No title

WATER RESEARCH A Journal of the International Water Association

Editor-in-Chief Mark van Loosdrecht Delft University of Technology Department of Biochemical Engineering Julianalaan 67 2628 BC Delft The Netherlands Tel: +31 15 27 81618 E-mail:[email protected]

Editors J. Block Université H. Poincaré, Nancy I France David Dixon University of Melbourne Australia Hiroaki Furumai The University of Tokyo Japan Xiaodi Hao Beijing University of Civil Engineering and Architecture China Gregory Korshin University of Washington USA Anna Ledin Formas Sweden Eberhard Morgenroth Swiss Federal Institute of Aquatic Science and Technology (EAWAG) Switzerland W. Rauch University Innsbruck Austria Maria Reis Universidade Nova de Lisboa/FCT Portugal Hang-Shik Shin Korea Advanced Institute of Science and Technology Korea Thomas Ternes Bundesanstalt für Gewässerkunde Germany Stefan Wuertz Univ. of California, Davis USA

Associate Editors Andrew Baker University of New South Wales Australia

Damien Batstone The University of Queensland Australia G-H. Chen The Hong Kong University of Science & Technology Hong Kong China Tom Curtis Univ. of Newcastle upon Tyne UK Ana Deletic Monash University USA Francis de los Reyes III North Carolina State University USA Rob Eldridge The University of Melbourne Australia Rosina Girones University of Barcelona Spain Stephen Gray Victoria University Australia Kate Grudpan Chiang Mai University Thailand E.E. Herricks University of Illinois - Urbana USA Peter Hillis United Utilities Plc UK H-Y. Hu Tsinghua University China P.M. Huck University of Waterloo Canada Bruce Jefferson Cranfield University UK Ulf Jeppsson Lund University Sweden Sergey Kalyuzhnyi Moscow State University Russian Federation Jaehong Kim Georgia Institute of Technology USA Jes La Cour Jansen Lund Institute of Technology Sweden G. Langergraber BOKU/Univ. of Natural Res. and Applied Life Scs. Austria S-L. Lo National Taiwan University Taiwan Dionisis Mantzavinos Technical University of Crete Greece

Y. Matsui Hokkaido University Japan A. Maul Université Paul Verlaine-Metz France Max Maurer EAWAG Switzerland How Yong Ng National University of Singapore Singapore Satoshi Okabe Hokkaido University Japan S.L. Ong National University of Singapore Singapore Jong M. Park Pohang University of Science & Technology Korea Susan Richardson U.S. Environmental Protection Agency USA Miguel Salgot University of Barcelona Spain David Sedlak University of California, Berkeley USA Jean-Philippe Steyer LBE-INRA France M. Takahashi Hokkaido University Japan Kai Udert EAWAG Switzerland V.P. Venugopalan BARC Facilities India E. von Sperling Federal University of Minas Gerais Brazil Hanqing Yu University of Science & Technology of China China J. Zilles University of Illinois Urbana USA A.I. Zouboulis Aristotle University Greece

Editorial Office Mailing address Water Research Radarweg 29 1043 NX Amsterdam The Netherlands E-mail: [email protected]

