such data may be available for regions within countries but may not always correspond to the official statistics or may be outdated. Global forestry data are available from FAO statistics and assessments ~z.g. FAO, 1995). However, information on the species planted and forest management are difficult to obtain. In assessments of trace gas fluxes it is generally important to know the amount of above- and below ground carbon in a certain forested area. Global data on carbon in vegetation can be obtained from Olson et al. (1985), for example, and carbon in soils from such sources as Batjes (1996). In summary, the economic and attribute data generally have to be inferred from aggregated country totals for the three land-use systems. Where the geographic distributions within countries are not directly available, data have to be distributed over a spatial grid or subnational regions. In this case surrogate distributions will have to be used (see section 2.2).
2.4. Uncertainties in the temporal distribution
Temporal patterns of trace gas fluxes vary in space. This poses difficulties for integration of
16
A.F. Bouwman, R.G. Derwent and F.J. Dentener
fluxes over spatial units. Spatial aggregation causes considerable loss of information on temporal flux patterns. However, the paucity of measurement data often makes generalizations unavoidable. Generalization is usually done by treating a landscape as a composite of representative soils or farms with average waste characteristics, management and weather conditions, or by treating populations as a group of identical members. Such generalizations may lead to errors in temporal distributions due to averaging procedures. The temporal pattern of estimates derived for a group of average farms may differ from the sum of all individual farms. Generally, different grazing systems co-exist within regions. For example, in dairy production systems part of the production takes place in stables only. The animal waste collected in the stables is at~plied to grassland or croplands at different times. Hence, the temporal pattern of gas fluxes is determined by the grazing systems occurring in the landscape considered. Errors caused by aggregation of groups of farms may be particularly large for N gas species. This was shown by Schimel et al. (1986), who analyzed the cycling and volatile loss of N derived from cattle urine at lowland and upland sites in a shortgrass steppe in Colorado, USA. The NH3 losses were measured in microplots representing three soil types typical for the shortgrass steppe landscape. Seasonal rates of urine and faeces deposition were mapped by landscape position, allowing for simulation of responses of animals to microclimate and forage availability, and differential use of upland and lowland pastures. This provided variation in the proportion of total excretion vulnerable to loss. Urine deposition was higher during the growing season when forage-N levels were high, and highest in lowland soils. Simple aggregation of the spatial patterns of deposition and loss would have resulted in a calculated loss of NH3 of a factor of 7 higher than for sophisticated stratification on the basis of the observed seasonal and spatial variability. Studies of gaseous fluxes are vulnerable to this type of error because fluxes can be intermittent and patchily distributed in space. Methane fluxes from rice fields are also extremely variable in time and space. Measurements for individual fields indicate diurnal and seasonal patterns caused by rice growth and development (e.g. Huang et al., 1998), which can best be described using process models (see above). Additional pulses caused by management practices are more difficult to describe in flux models or emission factor approaches because the statistical information on management is sparse and often absent, as discussed above. An attempt to distinguish seasonal variability in rice global cropping patterns was made by Matthews et al. (1991), who presented cropping calendars for rice production worldwide. This stratification serves as a basis for applying flux models with the corresponding data on soil, water and crop management. In summary, there is a problem in scaling-up of loss of information on temporal variability due to spatial aggregation or generalization. This problem may occur on any scale. Sophisticated and carefully chosen stratification schemes for the delineation of functional types within landscapes may help in reducing the aggregation loss of information on temporal variability. Temporal patterns can best be described by using process models.
3. Spatial and temporal resolution of current emission inventories and CTMs 3.1. Emission inventories In the previous sections we discussed a number of major problems that occur during the process of scaling-up data using different approaches on different scales. In this section we will present a number of global and regional inventories for selected trace gas species and sources of emissions which have been developed for scientific purposes. We will not discuss these
Towards reliable g l o b a l estimates o f emissions o f trace gases a n d aerosols
17
Table 3. Global inventories of emissions of trace gases and aerosols from aquatic and terrestrial ecosystems for a number
of gas species with a spatial resolution of 1o • 1o longitude-latitude representative for the period around 1990. Category
CO 2
CH4
N20
NO•
NH3
1 (m)
2 (m)
3 (h-d)
4 (y)
5 (y) 6(m)
2 (m)
CO
VOC
S/SO• Aerosols Black carbon
Land-use related sources
Crops, fertilized fields Animals (including enteric fermentation, animal waste) Biomass burning (including waste and fuelwood combustion Deforestation Post-clearing effects Landfills
7 (y)
7 (y)
7 (y)
2 (m)
4 (y)
7 (y)
4 (y)a
8 (y)
__
b
4(y) 2(m) 7 (y)
Natural sources
Soils under natural vegetation (including wetlands) Natural vegetation Oceans Lightning Volcanic activity
9 (y) 6(m)
2 (m)* 3 (d/m)3. 4 (y)* 10(h/m)*
11 (m)*
8 (y)
4 (m) 12(m)
8 (y) 8 (y)
13 (y) Wind erosion
14 (m)d*
The reference is indicated by a number and the temporal resolution in parenthesis by y (year), s (season), m (month) or h (hour). Inventories marked with an asterix (*) are model based; all other inventories are based on emission factor approaches. References: 1, Matthews et al. (1991); 2, Bouwman and Taylor (1996); 3, Yienger and Levy (I995); 4, Bouwman et al. (1997); 5, Lerner et al. (1988); 6, Fung et al. (1991); 7, Olivier et al. (1996); 8, Spiro et al. (1992); 9, Matthews and Fung (1987); 10, Guenther et al. (1995); 11, Nevison et al. (1995); 12, Lee et al. (1997); 13, Benkovitz and Mubaraki (1996); 14, Tegen and Fung (1995). a Inventory based on estimates of burnt dry matter burnt can also be used for other gases. b Inventory could be based on Bouwman et al. (1997). c Inventory is in fact based on emission factors for biomes coupled with a mechanistic model to produce temporal patterns of fluxes. d Soil dust emissions and transport are simulated on the basis of GCM-based wind fields.
inventories on the country or provincial (subnational) scale being prepared for non-scientific purposes (e.g. national communications in the United Nations Framework Convention on Climate Change). The inventories listed in Tables 3 and 4 represent data for the early 1990s or late 1980s. These lists are not intended to be complete but merely to illustrate the current "state-of-the-art" emission inventories. We have not presented earlier work, assuming that the methodology of early inventories is incorporated into the more recent ones. Some of the global inventories were based on regional data or inventories, and their spatial and temporal resolutions are not lower than those in the regional inventories. The reported spatial resolution for most regional and global inventories is 1~ 1o (Table 3). However, in many cases the real spatial resolution is much lower. For example, when inventories are based on the emission factor approach for vegetation types or biomes, the spatial detail is the biome and not the grid size. Emission factor approaches were used in many inventories, including all those for CH4, VOC, NO• and NH3. As discussed above, some of these inventories use simple rules or models to distribute fluxes over time. The most common temporal resolution of the inventories is one year. Some inventories have a monthly distribution; the inventory of NO• fluxes from soils has a temporal resolution of one day. This database was compiled by using the emission factor approach combined with
18
A.F. Bouwman, R.G. Derwent and F.J. Dentener
Table 4. Regional and continental inventories that include land-use related and biogenic emissions of a number of gas
species with different spatial and temporal resolutions. Region North America Europe
Europe, Russian Federation, United States of America Europe
Species/sources
Spatial scale
Temporal scale
Reference
CO, CH4, VOC, NOx. NH3, SO2, HCI for all known sources SO2, NOx, NH3, NMVOC, CH4, CO, N20, CO2 for all known sources SO2, NOx, NH3, NMVOC, CH4, CO, NzO , CO 2 for all known sources SO2, NO• NH3, VOC. CO for all known sources
80 • 80 km
h
1
Nuts regions, converted to 50• km grids + point sources 50• km grids + point sources
ya
2
ya
3
2~215 ~ grids (Ion. • lat.)
Ya
4
The temporal resolution is indicated by y (year), or h (hour). References: 1, EPA (1993); 2, EEA (1997); 3. UN (1995); 4. Veldt et al. (1991). a with time profiles for conversion to monthly or shorter time periods
a simple model based on temperature and precipitation data from one particular GCM. Some regional inventories include rules for distributing emissions in time, for example, on a daily or hourly basis (Table 4). National inventories will be produced in the framework in the IPCC Methodology for National Inventories. Most of these inventories will be compiled on the basis of default annual emission rates, as measurement data are not available in most countries. This temporal resolution of one year is similar to that of most of the global inventories.
3.2. Atmospheric models It is difficult to be definite about the current state of the art in CTMs since they continue to be developed as scientific understanding grows and as computers increase in soeed and capacity. Meteorological data with a time resolution of 1-6 h are typical of data used, while the spatial resolution in the models is typically a few degrees latitude and longitude. Models have typical runs of a few seasonal cycles: this is considered a mere snapshot when used for climate calculations. Model processes are usually handled with the same spatial and temporal resolution as the meteorological processes. It is important for two main reasons to accurately assess the trace gas fluxes between terrestrial and aquatic ecosystems and the atmosphere in CTMs. Firstly, CTMs need to describe these trace gas fluxes realistically so as to accurately assess the trace gas life cycle on the global or regional scale. Secondly, the CTMs may need to give an accurate representation of the trace gas flux for a particular ecosystem or region. In the first case, the spatial distribution of the flux may not be so crucial but it is important to achieve the correct total burden. In the second case, the flux to particular sensitive ecosystems may be a more important variable in the model than the total global flux. In considering model estimates of trace gas fluxes to terrestrial and aquatic ecosystems and their unce~ainties, there are a number of issues to consider. The CTM needs to describe the transport of the trace gas to the ecosystem and to present the trace gas to the ecosystem at the correct concentration level and on the correct time scale. Clearly, the greater the distance travelled from the point of emission and the smaller the area of the ecosystem, the greater the associated uncertainty. For regional-scale transport close to the planetary boundary layer, current CTMs should produce concentrations that are within the range o f - a factor of 4 or more for primary
Towards reliable global estimates of emissions of trace gases and aerosols
19
pollutants as monthly or seasonal averages in flat terrain 10-100 km downwind of sources (Jones, 1986). However, trace gas fluxes may often involve some form of chemical processing in the atmosphere downwind of the point of emission, which may contribute considerable additional uncertainty in modelled trace gas fluxes. Figure 3 illustrates some of the issues on validation of current generation CTMs against observational data for the short-lived trace gas, sulphur dioxide (SO2). The figure shows the annual average model SO2 concentration for the 5~ • 5 ~ grid square covering much of England along with the monthly mean observations for 19 monitoring stations. On this scale, there is significant spatial variability between the individual measurement sites, which in itself covers a range of up to a factor of 8. Such a range is likely to be significantly larger than the uncertainty in emissions. Furthermore, variability is significant at a finer time resolution e.g. daily or hourly. The uncertainty in coarse-resolution CTMs operating at 5 ~ x 5 ~ which approximates the state of the art CTMs, is likely close to a factor of 4 up and down for short-lived trace gases with significant ecosystem sources and sinks and existing in a complex terrain. The representation of the trace gas exchange processes in the CTMs at the ecosystem scale will introduce further uncertainties, the magnitudes of which are crucially dependent on the nature of the exchange process involved. Dry deposition processes are thought to be the simplest processes representing the concept of a dry deposition velocity. In this way, many of the problems of scaling trace gas fluxes can be side-stepped with a simple parameterization. Clearly, there is a huge gap in scale between the available dry deposition studies on the leaf or canopy scale and the coarse grid squares of the CTM. Wet deposition is a sporadic process which is difficult to describe adequately in models. The coarse spatial resolution of the models is certainly an issue but perhaps more important is their neglect of the detailed microphysical and chemical processes thought to be occurring in rain clouds. Simulated global- or regional-scale wet deposition fluxes are available with reasoSOa concentration
(ppb)
25
20 -
25 STOCHEM Ladybower Sunderland Newcastle Liverpool Leicester
-- -O-- ~ -- ~ ..... --t- ..... ...... x- .....
- -+-- --x-- - -0-- -- ~ ..... al$ ..... ...... [ 3 .....
..... O ..... Swansea
Bridge Race Bloomsbun/ Bamsley Leeds Birmingham East Southampton
...... 9..... Middlesborough
- -11(-- -- -q3- - -U-- ~ ..... < ) ..... ...... ~ , .....
Lullington Heath Birmingham Cardiff Bristol Hull Bexley
9 t~.3' ..-]r
-20
..-" /
~" I I
15-
10-
N
,.',_ 9 ~9 . \ >*.'-\ ~. i i .
~..... ~9 5 -
. o / t ,x,, x-.. ~ < . . . . ]~ ~ . ~... ~ . 0 " " " ... " x. " .~
.$1~'.'r, . .. ~. .' .. .. . . ~ .
.__ .
[]
.
..........
9 '. ~_..
. .. ~ l i .. .
. .Ik.^ .
i
~....
o..--. r F ' . : , . . . o . . . - " / i ".;~:.~".....-'" . # i . .". i " ~.. . - ' ~
~
.....i ~ . . . .~: ~
=-.-'~-:~'~--i-='.C_
i
Jan
I
Feb
~ i
Mar
" ~ - .=~ I
Apr
I
May
i
i
Jun
Jul
7-
~_.._ -~-I
Aug
I
Sept
i
Oct
15
-
10
-5
"'_-,m a:.'~.~-~," -~=~':":'-"~-.'='0
.......~ - ,, ~ _
-
9
. . .
-- . . . , , 0
-
--~ I
Nov
i
0
Dec
Month
Figure 3. Simulated concentration of SO2 using the STOCHEM model for the 5 x 50 grid square covering most of the U.K. and monthly mean observations for 19 monitoring stations. Source: Stevenson et al. (1998).
20
A.F. Bouwman, R.G. Derwent and F.J. Dentener
nable accuracy, but this accuracy deteriorates as spatial scales decrease to the catchment or landscape scales. Topography is a crucial factor in driving the orographic enhancement of wet deposition. In coarse-resolution CTMs the topography of all but the highest mountain ranges is necessarily averaged out, thus removing a major influence on model wet deposition fluxes to sensitive catchments. There is a consequential reduction in model estimates of cloud water deposition as the topography is smoothed out by model spatial resolution. Trace gas exchange with terrestrial and aquatic ecosystems is not always a one-way process, as emission and resuspension may occur simultaneously (see Conrad and Dentener, 1999). Ammonia emissions are difficult to represent accurately in models because they are sporadic and depend on local factors, which are highly variable. Soil moisture and animal husbandry are two such factors which are difficult to be specific about, but which have a significant influence on ammonia emissions (Bouwman et al., 1997). Resuspension of seasalts and wind-blown dust is often driven by high winds, which can be adequately represented in CTMs. However, the state of the terrestrial surfaces, whether wet or recently ploughed, may have a pronounced influence on resuspension, and these local factors are not often welldefined on the coarse scales used in the CTMs.
3.3. Comparison of CTMs with emission inventories With the exception of the spatial resoh".:.on of the emission inventories which meet the requirements of current CTMs, there are major inconsistencies to remain between the CTMs and the emission inventories which drive them. The most striking discrepancy between CTMs and inventories is in the temporal scale, which is generally one year for the inventories and 16 hours for the CTMs. Most CTMs include routines based on hypotheses on temporal flux distributions at the model scale, or assumptions on temporal patterns are provided with the emission inventories (see Table 4). Another way is to incorporate the trace gas flux model in the atmospheric model, as done for example in some CTMs for NOx from soils. For reactive species with short atmospheric lifetimes such as NH3, NOx and VOC, the temporal scale gap is a more serious problem than for long-lived species. An additional gap between inventories and CTMs is the number of VOC species; here, some of the mechanisms describing the chemistry in CTMs require a much larger number of species than included in current inventories. A general major problem is that it is not always possible to ensure that consistent land use and meteorological data are used throughout the modelling system including the emission inventorie~. Furthermore, there are scaling problems with all aspects of CTI~ input data, some of which are caused by limited computer resources, others by the focus of the modelling system and yet others by lack of current understanding. Turning to validation of emission inventories, the emission fields for long-lived trace gases can be tested using CTMs on the basis of concentrations, trends, and seasonality and spatial gradients of concentrations, as the chemistry is less crucial for long-lived species with fewer fluctuations over the year. For other species, deposition rates can be used to validate model results. A discussion of validation tools is, however, outside the scope of this paper. We refer to Heimann and Kaminski (1999) for a review of inverse modelling and atmospheric monitoring networks, Trumbore (1999) for a review of the use of isotopes and tracers in validation and scaling of trace gas fluxes, and to Sofiev (1999) for a discussion on validation and representativeness of measurement data. A review of the use of remote sensing techniques to determine atmospheric concentrations is given by Burrows (1999).
Towards reliable global estimates of emissions of trace gases and aerosols
4. Conclusions
and
21
recommendations
A comparison of the spatial and temporal scales of the present state-of-the-art CTMs and emission inventories for terrestrial and aquatic ecosystems indicates a wide gap when it comes to temporal resolution. The most common temporal resolution of emission inventories is one year, while CTMs describe processes with a time step of 1 to 6 hours. This discrepancy is particularly" important for gas species with a short atmospheric lifetime (less tt, an one day). It should be possible to produce estimates for most species and sources with a greater temporal resolution. However, the key problem involved in increasing the temporal resolution is the sparsity of data for use as a basis for flux estimates and a lack or even absence of independent data to validate fluxes. Available data may be appropriate to validate the temporal variability or the functional relationships between environmental conditions and fluxes. In general, it becomes increasingly difficult to find tools for validation as the level of detail of the temporal scale increases. In some cases such data are inadequate or even absent (e.g. deposition fluxes, concentrations of short-lived species). The spatial resolution of inventories in our review suggests the level of detail as being adequate for current CTMs. However, the real spatial resolution of most inventories is much lower than suggested by the 1~ reported. This is caused by the use of emission factors for biomes and functional types, and by the uncertainty and resolution of the environmental spatial data used. The major recommendations following from the examples discussed can be summarized as follows: - E m i s s i o n factor approaches. Where emissions are described with emission factor or regression approaches, variability can be used instead of the usual practice of averaging out the heterogeneity. This is done, for example, by presenting frequency distributions for regions or functional types, or the standard deviation for grid boxes,. In many cases the point-by-point uncertainty is not known. However, even the indication of the maximum and minimum values could be more helpful than the mean alone for sensitivity and quantitative uncertainty analysis. - F l u x m o d e l s . Flux models should be used where possible to replace traditional emission factor approaches. Firstly, models, which are descriptions of current process knowledge, are preferred above simple rules such as those applied in CTMs to produce temporal distributions. Secondly, intemal consistency of CTMs is improved by incorporating the flux models. In trace-gas flux models there is often an imbalance between the level of detail by which different processes are described. The relationship between scale, the model approach and the model parameters selected is very important in this respect. On a higher scale the data availability, generally poses a problem when using detailed process models. In this case, simplified or summary models are expected to interpret field experiments with limited information. The aim of simplifications is to make the model applicable to a wider area with limited data sets. Developing such ranges of models from the micro-scale to field scale and summary models to be used for extrapolation to other sites with different conditions is extremely useful. Summary models will also help to develop a better understanding of how to select the key variables to be used for specific scales. - Environmental d a t a . The spatial data on climate, oceans, soils, land cover and land use which are commonly used as a basis for scaling of trace fluxes have four general characteristics: (i) their uncertainty is regionally variable but generally unknown in the spatial distributions; (ii) data classifications are always aggregations (iii) classifications used are generally not easily translated into other classifications; (iv) classifications cannot be easily translated into properties or regulating factors of trace gas fluxes. In view of these
22
A.F. Bouwman, R. G. Derwent and F.J. Dentener
characteristics the use of common databases should be promoted. Geographic databases coupled with attribute files for the various map units distinguished is one way to at least describe the heterogeneity of the properties within a class. Examples of this approach are the soil database and the land-cover characterization discussed in this paper. Combining vegetation/land-cover data with climate and soil information may provide a basis for classification according to function. Finally, there is a need for compensation and recognition of so-called data collectors to encourage continued critical data collection, harmonization and analysis. - Functional types. Where distinct and easily identified differences in structure and composition of aquatic and terrestrial ecosystems coincide with the functions or management conditions relevant to trace-gas fluxes at the scale considered, the delineation of functionally different types or production/management systems provides a useful basis both for measurement strategies and scaling. Appropriate selection of classes may lead to reducing the number of sites to be sampled so as to derive a reliable flux estimate. Maps provide a useful basis for delineation, and in recent years remote sensing of ecosystem characteristics has been used increasingly for classification and modelling (see Estes and Loveland, 1999). Such approaches use the variability of a system or landscape instead of ignoring it, sometimes with unexpected consequences. It is very important to select appropriate stratification schemes for functional types, both for the scale of the exercise and the available spatial data. - Aggregation. Aggregation always leads to a loss of information. The variability in space is reduced and the uncertainty in the temporal patterns is increased by spatial aggregations. The problem of errors in temporal distributions as a result of spatial aggregation can be reduced by delineating functional types within a system. Scaling based on delineations with finer spatial data may be different from that derived from data with lower resolution. In general, it is better to aggregate model results than to aggregate the spatial data before modelling. Aggregation in the form of delineation of functional types as a basis for scaling generally decreases the uncertainty, and allows one to determine the uncertainties as discussed above. - Interannuai variability. Some processes in terrestrial and aquatic ecosystems show considerable year-to-year variation. Hence, in such systems with large interannual variability, inventories representing the long-range average have less value than time series of flux estimates. This paper has reviewed the uncertainties in estimating emissions from land-use-related, and natural terrestrial and aquatic sources. A comparison has also been drawn up between the available inventories and the requirements of state-of-the-art CTMs. We have shown a number of weaknesses and problems in current methods for estimating emissions. We have also presented several possibilities for improving flux estimates, hoping that these recommendations will stimulate further study and discussion on the reduction of uncertainties in flux estimates.
References
Allen, D.J., P. Kasibhatla, A.M. Thompson, R.B. Rood, B.G. Doddridge, K.E. Pickering, R.D. Hudson and S-J. Lin (1996) Transport-induced interannual variability of carbon monoxide determined using a chemistry and transport model. Journal of Geophysical Research 101:2865528669. Andreae, M.O. (1991) Biomass burning: its history, use and distribution and its impact on environmental quality and global climate. In: Levine, J.S. (Ed.) Global Biomass Burning. MIT Press, Cambridge, pp. 3-28.
Towards reliable global estimates of emissions of trace gases and aerosols
23
Asman, W.A.H. (1992) Ammonia emission in Europe: updated emission and emission variations. Report 228471008, National Institute of Public Health and the Environment, Bilthoven, Netherlands. Batjes, N.H. (1996) Total carbon and nitrogen in the soils of the world. European Journal of Soil Science 47:151-163. Batjes, N.H. (1997) A world data set of derived soil properties by FAO-Unesco soil unit for global modelling. Soil Use and Management 13:9-16. Batjes, N.H. and E.M. Bridges (1994) Potential emissions of radiatively active gases from soil to atmosphere with special reference to methane: development of a global database (WISE). Journal of Geop.hysical Research 99:16479-16489: Benkovitz, C.M. and M.A. Mubaraki (1996) Sulfur emissions from volcanic activity in 1985 and 1990. In: A.S. Lefohn, J.D. Husar and R.B. Husar (Eds.) Assessing historical global sulfur emission patterns for the period 1850 and 1990. US Department of Energy, Washington DC, pp. C1-C18. Bernsten T., I.S.A. Isaksen, W-C. Wang and X.-Z. Liang (1996) Impacts of increased anthropogenic emissions in Asia on tropospheric ozone and climate. Tellus 45B: 13-32. Bouwman, A.F. and J.A. Taylor (1996) Testing high-resolution nitrous oxide emission estimates against observations using an atmospheric transport model. Global Biogeochemical Cycles 10:307318. Bouwman, A.F., I. Fung, E. Matthews and J. John (1993) Global analysis of the potential for N20 production in natural soils. Global Biogeochemical Cycles 7:557-597. Bouwman, A.F., D.S. Lee, W.A.H. Asman, F.J. Dentener, K.W. van der Hoek and J.G.J. Olivier (1997) A global high-resolution emission inventory for ammonia. Global Biogeochemical Cycles 11:561-587. Brasseur, G.P., D.A. Hauglustaine, C. Granier, J.-F. Muller, P. Rasch, X.X. Tie and S. Walters (1997) MOZART: A global three-dimensional chemical-transport model of the atmosphere. Journal of Geophysical Research (in press). Burrows, J.P. (1999) Current and future passive remote sensing techniques used to determine atmospheric constituents. In: A.F. Bouwman (Ed.) Approaches to scaling of trace gas fluxes in ecosystems. Developments in Atmospheric Science 24. Elsevier, Amsterdam, pp. 313-347. Chin, M., D.J. Jacob, G.M. Gardner, Foreman-Fowler and P. Spiro (1996) A global three-dimensional model of tropospheric sulphate. Journal of Geophysical Research 101:18667-18690. Collins, W.J., D.S. Stevenson, C.E. Johnson and R.G. Derwent (1997) Tropospheric ozone in a global-scale three-dimensional Lagrangian model and its response to NOx controls. Journal of Atmospheric Chemistry 26:223-274. Conkright, M., S. Levitus and T. Boyer (1994) Worm Ocean Atlas 1994, Volume 1." Nutrients. NOAA Atlas NESDIS 1, US Department of Commerce, Washington, DC. Conrad, R. and F. Dentener (1999) The application of the compensation point concepts in scaling of fluxes. In: A.F. Bouwman (Ed.) Approaches to scaling of trace gas fluxes in ecosystems. Developments in Atmospheric Science 24. Elsevier, Amsterdam, pp. 203-216 Cramer, W. and A. Fischer (1996) Data requirements for terrestrial ecosystem modelling. In: B. Walker and W. Steffen (Eds.) Global change and terrestrial ecosystems, International GeosphereBiosphere programme Book Series, Cambridge University Press, Cambridge UK, pp. 529-565. Davidson, E.A. and W. Kingerlee (1997) A global inventory of nitric oxide emissions from soils. Nutrient Cycling in Agroecosystems 48:37-50. Denmead, O.T., R. Leuning, D.W.T. Griffith and C.P. Meyer (1999) Some recent developments in trace gas flux measurement. In: A.F. Bouwman (Ed.) Approaches to scaling of trace gas fluxes in ecosystems. Developments in Atmospheric Science 24. Elsevier, Amsterdam, pp. 67-84. Dentener F.J. and P.J. Crutzen (1993) Reaction of N205 on tropospheric aerosols: Impact on the global distributions of NOx, 03 and OH. Journal of Geophysical Research 98:7149-7163. EEA (1997) Corinair 94 summary report. European Topic Centre on Air Emissions. Report to the European Environment Agency, I20 pp. EPA (1993) Regional interim emission inventories (1987-1991) Report 454/R-93-021, U.S. Environmental Protection Agency, Research Triangle Park.
24
A.F. Bouwman,R.G. Derwentand F.J. Dentener
Estes, J.E. and T.R. Loveland (1999) Toward the use of remote sensing and other data to delineate functional types in terrestrial and aquatic systems. In: A.F. Bouwman (Ed.) Approaches to scaling of trace gasfluxes in ecosystems. Developments in Atmospheric Science 24. Elsevier, Amsterdam, pp. 123-150. FAO (1991) Digital soil map of the world. Worm Soil Resources Report 67. Food and Agricultural Organization of the United Nations, Rome. FAO (1995) Forest resources assessment 1990, global synthesis. Forestry Paper 124, Food and Agriculture Organization of the United Nations, Rome. Flatoy F. and O. Hov (1996) Three-dimensional model studies of the effect of NOx emissions from aircraft on ozone in the upper troposhere over Europe and the North Atlantic. Journal of Geophysical Research 101:1401-1422. Fowler, D. (1999) Experimental designs appropriate for flux determination in ten'estrial and aquatic systems. In: A.F. Bouwman (Ed.) Approaches to scaling of trace gas fluxes in ecosystems. Developments in Atmospheric Science 24. Elsevier, Amsterdam, pp. 99-121. Frolking, S.E., A.R. Mosier, D.S. Ojima, C. Li, W.J. Parton, C.S. Potter, E. Priesack, R. Stenger, C. Haberbosch, P. D6rsch, H. Flessa and K.A. Smith (1997) Comparison of N20 emissions from soils at three different agricultural sites: Simulations of year-round measurements by four models. Nutrient Cycling in Agroecosystems 52:77-105. Fung, I., J. John, E. Matthews, M. Prather, L.P. Steele, and P.J. Fraser (1991) Three dimensional model synthesis of the global methane cycle. Journal of Geophysical Research 96:13033-13065. Guenther, A., P. Zimmerman and M. Wildermuth (1994) Natural volatile organic compound emission rate estimates for U.S. woodland landscapes. Atmospheric Environment 28:1197-1210. Guenther, A., C.N. Hewitt, D. Erickson, R. Fall, C. Geron, T. Graedel, P. Harley, L. Klinger, M. Lerdau, W.A. McKay, T. Pierce, B. Scholes, R. Steinbrecher, R. Tallamraju, J. Taylor and P. Zimmerman (1995) A global model of natural volatile organic compound emissions. Journal of Geophysical Research 100:8873-8892. Heimann, M. and T. Kaminski (1999) Inverse modelling approaches to infer surface trace gas fluxes from observed atmospheric mixing ratios. ",: A.F. Bouwman (Ed.) Approache to scaling of trace gasfluxes in ecosystems. Developments in Atmospheric Science 24. Elsevier, Amsterdam, pp. 275295. Hoosbeek, M.R. and R.B. Bryant (1992) Towards the quantitative modelling of pedogenesis: a review. Geoderma 55:183-210. Huang, Y., R.L. Sass and F.M. Fisher Jr. A semi-empirical model of methane emission from flooded rice paddy soils. Global Change Biology 4:247-268. Hutchings, N.J., S.G. Sommer and S.C. Jarvis (1996) A model of ammonia volatilization from a grazing livestock farm. Atmospheric Environment 30:589-599. Jones, J.A. (1986) The seventh report of a working group on atmospheric dispersion. The uncertainty in dispersion estimates obtained from the working group models. Report NRPB-R199, National Radiological Protection Board, Chilton, Didcot, Oxfordshire, UK. Kraus, A.B., F. Rohrer, E.S. Grobler and D.H. Ehhait (1996) The global tropospheric distribution of NOx estimated by a three-dimensional chemical tracer model. Journal of Geophysical Research 101:18587-18604. Lapitan, R.L., R. Wanninkhof and A.R. Mosier (1999) Methods for stable gas flux determination in aquatic and terrestrial systems. In: A.F. Bouwman (Ed.)Approaches to scaling of trace gc's fluxes in ecosystems. Developments in Atmospheric Science. Elsevier, Amsterdam, pp. 27-66. Lee, D.S., I. K6hler, E. Grobler, F. Rohrer, R. Sausen, L. Gallardo-Klenner, J.G.J. Olivier, F.J. Dentener and A.F. Bouwman (1997) Estimations of global NOx emissions and their uncertainties. Atmospheric Environment 31:1735-1749. Leemans, R. and W.P. Cramer (1991) The IIASA database for mean monthly values of temperature, precipitation, and cloudiness on a global terrestrial grid, Research Report 91-18. International Institute for Applied Systems Analysis, Laxenburg, Austria, 62 pp. Lerner, J., E. Matthews and I. Fung (1988) Methane emission from animals: A global high-resolution data base. Global Biogeochemical Cycles 2:139-156.
Towards reliable global estimates of emissions of trace gases and aerosols
25
Levitus, S. and T. Boyer (1994a) Worm Ocean Atlas 1994, Volume 4: Temperature. NOAA Atlas NESDIS 4, US Department of Commerce, Washington DC. Levitus, S. and T. Boyer (1994b) Worm Ocean Atlas 1994, Volume 2: Oxygen. NOAA Atlas NESDIS 2, US Department of Commerce, Washington DC. Levitus, S., R. Burgett and T. Boyer (1994) Worm Ocean Atlas 1994, Volume 3: Nutrients. NOAA Atlas NESDIS 3, US Department of Commerce, Washington DC. Li, C., S. Frolking and T.A. Frolking (1992a) A model of nitrous oxide evolution from soil driven by rainfall events: I. Model structure and sensitivity. Journal of Geophysical Research 97:9759-9776. Li, Y. and J.S. Chang (1996) A three-dimensional global episodic tracer transport model 1. Evaluation of its transport processes by radon 222 simulations. Journal of Geophysical Research 101:2539125947. Loveland, T.R., D.O. Ohlen, J.F. Brown, B.C. Reed, Z. Zhu, J.W. Merchant and L. Yang and J.W. Merchant (1997) Western hemisphere land-cover progress toward a global land-cover characteristics data base. Pecora 13 Prr~reedings, American Society of Pbotogrammetry and Remote Sensing, Bethesda, MD (in press). Matson, P.A., P.M. Vitousek and D.S. Schimel (1989) Regional extrapolation of trace gas flux based on soils and ecosystems. In: M.O. Andreae and D.S. Schimel (Eds.) Exchange of trace gases between terrestrial ecosystems and the atmosphere. Wiley and Sons, Chichester, pp. 97-108. Matthews, E. (1983) Global vegetation and land use: New high-resolution data bases for climate studies. Journal of Climate and Applied Meteorology 22:474-487, Matthews, E. and I. Fung (1987) Methane emission from natural wetlands: Global distribution, area, and environmental characteristics of sources. Global Biogeochemical Cycles 1:61-86. Matthews, E., I. Fung and J. Lerner (1991) Methane emission from rice cultivation: Global distribution, area, and environmental characteristics of sources. Global Biogeochemical Cycles 5:3-24. Mosier, A.R. and W.J. Parton (1985) Denitrification in a shortgrass prairie: a modeling approach In: D.E. Caldwell, J.A. Brierley and C.L. Brierley (Eds.) Planetary Ecology. Van Nostrand Reinhold Co., New York, pp. 441-451. Moxim, W.J., H. Levy, H. and K. Kasibhatla (1996) Simulated global troposheric PAN: Its transport and impact on NOx. Journal of Geophysical Research 101:12621-12638. Mtiller, J-F. and G. Brasseur G. (1995) IMAGES: A three-dimensional chemical transport model of the global troposphere. Journal of Geophysical Research 100:16445-16490. Nevison, C.D., R.F. Weiss and D.J. Erickson III (1995) Global oceanic emissions of nitrous oxide. Journal of Geophysical Research 100:15809-15820. Olivier, J.GJ., A.F. Bouwman, C.W.M. van der Maas, J.J.M. Berdovski, C. Veldt, J.P.J. Bloos, A.J.H. Visschedijk, P.Y.J. Zandveld and J.L. Haverlag (1996) Description of Edgar Version 2.0: A set of global emission inventories of greenhouse gases and ozone depleting substances for all anthropogenic and most natural sources on a per country basis and on l~ ~ grid. Report 771060002, National Institute of Public Health and the Environment, Bilthoven, 141 pp. Olson, J., J.A. Watts and L.J. Allison (1985) Major world ecosystem complexes ranked by carbon in live vegetation: A database. Oak Ridge National Laboratory, Oak Ridge, TN, 164 pp., NDP-017. O'Neill, R.V. (1988) Hierarchy theory and global change. In: T. Rosswali R.G. Woodmansee and P.G. Risser (Eds.) Scales and global change, Wiley and Sons, Chichester, pp. 29-45. Pedersen, S., H. Takai, J. Johnsen and H. Birch (1996) Ammoniak og stov i kv~eg-, svine- og fjerkr~estalde II. Internal Report No. 65, Danish Institute of Animal Science, Horsens, Denmark, 54 PP. Penner, J.E., C.S. Atherton and T.E. Graedel (1994). Global emission~ and models of photochemically active compounds. In: R.G. Prinn (Ed.) Global Atmospheric-Biospheric Chemistry. Plenum Press, New York, pp. 223-247. Rodhe, H., J. Langner, L. Gallardo, and E. Kjellstrom, (1995). Global scale transport of acidifying pollutants. Water, Air and Soil Pollution 85:37-50. Roelofs, G-J. and J. Lelieveld (1995) Distribution and budget of 03 in the troposphere calculated with a chemistry general circulation model. Journal of Geophysical Research 100:20983-20998.
26
A.F. Bouwman,R.G. Derwentand F.J. Dentener
Schimel, D.S., W.J. Parton, F.J. Adamsen, R.G. Woodmansee, R.L. Senft and M.A. Stillwell (1986) The role of cattle in the volatile loss of nitrogen from a shortgrass steppe. Biogeochemistry 2:39-52. Seitzinger, S., J.-P. Malingreau, N.H. Batjes, A.F. Bouwman, J. Burrows, J.E. Estes, J. Fowler and R.L. Lapitan (1999) How can we best define functional types and integrate state variables and properties in time and space? In: A.F. Bouwman (Ed.) Approaches to scaling of trace gas fluxes in ecosystems. Developments in Atmospheric Science 24. Elsevier, Amsterdam, pp. 151-167. Sofiev, M. (1999) Validation of model results on different scales. In: A.F. Bouwman (Ed.) Approaches to scaling of trace gas fluxes in ecosystems. Developments in Atmospheric Science 24. Elsevier, Amsterdam, pp.233-255. Spiro, P.A., D.J. Jacob and J.A. Logan (1992) Global inventory of sulphur emissions with l~ ~ resolution. Journal of Geophysical Research 97:6023-6036. Stevenson, D.S., C.E.Johnson, W.J.Collins and R.G. Derwent (1998) .Three-dimensional (STOCHEM) model studies of regional and global scale formation of troposheric oxidants and acidifying substances. Turbulence and Diffusion Note No." 246, Atmospheric Processes Research, Met Office, Bracknell, UK. Tegen, I. and I. Fung (1995) Contribution to the atmospheric mineral aerosol load from land surface modification. Journal of Geophysical Research 100:18707-18726. Trumbore, S. (1999) Role of isotopes and tracers in scaling trace gas fluxes. In: A.F. Bouwman (Ed.) Approaches to scaling of trace gas fluxes in ecosystems. Developments in Atmospheric Science 24. Elsevier, Amsterdam, pp. 257-274. UN (1995) Strategies and policies for air pollution abatement. 1994 major review prepared under the convention of long-range transboundary air pollution. Report ECE/EB.AIR/44, United Nations/Economic Commission for Europe, Geneva, 138 pp. Unesco (1973) International classification and mapping of vegetation. Unesco, Paris, 93 pp. Veldkamp, J.G., V.F. Van Katwijk, W.S. Faber and R.J. Van De Velde (1996) Enhancements on the European land use database. Report 724001001, National Institute of Public Health and Environmental Protection, Bilthoven. 62 p. Veldt, C. (1991) Emissions of SOx, NOx, VOC and CO from East European countries. Atmospheric Environment 25A:2683-2700. Wagenet, R.J. and J.L. Hutson (1995) Consequences of scale dependency of chemical leaching models. In: A.L. EI-Kadi (Ed.) Groundwater models for resources analysis and management, CRC Press, Boca Raton FL, pp. 169-184. Wauben, V~.M.F., P.F.J. Van Velthoven and !,. Kelder (1997) A3D chemistry trar.sport model study of changes in atmospheric ozone due to aircraft emissions. Atmospheric Environment 31:18191836. Yienger, J.J. and H. Levy II (1995) Empirical model of global soil-biogenic NOx emissions. Journal of Geophysical Research 100:11447-11464.
Chapter 2
M E T H O D S F O R STABLE GAS FLUX D E T E R M I N A T I O N IN AQUATIC AND T E R R E S T R I A L SYSTEMS
R.L. Lapitan, R. Wanninkhof and A.R. Mosier
This Page Intentionally Left Blank
Approaches to scaling a trace gas fluxes in ecosystems A.F. Bouwman, editor 1999 Elsevier Science B.V.
METHODS FOR STABLE GAS FLUX DETERMINATION IN AQUATIC
AND TERRESTRIAL SYSTEMS
R.L: Lapitan l, R. Wanninkhote and A.R. Mosier 3 1Natural Resource Ecology Laboratory, Colorado State University, Fort Collins, CO 80523 USA 2 National Oceanographic and Atmospheric Administration, Miami, FL 33149 USA 3 Correspondence to: U.S. Department of Agriculture, Agricultural Research Service, P.O. Box E, Ft. Collins, CO 80522 USA
I. I n t r o d u c t i o n
Despite the world's keen awareness of the potential global warming effects of greenhouse gases, atmospheric loading of carbon dioxide (CO2), methane (CH4) and nitrous oxide (N20) from anthropogenic sources, and aquatic and terrestrial components of the biosphere continues at very high rates. Estimates made in 1994 showed increases of 1.6 ppmv yr -1 8.0 ppbv yr -1 and 0.8 ppbv yr -1 for CO2, CH4 and N20, respectively (Houghton et al., 1995). The present atmospheric concentrations of these gases and rates of increase could lead to irreversible climate change. To increase our confidence in projections of trace gas increases we must improve mass balance accounting of sources and sinks of these gases, on a large scale (> 1 km). The uncertainties in the estimates can be attributed to the wide spatial and high temporal heterogeneity at the surface (e.g. soil, vegetation, water) - atmosphere interface, inadequate accounting, and assessments of the source-sink strengths of these gases (Bouwman, 1990). The problems contributing to the latter factor are the unavailability of sensitive analytical devices for field measurements of trace gas fluxes, lack of effective sampling design for reducing variabilities in point measurements, and lack of proven mechanistic tools for reconciling flux measurements taken at different spatial and temporal scales. Additionally, it should be pointed out that existing models, used for extrapolating small-scale fluxes to regional and/or global scales, contained intrinsic uncertainties in their assumptions, parameterization, and analytical/numerical representations of the control processes that can further magnify the uncertainties of estimates. The primary considerations in the choice of method for measuring gas fluxes include the objectives of the study, type of ecosystem under study, cost, infrastructure, and logistical support, gas species in question, and availability of precise analytical instruments having appropriate response rates for measuring gas fluxes. Fluxes of trace gases from terrestrial systems have been measured by enclosure or micrometeorological techniques. Because of the enclosure and limited spatial resolution of the closed-chamber method it was found suitable for detecting small fluxes of trace gases (e.g. N20), studying processes, and identifying sources of spatial variations controlling gas fluxes (Hutchinson and Mosier, 1981; Mosier, 1989; Livingston and Hutchinson, 1995). Micrometeorological methods provide nondestructive, integrated measurements of gas fluxes over large areas, but generally require large, uniform fetch. Tower-based and airborne eddy flux correlation methods require expensive fast-response sensors and logistical support. Depending on the gas species in
30
R.L. Lapitan, R. Wanninkhof and A.R. Mosier
question, analytical instruments must be sensitively accurate to detect one-tenth of the mean concentration difference between updrafts and downdrafts for typical CO2 (-2x 10.6 kg m 2 s-l), CH4 (1 x 10-9 kg m -2 s-l), and N20 (4x 10-l~ kg m -2 s-1) fluxes (Denmead, 1979; Desjardins et al., 1993). It should be pointed out that the measured range of fluxes of CO2, CH4 and N20 gases vary among terrestrial ecosystems (Table 1). Thus, depending on the vegetation and physiographic characteristics of the system, accompanying analytical methods must be able to detect the wide range of fluxes of the gas species desired. The field methodologies for measuring trace gas fluxes in terrestrial systems have been developed and have changed little for the past 15 years. The strategies of field sampling and computational procedures for deriving trace gas fluxes using these methodologies have been well documented; such as, for chambers (Mosier, 1989; Hutchinson and Livingston, 1993; Livingston and Hutchinson, 1995), micrometeorological (Fowler and Duyzer, 1989; Denmead and Raupach, 1993; Desjardins et al., 1993), and aircraft-based (Desjardins and MacPherson, 1989; Desjardins, 1992; Desjardins et al., 1993; Choularton et al., 1995). More recent advances made in surface and atmospheric trace gas flux measurements were seen with the employment of fast-response high resolution spectrometers, such as the tunable diode laser differential absorption (TDL) and Fourier-transform infrared (FTIR) spectroscopy. Specifically, these sensitive spectrometers extend the capability of existing methods to detecting small in situ episodic fluxes and gradients of CH4 and N20 trace gases (Kolb et al., 1995). Hence, in this paper discussions of the methods of measuring gas fluxes in terrestrial systems are limited to providing a brief overview of the methods with examples taken from more recent studies employing advanced spectrometers. Descriptions of the individual methods are focussed on their merits and limitations as these would determine the suitability of the method's application to a system, given the extent of the spatial heterogeneity observed. The exchange of nonreactive trace gases between water or soil and the atmosphere are regulated by the same chemical, biological, and physical parameters. However some methods of measurement differ because aqueous systems are, in general, relatively more homogeneous than terrestrial systems and fluxes of scalar entities are often smaller. Mass balance approaches in the water column are often used to quantify air-water fluxes because micrometeorological techniques are difficult to employ. Refinements of the latter methodologies are continually being developed particularly in accurately estimating the gas transfer velocity. A fundamental treatment of the principles and governing equation of air-water gas exchange is helpful as a priori reference for discussions on how and wl~ere the different methods provide improvement in gas flux. The application of these methods to measuring gas fluxes, including problems and uncertainties, in aquatic systems is described in section 7.
2. Scales of spatial and temporal heterogeneities in gas flux measurements The scale of spatial heterogeneity in the landscape limits the extent of applicability of a method and validity of its assumptions for gas flux measurements. Since flux measurements of gases are one-dimensional, a priori knowledge of the extent of horizontal variabilities of the dynamic factors affecting surface-air gas exchange is important for accurate identification of sources of gas entities. Breaks in horizontal continuities of surface properties, spatial pattem of climate variabilities, and magnitude of gas exchanges between the surface and the atmosphere define the boundaries of natural systems. Types of land cover, mountain barriers, and sea-land transition account for the largest horizontal scale of variability in gas fluxes, extending to the order of > 1 km. It should be appreciated that gas flux measurements at this
Methods for stable gas flux determination in aquatic and terrestrial systems
31
scale are taken as the mean integrated response from the more commonly persistent features of the landscape. Atmospheric transport of gases encomp_'..ssing areas of this space scale follows large-scale atmospheric mixing phenomena, such as the vertical transport due to convective boundary layer flow, cloud mass flux, and synoptic-scale transport by trade winds and storm systems (Raupach and Finnigan, 1995). Around the transition edge between the land and marine environment, transport of gases to the atmospheric boundary layer is coupled with the heat and water exchanges associated with the cyclic land breeze - sea breeze system (Merrill, 1985). Qualitative in situ identification of gas sources has been made using hydrodynamic and radioactive tracers (Reiter, 1972;1978), but quantitatively, the temporal variations in gas concentrations and effects of atmospheric mixing (e.g., inversion) on the budget of gases in the atmospheric boundary layer can only be resolved using numerical models (Raupach, 1991; Raupach et al., 1992; Denmead et al.,1996). Within a region or at the ecosystem level, the composition and properties of the surface and surface cover align with the climate (i.e., temperature and precipitation) variabilities and land management systems. Horizontal gradient of scalar entities (e.g., heat, water vapor, and gas) can be of the order of 102 to 103 m depending on the persistence of uniform surface terrain, type of vegetation, and surface cover. Transport of gas entities downstream from the source is determined by the rate of turbulent mixing as wind blows steadily over the surface. In situ identification of the source and trajectory monitoring of gas fluxes can be accomplished using non-dispersive release gas (e.g., SF6) in conjunction with micrometeorological techniques and sensitive analytical devices (IAEA, 1992). The smallest scale of variability in gas flux measurements is of the order of < 1 m. The variations in gas flux at this space scale can be attributed to patchiness and type of ground cover, plant species composition, and differences in soil properties driving the biogeochemical processes effecting soil-air gas exchange. Transport of gas from the source can be laminar or turbulent depending on the prevailing wind field. In an unobstructed vegetative system, the sources and sinks of the gas can be inferred from the variations in gas concentration profile and vertical wind velocity within the canopy, such as following the inverse Lagrangian dispersion model developed by Raupach (1989). At this space scale, the use of enclosures permits isolation of the source of gas fluxes and eliminates the uncertainties associated with numerical calculations and estimates of the turbulent eddy transfer coefficients required by micrometeorological techniques. For reconciling flux measurements at various scales of heterogeneity, a flux footprint which describes the contributions, per unit emission, of each element in the area observed by the sensor located at a fixed height above the surface has been suggested (Shuepp et al., 1990; Horst and Weil, 1992). The flux footprint is determined by the surface properties (e.g. roughness length, vegetation height), wind speed, and atmospheric stability. Flux footprint can be obtained analytically (Schuepp et al., 1990) or numerical modelling approach (Leclerc and Thurtell, 1990; Horst and Weil, 1992); either approaches provide closely similar footprint estimates (Hargreaves et al., 1996). The estimated flux footprints for the different methods of measuring gas fluxes are given in Table 2. By weighting the area-integrated flux with the flux footprint, the sources of spatial variations in gas flux measurements from micrometeorological measurements can be identified (Hargreaves et al., 1996), potential errors and differences between micrometeorological methods can be determined (Wienhold et al., 1995), and comparative analysis between chambers and integrative methods of measuring gas fluxes can be made (Christensen et al., 1996). The temporal fluctuations in ambient atmospheric conditions can be as significant a source of uncertainties as the surface horizontal heterogeneities in gas flux measurements. Sudden,
Table 1. Annual atmospheric inputs(-)/sequestration (+), and typical observed fluxes of CO2, CH4 and N 2 0 from terrestrial and aquatic ecosystems. Numbers in parenthesis and Superscripted indicate references c. Ecosystems
Terrestrial Ecosystems - net emissions
from tropical land use - (uptake - respiration) Temperate croplands
Grassland Tropical Temperate (moist) Shortgrass Tallgrass b Pasture Savannas Tropical forest Temperate forest Tundra
Desert
CH4
COz Annual Input (1012kg y-I)
Method a
Net flux (10 "6 kg m "2 s "l)
Annual input (1012g y-l)
Method a
N20 Flux
Annual input
(10-11kg m-2 s-i)
(1012g yr-i)
Method ~
Flux (10-1 ikg m-2s-l)
-4.03 (34) 5.13 (34) A C C
0.12-0.18 (3~) 0.47-0.64 (31) 0.21-0.85 (38)
A C C A C
-0.001 - -0.14 (36) V 001-0.09 (37) S -0.02- -0.11 (37) 0.53-0.65 (3~) 0.69-0.97 (31)
0.7 (24)
C
0.017 (32)
3.2 (23)
A
0.02-0.14 (17)
A A
3.3-4.2 (26`27) 0.6-1.0 (23)
-1.4- -14.0 (l'2"s) -0.02- -0.16 (16'23)
A
-0.04- -0.12 (17)
A A
-0.003- -1.8 (28'29'3~ -0.03- -0.07 (23)
C
-4.0 (2) ,,.,.
5.6 (5) 0.4_17.1 (11) 0.4_5.6 (1~) - 10- -3 8 (19'20'21"22)
6.2 (12)
1.4 (5)
A A C D E A
0.2-0.7 (1~ 0.3-4.4 0~ -28.9(19) -57.8 (19) - 7 0 . 0 (3)
1.4(5)
t~
r~
Table 1. Continued. Ecosystems
CO2 Annual Input (1012kg y-i)
Method
CH4
Net flux (10 -6 kg m -2 s-I)
Annual input (1012g y-J)
Method
N20
Flux
Annual input
(10-1tkg m-2 s-l)
(1012g yr -I)
Method
Flux (lO-llkg m-2s-I)
Aquatic ecosystems Rice paddies
- 100(4`5,9)
Natural wetlands/ Swamps/marshlands Lake Oceans
-3 5- -84 (t2) C 7.3 (34)
-0.13
- 1- -4 (~) -0.005- -0.05 (34)
A,C C A
-361- -1100 (7'8) -600- - 7 0 0 0 (33) -11.1- -600 (~)
,..,.
-5.5- -600 (~) -0.004-
- 0 . 0 1 6 (34)
aA, closed chamber, B, open chamber, C, micrometeorology, D, aircraft-based sensors and E, convective boundary layer budget. Positive values of annual input and fluxes indicate u~take and negative values indicate emissions of the gas. v, vegetative stage, s, senescent stage. el, Denmead et al. (1979); 2, Galle et al. (1993); 3, Ritter et al. (1992); 4, Denmead (1993); 5, Seiler et al. (1984); 6, Cicerone and Oremland (1988); 7, Schtitz et al. (1989); 8, Sass et al. (1990); 9, Lauren and Duxbury (1993); 10, Keller et al. (1986); 11, Schtitz et al. (1990); 12, Aselman and Crutzen (1989); 13, Harris et al. (1982); 14, Houghton et al. (1983); 15, Lauenroth and Milchunas (1992); 17, Bronson and Mosier (1993); 19, Fan et al. (1992); 20, Mathews and Fung (1987); 21, Whalen and Reeburgh (1988); 22, Whalen and Reeburgh (1990); 23, Mosier et al. (1996); 24, Sommerfeld et al. (1993); 25, Houghton et al. (1992); 26, Steudler et al. (1989); 27, Crill (1991); 28, Schmidt et al (1988); 29, Bowden et al. (1990); 30, Brumme and Beese (1992); 31, Dugas et al. (1997); 32, Baldocchi et al. (1997); 33, Simpson et al. (1995); 34, IPCC (1995); 35, Meyers et al. (1996); 36, Mosier unpublished data; 37, Ham and Knapp (1997); 38, Ruimy et al. (1995).
r~
r~
k~
34
R.L. Lapitan, R. Wanninkhof and A.R. Mosier
Table 2. General characteristics of the current methods and corresponding formulations for calculating surface gas fluxes in terrestrial systems. Method
Footprint a (m)
Maximum instrument frequency response a'b (Hz)
Formulations for calculating gas flux (F)
Closed chamber
1
1 x 10 .3 - 3 x 10 .3
F=
Open-top chamber
1
1 x 10 .3- 3 x 10 .3
F = J (Cg(o) - Cg(i ))
Chamber
V dCg
Adt A
Mass balance
20
3 X 10 -3
1
F=
Xo Micrometeorological Eddy correlation
2 0 0 - 1 x 103
2.5 - 10.0
F=w'Cg'+ Cg p.
Flux gradients
2 0 0 - 1 x 10 3
2 X 10 -2
F=- K2
rl
1+q[3
E+
u
H c,T
du dC~ d[ln(z-d)] d[ln(z-d)] *-~'~hl
or
F =-
~cU.z dCg ~h dz
Eddy accumulation/ Eddy relaxation
2 0 0 - 1 x 10 3
Energy balance
200 - 1 xl 03
CBL budget Aircraft
3 x 10 .3
F=b(Cg -Cg) F=(R.~,-G ~Cg-C-g ) Cp(I+YXT§
3 x 104
2 x 10 .2
5 X 10 3 - 1.5 x 10 4
10.0
200
2 X 10 -2
-)
F = z L dCL - (CL+- C O)
dt
dZL.. dt
Non-isotopic tracer release ( S F 6 )
F=F~.~.6 "
dC~
dSF~
a Adapted from Denmead (1994). b Sampling rate for automated systems.
infrequent wind gusts can induce counter-gradient transport of gases, heat, and water vapor, and which can provide the greatest uncertainty in flux gradient / Bowen ratio measurements over cropland (Finnigan, 1979; Shaw et al., 1983; Meyers and Paw U, 1987) and forest (Denmead and Bradley, 1987) systems. Strong pulses in surface-air gas exchange due to episodic changes ( < l h ) in atmospheric conditions, such as a heavy downpour, may be neglected because of restricted measurements during the event. Nitric oxide (NO) is particularly susceptible, with emission rates typically increasing immediately following small rainfall events (Martin, 1996). On the other hand, CH4 uptake rates in soil are not strongly affected by small, short-term changes in temperature and soil water status (Mosier et al.,
Methods for stable gas flux determination in aquatic and terrestrial systems
35
1996). Because of these periodic short-term events, the question remains on the minimum sampling period to consider in time averaging of measurements for effectively estimating the actual magnitude of flux for the specific gas species being studied. On a larger temporal scale, appropriate time averaging has to separately consider daytime from nighttime measurements due to differential flux directions associated with different biological mechanisms governing gas (e.g., CO2) exchange (Denmead et al., 1996) and different daytime and nocturnal vertical mixing processes occurring in the atmospheric boundary layer (Merrill, 1985).
3. Recent developments in analytical methods Gas chromatography has been and still is the main analytical method commonly employed in measuring gas concentration. Its sensitivity (detection limit) depends on the type of detector installed characteristic of the gas being measured (Table 3). Gas chromatography and other commonly used analytical methods such as non-dispersive infrared analysis for measuring atmospheric gases were reviewed comprehensively by Crill et al. (1995). For methods, such as chambers and conditional sampling, that do not require fast response analytical devices for measuring instantaneous gas concentrations, gas chromatography is most appropriate for analyzing CO2, CH4 and N20. Sampling of gases using chambers can be automated for continuous monitoring of surface gas fluxes, including CH4 and N20 fluxes from rice paddies (Schtitz et al., 1989; Bronson et al., 1997). In conjunction with micrometeorological methods, however, fast-response analytical devices are required to measure gas concentrations. Fan et al. (1992) coupled a flame ionization detector (FID) detector with a tower- and an aircraft-based eddy correlation method for measuring CH4 fluxes, considering the small variance (< 5%) obtained between sensible heat fluxes taken at 20 Hz (sampling frequency of the sonic anemometer) and at 8 Hz (sampling frequency of the FID detector). They noted, however, potential underestimation of the actual magnitude of CH4 fluxes by 10% due to large noise in the signal and inadequate sensitivity of the detector to resolve < 0.1 ppbv CH4 concentrations. For CO2, open- and closed-path infrared (IR) analyzers provide adequate sensitivity and time constant for flux measurements using eddy correlation (Leuning and Moncrieff, 1990; Leuning and King, 1992), as well as flux gradient analysis (Denmead and Raupach, 1993; Wagner-Riddle et al., 1996a). For CH4 and N20, in ecosystems where surface-air exchange of these trace gases are low and instantaneous fluctuations of gas concentrations at very fine temporal resolution are required, tunable diode laser (TDL), and Fourier-transform infrared (FTIR) spectrometers offer high resolution and time constant to detect gas concentrations at pptv levels in a second or fraction of a second (Table 3). An overview of these advanced spectroscopic instruments can be found in the literature (Kolb et al., 1995). An example of a TDL currently being used for N20 measurements provides, at 10 Hz sampling rate, an instrument drift of <0.05 ppbv h ~ and precision of 2 ppbv and 0.1 ppbv at 0.1 s and 16 min. averaging time, respectively (Wienhold et al., 1996). Detailed specifications and resolutions of the current FTIR spectrometers employed in quantitative detection of trace gases are discussed by Galle et al. (1994), Griffith (1996), and Jaakkola et al. (1997). Another spectrometer called near-infrared diode lasers (NIRDL), with a detector based on high frequency modulation of the laser at a single wavelength (1.690 ~tm), showed high in situ sensitivity to CH4 at _<1 Hz response time suitable for eddy correlation and other micrometeorological techniques (Hovde et al., 1993).
q~
Table 3. Characteristics of the current analytical methods/instruments associated with each of measuring surface fluxes of CO2, CE4 and N20. Numbers in parenthesis and superscripted indicate references b. CO 2
Method
Analytical methods a'b Chamber
Detection limit
GC-FID (9) NDIR (8)
1.0 ppbv 3.0 ppmv
Flux gradients
IRGA (3,16)
0.1-3.0 ppmv
Eddy correlation
IR (~2)
N20
CH4
Response time
Analytical methods ~'b
Detection limit
Response time
Analytical methods a'b
GC-FID (10) GFCIR (19) FTIR (13)
10.0 ppbv 10.0 ppbv 2.0 ppbv
0. ls - min ls lmin
GC-ECD (~i) FTIR 03)
40.0 pptv 0.5 ppbv
0:ls - min
TDL (4,~)
5.0 ppbv
0.06 - 0.33
TDL (3,4,6,7)
6.0 pptv
0.06 - 0.1s
0.05 - 2s
TDL (5)
5.0 ppbv
0.06 - 0.3s
TDL (6,)
6.0 pptv
0.06 - 0. ls
0. l s - min 0.1s
Detection limit
Response time min lmin
Mass balance Eddy accumulation
0.03-0.20 ppmv
FTIR (2,13) NIRDL (18)
65.0 ppbv
ls Is
FTIR (2,13)
Is
Convective boundary layer budget Aircraft
IR (12) IR (12)
Non-isotopic tracer release (SF6) FTIR (2)
0.03 ppmv 0.20 ppmv
TDL (iv)
2s 0.05s GC-FID (10) GC-ECD (15) FTIR (2,13,~s)
10.0 ppbv 1.0 pptv 1.0 ppbv
0.1s - min 20s ls
GC-ECD (ll) FTIR (2,t3,~5)
< 1 ppbv
40.0 pptv 1.0 ppbv
< Is
min ls
al, chemiluminescence; ECD, electron capture detector; FID, flame ionization detector; GC, gas chromatograph; GFCIR, gas filter correlation infrared absorption analyzer; IR, infrared absorption spectroscopy; NDIR, non-dispersive infrared absorption; NIRDL, near-infrared diode laser; TDL, tunable diode laser; FTIR, fourier transform infrared spectroscopy. b 1, Desjardins et al. (1993); 2, Gosz et al. (1988); 3, Wagner-Riddle et el. (1996a); 4, Edwards et al. (1994); 5, Simpson et al. (1995); 6, Wienhold et al. (1994); 7, Wagner-Riddle et al. (1996b); 8, Komhyr et al. (1983); 9, Rasmussen and Khalil (1981); 10, Steele et al. (1987); 11, Wenthworth and Freeman (1973); 12, Desjardins and MacPherson (1989); 13, Galle et al. (1994); 14, Smith et al. (1994); 15, IAEA (1992); 16, Denmead and Raupach (1993); 17, Wienhold et al. (1993); 18, Hovde et al. (1993); 19, Harriss et al. (1985).
Methods for stable gas flux determination in aquatic and terrestrial systems
37
4. Current methods for measuring gas fluxes in terrestrial systems 4.1. Chamber technique The chamber method is still considered the method of choice for process-level studies of soil and microbiological factors controlling trace gas fluxes. The current types of chambers used vary in basal sampling area, from <1 m 2 (Hutchinson and Mosier, 1981), 15.9 m 2 (Ham et al., 1993; 1995) to 64 m 2 (Galle et al., 1994). Considerations of the basal area depend on the practicality of installing chambers as determined by the type of vegetation and the terrain of the sampling site (Livingston and Hutchinson, 1995). The two basic chamber designs are the closed ("static") and the open-top ("dynamic") system. The former system fully restricts gas, heat, and water vapor exchanges between the inside and outside the chamber. Such a scheme magnifies the changes in the concentration of the gas emitted inside the chamber. In the opentop chamber design, the concentration of the gas in the enclosed volume of air is maintained at ambient level through continuous air flow of external air, and a steady-state gas concentration gradient at the soil-air interface is established. Open chamber design is preferred over the closed chamber if extended, repeated measurements at a fixed location are desired. However, the static design provides better precision in detecting small fluxes of trace gases such as CH4 from forest soils (Crill, 1991) and N20 from agricultural systems (Hutchinson and Mosier, 1981). Gas fluxes for different chamber designs are calculated following the equations given in Table 2. Generally errors in flux measurements can be attributed to the chamber effects on perturbations of the natural conditions of the sampling site, modifications of the microclimate, pressure-induced gas flows in open chambers, and inhibiting effects of concentration build-up in closed-chamber designs. These problems and the mechanics of correcting them have been discussed in detail elsewhere (Denmead, 1979; Hutchinson and Mosier, 1981; Mosier, 1990; Hutchinson and Livingston, 1993; Livingston and Hutchinson, 1995). If all the necessary checks on the chambers have been secured, spatial average of gas fluxes obtained from several chambers aligned at grid points along the wind direction can provide area-wide estimates of surface flux density within the accuracy constrained by the magnitude of spatial variabilities observed (Christensen et al., 1996; Galle et al., 1994). With this sampling design, very close similarities between spatially-average, point measurements and integrated estimates of N20 fluxes taken by chamber and flux-gradient methods, respectively, were observed (Christensen et al., 1996).
4.2. Micrometeorologicai techniques Conventional ground-based micrometeorological methods provide one-dimensional (along the vertical axis) measures of gas, heat, and water vapor flux densities, and as such, assume absence of horizontal perturbations in the equilibrium exchange rates between the surface and the air. Generally surface flux measurements are affected by wind speed and direction, thermal stratification, and atmospheric stability. Profiles of wind velocity, air temperature, and humidity are required to derive the parameters of the gas flux equations including the parameters describing the surface roughness length (zo), eddy diffusivities for momentum (Kin), heat (KH), and water vapor (Kv), latent (H) and sensible heat (~E) fluxes, and atmospheric stability correction factors for momentum (~m) and heat (~h) transports. Hence, the limitations in resolution and real-time continuous monitoring capability of existing micrometeorological instruments are as much a major consideration in these methods as the
R.L. Lapitan, R. Wanninkhof andA.R. Mosier
38
sensitivity of analytical devices for accurately quantifying surface-atmosphere gas exchange. The theories behind each method have been elaborated in the literature (Denmead et al., 1977; Webb et al., 1980; Baldocchi et al, 1988; Fowler and Duyzer, 1989; Businger and Oncley, 1990; Desjardins et al., 1993; Denmead, 1994; Lenschow, 1995) and, in summary, the general characteristics and equations for calculating gas fluxes are shown in Table 2. The gas concentrations (Cg) are assumed to be taken from air samples pre-conditioned (e.g., through pre-heated lines) to a standard pressure and temperature or expressed as a mixing ratio with dry air. Otherwise, in situ measurements of Cg require corrections for atmospheric density variations due to H and ~,E (Wesely et al., 1989; Denmead, 1994). 4.2.1. Flux gradient
The flux gradient method calculates the flux of a gas species from measurements of its concentration at different heights from the surface. Over bare soils or bodies of water (e.g. ocean, lake), wind speed (u) approaches zero at the soil or water surface but over vegetated surfaces, u --> 0 at the canopy height (z) less the zero-plane displacement (d). The eddy
diffusivity of the gas species (Kg), corrected for ~m and ~h, Can be calculated from temperature and wind profile measurements (Paulson, 1970; Businger et al., 1971). Most applications of flux gradient technique were primarily on CO2 since commercially available, open- and/or closed-path infrared gas analyzers (Table 3) are adequately sensitive to detect profile differences in CO2 concentrations over cropland and forest systems (Denmead and Raupach, 1993; Wagner-Riddle et al., 1996a; Meyers et al., 1996). Most recently, flux gradient applica-
'~
40O0
I 'h
' t s*'I .
.
.
.
.
3000
~,
2O00
E 9a
10O0
x _=
400
~" O
3O0
Z"
2OO
100
i
,
1
0.60 -~" ,;,n '6
13.
o.40 0.20
m
ta.,
o.oo -0.20 3O o
I 0
.
8
0
r
0.40 ~
[=.
0.1~
230
2~o
270 Calendar
290
310
Day
Figure 1. Daily average N20 fluxes (Bowen ratio method) measured over a bare soil. The arrows represent occurrence of episodic events (I, II - irrigation, F = fertilization, S - sucrose addition). Gaps in CO2 and N20 flux curves represent nighttime period. Reprinted from Wagner-Riddle et al. ( 1 9 9 6 ) with permission.
39
Methods f o r stable gas f l u x determination in aquatic a n d terrestrial systems
Table 4. Fluxes of CH4 and N20 obtained from various terrestrial surfaces using TDL-or FTIR-based micrometeorological method. Surface
Method a
Mean CH 4 Flux (10 "1~kg m-2s-1)
Vegetated bog
EC-TDL EC-TDL
1.8 0.29
1 2
Pond
EC-TDL
3.6
2
-TDL
60.0
Bare soil with manure
FG-TDL FG-TDL
-0.05 -0.14
Cropland
CS-TDL EC-FTIR EC-TDL FG-FTIR FG-TDL
Rice paddies
N20 Flux (10 "11 kg m-2s-1)
Reference b
3 0.2-25.0
4 4
10.5 6.2 8.7 5.5 5.6
5 5,6 5,6 6 6
a See Table 3; CS, conditional sampling; EC, eddy correlation; FG, flux-gradient. b 1, Edwards et al. (1994) as cited in Wagner-Riddle et al. (1996b); 2, Rudd et al. (1993) as cited in Wagner-Riddle et al.
(I996b); 3, Simpson et al. (1995), as cited in Wagner-Riddle et al. (1996b); 4, Wagner-Riddle et al. (1996b); 5, Hargreaves et al. (1996); 6, Christensen et al. (1996).
tions have been extended for continuous (_< lh interval) flux determinations of CH4 (Wienhold et al., 1994; Wagner-Riddle et al., 1996a,b) and N20 (Wienhold et al., 1994; Christensen et al., 1996; Wagner-Riddle et al., 1996a,b) from profile concentrations measured with TDL and FTIR spectrometers (Hargreaves et al., I996; Christensen et al., 1996). TDL-based flux gradient (FG-TDL) method sensitively quantified the minima and maxima of N20 fluxes from non-event and episodic conditions (rainfall or substrate addition) continuously over a bare soil surface (Figure 1). Over vegetated surfaces, FG-TDL measurements of N20 flux closely agreed with FTIR-based flux gradient (FG-FTIR) measurements (Table 4). In summary from all these recent studies, the uncertainties in flux gradient estimates of gas fluxes are dictated by the accuracy of Kg estimates and existing spatial heterogeneity observed in the sampling area. Estimating Kg following the wind profile approach evidently alleviates the deficiencies of the energy balance method (see section 4.2.4). The high sensitivities of TDL and FTIR not only made CH4 and N20 flux measurements feasible using the flux gradient approach but also reduce the fetch requirement of the method. Gas sampling height intervals of
200 m to be able to detect small atmospheric gradients in gas concentrations (Denmead and Raupach, 1993). Reduction in sampling height from the surface also minimizes the potential errors associated with varying horizontal length scales (Lenschow, 1995). 4.2.2. Eddy correlation
Eddy correlation method relates the covariance of the instantaneous vertical wind velocity with the instantaneous fluctuations of gas concentration in the air, and as such, provides a direct measure of the vertical flux density of the gas from (or to in case of deposition) the underlying surface. While it is the most direct of all the micrometeorological methods for measuring gas fluxes, it can be the most difficult and technically demanding to operate considering the rigorous management of instrument operations before and during data collection. These include checks on instrument drifts, transient errors, vertical alignment, and siting geometry relative to the mast and other sensors (Businger, 1986). Also crucial in eddy
R.L. Lapitan, R. Wanninkhofand A.R. Mosier
40
correlation set-up are the sensor sampling height and fetch considerations. The minimum sensor (wind and temperature) height of 2 m requires a fetch of 200 m (Denmead and Raupach, 1993). Decrease in gas concentration gradient with height and thermal stratification affect estimates of gas flux density. Over a sampling length of time (15min - lh) determined suitably for averaging, the gas flux can be determined from the equation given in Table 2. More recent applications of eddy correlation coupled this approach to the fast response TDL and FTIR spectrometers that can sensitively measure trace concentrations of CH4 (Edwards et at., 1994; Simpson et al., 1995; Wagner-Riddle et al., 1996b) and N20 (WagnerRiddle et al., 1996a; Hargreaves et al., 1996; Christensen et al., 1996). Table 4 shows the dynamics of the TDL-based eddy correlation method (EC-TDL) for precisely determining trace gas fluxes within the spatial variabilities of gas exchanges observed in different terrestrial systems. With N20 fluxes in the cropland system, the EC-FTIR and the FG-based measurements were taken from the same mast and locations and thus, comparable. The data represent true N20 fluxes, i.e., surface integrated N20 fluxes weighted against the flux footprint (discussed more in a later section) calculated for the method, and showed close agreement of true fluxes estimated by the flux gradient and eddy correlation methods. The discrepancy between EC-TDL and EC-FTIR N20 fluxes was not accounted for by intrinsic (e.g. calibration and transient errors) differences between TDL and FTIR devices; rather, to biases associated with sampling site differences (Hargreaves et al., 1996).
4.2.3. Eddy accumulation and conditional sampling The eddy accumulation method, like the eddy correlation method, requires a fast-response wind sensor to monitor vertical wind velocity and, in addition, control the valves for switching air sampling and accumulation in the "up" (C +) and "down" ((7) reservoirs coinciding with the upward and downward drafts of wind, respectively. Air is pumped into the C + and U reservoirs at same rates proportional to the magnitude of vertical wind velocity (Desjardins, 1977). Because accumulated instead of instantaneous air samples are measured in the two reservoirs for concentrations of the desired gas species, slow-response high-resolution spectrometers can be used. The relaxed eddy accumulation method operates on the same principles and field set-up as the eddy accumulation method, except that air sampling rates at the C + and C- reservoirs are held constant (instead of linearly proportional to magnitude of vertical wind velocity); thus, simplifying flux calculations. At the end of a suitable sampling period (ts) the difference in the mean concentrations of the gas species (i.e., Cg+ and Cg-) taken from the C + and C~ reservoirs are used to calculate flux of the gas following the equations given in Table 2. The proportionality coefficient (b) is given by:
b
=
t (E ts)'
Eddyaccumulation
(1)
(Z" O"w) Relaxed eddy accumulation where e is the pump rate per unit vertical wind speed and ~w is the standard deviation of the vertical wind speed. Generally, the coefficient (z) ranged between 0.56 to 0.60 and, based on field experiments, is minimally affected by changes in atmospheric stability and turbulence intensity (Businger and Oncley, 1990; MacPherson and Desjardins, 1991; Baker et al., 1992; Ham and Knapp, 1997). Offsets in vertical wind velocity, mechanical failures of fast switches and flow rate circuitry, and low resolution of gas analytical instruments provide potential sources of errors in these techniques (Hicks and McMillen, 1984). The latter factor especially limits the
Methods for stable gas flux determination in aquatic and terrestrial systems
41
application of these techniques under conditions when the updraft and downdraft changes in the concentration of the trace gas are very small.
4.2.4. Bowen ratio-energy balance The Bowen ratio system measures air temperature (T), water vapor pressure (e), and desired gas concentration at two heights above the surface. The H and )~E components of the energy balance can be separately calculated from the finite difference (dT/dz) and (de/dz), respectively, or by residual knowing the total net radiation (Rnet), soil heat flux (G), and either H or )~E. The eddy diffusivity coefficients (e.g., KH) can be derived from the ratio between H and (dT/dz), and by invoking the similarity assumption, the diffusivity for the desired gas can be obtained. Using this assumption alleviates the need for ~m and q~h corrections in any atmospheric conditions with the energy balance approach. Gas flux can then be obtained from the product of Kg and (dCg/dz), or from the ratio of the measured gas concentration and temperature gradients given Rnet and G (Table 2). Under generally neutral conditions, good agreement between Kg estimated from flux profile measurements and Bowen ratio system was observed; except during nighttime and rainy conditions when due to low available energy, non-equivalence of flux and gradient polarities cause erroneous Bowen ratio measurements (Wagner-Riddle et al., 1996a,b). Figure ! shows a case of missing episodic CO2 and N20 fluxes (discontinuities in the curve) imposed by the constraints of the energy balance method. Some of the other uncertainties often encountered with this technique are those due to transient errors of the heat and moisture sensors, sampling errors during conditions of low available energy (Fritschen and Gay, 1979), inequalities in the exchange coefficients (Dugas et al., 1997; Meyers et al., 1996), and differences in horizontal length scale of air sampled at different heights. Sinclair et al. (1975) presented probability functions for evaluating uncertainties in Bowen ratio estimates of Kg over vegetated surfaces.
4.2. 5. Mass balance In contrast to the preceeding micrometeorological techniques, surface flux of gas using the mass balance equation (Table 2) is better evaluated if the fetch (X) is small. The mass balance method evaluates gas flux density to the height of the boundary layer, developed over the sampling site of known X, from the height integral of the product of the mean horizontal wind speed (h) and upwind gas concentration (c-ZT;,,,) corrected by the mean background concentration (c--~,) of the gas. Estimates c'f gas flux density are affected by the Zo, u, ~h, d~m, and magnitude of Cg(b) (Denmead and Raupach, 1993; Lenschow, 1995). A circular plot with the gas sampler at the center (Wilson et al., 1982; Denmead, 1983) and situated at a fixed height above the surface, where the ratio ; ' ~ / F for different atmospheric stability are similarly equal (Wilson et al., 1982;1983), have been suggested to alleviate the effects of changing wind direction and atmospheric conditions, respectively (Denmead and Raupach, 1993; Denmead, 1994). This method applies well to flux measurements of gases (e.g., NH3) that gives large Cg(u) relative to Cg(b) after nitrogen fertilizer applications (Fowler and Duyzer, 1989). It should be pointed out that the formulation for the mass balance method given in Table 2 neglects the diffusion transport of gas which can be about 10% of the contribution due to convective transport (Raupach and Legg, 1984). It may be necessary to empirically derive a correction factor to offset this discrepancy.
4.2.6. Aircraft-based measurements The micrometeorological techniques described above can all be potentially applied for aircraft
42
R.L. Lapitan, R. Wanninkhof and A.R. Mosier
1000--
3121 15:35- 16:30 C u 94/10 ~"~Hi /I / I
3/21 12:45- 13:33 C u 8110
3/21 9 : 4 5 - 11"10 C u 93/10
/
-
500
,,o%~...Hi
D'
o 358
N
1000 --
D
Ut 356
362
f"i
500
-- 9
/,
U ! 360
! 358
I_
I 360
'~
./.,,
,
I 362
!
D' U" I I_ 358 360
356
@@
-
&
I
,U
I
!
D
U
11i7 13:30- 15:00 C=" 2/10 ! I I 35-t 356 358 360
11/7 9:25 - 10:55 " 2-3110 I _1 356 358 360
~u
[' 11/7 D U 17:55 - 19:05
~u 9
" I-2/I0
356
I 358
I 360
ppmv
Figure 2a. Vertical profiles of CO2 concentrations, in the mixed layer of the atmosphere, measured by an aircraft-mounted fast-response, non-dispersive infrared analyzer on 21 March, 1991 and 7 Nov., 1992 over Iriomote Island, Japan. (The symbols D,U, Hi, and Cu represent downwind, upwind, height of the mixed layer, and cloudiness). Reprinted from Yamamoto et al. (1996) with permission.
i2
~
E~ (13 "IO :0
IO-
.m
.,a: 8
"o.
"-%.
,,,,
3i" !
2OO
i
9
9
=
!
25O
.
.
.
.
|
:3OO
.
.
.
.
Figure 2b. Vertical profiles of N20 mixing ratios in the upper troposphere and lower stratosphere measured by an aircraft-mounted tunable diode laser spectrometer. Reprinted from Wienhold et al. (1994) with permission.
Methods for stable gas flux determination in aquatic and terrestrial systems
43
mounted measurements to provide spatially-averaged, aerial assessments of gas fluxes from terrestrial and/or water surfaces (e.g. lakes, oceans). Of these methods, the eddy correlation and a modified Bowen ratio techniques have been evaluated for potential use of airborne measurements of CO2, CH4, and N20 gases (Desjardins et al., 1989; Desjardins and MacPherson, 1989). Details of aircraft-based eddy flux measurement technique, including instrumentation, problems, and corresponding corrections associated with surface flux density measurements, have been described in the literature (Lenschow 1986,1995; Desjardins et al., 1989; Desjardins and MacPherson, 1989; Mann and Lenschow, 1994). Part of the difficulties of aircraft-based measurements of trace gases have been alleviated with the use of fastresponse, open-path infrared analyzer and TDL spectrometer which can provide high resolution profiles of CO2 (Figure 2a) and N20 (Figure 2b), respectively, required for describing the vertical transport of these gases in the atmospheric boundary layer (Wienhold et al., 1993; Yamamoto et al., 1996). Still, potential errors that can be attributed to shorter horizontal flight lengths relative to the size of convective eddies associated with surface inhomogeneity, wider mesoscale variability relative to the measured scalar fluxes, and inadequately effective techniques for filtering random errors in measurements present additional complexities in aircraft-based measurements of flux at the convective boundary layer (Mann and Lenschow, 1994). 4.2. 7. Convective boundary layer budget
A way to extrapolate surface fluxes of trace gases, and other scalar entities such as beat and water vapor, to regional scale is to directly quantify the changes in the concentration of the gas in the upper, mixed layer of the convective boundary layer (CBL) of the atmosphere. CBL mass budgeting of trace gases may be accomplished by aircraft measurements along the trajectory path of convective flow from the source (Fowler, this volume) or inverse modelling approach (Raupach, 1991; Raupach et al., 1992; Denmead and Raupach, 1993). Both approaches assume build-up of gas concentration in a well mixed and clearly defined boundary layer, and entrainment of air from above the ceiling of the convective boundary layer (CBL). Gas concentrations at the CBL represent the average flux of gas species over a region from 108 - 109 klTl 2 a r e a scale. However, strong convective turbulence during the daytime can extend the height of the CBL to 103 m (Denmead and Raupach, 1993) with flux footprint extending to 104 m at the end of the day (Denmead, 1994; Denmead et al., 1996). The presence of deep convection limits the extent to which aircraft measurements can be applied. Modelling trace gas fluxes in the CBL has been extended to express the cumulative flux of the gas in terms of its near-surface concentrations (integrated CBLB); in contrast to the earlier inversemodelling approach of assessing surface fluxes from their concentrations in the mixed layer (Demnead et al., 1996). Applications of the integrated CBLB method of estimating CO2 fluxes above vegetated surfaces showed plausible agreement with locally measured fluxes of the gas (Figure 3). Refinements on the parameterization of entrainment and assumptions in the derivation of the governing equation, including parameterization of advection effects, are being undertaken to improve the accuracy of the method (Denmead et al., 1996). Additionally, attempts have been made to apply nocturnal boundary layer budgeting of CO2 fluxes; primarily since CO2 emission rates at night from respiration are higher than the daytime rates. The conceptual framework and potential problems of this method are discussed by Denmead et al. (1996; 1999). However, its utility has yet to be proven. 4.2.8. Tracer release
Good estimates of Kg are critical for accurately estimating surface gas fluxes; otherwise, it can
44
R.L. Lapitan, R. Wanninkhof and A.R. Mosier
0
~
r
!
'~
I
9
-20-
Ae
9
O~
1
//1 /
I
0 o
,..,
0
"7" -30 -
O r cO O
9
...a'-'~ -40
9
z~ A
_
IZI
o
-50 -60
-60
,,
-50
-40
b
-30
-20
-10
-
0
Measured Ic 0830-1530 (g m 2) o Bungendore pasture 1988 A Wagga wheat 1989 9Wagga wheat 1991
Figure 3. Integral fluxes of CO2 measured and estimated by the integral convective boundary layer budget (CBLIc) between 0830 and 1530 h for pasture (o) and wheat (A - 1989, 9- 1991) vegetations. Reprinted from Denmead et al. (1996) with permission.
be a significant source of uncertainties in flux-gradient and energy balance methods. Commonly, estimates of Kg are obtained using tracers (e.g. heat and water vapor) wherein respective fluxes and concentration gradients are measured simultaneously with gas concentration gradient. The use of tracers (e.g. SF 6, sulfur hexafluoride), in conjunction with sensitive analytical devices such as FTIR that has multi-component capabilities, can be used for estimating surface gas flux without the need for Kg(IAEA, 1992). The technique involves releasing SF6, of known release rate, at the same discrete points and pattern of the gas in question, and then measuring the gradients of SF6 and the gas simultaneously. Surface flux rate is then calculated from the ratio of the gradients and the tracer release rate (Table 2). To be successful, this technique requires consistent atmospheric conditions, well mixing of the tracer and gas plumes during release and sampling period, and complete tracer coverage of the area contributing to the gradient in gas concentration. Recent applications of the technique demonstrated close similarities of SF 6 tracer and NH3 gradients diurnal patterns, and suitability for measuring NH3 surface emissions (B. Galle, 1996; personal communication). This study also suggests that the pattern of dispersion and spatial coverage of the SF6 tracer can also be valuable in viewing the effective footprint under ambient atmospheric conditions during the sampling period. Another. valuable application of SF 6 tracers is in determining exchange rates of air due to leakage that modulates the build up of gas concentration inside enclosures. In a study involving megachambers, the actual rates of build-up of NH3 and N20 concentrations inside the chamber were predicted well with equations describing surface flux emissions and a SF6 decay function integrated into the equation (Mellqvist and Galle, 1998 ).
4.3. Flux footprint and reconciliation of flux measurements
The previous discussions described the unique merits of the different methods and showed that, individually, none provides more accurate quantitative descriptions of surface-air gas ex-
45
Methods for stable gas flux determination in aquatic and terrestrial systems
Table 5. Average N20 fluxes from measurements taken "~ different wind directions using chambers ~.nd different micrometeorological techniques. (Adopted from Christensen et al., 1996). Techniques
Closed chamber A = 0.008 m 2 A = 0.125 m 2 Open chamber A = 0.140 m 2 Flux Gradient
Analytical method a
GC GC
Number of days measurements
-20 -
3 7
5.25
-6
7
4.14 5.63
-26 < 1
7 2
(0.125 m 2)
-
4.50 5.61
1.6 1.6
GC FTIR b TDL
% Deviation from chamber method
Sensor height (m) (10 -ll kg m -2 s -l)
Mean
N20 flux
Eddy correlation
TDL
3.3
6.22
+31
5
Conditional sampling
FTIR
4.9
13.75
+ 145
1
aFTIR, Fourier-transform infrared spectroscopy; GC, gas chromatograph equipped with an electron capture detector; TDL, tunable diode laser. b FTIR measurement taken on the same day with TDL a~reed within 4%.
change over the other. Rather, they are complimentary and reconciling flux measurements made at various spatial and temporal scales can prove more effective in providing a broader understanding of surface flux and the control processes governing fluxes. With proper siting geometry of the chambers relative to the position of the micrometeorological sensors and wind direction, remarkably close agreement between chambers and flux gradient estimates of surface gas flux can be obtained (Christensen et al., 1996; Galle et al., 1994).This was especially the case when the surface integral flux of the gas was weighted against the flux footprint of the method (Figure 4). With the true flux obtained after weighting the upwind-integrated surface fluxes with the footprint, further comparisons indicated closely similar N20 fluxes obtained from different micrometeorological methods which, in general, compared well with chamber (closed and open designs) within the spatial variability observed in the field experiment (Table 5). Conditional sampling method gave significantly higher N20 flux but was not definitive considering only a day of measurement (Christensen et al., 1996; Hargreaves et al., 1996). From the same cropped field located in a reclaimed land in Denmark, the results from NH3 flux estimates using S F 6 tracers and FG/FTIR-based measurements demonstrated the usefulness of tracers in visualizing the actual footprint area and correcting micrometeorological techniques for comparative reconciliation of flux estimates (B. Galle, 1996; personal communication).
4.4. Future developments The advancements in laser spectroscopy have significantly elevated the levels of understanding of trace gas fluxes and the extent to which trace gas flux measurements can be carried out. The increased resolution and time constant of laser spectrometers had allowed short-term detection of small changes in fluxes, particularly of C H 4 and N 2 0 (Wagner-Riddle et al., 1996a,b; Christensen et al., 1997; Galle et al., 1994; Hargreaves et al., 1997; Simpson et al., 1998). However, the increased precision in analytical devices, although enhancing the capabilities of micrometeorological methods for detecting trace gas fluxes, does not readily translate to increased accuracy in field measurements. Oftentimes, the high spatial and temporal variabili-
46
R.L. Lapitan, R. Wanninkhof and A.R. Mosier
25O
9
o
~
n
m
.m
200
|
E
z
15o
0
i
~
m
.~
A
ann IO
onn
ql~N O
9
Nil
1011
n
II Omm mmlN
9 NI mm n n
9
z
50 0 12.00
l
I,,
I
15.00
18.00
21.00
-
24.00
Time of day on 20 August Figure 4. N 2 0 emissions detected using 0.125 m 2 closed chambers (A), open-flow chambers (e), and atmospheric gradient measured by FTIR (R), on 20 Aug., 1993, at Lammefjord regior Sealand, Denmark. Reprinted from Christensen et al. (1996) with permission.
ties inherent in the field experiment constrain the amount of information that can be inferred from flux measurements. To be able to detect very small gradients in N20, for example, flux measurements have to be averaged over longer period of times (Simpson et al., 1998). Such a scheme defeats the intent of measuring episodic fluctuations of trace gas fluxes. It therefore appears that a good understanding and possible control of the inherent variabilities in the field experiment through efficient and effective sampling designs is of utmost priority and necessary to be implemented before measurements are to be made. Weighting of the integrated upwind flux measurements of trace gases with the flux footprint have provided a step through which the scope of inferences from the experiment, such as proper siting geometry for chamber installation and reconciliation of flux measurements among chamber and micrometeorological methods, can be increased. Even with such a procedure, however, the levels of residual errors obtained remained high (Christensen et al., 1997), and therefore, has to be considered a first instead of the final step for resolving the spatial heterogeneity problem. Global estimates of trace gas fluxes cannot be made simply from extension of results from field-plot measurements of a trace gas in a system based on the land area coverage of the system. This is primarily because of the transient status of the latter and the danger of propagation and magnification of the errors obtained from field-plot measurements. Upscaling through modelling involved averaging of trace gas flux measurements across persistent land features which, although crude, has allowed inferences about the current atmospheric changes in trace gas concentrations within the constraints of measurement variabilities. Still, the key issues that need be resolved are the level of details that need be incorporated in the model at various levels of the scaling processes (Bonan, 1993), effective handling of the spatial heterogeneity, and non-linearity between biogeochemical processes and the variables involved (Jarvis, 1995). Appropriate scaling strategies appeared to be as much an important consideration in future trace gas flux studies as obtaining accurate measurements to gain a more comprehensive understanding of the contributions of aquatic and terrestrial syslems in atmospheric budget of trace gases.
Methods for stable gas flux determination in aquati, and terrestrial systems
47
5. Flux measurements across the air-water interface 5.1. Basic principles and equations Measurement of fluxes across the air-water interface is similar to those over terrestrial systems. There are several additional methods based on mass balance techniques in the relatively well mixed aqueous media. Direct flux measurements by eddy correlation (covariance), relaxed eddy accumulation, and flux gradient methods are more difficult over water because of small magnitude of the net fluxes and large corrections for interfering processes such as water flux and temperature gradients on the flux determination. Since there are some important differences between exchange from aqueous versus soil surfaces the general principles of measurements of air-water fluxes and the methodology is discussed with emphasis on the more recent methods and methods which are currently under development. The emphasis is on the exchange of CO2 but the principles are the same for any weakly soluble gas. The reader is referred to several excellent review articles and references therein for more detailed information on processes controlling air-water transfer, and discussion of the more mature measurement techniques (Liss and Merlivat, 1986; J~ihne and Monahan, 1995; Liss and Duce, 1997). The trend in air-water gas flux investigations is towards measurement on short time and small space scales to improve the mechanistic understanding of the air-sea gas exchange processes. Once the processes are better understood, extrapolation to larger scales can be done in a more robust fashion including use of remotely-sensed parameters such as wind, surface turbulence, water temperature and colour. The flux across the air-water interface (F) can be expressed in its basic form as F=kr(C w -aC,)
(2)
F = k r K o ( p X ., - p X , )
(3)
or in terms of partial pressures:
where the first two terms on the right hand side (kr) are the kinetic components, ot is the Ostwald solubility coefficient, k is the gas transfer velocity for a non-reactive gas, and r is the enhancement factor by chemical reaction of the gas at the interface. If bubble entrainment is important the formulation will change to include the supersaturation effect and gas transfer through bubbles (Woolf, 1997). The thermodynamic component (Cw - otCa) is the concentration gradient across the interface which can also be expressed in terms of partial pressures, or more correctly fugacities, as Ko (pX,,,- pXa). The fugacity is the partial pressure corrected for the non-ideality of the gas. For most atmospheric gases at ambient temperature and pressure the numerical value of the fugacity is similar (within 1%) to the partial pressure. The kinetic term k is controlled by molecular transport processes across the boundary layer. As outlined in Liss and Slater (1974), the resistance to transport (k-1) can be concentrated in the air boundary layer of approximately 1 cm thickness or in the water boundary within the top 1 mm of the water surface, depending on the solubility and reactivity of the gas. The relative importance of the air and water boundary in retarding gas transfer depends on the reactivity and solubility of the gases. For gases with (xr < 10, water side resistance will dominate and gas transfer will be controlled by processes in the water boundary layer (Figure 5). In this overview we will concentrate on these gases which include most of the greenhouse gases (e.g., CH4 and CO2) increasing in the atmosphere by anthropogenic perturbations (Table 6). Notable gases with err > 10 include water vapor, sulfuric acid, and ammonia. The gas transfer velocity is frequently parameterized with wind speed (Liss and Merlivat,
48
R.L. Lapitan, R. Wanninkhof and A.R. Mosier
10 10
5 i
|
i
i
! | |||1
,
,
,
,,,,,i
'
'
'
,
'''"1
|
I
l||l~.
=
4
1000
=_= o
"7 i._ i..
100
=.. o o
10
0.1
~=
0.01
J
0.01
0.1
,
,
i
J t ,j,I
1
i
l i Ill!
1
10
I
I
!
I I llll
100
I
~
1000
Figure 5. Ratio of liquid to air phase resistance ( rl rg"1) versus Ostwald solubility (c~). The ratio is also a weak function of surface turbulence which is not expressed in the Fig. Many gases of environmental interest (Table 6) have solubilities
1986; Wanninkhof, 1992). Although it is widely recognized that wind is not the only factor influencing k, it is the parameter that is most easily determined and it is an important contributor to surface turbulence. Gas transfer is also sensitive to surface cleanliness, and at higher wind speed to the entrainment of bubbles. Additionally, other factors such as wave age, boundary layer stability, fetch, and rain are thought to influence the kinetic control of gas flux, although the influence of these parameters is not fully understood. Wave parameters that are indicators of surface roughness such as mean square wave slope appear to have greater robustness than wind speed in predicting gas transfer under a range of environmental conditions. The methods of determining k often rely on the use of gases other than the gas of interest. Corrections to the k of the relevant gas can be performed with knowledge of the aqueous diffusion coefficient (D) or Schmidt number (Sc), defined as vD -1 where v is the kinematic viscosity of the water. The relationship between the k of two different gases is given as:
(Sc,/ )"
k,=k2~ / S c 2
(4)
where n (Schmidt number exponent) is a function of hydrodynamics. Experimental and theoretical studies show n equal to -0.66 for smooth surfaces and -0.5 for rough surfaces. The solubility of" the gas affects k when waves break, and k becomes a complex function of D, o~, bubble volume, bubble entrainment depth, and entrainment rate. Parameterizations of k in breaking wave regimes have been outlined in Asher et al. (1995) and Woolf (1997). The thermodynamic component of air-water fluxes is largely controlled by variations in pXw since, for most gases, the atmosphere is a slowly varying and well-mixed reservoir. Several aquatic processes influence pX, .... Gases (e.g., 02, CO2, DMS, and CH3Br) are produced or consumed in the biological photosynthesis or respiration cycle such that their concentrations in water vary on different time and space scales. The relative homogeneity of aqueous compared to terrestrial systems means that the flux measurements are less scale dependent. However, spatial variability does occur and temporal variability can cause the direction of the flux to change over a seasonal and sometimes shorter
49
Methods for stable gas flux determination in aquatic and terrestrial systems
Table 6. Solubilities of selected gases and temperature dependence of solubilities. Gas species
Ostwald solubility b
ct/dt c (%~
Air Concentration
Saturation levels d
SF 6 He Ne N2 H2 CO 02 Ar CH4 Kr CFC-12 CFC-11 Rn OCS N20 CO2 CC14 CH3I CH3Br DMS
6.6x 10 .3 9.4x 10 .3 l . l x l 0 -2 1.7x 10 -2 1.8x10 -2 2.5x 10 .2 3.3x10 2 3.7x10 "2 3.7x 10 .2 6.7x 10 -2 8.5x 10 -2 3.1x10 -~ 3.4x 10 -I 6.2x 10 ~ 6.9x 10 -I 9.4x 10 -I 1.4 4.9 5.2 13.3
3.0 0.0 0.3 1.3 0.4 1.4 1.5 1.5 1.7 2.0 3.6 4.1 3.3
3 ppt 5 ppm 18 ppm 78%
80-1000% 90-115%
0.04-.2 ppm 21% 1% 1.7 ppm 0.09 ppm 510 ppt 270 ppt
80-500% 75-110%
SO2 a H20
2.6x 101 inf.
90-115%
90-110% 90-110% 500%
2.7 2.5
500 ppt 0.3 ppm 360 ppm 130 ppt
80-200% 50-200%
3.4 3.0 3.3
10 ppt
50-200%
0-3%
a Nominal values from Liss and Slater (1974). bThe Ostwald solubility is the ml of gas dissolved per ml gas at in situ temperature. The values listed are for fresh water at 20~ c dot/dt [% ~ percent change in solubility per degree centigrade at 20~ d Typical saturation levels observed in surface waters in nature relative to the atmosphere.
cycles. Except in cases of extreme lateral stratification caused by strong salinity and temperature gradients, such as induced by river outflows, surface water gas concentrations are generally homogeneous on kilometer sc,~_es. However, variations in corcentration can be present on 10 to 30-m scale due to Langmuir cells, and in the ocean on 50 km scale due to mesoscale eddies. The variations in concentration often are accompanied by other surface manifestations, such as temperature anomalies, changes in chlorophyll or differences in surfactant concentrations often apparent by differences in surface roughness. Differences in environmental forcing (e.g. waves and wind speed) will cause spatial differences in fluxes as well. Waters are typically calmer near the lee shore and gas fluxes can be expected to be less. Shallow water bodies such as lakes and coastal oceans typically have steeper wave slopes and could have enhanced fluxes compared to open ocean conditions. No robust side by side comparisons of flux methodologies on different space scales have been performed such that notions on spatial variability of air-water gas fluxes is speculative. Variation of trace gas concentrations during the deliberate flux studies in lakes have been less than 5% from the mean, except close to shore where lower concentrations occasionally are encountered. This suggests that lateral variations of fluxes due to spatial variation in concentration is generally small. On global oceanic scales there are, of course, large spatial variations in trace gas fluxes due to large differences in production and consumption rates. In very broad brush, subtropical and polar gyres are on average sinks for CO2 while tropical regions are sources due to equatorial upwelling and heating of surface water. Temporal variations in fluxes on short time scales (less than a few days) are primarily controlled by changes of environmental forcing on k. Simple parameterizations of k suggest
50
R.L. Lapitan, R. Wanninkhof and A.R. Mosier
that it has a quadratic dependence on wind speed. Changes in the thermodynamic component can occur on time scales ranging from days to months by physical and biological processes. Temperature will change the solubility b~ 0-4% C -I (Table 6) thereby changing the concentration difference and flux (equation 2). Diel changes in surface water temperature can be as high as 2~ while annual changes can be as large as 25~ for inland waters and 5 -10~ over the ocean. The biological cycle exerts considerable control on temporal variations. Small diel variations in CO2 have been attributed to the photosynthesis/respiration cycle. Spring blooms can decrease concentrations of CO2 on time scales of weeks. For inland waters rapid changes in trace gas concentrations of CO2, CH4, and N20 occur over periods of days during fall overturn when bottom waters containing excess remineralization products mix to the surface. This section concentrates on methods of determining F and/or k rather than on the thermodynamic component. Several of the flux determinations, however, do rely on perturbation of the thermodynamic component. The methods described are roughly ordered in decreasing time and space scales of measurement (Table 7). The methods relying on carbon14 (14C) and radon-222 (222Rn) are well established and receive only cursory treatment. The more novel methods that rely on deliberate tracers and developmental work with the direct flux measurements are discussed in more detail.
5.2. Methods of measuring the flux across the air-water interface
5.2.1. Natural Z4C method Carbon-14 is a radioisotope with a decay constant (5) of 8270 years, and is produced in the stratosphere by spallation of 15N. Although the abundance in the atmosphere varies slightly with solar activity, it can be considered invariant for gas exchange determinations. The 14C that enters the ocean undergoes radioactive decay which causes the ocean to be slightly depleted in 14C. Assuming steady state, the decay of 14C in the ocean must equal the invasion rate such that -I
kho ('~c.-a'~co)
=
~(14c.)
(5)
where ho is the mean depth of the ocean (~- 3800 m). By measuring the 14C in the water (14Cw) and in the air (14Ca), the gas transfer velocity can be determined. The response time of this method is of the same order as the reciprocal of decay constant such that the k is applicable for millennium time scales over the global ocean. The global average k of 21 cm h l has an uncertainty of about 25%. A detailed explanation of the method can be found in Broecker and Peng (1982).
5.2.2. Bomb Z4C method Few measurements of 14C in the surface ocean are available because at the time that the measurement technique was perfected during the 1950s, the natural ~4C cycle was grossly disturbed by additional atmospheric input from nuclear bomb tests. Atmospheric values increased rapidly in the early 1960s to a factor of 2 over background values. This perturbation offered new opportunities to quantify the invasion of gases into the ocean by measuring the changes in atmospheric laC and in the upper ocean. From comprehensive measurements during global oceanographic surveys, an inventory of bomb 14C in the ocean yielded a global gas transfer velocity of 22 cm h -1 which is very similar to the natural 14C estimate (Broecker et al., 1985). The bomb laC inventory method has a response time of about a decade and can be used to obtain ocean basin estimates of gas transfer velocities. As shown in Broecker et al.
51
Methods for stable gas flux determination in aquatic and terrestrial systems Table 7. Air-water flux measurement techniques. Method
Gas
Time
Space
103 yr
10 s km 2
10 2 yr
10 7 k m 2
Natural radioactive
14C02 14C02 ' 222Rn
10 d
10 3 km 2
Deliberate tracers
SF6, 3He
30 d
103 km 2
All
<8 hr
<1 m 2
Eddy correlation, co-variance
All (CO2)
< 1 hr
1 km
Eddy accumulation
All (CO2)
< 1 hr
1 km
unproved
All
< lhr
5 km
unproved
Natural radioactive Bomb products
Helmets/enclosures regime
Gradient
Remarks
dyes, spores = non-volatile affects hydrodynamic
(1985) estimates of k vary by less than 15%, with the k for the Atlantic being slightly larger than those for the Pacific and Indian Oceans. Current observations of atmospheric 14C suggest that the rate of decrease is lower than anticipated. Oceanic flux estimate of k equal to 22 cm h l (Broecker et al., 1985) leads to suggest that the global gas transfer velocity obtained from 14C might be too high. However, Duffy and Caldeira (1995), using a 3-dimensional ocean model, concluded that the atmospheric and oceanic observations were compatible with the stated gas transfer velocity and that the apparent anomaly was caused by undersampling of the ocean. This issue will be resolved after the large number (104) of laC samples obtained from the World Ocean Circulation Experiment (WOCE) over the last decade are processed. 5.2.3. 222Rn
The 14C methods offer powerful constraints on the magnitude of global- or basin- averaged gas transfer velocities but they give little information on the variability of gas transfer on regional scales. Smaller time and space scales for the determination of k is desirable if we are to improve our mechanistic understanding of gas transfer in the field and to enable us to parameterize k with environmental forcing. The 222Rn method for determining k relies on the departure from isotopic equilibrium between the parent isotope 226Ra and its gaseous daughter, 222Rn, in the surface mixed layer of the ocean. If a water parcel is isolated, the parent and daughter will have the same activity. When in contact with the atmosphere, the 222Rn can escape. At steady state, the gas flux is proportional to the deficit of 222Rn activity (A222Rn) compared to the activity of 226Ra (A226Ra) and given as: F
=
h(A 226Ra- A 222Rn) =
k A 222Rnd-'
(6)
From Equation (6) it can be seen that the response time of the radon method is of the same order as the reciprocal of the decay constant of 222Rn; that is, 5.6 days. Attempts to relate the gas transfer of 222Rn to wind speed have limited success (Smethie et al., 1985). The main causes for the poor correlation are the variability of wind over the response time of the method, uncertainty in the analytical procedures, and the steady state assumptions which are rarely fulfilled over the ocean. Regional averages of 222Rn measurements taken over the ocean give higher k values in windy high latitude regions than in lower latitudes. Global averages are about 20% lower than the 14C observations. Roether et al. (1984) detailed the criteria necessary to obtain accurate k measurements with 222Rn. Careful measurements at a single location have yielded a consistent pattern of k with wind speed (Emerson et al., 1991). The method is not applicable for determining k in inland waters because of 222Rn emissions from the sediments.
52
R.L. Lapitan, R. Wanninkhof and A.R. Mosier
5.2. 4. Enclosure methods For inland waters enclosure methods have seen extensive use in which an air or water parcel is isolated from its surroundings and concentration changes are monitored. The chamber method (MacIntyre et al., 1995), which is also extensively used in soil and terrestrial flux studies (see section 4.1 .), consists of placing an enclosure over the water surface and measuring the change in concentration in the head space. The chamber can be either moored or free-floating with the lip just below the water surface so the water below the chamber can freely exchange with the main body of water. Different designs have been employed, including ones with removable tops, and designs in which wind is generated over the headspace with a fan. The head space can be ambient air, or it can be spiked or depleted with the gases to be measured. With the chamber method gas fluxes can be determined over a period of minutes to hours. The disadvantage of the method is that the surface turbulence controlling gas exchange under the chamber differs from the surrounding water. The gas transfer velocity is likely to be different, too. The bobbing motion of the chamber will create pressure fluctuations inside the box that can influence the gas transfer velocity as well. No systematic study of the impact of the dimensions or deployment methods have been performed but chambers appear to give values that are higher than other measurements. The chamber method is an excellent way to determine the relative rates of exchange for different gases (Conrad and Seiler, 1988). This is of particular interest to determine enhancement of exchange of acidic gases such as CO2 in alkaline waters. The gas transfer velocity is determined from the accumulation or depletion of gas in the head space under the chamber kdt=hH ! dCah (Cw-aCa )-I
(7)
where hH is the height of the chamber and C,h is the concentration in the headspace. Cw will not change with time if the water under the chamber replaces rapidly. Equation (4) can be integrated to yield k over a time interval, tl, t2: k=hH' lnl (C"' - aCah )t, ](t~ - t, )-I
L(C.. -at,,,. ),2 j
(8)
-
This method will work for most gases with water concentrations in disequilibrium with the head space. The disequilibrium can either be natural or induced by enriching or depleting the head space with the gas to be measured. The enclosure method in the water is similar to the chamber method except that the water parcel exposed to the air is isolated from the surrounding water and the concentration changes in water are measured rather than in air. Work by Torgersen et al. (1982) suggests that the inwater enclosure methods yield higher than average fluxes. This is probably because of the pumping action of the flexible sides on the water column. Determination of fluxes by this method is similar to the chamber method except that the air concentrations remain constant while water values change. A slightly different methodology involves isolating a water parcel in the water column and measuring the change in concentration of a biologically reactive gas in the container relative to that in the open water. This has been done with oxygen to estimate the biological productivity in the water column by the change in oxygen concentration contained in the bottles over time. The difference in O2 change over time between the bottle and the water column is proportional to the 02 flux (Langdon and Marra, 1995).
Methods for stable gas flux determination in aquatic and terrestrial systems
53
5.2.5. Opportunistic mass balance estimates
Several creative applications of natural perturbation of biogeochemical cycles of gases have been used to estimate gas transfer velocities. The perturbation can either be caused by changes in water temperature, or by some known biological or physical production or consumption rates. Gas transfer velocities in the sub-tropical gyres have been estimated by determining airwater disequilibrium of man-made halocarbons. During the spring, a mixed layer forms in the ocean and the surface water heats causing supersaturation of the halocarbons (Table 6). From the rate of heating and the difference in observed versus expected supersaturation caused by the temperature-induced decrease in solubility, the flux of the halocarbon can be estimated (J.H. Butler, 1996; personal communication). Similar large scale estimates can be performed using gases with known (diel) production and consumption rates such as methyl bromide, methyl iodide and DMS. Often these estimates are obtained from computer models which are tuned to reproduce the measured daily cycle of the reactive trace gases (Yvon et al., 1996). 5.2.6. Deliberate tracers
A quantitative way to perturb a mass balance is to add a known amount of gas to the water and subsequently follow the mass decrease though time. This has been done in lakes, rivers, and in the coastal ocean by adding the deliberate tracer sulfur hexafluoride (SF6) to the water. A small amount of SF6 is injected into the lake by bubbling the gas into the lake through a dispersing stone near the base of the mixed layer (epilimnion). It is mixed rapidly over one week period or less throughout the lake by wind and wave action. Once the concentration of the trace gas is homogeneous, samples are taken to determine the decrease in concentration over time. Gas transfer velocities are determined from k
=
F
(c~-ac.)