Publication information: Water Research (ISSN 0043-1354). For 2011, volume 45 is scheduled for publication. Subscription prices are available upon ­request from the publisher or from the Elsevier Customer Service Department nearest you or from this journal’s website (http://www.elsevier.com/locate/watres). Further information is available on this journal and other Elsevier products through Elsevier’s website (http://www.elsevier.com). Subscriptions are accepted on a prepaid basis only and are entered on a calendar year basis. Issues are sent by standard mail (surface within Europe, air delivery outside Europe). Priority rates are available upon request. Claims for missing issues should be made within six months of the date of dispatch. Author Enquiries: Please visit this journal’s homepage at http://www.elsevier.com/locate/watres. You can track accepted articles at http://www.elsevier.com/ trackarticle and set up e-mail alerts to inform you of when an article’s status has changed. Also accessible from here is information on copyright, frequently asked questions and more. Contact details for questions arising after acceptance of an article, especially those relating to proofs, will be provided by the publisher. Orders, claims, and journal enquiries: Please contact the Elsevier Customer Service Department nearest you: St. Louis: Elsevier Customer Service Department, 3251 Riverport Lane, Maryland Heights, MO 63043, USA; phone: (877) 8397126 [toll free within the USA]; (+1) (314) 4478878 [outside the USA]; fax: (+1) (314) 4478077; e-mail: [email protected] Oxford: Elsevier Customer Service Department, The Boulevard, Langford Lane, Kidlington OX5 1GB, UK; phone: (+44) (1865) 843434; fax: (+44) (1865) 843970; e-mail: [email protected] Tokyo: Elsevier Customer Service Department, 4F Higashi-Azabu, 1-Chome Bldg, 1-9-15 Higashi-Azabu, Minato-ku, Tokyo 106-0044, Japan; phone: (+81) (3) 5561 5037; fax: (+81) (3) 5561 5047; e-mail: [email protected] Singapore: Elsevier Customer Service Department, 3 Killiney Road, #08-01 Winsland House I, Singapore 239519; phone: (+65) 63490222; fax: (+65) 67331510; e-mail: [email protected] Application for membership of International Water Association should be made to: Executive Director, IWA, Alliance House, 12 Caxton Street, London SW1H 0QS, U.K. (Tel.: +44 207 654 5500; Fax: +44 207 654 5555; e-mail: [email protected]; website: http://www.IWAhq.org.uk). Registered Charity (England) No. 289269. Individual membership is available from £30 upwards. For details contact IWA.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 0 7 e1 9 2 0

Available at www.sciencedirect.com

journal homepage: www.elsevier.com/locate/watres

Review

Energy minimization strategies and renewable energy utilization for desalination: A review Arun Subramani a,*, Mohammad Badruzzaman a, Joan Oppenheimer a, Joseph G. Jacangelo a,b a b

MWH Americas Inc., 618 Michillinda Avenue, Arcadia, CA 91007, USA The Johns Hopkins University, Baltimore, MD 21205, USA

article info

abstract

Article history:

Energy is a significant cost in the economics of desalinating waters, but water scarcity is

Received 25 October 2010

driving the rapid expansion in global installed capacity of desalination facilities. Conven-

Received in revised form

tional fossil fuels have been utilized as their main energy source, but recent concerns over

27 December 2010

greenhouse gas (GHG) emissions have promoted global development and implementation of

Accepted 31 December 2010

energy minimization strategies and cleaner energy supplies. In this paper, a comprehensive

Available online 9 January 2011

review of energy minimization strategies for membrane-based desalination processes and utilization of lower GHG emission renewable energy resources is presented. The review

Keywords:

covers the utilization of energy efficient design, high efficiency pumping, energy recovery

Reverse osmosis

devices, advanced membrane materials (nanocomposite, nanotube, and biomimetic),

Nanotechnology

innovative technologies (forward osmosis, ion concentration polarization, and capacitive

Renewable energy

deionization), and renewable energy resources (solar, wind, and geothermal). Utilization of

Energy recovery

energy efficient design combined with high efficiency pumping and energy recovery devices

Water sources

have proven effective in full-scale applications. Integration of advanced membrane materials and innovative technologies for desalination show promise but lack long-term operational data. Implementation of renewable energy resources depends upon geographyspecific abundance, a feasible means of handling renewable energy power intermittency, and solving technological and economic scale-up and permitting issues. ª 2011 Elsevier Ltd. All rights reserved.

Contents 1. 2.

Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Minimization of energy usage for desalination processes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.1. Enhanced system design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2. High efficiency pumping . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3. Energy recovery . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.4. Advanced membrane materials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.4.1. Nanocomposite membranes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

* Corresponding author. Tel.: þ1 626 568 6002; fax: þ1 626 568 6015. E-mail address: [email protected] (A. Subramani). 0043-1354/$ e see front matter ª 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.watres.2010.12.032

1908 1908 1908 1910 1910 1910 1911

1908

3.