(9)
For experiments with SF6where (*Ca << Cw, F = h (dC,~/dt), and h is the effective depth of water exchanging with the atmosphere, equation (9) can be simply integrated to yield h l n / C , ,, ) ~:=(t~-t,) C..I,:
(10)
With the deliberate tracer method the optimal time interval can be chosen to determine the gas transfer velocity if frequent concentration measurements are made over time. For relating gas transfer to rapidly changing environmental forcing functions, short time intervals are advantageous. Longer time periods lead to larger concentration changes and, thus, greater precision of k measurement. For studies in unconfined systems such as rivers and oceans, concentration decreases can be caused by both gas exchange and by dispersion. Gas transfer velocities in these environments are determined by adding two tracers with different rates of escape. The initial work performed in rivers used the tracer combination of non-volatile tracer tritiated water and the volatile tracer, SSKr (Tsivoglou, 1967). The possible environmental impact of these radioactive tracers has limited its use. No suitable replacement for the non-volatile tracer has been found to date although recent work with biological spores and dyes looks promising (P.D. Nightingale, 1996; personal communication). An alternative is to use gases that have greatly different diffusion coefficients such as the gaseous tracers 3He and SF6. Dispersion will decrease the concentration of each gas but not their ratio while 3He will escape to the atmosphere at a rate about three times faster than SF6 because of its lower Sc. Taking the
54
R.L. Lapitan, R. Wanninkhof and A.R. Mosier
dispersion of the gases into account, the resulting equation for k is as follows" h
In( R'
) 82
1
(11)
H~ Scs~:6
where k3,, is the gas transfer velocity for 3He and R is equal to (3He) ( S F 6 ) "l . SC3H,, and ScsF6 are the Schmidt numbers of 3He and SF6 equal to 143 and 1065 at 20~ for seawater, respectively. The method has been used for several coastal ocean studies with surface areas of the patch growing to about 103 km 2 during a typical experiment of two to three weeks duration (Watson et al., 1991; Wanninkhof et al., 1993; Wanninkhof et al., 1997). The time interval over which the k can be determined is inversely related to the decrease in concentration, ranging from 1 day for stormy conditions with shallow mixed layer to 4 days for quiescent environments with thick mixed layer. The results from several coastal ocean studies show a good correlation with wind speed but the relationships are unique for each study area. Differences in methodology and environment (for instance, surfactant concentrations or breaking waves) are likely causes of the differences (Figure 6). The deliberate tracer methods are probably the most robust of the water side measurements since they do not perturb the hydrodynamic regime, and the environment can be chosen to optimize the scientific returns for the particular objective. The caveats to the interpretation of dual deliberate tracer results include the assumption that second order dispersion effects have negligible influence on the ratio of 3He and SF6, that the Schmidt number dependence between 3He and SF6 is known, and that the tracer patch is
80 O
70
m
60 E .o. o o to i
North Sea Georges Bank & FI shelf Steady Climatology
J
50
J J
40
I"
30
m
20
o
j.j
~"
0 0
9
0
10 0
5
U
10
( m s -1)
10
15
Figure 6. Relationship of gas transfer velocity, determinedwith the dual deliberatetracers aHe and SF6, and wind speed. The circles are a compilation of measurements performed on the North Sea (Watson et al., 1991; P.D. Nightingale, 1996; personal communication) and the squares are work on Georges Bank and the Florida Shelf (Wanninkhof et al., 1993; Wanninkhof et al., 1997). All values (k,600) have been normalized to Sc = 600 assuming n = 0.5. The curves are for reference and depict the climatological relationship and the relationship inferred for steady winds based on 14Cas suggestedby Wanninkhof(1992).
Methods for stable gas flux determination in aquatic and terrestrial systems
55
vertically well mixed and exposed to the water column. Equation (11) implicitly includes the assumption that the spread of the tracer patch is dominated by first order processes such that the ratio of the gases does not change. Eddy diffusion at the edges of the patch might invalidate this assumption. For determination of the gas exchange, samples should be taken near the center of the patch. The high cost of analysis of 3He has precluded determining how important the second order effects might be at the edges. The interpretation also assumes that the water column is well mixed, has a well defined mixed layer depth, and that the tracer patch remains exposed to the atmosphere for the duration of the experiment. Most studies to date have been performed in turbulent coastal environments where this is true. In the open ocean environment, mixed layer depths can vary over short time scales and stratification within the mixed layer is known to occur. The interpretation of the results depends on the Schmidt number exponent used. Either an exponent of 0.5 is used or the effect of bubbles on the Schmidt number is estimated based on laboratory studies of exchange of SF6 and 3He in presence of breaking waves (Asher and Wanninkhof, 1997). The effect of breaking waves on the calculated gas transfer velocity for SF6 or 3He is relatively small since they have similar solubilities. However, extrapolation of the results to gases with higher solubility (e.g., CO2) requires a large (up to 100%) and poorly quantified downward correction of k for environments with breaking waves and bubble entrainment.
5.3. Future developments: Air side measurements
A shortcoming of the methods for determining gas transfer velocities using the water side perturbation is the relatively slow response time in comparison with the variability in environmental forcing. This hampers efforts to parameterize k. Air-side measurements by eddy correlation/covariance, eddy accumulation, and flux gradient methods with measurement at 30 min. intervals may resolve this problem. These techniques have been used for CO2 in the past but, the results have been unreliable because of water vapor interference in the measurements and inability to correct for (ship's) motion. Improvement in sensors, drying techniques, and motion detectors make air side measurements over aqueous surfaces feasible. Only a limited amount of data has been obtained using the covariance technique over water. Similarly, only proof-of-concept work has been done with the gradient and accumulation techniques. The difficulty of the air side measurements over water are the relatively small magnitude of the fluxes compared to terrestrial ecosystems, and the difficulties associated with the moving platform (ship). For example, the annual average airocean CO2 flux is about 0.6 mol m 2 yr l (~ 10"9 kg m 2 S-1) and the range in C02 fluxes is about 5 times smaller than those typical for terrestrial fluxes (Table 1). New motion measuring systems on ships provide information to correct the anemometer data to obtain absolute wind speeds (referenced to the water surface). These systems generally consist of a gyrostabilized triaxial accelerometer set mounted near the ship's center of motion, a fast response triaxial accelerometer set, and triaxial angular rate unit mounted on the mast as an integral part of the anemometer base. -['he gyro system also provides pitch and roll information. For the eddy accumulation (EA) measurements this information has to be fed in real time to the valve switching unit such that the absolute up- and down-drafts are sampled in the appropriate containers. For the EA measurements small systematic errors in the determination of the vertical velocity can yield large uncorrectable errors in the flux. The advantage of the EA measurements is that the gas concentrations can be measured at leisure at very high precision in contrast with the eddy correlation measurements where gas concentrations have to be measured in real time at high frequency (> 1 Hz).
56
R.L. Lapitan, R. Wanninkhof and A.R. Mosier
60
'~,''i _ "
40
'7 t _
>,
E
-Q
E
I''
w'c'
9 z~
Net Webb
Q
' I'''
' I ' ' ' ' I ' ' ' ' I ' ' ' '
LX
m
yk
20
O
....
LX
/k
&
Q
9
9
0
,___.m
-
x Im
m
m
,.,.,.,
-20 -40
m
L
-
m
m
-
....
I ....
I,
,,,I,
,,
,I,,,
m ~1
. . . . .
I
....
-
Figure 7. Illustration of the corrections necessary to determine the CO2 flux over the ocean by co-variance using open path IR measurements. The solid squares are the measured co-variance signal w 'c ', the open triangles is the correction for heat and water vapor (the Webb correction), while the solid circles are the resulting CO2 flux. The CO2 fluxes are at the upper range encountered over the ocean. The results are based on preliminary observations during the ASGAMAGE experiment in 1996. (C. Fairall, 1996; personal communication)
Air-sea fluxes of CO2 by covariance techniques to date have been done with open path CO2 sensors. The signal is sensitive to water vapor by overlap of the CO2 and H 2 0 IR-absorption peaks, pressure broadening effects, and air density effects of H and XE on the concentration of the CO2 gas. This so-called Webb effect is typically an order of magnitude greater than the actual CO2 flux over the ocean. Figure 7 shows an example of the raw signal, the Webb effect, and the corrected signal for a recent study in the North Sea. In this case the environment was optimal for covariance measurements with very high CO2 fluxes. As the figure shows, even the sign of the corrected and uncorrected flux signal differ. At this point the air side measurements of CO2 and most other trace gases are only possible in situations with extreme partial pressure perturbation. Results within the realm of conventional geochemical constraints have been obtained for an experiment in the Great Lakes and on a tower in the North Sea (Figure 8). However, the values are about a factor of two higher than those inferred from relationships between the gas exchange and wind speed obtained from water side measurements. The question is, does there continue to be a systematic bias in the air side results or are the oceanic constraints invalid for the environment in which the air-side measurements were performed ? A way to get around the large Webb correction is to dry the gas prior to analysis. The drying necessitates a closed path sensor and lines leading from the intake near the micrometeorological sensors through the dryer and into the analyzer. Lag corrections become an important issue for the covariance and eddy accumulation measurements where a high frequency of response is necessary. The gradient method does not suffer from this constraint but spatial inhomogeneity along with the different footprint of the measurements at different heights can make the interpretation of the gradient method ambiguous.
57
Methods for stable gas flux determination in aquatic and terrestrial systems
200
.
.
.
9 150
"T L_
r
E o
,____,
100
.
.
.
.
.
.
II
Heat, CFT Climatology Steady CO eddy, lake
Zk
CO2 eddy, sea
I
'
'
'
'
i
'
'
II
II
II
'
/
--
ZX
0 0 ~0
II
/
/
s"
, .,,"
50
0
a_.. 0
!
5
U
10 1
10
(ms
)
15
2O
Figure 8. Compilation of eddy covariance measurements and controlled flux (heat) measurements versus wind speed. The solid circles are the results using heat as a proxy for gas exchange of Haussecker et al. (1995) normalized to k,600. Note, heat exchange will not be enhanced by bubble entrainment. The squares are results from work on Lake Ontario (Donelan and Drennan, 1995), and the triangles are the preliminary results from the North Sea tower work during ASGAMAGE (C. Fairall, 1996; personal communication). The two gas exchangewind speed relationships described in Figure 6 are added as reference.
An alternate method of performing direct flux measurements is the controlled flux technique in which heat is used as a proxy for a gas (Haussecker et al., 1995). In this method, the water surface is illuminated with an infrared source (controlled flux) and the dissipation of the temperature signal is measured with a sensitive infrared camera. Heat can, for these purposes, be considered a highly soluble gas with a Prandtl number (or Schmidt number for heat) of about 6. Since heat transfer is retarded in the air boundary layer, heat dissipation will occur through the water boundary layer and making the rate of dissipation through the aqueous boundary proportional to the air-water gas transfer. A summary of the measurements is shown in Figure 8. The future of gas flux measurements across the air-water interface clearly lies in the direct air-side measurements of gas fluxes. The short response time will elucidate the mechanism controlling the gas fluxes which may in turn facilitate robust extrapolations. Of particular interest are the possibilities to use remote sensing by active and passive radiometry to extrapolate gas transfer velocities and gas fluxes over large space and time scales. For process oriented studies, water side measurements will continue to be utilized. The major issues to resolve are the influence of bubbles and surfactants on air-sea gas exchange. The first step towards reconciling the apparent mismatch between the lower k values obtained with water side measurements compared to air side measurements is to perform these measurements sideby-side. The first such study has been performed as part of the European ASGAMAGE experiment from a stationary platform in the North Sea. Shipboard comparisons between deliberate tracer results and direct air-side measurements are planned for 1998 under the auspices of NOAA.
R.L. Lapitan, R. Wanninkhof and A.R. Mosier
58
Acknowledgments Salary support by NOAA/ERL for RW, from the U.S. Trace Gas Network (TRAGNET) for RL and USDA/ARS for AM during the preparation of this manuscript is gratefully acknowledged.
Nomenclature basal area of the chamber (m 2) proportionality coefficient specific heat of air at constant pressure (J g-~K-I) gas concentration in air immediately above the water surface concentration in the headspace (gm -3) Cah CBL Convective boundary layer C+ air sample reservoir in the 'up' position air sample reservoir in the 'down' position Cgas concentration (g m 3) Cg gas concentration measured in the C+reservoir/ Cg+ Sensor (g m "3) gas concentration measured in the C- reservoir/ Cg Sensor (g m -3) Cg(b) background gas concentration (g m -3) Cgo) gas concentration in the air going into the Chamber (gm -3) Cg(o) gas concentration in the air going out of the Chamber (gm -3) upwind gas concentration (g m 3) mean scalar gas concentration in the CBL (g m -3) Cc mean scalar gas concentration just above the CBL CL+ (g mE 3) gas concentration near the water surface Cw d zero-plane displacement (m) aqueous diffusion coefficient (m s-l) D water vapor pressure (gm -2) e latent heat flux (J m-2sl ) E EC eddy correlation method air flow rate (m sl ) f gas flux (g m ~ S"1) F FG flUX gradient method Fs~6 flux of SF 6 (g m "2 s"1) FTIR Fourier-transform infrared spectroscopy soil heat flux (J mZs "1) G h depth of the water in contact with the atmosphere (m) hH height of the helmet (m) mean depth of the ocean (m) ho sensible heat flux (J m-2s-l) H k gas transfer velocity for a non-reactive gas (m st ) Kg eddy diffusivity for gas (-) KH eddy diffusivity for heat (-) eddy diffusivity for momentum (-) Km solubility (g m -3) Ko eddy diffusivity for water vapor (-) Kv A b Cp Ca
C%u)
pXw pXa r Rnet Sc SF6 t T T+ TTDL X u U. w z zo Ze
partial pressures/fugacities of the gas in water partial pressures/fugacities of the gas in air enhancement factor by chemical reaction of the gas at the air-water interface (-) total net radiation (J m2s l ) Schmidt number (-) sulfur hexafluoride time (T) air temperature (K) air temperature measured by the top sensor (K) air temperature measured by the bottom sensor (K) tunable diode laser absorption spectroscopy upwind fetch (m) horizontal wind speed (m s -l) friction velocity (m s -l) vertical wind speed (m s-1) height of the sensor (m) surface roughness length (m) depth of the convective boundary layer (m)
bar indicates mean values prime indicates instantaneous deviation of the variable from the mean
Greek Symbols
c~ 13 8 c ), q ~: L d~h ~., Pa p, ~,~ x v
Ostwald solubility coefficient (-) ratio of the mean densities of dry and moist air (-) decay constant (s) pump rate per unit vertical wind speed (m 3 s-I) ratio of the mean densities of water vapor and air (-) ratio of the molecular weights of dry and moist air (-) yon Karman constant (-) latent heat of vaporization (J g-i K-i) atmospheric stability correction for heat (-) atmospheric stability correction for momentum (-) density of dry air (g m -3) density of dry air (g m -3) standard deviation of the vertical wind speed (m s-j) empirical coefficient (-) kinematic viscosity of the water (m 2 s-~)
Methodsfor stable gasflux determination in aquatic and terrestrial systems
59
References Asher, W. E., P.J. Farley, B.J. Higgins, L.M. Karle, E.C. Monahan and I.S. Leifer (1995) The influence of bubble plumes on air/seawater gas transfer velocities. Journal of Geophysical Research 101: 12027-12041. Asher W.E. and R. Wanninkhof (1997) A method for including the effects of bubble-mediated air-sea exchange processes in the analysis of purposeful dual-tracer gas transfer data. Journal of Geophysical Research (accepted). Aselman, I. and P.J. Crutzen, 1989, Freshwater wetlands: global distribution of natural wetlands and rice paddies, their net primary productivity, seasonality and possible methane emissions. Journal Atmospheric Chemistry 8:307-358. Baker, J.M., J.M. Norman and W.L. Bland (1992) Field-scale application of flux measurement by conditional sampling, Agricultural and Forest Meteorology 62:31-52. Baldocchi, D.D., B.B. Hicks and T.P. Myers (1988) Measuring biosphere-atmosphere exchanges of biologically related gases with micrometeorological methods. Ecology 69:1331-1340. Baldocchi, D.D., C.A. Vogel and B. Hall (1997) Seasonal variation of carbon dioxide exchange rates above and below a boreal jack pine forest. Agricultural and Forest Meteorology 83:147-170. Bonan, G.B. (1993) Do biophysics and physiology matter in ecosystem models? Climate Change 24:281-285. Bouwman, A.F. (1990) Exchange of greenhouse gases between terrestrial ecosystems and atmosphere. In: A.F. Bouwman (Ed.) Soils and the Greenhouse Effect.. John Wiley & Sons, Chichester, New York, Brisbane, Toronto, Singapore, pp. 61-127. Bowden, R.D., P.A. Steudler, J.M. Melilio and J.D. Aber (1990) Annual nitrous oxide fluxes from temperate forest soils in the northeastern United States. Journal of Geophysical Research 95:13997-14005. Broecker W. S. and T.-H. Peng (1982) Tracers in the Sea. Eldigio Press, Palisades, 690 pp. Broecker W. S., T.-H. Peng, G. Ostlund and M. Stuiver (1985) The distribution of bomb radiocarbon in the ocean. Journal of Geophysical Research 99: 6953-6970. Bronson, K.F. and A.R. Mosier, 1993, Nitrous oxide emissions and methane consumption in wheat and corn-cropped systems in Northeastern Colorado. In: L.A. Harper, A.R. Mosier, J.M. Duxbury and D.E. Roiston (Eds.) Agricultural Ecosystem Effects on Trace Gases and Global Climate Change. ASA Publ. 55, ASA-CSSA-SSSA, Madison, WI, pp. 133-144. Bronson, K.F., H.U. Neue, U. Singh and E.B. Abao, Jr. (1997) Automated chamber measurements of methane and nitrous oxide flux in a flooded rice soil: I. Fallow period emissions. Soil Science Society American Journal 61:988-993. Brumme, R. and F. Beese (1992) Effects of liming and nitrogen fertilization on emission of CO2 and N20 from a temperate forest. Journal of Geophysical Research 97:12851-12858. Businger, J.A., J.C. Wyngaard, Y. Izumi and E.F. Bradley (1971) Flux profile relationships in the atmospheric surface layer. Journal Atmospheric Science 28:181-189. Businger, J.A. (1986) Evaluating the accuracy with which dry deposition can be measured with current micrometeorological techniques. Journal Climate and Applied Meteorology 25:1100-1124. Businger, J.A. and S.P. Oncley (1990) Flux measurements with conditional sampling. Journal Atmosphere Ocean Technology 7:349-352. Christensen, S., P. Ambus, J.R.M. Arah, H. Clayton, B. Galle, D.W.T. Griffith, K.J. Hargreaves, L. Klemedtson, A.-M. Lind, M. Maag, A. Scott, U. Skiba, K.A. Smith, M. Welling and F.G. Wienhold (1996) Nitrous oxide emission from an agricultural field: comparison between measurements by flux chamber and micrometeorological techniques. Atmospheric Environment 30:4183-4190. Chourlarton, T.W., P.M.W. Gallagher, K.N. Bower, D. Fowler, M.S. Zahniser and A. Kaye (1995) Trace gas flux measurements at the landscape scale using boundary-layer budgets. Phil Trans. Roy. Meteorological Society London A 352:357-370. Cicerone, R.J. and R. Oremland, 1988, Biogeochemicai aspects of atmospheric methane. Global Biogeochemical Cycles 2:299-328.
60
R.L. Lapitan, R. Wanninkhofand A.R. Mosier
Conrad R. and W. Seiler (1988) Influence of the surface microlayer on the flux of nonconservative trace gases (CO, H2, CH4, N20) across the ocean-atmosphere interface. Journal Atmospheric Chemistry 6: 83-94. Crill, P.M. (1991) Seasonal patterns of methane uptake and carbon dioxide release by a temperate woodland soil. Global Biogeochemical Cycle 5:319-334. Crill, P.M., J.H. Butler, D.J. Cooper and P.C. Novelli (1995) Standard analytical methods for measuring trace gases in the environment. In: P.A. Matson and R.C. Harriss, (Eds.) Biogenic Trace Gases." Measuring Emissions from Soil and Water. Blackwell Sci, Ltd., London, Massachusetts, Australia, pp. 164-205. Denmead, O.T. (1979) Chamber method for measuring nitrous oxide emissions from soils in the field. Soil Science Society American Journal 43:89-95. Denmead, O.T. (1993) Sources and sinks of greenhouse gases in the soil-plant environment. Vegetatio 91:73-86. Denmead, O.T. (1994) Measuring fluxes of CI-h and N20 between agricultural systems and the atmosphere. In: NIAES, CH4 and N20." Global emissions and Controls from Rice Fields and Other Agricultural and Industrial Sources, pp. 209-234. Denmead, O.T. and E.F. Bradley (1987) On scalar transport in plant canopies. Irrig. Science, 8:131149. Denmead, O.T. and M.R. Raupach (1993) Methods for measuring atmospheric gas transport in agricultural and forest systems. In: L.A. Harper, A.R. Mosier, J.M. Duxbury and D.E. Rolston (Eds.) Agricultural Ecosystem Effects on Trace Gases and Global Climate Change. ASA Publ. 55, ASA-CSSA-SSSA, Madison, WI, pp. 19-43. Denmead, O.T., J.R. Freney and J.R. Simpson, 1979, Studies of nitrous oxide emission from a grass sward. Soil Science Society American Journal 43:726-728. Denmead, O.T., J.R. Simpson and J.R. Freney (1977) A direct field measurement of ammonia emission after injection of anhydrous ammonia. Soil Science Society American Journal 41:827828. Denmead, O.T., R. Leuning, D.W.T. Griffith and C.P. Meyer (1999) Some recent developments in trace gas flux measurement. In: A.F. Bouwman (Ed.) Approaches to scaling of trace gas fluxes in ecosystems. Developments in Atmospheric Science 24. Elsevier, Amsterdam, pp. 67-84. Denmead, O.T., M.R. Raupach, F.X. Dunin, H.A. Cleugh and R. Leuning (1996) Boundary layer budgets for regional estimates of scalar fluxes. Global Change Biology 2:255-264. Desjardins, R.L. (1977) Description and evaluation of a sensible heat flux detector. Boundary-Layer Meteorology 11:147-154. Desjardins, R.L. (1992) Review of techniques to measure C02 flux densities from surface and airborne sensors. In: G. Stanhill (Ed.) Monograph on Advances in Bioclimatology. Vol. 1, Springer-Verlag, New York, pp. 1-23. Desjardins, R.L. and J. I. MacPherson (1989) Aircraft-based measurements of trace gas fluxes. In: M.O. Andreae and D.S. Schimel, (Eds.) Exchange of Trace Gases between Terrestrial Ecosystems and the Atmosphere. John Wiley & Sons, Chichester, New York, Brisbane, Toronto, Singapore, pp. 135-152. Desjardins, R.L., J.I. MacPherson, P.H. Schuepp and F. Karanja (1989) An evaluation of aircraft measurements of CO2, water vapor and sensible heat. Boundary-Layer Meteorology 47:55-70. Desjardins, R.L., P. Rochete, J.I. MacPherson and E. Pattey (1993) Measurements of greenhouse gas fluxes using aircraft- and tower-based techniques. In: L.A. Harper, A.R. Mosier, J.M. Duxbury and D.E. Rolston (Eds.) Agricultural Ecosystem Effects on Trace Gases and Global Climate Change. ASA Publ. 55, ASA-CSSA-SSSA, Madison, WI, pp. 45-62. Duffy D.B. and K. Caldeira (1995) Three-dimensional model calculation of ocean uptake of bomb 14C and implications for the global budget of bomb Inc. Global Biogeochemical Cycles 9: 373-375. Dugas, W.A., D.C. Reicosky and J.R. Kiniry (1997) Chamber and micrometeorological measurements of CO2 and H20 fluxes from three Ca grasses. Agricultural and Forest Meteorology 83:113-133.
Methodsfor stable gasflux determination in aquatic and terrestrial systems
61
Duxbury, J.M., L.A. Harper and A.R. Mosier (1993) Contributions of agroecosystems to global climate change. In: L.A. Harper, A.R. Mosier, J.M. Duxbury and D.E. Rolston (Eds.) Agricultural Ecosystem Effects on Trace Gases and Global Climate Change. ASA Publ. 55, ASA-CSSA-SSSA, Madison, WI, pp. 1-18. Edwards, G.C., H.H. Nu~mann, G. den Hartog, G.W. Thurtell and G.E. Kidd (1994) Eddy correlation measurements of methane fluxes using a tunable diode laser at the Kinosheo Lake tower site during the Northern Wetlands Study (NOWES). Journal of Geophysical Research 99:1511-1517. Emerson, S., P. Quay, C. Stump, D. Wilbur and M. Knox (1991) 02, Ar, N2 and Rn-222 in surface waters of the subarctic ocean: Net biological production. Global Biogeochemical Cycle 5: 49-6. Fan, S.M., S.C. Wofsy, P.S. Bakwin, D.J. Jacob, S.M. Anderson, P.L. Kebabian, J.B. McManus and C.E. Kolb (1992) Micrometeorological measurements of CH4 and CO2 exchange between the atmosphere and subarctic tundra. Journal of Geophysical Research 97:16627-16643. Finnigan, J.J. (1979) Turbulence in waving wheat: II. structure of momentum transfer. BoundaryLayer Meteorology 16:213-236. Fowler, D. (1999) Experimental designs appropriate for flux determination in terrestrial and aquatic systems. In: A.F. Bouwman (Ed.) Approaches to scaling of trace gas fluxes in ecosystems. Developments in Atmospheric Science 24. Elsevier, Amsterdam, pp. 99-121. Fowler, D. and J.H. Duyzer (1989) Micrometeorological techniques for the measurement of trace gas exchange. In: M.O. Andreae and D.S. Schimel, (Eds.) Exchange of Trace Gases between Terrestrial Ecosystems and the Atmosphere. Wiley and Sons, Chichester, New York, Brisbane, Toronto, Singapore, pp. 189-207. Fritschen, L.J. and L.W. Gay (1979) Environmental Instrumentation. Springer Verlag, New York, 216p. Galle, B., L. Klemendtson and D.W. Griffith (1994) Application of an FTIR system for measurements of N20 fluxes using micrometeorological methods, an ultralarge chamber system and conventional field chambers. Journal of Geophysical Research 99: 16575-16583. Gosz, J.R., C.N. Dahm and P.G. Riser (1988) Long-path FTIR measurement~of atmospheric trace gas concentrations. Ecology 69:1326-1330. Griffith, D.W.T. (1996) Synthetic calibration and quantitative analysis of gas-phase FTIR spectra. Applied Spectroscopy 50:59-70. Ham, J.M. and A.K. Knapp (1997) Fluxes of CO2, water vapor, and energy from a prairie ecosystem during the seasonal transition from carbon sink to carbon source. Agricultural and Forest Meteorology (in press). Ham, J.M., C.E. Owensby and P.I. Coyne (1993) Technique for measuring air-flow and carbon dioxide flux in large, open-top chambers. Journal Environmental Quality 22:759-766. Ham, J.M., C.E. Owensby, P.I. Coyne and D.J. Bremer (1995) Fluxes of CO2 and water vapor from a prairie ecosystem exposed to ambient and elevated CO2. Agricultural and Forest Meteorology 77:73-93. Hargreaves, K.J., F.G. Wienhold, L. Klemedtson, J.R.M. Arah, I.J. Beverland, D. Fowler, B. Galle, D.W.T. Griffith, U. Skiba, K.A. Smith, M. Welling and G.W. Harris (1996) Measurements of nitrous oxide emission from an agricultural land using micrometeorological methods. Atmospheric Environment 30:1563-1571 . Harris, R.C., D.J. Sebacher and F.P. Day (1982) Methane flux in the Great Dismal Swamp. Nature 297:673-674. Harriss, R.C., E. Gorham, D.I. Sebacher, D.B. Bartlett and P.A. Flebbe (1985) Methane flux from northern peatlands. Nature 315:652-654. Haussecker H., S. Reinalt and B. Jaehne (1995) Heat as a proxi tracer for gas exchange measurements in the field: principles and technical realization. In: B. Jaehne and E. C. Monahan (Ed.)Air-Water Gas Transfer. Aeon Verlag & Studio, Hanau, pp. 405-413 Hicks, B.B. and R.T. McMillen (1984) A simulation of the eddy accumulation method for measuring pollutant fluxes. Journal Climate and Applied Meteorology 23:637-643. Horst, T.W. and J.C. Weil (1992) Footprint estimation for scalar flux measurements in the atmospheric surface layer. Boundary-Layer Meteorology 59:279-296.
62
R.L. Lapitan, R. Wanninkhofand A.R. Mosier
Houghton, J.T., B.A. Callander and S.K. Varney (1992) Climate Change, 1992, The supplementary report to the IPCC Scientific assessment, Intergovernmental panel on climate change. Cambridge Univ. Press, New York, 200 p. Houghton, J.T., L.G. Meira Filho, J. Bruce, Hoesung Lee, B.A. Callander, E. Haites, N. Harris and K. Maskell (eds) Climate Change 1994. Radiative Forcing of Climate Change and an Evaluation of the IPCC IS92 Emission Scenarios. Cambridge University Press, Cambridge, UK, 337 pp. Houghton, R.A., J.E. Hobbie, J.M. Melilio, B. Moore, B.J. Peterson, G.R. Shaver and G.M. Woodwell (1983) Changes in the carbon content of t,~rrestrial biota and soils between 1860 and 1980: a net release of CO2 to the atmosphere. Ecological Monograph 53:235-262. Hovde, D.C., J.A. Silver and A.C. Stanton (1993) Measuring atmospheric methane and water vapor using near-infrared diode lasers. In: A. Fried, D.K. Killinger and H.I. Schiff (Eds.) Tunable Diode Lase Spectroscopy, Lidar, and DIAL Techniques for Environmental and Industrial Measurements. Proceedings Society Photo-Optical Instrumentation Engineers, Vol. 2112, Washington, pp. 110117. Hutchinson, G.L. and G.P. Livingston (1993) Use of chamber systems to measure trace gas fluxes. In: L.A. Harper, A.R. Mosier, J.M. Duxbury and D.E. Rolston (Eds.) Agricultural Ecosystem Effects on Trace Gases and Global Climate Change. ASA Publ. 55, ASA-CSSA-SSSA, Madison, WI, pp. 63-78. Hutchinson, G.L. and A.R. Mosier (1981) Improved soil cover method for field measurement of nitrous oxide fluxes. Soil Science Society American Journal 45:311-316. IAEA (1992) Manual on measurement of methane and nitrous oxide emissions from agriculture. IAEA-TECDOC-674, Austria, 91 pp. Jaakkola, P., J.D. Tate, M. Paakkinainen, J. Kauppinen and P. Saarinen (1997) Instrumental resolution considerations for Fourier transform infrared gas-phase spectroscopy. Applied Spectroscopy 51:1159-1169. J~ihne B. and E.C. Monahan (1995) Air-Water Gas Transfer, Selected papers from the third international symposium on air-water gas transfer. Aeon Verlag & Studio, Hanau, Germany, 900 pp. Jarvis, P.G. (1995) Scaling processes and problems. Plant, Cell and Environment 18:1079-1089. Keller, M., W.A. Kaplan and S.C. Wofsy (1986) Emissions of N20, CH4 and CO2 from tropical forest soils. Journal of Geophysical Research 91:11791-11802. Kolb, C.E., J.C. Wormhoudt and M.S. Zahniser (1995) Recent advances in spectroscopic instrumentation for measuring stable gases in the natural environment. In: P.A. Matson and R.C. Harriss (Eds.) Biogenic Trace Gases." Measuring Emissions from Soil and Water. Blackwell Sci, Ltd, London, Massachusetts, Australia, pp. 259-290. Komhyr, W.D., L.S. Waterman and W.R. Taylor (1983) Semiautomatic nondispersive infrared analyzer apparatus for CO2 air sample analyses. Journal of Geophysical Research-Atmospheres 90:5567-5596. Langdon C. and J. Marra (1995) Measurements of net and gross O2 production, dark O2 respiration, and 14C assimilation at the Marine Light-Mixed Layers site (59~ 21 ~ in the northeast Atlantic Ocean. Journal of Geophysical Research 1tl0: 6645-6653. Lauren, J.G. and J.M. Duxbury (1993) Methane emissions from flooded rice amended with a green manure. In: L.A. Harper, A.R. Mosier, J.M. Duxbury and D.E. Rolston (Eds.) Agricultural Ecosystem Effects on Trace Gases and Global Climate Change. ASA Publ. 55, ASA-CSSA-SSSA, Madison, WI, pp. 183-192. Lauenroth, W.K. and D.G. Milchunas (1992) Short-grass steppe. In: R.T. Coupland (Ed.) Ecosystem of the Worm 8A: Natural Grasslands, Introduction and Western Hemisphere. Elsevier, London, New York, Tokyo, pp. 183-226. Leclerc, M.Y. and G.W. Thurtell (1990) Footprint prediction of scalar flux using a Markovian analysis. Boundary-Layer Meteorology 53:247-258. Lenschow, D.H. (1986) Aircraft measurements in the boundary layer. In: D.H. Lenschow (Ed.) Probing the Atmospheric Boundary Layer. American Meteorological Society, Boston, Massachusetts, pp. 29-55.
Methodsfor stable gasflux determination in aquatic and terrestrial systems
63
Lenschow, D.H. (1995) Micrometeorological techniques for measuring biosphere-atmosphere trace gas exchange. In: P.A. Matson and R.C Harriss (Eds.) Biogenic Trace Gases: Measuring Emissions from Soil and Water. Blackwell Sci, Ltd, London, Massachusetts, Australia, pp. 126163. Leuning, R. and K.M. King (1992) Comparison of eddy-covariance measurements of CO2 fluxes by open- and closed-path CO2 analyzers: Corrections for analyzer water vapor sensitivity and damping of fluctuations in air sampling tubes. Boundary-Layer Meteorology 53:63-76. Leuning, R. and J. Moncrieff (1990) Eddy-covariance CO2 flux measurements using open- and closed-path CO2 analyzers. Boundary-Layer Meteorology 59:297-311. Liss, P.S. and R.A. Duce (1997) The sea surface and global change. Cambridge University Press, Cambridge, pp. 519. Liss, P.S. and L. Merlivat (1986) Air-sea gas exchange rates: Introduction and synthesis, In: P. BuatMenard (Ed.) The Role of Air-Sea Exchange in Geochemical Cycling. Reidel, Boston, pp. 113-129. Liss, P.S. and P.G. Slater (1974) Flux of gases across the air-sea interface. Nature 247: 181-238, 1974. Livingston, G.P. and G.L. Hutchinson (1995) Enclosure-based measurements of trace gas exchange: applications and sources of errors. In: P.A. Matson and R.C. Harriss (Eds.) Biogenic Trace Gases: Measuring Emissions from Soil and Water. Blackwell Science Ltd, London, Massachusetts, Australia, pp. 14-51. Maclntyre S., R. Wanninkhof and J. P. Chanton (1995) Chapter 3, Trace gas exchange across the airwater interface in freshwater and coastal marine environments. In: P. A. Matson and R. C. Harriss (Ed.)Methods in Ecology- Biogenic Trace Gases." Measuring Emissions from Soil and Water. Blackwell Science, Cambridge, Mass, pp. 52-97 MacPherson, J.I. and R.L. Desjardins (1991) Airborne tests of flux measurements by the relaxed eddy accumulation technique. In: Proc. 7th AMS Symposium on Meteorological Observations and Instrumentation. Amer. Meteorological Society, New Orleans, LA, pp. 6-11. Mann, J. and D.H. Lenschow (1994) Errors in airborne flux measurements. Journal of Geophysical Research 99:14519-14526. Martin, R.E. (1996) Controls of annual emission of nitric oxide from soils of the Colorado shortgrass steppe. M.S. Thesis, Colorado State University, 81 p. Matthews, E. and I. Fung (1987) Methane emission from natural wetlands: Global distribution, area, and environmental characteristics of sources. Global Biogeochemical Cycles 1:61-86. Mellvqist, J. and B. Galle (1998) Application of open path DOAS and FTIR for a static chamber approach to measure emissions ofNH3, N20, and NO from a fertilized field. (submitted). Merrill, J.] '. (1985) Atmospheric pathways to the oceans. In: P. Buat-Menard (Ed.) The Role of AirSea Exchange in Geochemical Cycling. D. Reidel Publishing Company, Dordrecht, Bopston, Lancaster, Tokyo, pp. 35-63. Meyers, T.P. and K.T. Paw U (1987) Modeling the plant canopy micrometeorology within higherorder closure principles. Agricultural and Forest Meteorology 41:143-163. Meyers, T.P., M.E. Hall, S.E. Lindberg and K. Kim (1996) Use of the modified Bowen ratio technique to measure fluxes of trace gases. Atmospheric Environment 30:3321-3329. Mosier, A.R. (1989) Chamber and isotope techniques. In: M.O. Andreae and D.S. Schimel, (Eds.) Exchange of Trace Gases between Terrestrial Ecosystems and the Atmosphere. John Wiley & Sons, Chichester, New York, Brisbane, Toronto, Singapore, pp. 175-187. Mosier, A.R. (1990) Gas flux measurement techniques with special reference to techniques suitable for measurements over large ecologically uniform areas. In: A.F. Bouwman, (Ed.) Soils and the Greenhouse Effect.. John Wiley, Chichester, New York, Brisbane, Toronto, Singapore, pp. 289301. Mosier, A.R., W.J. Parton, D.W. Valentine, D.S. Ojima, D.S. Schimel and J.A. Delgado (1996) CH4 and N20 fluxes in the Colorado shortgrass steppe: 1. impact of landscape and nitrogen addition. Global Biogeochemical Cycles 10:387-39 ~ Paulson, C.A. (1970) The mathematical representation of wind speed and temperature profiles in the unstable atmospheric surface layer. Journal Applied Meteorology 9:857-861.
64
R.L. Lapitan, R. Wanninkhofand A.R. Mosier
Rasmussen, R.E. and M.A.K. Khalil (1981) Tropospheric methane (CH4): trends and seasonal cycles. Journal of Geophysical Research- Atmospheres 86:9826-9832. Raupach, M.R. (1989) Applying Lagrangian method for relating scalar concentrations to source distributions from concentration profiles in plant canopies. Agricultural and Forest Meteorology 47:85-108. Raupach, M.R. (1991) Vegetation atmosphere interaction in homogeneous and heterogeneous terrain: some implications of mixed-layer dynamics. Vegetatio 91:105-120. Raupach, M.R. and J.J. Finnigan (1995) Scale issues in boundary-layer meteorology: surface energy balances in heterogeneous terrain. Hydrological Processes 9:589-612. Raupach, M.R. and B.J. Legg (1984) The uses and limitations of flux-gradient relationships in micrometeorology (wind tunnel observations]. Agricultural Water Management 8 : 119-131 Raupach, M.R., O.T. Denmead and F.X. Dunin (1992) Challenges in linking atmospheric CO2 concentrations to fluxes at local and regional scales. Australian Journal Botany 40:697-716. Reiter, E.R. (1972) Atmospheric transport processes: Part III., Hydrodynamic tracers. TID No. 25731, National Technical Information Service. Reiter, E.R. (1978) Atmospheric transport processes: Part IV. Radioactive tracers. TID No. 27114, National Technical Information Service. Ritter, J.A., J.D.W. Barrick, G.W. Sachse, G.L. Gregory, M.A. Woerner, C.E. Watson, G.F. Hill and J.E. Collins (1992) Airborne flux measurements of trace species in an Arctic boundary layer. Journal of Geophysical Research 97:16601-16625. Roether W. and B. Kromer (1984) Optimum application of the radon deficit method to obtain air-sea gas exchange rates. In: W. Brutsaert and G. H. Jirka (Ed.) Gas Transfer at Water Surfaces. Reidel, Hingham, Mass., pp. 447-457 Ruimy, A., P.G. Jarvis, D.D. Baldocchi and B. Saugier (1995) Fluxes over plant canopies and solar radiation: a literature review. Advances in Ecological Research 26:1-68. Sass, R.L., F.M. Fisher and P.A. Harcombe (1990) Methane production and emission in a Texas rice field. Global Biogeochemical Cycles 4:47-68. Schlesinger, W.H. (1984) Soil organic matter: a source of atmospheric CO2. In: G.M. Woodwell (Ed.) The role of Terrestrial Vegetation in the Global Carbon Cycle." Measurement by Remote Sensing. Wiley and Sons, Chichester, New York, Brisbane, Toronto, Singapore, pp. 111-127. Schmidt, J., W. Seiler and R. Conrad (1988) Emission of nitrous oxide from temperate forest soils into the atmosphere. Journal Atmospheric Chemistry 1:95-115. Schuepp, P.H., M.Y. Leclerc, J.I. MacPherson and R.L. Desjardins (1990) Footprint prediction of scalar fluxes from analytical solutions of the diffusion equation. Boundary-Layer Meteorology 50:355-376. Schiitz, H., W. Seiler and R. Conrad (1989) Processes involved in formation and emission of methane in rice paddies. Biogeochemistry 7:33-53. Schiitz, H., W. Seiler and H. Rennenberg (1990) Soil and land use related sources and sinks of methane (CH4) in the context of the global methane budget. In: A.F. Bouwman (Ed.) Soils and the Greenhouse Effect. John Wiley and Sons, Chichester, New York, Brisbane, Toronto, Singapore, pp. 269-285. Seiler, W., A. Holzapfel-Pschom, R. Conrad and D. Scharfee (1984) Methane emission from rice paddies. Journal Atmospheric Chemistry 1:241-268. Shaw, R.H., J. Tavangir and D.P. Ward (1983) Structure of the Reynolds stress in a canopy layer. Journal Climate and Applied Meteorology 22:1922-1931. Simpson, I.J., G.W. Thurtell, G.E. Kidd, M. Lin, T.H. Demetriades-Shah, I.D. Flitcroft, E.T. Kanemasu, D. Nie, K.F. Bronson and H.U. Neue (1995) Tunable diode laser measurements of methane fluxes from an irrigated rice paddy field in the Philippines. Journal of Geophysical Research 100:7283-7290. Simpson, I.J., G.C. Edwards, G.W. Thurtell, G. den Hartog, H.H. Neuman and R.M. Staebler (1998) Micrometeorological measurements of methane and nitrous oxide exchange above a boreal aspen stand, Journal of Geophysical Research (accepted).
Methodsfor stable gasflux determination in aquatic and terrestrial systems
65
Sinclair, T.R., L.H. Allen, Jr. and E.R. Lemon (1975) An analysis of errors in the calculation of energy flux densities above vegetation by a Bowen-ratio profile method. Boundary-Layer Meteorology 8:129-140. Smethie W. M., T.T. Takahashi, D.W. Chipman and J.R. Ledwell (1985) Gas exchange and CO2 flux in the tropical Atlantic Ocean determined from 222Rn and pCO2 measurements. Journal of Geophysical Research 90: 7005-7022. Smith, K.A., A. Scott, B. Galle and L. Klemendtson (1994) Use of a long-path infra-red monitor for measurement of nitrous oxide flux from soil. Journal of Geophysical Research 99:16585-16592. Sommerfeld, R.A.A.R. Mosier and R.C. Musselman (1993) CO2, CH4 and N20 flux through a Wyoming snowpack and implications for global budgets. Nature 361:140-142. Steele, L.P., P.J. Fraser and R.A. Rasmussen (1987) The global distribution of methane in the troposphere. Journal Atmospheric Chemistry 5:125-171. Steudler, P.A., R.D. Bowden, J.M. Melilio and J.D. Aber (1989) Influence of nitrogen fertilization on methane uptake in temperate forest soils. Nature 341:314-316. Torgersen T., G. Mathieu, R. H. Hesslein and W. S. Broecker (1982) Gas exchange dependency on diffusion coefficient: direct 222Rn and 3He comparisons in a small lake. Journal of Geophysical Research 87: 546-556. Tsivoglou E.C. (1967) Tracer measurements of stream reaeration. Fed. Water Pollution Control Administration, U.S. Dept. of Interior, Washington D.C. Wagner-Riddle, C., G.W. Thurtell, K.M. Ling, G.E. Kidd and E.G. Beauchamp (1996a) Nitrous oxide and carbon dioxide fluxes from a bare soil using a micrometeorological approach. Journal Environmental Quality 25: 989-907. Wagner-Riddle, C., G.W. Thurtell, G.E. Kidd, G.C. Edwards and I.J. Simpson (1996b) Micrometeorological measurements of trace gas fluxes from agricultural and natural systems. Infrared Phys. Technology 37:51-58. Wanninkhof, R. (1992) Relationship between gas exchange and wind speed over the ocean. Journal of Geophysical Research 97: 7373-7381. Wanninkhof, R., W. Asher, R. Weppemig, H. Chen, P. Schlosser, C. Langdon and R. Sambrotto (1993) Gas transfer experiment on Georges Bank using two volatile deliberate tracers. Journal of Geophysical Research 98: 20237-20248. Wanninkhof, R., G. Hitchcock, W. Wiseman, G. Vargo, P. Ortner, W. Asher, D. Ho, P. Schlosser, M.L. Dickson, M. Anderson, R. Masserini, K. Fanning and J.-Z. Zhang (1997) Gas Exchange, Dispersion and Biological Productivity on the West Florida Shelf: Results from a Lagrangian Tracer Study. Geophysical Research Letters 24:1767-1770. Watson, A. J., R. C. Upstill-Goddard and P. S. Liss (1991) Air-sea exchange in rough and stormy seas, measured by a dual tracer technique. Nature 349: 145-147. Webb, E.K., G.I. Pearlman and R. Leuning (1980) Correction of flux measurements for density effects due to heat and water vapor transfer. Quarterly Journal Royal Meteorological Society 106:85-100. Wenthworth, W.E. and R.R. Freeman (1973) Measurement of atmospheric nitrous oxide using an electron capture detector in conjunction with gas chromatography. Journal Chromatography
29:322-324. Wesely, M.L., D.H. Lenschow and O.T. Denmead (1989) Flux measurement techniques. In: D.H. Lenschow and B.B. Hicks (Eds.) Global Tropospheric Chemistry: Chemical Fluxes in the Global Atmosphere. NCAR, Boulder, CO, pp. 31-46. Whalen, S.C. and W.S. Reeburgh (1988) A methane flux time series for tundra environments. Global Biogeochemical Cycles 2:399-409. Whalen, S.C. and W.S. Reeburgh (1990) A methane flux transect along the trans-Alaska pipeline haul road. Tellus 42B: 237-249. Wienhold, F.G., H. Fischer and G.W Harris (1996) Fast response tunable diode laser spectroscopy for trace gas flux measurements. Infrared Physics Technology 37:67-74.
66
R.L. Lapitan, R. Wanninkhofand A.R. Mosier
Wienhold, F.G., H. Frahm and G.W Harris (1994) Measurements of N20 fluxes from fertilized grassland using a fast response tunable diode laser spectrometer. Journal of Geophysical Research 99:16557-16567. Wienhold, F.G., M. Welling and G.W Harris (1995) Micrometeorological measurement and source region analysis of nitrous oxide fluxes from an agricultural soil. Atmospheric Environment 29:2219-2227. Wienhold, F.G., T. Zenker and G.W. Harris (1993) Dual-channel two-tone frequency modulation tunable diode laser spectrometer for ground-based and airborne trace gas measurements. In: A. Fried, D.K. Killinger and H.I. Schiff (Eds.) Tunable Diode Lase Spectroscopy, Lidar, and DIAL Techniques for Environmental and Industrial Measurement. Proceedings Society Photo-Optical Instrumentation Engineers, Vol. 2112, Washington, pp. 31-44. Wilson, J.D., V.R. Catchpole, O.T. Denmead and G.W. Yhurtell (1983) Verification of a simple micrometeorological method for estimating the rate of gaseous mass transfer from the ground to the atmosphere. Agricultural Meteorology 29:183-189. Wilson, J.D., G.W. Thurtell, G.E. Kidd and E.G. Beauchamp (1982) Estimation of the rate of gaseous mass transfer from a surface plot to the atmosphere. Atmospheric Environment 16:1861-1867. WMO (1995) Scientific assessment of ozone depletion." 1994. MWO Report No. 37, WMO, Geneva. Woolf D. K. (1997) Bubbles and their role in gas exchange. In: P. S. Liss and R. A. Duce (Ed.) The Sea Surface and Global Change. Cambridge University Press, Cambridge, pp. 173-206 Yamamoto, S., H. Kondo, M. Gamo, S. Murayama, N. Kaneyasu and M. Hayashi (1996) Airplane measurements of carbon dioxide distribution on Iriomote island in Japan. Atmospheric Environment 30:1091 - 1097. Yvon S.A., E.S. Saltzman, D.J. Cooper, T.S. Bates and A.M. Thompson (1996) Atmospheric dimethylsulfide cycling at a tropical South Pacific station (12~ 135~ A comparison of field data and model results. 1. Dimethylsulfide. Journal of Geophysical Research 101: 6899-6909.
Chapter 3
SOME RECENT DEVELOPMENTS IN TRACE GAS FLUX MEASUREMENT TECHNIQUES
O.T. Denmead, R. Leuning, D.W.T. Griffith and C.P. Meyer
This Page Intentionally Left Blank
Approaches to scaling a trace gas fluxes in ecosystems A.F. Bouwman, editor 9 Elsevier Science B.V. All rights reserved
SOME RECENT DEVELOPMENTS IN TRACE GAS FLUX MEASUREMENT TECHNIQUES
O.T.Denmead 1, R.Leuning 1, D.W.T.Griffith 2 and C.P.Meyer 3 CSIRO Land and Water, F.C.Pye Field Environment Laboratory., GPO Box 1666, Canberra, ACT 2601, Australia. 2Department of Chemistry, University. of Wollongong, Wollongong, NSW 2522, Australia. 3 CAIRO Division of Atmospheric Research, Private Bag 1, Aspendale, Vic 3195, Australia.
1. Introduction This contribution complements those of Lapitan et al. (1999) who deal in a general way with a wide range of methods for measuring surface gas fluxes in terrestrial systems, and Fowler (1999) who considers experimental designs for flux determination on various scales. It summarizes some recent developments in meteorological methods for inferring trace gas fluxes on scales not well catered for by conventional methods: in plant canopies, on small plots and at the regional scale. Chambers and most of the common micrometeorological flux techniques measure the net exchange of a gas between the land surface and the atmosphere. Interpretation and modelling omen necessitate more detailed information on the components of the flux: the canopy sources and sinks. Evapotranspiration and CO2 exchange are good examples of relevant exchange processes. The former has both soil and plant sources, each subject to different controls, while the latter has, by day at least, a soil source and a plant sink. Ammonia exchange is another example. The gas can be emitted and absorbed by both soil and foliage. Early attempts at flux measurement in plant canopies relied on gradient-diffusion approaches, but it is now apparent that this approach is inappropriate in the canopy space (Denmead and Bradley, 1987). The mean concentration profiles reflect more the distribution of sources and sinks than the direction of scalar transport. Alternative approaches to canopy transport have been developed in consequence, e.g., higher order closure models, large-eddy simulation, wavelet analysis and Lagrangian dispersion models. The first three approaches require rapid measurement of instantaneous scalar concentrations, which is not yet feasible for many trace gases. The inverse Lagrangian model developed by Raupach (1989b), however, permits the identification of sources and sinks of scalars in the canopy space from mean concentrations, which are usually much easier to measure. Lagrangian dispersion treats the canopy as an assembly of source elements, each releasing a plume of scalar material into the turbulent air flow within and above the canopy. The source strengths can be inferred from the mean concentration profiles and statistics of the turbulence. Asman et al. (1999) consider the use of isotopes as an alternative approach for identifying canopy sources and sinks. Fowler (1999) points out that the large point to point variability in soil gas emissions can make the scaling-up of chamber measurements very difficult. Alternative conventional micrometeorological methods integrate over large areas, smoothing out the small scale hetero-
70
O.T. Denmead, R. Leuning, D.W.T. Griffith and C.P. Meyer
geneity, but they require uniform land areas with lateral dimensions of hundreds of meters. Not only are such areas often difficult to find, but also many natural and man-made ecosystems are not so extensive. The different ecosystems that together make up wetlands, such as the peat wetlands illustrated in Figure 4 of Fowler t ~999), are an example of the former and landfills an example of the latter. There is scope for micrometeorological methods appropriate for smaller land areas, areas larger than the usual chamber size of, say, lm z, but with lateral dimensions of tens rather than hundreds of meters. Mass balance methods fill this gap. Mass balance approaches can also be used on a much wider scale to provide regional trace gas fluxes (Choularton et al., 1995; Yamamoto et al., 1996; Fowler, 1999). This application is discussed by Fowler (1999). Boundary layer budgeting techniques aim to provide estimates of the fluxes of scalars on regional scales. Two schemes will be described: one for the daytime convective boundary layer (CBL) and another for the nocturnal boundary layer (NBL). Both use time changes in atmospheric scalar concentrations to calculate spatially averaged surface fluxes. Fuller accounts are given in Raupach et al. (1992) and Denmead et al. (1996).
2. Inverse lagrangian dispersion methods for within canopy fluxes 2.1. Theoretical These methods, developed notably by Raupach (1989a, b) offer a relatively simple means for inferring fluxes of trace gases and their source-sink distributions within plant canopies. They provide a bridge between chamber or cuvette measurements on soil or individual foliage elements and whole canopy measurements on a field scale. They require measurements of the mean gas concentration profiles within the canopy and some knowledge of turbulence and Lagrangian time scales in that space. As for chambers, advantage is taken of the restricted air movement in the canopy, which makes for larger and more easily measurable concentration changes than in the unobstructed atmosphere above it. Within the canopy, the source or sink strength of the gas at levels z is designated by S(z). It is related to the vertical flux density F at any level by S ( z ) : dF(z)/dz,
(1)
F(h)= F(O)+ If S(z)dz
(2)
so that
where F(O) is the gas flux density at the ground surface and h is canopy height. The Lagrangian dispersion theory developed by Raupach (1989a) enables a prediction of the mean gas concentration profile, which Raupach designates by C(z), from knowledge of S(z). The inverse Lagrangian method described by Raupach (1989b) builds on that work to allow the inference of S(z) from C(z). The procedure uses a discretized, linear relationship between source and concentration profiles, in which the coefficients are determined by the statistics of the wind field. These statistics determine the contributions of gas emitted at particular heights to the concentrations developed at any height within the canopy. By summing the emissions from all source layers (denoted by the suffix j), the gas concentration at any height (denoted by suffix i) can be written as: c, - c . = Z D,jSAZJ
where CR is the concentration at a reference height above the canopy. The coefficients of the
(3)
Recent developments in trace gas flux measurement techniques
71
0800 1000 1100 1200 1300 1400 1500 1600 1700
g .
.
.
.
.
.
.
---
I~--- 5 m b ~*.~ 0800 1000 1100 1200 1300 1400 1500 1600 1700
]
E r"
i
h
.
.
.
.
.
.
.
.
I~- 25 ppmv--~
Figure 1. Mean 30-min profiles of water vapour pressure and CO2 concentration in and above a wheat crop. Each profile shows differences from the top measuring height of 2.15m. The crop height h was 0.76m and the leaf area index was 3.6. (From Denmead and Raupach, 1993).
dispersion matrix, the D;j, are calculated from profiles of crw, the standard deviation of vertical wind speed, and TL, the integral Lagrangian time scale within the canopy, as parameterized, for example, by Raupach (1989a). Once the coefficients are known, equation (3) becomes a set of linear equations that can be solved for the source profile Sj. Raupach (1989b) recommends that to ensure a stable solution for the source profile, redundant concentration data should be included, so that source densities Sj are sought in m layers with n measured concentration values C;, with n > m. The computer program developed by Raupach for inferring Sj. runs on a personal computer in a matter of seconds.
2.2.
Applications
Examples of the use of inverse Lagrangian analysis for determining sources of heat, water vapour and CO2 in crops of wheat and sugar cane are given by Denmead and Raupach (1993) and Denmead (1995). In these cases, the strengths of four source layers were inferred from concentrations measured at six heights within and two above the canopy. It is difficult to test the fine detail of the procedure because of lack of independent information on the withincanopy sources and sinks, but generally in these studies, fluxes predicted by the analysis for the whole canopy have been within 10 to 20% of fluxes above the canopy measured by conventional micrometeorological methods. Figures 1, 2 and 3 from Denmead and Raupach (1993) show some illustrative examples of the operation of the inverse Lagrangian technique. In this case sources and sinks of water vapour and CO2 were inferred for a wheat crop. Figure 1 shows concentration profiles for both gases, Figure 2 shows cumulative fluxes of both through the canopy and Figure 3 compares some predictions of the analysis with observations. For water vapour, the analysis
72
O.T. Denmead, R. Leuning, D. W.T. Griffith and C.P. Meyer L_
i
0.8 /
Z
1
L /;,"
0.6 (m)
i
1100
1300
1100
""
~
~z
0.4
i
1300
i
~ ~
~
~
0.2
""-.
10 O C T
0 I 0
I
I
I
100
200
300
-2
h E ( W i n -2)
I
1
-1
0
Fc ( m g C O 2 m - 2 s -1)
Figure 2. Flux densities of latent heat (XE) and COz (Fc) in the wheat crop of Figure 1, calculated by the inverse Lagrangian method. (From Denmead and Raupach, 1993).
predicts some evaporation at the soil surface and strong contributions from foliage in the top half of the canopy (Figure 2). For CO2, it predicts some soil respiration and strong assimilation in the top foliage. Figure 3a compares the predicted flux of water vapour at 0.19 m with mea-
60
I
40~iI
20
_l
0 400
E
' ' I Inverse method at z = O . 2 5 h
I
,
9
0. ,
'
' 9
9
9
I
9
I 9
9
' 9
9
Lysimeters ,
I
' a
t-
JI
,
_
b_
200
Id.l
9
0 0
~w.,~
9 I n v e r s e m e t h o d at z = h Energy balance "---- Eddy correlation
"7
I
-
/$'/
-
E 0 0
-1
-"
~i,\
9 ~/-.--.._ /
-
O'}
E
c-
-2 0900
1200 10 October
1500 1991
Figure 3. (a). Flux density, of latent heat at 0.19m in wheat crop of Figure 1, calculated by the inverse Lagrangian method (dots), and flux at the soil surface measured by minilysimeters; (b) Flux densities of latent heat and CO2 at the top of the wheat crop of Figure 1., calculated by the inverse Lagrangian method (dots) and by two conventional micrometeorological methods (lines). (From Denmead and Raupach, 1993).
73
Recent developments in trace gas flux measurement techniques
surements of evaporation at the soil surface made with minilysimeters. It is to be expected that the water vapour flux at 0.19 m would be close to evaporation from the soil surface, but slightly higher because of evaporation from lower leaves, which is, in general, what the analysis predicts. Figures 3b and 3c compare predictions of the cumulative fluxes of both gases at the top of the canopy with independent measurements of those fluxes in the air layer above the canopy obtained with conventional micrometeorological techniques. Apart from some discrepancies early in the day, the agreement between Lagrangian and conventional methods is excellent, as good as the mutual agreement between the two conventional methods. 2.3. Conclusions This relatively simple computational and measurement scheme obviously has much promise for trace gas flux measurement in plant canopies. The above applications show how useful it can be for gases with sources and/or sinks in the foliage, but it should prove particularly useful for gases like N20 that have only a soil source. The concentration gradients will be much larger than in the air above the canopy and the within-canopy analysis should be uncomplicated by contributions from the foliage. Other problems to which it might be applied are pathways of CH4 emission in wetland plant communities, the fate of NH3 released from effluent applied to crops and the sources and sinks of CO, NOx, 03 and stable isotopes in canopies. The method is, however, based on extensive homogeneous canopies where the source and sink pattern is constant in the horizontal. It should not be expected to work very well in heterogeneous canopies such as native or regrowth forests.
3. Mass balance methods 3.1. Theoretical Based on the conservation of mass, the general method equates the horizontal flux of gas across a face of unit width on the downwind edge of a designated area with the surface emission or absorption of the gas along a strip of similar width upwind. The horizontal flux density at any height is the product of horizontal wind speed u and gas concentration Cg. The total horizontal flux is obtained by integrating that product over the depth of the modified layer Z which is about 1/10 of the fetch X in neutral conditions, but usually less than that in unstable conditions and more in stable conditions. The average surface flux density is then given by
F -O/X)I~u(C.,-Cb)dz
(4)
where Cb is the upwind, background concentration and the overbar denotes a time average.
3.2. Applications One difficulty in applying equation (20) is that the term
uCg is the
mean of instantaneous
fluxes~
uC~ = u C~ + u'C'~
(s)
74
O.T.
4 r'-r--r--r'm,
( i , ,,
Denmead, R. Leuning, D. W.T. Griffith and C.P. Meyer
, i,,
i-~
3
iiiT!ii !i:i!i)i!iii[!ii!i?!!!iii!i !?i![)ii!ii
2 [ ~r "
0
._~ Q)
4
~",,,. Upwin, 4~\5"D~ ,:,
t._.._z.....t.....z~t ,
t
i
i i
~
",
~l-l-'-lii!tl 4
!
~tl
l
~.I I
s
a 1 ---
'
i
i
L ji
i
i
1
)i;i;!ii:~iiiiil))i)::i;iii
I
l
1
I
I
i
i
1
i
1
1
I [:[:[}?:il[::i[i::!i?[iii[iii[iii!iii[}?2!i!i~l I .. :.: ==========================================
I
;
1
I
I i
"
O
I
q ~:',i:i',i;i',i:,;i;:i~,i',ii;!i! : : Iil\ i ~
N \S
!il-i[ iiiiii[iiil ~ii!t iiii:i!iiiii!!!i!i!i!iiiiiii!iii! !ii i i !i i i i!iil!i i i!;i
"'~'"-
::::::::::::::::::::::::::::::::::::::::::::::::::
I I i I1 I I I II I I I l~:~:::i::~t:~:~:~:i:i:l~:i:i:~:::i~i:~:~:i:iI:~ 1 i I I / 0 10 0 10 0 10 0 10 0 10 20 30 CH 4 m i x i n g ratio ( p p m v ) I
~
1
1
t
2.0
l
1
t
I
i
1.5
:ii!iill !ii!!!.~
~
0
0
I
i C -
.......................
~ w~iiii~!~.~E
20
t
~ Aerodynamic /'
I
t
Mass balance
1.0 ~"
I
,,
!iii!ii!!ii!iiii!iii!iiiiiii!ii!!!ii:i::"::::
x ~.
,,
::::::::::::::::::::::::::::::::: .'.'.:.: ============================ ..:::.~
-,
~""..::i:!:::!:i:i:i:i:
_
~~
40
60
H o r i z o n t a l C H 4 flux d e n s i t y
(mg m-2s -1 )
0
~, 1500
1800
2000
2200
ii iii!ii!iiil
iilii!~iliiiiiiiiil;i;iii!;iiiiiiiiiiiiiiiiii ..... 0000
0200
Time
Figure 4 (a). Mean CI-I4concentrations on upwind (N and W) and downwind (S and E) boundaries of a 24m • 24m square on the surface of a landfill. (b) Horizontal flux densities of CI-I4 on upwind and downwind boundaries during run starting at 01:22 (stippled profiles in Figure 4a); (c) Surface flux densities of CH4 from landfill measured by mass balance method and a conventional aerodynamic method. The concentration profiles in Figure 4(a) were measured in the stippled interval. (From Denmead et al., 1 9 9 8 a ) .
where, as previously, the overbars denote time means and the primes, fluctuations about the means. The first term on the right of equation (5) represents the convective flux out of the treated area due to the mean flow of the wind. The second term represents a smaller turbulent diffusive flux. The first term, the product of the mean wind speed and the mean concentration, is the term that will be measured usually. Probably it will be enough to reduce it by an empirical correction of, say, 15% (Denmead, 1995). Another difficulty is that the effective gas concentration is the concentration in excess of background. Not only must the upwind concentration profile be measured as well as the downwind, but calculation of the flux through equation (4) involves the subtraction of experimentally determined data, an error-prone procedure. The technique is thus suited best to experimental situations where Cb is small and F large as in investigations of NH3 emissions from fertilizers, for which it has been much used (Denmead, 1995). Those conditions may not pertain in some natural ecosystems.
Recent developments in trace gas flux measurement techniques
75
A third problem with this approach is that the fetch X should be known precisely, whereas it is likely to vary with wind direction if the designated area is a rectilinear plot. This complication can be overcome by working with a circular plot and measuring u and Cg at its centre. Regardless of compass direction, the wind will always blow towards the centre of the plot and X will always be equal to the plot radius. It will often be possible to approximate small ecosystems by equivalent circles. A more elaborate but less arguable approach is to form a mass budget for an area whose geometry is known precisely. A recent application has been to the measurement of CH4 emission from the surface of a landfill (Denmead et al., 1998a). In that case, the designated area was a square, 24m x 24m. Concentration measurements were made at 4 heights with sampling arms extending the length of all 4 boundaries. Measurements of wind direction were used to calculate the vector winds U and V normal to the test plot. The surface flux density F was given by
Io Io" [ where the subscripts 1 and 2 denote the upwind boundaries and 3 and 4 the downwind, and A is the area of the test plot. Figure 4 illustrates the application. It shows (a) concentration profiles on upwind and downwind boundaries during consecutive runs when the wind was NW, (b) corresponding horizontal fluxes for one of the runs and (c) the favourable comparison between the surface flux density of CH4 calculated by the mass balance approach, equation (6), and that calculated by a conventional aerodynamic micrometeorological technique with correction for advection caused by the smallness of the test site (Denmead et al., 1998a).
3.3. Conclusions As employed in the above manner, the mass balance method can suffer from errors arising from the large numbers of gas analyses required for a flux determination and becomes unreliable in light winds and variable wind directions. On the other hand, it is non-disturbing, has a simple theoretical basis, is independent of atmospheric stability or the shape of the wind profile and smooths over surface heterogeneity. The applications of mass balance methods described by Choularton et al. (1995), Yamamoto et al. (1996), and Fowler (1999) are on much larger scales than the small plot methods discussed here, with fetches of 20 to 200 km, and involve fitting models of mixing processes in the atmospheric boundary layer to the observed concentration data.
4. Convective boundary layer budgeting 4.1. Theoretical The CBL consists of a shallow surface layer about 100m deep, in which vertical gas fluxes are nearly constant with height and concentration gradients are relatively large, and an overlying mixed layer where fluxes vary only slowly with height and concentrations are uniform because of large-scale mixing. The CBL is capped by a sharp temperature inversion. The mixed layer grows during the day through the input of heat at the ground, entraining air from above the inversion as it does so, and eventually extends up to 1 - 2 km. Conventional micrometeorological flux measurements are appropriate for the constant-flux surface layer. However, CBL budgeting techniques are based on the rate of change of gas concentrations in the mixed
o.T. Denmead, R. Leuning, D. W.T. Griffith and C.P. Meyer
76
layer which acts like a giant mixing chamber moving over the countryside with the mean wind. The method aims to provide regionally averaged rather than local surface fluxes. Gas concentrations in the CBL change through growth of the mixed layer, through the entrainment of air with a different concentration from above, and most importantly, through a flux of gas to or from the surface. By following the buildup or drawdown of gas concentration in the CBL and its height, and allowing for entrainment, an estimate can be made of the average surface flux from the landscape over which an air column moving at the mean wind speed has passed during the day. The bulk properties of the mixed layer are independent of small-scale heterogeneities, so it acts as a natural integrator of surface fluxes over heterogeneous terrain. The distance over which the CBL carries some "memory" of conditions upwind, i.e. its instantaneous footprint, is usually 5 to 30 km. Typically CB[ budgets estimate average fluxes over an area of 103 to 104 km 2 extending about 100 km upwind. Fuller accounts of CBL development and CBL budgeting techniques are given by Raupach et al. (1992) and Denmead et al. (1996). Denmead et al. (1996) developed an expression for calculating instantaneous surface flux densities F from observations of CBL height h, CBL gas concentrations Cm and the gas concentration in the free atmosphere just above the CBL, C+:
F = hdCm/dt - (C+ -C,.Xdh/dt - W+)
(7)
where t is time and W+ is the subsidence velocity. This equation can be integrated over the day to yield the cumulated regional flux I:
I(t) = I~ F(t)dt
(s)
I (t ) : h(t )[Cm (t ) - C +(t )l - h(O )[Cm (O) - C +(0)]+ (y / 2 )[h~ - h 2 ]- I~ W+(t )[Cm (t ) - C. (t )]dt In equation (8), y = dC+/dz, i.e., y is the rate of change of concentration with height just above the mixed layer at h+. Because of the difficulties of measuring dh/dt and dCm/dt accurately, the integral form of the CBL budget equation, equation (8), is easier to apply. To apply equation(8), Cm, C+ and y must be determined by analyzing air samples from within and above the mixed layer obtained with the aid of aircraft, kites or balloons. Such measurements are not often available, so the analysis presented here is based on measurements of near-surface concentrations Cs and some simplifications concerning C+ and 7". Cm can be inferred from C~ measured at height z,. through an aerodynamic resistance ra calculated from conventional micrometeorological similarity theory: Cm = C,. - ra F,
(9)
and l]vl[(Z m - d ) / ( z
r -
s
-d)]-[//(z m ku,
d, z s - d)
(10)
In equation (10), Zm is the height from the ground to the bottom of the mixed layer, d is the zero-plane displacement, k (=0.41) is the von Karman constant, u, is the friction velocity and ~, is the integrated form of the stability function for unstable conditions given by Paulson (1970):
Recent developments in trace gas flux measurement techniques
77
where L is the Monin-Obukhov length. For trace gases whose diurnal emissions are reasonably constant in time, equation (9) can be incorporated into equation (8) by recognizing that Ft I/t. Assuming that W§ is negligibly small and that 7"= 0 (i.e., there is a step change from Cm to a constant value of C+ at h), then
_ h(O[c (,)- c+
h(o)[c (o)- < (o)]
1 + [h(t)r (t)- h (O)G (O)l/ t
(12)
If production is controlled by the diurnal variation of light or temperature, as is the case for CO2 and N20, Denmead et al. (1996) assume that F(t) oc FA(t) where FA(t) is the available radiant energy. Then
I(t) -
h(t)[C~(t)- C+ (t)]- h(O)[C~(O)- C+(O)l 1 + hO)raO)FA O)/IA 0)-- h(O)G (O)FA(O)/IA (0)
(13)
where:
I A -~s
(13a)
is the integral flux of available energy at time t. A further simplification to the analysis is to assume that C+ has the current clean-air baseline value for the gas of interest.
4.2. Applications Denmead et al. (1996; 1998b) have used equation (12) and (13) to estimate regional fluxes of water vapour, CO2, CH4 and N20 in southeast Australia from ground-based measurements. In the latter study, a tower was erected in a 30 ha field, half of which was planted to lucerne and half to a crop of Triticale. The field was in a larger region comprising pasture (70%) and cropland (30%). Continuous measurements of mean, 30-min mixing ratios for COz, CH4 and NzO were made on air drawn from 7 heights between 0.5 m and 22 m on the tower using Fourier transform infrared spectroscopy and non-dispersive infrared gas analysis. Aerodynamic parameters for calculating ra via equation (10) were obtained with an eddy correlation system mounted at 22 m. Eddy fluxes of CO2 were also measured at 22 m and at 2 m above the lucerne and the Triticale. These and the gas concentration measurements allowed calculation of the corresponding fluxes of CH4 and N20 using the gradient - diffusion approach (Leuning et al,. 1998). Radiosonde ascents were used to determine h at least twice per day. Occasional measurements of Cm, C+ and y were made for each gas using gas sampling flasks attached to balloons or in an aircrat~. To evaluate the integral flux through equations (12) and (13), the starting time (t = 0) was set at 08:30 when h was 300 to 500 m, and the final time at 15:30 when h was 1.5 to 2.5 km. Figure 5(a) compares CBL estimates of the average regional CO2 flux between 08:30 and 15:30 with the eddy fluxes of CO2 measured during the same period. It might be expected that the CBL fluxes would be close to those observed at 22 m where the daytime footprint was 2 to 10 km covering a landscape similar to that of the region at large. However, the CBL fluxes were generally larger than the 22 m fluxes, being intermediate between those observed over the more productive lucerne and Triticale. This points to a problem with the ground-based observation system. When surface concentrations are extrapolated to the mixed layer through equation (9), the estimated Cm reflects the initial surface concentration and so biases the CBL flux estimate towards the local surface flux. This was confirmed by the direct balloon and air-
78
c~. T. Denmead, R. Leuning, D. W.T. Griffith and C.P. Meyer 2.5 "~.
2.0
~
1.5
"-"7,
~_'E "~ ~
ale O 0v
I -
1
I
I
I
I
I
I
I
CO 2
A
9
9
/Triticale
9 9
9
1.0
0 I
-0.5 500 ~
1
a
400
I
I
I
I
I
I
I
I
I
b
-
I
1
I
1
1
I
OH 4
I
1
I
I
9
-
x~ 'E 300 o
~_ 3:~ 200
Micromet
~o= ~oo t~ t-t'-v
~
Chambers
0
50
~
40
"5 E
30
I
I
1
L
I
1
l
I
l
1
I
I
I
I
I
1
I
I
I
I
_
C
N20
9
/
-o0 s ;z~
20
z
10
(3.
-C I
i
T
I
1
i
I
I
1
I
9
11
13
la
17
19
21
23
25
27
October 1995
Figure 5 (a). CBL estimates of the average regional CO2 flux in a rural area of southeast Australia, 08:3015:30, compared with local eddy fluxes for same period at 2m above a lucerne field, 2m above a Triticale field and 22m on a tower between both fields; (b) CBL estimates of CH4 flux in same region compared with chamber measurements in lucerne field and micrometeorological measurements at 22m; (c) CBL estimates of NzO flux in same region compared with chamber measurements in lucerne field and micrometeorological measurements at 22m. (From Denmead et al., 1998b).
craft observations of C,,, which for CO2 were several ppm higher than those predicted by equation (9). For CH4 and N20, with reasonably uniform animal sources in the observation region, equation (3) appeared to predict mixed layer contributions satisfactorily. A second difficulty with this observation system is that it is fixed in space (Eulerian coordinate system), whereas a system which moves with the mean wind is required (Lagrangian coordinate system). To reduce errors from this source, only days with steady wind directions are suitable for application of the CBL methodology, particularly in regions with heterogeneous land use. This requirement will usually restrict the potential data base significantly. Other restrictions arise from the lack of convective conditions, e.g. on cloudy days. Despite the many data points shown in Figure 5(a), only 7 days from a total of 19 in the study of Denmead et al. (1998b) met all requirements for strict application of the CBL analysis.
Recent developments in trace gas flux measurement techniques
79
A third problem is the high precision required in the concentration measurements. In the example study, the typical drawdown in the CO2 concentration of the mixed layer during the course of a day, C+ - Cm, was 3 ppm, while for CH4, the mixed layer enrichment was typically 20 ppb and for N/O, 2 ppb. A precision significantly better than 1% in absolute measurement is thus required. The CBL flux estimates for CH4 are compared with the average of nocturnal micrometeorological flux determinations at 22 m and with chamber measurements in the lucerne and Triticale in Figure 5(b). The latter were made by Meyer et al. (1998). The nocturnal micrometeorological flux estimates were used because they are the most reliable (see section 5.2) and because ruminant animals appear to produce CH4 equally throughout the day and night.The chamber measurements indicated that the soil itself was a CH4 sink with an average strength of-2 ng CH4 m 2 s-1. Both the CBL and micrometeorological measurements, however, showed a net upward flux indicating that the animal sources in the region dominate the soil sinks. The net CH4 emissions are consistent with a stocking rate between 0.5 and 1 cattle equivalents ha 1, the same range indicated by regional surveys of animal numbers. The CBL estimates of N20 emissions appeared to be intermediate between micrometeorological and chamber fluxes for the same observation periods (08:30 to 15:30) on 3 of the 4 days of measurement (Figure 5c). On the fourth day, October 23, the CBL estimates were higher than the locally measured fluxes, but that day followed 2 days of heavy rain which promoted a sharp increase in N20 emissions. As for CH4, one might expect the regional flux to be greater than the local soil flux because of the expected large regional contribution from animal dung and urine patches.
4.3. Conclusions For the simplified CBL budgeting scheme described here, the use of near-surface concentrations instead of mixed layer concentrations can bias the estimates towards local flux values. Measurements at more than one location in regions of heterogeneous land use would be a desirable improvement resulting in greater accuracy of flux estimates. Changing wind directions, unsuitable weather and the high precision required in gas concentration measurements are also limitations of the method. Neverthelesss, even in its simplified form, CBL budgeting is potentially very useful as a survey tool for estimating regional gas fluxes and in broadly homogeneous regions, appears to provide better than order of magnitude estimates from a relatively simple observation scheme. More elaborate approaches employing aircraft for direct measurements of concentrations in and above the CBL and perhaps a Lagrangian observation system could be expected to yield more precise results. Examples of the former are given in Wofsy et al. (1988), Betts et al. (1992), Ritter et al. (1992), and Choularton et al. (1995).
5. Nocturnal boundary layer budgeting 5.1. Theoretical At night when convective heating ceases, the CBL is replaced by the NBL, a shallow weakly turbulent layer which often extends to heights of only tens of meters, and is bounded by a lowlevel radiative inversion. The inversion inhibits vertical mixing so that emissions of gases from the surface are contained in a shallow air layer whose concentration changes appreciably. The surface flux can be calculated from:
80
O.T. Denmead, R. Leuning, D.W.T. Griffith and C.P. Meyer
F
- fz (dC/dt)dz,
(14)
Z being the height of the air layer whose concentration is affected by the emission. Equation (14) applies when concentration measurements are available up to the top of the inversion layer.
5.2. A p p l i c a t i o n s
Figure 6 shows an example of NBL budgeting for COz and CH4. The data come from the same campaign in which the CBL experiments described in Section 4.2 were performed. A helium filled balloon was used to carry an airline aloft in a series of vertical traverses to a height of 100m. A non-dispersive infrared gas analyzer system was used to measure CO2 concentrations in air pumped from the balloon height and a gas chromatograph was used to measure CH4. On this occasion, a temperature inversion developed early in the evening around 18:00 and most of the CO2 emitted between then and 22:00 was contained between the ground surface and a height of 40m (Figure 6a). The CO2 enrichment of that layer corresponded to an average surface emission rate of 0.05 mgCO2 m 2 s1, in agreement with eddy flux measurements described in Section 4.2. Significant CH4 enrichment was observed up to 60m (Figure 6b) and the calculated flux was 250 ng CH4 mZs l, much the same as the CBL flux estimates made at the site. Other examples of NBL budgeting for the CH4 flux from wetlands are given by Choularton et al. (1995). A more general NBL budgeting approach is needed when concentration measurements do not extend to the top of the inversion layer. Then equation (8) must be modified to include the flux Fz at the top of the layer: r
(15
-
Figure 7 from Leuning et al. (1998) shows the 3 terms in equation (15) for CO2, CH4 and NzO on one night during the trace gas experiment described in Section 4.2. Fluxes at 22 m and
100 t 80
~
I
60
-
"
-~9
-1-
-
800-
40
~,
\ 20
\~aoo
18001900 9
340
-
-
2200
900
~2300
2200\
I
350
\ -
~ ~
_
I
I~ ~
360
CO 2 concentration (ppm)
~'1
-
I
370 1.6
1.7
/
I
1.8
OH 4 concentration (ppm)
Figure 6. Profiles of (a) CO2 and (b) CH4 to 100m obtained with balloon at 18:00 and 22:00 in a rural area of southeast Australia. (From Denmead et al., 1998b).
Recent developments in trace gas flux measurement techniques
al
22
i
i
12.30~
~, 20 E
O o
i
I
i
I
32
16
& 12.14
~ 30 &
o
S 28
14
O 12.06 z
-'r 26 O
12
11.98
24
1
I
1
1600
30 ~ T ~ T ~ T ~
I
0.8
:o
e
E
o
._o
0 -O.4 18
21 00 03 Hour of day
06
==
I
I
q-
-F Flux @ O m
,~400
I0
o 21 O0 03 Hour of day
06
II
I,I
-
800
0
18
I
----9 Storage
I0
0.4
O)
C
"-" 1200 - -
% E
34
12.22
m 18 o
b
81
0 -400
8
21 00 03 Hour of day
06
Figure 7 (a), (b), (c). Nocturnal storage of CO2, N20 and CH4 between 0 and 22m on one night in a lucerne field; (d), (e), (f) . Rate of change of gas storage (0-22m), gas flux at 22m and the sum of both, representing the surface flux, for same period. (From Leuning et al., 1998).
changes in gas storage between the surface and 22 m were added to estimate the surface flux. In this instance, the storage term was ot~en very much larger than the flux at the top of the layer. Over 11 nights of measurement, it accounted for >60% of the surface flux of each gas. Fan et al. (1992) describe similar observations of CH4 and CO2 exchange over tundra made on a 12m tower. The storage term was 20 to 40% of the surface flux. The attraction of NBL budgeting is that the concentration gradients are larger than by day, hence more easily measurable. However, because the method is applicable only at night, inferences from it could lead to errors in assessing source and sink strengths. Examples include CO2 exchange with plant communities, where the day and night fluxes are of opposite sign, and N20 and NH3 emissions which are usually least at night. For trace gases whose emissions do not exhibit marked diurnal cycles, such as CH4 from animal sources, NBL budgets may give more reliable flux estimates than measurements in the day time.
5.3. Conclusions
The growth and height of the CBL are reasonably predictable, but not so for the NBL. Its shiffing height makes it difficult to work with a fixed sampling array and there will be times when the radiative inversion layer is impossibly deep or absent so that the method is not feasible at all. When a budget can be made, the requirement to measure concentration profiles rather than concentrations at a single point make it a more complicated procedure than CBL budgeting. Another weakness in comparison with CBL techniques is uncertainty about the extent of the surface that the budget represents. A rough estimate of its footprint is 1 to 5 km. On the other hand, the depth of the atmospheric mixing "chamber" is better defined, few
82
O.T. Denmead, R. Leuning, D. W.T. Griffith and C.P. Meyer
assumptions are required and the concentration changes usually will be larger and hence more easily detectable than in CBL budgeting. For trace gases whose emissions do not exhibit marked diurnal cycles, NBL budgets may be simpler alternatives than either CBL or conventional micrometeorological flux measurements by day. However, when there is a diurnal cycle in gas exchange, such as for CO2, errors could arise from assessing source and sink strengths from NBL budgets only. Of course, for CO2, the day and night fluxes represent the different processes of photosynthesis and respiration, so that these different approaches are not competitors, but supplement each other.
6. Summary Meteorological techniques for measuring trace gas fluxes on three important scales not well catered for by conventional methods have been discussed: an inverse Lagrangian dispersion method appropriate for the canopy scale, mass balance methods for small and heterogeneous ecosystems and boundary layer budgeting schemes for the regional scale. The inverse Lagrangian analysis offers a relatively simple measurement scheme for inferring fluxes of trace gases and their source-sink distributions within plant canopies. Inputs are the profiles of mean gas concentration and turbulence within and above the canopy. The analysis provides a bridge between chamber and cuvette measurements on soil and foliage elements and flux measurements on a field scale. Mass balance methods are appropriate for flux measurements in small ecosystems, tens of meters in lateral extent. Fluxes from areas of known geometry are calculated from the rate at which the wind transports gas across the upwind and downwind boundaries of the designated area. The method can fill the gap between chambers of, say, 1 m 2 in area and conventional micrometeorological methods representing, say, 104 m 2. It can suffer from errors arising from the large number of gas analyses required for a flux determination and may become unreliable when there are light winds and variable wind directions. On the other hand, it is nondisturbing, has a simple theoretical basis, smooths over surface heterogeneity and is independent of atmospheric stability or the shape of the wind profile. Convective and nocturnal boundary layer (CBL and NBL) budgeting techniques are discussed in the context of a recent experiment to estimate regional fluxes of carbon dioxide, methane, and nitrous oxide in a rural area of southeast Australia. CBL techniques estimate the average surface flux over regions of order 100 km 2 through the buildup or drawdown of gas concentration in the atmospheric mixed layer and its depth. An integral form of CBL budgeting was used to estimate daily fluxes. Input data were gas concentrations at 22 m and CBL heights obtained with radiosondes. The atmospheric gas concentrations above the CBL were assumed to be the current clean-air baseline values. It was concluded that even with this simplified observation scheme, CBL budgeting can be a very useful survey tool and in regions that are homogeneous in the large, can provide better than order of magnitude estimates of trace gas fluxes. NBL budgeting techniques follow the change of gas storage in the surface layer at night when low-level radiative inversions inhibit vertical mixing. The footprint is difficult to estimate, but is of order 1 to 5 km. On one occasion during tile experiment, balloon measurements were made up to a height of 100 m, but routinely, tower-based measurements were made to 22 m. It was concluded that for gases whose emissions do not exhibit marked diurnal cycles, NBL budgets may be simpler alternatives than either CBL budgets or conventional micrometeorological measureJnents made by day. When diurnal variation is large, both day and night measurements are needed to define the 24-hour flux.
Recent developments in trace gas flux measurement techniques
83
References Asman, W.A.H., M.O. Andreae, R. Conrad, O.T. Denmead, L.N. Ganzeveld, W. Helder, T. Kaminski, M.A. Sofiev and S. Trumbore (1999) How can fluxes of trace gases be validated between different scales? In: A.F. Bouwman (Ed.) Approaches to scaling of trace gas fluxes in ecosystems. Developments in Atmospheric Science 24. Elsevier, Amsterdam, pp. 85-97. Betts, A.K., R.L.Desjardins and J.I. MacPherson (1992). Budget analysis of the boundary layer grid flights during FIFE 1987. Journal of Geophysical Research 97:18533-18546. Choularton, T.W., M.W. Gallagher, K.N. Bower, D. Fowler, M. Zahniser and A. Kaye (1995). Trace gas flux measurements at the landscape scale using boundary-layer budgets. Philosophical Transactions of the Royal Society of London A 351:357-369. Denmead, O.T. (1995). Novel meteorological methods for measuring trace gas fluxes. Philosophical Transactions of the Royal Society of London A 351:383-396. Denmead, O.T. and E.F. Bradley (1987). On scalar transport in plant canopies. Irrigation Science 8:131-149. Denmead, O.T. and M.R. Raupach (1993). Methods for measuring atmospheric gas tr~sport in agricultural and forest systems. In: L.A. Harper, A.R. Mosier, J.M. Duxbury and D.E. Rolston (eds.) Agricultural Ecosystem Effects on Trace Gases and Global Climate Change. American Society of Agronomy Special Publication 55, Madison, WI, pp. 19-43. Denmead, O.T., M.R. Raupach, F.X. Dunin, H.A. Cleugh, and R. Leuning, (1996). Boundary-layer budgets for regional estimates of scalar fluxes. Global Change Biology 2:255-264. Denmead, O.T., L.A. Harper, J.R. Freney, D.W.T. Griffith, R. Leuning and R.R. Sharpe (1998a). A mass balance method for non-intrusive measurements of surface-air trace gas exchange. Atmospheric Environment (in press). Denmead, O.T., R. Leuning, D.W.T. Griffith, I.M. Jamie, M. Esler, H.A. Cleugh and M.R. Raupach (1998b). Estimating regional fluxes of CO2, CH4 and N20 at OASIS through boundary-layer budgeting. In: R. Leuning, O.T. Denmead, D.W.T. Griffith, I.M. Jamie, P. Isaacs, J. Hacker, C.P. Meyer, I.E. Galbally, M.R. Raupach and M.B. Esler (eds.) Assessing biogenic sources and sinks of greenhouse gases at three interlinking scales. Consultancy Report 97-56, CSIRO Land and Water, Canberra. Fan, S.M., S.C. Wofsy, P.S. Bakwin, D.J. Jacob, S.M. Anderson, P.L. Kebabian, J.B. McManus, C.E. Kolb and D.R. Fitzjarrald (1992). Micrometeorological measurements of CH4 and CO2 exchange between the atmosphere and subarctic tundra. Journal of Geophysical Research 97:16627-16643. Fowler, D. (1998). Experimental designs appropriate for flux determination in terrestrial and aquatic systems. In: A.F. Bouwman (Ed.) Approaches to scaling of trace gas fluxes in ecosystems. Developments in Atmospheric Science 24. Elsevier Science, Amsterdam, pp. 99-121. Lapitan, R.L., R. Wanninkhof and A.R. Mosier (1999) Methods for stable gas flux determination in aquatic and terrestrial systems. In: A.F. Bouwman (Ed.) Approaches to scaling of trace gas fluxes in ecosystems. Developments in Atmospheric Science 24. Elsevier, Amsterdam, pp. 27-66. Leuning, R., D.W.T. Griffith, O.T. Denmead and I.M. Jamie (1998). Air-land exchanges of CO2, CH4 and N20 during OASIS 1994 and 1995 measured using FITR spectroscopy and micrometeorological techniques. In: R. Leuning O.T. Denmead, D.W.T. Griffith, I.M. Jamie, P. Isaacs, J. Hacker, C.P. Meyer, I.E. Galbally, M.R. Raupach and M.B. Esler (Eds.) Assessing biogenic sources and sinks of greenhouse gases at three interlinking scales. Consultancy Report 97-56, CSIRO Land and Water, Canberra. Meyer, C.P., I.E. Galbally, D.W.T. Griffith, I.A. Weeks. I.M. Jamie and Y.P. Wang (1998). Trace gas exchange between soil and atmosphere in southem N SW using flux chamber measurement techniques. In: R. Leuning, O.T. Denmead, D.W.T. Griffith, I.M. Jamie, P. Isaacs, J. Hacker, C.P. Meyer, I.E. Galbally, M.R. Raupach and M.B. Esler (Eds.) Assessing biogenic sources and sinks of greenhouse gases at three interlinking scales. Consultancy Report 97-56, CSIRO Land and Water, Canberra.
O.T. Denmead, R. Leuning, D.W.T. Griffith and CP. Meyer
84
Paulson, C.A. (1970). The mathematical representation of wind speed and temperature profiles in the unstable atmospheric surface layer. Journal of Applied Meteorology 9:857-861. Raupach, M.R. (1989a). A practical Lagrangian method for relating scalar concentrations to source distributions in vegetation canopies. Quarterly Journal of the Royal Meteorological. Society 115:609632. Raupach, M.R. (1989b). Applying Lagrangian fluid mechanics to infer scalar source distributions from concentration profiles in plant canopies. Agricultural and Forest Meteorology 47:85-108. Raupach, M.R., O.T. Denmead, and F.X. Dunin, (1992). Challenges in linking atmospheric CO2 concentrations to fluxes at local and regional scales. Australian Journal of Botany 40:697-716. Ritter, J.A., J.D.W. Barrick, G.W. Sachse, G.L. Gregory, M.A. Woerner, C.E. Watson, G.F. Hill and J.E. Collins Jr. (1992). Airborne flux measurements of trace species in an Arctic boundary layer. Journal of Geophysical Research 97:16601-16625. Wofsy, S.C., R.C. Harriss and W.A. Kaplan (1988). Carbon dioxide in the atmosphere over the Amazon basin. Journal of Geophysical Research 93:1377-1387. Yamamoto, S., H. Koudo, M. Gamo, S. Murayama, N. Kaneyasu and M. Hayashi (1996). Airplane measurements of carbon dioxide distribution on Iriomote Island in Japan. Atmospheric Environment 30:1091-1097.
Chapter 4
H O W CAN FLUXES OF TRACE GASES BE VALIDATED B E T W E E N D I F F E R E N T SCALES?
W.A.H. Asman, M.O. Andreae, R. Conrad, O.T. Denmead, L.N. Ganzeveld, W. Helder, T. Kaminski, M.A. Sofiev and S. Trumbore
This Page Intentionally Left Blank
Approaches to scaling a trace gas fluxes in ecosystems A.F. Bouwman, editor 9 Elsevier Science B.V. All rights reserved
WORKING GROUP REPORT HOW CAN FLUXES OF TRACE GASES BE VALIDATED BETWEEN DIFFERENT SCALES?
W.A.H. Asman (Rapporteur), M.O. Andreae (Chairman), R. Conrad, O.T. Denmead, L.N. Ganzeveld, W. Helder, T. Kaminski, M.A. Sofiev and S. Trumbore
1. Introduction
Estimates of fluxes of trace gases and their spatial and temporal variability are needed for several reasons. Firstly, we need to know the magnitude and direction of the flux as well as the relative importance of the different source categories. Secondly, flux estimates or descriptions of the key processes regulating the fluxes at the scale considered are used to drive atmospheric models. Thirdly, flux estimates are used to validate the output of atmospheric models. The information on fluxes is needed at different spatial scales - ranging from the local (point) scale, to areas at the scale of a nature reserve (few km in diameter), provinces (up to about 100 km), to the country, continental and the global s c a l e - and temporal scale. In this report we focus mainly on the bottom-up approaches in scaling, where information on fluxes at lower scales is integrated to obtain information at a higher scale level. Firstly, we review selected flux measurement techniques at different scales. More detailed overviews are given by Lapitan et al. (1999), Denmead et al. (1999) and Fowler (1999). Secondly, methods used to derive fluxes using atmospheric models are discussed. This leads to a discussion on so-called "redundant' measurements. Finally, we present recommendations for future research on validation techniques applicable between different scale levels. The use of isotopes and tracers in scaling is extensively discussed by Trumbore (1999), and will not be reviewed here.
2. M e t h o d s to m e a s u r e fluxes at different scales
A variety of techniques are used to determine trace gas fluxes, including chamber, micrometeorological and airborne methods. In most cases these methods "see" different areas. For example, the chamber method represents only the area covered by the chamber. Alternatively, ground-based micrometeorological methods "see" a larger area, or footprint, depending on factors such as the height of the meteorological tower, surface roughness and atmospheric stability. Therefore, results of these methods, in fact, may only be compared for areas that do not show any spatial variation in the gas fluxes. As such homogeneous areas do not exist in practice, the comparison should be made for areas that are as homogeneous as possible, such as functional types (see Seitzinger et al., 1999). If the different methods used yield similar results, this indicates that they can be used successfully at their own, specific, scale levels. In theory, it is also possible to compare different methods for heterogeneous areas by stratifying landscapes on the basis of differences in e.g. land use and soils. Fluxes may then be measured
88
W.A.H. Asman et al.
according to a strategy based on the footprint of the selected technique. This procedure, however, may fail if non-linear processes regulate gas fluxes. Most methods applied at scales larger than 1 m 2 cannot be used continuously. For example, because the meteorological conditions are outside the range of conditions for which the technique was developed and tested. Estimates of fluxes over longer time periods (e.g., season, year) thus can only be obtained with tools (such as models) that integrate fluxes over time, using short-term measurements. This may cause measurement problems when the daily variation in fluxes is high. For example, the carbon dioxide (CO2) flux from terrestrial ecosystems is negative fluxes during daytime, and positive at night. Some methods can be used during a substantial fraction of the time, whereas others are restricted to very specific (meteorological) conditions that occur much less frequently. These issues are discussed in relation to long-term measurements of CO2 and water vapour fluxes by Baldocchi and Valentinti (1996). Irrespective of the techniques used, footprint analysis (e.g. Leclerc and Thurtell, 1990), and geostatistics (e.g. Ambus and Christensen, 1994) are critical steps in scaling up flux measurements. Micrometeorological measurements of oceanic fluxes pose additional problems. For many trace gases, the fluxes between the ocean and the atmosphere are very small compared to fluxes between terrestrial ecosystems and the atmosphere. This causes technical problems related to the detection limits of instruments.
2.1. Methods based on soil concentration profile measurements
Fluxes between soil and atmosphere car: be determined from the vertical gradient in gas concentration within the soil using a diffusion model. The effective diffusivity of the gas in the soil can be measured using tracers such as radon (Rn) (Born et al., 1990; Whalen et al., 1992; Koschorreck and Conrad, 1993). This method has been applied to measure soil uptake of methane (CH4) and could also be applied to determine fluxes of nitric oxide (NO) from upland soils to the atmosphere. Measurement of a concentration gradient, however, is difficult if the layer in which the major part of the concentration gradient occurs is thin. In the case of CI-I4 fluxes from aquatic sediments, for example, a marked CH4 consumption gradient may occur within 1-5 mm depth. The concentration gradient integrates diffusion, production and consumption processes. Hence, the net flux can be measured directly from these processes, as well as indirectly from the concentration gradient and the diffusivity (e.g. Galbally and Johansson, 1989; Remde et al., 1993; Rudolph et al., 1996). The recommended method to scale up is not the averaging of fluxes measured at different sites, but to establish the relation between gas fluxes and the processes controlling them, and the environmental conditions. This knowledge can then be combined with information on the spatiaJ and temporal variations in the controlling factors, allowing to estimate the flux at a larger scale (Matson et al., 1989). The footprint of the gas concentration profile method is 0.1-0.5 m. The method focuses on processes in soil, isolated from possible interference by the vegetation.
2.2. Chamber techniques
Under ideal circumstances (homogenous fetch, level and homogeneous terrain) chamber and micrometeorological methods give comparable results. Yet, the objectives of these two techniques differ. The chamber technique is often used to isolate the soil's contribution to the flux in systems containing tall vegetation. Conventional micrometeorological methods yield
How can fluxes of trace gases be validated between different scales ?
89
information on the total flux to or from an ecosystem, containing both soil and vegetation. They do not give information on processes such as adsorption of gases within the canopy. The Inverse Langrangian dispersion method described below aims to fill such scale gaps. The chamber technique may not give reliable results when the temperature, radiation, energy balance and gas concentration inside the chamber differ from those in the field. In addition, there is no turbulence inside the chamber (or a different turbulence, if fans are used). This is usually not critical for soils where the flux is controlled by soil processes and diffusion in the soil and not by air turbulence. The conditions inside the chamber are more critical when there is a canopy (grass, agricultural crops) for which the exchange is a function of turbulence. The chamber technique cannot be applied in oceanic systems, since the gas exchange velocity depends on turbulence and wind speed, which both are perturbed by or interfere with the chamber operation. Perturbation effects of chambers can be avoided by, for example, limiting the time period of operation or by maintaining an air flow through the chamber. Finally, the typical footprint of a chamber is 0.5-5 m, although megachambers of 30 m long have also been used (Smith et al., 1994).
2.3. Lagrangian dispersion method Inverse Lagrangian methods are not only used to measure the flux over a canopy, but also to infer sources and sinks within the canopy (Denmead et al., 1999). They bridge the "gap" between enclosures (soil chambers, leaf cuvettes) and towers. The method has been used to identify sources and sinks of heat, water vapour, C02, CH4 and ammonia (NH3). It should also be useful for gases stemming from soils only, such as N20, and to determine fluxes of carbon monoxide (CO), NO, ozone (03) and stable isotopes in the canopy. The footprint of the Lagrangian dispersion method is smaller than that of conventional micrometeorological measurements. Typically, it will correspond with three canopy heights upwind. The Lagrangian dispersion method is, however, based on extensive homogeneous canopies where the source and sink patterns are constant horizontally. Therefore, it is not likely to be appropriate for heterogeneou~ forests. Neither is the method expected to work under calm conditions, such as at night when gas fluxes is determined by the wind field, temperature inversion and other factors.
2.4. Micrometeorological tower methods Tower measurements, with conventional micrometeorological techniques, yield an integrated flux estimate over the upwind fetch. They are applied to obtain ecosystem fluxes and to study canopy processes. Multi-layer eddy correlation measurements can improve the definition of flux divergence in the canopy. Micrometeorological techniques have, in principle, a larger footprint than the chamber technique. The footprint can be enlarged by using a higher tower (i.e. elevation above the canopy), but this creates a lower signal to noise ratio and problems with convection and advection interfering with turbulent transport. In addition, very high towers are needed for forests, which poses practical problems. The footprints of the micrometeorological methods vary from tens to hundreds of metres, depending on the height of the instruments above the surface of investigation and the meteorological conditions. At night, when low wind speeds and temperature inversion may inhibit vertical transport, there can be substantial storage of emitted gas in the air between the surface and the level of the sensor. In such cases, additional measurements of the concentration profile are needed to account for changes in this storage, which at times may exceed the vertical flux through the
90
W.A.H. Asman et al.
top of the layer (Denmead et al., 1999). l~.!icrometeorological methods, so far, have not been very useful for measuring gas fluxes over oceans.
2.5. Airborne micrometeorological techniques Airborne eddy correlation techniques address scales of tens to hundreds of kilometres. Limitations of these methods are related to the limited availability of high-frequency and highprecision sensors and to operational limitations imposed by aircraft. Airbome eddy correlation techniques can represent large spatial scales, but the temporal resolution is limited. The instantaneous footprint of the method ranges from 1 to 10 km, depending on the cruising height of the aircraft. The effective footprint size is -100 km, taking into consideration that signals must be averaged over 20 minutes or more.
2.6. Convective boundary layer budgeting method When there is enough insolation, the earth's surface is heated and convection develops, leading to a well mixed convective boundary layer (CBL), capped by a sharp temperature inversion. The fluxes are derived from changes in concentration in the CBL, corrected for the inflow from the free troposphere. The latter is estimated from the concentration in the free troposphere and the changes in mixing layer height with time. The concentration in the CBL can only be measured with the aid of aircraft, kites or balloons. Such measurements, however, are often not available. Concentration at the CBL height then has to be estimated by extrapolating near-surface concentrations and by assuming constant concentration just above the mixing layer. The convective boundary layer method has been applied to derive fluxes of CO2, CH4 and N20 in Australia (Denmead et al., 1999). Its application over the oceans poses problems because of a weak boundary layer development, low flux rates and small concentration changes. The footprint is usually 5 - 30 km, but during the day it can range from 50 to 150 km. Therefore, the method can be used to help bridging the gap in footprints between micrometeorological tower and airbome measurements. The convective boundary layer method cannot be applied under conditions with low wind speeds, variable wind directions and non-convective conditions, and it requires high-precision concentration measurements. The applicability of the convective boundary layer method may be extended if used in combination with a Lagrangian observation system moving with the wind.
2.7. Nocturnal boundary layer budgeting method At night there is often no convection as a result of which the nocturnal boundary layer is often bounded by a radiative inversion at heights of typically tens of metres. The flux can be determined from concentration changes, that can be considerable due to the shallowness of the boundary layer. Vertical concentration profiles up to the height of inversion have to be measured as a function of time to obtain a good estimate of the budget. The nocturnal boundary layer budgeting method can not always be applied, in particular when the inversion layer is extremely deep or absent. Another weakness of the method is that the extent of the surface for which the budget is computed, remains uncertain, with estimated footprints ranging from 1 to 5 km.
How can fluxes of trace gases be validated between different scales ?
91
2.8. Airborne mass balance method The vertical flux is derived from the measured difference between the outflow and inflow of the gas over a given area, by measuring the change in concentrations. Vertical concentration profiles should be measured at a reasonable number of sites to avoid interference from plumes. The method should particularly yield good results when the upwind and downwind parts of the flight are over the sea, where the vertical concentration profile shows less fine structures than over land. The airborne mass balance method is appropriate for the long-lived atmospheric trace gases (CH4 and N20) and to estimate fluxes of CH4, N20, CO, "wintertimeCO2" and some of the less reactive volatile organic carbon compounds (VOC) (see Fowler, 1999). Measurements should be performed under conditions with steady boundary layer winds (315 m sl), with a well defined temperature inversion that is capping the boundary layer and in the absence of deep convection. The area that can be investigated depends on the length of the measurement (depending on the above conditions) and on the type of carrier used. For balloons and small aircraft the footprint can be as small as 3-5 km. The maximum footprint of the method is about 1000 km. The airborne mass budget method has potential for obtaining information on the N20 flux for upwelling areas in the oceanic margin. If the N20 concentration in sea water and the wind speed are measured simultaneously and in the same area, the gas exchange parameterizations of Liss and Merlivat (1986) and Wanninkhof (1992) could be verified with the airborne mass balance method. In addition, the method could be applied to determine fluxes over "trade wind" islands. This method represents the largest scale for which fluxes can be derived directly from measurements. At larger scales, gas fluxes have to be simulated with atmospheric models.
3. Application of atmospheric models to derive flux estimates Two distinct methods exist to derive fluxes using atmospheric models, i.e. forward simulations and inverse methods. In forward atmospheric modelling, emissions are used to drive the atmospheric model to predict concentrations and fluxes. Inverse methods use observations of atmospheric concentrations to derive flux estimates (Heimann and Kaminski, 1999). Both methods can be applied at all scale levels. The necessary level of detail of the descriptions of controlling processes depends on the spatial and temporal resolution required. For both methods the inflow and outflow for the model area considered need to be known from measurements or modelling. Data on inflow and outflow become more critical as the spatial scale decreases, being most critical for gases with long lifetimes, and less important for highly reactive gases.
3.1. Forward modelling The concentration and deposition fields simulated with forward modeling can be validated by comparing the model results with measured concentrations and fluxes. After validation the model can be used to interpolate in space and time between measurement sites. For a sound validation the measurement data should be representative of the spatial scale and the time period considered (Sofiev, 1999). Measurements should be representative for an area having the same size as the model's
92
W.A.H. A s m a n et al.
grid elements. In practice the area represented by a measurement site depends on the particular situation around the site. If there are strong signals from sources nearby, the measurement data only represent a limited area. The variability of the concentration of a trace gas within a model grid cell and the standard deviation of the model results should be taken into account. This variability can be quantified as a standard deviation of a point measurement in a model grid cell. A first estimate of the spatial standard deviation at one site can be obtained from the standard deviation of a time series at that site, although the spatial standard deviation is often likely to be greater. An estimate of the standard deviation of the model results can be obtained from a sensitivity analysis with the model, where model parameters are varied within likely ranges. Another estimate of the uncertainties of the model results for longer averaging periods can be found from the model results that are used to calculate the average. The time period represented by the model simulations and that of the measurements should be identical. In general, results represent a larger area when the averaging time is longer.
3.2. Inverse models
Inverse models use observations of atmospheric concentrations to derive flux estimates. The inverse problem can be formulated as: Ac = T ( q )
(1)
where /x c is the temporal change in the vector of the observed concentrations, q is the vector of sources and sinks, and T is the transport model. The transport model can not be used to describe the backward transport, as it is not possible to model backward diffusion. The unknown sources and sinks can be found by minimizing the difference between measured and modelled concentrations by varying the source and sink strengths. Similar to forward modelling, modelled and observed concentrations are compared and, therefore, inverse modelling has the same requirements regarding the representativeness of measurement sites as forward modelling. Inverse methods can provide an estimate of the unknown flux vector and an error covariance matrix quantifying the uncertainty in this flux estimate. In contrast to forward modelling, both are determined by objective minimization algorithms (e.g. Menke, 1989). In principle, the method can be applied both to reactive and inert trace gases. However, solving the inverse problem can become very complicated for a reactive trace gas, depending on the number of its reaction partners. For a number of trace gases (e.g., CO2, CH4, or F11), however, transport and chemistry in equation (1) can be approximated by a linear relationship (T), which links sources and sinks to the concentrations. In that case, the source and sink estimate as well as the uncertainties are directly given by algebraic equations. The relation of the inverse problem to scaling issues for unknown fluxes will be illustrated with two examples for the inversion of CO2 on different spatial scales. In the first example, a globally uniform flux field is to be inferred from atmospheric observations at a monitoring network of 25 stations. This is an overdetermined system, since there are 25 constraints for 1 unknown variable. Then, the flux can be determined with a relatively small uncertainty. In case of spatial inhomogeneity of the atmospheric measurements, however, the flux estimate will be biased. Assuming that all stations are located in one hemisphere and that the flux from that hemisphere is larger than that from the other one, the implied increase in the concentration on that hemisphere will be higher than the increase in the global average concentration. Consequently the inversion, which is based on this inhomogeneous network, will overestimate the global flux. In the second example the earth's surface is divided into more than 25 unknown source
How can fluxes of trace gases be validated between different scales ?
93
regions. This is an under-determined system. In this case, information on the unknown fluxes has to be combined with the atmospheric measurements, such as flux estimates on the basis of an emission inventory, process models or other measurements. There are now two constraints, being the modelled concentrations which should not deviate too much from the measured ones, and the flux which should not deviate too much from the a p r i o r i flux estimate. For this relatively high resolution of the fluxes, the bias in the estimate is reduced compared to the first example. But the uncertainty for the various flux components is higher. The reason for this is that in the first example the coupling of all the flux components together is an extremely strong constraint on the solution, but it does not take the spatial variability of the fluxes into account. So far, inverse modelling has been mainly applied to obtain global emission fluxes. Recently, the method has also been applied to find the CH4 emissions of Northwestern European countries, using measurements at 200 m height at the Cabauw meteorological tower in the Netherlands and a simple atmospheric transport model using trajectories (Vermeulen et al., 1998). The results were in good agreement with existing emission inventories. Inverse runs are also used to optimally locate future observation sites based on specific hypotheses for the sources and sinks of the trace gas to be tested (e.g. Hartley and Prinn, 1993).
4. Application of multiple measurements In many cases fluxes can be determined with different techniques, used simultaneously. If results of such multiple or "redundant" measurements are similar, the confidence in the results of the individual techniques is increased. In addition, instead of only measuring the net flux to or from an ecosystem, the components of the flux in the ecosystem (e.g. soil and canopy fluxes) can be measured. This reflects that the net flux depends on different processes and, therefore, is likely to give better possibilities for upscaling. The usefulness of multiple measurements can be illustrated for micrometeorological techniques in combination with other methods. Conventional micrometeorological methods measure only net exchange over a surface, but give no information regarding the underlying processes of trace gas production and consumption. For example, CO2 is taken up by plant canopies through photosynthesis, and it is released by root and heterotrophic respiration in soils and by respiration in above-ground components of plants. On time scales of hours to seasons, these components of the net C flux may be measured directly using a combination of techniques: (i) chamber measurements of soil respiration; (ii) below-canopy eddy covariance measurements of the CO2 flux inferring the respiration; (iii) measurement of leaf-level photosynthesis rates; and (iv) inference of the vertical distribution of sources and sinks within the canopy from CO2 profile measurements (Denmead, 1995). Comparison of the sum of below-canopy fluxes with the eddy flux measurement of the whole ecosystem flux requires some extrapolation. For example, the leaf-level photosynthesis rates depend on local light intensity, which will vary through the canopy. Estimating phototsynthesis for a whole canopy thus requires an estimate of total leaf area and variation of light intensity in the canopy. Similarly, CO2 emission rates from soils may will spatially (e.g., according to soil organic matter content, pH and drainage). For a gas that is both produced and consumed within an ecosystem, such as CO2, the net annual flux represents the difference of two very large numbers. Again, when measured for the whole ecosystem by eddy covariance, there is no information as to where, when and by which process C is being stored or released. Independent estimates of the rate of annual C flux (e.g. in a forest) may be obtained from investigation of the rate of tree growth (from tree ring
94
W.A.H. Asman et al.
widths or dendrometer band measurements plus allometry to relate the growth of the stem to the growth of the tree), litterfall, root growth and mortality, and soil C storage or release. Because annual increments of growth or C storage may be small and hard to detect (for example in soils), chronosequences (the substitution of space for time) or long-term observations (for rates of tree recruitment and mortality) are needed. Long-term estimates give rates averaged over several years to decades and therefore may not exactly represent the magnitude of the same flux during the specific year when eddy covariance measurements are made. Multiple approaches to determine the C budget for a single stand of black spruce boreal forest have been described in Goulden et al. (1998). Fire plays a dominant role in the C cycle of upland boreal ecosystems. C storage in trees and soil moss detrital layers was determined for sites with similar drainage conditions and vegetation but with differences in the time passed since the last fire. These measurements were used to estimate the rate of C sequestration in regrowing moss and trees recovering from fire. These estimates suggest the 120 year old black spruce stand in the flux tower footprint should be accumulating carbon at an average rate of 0.6+0.2 tons C ha-~yr~. The eddy covariance measurements showed that the ecosystem lost 0.5+0.3 tons C ha~yr ~, averaged over three years. To account for the difference between estimates, Goulden et al. (1998) had to infer a net loss of 0.8+0.5 tons C halyr l from humic materials stored deeper in the soil profile (below surface detritus but above the mineral soils). Three lines of independent evidence supported this conclusion. Automated chamber measurements showed that soil respiration increased through the summer/fall period as the deeper portion of the soil profile thawed and warmed (even as air temperatures were decreasing). The magnitude of increased respiration was consistent with a 0.8 ton C halyr l loss. Radiocarbon measurements of the respired CO2 showed that the source of soil respiration in the fall and winter was derived from decomposition of organic matter more than 30 and up to several hundred years old. Measurements of soil organic matter inventory and radiocarbon content showed that large amounts of carbon of that age were stored in humic materials at depths which thawed and warmed in the late summer and fall. Jar incubations showed that, when thawed and warmed, this deep soil organic matter decomposed rapidly. These multiple lines of supporting evidence build confidence that a net loss of C from deeper organic layers was occurring, which balanced the C gains in surface detritus and trees. Scaling of eddy covariance-based tower fluxes to larger regions is fraught with uncertainty, because of the heterogeneity of fluxes at the landscape scale. Using boreal forests as an example, a 120 yr old black spruce stand was not a strong annual source or sink of carbon based on eddy flux measurements. The average age of forest stands in the surrounding region was 40-50 yr, and these stands are accumulating carbon in growing trees and moss. Assuming losses of deep soil C do not vary much with stand age, the younger stands should be accumulating C. Over very large regions, however, burning of only 1% of the forested area (consistent with an average fire recurrence interval of 100 years) will offset the C stored in the other 99% of area that is regrowing forest or wetland (Rapalee et al., 1998). Thus over very large areas and very long time-scales (averaging over many fire cycles), the net storage of C in these ecosystems should be smaller than that measured in a single (unburned) forest stand. Since the area burned varies widely from year to year, the interannual variability in C storage for a region may reflect the occurrence of fire more than the response of vegetation and soils to shifts in regional climate.
5. Conclusions and recommendations Footprints in Figure 1 show that direct flux measurements can be made on space scales up to
95
How can fluxes of trace gases be validated between different scales ?
le-01
le+O0
le+01
le+02
le+03
le+04
le+05
le+06
footprint (m)
Figure 1. Footprints of different measurementmethods to derive surface fluxes. Methods and their footprints are: (1) Soil gas concentration profile method, 0.1-0.5 m; (2) Chamber technique, 0.5-5 m; (3) Micrometeorological tower measurements, 10-500 m; (4) Nocturnal boundary layer budgeting method, 1-5 km; (5) Convective boundary layer budgeting method, 5-30 km; (6) Airborne micrometeorological techniques, 10-100 km; (7) Airborne mass balance method, 3-1000 km. Not all methods presented can be applied under all meteorological conditions or at sea (see text).
1000 km. This suggests that there are no large gaps between the scale levels for which the different methods can be applied. However, this assumption is not correct, as most techniques can only be applied under certain meteorological conditions. Therefore, different techniques should be used simultaneously to assess the flux at a particular level of scale. It also means that models should be applied to interpolate in time between the periods for which flux measurements are available. This is not always possible. The nocturnal boundary layer budgeting method, for example, infers nighttime fluxes on scales of 1-5 km, which could lead to serious errors in assessing source and sink strengths if the daytime fluxes are not known. Examples where the night and day fluxes differ, include CO2 exchange between plants and the atmosphere (where day and night fluxes are of opposite sign), and N20 and NH3 emission by soils for which fluxes are usually smallest at night. This observational gap can be filled by conventional, tower-based, micrometeorological flux measurements at, say 20 m or more above the surface, or by aircraft. Generally, such measurements are difficult to organize. Flux measurements on a scale of 10-1000 km are rare, but there are several methods to cover this scale level. Two distinct methods are used to derive fluxes using atmospheric models, forward simulations and inverse methods. In forward atmospheric modelling emissions are used to drive the atmospheric model to predict concentrations and fluxes. Predicted concentrations and fluxes can be validated against measurements. Inverse methods use observations of atmos-pheric concentrations to derive flux estimates. Major recommendations for future research on validation are that: - Inverse modelling techniques for the estimation of fluxes at continental scales should be developed further. This requires model investigation of the optimal location of continental sampling sites, and the development of techniques to derive grid scale concentration estimates from point measurements at a station. - Measurement programmes should include multiple, or "redundant" approaches to constrain important mechanisms or processes and to reduce uncertainties in flux estimates.
96
W.A.H. Asman et al.
- Validation of flux measurements for large (country or continental) spatial scales requires the appropriate combination of data from satellite surface sensors and satellite borne, airborne and ground-based concentration measurements. This requires co-ordination of the different research groups working on these problems, as discussed in detail by Burrows (1999).
References Ambus, P. and S. Christensen (1994) Measurement of N20 emission from a fertilized grassland: An analysis of spatial variability. Journal of Geophysical Research 99:16549-16555. Born, M., H. D6rr and I. Levin (1990) Methane consumption in aerated soils of the temperate zone. Tellus 42B:2-8. Baldocchi, D. and R. Valentinti (eds.) (1996) Strategies for monitoring and modelling CO2 and water vapour fluxes over terrestrial ecosystems. Thematic number. Global Change Biology 2(3). Burrows, J.P. (1999) Current and future passive remote sensing techniques used to determine atmospheric constituents. In: A.F. Bouwman (Ed.) Approaches to scaling of trace gas fluxes in ecosystems. Developments in Atmospheric science 24. Elsevier, Amsterdam, pp. 315-347. Denmead, O.T. (1995) Novel meteorological methods for measuring trace gas fluxes. Philosophical Transactions of the Royal Society London A 351:383-396. Denmead, O.T., R. Leuning, D.W.T. Griffith and C.P. Meyer (1999) Some recent developments in trace gas flux measurement In: A.F. Bouwman (Ed.) Approaches to scaling of trace gas fluxes in ecosystems. Developments in Atmospheric Science 24. Elsevier, Amsterdam, pp. 67-84. Fowler, D. (1999) Experimental designs appropriate for flux determination in terrestrial and aquatic systems. In: A.F. Bouwman (Ed.) Approaches to scaling of trace gas fluxes in ecosystems. Developments in Atmospheric Science 24. Elsevier, Amsterdam, pp. 99-121. Galbally, I.E. and C. Johansson (1989) A model relating laboratory measurements of rates of nitric oxide production and field measurements of nitric oxide emission from soils. Journal of Geophysical Research 94:6473-6480. Goulden, M.L., S.C. Wofsy, J.W. Harden, S.E. Trumbore, P.M. Crill, S.T. Gower, T. Fries, B.C. Daube, S.-M. Fan, D.J. Sutton, A. Bazzaz and J.W. Munger (1998) Sensitivity of boreal forest carbon balance to soil thaw. Science 279:214-217. Hartley, D. and R. Prinn (1993) Feasibility of determining surface emissions of trace gases using an inverse method in a three-dimensional chemical tracer transport model. Journal of Geophysical Research 98:5183-5197. Heimann, M. and T. Kaminski (1999) Inverse modelling approaches to infer surface trace gas fluxes from observed atmospheric mixing ratios. In: A.F. Bouwman (Ed.) Approaches to scaling of trace gasfluxes in ecosystems. Developments in Atmospheric Science 24. Elsevier, Amsterdam, pp. 275295. Koschorreck, M. and R. Conrad (1993) Oxidation of atmospheric methane in soil: Measurements in the field, in soil cores and in soil samples. Global Biogeochemical Cycles 7:109-121. Lapitan, R.L., R. Wanninkhof and A.R. Mosier (1999) Methods for stable gas flux determination in aquatic and terrestrial systems. In: A.F. Bouwman (Ed.) Approaches to scaling of trace gas fluxes in ecosystems. Developments in Atmospheric Science 24. Elsevier, Amsterdam, pp. 27-66. Leclerc, M.Y. and G.W. Thurtell (1993) Footprint prediction of scalar fluxes using a Markovian analysis. Boundary-Layer Meteorology 52:247-258. Liss, P.S. and L. Merlivat (1986) Air-sea gas exchange rates: Introduction and synthesis, In: P. BuatMenard (ed.) The role of air-sea exchange in geochemical cycling. Reidel, Dordrecht, The Netherlands, pp. 113-129. Matson, P.A., P.M. Vitousek and D.S. Schimel (1989) Regional extrapolation of trace gas flux based on soils and ecosystems. In: M.O. Andreae and D.S. Schimel (eds.) Exchange of trace gases between terrestrial ecosystems and the atmosphere. Wiley and Sons, Chichester, pp. 97-108.
How canfluxes of trace gases be validated between different scales ?
97
Menke,W. (1989) Geophysical data analysis. Academic Press, San Diego, USA. Rapalee, G., S.E. Trumbore, E.A. Davidson, J W. Harden and H. Veldhuis (1998) Scaling soil carbon stocks and fluxes in a boreal forest landscape. Submitted to Global Biogeochemical Cycles. Remde, A., J. Ludwig, F.X. Meixner and R. Conrad (1993) A study to explain the emission of nitric oxide from a marsh soil. Journal of Atmospheric Chemistry 17:249-275. Rudolph, J., F. Rothfuss and R. Conrad (1996) Flux between soil and atmosphere, vertical concentration profiles in soil, and turnover of nitric oxide. 1. Measurements on a model soil core. Journal of Atmospheric Chemistry 23:253-273. Seitzinger, S., J.-P. Malingreau, N.H. Batjes, A.F. Bouwman, J. Burrows, J.E. Estes, J. Fowler, M. Frankignoulle and R.L. Lapitan (1999) How can we best define functional types and integrate state variables and properties in space and time ? In: A.F. Bouwman (Ed.) Approaches to scaling of trace gasfluxes in ecosystems. Developments in Atmospheric Science 24. Elsevier, Amsterdam, pp. 151167. Smith, K.A., A. Scott, B. Galle and L. Klemedtsson (1994) Use of a long-path infrared gas monitor for measurement of nitrous oxide flux from soil. Journal of Geophysical Research 99:1658516592. Sofiev, M. (1999) Validation of model results on different scales. In: A.F. Bouwman (Ed.) Approaches to scaling of trace gas fluxes in ecosystems. Developments in Atmospheric Science 24. Elsevier, Amsterdam, pp. 233-255. Trumbore, S. (1999) Role of isotopes and tracers in scaling trace gas fluxes. In: A.F. Bouwman (Ed.) Approaches to scaling of trace gas fluxes in ecosystems. Developments in Atmospheric Science 24. Elsevier, Amsterdam, pp. 257-274. Vermeulen, A.T., R. Eisma, A. Hensen and J. Slanina (1998) Transport model calculations of NWEurope methane emissions. Submitted to Climatic change. Wanninkhof, R. (1992) Relationship between gas exchange and wind speed over the ocean. Journal of Geophysical Research 98:20237-20248. Whalen, S.C., W.S. Reeburgh and V.A. Barber (1992) Oxidation of methane in boreal forest soils - A comparison of 7 measures. Biogeochemistry 16:181-211.
This Page Intentionally Left Blank
Chapter 5
EXPERIMENTAL DESIGNS APPROPRIATE FOR FLUX DETERMINATION IN TERRESTRIAL AND AQUATIC ECOSYSTEMS
D. Fowler
This Page Intentionally Left Blank
Approaches to scaling a trace gas fluxes in ecosystems A.F. Bouwman, editor 9 Elsevier Science B.V. All rights reserved
EXPERIMENTAL DESIGNS APPROPRIATE FOR FLUX DETERMINATION IN TERRESTRIAL AND AQUATIC ECOSYSTEMS
D. Fowler Institute of Terrestrial Ecology, Bush Estate, Penicuik, Midlothian EH26 0QB, UK
1. I n t r o d u c t i o n
Current interest in quantifying the exchange of trace gases and particles between terrestrial surfaces and the atmosphere has been stimulated primarily by observed increases in atmospheric concentrations in a range of trace atmospheric constituents. In some cases the chemical species involved have relatively short atmospheric lifetimes, for example gaseous ammonia (NH3) with a lifetime of a few hours. On the other hand nitrous oxide (N20) with slow atmospheric removal has an atmospheric lifetime in excess of 100 years. The search for the sources and sinks of each of the trace gases has shown with very few exceptions that sources or sinks of the trace gas in question in soils or vegetation play a major role in the atmospheric budget. Considering further the two examples, in the case of NH3 terrestrial surfaces are both sources and sinks depending on the relative concentrations at the surface (e.g. as apoplastic ammonium, NH4+) or in the atmosphere as NH3. Whereas for N20 terrestrial surfaces represent the major global source and a negligible sink, thus the methodology must be appropriate for bi-directional exchange. The techniques developed for measurement of trace gas fluxes are not unique to this scientific field, they are generally methods taken from the closely related fields of micrometeorology, environmental physics and plant ecology (Monteith, 1973; Woodward and Sheehy, 1983). The motives within these parallel areas of study were to understand the degree to which the exchanges of heat, water vapour and momentum were influenced by biological and physical processes and demonstrated at an early stage a major role of the vegetation in regulating the transpiration flux of water vapour above crop canopies. In most studies of land-atmosphere exchange processes to date the objectives have been to understand specific aspects of the underlying process for a particular surface. The up-scaling of the flux to much larger areas or for longer periods has generally been achieved through models. In developing some of the most elegant, yet simple models this approach has been highly successful, for example evapotranspiration of water from field crops calculated using the Penman-Monteith equation which is widely applied throughout the world. Complex treatments of the land-surface exchange of momentum, sensible and latent heat fluxes are included in a range of meteorological models. The land-atmosphere exchange processes are incorporated into models of long range transport and deposition of pollutants. Increasingly, these models are being adapted to include new developments in understanding land-atmosphere exchange of reactive trace gases (Sorteborg and Hov, 1996). An important difference between the studies of water vapour and energy fluxes and those of the trace gases, excepting the many orders of magnitude in the size of the fluxes, are the underlying links with the partitioning of energy at terrestrial surfaces. In the case of water and
102
D. Fowler
energy fluxes, a detailed study of the energy balance and the water fluxes show a strong dependence which is amenable to rigorous analysis of the underlying physical processes. In the case of the trace gases, while temperature and the energy balance at the surface may influence rates of exchange, the links are generally indirect, or conditional. For some of the reactive trace gases, chemical interaction with the surfaces of foliage and interaction with water films regulate rates of exchange (Flechard and F~'wler, 1998). The design of methods to measure fluxes of trace gases averaged over the range of scales 10 m 2 to 104 m 2 of terrestrial or aquatic surface include a well tried array of techniques from chambers to micrometeorology. For the second objective for this background paper, obtaining flux at a landscape scale or as a function of environmental characteristics requires more than simply obtaining fluxes at a flat, uniform and micrometeorologically ideal site. In this case, the objectives include identification of the quantitative influence of components of the landscape within the flux footprint on the observed rates of exchange. For this latter approach the measurement method is being used to investigate processes rather than simply obtain representative fluxes. The paper briefly outlines the main methods and some recent additions, drawing attention to the measurement strategies necessary for particular objectives. Lapitan et al. (1999) and Denmead et al. (1999) discuss the different measurement techniques used for determination of trace gas fluxes. The following section provides examples of the approaches available to quantify the influence of underlying processes on fluxes or contributions to the flux from components of the landscape.
2. Flux m e a s u r e m e n t at the 0.1-10 m 2 scale; enclosure methods
The measurement of trace gas fluxes at the 0.1 to 10 m 2 scale have been achieved primarily using enclosure methods. Such techniques include static (or closed) enclosures in which the rate of change of the trace gas concentration within the enclosure with time is monitored to determine the net exchange at the surface. Fc -
Ac V
t A
(1)
in which Fc is the net gas flux, Ac/t is the rate of change of trace gas concentration within the enclosure, V is the volume of enclosure, A is the area of source or sink (i.e. soil surface or leaf area). This method is particularly well suited to unreactive gases such as N20 and methane (CH4) for which the rates of reaction onte zhamber surfaces or with atmost-.heric gases within the chamber are small. The method has been widely applied to obtain estimates of fluxes of N20 (Smith et al., 1994; Skiba et aL, 1992) and CH4 in a wide range of soils (Sass et al., 1990; Schtitz et al., 1989; MacDonald et al., 1998). In particular, the static chamber method has been the method of choice for large numbers of individual measurements to quantify the spatial variability in trace gas fluxes over agricultural fields. For N20 flux measurement for example, estimates of the spatial variability in NzO flux within a 100 m 2 area of grassland has been shown to range over two orders of magnitude. Hence, determination of field-scale fluxes requires very large numbers of individual chambers to yield a statistically satisfactory flux estimate. Such application of chambers is not a practical method for the field scale. The alternative flux measurement method using enclosures is the dynamic method in which the difference in concentration between inlet and outlet (Co-C0 combined with the volume flow through the enclosure provides the flux which is then expressed per unit soil or leaf area within the enclosure:
Experimental designs appropriate f o r f l u x determination in terrestrial and aquatic ecosystems
103
Table 1. The loss of fertilizer N as NO for a range of N fertilizers and agricultural soils. Crop
N fertilizer
Bare soil, Germany
NaNO3 NH4NO3 NH4C1 NaNO3 NH4C1 Urea NaNO3 NH4C1 Urea NaNO3 NH4NO3 NH4C1 Urea 33% NaNO3, 67% limestone Manure NI-I4CI NaNO3 NH4C1 Urea NH4C1
Bare soil, Germany
Bare soil, Ge~many
Bare soil, Spain
Bare soil, Canada Bare soil Grass, Germany Grass, clover, dandelion, Germany
Plant incorporation experiment, plants as above, Germany Bare soil Soil mixed with plants Soil mixed Bare soil Perennial ryegrass, UK Perennial ryegrass, UK Grass ley, Sweden Bermuda grass, Texas, USA
Urine (NH4)2SO4 Ca(NO3)2 (NH4)2SO4
N application rate (kg ha -1)
% loss as NO
100
0.04 0.63 1.52 0.04 0.15 0.11 0 03 0.07 0.04 0.14 0.64 1.23 3.25 11 0.26 0.52 0.003 0.05 0.01
100
100
100
0-100 160-800 100 100
Reference
a
100
447 100 200 52
0.02 0.08 0.16 0.14 0.03 0.34 0.3 3.2
7 8 9 10
From Skiba et al. (1997). a 1, Slemr and Seiler (1984); 2, Slemr and Seiler (1991) (experiment 2, 15.7 to 3.8.1983); 3, Slemr and Seiler (1991) (experiment 4, 24.8 to 18.9.1983); 4, Sheperd et al. (1991); 5, Paul et al. (1993); 6, Slemr and Seiler (1991) (experiment 3, 4.8 to 14.8.1983); 7, Colboum et al. (1987); 8, Skiba et al. (1993); 9, Johansson and Granat (1984); 10, Hutchinson and Brams (1992).
r = (Co-C,)K A
(2)
where Co and C; are outlet and inlet concentrations respectively and 7,,. is the volume flow rate
of air through the chamber. In practice, both static and dynamic chamber methods are subject to a wide range of errors, including modification of environmental conditions within the chamber relative to those in the field. The modification includes: (i) the radiation budget in which almost all terms, in the short and long wavelength categories, are modified; (ii) the humidity (and if enclosure times are long, of CO2) and its feed back on stomatal function and the net exchange of CO2 and H20 with vegetation; (iii) modification of the turbulent structure of air and its profile through plant canopies; and, (iv) effect of pressure fluctuation on soilatmosphere exchange. The chamber methods have also been applied to measure nitric oxide (NO), nitrogen dioxide (NO2), ozone (O3), carbonyl sulphide (COS), dimethylsulphide (DMS), carbon disulphide (CS2), hydrogensulphide (H2S), sulphur dioxide (SO2) and a range of biogenic volatile organic compounds (VOC) (Slanina, 1997). In general the experimental design has been either: (i) process orientated - designed to investigate production, emission or mecha-
104
D. Fowler
nisms of exchange; (ii) designed for long-term flux measurement, to follow the sea-sonal or annual time course in fluxes of trace gases which show large temporal cycles; and, (iii) complementary flux measurements. Measurements of a particular trace gas flux to complement a much broader study, e.g. of a campaign to test methodologies of trace gas measurement, or the fate of applied fertilizer, or to define the net inputs or losses in ecosystem studies. Most measurements of trace gas fluxes to date were process oriented. In some studies a relatively short period of fluxes were measured and reported simply to identify a new source or sink for an atmospheric trace gas. For example, in the measurement of NO the detection of an emission flux from soil is relatively recent and early measurements simply identified the presence of this loss mechanism for fixed nitrogen from soil, its magnitude and likely importance to chemical processes in the atmosphere (Galbally and Roy, 1978). The use of enclosure methods was entirely appropriate for such studies, is much simpler than micrometeorological approaches and with an appropriate enclosure provides an unambiguous signal to interpret. The bulk of the published literature on emissions of NO from soil (a selection of which is provided in Table 1) have been observed using enclosure methods, but the range and complexity of chambers for the measurements is considerable. In general, dynamic chambers with substantial volume exchange rates and additional turbulent mixing where necessary, are preferred in the recent studies as these minimize the difference between ambient conditions and those within the enclosure (Figure 1). The use of chambers had focussed on the questions which can with minor modification be applied equally to most of the other trace gases: (i) what are the field emission rates of NO from soils subject to a range of fertilizer nitrogen inputs; (ii) which soil chemical and physical variables regulate NO fluxes, and can functional relationships between flux and the variables be established; and, (iii) can a set of readily obtained environmental variables be used to simulate long-term NO emission from soil to provide the basis for extrapolation to seasonal and field scale emission estimates, and validated against direct measurements. While progress with the fist two questions has been clear, few studies to date have been designed to address the third question. The major variables which regulate NO emission from soil have been identified from work listed in Table 1 and models of the process have been developed by e.g. Galbally and Johansson (1989). The extrapolation of NO fluxes to provide annual emissions over regions has however been based on a much smaller range of variables. For example, the approach of Williams et al. (1992) is based on a very simple exponential response of the NO emission rate to temperature: Fr
= A exp (B. Lo,,)
(3)
where F~vo~ is the flux in ng N m -2 s~, A is a factor to characterise soil properties and has units ofng N m 2 s1 and B is a temperature coefficient (0.071 + 0.007 ~ The values of A were obtained from land use maps and B was provided empirically from measured temperature responses of NO emission to soil temperature. While source strength estimates from up-scaling of this kind are necessary to compare the magnitude of this source term with other sources, the uncertainty introduced by the approach must be recognized. In particular, it is widely recognized that soil water, and in particular the water filled pore space strongly influences rates of gaseous exchange between soils and the atmosphere, has a major influence on NO fluxes. The measurements of Ludwig and Meixner (1994) over an arable soil showed that in very dry soils, the emissions of NO are small and that available soil nitrogen supply is also an important variable, whereas at intermediate soil water contents large emissions of NO were observed. The experimental design therefore needs to capture the range of physical, chemical and biological variability which can then be incorporated in subsequent models. The model extrapolation then remains within the boundaries of the parameter space validated by measurements.
105
Experimental designs appropriate forflux determination in terrestrial and aquatic ecosystems
perspex
chamber /air
'i [ charcoal
filter & fan
i,
fan
ilililil
baffle /
Ii "It(~ inlet air I .... .................................. 9 - -- -outlet air: ....
[ 1
air/
inlet--
I
outlet I
inlet
airl[
il il ..-:l
glasswool Drierite
1
soda lime
air
I
Drierite
I I
:i:i:i:i:i:i:i!i:i:!:i:!:i:i:i:i:i:i:i:i:~:i~!iiii:i:!:i:i:i:i: :::::::::::::::::::::::::::::::::::::::::::flame i::i::i::i::i::i
.
~so I
i!!!i!i!!!i!!!i!!!!!i!i!i!i!!!i!i-i!i!i!i!i!i!!!i!i!!!i!i!i!i!!
"/Z/////////////;;(~ V//////////////Z/,
I I I 1
-- --" PTFE tubing ....... nylon tubing
Nmolecular 20 trap: sieve 5A
lenoid
valve
I 1
r
(~ ~.(~ flowmeter NO NO, ,
i
1
Figure 1. a) Dynamic chamber design; b) Schematic figure of the dynamic flow-through chamber. The walls and all internal surfaces are fluorinated ethylene propylene (FEP) Teflon (From Kim et al., 1994).
The spatial variability has been addressed at the 1 m 2 scale using chamber methods and has been one of the limitations in the methodology for provision of fluxes at the field (104 m 2 to 106 m 2) scale for trace gases which show large spatial variability. Among the best examples of the spatial variability are provided by the literature on emissions of N20 and methane (CH 0. In the case of N20 emission the spatial heterogeneity is large because the combination of key variables which regulate N20 flux also shows large spatial variability. The denitrification process requires anoxic conditions which may occur in microsites or over a large volume of soil, and available soil nitrogen. For exchange with the atmosphere a diffusion pathway to the soil surface is required. In addition, microbiological processes are temperature sensitive. Given these constraints, and temporal and spatial variability in the variables, the observed range of emission fluxes in a single field of 20 to 500 ng N20-N m 2 s1 is not surprising (Christensen et aL, 1996). The scale of the variability in mean fluxes using simple static chamber methods may be seen in the diagram (Figure 2). Such variability can be quantified directly and used to deduce the number of chambers necessary to obtain field scale fluxes (Christensen et al., 1996).
D. Fowler
106
*
,26
/
.
//'s'///A
X
"~
-
28.4
21.3
o
14.2
-
71
0.0
~
~
0
0.0
l 142
71
! 28.a
21.3
i 35.5
a2 6
Distance (m) Figure 2. "Contour map" of N20 flux within 42 6 x 42.6 m grid sampling area, sho:~,ing three simulated positions of megachamber (values on contour lines in g N20-N ha .] d-~. (From Smith et al., 1994).
However, the fluxes over the field scale are more easily measured directly with current sensitive detectors including tunable diode laser absorption spectroscopy (TDLAS) which is ideal for micrometeorological flux measurements (Zahniser et al., 1995). Thus the chamber methods are limited in scale to 0.1 to 102 m2. The upper limit represents the megachamber approach such as that used by Smith et al. (1994) and also by Galle et al.
r
c o ".,=
5
m
~a "6 ~
E
2
3 Z
1
96
120
144
168
192
216
240
264
N=O flux (g N=O-N ha ~ d ~)
Figure 3. Frequency distribution of fluxes from 51 simulated random positions of megachamber (from Smith et aL, 1994).
Experimental designs appropriate forflux determination in terrestrial and aquatic ecosystems
107
(1994). The size of the large flexible wall chambers vary in the range 20 m e to 100 m 2 and provide a very useful means of integrating the very fine scale variations in emissions. The broader scale variations over a field, as illustrated in Figure 2, would limit the ability of even the large chambers to provide field scale flux estimates. An illustration of the spatial variability in N20 fluxes over agricultural grassland has been provided by Smith et al. (1994) from an experiment in which a range of different chamber sizes (0.008 m 2 to 62 m 2) were analyzed to map the fine scale variability in N20 flux over an area of 43 m x 43 m (as illustrated in Figure 2). A simulation of fluxes averaged over 62 m 2 at 51 random positions within the mapped area yielded a mean flux of 145 ng N20-N m 2 s-1 with a coefficient of variation of 25%. This mean flux may be compared with the measured mean value obtained using the megachamber at the 3 positions illustrated in Figure 2 that were possible during the study of 253 ng-N20-N m -2 s~ with a coefficient of variation of 67%. The spatial variability simulated over this area is illustrated in Figure 3. This experiment illustrates very clearly the difficulty in obtaining field scale fluxes using chambers for trace gases whose controlling variables show very large spatial variability. Other combinations of chamber measurements which reveal the quality of experimental design include CH4 fluxes over both wetlands (emission) and agricultural and semi-natural soils (deposition or oxidation). Methane emission from rice production has been extensively studied using chamber methods (Schiatz et al., 1989; Sass et al., 1990). While spatial variability in emission exists, the agronomic practices reduce the scale of variability and the more important focus for experimental design is the temporal variability in emission. An automated chamber technique designed for seasonal measurements but with sufficient time resolution to observe diurnal responses in CH4 flux to changes in temperature, water table and the net exchange of carbon dioxide (CO2) to the crop canopy has yielded key data to permit the up-scaling of CH4 emission from rice production (Schtitz et al., 1989). Variability in CH4 source strength at the 1 m 2 to 10 m 2 scale is illustrated by a horizontal transect of CH4 fluxes in peat wetlands (Figure 4). The data illustrated in Figure 4 are taken from peatlands in Scotland (MacDonald et al., 1998) but are very similar to data from other
Figure 4. Methane emission transect.
108
D. Fowler
Table 2. CI-h flux and mean soil physical and chemical characteristics from disturbed and undisturbed forest in Cameroon and Borneo. Wet lowland evergreen dipterocarp tbrest (Borneo)
Moist pre-montane semi-deciduous forest (Cameroon) Near primary forest
Old secondary tbrest
Primary forest
Old secondary forest
Young secondary forest
-27.2(18.6)
-17.5 (11.4)
-15.4 (6.4)
-13.9 (8.4)
-10.8 (9.5)
36.1 (6.3)
33.4 (3.7)
34.4 (5.1)
35.4 (4.9)
31.7 (7.0)
22.9 (0.4)
23.2 (0.6)
24.8
25.1
25.5
0.64 (0.12) 0.97 (0.06)
0.72 (0.12) 1.12 (0.17)
0.75 (0.06) 0.98 (0.09)
0.76 (0.07) 1.01 (0.06)
1.0 (0.14) 1.3 (0.12)
PH c (BaC12)
3.5 3.5
3.7 3.8
3.3
3.1
3.5
Organic carbon c (%)
10.3 3.5
4.5 1.4
3.6 (0.8)
3.3 (1.3)
3.0 (2.0)
Total N c (%)
0.27 0.20
0.21 0.11
0.33 _
0.25 _
0.22 _
CI~ flux a (ng m a s"l) Soil water content
(%) Soil temperature
(~ Bulk density b (g cm3)
From MacDonald (1998). a Median () standard deviation, all other figures are means. In Cameroon the number of CH4 flux measurements is 54, and in Borneo 20. b Upper figure is for 1-5 cm, lower figure tbr 6-10 cm. c Upper figure is tbr top 1-5 cm, lower figure tbr 5-50 cm.
northern peatlands (Bartlett et al., 1992). By contrast, rates of C H 4 oxidation in soils are confined within a narrow range as shown for a broad range of land uses in the UK, Cameroon and Borneo (Table 2). Thus for CH4, the spatial variability in fluxes depend primarily on whether the lanscape contains significant wetland areas (which are large potential sources), or whether wetlands areas (or saturated soils) are absent. In the latter case, CH4 oxidation rates in soil will determine the magnitude of the flux which will be small (1 to 20 ng CH4 m2 s-~), and -ve (i.e. deposition). In mixed landscapes with a small proportion (1% to 20%) occupied by saturated soils or wetland, predicting the net CH4 exchange and its temporal variability is a considerable challenge for experimental approaches or models. The detailed mechanistic studies of fluxes of reduced sulphur compounds in crop canopies and of VOC emissions from vegetation are necessarily limited in spatial scale. In the case of emissions of isoprene and other VOCs for example, it has been very important to establish the source strength of foliage as a function of the local irradiance on leaves and therefore the importance of position within the plant canopy (Steinbrecher et al., 1996). The data for both process study and extrapolation to annual and regional scale emissions has been based on chamber studies for VOCs. The field data using small enclosures over vegetation has provided mean emission fluxes per unit leaf dry weight and converted into standardized emission rates for the leaf temperature and irradiance (GOnther et al., 1991). An example of such studies from the work of Street et al. (1997) is presented in Table 3 for isoprene emission from 6 European tree species. The objectives of these chamber measurements include both improvements in understanding processes of trace gas production or consumption and the provision of key variables to model fluxes to canopies and the landscape scale. In many of the recent studies the response of the VOC emission flux to irradiance and leaf temperature are widely reported (GOnther et al., 1991; Pio et al., 1994).
109
Experimental designs appropriate for flux determination in terrestrial and aquatic ecosystems
Table 3. Summary of total summed, average emission rates of isoprene measured lbr various plant species during 19921994. Mean measured emission rates Standardized emission rates a ng h-I per gram dry.weight ng h -1 per gram dry.weight
Species / type / date Quercus petraea (n=4) (CO2 exp., 1992) Quereus petraea (n=l 1) (03 exp., 1992) Picea sitchensis (n=3) (Temp. exp., controlled env., 1993) Picea sitchensis (n=8-13) Field samples 1993
Picea sitchensis (n=4) (1 sample day, 1993) mean of 3 sample days June/July, 1992 and 1993 Hex europaeus (field samples)
Control Treated Control Treated 21.0 ~ 32.6 ~ Day (16 m) (13 m) Night (16 m) (13 m) Young Old Young (n=9) Old (n=12) Flowering (April 1993) Nonflowering (June 1993)
Quercus ilex (n-9) Field samples, Mediterranean, June 1993 Pinus pinea Field samples, Mediterranean, June 1993 Eucalyptus globulus Field samples, S. Europe, June 1994
Young (n=9) Old (n=-20) Young (n=9) Old (n=37)
2359 4758 9235 10193 10380 6769 10978 7611
2650 + 160 5080 + 1100 829 + 312 915 + 322 2391 6317 1083 + 956 145 + 120 135 :k 75 29 + 20 1562 + 418 218+ 98 1056 + 1090 904 + 559 71-705 b
5506 1082 7622 7396 281-668 b
1780
1866
29996 i 20402
18297
2898 + 1663 9301 + 4505 70509 + 46376 10435 + 9491
2258 7385 61398 17981
From Street et al. (1997) a Values presented for standard conditions of 30~ and 1000 gmol m2 s~ using current models. b Isoprene exceeded detector range, therelbre total estimated should be at least 1705 ng g-i h-l, and 1968 ng g~ h -1 (standardized), all gorse values + 25% on average.
The limitation o f enclosure techniques is their spatial coverage. M o s t o f the artefacts introduced by the enclosure itself can be o v e r c o m e with sophistication o f the equipment or quantified by experiment. The temporal changes can likewise be o v e r c o m e by automating the equipment. Some o f the most significant developments in understanding the mechanisms and providing annual emission fluxes have been provided by long-term automated enclosure methods. The spatial limit o f the enclosure method has recently been extended by the use o f the enclosed catchment at Risdahlsheia (Jenkins, 1997) with a ground area o f 400 m 2 to determine net ecosystem rates o f NO, 03, SO2 and CO2 exchange. Such enclosures present an opportunity to test hypotheses and to develop models describing the major controlling variables. There are also difficulties including the attribution o f the measured fluxes to c o m p o n e n t s o f a complex ecosystem and the same concerns over modification o f the environment. E v e n with these measurements the practical limit for enclosure techniques remains substantially below 103 m z.
3. Measuring fluxes over the field scale (104-106 m 2) The most widely used field scale methods are those using micrometeorological methods which extend from 102 to typically 106 m 2. The upper limit is generally constrained by the availability o f suitable fetch for measurement and cost and convenience. Another the problem is that as the
110
D. Fowler
footprint of the studied area expands, then the requirement for information on properties of the surface which may influence the measured flux increases with the square of the linear extent of the fetch. At the lower limit, fluxes of emitted trace gases have been measured using mass balance methods, especially for NH3 soil emissions (Denmead, 1983; Wilson et al., 1983). The experimental design constraints are similar in principle to those for the chamber studies. In the case of field scale fluxes, the early work followed closely the approach of micrometeorological studies of heat water and momentum, the objective was to obtain sufficient field data to identify the major variables regulating the flux so that scaling could be achieved using models. Seldom did the micrometeorological approaches provide mean fluxes data which could be used directly to represent annual fluxes. At its simplest the field data were used to parameterize components of a resistance network which were then applied to concentration data, e.g. for estimating fluxes of reactive trace gases SO2, NO2 and NH3. These fluxes are now almost invariably simulated using models containing spatially disaggregated land use, meteorological and concentration fields and progress in refining the methods down to finer and finer grid scales is rapid (Erisman and Baldocchi, 1994). For the trace gases N20, CH4 and CO, the focus is primarily on emissions and the scaling problems are larger than those for SO2, NO2 and NH3 as the terrestrial areas over which representative fluxes are required are much larger. Taking emissions of CH4 from high latitude wetlands for example, the terrestrial area of these peat dominated wetlands is of the order 3 to 9 x 1012 m ~ (Sebacher et al., 1986; Mathews and Fung, 1987). The mechanistic basis for extrapolation of emissions of the radiatively active gases is weaker than the deposition of short lived, reactive gases. The reason for this is that at present there are no models which from first principles are able to simulate all of the mechanisms and achieve fluxes of the same order as those measured. Thus, as is the case for VOCs, the response of the measured flux to e.g. temperature, water table and soil NO3 is coupled with mean measured fluxes for particular land use categories in standardized conditions as input in emission models. Thus the measurement database for key ecosystems is essential.
3.1. Eddy co-variance The methods available include Eddy Co-variance (or eddy correlation) in which the vertical flux density, F,., of a trace gas may be written as: ~ =wp.,
(4)
where the bar denotes an average over an appropriate measuring period, w is the vertical velocity and p.s. is the density of the trace gas. This may be considered as the sum of two components, the product of mean vertical windspeed w and the trace gas density A. and fluctuations about the means of the same quantities w' and p's: F.,= w p. + w'p'.,
(5)
where ps. is the gas density and w' and p's are the instantaneous vertical wind velocity and the departure from the mean concentration of the trace gas, respectively. It has been assumed by some that there is no mean vertical flow of air to complicate the above, very simple relation, However, in practice, sensible and latent heat fluxes cause vertical gradients in air density which result in an apparent vertical flow of air. This effect, described by Webb et al. (1980), has important implications for fluxes of some trace gases. The mean w signal resulting from these effects is too small to be detected in field measurements. However, the effect on trace gas fluxes has been estimated by Denmead (1983) from the work of Webb et al. (1980).
Experimental designs appropriate forflux determination in terrestrial and aquatic ecosystems
111
Equation 4 now becomes: F,=W'ps+(p.,/p~)[/2/(l+py)]E+(p.,/p)H/cpT
(6)
In equation (6) H is the sensible heat flux, E the latent heat flux,/2 equals the ratio of molecular weight of dry air to that of water vapour, or= p~/p~; p~ is the density of water vapour; pa is air density; the total air density p = p~ + p~; c, is the specific heat of air at constant pressure; and, T is the air temperature. The corrections are large for any of the trace gas species whose vertical flux is small in relation to the ambient concentration. Vertical fluxes are frequently normalized for ambient concentrations at a reference height S(z) The resulting quantity F]S(~), which is identical to the deposition velocity (vd) (Chamberlain, 1975), has been widely applied in trace gas studies, and the error in determining va in conditions of moderate sensible and latent heat fluxes becomes significant for deposition velocities smaller than 5 mm s-~. These problems, however, may be overcome by the measurement system. For example, if the sample gases are dried and brought to the same temperature or if the mixing ratios are determined for each sampling point, density problems are eliminated.
3.2. Flux-gradient methods 3.2.1. The aerodynamic method
The vertical transport of an entity towards the surface can be described as follows: F,. = -Pa K,~.Ss/Bz
(7)
where Ks is the transfer (diffusion) coefficient for the trace gas s; s (=ps/pa) is the mixing ratio of the gas with respect to dry air; z is height; and 6 s ~ z is the vertical gradient in air concentration in the constant flux layer. If the concentration decreases towards the surface, fis/& is negative by convention and the flux is towards the surface, and vice versa. The flux density for momentum, Fm (more commonly denoted z), may be written as F = PaKm~z~(6u / 6z)
(8)
where u is the wind velocity. The heat flux is then: F h = c p to a
Kmz~(60~z)
(9)
where 0 is the potential temperature. In neutral atmospheric conditions the eddy diffusivities for heat, water vapour, trace gases, and momentum are equal (K,,cz~ = KH~zj, etc.). In these conditions the eddy diffusivity may be determined from the wind profile equation, where: U(z) = U* l n z - d k Zo
(10)
K,,(z~--kU,(z-d)
(11)
and
where U is the windspeed at height z, U, is the friction velocity, k is von Karman's constant, d is the zero place displacement, and zo the roughness length. The measurement of a concentration gradient (6s/fiz) then enables an estimate to be made of the trace gas flux from equation 7, provided the distribution of sources and sinks is such that the value of d is the same for momentum and for the trace gas in question. The zero plane displacement (d) is in principle
112
D. Fowler
an unknown quantity. Over low vegetation (<0.5 m) d is normally small relative to the measurement height. However, over forests the uncertainty in d may be a major limitation for the analysis and interpretation of field measurements. The magnitude of the concentration gradient is small in most circumstances and commonly requires an accuracy in measurement of concentration of 1-4% to be able to detect fluxes. The procedures to correct flux-gradient flux measurements for the effects of strong stability (nocturnal and wintertime) and unstable conditions (summer and daytime) are described by Thom (1975) among others. The range of approaches to make the necessary corrections in generally yield a similar result providing cc,,rections to the flux between 15 to 20%. With much larger corrections, especially in strong stable conditions at night, the uncertainty in the flux/profile relationship becomes very large. Thus there is a practical limit to the application of flux/gradient analysis in strong stabilities which limit flux measurements in these very interesting conditions. 3.2.2. Bowen ratio methods (energy balance)
Another gradient technique which does not require wind velocity profile measurements is that based on the energy balance at the ground. The incoming net radiation (R,) is partitioned between sensible heat flux (H), latent heat flux (2E; E = evaporation; 2 is the latent heat of vaporization; the ratio H / 2 E is the Bowen ratio) and the soil heat flux (G). A further term corresponding to heat storage in the vegetation and air below the height of measurement can become significant for a forest canopy, so that Bowen ratio methods are best suited to low vegetation (grass, agricultural crops, etc.). R =H +2E+G
(12)
Rearranging equation 12 and substituting for H and 2E using equations equivalent to 7 and 9 yields: R , - G = K.~.(pcpSO/Sz + 2 p Sw./Sz)
(13)
where the gradient 8w/Sz is the vertical gradient in absolute humidity. Defining an effective temperature, To, for sensible and latent heat transport, where: T =0
A~
1
p~ 6z (l+r)
(14)
and 7 represents the ratio of the mean densities of water vapour and air. The eddy diffusivity for mass and heat transfer may be obtained as: X.,.-- (R, - G ) / ( p c; 8 Te/Sz)
(15)
and the trace gas flux (F,.) is given by:
t;= (R-G) ~s cp(1 + 7") 8T~
(16)
Energy balance methods therefore require measurements of vertical gradients of gas concentration, temperature and humidity, and these provide estimates of the fluxes of sensible heat, water vapour and the trace gas without the complication and uncertainty introduced by stability corrections. The greatest drawback, and a most important one for trace gas fluxes, is that substantial net radiation fluxes are required for the term (R, - G), as the error in the flux is directly proportional to the accuracy of the (R, - G) measurement. In cloudy, night or winter
Experimental designs appropriate for flux determination in terrestrial and aquatic eco~stems
113
conditions the available net radiation is frequently too small (<50 W m2) to permit satisfactory flux estimates. There are many variations in the formulation of Bowen ratio methods, some of which are described by Thorn (1975) and Woodward and Sheehy (1983). In the most commonly applied modified Bowen ratio technique, the sensible heat flux is measured directly by eddy correlation. This permits measurements of trace gas fluxes from vertical gradients in the trace gas concentration and air temperature and, again, is not subject to stability correction. The zero plane displacement (d) does not have to be evaluated for this method, but the measurements are limited by the assumption that sources and sinks for the trace gases and heat fluxes are distributed similarly. 3.2.3. M a s s balance m e t h o d s
The eddy correlation and flux-gradient methods are generally used to provide average fluxes over large areas of vegetation, generally with a fetch of between 100 and 1000 m. These methods require uniform surface and atmospheric conditions at the site. In contrast, mass balance methods measure the horizontal flux across a vertical plane from an emission area and can be considered as a plume measurement (Denmead, 1983). The flux F.~.into the atmosphere is given by: F.~. : 1/X U p.~.dz
(17)
where x is the fetch, p.s. is the density of gas in excess of background, and the term ups. is the time-averaged horizontal flux density at any level in the plume. The measurements of u and p.~. must cover the vertical zone where air concentrations are modified by the emission patch, and as a rule of thumb the upper height of measurement should be at about 0.1x. Atmospheric stability has a strong influence on the depth of the plume: it increases the upper boundary of the affected layer in unstable conditions by the additional vertical mixing and decreases it in stable conditions as shown by Denmead (1983).
3.3. Campaign measurements The fluxes measured by micrometeorological methods, like those using enclosures have primarily been process studies. Some large field studies such as those at the Halvergate grassland provided a combination of process research and intercomparison of method with measurements of nitric oxide (NO), nitrogen dioxide (NO2), nitric acid (HNO3), nitrous acid (HONO), NH3, SO2, 03 and N20 fluxes at a single site using the enclosure and a range of micrometeorological methods (Fowler et al., 1990; Hargreaves et al., 1992). The benefits of such approaches including rapid development of methods are that the relative magnitudes of individual fluxes at a single site may be compared directly (Figure 5). At this site for example, it is clear that substantial emission fluxes of NO and N20 (Remde et al., 1993) and large deposition fluxes of NH3 dominate the net gaseous exchange of reactive nitrogen. There is a strong demarcation in the published literature between individual gaseous nitrogen compounds in the estimation of net exchange, few studies combine a broad range of trace gas fluxes. Hence, both field studies and upscaling of fluxes tend to be compound specific. Consequently, at sites where considerable scientific advance may be provided by measurements of a range of species, the narrow focus on a few key parameters and one trace gas shows the progress in developing the larger picture. There should therefore be provision in the design of flux measurements for studies which include a broad range of trace gases, and where the overall understanding of processes is enhanced relative to the increased cost and workload.
114
D. Fowler
Figure 5. Halvergatenitrogen budget. (From Fowler et al.. 1990).
3.4. Monitoring micrometeorological methods Until recently the micrometeorological methods were applied as short campaigns, often with a focus on a single gaseous compound, but with some large collaborative studies, as for example, in the EUROTRAC experiments of BIATEX, HALVERGATE (grassland), LEENDE (heathland) and in the BOREAS experiments in North America. However, with the improved reliability of trace gas analysers, micrometeorological instruments and with greatly improved logging and data processing facilities continuous flux measurements have become possible (Erisman and Wyers, 1993). There are now 3 monitoring stations in Europe at which continuous measurement of SO2, NO2, NO, 03 and NH3 fluxes are being made, the same approaches could be used to monitor CH4 or N20 fluxes (Erisman et al., 1998). A much larger programme of continuous CO2 flux measurements is also in progress in the EUROFLUX project and similar long-term flux measurements are in progress in North America. Monitoring trace gas fluxes is now both technically feasible and very desirable for those ecosystems which show large temporal variability and which are poorly characterised. For the gases SO2, NO2, NO, 03 and NH3, progress in understanding the processes of exchange at the surface and the development of process-based models relies much more heavily on the longterm flux measurements than on campaigns (Flechard and Fowler, 1998). The practical application of deposition or emission models are also directed more to annual fluxes than to short term variability (Fowler et al., 1998).
4. Measurement of trace gases at the regional scale Boundary layer budget methods have been used to measure fluxes trace gases over spatial scales of 101 to 104 km 2 (Harriss et al., 1992; Ritter et al., 1992; Choularton et al., 1995). There are several approaches, including nocturnal box and daytime boundary layer box methods.
Experimental designs appropriate for flux determination in terrestrial and aquatic ecosystems
115
4.1. The nocturnal box method
Rates of change of concentration of a trace gas with time within a defined boundary layer depth may be used to deduce the rate of emission or loss at the surface according to: (18)
C(to) + FAt/zi = C(t)
in which F is the vertical flux which is assumed to be horizontally uniform and of constant magnitude, C(to) and C(t) are the concentrations at time zero and time t respectively while 8t is the time interval and zi the boundary layer depth. The boundary layer is assumed well mixed and the concentration increases with time to yield the flux according to: 6COt) 8t
-
F
(19)
zi
4.2. Daytime boundary layer box model
The well mixed daytime boundary layer in conditions with a defined temperature inversion capping the boundary layer, an absence of deep convection from clouds and a steady horizontal wind to advect the trace gases over a region permit the measurement of conservative trace gas fluxes at regional scales using aircraft over scales of up to 1000 km x 1000 km (depending on meteorological conditions and the capabilities of the aircraft). The budget of trace gases in the boundary layer can also be measured using aircraft measurements of eddy fluxes and mean concentration, as for example by Lenshow et al. (1981). In practice the rates of change of 03 concentration with time was the dominant term in the budget equation, and exceeded the eddy flux by more than an order of magnitude. The complication with 03 (and, or SO2, NOx, or NH3) is that chemical processes are involved. The photochemical production of 03 in this case was one of the dominant terms. In the case of a long-lived gas (CH4, N20) the boundary layer budget over timescales of hours are not influenced significantly by chemical processes in the atmosphere. Provided that the exchange of air at the upper surface of the boundary layer is minimized by the presence of a subsidence inversion and that the vertical profile in CH4 (or N20) concentration and wind velocity profile are quantified in the outflow region, the flux may be determined from simple mass balance approaches. The data in Table 4 show a simple mass budget approach for CH4 taken over 5 land trajectories of 126 km to 267 km length over the UK. These result in CH4 fluxes of 52 to 375 ng m 2 s1 averaged over the length of the trajectories. The work has been further extended to es-
Table 4. Summary of land trajectories and fluxes of CH4 estimated from the coastal flight around the northern half of Britain in the boundary layer on 29 November 1994. Fetch a
Time b
(kin)
(s)
193 126 267 174 185
1.7 x 104 1.14 x 104 2.4 x 104 1.57 x 104 1.67 x 104
CH4 emission flux c 52 142 144 220 375
Land trajectories d Portree Mallaig Oban Clyde Ayr
Land fetch ignoring small islands. b Elapsed time along low level trajectory. r Net methane flux from the land surface into the boundary, layer averaged over land fetch. d Start and end points for land trajectories.
Wick Black Isle Peterhead Forth Tyneside
116
D. Fowler
timate the source strength of the UK by measuring the boundary layer CIG and N 2 0 c o n c e n trations in the boundary layer upwind and downwind of the UK coastline in westerly airflow. The surface fluxes were obtained by solving the diffusion equation for a parcel of air advected along a trajectory within the boundary layer with a variable source strength at the surface. The budgets have been extended from CH4 and N20 to include CO and CO2. In the case of CO2 the interpretation is complicated by soil-vegetation-atmosphere exchanges as well as the combustion sources. In the early experiments the biological interactions were simplified by making measurements during the winter months during which the vegetation CO2 sink is very small relative to other terms, and soil respiration is minimized. In such conditions the dominance of combustion sources is clear and national CO2 fluxes may be readily monitored. In a recent study over the UK, CH4, N 2 0 , CO and CO2 fluxes were obtained at the country scale (24 x 106 ha) using the mass balance of the boundary layer to deduce average fluxes from the UK (Table 5). The data provide sufficient spatial detail to assess the relative magnitude of sources from different areas of the country and also source categories. In principle these methods are applicable for all the trace gases which have atmospheric lifetimes in excess of a few days but studies of large scale integration of fluxes in this way need to be focussed on specific research or policy questions. In particular, the countrywide greenhouse gas emission budgets are derived from activity and emission factor approaches with no independent top-down validation. The boundary layer budget approach measured using aircraft (Choularton et al., 1995) or an inverse modelling technique with trajectory based dispersion models and monitoring station data provide such checks on country or regional budgets.
5. Designing micrometeorological studies to address specific questions The standard tests on flux measurements by micrometeorological methods refer to the site uniformity and fetch requirements to obtain vertical fluxes representative of a given terrestrial surface (Arya, 1988). The broad conclusions are that a fetch of 100 m to 200 m is required for each 1 m depth of equilibrium boundary layer, but others (Munro and Oke, 1975; Gash, 1986) suggest that most of the adjustment in equilibrium surface layer depth is achieved over a substantially shorter fetch. Such arguments and concern over substantial errors from non-stationarity due to concentrations of the trace gas changing with time, or to the presence of horizontal gradients in concentration may have constrained the development of innovative approaches to flux measurement. For example, if the objective of a study is simply to obtain trace gas fluxes in the field which satisfy the boundary conditions for a particular analysis, a substantial investment of time locating possible sites and establishment of instruments is required. The data obtained from such a study may reveal useful new data, but whether or not relationships between the flux and features of the site which enable extrapolation of the findings to the landscape in general are often left to chance. The first requirement at the site is to define the footprint of the measure-
Table 5. Atmospheric mass budgets at regional scales. CO2
CO
CI-I4
N20
Upwind: background concentration in Atlantic air
367 ppm
105 ppb
1865 ppb
312.5 ppb
Downwind: boundary layer enhancement
2-20 ppm
10-100 ppb
20-150ppb
0.5-4 ppb
Mass budget are made by sampling of the boundaD, layer of-1 km by aircratt at 30 m, with profiles every 100 km along the trajectory. The method requires a clearly defined boundary layer, absence of deep convection and frontal activity along the trajectory, and steady boundary layer winds (3-15 m sl).
Experimental designs appropriate forflux determination in terrestrial and aquatic ecosystems
117
0 3 3 0 ~ 60
,..3O
270
90
210
150 180
Figure 6. a) Schematic diagram of the Loch More site - 1994, indicating the distribution of pool areas. The numbers indicate the percentage of the sector covered by pools; 1994 mast is at the centre of the circle; b) Sector dependence of methane flux, Loch More, May-June 1994.
ment tower to establish the surface area which contributes to the measured flux and the way this area changes with atmospheric conditions, notably stability, windspeed at a fixed measurement height (Schuepp et al., 1990). In this way the study is able to focus on particular elements of the landscape. An example of the way this feature of the method can be used is taken from CH4 flux measurements over wetlands. By placing the measurement tower at a boundary in the land-
118
D. Fowler
scape between an area with many open pools and a water table very close to the peat surface, and an area of open blanket bog with water table 20 cm below the surface (Figure 6), it was possible to study the fluxes over footprints of the bog surface with a range of water tables. In this way the relationship between CH4 flux and water table was established. The focus on this variable in the study also enabled the temperature response of the CH4 emission flux to be quantified for specific water table footprints (Fowler et al., 1995). The potential of the method therefore allow (i) long-term monitoring of fluxes to parameterize surface atmosphere exchange models over specific surfaces; (ii) campaign studies to quantify the effects of treatments, land use, or specific atmospheric or surface variables on trace gas fluxes; (iii) field measurements to integrate the small scale variability in landatmosphere exchange (e.g. N20); and, (iv) complementary trace gas fluxes, in which the measurements are made to help interpret a field experiment designed for quite different objectives. An example would be the measurement of net exchange of nitrogen compounds at sites where long-term carbon dioxide fluxes are being monitored, to study links between the nitrogen economy and cycling and net carbon exchange. In the absence of specific objectives which require large area fluxes, and which are focussed on much narrower responses to environmental variables, the scale appropriate for the study should be seriously examined first. Following such analysis it is quite probable that enclosure methods would provide more cost effective and faster answers to the scientific question.
References Arya, S. Pal. (1988) Introduction to micrometeorology. Academic Press. Bartlett, K.B., P.M. Crill, R.L. Sass, R.C. Harriss and N.B. Dise (1992) Methane emissions form tundra environments in the Yukon-Kuskokwim delta. Journal of Geophysical Research 97:1664516660. Chamberlain, A.C. (1975) The movement of particles in plant communities. In: J.L. Monteith (Ed.) Vegetation and the atmosphere, Vol. 1. Academic, New York, pp 155-210. Choularton, T.W., M.W. Gallagher, K.N. Bower, D. Fowler, M.S. Zahniser and A. Kaye (1995) Trace gas flux measurements at the landscape scale using boundary-layer budgets. Philosophical Transactions of the Royal Society of London A 351:357-370. Christensen, S., P. Ambus, J.R.M. Arah, H. Clayton, B. Galle, D.W.T. Griffith, K.J. Hargreaves, L. Klemedtsson, A-M. Lind, M. Maag, A. Scott, U. Skiba, K.A. Smith, M. Welling and F.G. Wienholds (1996) Nitrous oxide emission from an agricultural field. Atmospheric Environment 30:4183-4190. Colbourn, P., J.C. Ryden and G.J. Dollard (1987) Emissions of NO• from urine treated pasture. Environmental Pollution 46:253-261. Denmead, O.T. (1983) Micrometeorological methods for measuring gaseous losses of nitrogen in the field. In: J.R. Freney and J.R. Simpson (Eds) Gaseous loss of nitrogen from plant-soil systems. W. Junk, Amsterdam, pp. 133-157. Denmead, O.T., R. Leuning, D.W.T. Griffith and C.P. Meyer (1999) Some recent developments in trace gas flux measurement. In: A.F. Bouwman (Ed.) Approaches to scaling of trace gas fluxes in ecosystems. Developments in Atmospheric Science 24. Elsevier, Amsterdam, pp. 67-84. Erisman, J.W. and D.D. Baldocchi (1994) Modelling dry deposition of SO2. Tellus 46B: 159-171. Erisman, J.W. and G.P. Wyers (1993) Continuous measurements of surface exchange of SO2 and NH3; implications for their possible interaction in the deposition process. Atmospheric Environment 27A: 1937-1949.
Experimental designs appropriateforflux determination in terrestrialand aquatic ecosystems
119
Erisman, J.W., M.G. Mennen, D. Fowler, C.R. Flechard, G. Spindler, A. Gruner, J.H. Duyzer, W. Ruijgrok and G.P. Wyers (1998) Towards development of a deposition monitoring network for air pollution of Europe (European Commission DGXI 'LIFE' project). Report 722108015, National Institute of Public Health and the Environment, Bilthoven, the Netherlands. Flechard, C.R. and D. Fowler (1998) Atmospheric ammonia at a moorland site. II: Long-term surface/atmosphere micrometeorological flux measurements. Quarterly Journal of the Royal Meteorological Society 124:733-757. Fowler, D., C. Flechard, U.M. Skiba and J.N. Cape (1998) The atmospheric budget of oxidised nitrogen and it's role in ozone formation and deposition. New Phytology 139:11-23. Fowler, D.,K.J. Hargreaves, M.A. Sutton and R.L. Storeton-West (1990) Measurement of the exchange of atmopheric NO, NO2, 03 and NH3 with vegetated surfaces. In: EUROTRAC Annual Report, Part 4, BIATEX IFU, Garmisch-Partenkirchen, pp. 45-59. Fowler, D., K.J. Hargreaves, U. Skiba, R. Milne, M.S. Zahniser, J.B. Moncrieff, I.J. Beverland and M.W. Gallagher (1995) Measurements of CH4 and N20 fluxes at the landscape scale using micrometeorological methods. Philosophical Transactions of the Royal Society of London A 351:339-356. Galbally, I.E. and C. Johansson (1989). A model relating laboratory measurements of rates of nitric oxide production and field measurements of nitric oxide emissions from soils. Journal of Geophysical Research 94: 6473-6480. Galbally, I.E. and CR. Roy (1978) Loss of fixed nitrogen from soils by nitric oxide exhalation. Nature 275:734-735. Galle, B., L. Klemedtsson, and D.W.T. Griffith (1994) Application of a Fourier transform IR system for measurements of N20 fluxes using micrometeorological methods, an ultra large chamber system and conventional field chambers. Journal of Geophysical Research 99:16575-16583. Gash, J.H.C. (1986) Observations of turbulence downwind of a firest-heath interface. Boundary Layer Meteorology 36:227-237. Giinther, A.B., R.K. Monson and R. Fall (1991) Isoprene and monoterpene emission rate variability: observations with Eucalyptus and emission rate algorithm development. Journal of Geophysical Research 96:10799-10808. Hargreaves, K.J., D. Fowler, R.L. Storeton-West and J.H Duyzer, J.H. (1992). The exchange of nitric oxide, nitrogen dioxide and ozone between pasture and the atmosphere. Environmental Pollution 75:53-60. Harriss, R.C., S.C. Wofsy, D.S. Bartlett, M.C. Shipham, D.J. Jacob, J.M. Hoell Jr., R.J. Bendura, J.W. Drewry, R.J. McNeal, R.L. Navarro, R.N. Gidge and V.E. Rabine (1992) The Arctic boundary layer experiment (ABLE 3A) July-August 1988. Journal of Geophysical Research 97:1638316394. Hutchinson, G.L. and E.A. Brams (1992) NO versus N20 emission from an NH4+ amended bermuda grass pasture. Journal of Geophysical Research 97:9889-9896. Jenkins, A. (1997) CLIMEX project: Report on the third year of treatment May 1996-December 1996. NIVA/IH, Institute of Hydrology, Wallingford, U.K. Johansson, C. and L. Granat (1984) Emission of NO from arable land. Tellus 36B:27-37. Kim, D.S., V.P. Aneja and W.P. Robarge (1994 ) Characterization of nitrogen-oxide fluxes from soil of a fallow field m the Central Piedmont of North-Carolina. Atmospheric Environment 28:1129-1137. Lapitan, R.L., R. Wanninkhof and A.R. Mosier (1999) Methods for stable gas flux determination in aquatic and terrestrial systems. In: A.F. Bouwman (Ed.) Approaches to scaling of trace gas fluxes in ecosystems. Developments in Atmospheric Science 24. Elsevier, Amsterdam, pp. 26-66. Lenshow, D.H., R. Pearson Jr and B.B. Stankov (1981) Estimating the ozone budget in the boundary layer by use of aircraR measurements of ozone eddy flux and mean concentration. Journal of Geophysical Research 86:7291-7297.
120
D. Fowler
Ludwig, J. and F.X. Meixner (1994) Surface exchange of nitric oxide (NO) over three European ecosystems. In: Proceedings of. 6th European symposium on physico-chemical behaviour of atmospheric pollutants, Varese, Italy. Commission of the European Community, Directorate General for Science, Research and Development, Brussels, Belgium. MacDonald, J. (1998) Methane oxidation in temperate and tropical soils. Ph.D. Thesis, University of Edinburgh. MacDonald, J.A., D. Fowler, K.J. Hargreaves, U.M. Skiba, I.D. Leith and M.B. Murray (1998) Methane emission from peat wetlands: response to temperature, water table and transport. Atmospheric Environment 32:3219-3227. Matthews, E. and I. Fung, I. (1987) Methane emission from natural wetlands: global distribution, area and environmental characteristics of sources. Global Biogeochemical Cycles 1:61-86. Monteith, J.L. (1973)Principles of Environmental Physics. Arnold, London. Munro, D.S. and T.R. Oke (1975) Aerodynamic boundary-layer adjustment over a crop in neutral stability. Boundary-Layer Meteorology 9:53-61. Pad, J.W., E.G. Beauchamp and X. Zhang (1993) Nitrous and nitric oxides emissions during nitrification and denitrification from manure amended soil in the laboratory. Canadian Journal of Soil Science 73:539-553. Pio, C.A., T.V. Nunes and A.R. Valente (1994) Measurement of VOC emissions from Pinus pinaster species by bag enclosure and tracer techniques. In: P.M. Borrell, P. Borrell, T. Cvitas and W. Seiler (Eds.) Proceedings of the fourth symposium of the CEC/EUROTRAC project, SPB Academic Publishing bv. The Hague, the Netherlands, pp. 497-501. Remde, A., J. Ludwig, F.X. Meixner and R. Conrad (1993) A study to explain the emission of nitric oxide from a marsh soil. Journal of Atmospheric Chemistry 17:249-275. Ritter, J.A., J.D.W. Barrick, G. Sachse, G.L. Gregory, M.A. Woemer, C.E. Watson, G.F. Hill and J.E. Collins Jr. (1992) Airborne flux measurement of trace species in an arctic boundary layer. Journal of Geophysical Research 97:16601-16625. Sass, R.L., Fisher, F.M., Harcombe, P.A., Turner, F.T. (1990) Methane production and emission in agricultural wetlands. Global Biogeochemical Cycles 4: 47-68. Scheupp, P.H., M.Y. Leclerc, J.I. MacPherson and R.L. Desjardms (1990) Footprint prediction of scalar fluxes from analytical solutions of the diffusion equation. Boundary-layer Meteorology 50:355-376. Schiitz, H., A. Holzapfel-Pschorn, R. Conrad, H. Rennenberg and W. Seiler (1989) A three years continuous record on the influence of daytime, season and fertilizer treatment on methane emission rates from an Italian rice paddy field. Journal of Geophysical Research 94:16405-16416. Sebacher, D.I., R.C. Harriss, K.B. Bartlett, S.M. Sebacher and S.S. Grice (1986) Atmospheric methane sources: Alaskan tundra bogs, an alpine fen, and a subarctic boreal marsh. Tellus 38B: 1-10. Sheperd, M.F., S. Barzetti and D.R. Hastie (1991) The production of atmospheric NOx and N20 from a fertilized agricultural soil. Atmospheric Emironment 25A: 1961-1969. Skiba, U., D. Fowler and K.A. Smith (1997) Nitric oxide emissions from agricultural soils in temperate and tropical climates: Sources, control and mitigation options. Nutrient Cycling in Agroecosystems 48:139-153. Skiba, U., K.A. Smith and D. Fowler (1993) Nitrification and denitrification as sources of nitric oxide and nitrous oxide in a sandy loam soil. Soil Biology and Biochemistry 25:1527-1536. Skiba U., K.J. Hargreaves, K.A. Smith and D. Fowler (1992) Fluxes of nitric and nitrous oxides from agricultural soils in cool temperate climates. Atmospheric Environment 26A:2477-2488. Slanina, S. (Ed.) (1997) Biosphere-atmosphere exchange of pollutants and trace substances. Transport and chemical transformation of pollutants in the troposphere, Volume 4. Springer Verlag, Berlin, Heidelberg, New York. Slemr, F. and W. Seiler (1984) Field measurement of NO and NO2 emissions from fertilized and unfertilized soils. Journal of Atmospheric Chemistry 2:1-24.
Experimental designs appropriateforflux determination in terrestrial and aquatic ecosystems
121
Slemr, F. and W. Seiler (1991) Field study of environmental variables controlling the NO and NO2 emissions from soil and of the NO and NO- compensation points. Journal of Geophysical Research 96:13017-13031. Smith, K.A., H. Clayton, J.R.M. Arah, S. Christensen, P. Ambus, D. Fowler, K.J. Hargreaves, U. Skiba, G.W. Hams, F.G. Wienhold, L. Klemedtsson and B. Galle (1994) Micrometeorological and chamber methods for measurement of nitrous oxide fluxes between soils and the atmosphere: Overview and conclusions. Journal Geophysical Research 99:16541-16548. Sorteborg A. and O. Hov (1996) Two parametrizations of the dry deposition exchange for SO2 and NH3 in a numerical model. Atmospheric Environment 30:1823-1840. Steinbrecher, R., H. Ziegler, G. Eichstadler, U. Fehsenfeld, R. Gabriel, Ch. Kolb, R. Rabong, R. Schonwitz and W. Schurmann (1997) Monoterpene and isoprene emission in Norway spruce forests. In: S. Slanina (Ed.) Biosphere-atmosphere exchange of pollutants and trace substances. Transport and Chemical Transformation of Pollutants in the Troposphere, Volume 4. Springer-Verlag Berlin, Heidelberg, New York, pp. 352-371. Street, R.A., S.C. Duckham, C. Boissard and C.N. Hewitt (1997) Emissions of VOCs from stressed and instressed vegetation. In: S. Slanina (Ed.) Biosphere-atmosphere exchange of pollutants and trace substances. Transport and chemical transformation of pollutants in the troposphere, Vol. 4, Springer, Berlin, pp. 366- 371 Thorn, A.S. (1975) Momentum mass and heat exchange. In: J. Monteith (Ed.) Vegetation and the Atmosphere, ed. J. Monteith (Ed.) Academic, New York, pp 57-109. Webb, E.K., G.I. Pearman and R. Leuning (1980) Corrections of flux measurements for density effects due to heat and water transfer. Quarterly Journal of the Royal Meteorological Society 106:85-100. Williams, E.J., G.L. Hutchinson and F.C. Fehsenfeld (1992) NOx and N20 emissions from soil. Global Biogeochemical Cycles 6:351-388. Wilson, J.D., V.R. Catchpole, O.T. Denmead and G.W. Thurtell (1983) Verification of a simple micrometeorological method for estimating ammonia losses after fertilizer application. Agricultural Meteorology 29:283-290. Woodward, F.I. and J.E. Sheehy (1983) Principles and measurements in environmental biology. Butterworths, London, pp 129-133. Zahniser, M.S., D.D. Nelson, J.B. McManus and P.L. Kebabian (1995) Measurement of trace gas fluxes using tunable diode laser spectroscopy. Philosophical Transactions of the Royal Society of London A 351:371-382.
This Page Intentionally Left Blank
Chapter 6
T O W A R D THE USE OF R E M O T E SENSING AND O T H E R DATA TO DELINEATE FUNCTIONAL TYPES IN T E R R E S T R I A L AND AQUATIC SYSTEMS
J.E. Estes and T.R. Loveland
This Page Intentionally Left Blank
Approaches to scaling a trace gas fluxes in ecosystems A.F. Bouwman, editor 1999 Elsevier Science B.V.
TOWARD THE USE OF REMOTE SENSING AND OTHER DATA TO DELINEATE FUNCTIONAL TYPES IN TERRESTRIAL AND AQUATIC SYSTEMS
J.E. Estes I and T.R. Loveland 2 1 Department of Geography, University of California, Santa Barbara, California, CA 93106-4060 U.S.A. 2 U.S. Geological Survey, EROS Data Center, Sioux Falls, South Dakota, SD 57198, U.S.A.
I. I n t r o d u c t i o n
Spatially accurate thematic data, such as land cover, are required for studies addressing a broad range of issues. These issues, which span local to global scales, include biodiversity, desertification, deforestation, freshwater supply and quality, demography, and poverty (Htun, 1993). Spatial data are also important for investigations of ecosystems health, air quality, and all the issues being addressed by the international science community engaged in global change research. Availability of required data is, however, an important factor hindering both fundamental and applied research on these and other issues. It is generally true that adequate land cover maps, as well as other types of thematic data, do not exist for many areas of the world. This statement is equally true, depending upon scale, thematic content, and timeliness, for both the developed and developing nations (Estes and Mooneyhan, 1994). For some time, it has been recognized that there is considerable ignorance concerning the global distribution of vegetation types (Townshend et al., 1991). This lack of knowledge of the distribution and dynamics of our global vegetation cover types is an issue of increasing concern given the rate of transformation of the Earth's vegetation cover, and the profound impacts that these changes are having on our global environment and the sustainability of economic development. Changes in land surface cover associated with agriculture, human settlements, and other purposes are among the most pervasive and obvious impacts of human activities on the global environment. Local land cover changes can and do have profound implications for the functioning of local to global scale ecosystems, biogeochemical fluxes, and climate (IGBP, 1990). Despite the widespread recognition that mankind's intimate involvement in these fundamental transformations of the surface of the Earth have profound and lasting impacts, there is no reliable, comprehensive global data base of land cover changes (Townshend et al., 1991). Increasing attention is being given to the issue of establishing comprehensive, coordinated, operational science quality global measurement, mapping, monitoring, and modelling programs. However, to our knowledge, no such programs exist. Today, no civil organization that we are aware of has a global mapping charter; no civil agency globally has the resources, or the backing of its respective government, to aggressively develop a major, high resolution, science based global scale mapping effort. As Estes and Mooneyhan (1994) point out: (i) Large scale, science-based data sets do not exist for most of the Earth at the present time, even in highly developed countries; (ii) Development of such data sets is labor intensive, in terms of both scientific and technical personnel, and is, therefore, labor expensive; (iii) Although such data sets could support a wide variety of useful applications specific to a given locale, no
126
J.E.Estes and T.R. Loveland
single use can generally justify the cost of their development; (iv) In many developing countries, even well understood environmental changes with local causes and effects, that in the aggregate may represent a global cot~cern, often have very low priority with officials compared to such issues as food, health care, and safety of the people. Global change issues and environmental concerns are often treated as rumours from more fortunate neighbours; (v) In a number of countries, the high-resolution data sets needs by the world community are classified and are not permitted to leave the country in any form. In some instances where such data are exchanged with "friendly" nations, restrictive agreements limit access to these data; and (vi) Even in highly developed countries, where scientific understanding is widespread, it is often difficult to generate the political and financial support for the correction of widely recognized environmental problems. While the lack of comprehensive and coordinated programs is a major hindrance to longterm science and sustainable development initiatives, there are three encouraging movements underway that may improve this situation. First, the commitment to space-based remote sensing by many nations is leading to an unprecedented era of new, multi-resolution remotely sensed data sets. Second, there are a few project-level activities underway to develop some of the required spatial thematic data sets. These projects, largely funded by global change research programs, are typically one-time ~fforts. Third, there is now organi,zed dialog among representatives of national mapping agencies concerning the development of global map data. While these movements do not provide immediate solutions, they are all signs we are moving in the right direction. In spite of these positive steps, these facts remain: - We know little about the distribution of global vegetation types; - We have no well coordinated, operational, globally distributed measurements of key parameters associated with the determination of trace gas fluxes between terrestrial and aquatic ecosystems and the atmosphere; - No programs for operationally monitoring changes in patterns of surface cover exists at scales above local; and Without such programs, the modelling of global trace gas fluxes has inherent flaws. -
This paper will primarily focus on the role that remotely sensed data play in improving our ability to document the distribution of global land cover types. Given that the relationships between these cover types and trace gas fluxes can be unambiguously established, then the dynamics of these fluxes can be established. Brief attention will also be given to important aquatic systems, such as wetlands, and the land/water interface. A detailed discussion of functional aquatic types, including oceans, limnological, riverine, and other aquatic systems, is beyond the scope of this paper. The remainder of the paper addresses four points. First, a discussion of the role of land cover data in environmental assessments, and those issues important in the analysis of trace gas fluxes is given. Second, a general background on the interpretation of remotely sensed data is provided, and key issues in the use of such analyses to create map products are addressed. This discussion covers some of the misconceptions within the scientific community concerning the use of remotely sensed data for the creation of surface cover products. Third, the information factors important to consider in the creation of cover type maps are identified. Here we focus on issues such as: strategies for such mapping, legend development, large area mapping from remotely sensing data and accuracy assessment. Finally, we offer conclusion and some recommendations directed specifically at the creation of future thematic products from remotely sensed data for scaling trace gas fluxes between terrestrial and aquatic systems and the atmosphere.
Use of remote sensing data to delineate functional types in terrestrial and aquatic systems
127
2. Land cover data for environmental assessments
Turner and Meyer (1991) remind us of the significant distinction between land use and land cover. Lae~.d cover can be defined as the natural and artificial materials on the land surface (e.g. forests, crops, built-up structures), while land use is the activity for which the land is used (e.g. logging, farming, retailing). Natural scientists and environmental modellers more often require information on land cover to pursue their investigations because of the linkages to atmospheric and biotic systems. Social scientists, on the other hand, tend to be far more interested in land use, or the purpose to which land is manipulated. This is because land use links back to issues of human behaviour and social process. The link between cover and use, however, is important. Land use patterns, driven by a variety of social processes, result in land cover changes that affect biodiversity, water and radiation budgets, trace gas emissions, and other factors that, cumulatively, alter global climate and ultimately the biosphere (Riebsame et al., 1994). With remotely sensed data, land cover can be mapped, and arguably, used as a surrogate for land use. Environmental assessments require data that provide a means to determine the areal extent of various resources. These assessments also require site-specific descriptions of the types of land cover and their characteristics. It is important to note that most applications of land cover data need more than simple type names for a given location. They often require knowledge of attributes describing various properties of land cover (e.g. canopy density, leaf area, biomass, etc.). As a result, the collection of environmental baseline data involves intensive and costly effort to gather data prior to the initiation of a project (AAAS, 1983). Cost serves as an important incentive for encouraging the use of existing land cover and other spatial data sets. Focusing on the needs of the global change research community, IGBP (1992) states that there are two important basic technical requirements affecting the validity of global environmental forecasts and assessments. First, because the Earth system is complicated by multiple interactions and feedbacks, numerical models accounting for these interactions and feedbacks are needed. Second, a large amount of geographically-referenced, validated data will be needed to adequately parameterize these models. Geographically-referenced data needed include: (i) data for documenting and monitoring global change, such as land and sea temperature, and atmospheric concentrations of carbon dioxide and other trace gases; and, (ii) data which characterize important forcing functions. Land cover data relate to both items. The IGBP (1992) report concludes with the observation that land cover data are required by most IGBP core projects and these data are a critical, but missing, element in models of global ecosystems and hydrology. However, the report goes on to state that land cover data must be able to portray inter- and intra-annual vegetation dynamics, and that the question of appropriate spatial resolution is significant. The resolution must be such that it not only satisfies information requirements through spatial discrimination of land features and represents a level of spatial detail that permits flexible, sub-grid cell paramaterization, and coupled systems modelling, but also has a total data volume that can be accommodated in computationally complex models. Data sets with too fine of spatial detail would have costly data storage requirements, and would increase the time needed for modelling exercises. Thus, a balance between spatial detail and associated data volume efficiency must be found. An earlier IGBP report (1990) provides an interesting perspective on how to approach the challenge of multi-purpose land cover data. This report on core project science issues provides a perspective on the requirements for land cover information. An important point made is that requirements vary considerably for the different projects. The IGBP report speculated that some projects will require only broadly defined land cover types, while others will need more detailed information. The report proposed the concept of a comprehensive global land cover
128
J.E.Estes and T.R. Loveland
data base, which should contain the following sets of data planes: (i) basic calibrated remote sensing data; (ii) Preprocessed remote sensing data set; (iii) key ancillary data planes; (iv) attribute data sets on baseline and change; and, (v) parametric data sets on baseline and change (IGBP, 1990). Geographically referenced actual land cover data bases with a seasonality component are important for environmental modelling (IGBP, 1993). Land cover data are used to partition the landscape into geographic units corresponding to a broad suite of environmental parameters, such as surface roughness, albedo, latent and sensible heat flux, and associated biogeochemical processes and cycles. Changes in the distribution of land cover alter the regional, and possibly global, balance of these fluxes. Changes in land cover distribution include both the often cited conversions of one land cover type to another (e.g. forest to cropland, grassland to urban), but also the seasonal changes in land cover parameters. The seasonal development of foliage, for example, over a period of less than a few weeks can affect local weather (Bonan, 1995). Hall et al. (1995) provide a good review of the role of land cover characteristics data in land process models. For example, in land process models vegetation community composition information are used to partition the global landscape into functionally different strata. The specific land cover types represented in a data base correlate to the biological, thermodynamic, or chemical pathways corresponding to different vegetation associations. The seasonal dynamics of land cover are important because of their influence on patterns of latent heat flux throughout the year, as represented by changes in turbulent exchange parameters such as surface roughness, and radiation exchange variables such as albedo. Hall et al. (1995) suggest that land cover data are needed in which the categories are defined by modelling functional characteristics relating directly to properties such as energy, water, and nutrient cycling rather than by purely species characteristics. They even suggest a set of functional classes of land cover for use in International Satellite Land Surface Climatology Project (ISLSCP) science initiatives. The categories that Hall et al. (1995) recommend include coniferous forest, deciduous forest, broadleaf evergreen forest, tundra, woodland, savanna, grassland, desert, shrubs, cultivated, wetlands, freshwater areas, ice cover, and built areas. However, Hall et al. (1995) conclude that a single categorization of land cover types is unlikely to meet all modelling requirements. In an overview of the role of land cover in altering climate, Nemani and Running (1996) reiterate the range of roles land cover plays in determining available energy, and energy partitioning between sensible and latent heat fluxes through changes in albedo, roughness, and stomatal control of water losses. Land cover affects albedo, sensible heat, and latent heat through changes in aerodynamic resistance and stomatal resistance. Increase in leaf area contributes to lower canopy resistance and higher latent heat. Nemani and Running (1996) stress that the magnitude of energy and its partitioning into latent and sensible heat fluxes has important implications for ecology, hydrology, and atmospheric dynamics. A case for the importance of land cover parameters is also made by Copeland et al. (1996) through their statement that the problem with land cover data sets is not how representative the vegetation category is, but how valid the corresponding model parameters are for that geographical location. Copeland et al. (1996) point out that density, for example, is not often taken into account. Using a United States example, the land cover class "coniferous forests" include both the dense temperate rainforests of the Pacific Northwest, and the semi-arid open pinyon-juniper woodlands of the southwest. This problem of adequate representation points out the need for high quality spatial databases of land cover and other surface characteristics. From the above, we can conclude that land cover characteristics must be flexible, comprehensive, representative, and dynamic. In addition, they must capture environmental processes
| 29
Use of remote sensing data to delineate functional types in terrestrial and aquatic systems
Table 1. Land-cover characteristics input requirements and spatial scale for selected modelling applications and models. Model a
Classification Scheme b
Spatial scale
Associated attributes c
General Circulation Models NASA/GSFC
SiB
4 x 5 degrees
Simplified SiB
BATS
4.5 x 7.8 deg. 1.8 x 2.8 deg. 2 x 4 degrees
BATS set and NDVI derivatives
LEAF BATS
Nested Grids of 1, 10, 40 km Nested Grids of 4, 12, 36 km
LEAF Set and NDVI derivatives BATS Set and NDVI derivatives
Basic Classes
2.5, 5, 10km
Model specific
Anderson Level II
country level or 1 km
Model specific
Basic Biomes Anderson Level II Key species (oak, Hickory, etc.)
1-50 km 1-50 km 20 km
RHESSys Set, NDVI derivatives NDVI derivatives NDVI derivatives
Univ. of MarylandCOLA NCAR-CCM
SiB set and NDVI derivatives SSiB set and NDVI derivatives
Mesoscale Meteorological Models CSU-RAMS PSU-NCAR MM4 Hydrologic Models Watershed Precip./Runoff Agricultural Chem., Runoff Ecosystem Models RHESSys CENTURY Biogenic emissions
Source: Loveland et al. (1995). " COLA, Center for Ocean-Land-Atmosphere; CSU-RAMS, Colorado State University-Regional Atmospheric Modelling System; GSFC, Goddard Space Flight Center; NCAR-CCM, National Center for Atmospheric Research, Climate Community Model; PSU/NCAR-MM4, Penn State University-Nati0nal Center for Atmospheric Research; CENTURY (Parton et al., 1988). b BATS, Biosphere-Atmosphere-Transfer-Scheme (Dickinson et al., 1986); LEAF, Land-Ecosystem-AtmosphereFeedback; SiB, Simple Biosphere model (Sellers et al., 1986). c NDVI, Normalized Difference Vegetation Index; RHESSys, Regional Hydrological Ecosystem Simulation System.
at various spatial scales. The issues of mapping unit and scale present additional significant challenges. In her review of the role of spatial data in environmental simulation models, Kemp (1992) identifies several key data challenges facing modelling groups. While Kemp (1992) does not focus on specific data sets, she points out that a formidable task involves the range of spatial and temporal scales at which different processes are depicted in models. Kemp (1992) suggests that modelling has been hampered by the fact that critical environmental processes operate on many different time and space scales, and there may be scale thresholds at which critical processes change. This is complicated by the fact that discrete scales of measurement are typical with land cover, soils, and geological data, but that mathematical models typically use continuous data. As a result, thematic data, like the categories used with traditional land cover maps, must be converted to parameter values (e.g. stomatal resistance, surface roughness, leaf area index) which can be used in numerical calculations. This means that researchers must use measurements from limited field or laboratory studies and judgment to select the values that are associated with particular land cover classes. It must be remembered here that modelling disciplines (e.g. climate and land-atmospheric interactions, biogeochemistry, hydrology, and ecology) each have specific emphases in their models. As a result, the role of land cover varies between disciplines, and in fact, varies within specific disciplines (Steyaert et al., 1994). Table 1 presents a summary of the spatial and thematic land cover requirements of several types of environmental models. Below we examine land cover in relation to two types of models that consider trace gases: landatmosphere interaction models and ecosystem processes models.
130
J.E.Estes and T.R. Loveland
2.1. Land-atmosphere interactions modelling Land cover data are becoming increasingly important in climate models. Both general circulation models (GCM) used to estimate global climates under specific conditions, and mesoscale meteorological models operating at regional levels, make use of land cover data. Land cover provides a means to parameterize land-atmosphere interactions. The most detailed parameterizations of biophysical land-atmosphere interactions are the land surface schemes that are coupled with climate models. Land surface fluxes of energy, moisture, and momentum are required as boundary conditions in climate models. Land surface process models were developed to account for the effects of vegetation and soil on these fluxes. Analyses accomplished using these models show that the biophysical properties of vegetation and the physical properties of soils are important determinates of regional and global climates (Bonan, 1995). Several land surface parameterization models have been developed for the purpose of estimating suites of parameters that affect the interaction between land cover and the atmosphere. Aquatic types are treated as general cases, with only simple distinctions, fresh water and ocean water, handled. The Simple Biosphere Model (SiB) by Sellers et al. (1986) and the Biosphere Atmosphere Transfer Scheme (BATS) by Dickinson et al. (1986) were among the earliest efforts to provide realistic representation of landscape processes in atmospheric models. SiB, and in its updated form, SiB2, provides a means to model the biochemical mechanisms governing photosynthesis and how these mechanisms are tied to the plant stomatal functions that can be used to calculate photosynthesis and transpiration over large areas (Sellers et al., 1996). SiB2 relies on estimates of the fraction of photosynthetically active radiation, calculated directly from satellite vegetation indices. SiB2 requires land cover data for the assignment of time-invariant parameters: (i) morphological parameters (canopy height, leaf dimensions, leaf-angle distribution, root depths); (ii) optical properties (phytoelement reflectance and transmittance); and iii) physiological properties. These are treated as constants for each vegetation type. The time-varying vegetation parameters are generated from satellite data and include fraction of photosynthetically active radiation (FPAR), total leaf area index (LAI), canopy greenness fraction, and aerodynamic parameters. A land cover scheme is used to account for the dependency of the relationships between vegetation index, land surface parameters, and land cover type. Sellers et al. (1990) observes that the development, initialization, and validation of coupled land-atmosphere models must include interactions between the land surface and the atmosphere on daily, seasonal, and interannual time scales. Physical processes that must be modelled include the surface radiation budget, which determines the amount of radiation absorbed by the land surface and the surface heat budget describing the proportions of absorbed radiation used to heat the ground and generate the fluxes of sensible and latent heat into the atmosphere. Sellers (1993) later adds that the fundamental vegetation types needed to model the fluxes between the two systems are conifer forest, deciduous forest, shrubland, grassland, agriculture, and wetlands. If possible, Sellers suggests that the definitions be even more functional, for example, woody vs. non-woody, deciduous versus evergreen, needleleaf versus broadleaf, and C3 versus C4 grasses. Dickinson (1995) provides a useful overview of land cover scale issues in climate modelling. Traditionally, he states, single representations of land cover were used to describe large grid cells. The rationale for sticking with a single cover GCM mesh point were: (i) the percentages of individual land cover types at a single mesh point were very inaccurate because of the subjectivity used to create them; (ii) an adequate description of cover may be more important at the regional and global scales than at the local; and (iii) the variety of cover types
Use of remote sensing data to delineate functional types in terrestrial and aquatic systems
131
inferred for different mesh points in heterogeneous regions may act to represent heterogeneity on regional and global scales, or at least will provide a basis for study of climate affects of heterogeneous land surfaces. Dickinson (1995) explains that land cover data are used to estimate the various parameters that actually characterize the role of vegetation in the model. These parameters are given as data arrays, that is, as a number or numbers for each type of land cover. One of the serious criticisms of present treatments of land processes in climate models is their failure to include many of the most essential aspects of sub-grid-scale heterogeneity. Dickinson argues that at least three scales are needed to model land cover heterogeneity effects. On a very fine scale, canopy air interacts between different surfaces, so that surface roughness is not necessarily an average inferred from any single element. This type of heterogeneity should be provided as part of the overall land cover description. On a somewhat coarser scale, surfaces independently interact with an overlying homogeneous atmosphere. This is the scale of typical agricultural fields or small stands of forest. Finally, at scales of at least a few tens of kilometers and certainly at GCM resolution, individual surfaces are overlaid by a physical boundary layer that is essentially unaffected by individual neighbouring surfaces.
2.2. Ecosystems processes modelling Recently, researchers have begun to explore the important relationship between atmospheric processes modelling and ecosystem processes and functions, including biogeochemistry and net primary productivity. Matson and Ojima (1990) point out that land cover will become increasingly important in atmospheric chemistry studies because of a need for data that can be used for the synthesis of global N20, CH4 and CO2 fluxes. These authors suggest that there is a particularly strong need for land cover characteristics for high latitude ecosystems (Matson and Ojima, 1990). The classifications needed should represent a combination of vegetation, soils, hydrology, and biogeochemistry. The resolution needs to be broad enough to allow global and regional extrapolations, but fine enough to allow recognition of small-scale variations in vegetation and environmental factors. As with land-atmosphere interaction models, aquatic systems are treated generically in ecosystem processes modelling. Wetlands are generally not differentiated into functional types (e.g. wooded wetlands, herbaceous wetlands, fens, etc.), and different classes of water bodies are not distinguished. The IGBP Global Change in Terrestrial Ecosystems (GCTE) core project is focusing on determining the effects of changes in land cover on the functioning of ecosystems and biogeochemical cycles, along the entire spectrum from natural to intensively managed lands. The goal of GCTE researchers is to incorporate projections of land cover change into predictive models of land cover change (Steffen et al., 1992). GCTE scientists have called for the development of a new land cover classification scheme that is appropriate for ecological and biogeochemical modelling. In particular, these researchers call for a classification scheme that provides the ability to deal with the variability of canopy structure, including the distribution of height and vegetative cover among the components of a multi-layered canopy, and also permits accounting for differing nutrient regimes. Models of biogeochemistry processes and cycles are being developed and used to assess the influence of biogeochemical cycles on the physical climate system. Recognition of the need for such models has facilitated significant advances in the understanding of the global cycles responsible for the compositions of the atmosphere, oceans, and sediments on the surface of the planet. The IGBP Global Analysis and Integrated Modelling Task Force determined that it is essential to quantify the characteristic dynamics of these cycles and their
132
J.E.Estes and T.R. Loveland
controlling factors, in order to improve the understanding of global processes. This means that many types of data are needed for model input and validation of contemporary era models, including land cover statistics and land cover and vegetation maps derived from remotely sensed data (IGBP, 1994). In biogeochemical modelling, it is important to include land cover attributes describing community composition or vegetation types so that physiological differences in assimilation rates, carbon allocation, and nutrient use efficiency which influence CO2 uptake during photosynthesis can be modelled (Bonan, 1995). Models that include detailed canopy physiology, energy exchange, and microbial processes have been developed (e.g. the CENTURY model of Parton et al., 1988; BIOME-BGC developed by Hunt et al., 1996). These models integrate daily photosynthesis and leaf, stem, and root respiration to annual totals, taking into account irradiance, temperature, humidity, CO2, nitrogen, and soil moisture limits to photosynthesis and temperature effects on respiration. Biogeochemical models are used to address a variety of ecological processes. Schimel et al. (1991) review the relationship between the biogeochemical processes of ecosystems and the atmosphere. They state that the success at modelling the earth system will probably require separating two problems, i.e. simulation of the contemporary world, and prediction of its future course. The first will require the use of realistic and current representations of the landscape. This will require better data, and also the coupling of models of ecosystems to atmospheric and hydrological models. Increasing the linkage between components may increase the realism of simulations. Schimel et al. (1991) suggest that this may lead to increased predictability if the subsystems act to constrain each other's behaviour. While cautious in their perspective of the role of remote sensing, Schimel et al. (1991) recognized that successful use of remote sensing in extrapolation of models of biogeochemistry will require the rethinking of traditional modelling approaches. The key, they state, will be to use plant physiology as the interface between ecosystem properties and remote measurements. Kittel et al. (1996) provide another perspective regarding the challenges associated with ecological process models. They suggest that the accurate representation of the spatial distribution of driving variables and boundary conditions is needed for simulations of regional to global ecological dynamics. These authors caution that development of such representations can be difficult because of the paucity of extensive regional data sets and the coarse resolution needed to cover large domains. A particular challenge is that boundary conditions, such as land cover and soil type, can exert non-linear controls over ecological processes. When land cover and soils are heterogeneous with respect to the model simulation grid, grid averages may not adequately represent existing conditions. As a result, it is important to have data that have sufficient spatial detail for the statistical treatment of spatial heterogeneity and spatial coherence among variables that have nonlinear relationships to ecosystem processes. The ability to use implied sub-grid information (i.e., estimates of the percentages of different land cover types within a grid cell) instead of using finer grid cells is necessary so that the already high computation requirements of large simulations are kept at a manageable level. This can only be accomplished, they state, with: (i) the development of physically consistent model input data sets; (ii) smooth transfer of data between GIS and applications; and (iii) analysis of model results in both geographical and temporal contexts (Kittel et al., 1996). Speaking on behalf of the requirements of the ecological community, Ustin et al. (1991) stressed the need to develop land data sets of the highest possible resolution. The authors argue that pixel sizes can be enlarged by averaging but cannot shrink to accommodate ecosystem models, thus limiting the types of ecological models that can be applied.
Use of remote sensing data to delineate functional types in terrestrial and aquatic systems
133
One of the most modelled ecological parameters, global net primary productivity (NPP) assessment, is complicated because the models used differ markedly in approach and complexity and often yield incomparable estimates (Field et al., 1995). NPP is sensitive to many controls, including climate, soils, plant characteristics, disturbance regimes, and a number of other natural and anthropogenic factors. Field et al. (1995) suggest that the databases for evaluating NPP models are not adequate to support a comprehensive assessment. They indicate that vegetation characteristics have a number of effects on NPP. Maximum growth potential varies widely among plant species, and plants of infertile habitats are genetically incapable of realizing high rates of biomass accumulation. Estimation of NPP is complicated by the fact that the plant characteristics that affect this parameter vary both within and among species. They conclude that global NPP models are limited by the paucity of global data. Improvements in global data are a critical foundation for accurate NPP estimates for a wide array of global change scenarios.
3. Remote sensing and land cover mapping Remotely sensed data have been used land cover mapping for many years. The first intellectual framework for using remotely sensed data for mapping land cover was established by Anderson et al. (1976). This framework established the links between the scale or resolution of remotely sensed data, and the level of thematic land cover detail that can be consistently mapped. While still relevant, the widespread use of computer assisted image classification and the use of a wider variety of types of remotely sensed data have modified some of Anderson's basic tenants. For example, the difficulty of consistently mapping land cover at a given hierarchical level of a land cover legend when using computer assisted classification techniques, was not anticipated. In addition, the time for space substitution possible when satellite data with high temporal frequency are used can sometimes result in the interpretation of additional thematic detail not possible with single date imagery. With computer assisted classification, image segmentation algorithms are used to divide remotely sensed data into spectrally homogenous classes. These spectral classes must be related to land cover either using supervised (spectral data/land cover relationships are determined prior to image segmentation) or unsupervised (spectral data/land cover relationships are determined following image segmentation) techniques. An often-encountered misconception is that land cover can be interpreted solely from remotely sensed data. The relationship between spectral data and land cover is frequently ambiguous. This is especially true in regional to global scale surface cover mapping (Brown et al., 1993). When image analysts are creating land cover products, they typically rely on a variety of materials in the process. In addition to their own experiences, the analysts may rely on existing maps (e.g. soils, topographic, ecoregions, land cover), texts, historical photography, among others. Indeed analyst interpreters will filter the material found in these reference materials through their own experiences as they seek to delineate and label a given cover type. The fact that these ancillary data sets are often unavailable complicates the interpretation process. The interpretation and labeling process is important and relates to a second major misconception remote sensing analysts often encounter, i.e., the myth of automated image analysis. Although the term "automated interpretation" has been widely used in the remote sensing literature, interpretation is never automatic. Whether a researcher interested in the mapping or separation of land cover types uses either of the two most widely used forms of machine assisted analysis, supervised or unsupervised classification, the interpreter always labels the classes needed to drive the process.
134
J.E.Estes and T.R. Loveland
Another misconception is that the output images from analyses are in and of themselves "maps." The term "map" as used in this paper refers both to digital and analog (paper) products. Map does not refer solely to standard base cartographic products produced by national mapping organizations across the globe. While these maps typically include thematic layers (e.g. hypsography, hydrology, transportation, political subdivisions, and cover); with the except~ion of hypsography, they rarely cover these themes in the detail needed by the scientific community. Instead, these base cartographic themes are useful as frames of reference useful for interpreting thematic data. There are a wide variety of data types or themes that can be derived from remotely sensed data. Conversely given specific types of remotely sensed data, there are also themes that cannot be adequately derived. That is, given the spatial, spectral and temporal resolution of the remotely sensed data to be interpreted/analyzed in a given surface cover mapping effort, there are classes of cover that can and cannot be accurately delineated (for example, accuracy could be defined as greater than 85% correct at the 95% confidence interval). This makes it imperative that as a given surface cover classification scheme is developed for a given science purpose (e.g. scaling of trace gas fluxes), both remote sensing and biogeochemistry scientists should work closely together to ensure that the information required can be derived from the remotely sensed data. The global remote sensing land cover legend developed by Running et al. (1994) is an example of this kind of cooperation. This system was designed to be very simple so that the classes could be mapped consistently and accurately. With respect to the map products derived from remote sensing, scientists need to be aware of a number of issues. Since the surface of a sphere cannot be made to lie flat, forms of projections must be used. Essentially, this means that today, planes, cones, and cylinders are employed to transform data from spherical coordinates to planar, or "flat paper" coordinates. This is accomplished employing geometric surfaces (planes, cones, cylinders) and mathematical calculations. The issues of importance here are that in using this process distortions are unavoidable, and these distortions effect the ability to accurately represent areas and directions. With particular respect to the latter some projections maintain the integrity of areas at the expense of shape while others keep direction true while altering areas. Global data set developers have focused on equal area projections (i.e., interrupted Goode's homolosine, Lambert azimithal equal area, Albers equal area) because they minimize area distortions (Steinwand et al., 1995). It is important that researchers seeking to scale trace gas fluxes understand the attributes of the projections in which their data products are represented. It would, for example, not be appropriate to calculate the global area of a given cover class type that was represented on a Mercator projection. While this projection shows true compass direction, it is not an equal area projection. Because of the way they are constructed, all maps including those derived from remotely sensed data contain errors. Estes and Mooneyhan (1994) use the term "science-quality" with respect to maps where, in so far as practical and possible, the errors inherent in their production has been thoroughly documented. Again, science-quality maps exist for only a small fraction of the surface of the global and only for a very few themes. As far as we know today, only hypsography falls into this category at present. However, in the near future, a science quality global land cover data set should become available (Loveland and Belward, 1997).
4. Land cover mapping The previous two sections summarize the needs for land data in a variety of models and the role of remote sensing in land cover mapping. It is important to note that the required land
Use of remote sensing data to delineate functional types in terrestrial and aquatic systems
13 5
cover inputs fall into two groups: land cover categories, and biophysical parameters. The land surface characterization research community are aggressively pursing the development of methods for direct parameterization of key biophysical variables, including FPAR, LAI, albedo, and evapotranspiration. Issues of satellite data calibration, atmospheric corrections, and measurement physics are being addressed. Until direct parameterization can be applied operationally and with adequate precision and accuracy, land cover maps will be essential in environmental models. In this section, we will discuss issues important in the creation of land cover maps from remotely sensed data. We will focus primarily on mapping cover types that can aid in the analysis of trace gas fluxes. For this particular application, land cover provides the context for parameterizing trace gas fluxes, establishes their geographic locations, and contributes the area term for models. These three land cover elements (type, location, area), when coupled with time, give us a picture of the patterns, rates, magnitude, and dynamics of trace gas fluxes between terrestrial and aquatic systems and the air. There are many ways to characterize the surface of the earth. Several fundamental principles that can be stated with respect to land cover type mapping from remotely sensed data. First, at this time, for general purpose land cover mapping human image interpretation is superior to computer assisted classification of remotely sensed data. If the goal of an analysis is the delineation of a single class of cover with well defined and unambiguous spectral characteristics (e.g. agriculture or water), and we can choose the date of acquisition, the sensor, the spatial and spectral resolution, we can match the accuracy of human analysis by a well trained interpreter with machine assisted classification. If however, it is necessary to map many types of land cover (e.g. twenty to thirty classes) in an hierarchical land cover classification scheme, the trained analyst will out perform typical computer assisted analysis under most circumstances. However, the computer-assisted approach will result in more spatial precision than the manual interpretation. For maximum benefit, the two approaches should be integrated. Second, in general, the better the spatial resolution of the remotely sensed data employed, the more land cover types that can be correctly identified. The issue here again is a practical one: How much time does it take to do a detailed analysis of a very large area? There are the practical issues of cost versus information. That is, if we are only interested in a limited number (say 100 globally distributed) small 1 km by 1 km plots or ten 10 km by 10 km plots, then 1:500 scale data could be acquired and used to very accurately characterize a sample of cover types at a relatively reasonable cost. If on the other hand we are interested in global characterizations then from a practical standpoint there is not enough money, machine time and analysts available today to characterize the Earth's surface at such a scale. The best we are able to do today at a global scale is a resolution of 1.1 kilometer; and while this may change in the future, the question will be what scale/resolutions and number and types of surface cover categories are necessary to adequately model trace gas fluxes. Regarding spatial resolution and thematic detail, it is possible to compensate for coarse resolution satellite inputs if highfrequency multi-temporal satellite data are used. The addition of time permits characterization of land cover based on additional attributes, including seasonality and relative productivity (Loveland et al., 1991).
4.1. Strategies There are many issues that must be addressed when developing land cover map products. Here we focus on two that are key in the process. First, the land cover legend that provides the
136
J.EEstes and T.R. Loveland
categories or attribute definitions that will be assigned to the map units must be defined. Second, the mapping methodology must be determined. Both issues are complicated by the format of the source materials, and the ocale, intended uses, and required precision and accuracy of the final products required by the ends users. Development of mapping units involves both compilation and classification procedures. Past global land cover map units have relied heavily on compilation techniques. Large area mapping, whether done on a national, continental, or global scale, historically used compilation approaches which, because of the inconsistency of source materials, often prohibited the use of quantitative classification techniques. Compilation methods involve the synthesis of existing maps, statistics, and other data map units, often using manual mapping methods. Three types of compilation strategies are common. First, are the compilation of existing vegetation maps of vegetation, land cover, and land use into a consistent land cover legend. The Matthews (1983), Olson and Watts (1982), and Wilson and Henderson-Sellers (1985) maps are examples. Second, integrated terrain units are formed by using common occurrences of predefined landscape components, such as flora, fauna, soil, landforms, climate, and water, to form map units and descriptions (AAA,1983). The Geographic Belts and Zonal Types of Landscapes of the World map by Milanova and Kushlin (1991) is an example. Third, is the use of geographic regionalization methods such as those used by Bailey (1987) and Omernik (1987). Regionalizations capture the heterogeneity of the landscape into map units often called ecoregions. Ecoregions represent the repeating patterns or enduring features of an area, and may serve as models of the inherent environmental complexity of an area. Compilation methods have been the most common approach for large area mapping. An argument against the use of compilation strategies, however, is that they are difficult to apply in a consistent, objective, repeatable manner. Starting with existing land cover and vegetation maps, Matthews (1983) and Olson and Watts (1982) determined dominant vegetation for specific grid cells, and organized their descriptions into classification legends. The map units, because of the coarse scale of their databases, were inherently heterogeneous, but still relied on single vegetation type definitions. The Wilson and Henderson-Sellers (1985) database, on the other hand, provided secondary cover type codes to indicate other important landscape elements. Regionalization strategies are often based on straightforward compilation methods, but have the advantage that they are designed to recognize landscape variability. Pep!ies and Honea (1992) suggest that regionalization strategies are quite appropriate for global databases. Peplies and Honea state that regions, from a conceptual point of view, are basically models. Maps based on regions provide a means to simplify the many characteristics and elements that compose areas, and extracts those which are significant. Therefore, one reason for using a regional approach is to be more efficient in handling spatial data. The conterminous United States ecoregions framework developed by Omernik (1987) used classical geographic regionalization methods. This strategy was developed to provide a spatial framework for assessing regional water quality problems. Omernik's approach is based on the hypothesis that ecosystems display regional patterns that are reflected in spatially variable combinations of causal factors including mineral availability (soils and geology), vegetation, and physiography. In defining ecoregions, Omernik examined factors that either cause regional variations in ecosystems or integrate causal factors. Omernik argues that because the interrelationships among natural and anthropogenic factors vary spatially and temporally in a complex fashion, that mathematical and other models developed to predict land use/resource quality relationships are of questionable value. Instead, he relies on traditional geographic research methods. Bailey (1987) has also developed ecoregions maps for the United States, as well as the
Use of remote sensing data to delineate functional types in terrestrial and aquatic systems
137
entire Earth. The methods used by Bailey parallel those of Omernik. However, rather than define regions using a d hoc causal factors, Bailey follows a hierarchical regionalization framework in which each level is based on predetermined decision criteria. Meeus (1995) also advocates a regionalization strategy for mapping landscape. He suggests that it is not possible to identify a single approach for analyzing landscapes which suit all purposes. As a result, he suggests the use of a multidimensional landscape typology that include the interaction between human activities and natural systems. Meeus (1995) has applied his typology to map the regions of Europe. His approach is similar to ecoregionalization strategies and uses geological, ecological, agricultural, silvacultural, and visual criteria to determine the "major landscapes" in Europe at the highest level of abstraction. He described each mapped landscape in two different ways. First, he documented specific features of the individual elements, and second, the relations between them. This, he states, defines why one "regional" landscape differs from another. Such regionally specific landscapes are non-repeatable in time and space, but imply the essential characteristics that comprise spatial diversity. The second approach to mapping is the use of classification techniques. The classification of remotely sensed data, done either using manual image interpretation methods, or statistical classification techniques are commonly used for large area mapping. Also common is the classification of climate variables into vegetation groups. The latter is represented by the work of Box (1981) who uses ecophysiognomic life forms to represent the potential natural vegetation of the globe. Each form is defined primarily as a particular combination of the following characters: (i) general life-form class (e.g. tree, shrub, graminoid); (ii) general plant size; (iii) leaf type (e.g. broad, narrow/needle, absent); (iv) general leaf size; (v) structure of the photosynthetic surface (e.g. sclerophyllous); and (vi) seasonal photosynthetic habit (e.g. summergreen). Most remote sensing mapping practices follow carefully defined land cover legends and mapping methods. Recently, there has been some movement toward treating land cover mapping from remotely sensed data with strategies akin to the geographic regionalization approach. For example, Peplies and Honea (1992) suggest that one of the methodologies for regionalization that is appropriate for global change studies is the photomorphic unit concept. This notion is based on the idea that meaningful regions can be identified from remotely sensed imagery, in terms of similar patterns of tones, textures, and structures. They suggest that since remote sensing is commonly based on solar energy flow profiles, and since fundamental climatic concepts are the heat energy and water budgets, a "natural" relationship exists between global change and photomorphic regions. Finally, they state that regional models provide the tools for efficiently analyzing the composition of area in an inductive manner, and are useful for simplifying reality into manageable spatial units that provide some understanding of how various elements and processes operate within an area (Peplies and Honea, 1992). In an interesting twist, Aspinall and Veitch (1993) argue that the product of a remote sensing classification should be an "information surface" representing the probability of occurrence of relevant land cover properties. This approach forms a framework for combining relative values of being right or wrong (subjective probabilities) with the probabilities of being right or wrong (conditional probabilities). For trace gas studies, landscapes would be categorized according to the presence of variables affecting fluxes. Right or wrong equate to decisions over presence or absence and are expressed as probabilities. They claim their method reduces the dependence on establishing a close relationship between image classes (spectral ha bitat types) and land cover characteristics. Loveland et al. (1995) used a variation of the regionalization approach using 1-km Advanced Very High resolution Radiometer (AVHRR) data covering the conterminous U.S. They
138
J . E . E s t e s a n d T.R. L o v e l a n d
Table 2. Examples of the different thematic detail associated with five land cover legends.
IGBP
BATS
SiB2
USGS
Running
Evergreen needleleaf forest Evergreen broadleaf forest Deciduous needleleaf forest Deciduous broadleaf forests Mixed forests
Crops, mixed farming
Broadleaf-evergreen trees Broadleaf deciduous trees Broadleaf and needle:"af trees Needleleaf-evergreen trees Needleleaf-deciduous trees Short vegetation/C4grassland Shrubs with bare soil Dwarf trees and shrubs Agriculture or C3 grassland Water, wetland, ice/snow
Urban and built-up
Evergreen needleleaf vegetation Evergreen broadleaf vegetation Deciduous needleleaf vegetation Deciduous broadleaf vegetation Annual broadleaf vegetation Annual grass vegetatmn Non-vegetated land
Short grass
Open shrublands
Evergreen needleleaf trees Deciduous needleleaf trees Deciduous broadleaf trees Evergreen broadleaf forest Tall grass
Woody savannas
Desert
Savannas
Tundra
Grasslands
Irrigated crops
Permanent wetlands Croplands Urban and built-up Cropland/natural vegetation mosaic Snow and ice Barren or sparsely vegetated Water bodies
Semi desert Ice caps and glaciers Bogs and marshes Water bodies
Closed shrublands
Ocean Evergreen shrubs Deciduous shrubs Mixed shrubs and trees Mixed interrupted woodlands
Cropland and pasture Grassland Shrublands Mixed shrubland/grassland Deciduous broadleaf forest Deciduous needleleaf forest Evergreen broadleaf forest Evergreen needleleaf forest Mixed forest
Water bodies
Water bodies Herbaceous wetlands Wooded wetlands Barren or sparsely vegetated Herbaceous tundra Wooded tundra Mixed tundra Bare ground tundra Snow and ice
IGBP, Belward (1996); BATS, Dickinson et al. (1986): SiB2. Sellers et al. (1996); USGS, Anderson et al. (1976); Running, Running et al., (1994).
defined map units based on AVHRR-Normalized Difference Vegetation Index (NDVI) that they referred to as seasonal land cover regions. Seasonal land cover regions were defined as mosaics of land cover, seasonal properties, and relative productivity. Through documentation of the suite of attributes, they argued that the resulting data base was of sufficient flexibility to be tailored for use in a wide range of applications (Loveland et al., 1991).
4.2. Legends It often appears that there are as many land cover legends as there are applications of land cover data. Because of the variety of land cover legends used in global studies, this discussion will be limited to the most common. Table 2 contains listings of the land cover classes from several of the land cover legends discussed here, which relate to functional categorization of land cover types that could be employed in the study of trace gas fluxes. While there are several general purpose land cover legends, there are also many special purpose schemes developed for specific applications. The SiB, SiB2, and BATS landatmosphere interaction models, for example, have specific land cover categories. These legends have some "quirks" such as the SiB class named "Cropland and C3 Grassland" that represent the unique parameterization requirements of the affiliate model. Our research shows
Use of remote sensing data to delineate functional types in terrestrial and aquatic systems
139
that differences in legend philosophies and class definitions can have major implications in the spatial patterns of land cover and the corresponding estimates of land cover area. Figure 1 illustrates the differences in a given area, western Europe, through the representation of land cover in the SiB, BATS, IGBP, and Running land cover legends. The land cover legends developed and refined by Olson (1994) - Olson Global Ecosystems - is noteworthy because it was designed specifically for global applications. While not specifically developed for use with remotely sensed data, Olson has refined class definitions to permit more straightforward use with satellite imagery. An added strength of Olson is that he has developed attributes describing vegetation structure, seasonality, and climate for each category (Table 3). This legend now has 94 classes. Olson continues to add new classes and refine and improve definitions and attributes as new applications for the scheme arise. The Matthews (1983) legend, consisting of 35 categories, is important because of its longevity. Like Olson, the Matthews database has been used in many applications. As a result, it is pervasive and accepted by many modellers. Matthews is based on Unesco (1973) and therefore stresses potential rather than actual vegetation types. The land cover classification scheme developed by Anderson et al. (1976) for the U.S. Geological Survey, while not designed for global-scale studies, is worth noting because it is one of the few designed specifically for use with remotely sensed data. The Anderson system is based on several important considerations. First, it is hierarchical with subsequent levels of the hierarchy defined to be mapped by increasingly larger scales of remotely sensed data. Second, Anderson classes were developed to yield consistent accuracies of at least 85% at all levels of the hierarchy. Third, the land cover classes were defined to serve as surrogates for land use. The principles from which the legend was developed have been widely accepted. As a result, it is not uncommon to see elements of the Anderson system in other legends. Kfichler (1949) developed a framework for assembling potential natural vegetation maps. While his system was not designed for use with remotely sensed data, the philosophical approach is important. His approach was based on the philosophy that a classification of natural vegetation should avoid rigidity, yet permit world-wide comparison, and be adaptable to mapping. He identified three principals for vegetation classification: (i) the classification framework should be applicable to all types of vegetation in all regions of the earth; (ii) it should provide flexibility to be adapted to all map scales; and (iv) it should be clear and simple. In a more recent publication on vegetation mapping strategies, Kfichler (1967) stressed that a vegetation mapper's best approach to vegetation classification is to be adaptable and diversified in order to suit individual cases and to avoid rigidity. The Running et al. (1994) system is an example of a legend designed both for use with remotely sensed data, and use at the global scale. The Running strategy is based on definitions of three canopy components: vegetation structure (termed above ground biomass by Running), leaf longevity, and leaf type. Vegetation structure defines whether the vegetation retains perennial or annual above ground biomass, an issue for seasonal climate and carbon-balance modelling. It is also a determinant of the surface roughness length parameter that climate models require for energy and momentum transfer equations. Leaf longevity (evergreen versus deciduous canopy) is a critical variable in carbon cycle dynamics of vegetation, and affects seasonal albedo and energy transfer characteristics of the land surface. Leaf longevity indicates whether a plant annually must completely regrow its canopy, or a portion of it, with inferred consequences to carbon partitioning, leaf litterfall dynamics, and soil carbon. The leaf type(needleleaf, broadleaf, and grass) affects gas exchange characteristics. Olson (1994) recently modified the Olson and Watts (1982) global ecosystems framework to be more compatible with remotely sensed data. The global ecosystems framework was designed for use in carbon cycle studies and has 94 classes of land cover and land/water inter-
140
J.E. Estes and T.R. Loveland
Figure 1. Land cover of Western Europe as depicted using the IGBP (top) and SiB (bottom) land cover legends. The most significant difference between the two classifications relates to cropland. The IGBP legend includes a cropland mosaic class (medium gray) that includes 1-km regions with approximately equal areas of crops and woodlands or grasslands. The SiB legend only has a single cropland class (light gray). Dark gray represents forested lands. The land cover data were based on 1-km AVHRR data using methods described by Loveland et al. (1997).
141
Use of remote sensing data to delineate functional types in terrestrial and aquatic systems
Table 3. Olson's global ecosystem forest cover classes and corresponding attributes (Olson, 1994). Olson class
Structure
Foliage
Season
Coniferous forest Deciduous conifer forest
Needleleaf forest Needleleaf deciduous forest and woodland
Evergreen conifer Deciduous
Hot-cold
Broadleaf forest and woodland
Summergreen
Cool-cold
Broadleaf forest
Evergreen
Mild-warm
Deciduous broadleaf forest Evergreen broadleaf forest Cool rain forest Conifer boreal forest Cool conifer forest Cool mixed forest Mixed forest Cool broadleaf forest
Tall/medium forest, fjords
Evergreen conifer
Snow and rain
Medium-low dense-open taiga Tall-low forest and woodland Medium/tall forest Tall/medium forest
Narrow conifer Evergreen conifer Conifer and summer sreen Conifer/summer sreen/svergreen
Cold Cold Snowy
Medium/tall forest
Summer green
Cool-cold
Deciduous broadleaf forest
Medium/tall forest
Summer green/evergreen
Mild-hot
Conifer forest
Very tall-low forest, woodland
Evergreen trees and shrubs
Hot-mild
Montane tropical forest
Montane medium-low wood, herbs
Evergreen woods/herbs
Mild
Seasonal tropical forest
Medium forest and plantations
Evergreen, broadleaf
Hot, seasonal, moist
Dry tropical woods
Low open forest, woodlands
Drought deciduous
Hot-warm
Tropical rainforest Tropical degraded forest
Tall-medium closed forest Medium-low open forest and woodland
Evergreen broadleaf Evergreen broadleaf
Hot Hot
Dry evergreen woods Small leaf mixed woods
Woodland, low/dry forest Broadleaf evergreen Low-medium taiga, edges, groves Summer green, conifer
Dry Cold
Deciduous and mixed boreal forest
Medium-low tbrest and woodlands
Deciduous. mixed taiga
Cold
Narrow conifers
Dense-sparse woodlands with shrubs and herbs
Evergreen, mixed
Subpolar, mountain
Wet schlerophylic forest
Tall open eucalyptus forest
Broadleaf evergreen
Wet
Moist eucalyptus Rain green tropical forest
Medium open eucalyptus forest
Broadleaf evergreen
Mild
Medium forest, closed-open
Rain green
Hot-warm
face mosaics. Of the classification systems used for remote sensing, this offers the most detailed framework for aquatic systems. While different ocean and fresh water types are not provided, the system offers six classes of wetlands, and over twelve coastline categories. The coastline classes are difficult to map with remotely sensed data however. The IGBP recently developed a land cover legend (Belward, 1996). The legend consists of 17 general cover types selected based on the requirements of the IGBP Core Projects. It was designed specifically for use with 1-km remotely sensed data from AVHRR. The categories embrace the philosophy presented by Running et al. (1994) but with modifications to: (i) be compatible with classifications systems currently used for environmental modelling (e.g. SiB and BATS); (ii) provide, where possible, land use implications; and, (iii) represent landscape mixtures and mosaics. The IGBP legend retains key elements of Running, including climate-independent class definitions, and reliance on ancillary remotely sensed measures, such as vegetation greenness indices, as relative indicators of temporal dynamics of biophysical properties. A comprehensive legend for mapping aquatic systems is the Cowardin et al. (1979) classification scheme. The Cowardin system addresses wetlands, palustrine, and lacustrine systems. The detailed classes of Cowardin are typically mapped using large-scale aerial photography.
142
J.E.Estes and T.R. Loveland
4.3. Large-area land cover mapping using remotely sensed data In reviews of data that could be used to improve global land cover maps, IGBP scientists reached the consensus that AVHRR data were the logical data for continental to global-scale land cover mapping (IGBP, 1990). This choice was based on the following data characteristics: (i) an appropriate pixel size of the data in relation to the global coverage requirements; (ii) an ability for researchers to generate land cover classes of the type required; (iii) the past and future continuity of the AVHRR data; (iv) the inexpensive costs associated with data collection; and, (v) the daily global coverage. A follow-up IGBP report (1992) reiterated that when considering the need for global, spatially detailed data for many applications, 1-km data from AVHRR sensors are a prime source, because of the global coverage, and relatively high-resolution global view. The report acknowledges that there are other AVHRR products with more coarse resolution (the 4-km global area coverage and the 16-km global vegetation index); however, these products are not in a form suitable for global scientific applications. The summary of an Oak Ridge National Laboratory workshop on land cover concluded that satellite remote sensing offers the scientific community a globally consistent base of data from which the spatial dimensions and temporal fluxes of phenomena such as land cover change can be made within some accepted accuracy limit (Dale, 1990). They recognized problems in using remote sensing including: (i) data gaps exist; (ii) clouds, smoke and atmosphere obscure images; (iii) older data already are discarded; (iv) high-resolution data are expensive; (v) high investments needed for processing facilities; (vi) direct information on biomass not directly achievable; and, (vii) future availability of monitoring data is questionable. However, several strengths were also identified, such as: (i) unbiased data; (ii) high positional accuracy; (iii) high temporal frequency; (iv) easily incorporated into a GIS; (v) extended spectral coverage offers detection of vegetation features not seen by human eye; (vi) a base for checking other data. Data from the AVHRR has been used for comprehensive continental to global mapping (Achard and Estreguil, 1995; Defries et al., 1995; Loveland and Belward, 1997). Ehrlich et al. (1994) provide a comprehensive overview of the role of AVHRR for environmental studies, including land cover mapping, vegetation dynamic studies, tropical forest monitoring, vegetation production estimation, fire risk assessment, and biophysical parameter estimation. They claim that AVHRR daily coverage has shown to be of great value to land cover scientists because it increases the chance of producing cloud-free composites. A good fraction of the researchers have found that AVHRR channels 3 and 4 provide useful information on land cover in the tropics. Based on their review of recent applications, the authors conclude that an area of major application potential for AVHRR 1-km data involves land cover mapping, since little quantitative information about land surface is available at regional or global scales (Ehrlich et al., 1994). At present, mapping land cover using remotely sensed data boils down to stratifying the spectral-temporal radiance values recorded by the sensor into discrete land cover categories. Biophysical land cover characteristics are then assigned to each category. Programs to develop consistent large-area 1-km resolution AVHRR data sets abound. For example, the U.S. Geological Survey has been developing biweekly AVHRR products for the conterminous U.S. beginning in 1989 (Eidenshink, 1992). At the global level, the IGBP (1992) has served as the catalyst for the development of a global 1-km AVHRR land data set. An international effort, consisting of 30 AVHRR ground station operators, the USGS, NASA, NOAA, and the European Space Agency have been gathering all daily 1-km AVHRR over land areas since April 1992. The USGS, as part of the IGBP initiative, and the NASA EOS Pathfinder Project is processing these daily data into global 1-km 10-day composites based on
Use of remote sensing data to delineate functional ,~)pes in terrestrial and aquatic systems
143
a maximum value compositing strategy (Eidenshink and Faundeen, 1994). These data were developed specifically to be used for global land cover mapping. Similar data sets, prepared from 4 km AVHRR data, have been assembled as part of the NASA Pathfinder Program (James and Kalluri, 1994) and through European Community projects (Malingreau and Belward, 1994). While global land cover studies have focused on the use of AVHRR data, Landsat data, with 80 m and 30 m resolution, and Satellite Pour L'Observation de la Terre (SPOT) high resolution visible data with 20m resolution, has been used for regional studies of land cover change. For example, Skole and Tucker (1993) have used Landsat data spanning a period of nearly 20 years to map and measure deforestation in the humid tropical forests. Landsat, and high resolution data from SPOT satellites, are particularly important for regional studies of land cover and land cover change. While the high spatial resolution is advantageous for land cover chap ge studies, extremely large data volumes and relatively high data costs have typically been considered as barriers to the use of Landsat, SPOT and similar sensor data for global studies. However, a new mapping initiatives proposed to contribute to both international science activities and national resource management programs calls for the use of high resolution satellite data for repetitive mapping of global forest lands (Skole et al., 1997). Clearly, technical advancements in computing, mapping methods, and data costs, combined with an urgent need for increasingly higher resolution land cover data sets, are motivating the initiation of projects with scopes that were previously considered impossible.
4.4. Accuracy considerations Estes and Mooneyhan (1994) remind us that science quality global land cover maps require that we understand the error terms of the produced data sets. The error terms associated with areal extent of global land cover far exceed the errors in precision with which many land related global change measurements are made. Fisher and Langford (1996) state that the rate of misclassification of land cover derived from satellite imagery is notoriously high, and inaccurate land cover could clearly have profound effects on the accuracy of model results. They speculate that the overall accuracy in a classified remotely sensed image rarely exceeds 90% but commonly is better than 80%. However, they recognized that the classification errors at the pixel level in a raster data set can be large (classification accuracy) without impacting the accuracy of estimates of regional amounts (inventory accuracy) due to offsetting errors in area. They conclude that models in which land cover data are used must consider the accuracy of such data sets. There are a number of studies which have dealt with methods for assessing the accuracy of landcover maps (Congalton et al., 1983; Congalton, 1991; Rosenfeld et al., 1982; Stehman, 1992; and Van Deusen, 1996). Belward (1996) states that accuracy assessments of digital land cover classifications are typically based on contingency tables, or confusion matrices, where accuracy is expressed in terms of errors of omission and commission, or in terms of agreement analysis using the Kappa test statistic (Stehman, 1996b; Congalton, 1991; Rosenfeld and Fitzpatrick-Lins, 1986). Contingency tables are created by comparing on a class by class basis the land cover classification with an independent data source - field observations, existing maps, higher resolution imagery - collected using a statistically valid sampling strategy (Robinson et al., 1983; Stehman, 1996a, Rosenfield and Fitzpatrick-Lins, 1986). While such methods are well established, they have typically only been applied to local scale classifications, occasionally to subnational regional scale work, and only in isolated instances on scales greater than these (Hay, 1988; Manshard, 1993; Estes and Mooneyhan,
144
J.E.Estes and T.R. Loveland
1994; Jeanjean et al., 1996). In essence, there is no agreed upon state-of-the-art for the validation of thematic products ("maps") derived from remotely sensed data. There are only state-of-the-practice methodologies. While a number of these are included in the references discussed above, Belward (1996) discusses the issue of selection of a scheme to validate the IGBP DISCover product. This product is derived from the land cover characteristics database being developed under IGBP auspices. This paper should be read by anyone interested in issues of large area land cover validation. Belward (1996) states the assumptions made and the options considered in the selection process. Details of the methodology are discussed as well. Based upon considerations of cost, resources, and time, a stratified random sampling procedure was adopted (Belward, 1996).
5. C o n c l u s i o n s
and recommendations
Researchers have a wide variety of options available when seeking to utilize remotely sensed data for the scaling of trace gas fluxes between terrestrial and aquatic ecosystems and the atmosphere. Currently an important application of remote sensing for improving our understanding of trace gas fluxes is the classification of land cover types. As we look to the future, we anticipate several trends in the way land cover type data are used for measuring, mapping, monitoring and modelling trace gas flux. These include: - The ways land cover data/information are used will become more diverse as models improve and become more sophisticated; - A strong move towards the use of quantitative land cover variables along with nominal land cover classes; - Data will be of increasingly higher (finer) spatial resolution, and will most likely must be multi-scalar; - Data set development will become cyclic so temporal issues can be more routinely addressed; - There will be continuity in the base data/information content between historical and future land cover products so that long-term trends can be more efficiently and effectively analyzed; and, - Validation protocols will become a routine element of science driven land cover characterization studies. We are in an era where there are, and will continue to be, large volumes of remotely sensed data available for environmental applications. The ability to acquire data at scales from site specific to globally comprehensive exists, as does the means to verify or validate the result of a given classification based on the extraction of information from remotely sensed data. However, the means in which remotely sensed data are used will continue to have significant resource, time, cost, and accuracy implications. As such it is critical that researchers understand the implications of the choices they make when they use remotely sensed data to map land cover for the purpose of studying trace gas fluxes. With each choice, there are a series of questions that must be addressed. In reality, the first series of questions involves the choices which must be made in laying out any study. These questions involve the elucidation of the goals and objectives of the overall study. Some extremes here, for example, are the modelling all trace gas fluxes from 1970 to the present at a global scale; to the estimation of a single trace gas for a given date at a local scale? Based on a clear understanding of the specific, the following must be determined:
Use of remote sensing data to delineate functional types in terrestrial and aquatic systems
145
Selection or development of the classification system to be used; Determination of the overall attributes of the remote sensing systems, techniques, and methodologies to be employed; and, Definition of an appropriate validation strategy.
-
-
-
With regard to the selection or development of a classifications system, the user must understand the type and level of detail of the information that must be extracted from the the remote sensing and ancillary data to be employed. Questions here include the following: - Is the use of an existing land cover classification system appropriate or acceptable? - Is a new system needed ? - If the former, are their studies illustrating the use of remote sensing for deriving the individual classes which make up the system ? With what accuracy were the individual classes derived in these studies ? - Is this level of accuracy appropriate for the study at hand ? If a new classification system is required, a new series of questions must be asked. -
-
Questions, which need to be answered include: - What is an appropriate land cover classification scheme ? Are the individual classes selected mutually exclusive ? - Can these classes of land cover be extracted from remotely sensed data ? - How efficiently, effectively and accurately can these classes be extracted ? -
Next, a series of choices associated with the relationship between the required information and the technical specification of the source data for the analysis must be addressed. Here the user must also consider, among other things, the date of acquisition; image scale or pixel size, and the spectral region or regions and bandwidths to be employed. All these choices really involve questions regarding the most appropriate and cost effective and personnel efficient means to proceed. All choices are important and influence the types of information that can accurately be extracted from analysis of the remotely sensed and other data used in a given study. Choice of an accuracy assessment strategy is also very important. If you do not know both the area w~ighted and the by class accuracies of a given land cover product, there will be unacceptable uncertainties of unknown magnitudes in key model parameters. However, care must be exercised when setting accuracy goals. Unreasonable goals can drive the choice of a verification strategy to unacceptable cost levels given current state of the practice techniques. It is not unreasonable to say that if a goal of 95% per class accuracy was set for a product, the cost of verification could easily approach and perhaps exceed the costs associated with product development. Based on all of the above we recommend that multidisciplinary approaches be taken in any effort directed at the use of remotely sensed data for the production of land cover data for the scaling of trace gas fluxes between terrestrial and aquatic ecosystems and the atmosphere. The choices involved in any efforts of this type are complex and the state of the art, state of the practice is evolving. New understandings on the land, aquatic and atmospheric science side are matched with technology and methodological advances in the remote sensing area. It is difficult enough to keep abreast of any one of the areas of science or of remote sensing let alone all. Townshend et al. (1991) showed a 100 ~~, variation in land cover classes in some 16 global land cover products created between 1954 and 1985. This is alarming, and we should be concerned. When DeFries et al. (1995) point out that when data from three global land cover
146
J.E.Estes and T.R. Loveland
classifications that are commonly used in modelling today are compared, the spatial agreement is about 26% of the global lancl surface, we should be worried. Such variations are important and the uncertainty they represent must be reduced. The current IGBP DISCover effort is an attempt to do just that. Our final recommendation is that we continue and perhaps even expand this type of internationally coordinated and financed effort with the goal of producing validated land cover products on a regular basis at the variety of scales required for scientific investigations. Remote sensing is a powerful tool for the collection of environmental data. When combined with other "more conventional" data, they can be analyzed to extract a wide variety of environmental themes. Care, however must be exercised. The acquisition and processing of these data are expensive and can be difficult. Yet, the processing of remotely sensed data provides the only means we have today to derive internally consistent global land cover data whose accuracy can be statistically validated. As processing power improves and data collection becomes more routine, we look forward to a time when global land cover data will be updated routinely and input to a wide variety of models. Models that will improve our fundamental understanding of the dynamics of the complex interconnected system that supports and sustains life on Earth.
References AAAS (1983) Resource inventory and baseline study methods for developing countries. American Association for the Advancement of Science, Washington, D.C., 539 pp. Achard F. and C. Estreguil (1995) Forest classification of southeast Asia using NOAA AVHRR Data. Remote Sensing of Environment 54:198-208. Anderson, J.R., E.E. Hardy, J.T. Roach and R.E. Witmer (1976) A land use and land cover classification system for use with remote sensor data. Professional Paper 964, U.S. Geological Survey, Reston, VA: U.S. Geological Survey, 28 pp. Aspinall, R., and N. Veitch (1993) Habitat mapping from satellite imagery and wildlife survey data using a bayesian modeling procedure in a GIS. Photogrammetric Engineering and Remote Sensing 59:537-543. Bailey, R.G. (1987) Suggested hierarchy of criteria for multiscale ecosystem mapping. Landscape and Urban Planning 14:313-319. Belward, A.S. (Ed.) (1996) The IGBP-DIS global 1 km land cover data set (DISCover): Proposal and Implementation Plans. IGBP-DIS Working Paper No. 13, IGBP-DIS Office, Meteo-France, Toulouse, France, 61 pp. Bonan, G.B. (1995) Land-atmosphere interactions for climate system models: coupling biophysical, biogeochemical, and ecosystem dynamical processes. Remote Sensing of Environment 51:57-73. Box, E.O. (1981) Predicting physiognomic vegetation types with climate variables. Vegetatio 45:127139. Brown, J.F., T.R. Loveland, J.W. Merchant, B.C. Reed and D.O. Ohlen (1993) Using multisource data in global land characterization: concepts, requirements, and methods. Photogrammetric Engineering and Remote Sensing 59:977-987. Congalton, R.G. (1991) A review of assessing the accuracy of classifications of remotely sensed data. Remote Sensing of Environment 37: 35-46. Congalton, R.G., R.G. Oderwald and R.A. Mead (1983) Assessing landsat classification accuracy using discrete multivariate analysis statistical techniques. Photogrammetric Engineering and Remote Sensing 49:1671-1678. Copeland, J.H., P.A. Pielke and T.G.F. Kittel, 1996) Potential climatic impacts ol vegetation change: A regional modeling study. Journal of Geophysical Research 101:7409-7418.
Use of remote sensing data to delineatefunctional types in terrestrial and aquatic systems
| 47
Cowardin, L.M., V. Carter, F.C. Golet, and E.T. LaRoe (1979) Classification of wetlands and deepwater habitats of the United States. FWS/OBS-79/31, U.S. Fish and Wildlife Service, Washington, D.C., 131 pp. Dale, V.H. (1990) Report on a workshop on using remote sensing to estimate land use change. Environmental Sciences Division Publication 3397. Oak Ridge National Laboratory, Oak Ridge, Tennessee, 54 pp. DeFries, R., M. Hansen and J.G.R. Yownshend (1995) Global discrimination of land cover types from metrics derived from AVHRR pathfinder data. Remote Sensing of Environment 54:209-222. Dickinson, R.E. (1995) Land processes in climate models. Remote Sensing of Environment 51:27-38. Dickinson, R.E., A. Henderson-Sellers, P.J. Kennedy M.F. and Wilson (1986) Biosphere-atmosphere transfer scheme (BATS) for the NCAR Community Climate Model. NCAR Technical Note NCAR/TN-275+STR, Boulder, Colorado. Ehrlich, D., J.E. Estes, and A. Singh (1994) Applications of NOAA-AVHRR 1-km data for environmental monitoring. International Journal of Remote Sensing 15:145-161. Eidenshink, J.C. (1992) The 1990 conterminous U.S. AVHRR data set. Photogrammetric Engineering and Remote Sensing 58:809-813. Eidenshink, J.D., and J.L. Faundeen (1994) The 1 km AVHRR global land data set: First stages in implementation. International Journal of Remote Sensing 15:3443-3462. Estes, J.E., and D.W. Mooneyhan (1994) Of maps and myths. Photogrammetric Engineering and Remote Sensing 60:517-524. Field, C.B., J.T. Randerson and C.M. Malmstrom (1995) Global net primary production: Combining ecology and remote sensing. Remote Sensing of Environment 51:74-88. Hall, F.G., J.G.R. Townshend and E.T. Engman (1995) Status of remote sensing algorithms for estimation of land surface state parameters. Remote Sensing of Environment 51:138-156. Hay, A.M. (1988) The derivation of global estimates from a confusion matrix. International Journal of Remote Sensing 9:1395-1398. Htun, Nay (1993) The driving forces of global change, paper presented at Aspen Global Change Institute's fourth annual Walter Orr Roberts Memorial Public Lecture Services, Aspen, Colorado. Hunt, E.R., Jr., S.C. Piper, R. Nemani, C.D. Keeling, R.D. Otto and S.W. Running (1996) Global net carbon exchange and intra-annual atmospheric CO2 concentrations predicted by an ecosystems process model and three-dimensional atmospheric transport model. Global Biogeochemical Cycles 10:431-456. IGBP (1990) The International Geosphere Biosphere Pro-gramme: A study of global change- The initial core projects. Global Change Report No. 12, International Geosphere-Biosphere Programme, Stockholm, Sweden. IGBP (1992) Improved global data for land applications. Global Change Report No. 20, International Geosphere-Biosphere Programme, Stockholm, Sweden, 87 pp. IGBP (1993) Biospheric aspects of the hydrological cycle (BAHC): The operational plan. Global Change Report No. 27. International Geosphere-Biosphere Programme, Stockholm, 103 pp. IGBP (1994) IGBP global modeling and data activities 1994-1998. Global Change Report No. 30. International Geosphere-Biosphere Programme, Stockholm, 87 pp. James, M.E. and S.N.V. Kalluri (1994) The pathfinder AVHRR land data set - An improved coarse resolution data set for terrestrial monitoring. International Journal of Remote Sensing 15:33473363. Jeanjean, H., F. Achard and J.-P. Malingreau (1996) Large scale tropical forest change monitoring using multiple resolution satellite data: from hot spot detection to global deforestation assessment? Presented at the Spatial Accuracy Assessment in Natural Resources and Environmental Sciences: 2nd Symposium, Fort Collins, Colorado. Kemp, K.K. (1992) Spatial models for environmental modeling with GIS. In: Proceedings: 5th International Symposium on Spatial Data Handling, v 2. International Geographical Union, Charleston, South Carolina, pp. 524-533.
148
J.E.Estes and T.R Loveland
Kittel, T.G.F., D. Ojima, D.S. Schimel, D.S. McKeown, R. Bromberg, T. Painter, T. Rosenbloom and W. Parton (1996) Model GIS integration and data set development to assess terrestrial ecosystem vulnerability to climate change. In: M.F. Goodchild, L. Steyaert and B. Parks (Eds.) GIS and Environmental Modeling." Progress and Research Issues, GIS World, Inc., Fort Collins, Colorado, pp. 293-297. Ktichler, A.W. (1949) A physiognomic classification of vegetation. Annals of the Association of American Geographers 39:201-210. KiJchler, A.W. (1967) Vegetation mapping. Ronald Press Company, New York, 472 pp. Loveland, T.R. and A.S. Belward (1997) The IGBP-DIS Global 1 km land cover data set, DISCover: First Results. International Journal of Remote Sensing 18:3291-3295. Loveland, T.R., J.W. Merchant, D.O. Ohlen and J.F. Brown (1991) Development of a land cover characteristics database for the conterminous U.S. Photogrammetric Engineering and Remote Sensing 57:1453-1463. Loveland, T.R., J.W. Merchant, B.C. Reed, J.F. Brown, D.O. Ohlen, P. Olson and J. Hutchinson (1995) Seasonal land cover regions of the United States. Annals of the Association of American Geographers 85:339-355. Loveland, T.R., D.O. Ohlen, J.F. Brown, B.C. Reed, Z. Zhu, L. Yang and J.W. Merchant (1997) Western hemisphere land cover: progress toward a global land cover characteristics data base. In: Pecora 13 Proceedings, American Society of Photogrammetry and Remote Sensing, Bethesda, MD (in press). Malingreau J.P. and A.S. Belward (1994) Recent activities in the European Community for the creation and analysis of global AVHRR data sets. International Journal of Remote Sensing 15:3397-3416. Manshard, W. (1993) Geography and the international global change programmes. European Review 1:309-317. Matthews, E. (1983) Global vegetation and land use: New high resolution data bases for climate studies. Journal of Climatology and Applied Meteorology 22:474-487. Matson, P.A., and D.S. Ojima (1990) Terrestrial biosphere exchange with global atmospheric chemistry: terrestrial biosphere perspective of the IGAC project. IGBP Global Change Report 13, International Geosphere Biosphere Programme, Stockholm, Sweden, 103 pp. Meeus, J.H.A. (1995) Pan-European landscapes. Landscape and Urban Planning 31:57-79. Milaova, E.V. and A.V. Kushlin (1991) Report on methodology of compiling the maps of the present status of landscapes. Moscow State University, Faculty of Geography, Moscow, Russia, 39 pp. Nemani, R.R. and S.W. Running (1996) Satellite monitoring of global land cover changes and their impact on climate. Climatic Change 31:393-413. Olson, J.S. (1994) Global ecosystems framework: Definitions. USGS EROS Data Center Internal Report, Sioux Falls, South Dakota, 37 p. Olson, J.S. and J.A. Watts (1982) Major world ecosystem complex map. Oak Ridge, Tennessee: Oak Ridge National Laboratory. Omernik, J.M. (1987) Ecoregions of the conterminous United States. Annals of the Association of American Geographers 77:118-125. Parton, W.J., J.W.B. Stewart and C.V. Cole (1988) Dynamics of C, N, P, and S in grassland soils: A model. Biogeochemistry 5:109-131. Peplies, R.W. and R.G. Honea (1992) Some classic regional models in relation to global change studies. 88th Annual Meeting, Association of American Geographers (Unpublished), San Diego, California, 24 pp. Riebsame, W.E., W.J. Parton, K.A. Galvin, I.C. Burke, L. Bohren, R. Young, and E. Knop (1994) Integrated modeling of land use and cover change. BioScience 44:350-356. Robinson, J.W., F.J. Gunther W.J. and Campbell (1983) Ground truth sampling and LANDSAT accuracy assessment, Document N83-26161, National Technical Information Services. Rosenfield, G.H. and K. Fitzpatrick-Lins (I986) A coefficient of agreement as a measure of thematic classification accuracy. Photogrammetric Engineering and Remote Sensing 52:223-227.
Use of remote sensing data to delineatefunctional types in terrestrial and aquatic systems
149
Rosenfield, G.H., K. Fitzpatrick-Lins and H.S. Ling (1982) Sampling for thematic map accuracy testing. Photogrammetric Engineering and Remote Sensing 48:131-137. Running, S.W., T.R. Loveland and L.L. Pierce (1994) A vegetation classification logic based on remote sensing for use in global biogeochemical models. Ambio 23:77-81. Schimel, D.S., T.G. Kittel and W.J. Parton (1991) Terrestrial biogeochemical cycles: Global interactions with the atmosphere and hydrology. Tellus 43AB: 188-203. Sellers, P.J. (Ed.) (1993) Remote sensing of the land surface for studies of global change: Models algorithms- experiments. In: ISCLSP Workshop Report. International Satellite Land Surface Climatology Project, Columbia, Maryland, 106 pp. Sellers, P.J., S.I. Rasool and H.-J. Bolle (1990) A review of satellite data algorithms for studies of the land surface. Bulletin of the American Meteorological Society 71:1429-1447. Sellers, P.J., Y. Mintz, Y.C. Sud and A. Dalcher (1986) A Simple Biosphere Model (SiB) for use within general circulation models. Journal of Atmospheric Science 43:505-531. Sellers, P.J., S.O. Los, C.J. Tucker, C.O. Justice, D.A. Dazlich, G.J. Collatz and D.A. Randall (1996) A revised land surface paramterization (SiB2) for atmospheric GCMs - Part II: The generation of global fields of terrestrial biophysical parameters from satellite data. Journal of Climate 9:706737. Skole, D. and C. Tucker (1993) Tropical deforestation and habitat fragmentation in the amazon: Satellite data from 1978 to 1988. Science 260:1905-1910. Skole, D.L., C.O. Justice, J. Townshend and A.C. Janetos (1997) A land cover change program: Strategy for an international effort. Mitigation and Adaptation Strategies for Global Change (in press). Steffen, W.L., B.H. Walker, J.S.I. Ingram and G.W. Koch (1992) Global change and terrestrial ecosystems: The operational plan. IGBP Global Change Report No. 21, International Geosphere Biosphere Programme, Stockholm, Sweden, 95 pp. Stehrnan, S.V. (1992)Comparison of systematic and random sampling for estimating the accuracy of maps generated from remotely sensed data. Photogrammetric Engineering and Remote Sensing 58:1343-1350. Stehman, S.V. (1996a) Cost-effective, practical sampling strategies for accuracy assessment of largearea thematic maps. In: Proceedings, Second International Symposium on Spatial Accuracy in Natural Resources and Environmental Science. USFS General Technical Report RM-GTR-277. U.S. Forest Service, Fort Collins, Colorado, pp. 485-492. Stehman, S.V. (1996b) Estimating the kappa coefficient and its variance under stratified random sampling. Photogrammetric Engineering and Remote Sensing 62:401-407. Steinwand, D.R., J.A. Hutchinson and J.P. Snyder (1995) Map projections for global and continental data sets and an analysis of pixel distortion caused by reprojection. Photogrammetric Engineering and Remote Sensing 61:1487-1498. Steyaert, L.T., T.R. Loveland, J.F. Brown and B.C. Reed (1994) Integration of environmental simulation models with satellite remote sensing and geographic information systems technologies: Case studies. In: Proceedings: Pecora 12 Symposium on Land Information from Space-Based Systems, American Society of Photogrammetry and Remote Sensing, Bethesda, Maryland, pp. 407-417. Townshend, J., C. Justice, W. Li, C. Gurney and J. McManus (1991) Global land cover classification by remote sensing: present capabilities and future possibilities. Remote Sensing of Environment 35:243-255. Turner, B.L., II and W.B. Meyer (1991). Land use and land cover in global environmental change: Considerations for study. International Social Science Journal 130:669-677. Unesco (1973) International classification and mapping of vegetation. Paris, France: Unesco, 93 pp. Ustin, S.L., C.A. Wessman, B. Curtiss, E. Kasischke, J. Way and V.C. Vanderbilt (1991) Opportunities for using the cos imaging spectrometers and synthetic aprture radar in ecological models. Ecology 72:1934-1945. Van Deusen, P.C. (1996) unbiased estimates of class proportions from thematic maps. Photogrammetric Engineering and Remote Sensing 62:409-412.
150
J.E.Estes and T.R. Loveland
Wilson, M.F. and A. Henderson-Sellers (1985) A global archive of land cover and soils data for use in general circulation models. Journal of Climatology 5:119-143.
Chapter 7
HOW CAN WE BEST DEFINE FUNCTIONAL TYPES AND INTEGRATE STATE VARIABLES AND PROPERTIES IN TIME AND SPACE?
S.P. Seitzinger, J.P. Malingreau, N.H. Batjes, A.F. Bouwman, J.P. Burrows, J.E. Estes, D. Fowler, M. Frankignoulle and R.L. Lapitan
This Page Intentionally Left Blank
Approaches to scaling a trace gas fluxes in ecosystems A.F. Bouwman, editor 9 Elsevier Science B.V. All rights reserved
WORKING GROUP REPORT HOW CAN WE BEST DEFINE FUNCTIONAL TYPES AND INTEGRATE STATE VARIABLES AND PROPERTIES IN TIME AND SPACE?
S.P. Seitzinger (Rapporteur), J.P. Malingreau (Chairman), N.H. Batjes, A.F. Bouwman, J.P. Burrows, J.E. Estes, D. Fowler, M. Frankignoulle and R.L. Lapitan
I. I n t r o d u c t i o n
The charge to this working group was to address the question: How can we best define functional types and integrate state variables and properties in time and space? In addressing this question the working group, with input at intervals from members of other working groups, discussed a wide variety of issues. The challenge the group faced was to work toward a common lexicon so that participants from different disciplines could come to a common understanding of the question and the magnitude of the task involved in formulating an answer. What quickly became apparent was: (i) the need to produce a definition for functional type; and (ii) to select an approach that would allow us to address the potential value of functional types for the scaling of trace gas fluxes. This lead to a discussion and general agreement within the working group that modelling of fluxes could be enhanced if a paradigm could be developed wherein geo-spatially defined areas of the earth's surface could be defined that behaved in a similar manner with respect to the processes controlling or influencing the fluxes of traces gases. If these areas could be identified through time, and if the state variables and properties controlling fluxes could be attached as attributes, then modelling over space and through time could potentially be improved. The definition of functional type developed by the working group and used in the current report is as follows: A functional type is an area on the earth's surface which behaves in a defined manner with respect to processes controlling the flux of a particular trace gas or group of them. The material in this chapter starts with a brief discussion of the overall rationale for studying trace gases. The need for improved approaches for obtaining spatially explicit global estimates of trace gas fluxes is then developed. The practical and theoretical utility of a functional type approach and the basic elements of that approach follow. The application of this type of approach is illustrated using methane as a specific example. The application of this approach to a subset of trace gases is then examined to explore the potential for generalization of the concept. Tools for identifying the geographical distribution of functional types are presented including global maps, geographic information systems (GIS) and remote sensing technologies, along with the utility of global circulation and chemical transport models. The chapter ends with a series of conclusions and recommendations.
154
s.P. Seitzinger et al.
2. Background 2.1. The coupled atmosphere/land/ocean system The atmosphere, land and oceans form a coupled system, in which physical, chemical and biological processes occur. These determine the composition of the atmosphere and as a consequence the environment experienced by the biosphere. Anthropogenic activity has increased the emissions and deposition of many trace atmospheric constituents, which are changing the chemistry of the earth/atmosphere system on a global scale. The magnitude of the anthropogenic impact on the composition of the trace atmospheric constituents has been growing since the industrial revolution and is predicted to continue to increase as a result of future population increases and the elevated standard of living. There are a number of potential current and future consequences of these changes including global warming and depletion .~f stratospheric ozone. Two of the groups of gases that are involved in global change include those that absorb a significant amount of infrared radiation, and those that react chemically in the stratosphere to destroy ozone. Arrhenius, on the basis of hand calculations, pointed out a little over a century ago, that the doubling of the atmospheric burden of carbon dioxide would result in a significant increase in the temperature at the earth's surface (Arrhenius, 1896). This effect has been termed the "greenhouse effect". It is now well known that a family of trace gases, which absorb significant amounts in the infrared region, are also "greenhouse gases" (e.g. CO2, CH4, N20). Such gases typically have relatively long tropospheric lifetimes. An exception however is tropospheric ozone (O3), which is both transported to and produced within the troposphere. Aerosols, in contrast to greenhouse gases, tend to cool the atmosphere by scattering incoming solar radiation. Another group of gases, halocarbon compounds, is involved in stratospheric ozone depletion. These gases (chlorofluorohydrocarbons and halones) are released into the troposphere and transported to the stratosphere where they yield significant amounts of stratospheric halogen atoms, mainly C1 and Br. These radicals participate in a series of homogeneous and heterogeneous cataljtic cycles which destroy stratospheric ozone. Additionally, N20 reacts photochemically in the stratosphere in a series of reactions that destroy ozone.
2.2. Need for an improved approach for estimating global fluxes of trace gases Accurate assessments of the sources, sinks, impacts and consequences of changes in trace gas fluxes are needed for the development of effective environmental policies that will reduce the adverse consequences of human activities. Such policies are required to regulate anthropogenic activities and thereby control emissions from anthropogenic sources. In compliance with international agreements, national governments now are required to prepare emissions inventories resulting from anthropogenic activity (IPCC, 1995; 1997). However, despite the need for accurate assessments of the sources and sinks of trace gases at local as well as regional and country scales, the available data and approaches currently used often do not match the scale of the questions asked For example, chemistry-transport models (CTM) are important tools in the assessment of sources and sinks of trace gases on the global scale. CTMs, coupled with general circulation models (GCMs), are used to simulate the resultant atmospheric behaviour of selected scenarios. The spatial variations of atmospheric concentrations of trace gases are a result of
How can we best define functional types and integrate state variables and properties in time and space ?
15 5
numerous sources within a region. CTMs coupled to GCMs can provide broad scale spatial information on the magnitude of total sources and sinks within a region. However, they generally can not determine the magnitude and/or location of specific sources from the spatial variation in atmospheric concentrations. This is because the source and sink fluxes have been integrated over weeks to years and over hundreds or thousands of kilometers. On the other hand, measurements of trace gas fluxes are generally on small scales (cm to m) and over short time periods (minutes to hours). In addition they can provide detailed information on the factors controlling fluxes for a particular parcel of land or aquatic (rivers, lakes, marine) environment. Extrapolation of those measurements to other locations with slightly different characteristics is generally problematic. However, neither measurements nor models can be made at all scales and for all places, although spatially explicit estimates of trace gas fluxes are needed at many scales and locations, in order to evaluate the relative contribution of various ecosystems and human activities to trace gas fluxes. In fact, methodologies currently used to estimate the global distribution of trace gas fluxes often rely on the application of constant emission factors, or an average flux rate, to broad land cover/land use categories (e.g. wetlands, agricultural lands, estuaries). And the spatial distribution of those land cover/land uses is often at a very large scale (e.g. per country or per latitudinal zone). This approach severely limits our ability to estimate variation in trace gas fluxes through space and time, except at very gross scales. The above considerations lead the working group to explore the utility of a functional type approach for estimating spatially explicit, time varying global fluxes of trace gases.
3. A functional approach for estimating global fluxes of trace gases 3.1. Proposed approach A functional type approach to the estimation of trace gas fluxes combines: (i) functional relationships between environmental variables and trace gas fluxes with, (ii) spatially explicit information on the global distribution of controlling variables across the temporal and spatial scales of interest (Figure 1). Information used to identify the controlling variables and develop the functional relationships can be obtained from detailed process-based studies and modelling efforts. Information on the spatial distribution of the controlling variables is obtained from various sources including both ground-based measurements and remote sensing.
3.2. Practical and theoretical utility of a functional type approach The identification of a functional type does not necessarily fall within the familiar categories such as forests, wetlands, agricultural soils, estuaries and rivers. Rather it relies on the spatial and temporal distribution of controlling variables which may cut across many ecosystem types. This requires spatially disaggregated data sets for the driving variables, in contrast to the more conventional approach of identifying regions based on the vegetational composition. However, the working group also noted that, realistically, development of a mechanistic approach requires that the input parameters need to be ones that are detectable from remote sensing (e.g. surface temperature, flooding regime) or relatively easily measured ground/field parameters (physical or chemical soil propo,'ties). Current remote sensing capabilities, the use of general circulation models, new remote sensing capabilities coming on line in the near future, as well as ideas for focussing efforts for developing new capabilities were discussed.
156
s.P. Seitzinger et al.
Figure 1. Dynamic relationships between key parameters and variables in defining functional types for assessing trace gas fluxes in aquatic and terrestrial systems.
Obviously, development of the functional relationships between key variables and the flux of a trace gas is no small task and a single relationship will not necessarily be applicable at all scales and for all places. Nor can one technique or approach provide the degree of information required to address all current and future issues. While there is considerable information available from previous studies on some of the factors controlling the flux of a particular trace gas, considerable additional efforts will be needed to refine that understanding across a range of systems. The group noted that for some gases extensive data exist which are valuable for the estimation of emission or deposition fluxes. However, the working group could not identify a trace gas for which there exists a comprehensive set of relationships between the net flux and the major environmental variables which regulate it. Furthermore, in the cases with valuable relationships (e.g. between the net flux and soil temperature), the data sets to apply these relationships across terrestrial or aquatic ecosystems globally are patchy at best and in many cases are missing. No single functional type scheme is appropriate for all trace gases, due to the differences in controlling factors for the various gases (for example, soil versus water-based processes, those dominated by vegetation type, aerobic vs. anaerobic processes, chemical vs. biological source/sink). However, a number of trace gases may have the same or similar major functional types because of similarities in the factors controlling the production and consumption of those gases. An example is those trace gases that are regulated by the major physical soil variables (e.g. temperature and soil water). Identification of functional types which apply to more than one trace gas or ~ource/sink category of a single trace gas can make progress in simulating trace gas fluxes using functional types much more efficient. Given the degree of information needed to move towards a functional type approach for estimating trace gas fluxes across regional or global scales, it may be that expanded opportunities for not only measurement and modelling, but mapping and monitoring from remotely sensed data, can provide useful data and information. Measurements, maps and models at many scales are required to simulate trace gas fluxes at the global scale and
How can we best define functional types and integrate state variables and properties in time and space ?
157
monitoring is required if we are to predict changes in fluxes due to human and natural changes. These dynamic activities are designed to generate the necessary criteria for defining the functional types. It is anticipated that the information generated will feed back into prioritization of sampling locations and experimental designs to facilitate more accurate assessments of trace gas fluxes at global a., well as smaller scales.
3.3. Steps towards development of a functional type approach The working group explored the use of the functional type concept, as defined above, to estimate trace gas fluxes. To accomplish this the working group formulated a series of specific questions to guide the development of a functional type approach. These questions included: (i) What are the major variables and parameters controlling the magnitude of the flux of a particular trace gas? (ii) Can we use these major variables and parameters to define functional types for a particular trace gas? or put simply - Can the approach provide a better estimate of trace gas flux and reduce the complexity of the scaling process? (iii) How can the observable parameters and variables best be used to identify the defined functional type? (iv) What observing technologies or data sets are currently available, or could be available in the near future, to measure, map, model, and monitor the parameters used in the functional type approach? (v) Can we define functional types by a limited set of variables in a way that is useful for groups of trace gases, not just for a single gas? (vi) What is the role of remote sensing in applying the functional type concept? and, (vii) Can we use a functional type approach to help set priorities for trace gas research programs?
4. Application of functional type approach 4.1. Methane as an example The working group explored the process of developing a functional type approach for the upscaling of CH4 emission and deposition (oxidation) fluxes globally as an example to show how a functional type approach may be applied to replace the use of look-up tables and fixed emission factors. While only data readily at hand during the workshop were used for this exercise, it proved very useful in refining the concept and some of its current limitations. The results are presented below. 4.1.1. Current a p p r o a c h to methane emission estimation
There is a wide range of methane sources, including biogenic and non-biogenic sources, both natural and anthropogenic. A number of global budgets of methane have been constructed from both bottom-up and top-down approaches with varying degrees of spatial resolution for anthropogenic and natural sources (Fung et al. 1991; The and Beck 1995; Lelieveld and Crutzen, 1993; Olivier et al., 1996; Hein et al., 1997, reviewed in Van Amstel et al., 1997). A variety of approaches have been used in those analyses to estimate sources, however, in general, constant emission factors or average rates for a source have been used in combination with, for example, estimated areal extent or number of animals of each source (see Bouwman et al., 1999). In general, a mechanistic approach has not been used to estimate source strength. For example, natural sources of methane are primarily from wetlands. Estimates of the global methane emissions from natural wetlands range from approximately 90 to 237 Tg per
O~
Table 1. Approach for estimating C H (First approximation). Land cover/land use
and sinks: land cover/use, magnitude and certainty of source/sink, data needs and availability, and sensitivity of source.
4 sources
Magnitude a (1012g CH4y 1)
Uncertaintyb (1-4)
Data needed
Major Variables
Approach
Area; net primary production; temperature; flooding regime Area; Fertilizer/organic amendments, net primary production Water management Area Water table Time course Surface temperature Net primary production Nutrient loading Threshold for C H 4 production Population Type of treatment Sort of waste
Remote sensing
Sensitivityc of source to:
Availability
Scale/ Resolution
(1-4) Human Impacts
(1-4) Climate Change
Aquatic sources/wetlands
Tropical wetlands
70-190
3 m, s
Rice paddies
50-110
2 m
High Latitude wetlands
Estuarine/nearshore
20-160
3 m, s
?
3-4 m, o
Human sewage
30-60
3-4 o, s, m
Enteric fermentation
65- 100
2
Remote sensing Country data
Country data Remote sensing Met. Modeling Met. Modeling Climate/weather data. Remote sensing Watershed land use 9 Country and subnational
1 km
1
2
1 km
1
2
1 km
1
1
Watersheds
1
1
Yes Little info
Administrative units
1
3-4
FAO + various sources
Region
1
3
FAO IRRI (Hueke and Hueck 1997); Matthews et al. ( 1991 ) Matthews et al. ( 1991 )
various
A n i m a l s / l i v e s t o c k production O, S
Number of cattle, type; management data (feeding, age, milk production weight, etc.)
Country/subnational statistics
Table 1. Continued.
Animal wastes
20-30
Wild animals
2-6
Ranching
9
2 o, s 3-4 s, o, (m) 4 s, o
Source, type; waste management data Number of animals by category; feed, weight, age, etc. Number of sheep, cattle, goats; feeding, weight, age
Country/subnational statistics
FAO + various sources
Region
3
Regional inventories
Various sources, e.g. WNF
Ecosystems
3
National inventories; remote sensing
Region/ ecosystems
2-3
,,.,.
Other sources Landfills Biomass burning
15-75 30-100
2-3 o, s 2-3 m, o
Location Type, age, composition Carbon loading; Vegetation type; Fire frequency; Weather etc.
Country, Subnational Field data Remote sensing
IPCC
Country
1
4
Sporadic data IGBP-DIS
Region/ ecosystem
1
I-2
1:5M 1:5M 1 : 10M
1-2
o~ r~
,..,.
Meteo. Data Sinks
Oxidized soils
0-40
2-3 s, m, o
Soil type (chemical and physical properties), N status; land use; land degradation status
FAO (1995) Batjes (1997) Oldeman et al.( 1991 )
3 ,...,.
a Range of estimates from Fung et al. (1991), Hein et al. (1997), The and Beck (1995), Lelieveld and Crutzen (1993), IPCC (1994), Olivier et al. (1996). b Uncertainty scale: 1, well known; 2, moderately well known; 3, highly uncertain; 4, unknown; Additional codes indicate the source of uncertainty, i.e. m, measurements; s, spatial distribution; o, other info (e.g. management data on diets, composition of herd). c Sensitivity scale: 1, highly sensitive; 2, moderately sensitive; 3, minimally sensitive; 4, not sensitive. o~
2 o~
160
s.P. Seitzinger et al.
year (reviewed in Van Amstel et al., 1997). The approaches adopted to estimate the large areal source strength of both the high latitude and the tropical wetlands have primarily been based on emission factors or average rates for wetlands and do not use existing knowledge of the response of methane emissions to soil temperature or characteristics of the wetland area. Identification of important parameters and functional relationships between environmental variables and methane flux are required in order to provide more accurate and spatially explicit information on current methane fluxes, and to estimate past and future changes in methane fluxes as a function of global change scenarios. 4.1.2. Results o f a first attempt to use the functional type approach for methane
The results of the attempt to develop a fur~tional type approach for methav~ are summarized in Table 1. First of all the source/sink categories as traditionally identified (e.g. rice paddies, tropical and high latitude natural wetlands) were listed along with the approximate magnitude of the source strength from various studies based on constant emission factors, average rates, or look-up tables. An estimate of the degree of uncertainty in each of those source/sink strengths was assigned. The next step was to identify the data needed to develop a functional type approach for each of those source/sink categories. The major variables controlling the emission/deposition of methane from each source/sink were listed. Theoretically these variables could be used to develop functional relationships to estimate the magnitude of the flux for a geo-referenced site where the value of those variables is known. The best approach currently available to obtain geo-referenced data for each variable was added to the list (e.g. remote sensing, meteorological modelling, ground based regional inventories). Existing data bases for each of those variables were noted as was the resolution or scale of those data sets. Finally, a ranking was assigned to each source indicating the relative sensitivity of that source to human impacts or climate change. This ranking in combination with existing estimates of source/sink strength was considered to be important for identifying sources that should receive high priority for further measurements, models, mapping, and monitoring. A typical example of needed information is emission from estuarine environments: they are highly sensitive to both human impact and climate change, they depend on numerous variables but no order of magnitude is so far available. For some of the elements of the budget, it is clear that this approach is potentially very valuable. Taking the case of high latitude wetlands as an example, the primary driving variables for emission fluxes were thought to be well established (including surface temperature, net primary production and water table) and the general circulation models are able to provide appropriate fields for some of these over the entire regions and with temporal resolution sufficient to simulate the seasonal changes. In principle the key land use information necessary to identify the areas with a water table at the surface is possible using remote sensing methods. It therefore may be possible to develop the elements of a mechanistic approach to provide the spatially and temporally disaggregated emissions of methane from the high northern wetlands, as expressed by the following relationship: Time varying emission per gridcell = f(time varying: temperature, water table, .. in gridcell)
(1)
The same approach can also be adopted for the deposition flux of methane to oxidizing soils. In this case the approach is simpler than that for methane emission because the absolute range of the flux is much smaller and over unmanaged soils is controlled by a much reduced range of variables.
How can we best define functional types and integrate state variables and properties in time and space ?
161
Table 2. Processes and major regulating factors for production, consumption and exchange of CO2, CH4, N20, NO and NH 3
(first approximation). Gas Species
Processesof formation / consumption
Major regulating factors
C02
Decomposition Organic matter accumulation Calcification Ventilation / exchange Upwelling
Net primary production (integrated over depth for marine systems) N inputs Temperature Organic matter input from outside the ecosystem C/N ratio Surface concentration Wind speed, sea state Light penetration Plankton species
CH4
Methanogenesis Methane oxidation
Net primary production Redox potential, pH Inundation Soil drainage Precipitation Temperature Soil type
N20/NO
Decomposition Ammonification Denitrification Nitrification
Net primary production Redox potential, pH Soil drainage Precipitation Temperature C/N ratio Anthropogenic N input
NH3
Decomposition Ammonification Volatilization Canopy reabsorption
Net primary production Temperature Precipitation pH Wind speed Anthropogenic N input N type Total N cycling in the system
4.2. The functional approach as applied to a group of trace gases
After having made an inventory of possible ways to improve the current estimates for CH4 fluxes from ecosystems by using the functional type approach, the use of the functional type approach was also explored for a range of gases treated as a group. The gases included in the group were CO2, CH4, N 2 0 , NO and NH3. The processes that are responsible for fluxes of these gases were listed along with the major process regulators (Table 2) and the data sources (Table 3). Although the inventory was intended to cover both the terrestrial and aquatic ecosystems, its focus is primarily on the terrestrial systems. The regulating factors given are those that could be used at the global scale. At a smaller level of scale more detailed information is required to describe process regulators. Therefore, the list is a minimum set of parameters needed. These process regulators are the basic elements of the functional types at the selected global scale. Net primary production (NPP) is controlling the production of most of the gas species listed, as all the processes responsible for their formation are microbially mediated. Other physical regulators, such as temperature, control the transformation at many different levels in the soil system. Most of the processes, such as decomposition, methanogenesis, methane oxidation, organic matter accumulation, nitrification and denitrification, occur both in terrestrial and aqua-
162
S.P. Seitzinger et al.
Table 3. Major regulating factors from Table 2 and their data sources. Regulating factor
Source
Net primary production Surface temperature Precipitation Inundation C/'N pH, soil drainage N content
Remote sensing Remote sensing Remote sensing, GCM, databases Remote sensing Models, observations Soil databases Deposition data, deposition models Agricultural statistics Human population statistics Meteorological data
Wind speed
tic environments, and the required set of input variables is comparable for the description of the processes. The exchange of NH3 between canopies and the atmosphere and the process of calcification are not microbially mediated, and they require different sets of input variables. A major difference between terrestrial and aquatic systems is that in the oceans production and exchange are not necessarily occurring az the same time and the same place. Therefore, the emission of gases to the atmosphere also requires information on the concentrations in the surface water and parameters needed to describe the exchange. It is clear that for the description of functional types for the group of gas species used here there is considerable overlap in the regulators. Hence, the examples demonstrate that it is possible to compile common lists of parameters for groups of gas species, and that the definition of functional types for such groups is also feasible. However, the requirements for the description of functional types for species other than those listed here should be analyzed to see if this commonality also occurs for other species including those for which the processes are not completely biologically mediated.
5. Tools Identification of processes and major variables controlling the flux of a trace gas or set of gases, followed by development of equations to estimate the flux based on those variables, must go hand in hand with identification and development of geo-spatially referenced data bases of those variables. The resultant regional or global source and sink fluxes should then be integrated into CTMs, with refinements on both ends to resolve differences. 5.1. Global databases and geographic information systems Today a variety of organizations and institutions are working towards the creation of global scale information on key environmental parameters. Work is moving forward to produce digital cartographic products at scales of 1:1,000,000 (-~1:1 km). For example maps of land cover, hypsography, surface drainage, soils, population, transportation, political administrative boundaries, and transportation networks are all either in the process of production and/or update. Most of these products are being developed, in part, through the processing and analysis of remotely sensed data (Estes and Loveland, 1999). All of these products will be available in formats and data structures which will facilitate their use by researchers employing geographic information systems (GIS) technologies. GIS's are increasingly being employed to organize, integrate and process spatial data concerning a wide
How can we best define functional types and integrate state variables and properties in time and space ?
163
variety of environmental features and processes. GIS's are also increasingly being employed to support a variety of modelling activities, whether there is a loose coupling, a tight coupling or the model is written employing the analytical capability of the GIS (embedded coupling). GISs provide a powerful tool for increasing our understanding of issues related to the scaling of trace gas fluxes. Some useful references to "available" global data bases include: Van Woerden et al. (1995); ICSU World Data Centers (e.g. ICSU, 1996. Guide to the World Data Center System: Geophysical, Solar and Environmental); UNEP-GRID (http://www. grid.unep.ch/mdb_datasets.html); International Council of Scientific Unions, Paris (http:// www.ngdc.noaa.gov/wdcmain.html); and IGBP-DIS (http://www.cnrm.meteo.fr:8000/igbp/).
5.2. R e m o t e sensing of surfaces and the a t m o s p h e r e
The range of remote sensing devices and techniques and the number of earth observation platforms is in a continuous phase of growth. Yet, as capable as these systems are at creating data, the question we must ask is are they capable of providing the data needed to model trace gas fluxes between the land, atmosphere and the oceans? The answer to this question is complex. In some cases sensors can provide direct measurements of important state variables and properties. In other instances these systems can be used to identify surrogate data from which information concerning state variables and properties may be derived. In both instances ancillary data is typically employed to add information to the remotely sensed data to improve the accuracy of the labelling of features relevant to given studies. However, potential users of remotely sensed data should be aware that the use of remote sensing data is not straight forward in and of itself. Once acquired considerable technological skill and scientific understanding is required to effectively process these data into the information a given trace gas modeller may need. Its access and the know-how associated with its processing and analysis all complicate matters for the non-specialist facing a specialized field of growing complexity. The working group believes that three approaches can be used to improve upon the current use of remote sensing methods and techniques to better assess the fluxes of trace gases between the surface and the atmosphere. Two of them look at features, variables, or processes representing sources and sinks (land cover and regulating factors of trace gas fluxes, respectively), and the third approach focuses on atmospheric transport: - L a n d cover and land use. The predominant approach employed in the past to use remotely sensed data to estimate trace gas fluxes has been to use the more familiar land cover or land use classes in combination with an emission factor for each relevant land cover/use class. For example, wetlands or a few subclasses of that category would represent a traditional "functional type" for methane production; changes in forest cover (i.e. from deforestation to the regrowth of a permanent or seasonal cover) provide the base for CO2 release or sequestration calculations. This empirical approach is based on the assumption that the relationships which have been experimentally determined between the functioning of an ecosystem and trace gas fluxes hold true over the whole domain of distribution of that ecosystem (for example all wetlands in a climatic zone) or for any occurrence of the process under study (i.e. all conversion of forest type A into a land cover B). Obviously, improvements can be made here simply by improving the identification and mapping of the seasonal or longer term variability of land cover or land use classes. Of particular interest here are those classes having a close relationship with changes in trace gas fluxes. Measuring changes in land cover/use globally may not always be easily conducted by comparing global maps obtained at two different periods of time. The level of detail required for identifying and measuring change with a good level of confidence must be based on the
164
S.P. Seitzinger et al.
application of remote sensing techniques at a much finer resolution; sampling techniques can also be guided by a stratification based upon a pre-assessment of the most significant changes in terms of trace gas fluxes. Much international effort is currently devoted to that task; yet while global assessments of land cover are now well underway the ability to globally identify and measure changes in the land cover is still not sufficiently developed. Future earth observation satellite products will be dedicated to that task. For example the Earth Observing System (EOS) MODIS team is preparing to provide landcover products at regular intervals. Regulating factors of trace gas fluxes. The application of remote sensing techniques to obtain better direct measurement of those targeted variables and parameters that control the flux of trace gases is entirely driven by the requirements expressed in the functional type analysis (section 3). To illustrate this point we use the example of methane; the areas of the world where anaerobic conditions are leading to the formation of methane could be actively sought using a range of remote sensing tools adapted to that particular problem. A complete data set might provide separate maps or attributes for water logged areas, a spatial assessment of water regimes, biomass distribution, temperature in temperate areas, field management practices for rice paddies, etc. (Table 1). The same observing system would look at seasonal changes and interannual variability in those parameters. Such information can only be extracted from an ad-hoc set-up which would combine remote sensing data from systems such as SPOT, Thematic Mapper, IRS, ERS-1 and 2, and RADARSAT, meteorological models and GIS facilities. With respect to carbon balance, a dedicated remote sensing effort could target areas which contribute most to sequestration (i.e., forest regrowth areas on land) or release (i.e., deforestation, burned areas). Long time series of satellite data could also try to identify whether relative changes have occurred in some basic processes such as photosynthetic activity (through trend analysis or changes in seasonal amplitude of a "vegetation index"). AVHRR data and data from the MODIS instrument, when it becomes available, also could be used here. Biomass burning which is considered as an important emission process with a significant impact on atmospheric chemistry can also represent, as it does today, the focus of dedicated remote sensing experiments using systems such as AVHRR and ATSR. Such experiments would address questions associated with the occurrence of fire, the combustion process, and the estimation of burnt biomass. It is important to add that the coupling of remotely sensed data for control variables with trace gas fluxes requires "ground truthing" including ground-based measurements of the trace gas fluxes in each functional type. For comparisons of ground-based and remote sensing measurements, both measurements have to be made at the (same?) level of spatial and temporal resolution. For accurate assessment of trace gas flux, we need to employ remotelybased sensors with a field of view (footprint) closely similar to the extent of spatial distribution of the sources of the gas species in question. All of the above require groundbased measurement methods that are coupled with sensitive instrumentation. Remote sensing techniques must be fine-tuned in terms of spectral bands, resolution, frequency of data acquisition and data processing/analysis techniques to meet the requirements expressed by the functional type analysis. A certain dose of trial and error and of technological opportunism (=using remote sensing tools not designed for the task) will still exist for several years to come. Yet, it can be expected that the selective and systematic application of a range of remote sensing techniques to a well defined set of issues will significantly improve global assessment of trace gas fluxes. Such an approach can lead to a better definition of earth observation systems of the future which would be better suited for the task of improving our understanding of the scaling of trace gas fluxes. -Atmospheric concentrations. One of the recent developments is the remote observation of the integrated atmospheric gas concentration over a vertical profile. Combining the -
How can we best define functional types and integrate state variables and properties in time and space
?
165
estimated global distribution of trace gas fluxes derived from the functional type approach (Section 3) with measurements of the distribution of gas concentrations in the atmospheric profile is an important step in assessing the accuracy of the flux estimates as well as our understanding of atmospheric behaviour of trace gases. Atmospheric behaviour is simulated by chemical and transport numerical models (CTMs) which parameterize the relevant processes occurring within the atmosphere. One of the most important methods to validate a particular CTM and GCM is comparison with atmospheric measurements. The amount of gas at any location at a given time represents a stationary state between production and loss. Changes in either the production or the loss process result in changes in the observed stationary state. For long lived tropospheric species (e.g. CO2, N20 and CH4) a reasonably accurate picture of the global distribution results from a set of appropriately located ground based stations. Regions of relatively high and low abundance can be identified, enabling hot spots and low spots to be observed, provided the global distribution of such species can be determined at a sufficiently high precision from space instrumentation. For species having relatively short tropospheric lifetimes and or high variability (e.g. 03, CO, NO2, H20, VOC and aerosols), the only feasible method to obtain their global tropospheric distribution is from the employment of the appropriate instrumentation on board space based platforms. The number of species which may be retrieved from remote sensing measurements is much more limited in the troposphere than in the stratosphere. This is because unlike the stratosphere and mesosphere, the troposphere contains large amounts of water vapor and high pressure. This restricts significantly the use of mid-infrared, far infrared and microwave spectroscopy. Another complication for remote sensing in the troposphere is the presence of clouds, which exhibit large variability. Finally retrieval of the tropospheric amounts of species such as 03 and NO2, having significant stratospheric abundance, requires additional information about their stratospheric distributions. The above reasons make the remote sensing of trace constituents in the troposphere from space based platforms intrinsically more scientifically and technically challenging than in the stratosphere or mesosphere. However the development of new sensors and the availability of fast computers is now enabling remote sensing techniques to be successfully applied to this task. In the past instruments such as TOMS, SBUV, SAGE and ATMOS and those on the UARS satellite were developed by NASA with the primary objective of measuring upper atmospheric constituents. Similarly AVHRR is primarily a surface mapper. However, it has been shown that the retrieval of tropospheric 03 amounts and aerosol from the inversion of, respectively, TOMS and AVHRR measurements is successful. Similarly the SAGE vertical profile data products (e.g. 03, aerosol) extend from the tropopause down to the cloud top. ATMOS a shuttle borne solar occultation experiment utilizing a mid-infrared Michelson Interferometer has successfully measured a series of gases in the upper troposphere. Future instruments (e.g. GOME-2, MOPITT and TES) will address the identification in the column of more species with a footprint which is suited for modelling.
6. Conclusions and recommendations The functional type approach developed in this chapter offers opportunities for improving the estimation of trace gas fluxes. The proposed method combines: (i) functional relationships between environmental variables and trace gas fluxes with, (ii) spatially explicit information on the global distribution of controlling variables across the temporal and spatial scales of
166
s.P. Seitzinger et al.
interest. Information used to identify the controlling variables and develop the functional relationships can be obtained from detailed process based studies and modelling efforts. Information on the spatial distribution of the controlling variables is obtained from various sources including both ground based and remote sensing measurements. There is now a need for further investigations of the applicability of the functional type approach to a range of trace gases, processes and ecosystems. The method should be refined through testing in a wide range of conditions. It is also important to identify common grounds between the functional type method and more classical approaches such as those consisting of interpreting land cover and land use classes in terms of emission rates. Remote sensing approaches to aquatic and terrestrial surface monitoring will be a main source of data at global and regional scales. Those tools need to be optimized with respect to the objectives of trace gas assessments. Compromises will have to be explicitly made between what is desired and what is feasible. The functional type approach also helps in systematically exploring the requirements for data. Requirements for measurement can now be expressed in very specific terms. These requirements should be properly channeled to programmes leading to the definition, development, launch and operation of space observing instruments and platforms of the future.
References Arrhenius, S. (1896) On the influence of carbonic acid in the air upon the temperature of the ground. The Philosophical Magazine, Series 5, volume 41, No. 251 (Reprinted in Tisglow 1992 3:3-26). Batjes, N.H. (1997) A world data set of derived soil properties by FAO-UNESCO soil unit for global modelling. Soil Use and Management 13:9-16. Bouwman, A.F., R.G. Derwent and F.J. Dentener (1999) Towards reliable bottom-up estimates of temporal and spatial patterns of emissions of trace gases and aerosols from land-use related and natural sources. In: A.F. Bouwman (Ed.) Approaches to scaling of trace gas fluxes in ecosystems. Developments in Atmospheric Science 24. Elsevier, Amsterdam, pp. 1-26. Burrows, J.P. (1999) Current and future passive remote sensing techniques used to determine atmospheric constituents. In: A.F. Bouwman (Ed.) Approaches to scaling of trace gas fluxes in ecosystems. Developments in Atmospheric Science 24. Elsevier, Amsterdam, pp. 313-347. Estes, J.E. and T.R. Loveland (1999) Toward the use of remote sensing and other data to delineate functional types in terrestrial and aquatic systems. In: A.F. Bouwman (Ed.)Approaches to scaling of trace gasfluxes in ecosystems. Developments in Atmospheric Science 24. Elsevier, Amsterdam, pp. 123-150. FAO (1995) Digital Soil Map of the World and Derived Soil Properties (version 3.5). CD-ROM, Food and Agriculture Organization of the United Nations, Rome. Fung, I., J. Jones, J. Lerner, E. Matthews, M. Prather, L.P. Steele, and P.J. Fraser (1991) Threedimensional model synthesis of the global methane cycle. Journal of Geophysical Research 96:13011-3065. Hein, R., P.J. Crutzen, and M. Heimann (1997) An inverse modeling approach to investigate the global atmospheric methane cycle. Global Biogeochemical Cycles 11:43-76. Huke, R.E. and E.H. Huke (1997) Rice area by type of culture: South, Southeast, and East Asia. A revised and updated database. International Rice Research Institute, Manilla, Philippines. IPCC (1995) IPCC Guidelines for National Greenhouse Gas Inventories. Intergovernmental Panel on Climate Change/Organization for Economic Cooperation and Development, OECD/OCDE, Paris. IPCC (1997) Guidelines for National Greenhouse Gas Inventories. Intergovernmental Panel on Climate Change/Organization for Economic Cooperation and Development, OECD/OCDE, Paris.
How can we best definefunctional types and integrate state variables and properties in time and space ?
167
Lelieveld, J. and P.J. Crutzen (1993) Methane emissions into the atmosphere, an overview. Proceedings of the Interatnional IPCC workshop on methane and nitrouse oxide, Amersfoort, The Netherlands, Report 481507-003, National Institute of Public Health and the Environment, Bilthoven, the Netherlands. Matthews, E., I. Fung and J. Lerner (1991) Methane emission from rice cultivation: Geographic and seasonal distribution of cultivated areas and emissions. Global Biogeochemical Cycles 8:3-24. Oldeman, L.R., R.T.A. Hakkeling and W.G. Sombroek (1991) Worm Map of the Status of Human-induced Land Degradation." An Explanatory Note (2nd revised edition). ISRIC, Wageningen and UNEP, Nairobi. Olivier, J.G.J., A.F. Bouwman, C.W.M. van der Maas, J.J.M. Berdowski, C. Veldt, J.P.J. Bloos, A.J.H. Visschedijk, P.Y.J. Zandveld and J.L. Haverlag (1996) Description of EDGAR 2.0: A set of global emission inventories of greenhouse gases and ozone-depleting substances for all anthropogenic and most natural sources on a per country basis and on a 1 x 1 degree grid. Report 771060-002, National Institute of Public Health and the Environment, Bilthoven, The Netherlands. The, T.H.P. and J.P. Beck (1995) Scenario studies on effects of methane emissions using a 3-D tropospheric model. Report 722501-003, National Institute of Public Health and the Environment, Bilthoven, The Netherlands. Van Amstel, A.R., C. Kroeze, L.H.J.M. Janssen, J.G.J. Olivier, and J.T. van der Wal (1997) Greenhouse gas emission accounting: Pr:'.iminary study as input to a joint !nternational IPCC Expert Meeting/CKO-CCB Workshop on comparison of top-down versus bottom-up emission estimates. WIMEK/RIFM report 728001-002, National Institute of Public Health and the Environment, Bilthoven, the Netherlands. Van Woerden, J.W., J. Diederiks and K. Klein Goldewijk (1995) Data management in support of integrated environmental assessment and modelling at RIVM (Including the 1995 RIVM catalogue of international data sets). Report No. 402001006, National Institute of Public Health and the Environment, Bilthoven, the Netherlands.
This Page Intentionally Left Blank
Chapter 8
M O D E L L I N G CARBON DIOXIDE IN THE OCEAN: A REVIEW
D. Archer
This Page Intentionally Left Blank
Approaches to scaling a trace gas fluxes in ecosystems A.F. Bouwman, editor 9 Elsevier Science B.V. All rights reserved
MODELLING CARBON DIOXIDE IN THE OCEAN: A REVIEW
D. Archer Department of the Geophysical Sciences, University of Chicago, Chicago, IL 60637, USA
1. I n t r o d u c t i o n
Modelling in the natural sciences plays two roles depending on the maturity of our understanding of the problem. In an initial stage, a conceptual or mathematical model is used to try to differentiate between competing hypotheses by comparing predictions of the model with measured data. Because the aim is to evaluate the natural system by comparison with model predictions, we can call this a "diagnostic" endeavor. An example of this might be the stagnant film versus the boundary layer models for kinetics of gas exchange across an air/water interface, where the controlling mechanisms for gas evaporation from solution are still uncertain. As our understanding of nature progresses, the more mature "prognostic" stage uses an established and successful model to extrapolate data to wider spatial or temporal coverage than is practical to measure in the field. Uptake of fossil fuel carbon dioxide (CO2) by dissolution in the oceans is an example of this stage of scientific development. This review presents an overview of the science of modelling CO2 fluxes in aquatic systems, primarily in the oceans, discussed within the framework of these two stages of scientific development. First the nearly instantaneous molecular scale thermodynamic calculations of CO2 aqueous chemistry will be discussed. Subsequently the millimeter scale processes which govern gas transfer kinetics across the air-sea interface, the 100 m scale dynamics of mixing and biogeochemistry in the upper ocean, and the mesoscale eddies and gyre circulation will be discussed. The paper concludes with global spatial scale 100,000 year time scale models for CO2 in the atmosphere and oceans (Figure 1). An underlying motivation for these topics is to understand and predict fossil fuel uptake into the oceans at present and in the coming centuries. While some parts of the discussion are applicable to any gas and any aquatic system, for other parts of this review the focus on CO2 in the oceans becomes more transparent.
2. M o l e c u l a r s c a l e
On a molecular level, the partitioning of gases between gaseous and aqueous phases is governed by thermodynamics in the form of Henry's law, which states that in ideal (dilute) systems the partial pressure in the gas phase will be proportional to the concentration of the aqueous species in solution. The value of the Henry's law constant for a given gas depends on temperature and salinity through decreasing solubility of CO2 at higher temperature and salinity, respectively. In addition, for dissolved CO2, there is aqueous phase pH equilibrium chemistry between CO2 (aqueous), H2CO3 (carbonic acid), HCO3- (bicarbonate), and CO32(carbonate ion). Partitioning between CO2 and H2CO3 is difficult to detect analytically, so the two species are typically treated as a single species, called H2CO3", which is taken to be in
172
D. Archer
equilibrium both with gaseous CO2 and dissolved HCO3-. The gas phase equilibrium condition for a water sample is often referred to as the pCO2 of the solution; note that in disequilibrium the pCO2 of the solution differs from the pCO2 of the gas phase. The solubility of CO2 gas, and the equilibrium dissociation constants for H2CO3"vary as a function of temperature and salinity, such that a 1~ increase in sea water temperature leads overall to a 4.23% increase in the equilibrium gas phase partial pressure (pCO2) of a sea water sample (Chipman et al., 1992), depending on the choice of thermodynamic constants, while a 1 ppt increase in salinity at 25~ generates a 3.2% increase in pCO2. The pH equilibrium model has been first order successful in predicting the behaviour of dissolved CO2 in sea water, while successive studies have improved the consistency of the dissociation (Roy et al., 1993) and solubility (Weiss, 1974) constants for sea water. The carbonate buffer system has two degrees of freedom, so that simultaneous measurements of three or more parameters can be used to test the thermodynamic constants in the field; this also appears to work fairly well (Millero et al., 1993). The pCO2 of a water sample can typically be directly measured much more precisely (Chipman et al., 1992) than it can be calculated from other measurements such as the total dissolved CO2 concentration (Chipman et al., 1992), spectrophotometric pH (Clayton and Byrne, 1993), or the titration alkalinity (Bradshaw et al., 1981). However pH equilibrium calculations are universally used in numerical models of upper ocean carbon cycling, and they appear to be sufficient for that purpose, because the CO2 flux is in general limited by circulation rather than by gas exchange, which would be sensitive to the sea surface disequilibrium (Sarmiento et al., 1992).
3. Microscale On a spatial scale of millimeters to a meter at the sea surface, the flux of a gas across the airsea boundary is governed by water and air motions near the surface, and the rate limiting mechanism is usually in the water. Excellent and comprehensive reviews of this topic can be found in Liss and Duce (1997). The conceptual picture begins with a well mixed large reservoir of sea water below the sea surface, and a well mixed atmospheric reservoir in the boundary layer above. Chemical fluxes across the sea surface alter the concentrations of solutes at the sea surface, which generate a chemical boundary layer in the water. Continued air-sea flux is limited by replenishment of this sea surface microlayer, either by molecular diffusion or by fluid flow renewing the sea surface microlayer with bulk fluid from the sea surface mixed layer. The sea surface microlayer is also thought to be cooler than the bulk sea surface mixed layer, because of evaporation, while the air immediately above the sea surface is thought to be more humid than the bulk atmospheric boundary layer. The air-sea flux is driven by these altered microlayer characteristics, rather than by the conditions of the bulk reservoirs above and below the microlayer which can be more easily measured. The sea surface microlayer is distinguished by enrichment in organic matter, and by distinct planktonic communities (the neuston). The limitation of the gas transfer by stagnation at the sea surface has been described and parameterized by two mechanistically distinct but functionally nearly indistinguishable models, the boundary layer model and the surface renewal model (Figure 2). In the boundary layer model, the transition from turbulent transport to diffusive transport near the sea surface is conceived to occur at a transition depth below the sea surface, demarking a stagnant layer typically 10 - 100 pm thick. Gas exchange rates increase with increasing wind speeds, and this behaviour is parameterized as a decreasing thickness of the boundary layer in more turbulent conditions. The surface renewal model imagines that the stagnant layer is replaced with bulk
173
Modelling C02 in the ocean: A review
Molecular Scale
PCO-z<->C02(aq) COz(aq)+ H20<->H-zCO3
K0
H2003"<-> HCOs-. H+
K, ~
H-zCOa+C02(aq)= t-~CO.a"
HCOa"<->CO~+ H+
I lOO#m ~
COz
K2
~
~GasExchange BoundaryLayerIk~
1OOm
[
Wind ~
}
~, / r
Convection l U g h ,
Bubbles
~
J Global, IO00 years
L __ ._o.
f-C02~
100 km
NO3+ P04
[
[
'f
.i ":
v "
~
"l./,,.. :
I\ f
z ~~
o0,
"
sat,n
Burial
Figure 1. An overview of the fundamental processes controlling the partitioning of CO2 in the oceans and exchange with the atmosphere on a wide range of space and time scales.
sea water with some frequency which increases with increasing water turbulence. One potential means of differentiating the two models is in the dependence of the gas fluxes on the diffusion coefficient of the gas; the surface renewal model predicts a square root dependence of flux on D (exponent 1/2), while the boundary layer models give an exponent of 2/3 when the sea surface is treated as a smooth rigid wall (Deacon, 1977) or 1/2 for a free surface (Ledwell, 1984), the difference arising from the effect of the wall on the turbulent flow. In lab experiments, the observed diffusion dependence decreases from 2/3 order when the surface is calm (low mean square wave slope) to 1/2 order when the wave slope increases (J~ihne, 1985), suggesting that the boundary layer model might be appropriate at low wind speeds, but leaving the question open at high wind speeds. The relationship between exchange rates of different gases is important not only to distinguish between these models, but also because
174
D.
Archer
several gas exchange measurement techniques, such as the radon method (Smethie et al., 1985), or the simultaneous deliberate injection of multiple tracers (Watson et al., 1991), depend on being able to relate the gas exchange rates of different gases. Films of organic molecules at the sea surface have been found to exert a strong influence on the rate of gas exchange (Frew, 1997). The model of Davies (1966) explains this as a result of surface tension exerted by the film, which tends to hold the sea surface molecules together, suppressing the smaller and slower turbdlent eddies near the sea surface. Increasing sea surface film concentration in laboratory experiments has been shown to decrease gas exchange velocity by establishing a critical wind speed, below which gas exchange is low and relatively insensitive to winds, and above which gas exchange increases with wind speed, but never to the value which would be expected under clean conditions. The critical wind speed is found to coincide with the onset of capillary waves at the sea surface (J~.hne et al., 1987), and laboratory data using a variety of surface films and turbulence conditions collapses to a single line when plotted against mean square wave slope (Frew, 1997). Another poorly constrained physical process in gas exchange is the effect of bubbles. Here the primary distinction to be drawn is between bubbles that dissolve completely, delivering the original load of gases entirely to solution, and bubbles which survive to rise to the surface again, fractionating the gases according to gas solubility and the kinetics of gas exchange. The effect of bubbles on the gas exchange model is to (i) increase the rate constant for exchange, by increasing the effective area of the sea surface, and (ii) increase the average pressure of the atmosphere that the ocean sees, because gases dissolve at greater pressure (depth) than atmospheric (Memery and Merlivat, 1985). Bubble injection of gases can lead to a steady state supersaturation of the gas at the sea surface, with bubble invasion balanced by surface evasion. The effect of bubbles is greatest for relatively insoluble gases such as oxygen, and less for the more soluble CO2 (the high solubility can be attributed in part to the aqueous form H2CO3), because the concentration of the more soluble gas equilibrates more quickly between the bubble and the liquid phase. The dependence of bubble gas exchange on solubility as well as on diffusion coefficient complicates gas exchange measurements in the field which rely on relating the exchange rates of different gases, because they introduce solubility as a second variable. It is observed that only smaller bubbles (<150 ~tm) are typically mixed below about 1 meter depth (the pressure at which corresponds to a 10% atmospheric supersaturation at the sea surface). Deep bubbles are thought to be most important to the steady state supersaturation at the sea surface, while the more numerous and larger shallow bubbles dominate the bubbleinduced increase in the piston velocity (Woolf, 1997). Another potentially important effect is the temperature gradient at the sea surface, generated by cooling associated with evaporation and surface IR radiation. The deviation of sea surface CO2 solubility relative to the bulk temperature of the mixed layer has been estimated to contribute 0.7 Gton C yr ~ to the net atmosphere - ocean CO2 flux (Robertson and Watson, 1992). Global models of CO2 uptake are not particularly sensitive to this correction, but global estimates of air-sea CO2 fluxes based on measured sea surface mixed layer pCO2 values (Tans et al., 1990; Takahashi et al., 1997) clearly are. Finally, the reaction of CO2 with H20 to form H2CO3, which increases the solubility of CO2 gas and enables it to participate in the carbonate buffer system of sea water, may play a role in the kinetics of CO2 transfer. The kinetics of hydration are relatively slow (Johnson, 1982) but if significant concentrations of the enzyme carbonate anhydrase were present in the surface microlayer, the buffer chemistry of sea water might enhance the gas exchange flux by catalytically promoting the hydration of aqueous CO2 within the surface boundary layer (Wanninkhof, 1992). Gas exchange rates have been compiled from a variety of field and laboratory sources, using a variety of techniques. These are discussed in detail in another paper in this volume
Modelling C02 in the ocean: A review
175
Air Diffusive
Boundary Layer
Turbulent Mixing
Sea
Molecular Diffusion
Surface Film Renewal
Figure 2. Two competing models for describing the processes limiting the rate of gas transfer between surface waters and the atmosphere.
(Lapitan et al., 1999) and will only be summarized here. In general, there is considerable scatter in the data which limits the precision to which the present-day ocean uptake of CO2 can be estimated from field data (Takahashi et al., 1997). There may also be a systematic discrepancy between measurement of inert gases (e.g. radon, SF6, and He) and 14C (Liss and Merlivat, 1986; Wanninkhof, 1992). Models for gas exchange on a millimeter to meter scale are therefore still in the diagnostic stage, where research focuses on identifying the relevant mechanisms to encode into the models.
4. Mixed layer scale The next spatial scale to consider encompasses the top 10-250 meters of the water column, and a "local" horizontal scale of hundreds of meters to kilometers. On this scale the relevant dynamics are the establishment and destruction of thermal stratification which inhibits mixing of surface waters with subsurface waters below. Photosynthesis is confined to the top zone
176
D.
Archer
because only the top 75-150 meters receive enough sunlight to support net photosynthesis. The net effect of biological activity is to capture CO2 and dissolved nutrients like NO3 ~ and PO43- and transform them into particulate and dissolved organic matter (reduced carbon). Particulate matter can be removed from its parent water mass by sinking, which results in a net depletion of sea surface water in the nutrients and in CO2. In the steady state the downward nutrient flux in sinking particles is regulated by the supply of nutrients to the sea surface, although this is a complicated story. Seasonal excursions in the depth of mixing at the sea surface, with deep mixing in winter and stratification in summer, lead to a seasonal cycle of nutrient entrainment into the euphotic zone. In addition, the shear generated by wind-driven currents leads to turbulent mixing at the base of the well-defined mixed layer. The physics which describe the dynamics of the mixed layer are therefore thermal (free) convection, both as an adjustment to static stability and with the added possibility of further entrainment derived from the kinetic energy of convection; wind-driven (forced) convection, which appears to generate mixing through shear at the base of the mixed layer; turbulence and Langmuir circulation within the mixed layer; and turbulent mixing below the mixed layer. In many parts of the ocean these processes can be described as a one-dimensional vertical balance driven by surface heat, fresh water, and momentum fluxes. This observation has led to a long history of mixed layer models and their application to modelling of sea surface biological activity and CO2. These models have been extensively reviewed and compared elsewhere (Price et al., 1978; Martin, 1985; Martin, 1986; Large et al., 1994; Archer, 1995), so a few salient points will be summarized here. The models can be divided into several classes. Bulk mixed layer models are based on the assumption that the sea surface mixed layer is completely homogeneous, an assumption which is not bad but not strictly correct either. Often heterogeneity in momentum, turbulence, or even in some short lived chemical species can be observed in the sea surface "mixed layer", especially under conditions of deep convective mixing (Lal and Lee, 1988). Bulk mixed layer models either track the influx and dissipation of turbulent kinetic energy (Kraus and Turner, 1967; Gaspar et al., 1990) or they pay attention to current shear at the mixed layer base (Price et al., 1986). A second class of models does not predict a priori the degree of mixing in the mixed layer, but rather resolves the statistics of turbulence in the upper water column and allows the model to predict mixing (Mellor and Yamada, 1974; Large et al., 1994). The challenge here is the tendency to under predict mixed layer homogeneity due to the neglected two or three dimensional structure such as organized convection and Langmuir circulation. The models are in general reasonably able to reproduce the available pool of ship-based and time series mixed layer physical observations. Discrepancies still exist and discussion continues, but the models are ubiquitously deployed in global ocean circulation and biogeochemistry models. In general, then, we can conclude that the mixed layer physical models have largely reached the prognostic stage of scientific maturity. In contrast, several chemical issues remain mired in the diagnostic stages. A long-standing issue recently resolved (in my opinion) is the limitation of biological uptake of nutrients in some parts of the upper ocean. The observation is that vast tracts of the upper ocean, in particular the equatorial and North Pacific, and the Southern Ocean, are replete with nutrients and sunlight in surface waters, and yet the rate of biological export of nutrients as sinking particles is relatively low. The residence time of nitrate in surface waters of the equatorial Pacific, for example, is several years while in the North Atlantic the spring bloom consumes and exports the available surface nutrients in a matter of days. This observation has been shown fairly convincingly to be the result of limitation by iron (Martin and Fitzwater, 1988; Coale et al., 1996). Iron limits production ultimately because of its insolubility. Nitrate and phosphate are present in sea water in roughly the ratio of their utility to phytoplankton,
Modelling C02 in the ocean. A review
177
because they tend to travel together through the biological system. Iron, in contrast, is scavenged from sea water by precipitation or adsorption onto particles when its concentration approaches or exceeds its solubility. Therefore a significant or even dominant portion of the source of iron at the sea surface is from wind-blown continental dust. Another source is sediments in coastal areas. In surface waters particulate or dissolved Fe 3+ can be transformed into the more soluble but oxidatively unstable form Fe 2+ by reaction with UV light, leading to a diurnal signal in iron speciation (Johnson et al., 1994). The effect of iron on a planktonic ecosystem appears to be to promote or encourage the larger diatoms to bloom (Martin and Fitzwater, 1988; Coale et al., 1996). With size comes the ability to produce particles large enough to escape from the mixed layer by sinking. Therefore modelling the effect of iron on upper ocean carbon cycle requires simulating and predicting (i) atmospheric deposition of dust; (ii) dissolution of iron from the particles; (iii) upper ocean photochemistry of iron; (iv) scavenging and precipitation removal of iron; (v) biological sources and sinks; (vi) the global ocean distribution of dissolved iron, to capture the upwelling component; (vii) the ecosystem response to changes in the iron supply; and, (viii) the impact of the ecosystem response on the export of carbon from the mixed layer. Many of these tasks have not yet even begun, leading us to the clear conclusion that a fundamental portion of chemical oceanography is still very much in the diagnostic stage of scientific development. Another mysterious issue has been revealed at the JGOFS time-series stations in the subtropical oceans, concerning the cycling of nutrients in the upper ocean. In the steady state, the upward flux of nutrients by mixed layer entrainment, turbulent mixing, and mesoscale motions (next section) ought to account for the observed rate of carbon export as sinking particles (measured by a variety of techniques, such as sediment traps, or sea surface oxygen fluxes). In both the time series stations Hawaii (Karl and Lukas, 1996) and Bermuda (Michaels and Knap, 1996), this balance cannot be made, as upward nutrient fluxes are inadequate to balance observed carbon export production (Hayward, 1987; Jenkins, 1988). Possible mechanisms include in situ nitrogen fixation, which is clearly diagnosable from nitrate / phosphate data (Gruber and Sarmiento, 1997), atmospheric deposition (Owens et al., 1992), and biological transport of nutrients by active buoyancy regulation among plankton (Villareal et al., 1996). Given that the mechanisms to account for these observations are still unknown, models of upper ocean nutrient dynamics are clearly still in the diagnostic stage. Our ignorance in this area is a major impediment to predicting future ocean CO2 uptake in light of changes in global climate.
5. Mesoscale The next larger spatial scale to consider ranges from 10-200 kilometers in the horizontal, and to several kilometers in the vertical. This is called the mesoscale, and the physics of circulation at this scale is analogous to storms in the atmosphere. A crucial length scale for ocean flow is the Rossby radius of deformation, which represents the distance which information can be carried by internal gravity waves within the time scale of the rotation of the Earth. A spike perturbation in sea surface elevation and subsurface density structure, for example, would excite gravity waves and "run downhill" in all directions, but after a period of time determined by the local vertical rotation rate, the now smeared out hill of sea surface and subsurface anomaly would be confined by rotation and geostrophy. The spatial scale of gravity wave information flow, the Rossby radius of deformation, ranges from -100 kilometers in low latitudes to -10 km in higher latitudes, because of the variable projection of the Earth's rotation on local vertical. Density and flow structures generated by the mean flow are
178
D.
Archer
stretched and distorted by the flow, spawning ever smaller features, until the spatial scale of the Rossby radius is reached; these features are called mesoscale eddies. Features which are generated smaller than the Rossby radius quickly spread out by gravity waves, returning to the mesoscale size scale. Eddies can affect the internal cycling of the ocean by distorting the depth of the thermocline, bringing nutrient rich subsurface waters within the reach of sunlight from the sea surface. Time series shipboard observations occasionally observe anomalous nutrient injections to the euphotic zone (Jenkins, 1987), large enough that only a few of these events per year could supply the entire nutrient demand. The eddy pumping mechanism is certainly part of the "missing" nutrient flux to the euphotic zone in the subtropics described in the last section (McGillicuddy and Robinson, 1998). However, the story is far from complete, because it not clear that the eddy pumping mechanism, which lifts density surfaces through the -100 meter depth of sunlight penetration, can generate chemical anomalies at the sea surface such as high sea surface oxygen supersaturation (Spitzer and Jenkins, 1989) and 3He data (Jenkins, 1988). Summarizing material from this, the previous, and the next section, sources of nutrients and carbon at the sea surface include (i) seasonal and faster changes in the depth of the mixed layer; (ii) large-scale upwelling in equatorial and subpolar oceans; (iii) transient uplift of the thermocline in mesoscale eddies; (iv) irreversible mixing required to bring subsurface geochemical signatures such as 02 and 3He to the sea surface; and, (v) nitrogen fixation. These mechanisms each have distinct signatures on the tracers 02 and 3He, but all of the data have yet to be reconciled. Mesoscale processes are certainly a crucial part of the geochemical cycles of gases and associated biological tracers in the upper ocean, but models of ocean gas chemistry at the mesoscale are still working in the diagnostic stage, attempting to understand which basic processes are important and which are not.
6. Gyre and global scale On the scale of an entire ocean basin, ocean circulation organizes into wind-driven gyres of cyclic flow. Below the surface mixed layer lie the waters of the thermocline, which communicate with the atmosphere on time scales of years to decades. Contact between the thermocline and the atmosphere occurs mostly in winter, when seasonal cooling extends the reach of surface mixing to its greatest depth. During spring warming, the relict winter mixed layer remains subsurface and now isolated from communication with the atmosphere. The stratification of that water mass determines its potential vorticity, which plays a fundamental role in the physics of thermocline circulation (Luyten et al., 1983). The constraint that fluid parcels move along lines of constant potential vorticity generates the pathways of fluid flow, including some regions of the thermocline, in theory, called "shadow zones", which are only ventilated by diffusion, not by direct advection. Below the thermocline, intermediate and deep water masses form primarily at high latitudes and advect through the entire deep ocean. The processes of deep high latitude convection (brine rejection during sea ice formation, convection in ice-free polynias in the Southern Ocean, as examples) are severely underresolved in global models, but the global deep sea ventilation rate, as indicated by the distributions of tracers such as natural and bombproduced 14C, freons, and bomb tritium, can be reproduced by clever parameterization. Ocean uptake of anthropogenic CO2 has been simulated using the global models such as the Princeton model (Sarmiento et al., 1992) or the Hamburg model (Maier-Reimer and Hasselmann, 1987). In both of these studies the uptake of fossil fuel CO2 was simulated as an abiotic perturbation to the natural system. An alternative approach has been to prescribe rather
Modelling C02 in the ocean." A review
179
than predict the physics of ocean circulation using box models with parameterized physics intended to capture the essence of the real ocean circulation using vertical diffusion, advective formation of deep water, and perhaps a diffusive, outcropping thermocline (Oeschger et al., 1975; Siegenthaler and Oeschger, 1987). These models, when tuned to reproduce the same transient tracer data as are used to validate the more physical global circulation models, tend to give about the same present-day uptake of fossil fuel (1.7 - 2.8 Gton C yr l ) as is predicted by the circulation model studies (Sarmiento and Sundquist, 1992). Still unresolved is the potential for coupling or feedback between ocean circulation and biology, and the effect of biological changes on uptake of fossil fuel CO2. Global warming is expected to have at least some impact on the circulation of the ocean, resulting, for example, in a shutdown of deep convection in a 4x CO2 coupled atmosphere ocean model experiment (Manabe et al., 1994). Simulations of fossil fuel uptake under these conditions are only beginning to appear (Maier-Reimer et al., 1996; Sarmiento and Qu6r6, 1996). The two available model predictions are radically different from each other because of the way that they treat the biological response to the oceanographic changes. Biological uptake and export of nutrients and carbon in the Hamburg model is predicted using a kinetic rate law for nutrient uptake (with a latitude dependence) which has been tuned to reproduce the present day distribution of nutrients at the sea surface under present day flow. The Princeton model restores the sea surface nutrient concentration to the observed value for present day simulation, and for the future, maintains the biological export production rate even in the face of changing flow and nutrient conditions. In short, biology in the Hamburg model attempts to respond to changes in nutrient supply, although no one, the model least of all, knows what this response will be, while the Princeton model maintains constant biological production in the face of changing circulation and climate. The Hamburg model predicts little effect of changing ocean circulation on fossil CO2 uptake, while the Princeton model predicts a much larger response. Clearly we need to be able to predict the biological response to changing climate forcing in order to predict future ocean CO2 uptake.
7. Geologic scale On a geological time scale, we must consider the interaction of the chemistry of the ocean with the dissolution of rocks on land (chemical weathering) and se&mentation in the ocean. Here an essential distinction is between weathering of igneous rocks, the calcium component of which can be idealized as CaO, and CaCO3. The former reacts with two CO2 molecules from the atmosphere to generate 2 HCO3 in river water, which is delivered to the ocean. Alkalinity is removed from sea water by burial of CaCO3, which consumes two alkalinity equivalents, retuming one carbon to CO2. This series of reactions results in the net transfer of one carbon from the atmosphere to the lithosphere as CaCO3. In contrast, weathering of CaCO3 consumes one CO2 from the atmosphere to form 2 HCO3, but it releases that CO2 upon recrystalization of CaCO3 in the ocean, resulting in no net flux from the atmosphere to the lithosphere. On time scales of thousands to tens of thousands of years, the CaCO3 weathering reaction buffers or regulates the pCO2 of the atmosphere as a function of the chemistry of the oceans. Weathering runoff products added to the ocean tend to drive the ocean to higher concentrations of CO32- (higher pH). CaCO3 is produced biologically in the ocean, and rains steadily to the sea floor. CaCO3 shells which land in the abyss of the ocean dissolve because of a pressure effect on CaCO3 solubility, while CaCO3 which lands on topographic highs is removed from the ocean by burial in the sediments. An increase in the ocean [CO32-] acts to
180
D.
Archer
increase the area on the sea floor where CaCO3 can be buried, increasing the global burial rate of CaCO3. This dynamic sets up a negative feedback which regulates the [CO32] of the ocean, on a global average, to maintain parity between terrestrial weathering and ocean burial of CaCO3. This mechanism, called CaCO3 compensation, regulates the pH of the ocean and thereby the pCO2 of the atmosphere, on time scales of 5-10 kyr. Models of this process must resolve the dissolution of CaCO3 from deep sea sediment, a process which is regulated by the diffusion of carbonate species through the sediment pore water and by the acidification of the pore water by organic carbon diagenesis (Boyle, 1983; Broecker and Peng, 1987; Sundquist, 1990; Archer et al., 1997). A recent simulation of fossil fuel uptake into an ocean circulation and carbon cycle model (Archer et al., 1997) predicted that 70-80% of the net CO2 emission (fossil fuel plus whatever eventually happens to the terrestrial biota) will be absorbed by dissolution in the ocean on a time scale of hundreds of years. Chemical neutralization by reaction with CaCO3 in the oceans and on land will account for another 20% on time scales of 5-8 kyr. 7-8% of the CO2 release is projected to remain in the atmosphere for hundreds of thousands of years, awaiting neutralization by reaction with igneous rocks on land (Berner et al., 1983). Another regulation mechanism has been identified associated with the weathering of igneous rocks (Walker et al., 1981). The CO2 flux from the atmosphere by igneous rock weathering must be balanced on geologic time by degassing of CO2 by volcanic and metamorphic decarbonation. The regulation mechanism in this case is the rate of terrestrial weathering, which is limited by the flushing rate of liquid water over the rocks (runoff). Runoff is likely to be positively correlated with global mean temperature, which is in turn driven by atmospheric CO2 concentration. So the atmospheric CO2 concentration is controlled by the requirement for balance between the degassing source of CO2 to the atmosphere and the igneous rock weathering sink. This mechanism is thought to act with an e-folding time of 100-400 kyr. Models of the igneous rock weathering cycle and atmospheric CO2 include (Walker et al., 1981; Berner et al., 1983; Berner, 1994). These two models are not at all in conflict with each other, fundamentally because the pH equilibrium reactions of the carbonate buffer in sea water have two degrees of freedom (according to the phase rule F = 2 + C - P, where F is degrees of freedom, C is the number of components, here CO2, CaCO3, and H20, and P is the number of phases, here gas, solution, and crystalline CaCO3). The CaCO3 compensation mechanism in effect specifies the concentration of CO32 in the ocean, while the igneous rock mechanism specifies the CO2 concentration in the atmosphere. Both of these demands can be met simultaneously, with no conflict, by varying independently the alkalinity and total CO2 concentration of the oceans. Our confidence in predictions of CO2 uptake on geological time scales is badly marred by our inability to explain the observation that atmospheric CO2 during glacial time was 30% lower than preanthropogenic Holocene values (Barnola et al., 1987). The terrestrial biosphere was if anything smaller than today (Shackleton, 1978); the wrong direction of change to explain the atmospheric data, leaving the ocean as the only conceivable source for the variations. A temperature related increase in CO2 solubility was nearly offset by a salinity related solubility decrease, as ice sheets withdrew fresh water from the oceans. Candidate mechanisms include an increase in the biological pump which sequesters carbon in the deep sea (Sarmiento and Toggweiler, 1984) or a change in the pH balance of the entire ocean (Archer and Maier-Reimer, 1994; Sanyal et al., 1995). In either case, changes in the biology and the circulation of the ocean must have been ultimately responsible, and until we understand how this coupling operated i~ the past, we will be unable to predict with any confidence how atmospheric CO2 in the future will respond to changes in Earth's climate.
Modelling C02 in the ocean."A review
181
References Archer, D. (1995) Upper ocean physics as relevant to ecosystem dynamics: A tutorial. Ecological Applications 5: 724-739. Archer, D., H. Kheshgi and E. Maier-Riemer (1997) Multiple timescales for neutralization of fossil fuel CO2. Geophysical Research Letters 24:405-408. Archer, D.E. and E. Maier-Reimer (1994), Effect of deep-sea sedimentary calcite preservation on atmospheric CO2 concentration. Nature 367: 60-264. Barnola, J.M., D. Raynaud, Y.S. Korotkevich and C. Lorius (1987) Vostok ice core provides 160,000 year record of atmospheric CO2. Nature 329:408-414. Berner, R.A. (1994) GEOCARB II: A revised model of atmospheric CO2 over Phanerozoic time. Americal Journal of Science 294:56-91. Berner, R.A., A.C. Lasaga and R.M. Garrels (1983) The carbonate-silicate geochemical cycle and its effect on atmospheric carbon dioxide over the past 100 million years. Americal Journal of Science." 283:641-683. Boyle, E.A. (1983) Chemical accumulation variations under the Peru Current during the past 130,000 years. Journal of Geophysical Research 88: 667-7680. Bradshaw, A.L., P.G. Brewer and D.K. Shafer (1981) Measurements of total carbon dioxide and alkalinity by potentiometric titration in the GEOSECS program. Earth and Planetary Science Letters 55:99-115. Broecker, W.S. and T.H. Peng (1987) The role of CaCO3 compensation in the glacial to interglacial atmospheric CO2 change. Global Biogeochemical Cycles 1" 15-29. Chipman, D.W., J. Marra and T. Takahashi (1992) Primary production at 47~ and 20~ in the North Atlantic Ocean: a comparison between the ~4C incubation method and mixed layer carbon budget observations. Deep Sea Research 40" 151-170. Clayton, T.D. and R.H. Byrne (1993) Spectrophotometric pH measurements" Total hydrogen ion concentration scale measurements and at-sea results. Deep Sea Research 40:2315-2329. Coale, K.H., K.S. Johnson, S.E. Fitzwater, R.M. Gordon, S. Tanner, F.P. Chavez, L. Ferioli, C. Sakamoto (1996) A massive phytoplankton bloom induced by an ecosystem-scale iron fertilization experiment in the equatorial Pacific Ocean. Nature 383:495-501. Davies, J.T. (1996) The effects of surface films in damping eddies at a free surface of a turbulent liquid. Proceedings of the Royal Society of London A290:515-526. Deacon, E. (1997) Gas transfer to and across an air-water interface. Tellus 29:363-374. Frew, N.M. (1997) The role of organic films in air-sea gas exchange. In: P.S. Liss and R.A. Duce (Eds.) The sea surface and global change, Cambridge University Press, Cambridge, pp. 121-172. Gaspar, P., Y. Gregoris and J.M. Lefevre (1990) A simple eddy kinetic energy model for simulations of the oceanic vertical mixing: tests at station Papa and the long-term upper ocean study site. Journal of Geophysical Research 95:16179-16193. Gruber, N. and J.L. Sarmiento (1997) Global patterns of marine nitrogen fixation and denitrification. Global Biogeochemical Cycles 11:235-267. Hayward, T.L. (1987) The nutrient distribution and primary production in the central North Pacific. Deep Sea Research 34:1593-1627. J~ihne, B. (1985) Transfer processes across the free water surface. University of Heidelberg, Heidelberg. J~ihne, B., K.O. M~innich, R. B6singer, A. Dutzi, W. Huber and P. Libner (1987) On the paramterization of air-water gas exchanger Journal Geophysical Research 92" 1937-1949. Jenkins, W.J. (1987) 3H and 3He in the Beta Triangle: observations of gyre ventilation and oxygen utilization rates. Journal of Physical Oceanography 17:763-783. Jenkins, W.J. (1988)Nitrate flux into the euphotic zone near Bermuda. Nature_331"521-524. Johnson, K.S. (1982)Carbon dioxide hydration and dehydration kinetics in seawater. Limnology and Oceanography 27:849-855. Johnson, K.S., K.H. Coale, V.A. Elrod and N.W. Tindale (1994) Iron photochemistry in seawater from the equatorial Pacific. Marine Chemistry 46:319-334.
182
D. Archer
Karl, D.M. and R. Lukas (1996) The Hawaii Ocean Time-series (HOT) program" Background, rationale and field implementation. Deep Sea Research H 43:129-156. Kraus, E.B. and J.S. Turner (1967) A one-dimensional model of the seasonal thermocline II. The general theory and its consequences. Tellus 19"98-105. Lal, D. and T. Lee (1988) Cosmogenic 32p and 33p used as tracers to study phosphorus recycling in the upper ocean. Limnology and Oceanography 33:752-754. Lapitan, R.L., R. Wanninkhof and A.R. Mosier (1999) Methods for stable gas flux determination in aquatic and terrestrial systems. In" A.F. Bouwman (Ed.) Approaches to scaling of trace gas fluxes in ecosystems. Developments in Atmospheric Science 24. Elsevier, Amsterdam, pp. 27-66. Large, W.G., J.C. McWilliams and S.C. Doney (1994) Oceanic vertical mixing: A review and a model with a nonlocal boundary layer jparameterization. Review of Geophysics_32:363-404. Ledwell, J.R. (1984) The variation of the gas transfer coefficient with molecular diffusivity. In W. Brutsaert and G.H. Jirka (Eds.) Gas Transfer at Water Surfaces, D. Reidel, Hingham, MA, pp. 293-302. Liss, P.S. and R.A. Duce (Eds.) (1997) The Sea Surface and Global Change, Cambridge University Press, Cambridge, 519 pp. Liss, P. S. and L. Merlivat (1986) Air-sea gas exchange rates: Introduction and synthesis. In P. BuatMenard (Ed.) The role of air-sea exchange in geochemical cycling, D. Reidel, Dordrecht, pp. 113127. Luyten, J.R., J. Pedlosky and H. Stommel (1983) The ventilated thermocline. Journal of Physical Oceanography 13:292-309. Maier-Reimer, E. and K. Hasselmann (1987) Transport and storage of CO2 in the ocean-- an inorganic ocean-circulation carbon cycle model. Climate Dynamics 2:63-90. Maier-Reimer, E., U. Makolajewicz and A. Winguth (1996) Future ocean uptake of CO2: interaction between ocean circulation and biology. Climate Dynamics 12:711-721. Manabe, S., R.J. Stouffer and M.J. Spelman (1994) Response of a coupled ocean-atmosphere model to increasing atmospheric carbon dioxide. Ambio 23:44-49. Martin, J.H. and S.E. Fitzwater (1988) Iron deficiency limits photoplankton growth in the north-east Pacific subarctic. Nature 331:341-343. Martin, P.J. (1985) Simulation of the mixed layer at OWS November and Papa with several models. Journal of Geophysical Research 90:903-916. Martin, P.J. (1986) Testing and comparison of several mixed-layer models. Tech Rep 143, Naval Ocean Research and Development Activig,, Stennic Space Center, Mississipi. McGillicuddy, D.J. and A.R. Robinson (1998) Eddy induced nutrient supply and new production in the Sargasso Sea. Deep Sea Research 144:1427-1450_ Mellor, G. L. and T. Yamada (1974) A hierarchy of turbulence closure models for planetary boundary layers. Journal Atmospheric Science 31:1791-1806. Memery, L. and L. Merlivat (1985) Modelling of gas flux through bubbles at the air-water interface. Tellus 37B:272-285. Michaels, A.F. and A.H. Knap (1996) Overview of the U.S. JGOFS Bermuda Atlantic Time-series Study and Hydrostation S program. Deep Sea Research H 43:157-198. Millero, F.J., R.H. Byrne, R. Wanninkhof, R. Feely, T. Clayton, P. Murphy and M. Lamb (1993) The internal consistency of CO2 measurements in the equatorial Pacific. Marine Chemistry 44:269-280. Oeschger, H., U. Siegenthaler, U. Schotterer and A. Gugelmann (1975) A box diffusion model to study the carbon dioxide exchange in nature. Tellus 27:168-192. Owens, N.J.P., J.N. Galloway and R.A. Duce (1992) Episodic atmospheric nitrogen deposition to oligotrophic oceans. Nature 357:397-399. Price, J.F., C.N.K. Mooers and C. Van Leer (1987) Observations and simulation of storm-induced mixed-layer deepening. Journal of Physics, 1Oceanography 8:582-599 Price, J.F., R.A. Weller and R. Pinkie (1986) Diurnal cycling: observations and models of the upper ocean response to diurnal heating, cooling and wind mixing. Journal of Geophysical Research 91:8411-8427.
Modelling C02 in the ocean: A review
183
Robertson, J.E. and A.J. Watson (1992) Thermal skin effect of the surface ocean and its implications for CO2 uptake. Nature 358:738-740. Roy, R.N., L.N. Roy, K.M. Vogel, C. Porter-Moore, T. Pearson, C.E. Goode, W. Davis, F.J. Millero (1993) The dissociation constants of carbonic acid in seawater at salinities 5 to 45 and temperatures 0 to 45~ Marine Chemistry 44:249-267. Sanyal, A., G. Hemming, G. Hansen and W. Broecker (1995) Evidence for a higher pH in the glacial ocean from boron isotopes in foraminifera. Nature 373:234-237. Sarmiento, J.L. and C. Le Qu6r4 (1996) Oceanic carbon dioxide uptake in a model of century-scale global warming. Science 274:1346-1350. Sarmiento, J.L., U. Siegenthaler and J.C. Orr (1992) A perturbation simulation of CO2 uptake in an ocean general circulation model. Journal of Geophysical Research 97:3621-3645. Sarmiento, J.L. and E.T. Sundquist (1992) Revised budget for the oceanic uptake of anthropogenic carbon dioxide. Nature 356:589-593. Sarmiento, J.L. and R. Toggweiler (1984) A new model for the role of the oceans in determining atmospheric pCO2. Nature 308:621-624. Shackleton, N.J. (1978) Carbon 13 in Uvigerina: tropical rainforest history and the equatorial Pacific carbonate dissolution cycles. In N.R. Andersen and A. Malahoff (Eds.) The Fate of Fossil Fuel C02 in the Oceans, Plenum Press, New Yo,'k, pp. 401-428. Siegenthaler, U. and H. Oeschger (1987) Biospheric CO2 emissions during the past 200 years reconstructed by deconvolution of ice core data. Tellus 39B: 140-154. Smethie, W.M., T.T. Takahashi, D.W. Chipman and J.R. Ledwell (1985) Gas exchange and CO2 flux in the tropical Atlantic Ocean determined from 222Rn and pCO2 measurements. Journal of Geophysical Research 90: 7005-7022. Spitzer, W.S. and W.J. Jenkins (1989) Rates of vertical mixing, gas exchange, and new production: estimates from seasonal gas cycles in the upper ocean near Bermuda. Journal of Marine Research 47:169-196. Sundquist, E.T. (1990) Influence of deep-sea benthic processes on atmospheric COR. Phil Transactions Royal Society of London A 331:155-165. Takahashi, T., R.A. Feely, R. F. Weiss, R.H. Wanninkhof, D.W. Chipman, S.C. Sutherland and T.T. Takahashi (1997) Global air-sea flux of CO2: An estimate based on measurements of sea-air pCO2 difference. Proceedings of the National Academy of Sciences USA 94:8292-8299. Tans, P.P., I.Y. Fung and T. Takahashi (1990) Observational constraints on the global atmospheric carbon dioxide budget. Science 247:1431-1438. Villareal, T.A., S. Woods, J.K. Moore and K. Culver-Rymsza (1996) Vertical migration of Rhizosolenia mats and their significance to NO3- fluxes in the central North Pacific gyre. Journal of Plankton Research 18:1103-1121. Walker, J.C.G., P.B. Hays and J.F. Kasting (1981) A negative feedback mechanism for the long-term stabilization of Earth's surface temperature. Journal Geophysical Research 86:9776-9782. Wanninkhof, R. (1992) Relationship between wind speed and gas exchange over the ocean. Journal of Geophysical Research 97:7373-7382. Watson, A., R. Upstill-Goddard and P.S. Liss (1991) Air-sea exchange in rough and stormy seas, measured by a dual tracer technique. Nature 349:145-147. Weiss, R. (1974) Carbon dioxide in water and seawater. The solubility of a non-ideal gas. Marine Chemistry 2:203-215. Woolf, D.K. (1997) Bubbles and their role in gas exchange. In: P. S. Liss and R. A. Duce (Eds.) The sea surface and global change, Cambridge University Press, Cambridge, pp. 173-207.
This Page Intentionally Left Blank
Chapter 9
SIMULATION MODELS OF T E R R E S T R I A L TRACE GAS FLUXES AT SOIL M I C R O S I T E S TO GLOBAL SCALES
D.S. Schimel and N.S.. Panikov
This Page Intentionally Left Blank
Approaches to scaling a trace gas fluxes in ecosystems A.F. Bouwman, editor 9 Elsevier Science B.V. All rights reserved
SIMULATION MODELS OF TERRESTRIAL TRACE GAS FLUXES AT SOIL MICROSITES TO GLOBAL SCALES
D.S. Schimel I and N.S. Panikov 2 1 National Center for Atmospheric Research, Boulder, Colorado, CO 80303 USA z Institute of Microbiology, Russian Academy of Sciences, Moscow 117811, Russia
I. Introduction The contribution of terrestrial ecosystems to the global budgets of most trace gases is much more important than the contribution of marine systems. This difference stems from the fact that aquatic systems are often more homogeneous and characterized by almost complete compensation between consumption and production of most trace species within the water column (see Middelburg et al., 1999). The most striking evidence of the importance of the role of terrestrial systems in the global budgets of trace gases stems from the latitudinal distribution of atmospheric concentrations of methane (CH4) and nitrous oxide (N20). Maximum values are reached in the Northern Hemisphere, where the extent of terrestrial systems is much larger than in the Southern Hemisphere. Mathematical models play an important role in natural sciences in understanding complex phenomena. This is done through testing different hypotheses by comparing the model's output with measured data. Models are also used for extrapolation of measured data to wider temporal and spatial coverage. The aim of this paper is to review the application of mathematical models at different scales in studies on trace-gas exchange between the soil and the atmosphere. We will emphasize models describing fluxes and their variation in space and time on the basis of the underlying mechanisms (mechanistic models). There are various classifications for grouping models simulating gas exchanged between terrestrial ecosystems and the atmosphere. The different classes include deterministic and stochastic, empirical and mechanistic, numeric and analytic, and dynamic and static formulations. We will skip over the respective definitions because they are well known and are of general application in any scientific field. The majority of models used in terrestrial flux studies are deterministic, numerical, dynamic simulations. Stochastic models are used sometimes to deal with statistical matters, e.g. to assess the confidence limits of scaled-up estimates of fluxes measured with chambers (Roulet et al., 1992; Panikov et al., 1995). Important differentiation of models comes, however, from the consideration of spatial scale. Some models ignore the spatial inhomogeneity of soils and ecosystems. Other, more complicated models, take into account spatial gradients at microscales to consider gas transport and variations in substrate for microorganisms; still others are linked to Geographic Information Systems for regional and global calculations (Schimel et al., 1994). There is a continuum of models filling the space between two extremes. One pole is occupied by simulations based on purely empirical relationships, e.g. regression models describing fluxes as a function of climatic parameters, the regression coefficients being fitted to data. The opposite extreme is represen,=d by highly structured and developed mechanistic
188
D.S. Schimel and N.S. Panikov
models that mimic the real processes, e.g. models describing the coupled tropospheric chemistry of methane (CH4), carbon monoxide (CO), ozone (O3), nitrogen oxides (NOx) and hydroxyl radicals (OH) (Crutzen and Zimmerman, 1991). The majority of models describing gas fluxes in terrestrial ecosystems is located between two mentioned extremes and is frequently referred to as semi-empirical or semi-mechanistic. This means that they contain both types of coefficients (mechanistic and fitted), and only some of them can be interpreted in strictly mechanistic terms. A good example of such compromise are the SOMNET (Soil Organic Matter Network) models describing primary productivity and decomposition in terrestrial ecosystems. Two of the most popular and developed are the CENTURY model (Parton et al., 1994) and DAYSY (Jensen, Nielsen, Royal Vet. and Agricultural University, Denmark), which are intensively used in applied ecology to predict the effects of land use, pollution, and climatic changes in particular ecosystems. Input parameters are weather data
-Arthrobacter
~
Pseudomonas
......
Bacillus
100
4
80 +
.......
-3
) o c~
m
60 -
~
0
~
._ m ,,
9
0
~-
.
.
.
.
.
.
0
- .
.
,
tO
.
.
2
......
1'
.
1
2
~'E ._>,
.
,,
C02
3
flux
4
Plant
A
,
.
.
.
1 6
litter
,
:'
.
5
;:
1.5
1.4
,".'
t
"o o~ x"
0
1.3
"13 @" --! ,.= r
-'~ (3 3
1.2
-2 ~- . . . . . . . . . . . . . .
-
0
1
0
1
2
3
4
5
-
-1.1
6
20-
0
oo -20
-40 2
3
4
5
Years
Figure 1. Simulation of microbial population dynamics and carbon dioxide (CO2) exchange in a tundra ecosystem under anticipated warming. The assumed changes in the climatic and ecosystem parameters are: increase of the January temperature from -24 to -14~ increase of the number of frost-flee days from 81 to 135, and increase of the index of degree-days from 293 to 1365.
Simulation models of terrestrial trace gas fluxes at soil microsites to global scales
189
(rainfall, air temperature, irradiation), soil characteristics (texture, pH, thermal and hydraulic conductivity), plant growth and land use parameters (crop rotation, fertilizer use, irrigation/ drainage, N deposition, etc.). Model outputs include soil temperature and water, carbon and nitrogen dynamics, and the rates of biogeochemical processes including gas fluxes (phytomass production, decomposition of organic matter, soil respiration, nitrification, denitrification). These models are called process-oriented (Paustian, 1994) for the reason that they focus on the processes and do not explicitly represent the metabolism and population dynamics of the organisms (microorganisms, animals, plants) catalyzing the processes in question. One step forward to tackle biological mechanisms is provided by the so-called organismoriented models that simulate the flows of matter or energy through different functional or taxonomic groups of soil organisms (Paustian et al., 1990). These groups can be very general (e.g. bacteria-fungi) or represent more specifically detrital food webs (Hunt et al., 1987). The next step in such developments is to simulate the biological mechanisms known from contemporary physiological and molecular studies. The main problem to be resolved here is how to avoid either oversimplification or excessive intricacy (in order to keep such models manageable). A compromise has been found in the concept of the coordinated microbial biosynthesis and construction used in synthetic chemostat model (Panikov, 1991; 1996) stemming from an idea put forward by Powell (1967). Independently, chemical approaches led to construction of cybernetic growth models having similar (but not identical) mathematical structure (Kompala et al., 1986; Turner et al., 1989). These models turned out to be rather efficient research tools for examiration of complex dynamic beha,,iour of soil organisms catalyzing transformation of organic matter: growth and survival, differentiation and extinction, transient dynamics, variations in cell composition, synthesis of by-products, competition between organisms with different life strategy, etc. The models of this kind were able to generate realistic patterns of mass and energy flow (primary productivity, decomposition rates, soil respiration) under present-day conditions and in response to certain kinds of perturbations such as pollution, fertilization and drying-rewetting of soil (see Figure 1 for illustration). However, their obvious disadvantage is that they are too complex to be used over expanded spatial scales.
2. Mechanistic models at small scales 2.1. General principles
Trace gas exchange with the atmosphere occurs through a set of coupled processes, including (i) production of substrate(s) for processing by trace gas-producing organisms: acetate or carbon dioxide + molecular hydrogen (CO2+H2) for methanogenic bacteria, ammonium (NH4+) and nitrite (NO2) for nitrous oxide (NzO)-producing nitrifying bacteria, nitrate (NO3-) and available organic compounds for NzO-producing denitrifying microorganisms, oxidizible organic compounds for respiring organisms (microbes, plant roots, animals) producing CO2, and pyruvate for isoprene formation in plants. (ii) conversion of substrate to respective gaseous species in parallel with the biological or chemical uptake of gases within an ecosystem, e.g. consumption of CH4, N20, or CO2 by methanotrophs, denitrifying/nitrifying bacteria, and autotrophic organisms, respectively, chemical immobilization of CO2 in insoluble carbonates, etc. (iii) mass-transfer of produced gas to the free atmosphere, which includes three main mechanisms: molecular diffusion, vascular gas transfer, and ebullition.
190
D.S. S c h i m e l a n d N.S. P a n i k o v
2.2. S u b s t r a t e p r o d u c t i o n
Substrates are formed through one of three processes: (i) decomposition (hydrolytic breakdown of plant litter, oxidation, fermentation); (ii) nitrogen mineralization, and (iii) photosynthesis + photorespiration. However, at the ecosystem level the generation of almost all substrates is initiated by primary production of organic matter through plant photosynthesis or (occasionally) bacterial chemosynthesis. The chain of events leading to formation of immediate precursors of trace gases can be rather long. For example, methanogenic substrates (acetate or CO2+H2) are produced at the end of the following sequence of processes (Figure 2): photo-assimilation of CO2 in plant leaves; photosynthate m o v e m e n t to stems and roots; plant litter formation and rhizodeposition (root exudation and sloughing); breakdown of plant litter; fermentation of monomeric organic compounds (like sugars) into volatile fatty acids, H2, and CO2; and synthrophic conversion of propionate and butyrate to acetate. It is worthwhile to note that plants, the primary producers, are frequently neglected in models of terrestrial ecosystem production of CH4, CO2, and other gases. There are two main flows of Csubstrates from plants: (i) via plant litter formation, which is represented by resistant macro-
Figure 2. Schematic representation of biological ; rocesses in wetland soil responsible fcr methane formation, uptake and emission. The symbols indicate: - ci, carbon-reservoirs:c~, green phytomass; c2, belowgroundphytomass(roots and rhizoid); c3, plant litter; Ca, CO2; c5, low molecularweight C-compounds;c6, volatilefatty acids; c7, CH4. - xi, biocatalysts:x~, aerobic soil microorganisms;x2, fermenting microorganisms;x3, methanogens;x4, methanotrophs;xs, protozoa (microscopicanimalsfeeding on microbial cells);x6, hydrolyticenzymes. The arrows indicatethe followingbasic processes: - Plant-mediated: c4--+cb plant photosynthesis; c~--+c2, transport of C-compounds (photosynthates) from leaves to roots; c ~ c 3 , plant litter formation; c2---}c5, root exudation; c2---}c4,root respiration. - Microbial community: c3-+c5, depolymerizationof plant litter; c5-~c6, fermentation (anaerobic conversion of sugars to acetate and other volatile compounds); c6--).c7, CH4 formation; c5-~c4, total microbial respiration; c7---).c4, C H 4 consumption/oxidation. - General: 1, gas moleculardiffusion; 2, gas vascular transport; 3, biosynthesisof hydrolytic enzymes, 4; protozoan grazing; 5, oxygenuptake for respiration.
Simulation models of terrestrial trace gas fluxes at soil microsites to global scales
191
molecules like lignocellulose and is maximal at the end of the growing season (fall); and, (ii) continuous supply of readily available C monomers (root and foliage exudation) which is maximal during early ontogenetic phases. Production of gas substrate(s) and gases and mass-transfer processes in the soil or water column should not be considered in separation. A better approach would be the construction of a holistic model based on Figure 2 or similar schemes, which represent the entire C balance of a wetland ecosystem. In part this would be an improvement because some of the chemical transformations are sensitive to product and reactant concentrations, which are largely influenced by transport. Cao et a l . ' s (1995) model of CH4 emissions accounts for carbon mass flow to the methanogenic community from primary producers (plants) via the decomposition loop and rhizodeposition. Plant photosynthesis and carbon allocation to roots was simulated with the International Benchmark Sites Network for Agro-Technology Transfer (IBSNAT) model, CH4 oxidation and rhizodeposition was assumed to be a linear function of root mass, and decomposition was described by a first-order rate equation. The simulation accounted for about 90% of the seasonal variation of CH4 emission from a particular site used for model calibration. Even better agreement with observations has been achieved in a one-dimensional model (Walter et al. 1996) that explicitly describes the formation of methanogenic substrate via decomposition of plant litter and rhizodeposition as well as the main gas-transfer mechanisms including diffusion, ebullition, and vascular transport. The results simulated with this model explained close to 100% of the variation in three years of observations of CH4 emissions from a Michigan peatland. The link of microbial activity to plant components has been extended to the global scale by Christensen et al. (1996). They assumed that CH4 emission from waterlogged soils is proportional to heterotrophic respiration and net primary production (NPP). Applied on a 1~ grid using standard climatological and wetland distribution data sets, this approach provides a simulation of seasonal dynamics and a circumpolar estimate of a total CH4 emission. The simulation of N20 emission requires consideration of the combined N and C cycles, because the substrates of denitrification include electron acceptors (nitrate) as well as oxidizable C substrates. In the case of modelling of CO2 emissions the situation should be less complicated if supply of mineral elements and N do not limit decomposition of organic matter, particularly in terrestrial ecosystems. Otherwise, the model must include a description of the influence of limiting mineral compounds and their sources and sinks.
2.3. Conversion of substrate to gaseous product
Examples of processes where trace gases are formed include: Nitrification Denitrification Methanogenesis Methanotrophy NO production Isoprene synthesis
NH4+--~(approx. 99%) NO3-+ (approx. 1%) N20 NO3--+ N20--4 N2 C02 + H2 --~ CH4 or acetate --~ CH4 CH4 ~ CO2 + H20 NO3 --~ NO pyruvate ...--~ ...--~ isoprene (C5H8)
Commonly, in model descriptions the reaction rates of the processes listed are assumed to depend on the concentration of the substrates (Table 1). The hyperbolic Michaelis-Menten equation (3) is the most popular and adequate expression to relate reaction rate to concentration. The equations of chemical kinetics are easily derived from it under specified conditions:
192
D.S. Schimel and N.S. Panikov
Table 1. Generalized equations to relate substrate concentration and trace gas production (atter Schimel and Potter, 1995; Panikov, 1995). Model type
Formulation (dG/dt)
Zero order
V max
First order
K x s; k =
Michaelis-Menten (mixed order)
(1)
(vmaX/Ks)
(2)
V maxs
(3)
K,. + s Summative Michaelis-Menten (heterogeneity of biocatalysts)
vlmax S
v2max S
+
K.,.~ + s
K.,.2 + s
Multiplicative Michaelis-Menten (control of co/wersion rate by several substrates)
ma,( S, /(
Threshold effects
vmax(s--s* )
(4)
~tmax
~
+...-
,=~ K~, + s
" / ':
K,.~ + s~
(5)
'
K.,. 2 + s 2
,=~ K,..i + s, (6)
K., + s - s * k, phenomenological first-order rate constant (hl); s, substrate concentration (g m-3); 1/m~, maximum reaction rate (g h-1 m3); Ks, saturation constant (g m3).
12 ~ ~....
10
~
~ a -7e-
2 .
.
.
.
.
.
.
100
.
.
.
.
.
.
.
.
150
.
.
.
.
.
200
.
.
.
.
250
.
. 300
Dissolved CH4, mM 1 -
0.8
0.6
-~
~r 0.4
A 0.1
.
.
.
.
100
Ks,
1000
mM
Figure 3. The effects of CH4 concentration on the methanotrophic activity in a Sphagnum-Eriophorum peat soil (from Dedysh and Panikov, 1997). Upper panel: original experimental data fitted to equation (4). Bottom panel: relation between the distribution of soil methanotrophic activity and affinity to CH4, with three peaks corresponding to three functional groups of organisms or uptake systems with high affinity (Ks=0.3 mM), low affinity (KF200 mM) and intermediate affinity (K,=3 mM) to CH4, respectively.
Simulation models o f terrestrial trace gas fluxes at soil microsites to global scales
193
at low substrate concentration (e.g. substrate concentration much smaller than the saturation constant, i.e. s<>Ks) we arrive at zero-order equation (1). The sum of two or more equations (3) actually stands for heterogeneity of microbial catalysts represented by a mixture of micro-bial species differing in their affinity to the substrate. To assess this heterogeneity, we should fit equation (4) to experimental data relating the reaction rate to s. Then the best-fit n-value would be the measure of microbial heterogeneity with respect to parameter Ks. An example is given in Figure 3 showing that the best fit to available experimental data was provided when the presence was assumed of three groups of soil methanotrophs with Ksvalues 0.3, 3, and 200 mM of dissolved methane. The multiplicative form of Michaelis-Menten equation (5) describes the simultaneous limita-tion of the reaction rate with several substrates. Examples are the dependence of N20 formation on nitrate and C substrate concentration and the dependence of methanotrophic activity on CH4 and O2 concentration. Finally, the threshold substrate concentration s* has to be introduced (equation 6) if the studied ,~onversion reaction is the sum of two processes of opposite directions, e.g. uptake and leakage. For s = s * the net reaction rate is zero because the two processes compensate each other. Microbial population dynamics is described in a limited number of simulation models, although such dynamics are very important to improve our understanding of the observed flux variations related to climate change (Panikov, 1994). Population dynamics are described by differential equations that account for sources (growth, immigration) and sinks (autolysis, grazing, washing out) of microbial cells in the soil, based on the following assumptions: (i) the reaction rate is proportional to biomass of the microorganisms considered; (ii) the yield factor (Y) relating biomass increase (Ax) and substrate uptake (As) is constant (Y=Ax/As); (iii) the growth rate does not depend on population density or competitive interactions with other microorganisms (the cells are immobile and homogeneously distributed in the soil). However, the above assumptions are often not realistic. In particular, the reaction rate should be related to "active" biomass (rather than to total biomass according to the first assumption) by introducing a special physiological state function (Panikov, 1991). This function is also useful to describe survival of starving microbes and variation of their metabolic activity. The apparent yield factor varies because of maintenance requirements and other non-productive respiration reactions (the higher the maintenance/unproductive catabolism, the stronger the deviation from assumption ii). Finally, the effects of population density, competition and heterogeneity often cannot be neglected (assumption iii). For example, to simulate seasonal dynamics of tundra soil respiration and organic matter decomposition the competitive interactions between populations with different functional characteristics had to be accounted for, including their growth rate, affinity to common substrate, survival ability, and response to temperature (Panikov, 1994). In tundra ecosystems, the short-term and long-term effects of temperature are not identical, the latter being dependent on structural changes in the microbial community. Elevated temperature and input of dead organic matter will reduce the pressure of environmental stress, resulting in stabilization of the tundra microbial community. Increased temperature can lead to acceleration of aerobic decomposition of dead organic matter (plant litter, soil humus), i.e. tundras may tum from a sink into a source of CO2. Competition effects are essential for modelling the soil anaerobic community. The methanogenic, actogenic, and sulfidogenic bacteria compete with each other for molecular hydrogen and the outcome of this competition depends on the affinity to H2, the temperature, and population density of the various populations. Competitive interactions determine the pattern of gas formation not only quantitatively, but also in qualitative terms. For example,
194
D.S. Schimel and N.S. Panikov
ecosystems can be a source of CH4 (if methanogenic bacteria prevail), or H2S and other sulfides (under domination of sulfate reducing bacteria), or acetic acid (if a large population of acetogens is present). Kinetic (mechanistic) models of the methanogenic community are discussed in numerous papers and books, because of the importance of methanogenesis in general microbiology and in the biotechnology of waste water treatment (e.g. Stams, 1994; Westerman, 1994; Buffiere et al., 1995). The biochemistry of anaerobic digestion can be described schematically by four major steps: Plant litter
Hydrolysis
>Monomers
F. . . .
tation
>
Organic acids, H z, CO:
Ace,og.... ,s >Acetate
Methanog . . . .
,s >CH 4
$ CH 4
This anaerobic metabolic pathway including its organisms is very sensitive to elevated concentrations of intermediate products (H2, formate, acetate). Their accumulation above some threshold level inhibits activity of syntrophic bacteria which convert propionate, butyrate, ethanol, and probably other reduced compounds to acetate, CO2, and H2 (e.g., CH3CHzCOOH + H20 ~ aCH3COOH + bCO2 + cH2). In fact, these reactions are thermodynamically feasible only at low hydrogen and formate concentrations as can be seen from a calculation of Gibb' s energy: AG=(AH~
~ T) +RT ln([acetate] [HCO3] [H+] [H2]/[propionate])
(7)
where AH~ and zlS~ are standard enthalpy and entropy of reaction, T is absolute temperature, R is the gas constant, and values in square brackets are concentrations. At room temperature the AG-value is negative (reaction is allowed) only if the partial pressure of H2 is less than 10 .4 atm. There are several mathematical models describing CH4 formation in an anaerobic community assuming a conditional halt on syntrophic growth for positive AG-value (Zavarzin et al., 1990).
2.4. Trace gas transport In addition to the biochemistry, as noted above, transport is also critical for the flux of trace gases to the atmosphere. There are three main transport mechanisms: (i) molecular diffusion, (ii) vascular transport of gas through plant roots, and (iii) ebullition. Diffusion is usually modelled using Fick's first law: J - -~ O~ ~ s/~ z
(8)
where J is the diffusive flux, ~b is porosity, Ds is the diffusion coefficient, and 8 s/8 z is the concentration gradient over the distance z. Vascular transport can be described as a diffusion process through gas-filled channels. Vascular transport is two or three orders of magnitude more rapid than Fickian flow in water. Careful measurements are needed to estimate the parameter ~bas a function of root mass and porosity of specific plants. To simulate ebullition we can use approaches developed in chemical technology (Bowers and Noyes, 1985). Fluctuations in dissolved gas concentration (c/~) provide microscopic shortlived cavities called bubble germs. Whey cL attains saturation, the germ is converted to a nucleus, which starts to grow into a bubble. The dynamics of the bubble radius, r, follows equation (9):
Simulation models o f terrestrial trace gas fluxes at soil microsites to global scales
195
Table 2. Mathematical expressions used to relate biogeochemical reaction rates to temperature.
Model type
Formulation for reaction rate k
Eyring
d(lnk)/dT = (Atfl+RT)/T
Arrhenius
d(lnk)/dT = EJT, (AH*+RT ~ AH = E~) In k = In A - E,,/RT or k = A exp(-E,,/RT)
(11 ) (11 a) (llb)
Exponential (Qi0) model
d(lnk)/dTc = or, Qi0 = exp(10%t) In k = In A + a t e k = A exp(etT,.)
(12) (12a) (12b)
Account of reversible temperature denaturation
k = A exp(a Tc)/{ 1+exp(AS"/R-AG~ k = A exp(etT~.)/{ l+exp[fl( l-O/T)]} /3= AG~ 69= A/F/AS"
Account of heterogeneity of the soil
~
or
(13) (13a) (13b)
A,e~'r( 1 + e ~ir)i~
community
(1 O)
'
i = 1, 2 ..... n
(14)
t=l T is absolute temperature (K), Tc is temperature (~ R is the gas constant, AH* is the enthalpy of activated complex formation, Ea is the activation energy, A is the integration constant, interpreted as the frequency of collisions of reacting molecules, AG ~ AS~ and AH~ are the standard Gibbs free energy, enthalpy, and entropy of denaturation reaction; a,/3 and O are kinetic parameters.
dr/dt=3RTk(cL -CL *-2 Us/r)/(2Pr+4s)
(9)
where k is volumetric mass-transfer rate, CL* is gas solubility, U is the Henry constant, s is the surface tension, and P is external pressure (usually P=I atm). There is some critical r* (r*=--2Us/[CL-CL*]) such that if r < r*, the nucleus will be annihilated (dr/dt < 0), whereas nuclei having r > r* will grow (dr/dt < 0) into a bubble.
2.5. Effects of environmental factors on gas emission
Above we have discussed "proximal" environmental factors including primary productivity of ecosystems and substrate concentration. Now we will cover other factors, including temperature, water content and soil pH. 2.5.1. Temperature
Temperature effects are usually described by an exponential function (Q10, equation 12 in Table 2). Other models are summarized in Table 2. In chemical kinetics the dependence of the reaction rate on temperature is explained by transition-state theory developed by Eyring in 1930-1935. It is based on the use of thermodynamics and principles of quantum mechanics. The reaction proceeds through a continuum of energy states and must surpass the state of maximum energy, when the transient activated complex is formed. Then the dependence of reaction rate on absolute temperature, T, is expressed by equation (10). As can be seen from Table 2, other generally used equations (11) and (12) are no more than approximations of the mechanistically correct equation (10). These are more or less valid for a chemical reaction only in a limited range of temperatures. For biological reactions we have to account for the thermal denaturation of some key enzymes by using equation (13), where parameter O is generally close to the so-called "optimal temperature". Finally, the response to temperature variation of the soil community should bo described by equation (14), which is the sum of responses of various organisms (ideally, using a continuum of temperature adaptations, or at least several groups like psychrophiles, mesophiles, and thermophiles). It is surprising that
196
D.S. Schimel and N.S. Panikov 1 . 2
9
1 t~ r v ,r
-
0.8
-
0.6
.
C H 4 form
ation
o} E 0.4
,
CH4
.....~
0.2
0
.
0
10
20 Incubation
30 tern perature,
.
.
uptake
.
40
50
~
Figure 4. The effects of temperature on the methanotrophic and methanogenic activity of a SphagnumEriophorum peat soil (Panikov, unpublished). The rates of methane generation (open dots) and methane uptake (closed dots) were measured in a peat column continuously flushed with pure N2 and an air-CH4 mixture, respectively, at different temperatures. The curves were calculated according to equation (14).
equation (14), even at n=2, provides a fairly good fit to experimental curves (see Figure 4 as an example). This can be explained by the fact that the distribution of soil microorganisms with regard to parameter O is bimodal. Therefore, the identification procedure may be used for assessment of the relative contribution of, say, psychrophiles and mesophiles to CH4 formation and uptake. It is interesting that the anaerobic process of methanogenesis is much more responsive to temperature than CH4 uptake. The mechanistic basis for this difference is not clear, but the ecosystem consequences are clear: soil warming will accelerate emission (which is the difference between production and consumption), in spite of the simultaneous stimulation of the both opposing processes. 2.5.2. Soil moisture
The effects of soil moisture are approximated by numerous empirical expressions. For example the dependence of the soil respiration rate, R, on the soil moisture content (/4/) is described by Flanagan et al. (1993): R(W,T)=a3 W/(al+W) a2/(a2+W)
(15)
where al, a2, and a3 are site-specific empirical parameters. Two main mechanistic considerations are involved in this relationship: (i) the limitations of biological reaction by availability of water under dry conditions; and, (ii) the restricted supply of oxygen under submerged conditions. It is interesting to note that in detailed mechanistic models both phenomena should be simulated separately. For example, to account for the effects of water on methane exchange we can express the dependence of methanogenic and methanotrophic activity on 02 concentration and add a differential equation for oxygen mass-transfer. 2.5.3. p H o f the soil solution
The pH of the soil solution affects practically all known metabolic reactions. The pH curves can be described by equations used in enzymatic kinetics (Cornish-Bowden, 1976):
Simulation models of terrestrial trace gas fluxes at soil microsites to global scales
V= Vm
]2 K'[H+] [H + +K,([H+]+K2)
197
(16)
where v is the observed reaction rate, Vm is the upper limit of v, and [H +] is the proton concentration. Equation (16) is based on the assumption that the enzyme's active center is a dibasic acid, HzE, which undergoes a 2-step dissociation (H2E ~ H E ~--> E 2-) with equilibrium constants K1 and /(2, and only H E is active. Although the real mechanism is much more complex, this equation is useful as it provides a good fit to experimental data and allows us to make unbiased comparisons between different organisms under different experimental conditions. The shape of the pH profile is determined by two p K values, where the optimal pH is found as (pK2 +pK~)/2, while the difference pK2 -pK1 determines the width of the peak at 50% of maximal activity.
3. Scaling trace gas fluxes: a regional-global modelling perspective Field and laboratory studies of trace gas fluxes have, as their most distinguishing characteristic, extremely high variability and sensitivity to environmental conditions. This is manifest in a number of ways. Spatial variability within field sites is characteristically very high (Schimel et al., 1986; Mosier et al., 1997). Flux rates may change non-linearly with small changes in conditions. For example, N20 fluxes may change dramatically when soil moisture levels cross critical thresholds (Parton et al., 1987), resulting in transitions between nitrification and denitrification as the dominant process. Methane exhibits similar behaviour in response to anoxia and substrate availability. As a consequence of the observed complexity of the regulation of trace gas production, many researchers have concluded that prediction of trace gas fluxes is a nearly impossible task, likely to require detailed information on soil properties, soil microclimate, microsites within soils, and microbial population dynamics at all locations where predictions are to be made. By contrast, large scale analyses, exemplified by the work of Matson and Vitousek (1990) and model studies by Li et al. (1996) and Parton et al. (1996) suggest that large-scale patterns of trace gas fluxes, especially when aggregated to weekly or longer time scales, may have a strong element of predictability. This is because at some time scales, integrated fluxes may be a function of substrate availability and "average" biophysical conditions (Valentine et al., 1994). When large aggregates of data on methane are examined, NPP emerges as a reasonable predictor, together with water table depth (Roulet et al., 1992; Valentine et al., 1994). Synthesis of nitrogen trace gas flux data from agricultural systems shows that fertilizer losses to N20 lie within a relatively narrow range as a percentage of fertilizer applied (Cole et al., 1996; Nevison et al., 1996; Nevison and Holland, 1997). In natural ecosystems, Matson et al. (1990) have shown that N20 fluxes scale with N availability within ecosystems across both macrogradients and smaller topographic gradients. This contrast, of extreme variability at some scales, and an unexpected degree of predictability at certain other scales, is not uncharacteristic of geophysical systems. For example, weather becomes unpredictable beyond a few days, but the seasonal cycle is highly predictable. In complex dynamical systems, it can be extremely fruitful to look for scales at which large-scale constraints provide a measure of predictability. To explore this, a version of the CENTURY ecosystem model (Parton et al., 1995; Schimel et al., 1996) was integrated globally and to analyze relationships between environmental constraints and trace gas fluxes. The CENTURY model has been described extensively elsewhere (Parton et al., 1993; 1995; Schimel et al., 1996; 1997). Trace gas fluxes are simulated as a consequence of two
198
D.S. Schimel and N.S. Panikov
processes. First, a fraction of N mineralization is lost before being added to the soil inorganic N pool. Conceptually, this represents N2 and N20 production from nitrification. Second, a fraction of the standing pool of nitrate is assumed to be lost, an assumption representing denitrification. Generally, N gas losses resulting from "nitrification" exceed those from "denitrification". The trace gas flux coefficients are calibrated against field observations and against the model of Parton et al. (1996). N gas losses are not partitioned into separate species (N20, N2, and NO) and so this submodel captures the effects of trace gas flux on ecosystem N budgets, but is not very useful for understanding atmospheric consequences, except in a qualitative way. As will be noted below, this model, although very simple, is fairly accurate relative to observations at large spatial scales. Results in Figure 5 show patterns of N cycling from a global CENTURY simulation. In the long-term, the amount of nitrogen cycling in an ecosystem is controlled by the balance between inputs, which in CENTURY are simulated as a function of atmospheric inputs and biological N fixation (Ojima et al., 1994; Schimel et al., 1997), and losses to trace gases and leaching. Figure 5a shows that trace gas losses scale with N inputs, but that there are differences between major ecosystem types. Specifically, savanna and grassland ecosystems (coded as grey and light grey) have lower than-expected losses relative to inputs. This occurs not because of aridity, but because of substantial simulated losses to fire and herbivory (Ojima et al., 1994; Schimel et al., 1997). In systems with frequent disturbance, less N is retained and cycled as a fraction of inputs, and with reduced N turnover, trace gas losses are also reduced. Thus, while trace gas losses generally scale with N inputs, the relationship is quantitatively influenced by ecosystem type-specific disturbance patterns. We normally think of N gas losses as being strongly controlled by soil moisture and temperature (e.g. Li et al., 1996; Parton et al., 1996). Figure 5b shows the expected relationship between biophysical controls and gas losses. As evapotranspiration (ET) increases, trace gas losses increase. Note that ET is not a direct but an indirect control over trace gas fluxes, as ET itself increases in general as precipitation, soil moisture, and soil temperature increase (Schimel et al., 1997). This relationship shows an interesting structure. Both grassland (light grey) and savanna (dark grey) points show a bifurcation, reflecting the differences in the typical N content of litter in tallgrass (usually tropical C4, wide C:N ratios) and shortgrass (C3/C4 mix: narrower C:N ratios) grasslands and savannas. In tallgrass systems, lower litter quality results in less N mineralization (and hence lower trace gas losses) compared to shortgrass systems under similar ET conditions. Thus, while N gas losses are predicted to scale with moisture and temperature, the slope of the relationship will be influenced by ecosystem-specific conditions affecting the microbial mineralization-immobilization balance. Finally, Figure 5c shows N gas losses as a function of N mineralization rate. This figure also shows regressions computed from Matson and Vitousek (1990) based on data from the Amazon basin. N gas losses appear as a relatively simple function of N turnover for forest and some grassland/savanna ecosystems. The l esults appear in reasonable agreement with similar relationships scaled from observations, subject to assumptions about the ratio of N20 to N20+N2+NO, the quantity this version of CENTURY simulates (Schimel et al., 1997). Note however, that several ecosystem types have notably higher than expected fluxes. These are productive but highly seasonal grassland and savanna ecosystems. These systems appear in CENTURY as having high (approximating 5% of N turnover) N gas losses, corresponding to ecosystems that have observed high NO and possibly NH4 fluxes (Martin et al., 1998). In seasonal ecosystems with high evaporative demand, rapidly fluctuating soil moisture conditions appear to favour high rates of NO flux. Clearly, additional observations and process studies are required to test the relationships simulated in this modelling study. In conclusion, although CENTURY simulates a high degree of variability from cell to cell
Simulation models o f terrestrial trace gas fluxes at soil microsites to global scales
Figure 5.
199
(a) Sum of annual N gas fluxes (N2+NO+N20) vs N inputs from a global CENTURY model simulation. N inputs result from wet and dry deposition and biological nitrogen fixation. Light-grey points are grassland ecosystems, dark grey points are forests and black points are "mixed" ecosystems such as savannas. (b) N trace gases vs evapotranspiration (ET). (c) N trace gases vs annual N mineralization (NMIN). Lines indicate regressions computed from Matson and Vitousek (1990) based on data from the Amazon Basin. Trace gas fluxes, N input and N mineralization in g N m 2 yr-~; ET in cm yr ~.
200
D.S. Schimel cnd N.S Panik.~v
in biogeochemical processes, several large-scale patterns also emerge clearly and can be considered hypotheses. Large-scale patterns of trace gas flux should scale with large-scale constraints on the N cycle. These constraints include controls over N inputs, biophysical constraints over microbial activity, and overall N turnover. The model also suggests that these constraints cannot be considered independently of ecosystem or even species-specific properties of vegetation that influence nutrient cycling (litter quality, plant N uptake patterns). Large-scale monthly-to-annual patterns of trace gas fluxes may be predictable from a fairly small set of physical and ecosystem characteristics. Testing this will require a combination of the spatial sampling approaches advocated in Matson and Vitousek (1990), to determine climatological and ecosystem-specific effects with the long-term sampling strategy of Mosier et al. (1996, 1997) to determine effects of interannual climate variability. This type of research aimed at large scale controls is a crucial complement to organism and physiological-level investigations.
References Bowers, P. and R. Noyes (1985) Chemical oscillations associated with gas formation. In: R. Field and M. Burger (Eds.) oscillations and travelling waves in chemical systems, Wiley and Sons, New York, pp. 511-531. Buffiere, P., J.-P. Steyer, C. Fonade and R. Moletta (1995) Comprehensive modelling of methanogenic biofilms in fluidizied bed systems: mass transfer limitations and multisubstrate aspects. Biotechnology and Bioengineering 48: 725-36. Cao M., J.B. Dent and O.W. Heal (1995) Modelling methane emissions from rice paddies. Global Biogeochemical Cycles 9:183-95. Christensen, T.R., I.C. Prentice, J. Kaplan, A. Haxeltine and S. Sitch (1996) Methane flux from northern wetlands and tundra: an ecosystem source modelling approach. Tellus 48B:651-660. Cole, V., C. Cerri, K. Minami, A. Mosier, N. Rosenberg, D. Sauerbeck, J. Dumanski, J. Duxbury, J. Freney, R. Gupta, O. Heinemeyer, T. Kolchugina, J. Lee, K. Paustian, D. Powlson, N. Sampson, H. Tiessen, M. van Noordwijk and Q. Zhao (1996) Agricultural options for mitigation of greenhouse gas emissions. In: R.T. Watson, M.C. Zinyowera, R.H. Moss and D.J. Dokken (Eds.) Climate Change 1995 - Impacts, Adaptations arm Mitigation of Climate Change." Scientific-Technical Analyses. Contribution of Working Group II to the Second Assessment Report of the Intergovernmental Panel on Climate Change. Cambridge University Press, Cambridge, UK., pp. 745-771. Cornish-Bowden, A. (1976) Principles of enzyme kinetics. Butterworths, London, Boston. Crutzen, P.J. and P.H. Zimmerman (1991) The changing photochemistry of the troposphere. Tellus A and B 43:136-151. Dedysh, S.N. and N.S. Panikov (1997) Effect of methane concentration on the rate of its bacterial oxidation in a Sphagnum peat. Microbiology 66:563-68 (English translation from Mikrobiologiya) Flanagan, W., L.T. Ramsey and E. Kostlan (1993) Annual CO2 emission from forest floors predicted by simulation models including climate change. In: K. Stamnes (Ed.) Atmospheric Radiation. Proceedings of the International Society for Optical Engineering 2049: 37-55. Hunt, H.W., D.C. Coleman, E.R. Ingham,, R.E. Ingham, E.T. Elliott, J.C. Moore, I. Rose, C. Reid and C.R. Morly (1987) The detrital food web in a shortgrass prairie. Biology and Fertility of Soils 3:57-68. Kompala, D.S., D. Ramkrishna, N.B. Jansen, G.T. Tsao (1986) Investigation of bacterial growth on mixed substrates: experimental evaluation of cybernetic models. Biotechnology and Bioengineering 28:1044-1055. Li, C., V. Narayanan and R.C. Harriss (1996) Model estimates of nitrous oxide emissions from agricultural lands in the United States. Global Biogeochemical Cycles 10:297-306.
Simulation models of terrestrial trace gasfluxes at soil microsites to global scales
201
Martin, R.E., M.C. Scholes, A.R. Mosier, D.S. Ojima, E.A. Holland and W.J. Parton (1998) Controls on annual emissions of nitric oxide from soils of the Colorado shortgrass steppe. Global Biogeochemical Cycles 12:81-91. Matson, P.A. and P.M. Vitousek (1990) Ecosystem approach to a global nitrous oxide budget. BioScience 40: 667-672. Matson, P.A., P.M. Vitousek, G.P. Livingston and N.A. Swanberg (1990) Sources of variation in nitrous oxide flux from Amazonian ecosystems. Journal of Geophysical Research 95:1678916798. Middelburg, J.J., P.S. Liss, F.J. Dentener, T. Kaminski, C. Kroeze, J.-P. Malingreau, M. Nov~ik, N. Panikov, R. Plant, M. Starink and R. Wanninkhof (1999) Relations between scale, model approach and model parameters. In: A.F. Bouwman (Ed.) Scaling of trace gas fluxes. Developments in Atmospheric Science 24. Elsevier, Amsterdam, pp. 217-232. Mosier, A.R., W.J. Parton, D.W. Valentine, D.S. Ojima, D.S. Schimel and J.A. Delgado (1996) CH4 and N20 fluxes in the Colorado shortgrass steppe: 1. Impact of landscape and nitrogen addition. Global Biogeochemical Cycles 10:387-399. Mosier, A.R., W.J. Parton, D.W. Valentine, D.S. Ojima, D.S. Schimel and O. Heinemeyer (1997) CH4 and N20 fluxes in the Colorado shortgrass ~teppe 2. Long-term impact of land use change. Global Biogeochemical Cycles 11:29-42. Nevison, C. and E.A. Holland (1997) A reexamination of the impact of anthropogenically fixed nitrogen on atmospheric N20 and the stratospheric O3 layer. Journal of Geophysical Research 102:25519-25536. Nevison, C.D., G. Esser and E.A. Holland (1996) A global model of changing N20 emissions from natural and perturbed soils. Climatic Change 32:327-378. Ojima, D.S., D.S. Schimel, W.J. Parton and C.E. Owensby (1994) Long- and short-term effects of fire on nitrogen cycling in tallgrass prairie. Biogeochemistry 24:67-84. Panikov, N.S. (1991) A synthetic chemostat model as a means of describing the complex dynamic behavior of microorganisms. Microbiology 60:299-307 (English translation from Mikrobiologiya). Panikov, N.S. (1994) Response of soil microbial community to global warming: simulation of seasonal dynamics and long-term succession in typical tundra. Microbiology 63:389-404 (English translation from Mikrobiologiya). Panikov, N.S. (I995) Microbial Growth Kinetics. Chapman and Hall, London, Glasgow. Panikov, N.S. (1996) Mechanistic Mathematical Models of Microbial Growth in Bioreactors and in Natural Soils: Explanation of Complex Phenomena. Mathematics and Computers in Simulations 42:179-86. Panikov, N.S., M.V. Sizova, V.V. Zelenev, G.A. Machov, A.V. Naumov and I.M. Gadzhiev (1995) Methane and carbon dioxide emission from several Vasyugan wetlands: spatial and temporal flux variations. Ecological Chemistry 4:13-23. Parton, W.J., J.M.O. Scurlock and D.S. Ojima (1993) Observations and modeling of biomass and soil organic matter dynamics for the grassland biome worldwide. Global biogeochemical cycles 7:785809. Parton, W.J., D.S. Ojima, C.V. Cole and D.S. Schimel (1994) A general model for soil organic matter dynamics: Sensitivity to litter chemistry, texture and management. In: R.B. Bryant and R.W. Arnold (Eds.) Quantitative Modeling of Soil Forming Processes. Soil Science Society of America, Madison, WI., pp. 147-167. Parton, W.J., D.S. Schimel, C.V. Cole and D.S. Ojima (1987) Analysis of factors controlling soil organic matter levels in Great Plains grasslands. Soil Science Society of America Journal 51:11731179. Parton, W.J., A.R. Mosier, D.S. Ojima, D.W. Valentine, D.S. Schimel, K. Weier and A.E. Kulmala (1996) Generalized model for N2 and N20 production from nitrification and denitrification. Global Biogeochemical Cycles 10:401-412. Parton, W.J., J.M.O. Scurlock, D.S. Ojima, D.S. Schimei, D.O. Hall and SCOPEGRAM Group Members (1995) Impact of climate change on grassland production and soil carbon worldwide. Global Change Biology 1:13-22.
202
D.S. Schimel and N.S. Panikov
Paustian, K. (1994) Modelling soil biology and biochemical processes for sustainable agricultural research. In: C.E. Pankhurst, B.M. Doube, V.V.S.R. Gupta and R. Grace (Eds.) Soil Biota. Management in Sustainable Farming Systems. CSIRO Information Services, Melbourne, pp. 182193. Paustian, K., O. Andr~n, M. Clarholm, A.-C. Hansson, G. Johansson, J. Lagerl6f, T. Lindberg, R. Pettersson and B. Sohlenhuis (1990) Carbon and nitrogen budgets of four agro-ecosystems with annual and perennial crops, with and without N fertilisation. Journal of Applied Ecology 27:60-84. Powell, E.O. (1967) The growth rate of microorganisms as a function of substrate concentration. In: E.O. Powell, C. Evans, R. Strange and D Tempest (Eds.) Continuous cultivation of microorganisms, H.M. Stationary Office, Salisbury, pp. 34-55. Roulet, N., T. Moore, J. Bubier and P. Lafleur (1992) Northern fens: methane flux and climate change. Tellus 44B: 100-105. Schimel, D.S. (1986) Carbon and nitrogen turnover in adjacent grassland and cropland ecosystems. Biogeochemistry 6:239-243. Schimel, D.S. and C.S. Potter (1995 ) Process modelling and spatial extrapolation. In: P. Matson and R. Harriss (Eds.) Biogenic Trace Gases." Measuring Emission from Soil and Water. Blackwell, Oxford, pp. 358-383. Schimel, D.S., B.H. Braswell, R. McKeown, D.S. Ojima, W.J. Parton and W. Pulliam (1996) Climate and nitrogen controls on the geography and time scales of terrestrial biogeochemical cycling. Global Biogeochemical Cycles 10:677-692. Schimel, D.S., B.H. Braswell Jr., E.A. Holland, R. McKeown, D.S. Ojima, T.H. Painter, W.J. Parton and J.R. Townsend (1994) Climatic, edaphic, and biotic controls over storage and turnover of carbon in soils. Global Biogeochemical Cycles 8:279-293. Schimel, D.S., VEMAP Participants and B.H. Braswell (1997) Continental scale variability in ecosystem processes: models, data, and the role of disturbance. Ecological Monographs 67:251-271. Stams, A.J.M. (1994) Metabolic interactions between anaerobic bacteria in methanogenic environments. Antony van Leeuwenhoek 66:271-94. Turner, B.G., D. Ramkrishna and N.B. Jansen (1989) Cybernetic modeling of bacterial cultures at low growth rates: single-substrate systems. Biotechnology and Bioengineering 34:252-261. Valentine, D.W., E.A. Holland and D.S. Sch:mel (1994) Ecosystem and physiological controls over methane production in northern wetlands. Journal of Geophysical Research 99:1563-1571. Walter, B.P., M. Heimann, R.D. Shannon and J.R. White (1996) A process-based model to derive methane emission from natural wetlands. Geophysical Research Letters 23:3731-3734. Westerman, P. (1994) The effect of incubation temperature on steady-state concentration of hydrogen and volatile fatty acids during anaerobic degradation in slurries from wetland sediments. FEMS Microbiology Ecology 13:295-302. Zavarzin, G.A., V.V. Kalashnikov, V.V. Kevbrin and S.T. Petrov (1990) Simulation of methanogenic community. Proceedings USSR Academy of Sciences, Ser. Biology 1:116-124.
Chapter 10
THE A P P L I C A T I O N OF C O M P E N S A T I O N POINT C O N C E P T S IN SCALING OF FLUXES
R. Conrad and F.J. Dentener
This Page Intentionally Left Blank
Approaches to scaling a trace gas fluxes in ecosystems A.F. Bouwman,editor 9 ElsevierScience B.V. All rights reserved
THE APPLICATION OF COMPENSATION POINT CONCEPTS IN SCALING OF FLUXES
R. Conrad 1 and F.J. Dentener 2 t Max-Planck-Institutflir terrestrischeMikrobiologie,Karl-von-Frisch-Strasse,D-35043 Marburg, Germany 2 Utrecht University, Institute for Marine and Atmospheric Research, Princetonplein 5, 3584 CC Utrecht, the Netherlands
I. Introduction Exchange of trace gases between the biosphere and atmosphere has traditionally been described in models by one-directional fluxes. Generally either emission or uptake by soils and vegetation through dry deposition wer,, taken into account. However, fo., many trace gases of biogenic origin the situation is more complicated, as uptake and emission may occur simultaneously. The direction of the net flux depends on the environmental factors regulating the consumption and production of the trace gas considered. The compensation point is a useful concept for understanding atmosphere-biosphere exchange of trace gases and for describing the regulators of fluxes in models. Compensation points have been described for photosynthetic assimilation of carbon dioxide (CO2), where the compensation point characterizes a condition under which CO2 assimilation is compensated by simultaneous respiratory CO2 production, resulting in zero net CO2 exchange between plant and atmosphere. Similarly, for trace gases the compensation concentration is defined as the concentration for which the uptake reaction is equal to the simultaneously operating production rate, and the net flux is zero. For most situations the uptake reaction can be best described by first order kinetics whereby uptake is proportional to the trace gas concentration in the air, whereas the production is independent of the concentration of the trace gas (Conrad, 1994). This can be described as follows: J=i'-kxm
P = k x mc if J = 0
(1) (2)
where J = net flux; P = production rate; k = first order uptake rate constant or exchange velocity; m = ambient (atmospheric) concentration (at a specified reference height); m c compensation concentration. Thus, the compensation concentration can be used to characterize the direction of the flux at a given ambient concentration. The flux is an emission (J > 0) when the ambient concentration is lower than the compensation concentration (m < mc), or a deposition (J < 0) when m > me. When the fluctuations in the ambient concentration overlap with the variations of compensation concentrations, the compensation concentration is a critical variable for the regulation of the flux. This is the case for many trace gases (Conrad, 1994). Knowledge of the compensation concentration allows parameterization of fluxes from simultaneous production and consumption processes, as me = P/k, thus providing information about the mechanisms that are ultimately responsible for the ensuing flux. If mc and either P or k are known, the resulting net flux can usually be calculated (provided that some additional parameters such as soil density and gas diffusivity are known). This has been shown for the
206
R. Conrad and F.J. Dentener
exchange of nitric oxide (NO) between s~';,1 and atmosphere (Galbally and Johansson, 1989; Remde et al., 1993; Rudolph et al., 1996b). Compensation points can be measured by combining appropriate micrometeorological techniques and concentration measurements. For example, aerodynamic gradient methods have been used to measure exchange fluxes of ammonia (NH3) between a plant canopy and the atmosphere. The atmospheric concentration at which there is no net exchange is then defined as the compensation concentration (e.g. Sutton et al., 1995). The compensation point concept provides a useful theoretical framework for measuring and modelling trace gas fluxes between the biosphere and the atmosphere. However, it must be carefully evaluated whether the mathematical formulations described above can be usefully applied, since production and consumption processes should be homogeneously distributed at the scale considered. For example, the concept has been successfully applied for dihydrogen (H2) exchange between legume fields and the atmosphere (Conrad and Seiler, 1980b), but the parameters measured are only valid at the scale of the field. They do not apply when only parts of the system (e.g. soil without plants) are considered (Conrad, 1996a), because production and consumption of H2 are spatially separated. For example, H2 is produced by N2-fixing bacteroids in plant root nodules, while H2 is consumed by H2-oxidizing bacteria and abiotic enzymes in the soil (Conrad and Seiler, 1981; Schuler and Conrad, 1991). Compensation points are found in soils, vegetation and water bodies (oceans/lakes). This paper addresses the use of compensation points in understanding measured atmosphere-biosphere exchange fluxes and describing those fluxes in models. We only consider compensation points for terrestrial systems.
2. Compensation concentrations in soil and vegetation 2.1. Occurrence of compensation concentrations The example of H2 exchange shows that the compensation concentration can be governed by a composite of soil and vegetation processes. For simplicity, we shall also consider those processes taking place in plant roots (such as H2 production by legumes) as soil processes. However, it is still necessary to differentiate between the pedosphere (soil) and the phyllosphere (active above ground biomass). The total mass of leaves and the leaf area index depend on the type of vegetation. Especially in forests with a dense canopy there is a stratified system in which a trace gas (e.g. NO) may undergo chemical reactions in the atmosphere below the canopy before it is emitted into the tropospheric boundary layer or deposited from the troposphere onto the canopy or soil (Duyzer et al., 1983; 1995; Kramm et al., 1991). In addition, the net flux between troposphere and biosphere may be affected by production and consumption processes occurring both in the soil and the plant canopy. The direction of the flux is determined by the compensation point, which may be different for soils and plants. Our current knowledge base of compensation points is compiled in Table 1 for different trace gases. In soil, compensation concentrations have been described for all the trace gases listed, except for methane (CH4) and ammonia (NH3). For NH3 we may assume that compensation concentrations do exist, since biogenic soil ammonium (NH4+) should readily equilibrate with soil hydroxyl anions (OH) to form NH3. Indeed measurements by Langford et al. (1992) suggest that at some locations soil NH4+ and soil pH are consistent with measured atmospheric NH3 concentrations. However, to our knowledge explicit measurements of soil NH3 compensation points are still lacking.
207
The a p p l i c a t i o n o f c o m p e n s a t i o n p o i n t c o n c e p t s in s c a l i n g o f f l u x e s
Table 1. Literature reports a demonstrating production and consumption processes in upland soils and plants (phyllosphere) and the existence of compensation concentrations. Trace gas
Soil Production
Hz
Consumption
Vegetation Compensation
Production
Consumption
Compensation
1,2
3,4
1,5
n.r.
n.r.
n.r.
CO CH4 OCS
6,7 14-16 20-22
3,8 17-19 23, 24
3,7 n.r. 24
9-11 n.r. 25
12,13 n.r. 26-28
n.r. n.r. 29
NzO NO
18,30,31 40,41
32-34 42-44
35 45-47
36-38 37.38,48-50
39 48-50
n.r. 50
NO2
51-53 58,59
51-53 60
52 n.r.
48,54,55 61,62
55-57 62,63
54,55,57 62,64
NH3
n.r. = not reported. a 1, Conrad and Seiler (1980b); 2, Schuler and Conrad (1991); 3, Liebl and Seiler (1976); 4, Conrad and Seiler (1981); 5, Conrad and Seiler (1979); 6, Conrad and Seiler (1982); 7, Conrad and Seiler (1985); 8, Conrad and Seiler (1980a); 9, Seiler et al. (1978); 10, Ltittge and Fischer (1980); 11, Tarr et al. (1995); 12, Bidwell and Bebee (1974); 13, Peiser et al. (1982); 14, Sexstone and Mains (1990); 15, De Groot et al. (1994); 16, Yavitt et al. (1995); 17, King (1992); 18, Conrad (1995); 19, Conrad (1996a); 20, Bremner and Steele (1978); 21, Aneja et al. (1979); 22, Adams et al. (1981); 23, Castro and Galloway (1991); 24, Lehmann and Conrad (1996); 25, Feng and Hartel (1996); 26, Taylor et al. (1983); 27, Brown and Bell (1986); 28, Protoschill et al. (1996); 29, Kesselmeier and Merk (1993); 30, Bouwman (1990); 31, Granli and Bockman (1994); 32, Ryden (1981); 33, Slemr et al. (1984); 34, Donoso et al. (1993); 35, Seiler and Conrad (1981); 36, Weathers (1984); 37, Dean and Harper (1986); 38, Klepper (1987); 39, Lensi and Chalamet (1981); 40, Davidson (1991); 41, Williams et al. (1992); 42, Conrad (1996b); 43, Rudolph et al. (1996a); 44, Dunfield and Knowles (1997); 45, Johansson and Galbally (1984); 46, Remde et al. (1989); 47, Rudolph et al. (1996b); 48, Johansson (1989); 49, Weber and Rennenberg (1996a); 50, Wildt et al. (1997); 51, Slemr and Seiler (1984); 52, Slemr and Seiler (1991); 53, Baumg~h'-tneret al. (1992); 54, Rondon et al. (1993); 55, Weber and Rennenberg (1996b); 56, Segschneider et al. (1995); 57, Thoene et al. (1996); 58, Fenn and Hossner (1983); 59, Milchunas et al. (1988); 60, Buijsman and Erisman (1988); 61, Sutton et al. (1995); 62, Mattsson and Schjoerring (1996); 63, Hutchinson et al. (1972); 64, Husted et al. (1996).
The situation is more complicated for CH4, since consumption processes occur in deeper soil layers, whereas production processes occur in the surface soil (see Conrad, 1996a). Due to differences in the vertical distribution of CH4 production and consumption, it is difficult to attribute a compensation concentration which would be useful at the field scale. More research is needed on this subject. In the phyllosphere, both production and consumption processes have been described for all the trace gases except for H2 and CH4 (Table 1). Hence, plants probably do not interfere with the fluxes of H2 and CH4 in upland sites with well aerated soils. This is different in wetlands, where the vascular system of aquatic plants may serve as a venting system for the emission of H2 and C H 4 ( a n d of other trace gases, such as nitrous oxide, N20) from the submerged soil into the atmosphere (Conrad, 1995; 1996a). Plant compensation concentrations for carbon monoxide (CO) have not yet been measured, but they should theoretically exist, since plants can both produce and consume CO (Table 1). The reason why, so far, detection of CO compensation points has failed is that CO exchange with plants has generally been measured in flushed chambers. The CO concentration observed at the outlet of the chamber was probably much less than the compensation concentration which would have been be reached in a closed system at steady state between production (a light-dependent reaction) and consumption (a dark reaction). The concentration increase of CO caused by the plants was always <100 ppbv CO relative to ambient CO (Seiler et al., 1978; Tarr et al., 1995). Reports on production and consumption of N20 in plants are scarce and generally restricted to laboratory experiments, partially using cellular and subcellular systems (Table 1). In con-
208
R. Conrad and F.J. Dentener
Table 2. Typical ambient concentrations and compensation concentrations observed for soil and vegetation for selected trace gases. Trace gas
Ambient concentration
Compensation concentration ( p p b v )
(ppbv)
HE CO OCS
N20 NO
NO2,NOx NH3
500 - 700 70 - 170 0.5 310 <0.1 - 16 <0.1 - 24 0.1 - 13
Soil
Vegetation
<5 - 900 <5 - 1200 15 - 70 >500 0.3 - > 1 0 0 0 <0.1 - 16 n.r.
n.r. 70 - > 2 5 0 0 - 0.5 n.r. 0.3 - 3 0.1 - 1.6 1- 9
References for the individual trace gases are indicated in Table 1. n.r. = not reported.
trast, plants were found to be inert with respect to N 2 0 under field conditions (Seiler et al., 1978). However, more measurements are required, because only a limited number of plants has been tested. Plant compensation concentrations have been described for all the other trace gases listed in Table 1, i.e. NO, N O 2 and carbonyl sulphide (OCS). Metabolism of NO in plants has only recently been demonstrated (Wildt et al., 1997). Compensation points for atmospheric trace gases other than those listed in Table 1 may exist also. Likely candidates are dimethyl sulfide and volatile organic compounds. 2.2. Magvitude of compensation concentrations
For most of the trace gases listed in Table 2, the ranges in compensation concentration overlap with their atmospheric concentrations at the Earth's surface. This is not unexpected, given the important role of soils and vegetation in determining the atmospheric concentrations and budgets of these gases. Changes in compensation concentrations due to changes in environmental conditions could be critical for the direction and magnitude of the ensuing flux. At a first glance there are only two exceptions, i.e. N 2 0 and OCS, which have much higher compensation points than the atmospheric concentrations. However, as discussed below, the range of compensation concentrations will probably have to be revised as new research data are published. The compensation concentrations of N20 seem to be significantly higher than the ambient ones. This is in agreement with the observation that most soils are a source for atmospheric N20. However, an increasing number of studies (see Table 1) report occasional N 2 0 consumption by soils, suggesting that during uptake the compensation concentrations are lower than the ambient ones, i.e. <310 ppbv N20. There is only one study in which the existence of N 2 0 compensation concentrations in soil was demonstrated (Seiler and Conrad, 1981), and the soil studied, in fact, was acting as a net source for atmospheric N20. The compensation concentrations which Lehmann and Conrad (1996) reported for OCS were much higher than the ambient ones. Soils may contain two OCS consumption activities, one being dominant at lower OCS concentrations than the other. As the lower range activity was not covered by the measurements of Lehmann and Conrad (1996), the reported compensation concentration may have been overestimated (this was confirmed by recent experiments by J. Kesselmeier, 1997, personal communication, and our laboratory). These experiments indicated that environmentally relevant soil compensation concentrations are
The application of compensation point concepts in scaling offluxes
209
probably <2 ppbv and partially overlapping with the ambient concentration of 0.5 ppbv OCS. Considering the existence and the magnitude of compensation concentrations among soil and vegetation strata (Tables 1 and 2), one may conclude that the flux of H2 and N20 is controlled only by the soil and not by the vegetation, whereas fluxes of all other trace gases are regulated by both soil and vegetation. Moreover, in general the compensation concentrations of vegetation are somewhat lower than those observed for soils. This seems to be especially the case for OCS, NO and NOz/NOy. Also it is not unlikely for NH3, although there are no explicit studies of NH3 compensation concentrations for soils. Indeed, Langford et al. (1992) showed that atmospheric NH3 concentrations were mostly lower than those in equilibrium with soil ammonium, indicating a lower compensation point of the canopy than that of the soil. For CO, on the other hand, it may well be that plants under daylight conditions have much higher compensation concentrations than those indicated in Table 2, and that these values are much higher than those found for soils. If the compensation concentration for a given trace gas in the soil is significantly different from that for the plant canopy, the exchange of the trace gas between the biosphere (i.e. soil plus vegetation) and the atmosphere is a complex process. For example, OCS compensation concentrations are 2.0 ppbv for the soil and 0.1 ppbv for the vegetation at an atmospheric (above vegetation) OCS concentration of 0.5 ppbv. In this case OCS would be simultaneously deposited from the troposphere onto the vegetation, and emitted from the soil. A significant part of the emitted OCS would also be taken up by the vegetation. Hence, the total soil plus vegetation system could still act as a sink for atmospheric OCS. The direction of the net flux, however, will depend on the complex interaction of the compensation points and time scales involved with turbulent transport through the canopy and exchange with soil and vegetation. The OCS concentration in the atmosphere below the canopy would be higher than that above the canopy and the apparent compensation concentration of the total system would be higher than that of the vegetation alone. Alternatively, compensation concentrations of CO are 50 ppbv for soil and 500 ppbv for the vegetation, at an ambient CO concentration of 200 ppbv, would lead to simultaneous emission fi'om the vegetation into the troposphere and deposition onto the soil. Overall, the total soil-vegetation system would probably act as a source for atmospheric CO, despite the sink activity of soil. The apparent compensation concentration of the soil-vegetation system would probably be lower than that of the vegetation alone. Again, the interaction of soil and vegetation depends on the actual conditions in the field, in particular on the density of the vegetation and the gas transport within the canopy. Gas exchange between vegetation and the atmosphere also depends on (photo-)chemical reactions occurring in the atmosphere below the canopy. For example, NO and NOz/NOy are in photochemical equilibrium, but their concentrations are also affected by production and consumption processes occurring in the soil or the plants. Consequently, sophisticated micrometeorological/chemical models are required to estimate the net flux (Kramm et al., 1991; Duyzer et al., 1983; 1995). The magnitude of the flux is determined by the difference between compensation concentration and atmospheric concentration. Hence, fluxes of H2, OCS and N20 are controlled by the processes regulating the compensation concentration, since the ambient concentrations of these trace gases are relatively constant, due to their long atmospheric lifetimes. However, the ambient concentrations of CO and especially of NO, NOz/NOy and NH3 are highly variable. Local sources, such as automobiles, may create temporal and spatial fluctuations in the atmospheric concentrations of these gases. These may be much larger than variations in the compensation concentrations, causing major fluctuations in the net flux. In the latter case compensation concentrations still contribute to the control of the magnitude and
210
R. Conrad and F.J. Dentener
Table 3. Likely effects of environmental conditions on the compensation concentrations in soil and vegetation. Condition/event
H2
CO
OCS
Light Temperalure Precipitation N-Fertilization Soil type
+ +/+/+/-
+ + ? +/-
? ? 9 9 +/-
Light Temperature Precipitation N-Fertilization Plant type
n.a. n.a. n.a. n.a. n.a.
+ + 9 '~ +/-
? ? ? ? +/-
N20
NO
NO2/NOy
NH3
0 +/+/_ + +/-
9 +
0 +
+ +/-
+ +/-
+ +/-
+ + + +/-
Soil 0 + + + +/Plants n.a. n.a. n.a. n.a. n.a.
+ ? ? + +/-
References for the individual trace gases are indicated in Table 1. +, increase; - , decrease; 0, no effect; ?, not known; n.a., not applicable.
direction of fluxes. It is important, therefore, to know how and on which temporal and spatial scales the compensation concentration is influenced by the environmental conditions.
2.3. Environmental control of compensation concentrations Some environmental conditions that affect compensation concentrations change relatively slowly over time, whereas other conditions such as light intensity and temperature can vary rapidly. Rain and fertilizer applications are short-term events, but their effects may extend over prolonged periods. Microbial populations and plant biomass, on the other hand, often change only with the season, while some of the soil characteristics, such as pH, bulk density, humus content, and clay content, are virtually constant in time, at least at time scales shorter than several months. Likely effects of environmental variables on the compensation concentrations in soil and vegetation are listed in Table 3. There may be other environmental factors influencing the compensation points, but those listed in Table 3 are the only ones which are available from meteorological forecast models, observations and other databases. Other factors, such as soil moisture, nutrient availability, primary productivity, plant transpiration, are implicitly related to the factors presented in Table 3. Most of the effects listed have not been measured directly, but have been deduced (via mc = P / k) from our knowledge of the behaviour of the production and consumption processes in soil and plants (see references in Table 1).
3. Compensation points and scaling of fluxes As discussed in the previous section, the fluxes of H 2 , O C S , NzO (and C H 4 ) c a n be evaluated relatively easily. These gases have long atmospheric lifetimes and rather constant ambient concentrations. Therefore, it is sufficient to evaluate the magnitude, temporal and spatial variations of the compensation points and exchange velocities, to estimate fluxes to and from the atmosphere. However, for the other trace gases it is necessary to assess the complex cooccurrence of atmospheric and compensation concentrations; atmospheric models are
The application of compensation point concepts in scaling offluxes
211
probably the only tool for estimating the associated exchange fluxes. In some cases, these models should include a parameterization of turbulent transfer through the canopy, and chemistry taking place within the canopy. The amount of living biomass above the surface is one of the determining factors for this canopy-exchange parameterization. A further point of concern is the spatial scale at which the flux estimate is to be made, relative to the spatial scale at which a compensation concentration can be applied. For example, in the case of soil NO emissions there may be a large amount of small hot-spots, exhibiting rather high compensation concentrations in comparison with the atmospheric concentrations. In such a case it is necessary to determine the fraction of the area (A) covered by these hot spots together with the exchange velocity, to evaluate the area average emission flux FE: Fz= A x k x (mc-m)
(3)
Models should describe this emission flux as a boundary condition, and apply their traditional deposition flux FD calculation only to the fraction of the area not covered by these emitting spots:
FD=(I-A) x k x (m-mc)
(4)
A similar situation applies to NH3 emission following the decomposition of animal manure. On a very limited area, where animal manure is stored or where animal excreta are deposited in the field, a high compensation concentration is maintained by bacterial decomposition of the manure. Therefore, in practice it is not useful to apply NH3 compensation points to calculate fluxes from animal manure. The application of compensation points to calculate (bi-directional) fluxes to or from the atmosphere is probably most useful when the compensation concentrations overlap with the range of atmospheric concentrations (see Table 2), and the areal extent of soil and vegetation causing compensation concentrations corresponds to the area for which a flux estimate has to be made. In the case of atmospheric models this area is associated with spatial scales of 10-100 km for regional scale models, and 100-500 km for global scale models (Sofiev, 1999). If the factors controlling the compensation points are distributed heterogeneously within the model area, a parameterization has to be developed. The example of a simple parameterization given above is using separate and independent emission and deposition fluxes. The resolution and availability of the data used to calculate compensation points can be a limiting factor. At present most data (e.g. from meteorological forecast models) are available on spatial scales of about 50 km. In some areas the real resolution may be lower, since the meteorological variability of these models may partly depend on the variability of surface characteristics. Strong fluctuations of e.g. precipitation occur on even smaller scales, but the influence of these fluctuations on compensation points may be levelled out by a slow response of the soils and vegetation to this pulsing. More research is needed on this topic. Another point of concern is the non-linearity of the response of processes to a certain forcing. For example, use of an average temperature assumed to be valid for a larger region will lead to erroneous results in case of a non-linear response to temperature of a compensation point. Finally, in many parts of the world databases concerning soil and vegetation characteristics are still inaccurate (Bouwman et al., 1999), and they may not be adequate to describe the basic parameters needed for estimating bi-directional fluxes. It may be easier to describe the temporal variability of fluxes using compensation points than their spatial heterogeneity, although spatial and temporal variability are often closely related. Meteorological models provide output on time scales of hours, which may be sufficient to describe the temporal variability of compensation points. Early attempts to apply
212
R. Conrad and F.J. Dentener
compensation points in models (e.g. Dentener and Crutzen, 1994) have shown that bidirectional exchange fluxes can be important on global scales, but failed to address the uncertainties associated with spatial and temporal heterogeneity. Given the constraints mentioned above, a parameterization of compensation points applicable in atmospheric transport models should be a function of static factors (e.g. soil characteristics) and fluctuating factors dependent on variables available in meteorological models (e.g. temperature, radiation and rainfall).
4. Conclusions Measurements have shown that compensation points exist for a host of biogenic trace gases in soils and vegetation. Compensation points generally are in the range of concentrations measured in the atmospheric mixing layer. For some trace gases, soils and vegetation may have different compensation concentrations. The resulting magnitude and direction of the atmosphere-biosphere exchange flux will depend on the interaction of the respective compensation points and the rate of turbulent transfer through the canopy. This situation may be complicated by atmospheric chemistry occurring between the soil and the canopy. Models of gaseous transport through the canopy and exchange with the boundary layer are needed to adequately describe this exchange process. Compensation points vary with environmental factors such as light, temperature, precipitation and soil and vegetation characteristics. In most cases, we know the direction of these responses, but the possibilities for wider application of this small amount of measurements need to be assessed. Theoretical models describing the fundamental processes governing compensation points are needed to permit extrapolation of the available measurements. Major advances in applying compensation points for scaling of fluxes are probably to be made in this area. Application of compensation points in models is most useful when the compensation concentrations are in the same range as the atmospheric concentrations, and the spatial scale of the soil and vegetation properties governing compensation concentrations is similar to the spatial scale of the atmospheric model. Ideally, the compensation point should be described as a function of soil and vegetation parameters and meteorological variables provided by the model.
References Adams, D.F., S.O. Farwell, M.R. Pack and E. Robinson (1981) Biogenic sulfur gas emissions from soils in eastern and southeastern United States. Journal of the Air Pollution Control Association 31:1083-1089. Aneja, V.P., J.H. Overton, L.T. Cupitt, J.L. Durham and W.E.Wilson (1979) Carbon disulphide and carbonyl sulphide from biogenic sources and their contributions to the global sulphur cycle. Nature 282:493-496. Baumg~irtn~,r, M., E. Bock and R. Conrad (19~2) Processes involved in uptake and release of nitrogen dioxide from soil and building stones into the atmosphere. Chemosphere 24:1943-1960. Bidwell, R.G.S. and G.P. Bebee (1974) Carbon monoxide fixation by plants. Canadian Journal of Botany 52:1841-1847.
The application of compensationpoint concepts in scaling offluxes
213
Bouwman, A.F. (1990) Exchange of greenhouse gases between terrestrial ecosystems and the atmosphere. In: A.F. Bouwman (Ed.) Soils and the Greenhouse Effect, Wiley and Sons, pp. 61127. Bouwman, A.F., R.G. Derwent and F.J. Dentener (1999) Towards reliable bottom-up estimates of temporal and spatial patterns of emissions of trace gases and aerosols from land-use related and natural sources. In: A.F. Bouwman (Ed.) Approaches to scaling of trace gas fluxes in ecosystems. Developments in Atmospheric Science 24. Elsevier, Amsterdam, pp. 1-26. Bremner, J.M. and C.G. Steele (1978) Role of microorganisms in the atmospheric sulfur cycle. Advances in Microbial Ecology 2:155-201. Brown, K.A. and J.N.B. Bell (1986) Vegetation - the missing sink in the global cycle of carbonyl sulphide (COS). Atmospheric Environment 20:537-540. Buijsman, E. and J.-W. Erisman (1988) Wet deposition of ammonium in Europe. Journal of Atmospheric Chemistry 6:265-280. Castro, M.S. and J.N. Galloway (1991) A comparison of sulfur-free and ambient air enclosure techniques for measuring the exchange of reduced sulfur gases between soils and the atmosphere. Journal of Geophysical Research 96:15427-15437. Conrad, R. (1994) Compensation concentraf~n as critical variable for regulatir~g the flux of trace gases between soil and atmosphere. Biogeochemistry 27:155-170. Conrad, R. (1995) Soil microbial processes involved in production and consumption of atmospheric trace gases. Advances in Microbial Ecology 14:207-250. Conrad, R. (1996a) Soil microorganisms as controllers of atmospheric trace gases (H2, CO, CH4, OCS, N20, NO). Microbiological Reviews 60:609-640. Conrad, R. (1996b) Metabolism of nitric oxide in soil and soil microorganisms and regulation of flux into the atmosphere. In: J.C. Murrell and D.P. Kelly (Eds.): Microbiology of Atmospheric Trace Gases: Sources, Sinks and Global Change Processes, Springer Verlag, pp. 167-203. Conrad, R. and W. Seiler (1979) Field measurements of hydrogen evolution by nitrogen-fixing legumes. Soil Biology and Biochemistry 11:689-690. Conrad, R. and W. Seiler (1980a) Role of microorganisms in the consumption and production of atmospheric carbon monoxide by soil. Applied and Environmental Microbiology 40:437-445. Conrad, R. and W. Seiler (1980b) Contribution of hydrogen production by biological nitrogen fixation to the global hydrogen budget. Journal of Geophysical Research 85:5493-5498. Conrad, R. and W. Seiler (1981) Decomposition of atmospheric hydrogen by soil microorganisms and soil enzymes. Soil Biology and Biochemistrv 13:43-49. Conrad, R. and W. Seiler (1982) Arid soils as a source of atmospheric carbon monoxide. Geophysical Research Letters 9:1353-1356. Conrad, R. and W. Seiler (1985) Influence of temperature, moisture and organic carbon on the flux of H2 and CO between soil and atmosphere. Field studies in subtropical regions. Journal of Geophysical Research 90:5699-5709. Davidson, E.A. (1991) Fluxes of nitrous oxide and nitric oxide from terrestrial ecosystems. In: J. E. Rogers and W.B. Whitman (Eds.).Microbial Production and Consumption of Greenhouse Gases." Methane, Nitrogen Oxides, and Halomethanes, American Society for Microbiology, pp. 219-235. Dean, J.V. and J.E. Harper (1986) Nitric oxide and nitrous oxide production by soybean and winged bean during the in vivo nitrate reductase assay. Plant Physiology 82:718-723. De Groot, C.J., A. Vermoesen and O. Van Cleemput (1994) Laboratory study of the emission of N20 and CH4 from a calcareous soil. Soil Science 158:355-364. Dentener, F.J. and P.J. Crutzen (1994) A global model of the ammonia cycle. Journal of Atmospheric Chemistry 19:331-369. Donoso, L., R. Santana and E. Sanhueza (1993) Seasonal variation of N20 fluxes at a tropical savannah site: soil consumption of N20 during the dry season. Geophysical Research Letters 20:1379-1382.
Dunfield, P. F. and R. Knowles (1997) Biological oxidation of nitric oxide in a humisol. Biology and Fertility of Soils 24:294-300.
214
R. Conrad and F.J. Dentener
Duyzer, J.H., G.M. Meyer and R.M. Van Aalst (1983) Measurement of dry deposition velocities of NO, NO2 and 03 and the influence of chemical reactions. Atmospheric Environment 17:2117-2120. Duyzer, J., H. Weststrate and S. Walton (1995) Exchange of ozone and nitrogen oxides between the atmosphere and coniferous forest. Water Air and Soil Pollution 85:2065-2070. Feng, Z. and P.G. Hartel (1996) Factors affecting production of COS and CS2 in Leucaena and Mimosa species. Plant and Soil 178:215-222. Fenn, L.B. and L.R. Hossner (1985) Ammonia volatilization from ammonium or ammonium-forming nitrogen fertilizers. Advances in Soil Science 1:123-169. Galbally, I.E. and C. Johansson (1989) A model relating laboratory measurements of rates of nitric oxide production and field measurements of nitric oxide emission from soils. Journal of Geophysical Research 94:6473-6480. Granli, T. and O.C. Bockman (1994) Nitrous oxide from agriculture. Norwegian Journal of Agricultural Science, Supplement 12: 7-128. Hutchinson, G.L., R.J. Millington and D.B. Peters (1972) Atmospheric ammonia: absorption by plant leaves. Science 175:771-772. Husted, S., M. Mattsson and J.K. Schjoerring (1996) Ammonia compensation points in two cultivars of Hordeum vulgare L during vegetative and generative growth. Plant, Cell and Environment 19:1299-1306. Johansson, C. (1989) Fluxes of NOx above s.Jll and vegetation. In: M.O. Andrea~ and D.S. Schimel (Eds.) Exchange of Trace Gases between Terrestrial Ecosystems and the Atmosphere. Wiley and Sons, pp. 229-246. Johansson, C. and I.E. Galbally (1984) Production of nitric oxide in loam under aerobic and anaerobic conditions. Applied Environmental Microbiology 47:1284-1289. Kesselmeier, J. and L. Merk (1993) Exchange of carbonyl sulfide (COS) between agricultural plants and the atmosphere - studies on the deposition of COS to peas, com and rapeseed. Biogeochemistry 23:47-59. King, G.M. (1992) Ecological aspects of methane oxidation, a key determinant of global methane dynamics. Advances in Microbial Ecology 12:431-468. Klepper, L.A. (1987) Nitric oxide emissions from soybean leaves during in vivo nitrate reductase assays. Plant Physiology 85:96-99. Kramm, G., H. Miiller, D. Fowler, K.D. H6fken., F.X. Meixner and E. Schaller (1991) A modified profile method for determining the vertical fluxes of NO, NO2, ozone, and HNO3 in the atmospheric surface layer. Journal of Atmospheric Chemistry 13:265-288. Langford, A.O., F.C. Fehsenfeld, J. Zachariassen and D.S. Schimel (1992) Gaseous ammonia fluxes and background concentrations in terrestrial ecosystems of the United States. Global Biogeochemical Cycles 6:459-483. Lehmann, S. and R. Conrad (1996) Characteristics of turnover of carbonyl sulfide in four different soils. Journal of Atmospheric Chemistry 23:193-207. Lensi, R. and A. Chalamet (1981) Absorption de l'oxyde nitreux par les parties aeriennes du mais. Plant and Soil 59:91-98. Liebl, K.H. and W. Seiler (1976) CO and H2 destruction at the soil surface. In: H.G. Schlegel; G. Gottschalk and N. Pfennig (Eds.) Microbial Production and Utilization of Gases, E. Goltze, G6ttingen, pp. 215-229. L~ittge, U. and K. Fischer (1980) Light-dependent net CO-evolution by C3 and C4 plants. Planta 149:59-63. Mattsson, M. and J.K. Schjoerring (1996) Characteristics of ammonia emission from barley plants. Plant Physiology and Biochemistry 34:691-695. Milchunas, D.G., W.J. Parton, D.S. Bigelow and D.S. Schimel (1988) Factors influencing ammonia volatilization from urea in soils of the shortgrass steppe. Journal of Atmospheric Chemistry 6:323340. Peiser, G.D., M.C.C. Lizada and S.F. Yang (1982) Dark metabolism of carbon monoxide in lettuce leaf discs. Plant Physiology 70:397-400.
The application of compensationpoint concepts in scaling offluxes
215
Protoschill-Krebs, G., C. Wilhelm and J. Kesselmeier (1996) Consumption of carbonyl sulphide (COS) by higher plant carbonic anhydrase (CA). Atmospheric Environment 30:3151-3156. Remde, A., J. Ludwig, F.X. Meixner and R. Conrad (1993) A study to explain the emission of nitric oxide from a marsh soil. Journal of Atmospheric Chemistry 17:249-275. Remde, A., F. Slemr and R. Conrad (1989) Microbial production and uptake of nitric oxide in soil. FEMS Microbiology Ecology 62:221-230. Rondon, A., C. Johansson and L. Granat (1993) Dry deposition of nitrogen dioxide and ozone to coniferous forests. Journal of Geophysical Research 98:5159-5172. Rudolph, J., M. Koschorreck and R. Conrad (1996a) Oxidative and reductive microbial consumption of nitric oxide in a heathland soil. Soil Biology and Biochemistry 28:1389-1396. Rudolph, J., F. Rothfuss and R. Conrad (1996b) Flux between soil and atmosphere, vertical concentration profiles in soil, and turnover of nitric oxide. 1. Measurements on a model soil core. Journal of Atmospheric Chemistry 23:253-273. Ryden, J.C. (1981) N:O exchange between a grassland soil and the atmosphere. Nature 292:235-237. Schuler, S. and R. Conrad (1991) Hydrogen oxidation in soil following rhizobial H2 production due to N2 fixation by a Vicia faba-Rhizobium leguminosarum symbiosis. Biology and Fertility of Soils 11:190-195. Segschneider, H.J., J. Wildt and H. F6rstel (1995) Uptake of 15NO2 by sunflower (Helianthus annuus) during exposures in light and darkness: Quantities, relationship to stomatal aperture and incorporation into different nitrogen pools within the plant. New Phytologist 131:109-119. Seiler, W. and R. Conrad (1981) Field measurements of natural and fertilizer induced N20 release rates from soils. Journal of the Air Pollution Control Association 31:767-772. Seiler, W., H. Giehl and G. Bunse (1978) The influence of plants on atmospheric carbon monoxide and dinitrogen oxide. Pure and Applied Geophysics 116:439-451. Sexstone, A.J. and C.N. Mains (1990) Production of methane and ethylene in organic horizons of spruce forest soils. Soil Biology and Biochemistry 22:135-139. Slemr, F. and W. Seiler (1984) Field measurements of NO and NO2 emissions from fertilized and unfertilized soils. Journal of Atmospheric Chemistry 2:1-24. Slemr, F. and W. Seiler (1991) Field study of environmental variables controlling the NO emissions from soil, and of the NO compensation points. Journal of Geophysical Research 96:13017-13031. Slemr, F., R. Conrad and W. Seiler (1984) Nitrous oxide emissions from fertilized and unfertilized soils in a subtropical region (Andalusia, Spain). Journal of Atmospheric Chemistry 1:159-169. Sofiev, M. (1999) Validation of model results on different scales. In: A.F. Bouwman (Ed.) Approaches to scaling of trace gas fluxes in ecosystems. Developments in Atmospheric Science 24. Elsevier, Amsterdam, pp. 233-255. Sutton, M.A., J.K. Schjoerring and G.P. Wyers (1995) Plant-atmosphere exchange of ammonia. Philosophical Transactions of the Royal Society, London, A 351:261-278. Tarr, M.A., W.L. Miller and R.G. Zepp (1995) Direct carbon monoxide photoproduction from plant matter. Journal of Geophysical Research 100:11403-11413. Taylor Jr., G.E., S.B. McLaughlin, D.S. Shriner and W.J. Selvidge (1983) The flux of sulphurcontaining gases to vegetation. Atmospheric Environment 17:789-796. Thoene, B., H. Rennenberg and P. Weber (1996) Absorption of atmospheric NO2 by spruce (Picea abies) trees .2. Parameterization of NO2 fluxes by controlled dynamic chamber experiments. New Phytologist 134:257-266. Weathers, P.J. (1984) N20 evolution by green algae. Applied and Environmental Microbiology 48:1251-1253. Weber, P. and H. Rennenberg (1996a) Exchange of NO and NO2 between wheat canopy monoliths and the atmosphere. Plant and Soil 180:197-208. Weber, P. and H. Rennenberg (1996b) Dependency of nitrogen dioxide (NO2) fluxes to wheat (Triticum aestivum L) leaves from NO2 concentration, light intensity, temperature and relative humidity determined from controlled dynamic chamber experiments. Atmospheric Environment 30:3001-3009.
216
R. Conrad and F.J. Dentener
Wildt, J., D. Kley, A. Rockel, P. Rockel and H.J. Segschneider (1997) Emission of NO from several higher plant species. Journal of Geophysical Research 102:5919-5927. Williams, E.J., G.L. Hutchinson and F.C. Fehsenfeld (1992) NOx and N20 emissions from soil. Global Biogeochemical Cycles 6:351-388. Yavitt, J.B., T.J. Fahey and J.A. Simmons (1995) Methane and carbon dioxide dynamics in a northern hardwood ecosystem. Soil Science Society of America Journal 59: 796-804.
Chapter 11
R E L A T I O N S B E T W E E N SCALE, M O D E L A P P R O A C H AND M O D E L P A R A M E T E R S
J.J. Middelburg, P.S. Liss, F.J. Dentener, T. Kaminski, C. Kroeze, J.-P. Malingreau, M. Novfik, N.S. Panikov, R. Plant, M. Starink and R. Wanninkhof
This Page Intentionally Left Blank
Approaches to scaling a trace gas fluxes in ecosystems A.F. Bouwman, editor 9 Elsevier Science B.V. All rights reserved
WORKING GROUP REPORT RELATIONS BETWEEN SCALE, MODEL APPROACH AND MODEL PARAMETERS
J.J. Middelburg (Rapporteur), P.S. Liss (Chairman), F.J. Dentener, T. Kaminski, C. Kroeze, J.-P. Malingreau, M. Novfik, N.S. Panikov, R. Plant, M. Starink and R. Wanninkhof
1. I n t r o d u c t i o n
Many scientific problems and environmental issues are posed or occur at temporal and spatial scales that cannot be covered by direct measurements or experiments. Relevant spatial scales may vary from micrometers for microbial processes to the global scale for atmospheric processes (i.e. almost 14 orders). Similarly, the dynamics of reactive gases occur within much shorter time scales than that of a relative non-reactive gas, such as carbon dioxide. Moreover, components may show variability at various time and spatial scales. For instance, carbon dioxide exhibits not only diel (light/dark) and seasonal variability, but also interannual (e.g. ENSO events) and longer-term variations. There is also a relation between the temporal and spatial scales of interest, but it is not necessarily linear. The various scales may overlap, but often there are significant gaps. Mathematical modelling is one way of attempting to bridge between scales. A model can be defined as a simplified representation of nature or a system that can be used, inter alia, to improve our understanding of processes or to simulate and evaluate the response to an imposed forcing. Depending on the accuracy and precision wanted and the level of detail required, models can be formulated to operate at various scales. As an example, a geographical map for an aircraft pilot contains much less detail of the landscape than that for a car driver, whereas that for a hiker contains even more detail. Nevertheless they serve the same function (navigation) and are just made to cover the appropriate scale. Models can be classified in various ways, i.e. according to discipline (physics, biology, chemistry, economy), to scale (molecular, micro-scale, measurement scale, field scale, regional, global) or to their mathematical nature. Conceptual models are those in which we formalize some theoretical ideas and which are mainly used for diagnostic purposes, i.e. they serve to increase our understanding. Empirical models rely heavily on data or parameterizations derived therefrom, but do not necessarily depend on a detailed understanding of the underlying processes. Mechanistic models are based on detailed process knowledge and need calibration before they can be applied to measurements. They can be used for diagnostic or prognostic purposes. Because of the complexity of nature, mechanistic models usually contain some processes that have been parameterized via empirical relations. Moreover, mechanistic models may either be deterministic if all model parameters have a fixed value or stochastic if some model parameters are in the form of probability density functions. The starting point for the discussion was the question "what is the relation between scale, the model approach and the model parameters selected". The discussion was constrained within the overall objective of scaling trace gas exchanges between terrestrial and aquatic
220
J.J. Middelburg et al.
systems on the one hand and the atmosphere on the other. This report will present a number of examples that are instructive for the problems encountered in dealing with the scaling issue. These examples illustrate the intimate relationship between scale and the model approach chosen. We first discuss the rationale of adopting a certain scale. This will be followed be an evaluation of the use of models in up- and down-scaling, with an emphasis on the balance between required detail and feasibility of validation and computation. We close with a comparison of scaling problems inherent to marine and terrestrial ecosystems.
2. W h y w o r k at a certain scale?
Trace gas fluxes between terrestrial and aquatic ecosystems and the atmosphere are studied on a range of scales. On the one hand, models of global atmospheric transport currently need input emission estimates on a horizontal grid of 8 degrees latitude by 10 degrees longitude. On the other hand, models of microbial processes simulate processes at a scale of centimeters. In this section we focus on the choice of scale for flux estimate studies. Model building involves an iterative number of steps (Figure 1). First, the questions to be answered need to be defined. Next, a conceptual model is developed, which encompasses identification of the processes involved a,.:t the system boundaries. Once the processes have been identified, the input-output relations can be quantified by combining available data with theoretical knowledge. After verification and validation of the model, we can start employing it to analyse the system being studied. Sensitivity and uncertainty analysis may add to the knowledge of the model's behaviour. This may lead to a better understanding of the essential features of the system, which in turn may allow construction of a summary or meta model giving a simplified representation. Within the process of model building several scales are involved. To begin with the scientific question to be answered is associated with specific temporal and spatial scales. The scales of the processes involved are usually different from the scale of the questions to be answered. Further, the data available may add another scale to the problem. However, the operating scale of a model may not only be the result of scientific reasoning alone, but may also be influenced by a number of limitations, such as computer resources, the modellers' understanding of the respective processes and the limited availability of data, etc. In the remainder of this section we address the question of the degree to which each of the abovementioned scales and limitations affects the scale at which the model operates. This will be done by discussing a number of examples of studies that span the entire range of scales involved in the biogenic production and emission of nitrous oxide.
2.1. Agricultural emissions of nitrous oxide at the national level
Nitrous oxide (N2O) is one of the most important climate-active gases. Atmospheric concentrations have been increasing as a result of agricultural activities, for the most part. Due to improved modelling and data availability, uncertainties in the global source strength are decreasing. For instance, a recently developed method for estimating agricultural emissions indicates that previous studies had underestimated this source (Mosier et al, 1998). This methodology for estimating N20 emissions from agricultural land at the national scale was developed for the IPCC Guidelines for National Greenhouse Gas Inventories (IPCC, 1997). The IPCC Guidelines were developed to assist countries that are parties to the United Nations Framework Convention on Climate Change (UNFCCC) to report their national, annual green-
Relations between scale, model approach and model parameters
~
221
ConceptualModel ~ I MathematicalEquations~ ModelConstruction-~ Models
& ~ i Sensitivity UncertaintyAnalysis~/
~
Systems Analysis
Predictions
Figure 1. Stages in model development.
house gas emissions to the IPCC Secretariat. The basic assumption made is that these methods for estimating emissions are applicable to any country in the world, irrespective of climate, soil or land-use conditions. The IPCC methodology for estimating N20 emissions on a national scale aims at assessing the full nitrogen cycle. It distinguishes between (i) direct emissions of N20 from agricultural soils, as induced by nitrogen additions, (ii) N20 produced in animal production systems, and (iii) indirect emissions from aquatic systems or remote soils taking place after the nitrogen leaves the fertilized fields through leaching, runoff, volatilization, or as sewage. The national input data needed to estimate the N20 emissions can all be obtained from readily available databases, such as those of the U.N. Food and Agricultural Organization (FAO). The bacterial processes that lead to agricultural emissions of N20 are nitrification and denitrification. Nitrification and denitrification are influenced by environmental parameters, including the availability of nitrogen, oxygen and carbon, as well as soil acidity and temperature. As a result, the observed fluxes show large spatial and temporal variability. In
Table 1. Comparison of six models/flux estimates of N20 (one per row) with respect to the scale of the scientific questions to be answered (first column), the scale of the processes involved (second column), the scale of the data available (third column), practical limitations (fourth column), and the scale of the operational model or study (last column). Scientific question to be answered and its spatial and temporal scale
Scale of processes
Scale of data
Limitations
Scale of the model or flux estimate (Reference)
Global inventory of N20 emissions for 3-D atm. Transport model; 8 x 10 degrees and monthly
spatial: ranging from mirosites (
spatial: ranging from microsites (
spatial representativeness of data availability of data on grid scale
1 degree x 1 degree temporal: monthly - annual (Bouwman et al., 1995)
Global estimate of N20 emissions on the finest possible spatial and temporal resolution based on atmospheric observations
Atmospheric transport: spatial: ranging from molecular to hemispheric - temporal: seconds to years sources and sinks: spatial: ranging from microsites to hemispheric - temporal: seconds to decades
atmospheric measurements: spatial: 11 monitoring sites - temporal: weekly (?) source: see above sinks: laboratory mainly
- computer resources availability of atmospheric observations - a priori flux estimates and their uncertainties
spatial: ranging from soil aggregates (
- spatial: ranging from soil aggregates (
- spatial representativeness of data availability of measurements on national scale
Emissions of N20 from the Somalian upwelling region between 21 June and 25 August, 1992.
spatial: ranging from microsites (
spatial: 100 cm 2 - temporal: seconds
spatial and temporal representativeness of measurements
Dentrification in the river Schelde for a nitrogen budget of the river and riverbanks on an annual basis
spatial: ranging from sediment aggregates (
- spatial: ranging from laboratory to field scale - temporal: ranging from minutes to month
data availability - process knowledge
Processes on microbial level that control N20 formation
spatial: micro-organism (<micrometer) - temporal: seconds
spatial: micro-organism (<micrometer) - temporal: seconds
-
-
-
-
-
-
-
-
s
p
a
t
i
a
l
:
-
spatial: 8 degrees by 10 degrees - temporal: monthly (scale of the TM2 model: Heimann, 1995)
-
-
National annual emissions of N20 from agriculture
-
-
-
-
1
0
0
-
-
-
-
-
spatial: national temporal: annual (IPCC, 1997) -
-
spatial: entire basin (84,000 km2) temporal: 21 June - 25 August, 1992 (De Wilde and Helder, 1997)
-
-
spatial: 15 model compartments ofc. 10 km x 50 m - temporal: daily (Soetaert and Herman, 1995)
-
spatial: micro-organism (<micrometer) - temporal: seconds (yon Schulthess et al., 1994) -
t,,3 1",3 1',,3
Relations between scale, model approach and model parameters
223
particular, there are units of high bacterial activity, at the level of soil aggregates, often referred to as 'hot spots'. Fluxes of N20 are measured at the scale of soil aggregates (< cm 2) in the laboratory, at the point scale (< 1 m 2) in the field using chambers, and at field scale (< 100 ha) by use of micrometeorological techniques (Lapitan et al., 1999). Most estimates of fluxes of N20 from agricultural systems have been obtained by chamber measurements in agricultural fields in North America and Europe (Bouwman et al., 1999). Relatively few data are available for soils from other world regions particularly tropical and aquatic systems along the aquatic continuum river-lake-estuary-sea. From available measurement data for agricultural fields, Bouwman (1996) inferred that 1.25 (0.2-2.3)% of each kg of nitrogen applied to soils is emitted as N20. His analysis includes only those experiments where N20 had been observed for at least a year. He concluded that the data sets were too limited in number and geographical coverage to derive quantitative relations between the N20 flux and environmental factors (e.g. management, crop, fertilizer type and climate) that could facilitate extrapolating the results to other scales by modelling the processes involved. The IPCC methodology, which we consider here as an example, aims at estimating emissions at the national and annual scales. It recommends using for direct soil emissions of N20 from a country the same emission factor as proposed by Bouwman (1996) and others referenced in Mosier et al. (1998) for chamber experiments, i.e. 1.25 (0.2-2.3)% of the N input to a country's agricultural land. This emission factor is the best documented factor in the method. Other emission factors (e.g. direct emissions from crop residues) in the IPCC method are, in part, indirectly derived from this emission factor, because for these fluxes few measurements are available. The national scale was set by policy makers for the purpose of international negotiations. Obviouslyj there is a large gap between the national scale of the IPCC method and the scale of the available data and processes involved. Due to the unavailability of data on the national scale, emission factors had to be used on a scale for which they were not derived. This essentially assumes that the relations found in the chambers hold also for the field and the national scales. It may be worthwhile, however, to investigate possible scaling errors. Aggregation to the national scale is further complicated by the fact that country sizes differ by orders of magnitude and it can be questioned to what extent the size of the country affects the accuracy of the result. A promising way to improve the national inventory methodology for N20 emissions from agriculture seems to use process based models to produce national emission estimates (Mosier et al., 1998). 2.2. Modelling N20 on different scales
Table 1 overviews some scale issues that relate to approaches to estimating emissions of N20. The models and flux estimates range spatially from the micro-site to the whole globe, and temporally from seconds to annual. Meanwhile, the data available for these measurements and models generally do not exceed the field scale. There are different reasons for choosing a certain model scale. Clearly, the questions to be answered determine the scale of the modelling to a large extent. In the case of the national greenhouse gas inventories and in modelling, for instance, the Schelde watershed, policy requirements set the scale of the estimate, regardless of the scale of the data available. For both the Somalian upwelling region, which may be considered a biogeochemical province, and the Schelde watershed, the scientific questions were formulated at the scale of the entire upwelling zone and watershed, respectively. In global inverse modelling different limitations apply, including the sparcity/scarcity of the atmospheric observational network and computer
224
J.J. Middelburg et al.
Table 2. Nesting of spatial resolutions for trace gas fluxes assessment using remote sensing.
Resolution
Spatial scale
Frequency of observations
Mode
Target
Coarse (xooo m)
Global
High
Detection of changes
Processes via dynamics of surface variables, e.g. NPP via FPAR, ET via T
Fine (xo m)
Local
Episodic
Measurement of changes
Features, objects, classes, land use
resources (both memory and cpu-time). In this case the scale of modelling is mainly determined by technical and data limitations, and not by the scientific questions to be answered or the processes involved. In contrast, the scale of micro-scale models seems to be exclusively determined by the scientific questions and processes involved.
3. Up and down-scaling In the previous section we have shown that there are various valid reasons to work at a certain scale, although the question is posed at another scale. In order to bridge this gap, ways have been developed to pass information from one scale to another. There are two principal ways to obtain information at a number of scales, namely empirically via remote sensing techniques and via modelling. Remote sensing techniques cover spatial scales varying from tens of meters up to hundreds of kilometers, either at regular intervals or episodically (Table 2). Examination of a landscape or a process at various resolutions through time yields information related to spatial variability (patchiness). The modelling approach is most widely used and the most generic method, but there are some problems due to (i) coupling of models (aggregation problems, feed-backs mechanisms and non-linear and stochastic processes) and (ii) scale dependent (dominant) processes. The problems inherent in the coupling of models operating at various scales will be illustrated via a number of examples.
3.1. Problems inherent to coupling of models Consider a well-tested mechanistic model that accurately predicts a flux for a well-defined region and a well-defined time domain. The model under consideration is based on measurements, which are performed in such a way that they are the integral of the heterogeneous properties of the underlying sub-grid processes (Figure 2). The region is defined by a unique combination of key-factors controlling the processes and can therefore be called a 'functional type'. In the Eulerian context the functional type may be called a 'grid cell'. The key-factors, and hence the functional types, are scale dependent. Examples of keyfactors determining functional types are soil type, climate and land use. In the next step the model is used to calculate the flux to or from a larger region. This region thus consists of several grid-cells. The key-factors provide a proxy for the parameters used in the model. At a certain point and during a particular time-interval, a grid-cell parameter is constant. When integrating to the larger region, the parameters are probably no longer constant (disregarding the time aspect). If the model is applicable to other grid-cells within the larger region, the integrated flux can in principle be expressed by the sum of the grid-cell fluxes:
F= ZSz,
(1)
225
Relations between scale, model approach and model part,meters
1
Functional type or Grid cell Lower levels
I _
Regional Level
Figure 2. Superposition of various model levels. Functional type is represented as a grid cell at the regional level.
where F is the integrated flux, and f is the flux for grid-cell i and zi the areal fraction within the larger region. In principle, equation 1 holds for uni- and bi-directional fluxes. Problems arise when lateral and/or vertical interactions between grid-boxes occur. Consider the following example where the actual flux is a function of the atmospheric concentration (Figure 3). A widely used parameterization for the calculation of the flux is by multiplication of an exchange velocity (k) by a concentration gradient: F = k x (Co-Catm)
(2)
where Co is the concentration at the surface of the grid-box, and Calm is the atmospheric concentration. Co is defined at grid-cell level, whereas Calm is known only for the larger region, e.g. the grid-cell of an atmospheric transport model. However, the 'true' value of Calm is also affected by the flux itself; i.e. a feedback mechanism is operating. Another example of the aggregation problem is the flux to or from a single leaf. When calculating the flux for the whole canopy, shading by other leaves will also influence the flux from a single leaf. As a consequence of aggregation, the description of fluxes becomes complicated when large-scale data (model outputs) are used for calculating fluxes on a finer grid-scale. This has initiated research on nesting of models with different resolutions. The scaling up of fluxes is further complicated when the processes on the finest grid-scale and in the larger region are non-linearly coupled, e.g. a coupled biosphere and atmospheric model. If the biosphere model contains non-linear processes, which are subject to feedback by the atmospheric model, the exchange fluxes within the coupled system may lead to unrealistic values. In that case it is probable that a fundamental feedback process is missing in the biosphere model. An example is the use of the calculated temperature from a meteorological transport model to calculate fluxes at a finer grid-scale. This is straightforward as long as the temperature dependency of the model processes behave linearly, since the average temperatu-
226
J.J. Middelburg et al.
Catm,1 ..." Ca!m,3~ ~ --Catm,211::"--(~a;m-
I Regional Level Box with Catm - avg (Catm,i) Figure 3. Coping with the difference in scale between model domains: regional level box.
re can be used to calculate the grid-cell fluxes. In reality almost all processes, including biological ones, show a non-linear behaviour, and the large-scale temperature should be disaggregated to the grid-box scale. A potential way of doing this is by using the modelsimulated temporal variability as a measure of the spatial variability. In meteorological terms temperature fluctuations may be associated with depressions which have a spatial extent of several hundreds of kilometers and a temporal extent of hours to days. Meteorological conditions may cause a stochastic distribution of emission rates. For instance, the emission of nitric oxide (NO) is stimulated from soils by episodic rainfall events. These emissions can essentially be described by a non-linear step function. Modelled rainfall data are normally available only on an aggregated spatial and temporal scale. Thus statistics on temporal and spatial variability are needed to accurately calculate NO emissions from soils. Similarly, gas exchange between the ocean and the atmosphere depends strongly and nonlinearly on wind speed. This non-linear response is caused not only by non-linear wind speed gas transfer velocity relationships, but is a'.~o due to non-linear relations between wind speed and water column mixing, hence biological activity and trace gas production.
3.2. Scale dependent parameters So far, temporal aspects have been ignored. However, the model parameters may change as a function of time. The model is based on measured data collected during a certain period. If the model is outside this time domain, the validity of the model is no longer assured. However, the time response of a model may implicitly be accounted for by a parameter already present in the model. For example, the seasonal time-scale may be represented by the temperature, or by the precipitation in the wet or dry season. Other time aspects, however, may not be taken into account. For instance, if the nitrogen loading of an ecosystem is a key-factor which is constant during the time-span of the measurements, the model which is based on these measurements, will not be able to calculate for fluxes when the nitrogen loading is changed abruptly or considerably. Essentially, the same problem occurs when the model is tested against data in one area and then applied to regions for which measurements are not available. If indeed the nitrogen loading turns out to be a key-factor, serious model prediction errors will be made in regions which have a different nitrogen loading compared with the grid-cell for which the model is valid. In the background paper by Archer (1999), for carbon dioxide
Relations between scale, model approach and model parameters
227
uptake by the ocean it is argued that on smaller scales gas exchange at the sea-surface is the determining factor, while on larger scales t~.e oceanic circulation may be the dominant factor. Denitrification in marine sediments may serve as another example. In marine sediments denitrification depends mainly on the input of labile organic carbon and bottom water concentrations of oxygen and nitrate, though numerous other factors, such as temperature, bioturbation, irrigation, sediment accumulation rate, may also contribute. On the global scale denitrification depends primarily on the flux of labile organic matter to the sediments, whereas at a regional scale nitrate and oxygen are usually important as well. Moreover, in organiccarbon-rich coastal sediments, rates of denitrification vary seasonally because of seasonally varying bottom-water concentrations (oxygen and nitrate) and temperature, while seasonal variability of carbon inputs can be neglected. This scale dependence requires cross-system validation and Monte-Carlo type sensitivity analyses.
4. How much detail in a black box? The concept of a black box has been widdy used in studies of elemental budgets at various scales. According to the simplest definition, a black box is a segment of any (natural) system whose mass/energy inputs and outputs can be evaluated without paying attention to processes operating inside, i.e. to processes consuming the inputs or generating the outputs. There is usually a three-dimensional aspect to black boxes, since they are typically defined as a volume of space at the land-atmosphere or ocean-atmosphere interface. Hydrologically self-contained catchments with one dominating output via surface discharge can serve as an example. Chemical, biological and physical black boxes are also widely used. In modelling, a black box can be regarded as an information source containing the underlying process information and interactions at different integration levels. The detailed processes and interactions are not explicitly considered, but the net result is passed on to the model via simple functions. Black boxes are used because of a lack of understanding of the interacting processes, in order to keep the model structure comprehensible or because of limited computer resources. Empirical models of trace gas fluxes are essentially black box type models. Mechanistic models are more "grey" since some processes are explicitly considered while others are parameterized, ".e. blackened. In this section we discuss the detail to which it is necessary to open the box. As a general rule it appears that (i) if possible, opening of the black is to be avoided, and (ii) in reality, 'grey' boxes outnumber the black boxes since there is at least some conceptual understanding included in measurements or models underlying the black box. The black box is opened when extreme patchiness creates the need to quantify sub-grid scale variability or to assess the physical, chemical and biological processes operating on a smaller scale. Opening of the black box often requires use of additional research tools and collaboration between modellers and pr0cess-oriented specialists. To obtain insight into the underlying processes, theoretical concepts and laws, and high-quality, high resolution data must be combined. When extrapolating from point measurements to a larger scale, the application of a suitable model is to be preferred over extensive acquisition of additional field data. Up-scaling by means of a model will rarely require opening of the initial black box, although such a case can not a priori be ruled out. By contrast, down-scaling is crucially linked to opening black boxes. Whether or not opening the box is a more effective research strategy, compared to undertaking field measurement campaigns in ecosysteas previously not studied, rema:qs to be seen. In fact, a cost-benefit analysis should be performed for each particular case. Proper understanding of the underlying processes may prove to necessitate time-consuming
228
J.J. Middelburg et al.
measurements in the laboratory. A stable isotope study of biogenic emissions of hydrogen sulphide from spruce die-back affected ecosystems of Central Europe (Novfik et al., 1996) can serve as an example. The black-box approach using soil-column incubations in the laboratory indicated a loss of 8 to 16 % of the incoming sulphur by re-volatilization. The 534S signature of the emission was significantly lower than that of the wetting solution (+ 14 vs. +34 %0 CDT). Such an isotope shift would correspond to biogenic emission of up to 20 kg S ha -I yr l. The up-scaling to the landscape level is feasible only by preserving the black box. It would be very difficult to decouple the individual processes of sulphur cycling within the soil and measure their isotope fractionation under varying temperatures and wetting conditions. Up-scaling thus requires use of the same technique (i.e. 34S labelling of rainfall) at the catchment level. In fact, such an experimental design is currently being employed in a roofed-catchment project at Lake Garsjon, southern Sweden. However, if it was feasible to open the black box then the upscaling would become more fundamentally sound and thus more reliable. The perils inherent to opening a black box are twofold. Firstly, a missing gas source/sink is intuitively ascribed to a specific process occurring within the black box. Concentrating on this process may divert attention from complementary processes which may be responsible for a sizeable proportion of the flux discrepancy. Although there are no short-cuts in such a case, the following second peril is preventable if care is exercized. For example, if the most important process has indeed been identified and a field or laboratory study conducted, the selection of study sites may be biased towards those with above average fluxes due to detection limits concerns. However, by proper site selection this problem can be avoided. Proper scaling may be hindered as a consequence of tailoring research projects to the needs of a process-level investigation while not enough information is collected for representative and more abundant ecosystems. This so-called hot-spot bias has been identified in marine sedimentary biogeochemical databases in which high deposition sites are over-represented, and compiled inventories for trace-gas fluxes that are biased towards temperate high-flux environments.
5. Marine vs. terrestrial ecosystems So far, our discussion of scaling procedures has been rather generic, because most scaling problems and approaches are universal and not dependant on the specific system being investigated. Marine and terrestrial ecosystems are however different in terms of heterogeneity, and the scientific approaches and measurement techniques being applied. We will first summarize different experimental and modelling approaches and end with a summary of differences in heterogeneity. The latter accounts in part for the different approaches chosen.
5.1. Measurement techniques Similar flux measurement techniques can be applied to the study of gas transfer between water and air, and between terrestrial systems (soil and vegetation) and air. There are however important differences in application of the technique caused by different magnitudes of concentration differences, and different degrees of heterogeneity. Here the focus is on the techniques used over the ocean with comparisons, where applicable, to terrestrial systems. For measurements over the ocean and other aquatic systems the flux (F) is commonly expressed in terms of a kinetic driving force, the gas transfer velocity (k), and the partial pressure difference (pXw-pXa) between water, pX,,, a,~d air, pXa:
Relations between scale, model approach and model parameters
F = k Xo (pXw-pXa)
229 (3)
where Ko is the solubility of the gas in water. Notice the similarity with equation 2. Thus the problem of determining the flux is separated into two components. Gas transfer velocities are commonly determined independently from the partial pressure differences. Only limited measurements of gas transfer velocity have been performed in the natural environment and for estimation of fluxes a parameterization of gas transfer with wind speed is often used. Several formulations have been developed relyi~,g on a variety of measuremer.ts in natural and laboratory environments (Lapitan et al., 1999). They generally show a non-linear increase with wind speed, but the formulations differ by up to a factor of two at any given wind speed. The formulation expressed above can be applied because on the scale of measurement (1 m 1 km) the partial pressure of surface water is reasonably homogeneous, in sharp contrast with terrestrial systems where spatial heterogeneity can extend down to the sub-centimeter space scale. However, surface water partial pressures do vary because of changes in physical and biological properties over a range of time and space scales. As an example let us consider CO2 partial pressures (pCO2) in a subtropical gyre in the ocean, pCO2 levels range from roughly 320 gatm in the spring to 400 gatm in the winter. As comparison the pCO2in water saturated air is roughly 350 gatm with an annual variation of 10-15 gatm depending on latitude. This means that large parts of the ocean can be either a sink (pCOzwpCO2a), depending on the time of year. In addition, the gas transfer velocity will change in response to variations in wind speed. On average, the global ocean is undersaturated by about 7-9 gatm yielding a net flux of approximately 0.4 mol mZyr l in to the ocean (or about 1.5 to 2 Pg yrl). On local scale this ~,ax can vary depending on the magnitude of the gas transfer velocity and partial pressure differences from roughly-7 mol m 2 yr ~ to 5 mol m2yr -I This is in sharp contrast with terrestrial systems over which fluxes of 3 x 103 mol m 2 yr l are not uncommon. Although this example is specifically for CO2, the same relative differences in fluxes for trace gases such as CH4, and N20 between ocean and terrestrial biosphere are common. The small relative magnitude of the oceanic flux makes direct flux measurement using micro-meteorological techniques difficult. These are frequently employed for determining fluxes in terrestrial systems with reasonable agreement with other techniques. For the ocean, however, the initial measurements gave fluxes that were an order of magnitude higher than those from other techniques and were irreconcilable with global mass balance constraints on oceanic carbon uptake. Recently, improvements to eddy correlation, gradient, and eddy accumulation techniques have led to new attempts to perform direct measurements over aqueous systems. Signal to noise of the measurements is still about 0.5-1 mol mZyr l, making such measurements feasible, at best, only in regions with large pCO2 differences. Preliminary results suggest that the eddy correlation t,, chniques still yield measurements that are about a factor of two higher than other techniques, but such differences are within the range of uncertainties of all aqueous flux measurements. The large interest in the micro-meteorological flux measurement techniques over aqueous systems is two-fold. Firstly, they offer a direct flux estimate rather than using the product of an often ill-defined gas transfer velocity and a patchy partial pressure difference field. Secondly and more importantly, they allow for making measurements at much smaller time and space scales than are possible with the "water side techniques". For the micrometeorological measurements the time scale of measurements is about 89hour and space scale, depending of height above the surface, is 200 m - 10 km (a general rule of thumb is that the footprint is approximately 100 to 1000 times the distance of the measuring device from the water surface). For water side measurements the time scale is eventually at best half a day and
230
J.J. Middelburg et al.
space scales are 10 km upward. Hence, with the micro-meteorological studies we can potentially investigate the forcing mechanisms of air-sea exchange in greater detail. The end result of such work should be more robust parameterization of the gas transfer with environmental forcing such as wind and surface turbulence. This in turn will lead to more reliable ways to scale up local measurements to regional dimensions, particularly if functionalities can be found with remotely sensed parameters such as those measured with satellite-based scatterometers.
5.2. Heterogeneity Generally marine ecosystems are viewed as being more homogenous than terrestrial ones. In the ocean, single-point water flux measurements made every 10-100 m from the ship display variation generally no more than 20%. In terrestrial ecosystems chamber single-point gas flux variations are generally much higher depending on the particular gas species and the type of the ecosystem. The spatial variability increases in the order: arable lands > meadow > natural forest > man-disturbed environment. As a rule the fluxes of trace gases like CH4, N20 display log-normal spatial distributions, extremely patchy with "hot-spots" of intensive emission as compared with more-or-less even or normally distributed emission patterns for gases like CO2. However, the homogeneity of marine ecosystems should not be underestimated. The application of remote sensing has revealed strong patchiness of oceanic water related to upwelling phenomena and currents. This inhomogeneity has much bigger scale as compared to terrestrial systems, being displayed within 102-104 m scale as compared with 10 -3-10~ m in terrestrial habitats. Table 3 presents the relevant characteristics of the open ocean and a forest. The degree of patchiness of gases and 2.as fluxes in the ocean depends on the relative importance of turbulent mixing and the gas life time, i.e. reactivity. Nitrous oxide exhibits less patchiness than a more reactive gas such as dimethyl sulphide (DMS). Carbon dioxide shows even less spatial heterogeneity since it is well buffered. The diel cycle of carbon dioxide in the ocean is rather limited because of this buffering mechanism, whereas in terrestrial systems it may be very pronotmced. The net carbon dioxide fluxes in these ecosystems are, however, of similar magnitude.
Table 3. Comparison between the ocean and a forest in terms of trace gas and mixing dynamics. Physical property
Ocean
Forest
Mixing of horizontal layers and vertical convection
High
Low
Spatial gradients of dissolved trace gases and their precursors
Gradual
Steep
The main mechanisms of gas transfer
Diffusion Bubbles
Diffusion Ebulli'ion Vascular transport
The rate of diffusive gas transfer
Low
High
The relationship between the rates of production and consumption of trace gases
Almost complete compensation
Lack of compensation
Net gas fluxes
Low
High
Gas turnover time
High
Low
Spatial organization of biota
Unstructured dominance of free-swimming organisms; vertical stratification over tens of meters; position of chemocline is well defined
Structured dominance of rooted immobile organisms; vertical stratification over cm to meters; mosaic-type of chemoclines
Relations between scale, model approach and model parameters
231
6. Conclusions and recommendations There are a variety of different reasons why modellers chose to work at a certain scale. Ideally, the scale of modelling should coincide with the scale of the questions to be answered and the processes involved. However, the scale of the information available and that of the information required usually do not match because of data limitation, computer resources and environmental questions posed by policy makers. This requires development of modelling strategies which allow coupling of models working at different scales and originating from different scientific disciplines. To this end we should give more attention to aggregation problems, non-linear effects as a result of coupling, and advanced methodology for nesting, both on-line and off-line. Moreover, these coupled, scale-bridging, models should be tested at various scales so that the uncertainties will be constrained and the most important parameters at the scale of interest made apparent. As a general rule, for deriving flux estimates we should not resolve the processes on scales smaller than needed to answer the questions raised, because (i) this would be a waste of resources, and (ii) could result in an imbalance in the detail and the level of scale between different model components which does not necessarily improve the results. However, for predictions beyond present-day conditions it is mandatory to include detailed knowledge of the most important processes. Validation of such a model is, of course, rather limited, but deserves attention given our changing environment.
References Archer, D. (1999) Modelling carbon dioxide in the ocean: A review. In: A.F. Bouwman (Ed.) Approaches to scaling of trace gas fluxes in ecosystems. Developments in Atmospheric Science 24. Elsevier, Amsterdam, pp. 169-183. Bouwman, A.F., J.G.J. Olivier and K.W. van der Hoek (1995). Uncertainties in the global source distribution of nitrous oxide. Journal of Geophysical Research 100:2785-2800. Bouwman, A.F. (1996). Direct emission of nitrous oxide from agricultural soils. Nutrient Cycling in Agroecosystems 46:53-70. Bouwman, A.F., R.G. Derwent and F.J. Dentener (1999) Towards reliable bottom-up estimates of temporal and spatial patterns of emissions of trace gases and aerosols from land-use related and natural sources. In: A.F. Bouwman (Ed.) Approaches to scaling of trace gas fluxes in ecosystems. Developments in Atmospheric Science 24. Elsevier, Amsterdam, pp. 1-26. De Wilde, H. and W. Helder (1997). Nitrous oxide in the Somali Basin: the role of upwelling. Deep Sea Research H 44:1319-1340. Heimann, M. (1995) The TM2 tracer model, model description and user manual. DKRZ Report No. 10, Ger. Clim.Comput. Centr. Hamburg, 47 pp. IPCC (1997). The 1996 Revised IPCC Guidelines for National Greenhouse Gas Inventories. OECD, Paris. Lapitan, R.L., R. Wanninkhof and A.R. Mosier (1999) Methods for stable gas flux determination in aquatic and terrestrial systems In: A.F. Bouwman (Ed.) Approaches to scaling of trace gas fluxes in ecosystems. Developments in Atmospheric Science 24. Elsevier, Amsterdam, pp. 27-66. Mosier, A., C. Kroeze, C. Nevison, O. Oenema, S. Seitzinger and O. van Cleemput (1998). Closing the global atmospheric N20 budget: nitrous oxide emissions through the agricultural nitrogen cycle. Nutrient Cycling in Agroecosystems 52:225-248. Nov~ik, M., S.H. Bottrel, D. FoltovL F. Buzek., H. Groscheovfi., K. Zfik (1996) Sulfur isotope signals in forest soils of Central Europe along an air pollution gradient. Environmental Science and Technology 30:3473-3476.
232
J.J. Middelburg et al.
Soetaert, K. and P.M.J. Herman (1995). Nitrogen dynamics in the Westerschelde estuary (SW Netherlands) estimated by means of the ecosystem model MOSES. Hydrobiologia 311: 225-246. Von Schulthess, R., D. Wild and W. Gujer (1994) Nitric and nitrous oxide from denitrifying activated sludge at low oxygen concentration. Water Science and Technology 30:123-132.
Chapter 12
VALIDATION OF M O D E L RESULTS ON DIFFERENT SCALES
M.A. Sofiev
This Page Intentionally Left Blank
Approaches to scaling a trace gas fluxes in ecosystems A.F. Bouwman, editor 9 Elsevier Science B.V. All rights reserved
VALIDATION OF MODEL RESULTS ON DIFFERENT SCALES
M.A. Sofiev Institute of Program Systems, c/o. Sroiteley str. 4, bid. 1, app. 18, Moscow 117311, Russia
I. I n t r o d u c t i o n
One of the most difficult problems in numerical simulations of atmospheric processes is the validation of calculated results against actual measurements. All models are based on sets of assumptions and simplifications of either complicated or poorly known processes. Therefore, it is necessary to carefully assess the accuracy of each model component to reach a balanced configuration. Since such a procedure still does not guarantee a proper reproduction of reality, validation of the total model is also important. The testing procedures used during model development differ from those used for model validation (Figure 1). During model development the quality of separate model components is examined; this requires specific measurement data, analytical solutions of test problems and peer review of scientific approaches. Model validation is the numerical evaluation of the modelled quality without consideration of such aspects as internal model structure, underlying principles and scientific approaches. For validation it is essential that measurement data are accompanied by a description of the data's precision and applicability for the model considered. There are three main objectives of model validation: (i) quantitative evaluation of the precision of the model results; (ii) improvement of the understanding of a particular phenomenon; and, (iii) assessment of the general factors regulating atmospheric transport and chemistry processes. A quantitative evaluation of the model's precision is especially important for further application of model results, for example, in the construction of scenarios on emission reduction and associated abatement costs on the basis of anthropogenic loads and ecosystem sensitivity, as calculated with mathematical models. Clearly, scenario calculations should be based on input data of known quality, as the answer has to be formulated in concrete figures which can be used in further assessments. Qualitative conclusions in terms of a "reasonable agreement" or "acceptable precision" are therefore not appropriate. The second set of objectives of model validation is related to understanding particular observations, such as trends, and maximum and minimum values. Model validation generally results in an expression of the reliability of the model simulations, the model's ability to reproduce a particular process and an indication of possible causes of disagreement. The output of the validation procedure is a key input to further model development, and a basis for the design of measurement strategies. The last set of objectives is related to the fundamental description of processes, which is particularly important in exploring new investigation areas, for example, in heavy metal or persistent organic pollution problems. In addition to the above approaches based on modelmeasurement comparison, such description implies a direct comparison between models.
236
M.A. Sofiev
of model r e s u ~ ,•.o.mparison withmeasurements and "state
1~(,
I/ V
Measurement data sets for development
Verification measurementdata sets State of the art knowledge
Analytical solutions Generally accepted scientific approaches
j MODEL DEVELOPMENT
K.
J MODEL VALIDATION
Figure 1. Simplified scheme of verification during the model development and validation. The results indicate the processes reproduced similarly by most of the models considered, which can be seen as the "current state of knowledge" on a particular process. Particularly in large-scale modelling there are many examples where the amount and quality of available measurements is insufficient for complete and reliable model evaluation. In such cases an additional artificial data set, reflecting what is referred to as "generally accepted" estimates, can partially fill the data gap. An example of such analysis is an estimation of the transport distance of a pollutant. This parameter cannot be easily measured but it can be inferred from the analysis of the output of different models. Where there is agreement of results from several independent models, it is possible to speak about the "generally accepted value" of the transport distance. This paper will discuss the current status of validation methods for atmospheric pollution models applied to different scales. The focus is on evaluating the complex models rather than on testing the separate model components. The scales and the corresponding types of models considered are locally, regionally and globally oriented. The differences between the type of observational data and their use in vaiidating models at different scales will also be emphasized. Short-range (local) models are validated against available monitoring data obtained from specially developed field experiments with large sample volumes appropriate for detailed analysis. The validation of long-range models requires careful consideration of the representativeness of the measurement for the appropriate scales.
2. Classes of atmospheric chemistry transport models There are different ways to classify models of air pollution distribution. Here, models will be classified on the basis of their spatial coverage, which is closely related to their spatial and temporal resolution. Three groups of models can be distinguished: (i) global and hemispheric models with a spatial coverage of more than 10000 x 10000 km, a spatial resolution of several degrees latitude x longitude and averages of the output data over a month or longer periods; (ii) region,q models, with a spatial coverage of about 1000 x 1000 km, a spatial resolution of about 100 Ion and averages over several days; and (iii) local models with a coverage of up to 100 x 100 km, a spatial resolution of several kilometers or less, and averages over one hour or less.
Validation of model results on different scales
237
Global models are generally developed for evaluation of mass balances of a substance, transformation of a set of compounds in the atmosphere; assessment of the main pathways and distribution of emissions, and deposition and interactions between the Northern and Southern hemispheres. Correspondingly, the methods for validation should consider those parameters appropriate for the temporal and spatial scale of the model. Global models generally operate with seasonally, annually or multi-annually averaged data. The spatial resolution of the results comprises several degrees. Therefore the availability of representative monitoring data is a major problem. In addition, global models often include complicated chemistry schemes for the analysis of multiple substances (e.g. NO, hydrocarbons, ozone). Regional models also simulate the distribution of pollution on a large scale and in the long term. They generally include complex no:,-linear chemical transformations of the substances considered. Hence, validation of regional models requires information on multiple substances from long-term measurements. Similar to global models, a problem in applying regional models is to select measurement data that represent large areas and long time periods. The validation of global and regional models is normally carried out at several measurement sites, which should represent all the main characteristics of the regions covered by the calculations, such as climatic zones, soils and water surfaces. The model is used for extrapolating to areas where no measurement data are available. Ideally, the period of the validation corresponds to that of the simulations. Contrary to the first two types of models, local models are mainly used for the evaluation of short-term concentrations averaged over time periods of minutes to hours. The major goal is to make a short-term forecast of the pollution distribution around the emission sources (both continuous and instantaneous). Local models are also used to assess the distribution of accidental pollutant release and emission of hazardous substances. Except for ozone models, the description of the chemistry of these models is of secondary importance. One of the main problems of local models is to correctly describe the pollution pattern :,,ear the emission source for different atmospheric conditions. As local models describe near-source processes, they are often constructed using a non-hydrostatic approach (contrary to hydrostatic approaches where air is considered as a liquid in quasi-equilibrium). For their validation, the model should produce a three-dimensional description of pollution plumes in the complete calculation domain. Validation is done for a number of typical conditions, for example, for several types of meteorological stability, so that the results can be extrapolated to other episodes. The classification of atmospheric models presented above is not strict. Current fast improvement of computer power results in increasing comprehensiveness of the models. The global models of today are now as comprehensive and detailed in their process descriptions as the regional models of some years ago. In some cases modem computers allow one to apply regional models to global or hemispheric calculations (Galperin et al., 1995). Another way of increasing the comprehensiveness without losing the spatial coverage is to use zooming models, which actually work at several different scales simultaneously (e.g. Moussiopoulos, 1994; Galperin et al., 1996). Currently, there are some efforts to compare models based on different approaches and on different scales. In the field of local modelling, the initiative, "Harmonisation within Atmospheric Dispersion Modelling for Regulatory Purposes" (Olesen, 1995b), should be mentioned. Comparison of regional models was done within the ozone model comparison project (Hass et al., 1994), and acid models were compared by the EMEP / MSC-W (CoOperative Programme for monitoring and evaluation of the long-range transmission of air pollutants in Europe / Meteorological Synthesising Centre West) and EUMAC (European Modelling of Atmospheric Consistence) model comparison (Hass et al., 1996). Heavy Metal
238
M.A.
Sofiev
Model Intercomparison was done in EMEP (Sofiev et al., 1996) and some others. Most of these projects concentrated on the development of the best available measurement data set and evaluation of the overall model behaviour.
3. Statistical techniques The first step in all model validation exercises is the calculation of the mean or median values of observations and model results. They are calculated in a general way as follows: -
1
a4
x - --~--~ x, m
(1)
t=l
if arithmetic averaging is chosen, and x = med(x,M)
(2)
for the median. Here M is the sample size and x either measured or modelled values. The use of median values is preferable if there is a possibility of values falling outside the main trend or far from the average values (outliers), because this method is robust to their existence and it yields results that are not sensitive to outliers. Contrarily, the arithmetic mean is more appropriate if the effectiveness of the estimation is more important than its robustness. The correlation coefficient between two sets is also taken as sample estimate: r
-
coy(x, y ) -
~
--
(x - x)(y - y )
(3)
The correlation coefficient is not a robust estimate and should always be applied with caution. The following example from the FMEP Heavy Metal model Intercomparison (Sofiev et al., 1996) illustrates this point. One of seven participating models produced values for 10 monitoring sites, as indicated in Figure 2. The extremely high value of the correlation coefficient (0.985) is mainly due to the monitoring site GB2. The correlation coefficient decreases to 0.958 by excluding this point. This value is high, but the small size of the sample (9 pairs) results in large uncertainties in its determination, which is reflected by the wide 67% confidence range (0.73 - 0.99) (see equation 7). The 67% confidence range means that the true value is inside this range with a probability of 0.67 (such a range for normal distributions equals 2 x the standard deviation). The regression analysis is generally based on linear assumptions of the model-measurement relationship considered to be distributed by the first-order polynomial additive, normally distributed, stochastic component ~.
y,=Ax,+B+g,,
i=l,M
(4)
On the basis of these assumptions, the method of least squares leads to the common estimates of the regression coefficients: AMLs = x y -
Xy --2
'
BMI,s = y - AMI.s x
(5)
x 2 -x
All estimates presented above are finite-size sample estimates of the parameters considered. Their standard deviations can be calculated as follows:
239
Validation of model results on different scales
-o i2
dB2
:3 O
-~ io o
G'B1 ~ NL3 ,i,
N2 ,F1 9GB4
Mean obs.=3.52 Mean calc.=3.5 Corr.=0.985
,, GB3 DK1
2
4
6
8
i0
12
Observed
Figure 2. Measured and calculated lead concentration in ~g 1~ in precipitation. Abbreviations of the stations are the notation~ accepted in OSPARCOM monitorin? network (Oslo and Paris Commission, !994). Source: Sofiev et al. (1996).
1 o- E = - ~ o - x
(6)
for arithmetic mean E(x); 2 1 ~
O'r
rx
(7)
JM
for correlation coefficient rx; o~.
1
o~. =
~
-~ x-
(8)
x/Met,.
for regression slope A; and O"B
~
,/7 ,/7 +x-2
(9)
for the regression bias. The notations here are similar to those of equations (1)-(5). In some cases equation 4 is not correct, and application of the method of least squares might give uncertain results. Some reasons for that and possible solutions for a more robust methodology of linear regression are discussed by Sofiev (1994), Sofiev and Galperin (1994) and Sofiev et al. (1994). In addition to the above simple model evaluation procedures, the ability of a model to describe the measured time trends (both short-term and long-term) can be quantitatively evaluated by the mean square deviation, maximum absolute deviation and the maximum relative deviation between modelled and measured trends. These are calculated as follows: p _ ~1 N
N ~-~( y' - x , ) l=,
2
(10)
M.A. Sofiev
240 for the mean square deviation;
(11)
Q = max ( [ y , - x, ,i = 1,X) for maximum absolute deviation, and
IT,- x,I
R-max lY, +x,l'
(12)
i : l,N, y~ + x~ >0 'v'i)
for the maximum relative deviation. N is the number of elements in a set. There are other parameters which may be used to compare two data sets, but equations (10)-(12) are probably the most common with clear physical meaning. Equation (10) is closely related to the method of least squares and is applicable to the estimation of overall agreement of model results and observations. Maximum values of absolute deviation in equation (11) indicate the ability of a model to reflect peaks. This estimate may be normalised by the mean value over the measurement data set. The relative deviation in equation (12) is more sensitive to low values where small absolute errors result in high relative deviations, reflecting the model's quality in remote regions and for episodes with low concentrations. Kriging analysis can be used for interpolation of measurement data. The number of stations and their locations must be adequate to represent the heterogeneity of the region considered, to resolve major emission sources and characteristics of the terrain, such as relief and large water basins. After smoothed observed values have been obtained, they can be projected to a model grid and compared with the calculated results. Probably the best approach here would be the spatial analysis of the areas of differences, their location and final creation of a map showing areas of the model correspondence with measured data. This map can be useful both for model improvement and further use of the model data. An example of applied kriging analysis can be found in Berge et al. (1994).
4. Model-measurement comparison The comparison of model output with available measurements is the most widely used method for model validation. Practically all models are validated with some sets of observational data which are considered as an "ideal" reference. Nevertheless, there is a substantial difference in the approaches to model-measurement comparison used for regional and global models on the one hand and local models on the other. For each group of models, model validation against observations will be discussed first, with emphasis on the requirements of measurement data. After this, the different validation techniques will be summarized for each group of models. The precision of non-measurable model results for the group of regional and global models will be discussed briefly. For local models, a tracer experiment will be described as an example of a systematic method of model validation.
Table 1. Estimate of the representativeness (D) of monthly measurements for a large area (about
100 km linear size). Type of measurement
Minimum D
Maximum D
Mean value 23%
air concentrations
20%
28 %
NO2 air concentrations
27%
34%
31%
SO4=, NO3, NH4+ concentration in precipitation
14%
30%
25%
SO 2
Source: Galperin and Sofiev (1994).
241
Validation of model results on different scales
4.1. Regional and global models 4.1.1. Measurement data
One of the characteristics of large-scale models is that they cannot reproduce high concentrations close to pollution sources. They cover large areas, including remote regions where concentrations are sometimes at the background level, close to the device detection limit. This raises the problem of calibrating the monitoring device and estimating its precision. Nowadays virtually all large-scale monitoring campaigns include quality assurance of the measurement technology and devices. For example, a regular quality assurance of the monitoring sites and their intercalibration are performed within EMEP each year (e.g. Fahnrich et al., 1993; Semb et al., 1994). For substances like sulphur and nitrogen oxides, the precision of measurement devices is high and corresponding errors are negligible. For other species such as zinc and cadmium, the standard deviation and detection limit is in the same range as the measured value itself. For some heavy metals and their compounds (e.g. mercury-containing species) and persistent organic pollutants, the measurements may only show the order of magnitude of their concentration or a mere detection of their presence. The problem of representativeness of point measurements for some of the surrounding areas and time periods was considered by several investigators. The representativeness of point measurements can be described on the basis of differences between data from closely located stations and the standard deviation of the sample mean. The deviation of the mean concentration in a grid cell depends on the size of the cell, the averaging period, and the type of substance (Ebel et al., 1994; Galperin and Sofiev 1993; Seilkop, 1994; Sirois and Vet, 1994). Table 1 presents the sum of squares of regular and chaotic differences between monthly values measured at two independent closely located monitoring sites, which is calculated as: D=
+or A-
(13)
where A is the mean value of relative differences between station data A, and era is its standard deviation. The distance between stations was about 100 km and several pairs of stations were considered for each compound. This allows for calculating the maximum, minimum and mean value of the representativeness, D. For daily measurements, the estimate of the representativeness taken from Seilkop (1994) shows the 95% confidence limits for individual summer air pollutant concentrations measured within 80-km grid cells and presented as a percentage of the grid-cell mean (Table 2). The uncertainty caused by limited representativeness of point measurements for a grid cell can be very important. Comparing daily measurements with model calculations carried out with a spatial resolution of about 100 km is not meaningful (Table 2). This uncertainty drastically decreases when measurements are averaged over longer time periods (Table 1).
Table 2.95% percentile range for daily concentrations measured at points within the 80-km grid cell. Type of measurement SO 2
air concentration
Lower limit
Upper limit
-76%
+172%
SO4= air concentration
-40%
+56%
03 maximum daily concentrations
- 17%
+20%
Source: Seilkop (1994).
M.A. Sofiev
242
4.1.2. Validation methods Generally, the number of statistical tools currently used for model validation is rather limited. Expressions of the validity include the correlation coefficient describing the agreement between measurements and the model simulations, various types of confidence intervals and percentiles, and linear regression analysis commonly based on the least squares method. Apart from statistical techniques, a widely used method is qualitative analysis of various graphs showing temporal variations of the concentrations at the stations, with a further qualitative evaluation of the model's ability to reproduce the peaks and lowest values as well as the main observed trends. The main advantage of the above methods is their simplicity. They are also fairly useful for interpretation of the results. The main disadvantage is that they sometimes lead to inaccurate or subjective conclusions, as they are based on assumptions on the type of distribution function of the data considered. First of all, they are supposed to be normally distributed N(x, a, or) according to the Gaussian distribution function (Figure 3): l
(x-=?
N(x,.,o ) - 4 ~
e 2~
(14)
where a is the mathematical expectation of the stochastic variable x , and cr the standard deviation of x. This function permits negative values of x, which is incorrect in the case of atmospheric concentrations and many other physical parameters. However, this function is acceptable for some cases where a > > or. In many other situations essentially positive functions should be used, such as the widely used gamma distribution (Figure 4): 1
x-1 -x e x>o O,
(15)
x_
where 1-'(2)is Eulerian gamma function calculated as: F(2 ) - I y~-'e-"dy
(16)
0
Another example of a function allowing only for positive values of x is the log-normal distribution (Figure 5) presented by: I p(,)_
1 o
- (~~.,-~,)~ e
0,
~>0
(17)
x<0
Equation (17) characterizes the logarithm of the normally distributed stochastic variable
N(x,a,a 2). An important aspect of the log-normal function is that the combination of two independent log-normal variables yields another log-normal variable. The selection of a specific distribution function is difficult because, qualitatively, the behaviour of the gamma and log-normal functions are rather similar (equations 14-16). However, their detailed quantitative analysis is, for many practical purposes, not obligatory. It may suffice to assure positive definity of the unimodal function with prescribed ranges of mathematical expectation and standard deviation. For such cases the choice can be based on the properties of a particular distribution function. Hence, with regard to convolution, the gamma type of function is closed, which means that the resultant of two gamma functions will
243
Validation of model results on different scales 0.6
Gauss distribution density
0.4 .
-2
-1
"=~m'ot =3, cy =1 ot =3, ~ =2
~ /
0
/m,
1
2
/
I
3
~
~ -33,,; - 3 4
4
5
6
7
8
Figure 3. Characteristic shapes of normal distribution functions (Gauss function). 0.4
Gam ma-distribution density
~ X
=2 X=3
0.2
~;L
=4
--X
=5
6
7
/ j
-2
-1
0
1
/
2
3
4
5
8
Figure 4. Characteristic shapes of the F distribution function. 0.2
-
Log-normal distribution density
~c~
=3, oc =3, ot =3, --or =3,
-2
-1
0
1
2
3
4
Figure 5. Characteristic shapes of the log-normal distribution function.
5
6
1
cy =1 I cy =2 cy =3 cy =4
7
8
244
M.A. Sofiev
yield the following one gamma function: g~, * gx2 = g~.+~2
(18)
This result implies that with regard to averaging, the gamma distribution family is also closed. For example, if daily mean concentrations have a gamma distribution, we can expect monthly averaging to also have this type of distribution function (but with different parameters). In addition, the gamma function is closely linked with )(9 distribution and is therefore convenient in cases of squares of normally distributed functions. These features can considerably simplify a further analysis if, for example, different averaging periods are concerned (initial F-functions ensure similar types of distribution densities for all averaged values). A few problems may occur. Firstly, it should be stressed that the distribution functions used for modelled and measured data also determine the appropriate statistical equations for the regression analysis. Applicability and precision of all other techniques should be assessed because most commonly used methods are optimal for normally distributed variables. Using strongly different assumptions may lead to inaccurate results of statistical analysis. Secondly, the analytical methods used must be robust. Good examples of non-robust statistics are the arithmetic mean and the correlation coefficient (see section 3). It is often possible to modify these methods to create more robust statistics (Huber, 1981) or use a different method, such as the median. Thirdly, observations are generally point measurements, while the simulated values are averaged over large areas in regional and global models. Limited representativeness of the observed values requires consideration of data from each station as a realization of a stochastic process with an unknown mean and non-zero standard deviation. Having a data set for each station for a certain time period, and assuming an ergodic process, it is possible to restore the mean values for each grid cell and then compare them with the calculated ones. The term "ergodic" means that the average over a time period represents the mean value for the grid cell where the station is located. In reality this is not always correct, although this assumption is implied in all model-measurement comparison exercises. For example, without this assumption the comparison of monthly values obtained from point measurements with corresponding modelled concentrations averaged over a grid cell would be meaningless. Another approach to this task is suggested in several papers (e.g. Pedersen, 1994; Journel and Huijbregts, 1978), where the measurement network was used for the creation of a smoothed map of monitored parameters. This can then be compared with the smoothed model output pattern. Although this method is promising, it is based on spatial correlograms, requiring a rather dense network of monitoring stations, which is not always available. 4.1.3. Precision of non-measured model results Model validation is often hampered by the lack of available measurement data. The measured properties, such as the concentration of sulfur dioxide (SO2) in air, are, particularly for largescale models, not important from an environmental point of view. It is more crucial to know the deposition of sulphate, which requires data on both wet and dry deposition. However, dry deposition is practically non-measurable. Hence, validation of modelled SO2 concentrations is of no value for assessing simulated deposition rates. The mass balance approach can be used to constrain the values obtained for substances or parameters not measured. For example, sulphur compounds, including SO2 and SO4= in air, and sulphate in precipitation, form a set of species linked by chemical reactions. For validation on the basis of the mass balance approach wet and dry deposition fluxes need to be
Validation of model results on different scales
245
known too. All concentrations and wet deposition are measurable; their precision can be evaluated by direct comparison. The quality of simulated dry deposition can be qualitatively assessed if the following requirements are met: (i) a large enough calculation region with the bulk of the emitted substance deposited inside the region; (ii) a similar precision for all calculated components (at least the size of the errors should be the same); (iii) a spatial distribution of errors is chaotic without definite structure. Under these conditions the mass balance approach leads to an absolute precision of the dry deposition flux of the same order of magnitude as that of the different items in the mass balance. This approach cannot be used when the mass fraction of a non-measured substance is small, because then the relative uncertaint~ will be high and conclusion:, on the precision of the model reflection of the particular substance cannot be drawn.
4.2. Local models
4.2.1. Measurement data
One of main goals of local models is on-line monitoring of short-term atmospheric concentrations in the vicinity of large emission sources, such as areas with intensive agriculture, biomass burning or industrial centres and cities. Contrary to large-scale models, the parameters simulated by local models are mostly the same as the measured ones, i.e. concentrations in air and precipitation. In addition, the concentrations are high in near-source areas, so the precision of measurement devices is generally not a limiting factor. Because of the short time period associated with the pollution transport through the model domain, the chemical transformations can often be neglected. The main output parameter of local models is the maximum concentration in the pollution plume. Measurements of plume concentrations by the monitoring stations are quite uncertain for several reasons. Since the observational points are located close to the source, and the averaging period is short (less than about 10-12 minutes) with medium-range turbulence affecting the shape of a pollution plume, non-continuous releases of a pollutant may arrive at a specific measurement site. In addition, the pollution cloud near its source normally expands to an angle of about 20 degrees, and even a small distortion of the plume or a shift in the transport direction can result in large changes in the concentrations at the monitoring site. So if a monitoring site is exposed to a pollution event, the concentration is high, while if the site is not exposed, the concentration is zero or very low. This "yes-no" structure of measurement data in combination with disturbances of shape and direction of the pollution plume create additional stochastic variability in the distribution pattern. 4.2.2. Validation methods
As for large-scale models, there are several tools for the validation of local models, including the "model validation kit" developed by Olesen (1995a,b) presented here. This kit was developed in the framework of the initiative on "Harmonization within Atmospheric Dispersion Modelling for Regulatory Purposes". The kit includes three sets of measurement data obtained from experiments at different locations under different conditions, i.e. Kincade, Copenhagen and Lillestrom. In addition, the kit contains a set of statistical methods for the comparison of model results with measurements, and tools for the graphical interpretation of the results. The procedure was applied to five models from different countries participating in this initiative. Another example of standardization efforts is the "Standard practice for statistical evaluation of atmospheric dispersion models" (Irwin, 1997). The methodology presented in
M.A. Sofiev
246
that document is based on the approach developed by Lee and Irwin (1995). The examples of statistical techniques presented below are mostly based on these approaches. The most important criterion for assessing the quality of local models is their ability to reproduce the maximum concentrations in the pollution plume. Large amounts of measurement data allow us to determine the maxima of cross-wind arcs (where the monitoring sites are located). Altematively, the maxima may be calculated from an assumed distribution function (e.g. Gaussian). The calculated maximum concentration corresponds to the centerline of the plume. Cross-wind integrated concentrations can be used instead of maximum values. A Gaussian distribution should be used with some caution, because the maxima may strongly deviate from those described by the Gaussian distribution function. The model performance can be quantitatively evaluated in terms of the fractional bias FB:
FB = 2 cm - c ,
(19)
Cm + Co
where Cm and Co correspond to modelled and observed concentrations for the upper quantiles of values (e.g. for a quartile). The quantile analysis can be used explicitly, as in the Model Validation Kit. By definition the quantile Kp of the order p for the stochastic variable X with the distribution function F(x), is the number x for which the following inequalities are true:
F(Kp) <_p,
F(Kp +O)_> p,
0
(20)
The quantile Ko5 is referred to as the median of X while Ko25 and Ko75 are lower and upper quartiles (see also Figure 6). When sufficient measurement data are available, there is no need to use artificial reference data sets. Models are not directly compared with each other, but the overall result of the comparison is derived from individual model-measurement comparison results for each model. The completeness of such a ranking depends upon the completeness of the measurement data. Thus, all three monitoring data sets in the "Model Validation Kit" were obtained for rather homogeneous terrains using only one source of an inert tracer. The extrapolation of the evaluation results outside these limits is therefore uncertain if no additional measurement data are available. The problem of low concentrations produced by models which often correspond to zero concentrations observed at the stations (and vice versa) can be resolved by some artificial modifications of the data sets compared. The relative error in such pairs (zero observed, small non-zero calculated) is infinitely high, but from a practical point of view the concentrations can be considered to be equal and the ratio between observed and calculated concentrations for such pairs can be assumed to be 1. For validation of local models the measurement data should represent the actual environment to which the model is applied. In practice, however, local models are generally validated against available measurement data sets which are not necessarily obtained in the application area of the model. Application of local models in other areas with different source types often occurs without additional calibration. Therefore it is essential that such data sets represent conditions (in particular, meteorological ones) which are also met in the model application region.
4.2.3. The ETEX experiment Two of the most wide-ranging model validation projects carried out under real conditions were the Atmospheric Transport Model Evaluation Study (ATMES), initiated after the Cher-
247
Validation o f model results on different scales
0.75 0.5
P Kp
Ko. 5
K0.75
x
Figure 6. Illustration of quantile definition.
nobyl accident in April 1986 (Klug et al., 1992), and the European Tracer Experiment (ETEX), carried out several years later. The purpose of ETEX was to compare the capability of forecasting models to predict the large-scale distribution of compounds emitted during accidental releases (Graziani e t al., 1997). In ETEX there were two releases of an inert tracer. The first purpose of the models was to predict the distribution of the cloud over Europe, with a time resolution of three hours, and the concentrations at about 160 monitoring sites. The next step would include the model run on the basis of diagnostic meteorological data obtained from different sources. All data were collected, analyzed and organized in the distributed information system based on GIS (Geographical Information System) and Web technology (Van Liedekerke and Jones, 1995). ETEX was a unique experiment, being in fact an application of regional models to shortterm episodes. The calculations covered E arope as a whole, which is the common coverage of regional models. However, the temporal resolution of three hours, the fine spatial resolution (160 monitoring sites in Central and northern Europe) and the fact that the tracer was released from one point only, would fit better to the characteristics of local models. The evaluation of the ETEX data and comparison of models were done with a limited number of simple methods described in the previous sections of this chapter. More comprehensive analyses will probably be made in the future. A rather promising methodology for evaluating the model's ability to reproduce the general cloud pattern was practically not applied. This method concerns the evaluation of the "yes / no" problem, for example, in studying accidental releases, often more important than the representation of the maximum concentration. Numerical characteristics of the "yes / no" criterion can be taken from the probability theory. For example, the model's quality can be considered a combination of the probability of missing non-zero values (zero calculated values for non-zero values measured): = p ( C = 0] o > 0)
N(C
< c~ ..... d , O > c . .... d)
(21)
N s , e.,
and probability of false alert (non-zero calculated values for zero measured): = p(C
>
0 ] o = 0)
N(C
> c~, ..... a , O < c c .... a )
(22)
N s, e.,
Here p is the probability, while C and 0 are the calculated and observed concentrations,
248
M.A. Sofiev
respectively; Nsites is the total number of monitoring sites, and iV(. ) the number of sites with missing non-zero values or non-zero calculated values for zero measured. One important problem is the choice of co .... d, which should exceed the background concentration and be essentially positive even for purely anthropogenic pollutants with zero background concentration. The value of co .... d depends upon several factors, including the measurement detection limit, averaging period of measurements, the scale of the calculations and the range of the concentrations considered. In addition to the reflection of background concentrations and finite measurement precision, co .... d also allows for smoothing the influence of several second-order factors, such as finite precision of the model advection scheme (for Eulerian models it may be so-called numerical viscosity) and local turbulent distortions of the pollution plume. The reason for its introduction is to some extent similar to that of the last modification of Model Validation Kit, where the ratio between small observed and calculated values is assumed to be 1. In more complicated cases the equations (21) and (22) may be extended in order to consider more than two levels of discretisation. Considering a set: { c ~ , i = 1,n Ic,~
cb}
(23a)
instead of CGround we obtain a sequence of n-1 ranges, which allows for representing the model quality as a probability that modelled concentration are in the same range as the observations: Pmatch-- p [ C c (fiG c G ) O c (cCi" c G )] ,...., N ( C G ~-- C ~ c i+l, G c ,G ~-- O ~ c 'G) , i - 1,n , i+l ' t+l Nme.~.
(23b)
It is clear that equations (21) and (22) represent a particular case of equation (23b) for n=3, cz~ c Y = c o .... a, c3~ Pmatch---- 1-pm,.,',-pFA.
5. Model to model comparison The knowledge on the physical and chemical processes for different substances involved may be incomplete in new areas of research. In traditional areas like acidification, such a knowledge base has been created mostly on the basis of a variety of studies, analyzed, compared and integrated into coherent estimates. An attempt has therefore been made to build up the knowledge base in the relatively new field of research on heavy metals and persistant organic compounds in a more systematic way. This has been done in the form of intercalibration projects carried out to evaluate the available knowledge, including descriptions of processes generally accepted and not contradicting the observations. Another problem of new fields of work or research is the scarcity of measurement data. For toxic substances the potentially critical concentrations number up to figures lower than those of acidifying pollutants. As a result, the measurements are expensive, and their quality often leaves a lot to be desired. In such cases, numerically described state-of-the-art knowledge can be used as a reference in addition to measurement data. The model validation procedure then consists of comparing model results with available measurements and state-of-the-art artificial reference data sets.
5.1. Creation of average estimates
The results obtained with different models covering the same area and time period, and simulating the distribution and chemistry of the same substances, can be averaged so as to
249
Validation of model results on different scales
create a new data set covering the comnlon area and time period for all models from the output data of all models. This new data set, created by statistical averaging procedures, can be considered a new model, which is referred to as the "Statistical Average Model" (SAM). The original models are also made on the basis of certain simplifications and averaging of actual parameters in time and space. SAM, then, represents one more averaging step i.e. the mean taken over different approaches to the description of the same items in the data set. Averaging the results from different models allows one to smooth errors produced by those models. However, the opposite may also occur when one model, producing results strongly deviating from the others, may show considerable effect on the group mean. Robust averaging procedures are recommended to avoid such problems. The simplest example of a robust procedure is the statistical concept of the median value (equation 2). The main advantage of artificial data sets is that they reflect the common features of most participating models; they also represents the "state of the art" values. The Statistical Average Model SAM has a low sensitivity to variations in input parameters of individual models and to variations in their basic assumptions. Within the Heavy Metal Model Intercomparison Project (HMMIP) experience, the agreeme,~t between SAM and measurements appeared to be one of the best in comparison with the individual models (Sofiev et al., 1996). However, this does not necessarily occur in practice. For example, when all models deviate from the measurements in the same direction, SAM will also deviate, indicating a systematic error in all the model assumptions or measurements.
5.2. Model comparison using artificial reference data Artificial reference data sets as developed, for example, in the form of SAM or via kriging of measurement data, allow for additional validation of models. Since maps of concentrations or deposition result from both kriging and SAM, validation against these reference points amounts in fact to a comparison of two model data sets. Sofiev et al. (1996) proposed two procedures: (i) Evaluation of general characteristics of the models; and (ii) Direct comparison of maps. 5.2.1. Comparison o f the integral model characteristics
The integral model characteristics behind the simulated pollution pattern may be more important than the values themselves. In HMMIP these characteristics, deemed both fairly stable and robust to variations in input data, included (i) correlation radius as a measure of transport distance; (ii) the variation in correlation radius, where the direction is a measure of meteorological and emission anisotropy; (iii) emission-deposition cross-correlation function; and (iv) mass budget tables. The correlation radius for each substance can be calculated from the autocorrelation function (ACF) of an (infinite) matrix M, where ACF is also a matrix consisting of correlation coefficients between non-shifted and shifted maps M: A C F ( i , j ) = corr[M(i + ~ , j + ~),M((,~:)],
i , j , ~ , ~ e (-c~,c~)
(24)
where corr indicates the correlation coefficient, and M ( . ) the value of a map cell with corresponding co-ordinates. Equation (24) is written for infinite definition ranges. However, the actual map is a finite-size object. He:ice, the correlation can be calculated only for the common area of non-shifted and shifted maps, resulting in non-zero variance of the estimates obtained. This variance can be estimated as follows:
M.A. Sofiev
250
+2*
~ ZACF2(q,~ ) ~2+~2
(2s)
where N~ is the number of cells in the common grids. The summing is done within the circle, with tad (i,j) = 02 + j2)0.5 According to the standard definition, the correlation radius r corresponds to the lag (i,j), where the ACF equals Lie or, in a more general form: r(|
- a / Z + Y~ln c
ACF(x, y) = c, sin O =
X2 + y2
(26)
As the correlation radius is a function of direction O, its variation can be easily expressed as the standard deviation. Another parameter used in model comparison is the histogram H of a map M((, ~):
H - {H(al,al+,)H(a~,al+~) - Num~4(~,{)e (al,a~+~l = 1,L-1}]
(27)
where A={a~ is the set of histogram ranges, Num the number of grid cell values within the corresponding range (al, a~+l) and L the total number of ranges of the histogram. The cross-correlation function can be obtained from equation (24) by introducing two different maps into the equation:
C C F ( i , j ) - corr[M~(i+q,j+~),Mz(i,j)] ,
i,j,q,~ e (-oo, oo)
(28)
The above remarks on the finite-size objects and corresponding standard deviation (equation 25) are also valid for this function (equation 28). The total pollution mass distribution or mass budget is an important result of models because of its use for validation. Large-scale models generally produce so-called mass budget tables (see e.g. Barrett et aL, 1995). Local models commonly do not consider mass budgets, except where deposition of pollutants is important. It is obvious that local mass balances cannot be measured, hence, artificial references are inevitable. The classical form of the mass budget table is a two-dimensional matrix
B - { b~,~ l k - l,K , 1-1,L }
(29)
where K and L are the numbers of independent sources and receptors of the polluting substance considered, respectively, and bk,l is the amount of pollutant emitted from the k-th source and deposited onto the/-th receptor. Evidently, this value is determined by the rate and intensity of the emission, the area of the receptor and the air transport conditions. Comparison of budget matrices can include most of the standard methods of non-parametric statistics and regression analysis discussed in section 3. They give a quantitative estimate of the matrix differences. However, in simple comparisons the physical meaning of the mass budgets is not considered. A better understanding of model behaviour can be obtained, for example, by comparing the origin of the deposited substance or transport outside the model boundary. Local deposition near each emission source can be represented by the deposition from this source onto the area (receptor) where it is located. A one-dimensional vector can be constructed from the depositions onto corresponding sources, which, in turn, can be used to compare model behaviour on a small scale. Contrary to local deposition, the fraction of the pollutants transported outside the calculation domain is the result of the integral properties of the large-scale model.
Validation of model results on different scales
251
5.2.2. Direct comparison o f model results The direct comparison of model results in the form of concentration and deposition maps provides information about the differences and similarities of the model applications considered. This contrasts with integrated parameters, which do not depend on the particular application. The statistical tools used for direct model comparisons are different from those previously described. Many of the above procedures were anisotropic, i.e. they implied the existence of "reference data" considered as correct, and some simulation of these true values. If different models are compared without further reference data, their quality cannot be assessed. The most crucial change to be made will be in the regression methods, which should be symmetrical (see least squares method). If x and y are two estimates of one unknown variable and there is no a priori information on their quality, the best estimate of the variable will be the arithmetic mean (equation 1). So, the independent variable x in the standard method of least squares should be substituted by: Z
w
x +y 2 .
(30)
~
the basis for the regression analysis. The method of least squares should then be applied to new data pairs, either (x,z) or (y,z). After substituting the results to the initial variables (x,y), we obtain formulas for the symmetrical regression estimates for slope A and bias B: 2
_
m
A - Cry + cov(x, Y), B = y - Ax o'~ + cov(x, y)
(31)
where cov denotes covariation and cr is the standard deviation. Equation (31) is symmetrical for x-to-y switching, reflecting the symmetry of the data sets. Equation (30) can be written in a more general form: z = a x+/3 y ,
I a+/3=1
(32)
where a and/3 reflect the prescribed trust given to the corresponding data set. Evidently, the method of least squares (equation 11) corresponds to a =I, f l - 0 . Application of non-zero weight coefficients will be useful for model comparison with SAM, where the reliability of SAM is considered to be higher than that of the original models, but lower than that of measurements. An important characteristic of direct model comparison is the large volume of data (maps generally contain thousands of values), making more sophisticated analytical techniques needed for model-to-model than for model-measurement comparisons. Such techniques include a full set of spectral analysis (for example, a Fourier transformation) and a cluster analysis of the spatial distributions. Although these methods are quite promising, they are not common for the currently used model validation routines, and there is very little experience in their application. 5.2.3. Ranking procedure. Interpretation of the results Model comparison becomes more complicated if the number of models exceeds 3 or 4. Analysis following the scheme "each-to-each" requires a lot of resources and produces a large number of different results which have to be analyzed and interpreted. One possible way to systematically compare model results is by using a ranking procedure, which includes a number of steps: (i) separation of the model-measurement and model-to-model comparisons; (ii) creation of an artificial reference data s,~ (e.g. in a form of SAM) for the model-to-model
252
M.A. Softer
Figure 7. Scores on the basis of model-measurement comparison and comparison of model to the Statistical Average Model (SAM) from the Heavy Metal Model Intercomparison Project (Sofiev et al., 1996).
comparison; (iii) comparison of all models with measurements and with SAM. It is essential that SAM is also validated against available measurements; and (iv) The results of both routines are transformed to relative units and normalized to 1, which can be used as ranks. Their sum or mean value for each model yield the overall score of the model with regard to measurements and SAM. Figure 7 gives an example of the final scores of the Heavy Metal Model Intercomparison Project, showing that two out of seven models manifested both bad agreement with measurements and large deviation from SAM (models 2 and 6). For other models the deviations are often within the 67% confidence ranges of corresponding values, indicating that differences are not statistically significant. It should also be noted that SAM gives good agreement with measurements. The interpretation of the total scores is difficult. While the model-measurement comparison shows the model quality in a direct way, the model deviation from SAM only indicates that corresponding characteristics of this particular model differ considerably from those of most other models. The reason for the deviation can also be that this particular model includes processes or factors, ignored by other models. Some firm conclusions can only be drawn when model results are compared with both SAM and available measurements (see Figure 7, models 2 and 6).
6. Summary and conclusions The focus of this paper is on quantitative model validation as a basis for further case-specific assessment of the ability of a model to perform a particular task. Modem validation algorithms should also give an estimate of the reliability of the model results. The algorithms of the Quality Assurance model are, on all scales (local, regional, global), based mainly on the comparison with available observations. The methodology of direct model-to-model comparison is not well developed and accepted so far, but could contribute to model validation and characterization. The validation procedures and the requirements of the measurement data differ considerably for regional and global models on the one hand and local models on the other. The first group requires measurements representative of large areas and long time periods in a network that covers ideally all major attributes within the domain considered, including
Validation of model results on different scales
253
climatic zones, relief, large water basins, and main sources of emissions. The second group of models does not require spatially representative measurement data. A proper temporal resolution and a more detailed representation of the three-dimensional pattern of the pollution distribution close to the source are more crucial. Since local models are generally tested on commonly available experimental data obtained at various locations (sometimes outside the model application area), it is essential that such data sets represent the actual conditions (in particular, meteorological) of the pollution distribution in the model application region. The methods currently used for model validation consist of a limited number of standard tools, practically without analysis of the precision of model output. The paper gives a few examples of how non-trivial validation methods like autocorrelation or kriging analyses are applied as well as several promising techn. '~ogies which are still being devel.~ped.
References Barrett, K., O. Seland, A. Foss, S. Mylona, H. Sandnes, H. Styve and L. Tarrason (1995) European transboundary acidifying air pollution. Ten years calculated fields and budgets to the end of the first sulfur protocol. EMEP/MSC-W report 1/95, Co-Operative Programme for monitoring and evaluation of the long-range transmission of air pollutants in Europe, Meteorological Synthesizing Center West, Oslo, 71 pp. Berge, E., J. Schaug, H. Sandnes and I. Kvalvagnes (1994) A comparison of results from the EMEP/MSC-W acid deposition model and the EMEP monitoring sites during the four seasons of 1989. Proceedings of EMEP workshop on the Accuracy of Measurements. EMEP/CCC report 2/94, Co-Operative Programme for monitoring and evaluation of the long-range transmission of air pollutants in Europe, Chemical Co-ordinating Centre, Lillestrom, pp. 209-266. Ebel, A., H. Hass and H. Petry (1994) Correlation distances for air pollutants and implications for mesoscale modelling. Proceedings of F.'.IEP workshop on the Accuracy of Measurements. EMEP/CCC report 2/94, Co-Operative Programme for monitoring and evaluation of the longrange transmission of air pollutants in Europe, Chemical Co-ordinating Centre, Lillestrom, pp.201207. Fahnrich, B., J.E. Hanssen and Nodop, K. (1993) Comparison of measuring methods for nitrogen dioxide in ambient air, Kleiner Feldberg, Federal Republic of Germany, 21 April- 31 May 1991. EMEP/CCC report 3/93, Co-Operative Programme for monitoring and evaluation of the longrange transmission of air pollutants in Europe, Chemical Co-ordinating Centre, Norwegian Institute of Air Research, Lillestrom, 44 pp. Galperin, M. and M. Sofiev (1994) Errors in the validation of models for long-range transport and critical loads stipulated by stochastic properties of pollution fields. Proceedings of EMEP workshop on the Accuracy of Measurements. EMEP/CCC report 2/94, Co-Operative Programme for monitoring and evaluating the long-range transmission of air pollutants in Europe, Chemical Co-ordinating Centre, Lillestrom, pp. 162-179. Galperin, M., M. Sofiev and O. Afinogenova (1995) Long-term modelling of airborne pollution within the Northem Hemisphere. Journal of Water, Air and Soil Pollution 85:2051-2056 Galperin, M., M. Sofiev and T. Cheshukina (1996) An approach to zoom modelling of acid deposition on the basis of sulfur compounds evaluat.on for Sankt-Peterburg region. In: ?re-prints of the 4th Workshop on harmonization within Atmospheric Dispersion Modelling for Regulatory Purposes, Add. 1, Oostende, Belgium, pp. 1-7. Graziani, G., W. Klug and S. Mosca (1997) The European Tracer Experiment ETEX: A comparison of long-range atmospheric dispersion models in different weather conditions. In: Proceedings of the 22d NATO/CCMS International Technical Meeting on Air Pollution Modelling and its Applications. Clermont-Ferrand, France, June 1997, pp. 140-153.
254
M.A. Softer
Hass, H., E. Berge, I. Ackerman, H.J. Jakobs, M. Memmersheimer and J-P. Tuovinen (1996) A diagnostic comparison of EMEP and EURAD model results for wet deposition episodes in July 1990. Co-Operative Programme for monitoring and evaluation of the long-range transmission of air pollutants in Europe. EMEP / MSC-W Note 4/96, Meteorological Synthesizing Center West, 35 pP. Hass, H., E. Berge, P. Builtjes, A. Ebel, H.J. Jakobs, M. Memmesheimer, D. Simpson and R. Stern (1994) A comparison of long-range transport models applied to a European summer episode. In: P.M. Borrell, P. Borrell, T. Cvitach and W. Seiker (Eds.). Proceedings of EUROTRAC Symposium '94. SPB Academic, The Hague, 857 pp. Huber, P.J. (1981) Robust Statistics. Wiley series in probability and mathematical statistics. Wiley and Sons, New York, 304 pp. Irwin, J.S. (1997) Standard practice for statistical evaluation of atmospheric dispersion models (Draft issued on April 21 1997), North Carolina, 29 pp. Journel, A.G. and C.J. Huijbregts (1978) Mining Geostatistics. Academic Publishing Inc., London, 215 pp. Klug, W., G. Graziani, G. Grippa, D. Pierc-~ and C. Tassone (1992) Evaluation of Long Range Atmospheric Models using Environmental Radioactivity Data from the Chernobyl Accident." ATMES Report. Elsevier Science Publishers, Barking, UK, 366 pp. Lee, R.F. and J.S. Irwin (1995) A methodology for a comparative evaluation of two air quality models. Workshop Operational Short-Range Atmospheric Dispersion Models for Environmental Impact Assessments in Europe. International Journal of Environmental Pollution 5:723-733. Moussiopoulos, N. (1994) (Ed.) The EUMAC zooming model. Model structure and applications. EUROTRAC Special Publications, Garmish-Partenkirchen, 267pp. Olesen, H.R. (1995a) Data sets and protocol for model validation. Workshop on Operational ShortRange Atmospheric Dispersion Models for Environmental Impact Assessments in Europe. International Journal of Environmental Pollution 5:693-701. Olesen, H.R. (1995b) The Model Validation Exercise at Mol. Overview of Results. Workshop on Operational Short-Range Atmospheric Dispersion Models for Environmental Impact Assessments in Europe. International Journal of Environmental Pollution 5:761-784. Oslo and Paris Commissions (1994). Calculation of Atmospheric Input of Contaminants to the North Sea, 1987-1992. Oslo and Paris Commissions, London, 27 pp. Pedersen, U. (1994) Improvement of spatial correlation structure by use of anisotropic v~riogram analysis. Proceedings of EMEP workshop on the Accuracy of Measurements. EMEP/CCC report 2/94, Co-Operative Programme for monitoring and evaluation of the long-range transmission of air pollutants in Europe, Chemical Co-ordinating Centre, Lillestrom, pp. 190-200. Seilkop, S.K. (1994) Representativeness of a site in an 80 km grid. Proceedings of EMEP workshop on the Accuracy of Measurements. EMEP/CCC report 2/94, Co-Operative Programme for monitoring and evaluation of the long-range transmission of air pollutants in Europe, Chemical Co-ordinating Centre, Lillestrom, 208 pp. Semb, A., J. Schaug and J.E. Hansen (1994) Accuracy and precision in the EMEP data. A first evaluation on the basis of available information. Proceedings of EMEP workshop on the Accuracy of Measurements. EMEP/CCC report 2/94, Co-Operative Programme for monitoring and evaluation of the long-range transmission of air pollutants in Europe, Chemical Co-ordinating Centre, Lillestrom, pp. 53-66. Sirois, A. and R.J., Vet (1994) Comparability of precipitation chemistry measurements between the Canadian air and precipitation monitoring network (CAPMoN) and 3 other North American networks. Proceedings of EMEP workshop on the Accuracy of Measurements. EMEP/CCC report 2/94, Co-Operative Programme for monitoring and evaluation of the long-range transmission of air pollutants in Europe, Chemical Co-ordinat ~.g Centre, Lillestrom, pp. 88-114. Sofiev, M. (1994) Statistical Properties of the Model Verification Problem and Special Methods for Comparison of Measured and Calculated Data. In: P. M. Borrell, P. Borrell, T. Cvitas, W. Seiler (Eds.) Proceedings of Eurotrac Symposium '94. SPB Academic Publishing, The Hague, pp.869873.
Validation of model results on different scales
255
Sofiev, M. and M. Galperin (1994) Robustness of Methods for Comparison of Measured and Calculated Data. Proceedings of EMEP workshop on the Accuracy of Measurements. EMEP/CCC report 2/94, Co-Operative Programme for monitoring and evaluation of the long-range transmission of air pollutants in Europe, Chemical Co-ordinating Centre, Lillestrom, pp.315-341. Sofiev, M., L. Gusev and I. Strijkina (1994) Results of MSC-East current model calibration with measurement of SOx, NOx, NHx 1987-93. EMEP/MSC-E report 4/94, Co-Operative Programme for monitoring and evaluation of the long-range transmission of air pollutants in Europe, Meteorological Synthesizing Center East, Moscow, 125 pp. Sofiev M., A. Maslyaev and A. Gusev (1996) Heavy metal model intercomparison. Methodology and results for Pb in 1990. EMEP/MSC-E report 2/96, Meteorological Synthesizing Center East, Moscow, March 1996. Van Liedekerke, M. and A. Jones (1995) The ETEX Information System : Where GIS and WWW meet. Paper presented at the 15th Annual ESRI User Conference, May 22-26, Palm Springs, CA.
This Page Intentionally Left Blank
Chapter 13
ROLE OF ISOTOPES AND TRACERS IN SCALING TRACE GAS FLUXES
S.E. Trumbore
This Page Intentionally Left Blank
Approaches to scaling a trace gas fluxes in ecosystems A.F. Bouwman, editor 9 Elsevier Science B.V. All rights reserved
ROLE OF ISOTOPES AND TRACERS IN SCALING TRACE GAS FLUXES
S.E. Trumbore Department of Earth System Science, University of California, Irvine, CA 92697-3100, USA
I. Introduction Isotopes of an element undergo mass-dependent fractionation during physical, chemical or biological transformation. The isotopic signature of a gas emitted from land or ocean surfaces thus reflects the dominant mechanisms involved in its production and transport to the atmosphere. The isotopic signature of gases measured in the atmosphere will reflect the overall balance of surface sources and sinks, as well as modifications that occur during atmospheric chemical reactions. If isotopic signatures differ significantly among sources and sinks, the global budget for isotopic species provides a powerful constraint for the overall trace gas budget. Hence, isotopes provide a means to scale processes from the site to the globe. Isotopic data have been used to constrain the global budgets of trace gases like CO2, CH4, CO, and N20. For global and regional scales, these efforts have focused on two time scales. The first assumes a relatively well-mixed atmosphere (for long-lived trace gas species), and uses observations of change in the isotopic signature of atmospheric trace gases to determine how the global balance of sources and sinks have changed over the past years to millennia. Direct measurements of atmospheric isotopic composition are extended back in time using air sampled from ice cores and tim. These analyses provide estimates of the global importance of various sources and sinks, but not their spatial distribution. The second approach takes advantage of spatial variations in the concentration and isotopic signature of trace gases caused by proximity to local or regional sources or sinks. Recent efforts have expanded networks that monitor the isotopic composition of trace gases at many sites distributed over the globe. When combined with two- or three-dimensional tracer transport models, these observations may be "inverted" to infer the regional or zonal distribution of sources and sinks over seasonal to interannual time scales. Tracer gases are defined here as gases that have very simple source-sink characteristics and relatively well-known atmospheric budgets. For example, the spatial distribution of a gas that is not strongly influenced by biological or chemical processes predominantly reflects the rates at which it is mixed and transported by winds. Tracer gases are used to calibrate transport models that in turn are used to infer the latitudinal distribution of sources and sinks from the variations in the mixing ratio and isotopic signatures of gases in the atmosphere. Examples of tracer gases are radon, 85Kr, sulfur hexafluoride (SF6), and selected chlorofluorocarbons. Long-lived trace gases with known recent histories of atmospheric change, such as CO: or N20, are also useful as tracers of mixing in the upper troposphere and stratosphere (e.g. Boering et al., 1996). The use of these gases to test tracer transport models was reviewed in the 1995 IPCC report (Prather et al., 1995), and will not be discussed in detail here. This paper will review some of the recent advances in the use of isotopes to infer the global and regional balance of sources and sinks for four trace gases, CO2, CH4, CO, and N20. These
S.E. Trumbore
260 Table 1. Abundances of the isotopes discussed in this chapter. Element Hydrogen (H) Deuterium (D) Carbon (C) Nitrogen (N) Oxygen (O)
Mass
Abundance (%)
1 2 12 14 14 15 16 17 18
99.985 0.0148 98.89 <101~ 99.63 0.366 99.76 0.038 0.204
will serve as examples of both the power of using isotopes to constrain trace gas budgets, and the complications that affect their interpretation. Use of isotopes and tracers to determine the importance of microbial and physical transport processes at the plot scale will not be discussed in detail here, except in the context of how they will influence the isotopic signature of a specific trace gas source. This process-level understanding of how the environmental conditions (inundation, temperature, plant composition) determines the isotopic signature of the emitted trace is key to scaling from the plot (at which measurements are made) to larger scales. Table 1 summarizes the abundances of isotopes of H, C, N, and O that will be discussed in this chapter. For a review of analytical methods for measuring stable isotopes, see Trumbore (1995), and references therein. Isotopic data are commonly reported as the deviation in parts per thousand in the ratio of the rare to common isotope in the sample (Rx, where R equals 13C/12C, 15N/14N, 170/160, 180/160, or D/H), to that in a commonly accepted standard (Rstd):
8
=
I Rx
- Rstd
Rstd
1 X 1000 ,for example, 6~3C = ~"~2C!x~3 ( C ~" C)"d
xl000
(l)
3
Most biological, physical, and chemical processes occurring at the earth's surface involve mass-dependent fractionation of isotopes. That is, the discrimination against ~80 by a process like diffusion, dissolution, or microbial transformation, is twice that against 170, because the mass difference between ~80 and 160 is twice that between 170 and 160. Recently, however, mass independent fractionation of isotopes in atmospheric trace gases has been observed. In such cases, fractionation of 180 versus ~~ is less than twice that of 170 and 160. Mass independent fractionation occurs in gas phase reactions not at thermodynamic equilibrium (eg. reactions with radicals) (Thiemens et al., 1997). While not always important for understanding sources and sinks at the earth's surface, mass independent fractionation provides a signature of chemical processing in the atmosphere that may be useful for understanding budgets of shorter-lived, more reactive gases like CO (Brenninkmeijer and Roeckmann, 1997). Isotopic measurements of atmospheric trace gas species are not routinely made in situ, though spectroscopic methods are under development. Normally, air samples are collected in the field, and the gas of interest is separated and purified in the laboratory, either chromatographically and/or cryogenically on a vacuum line. The isotopic measurement is made using a mass spectrometer. Sampling at the chamber scale in terrestrial ecosystems involves either observation of the changes in headspace concentration and isotopic content over time, or trapping of the trace gas that has accumulated in the chamber headspace.
Role of isotopes and tracers in scaling trace gas fluxes
261
At the scale of small towers (which integrate up to 1 km2), large volume samples needed for isotopic analysis may be collected by relaxed eddy accumulation methods. An alternative method is the collection of samples at several heights in and above the canopy; the correlation of gas concentration with isotopic signature yields a mixing curve between free tropospheric and ecosystem source characteristics for non-reactive species like CO2. At larger scales, inferences about sources are derived by combining transport models with atmospheric sampling networks.
2. Recent advances in the use of isotopes and tracers in scaling of gas fluxes 2.1. Carbon dioxide
Carbon dioxide has three isotopes that have been used as tracers: stable isotopes 13C and 180, and the radioactive isotope, 14C. In addition, the links between atmospheric O2 variations and organic carbon make O2 a tracer for the global CO2 cycle. Each of these species provides slightly different information about the sources and sinks of anthropogenic CO2. The discrimination against ~3C by biological uptake during photosynthesis is much larger than that associated with dissolution of CO2 in the ocean. Hence, the overall ~3C signature of atmospheric CO2 will reflect the relative strengths of biosphere and oceanic sinks. Variations in the 180 of CO2 primarily reflect isotopic exchange of O with plant and soil water, and latitudinal gradients may provide information on gross photosynthesis and respiration in terrestrial ecosystems (Ciais et al., 1997). Atmospheric O2 primarily reflects the production and destruction of organic matter in the marine and terrestrial biosphere, including fossil fuel and biosphere burning (Keeling et al., 1993). Radiocarbon measurements in the atmosphere, which have been corrected for isotopic fractionation effects, reflect the "age" distribution of C sources to the atmosphere. Fossil fuels, which have zero radiocarbon, are thus distinguished readily from biosphere and surface ocean sources (Levin et al., 1992; Manning and Melhuish, 1994) . Further, the fate of bomb 14C, which was released in the atmosphere in the early 1960's and subsequently taken up by ocean and biosphere reservoirs, is useful for determining the rate of exchange of C among atmosphere, biosphere and ocean reservoirs on decadal time scales. Taken all together, CO2 mixing ratios, 13C, 180, 02 and pre- and post-bomb 14C should provide powerful constraints on the distribution of anthropogenic CO2 among atmosphere, ocean and biosphere reservoirs. Why, then, is there still considerable uncertainty over the fate of anthropogenic CO2 ? Over the last decade to century, 13C, 14C and 02 from direct air sampling and air archived in ice cores has been used to interpret the global partitioning of CO2. Figure 1 illustrates the use of 02 to distinguish the contributions of biosphere and ocean to explaining both CO2 and 02 in the atmosphere as a whole over time. Oxidation of organic matter (eg. by fossil fuel burning or by aerobic decomposition or plant respiration) will increase CO2 and decrease 02 in a fixed ratio (roughly 1:1.1), while CO2 is far more affected by ocean exchange than 02 (because of the greater solubility of CO2). Hence the relative changes in CO2 versus 02 (assuming the quantity of fossil fuel burned is known) may be deconvolved into that due to ocean and biosphere uptake. Because the marine biosphere is thought to be small and relatively constant in size over the past several decades (compared to the vast changes occurring in terrestrial ecosystems), the uptake of CO2 by the biosphere inferred in Figure 1 for the early 1990's is interpreted as mostly taking place on land. The interpretation of 13C data is in essence similar to that for 02; net storage of C in the biosphere has more discrimination against the heavy isotope (greater slope of CO2 change to
262
S.E. Trumbore
01
-100
-
-120
-
-140
-
-160
-
-180
-
E I_
Q. A
Z
0
Changeexpectedif all fossil fuel remained in the atmosphere ", ~ (slopeequivalentto 1.1 mole02 consumed 9 9 ~9 per mole 002 r e l e a s e d ) Observed ",. ~ s,ooe
o,I
land biosphere (same slope as Oceanuptake fossil fuel buming) (slope= O)
r.x3
I 352
I
I
I
356
360
I
I
364
CO ppm 2
Figure l.The trends in CO2 and 02 changes in the atmosphere over several years may be used to deconvolvethe roles of ocean and biosphere net exchange. CO2 release and 02 consumptiongive rise to a slope of- 1.1 for fossil fuel burning (or + 1.1 for net uptake of C by the biosphere and net release of 02). Oceans are assumed to take up CO2 but not affect atmospheric O2. If the release of CO2 by fossil fuel burning is known independently, the relative roles of ocean and biosphere uptake of C may be solved for graphically. After Keeling et al. (1995).
13C change) than uptake by dissolution in the ocean. However, interpretation of 13C data is complicated by two major factors (Fung e: al., 1997). First, the photosynthetic pathways used by plants fractionate isotopes differently. Plants using C3 photosynthesis discriminate much more against the heavier ]3C isotope than do plants with the C4 (or CAM) pathway. Hence, a greater or lesser importance of the oceanic sink is inferred depending on the global partitioning of net photosynthesis among C3 and C4 plants. The second complication, which has been dubbed "isotope disequilibrium", takes account of the fact that the lac in atmospheric CO2 is decreasing at a rate o f - 0.6 per mil per year (in the 1980's) due to the accumulation of 13C-depleted fossil-fuel-derived CO2. This effect is illustrated in Figure 2. Since the lag time between photosynthesis and respiration may be years to decades for organic matter tied up in tree biomass and soil organic matter, carbon respired from terrestrial ecosystems today is enriched in ~3C compared to the ~3C taken up by photosynthesis. A larger isotope disequilibrium gives rise to the same isotopic signature (accumulation of 13C-enriched CO2 in the atmosphere) as a net biospheric sink of carbon. Failure to account for isotope disequilibrium caused early efforts to invert atmospheric 13CO2global distributions to overestimate significantly the strength of a northern terrestrial carbon sink (Ciais et al., 1995b; Fung et al., 1997). Lotope disequilibrium effects are also expected for ocean-atmosphere exchange (Tans et al., 1993). Radiocarbon in atmospheric CO2 provides information on the residence time of C in different reservoirs. Separate global radiocarbon budgets must be constructed for the periods before and after-1950. For pre-1950 radiocarbon, production by spallation in the upper troposphere and stratosphere should balance loss by radioactive decay in all organic and inorganic carbon reservoirs. For reservoirs that exchange C slowly with the atmosphere (over centuries to millennia), the degree to which ~4C is depleted because of radioactive decay (half-
Role of isotopes and tracers in scaling trace gas fluxes
263
Figure 2. The solid line shows 13Cchanges in ice cores (open circles; Friedli et al. [1986]) and the atmosphere (Cape Grim, open triangles; from R.J. Francey, C.E. Allison, E.D. Welch, I.G. Enting, H.S. Goodman, In situ carbon-13 and oxygen-18 ratios of atmospheric CO2 from Cape Grim, Tasmania, Australia, 1982-1993; CDIAC data base). The effect of a significant time lag between uptake of CO2 from the atmosphere by photosynthesisor dissolution and subsequent release by respiration or degassing is shown. Significant lags have been shown to exist in ~4Cmeasurementsof C respired by terrestrial ecosystems(Gaudinski et al., 1998).
life = 5730 years) reflects the average amount of time since the reservoir has been isolated from exchange with atmospheric CO2. Addition of fossil fuel C the atmosphere dilutes the 14C of atmospheric CO2. This effect (dubbed the "Suess" effect), is observed in the decreasing trend of 14C in cellulose of tree rings that grew between the late 1800's and about 1950 (Figure 3). Deconvolution of the pre1950 atmospheric CO2 record has been accomplished using both ~3C and ~C data together. The amount of C derived from fossil fuel burning in the atmosphere may be determined from the ~4C dilution. Given the ~3C signature of fossil fuel-derived CO2, the corresponding decrease in atmospheric ~3CO2 may be calculated. If the atmospheric 13CO2 decrease exceeds that predicted, there is a net source of biospheric C (which is depleted in 13C but not lac) to the atmosphere. These deconvolution studies were used to show the importance of net sources of CO2 from the biosphere between 1850 and 1900 (e.g. Siegenthaler and Oeschger, 1987). The second constraint provided by the radiocarbon tracer is the global distribution of 14C produced by atmospheric testing of thermonuclear weapons, called bomb 14C (Figure 3). Since the time of peak production in 1963, the atmospheric ~4CO2 levels have declined, reflecting net uptake by the ocean and terrestrial C reservoirs, and continued addition of 14C-free fossil fuel CO2 to the atmosphere. The major net sink for bomb ~4C has been oceanic uptake, and the time-dependent distribution of bomb ~4C in the oceans has been used to constrain models of ocean ventilation, mixing and transport. Radiocarbon in terrestrial ecosystem respiration may be used as a direct measure of the magnitude of isotope disequilibrium (Figure 2) in ~3C. For example, the ~4C content of CO2 respired from temperate forest soils is greater than contemporary atmospheric values, indicating an average lag time of several years between C fixation by photosynthesis and release by decomposition (Gaudinski et al., 1998). Measurements of 180 in atmospheric CO2 show large latitudinal gradients within each he misphere and differences in annually averaged 180 values between northern and southern he-
264
S.E. Trumbore
Figure 3. History of radiocarbon in the atmosphere, showing the Suess effect (a dilution of atmospheric 14Cby addition of 14C-free carbon released from fossil fuels), and the changes observed due to the bomb ~4Cspike. Data are from Stuiver and Quay (1981), Burcholadze et al. (1989), Manning and Melhuish (1994) and Nydal and LOvseth(1996).
mispheres. To maintain these gradients against atmospheric mixing requires that isotopic exchange with the surface be rapid. Francey and Tans (1987) inferred that exchange of CO2 with leaf and soil water would explain the observed latitudinal trends. Since the rates at which CO2 will equilibrate with leaf water depend on factors like leaf resistance to gas transfer, seasonal and interannual variations in the 180 in CO2 offer the opportunity to constrain gross, rather than net CO2 fluxes associated with photosynthesis and respiration (see review by Ciais et al., 1995a). The latitudinal distribution of sources and sinks of CO2 have been inferred from inverse modelling using data from a global network of stations monitoring 13CO2 and CO2. Tans, Fung and Takahashi (1990) compared the observed latitudinal gradient in CO2 mixing ratios in the 1980's with predictions using a tracer transport model coupled with estimates of the spatial distribution of fossil fuel sources and ocean sources and sinks. Ciais et al. (1995a,b) compared the observed latitudinal gradient of both CO2 and 13CO2 with those expected from an atmospheric tracer model with a prescribed spatial distribution of sources and sinks. The overall approach is shown in Figure 4. Both studies infer the existence of a significant net sink of anthropogenic CO2 in the terrestrial biosphere in the northern hemisphere during the late 1980's, which likely intensified in the early 1990's. Keeling et al. (1995) infer large interannual variability in the strength of this C sink from changes in the rate of CO2 accumulation (and 13CO2 decrease) in the atmosphere, which correlates with variations in climate. A net sink of C in the biosphere is also inferred from atmospheric 02 in the early 1990's (Figure 1). Some disagreement between data sets obtained by different sampling networks (e.g. Francey et al. (1995) versus Keeling et al. (1995)) highlights the importance of intercalibration efforts among isotope laboratories. An uncertainty in the interpretation of interhemispheric gradients of trace gases arises from what has been dubbed the "rectifier" effect (Denning et al., 1995). This effect arises because of the covariance of seasonal CO2 exchange by terrestrial ecosystems (net drawdown in the
265
Role of isotopes and tracers in scaling trace gas fluxes 5
r to r
v
r-
F| ~-
Predicted if all fossil fuel CO 2 remains in
3
L
the atmosphere
2
11_
4
,, .. " ,,""
~
... .....
9
E
o E ,,=... E
0
9 g:
"1o
r
,,"
- ' ' ~ o n s
(smoothed)
1 0 -
-80
-40
0
40
80
40
80
Latitude
E._.
0.10
NE o
o
~E ~ 0.00 o
-o.~
-0.10
Predicted if all foss
2Z; e2
.o
""
.~_ -0.20 > ra0 -80
-40
0 Latitude
Figure 4. The approach used in interpretation of the latitudinal distribution of C O 2 s o u r c e s and sinks from comparison of the observed spatial distribution of CO2 and ~3CO2 with that predicted from 3D transport models with prescribed source distributions. The difference between predicted and observed values must be combined with oceanic and biospheric fluxes. After Fung et al. (1997).
spring/sununer and release in the fall/winter) with the height of the planetary boundary layer (highest in summer, lowest in winter). In the annual mean, these effects give rise to a net positive concentration anomaly in the air near the surface (which is where flask samples are taken). Including the rectifier effect, tracer models predict a larger interhemispheric gradient for a given distribution of sources by latitude. Since a northern hemisphere sink is predicted because observed interhemispheric gradients are smaller than predicted (see Figure 4), including the rectifier effect tends to increase the size of inferred northern hemisphere sink for CO2. Because the rectifier effect is largely biological in origin, it will also affect predictions of the latitudinal trend in 13CO2 (Fung et al., 1997). Fung et al. (1997) summarize the present state of understanding of the C cycle based on interpretation of 13CO2 data. They conclude that uncertainties in: (i) the relative abundance of C3 and C4 plants, and (ii) the magnitudes of isotope disequilibrium effects in both ocean and terrestrial C reservoirs presently limit the ability to determine the partitioning of anthropogenic CO2 among biosphere and ocean sinks. The addition of atmospheric 02 (which will presumably be less affected by the type of plant tissue being respired and the time it was fixed from the atmosphere) wiR provide better constraints on this partitioning, especially as the number of monitoring stations and the length of the record increases. However, factors such as the rectifier effect, and questions a~ to the constancy of the ratio of CO2 released to 02 consumed for different types of organic matter decomposition, require further investigation. In summary, the different isotopes that can be measured in atmospheric CO2 constrain the
266
S.E. Trumbore
overall global balance of fluxes among atmosphere, biosphere, ocean and fossil fuel reservoirs. Most of the work to date has focused on (i) long-term trends in 13C02 recorded in ice cores; (ii) inversion modelling of 13C02 from a global monitoring network to infer the latitudinal distribution of sources and sinks; and, (iii) modelling of the fate of bomb-14C over the past several decades. Interpretation of lSo in C02 and in particular, the design of sampling strategies to infer net C02 source and sink distributions over regional and smaller spatial scales are under development. Uncertainties associated with inversion modelling remain because of issues like the isotope disequilibrium, the changing distribution of C3 and C4 plants, and the sampling biases like the rectifier effect. 2.2. Methane
The use of Deuterium (D), ~3C, and 14C to constrain the global CH4 budget has been summarized by several authors, most recently by Tyler (1991), Wahlen (1993) and Manning et al. (1989). The isotopic signature of tropospheric methane represents a balance between various sources. The largest natural sources of CH4 to the atmosphere are anaerobic microbial production in wetlands, rice paddies and ruminant stomachs. The isotopic signatures of C in the methane produced by these processes depends on the C in the organic matter substrate from which CH4 is produced, and the pathway of CH4 production (acetate fermentation versus CO2 reduction by H2). The D/H ratio of CH4 represents a combination of the D/H ratio of the water where methane is produced and the CH4 production pathway. A further complication affecting CH4 isotopes is the potential for CH4 to be microbially oxidized to CO2 within the soil, sediment or water column prior to being emitted into the atmosphere. Microbial sources of CH4 in general are highly depleted in the heavier ~3C and D isotopes compared to atmospheric CH4. The most 13C-enriched CH4 is produced from incomplete combustion during biomass burning and from thermogenic natural gas. For the isotopic budget of methane to balance, the flux weighted sum of sources for each isotope (12C, 13C, H and D) must equal the sinks. The major methane sink is destruction by photochemical reactions in the atmosphere, with roughly 5% of the annual loss due to microbial uptake in well-drained soils. The isotopic flux due to the atmospheric sinks will depend on the kinetic isotope effect (the degree to which the lighter isotope reacts faster) and the isotopic composition of atmospheric methane. For the major CH4 sink, reaction with the OH radical, the kinetic isotope effect is roughly 5 per mil in ~3C (Cantrell et al., 1990; Tyler, 1991). Hence the sum of methane sources must be such that the 13CH4 is --~ -52 per mil (- 5 per mil lighter than the measured atmospheric value of-~-47%o). One source of uncertainty in the present estimates of the global CH4 budget from isotopic constraints are the isotopic signatures and relative importances of different methane sink processes. The sinks for atmospheric methane include reaction with OH and C1 radicals (Habstraction reactions that initiate oxidation of CH4 to CO2), and microbial oxidation in soils. The reaction with C1 has been noted only recently, and may influence vertical changes in the isotopic signature of CH4 in the upper troposphere (Saueressig et al., 1996; Bergamaschi et al., 1996; Gupta et al., 1996). However, the kinetic isotope effect (KIE) associated with the C1 reaction is uncertain. The reaction with C1 is of interest because it is increasing in importance as a CH4 sink as the amount of CI produced from photolysis of chlorofluorocarbons released by humans increases over time. Changes in the atmospheric OH distribution in response to human influences on the chemistry of the stratosphere will also affect the isotopic signature of atmospheric methane. Meth~e oxidation in aerobic soils, while thought to be a relatively minor proportion of the total annual CH4 sink, can affect the isotopic budget significantly because it fractionates the
Role of isotopes and tracers in scaling trace gas fluxes
267
isotopes more than the reaction with OH. The fractionation factor for aerobic methane oxidation has been determined in only a few soil types and at only a few sites globally. In addition to determining how the fractionation factor may vary with factors like soil type and climate, the global estimate of the effect of the soil sink on the methane isotopic budget needs refinement. Radiocarbon measurements of atmospheric CH4 have been used to estimate the fossil fuel contribution of CH4 sources. Fossil fuel emissions contain no radiocarbon. Most biogenic sources of CH4 have radiocarbon signatures close to that of contemporary atmospheric CO2, indicating the labile nature and recent origin of the substrates used for CH4 production (Chanton et al., 1995) . Interpretation of the lac in atmospheric CH4 is complicated by the increased dominance of 14CH4released from pressurized water nuclear reactors (PWR). Using various methods to correct for the PWR source of laC, including analysis of archived air samples, the lac budget approach estimates some 16-21% of the total CH4 sources from fossil fuels (Wahlen, 1993). Direct estimates of the fossil fuel contribution to the CH4 budget are significantly smaller than those inferred from radiocarbon data (Cicerone and Oremland, 1988). The reason for the difference has been attributed to either a source of ~4C-depleted methane from the biosphere or underestimation of natural gas seepage. Recently, Zimov et al. (1997) demonstrated that CH4 generated in Siberian "kashka" lakes had ~4C-depleted signatures that reflected the age of the thawed sediments that are the likely subst~'ate for CH4 production. However, their estimate of the global importance of this source is too small to make up the difference between estimates of the "old" CH4 source. Ice cores record large changes in the tropospheric mixing ratio of CH4 during the past -250,000 years (Chappellaz et al., 1993). These changes may be explained by variations in the sources of methane to the atmosphere, or by changes in the lifetime of methane in the atmosphere related to the amount of OH available for reaction with CH4. Isotopic measurements of methane in ice cores are currently being attempted and should help to distinguish between source and sink changes. Similarly, measurements in firn ice and rapidly accumulating ice cores should yield information as to the changes in CH4 sources related to human activities over the past century. Continuous records of CH4 isotopes are available for a few sites worldwide, and observations onboard ships are giving better measures of the latitudinal distribution of CH4 isotopes in clean marine air (D. Lowe, personal communication, 1997). Observations of the D in CH4 are particularly sparse because of the difficulty of measuring D/H ratios in CH4. These studies have shown (i) the importance of biomass burning as a CH 4 source in the southern hemisphere; (ii) the seasonal influence of changes in the rate of CH4 reaction with OH; and (iii) a decrease in the 13C of atmospheric methane with time (Gupta et al., 1996; Manning and Melhuish, 1994; Quay et al., 1991). 2.3. Carbon monoxide
The lifetime of carbon monoxide (CO) in the atmosphere is of the order of months, shorter than that of either CO2 or CH4. Hence, its mixing ratio is low (- 50 -200 ppbv) and highly variable in space and time. The major sources of CO include direct surface emissions and oxidation of non-methane hydrocarbons (together accounting for-75% of sources) and oxidation of CH4 (25% of sources). The major sink is reaction with OH, with a smaller sink in soils. The KIE for ~3C associated with these sinks are "-3 and --6 per mil, respectively (Stevens and Wagner, 1989).
268
S.E. Trumbore
Isotopes of 13C and 180 in atmospheric CO and its sources were first measured by Stevens and Krout (1972). The major sources of CO are distinct in their stable isotopic signatures (see Manning et al., 1997 and Brenninkmeijer and Roeckmann, 1997 for recent reviews). Biomass burning is the most 13C-enriched source (--21 per mil), with a signature fixed by the 513C value of the plant material being burned. The most 13C -depleted source (--32 per mil) is oxidation of atmospheric methane. Oxidation of non-methane hydrocarbons (-32 per mil) and direct emissions from fossil fuel combustion (--25 per mil) have intermediate 513C values. The 180 of fossil fuel-derived CO is from atmospheric 02, and enriched by 10-30 per mil relative to background C~80 values. Mass-independent fractionation of O isotopes have been observed for photochemical and radical chain reactions leading to CO production (e.g. from non-methane hydrocarbons; Roeckmann et al., 1998). A seasonal CO budget for the southern hemisphere has been derived using mixing ratio and isotopic information based on remote air sampling stations (Manning et al., 1997). Seasonal variations in the amount of CH4 oxidation (which would tend to yield lighter 13CO when OH mixing ratios are high in the austral summer) are offset by changes in the overall isotopic signature of surface-derived CO sources (increased biomass burning, yielding more 13Cenriched CO in the austral summer). Stable isotopes in CO from air sampled at non-remote stations will reflect the local to regional balance of CO sources. Conny et al. (1997) used a three-dimensional regional transport/diffusion model with prescribed CO and 13CO sources to predict the variability in CO mixing ratios and isotopes at four sampling stations in Brazil. The model predicted large site to site and week to week variability in CO mixing ratios and isotopes during the biomass burning season, depending on how efficiently burning plumes were transported to the sampling station. The authors caution that such large variability means that CO isotopes measured for any given day should not be taken as representative of the regional balance of CO sources. The radiocarbon content of atmospheric CO yields little direct information about the global or regional CO budget, since it is dominated by the production of radiocarbon in the stratosphere and upper troposphere. Instead, the global distribution of 14CO is used to infer the distribution of OH in the atmosphere (see Volz et al., 1982; Mak et al., 1994).
2.4. Nitrous oxide
The use of stable isotopes of N and O to constrain the global N20 budget is still in .its early stages. N20 is an important greenhouse gas and it is broken down in the stratosphere into NO, which catalyzes the destruction of stratospheric ozone. N20 is increasing in the atmosphere at a rate of about 0.25% per year, though the exact cause of the increase is still not determined. The major sources of N20 are microbial nitrification and denitrification in soils and in oceans, and the major sinks are photolysis and oxidation by O(ID) in the stratosphere. Attempts to construct a global budget for N20 presently show estimated sources - 30% less than estimated sinks (WMO, 1995). Kim and Craig (1993) attempted the first budget for N20 constrained by stable isotopes. Their budget was influenced by the value they assigned to the KIE associated with N20 sinks. Destruction of N20 by photolysis (90% of the sink) and reaction with O(1D) (10% of the sink) discriminates against the heavier isotopes, resulting in enrichment of 15N and 180 in the surviving N20 that mixes down into the troposphere from the stratosphere. Kim and Craig hypothesized that stratospheric N20, which is enriched in 15N and 180 compared to average tropospheric values, should be balanced by addition of isotopically lighter (depleted in 15N
Role of isotopes and tracers in scaling trace gas fluxes
269
and ~80) N20 produced by nitrification and denitrification sources in soils and oceans. Recent analyses by Rahn and Wahlen (1997) suggest that the stratospheric enrichment of 15N and 180 in N20 reported by Kim and Craig was too large, and that budgets of N20 derived from isotopic constraints are roughly in agreement with the IPCC budgets for N20 (Prather et al., 1995). An interesting issue raised by the results of Rahn and Wahlen (1997) is that the fractionation factors they derive for the destruction of N20 in the stratosphere are larger than those predicted from laboratory simulations of photochemical and O(1D) reactions. These differences may derive from the wavelength used for photolysis determinations, coupled with strong wave-length dependence of the fractionation factor for photolysis (Yung and Miller, 1997). Cliff and Thiemens (1996) have observed mass-independent fractionation of O isotopes in N20. The characterization of the isotopic signatures of various N20 sources also requires more effort. For example, the value for N20 emitted from tropical forest soils (the largest natural source of N20 ) was based on only 5 determinations (in 3 locations). The range of variations was quite large and most likely reflects differences in the pathways of N20 production (nitrification versus denitrification) and consumption (reduction during denitrification of N20 to N2) in soils. These different processes produce markedly different isotopic signatures for N20 emissions (P6rez et al., 1997). More work on characterizing the isotopic signatures of N20 sources is needed if isotopic constraints will be useful in determining the N20 budget. Given the long lifetime of N20 (100-150 years), differences in the N20 signature in tropospheric air due to an increase in a specific source may take decades of longer to be measurable. Hence measurements of N20 isotopes in firn air may provide the best evidence for how the balance of N20 sources may have changed since pre-industrial times. 2.5. Tracers
Some of the uncertainties in interpreting observational constraints such as the interhemispheric gradient of a trace gas, are related to the rate and mechanism of transport of air across the equator (or the stratosphere-troposphere boundary, in the case of N20 and CH4). For example, a north-south gradient may be smaller than expected, either because the source-sink distribution with latitude that was assumed is incorrect, or because mixing of air across the equator is faster than assumed. Interhemispheric transport rates are largely deduced from the observed latitudinal distribution of anthropogenically released halocarbons, such as CFC's. Since most of the release is in the northern hemisphere for these gases, the overall atmospheric mixing ratio and the difference between mixing ratios in the northern and southern hemisphere are determined by (i) the atmospheric lifetime of the gas, and, (ii) the interhemispheric mixing ratio. For some gases (like 85Kr emitted from nuclear power plants), the lifetime is well known, allowing estimation of the interhemispheric mixing rate. The "rectifier" effect, discussed above, raises questions of how representative trace gases sampled in the planetary boundary layer are of free tropospheric (or average tropospheric) air. In addition, differences in the vertical distribution of sinks in the atmosphere will affect the representativeness of a surface sample of the tropospheric average. For example, methane isotopic and CO values change with height in the troposphere (Gupta et al., 1996; Brenninkmeijer and Roeckmann, 1997). As pointed out in Prather et al., (1995), rigorous comparison and testing of various 3D tracer transport models is in its beginning stages. The intercomparisons are based on the
270
S.E. Trumbore
transport of Rn-222 (a noble gas emitted from the earth's surface and with a half-life with respect to radioactive decay of 3.8 days). While observations of 21~ (the ultimate daughter product of Rn-222 decay) deposition are available to test the long-term average Rn-222 transport patterns predicted by climate models, measurements of Rn-222 in the atmosphere (and particularly in the upper troposphere) are few. Other potential tracers for which observation networks exist (like D and 180 in precipitation) have been predicted only by a few 3D models. Given the importance of transport in inversion modelling, more work using tracer data to improve and test transport models is needed for improved interpretation of trace gas budgets.
3. Areas for future research 3.1. Expansion of monitoring networks Recent technological advances in continuous flow stable isotope ratio mass spectrometry (CFIRMS) have enabled more rapid and sensitive measurements of stable isotopes in atmospheric trace gases. Inlet systems that provide chromatographic separation, followed by combustion, mean that measurements may be more rapid since vacuum-line separation methods are no longer needed. Greater sensitivity means that analyses may be made on much smaller whole-air samples. These methods are still under development, and there may be problems of analytical precision for some of the gases involved. However, in the near future, it is a reasonable guess that the capacity for measuring stable isotopes in air samples will be far greater than in the past. Similarly, the use of mass spectrometric methods to measure atmospheric 02 variations have increased capacity for monitoring of atmospheric 02. Where should extra monitoring capacity be employed to best constrain trace gas budgets? Are extra resources best put into enhanced atmospheric monitoring if uncertainties in interpretation (such as the global distribution of C3 vs. C4 plants, and the isotope disequilibrium questions raised by Fung et al., 1997) ultimately limit interpretation of the atmospheric data? Recently Accelerator Mass Spectrometry (AMS) measurements of 14C have improved in precision so as to come close to the best ~4C counting labs. Is it worth expanding existing atmospheric monitoring of radiocarbon so as to better account for spatial and temporal variations in the addition of fossil fuels to the atmosphere? This might be a potentially important verification tool for enforcing global CO2 emissions limitations. Studies using tracer transport models are needed to predict the locations where enhanced sampling will be most effective in constraining trace gas budgets. For example, does the "rectifier effect" potentially influence the interpretation of interhemispheric gradients in methane and nitrous oxide? Measurements of an inert and short-lived gas such as 222Rn would be useful, especially in determining the rate of vertical transport of gases emitted from the earth's surface to the upper troposphere by deep convective processes. These studies should be combined with regional predictions of isotopic fluxes derived from process-level models.
3.2. Model development In terrestrial ecosystems, regional to global scale process models that predict the emissions of trace gases given variables like vegetation type, soil type, temperature and precipitation are being developed (for an example using methane, see Potter et al., 1997). The same factors that predict the magnitude of the trace gas flux (for example, temperature and soil water content for N20; ecosystem productivity, vegetation type and water table depth for CH4), may also be
Role of isotopes and tracers in scaling trace gas fluxes
271
used to estimate the isotopic signature of the emitted trace gas. Comparison of isotopic values predicted by models with measurements will provide a stringent test of the model, which may then be used to scale from the plot to the region. Basic research is also needed to tie the isotopic signature of the trace gas to the processes producing, transporting, and consuming it. Particularly problematic are the development of predictive process models for gases like CH4 and NzO, which are both produced and consumed in soil and sedimentary environments from which they are emitted. The variations in some trace gases may be linked to others. Just as 02 may be used as an "isotope" for CO2, some trace gas "assemblages" may be used to characterize a given source. For example, biomass burning sources show predictable relationships between gases like CH4, CO2 and CO. The ratio of NO to N20 emi*ted from tropical soils has been ~sed as a clue for determining which process (nitrification versus denitrification) is producing these gases.
3.3. Reducing uncertainties Much remains to be determined about how production and consumption pathways combine with transport mechanisms to determine the isotopic signature characteristic of a given set of site conditions. Data collection needs to be tied to process models so as to best evaluate the causes of spatial and temporal variability in the isotopic signatures of trace gas sources and sinks. An obvious need is the quantification of isotope disequilibrium in terrestrial and ocean systems. This has ramifications beyond the interpretation of atmospheric ~3C data. For example, the lag time between photosynthesis and respiration in the terrestrial biosphere is a measure of the amount of C that can be temporarily stored given an increase in plant productivity (and how long such a sink would last).
3.4. Expansion of methods to look at other trace gases The improved sensitivity of stable isotope and radiocarbon analyses allow for studying the isotopic signatures of trace gases not measured to date. For example, questions about the relative role of anthropogenic emissions for gases like methyl bromide, acetone, or carbonyl sulfide, could be answered with relatively few radiocarbon measurements (assuming the anthropogenic emissions derive C in these gases from fossil fuel as opposed to biospheric C). While these measurements are analytically quite challenging (since the mixing ratios for the gases in question are in the 10-100 pptv range, and the AMS measurement of radiocarbon still requires something like 20 lag of carbon), advances in automation of chromatographic collection techniques mean that the measurements are feasible, if difficult.
3.5. Mass independent fractionation of isotopes The presence of mass-independent fractionation for some reactions involving isotopes of CO2, N20 and CO provides an opportunity to assess the relative importance of these reactions compared to the strength of biogenic and oceanic sources. At the same time, the processes leading to mass-independent fractionation of isotopes need to be better understood, since they limit our ability to predict KIEs for trace gases consumed by reaction with free radicals in the atmosphere.
272
S.E. Trumbore
4. Conclusions Isotopes are powerful tools for constraining the global budgets of trace gases. However, significant uncertainties limit their usefulness in answering questions involving relatively small perturbations of trace gas budgets. At the plot scale (not discussed here), there is no doubt that isotopic tracers aid in determining the balance of production, consumption and transport involved in trace gas emission from ecosystems. At the global scale, changes in the average content of the atmosphere over time of isotopic signatures in long-lived trace gases can significantly constrain past changes in sources and sinks. What remains most uncertain is the usefulness of isotopes for constraining sources and sinks at intermediate scales. Confidence is gained if sources and sinks are derived using more than one isotopic method. Future advances in instrumentation will allow greater spatial and temporal coverage in sampling of isotopes in the atmosphere and for characterizing trace gas source processes.
References Boering, K.A., S.C. Wofsy, B.C. Daube, H.R. Schneider, M. Lowenstein, and J.R. Podolsky (1996) Stratospheric mean ages and transport rates from observations of carbon dioxide and nitrous oxide. Science 274:1340-1343. Bergamaschi, P., C. Bruhl, C.A.M. Brenninkmeijer, G. Saueressig, J.N. Crowley, J.U. Grooss, H. Fischer and P.J. Crutzen (1996) Implications of the large carbon kinetic isotope effect in the reactions CH4+CI for the C-13/C-12 ratio of stratospheric CH4. Geophysical Research Letters. 23:2227-2230. Brenninkmeijer, C.A.M. and T. Roeckmann (1997) Principal factors determining the 180/~60 ratio of atmospheric CO as derived from observations in the southern hemispheric troposphere and lowermost stratosphere. Journal of Geophysical Research 102:25477-25486. Burcholadze, A.A., M. Chudy, I.V. Eristavi, S.V. Pagava, P. Povinec, A. Sivo and G.I. Togonidze (1989) Anthropogenic ~4C variations in atmospheric CO2 and wines. Radiocarbon 31:771-776. Cantrell, C. A., R. E. Shetter, A.H. McDonald, J.G. Calvert, J.A. Davidson, D.C. Lowe, S.C. Tyler, R.J. Cicerone and J.P. Greenburg (1990) Carbon kinetic isotope effect in the oxidation of methane by hydroxyl radicals. Journal of Geophysical Research 95:22455-22462. Chanton, J.P., J.E. Bauer, P.A. Glaser and D.I. Siegl (1995) Radiocarbon evidence for the substrates supporting methane formation within northern Minnesota peat|ands. Geochimica et Cosmochimica Acta 59:3663-3668. Chappellaz, J.A., I.Y. Fung and A.M. Thompson (1993) The atmospheric CH4 increase since the last glacial maximum. Tellus 45B:228-241. Ciais, P., P.P. Tans, A.S. Denning and R.J. Francey (1997) A three-dimensional synthesis study of delta O-18 in atmospheric CO2. 2.Simulations with the TM2 transport model. Journal of Geophysical Research 102:5873-5883. Ciais, P., P.P. Tans, M. Trolier, J.W.C. White and R.J. Francey (1995a) A large northern hemisphere terrestrial CO2 sink indicated by 13C/12C of atmospheric CO2. Science 269:1098-1100. Ciais, P., P.P. Tans, J.W.C. White, M. Trolier, R.J. Francey, J.A. Berry, D.R. Randall, P.J. Sellers, J.G. Collatz and D.S. Schimel (1995b) Partitioning of ocean and land uptake of CO2 as inferred by 813C measurements from the NOAA/CMDL global air sampling network. Journal of Geophysical Research 100:5050-5070. Cicerone, R.J. and R.S. Oremland (1988) Biogeochemical aspects of atmospheric methane. Global Biogeochemical Cycles 2:299-328. Cliff, S.S. and M.H. Thiemens (1997) The ~80/~60 and 170/160 ratios in atmospheric nitrous oxide: a mass-independent anomaly. Science 278:1774-1776.
Role of isotopes and tracers in scaling trace gas fluxes
273
Conny, J.M., R.M. Verkouteran and L.A. Currie (1997) Carbon-13 composition of tropospheric CO in Brazil: A model scenario during the biomass burning season. Journal of Geophysical Research 102:10683-10693. Denning, A.S., I. Fung and D. Randall (1995) Strong simulated meridional gradient of atmospheric CO2 due to seasonal exchange with the terrestrial biosphere. Nature 376:240-242. Francey, R.J. and P.P. Tans (1987) Latitudinal variation in oxygen-18 of atmospheric CO2. Nature 327:495-497. Francey, R.J., P.P. Tans, C.E. Allison, I.G. Enting, J.W.C. White and M. Trolier (1995) Changes in oceanic and terrestrial carbon uptake since 1982. Nature 373:326-330. Friedli, H., H. L6tscher, H. Oeschger, U. Siegenthaler and B. Stauffer (1986) Ice core record of the ~3Cfl2C ratio of atmospheric CO2 in the past two centuries. Nature 324:237-238. Fung, I., C.B. Field, J.A. Berry, M.V. ThomF;on, J.T. Randerson, C.M. Malmstr6m, P.M. Vitousek, G.J. Collatz, P.J. Sellers, D.A. Randall, A.S. Denning, F. Badeck and J. John (1997) Carbon 13 exchanges between the atmosphere and biosphere. Global Biogeochemical Cycles:ll:507-533. Gaudinski, J., S.E. Trumbore and E.A. Davidson, E. Belk and S. Zheng (1998) Partitioning soil respiration using a radiocarbon mass balance approach. Submitted to Biogeochemistry. Gupta, M., S. Tyler and R. Cicerone (1996) Modeling atmospheric delta(CH4)-C13 and the cases of recent changes in atmospheric CH4 amounts. Journal of Geophysical Research:lOl:22923-22932. Keeling, C.D., T.P. Whorf, M. Wahlen and J.v.d. Plicht (1995) Interannual extremes in the rate of rise of atmospheric carbon dioxide since 1980. Nature 375:660-670. Keeling, R., R.P. Najjar, M.L. Bender and P.P. Tans (1993) What atmospheric oxygen measurements can tell us about the global carbon cycle. Global Biogeochemical Cycles 7:37-67. Kim, K.-R. and H. Craig (1993) Nitrogen-15 and oxygen-18 characteristics of nitrous oxide - a global perspective. Science 262:1855. Levin, I., R. B6singer, G. Bonani, R.J. Francey, B. Kromer, K.O. Miinnich, M. Suter, N.B. Trivett and W. W61fli (1992) Radiocarbon in atmospheric carbon dioxide and methane: global distribution and trends. In: E. Taylor, A. Long, and R. Kra (Eds.) Radiocarbon after four decades, Springer-Verlag, New York, pp.503-518. Mak, J E, C.A.M. Bemninkmeijer and J. Tamaresis (1994) Atmospheric laco observations and their use for estimating carbon monoxide removal rates. Journal of Geophysical Research 99:2291522922. Manning, M.R., C.A.M. Brenninkmeijer and W. Allen (1997) Atmospheric carbon monoxide budget of the southern hemisphere: Implications of 13C/12C measurements. Journal of Geophysical Research 102:10673-10682. Manning, M.R., D.C. Lowe, W.H. Melhuish, R.J. Sparks, G. Wallaca, C.A.M. Brenninkmeier and R.C. McGill (1989) The use of radiocarbon measurements in atmospheric studies. Radiocarbon 32:37-58. Manning, M.R. and W.H. Melhuish (1994) Atmospheric A 14C record from Wellington, NZ. In: T.A. Bowden, D.P. Kaiser, R.J. Sepanski, and F.W. Stoss (Eds.) Trends '93.'A compendium of data on Global Change, ORNL/CDIAC-65, Oak Ridge National Laboratory Carbon Dioxide Analysis and Information Center, Oak Ridge National Laboratory, Oak Ridge, pp. 193-201. Nydal, R. and K. L6vseth (1996) Carbon-14 measurements in atmospheric CO2 from northern and southern hemisphere sites, 1962-1993. ORNL/CDIAC-93, NDP-05 7, Carbon Dioxide Information Analysis Centre, Oak Ridge National laboratory, Oak Ridge, Tennessee, 67 pp. P6rez, T., S. Trumbore, S. Tyler, E.A. David~on and M. Keller (1997) Stable isotopes of 15N and iSo in nitrous oxide emitted from tropical forest soils in the rainy season. EOS 78:270. Potter, C.S. (1997) An ecosystem simulation model for methane production and emission from wetlands. Global Biogeochemical Cycles 11:495-506. Prather, M., R. Derwent, D. Ehhalt, P. Fraser, E. Sanhueza and X. Zhou (1995) Other trace gases and atmospheric chemistry. In: J.T. Houghton, L.G. Meira Filho, J. Bruce, Hoesung Lee, B.A. Callander, E. Haites, N. Harris and K. Maskell (Eds.) Climate change 1994. Radiative forcing of climate change. And." An evaluation of the IPCC IS92 emission scenarios. Cambridge University Press, pp. 73-126.
274
S.E. Trumbore
Roeckmann, T., C.A.M. Brenninkmeier, P. Neeb and P.J. Crutzen (1998) Ozonolysis of nonmethane hydrocarbons as a source of the observed mass independent oxygen isotope enrichment in tropospheric CO. Journal of Geophysical Research 103:1463-1470. Quay, P.D., S.L. King and J. Stutsman (1991) Carbon isotopic composition of atmospheric methane: fossil fuel and biomass burning source strengths. Global Biogeochemical Cycles 5:25-47. Rahn, T. and M. Wahlen (1997) Stable isotopes in stratospheric nitrous oxide. Science 278:17761778. Saueressig, G., P. Bergamaschi, J.N. Crowley, H. Fischer and G.W. Harris (1996) D/H kinetic isotope effect it'. the reaction CH4+CI. Geophysical Research Letters 23:3619-3622. Siegenthaler, U. and H Oeschger (1987) Biospheric CO2 emissions during the past 200 years reconstructed by deconvolution of ice core data. Tellus 39B: 140-154. Stevens, C. M. and L. Krout (1972) Method for the determination of the concentration of and the carbon and oxygen isotopic composition of atmospheric carbon monoxide. International Journal of Mass Spectrometry and Ion Physics 8:265-275. Stevens, C.M and A.F. Wagner (1989) The role of fractionation effects in atmospheric chemistry. Zeitschrift fiir Naturforschung A 44:376-384. Stuiver, M. and P.D. Quay (1981) Atmospheric ~4C changes resulting from fossil fuel release and cosmic ray flux variability. Earth and Planetary Science Letters 53:349-362. Tans, P.P., J.A. Berry and R.F. Keeling (1993) Oceanic ~3C/z2Cobservations; a new window on ocean CO2 uptake. Global Biogeochemical Cycles 7:353-368. Tans, P.P., I.Y. Fung, and T. Takahashi (1990) Observational constraints on the global atmospheric CO2 budget. Science 247:1431-1438. Thiemens, M.H., T.L. Jackson and C.A.M. Brenninkmeijer (1995) Observations of mass-independent oxygen isotopic composition in stratospheric CO2 and the importance for ozone chemistry and the Martian atmosphere. Geophysical Research Letters 22:255-257. Yrumbore, S.E. (1995) Use of isotopes and tracers in the study of emission and consumption of trace gases in terrestrial environments. In: P. Matson and R. Harriss (Eds.) Biogenic trace gases." measuing emissions from soil and water, Blackwell, Oxford, pp. 291-326. Tyler, S.C. (1991) The global methane budget. In: J.E. Rogers and W.B. Whitman (Eds.) Microbial production and consumption of Greenhouse gases." methane, nitrogen oxides, and halomethanes. American Society for Microbiology, Washington, DC, pp. 7-38. Volz, A., D.H. Ehhalt and R. Derwent (1993) Seasonal and latitudinal variation of 14CO and the tropospheric concentration of OH radicals. Journal of Geophysical Research 86:5163-5171. Wahlen, M. (1993) The global methane budget. Annual Review of Earth and Planetary Sciences 21:407-426. WMO (1995)Global Ozone Research and Monitoring Project. Report No. 37, World Meteorological Organization, Geneva, pp. 2-20. Yung, Y.L. and C.E. Miller (1997) Isotopic fractionation of stratospheric nitrous oxide. Science 278:1778-1780. Zimov, S.A., Y.V. Voropaev, I.P. Semiletov, S.P. Davidov, T.L. Chapin, M. Chapin, S. Tyler and S. Trumbore (1997) North Siberian lakes: A methane source fueled by Pleistocene carbon. Science 327:806-802.
Chapter 14
INVERSE M O D E L L I N G APPROACHES TO INFER SURFACE TRACE GAS FLUXES F R O M OBSERVED A T M O S P H E R I C MIXING RATIOS
M. Heimann and T. Kaminski
This Page Intentionally Left Blank
Approaches to scaling a trace gas fluxes in ecosystems A.F. Bouwman,editor 9 Elsevier Science B.V. All rights reserved
INVERSE MODELLING APPROACHES TO INFER SURFACE TRACE GAS FLUXES FROM OBSERVED ATMOSPHERIC MIXING RATIOS
M. Heimann 1 and T. Kaminski Max-Planck-Institut for Meteorologie, Bundesstrasse 55, D-20146 Hamburg, Germany
I. Introduction The accurate temporal and spatial quantification of sources and sinks of radiatively or chemically active atmospheric trace gases, constitutes a considerable scientific challenge. Reliable quatificative estimates of sources and sinks are needed for two purposes. Firstly, they provide critical data for the evaluation of process-based prognostic models, which are used to predict the evolution of the atmospheric composition as functions of anthropogenic impacts and environmental changes. Secondly, in the context of international negotiations to curb the emissions of greenhouse gases, an accurate quantification is indispensable to verify reduction targets claimed by individual nations or groups of nations. One approach to quantifying sources and sinks involves the extrapolation of local flux measurements using geographically referenced databases of properties of the surface (e.g. vegetation cover, topography, soil properties) in conjunction with climatic variables (e.g. temperature, precipitation, insolation) and databases of anthropogenic activities (e.g. statistics of land use, energy consumption, population, agricultural practices) as discussed in Bouwman et al. (1999). As sources and sinks of trace gases are also reflected in the spatial distribution and temporal variation of their atmospheric mixing ratio, an alternative approach consists of inverting observed atmospheric mixing ratios into a spatial and temporal distribution of the trace gas sources. In order to do this the atmospheric transport from the source regions to the observation sites has to be described using simulation models of atmospheric transport. During recent decades several global networks of trace gas monitoring stations have been developed, e.g. by the Climate Monitoring and Diagnostics Laboratory of the U.S. National Oceanic and Atmospheric Administration (NOAA/CMDL), the Scripps Institution of Oceanography, La Jolla, California, and the CSIRO of Australia, which routinely monitor the composition of the atmosphere with increasing accuracy and temporal and spatial resolution. These networks are being supplemented by measurements from airplanes, ships and buoys, and data from satellite-based remote sensing instruments. The inference of the distributions of sources and sinks and their temporal variations in a consistent way from all the observations i',~ conjunction with a model of atmospheric transport constitutes an inverse problem of considerable complexity. Ultimately, it requires the design and implementation of a global observing system in which a model of the surface sources is optimized in a consistent way by the different observations, including atmospheric concentrations, isotopic composition and surface features observed from satellites.
Now at Max-Planck-Institut for Biogeochemistry, PF 100164, D-07745 Jena, Germany
278
M. Heimann and T. Kaminski
In many respects the design of observational networks for monitoring of trace gases and the mathematical optimization procedures used in such as system are similar to those currently used in the assimilation of meteorological observations in weather forecasting. The main difference is that the aim of meteorological data assimilation is to find more realistic, dynamically consistent fields of the different meteorological variables to define the initial state of the atmosphere from which a forecast is subsequently computed. Contrary, an observing system for the trace gases would be designed to optimize the surface sources which constitute the boundary conditions. At present such monitoring systems for the trace gases are not in place. This is mainly caused by three reasons: (i) the interest in atmospheric trace gases is rather recent; (ii) techniques of accurate measurements on a routine basis have only recently become feasible; and, (iii) the existing monitoring networks are therefore much less dense than the networks of the meteorological agencies. In spite of these problems there are many pilot studies, in which global and regional scale sources and sinks of atmospheric trace gases have been estimated from a limited number of observational data using a variety of inverse approaches. Most of these attempts have been restricted to long-lived trace gases (i.e. gases with life times longer than one month), trace gases for which atmospheric chemical transformations or removal processes are either absent or relatively well understood. Examples of such gases include carbon dioxide (CO2), methane (CH4), nitrous oxide (N20), halocarbons, and carbon monoxide (CO). The main reason for this restriction is that the mathematical inverse problem of these gases is either linear or may be linearized. Short-lived reactive species such as NO• have not received much attention so far, because their atmospheric chemistry is nonlinear and depends on many other coupled species, which makes the problem difficult to solve. However, highly nonlinear cases have been addressed in related fields, such as inversions of oceanic biogeochemical processes as discussed in e.g. the proceedings of the "Workshop on Inverse Methods in Global Biogeochemical Cycles, held in Heraklion, Greece, March 18-20, 1998; to be published as AGU monograph in 1999 (http:// www.mpimet.mpg.de/gbc/heraklion). Some important difficulties of the inversion problem include: Current atmospheric transport models are not perfect. - The observational network is sparse, i.e there are only a small number of monitoring stations. Furthermore, at some stations the sampling frequency is low, and there are often temporal gaps in the observations. - Technically, the "inversion" of the atmospheric transport model is not trivial and requires much larger computing resources than running the model in the forward mode. Individual measurements are often not representative of the appropriate temporal and spatial scale of the transport model. -Individual observations are of limited accuracy and precision and, furthermore, observations from different networks are often not easily compared because of differences in measurement techniques and uses of different standards. -
-
The present review focuses on global approaches. Section 2 addresses the problem of modelling atmospheric transport. In section 3 we discuss the mathematical and technical difficulties of the inversion problem. Secuon 4 describes the results of two case studies to demonstrate the current state-of-the-art. Finally, a brief overview is given of recently proposed strategies to address the regional (continental) problem with considerably higher temporal and spatial resolution.
Inverse modelling approaches to infer surface trace gas fluxes from observed atmospheric mixing ratios
279
2. Atmospheric transport 2.1. The continuity equation The link between sources and sinks of a trace gas and its mixing ratio for a specific location and time period is provided by atmospheric transport. In mathematical terms we have a twoor three-dimensional, time dependent field Q(x, t, 2') describing the sources and sinks of the trace gas considered. Generally the source may also depend on the mixing ratio of the gas itself and, in chemically more complex cases, on the mixing ratios of other coupled constituents. The atmospheric transport translates the source field Q(x, t, z) into a threedimensional, time dependent field of mixing ratios Z (x, t). The sources and the mixing ratios are related by the continuity equation: O Ot
PZ + V . v p z = Q(x,t, Z)
(1)
where 9 is the air density and v the wind vector, both being also three dimensional timevarying quantities. If trace gas emission and uptake are linearly related to the mixing ratio of the trace gas then the continuity equation is also linear. In the case of the long lived trace gases this is usually correct, or a linear expansion around a mean background state is assumed. For example, for methane this is a good approximation (Hein et al., 1996). In this case the continuity equation can be written as: 6
--~ P Z + V . v p z - k ( x , t ) Z = Q(x,t)
(2)
where k(x, t) describes a first order reaction expressed by the spatially and temporally varying reaction constant. The solution of equation (2) for the atmospheric mixing ratio 2'(x, t) given trace gas sources, Q(x, t), and reaction coefficients, k(x, t), for a finite time interval requires the specification of an initial mixing ratio field Z(x, 0) and boundary conditions at the borders of the spatial computing domain. For examplo, in the case of a global model for the troposphere suitable boundary conditions, e.g. prescribed trace gas sources or mixing ratios, have to be specified at the tropopause and the Earth's surface. In practice the continuity equation has to be solved numerically in discretized form (see section 2.2 below). Computing the spatio-temporal distribution of the mixing ratio Z(x, t) from prescribed sources and sinks constitutes a forward model run. In the linear case this may be written formally as a matrix equation: =
(3)
where tilde symbol "-~" denotes quantities on the full temporal and spatially discrete resolution of the transport model, the column vector of the mixing ratios m[ includes all model grid points in space and time of the simulation, and likewise the column vector of source values ql includes all grid points of the source in space and time. It also includes any initial and boundary condition terms. The matrix T represents the transport model code. In most cases a r e d u c e d p r o b l e m is of ;nterest: We are not interested it, the mixing ratio field on the full model resolution at every time step of the model, but in time averaged mixing ratios (e.g. weekly, monthly or annual means) at a finite number of observation locations.
280
m. Heimann and T, Kaminski
Also the sources of interest are often time averaged quantities, possibly also averaged over larger spatial domains (see Sofiev, 1999). If the averaging operations both in the space of observations and of sources are linear, we have a linear "model equation" similar to (3):
ml-T.q I
(4)
Here, the elements of ml represent the observations (in time and space), and the elements of ql are the "source components" (in time and space) which include the (possibly spatially and temporally averaged) initial and boundary conditions. The elements of the matrix T may be interpreted as the sensitivities of the observables with respect to the source components. For example, the element T;j denotes the derivative of the mixing ratio at the time-space location i with respect to the source component j. Formally, the matrix T is closely related to the Green's function of the reduced problem.
2.2. Models of atmospheric transport Numerical models of atmospheric transport solve the continuity equation for a passive atmospheric trace constituent on a discrete model grid. Typical spatial resolutions in presentday global models are on the order of a fe',~ degrees latitude and longitude at~d 10 to 30 layers in the vertical dimension. Likewise, the temporal dimension is also discretized with time steps ranging from a few tens of minutes to a few hours. Atmospheric transport can be computed "on-line" as part of an atmospheric general circulation model which provides the meteorology (windspeeds, air properties, subgrid scale transport through clouds, convection or diffusion) (e.g. Erickson et al., 1996). Alternatively, the transport may be calculated "off-line" by reading the meteorological fields from stored output of a general circulation model simulation or from analyzed fields of a weather forecast model. In the latter case subgrid scale transport (see below) has to be recomputed in the transport model, which constitutes a difficult task (see e.g. Heimann, 1995). By using archived analyses from weather forecast models, the off-line approach has the advantage that the "real" meteorology is used in the simulations, i.e. the model describes the actual atmospheric circulation fields prevailing during the particular time period when the measurements were obtained. In contrast, the on-line general circulation model approach yields simulation results that can only be compared to the observations in a statistical sense, e.g. as monthly or annual means. The models typically split the transport into two main components: "advection", i.e. the transport resolved on the model grid, and subgrid scale transport including the effects of all processes on temporal and spatial scales not resolved on the model grid (e.g. transport through cumulus clouds, thermal convection, diffusion or boundary layer mixing). Formally this can be presented as: _
m
6t P Z + V . v P Z + C O N V E C ( z )
m
= Q(x,t,z)
(5)
where the overbars indicate quantities averaged over the model grid. The second term on the left represents the resolved "advection", while C O N V E C denotes the subgrid scale transport processes. 2.2. I. A d v e c t i o n
Many numerical schemes have been developed in the past for the numerical solution of the part of equation (5) which denotes the grid-resolved transport. Important properties of
Inverse modelling approaches to infer surface trace gas fluxes from observed atmospheric mixing ratios
281
numerical advection schemes are accuracy, linearity, mass conservation, small numerical diffusion, computational efficiency, and positive definiteness. A review of some of the more commonly used Eulerian numerical schemes has been given by Rood et al. (1987). Examples of more advanced techniques are discussed by Spee et al. (1997) and Spee (1998). Spectral techniques, which used to be popular in the atmospheric general circulation modelling community, expand the tracer mixing ratio field into spherical harmonics. This approach is not very well suited for tracer transport because the spectral transformation may generate negative concentrations. There are also lagrangian schemes in which the atmosphere is represented by a series of discrete air parcels, each containing an amount of tracer in accordance with the local tracer mixing ratio. These parcels are moved around the atmosphere by the specified wind velocity and keep their identity over the entire length of the model integration time period (e.g. Penner et al., 1991). Lagrangian schemes have much less numerical diffusion as compared to Eulerian techniques. However, it is difficult to adequately represent subgrid scale mixing processes in such a scheme. Furthermore, because the parcel distribution tends to become non-uniform with time, the schemes need a large number of parcels in order to represent the tracer fields with sufficient spatial resolution. This generally makes the schemes computationally expensive, except in situations where the transport of several tracers is to be computed simultaneously, because in this case the air parcel movement has to be computed only once for all tracers. In semi-lagrangian schemes the tracer is transported during each time step in a lagrangian fashion, i.e. attached to discrete air parcels. After each time step the three-dimensional tracer mixing ratio fields are reconstructed by interpolation from the parcels to the regular model grid. This interpolation tends to have problems with mass conservation (Rasch and Williamson, 1990). Despite the different numerical techniques employed in current atmospheric transport models, the advection is not considered to be the most critical component in need of improvement. 2.2.2. Subgrid scale transport processes
In the course of the discretization of the basic continuity equation on the model grid, the effects of all transport processes on smaller spatial and temporal scales must be described ("parameterized") as functions of the values of the meteorological variables on the resolved scale. Some of these processes include vertical transport through cumulus clouds ("wet convection"), thermally driven dry convection, turbulent diffusion and vertical mixing in the surface and boundary layers. A comparison of several vertical subgrid scale transport parameterizations can be found in Mahowald et al. (1995). Parameterizations for these processes also exist in atmospheric general circulation models and in weather forecast models. In principle these could be transferred to the transport models. However, except for water vapor, the transported quantities (energy, momentum) have different properties as compared to trace gases (source distributions, lifetimes) making such a transfer difficult to implement. A very critical subgrid scale transport process is boundary layer mixing over the continents. Most atmospheric observations are being made close to the surface in the planetary boundary layer, and most of the long-lived trace gas sources are at the Earth's surface. The height of the layer, into which the trace gas is emitted, and the exchange processes with the overlaying free troposphere critically determine the simulated mixing ratio in this layer. In the case of a trace gas with strong diurnal and seasonal sources, such as CO2, systematic
282
m. Heimann and T. Kaminski
temporal changes of the transport (e.g. height of the boundary layer during day and night and also during the different seasons of the year) are responsible for "rectifier effects", i.e. spatial structures in the mean annual mixing ratio field even if the trace gas source on annual average were balanced at every grid point to zero (Denning, 1995; Denning et al., 1996). These structures arise because of the temporal covariance between source and atmospheric transport (Keeling et al., 1989; Denning et al., 1995; Denning et al., 1998; Law et al., 1996; Rayner and Law, 1996). Hence, observed spatial structures in the mean annual mixing ratio fields reflect a combination of true sources and sinks and of the "rectifier effects". An accurate representation of the "rectifier effect" in the atmospheric transport model is, therefore, an indispensable necessity. Unfortunately, at present there is no way to independently verify the simulated rectifier effect, which, at least for CO2, constitutes a serious limitation in current inversion studies.
2.3. Validation of atmospheric transport models An assessment of the realism of present atmospheric transport models is beyond the scope of this review. For a detailed discussion on validation of atmospheric models we refer to Sofiev (1999). A critical quantity of global atmospheric transport models is their description of the meridional interhemispheric transport, which can be validated by means of simulations with trace gases of known surface sources, such as the radioactive Krypton-85 (Jacob et al., 1987; Heimann and Keeling, 1989), CFC-11 (Prather et al., 1987) or sulfurhexaflueride (SF6) (Levin and Hessheimer, 1996). A consistent intercomparison of global atmospheric transport models with SF6 has recently been undertaken within the phase 2 of the TRANSCOM project (Denning et al., 1998). As discussed above, at present there is no method available to validate the simulated "rectifier effect". The way the different models simulate this effect has been explored in phase 1 of the TRANSCOM project (Rayner and Law, 1995; Law et al., 1996).
3. Methodological aspects of the inversion problem 3.1. The mathematical problem The inverse problem consists of inverting equation (4) to find a solution for the source components q[. The structure of equation (4) shows that this involves two tasks: Firstly, the matrix T has to be computed; and secondl, ', the linear equation system has tc be solved for the unknown source components. Depending on the number of observables and the number of source components, the matrix T may be very large, hence its computation may be very expensive. But also the solution of equation (4) for the source components is not trivial. Typically there are only a limited number of observations in space and time. On the global scale, currently most of the gases under consideration are monitored at less than 10 stations continuously or quasi-continuously and at less than a few hundred stations with a sampling frequency of less than 1-2 observations every week. For some of the gases (e.g. CO) remote sensing observations from space craft are available, but these typically are of limited accuracy and often represent only the vertical integral. In addition, most source and sink processes are strongly heterogeneous in space and time. To represent this heterogeneity adequately, one is tempted to choose a high spatial
Inverse modelling approaches to infer surface trace gas fluxes from observed atmospheric mixing ratios
283
resolution for the source components, herce the matrix T becomes rectangular with a much larger number of columns than of rows. In this case the source inference problem is highly underdetermined. But even if one restricts the number of source components to a smaller number than the observations, the equation system might still be "ill-conditioned" because of the diffusive nature of the atmospheric transport. Hence, additional restrictions (simplifications, assumptions, "a priori" information from other data sources) are needed in order to make the problem mathematically well-conditioned. Some of these are discussed in the subsections below.
3.2. Reducing the spatial resolution A reduction of the spatial resolution is obtained by dividing the globe up into only a small number of regions (typically 10-100). The surface sources for each of these regions are then prescribed in their spatial and temporal pattern, whereas their overall magnitude is left as an unknown scaling parameter to be determined in the inversion procedure. For high-resolution, computationally expensive atmospheric transport models this is the only way to make the determination of the matrix T manageable. The matrix T can be determined by brute force, whereby the atmospheric transport model code is run for each of the source regions separately. T is then obtained by recording the contributions from the different source components at the observation sites. In this "synthesis inversion" the source regions may be specified as a simple geometric breakdown of the globe, e.g. in latitudinal bands. This approach has been applied e.g. by Brown (1993; 1995) to deduce the sources of CH4 for latitudinal bands. A slightly different variant of the "synthesis inversion" consists of decomposing the source field into several, possibly spatially overlapping components that represent different source processes. This has been applied for of CH4 by e.g. Hein et al. (1996). Here the different source and sink processes of CH4 are specified by their global spatial and temporal patterns (e.g. emissions from peats and bogs, from coal mining, oil or gas production, waste disposal, rice paddies, cattle). For CO2 this approach has been used by Enting et al. (1995), Bousquet (1997) and Rayner et al. (1998). The "synthesis inversion" approach typically yields a relatively stable iIlversion (possibly even an overdetermined equation system), since in most cases the number of source components chosen is less than the number of observations. However, the use of predefined "rigid" spatio-temporal source patterns strongly influences the resulting solution.
3.3. Simplifications in the temporal domain The atmosphere has a limited "memory". An impulse input of a conservative tracer released at a specific location and time eventually becomes homogeneously mixed. The longest mixing time defines a time horizon, beyond which any source or sink contributes only to the global background mixing ratio of the tracer. Within the troposphere a pulse input into one hemisphere decays by a factor of 1/e within approximately one year (Weiss et al., 1983). Mixing times into the stratosphere, however, are much longer. Hence, for a tropospheric tracer, a conservative value for the time horizon is 3 to 4 years (Heimann and Keeling, 1989). For example, for the simulation of a transient tropospheric tracer, such as the balocarbons during the 1980's, the history of the F11 sources prior to 1977 is irrelevant, and only their cumulative global integral determines the atmospheric background F11 mixing ratio (Bloomfield et al., 1994).
284
m. Heimann and T. Kaminski
This implies that the elements in the sensitivity matrix T which refer to sources more than 4 years prior to the time of the observation are given simply by the increment in the globally averaged mixing ratio per unit input of the tracer. Hence the atmospheric transport model must be run for only 4 years for each of the source components in order to determine the sensitivity matrix T. A further simplification arises if one assumes that the large-scale atmospheric transport does not change significantly from year to year. This appears to be the case for SF6, which shows that the mixing ratio differences between the hemispheres remained relatively constant over the last decade, when emissions did not change significantly (Maiss and Levin, 1994; Levin and Hessheimer, 1996). In this case one may use the same wind fields year after year. As an additional check one can repeat the simulation with the meteorology from another year (e.g. Knorr and Heimann, 1995). Finally, an additional, considerable simplification can be achieved by addressing only the "quasi-stationary" problem by assuming that all sources and sinks may include a seasonality, but are invariant from year to year. If the transport is also assumed to be identical year after year, the atmospheric mixing ratio at any location, x, as a function of time, t, can be expressed as an offset, Z0(x), a globally uniform linear trend a, and a seasonal cycle S(x, ~-): Z(x,t) = Z0(X) + a,t + S(x, z')
(6)
where r denotes the time since the beginning of the year. Many global inversion studies of CO2 (e.g. Enting et al., 1995) and CH4 (e.g. Hein et al., 1996) employed this approximation. The extension to the interannually varying case has been addressed by Bloomfield et al. (1994) for CFC-11 and Rayner et al. (1998) for COz.
3.4. Bayesian approaches Bayesian approaches to the inverse problem provide a means to include a priori information on the unknown source components in the inversion procedure (see e.g. Tarantola, 1987). They are based on a formulation of the problem in terms of probability distributions in the joint space of sources and concentrations. In practice these probability distributions are assumed to be Gaussian. An a p o s t e r i o r i s o u r c e estimate is derived, which is optimal in the sense that it is as close as possible to the prescribed a priori sources, while the resulting simulated concentrations are as close as possible to the observations. Thereby "close" is defined relative to specified uncertainties in both the observations and the a priori source estimates. In an otherwise underdetermined inverse problem, the Bayesian approach yields a unique solution from all source configurations that are consistent with the observations. In an illconditioned inverse problem the Bayesian approach limits the amplification of errors in the observations when inferring source combinations that are poorly constrained by the observations (see Enting, 1993). The required a priori information, i.e. the a priori sources, may be provided by interpolating direct flux measurements, or they may be obtained from prognostic source models. In both cases quantifying the uncertainties of the a priori estimates is crucial and should reflect the understanding of the underlying source processes. Bayesian inversions have been carried out by Enting et al. (1995), Bousquet (1997), Rayner et al. (1998) and Kaminski (1998) for CO2 and for CH4 by Hein and Heimann (1994) and Hein et al. (1996).
Inverse modelling approaches to infer surface trace gas fluxes from observed atmospheric mixing ratios
285
3.5. Technical approaches In the case of scant networks, adjoint models provide an efficient tool to compute the matrix T in equation (4). Instead of computing the partial derivatives T~j by brute force with forward transport model runs for each source component j, the adjoint model begins at the location and time of the observations (i). Starting with an infinitesimal mixing ratio deviation at the observation point i, the adjoint model works its way backward in time. Thereby it determines the necessary changes in the source components at earlier time points that would induce the infinitesimal mixing ratio deviation at i. In this way the adjoint model essentially propagates the sensitivity of the modelled concentrations from the observational sites i backwards to the sources (Errico, 1997; Corliss and Rall 1996). Using forward model runs to compute T;j the required computational resources are proportional to the number of source components and relatively independent of the number of observations, while for the adjoint model the required computational resources are proportional to the number of observations and relatively independent of the number of source components. The major difficulty of the adjoint approach is the time consuming construction of the adjoint model. This can be overcome, however, by using tools for automatic generation of adjoint code, which are being developed, e.g. Odyss6e (Rostaing et al., 1993) or the Tangent and Adjoint Model Compiler (TAMC) (Giering, 1996). Using the TAMC, Kaminski et al. (1996) developed the adjoint of the atmospheric transport model TM2 (Heimann, 1995), which allowed an efficient computation of the matrix T for an observational network of approximately 30 stations with monthly observations and monthly sources on the 8 by 10 degree horizontal model grid, i.e. a matrix for observations at 30x 12 = 360 time-space locations and 12x24x36 = 10368 source components (a matrix with 3.7 million elements). 3.6. Representativeness of individual measurements In most global studies performed so far, the modelled monthly mean mixing ratio is compared to estimates of the monthly mean mixing ratio at the observing sites. The mixing ratio computed by a model is "representative" for the spatial and temporal resolution of the model grid. On the other hand, individual measurements at a monitoring site in general reflect also local transport and source processes on finer scales than resolved by the model (Sofiev, 1999). The comparison between observations and model results, therefore, requires considerable care in order to avoid potentially serious biases. In the observations one typically tries to eliminate the influences of local contamination by following a data selection protocol designed to sample so-called "background air" under "baseline conditions". This is achieved for example by specifying the time of day and other requirements (minimum wind speed, particular wind directions, other meteorological conditions) under which air sampling is performed. Where possible also the information from other, concurrently measured tracers such as e.g. Radon-226 and Radon-222 can be used (Polian et al., 1986). An additional elaborate data screening is used after the sampling in order to eliminate "outliers" believed to represent local ("polluted") samples before computing monthly means. In many cases these procedures may yield observations that are representative to the simulations, but it is difficult to estimate to what extent these approaches in fact eliminate the local unwanted source contributions. The data selection procedures by themselves may also induce biases, because potentially air is sampled that reflects a different model grid box than the one containing the observing station.
286
m. Heimann and T. Kaminski
!101t,
1...... 16
i., "
.I,.- - i .... :-
e~
I .......................... : I : ilit
liiO
' Iliil
i
@ i
1
......
,.I./
-
l
Figure 1. Global station network of the Climate Monitoring and Diagnostics Laboratory of the National Oceanic and Atmospheric Administration(from Conwayet al., 1994).
In the models a sampling strategy should be used that mimics the procedures employed at the observing stations. For example, it is not meaningful to compute monthly averages in the model by continuous sampling if the model exhibits a diurnal cycle which includes night-time inversions, while the observations are only taken in the afternoon when vertical mixing is strongest. Unfortunately, the implementation of adequate sampling strategies in the models is rather cumbersome, as it will be different for each station. A thorough demonstration of the potential biases incurred by different sampling procedures in the model and the observations has been given by Ramonet and Monfray (1996). 3.7. Inhomogeneity of sampling networks In many cases, the spatial inhomogeneity of the current observational networks is obvious. For example, the CO2 monitoring network maintained by the U.S. National Oceanographic and Atmospheric Administration's Climate Monitoring and Diagnostic Laboratory (Conway et al., 1994) in which the tropics and in particular the tropical land masses are undersampled (Figure 1). Also, sampling of oceanic regions has been favoured by choice of observational sites and by the definition of baseline conditions (see the previous section). On the other hand, in most inversion studies one chooses a relatively small number of unknown source components (see section 3.2). In combination with inhomogeneous sampling, a low spatial resolution of source estimates is likely to yield a biased estimate of the inversion as demonstrated by Snieder (1993). For inverse problems in seismic tomography, Snieder (1993) and Trampert and Snieder (1996) demonstrate how to reduce the bias at the cost of increasing the uncertainty. For inversion of the global atmospheric transport and the current global networks, the magnitude of this bias is yet to be quantified.
3.8. Calibration problems between different measurement networks It is difficult to compare observations from different measurement agencies, because of differences in measurement techniques and standards used. This is a serious problem, because,
Inverse modelling approaches to infer surface trace gas fluxes from observed atmospheric mixing ratios
287
in the case of long-lived species, small mixing ratio differences between different stations imply relatively large sources. Hence, differences between networks may induce substantial biases in the inferred source distributions. Discrepancies between the networks could be reduced by systematic intercomparisons of standards, measurement techniques, sampling procedures and sample handling. Several intercomparison efforts are currently underway within the different trace gas species measurement communities. Unfortunately, these tedious efforts are scientifically not very rewarding and are therefore often lacking sufficient funding. It must be stressed, however, that such efforts are very important and should receive highest priority. As an intermediate step it is also possible to short-cut detailed intercomparison procedures by merging the existing datasets in a way that takes possible differences into account. In the case of CO2 such an effort has been undertaken by NOAA-CMDL in the creation of the GLOBALVIEW dataset (Masarie and Tans, 1995; NOAA/CMDL, 1996). Alternatively, one can of course also use only the observations from a single network. Or, it is also possible to include the calibration offset between different networks as unknown parameters to be determined in the inversion.
3.9. Uncertainties
In principle, an inversion is subject to two sources of errors: (i) observations are of finite precision; and, (ii) models are imperfect. A third type of error, the representation error, arises from the fact that the observations typically are not representative for the spatial and temporal scale of the model predictions. For example, the observations are point measurements in time and space, while the model predicts the mixing ratio averaged over an entire model grid box and model time step. Depending on the point of view the representation error may be subsumed either under the model or the observational error. In general, these errors are not known (otherwise one could apply explicit corrections), hence they induce uncertainties in the source fields that are determined by the inversion. The formal treatment of these uncertainties is in terms of probability distributions. Usually, for the sake of computational convenience, Gaussian distributions are assumed: The observations are characterized by a set of mean values and a covariance matrix, quantifying the uncertainties in these values and their interdependence. Other Gaussian distributions characterize the uncertainties due to the model error. For alinearization of the transport model, one can derive a Gaussian distribution of the source field that is consistent with the sum of the uncertainties induced by the above-mentioned errors (see Sofiev, 1999). Due to this Gaussian assumption, the actual computations are then manipulations of means and covariance matrices. In Bayesian inversions, the a priori information is also quantified in terms of a mean and a covariance matrix. The inversion then derives a posterior mean field and covariance matrix. The information in the atmospheric observations then is reflected in a change of this mean and a reduction of the uncertainties. In practice, errors in the transport model are very hard to quantify. Potential approaches are either model intercomparisons or checking against observed concentrations for tracers or isotopes with well known source distributions (see Trumbore, 1999). Because of these difficulties, transport model errors are usually neglected in inversion studies. Similar problems complicate the explicit inclusion of the representation error. Selection of observations according to their representativeness (see section 3.6) may be an approach to minimize this error.
288
M. Heimann and 7". Kaminski
4. Examples Enting et al. (1995) presented a Bayesian synthesis inversion for CO2. Using observations of atmospheric CO2 from the global networks of the NOAA-CMDL (Conway et al., 1994) together with 13C/12Cisotopic ratio measurements from CSIRO (Francey et al., 1995), they solved the quasi-stationary problem (see section 3.3) for the magnitudes of about 20 unknown source components. Since the study by Enting et al. (1995) a number of research groups have
_
~
~
. . - - _ ~ -
_ . e = ~ - . - . . . - -
-
--~,,..-,-,-,-,-,-,-,-m~
~
_
o.o ~
__.,"o.o
"1~
"~? o,,,
,,,
~ o.a
o.ot) .T
-0.3
'~ ,~ ..,~.,~.~.~...,.
o.a
~'~"
~
rP~
-0.3
o.e
'~,-,--~ I?o.o"~
q o.a
9
- 1.0
-0.~
.,v~
-0.5
~
a
o.o
"-
~ - - a I ,,.,,,-0._~ .-
-0.7 0.5
0.6
-
0.3
'k
..~.
-0.7
-0.8
~
~
-0.3
J
-0.1
b Figure 2. a) A priori long-term annual mean source distribution as specified in the inversion study of Rayner et al. (1998); b) a posteriori sources inferred by the inversion procedure. Numbers represent source fluxes of CO2 in Gt C yr 1. Fluxes into the atmosphere are indicated by positive numbers.
Inverse modelling approaches to infer surface trace gas fluxes from observed atmospheric mixing ratios
289
Figure 3. a) A priori; b) a posteriori; and, c) difference of the annual mean source flux fields of CO2 as resolved on the full model grid (8 ~ latitude by 10~ longitude) in the study of Kaminski (1998), using an adjoint model of atmospheric transport. Only the atmosphere-ocean and atmosphere-terrestrial biosphere CO2 fluxes are presented; the anthropogenic CO2 flux from the combustion of fossil fuels has been excluded. Fluxes into the atmosphere are indicated by positive numbers.
290
m. Heimann and T. Kaminski
been refining the inversion approach in various directions. As examples, we briefly discuss the long-term average annual mean source fluxes of CO2 inferred in the studies of Rayner et al. (1998) and Kaminski (1998). By resolving the interannual variability in the sources, Rayner et al. (1998) removed the quasi-stationary assumption. For the period from 1980 to 1995, in a Bayesian synthesis inversion they inferred the monthly magnitudes of about 30 source patterns from global observations of CO2, supplemented by one station's time series of the isotopic CO2 composition and the station's linear trend in the oxygen to nitrogen ratio. The a priori and a posteriori estimates for the long term annual mean atmosphere-ocean and atmosphereterrestrial biosphere fluxes of CO2 are presented in Figure 2. Maintaining the quasi-stationary assumption, in contrast, Kaminski (1998) removed the simplifications in the spatial domain by means of an adjoint approach (see section 3.5). In principle this approach allows the computation of adjustments to the a priori source field resolved on the full horizontal grid of the transport model (8 ~ latitude by 10~ longitude). In this study observations of the atmospheric CO2 mixing ratio from 25 stations of the NOAACMDL (Conway et al., 1994) network for the period from 1981 to 1986 were used (i.e. a subset of the stations shown in Figure 1). The highly underdetermined nature of this problem necessitated a Bayesian approach (see section 3.4). The a priori source fields were obtained from energy use statistics for the fossil fuel CO2 source, statistics of land use change and spatially explicit carbon cycle simulation models for the ocean and the terrestrial biosphere. Figure 3 shows the a priori and a posteriori estimates for the annual mean source fields and their difference. Figure 3b shows an overall pattern of the non-fossil fuel CO2 source fields which is similar to the a posteriori source field of Rayner et al., (1998) (Figure 2), with a strong sink for CO2 in the northern mid-latitudes and a smaller sink in the southern hemisphere oceans. The tropical oceans and in particular the equatorial Pacific reflect regions with CO2-outgassing. Despite the considerable spatial and (not shown) temporal resolution of the source fluxes of CO2 as inferred in the two studies, it must be realized that the a posteriori solution is determined considerably by the a priori information. The extent to which the atmospheric measurements and the transport model provide additional information can be investigated by analysis of the Singular Vector Decomposition (SVD) (Menke, 1989) of the model matrix T in equation (4) (Kaminski, 1998). The reduction of the a priori uncertainty induced by the inversion provides also a measure to assess the relative importance of the a priori information. Figure 4 presents the a priori and a posteriori uncertainty and the relative reduction for the annual mean CO2 fluxes from the study of Kaminski (1998). In most regions the uncertainty is only slightly reduced (at most 20%). Clearly the uncertainty reductions are largest close to the monitoring stations. If averaged over larger regions, e.g. zonally or over continents the percentage uncertainty reduction becomes larger. In particular the large scale features of the resulting solution as described above are found to be significant (Kaminski, 1998). However, this assumes uncorrelated a priori uncertainties between the different gridboxes. This assumption is difficult to assess in the present example, since the a priori flux fields were partly derived from oceanic and terrestrial carbon cycle models, for which estimates of their errors and associated covariance structure are not readily available. The relatively small uncertainty reduction provided by the atmospheric observations is disappointing. For the case of CO2 it demonstrates clearly, that the present atmospheric background monitoring networks do not allow a regional determination of the sources and sinks of CO2 without significant a priori information on their spatio-temporal distribution and magnitude. On the other hand, maps such as Figure 4 provide a means to assess the merits of
Inverse modelling approaches to infer surface trace gas fluxes from observed atmospheric mixing ratios
291
Figure 4. a) A priori and b) a posteriori uncertainty of the annual mean source flux field of C O 2 corresponding to Figure 3a and 3b; c) percentage reduction between the a priori and a posteriori uncertainty.
292
M. Heimann and T. Kaminski
individual stations of the monitoring networks. On the basis of this tool, strategies for the optimization of the networks may be devised.
5. New approaches Recently, a series of new projects have been initiated or proposed to determine the regional sources and sinks with a much higher spatial and temporal resolution than is possible with the existing global approaches as discussed in the previous sections. These initiatives include "Carbon America" (Tans et al., 1996), the Large-scale Biosphere Atmosphere (LBA) experiment over Amazonia (Anonymous, 1997), COBRA (Wofsy et al., 1997) and "Eurosiberian Carbonflux" (Heimann et al., 1997). These new proposals call for a considerable extension of the monitoring networks. As discussed above, current observational networks are heavily biased toward oceanic areas. A better and more detailed regional determination of continental sources requires observations closer to these sources. However, for continental monitoring stations it may be difficult to obtain representative measurements, because of the complex meteorology and the typically strong and heterogeneous terrestrial sources and sinks. The new proposals also call for an extension of the measurement techniques used. In particular they include measurements in the vertical dimension by aircraft and, if possible, observations from remote sensing platforms. In addition, a multi-tracer approach may be chosen, in which several atmospheric species are measured simultaneously. For example, in the case of the Eurosiberian Carbonflux project the following tracers are to be measured at several sites on bi-weekly vertical profile flights over a time period of at least three years: CO2 and its carbon and oxygen isotopes, CH4 and its carbon isotopes, CO, Radon-222 (a radioactive noble gas with a half-life of 3.8 days), SF6 and others. Since all tracers are subject to the same atmospheric transport, the problem of representativeness of a single measurement with respect to the modelled grid-averaged mixing ratio may be addressed in an effective way. The modelling of the atmospheric transport over the continental sources and sinks also poses a substantial challenge. For properly describing the heterogeneity of the terrain and the induced complicated meteorology, meso-scale models are required. Such models must have a horizontal grid resolution of 10-50 km covering an total area of up to 25x 106 km 2, and need to resolve the diurnal cycle of boundary layer mixing and convection in considerable detail. In order to run these models the large-scale meteorology has to be specified from weather forecast analyses. An additional problem is the specification of the boundary conditions for the tracer(s) under consideration, which will have to be deduced from the output of a global, low-resolution simulation. An example of such a modelling system has been described for simulations of CO2 in the arctic region (Engardt, 1997; Engardt and Holmen, 1997) based on the MATCH model (Robertson et al., 1996). How will these models perform in an inverse approach? Even if the difficulties to realistically describe the transport over the continental regions can be overcome, it is not clear if this approach yields reliable regional source flux estimates. This is because the large and mostly unknown heterogeneity of the source flux distribution might require a sampling density in space and time that is not feasible. Clearly, without progress in these new approaches the second goal of inversion studies mentioned in the introduction, i.e. the quantification of regional flux estimates for the verification of national greenhouse gas reduction targets will remain elusive. Nevertheless, it is hoped, that the new projects will provide some insight into the regional source estimation problem.
Inverse modelling approaches to infer surface trace gas fluxes from observed atmospheric mixing ratios
293
References Anonymous (1997) The large-scale biosphere-atmosphere experiment in Amazonia- LBA. Http://www.tor.ec.gc.ca/GEWEX/GHP/LBA/_96.htmi. Bloomfield, P., M. Heimann, M. Prather and R. Prinn (1994) Inferred lifetimes. In: J.A. Kaye, S.A. Penkett and F.M. Ormond (Eds.) Concentrations, lifetimes and trends of CFCs, halons and related species. NASA Reference Publication 1339, NASA, Washington. Bousquet, P. (1997) Optimisation des flux nets de CO2: assimilation des mesures atmosph6riques en CO2 et en 8~3C dans un mod61e de transport tridimensionnel. Ph.D. thesis, Universit4 Paris VI. Bouwman, A.F., R.G. Derwent and F.J. Dentener (1999) Towards reliable bottom-up estimates of temporal and spatial patterns of emissions of trace gases and aerosols from land-use related and natural sources. In: A.F. Bouwman (Ed.) Approaches to scaling of trace gas fluxes in ecosystems. Developments in Atmospheric Science 24. Elsevier, Amsterdam, pp. 1-26. Brown, M. (1993) Deduction of emissions of source gases using an objective inversion algorithm and a chemical transport model. Journal of Geophysical Research 98:12639-12660. Brown, M. (1995) The singular value decomposition method applied to the deduction of the emissions and the isotopic composition of atmospheric methane. Journal of Geophysical Research 100:425446. Conway, T.J., P.P. Tans, L.S. Waterman and K.W. Thoning (1994) Evidence for interannual variability of the carbon cycle from the Na,~onal Oceanic and Atmospheric Administration Climate Monitoring and Diagnostics Laboratory global air sampling network. Journal of Geophysical Research 99:22831-22855. Corliss, G.F. and L.B. Rail (1996) An introduction to automatic differentiation. In M. Berz, C. Bischof, G. Corliss and A. Griewank (Eds.) Computational Differentiation: Techniques, and Tools, SIAM, Philadelphia, Penn., pp 1-18. Denning, A.S. (1995) Investigations of the transport, sources, and sinks of atmospheric CO2 using a general circulation model. Ph.D. Thesis, Colorado State University, Fort Collins, Colorado. Denning, A.S., I.Y. Fung and D. Randall (1995) Latitudinal gradient of atmospheric CO2 due to seasonal exchange with land biota. Nature 376:240-243. Denning, A.S., D.A. Randall, G.J. Collatz, P.J. and P.J. Sellers (1996) Simulations of terrestrial carbon metabolism and atmospheric CO2 in a general circulation model. 2. Simulated CO2 concentrations Tellus 48B:543-567. Denning, A.S., M. Holzer, K.R. Gurney, M. Heimann, R.M. Law, P.J. Rayner, I.Y. Fung, S.-M. Fan, S. Taguchi, P. Friedlingstein and Y. Balkansky (1998) Three-Dimensional transport and concentration of SF6: Preliminary results of a model intercomparison study (TransCom). Tellus, submitted. Engardt, M. (1997) Climate change and carbon dioxide fluxes in the high latitude Northern Hemisphere. Dep. of meteorology, Stockholm University, Ph.D. Thesis, 104pp. Engardt, M. and K. Holmen (1997) Towards decuding regional sources and sinks from atmospheric CO2 measurements at Spitsbergen. Physics and Chemistry of the Earth 21:523-527. Enting, I.G. (1993) Inverse problems in atmospheric constituent studies. III: Estimating errors in surface sources. Inverse Problems 9:649-665. Enting, I.G., C.M. Trudinger and R.J. Francey (1995) A synthesis inversion of the concentration and 138C of atmospheric CO2. Tellus 47B:35-52 (1995. Erickson III, D. J., P.J. Rasch, P. P. Tans, P. Friedlingstein, P. Ciais, E. Maier-Reimer, K. Six, C.A. Fischer and S. Waiters (1996) The seasonal cycle of atmospheric CO2: A study based on the NCAR Community Climate Model (CCM2). Journal of Geophysical Research 101:15079-15097. Errico, R. (1997) What is an adjoint model? Bulletin of the American Meteorological Society 78:2577-2591. Francey, R.J., P.P. Tans, C.E. Allison, I.G. Enting, J.W.C. White and M. Trolier (1995) Changes in oceanic and terrestrial carbon uptake since 1982. Nature 373:326-330. Giering, R. (1996) Tangent linear and Adjoint Model Compiler, Users manual, Max Planck Institute, Bundesstrasse 55, 20146, Hamburg, Germany (http://puddle.mit.edu/--~ralf.tamc.tamc.html).
294
M. Heimann and T. Kaminski
Heimann, M. (1995) The global tracer model TM2. Technical Report No. 10, Deutsches Klimarechenzentrum, Hamburg, 47 pp. Heimann, M. and C.D. Keeling (1989) A three dimensional model of atmospheric CO2 transport based on observed winds: 2. Model description and simulated tracer experiments. In: D.H. Peterson (Ed.) Aspects of Climate Variability in the Pacific and the Western Americas, Geophysical Monograph 55, Americal Geophysical Union, Washington, USA, pp. 237-275. Heimann, M., E.-D. Schulze, P. Monfray, P. Ciais, I. Levin, G. Dedieu, K. Homen, A.J.H. Meier and N. Vygodskaya (1997) Eurosiberian Carbonflux, Proposal to the Environment and Climate RTD Programme of the European Commission. http://www.bgc-jena.mpg.de/projects/sibir/flux01, html Hein, R. and M. Heimann (1994) Determination of global scale emissions of atmospheric methane using an inverse modeling method. In: J. Van Ham, L.J.H.M. Janssen and R.J. Swart (Eds.), NonC02 Greenhouse Gases, Kluwer, Dordrecht, the Netherlands, pp. 271-281. Hein, R., Crutzen, P.J. and M. Heimann (1996) An inverse modeling approach to investigate the global atmospheric methane cycle. Global Biogeochemical Cycles 11:43-76. Jacob, D.J., M.J. Prather, S.C. Wofsy and M.B. McElroy (1987) Atmospheric distribution of 85Kr simulated with a general circulation model. Journal of Geophysical Research 92:6614-6626. Kaminski, T. (1998) On the benefit of the adjoint technique for inversion of the atmospheric transport employing carbon dioxide as an example of a passive tracer. Ph.D. Thesis, Max-Planck-Institut f'tir Meteorologic, Hamburg, Germany. Kaminski, T., R. Giering and M. Heimann (1996) Sensitivity of the seasonal cycle of CO2 at remote monitoring stations with respect to seasonal surface exchange fluxes determined with the adjoint of an atmospheric transport model. Physics. and Chemistry of the Earth 21:457-462. Keeling, C.D., S.C. Piper and M. Heimann (1989) A three dimensional model of atmospheric CO2 transport based on observed winds: 4. Mean annual gradients and interannual variations. In: D.H. Peterson (Ed.) Aspects of Climate Variability in the Pacific and the Western Americas. Geophysical Monograph 55, American Geophysical Union, Washington (USA), pp. 305-363. Knorr, W. and M. Heimann (1995) Impact of drought stress and other factors on seasonal land biosphere CO2 exchange studied through an atmospheric tracer transport model. Tellus 47B:471489. Law, R.M. et al. (1996) Variations in modeled atmospheric transport of carbon dioxide and the consequences for CO2 inversions. Global Biogeochemical Cycles 10:783-796. Levin, I. and V. Hesshaimer (1996) Refining of atmospheric transport model entries by globally observed passive tracer distributions of 85Krypton and sulfur hexafluoride (SF6). Journal of Geophysical Research 101:16745-16755. Mahowald, N.M., P.J. Rasch and R.G. Prinn (1995) Cumulus parameterizations in chemical transport models. Journal of Geophysical Research 100:26173-26189. Maiss, M. and I. Levin (1994) Global increase of SF6 observed in the atmosphere. Geophysical Research Letters 21:269-272. Masarie, K.A. and P.P. Tans (1995) Extension and integration of atmospheric carbon dioxide data into a globally consistent measurement record. Journal of Geophysical Research 100:1159311610. Menke, W. (I 989) Geophysical Data Analysis, Academic Press, San Diego, CA. NOAA/CMDL (1996) Cooperative Atmospheric Data Integration Project - Carbon Dioxide (Globalview-CO2), CD-ROM, NOAA/CMDL, Boulder, Colorado. Penner, J.E., C.S. Atherton, J. Dignon, S.J. Ghan, J.J. Walton and S. Hameed (1991) Tropospheric nitrogen: A three-dimensional study of sources, distributions, and deposition. Journal of Geophysical Research 96:959-990. Polian, G., G. Lambert, B. Ardouin and A. Jegou (1986) Long-range transport of continental radon in subantarctic and antarctic areas. Tellus 38B:178-189. Prather, M.J., M.B. McElroy, S.C. Wofsy, G. Russel and D. Rind (1987) Chemistry of the global troposphere: fluorocarbons as tracers of air motion. Journal of Geophysical Research 91:66716681.
Inverse modelling approaches to infer surface trace gas fluxes from observed atmospheric mixing ratios
295
Ramonet, M. and P. Monfray (1996) CO2 baseline concept in 3-d atmospheric transport models, Tellus 48B:502-520. Rasch, P. and D. Williamson (1990) Computational aspects of moisture transport in global models of the atmosphere. Quarterly Journal of the Royal Meteorological Society 116:1071-1090. Rayner, P.J. and R.M. Law (1995) A comparison of modeled responses to prescribed CO2 sources. Technical Paper 36, CSIRO Divison of Atmospheric Research, Aspendale, Victoria, Australia. Rayner P.J., I.G. Enting, R.J. Francey and R. Langenfelds (1998) Reconstructing the recent carbon cycle from atmospheric CO2, 813C and O2/N2 observations. Tellus, submitted. Robertson, L., J. Langner and M. Engardt (1996) MATCH - Meso-scale atmospheric transport and chemistry model. Basic model description and control experiments with 222Rn. Swedish Meteorological and Hydrological Institute, Report RMK-70, 37pp. Rood, R.B. (1987) Numerical advection algorithms and their role in atmospheric transport and chemistry models. Review of Geophysics 25:71-100. Rostaing, N., S. Dalmas and A. Galligo (1993) Automatic differentiation in Odyss6e. Tellus 45A:558568. Snieder, R. (1993) Global inversions using normal modes and long-period surface waves. In H. M. Iyer and K. Hirahara (Eds.) Seismic tomography." Theory and practice, Chapman and Hall, London. Sofiev, M. (1999) Validation of model results at different scales. In: A.F. Bouwman (Ed.) Approaches to scaling of trace gas fluxes in ecosystems. Developments in Atmospheric Science 24. Elsevier, Amsterdam, pp. 233-255. Spee, E.J. (1998) Numerical methods in global transport-chemistry models. Ph.D. thesis, University of Amsterdam, 150 pp. Spee, E.J., J.G. Verwer, P.M. de Zeeuw, J.G. Bom and W.H. Hundsdorfer (1997) A numerical study for global atmospheric transport-chemistry problems. Report MAS-R9702, Centrum voor Wiskunde en Informatica, CWI, Amsterdam, 27pp. Tans, P.P., P.S. Bakwin and D.W. Guenther (1996) A feasible global carbon cycle observing system: a plan to decipher today's carbon cycle based on observations. Global Change Biology 2:309-318. Tarantola, A. (1987) Inverse problem theory - methods for data fitting and model parameter estimation. Elsevier, Amsterdam. Trampert, J. and R. Snieder (1996) Model estimations biased by truncated expansions: Possible artifacts in seismic tomography. Science 271:1257-1260. Trumbore, S. (1999) Role of isotopes and tracers in scaling trace gas fluxes. In: A.F. Bouwman (Ed.) Approaches to scaling of trace gas fluxes in ecosystems. Developments in Atmospheric Science 24. Elsevier, Amsterdam, pp. 257-274. Weiss, W., A. Sittkus, H. Stockburer and H. Sartorius (1983) Large-scale atmospheric mixing derived from meridional profiles of Krypton-85. Journal of Geophysical Research 88:8574-8578. Wofsy, S.C., B.C. Daube, R.F. Keeling, B.B. Stephens, P.P. Tans, P.S. Bakwin and J.L. Stith (1997) CO2 budget and rectification airborne study - North America (COBRA-NA). Proposal to the U.S. Deptartment of Energy, 18 pp.
This Page Intentionally Left Blank
Chapter 15
H O W SHOULD THE U N C E R T A I N T I E S IN THE RESULTS OF SCALING BE I N V E S T I G A T E D AND DECREASED
?
R.G. Derwent, A R Mosier, S. Bogdanov, J.H. Duyzer, V. Gargon, S. Houweling, M.A. Sofiev, H. Denier van der Gon, F. Wania, R. Wanninkhof
This Page Intentionally Left Blank
Approaches to scaling a trace gas fluxes in ecosystems A.F. Bouwman, editor 9 Elsevier Science B.V. All rights reserved
WORKING GROUP REPORT HOW SHOULD THE UNCERTAINTIES IN THE RESULTS OF SCALING BE INVESTIGATED AND DECREASED .9
R.G. Derwent (Rapporteur), A R Mosier (Chairman), S. Bogdanov, J.H. Duyzer, V. Gargon, S. Houweling, M.A. Sofiev, H. Denier van der Gon, F. Wania and R. Wanninkhof
I. I n t r o d u c t i o n
To address these questions we first set the context of our discussions by defining the terms: results, scaling and uncertainty. Our result is an annual trace gas flux for the scale, based on area of interest (Table 1). Yearly flux is the preferred unit of addition and subtraction from a budget point of view. Naturally, measurements are often made at a smaller temporal scale, for example, over a particular season only, and then extrapolations are required to generate an annual flux, which may or may not increase the uncertainty beyond that in the field-scale flux. We use specific examples of a gas that is ~ong-lived in the atmosphere (car: on dioxide, CO2) and one that is reactive and short-lived (nitric oxide, NO), as chemical end points so that the concepts that follow can be used generally for all short- or long-lived compounds. Scales extend from the point measurement of a trace gas at a scale of < m 2 to the globalscale (Table 1). Consideration is given here to the scaling up of trace gas from the smaller to the larger scale (bottom-up) and from the larger to the smaller scale (top-down). Our working definition of uncertainty is that it represents the result of an objective analysis of the variability in an annual trace gas flux, on a statistical basis. For example the global atmospheric increase in CO2 is estimated to be 3.2 + 0.2 Pg C (1 Pg=10 ~5 g yrl), see Table 2. Although this "simple error analysis" may become difficult as scale size increases and definition of boundary conditions decreases, an estimate of the variability of a trace gas flux presents a quantitative estimate of the uncertainty of the flux. The term uncertainty may otherwise have a variety of meanings. It may also be easier to define uncertainties for present or past events than for predicted future events. Uncertainty can also be viewed subjectively, in relative terms, depending upon how the information is to be used. For example, site selection for a mea;urement campaign may not nece/sarily be random,
Table 1. Identification of the major scales employed in the investigation of trace gas fluxes from terrestrial and aquatic ecosystems, togelher with their typical spatial and temporal scales. Scale
Time
Space
Point Field
Minutes-hours hours-days
0.01-0.01 km O. 1- 1 km
Local
days-months
1-1 O0 km
Regional
weeks-years
100-1000 km
years-decades
> 10 000 km
Global
300
R.G. D e r w e n t et al.
T a b l e 2. Average annual global budgets and fluxes of CO2 for the period
1980 to 1989 based on Houghton et al. (1995). Source/flux
Pg C yr -!
CO2 sources
Fossil fuel and cement Changes in tropical land use Total emissions from human activities
5.5• 1.6• 7.1•
P a r t i t i o n i n g b e t w e e n the e n v i r o n m e n t a l reservoirs
Storage in the atmosphere Oceanic uptake Uptake by northern hemisphere regrowth Additional sinks from CO2 and nitrogen fertilization
3.2• 2.0• 0.5• 1.4•
but selected on the basis of real, physical or economic constraints. For some issues, a relative (0 to 1 scale) may provide a satisfactory methodology for handling uncertainty.
2. Scales a n d t r a c e gas b u d g e t s
Considerable uncertainties remain in the regional and global budgets for a number of atmospheric trace gases because of the enormous spatial heterogeneity and temporal variability of the factors controlling trace gas fluxes in both aquatic and terrestrial ecosystems. Indeed, oceanic and terrestrial phenomena are expressed over scales ranging across many orders of
ecological complexity
10 I
(number of state variables)
O
High complexity process models
\
O
-1
....):~i................:9 ..... .:.: ....................-
.......::( . ............
space Extrapolation
Regionalmodels
.~ 1 ~ ~.. 3 ~
........q ill .......... O
....... :...i
log (1 O0 k m )
time 5 " ~
Generalcirculation models
log ( y e a r s )
F i g u r e 1. Arrangement of different types of models according to their characteristic spatial and temporal scale, compared with the scales over which models are extrapolated. Note that space and time scales are log scales. Modified from Murphy et al. (1993).
How should the uncertainties in the results of scaling be investigated and decreased ?
301
Figure 2. Diagramshowingthe top-downscaling approach with its hierarchyof models (inverse models)and the bottom-up approachwith the various forward(direct) models. magnitude, from seconds to millennia, and from microns to thousands of kilometers (Table 1) and variability is observed on all these scales. Techniques used for extrapolating measurements or properties and constraining results between different temporal and spatial scales are referred to as "scaling". Two approaches to scaling are used, i.e., the bottom-up and top-down scaling. Models are widely used tools in bottom-up scaling approaches. The range of possible models that can be used and coupled together is large, and models may cover many different scales (Figure 1). There is a need for a hierarchical approach to modelling ecological (aquatic and terrestrial) systems for crossing scales. Process models of high ecological complexity, probably imply high resolution in time and space. Ecological complexity is defined here as the number of components included in ecological process models. Regional models tend to have a lower temporal and spatial resolution and hence a reduced ecological complexity. Highly aggregated or parameterized models are generally associated with atmospheric or oceanic general circulation models. Figure 1 shows the scales over which the models are built compared with the scales over which models are extrapolated. Figure 2 illustrates on a circular diagram the different steps in scaling, the top-down approach with its hierarchy of models (inverse models) and the bottom-up approach with the various forward (direct) models. At each given scale, it is crucial to validate any model used with an observational data set. Measurement data should also be accompanied with further information concerning precision and applicability to validation at a given scale. Three main reasons can be invoked for conducting model validation: (i) to obtain a quantitative evaluation of the precision of the model output results; (ii) to gain an understanding of particular trends, minima and maxima, and so on; and (iii) to assess existing effects and to find out some "state of the art" understanding of phenomena (Sofiev, 199q). Here, uncertainty is investigated at a given scale using a mass balance approach. Uncertainty at a given scale depends, in part, upon the scaling approach, top-down or bottom-up, which has been chosen to reach selected objectives. We will illustrate this strategy with examples of a long-lived radiatively active gas such as CO2, and a short-lived reactive gas such as NO.
302
R.G. Derwent et al.
3. Calculating trace gas budgets on different scales 3.1. Mass balances and budgets A trace gas budget for a particular environmental compartment or reservoir can be formulated by drawing an imaginary boundary around the compartment and constructing a mass balance for the exchanges across the boundaries and the intemal sources and sinks. The basic equation for the trace gas mass balance in the compartment is: U
M
dc
(1)
F,n-F~,u,+~E,-~LI-V-t=l
/=1
dt
where c is the mean trace gas concentration in mass or molecules per unit volume, V is the compartment or reservoir volume, Fin is the advection flux of the trace gas into the compartment, Four is the advection flux of the trace gas out of the compartment, ZE is the sum of the internal source terms for the trace gas, EL is the sum of the internal sinks. At each scale above point-scale, the uncertainty can be reduced by constraining the Fin-Fout of that particular scale. The most obvious way to apply that constraint is by using trace gas measurements that encompass the complete scale size and another option is inverse modelling. Micrometeorological techniques allow measurement of field-scale fluxes which can constrain the fluxes derived from scaling up from the point measurement, while aircraft measurements may do the same for the local- or regional-scales. Inverse modelling using surface trace gas measurements from an appropriately-sited baseline station or network with meteorological data such as wind trajectories may constrain the local- or regional-scale trace gas flux. Such constraints are generally easier to apply at the local-scale rather than at the regional-scale but there is no fundamental difference. Such techniques, from the global-scale perspective, may give constraints to the trace gas budget from the regional-scale. An example of putting constraints on the regional-scale trace gas fluxes is given by Fowler (1999), using aircraft measurements around the United Kingdom. However, it is generally harder to put constraints at the regional-scale. Studies at the regional-scale call for creativity. For Europe, baseline trace gas monitoring stations and meteorological analyses are available and have been useful in defining European source strengths for a range of trace gases (Simmonds et al., 1993; Veltkamp et al., 1995; Hensen et al., 1995; Simmonds et al., 1996; Vermeulen et al., 1997). However, it does not follow that other regional-scale trace gas fluxes could be likewise constrained since it might not necessarily be feasible. Other options to reduce uncertainties include: (i) the use of multiple trace gases and their correlations with time and space; and (ii) isotopic signatures (Trumbore, 1999). On the global-scale, where the environmental compartment or reservoir contains the entire global atmosphere, the inflow and outflow are both by definition zero. For long-lived, wellmixed gases such as CO2 the annual growth in the atmospheric burden can be determined accurately from the network of measurements of atmospheric concentrations and there is no need for scaling-up when estimating that quantity. Consequently, the balance on the sinks and sources must equal the change in mass deduced from the concentration changes. This "constraint" limits the degree of uncertainty in the mass balance estimates. On the point-scale, that is at one single point in the atmosphere, the number of sources and sinks at tbat point is limited and can be accessed accurately by measurements or estimation methods and so there is no need for scaling. On intermediate-scales, depending on the scale and the character of a region, there are often a large number of sources and sinks which have to be assessed and which are not
How should the uncertainties in the results of scaling be investigated and decreased ?
303
accessible to direct measurement (because of their spatial extent or because of their number). Trace gas fluxes thus have inherently the largest uncertainty at the intermediate-scale. It is possible that the uncertainty increases with the number of source and sink "units" that have to be assessed and summed. For trace gases with relatively short atmospheric residence time, the situation is different. Atmospheric concentrations will be heterogeneous and global inventories can no longer be accurately assessed readily from a small number of atmospheric concentration measurements. However, it is conceivable that there are regions for which the inflow and outflow terms could be neglected, if the short-lived trace gas has a distinct, spatially limited source region, for example. A mass balance encompassing this source region and its immediate surroundings, including the regions to which the gas can be transported from that source, could be constructed in such a way that the advective terms of the mass balance would not have to be quantified.
3.2. Up-scaling and aggregation At the point (process) level we have a trace gas emission from a process, i, of which the observed intensity, E, is given by: E, : A, e ,(*)
(2)
where A; is the elemental size, c; is the emission factor for the considered process. The emission may depend on several continuous environmental parameters, such as temperature, pressure and humidity, and these are indicated by * in the above equation. Let the accuracy of the point-scale measurements of the emission flux be ere and assume that for point-scale measurements, the size of an element is known exactly. Then, for the larger scales we have:
El~ents E F,e,a / ,o~ / ,.~ / ~,,,h =
elements ~ ,( * ) ~
t=l
A~
(3)
j=l
and several conclusions may be noted: (i) the uncertainty of the sizes of elements becomes non-zero starting from the field-scale and so the uncertainty of any mapping increases whilst the spatial scale increases; (ii) qE also increases with spatial scale since the ranges of the continuous environmental parameters, or indeed any new parameters which may influence emissions, also increases whilst the spatial scale increases. Other processes (such as losses due to sinks and chemical transformations) can be considered in the same manner. Hence uncertainty in bottom-up scaling increases as the spatial scale increases. The only exclusion is the global-scale, where we have an additional constraint since Fm and Fo,t are zero. This constraint may simplify the processes of performing validation and reducing the uncertainties.
3.3. Down-scaling and disaggregation Let us consider the inverse modelling as the process of downscaling. E = Te
(4)
where E is the emission flux vector to be calculated, e is the measured concentration vector,
304
R.G. Derwent et al.
T is the inverted transport matrix. Let Re and 1~ correspond to the covariance matrices of the emission fluxes and concentrations, respectively, then they will be connected by the equation: RE = TRcT T
(5)
where T T is the transposed matrix T. The uncertainties in the transmission matrix also result in an additional variability of the emission fluxes. These two effects result in increasing uncertainty from top to bottom. The sequential application of scaling "bottom-up" (or "top-down") results in increasing uncertainties whilst going along the chain. Only the inclusion of additional information at each scale reduces uncertainties for a particular scale and gives an improved basis for the next steps in scaling.
3.4. Identifying and estimating uncertainties in scaling 3.4.1. Point scale
At the point-scale, there is no uncertainty in the elemental size since it is defined as the size of the measurement (e.g., a leaf, the enclosed m 2 of rice paddy). The uncertainty in the emission measurement is derived from replicate measurements and typically in the order of 1-10%. Depending on the continuity of the measurements, this can have a time resolution of the order of minutes to a year. At the point-scale, uncertainties in flux estimates are usually at their least and are governed by the experimental design and the accuracy of the trace gas observations required. A processbased model is constructed which explains much of the variance in the observed trace gas fluxes and, if the explanatory variables have been thoughtfully chosen, is suitable to extrapolate the fluxes from the point to the field-scale. 3.4.2. Field scale
The uncertainty in the elemental size now appears but may still be minimal (e.g. the size of an agricultural field is still well defined, the number of leaves in a stand of trees will already be based on an estimate with a standard deviation). So, the uncertainty at the point-scale is passed on to the field-scale, enlarged by the uncertainty in the elemental pool size and by spatial variability at the field-scale. The uncertainty is therefore higher at the field-scale compared with the point-scale. At these scales, the scaling of the trace gas fluxes is a relatively straight-forward process and will involve sub-dividing the field up into a number of elements. The process model would then be applied within each element and the field-scale flux would be obtained by summing over all the elements. Since the elements are not obviously dissimilar to the experimental sites, the model would be applied without change but employing the particular input data for the explanatory variables appropriate to each element. The uncertainties involved in scaling are usually at their least in moving from the point to the field-scale. They may be investigated by: (i) field-scale model evaluation campaigns; (ii) field-scale budget studies; (iii) isotopic methods can be used to increase the sensitivity of field-scale budget studies by introducing a unique label for the exchange of the trace gas with the ecosystem in question; and, (iv) using multiple trace gases to provide a cross check on the relative fluxes of the gases. By increasing the density of point-scale measurements within a field-scale study on a campaign basis, it should be possible to check the validity of the process-based model
How should the uncertainties in the results of scaling be investigated and decreased ?
305
employed, together with the aggregation procedures used to scale up to the field-scale. By making observations upwind and downwind of the studied field, it is possible to infer fieldscale trace gas fluxes which can be compared with the results of the aggregation and scaling procedures from the lower scale. 3.4.3. Local scale
In principle, the same procedures can be used to investigate uncertainties in scaling from the field- to local-scales as from the point- to field-scales, involving model evaluation campaigns and local-scale budget studies. The uncertainty in the element distribution increases (based on e.g. mapping, census), together with the spatial variation in the parameters (such as soil types and vegetation composition). In the case of the ecosystem trace gas flux at the local-scale, no other uncertainties are involved. However, when trying to constrain the flux at the local scale by fixing the Four, other sources and sinks may also contribute to the budget of the trace gas being studied. Uncertain-ties in these sources and sinks are largely independent and are transferred to the uncertainty of the trace gas flux at the local-scale. Scaling from the field scale to the local scale will involve using the process-based models for each of the fields within a locality. Since biological species composition and meteorological conditions are homogeneous over the locality by definition, the process-based models can be used at the local scale without significant adjustment. Data for the explanatory variables in the process-based models have to be obtained for each field and the aggregation process may have to be handled within a geographical information system. It may be that not all data on the explanatory variables are available, in which case new variables or surrogate data would have to be used. Depending on the definition of the locality, there might be up to hundreds of fields within the locality. For such techniques to be effective, the chosen ecosystem would necessarily need to be the dominant source or sink of the trace gas within the locality. 3.4. 4. Regional scale
Generally speaking the regional scale is not much different from the local scale, it is larger and therefore uncertainties grow accordingly. Uncertainty increases because it is most likely that the number of sources and sinks involved in the total flux (budget) increases, uncertainty in the element distribution increases (mapping, census) and spatial variation within the source and sink elements increases because soil composition, agricultural management will vary over the study region much more than at the local scale. Alternatively, the regional scale could be defined as that scale over which parameters may (but not necessarily) exceed the range covered by the field- and point-scale measurements (climate, temperature, light, etc.). At the local-scale this range is not exceeded. Uncertainties are generally the largest and confidence is the least in the scaling up of trace gas fluxes from the local- to the regional-scale (Figure 2), despite the approach being, in principle, much the same. Depending on the definition of the region, there might be up to thousands of localities within the region. Uncertainties enter into the scaling because coverages are not always available within the geographical information system for all of the explanatory variables in the process-based model. Surrogate variables may have to be constructed to fill in gaps in coverages. The process-based model may have to be simplified and perhaps recast using an entirely different and inferior set of variables. To investigate the uncertainties in the scaling up from the local- to the regional-scale, we
306
R.G. Derwent et al.
rely on the same validation procedures as employed at the smaller scales. This validation is, however, much more difficult than at the smaller scales and makes a significant contribution to the level of confidence in regional-scale ecosystem trace gas flux estimates. There are a number of difficulties associated with the validation of ecosystem fluxes for the trace gases at the regional-scale, as follows: (i) other sources and sinks for the trace gas may be operating on the regional-scale in addition to the chosen ecosystem source; there may be, for example, an animal enteric fermentation source of methane associated with the use of organic fertiliser in rice paddies; (ii) there may be significant changes in meteorology within the time-scale of the transit time across the region; (iii) it is difficult to establish accurately the upwind boundary conditions for the trace gas for each wind direction; (iv) large sources within the region may disguise the presence of smaller sources further upwind; (v) the large spatial scales involved with regional-scale validation makes experiments more onerous, difficult to replicate and expensive to maintain over extended periods; and, (vi) cross-checks with other pollutants may become unsatisfactory because of divergent fates and behaviour between the individual trace gases on the regional scale. 3.4.5. Global scale
From the bottom-up approach, uncertainty would be anticipated to be highest at the globalscale. This is not always the case because for some trace gases the global-scale ecosystem flux is tightly constrained by the global trace gas budget. Under these conditions, scaling up the uncertainties from the regional-scale to the global-scale is not relevant. However, in situations where the global budgets cannot be constructed with any confidence, the global constraint on total ecosystem flux does not apply and uncertainty is indeed highest at the most aggregated scale (Figure 2).
4. Investigating and reducing uncertainties in ecosystem fluxes The conceptual framework of uncertainties in results of scaling can be illustrated using examples of different environmentally important trace gases, with different atmospheric lifetimes. We use CO2 as an example of a long-lived trace gas and NO as a short-lived gas.
4.1. Carbon dioxide Here we u s e CO2 as an example of a trace gas which is long-lived and well-mixed throughout the lower atmosphere. Assessment of the uncertainties in the scaling of net ecosystem exchange fluxes of CO2 is of paramount importance for the implementation of possible mitigation strategies. Because of the rich data set of atmospheric CO2 observations, it is possible to construct a global budget which constrains the global net ecosystem exchange flux. For convenience the oceanic and terrestrial reservoir are discussed separately because of the differences in heterogeneity and net source and sink strengths. CO2 fluxes between the ocean and the atmosphere are investigated by either mass balance constraints in the ocean, or by determining the flux from air-water partial pressure differences (ApCO2) (see Archer, 1999). Mass balances can either be performed in the surface mixed layer (20-100 m), basin wide, or for the whole ocean. Fluxes are determined from the ApCO2 multiplied by the solubility, Ko and the gas transfer velocity, k: F = k Ko ApCO2
(6)
How should the uncertainties in the results of scaling be investigated and decreased ?
307
At the poim-scale, fluxes are generally determined by measuring the ApCO2 and estimating the k from a parameterization with wind speed which has an uncertainty of a factor of two based on the range of derived relationships (Liss and Merlivat, 1986). The partial pressure difference ApCO2 can be measured with an accuracy of 1-2 ].tatm while the range of ApCO2 over the open ocean ranges from roughly -80 ~atm for sinks during biological blooms to 150 ~atm for sources in the Equatorial upwelling region. In coastal areas and inland waters the range can increase t o - 2 5 0 j.tatm in intense bloom situations to 350 j.tatm during coastal upwelling. On average, the global disequilibrium has to be -7 to -10 ~tatm to satisfy the global constraint of 2 • 1015 g C yr -1 uptake by the ocean. The error in CO2 measurement is absolute such that the relative error in ApCO2 will range from 3% to over 100%. The greatest errors correspond to situations with low fluxes such that error estimates on point-scale fluxes from ApCO2 alone is about 3-10% with little possibility in the near future of reducing this uncertainty. Scaling from the point-scale to the local-scale and field-scales is performed by taking ApCO2 measurements over an extended area and over several seasons. Lack of temporal coverage contributes significantly to the uncertainty in the scaling. Accurate wind speed records can now be obtained on these scales from satellite observations and numerical weather prediction models. In addition to the average wind speed, the variability in the winds has to be known since the relationship between gas exchange and wind speed is thought to be nonlinear. Independent verification of uncertainty is difficult and can be best estimated from the variability in the parameters that influence pCO2 such as temperature and salinity. A promising technique to determine fluxes over the ocean, and thereby reduce uncertainties, are micrometeorological measurements such as eddy correlation, relaxed eddy accumulation, and gradient measurements. Because of the small magnitude of the fluxes, these direct flux measurements (see Fowler, 1999; and Lapitan et al., 1999) are singularly difficult to perform over the ocean. Depending on the height of measurement, these trace gas flux measurements can cover ranges from 1 to 10s of km. Scaling up CO2 fluxes from the local- and field-scales to regional-scales is frequently done by interpolating data using parameterizations with temperature and more recently ocean colour. Global maps of these parameters on short (day-month) time scales are available. Uncertainties are determined by the robustness of the parameterizations that are empirical rather than obtained from first principles. Verification on the regional-scale is commonly performed using simple process models embedded in basin-wide ocean circulation models. Global-scale estimates are scaled up from regional measurements in a similar way as regional extrapolations. The uncertainty of global oceanic exchange is reduced because of independent observational and modelling constraints. The global observational constraints are based on uptake and partitioning of carbon isot%es 13C and ]4C between ocean and atmosphere (Trumbore, 1999). Ocean circulation models (OCMs) tuned to mimic the observed 14C distributions or column burdens can be used to downscale observations to the regional-scale although this work is still at a rudimentary stage. Recently, independent global CO2 constraints using changes in O2/N2 ratios in the atmosphere have been implemented decreasing the uncertainty on the global scale (Bender et al., 1996; Keeling et al., 1996). Scaling terrestrial CO2 exchange is in many ways more difficult than for ocean exchange because of the extreme heterogeneity and small net fluxes compared to the large variable gross exchanges. On the regional-scale even the direction of fluxes is uncertain. For instance, it is not clear whether the rain-forests are a net source or sink for CO2, and the magnitude of the sink strength of the Northern Boreal forests is uncertain. The large uncertainty in trace gas exchange fluxes at the point-scale is to a large extent caused by temporal variability that
308
R.G. Derwent et al.
ranges from diumal changes in photosynthetic uptake to interannual changes in climate. Pointscale measurements are seldom performed continuously and flux measurements have to be interpolated with significant uncertainty, even at this level. Local-scale constraints on uncertainty can be imposed by micrometeorological measurements but frequently this upscaling step cannot be done with sufficient temporal resolution to offer a strong constraint. At larger scales even fewer validation checks can be performed for terrestrial systems. Detailed process-oriented modelling approaches using functional types (e.g. the CASA model) incorporated into three-dimensional atmospheric transport models, provide a new way to increase the scale of observations through to the regional- and globalscales. On the global-scale the same constraints are in place as for the ocean. The combined net uptake by the ocean and terrestrial sinks is well known from the response of the global burden to the perturbation of the CO2 system by fossil fuel input; it is the partitioning between the oceanic and terrestrial sinks that is uncertain. As for the ocean, the global constraints are improving through the application of carbon isotopic ratios and determination of 02/N 2 ratios.
4.2. Nitric oxide
Nitrogen oxides are produced in several ecosystem processes mainly as nitric oxide (NO). In the atmospheric boundary layer NO is involved in many reactions. In this chapter, the sources and sinks of NO are discussed as an example of a highly reactive trace gas. Firstly, the most important sources of NO on a global scale are discussed (Table 3), then the atmospheric fates of the nitrogen oxides are described briefly. Finally, the processes by which these gases are removed from the atmosphere are described.
4.2.1. Sources of NO On the global scale, human activities as well as natural biogenic processes are important sources of NO. Some natural sources may even be linked to human activities. Table 3 starts with lightning, an important natural source, although not much information is available on the production of NO during lightning events. The production on the global scale is, however, relatively small although in background areas and in the absence of human activity, lightning can be important. Microbial denitrification and nitrification processes cause emissions from soils, with usually these processes taking place in the uppermost soil layer at depths of only a few centimeters. The production rate of NO has shown to be dependent on several variables. Strong correlations are observed with soil moisture content and temperature. The process seems to be understood relatively well on an ecosystem scale although the number of ecosys-
Table 3. Global budgets and their uncertainties lbr the nitrogen oxide (NO.,,) species. Source type
Global emissions (1012 g N yrl )
(1012 g N yr-l)
Uncertainty
Lightning
6
3
Soils
10
5
Biomass burning
8
2
Fossil fuel combustion
21
1
Total
45
14
Sinks
55
30
How should the uncertainties in the results of scaling be investigated and decreased ?
309
tems that have been studied is limited. Consequently there is an urgent need to carry out measurements in different ecosystems rather than long-term, detailed, studies on one site. The soil NO emission rate is also dependent on parameters such as ecosystem type, nutrient availability, and especially soil texture. In some areas the input from the atmosphere through dry and wet deposition is quite important in the soil nitrogen budget (Davidson and Kingerlee 1997; Delmas et al., 1997; Veldkamp and Keller, 1997). Through this mechanism, soil emissions are related to human activities. Global estimates of NO emissions from soil are rather crude and rely upon a limited number of measurements in a small number of ecosystems. Although the detailed geographical information on soil types and environmental parameters is lacking, more information about NO and soils is needed. As with the burning of fossil fuels, NO is also formed in the burning of biomass. A limited number of studies have been performed and little is known about the influence of biomass composition. Another important uncertainty is the number of fires that take place in a certain region, though increasingly, remote sensing methods are now being used to obtain better information on fires. Much more information is, however, available on the most important global source, fossil fuel burning. Emission factors are usually available as a function of fuel type and statistical information is available on fuel use. Therefore the uncertainty in the estimate of the emission from this source category is relatively small (Table 3). Summarizing, we can conclude that the largest uncertainties in estimates of global emissions of NO lie with the emissions of NO from soils and from lightning. 4.2.2. Atmospheric processes o f NO
NO may be rapidly converted by ozone (03) to nitrogen dioxide (N02). In the n0ctumal boundary layer NO will deplete 03 and participate in a series of other reactions such as reactions with NO2. During the day nitrogen oxides play a dominant role as a catalyst in ozone formation. In this process NO and NO2 are converted into one another rapidly: NO + 0 3 ~ NO2 In principle, this interconversion does not produce any ozone until hydrocarbons become involved. Through reactions of hydroxyl (OH) radicals, hydrocarbons are destroyed and radicals are formed that may shift, through reaction with NO, the above photostationary state towards ozone production. Note that no NO• (- NO + NO2 ) is lost in this process. It is therefore often more important to discuss the fate of NO• rather than NO or NO2 individually. NO and NO2 participate in several more reactions in which more stable products such as peroxyacetylnitrate (PAN), nitric acid and nitrate aerosol are formed and NOx is lost. As a result of these reactions, the atmospheric lifetime of NOx is typically less than one day. Usually, the term NOy is introduced in which apart from NO and NO2 also the secondary nitrogen oxides such as PAN, nitric acid and nitrate aerosols are represented. It is through these components that nitrogen oxides are finally removed from the atmospheric circulation. The atmospheric residence time of NOy is nearly 10 days. 4.2.3. Loss processes and the budget on a global scale
As outlined above, NO is hardly dry-deposited before being converted into stable products. Its products starting with NO2 are subject to several processes. Two major loss routes are important; dry and wet deposition. Dry deposition, the uptake of gases and aerosols at the surface in dry conditions is important for NO2 and for the secondary products such as PAN,
310
R. G. Derwent et al.
nitric acid and the nitrates. Wet deposition is equally important for nitrate aerosols. The uncertainty in estimates of the deposition fluxes on a global scale is high. Conversion rates of NO and NO2 can be estimated quite well but the uncertainty in dry deposition rates is large. Wet deposition rates can be accurately assessed and the representativity of point measurements is reasonable in large regions such as Europe or North America. In other regions, knowledge is limited because of the lack of measurements. Another sink is the free troposphere and estimates of this sink are up to 10% of the NOx budget. The estimated total of all loss processes of nitrogen oxides on the global-scale amounts to 55 Tg yr-1. The uncertainty in this number however is large and could amount to + 50%. Consequently a gap exists between the estimated global emissions and the estimated sum of all loss processes. So in contrast to CO2, there is no constraint available that can be derived from the global budget for oxidized nitrogen. 4.2. 4. B u d g e t s on other scales
The first scale to consider would be the point-scale measurements. The problems and drawbacks of point-scale measurements are considered elsewhere in Lapitan et al. (1999) and Asman et al. (1999). On the lowest scale, enclosure measurements can be carried out well. In setting up the experiment attention has to be paid to the specific reactivity of NO. Especially the reaction with 03 can easily cause significant bias. Provided that residence times in the enclosure and sampling lines are small, suitable corrections can be made. With a good set up, estimates of the deposition velocity of NO2 and the emissions of NO can made on various locations. These methods can be used to study the influence of, for example, vegetation type or state and soil texture. In addition the influence of environmental parameters such as soilmoisture content or temperature can be studied. The results of such studies will need to be generalized to extrapolate to a larger scale. As a first approximation it could be assumed that, for example, the influence of soil temperature has a similar effect on soil emission rates in other ecosystems. On the field scale the use of micrometeorological methods may provide flux estimates of NO and NO2. With suitable selection of measurement sites, experimental conditions, etc. problems associated with fetch, instationarity (conversions), advection, etc., can be avoided. Conversion of NO and NO2 between the earth and the observation height may bias flux estimates. Using a detailed model simulation, Duyzer (1992) showed that in most conditions with low measurement heights these problems might be small, although experimental evidence is not available at this stage. Measurements can be carried out on different temporal scales, from hourly to annual averages. Flux footprint techniques are currently used to link point measurements carried out in the fetch area to field-scale measurements (Fowler, 1999). On larger scales, the reactive nature of NOx gases becomes a problem as on these scales, atmospheric conversion processes are important. Assessing budgets on these scales requires knowledge of concentrations of several gases including nitrates in aerosols and PAN. These concentrations are required to calculate the exchange of gases with other regions. In the absence of these measurements, the uncertainty of NOx budgets is large. On this scale the dry deposition rates of all secondary products of NO• gases is also a major source of uncertainty (see for example, Conrad and Dentener [1999] for a discussion on compensation concentrations in gas exchange between soils and canopies and the atmosphere).
How should the uncertainties in the results of scaling be investigated and decreased ?
311
5. Conclusions
For a relatively well-mixed and long-lived trace gas, uncertainties in trace gas fluxes to ecosystems appear to be greatest at the regional scale (Figure 2). This is certainly true of the main greenhouse gases: CO2, methane (CH4)and nitrous oxide (N20). That the global-scale is not the most uncertain, is a consequence of the constraints which can be applied through accurate trace gas observations, coupled with a thorough and complete understanding of trace gas life cycles. For a relatively short-lifetime trace gas, a different picture emerges (Figure 2). Such trace gases have markedly heterogeneous spatial distributions and it is often not possible to constrain global budgets with enough accuracy to reduce the uncertainty in trace gas budgets. Uncertainty in trace gas fluxes therefore continues to grow with increasing spatial scale, from the point-scale to the global-scale.
6. Recommendations
For the long-lived and well-mixed trace gases, the priority for future research should be given to reducing uncertainties in regional-scale trace gas fluxes. This can be done by: -
Comparison of regional-scale models with observations; Direct measurements of regional scale fluxes by mass balance; Studies of multiple trace gases; Application of isotopes; Applying estimation methods for uncertainties based on Monte-Carlo and boot-strapping techniques; Giving greater emphasis to the estimation of uncertainties.
-
-
For the short-lived trace gases, priorities for research, by necessity, have to be directed both towards the improvement of spatial and temporal resolution in trace gas ecosystem fluxes. This will involve: -
Increasing the accuracy and coverage of baseline monitoring of trace gas concentrations; Extending atmospheric process studies to include a wider range of conditions and seasons; Studies of multiple trace gases; Developing more detailed and reliable parameterizations of ecosystem exchange pro-cesses such as dry and wet deposition and air-sea exchange; - Improving the accuracy and completeness of trace gas emissions, particularly for human activities such as fuel combustion and biomass burning.
Few short-lived trace gases have adequate enough global budgets so that it is still not yet possible to assess reliably the relative importance of ecosystem exchange processes.
References
Archer , D. (1999) Modelling carbon dioxide in the ocean: A review. In: A.F. Bouwman (Ed.) Approaches to scaling of trace gas fluxes in ecosystems. Developments in Atmospheric Science 24. Elsevier, Amsterdam, pp 169-183.
312
R. G. Derwent et al.
Asman, W.A.H., M.O. Andreae, R. Conrad, O.T. Denmead, L.N. Ganzeveld, W. Helder, T. Kaminski, M.A. Sofiev, S. Trumbore (1999) How can fluxes of trace gases be validated between different scales? In: A.F. Bouwman (Ed.) Approaches to scaling of trace gas fluxes in ecosystems. Developments in Atmospheric Science 24. Elsevier, Amsterdam, pp. 85-97. Bender, M., T. Ellis, P. Tans, R. Francey and D. Lowe (1996) Variability in the O2/N2 ratio of southern hemispheric air, 1991-t 994: Implications for the carbon cycle. Global Biogeochemical Cycles 10:9-21. Conrad, R. and F.J. Dentener (1999) The application of the compensation point concepts in scaling of fluxes. In: A.F. Bouwman (Ed.) Approaches to scaling of trace gas fluxes in ecosystems. Developments in Atmospheric Science 24. Elsevier, Amsterdam, pp. 203-216. Davidson, E.A. and W. Kingerlee (I997) A global inventory of nitric oxide emissions from soil. Nutrient Cycling in Agroecosystems 48:37-50. Delmas, R., D. Serca and C. Jambert (1997) Global inventory of NOx sources. Nutrient Cycling in Agroecosystems 48:51-60. Duyzer, J.H. (1992) The influence of chemical reactions on surface exchange of NO, NO2 and 03: results of experiments and model calculations. In: X. Schwartz and Y. Slinn (Eds.) Proceedings of the Fifth International Conference on "Precipitation scavenging and atmosphere-surface exchange processes", Richland, Washington, pp 1105-1114. Fowler, D. (1999) Experimental designs appropriate for flux determination in terrestrial and aquatic systems. In: A.F. Bouwman (Ed.) Approaches to scaling of trace gas fluxes in ecosystems. Developments in Atmospheric Science 24. Elsevier, Amsterdam, pp. 99-121. Hensen, A., W.M. Kieskamp, A.T. Vermeulen, W.C.M. Van Den Bulk, D.F. Bakker, B. Beemsterboer, J.J. Mols, A.C. Veltkamp and G.P.Wyers (1995) Determination of the relative importance of sources and sinks of carbon dioxide. Report ECN-C-95-035, Netherlands Energy Research Foundation Petten, The Netherlands. Houghton, J.T., L.G. Meira Filho, J. Bruce, H. Lee, B.A. Callander, E. Haites, N. Harris and K. Maskell (1995) Climate Change 1994." Radiative forcing of climate change and an evaluation of the IPCC IS92 emission scenarios, Cambridge University Press, Cambridge, 339 pp. Keeling, R.F., S.C. Pipier and M. Heimann (1996) Global and hemispheric CO2 sinks deduced from changes in atmospheric 02 concentrations. Yature 381:218. Lapitan, R., R. Wanninkhof and A.R. Mosier (1999) Methods for stable gas flux determination in aquatic and terrestrial systems. In: A.F. Bouwman (Ed.) Approaches to scaling of trace gas fluxes in ecosystems. Developments in Atmospheric Science 24. Elsevier, Amsterdam, pp. 27-66. Liss, P.S. and L. Merlivat (1986) Air-sea exchange rates: Introduction and Synthesis. In: P. BuatMernard (Ed.) Role of air-sea exchange in geochemical cycling, Reidel, Boston, pp. 113-129. Murphy, E.J., J. Field, B. Kagan, C. Lin, V. Ryabchenko, J. Sarmiento and J. Steele (1993) Global extrapolation. In: G.T. Evans and M.J.R. Fasham (Eds.) Towards a model of ocean biogeochemicalprocesses. Springer Verlag, Heidelberg, pp. 21-46. Simmonds, P.G., D.M. Cunnold, G.J. Dollard, T.J. Davies, A. McCulloch and R.G. Derwent (1993) Evidence for the phase-out of CFC use in Europe over the period 1987-1990. Atmospheric Environment 27A: 1397-1407. Simmonds, P.G., R.G. Derwent, A. McCulloch, S O'Doherty and A. Gaudry (1996) Long-term trends in concentrations of halocarbons and radiatively active trace gases in Atlantic and European air masses at Mace Head, Ireland from 1987 to 1994. Atmospheric Environment 30:4041-4063. Sofiev, M. (1999) Validation of model results at different scales. In: A.F. Bouwman (Ed.) Approaches to scaling of trace gas fluxes in ecosystems. Developments in Atmospheric Science 24. Elsevier, Amsterdam, pp. 233-255. Trumbore, S. (1999) Role of isotopes and tracers in scaling trace gas fluxes. In: A.F. Bouwman (Ed.) Approaches to scaling of trace gas fluxes in ecosystems. Developments in Atmospheric Science 24. Elsevier, Amsterdam, pp. 257-274. Veldkamp, E. and M. Keller (1997) Fertiliser induced nitric oxide emissions from agricultural soils. Nutrient Cycling in Agroecosystems 48:69-77.
How should the uncertainties in the results of scaling be investigated and decreased ?
313
Veldkamp, A.C., R. Eisma, A.T.Vermeulen, W.M. Kieskamp, W.C.M. Van Den Bulk, B. Beemsterboer, O. Zwaagstra, J.J. Mols, A. Hensen and G.P. Wyers (1995) Validation of methane source strengths. Report ECN-C-95-034, Netherlands Energy Research Foundation, Petten, The Netherlands. Vermeulen, A.T., B. Beemsterboer, W.C.M. Van Den Bulk, R. Eisma, A. Hensen, W.M. Kieskamp, J.J. Mols, J. Slanina, A.C. Veltkamp, G.P.Wyers and O. Zwaagstra (1997) Validation of methane emission inventories for NW-Europe. Report ECN-C-96-088, Netherlands Energy Research Foundation, Petten, The Netherlands. Wanninkhof, R. (1992) Relationship between gas exchange and wind speed over the ocean. Journal of Geophysical Research 97:7373-7381.
This Page Intentionally Left Blank
Chapter 16
CURRENT AND FUTURE PASSIVE REMOTE SENSING TECHNIQUES USED TO DETERMINE ATMOSPHERIC CONSTITUENTS
J.P. Burrows
This Page Intentionally Left Blank
Approaches to scaling a trace gas fluxes in ecosystems A.F. Bouwman, editor 9 Elsevier Science B.V. All rights reserved
CURRENT AND FUTURE PASSIVE REMOTE SENSING TECHNIQUES USED TO DETERMINE ATMOSPHERIC CONSTITUENTS
J.P. Burrows Institute of Environmental Physics and Remote Sensing, University of Bremen, Postfach 330440 28334 Bremen, Germany
I. Introduction Atmospheric pollution is unfortunately not new, for example in the 17 th century Evelyn discussed in his work Fumifugium, "the inconvenience of the aer and smoake of London" (Finlayson-Pitts and Pitts, 1986, and references therein). The word smog is derived from the words smoke and fog and was originally used to describe the conditions often experienced in London in winter. Emissions from the buming of coal containing sulphur resulted in a smog having adverse health effects and being chemically reducing. Summer smog was first observed in Los Angeles, Califomia (Haagen-Smit, 1952; Haagen-Smit et al., 1952). It is also detrimental for health but is oxidizing in character. Since the late 1940' s, it is recognized that atmospheric pollution has not only local but also global impacts, and as a consequence environmental issues at both regional and global scales have become matters of scientific debate and public concern. The scientific community has responded to the need to identify and assess potential environmental hazards in a variety of ways. For example, much effort has gone into the study of the physical and chemical processes determining the behaviour of the atmosphere. Similarly a hierarchy of atmospheric models have been developed to simulate the current state of the atmosphere, to predict its future behaviour and to estimate response to both natural and anthropogenically induced change. The vast majority of the constituents considered as pollutants are present in trace amounts in the unpolluted atmosphere, the most notable exceptions being the chlorofluorocarbon compounds (CFCs) and halons, which have no known natural sources. The assessment of the impact and consequences of increasing emissions of constituents into the atmosphere is not trivial because of the inherent non-linear and complex nature of the atmosphere. Therefore, detailed knowledge about the elementary atmospheric processes is required. The measurement of the composition and trends in the mixing ratios of atmospheric constituents (gases, aerosols and clouds) enables to test our understanding of the biogeochemical cycles within the atmosphere. Such measurements may also be used as an early warning signal: of the potential negative consequences resulting from a specific anthropogenic activity. For long-lived atmospheric species a limited number of measurement stations around the globe may provide an adequate monitoring network. However, for short-lived species and species having sources that are variable in time and space, the global measurement of concentrations can best be made from remote sounding instrumentation aboard orbitting space-based platforms. The development of remote sensing techniques for atmospheric constituents (gases,
318
J.P. Burrows
aerosols and clouds) and parameters is one of the most exciting developments in the environmental sciences during the past 25 years. Using these techniques it is possible to monitor the composition and behaviour of the global atmosphere on both short and long time scales. Remote sensing data will improve both the prediction of weather patterns and establish the importance of changing atmospheric composition for global climate. The number of applications of remote sensing data is growing rapidly. Current and future generations of instrumentation will provide data of great importance for global change issues. In this study a brief overview is given of our current understanding of the atmosphere and environmental processes causing atmospheric changes. The relevance and use of passive remote sounding of the atmosphere from space is then discussed. Finally, some recent measurements by remote sensing techniques of some important tropospheric constituents are described.
2. The earth's atmosphere and environmental concerns The composition of the earth's atmosphere is different from that of neighboring planets such as Mars and Venus, which are apparently lifeless. Fossil records indicate that the atmosphere evolved to its present composition as a result of life. The atmospheric increase of the concentration of oxygen (02) and ozone ((a:) since 4600 million years before present indicates that the build up of oxygen resulted from photosynthesis after the appearance of life (Figure 1). The amount of oxygen as shown in the Figure 1, is estimated from the analysis of the geological records on the basis of the chemical composition of fossils. Life on Earth could not have existed on land until sufficient ozone was there to protect the biosphere from harmful short-wave radiation. The amount of ozone, shown in Figure 1, is calculated using a simple photochemical model (Wayne, 1992). According to the Gaia hypothesis the biosphere has played an important role in determining the composition of the atmosphere since life on Earth began (Lovelock, 1979). This hypothesis also suggests that the biosphere maintains favourable conditions for life on Earth. On the geological time scale the impact of anthropogenic activities on the atmosphere has been insignificant. However, since the industrial revolution the energy and food requirements for the increasing human world population have risen dramatically leading to increasing injection of a number of trace gases into the atmosphere, the most significant being carbon dioxide (C02) and methane (CH4). In following the earth's atmosphere is described, thereafter a discussi6n is presented of the environmental processes which drive atmospheric change. 2.1. The earth's atmosphere The earth's atmosphere is a complex system. It consists of a set of layers which differ in their temperature gradient with respect to altitude. Figure 2 shows typical temperature and pressure profiles for mid-latitudes. The sign rate of temperature change in the atmosphere as a function of height enables regions of positive and negative gradient or lapse rate to be defined. Starting at the earth's surface, the temperature decreases up to the region known as the tropopause. The latter separates the troposphere, which is vertically well mixed, from the stratosphere, which is characterized by slow vertical mixing. In the stratosphere the temperature increases from the tropopause to the mesopause, which separates the stratosphere from the mesosphere. Above the mesosphere, the temperature increases again in the thermosphere.
Current and future passive remote sensing techniques used to determine atmospheric constituents
~,
-,
oO
~
/
s
I
~"
o "~ "~ E
L -~ --_ 1~
~o~
|/
/
I
~ "'~
= *~ ~" ~
#
..,
o
~,, &
t::
_ r~
=
-~ ~"
~'
~
"s .~ o ~
~ ~
.s*)"ff:5
~ ~
o
z
o =
•
l
~ I 0 -
"~"
~
I
I
~"
~
I
Geological period
tl ]11
.~ [u_- = .~. 121&l~ ~ ~
P. . . . . bri. . . . .
~/~o~]<sx~o-~PaL beforelife
~
.~I Z ~1
1~3151~1&1 ~
1
I
m m i
i
o~ "
Extinction of dinosaurs-65 MYr. Emerg. . . . . f . . . . S MYr. Early civilization~0.005 MYr.
w,
o u. 10_3 i_.
.t-
-
=
/ '
~
319
I
i
i
i
~ ~
I
9
Millions of years before present
~
Figure 1. The evolution of oxygen, ozone and life on earth. Source: Wayne (1992).
The increase in temperature in the stratosphere results mainly from the absorption of solar radiation between 200 and 300 nm by the stratospheric ozone layer. In the thermosphere a different but related mechanism results in a temperature increase, caused by the absorption of short wavelength solar radiation typically below 200 nm by molecules, atoms and ions. In the vacuum ultra violet region, the solar output does not obey Planck's black body theory and is
US S t a n d a r J
,0o
160
180
Atmosphere
(1976)
Ternperoture [;]40 200 220
,,v1
260
280
(heated , by, .oxygen of solar U . , , absorption , . , , , ,,
r ~
Mesopouse
80 1 Mesosphere
T
-
40
Strctosphere
(he~tea by ozone obsorption of
Troposphere
/r
/
\
\
~
\
(hected by surface ol:~Jorption o# solar visible)
0 10--6
10 - 4
10--2 100 Pressure [hPo]
102
104
Figure 2. Atmospheric pressure and temperature profiles for mid latitudes (American Standard Atmospheric Models).
320
J.P. Burrows
higher during periods of solar activity. As a result the temperature of the thermosphere is regulated by the solar cycle. The pressure of the atmosphere is highest at the earth's surface and decreases with height according to the barometric formula. The height of the tropopause varies between about 8 km at the poles and 16 km in the tropics. The mesopause occurs at typically about 45 km. Between 80 and 90% of the atmospheric mass are contained within the troposphere. The mesopause has typically a pressure of a few millibar. The troposphere and stratosphere therefore contain over 95% of the mass of the atmosphere. The behaviour of the stratosphere and troposphere are coupled through gas and particle exchange. Overall, the conditions experienced by the biosphere at the earth's surface are determined in a complex manner by the physical and chemical processes occurring in these regions. 2.2. Relevant environmental issues
The physics and chemistry of the atmosphere are not in themselves environmental issues but rather matters of scientific curiosity. However, knowledge of these systems gained from measurements of the distributions of trace atmospheric constituents has contributed significantly to the public recognition of several important contemporary environmental issues, for example: (i) stratospheric ozone depletion; (ii) global warming and climate change; (iii) global increase of tropospheric ozone; (iv) air quality; (v) biomass burning; and, (vi) acidic deposition. Possibly the most dramatic example of the role of science in the understanding and prediction of an environmental issue was the discovery that stratospheric ozone is depleted by the tropospheric release of chlorofluorocarbon compounds (CFCs). This story had many twists and turns. The first measurements of CFCs in the troposphere were made by Lovelock et al. (1973). Subsequently it was proposed that their presence might lead to a depletion of the stratospheric ozone layer (Molina and Rowland, 1974), followed about a decade later by the discovery of the so-called ozone hole over Antarctica and the recognition that this resulted from chlorine chemistry (Farman et al., 1985). The consequences of a reduction of the stratospheric ozone layer affect both the stratosphere and the troposphere. The assessment of our understanding of the role of global ozone has become a significant international scientific activity (e.g. WMO, ! 995, 1998). After much scientific debate and public discussion, international agreements have resulted (the Montreal protocol and its amendments) and controls on the production of CFCs have been introduced. Only recently it has been recognized that global warming can influence both the intensity and duration of the stratospheric ozone loss (Shindell et al., 1998; Dameris et al., 1998). Although already discussed over a century ago by Arrhenius and others (Arrhenius, 1896), the issue of global warming caused by the injection of the so-called greenhouse gases such as CO2 into the atmosphere, has become prominent in recent years. This is because of the rapid increase in atmospheric CO2, associated with the combustion of fossil fuels in the second half of the twentieth century. The recognition that other species can behave in a similar manner but often more effectively than CO2 has resulted in the definition of the "global warming potential" of trace gases. The list of greenhouse gases now comprises many species, including CO2, CH4,nitrous oxide (N20), CFCs and tropospheric ozone. The governments of the world, concerned with the potential negative consequences of global warming, have mandated that evalu.~tions be made to provide national and intemational policymakers with an accurate assessment of our current understanding of climate change (e.g.
Current and future passive remote sensing techniques used to determine atmospheric constituents
321
Houghton et al., 1995). At the recent earth summit at Kyoto, one of the most important topics on the agenda was concerned with this issue. A number of international agreements resulted which intend to decrease the atmospheric loading of the greenhouse gases. In the troposphere ozone plays a number of important roles: (i) the photolysis of ozone initiates the production of the most important tropospheric oxidizing agent, hydroxyl radicals (OH), and thereby determines to a large extent the oxidative capacity of the troposphere; (ii) ozone acts as a greenhouse gas; and, (iii) et elevated levels ozone is a health hazard. A little over 30 years ago the troposphere was perceived to be relatively chemically inactive. Although ozone was known to play an important role in local air pollution (HaagenSmit, 1952), it was believed that tropospheric ozone originated from the stratosphere and that most of it was destroyed by contact with the earth's surface (i.e. dry deposition). In the mean time, it has been recognized that the troposphere has a large global photochemical source (Crutzen, 1973) and measurements indicate a global increase in the amount of tropospheric ozone (Volz and Kley, 1988). Consequently, the study of the physics and chemistry of tropospheric ozone and its precursors has become an important research area. The abundance of tropospheric ozone has a major impact on air quality. After its discovery in Los Angeles, summer smog has become an issue for many urban and rural locations in the northern and southern hemispheres. Anthropogenic emissions from industrialized (fossil fuel combustion) and rural areas (biomass burning) are considered to be the main causes of the poor air quality. Under typically anticyclonic conditions ozone, aerosols and other toxic compounds can be generated in amounts which are biologically hazardous. Recently air quality laws have been introduced in many countries to limit the amounts of aerosol and ozone in the urban and rural atmosphere. Biomass burning has now been identified as a significant source of many atmospheric trace gases and aerosols (e.g. Watson et al., 1990, and references therein). The practice of biomass burning is increasing with growing human population, particularly in the tropics, and it has become a major global environmental issue in its own right (Levine, 1991). In the unpolluted planetary boundary layer, dimethyl sulphide (DMS) and other sulphur containing compounds are emitted by a variety of sources including oceans and volcanoes. These sulphur compounds are oxidized to form sulphur dioxide (SO2), which reacts in the gas and liquid phases with the hydroxyl radical (OH) and hydrogen peroxide (H202), respectively, to produce sulphuric acid (H2SO4). The latter compound has a low vapour pressure and forms cloud condensation nuclei (CCN). In a similar manner nitrogen dioxide (NO2) reacts with OH to form nitric acid (HNO3) in the gas phase. The acid anhydride dinitrogen pentoxide (N205) formed by the reaction of NO2 with the nitrate radical (NO3) readily produces HNO3 on reaction with water (H20) in the liquid phase. HNO3 has a high solubility in water and, therefore, accumulates in tropospheric aerosols and cloud drops. Hence, in an unpolluted atmosphere the natural sources of SO2 and NO2 provide a mechanism by which the pH of aerosol and rain is expected to be slightly acidic. In tropical rain forests organic acids produced as a result of biogenic emissions can provide a further natural source of acid. The combustion of fossil fuels results in the release of large amounts of NO and NO2 (together denoted as NOx) and SO2 into the planetary boundary layer and the free troposphere. With increasing use of fossil fuels since the industrial revolution there has been a corresponding increase of the amount of acid deposition. Acidification was first recognized as an important ecological issue in the 1970's. Public interest in the consequences of acid deposition peaked in the middle of the 1980' s with the concern over forest die-back caused by
322
J.P. Burrows
acidification. However this remains an important environmental issue and research area (e.g. Heij and Erisman, 1995). Although our understanding of the chemistry of the troposphere has advanced significantly in the last 25 years, there have been continual surprises and discoveries indicating that our knowledge and predictive capability remains at an elementary stage of development. Examples are the observations of the springtime loss of ozone at high latitudes (Barrie et al. (1988), the satellite-borne observations of the large-scale production of tropical tropospheric ozone (Fishman et al., 1990; 1992) and recent discoveries of large clouds of bromine oxide in the troposphere both at the ground and from space (Richter et al., 1998a, and references therein). Common to all the environmental issues described above is the fact that they are of international rather than national significance. They impact on the regional and global rather than the local scale. To a large extent these environmental issues are the consequence of the higher standards of living and the increase in the world population over the last two centuries, which resulted in dramatic changes in the nature of the earth's surface and increasing and different emissions of trace gases to the atmosphere. Management of the anthropogenic influence on global change is a challenging task for the future. Accurate global knowledge of the amounts, trends, transport and chemistry of atmospheric constituents is essential information required for this task.
3. Assessment and prediction needs for the earth-atmosphere system The assessment of a given environmental issue relates the current scientific information and understanding of this issue to those who need it as a basis for policy and decision making. In this sense it bridges the research and the decision making communities. In order to assess accurately the impact of anthropogenic activity on atmospheric ciemistry and potential climate change, a detailed understanding is required of: (i)the physical and chemical processes determining the behaviour of the atmosphere; and, (ii) the biological, physical and chemical processes controlling the emission and deposition of trace gases at the earth's surface. Knowledge about these processes is gained through field measurements, laboratory studies, and modelling of trace gas fluxes and atmospheric processes. Of particular importance for assessment purposes are: (i) the accurate measurement of trace atmospheric constituents; and, (ii) the continuity of the measurements to generate data products which are readily comparable over several decades. The number and type of measurements of a species required to provide a true global representation depends on the atmospheric lifetimes of the species involved. Table 1 presents a list of atmospheric trace constituents and parameters, their atmospheric lifetimes or cycling time, and their relevance for a number of global environmental issues. The time scales of atmospheric processes range from hours to years (Table 1). For example, a pollution episode may only last for a week or less, but the consequences of such episodes may last much longer. This reflects the fact that processes such as the emission rate to the pianetary boundary layer, homogenous (e.g. radical-molecule, radical-radical, photolysis) and heterogeneous reactions, including both wet and dry deposition, determine the production and removal rates of species within the atmosphere. For climate research long-term series of trace gas, cloud and aerosol measurements are required. CO2 is only one of several gas species known to be increasing in the troposphere. For example, the amount of CH4 has almost doubled since pre-industrial times. For long-lived species such as CO2, CH4 or N20 a reasonable estimate of the global tropospheric amount can
323
Current and future passive remote sensing techniques used to determine atmospheric constituents
Table 1. List of atmospheric trace constituents and parameters and major relevant environmental issues.
Species/parameter Ozone (03) Nitrogen oxides (NOx)b Water vapour (H20) Carbon monoxide (CO) Methane (CH4) Nonmethane hydrocarbons (NMHC) Chlorofluorcarbons (CFCs) Chlorofluorhydrocarbons (HCFCs) Carbon dioxide (COz) Nitrous oxide (N20) Sulphur dioxide (SO1) Ammonia (NH3) Aerosol Clouds Polar stratospheric cloud Surface spectral reflectance UV radiation
Environmental issuea
Approximate lifetimes or cycling time
Well mixed in the atmosphere
TO, GW TO, AD TO, GW TO TO, GW, SO GW, SO GW, SO GW GW GW, SO AD AD GW GW, TO SO GW, TO. SO TO, SO, GW
1 week 1 day 1 week 2 months 10 years Hours-weeks Many years Many years > 100 years > 100 years Hours Hours Hours-weeks Hours to days Days to weeks Months Highly variable
No No No No Yes No Yes Yes Yes Yes No No No No No n.a. n.a.
a AD, atmospheric deposition; GW, global warming; SO, stratospheric ozone chemistry; TO, tropospheric ozone chemistry. b NOx refers to the sum of NO and NO2. These species are rapidly interconverted during the day by photolysis of NOE and the reaction of NO with O3.
be obtained from a network of ground-based measurement stations. The data from these networks indicate that the rate of increase of atmospheric CH4 has decreased recently (e.g. Dlugokencky et al., 1998). The origin of such changes remains unclear and a matter of much debate due to the lack of correlative global measurements. For short-lived, highly reactive atmospheric species the only feasible approach for obtaining their global distributions is to make regular measurements from space-based platforms. Provided they have sufficient accuracy, global space-based measurements of species with relatively long atmospheric lifetimes can be used to determine source and sink regions and possibly to infer trends. In order to meet the need for global observations of atmospheric species, an optimized and integrated measurement system is required. Ground-based, shipboard, balloon- and aircraftbome measurements often have much higher spatial resolution than satellite-bome measurements. However, the appropriate combination of measurements at different scales leads to well-calibrated and validated data sets. These are required for the next phase in developing our understanding of atmospheric and climate change.
4. Passive remote sensing of atmospheric constituents One of the first examples of the modem application of remote sensing of the composition of the atmosphere was the assertion by Hartley in 1880 that the UV absorption in the solar absorption spectrum was attributable to ozone. In this manner the earth's stratospheric layer was discovered. Since this pioneering work and especially in the past 30 years, there has been rapid progress in the development of atmospheric remote sensing techniques. The emissions from the aurora borealis and other night and day glow emissions in the visible wavelength range intrigued the philosophers of science since the middle ages (e.g. The Kings mirror, Konigsspielet, a chronicle of the middle ages; Roach and Gordon, 1973; Bone, 1991; Brekke and Egland, 1994, and references therein). However, it was the development of
324
J.P. Burrows
atomic and molecular physics in this century, which has enabled air glows to be attributed to transitions from excited atoms, molecules and ions in the upper atmosphere. Excitation occurs for example by the absorption of short wavelength radiation, as a result of the photodecomposkion of molecules and also from chemiluminescent chemical reactions. The identification of such features enabled "the composition of the upper atmosphere to be investigated much earlier than the lower atmosphere. There are two categories of remote sensing, active and passive. Passive techniques utilize electromagnetic radiation emitted from or transmitted through the atmosphere, the radiation source being for example the black body emission from the earth's surface, or solar and stellar irradiances. The most critical part of a passive remote sensing instrument is its detector. In contrast, active remote sensing systems have their own radiation source and a detector, for example, radar and lidar techniques. Satellite sounding instruments generally employ one of two types of viewing geometry, i.e. nadir viewing or limb viewing. Nadir viewing instruments observe a selected solid angle centered about a given spot on the Earth. Spatial coverage is maintained by a scanning system (generally across-track or conical) as the satellite moves. Limb viewing instruments scan vertically the earth's atmosphere, observing large horizontal paths at different altitudes. Limb viewing generally yields high vertical resolution and ability to observe higher in the atmosphere than nadir sounding instruments. At low altitudes, the horizontal resolution of the limb observation is often limited by multiple scattering and other effects. In addition, the likelihood of clouds interfering with the observation restricts measurements below 10 km. Therefore, limb observations have been primarily used for sounding the stratosphere and lower mesosphere.
4.1. Remote sensing of gases Whether active or passive, the retrieval of information about atmospheric trace gases relies on the knowledge of the absorption, emission and scattering of electromagnetic radiation in the atmosphere. Gases exhibit characteristic fingerprint spectra in emission or absorption: (i) rotational transitions are observed primarily in the far infrared and microwave spectral regions; (ii) vibrational rotational transitions may be observed in the infrared; and, (iii) electronic transitions are mainly in the UV, visible and near infrared spectral regions. The advantages and disadvantages of each spectral region are discussed briefly below. Various instruments utilizing remote sensing techniques to determine atmospheric composition are discussed in the sections. The characteristics of the different instruments and their objectives are summarized in Table 2. 4.1.1. Remote sensing in the far infrared and microwave spectral regions All gases having a dipole moment exhibit active rotational transitions in absorption and emission in the microwave (mm) and far infrared (sub-mm) spectral regions. The doppler line widths of rotational spectra are relatively narrow and the observed atmospheric emissions or absorptions are pressure-broadened by collisions with air molecules. On the one hand this enables the amount of gas at different pressures to be retrieved from an understanding of the line shape of the spectral features. On the other hand this results in the broadening of the absorption or emission line as the pressure in the atmosphere increases (Figure 2) leading to a loss of specificity in the lower atmosphere. This limitation is compounded by strong atmospheric absorption by water vapor and carbon dioxide and results in saturation in many parts of the spectral region.
Current and future passive remote sensing techniques used to determine atmospheric constttuents
325
Therefore, ground-based measurements are restricted to atmospheric windows where the water vapour absorptions are relatively small. For limb space-based atmospheric sounding, this is less of a problem but retrieval of trace gases in the upper troposphere are restricted to a limited number of candidates. For many gases such measurements are optimal for sounding the stratosphere and mesosphere. For the lower stratosphere and upper troposphere, some interesting possibilities exist, but the range of application is more more restricted. Passive remote sounding instruments for both the mm-wave and sub-mm or far-infrared spectral regions have been developed for atmospheric remote sounding. Some examples are briefly described below. - Microwave. Millimeter and sub-millimeter wave radiometry have been very successfully used from a variety of platforms to measure stratospheric gases. By using microwave local oscillators (LOs) and filter banks, or appropriate spectrometers, the observed brightness temperatures can be measured at various frequencies or across an emission line from a particular molecule. The sensitivity of this technique has improved significantly during the last two decades, as the detector noise has been reduced. The first retrievals exploiting this technique were made from the ground based systems: vertical profiles being inverted from the pressure broadening of the observed emission line. However observations from platforms higher in or above the atmosphere have many advantages. Aircraft measurements typically use a fixed viewing angle, while balloon and satellite experiments scan the limb of the atmosphere. A significant American and European effort has gone into the development of microwave and sub-millimeter sounding of the stratosphere. Some highlights of these programs have been measurements of elevated C10 amounts, including ground-based measurements (deZafra et al., 1987), balloon-borne measurements (Waters et al., 1981, 1984; Stachnik et al., 1992) and aircraft-borne measurements (Crewell et al., 1993; Urban et al., 1998). Recently OH has been observed from an aircraft (Titz et al., 1995). The Jet Propulsion Laboratory (JPL) research team have successfully flown the Microwave Limb Sounder (MLS) aboard the Upper Atmospheric Research Satellite (UARS), which has been measuring stratospheric profiles of C10, 03, H20 and HNO3 since 1992 (Waters et al., 1993) (Table 2). In total some 140 publications have been published on the basis of MLS data. The Microwave Atmospheric Sounder (MAS) flew three times aboard the Shuttle also measuring C10, 03 and H20 and has also provided a unique set of observations (Hartmann et al., 1996). - Far-infrared. The spectral region from 7 to 200 cm l has been used to study molecular species in the stratosphere and mesosphere. In the far infrared spectral region vibrations having weak force constants and rotations of light molecules are active. For these stratospheric measurements, Fourier Transform Spectrometers (FTS) with relatively high spectral resolutions have been used. Two balloon-borne instruments have been flown a number of times (e.g. Carli et al., 1984; Chance et al., 1989). More recently an FTS instrument has been developed for flight on a high flying aircraft SAFIRE. (Table 2). In summary, with the possible exceptions of H20 and 03, trace gas measurements made thus far have exploited these long wavelength ranges exclusively for stratospheric measurements. 4.1.2. Remote sensing in the mid-infrared
In the mid-infrared, molecules with a dipole moment have active vibrational rotational transitions, which can be observed both in emission and absorption. For the observation of tropospheric species, pressure broadening of the spectral features is considerably less proble-
326
J.P. Burrows
Table 2. The different instruments discussed in this paper, coverage of their measurements, species measured and the
satellite. The list is not intended to be complete, but merely to illustrate the currently available instrumentation. Instrument
Name
Height of measurement a
Target Species
ATMOS
Atmospheric Trace Molecule Spectroscopy
ST and upper Tr
03, NO• N205
ATSR
Along Track Scanning Radiometer
AVHRR
Advanced Very High Resolution Radiometer. Four channels on the first 4 platforms listed, five channels on the last 5 platforms Backscatter Ultraviolet Ozone Experiment Cryogenic Limb Array Etalon Spectrometer
TR, sea surface TR
BUV CLAES
ST, TR, profiles ST
Space Shuttle Spacelab-3 CIONO2, HCI, HF, (1985), ATLAS-I,2 and 3 CH4, CFCs, e t c . (1992,1993 and 1994) Aerosols, clouds, sea ESA-ERS1,2 (1991-present) surface temperature Smoke, fire, clouds TIROS-N, aerosols, vegetation NOAA-6,-8, - 10; -7, -9, -11, -12, -13 (1978- present) 03 Nimbus-4 (1970-1974)
Temperature, N20, NO, NO2, HNO3, CF2C12, CFCI3, HCI, 03, CIONO2, CO> H20, and CH4 03, OH, HO2, HCI, HF, HNO3, CIONO2, HOCI,, etc. 03, NO2, H20 BrO, OCIO, SO> HCHO and clouds and aerosols 03, NO2
FTS
Fourier Transform Spetrometer
ST
GOME
Global Ozone Monitoring Experiment
TR and ST
GOMOS
Global Ozone Monitoring by Occultation of Stars Halogen Occultation Experiment
Upper TR, ST and ME ST CO 2, H20, 03, NO2, HF, HCI, CH4,NO ST 03, NO2, N20, H20, CF3CI, CH4, CIONO2, T and P ST and TR 03, N20, H20, CH4, CO and CO2 ST, ME CO2, H20, CO, N20, CH4, NO, NO2, N205, HNO3, 03 ST CO2, HNO3, 03, H20, NO2 ST CO2, 03 TR CO
HALOE ILAS I, II
Improved Limb Atmospheric Spectrometer
IMG
Interferometric Monitor for Greenhouse Gases Improved Stratospheric and Mesospheric Sounder
ISAMS
LIMS LRIR MAPS MAS MERIS
MIPAS
MLS MOPITT OMI POLDER
Limb Infrared Monitor of the Stratosphere Limb Radiance Inversion Radiometer Measurement of Air Pollution from satelllites Microwave Atmospheric Sounder Medium Resolution Imaging Spectrometer for Passive Atmospheric Sounding Michelson Inferometer for Passive Atmospheric Sounding Microwave Limb Sounder Measurement of Pollution in the Troposphere Ozone Monitoring Instrument Polarization and Directionality of the Earth's Radiance
ST
CIO, 03, H20
TR
H20, clouds and aerosol
Upper TR
Platform
03, NO,,, N205
UARS (1991-1993)
Geophysica - a high-flying aircraft ESA-ERS- 1 (1995-present)
ESA ENVISAT (2000) UARS (1991-present) ADEOS (1996-97) ADEOS II (1999) ADEOS (1996-97) UARS (1991-1992)
Nimbus 7 (1978-79) Nimbus 6 (1975) STS-2 (1981); Space Shuttle (1984 and 1994) Space shuttle ATLAS 1, 2 and 3 (1992, 1993, 1994) ESA-ENVISAT (2000)
ESA ENVISAT (2000)
CIONO2, CH4, CFCs, etc.; temperature ST CIO, 03, H20, HNO3 TR, profiles Total column of CO, CH4 + CO profiles TR 03, SO2,N O 2 , TR Polarization, aerosols, clouds
UARS (1991-present) NASA AM-1 (1999) EOS-CHEM (2003) ADEOS- 1 (1996-97)
Current and future passive remote sensing techniques used to determine atmospheric constituents
Table 2. Continued. Instrument Name
SAGE I SAGE II
SAGE III
Stratospheric Aerosol and Gas Experiment I Stratospheric Aerosol and Gas Experiment II
Height of measurementa
Species measured
Platform
Upper TR and ST profiles
03, NO2, aerosols
NASA Atmospheric Explorer Mission (1979-81) NASA Earth Radiation Budget Satellite (1984 present) Meteor 3M ( 1999); International Space Station (2002)
Stratospheric Aerosol and Gas Experiment III Stratospheric Aerosol Measurement II Stratospheric and Mesospheric Sounder
ST ST, ME
SBUV
Solar Backscatter Ultraviolet Ozone Experiment
ST, TR, profiles
SBUV-2
Solar Backscatter Ultraviolet Ozone Experiment 2
ST. TR, profiles
SAM II SAMS
SCIAMACHY Scanning Imaging Absorption Spectrometer for Atmospheric Cartography
SCR SME
Selective Chopper Radiometer Solar Mesospheric Experiment
TES
Tropospheric Emission Spectrometer
TOMS
Total Ozone Monitoring Spectrometer
327
03, NO2, H20, aerosols 03, OclO, BrO. NO2, NO 3 aerosols Aerosols CO2, H20, CO, N20, CH4, NO 03 03
TR, ST, and 03, 02, O2(IA), 04, ME total NO, NO2, N20, columns BrO, OclO CO, and profiles H20, S02, HCHO, CO, CO2 and CH4, cloud, aerosols, pressure, temperature ST CO2, T ST, ME 03, O2(IA), NO2 Profiles TR total Various incl. columns HNO3, O3, NO, and profiles H20 ST, TR, 03 profiles
Nimbus-7 (1979-90) Nimbus-7 (1979-90) Nimbus-7 (1979-90) NOAA -9 (1985-present) 11(1989-95)-14 (1995present) ESA-ENVISAT (2000)
Nimbus 4-5 (1970-75) NASA (1983) NASA-EOS-CHEM (2003)
Nimbus 7 (1979-92) ADEOS (1996-97) Earth Probe (1996-) Meteor (1992-94)
aME, mesosphere; ST, stratosphere; TR, troposphere.
matical than at longer wavelengths. This is because the Doppler line width is proportional to the frequency of the light and is therefore larger in the mid-infrared than at longer wavelengths. Nevertheless, broadening of lines results in some loss of specificity of the measurements. In addition, in the mid infrared strong absorptions by H 2 0 and CO2 restrict the available spectral windows. Passive remote sensing by mid-infrared spectroscopy has been successfully applied to the measurement of a large number of stratospheric trace constituents and some upper tropospheric constituents. Initially measurements were made from mountain tops (e.g. Zander, 1981), and balloon and aircraft experiments were subsequently developed (Fischer et al., 1980; Murcray et al., 1975; 1979; Coffey et al., 1981; Brasunas et al 1988; Kunde et al., 1988). Both absorption and emission experiments have been made using a number of different instruments. The Limb Radiance Inversion Radiometer (LRIR) and Limb Infrared Monitor of the Stratosphere (LIMS) are both infrared radiometers which were flown aboard Nimbus 6 and 7, respectively, and recorded data in 1978 and 1979 (Gille et al., 1980; Gille and Russell,
328
J.P. Burrows
1984). The six channels of LIMS observed emission by CO2, HNO3, 03, H20 and NO2 from 15-65 km. Starting with the selective chopper radiometers (SCR) on Nimbus 4 and 5, the Department of Physics at the University of Oxford developed a series of instruments observing infrared emission from the atmosphere. A pressure modulated instrument flew on Nimbus 6 and the Stratospheric and Mesospheric Sounder (SAMS) flew on Nimbus 7 (Table 2). SAMS measured in limb viewing geometry the gases CO2, H20, CO, N20, CH4 and NO in the stratosphere and mesosphere. An improved version ISAMS flew aboard the UARS and added channels for NO2, N205, HNO3, 03, and H20 (e.g. Barnet et al., 1992) (Table 2). During the last 15 years Fourier transform spectrometers have been used successfully to sound the stratosphere and upper troposphere. One of the most important successes has been the Atmospheric Trace MOlecule Spectroscopy (ATMOS) project (e.g. Farmer et al., 1987; Gunson et al., 1996). The ATMOS instrument flew aboard Spacelab 3 and the Atmospheric Laboratory for Applications and Science (ATLAS) Space Shuttle missions (Table 2). ATMOS performed solar occultation measurements and a variety of trace gases in the upper troposphere and stratosphere have been retrieved. Since its launch in 1991 the Upper Atmospheric Research Satellite (UARS) has circled the Earth in a low earth non sun-synchronous orbit. The UARS flew three infrared experiments. In addition to ISAMS (described above) the Cryogenic Limb Array Etalon Spectrometer (CLAES) and the Halogen Occultation Experiment (HALOE) make infrared measurements designed to yield information about stratospheric and tropospheric trace constituents. The CLAES used a high resolution etalon to measure the limb emission in the infrared (Roche et al., 1982; Roche and Kumer, 1989). The target gases and parameters were N20, NO, NO2, HNO3, CF2C12, CFC13, HC1, 03, C1ONO2, CO2, H20, CH4 and temperature (UARS, 1987). HALOE used broad band filter radiometry to measure CO2, H20, 03 and NO2 and gas filter correlation radiometry to measure HF, HC1, CH4 and NO (Baker et al., 1986) (Table 2). All four stratospheric remote sensing missions (MLS, ISAMS, CLAES and HALOE) aboard the UARS have achieved their goals. Currently the HALOE and MLS are still measuring after eight years and have produced a unique record about the stratosphere. The Japanese space agency NASDA launched its ADEOS (Advanced Earth Observing Satellite) in 1996. The payload included ILAS (Improved Limb Atmospheric Spectrometer), IMG (Interferometric Monitor for Greenhouse gases) and a TOMS for atmospheric sensing. IMG and ILAS are nadir and limb sounding infrared instruments (Table 2). Early measurements show very promising results. The retrieval algorithms are currently being optimized. The only experiment flown up to the present, which specifically uses infrared information to probe the lower troposphere is the Measurement of Air Pollution from Satellites (MAPS) experiment. The MAPS instrument is a nadir sounding gas correlation, which makes global measurements of CO in the middle and upper troposphere. It flew three times between 1981 and 1994 on the NASA Space Shuttle (Reichle et al., 1986; 1990; Connors et al., 1991). Validation of MAPS was made using ground-based passive remote sensing instruments (Pougatchev et al., 1998). 4.1.3. Remote sensing in the UVvisible and near-IR
In contrast to the longer wavelengths, the source of radiation for passive remote sounding of the atmosphere in the ultraviolet, visible and near and short-wave infrared regions is the sun. The sun's maximum emission is around 580 nm. Beyond 300 nm the sun corresponds fairly well to a black body having a temperate around 5800 K. Absorption by atoms in the sun produce the well known Fraunhofer structure. This has many strong features in the ultraviolet
Current and future passive remote sensing techniques used to determine atmospheric constituents
329
and visible but is less significant at longer wavelengths. Below 300 nm the solar emission deviates from black body behaviour. Moving further into the ultraviolet, the solar output is determined to a large extent by processes occurring at the edge of the sun. These processes are often correlates with solar activity. In 1957 it was proposed that satellite measurements of back scattered ultraviolet (BUV) radiation from the terrestrial atmosphere could be used to deduce ozone profiles on a global basis (Singer and Wentworth, 1957). The method relies on two effects, i.e. the scattering of light at short wavelengths and the absorption of ozone. Rayleigh scattering of light by air molecules has a strong dependence on wavelength, whereby the intensity of scattered light is a function of the inverse of the fourth power of the wavelength. Similarly, ozone absorption is strongly wavelength-dependent. These effects combine and as a result the penetration depth of light in the atmosphere varies strongly between the ozone maximum absorption in the Hartley band around 250 nm and its minimum beyond 380 nm in the Huggins bands. The numerical technique for the determination of vertical profile information was also studied for the determination of total ozone in the atmosphere by NASA (Dave and Mateer, 1967; Mateer et al., 1971). The development of this retrieval techniques has continued up to the present (Bhartia et al., 1996). The earliest measurement utilizing the BUV technique was undertaken by Rawcliffe and Eliot (1966) using a photometer observing at 284 nm. Ozone distributions were determined utilizing measurements from the USSR COSMOS satellites, which flew a double monochromator in 1965 and 1966 (Krasnopol'skiy, 1966; Iozenas et al., 1969a,b). The Backscatter Ultraviolet atmospheric ozone experiment (BUV) was the first of a series of instruments made by NASA and later NOAA, which has successfully made long-term measurements of the BUV for the vertical profile and total amount of ozone (Heath et al., 1973) (Table 2). BUV was launched aboard the Nimbus 4 satellite into a circular polar orbit at an altitude of 1100 km. This orbit is sun-synchronous and the satellite crosses the equator in an ascending mode every 107 minutes close to local noon. This instrument concept was developed and resulted in the SBUV (Solar Backscatter Ultraviolet) and TOMS (Total Ozone Mapping Spectrometer) being launched aboard Nimbus 7 (Heath et al., 1975). The SBUV instrument was further improved to the SBUV-2 and has been flown by NOAA on a series of satellites (Frederick et al., 1986) (Table 2). After Nimbus 7 (1979-1992) TOMS has also been flown on the Russian Meteor Platform (1992-1994), as part of the Japanese ADEOS satellite (1996-1997) and aboard Earth Probe (1996-present) (Table 2).These measurements have been used to derive a unique ozone data set. Readers are referred to the literature about the T O M S data record for more details. Recognizing the importance of long term calibration and validation of space based instrumentation, NASA developed the SSBUV (Shuttle SBUV), which flew 8 times on the shuttle to calibrate radiometrically the BUV instruments (Hilsenrath et al., 1988; 1996). The attention paid to the detail of the calibration of the NASA and NOAA BUV instruments has established the quality of these data sets (Hilsenrath et al., 1995). The Stratopsheric Aerosol and Gas Experiment I (SAGE I) instrument flew from 1979 to 1981 on the NASA Atmospheric Explorer Mission (Table 2). It is a satellite-borne spectrometer that measures the absorption of the sunlight by ozone with four channels centred at 0.385, 0.45 0.6 and 1.0 ~tm (McCormick et al., 1979; Chu and McCormick, 1979). SAGE II is a seven channel instrument from the same team (Maudlin et al., 1985), which was launched on NASA Earth Radiation Budget Satellite (ERBS) and is still working today. A third generation SAGE has been developed, which will be launched on board the Russian Meteor3M in 1999 and the international space station in 2002 (Table 2). SAGE determines atmospheric absorption in occultation and measures at sunrise and sunset. So far the SAGE
330
J.P. Burrows
series has provided reliable data about a number of atmospheric constituents. An interesting set of measurements were made by the Solar Mesospheric Explorer (SME), which was flown by NASA in 1983. It contained limb scanning ultraviolet, visible and near IR channels for measuring stratospheric and mesospheric 03, O2(lA) and NO2 (Rusch et al., 1984; Thomas et al., 1984; Mount et al., 1984) (Table 2). The Global Ozone Monitoring Experiment (GOME) represents the entry of the European Space Agency (ESA) into the measurement of global distributions of atmospheric constituents (Burrows et al., 1991; 1993; 1998b, and references therein). GOME is a small scale version of the SCanning Imaging Absorption spectroMeter for Atmospheric CHartographY (SCIAMACHY) and was originally named SCIA-mini. Both GOME and SCIAMACHY were proposed in 1988. GOME flies aboard the ESA's second earth research satellite (ERS-2), which was launched in April 1995. SCIAMACHY is a joint German, Dutch and Belgian national contribution to the ESA-ENVISAT payload to be launched in 2000 (Burrows et al., 1990; Burrows et al., 1991; 1995; Bovesmann et al., 1998) (Table 2). GOME is a nadir sounding spectrometer which observes simultaneously either the upwelling radiance from the top of the atmosphere or the extra-terrestrial solar irradiance between 240 and 790 nm. The resolution of the measurements is chosen to be suitable for the application of differential optical absorption spectroscopy (DOAS) technique, which was developed for long-path measurements and zenith sky observations (e.g. Platt and Perner, 1980; Mount et al., 1987; Eisinger et al., 1997). 4.2. Remote sensing techniques for aerosols and clouds
Molecular or Rayleigh scattering of electromagnetic radiation is strongly dependent on its wavelength (~-4), whereas scattering by aerosols and clouds obeys approximately Mie theory and has a relatively weak dependence on wavelength (~-l). Aerosols have been identified by their scattering effects both actively and passively. The first passive remote sensing experiment to measure successfully the abundance of atmospheric aerosols from space was the Stratospheric Aerosol Measurement (SAM II) aboard Nimbus 7 (McCormick et al. 1979). This experiment was a single channel radiometer observing in solar occultation and was the forerunner of SAGE. Stratospheric aerosols have also been measured by their infrared absorptions (e.g. HALOE). Tropospheric aerosol and smoke have been retrieved from several different types of remote sensing data. For example, algorithms to retrieve tropospheric aerosol have been developed for data from the Advanced Very High Resolution Radiometer (AVHRR) data over the ocean (Ignatov et al., 1995). Biomass burning smoke has not only been observed by AVHRR but also from TOMS data (Hsu et al., 1996). This algorithm has also been successfully applied to GOME data (Gleason et al., 1998). In the ultraviolet, visible and near-infrared spectral regions, radiation is strongly scattered by clouds, enabling their presence to be detected, provided that the spatial resolution is sufficiently high and the difference between the effective cloud albedo and the spectral reflectance of the earth's surface is suffici~,ntly large. Several algorithms have been developed, which aim at utilizing the oxygen absorption for the determination of cloud top height (Guzzi et al., 1996; 1998 and references therein). Clouds emit long-wavelength infrared radiation, the spectrum of which depends on their temperature. Observations of the infrared emissions by clouds with instruments on the METEOSAT platform and related meteorological satellites are routinely used to estimate the cloud top height and cover.
Current and future passive remote sensing techniques used to determine atmospheric constituents
331
The main objective of the ATSR (Along Track Scanning Radiometer) aboard ERS-1 and ERS-2 is the measurement of sea surface temperature. In addition, it has been used to determine aerosol and cloud (Table 2). The French POLDER (Polarization and Directionality of the Earth's Radiance) instrument aboard the NASDA platform ADEOS-1 measured polarization parameters (Table 2), which are being used to study polarization, aerosols and clouds in the atmosphere (Deschamps et al., 1990). 4.3. Validation of retrieved parameters from space based remote sensing measurements One of the most important aspects of any remote sensing experiment is the validation of retrieved parameters. Without a rigorous validation, the use of remote sensing data is of minimal value. Validation usually involves the measurement of the same parameter by an independent method, for example an in-situ measurement or an alternative remote sensing measurement from ground, ship or aircraft platforms, as required. Validation necessitates in-flight calibration of instruments in space. As the performance of instruments often degrades during their life in space, validation of data products is required systematically throughout the lifetime of a mission. This may be achieved by utilizing data from networks and organizing validation campaigns. A good example of the former is the NDSC (Network for the Detection of Stratospheric Change). This has been established to provide long-term measurements of the stratospheric composition at a selected set of locations. A similar network is needed for the troposphere, but requires more measurement sites than in the stratosphere because of the high variability of the troposphere. Unfortunately the value and importance of validation is not always recognized by the relevant governmental agencies. Often the majority of the funds designated for a particular mission have been used up during the industrial fabrication and launch of an instrument. The consequences of poorly calibrated or validated data is a less than optimal exploitation of data products. In summary, validation of the parameters retrieved from remote sensing measurements is an essential part of a mission. Validation measurements are required throughout the lifetime of a mission. Validation is best achieved by the comparison of the retrieved parameter with independent measurements. The latter may be in situ measurements or independent remote sensing measurements.
5. Remote sensing measurements of trace gases in the troposphere The retrieval of trace constituents in the trt, posphere is more difficult than in the stratosphere or mesosphere, because (i) pressure broadening and strong tropospheric absorptions makes the application of microwave, sub-millimeter and infrared techniques difficult if not impossible to invert; (ii) for nadir sounding, multiple scattering in the ultraviolet and visible wavelengths smears the information about absorption from the lower atmosphere; and, (iii) limb sounding has a horizontal resolution of the order of 400 km, implying that tropospheric clouds are nearly always present in the field of view, effectively restricting the limb measurements to above the cloud top. The most reliable and self-consistent approach for passive remote sensing of the troposphere is, therefore, the simultaneous use of limb and nadir sounding of the atmosphere (e.g. as pioneered by SCIAMACHY and related instruments).
332
J.P. Burrows
5.1. Tropospheric ozone
For tropospheric ozone, the residual technique was pioneered by Fishman et al. (1990), who used total 03 column, estimated from TOMS, combined with the co-located stratospheric ozone profiles from SAGE and SBUV in a series of studies to demonstrate the presence of elevated tropospheric ozone for large-scale pollution events (Fishman e t a l . , 1991; 1992; 1996). Figure 3 shows results of one of the studies done by Fishman and coworkers. These studies stimulated the investigation of the use of TOMS data for the determination of the tropospheric ozone column amount by related techniques. Hudson, Thompson and coworkers have developed and refined a technique called the tropical tropospheric ozone (TTO) method (Hudson etal., 1995; Kim et al., 1996; Hudson and Thompson, 1998). This technique utilizes a Fourier analysis to identify the range of latitudes for which the method is applicable by using the recognition of a planetary wave pattern to estimate stratospheric and background tropospheric ozone. Ziemke et aL (1998) have developed two different residual techniques. The first combines TOMS and stratospheric ozone information from the UARS instruments HALOE and MLS. The second approach identifies high clouds in the tropical region and assumes that such columns contain only information on stratospheric ozone. The use of such condensation cloud differential information (CCD) has given the technique its name.
Figure 3. Tropospheric ozone determined by the Tropospheric Ozone Residual technique on October 3 and 6, 1992. D.U., Dobson unit; 1 D.U. = 2.69 x 1016molecules per cmz. Source: Fishman et al. (1996).
Current and future passive remote sensing techniques used to determine atmospheric constituents
333
The GOME and SCIAMACHY ultraviolet measurements in nadir viewing geometry are similar to those of SBUV and SBUV-2. However, in contrast to the NASA instruments, GOME and SCIAMACHY observe the entire spectrum at a spectral sampling of 0.1 nm and a spectral resolution of 0.23 nm, whereas SBUV and SBUV-2 instruments measure consecutively 12 selected wavelengths. Each set of 12 SBUV wavelength measurements takes about 30 s, and the spectral resolution of each individual measurement is 1 nm. During the development of the SCIAMACHY proposal the idea of utilizing additional information in the temperature dependence of the ozone Huggins bands was proposed. The potential use of this technique for the retrieval of tropospheric information about ozone is described by Chance et al. (1991), Rozanov et al. (1992; 1993), Munro et al. (1993), Burrows et al (1994), Chance et al. (1997), Rozanov et al (1998) and DeBeek et al. (1998). Currently one issue, limiting the exploitation of the retrieval of ozone profiles from the GOME data set is the presence of systematic radiometric calibration errors, which have been identified but not yet eliminated from the operational geophysical irradiance and radiance data products. Some schemes have been developed, which remove these errors (Bramstedt et al., 1998; Hoogen et al., 1998; Eichmann et al., 1998) but these have not yet been implemented operationally. Above the ozone maximum the vertical resolution of GOME data for the ozone profile is of the order of 6 km. In the lower stratosphere and troposphere, accurate profiles with an effective vertical resolution of around 10 km have been retrieved from GOME data (Munro et al., 1998; Burrows et al., 1998b; Hoogen et al., 1999). However the removal of systematic errors in the GOME data set is critical with respect to the accuracy of tropospheric ozone retrievals from this data. The higher information content of the GOME profile information about ozone, as compared to the SBUV, arises from a variety of reasons. For example the higher effective signal to noise ratio of the measurements, the measurement of additional spectral features, such as the temperature dependent Huggins bands and the relatively temperature independent Chappuis bands, all play a role. Using a priori information about the temperature and height of the tropopause combined with the GOME measurements enables the tropospheric and lower stratospheric columns of ozone to be derived. Accurate knowledge about the scattering characteristic of tropospheric clouds and the earth's albedo is important in this respect. As a result of primarily multiple scattering in the lower atmosphere, the number of pieces of independent information retrievable in the troposphere and lower stratosphere from nadir viewing is limited. A simple way to obtain the tropospheric excess amount of a gas in the tropics is by assuming that locally the stratosphere is longitudinally homogeneous. This is similar to the residual approaches mentioned above. This has been used to identify the amount of excess tropospheric ozone produced over and in the plume downwind of Indonesia from the forest fires in September 1997. This excess column amount of ozone is attributed to the troposphere and is shown in Figure 4.
5.2. Nitrogen Dioxide At the earth's surface and in the lower atmosphere large amount of oxides of nitrogen are released by a variety of natural phenomena, chemical processes and anthropogenic activities. The major source of stratospheric NO and NO2 is nitrous oxide (N20). N20 is released into the troposphere by the biological reduction of NO3 and the oxidation of NH4 + in soils. Due to its long tropospheric lifetime, significant quantities of N20 are transported to the stratosphere, where it is destroyed by photolysis and by reaction with excited oxygen atoms to produce NO,
3 34
J.P. Burrows
Figure 4. Excess tropospheric columns of ozone (upper panel), nitrogen dioxide (middle panel) and formaldehyae (lower panel), derived from GOMF data. Source: A. Ladst~tter-WeiBenma)'er and J.P. Burrows, 1998, personal communication.
Current and future passive remote sensing techniques used to determine atmospheric constituents
335
Figure 5. Global total column density of nitrogen dioxide derived from GOME data. The white colours indicate areas where no measurements were made. Source: A. Richter, L. Hild and J. P. Burrows, 1998, personal communication. which in turn reacts with 03 to produce NO2. The second source of stratospheric oxides of nitrogen is downward transport from the mesosphere. The stratospheric column of NO2 has a strong seasonal cycle (see for example Eisinger et al., 1996; Richter et al., 1998b,c; Wittrock et al., 1998). In the polar vortex in winter and spring, NO2 amounts are low because of the enhanced formation rate of N205 at the low temperatures and its subsequent heterogeneous removal to form aerosols containing nitric acid and polar stratospheric clouds. In summer the polar latitudes have high stratospheric NO2 values. This is because of the thermal instability of N205 and the increased daytime photolysis of NO2 precursors such as N205 and C1ONO2. In remote and unpolluted regions of the planetary boundary layer, natural sources of NO• (NO and NO2) such as lightning, result in relatively small mixing ratios, typically being less than 20 pptv. In contrast, the amount of NO• in downtown city air is often above 100 ppbv. Thus NO2 has large tropospheric variability. Both the distribution of sources and the lifetime of NO2 are very different in the troposphere compared to the stratosphere. Figure 5 presents a composite picture of the total column density of NO2 retrieved from GOME observations on 15, 16 and 17 September 1997. The total column is clearly showing the differences in both stratospheric and tropospheric NO2 patterns. The low stratospheric amounts of NO2 in the polar vortex above Antarctica and the high values above the Arctic are readily observable. Elevated tropospheric NO2 can be seen over Europe, the United States, the Middle East oil fields and other industrial regions. In addition, the production of NO2 from biomass burning in the southern hemisphere and its transport around the globe is clearly visible. The retrieval of the tropospheric c~'!umn of NO2 requires the subtraction of the stratospheric column. This can be achieved by assuming local longitudinal homogeneity of the stratosphere. Figure 4 shows the excess tropospheric NO2 over Indonesia during the fires of September 1997 observed using GOME data (Burrows et al., 1998b).
336
J.P. Burrows
5.3. Formaldehyde Similar to several other trace species, formaldehyde (HCHO) occurs in elevated concentrations as a result of pollution in the lower troposphere. Formaldehyde is generated during the oxidation of hydrocabons in copious amounts. However, as it is photolyzed and reacts with OH, it has a relatively short tropospheric lifetime. The observation of formaldehyde from GOME over industrial and biomass burning regions has been discussed by Perner et al. (1998) and Burrows et al. (1998b). Thomas et al. (1998) have also observed formaldehyde over Borneo using GOME data. An example of the plume of excess tropopsheric HCHO from Indonesian fires in 1997 is shown in Figure 4.
5.4. Sulphur dioxide Sulphur dioxide is both emitted into thetroposphere and also formed during oxidation of dimethyl sulphide (DMS) and other sulphur containing species produced in the biosphere. Important sources of atmospheric SO2 are volcanoes. However, the major single global source is probably the combustion of sulphur-containing fossil fuels. In the stratosphere there are two important sources of sulphur dioxide, i.e. injection by volcanic eruptions and oxidation of carbonyl sulphide (COS), which is transported from the troposphere. SO2 has a short lifetime in the troposphere where it is oxidized in both the gas and liquid phases to form H2SO4. Its gas phase oxidation by OH leads to the formation of cloud condensation nuclei. In rain and aerosols it is oxidized by H202 to sulphuric acid. In the dry lower stratosphere the lifetime of SO2 is expected to be longer compared to that in the troposphere.
Figure 6. Total column density of sulphur dioxide over Eastern Europe averaged over the period 15-29 February 1996 derived from GOME data. D.U., Dobson unit; 1 D.U. = 2.69 • 1016moleculesper cm2. Source: M. Eisinger and J.P. Burrows, 1998,personal communication.
Current and future passive remote sensing techniques used to determine atmospheric constituents
337
SO2 was first retrieved from space-based measurements using TOMS data (Krueger, 1983; Krueger et al., 1995; 1996). Successful observations have been made on volcanic eruptions such as E1 Chichon and Pinatubo which injected SO2 not only into the stratosphere, but also into the troposphere. Recently GOME measurements have been used to observe tropospheric SO2 from both volcanic plumes and fossil fuel burning (Eisinger et al., 1998; Eisinger and Burrows, 1998a,b). Figure 6 shows the retrievals of the total column of SO2 observed over Eastern Europe in the first half of February 1996.
5.5. Halogen oxides GOME is the first space-borne remote sensing instrument, whose measurements are suitable for halogen oxides in the lower atmosphere. Its primary halogen targets are BrO and OC10, and under specific circumstances C10 may be retrievable. This is very complementary to the microwave instruments used to measure stratospheric C10. As these halogen radicals play an important role in the catalytic removal of stratospheric 03, one of the original aims of GOME was to measure their stratospheric abundance. While OC10 has only been observed by GOME mainly in the stratospheric polar vortex, BrO has been observed globally from GOME observations (Eisinger et al., 1996; Hegels et al., 1998) as large clouds of BrO occur in the troposphere both in the southern hemisphere (T. Wagner and U. Platt, 1998, personal communication) and in the northern hemisphere (Richter et al., 1998a). Examples of observations of northern hemispheric clouds of BrO are presented in Figure 7, showing a polar projection of the average total column amount of BrO in the months of March and April 1997. The large BrO cloud observed by GOME does not correlate with the stratospheric polar vortex or other dynamical behaviour. This cloud is attributed to tropospheric production of BrO in the boundary layer. The BrO above the Hudson Bay is clearly visible in late winter and early spring. The region of BrO production appears to move northwards from spring to summer but in late summer it is no longer visible (Richter et al. 1998a). The mechanism for the production of BrO is not yet well explained but is probably a natural process. The BrO is considered to be in large part responsible for the spring time low ozone episodes first reported some 10 years ago (Barrie et al., 1988, and references therein).
6. Planned future tropospheric measurements The need to study the change of atmospheric composition over long time intervals requires the continuity of measurements. This argues strongly for long-duration missions, which make the same well-calibrated measurements over many years. Passive remote sensing experiments utilizing absorption spectroscopy such as TOMS, SBUV, SAGE, GOME and SCIAMACHY (Table 2) are well suited for this task. In the next few years a number of important atmospheric remote sensing missions are planned by NASDA, NASA and ESA. NASDA has constructed a second ADEOS satellite which will have POLDER and ILAS-II (Improved Limb Atmospheric Sounder) on board. The NASA-AM-1 platform will have the MOPITT instrument (Measurement of Pollution in the Troposphere), which aims to measure vertical profiles of carbon monoxide and methane in the troposphere. MOPITT is a Canadian instrument supported by an international science team. NASA-AM is due for launch in the middle of 1999.
33 8
J.P. Burrows
Figure 7. Polar projection of the GOME retrieval of tropospheric BrO in March and April 1997. Source: A. Richter and J. P. Burrows, 1997, personal communication.
Current and future passive remote sensing techniques used to determine atmospheric constituents
339
The ESA-ENVISAT is planned for launch in 2000. It has a large payload, including SCIAMACHY, MIPAS (Michelson Interferometer for passive atmospheric Sounding), GOMOS (Global Ozone Monitoring by Occultation of Stars) and MERIS (MEdium Resolution Imaging Spectrometer). MIPAS, MERIS and GOMOS are ESA-developed instruments, whereas SCIAMACHY is a joint contribution of Germany, the Netherlands and Belgium to ENVISAT. GOMOS uses UV visible spectroscopy primarily to measure 03, NO2 and NO3 in the stratosphere. As indicated by its name, it uses the stellar occultation technique and makes measurements mainly at night. GOMOS is similar to SAGE or HALOE, but it uses a variety of stars as light sources instead of the sun. This enables many more star rise and star sets to be observed, when observing from a sun synchronous platform such as ENVISAT. Daytime measurements by GOMOS are influenced by solar radiation scattered from the atmosphere. MIPAS observes the infrared atmospheric emission both by day and night. GOMOS and MIPAS will both measure in the upper troposphere. MERIS observes during the day in a selected set of visible and near-infrared channels at a relatively high spatial resolution. MERIS data will also be used to retrieve albedo, aerosols and water vapor. The SCIAMACHY utilizes near simultaneous limb and nadir measurements of the scattered light in the atmosphere between 240 and 2400 nm to determine the amounts and distributions of tropospheric constituents. The target species and parameters are 03, NOR, N20, BrO~ CO, H20, SO2, CO, CO2, CH4, aerosols temperature and pressure. For the longlived gases such as N20, CH4 and CO2 the scientific objective is to measure the small gradients, which define source and sink regions. The NASA EOS-CHEM mission to be flown in 2003 will include the instruments TES (Tropospheric Emission Spectrometer, a nadir sounding Michelsen interferometer) for measurement of a variety of tropospheric gases such as 03, HNO3, and NO, and the OMI (Ozone Monitoring Instrument, a Dutch contribution to EOS-CHEM) with objectives similar to those of GOME.
7. Conclusions In the last 25 years remote sensing of atmospheric constituents has established itself as an important research field. Global remote sensing observations are essential to understand the natural processes which determine the global behaviour of the atmosphere and to assess the impact of human activity on the atmosphere. In addition, remote sensing of the atmosphere provides data needed to assess the impact of international agreements designed to limit the environmental impact of industrial activity. Following the great success in developing an adequate measurement strategy for stratospheric constituents, the challenge is to accurately measure the tropospheric composition. This is technically much more difficult than measurements in the upper atmosphere. The use of assimilation techniques to maximize the information retrieved from tropospheric sounding is foreseen as essential. A number of experiments have been or will be developed which are designed to determine the concentration of constituents in the troposphere. Well validated global data about the distributions of atmospheric constituents, obtained from remote sensing instrumentation, is essential for testing our knowledge and understanding of the atmosphere. High vertical, and horizontal spatial resolution as well as an appropriate temporal resolution is required. The parameterizations used in chemical and transport models (CTMs) of the atmosphere can only be validated by using such global data. One of the most critical parameterizations for CTMs in this respect is the fluxes of species into or out of
340
J.P. Burrows
atmosphere from the earth's surface. As the new generation of advanced data products become available, it is to be expected that the current generation of CTMs will be improved dramatically. This is a challenging task for the years ahead.
Acknowledgements This study has in part been supported by the University and State of Bremen. I would like to thank my close scientific colleagues P.J. Crutzen and D. Perner, without whom the GOME and SCIAMACHY projects would not have been started or realized. I would like to express my gratitude to all the scientists and engineers, who have worked on GOME and SCIAMACHY, in particular the ESA scientists and engineers, who supported and participated in the development of the GOME instrument (Drs C.J. Readings, Dr. P. Dubock Dr. A. Hahne, Dr. J. Callies) as well as the German, Dutch and Belgian governments and ESA for supporting the development of SCIAMACHY. I take this opportunity to express my thanks to all the scientists in my research group at the University of Bremen, who have generated many of the figures shown and provided much stimulating scientific discussion. Finally I would like to dedicate this study to my late father-in-law, K. Holtkotte, who died during the writing of this manuscript.
References Arrhenius, S. (1896) On the influence of carbonic acid in the air upon the temperature of the ground. Philosophical Magazine and Journal of Science. 41:237. Baker, R. L., L.E. Maudlin III and J.M. Russell III (1986) Design and performance of the halogen occultation experiment (HALOE) remote sensor. Society of Photo-OpticaL Instrumentation and Engineering Proceedings 685:181-191. Barnett, J.J., P.E. Morris, T.J. Nightingale, C.W.P. Palmer, G.D. Peskett, C.D. Rodgers, F.W. Taylor, P. Venters, R.J. Wells, J.G. Whitney, J. Ballard and J. Knight (1992) The Improved Stratospheric and Mesospheric Sounder on the Upper Atmosphere Research Satellite. Society of Photo-Optical Instrumentation and Engineering Proceedings 1715:527. Barrie, L.A., J.W. Bottenheim, R.C. Snell, P.J. Crutzen and R.A. Rasmussen (1988) Ozone destruction and photochemical reactions at polar sunrise in the lower Arctic atmosphere. Nature 334:138-140. Bhartia, P.K., R.D. McPeters, C.L. Mateer, L.E. Flynn and C. Wellemeyer (1996) Algorithm for the estimation of vertical ozone profiles fr~-.m the backscattered ultraviolet technique. Jo:trnal of Geophysical Research 101:18793-18806. Bone N. (1991) The Aurora: sun-earth interactions. Ellis Horwood, New York, London. Bovesmann, H., J.P. Burrows, M. Buchwitz, J. Frerick, S. NoEl, V.V. Rozanov, K.V. Chance, A.P.H. Goede (1998) SCIAMACHY- Mission Objectives and Measurement Modes. Journal of Atmospheric Science (in press). Bramstedt, K., K.-U. Eichmann, M. Weber, V.V. Rozanov, R. Hoogen, R. de Beek, M. Buchwitz, T. Kurosu and J.P. Burrows (1998) Ozone Profiles from the GOME Satellite Data- Part II: First Results from the Arctic Winter Campaign Proceedings of the 4th European Symposium on Stratopsheric Ozone, Schliersee, 22-26 September 1997, pp. 189-192. Brasunas, J.C., V.G. Kunde and L.W. Herath (1988) Cryogenic fourier spectrometer for measuring trace species in the lower stratosphere. Applied Optics 27:4964-4976. Brekke, A. and A. Egland (1994) The Northern Lights: their heritage and science. Grondahl og Dreyers Forlag AS, Oslo.
Current andfuture passive remote sensing techniques used to determine atmospheric constituents
341
Burrows, J.P., W. Schneider and K. V. Chance (1991) GOME and SCIAMACHY: Remote Sensing of Stratospheric and Tropospheric Trace Gases. In: J.A. Pyle and N.R.P. Harris (Eds.) Polar stratospheric ozone. Proceedings of the 1st European Workshop, 3-5 October 1990, Schliersee, Germany. CEC Air Pollution Research Report 34, CEC, Brussels, pp. 99-102. Burrows, K., V. Chance, A.P.H. Goede, R. Guzzi, B. J. Kerridge, C. Muller, D. Perner, U. Platt, J.-P. Pommereau, W. Schneider, R. J. Spurr and H. van der Woerd (1993) Global Ozone Monitoring Experiment. In: T. D. Guyenne and C. J. Readings (Eds.) Interim Science Report ESA SP-1151, European Space Agency, Paris. Burrows, J.P., D. Diebel, B. J. Kerridge, R. Munro, D. Pemberton, U. Platt, H. Frank (1994) A study of the methods for the retrieval of atmospheric constituents, Final Report ESA Contract 9687/91/NL/BI. Burrows, J.P., E. H61zle, A.P.H. Goede, H. Visser and W. Fricke (1995) SCIAMACHY - Scanning Imaging Absorption Spectrometer for Atmospheric Chartography. Acta Astronautica 35:445-451. Burrows, J.P., M. Eisinger, A. Richter, V. Rozanov, D. Perner and E. Hegels (1998a) Global Ozone Monitoring Experiment (GOME): Measurements of OCIO over Antarctica 1995. In: R.D. Bojkov and G. Visconti (Eds.) Atmospheric ozone. Proceedings of the XVIII Quadrennial Atmospheric Ozone Symposium 1996, L'Aquila, Italy, 12-21 September 1996, pp. 661-664. Burrows, J.P., W. Schneider, J. C. Geary, K. V. Chance, A. P. H. Goede, H. J. M. Aarts, J. de Vries, C. Smorenburg and H. Visser (1990) Atmospheric remote sensing with SCIAMACHY. Digest of topical meeting on optical remote sensing of the atmosphere 1990 No. 4, Optical Society of America, Washington, D.C., pp. 71-75. Burrows, J.P., M. Weber, M. Buchwitz, V. V. Rozanov, A. Ladst~itter-WeiBenmayer, A. Richter, R. De Beek, R. Hoogen, K. Bramstedt and K.U. Eichmann (1998b) The Global ozone monitoring experiment (GOME): Mission concept and first scientific results. Journal of Atmospheric Science (in press). Carli, B., F. Mencaraglia and A. Bonetti (1984) Submillimeter high resolution FT spectrometer for atmospheric studies. Applied Optics 23:2594-2603. Chance, K.V., J.P. Burrows and W. Schneider (1991) Retrieval and molecule sensitivity studies for the global ozone monitoring experiment and the scanning imaging absorption spectrometer for atmospheric chartography. In: J.L. McElroy and R.J. McNeal (Eds.) Remote Sensing of Atmospheric Chemistry. Society of Photo-Optical Instrumentation and Engineering Proceedings 1491:151-168. Chance, K.V., D.G. Johnson and W.A. Traub (1989) Measurement of stratospheric HOCI: concentration profiles, including diurnal variation. Journal of Geophysical Research 94:1105911067. Chance, K., J.P. Burrows, D. Perner and W. Schneider (1997) Satellite measurements of atmospheric ozone profiles, including tropospheric ozone from UV/visible measurements in the nadir geometry: a potential method to retrieve tropospheric ozone. Journal Quantitative Spectroscopy and Radiative Transfer 57:467-476. Chu, W. and M.P. McCormick (1979) Inversion of stratospheric aerosol and gaseous constituents from space craft solar extinction data in the 0.38-1.0 micron region. Applied Optics 18:1404-1414. Coffey, M.T., W.G. Mankin and A Golden (1981) Simultaneous spectroscopic determination of the latitudinal, seasonal, and diurnal variability of N20, NO, NO2 and H N O 3. Journal Geophysical Research 86:7331-7341. Connors, V.S., D.R. Cahoon, H.G. Reichle Jr. and H.E. Scheel (1991) Comparison between carbon monoxide measurements from spaceborne and airborne platforms. Canadian Journal of Physics 69:1128-1137. Crewell, S., K. Ktinzi, H. Nett, T. Wehr and P. Hartogh (1993) Aircraft measurements of CIO and HCI during EASOE 1991/92. Geophysical Research Letters 21:1267-1270. Crutzen, P.J. (1973) A discussion of the chemistry of some minor constituents in the stratosphere and the troposphere. Pure and Applied Geophysics 106-108:1385-1399. Dameris, M., V. Crewe, R. Hein, C. Schnat, C. Brtihl and B. Steil (1998) Assessment of the future development of the ozone layer. Geophysical Research Letters 25:3579-3583.
342
J.P. Burrows
Dave, J.V. and C.L. Mateer (1967) A preliminary study of the possibility of estimating total atmospheric ozone from satellite measurements. Journal of Atmospheric Science 24:414-427. De Beck, R., R. Hoogen, V. Rozanov and J.P. Burrows (1998) Ozone profile from GOME satellite data I: algorithm development. Proceedings of the third ERS Symposium Florence 18th-23rd of March 1997. Report ESA SP 414, pp. 749-754. Deschamps, P.Y., M. Herman, A. Podaire, M. Leroy, M. Laporte, P. Vermande (1990) A spatial instrument for the observation of polarization and directionality of earth reflectances, IGARSS 90 Conference Proceedings, IGARSS IEEE, Washington, D.C. DeZafra, R.L., M. Jaramillo, A. Parrish, P. Solomon, R. Connor and J. Barret (1987) High concentrations of chlorine monoxide at low altitudes in the Antarctic spring stratosphere: diurnal variation. Nature 328:408-411. Dlugokencky, E.J., K.A. Masarie, P.M. Lang and P.P. Tans (1998) Continuing decline in the growth of the atmospheric methane burden. Nature 393:447-450 Eichmann, K.-U., K. Bramstedt, M. Weber, V. Rozanov, R. de Beck, R. Hoogen and J.P. Burrows (1998) Ozone profile from GOME satellite data II: Validation and application. Proceedings of the third ERS Symposium Florence 18th-23'd of March 1997. Report ESA SP 414, pp. 755-758. Eisinger, M. and J.P. Burrows (1998a) Observations of SO2 from volcanic plumes. Earth Observation Quarterly 58:16-18. Eisinger, M. and J.P. Burrows (1998b) Tropospheric sulphur dioxide observed by the ERS-2 GOME instrument. Geophysical Research Letters (in press). Eisinger, E., J.P. Burrows and A. Richter (1996) Studies on the precision of GOME irradiance and radiance products and GOME measurements of OCIO and BrO over Antarctica. Report ESA WPP108, pp. 93-109. Eisinger, M., J.P. Burrows, A. Richter and A. Ladst~itter-WeiBenmayer (1998) SO2, OCIO, BrO and other minor trace gases from the Global Ozone Monitoring Experiment (GOME). Space at the service of our Environment. Proceedings of the third ERS Symposium Florence 18-23 March 1997. Report ESA SP 414, pp. 647-652. Eisinger, M., A. Richter, A. Ladst~itter-WeiBenmayer and J.P. Burrows (1997) DOAS Zenith Sky Observations: 1 BrO Measurements over Bremen (53~ 1993-1994. Journal of Atmospheric Chemistry 26:93-108. Farman, J.C., B.G. Gardner and J.D. Franklin (1985) Large losses of total ozone in Antarctica reveal seasonal CIOx and NOx interaction. Nature 315:207-210 Farmer, C.B., O.F. Raper and F.G. O'Callaghan (1987) Final report on the first flight of the ATMOS instrument during the Spacelab 3 mission, April 29-May 6 1985. Publication 87-32, Jet Propulsion Laboratory, Passedena, California, 45 pp. Finlayson-Pitts, B.J. and J.N. Pitts Jr. (1986) Atmospheric chemistry." Fundamentals and experimental techniques. Wiley and Sons, Chichester, New York. Fischer, H., F. Fergg and D. Rabus (1980) Radiometric measurements of stratospheric H20, HNO3 and NO2 profiles. Proceedings of the International Radiation Symposium, Fort Collins, August 1980. Fishman, J., C.E. Watson, J.C. Larsen and J.A. Logan (1990) Distribution of tropospheric ozone determined from satellite data. Journal of Geophysical Research 95:3599-3617. Fishman, J., K. Fakhruzzaman, B. Cros and D. Nganga (1991) Identification of widespread pollution in the southern hemisphere deduced from satellite analyses. Science 252:1693-1696. Fishman, J., V.G. Brackett and K. Fakhruzzaman (1992) Distribution of tropospheric ozone in the tropics from satellite and ozonsonde measurements. Journal of Atmospheric and Terrestrial Physics 54:589-597. Fishman, J., V.G. Bracket E.V. Browell and W.B. Grant (1996) Tropospheric ozone derived from TOMS/SBUV measurements during TRACA-A. Journal of Geophysical Research 101:2406924082. Frederick, J.E., R.P. Cebula, and D.F. Heath (1986) Instrument characterization for the detection of long term changes in stratospheric ozone: an analysis of the SBUV/2 radiometer. Journal of Atmospheric and Oceanic Technology 3:472-480.
Current andfuture passive remote sensing techniq,,es used to determine atmospheric constituents
343
Gille, J.C. and J.M. Russell III (1984) The limb Infrared Monitor of the stratosphere: experimental description, performance and results. Journal of Geophysical Research 89:5125-.5140. Gille, J.C., P.L. Bailey, R.A. Craig, F.B. House and G.P. Anderson (1980) Sounding the stratosphere and mesosphere by infrared limb scanning from space. Science 208:397. Gleason, J.F., N. Christina Hsu and O. Torres (1998) Biomass burning smoke measured using backscattered ultraviolet radiation: SCAR-B and Brazilian smoke interannual variability. Journal of Geophysical Research (in press). Gunson, M.R., M.M. Abbas, M.C. Abrams, M. Allen, L.R. Brown, T.L. Brown, A.Y. Chang, A. Goldman, F. W. Irion, L.L. Lowes, E. Mahieu, G. L. Manney, H.A. Michelsen, M.J. Newchurch, C. P. Rinsland, R.J. Salawitch, G.P. Stiller, G.C. Toon, Y.L. Yung and R. Zander (1996) The Atmospheric Trace Molecule Spectroscopy (ATMOS) experiment: Deployment on the ATLAS space shuttle missions. Geophysical Research Letters 23:2333-2336. Guzzi R., E. Cattani, M. Cervino, C. Levoni, F. Torricella, J.P. Burrows and T. Kurosu (1998) GOME cloud and aerosol data products algorithms development. Final Report ESA Contract 11572/95/NL/CN. Guzzi, R., M. Cervino, F. Torricella, J.P. Burrows, T: Kurosu, V. Rozanov, K. Chance and P. Watts (1996) Final Report- A study of cloud detection. ESA Contract 10997/94/NL/CN. Haagen-Smit, A.J. (1952) Chemistry and physiology of Los Angeles smog. Industrial Engineering and Chemistry. 44:1342. Haagen-Smit, A.J., E.F. Darley, M. Zailin, H. Hull and W. Noble (1952) Investigation on injury to plants from air pollution in the Los Angeles area. Plant Physiology 27:18. Hartmann, G.K., R.M. Bevilacqua, P.R. Schwartz, N. K/impfer, K.F. Ktinzi, A. Berg, W. Boogaerts, B.J. Connor, C.L. Croskey, M. Daehler, W. Degenhardt, H.D. Dicken, D. Goldizen, D. Kriebel, J. Langen, A. Loidl, J.J. Olivero, T. A. Pauls, S.E. Puliafito, M.L. Richards, C. Rudin, J.J. Tsou, W.B. Waltmann, G. Umlauft and R. Zwick (1996) Measurements of 03, H20, and CIO in the middle atmosphere using the millimeter wave atmospheric sounder (MAS). Geophysical Research Letters 23:2313-2317. Heath, D.F., C.L. Mateer and A.J. Kruger (1973) The Nimbus-4 Backscatter Ultraviolet (BUV) atmospheric ozone experiment - two years' operation. Pure and Applied Geophysics 106108:1238-1253. Heath, D.F., A.J. Krueger, H.A. Roeder and P D. Henderson (1975) The solar baekscatter ultraviolet and total ozone mapping spectrometer (SBUV/TOMS) for Nimbus-G. Optical Engineering 14:323-332. Hegels, E., D. Perner, P.J. Crutzen, J.P. Burrows, A. Ladst/~tter-WeiBenmayer, M. Eisinger, J. Callies and W. Balzer (1998) Global Ozone Monitoring Experiment: first BrO measurement. In: R. D. Bojkov and G. Visconti (Eds.) Atmospheric Ozone. Proceedings of the XVIII Quadrennial Atmospheric Ozone Symposium 1996, L'Aquila, Italy 12-21 September 1996, pp. 293-296. Heij, G.J. and J.W. Erisman (1995) (Eds.)Acid rain research: Do we have enough answers ?. Studies in Environmental Science 64. Elsevier, Amsterdam, 502 pp. Hilsenrath, E., D. Williams and J. Frederick (1988) Calibration of long term data sets from operational satellites using the space shuttle. Society of Photo-Optical Instrumentation and Engineering Proceedings 924:215-222. Hilsenrath, E., P.A. Newman, R.B. Cebula, P.W. DeCamp, T.J. Kelly and L. Coy (1996) Ozone change from 1992 to 1993 as observed from SSBUV on the ATLAS-1 and ATLAS-2 missions. Geophysical Research Letters 23:2305-2308. Hilsenrath, E., R.P. Cebula, M.T. Deland, K. Laamann, S. Taylor, C. Wellemeyer and P.K. Bhartia (1995) Calibration of the NOAA-11 solar backscatter ultraviolet (SBUV/2) ozone data set from 1989 to 1993 using in-flight calibration aata and SSBUV. Journal of Geophysical Research 100:1351-1366. Hoogen, R., V.V. Rozanov and J.P. Burrows (1999) Ozone profiles from GOME satellite data: algorithm description and first validation. Journal of Geophysical Research (in press).
344
J.P. Burrows
Hoogen, R., V. Rozanov, R. de Beek, K. Bramstedt, K.-U. Eichmann, M. Weber, T. Kurosu, and J.P. Burrows (1998) Ozone Profiles from GOME satellite Part 1: Advances in retrieval algorithm development. Proceedings of the 4th European Symposium on Stratospheric Ozone, Schliersee 2226 September 1997, pp. 677-680. Houghton, J.T., L.G. Meira-Filho, J. Bruce, Hoesung Lee, B.A. Callander, E. Haites, N. Harris and K. Maskell (Eds.) (1995) Climate Change 1994. Radiative forcing of climate change and an evaluation of the IPCC IS92 emission scenarios. Cambridge University Press, Cambridge, 339 pp. Hsu, N.C., J.R. Herman, P.K. Bhartia, C.J. Seftor, O. Torres, A.M. Thompson, J.F. Gleason, T.F. Eck and B.N. Holben (1996) Detection of biomass burning smoke from TOMS measurements. Geophysical Research Letters 23: 745-748. Hudson, R.D. and A.M. Thompson (1998) Tropical ozone from total ozone mapping spectrometer by a modified residual method. Journal of Geophysical Research 103:22129-22146. Hudson, R.D., J. Kim and A.M. Thompson (1995) On the derivation of tropospheric column ozone from radiance measured by the total ozone mapping spectrometer. Journal of Geophysical Research 100:11138-11145. Ignatov, A.M., L.L. Stiwe, S.M. Sakerin, G.K. Korotaev (1995) Validation of the NOAA/NESDIS satellite aerosol product over the North Atlantic in 1989. Journal of Geophysical Research 100:5123-5132. Iozenas, V.A. et al. (1969a) Studies of the earth's ozonosphere from satellites. Izvestiya Akademii Nauk SSSR Fizika Atmosfery I Okeana 5: 77-82. Iozenas, V.A. et al. (1969b) An investigation of the planetary ozone distribution from satellite measurements of ultraviolet spectra. Izvestiya Akademii Nauk SSSR Fizika Atmosfery I Okeana 5:219-233. Kim, J.-H., R.D. Hudson and A.M. Thompson (1996) A new method for deriving time averaged tropospheric column ozone over the tropics using total ozone monitoring spectrometer (TOMS) radiances: intercomparison and analysis using TRACE A data. Journal of Geophysical Research 101:24317-24330. Krasnopol'skiy, V.A. (1966) The ultraviolet spectrum of solar radiation reflected by the terrestrial atmosphere and its use in determining the total content and vertical distribution of atmospheric ozone. Russian Geomagnetism and Aeronomy 6:236-242. Krueger, A J. (1983) Sighting of El Chichon sulfur dioxide with the Nimbus 7 total ozone mapping spectrometer. Science 220:1377-1378. Krueger, A.J., C.C. Schnetzler, L.S. Walter (1996) The December 1981 eruption of Nyamuragira Volcano (Zaire) and the origin of the "mystery cloud" of early 1982. Journal of Geophysical Research 101:15191-15196. Krueger, A.J., L.S. Walter, P.K. Bhartia, C.C. Schnetzler, N.A. Krotkov, I. Sprod and G.J.S. Bluth (1995) Volcanic sulfur dioxide measurements from the total ozone mapping spectrometer (TOMS) instruments. Journal of Geophysical Research 100:14057-14076. Kunde, V.G., J.C. Brasunas, W.C. Maguire, J.R. Herman, S.T. Massie, M.M. Abbas, L. Herath and W.A. Shaffer (1988) Measurement of nighttime stratospheric N205 from infrared emission spectra. Geophysical Research Letters 15:1177-1180. Levine, J. (Editor) (1991) Global biomass burning." atmospheric, climatic and biospheric implications. MIT Press, Cambridge, Mass. Lovelock, J.E. (1979) Gaia: A new look at life on earth. Oxford University Press, Oxford. Lovelock, J.E., R.J. Maggs and R.J. Wade (1973) Halogenated hydrocarbons in and over the Atlantic. Nature 241:194-196 Mateer, C.L., D.F. Heath and A.J. Kruger (1971) Estimation of total ozone from satellite measurements of backscattered Earth radiance. Journal of Atmospheric Science 24:414-420. Maudlin III, L.E., N.H. Zaun, M.P. McCormick, Jr., J.H. Guy, and W.R. Vaughn (1985) Stratospheric aerosol and gas experiment II instrument: a functional description. Optical Engineering 24:307312.
Current andfuture passive remote sensing techniques used to determine atmospheric constituents
345
McCormick, M.P., P. Hamill, T.J. Pepin, W.P. Chu, T.J. Swissler and L.R. McMaster (1979) Satellite studies of the stratospheric aerosol. Bulletin of the American Meteorological Society 60:10381048. Molina, M.J. and F.S. Rowland (1974) Stratospheric sink for chlorofluormethanes: chlorine atom catalyzed destruction of ozone. Nature 249:810-814. Mount, G.M., R.W. Sanders, A.L. Schmetekopf and S. Solomon (1987) Visible and near ultraviolet spectroscopy at McMurdo station Antarctica 1: Overview and daily variations of NO2 and 03 during austral spring 1986. Journal of Geophysical Research 92:8320-8340. Mount, G.M., D.W. Rusch, J.F. Noxon, J.M. Zawodny and C. Barth (1984) Measurement of stratospheric NO2 from the solar mesospheric explorer satellite: 1. An overview of the results. Journal of Geophysical Research 89:1327-1339. Munro, R., B.J. Kerridge, K.V. Chance and J.P. Burrows (1993) Ozone Profile Retrieval Information from the ESA GOME Instrument. Proceedings of the Quadrennial Ozone Symposium, June 4-13 1992, Charlottesville, Virginia, USA. Munro, R., R. Siddams, W.J. Reburn and B.J. Kerridge (1998) Direct measurement of tropospheric ozone distributions from space. Nature 392:168-171. Murcray, D., D.B. Barker, J.N. Brooks, A. Goldman and G.W.J. Williams (1975) Seasonal and latitudinal variation of the stratospheric concentration of HNO3. Geophysical Research Letters 2:223-225. Murcray, D.G., A. Goldman, F.H. Murcray, F.J. Murcray and W.J. Williams (i979) Stratospheric distribution of C1ONO2. Geophysical Research Letters 6:857-859. Perner, D., T. Kltipfel, E. Hegels, P.J. Crutzen and J.P. Burrows (1998) First results on tropospheric observations by GOME on~ ERS-2. Proceedings of the third ERS Symposium Florence 18-23 March 1997, Space at the service of our Environment. Report ESA SP 414, pp. 647-652. Platt, U. and D. Perner (1980) Direct measurements of atmospheric HCHO, HONO, 03, NO2, and SO2, by differential optical absorption spectroscopy in the near UV. Journal of Geophysical Research 85:7453-7465. Pougatchev, N.S., N.B. Jones, B.J. Connor, C.P. Rinsland, E. Becker, M.T. Coffey, V.S. Connors, P. Demoulin, A.V. Dzhola, H. Fast, E.I. Grechko, J.W. Hannigan, M. Koike, Y. Kondo, E. Mahieu, W.G. Mankin, R. L. Mittermeier, J. Notholt, H. G. Richle Jr., B. Sen, L. P. Steele, G.C. Toon, L. N. Yurganov, R. Zander and Y. Zhao (1998) Ground based infrared solar spectroscopic measurements of carbon monoxide during the 1994 MAPS flights. Journal of Geophysical Research 103:19317-19325. Rawcliffe, R.D. and D.D. Elliot (1966) Latitude distributions of ozone at high altitudes deduced from satellite measurement of the earth's radiance at 2840 A. Journal of Geophysical _Research 71:5077 5089. Reichle Jr., H.G., V.S. Connors, J.A. Holland, W.D. Hypes, H.A. Wallio, J.C. Casas, B.B. Gormsen, M.S. Saylor and W.D. Hesketh (1986) Middle and upper tropospheric carbon monoxide mixing ratios as measured by a satellite-borne remote sensor during November 1981. Journal of Geophysical Research 91:10865-10887 Reichle Jr., H.G., V.S. Connors, J.A. Holland, R.T. Sherrill, H.A. Wallio, J.C. Casas, E.P. Condon, B.B. Gormsen and W. Seiler (1990) The distribution of middle tropospheric carbon monoxide during early October 1984. Journal of Geophysical Research 95:9845 9856. Richter, A., F. Wittrock, M. Eisinger and J.P. Burrows (1998a) GOME observations of tropospheric BrO in the Northern Hemispheric Spring and Summer 1997. Geophysical Research Letters 25:2683-2686. Richter, A., M. Eisinger, A. Ladst~itter-WeiBenmayer and J.P. Burrows (1998b) DOAS Zenith Sky Observations: 2 Seasonal variation of BrO over Bremen (53~ 1994-1995. Journal of Atmospheric Chemistry (in press). Richter, A, M. Eisinger, A. Ladst~itter-WeiBenmayer, F. Wittrock and J.P. Burrows (1998c) Ground based UV/Vis measurements of 03, NO2 and BrO over Bremen (53N 9E). In: R.D. Bojkov and G. Visconti (Eds.) Atmospheric ozone. Proceedings of the XVIII Quadrennial Atmospheric Ozone Symposium, L'Aquila, Italy 12-21 September 1996, pp. 591-594.
346
J.P. Burrows
Roach, F.E. and J.L. Gordon (1973) The light of the Night Sky. Geophysics and Astrophysics Monographs, D. Reidel, Dordrecht. Roche, A.E. and J.B. Kumer (1989) Cryogenic Limb Array Etalon Spectrometer (CLAES). Society of Photo-Optical Instrumentation and Engineering Proceedings 973:324-335. Roche, A.E, P.B Forney, J.B. Kumer, L.G. Naes and T.C. Nast (1982) Performance analysis for the cryogenic etalon spectrometer on the upper atmospheric research satellite. Society of PhotoOptical Instrumentation and Engineering Proceedings 354:46-54. Rozanov, V.V., Yu. M. Timofeyev, A.V. Polyakov, Burrows J.P. and K.V. Chance (1992) On the possibilities of precision measurements of 03 and NO2 from space by solar radiation absorption spectra. Izvestiya Akademii Nauk SSSR Fizika Atmosfery I Okeana 28:500-505. Rozanov, V.V., Yu. M. Timofeyev, M.S. Biryulina, J.P. Burrows, R.J.D. Spurr and D. Diebel (1993) Accuracy of atmospheric constituent retrieval from multichannel remote sensing instruments. Proceedings of the International Radiation Symposium 1992, pp. 394-397. Rozanov, V.V., T. Kurosu and J.P. Burrows (1998) Retrieval of atmospheric constituents in the UVvisible: a new analytical approach for the calculation of weighting functions. Journal of Quantitative Spectroscopy and Radiative Transfer, in press. Rusch, D.W., G.H,. Mount, C.A. Barth, R.J. Thomas and M.T. Callan (1984) Solar mesospheric explorer ultraviolet spectrometer: measurements of ozone in the 1.0-0.1 mbar region. Journal of Geophysical Research 84:11677-11687. Shindell, D.T., D. Rind and P. Lonergan (1998) Increased polar stratospheric ozone losses and delayed eventual recovery owing to increasing greenhouse gas concentrations. Nature 392:589592. Singer, S.F. and R.C. Wentworth (1957) A method for the determination of the vertical ozone distribution of ozone distribution from a satellite. Journal of Geophysical Research 62:299-308. Stachnik, R.A., J.C. Hardy, J.A. Torsala, J.W. Waters and N.R. Erickson (1992) Sub-millimeter heterodyne measurements of stratospheric CIO, HCI, 03 and HCI, First results. Geophysical Research Letters 19:1931-1934. Thomas, R.J., C.A. Barth, D.W. Rusch and R.W. Sanders (1984) Solar mesospheric near-infrared spectrometer: measurements of 1.27 mm radiances and inference for mesospheric ozone. Journal of Geophysical Research 89:9569-9580. Thomas, W., E. Hegels, S. Slijkhuis, R. Spurr and K. Chance (1998) Detection of biomass burning products in South East Asia from backscatter data taken by the GOME Spectrometer. Geophysical Research Letters 25:1317-1320. Titz, R., M. Birk, D. Hausamann, R. Nitsche, F. Schrier, J. Urban, H. Ktillmann and H.P. R6ser (1995) Observation of stratospheric OH at 2.5 THz with an airborne heterodyne system. Infrared Physics and Technology 36:883-891. UARS (1987) Upper Atmosphere Research Satellite: Project Data Book April 1987. Prepared by General Electric Astro-Space Division Valley Forge for the NASA Goddard Space Flight Center. Urban, J., A.H.P. Goede, H. Ktillmann, K. Ktinzi, G. Schwaab, N. Whybom and J. Wohlgemuth (1998) Chlorine activation and ozone destruction in the Arctic winter stratosphere 1996, as measured by an Airborne-Submillimetre SIS Radiometer. Geophysical Research Letters (in press). Volz, A. and D. Kley (1988) Evaluation of the Montsouris series of ozone measurements made in the nineteenth century. Nature 332:240-242. Waters, J.W., J.C. Hardy, R.F. Jarnot and H.M. Pickett (1981) Chlorine monoxide radical, ozone and hydrogen peroxide by microwave limb sounding. Science 214:61-64. Waters, J.W., J.C. Hardy, R.F. Jamot, H.M. Pickett and P. Zimmermann (1984) A balloon borne microwave limb sounder for stratospheric measurements. Journal of Quantitative Spectroscopy and Radiative Transfer 32:407-433. Waters, J.W., L. Froideveaux, W.G. Read, G.L. Manney, L.S. Elson, D.A. Flower, R.F. Jamot and R.S. Harwood (1993) Stratospheric CIO and ozone from the microwave limb sounder on the Upper Atmospheric Research Satellite. Nature 362:597-602.
Current andfuture passive remote sensing techniques used to determine atmospheric constituents
347
Watson, C.E., J. Fishman and H.G. Reichle jr. (1990) The significance of biomass burning as a source of carbon monoxide and ozone in the southern hemisphere tropics: a satellite analysis. Journal of Geophysical Research 95:16443-16450. Wayne, R.P. (1992) Chemistry of Atmospheres. Oxford University Press, Clarendon Press, Oxford. Wittrock, F., M. Eisinger, A. Ladst/itter-WeiBenmayer, A. Richter and J.P. Burrows (1998) Ground based UV/Vis measurements of O3, NO2, OCIO and BrO over Ny-Aalesund (79N, 12z. In: R. D. Bojkov and G. Visconti (Eds.) Atmospheric ozone. Proceedings of the XVIII Quadrennial Atmospheric Ozone Symposium 1996, L'Aquila, Italy, 12-21 September 1996, pp. 623-626. WMO (1995) Scientific assessment of ozone depletion. World Meteorological Organization Global Ozone Research and Monitoring Project. Report No. 37, World Meteorological Organization, Geneva. WMO (1998) SPARC/IOC/GAW Assessment of trends in the vertical distribution of Ozone, World Meteorological Organization Global Ozone Research and Monitoring Project. Report No. 43, World Meteorological Organization, Geneva. Zander, R. (1981) Recent observations of HF and HCI in the upper stratosphere. Geophysical Research Letters 8:413-416. Ziemke, J.R., S. Chandra and P.K. Bhartia (1998) Two new methods for deriving tropospheric column ozone from TOMS measurements: Assimilated UARS MLS/HALOE and convective-cloud differential techniques Journal of Geophysical Research 103:22115-22127.
This Page Intentionally Left Blank
349
PARTICIPANTS AND CONTRIBUTING AUTHORS 1 WITH FIELDS OF RESEARCH
Andreae, M.O. Max Plack Institute for Chemistry Biochemistry Department Postfach 3060 D-55020 Mainz, Germany Email: [email protected]
Bouwman, A.F. National Institute of Public Health and the Environment P.O. Box 1 3720 BA Bilthoven, Netherlands Email: [email protected]
Biogeochemistry, atmospheric chemistry
Soil science, biogeochemistry, global emission inventories
Archer, D. Department of Geophysical Sciences University of Chicago Chicago I11 60637 USA Email: [email protected]
Burrows, J.P. Universit~it Bremen Institut fur Umweltphysik/Fernerkundung Fachbereich 1, Postfach 330440 D-28334 Bremen, Germany Email: [email protected]
Oceanography, C cycle modelling in oceans
Atmospheric physics and chemistry, remote sensing
Asman, W.A.H. Assenvej 399 4000 Roskilde Denmark Email: [email protected]
Conrad, R. Max Planck Institute for Terrestrial Microbiology Karl von Frisch Strasse D-35043 Marburg/Lahn Germany Email: [email protected]
Atmospheric ammonia (chemistry, meteorology)
Microbiology
Batjes, N.H. International Soil Reference and Information Centre P.O. Box 353, 6700 AJ Wageningen Netherlands Email: [email protected]
Denier van der Gon, I-I.A.C. Department of Soil Science and Geology Wageningen Agricultural University P.O. Box 37, 6700 AA Wageningen Netherlands Email:[email protected]. wau.nl Methane fluxes from rice fields/tropical wetlands, measurements
Soil science, regional~global soil databases and applications Bogdanov, S. National Institute of Meteorology and Hydrology 66 Tsarigradsko Chaussee Sofia 1784 Bulgaria Email: [email protected]
Denmead, O.T. CSIRO Land and Water GPO Box 1666 Canberra ACT 2601 Australia Email: [email protected]
Trace gas emissions from agriculture and other human activities
Micrometeorology, flux measurement at local and regional scales
t Authors who were not at the workshop are indicated with *
350 Dentener, F.J. Institute for Marine and Atmospheric Research Utrecht University Princetonplein 5 3584 CC Utrecht, Netherlands Email: [email protected]
Ganzeveld, L. Institute for marine and Atmospheric Research University Utrecht Princetonplein 5 3584 CC Utrecht, Netherlands Email: [email protected]
Global tropospheric modelling, atmospheric chemistry
Atmosphere-biosphere exchange of trace gases at the global scale
Derwent, R.G. Meteological Office London Road, Bracknell RG 12 2SZ Berkshire U.K. Email: [email protected]
Garqon, V. LEGOS, UMR 5566 CNRS/CNES/UPS, GRGS 18 av. E. Belin 31400 Toulouse, France Email: [email protected], fr
Regional and global scale modelling acid rain and ozone formation, validation
Local~regional ocean C cycle modelling, oceanographic monitoring instrument development
Duyzer, J.H. TNO Institute of Environmental Sciences, Energy Research and Process Innovation P.O. Box 342, 7300 AH Apeldoorn Netherlands Email: [email protected]
Griffith, D.W.T.* Department of Chemistry University of Wollongong Wollongong NSW 2522 Australia Email: [email protected]
Photochemical ozone production, modelling, monitoring of gas concentrations
Atmospheric chemistry, FTIR techniques, surface exchange of trace gases
Estes, J.E. Remote Sensing Research Unit Department of Geography University of California, Santa Barbara CA 93106-4060, USA Email: [email protected]
Heimann, M.* Max Planck Institute for Meteorology Bundesstrasse 55 D-20146 Hamburg Germany Email: [email protected]
Remote sensing, GIS, land use~cover mapping at local to global scales
Atmospheric modelling, inverse modelling, atmospheric chemistry
Fowler, D. Institute of Terrestrial Ecology Bush Estate, Penicuik EH26 0QB Midlothian United Kingdom Email: [email protected]
Helder, W. Department of Marine Chemistry and Geology Netherlands Institute for Sea Research P.O. Box 59 1790 AB Den Burg (Texel), Netherlands Email: [email protected]
Environmental physics, micrometeorology, field measurements of trace gases
Marine biogeochemistry, N-cycling
Frankignoulle, M. Universit6 de Li6ge Mecanique des Fluides Geophysiques Unit6 d'Oceanographie Chimique B 4000 Sart Tilman, Belgium Email: [email protected]
Houweling, S. Institute for Marine and Atmospheric Research University Utrecht Princetonplein 5 3583 CC Utrecht, Netherlands Email: [email protected]
Marine chemistry, fieM measurements, coastal ocean
Global modelling of tropospheric chemistry, non-methane hydrocarbons, inverse modelling
List of participants and contributing authors withfields of research
3 51
Kaminski, T. Max Planck Institute for Meteorology Bundesstrasse 55 D-20146 Hamburg Germany Email: [email protected]
Meyer, C.P.* CSIRO Division of Atmospheric Research Private Bag 1 Aspendale Vic 3195 Australia Email:
Inversion of atmospheric transport, global scale, adjoint modelling
Exchange of trace gases, atmospheric chemistry, emission inventories
Kroeze, C. LUW-WIMEK P.O. Box 9101 6700 HB Wageningen Netherlands Email: [email protected]
Middelburg, J. Netherlands Institute of Ecology Centre for Estuarine and Coastal Ecology Korringaweg 7 4401 NT Yerseke, Netherlands Email: [email protected]
Biogeochemistry, environmental systems analysis modelling
Marine biogeochemistry, modelling and measurements in sediments
Lapitan, R.L. USDA-ARS-NPA 301 S. Howes St., Fort Collins P.O. Box E CO 80522 U.S.A. Email: [email protected]
Mosier, A.R. USDA-ARS 301 S. Howes St., Fort Collins P.O. Box E. CO 80523 U.S.A. Email: [email protected]
Micrometeorology, soil physics
Soil science, N and C cycle, nitrification/denitrification, elevated C02 and soil processes
Leuning, R.* CSIRO Land and Water GPO Box 1666 Canberra ACT 2601 Australia Email: [email protected],au
Nov~ik, M. Czech Geological Survey Geologicka 6 15200 Prague 5 Czech Republic Email: [email protected]
Land-atmosphere exchanges of radiation, heat and water, local and regional trace gas fluxes
Stable isotope geochemistry
Liss, P.S. University of East Anglia School of Environmental Sciences NR4 7TJ Norwich England Email: [email protected]
Panikov, N.S. Institute of Microbiology Russian Academy of Sciences Prosp. 60-1etija Octjabrja 7 Moscow 117811 Russia Email: [email protected]
Environmental chemistry, air-sea exchange
Soil microbiology, wetland, trace gas modelling
Malingreau, J.P. CCR Unit- SDME 10/85 European Commission 200 rue de la Loi 1049 Brussels, Belgium Email: [email protected]
Plant, R.A.J. Department of Soil Science and Geology Wageningen Agricultural University P.O. Box 37 6700 AA Wageningen, Netherlands Email: [email protected]
Remote sensing, biomass burning
Physical geography, land use effects on N20 emissions
352 Schimel, D.S.* NCAR, UCAR Climate System Modeling Program, Climate and Global Dynamics Division CO 80303 USA Email: [email protected]
Wanninkhof, R. NOAA/AOML Ocean Chemistry Division 4301 Rickenbacker Causeway Miami, FL 33149 USA Email: [email protected]
Biogeochemistry, trace gas production in ecosystems
Air-sea exchange, CO: uptake by oceans
Seitzinger, S.P. Inst. of Marine & Coastal Sciences Rutgers/NOAA CMER Program NJ 08901-g521 USA Email: [email protected]
Coastal marine biogeochemistry ~, P, C), water-shed land use effects on N/P input to coastal zone Sofiev, M.A. Institute of Program Systems c/o. Stroiteley str. 4 Bid. 1, App. 18 Moscow 117311 Russia Email: [email protected]
Long-range atmospheric transport modelling, model validation and intercomparison Starink, M. Netherlands Institute of Ecology Centre for Estuarine and Coastal Ecology Korringaweg 7 4401 NT Yerseke Netherlands Email: [email protected]
Microbiology Trumbore, S. Earth System Science UC Irvine 19172 Jamboree Road Irvine CA 92717-3100 U.S.A. Email: [email protected]
Isotope geochemistry Wania, F. Wania Environmental Chemsits Corp. 280 Simcoe Street, Suite 404 Toronto, Ontario M5T 2Y5 Canada Email: [email protected]
Global~regional modelling of persistant organic pollutants and of their air-surface exchange
INDEX
a posteriori estimates 284, 288, 289-291 a priori est'.'mates 284, 288-291 ADEOS, see remote sensing-Advanced Earth Observation Satellite adjoint model 285,289-292 Advanced Very High Resolution Radiometer 137, 138, 140, 142, 164, 326, 330 aerodynamic gradient method, s e e flux measurement aerosols, emission inventory 17 aggregation 5, 7, 12-14, 15, 16, 22, 197-200, 224-225, 303 -modelling 197-200 aircraft-based measurement 42, 43 aircraft-based sensors 32-33 ammonia, -compensation concentration 208 -emission inventory 17 -environmental controls 161-162 -processes 161-162 analytical devices 35-36 animal, -categories 15 -housings 7 -manure 15 -production 15 -waste 7, 15 animal waste management 7 aquatic systems, -bomb 14C 50, 51 -carbon dioxide fluxes 32-33 -chamber method 52 -eddy accumulation 51, 55, 56 -eddy correlation 51 -eddy covariance 57 -flux measurement methods 47-57 -fugacity 47 -gas transfer velocity 9, 47, 48, 50, 54, 174 -gradient method 51 -kinematic viscosity 48 -langrnuir cells 49 -mass balance 53 -mesoscale eddies 49 -remote sensing 57 -saturation level 49 -Schmidt number 48, 54, 55
aquatic systems, continued -surface roughness 48 -tracers, 51, 53 -waves 48, 49, 55 -Webb correction 56 arithmetic mean 239 ATMOS, see remote sensing- Atmospheric Trace Molecule Spectroscopy atmosphere, -memory 76, 283 -mesopause 318 -oxygen 318-319 -ozone 318-319 -pressure 318-319 -remote sensing 317-347 -stratosphere 318-319 -temperature 318-319 -thermosphere 319 -tropopause 318 -troposphere 318-319, 321 atmospheric measurements, -monitoring networks 19, 278, 286-287, 292 -outliers 285 -representativeness 240-241,278, 287 atmospheric model 4-6, 18-21, 91,235-237, 275295 - s e e also chemistry transport model atmospheric transport, -memory 283 -rectifier effect 264, 266, 269, 282 ATSR, see remote sensing- Along Track Scanning Radiometer autocorrelation 249 AVHRR, s e e Advanced Very High Resolution Radiometer biomass burning 7, 159, 321 biome, emission factors 7, 8 black box model 227-228 black carbon, emission inventory 17 bomb ~4C 50, 51 Bowen ratio method, s e e flux measurement Bowen ratio, -latent heat flux 41 -net radiation 41 -sensible heat flux 41
354 Bowen ratio, continued -soil heat flux 41 BUV, see remote sensing- Backscatter Ultraviolet Ozone Experiment canopy flux 69, 70-73 carbon dioxide -convective boundary layer budgeting 78 -emission inventory 17 -environmental controls 161-162 -global budget 300 -infrared analyzer 42, 43, 56 -isotopes 261-266 -modelling 171-183 -nocturnal boundary layer budgeting 80, 81 -oceans 171-183 -processes 161 - 162 -profiles 42, 43 -uncertainty 306 carbon dioxide cycle, uncertainty 265 carbon dioxide flux, -ecosystems 32-33 -geologic scale 179 -gyre 178 -mesoscale 177 -microscale 172-175 -mixed layer 175-177 -molecular scale 171-172 carbon monoxide, -compensation concentration 208 -emission inventory 17 -isotopes 267-268 -variability 268 carbonate buffer 171, 174, 179-180 carbonyl sulphide, 208, 336 -compensation concentration 208 CBL, s e e convective boundary layer CFC, s e e chlorofluorocarbons CH4, s e e methane chamber method, s e e flux measurement chemistry transport model 3, 4-6, 18-21, 91, 275295 -advection 280, 281 -bayesian approach 284, 288 -chemical mechanisms 4 -comparison - method of least squares 251 -comparison - ranking procedure 251-252 -continuity equation 279 -coverage 236 -global 236, 237 -inverse approaches 275-295 -local 236, 237 -off line 4, 280 -on line 4, 280
chemistry transport model, continued -precision 244-245 -regional 236, 237 -reliability 235 -resolution 236 -spatial scale 4, 18-21,236-238, 278 -subgrid transport 280, 282 -temporal scale 4, 18-21 -validation 235-255,282 -verification 236 -wind fields 284 -chlorofluorocarbons, 4 CLAES, see remote sensing - Cryogenic Limb Array Etalon Spectrometer climate data, uncertainty 12 climate model 130-131 closed-path infrared analyzers 35-36 CO, s e e carbon monoxide C 0 2 , s e e carbon dioxide compensation concentration 205-216 -ambient concentration 205 -ammonia 208 -carbon monoxide 208 -carbonyl sulphide 208 -dihydrogen, 208 -environmental controls 210 -nitrogen oxides (NOx) 208 -nitrous oxide 208 -plant canopy 208-209 -soil 208-209 computer assisted classification 133, 135 concentration data, redundant 71 conditional sampling 39, 40-41, 45 -nitrous oxide 39, 45 continuity equation 279 convective boundary layer budgeting, s e e flux measurement correlation coefficient 8, 239 crop production 15 cross-correlation 250 CTM, s e e chemistry-transport model data, -animal production 15 -crop production 15 -environmental 22 -forestry 15 -geographical 11-14 -georeferenced data 127, 134 -land cover 127-129 -land use 13-14, 15 -livestock production 15 -oceans 12-13 -projection 134
Index
data, continued -quality 14, 134 -soils 13 -surrogate 305, 306 -vegetation 13-14 database, -mask files 13 -projection 134 daytime boundary layer budgeting, s e e convective boundary layer budgeting deposition 3, 20, 205 -parametrization 20 detection limit, measurement methods 36 diffusion 194 dihydrogen, compensation concentration 208 dimethylsulphide 4, 321,336 DMS, s e e dimethylsulphide DOAS, see remote sensing- Differential Optical Absorption Spectroscopy dry deposition 20 ebullition, s e e gas exchange ECD, s e e electron capture detector economic data, -animal production 15 -crop production 15 -forestry 15 -livestock production 15 ecoregions 136 electron capture detector 36, 45 emission estimates 6-16 -databases 11-14 -uncertainty 6-16 emission factor 6, 7, 8, 17, 21 -biomes 7 emission inventory 3, 16-18 -aerosols 17 -ammonia 17 -black carbon 17 -carbon dioxide 17 -carbon monoxide 17 -global 16-18 -methane 17 -nitrogen oxides (NOx) 17 -nitrous oxide 17 -regional 18 -spatial resolution 16-18 -sulphur dioxide 17 -temporal resolution 16-18 -validation 20-21 -volatile organic compounds 17 emission -measurement data 7 -scenario 3
3 55
emission, continued -spatial patterns 6 -temporal patterns 6, 15-16 -uncertainty 158-159 -variability 22 enclosure, s e e flux measurement - chamber method energy balance method 34, 41, 112-113 environmental controls 195-197, 198 environmental data 22 Eulerian coordinate system 78 experimental design 101-121 farm model 11 fertilizer use 15 fertilizers, nitric oxide flux 103 fetch 39, 75, 89, 110 FID, s e e flame ionization detector fires, flaming 7 fires, smouldering 7 flame ionization detector 35-36 flux measurement, -aerodynamic gradient method 111-112, 206 -airborne 90 -aquatic systems 50-57 -Bowen ratio method 34, 41, 112-113 -campaigns 113-114 -canopy height 38 -chamber - experimental design 101-109 -chamber method 29, 32-33, 34, 37, 46, 52, 87, 88, 89, 101-109 -chamber method - aquatic systems 52 -chamber method - design 105 -chamber method - errors 103 -chamber method- footprint 88, 89 -chamber method - spatial coverage 109 -chamber method 29 -chemiluminescence 36 -closed chamber method 29, 32-33, 34, 37, 46, 102 -convective boundary layer budgeting 32, 33, 34, 43, 44, 75-79, 82, 83, 90, 115-116 -convective boundary layer budgeting Eulerian coordinate system 78 -convective boundary layer budgeting footprint 76, 90 -convective boundary laye," budgetingLagrangian coordinate system 78 -convective boundary layer budgeting methane 77, 78, 115-116 -convective boundary layer budgeting nitrous oxide 77, 78 -detection limit 36 -dynamic chamber method, s e e open chamber
356 flux measurement, continued -eddy accumulation- aquatic systems 51, 55, 56 -eddy accumulation method 34, 36, 40-41, 51, 55, 56 -eddy correlation - aquatic systems 51 -eddy correlation method 29, 34, 36, 39-40, 51, 89, 90 -eddy covariance - aquatic systems 57 -eddy co-variance method 39-40, 57, 94, 110 -eddy diffusivity- tracers 44 -eddy diffusivity 37, 38, 39,44 -eddy relaxation method 34 -energy balance method 34, 41, 112-113 -experimental design 101-121 -fetch 3 9 , 75, 89, 110 -field scale 109 -flux gradient method 34, 36, 38-39, 51, 111113 -footprint 31, 44-45, 46, 76, 87, 88, 89, 95, 109, 117, 229 -inverse Lagrangian dispersion method 70-73 -inverse Lagrangian dispersion method carbon dioxide 72 -inverse Lagrangian dispersion method latent heat 72 -Lagrangian dispersion method 89 -mass balance - aquatic systems 53 -mass balance - turbulent diffusive flux 74 -mass balance method 34, 41, 53, 73-75, 91, 113 -mass balance method - convective flux 74 -mass balance method - fetch 41, 75 -mass balance method - footprint 91 -mass budget 115,250 -megachamber 44, 89, 105-106 -methane 38 -micrometeorological method 29, 32-33, 34, 37-44, 87, 89, 109-114 -monitoring 114 -nitrous oxide 38, 39 -nocturnal boundary layer budgeting 43, 8082, 90, 115 -nocturnal boundary layer budgeting - carbon dioxide 80, 81 -nocturnal boundary layer budgetingfootprint 90 -nocturnal boundary layer budgetingmethane 80, 81 -nocturnal boundary layer budgeting - nitrous oxide 81 -open chamber method 32-33, 34, 37, 46, 102103, 105 -regional scale 114-116
flux measurement, continued -relaxed eddy accumulation 40 -soil concentration profile 88 -static chamber, s e e closed chamber -strategy 101 - 121 -surface roughness length 37 -tracer method 34 -tunable diode laser 106 -water-side 229 -wind speed 38 -zero plane displacement 38 footprint 31, 44-45, 46, 76, 87, 88, 89, 95, 109, 117, 229 -airborne mass balance 95 -airborne micrometeorological method 95 -atmospheric stability 31 -chamber method 95 -convective boundary layer budget 43, 90, 95 -micrometeorlogical method 95 -nocturnal boundary layer budgeting 90, 95 -roughness length 31 -soil gas concentration profile 95 -vegetation 31 -wind speed 31 forestry 15 formaldehyde, remote sensing 334, 336 fourier transform infrared spectrometer, nitrous oxide 39, 45, 46 fractional bias 246 friction velocity 76 FTIR, s e e Fourier-transform infrared spectrometer FTS, s e e remote sensing - Fou:ier Transform Spectrometer fugacity 47 functional strata 128 functional type 7, 16, 22, 128, 153-168,224 -environmental controls 155-157, 164 -methane 157-161 -remote sensing 155, 163 gamma distribution 242, 243 gas chromatography 35-36, 45 gas diffusivity 205 gas exchange, ebullition 55, 174, 194 gas filter correlation infrared absorption analyzer 36 gas flux, s e e trace gas flux gas transfer, modelling 175 gas transfer velocity, -aquatic systems 9, 47, 48, 50, 54, 174 -wind speed 9 gaussian distribution 242, 243 GC, s e e gas chromatography
357
Index
GCM, s e e global climate models generalization 5, 15, 16 geographic data 11-14, 125-150 -land use 13-14 -oceans 12-13 -soils 13 -vegetation 13-14 georeferenced data 127 -projection 134 GFCIR, s e e gas filter correlation infrared absorption analyzer global area coverage 142 global emission inventories 16-18 global models, validation method 242-244 global vegetation index 142 GOME, s e e remote sensing - Global Ozone Monitoring experiment GOMOS, s e e remote sensing - Global Ozone Monitoring by Occultation of Stars gradient method, aquatic systems 51 H2, s e e dihydrogen H202, s e e hydrogen peroxide half-life 262-263 HALOE, s e e remote sensing - Halogen Occultation Experiment halogen oxides, remote sensing 337, 338 Henry's law 171 heterogeneity, s e e variability histogram 250 hot spots 11,230 HNO3, s e e nitric acid hydrogen peroxide 321 hydroxyl radical 321 ILAS, s e e remote sensing - Improved Limb Atmospheric Spectrometer IMG, s e e remote sensing - Interferometric Monitor for Greenhouse gases infrared absorption spectroscopy 35-36 infrared analyzer 42, 43, 56 infrared analyzer, carbon dioxide 42, 43 interannual variability 22, 200 inverse Lagrangian dispersion method, s e e flux measurement inverse modelling 277-295, 91, 92-93 - a p o s t e r i o r i estimates 284, 288, 289-291 - a p r i o r i estimates 284, 288-291 -adjoint model 285,289-292 -advection 280, 281 -brute force 283 -isotopes 292 -rectifier effects 282 -subgrid transport 280, 28I
inverse modelling, continued -synthesis 283 -uncertainty 287, 290-292 IR, s e e infrared absorption spectroscopy ISAMS, s e e remote sensing - Improved Stratospheric and Mesospheric Sounder isoprene, flux 109 isotopes 50, 259-272 -abundancies 260 -carbon 50 -carbon dioxide 261-266 -carbon monoxide 267-268 -common isotopes 260 -fossil fuel 266 -global methane budget 266 -global nitrous oxide budget 268 -half life 262-263 -kinetic isotope effect 266, 268, 271 -mass spectrometer 260, 2"70, 271 -methane 266-267 -nitrous oxide 268-269 -radioactive decay 262 -rare isotopes 260 kinematic viscosity 48 kinetic isotope effect 266, 268, 271 Lagrangian coordinate system 78 Lagrangian dispersion method 89 land cover, -aerial photography 141 -changes 125 -classification 135-137, 141 -classification schemes 129, 131 -data 127-129 -functional type 128 -geographically referenced data 127 -legends 137, 138-141 -mapping 133-144 -mapping- regionalization 137 -mosaics 141 -seasonal dynamics 128 -types 7 land use data 13-14, 15 landfills 159 laser spectroscopy 45 life form 137 LIMS, s e e remote sensing - Limb Infrared Monitor of the Stratosphere livestock production 15, 158 local models, measurement data 245 -validation method 245-248 log-normal distribution 242, 243
358 LRIR,
remote sensing - Limb radiance Inversion Radiometer
see
mapping, -dominant vegetation 136 -land cover 134-144 maps, science quality 134 MAPS, s e e remote sensing - Measurement of Air Pollution from Satellites MAS, s e e remote sensing - Microwave Atmospheric Sounder mass balance method, s e e flux measurement mass spectrometer 260, 270 maximum absolute deviation 240 mean 238 mean square deviation 239-240 measurement data, outliers 285 measurement data, representativeness 7, 240241,278,287 mechanistic model 187-202 median 238 megachamber 44, 89, 105-106 MERIS, s e e remote sensing - Medium Resolution Imaging Spectrometer for Passive Atmospheric Sounding mesopause, s e e atmosphere mesosphere, s e e atmosphere methane flux 32, 107, 108 -conditional sampling 39 -eddy correlation 39 -flux gradient 39 -forests 108 -mass budget 115, 116 -spatial variability 108 -transect 107 methane, -aquatic systems 158 -aquatic systems 158 -biomass burning 159 -convective boundary layer budgeting 77, 78 -emission inventory 17 -environmental controls 161-162 -forests 108 -functional type 157-161 -isotopes 266-267 -landfills 159 -livestock production 158 -mass budget 115, 116 -nocturnal boundary layer budgeting 80, 81 -oxidation 159 -processes 161-162, 190 -spatial variability 108 -transect 107 -uncertainty 266-267
method of least squares 240, 251 micrometeorological method, s e e flux measurement MIPAS, s e e remote sensing - Michelson Inferometer for Passive Atmospheric Sounding MLS, s e e remote sensing - Microwave Limb Sounder model, - s e e also chemistry transport models, trace gas flux model -validation, s e e validation -climate 130-131 -functional type 224-225 -hierarchy 301 monitoring networks, atmosphere 19, 278, 286287, 292 monitoring stations, sulphur dioxide 19 MOPITT, s e e remote sensing - Measurement of Pollution in the Troposphere multiple measurement 71, 87, 93-94, 95 N20, s e e nitrous oxide natural 14C 50 NBL, s e e nocturnal boundary layer NDIR, s e e non-dispersive infrared absorption NDVI, s e e Normalized Difference Vegetation Index near-infrared diode laser 35-36 net primary production 133 NH3, s e e ammonia NIRDL, s e e near-infrared diode laser nitrate radical 321 nitric acid 321 nitric oxide, -emission inventory 17 -environmental controls 161-162 -field scale 310 -global sources 308 -point scale 310 -processes 161 - 162 -temporal scale 12 nitric oxide flux, -agricultural soils 103 -extrapolation 104 -soil 7, 103 nitrogen oxides, -atmospheric processes 309, 333-335 -compensation concentration 208 -remote sensing 333-335 -uncertainty 307 nitrous oxide flux, -agricultural soils 103,220-223 -conditional sampling 39
359
Index
nitrous oxide flux, continued -eddy correlation 39 -flux gradient 39 -modelling 222-224 -soils 8 -variability 106 nitrous oxide, -compensation concentration 208 -convective boundary layer budgeting 77, 78 -emission inventory 17 -environmental controls 161-162 -fourier transform infrared spectrometer 46 -isotopes 268-269 -nocturnal boundary layer budgeting 81 -processes 161-162 -profiles 42, 43 -tunable diode laser 42,43 -uncertainty 268-269 NO, s e e nitric oxide, nitrogen oxides Non-dispersive infrared absorption, 36 nonmethane volatile organic compounds 18 Normalized Difference Vegetation Index 138 NOx, s e e nitrogen oxides, nitric oxide NPP, s e e net primary production nutrient balance models 11 03, s e e ozone ocean data 12-13 -uncertainty 12-13 ocean flux, micrometeorological method 229 ocean, -calcium carbonate 179 -carbonate buffer 171, 174 -carbonate buffer 179-180 -gas transfer 172, 175 -geologic scale 179 -gyre 178 -mesoscale 177 -microscale 172-175 -mixed layer 175-177 -molecular scale 171-172 -processes 173 -trace gas flux modelling 8 -modelling 171-183 OCS, s e e carbonyl sulphide OH, see hydroxyl radical OMI, s e e remote sensing - Ozone Monitoring Instrument open chamber method, s e e flux measurement Ostwald solubility coefficient 47, 48, 49 outliers 238, 285 ozone 4, s e e also atmosphere -layer 320 -remote sensing 332-333,334
POLDER, s e e remote sensing - Polarization and Directionality of the Earth's Radiance quantile 246, 247 quartile 246, 247 r2, s e e correlation coefficient radiative forcing 3 radioactive decay 262 radon 51 rectifier effect 264, 266, 269, 282 redundant concentration data 71 redundant measurement 71, 87, 93-94, 95 regional emission inventories 18 regional model 197-200 regional model, validation method 242-244 regression, -analysis 238 -bias 239 -coefficients 238 -models 8 -slope 239 relaxed eddy accumulation, s e e flux measurement remote sensing 13-14, 125-150, 164-165,317347 -Advanced Very High Resolution Radiometer 137, 138, 140, 142, 164 -aquatic systems 57 -atmospheric composition 165 -calibration 135 -land cover 125-150 -land cover mapping 133-134 -Landsat 143 -Normalized Difference Vegetation Index 138 -Pathfinder 143 -SPOT 143, 164 -validation 96, 331 -vegetation data 13-14 remote sensing of atmosphere, -Advanced Earth Observation Satellite 326, 328, 329 -Advanced Very High Resolution Radiometer 326, 330 -Along Track Scanning Radiometer 326, 331 -atmosphere 317-347 -Atmospheric Trace Molecule Spectroscopy 326,328 -Backscatter Ultraviolet Ozone Experiment 326, 329 -Cryogenic Limb Array Etalon Spectrometer 326, 328
360 remote sensing of atmosphere, continued -Differential Optical Absorption Spectroscopy 328 -Dobson unit 332, 336 -ENVISAT, 326-327, 339 -far infrared 324-325 -formaldehyde 334, 336 -Fourier Transform Spectrometer, 325 -Global Ozone Monitoring by Occultation of Stars 326, 339 -Global Ozone Monitoring Experiment 326, 329, 330, 333,334, 335, 336, 337, 338 -Halogen Occultation Experiment 326, 328, 329,332 -halogen oxides 337, 338 -Improved Limb Atmospheric Spectrometer 326, 328, 337 -Improved Stratospheric and Mesospheric Sounder 326, 328 -Interferometric Monitor for Greenhouse gases 326 -Limb Infrared Monitor of the Stratosphere 326 -Limb Radiance Inversion Radiometer 326 -Measurement of Air Pollution from Satellites 326,328 -Measurement of Pollution in the Troposphere 326, 337 -Medium Resolution Imaging Spectrometer for Passive Atmospheric Sounding 326, 339 -Michelson Inferometer for Passive Atmospheric Sounding 326 -microwave 324-325 -Microwave Atmospheric Sounder 326 -Microwave Limb Sounder 325,326, 332 -mid-infrared 325-328 -Nimbus 326, 327, 328 -nitrogen dioxide 323 -ozone 332-333 -Ozone Monitoring Instrument 326, 339 -Polarization and Directionality of the Eartll's Radiance 326, 331,337 -SAFIRE 325 -satellites 326-327 -Scanning Imaging Absorption Spectrometer for Atmospheric Cartography 327, 329, 333,337,339 -Selective Chopper Radiometer 327, 328 -Solar Backscatter Ultraviolet Ozone Experiment 327, 328, 329, 333,337 -Solar Mesospheric Experiment 327 -Space shuttle 326, 328
remote sensing of atmosphere, continued -Stratospheric Aerosols and Gas Experiment 327, 329, 330, 337 -Stratospheric Aerosols Measurement 327, 328 -Total Ozone Monitoring Spectometer 327, 328, 329, 330, 332, 337 -trace gases 324-339 -Tropospheric Emission Spectrometer 327 -tropospheric ozone residual technique 332 -Upper atmospheric Research Satellite 325, 326, 327, 328, 332 representativeness 7, 240-241,278, 287 robust statistical method 8, 238 SAGE, s e e remote sensing - Stratospheric Aerosols and Gas Experiment SAM, s e e remote sensing - Stratospheric Aerosols Measurement SBUV, s e e remote sensing - Solar Backscatter Ultraviolet Ozone experiment scale, -flux models 10, 11 -model approach 219-232 -model parameters 219-232 -modelling 187-202, 219-232 -trace gas budget 300, 302 -trace gas flux 299 scaling, -bottom-up 5,303 -top-down 5,275-295,301,304 Schmidt number 48, 54, 55 SCIAMACHY, s e e remote sensing - Scanning Imaging Absorption Spectrometer for Atmospheric Cartography science quality maps 134 SCR, s e e remote sensing - Selective Chopper Radiometer SF6 44, 53 SME, s e e remote sensing - Solar Mesospheric Experiment SO2, sulphur dioxide soil, -data 13 -fertility 8 -moisture 8 -organic matter 8 -oxygen 8 -processes 190 -profile data 13 -temperature 8 spatial distribution of controlling factors, -surrogate 11, 14 -uncertainty 11
Index
spatial resolution 16-18 spatial variability, gas flux 10, 102, 105, 106 stable isotopes, 228 standard deviation 238, 239 statistical average model 249, 252 statistical methods 238-240 -robustness 8, 238 stratification 5, 8, 16, 151-167 stratosphere, s e e atmosphere sub-grid heterogeneity 131,132 substrate production 189-191 sulphur dioxide 4, 17, 19, 321 -emission inventory 17 -monitoring stations 19 -remote sensing 336-337 sulphur, stable isotopes 228 sulphuric acid 321 summary model 10, 11,306 surface concentration anomalies 9 surrogate spatial distribution 14 TDL, s e e tunable diode laser temporal distribution of emissions 15-16 temporal resolution 16-18 temporal variability, trace gas flux 10, 107 terrain units 136 terrestrial ecosystems, -carbon dioxide fluxes 32-33 -environmental controls 195-197, 198 -modelling 187-202 -regional modelling 197-200 TES, s e e remote sensing- Tropospheric Emission Spectrometer thermosphere, s e e atmosphere TOMS, s e e remote sensing - Total Ozone Monitoring Spectometer TOR, s e e remote sensing - tropospheric ozone residual technique trace gas exchange, s e e gas exchange trace gas flux model 5, 6, 10, 11, 21,219-232, 171-202, 219-232 -aggregation 301 -black box 227-228 -coupling of models 224-226 -development 221 -empirical 187-188 -environmental simulation 129, 130-133 -gas transfer 175 -mechanistic 187-202 -microbial population 188, 193 -non-linear processes 225-226 -ocean 8, 171 - 183 -parameters 219-232 -process 301
3 61
trace gas flux model, continued -process 8 -regional scale 197-200 -regional scale 301 -regression 8, 187, 199 -scale 300 -soil 9 -spatial scale 211, 219-231 -stability-dependent method 9 -substrate production 189-191 -summary model 10, 11,306 -terrestrial systems 187-202 -types 187-189 -validation 301 trace gas flux, -aggregation 225 -canopy 69, 70-73 -carbon dioxide 32 -compensation concentration 205-216 -deposition 205 -ecosystems, 30-31 -environmental controls 155-157, 164 -environmental controls 161-162 -environmental controls 195-197, 198 -exchange 20 -functional type 153-168 -functional type 153-168 -heterogeneity s e e variability; trace gas fluxvariability -hot spots 11,230 -integration 224-225 -isoprene 109 -mass budget 115, 116 -methane 107, 108 -methane 32 -model 6, 21, 171-202, 187-202, 219-232 -model approach 219-232 -model development 221 -model, s e e model -net flux 205 -nitrous oxide 32 -processes 161 -pulses 16 -regional modelling 197-200 -regulating factors 8 -spatial variability 10, 102, 105, 106, 230 -temporal variability 10, 107 -variability 10, 11, 30-35, 102, 105, 106, 107, 108,230 trace gas, substrate 192 -transport 194-195 tracer 51, 53, 54, 174, 269-270 -aquatic systems 51, 53
362 tracer, continued -deuterium 270 -eddy diffusivity 44 -lead 270 -method 34 -oxygen 270 -radon 270 -release 43-44 -SF6 44, 53, 54 transport models, s e e chemistry transport models tropopause, s e e atmosphere troposphere, s e e atmosphere tunable diode laser 35-36, 39, 40, 42, 43, 44, J06 -nitrous oxide 39, 42, 43 UARS, s e e Upper atmospheric Research Satellite uncertainty, -aggregation 303 -carbon dioxide 306 -carbon dioxide partitioning 265 -climate data 12 -disaggregation 304 -economic data 15 -emission factor 7-8 -emissions 6-16 -field scale 304 -fluxes 31, 34, 44 -global budgets 300 -global scale 306 -inverse modelling 287, 290-292 -local scale 305 -methane budget 266-267 -methane fluxes 158-159 -models 235-255 -nitrogen oxides 307 -nitrous oxide budget 268-269 -ocean data 12-13 -point scale 304 -process models 8-11 -reduction of uncertainty 306 -regional scale 305 -regression model 8 -soil data 13 -spatial distribution 11-14 -statistical analysis 238-240 -temporal distribution 15-16 -trace gas fluxes 158-159 -vegetation data 13-14 validation 19, 20-21, 95,257-274, 282 -arithmetic mean 239 -autocorrelation 249 -correlation coefficient 239 -cross-correlation 250
validation, continued -ecosystem flux 306 -fractional bias 246 -histogram 250 -mass budget 250 -maximum absolute deviation 240 -mean square deviation 239 -method 242-244 -method of least squares 240 -model comparison 248-252 -quantile 246, 247 -quartile 246, 247 -ranking procedure 251-252 -regression analysis 238 -regression bias 239 -regression coefficients 238 -regression slope 239 -remote sensing 96, 331 -standard deviations 238, 239 -trace gas flux model 301 -tracers 247 variability, -carbon monoxide 268 -emissions 22 -interannual 22, 200 -methane flux 108 -nitrous oxide flux 106 -trace gas flux 10, 11, 30-35, 102, 105, 106, 107,108, 230 vascular transport 194 vegetation, -classification schemes 129, 131,133, 135137, 141 -data 13-14 -dominant 136 -functional type 128 -heterogeneity 131 -landscape variability 136 -legends 137, 138-141 -mapping 133-144 -mapping - regionalization 137 -net primary production 133 -remote sensing 13-14, 125-150 -sub-grid heterogeneity 131, 132 -types 125 -uncertainty 13-14 VOC, s e e volatile organic compounds volatile organic compounds 7, 17, 18 -emission inventory 17 Von Karman constant 76, 111 waves, see aquatic systems Webb correction 56 wet deposition 20