4. 5.

1.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 0 7 e1 9 2 0

2.4.2. Nanotube membranes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.4.3. Biomimetic membranes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.5. Innovative technologies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.5.1. Forward osmosis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.5.2. Ion concentration polarization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.5.3. Capacitive deionization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Renewable energy utilization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.1. Solar energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.1.1. Solar thermal processes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.1.2. Solar electromechanical process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2. Wind energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3. Geothermal energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.4. Hybrid systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.5. Design and implementation of renewable energy systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Research needs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Introduction

In response to increasing water demand, many municipalities and water suppliers are considering more energy intensive seawater desalination to supplement inadequate freshwater sources (GWI, 2010). Desalination has been successfully implemented to provide additional water to communities experiencing shortages by applying processes developed over the last 40 years (Gleick, 2006). It is estimated that the capital expenditures for new desalination plants will exceed $17 billion by the year 2016 out of which more than $13 billion is expected to be targeted for seawater reverse osmosis (RO) (GWI, 2010). Desalination processes are broadly categorized as thermal or membrane-based technologies based on the separation process adopted (Greenlee et al., 2009). Although thermal desalination has remained the primary technology of choice in the Middle East, membrane processes have rapidly developed since the 1960’s (Loeb and Sourirajan, 1963) and currently surpass thermal processes in new plant installations (Greenlee et al., 2009). Membrane-based desalination technologies are favored over thermal-based desalination in regions where the cost of energy for steam production is high (NREL, 2006). The energy requirements for seawater desalination using thermal-based technologies are on the order of 7e14 kWh/m3 when compared to 2e6 kWh/m3 for membranebased technologies (Veerapaneni et al., 2007; Semiat, 2008; Anderson et al., 2010). The energy requirements are lower for RO, but energy consumption still remains the major operational cost component due to the high pressure pumps required to feed water to the RO process. These pumps are responsible for more than 40% of the total energy costs (Service, 2006; Souari and Hassairi, 2007). Reducing energy consumption is, therefore, critical for lowering the cost of desalination and addressing environmental concerns about GHG emissions from the continued use of conventional fossil fuels as the primary energy source for seawater desalination plants. A large number of energy minimization approaches

1911 1912 1912 1912 1913 1913 1913 1913 1913 1915 1915 1916 1916 1916 1917 1917 1917 1917

and renewable energy alternatives are rapidly being developed, investigated and implemented around the globe (Charcosset, 2009). Thus, providing an updated, comprehensive review of sustainable design and operational strategies to reduce energy usage and GHG emissions is warranted. This paper critically reviews in a holistic manner the latest developments and technologies for reducing energy consumption by reverse osmosis desalination processes and addresses strategies for integrating renewable energy as a source of alternative clean energy supply. The paper is organized by a discussion about energy minimization strategies for desalination followed by a discussion on the utilization of renewable energy resources to reduce GHG emissions.

2. Minimization of energy usage for desalination processes Factors influential in minimizing energy usage in desalination processes using RO membranes can be classified according to enhanced system design, high efficiency pumping, energy recovery, advanced membrane materials, and innovative technologies. Each of these factors is described in more detail below.

2.1.

Enhanced system design

The design and configuration of membrane units have a significant effect on the performance and economics of an RO plant (Wilf and Bartels, 2005). In the past, membrane units for seawater were usually configured as two stages with six elements per pressure vessel. The two-stage system resulted in a high feed and concentrate flow, which reduced concentration polarization at the expense of a greater feed pressure needed to compensate for the increased pressure drop across the RO train. Design efforts to reduce power consumption resulted in the use of single-stage configurations for high salinity feed water

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 0 7 e1 9 2 0

applications, and in some cases, the use of seven (or) eight elements per pressure vessel (Wilf and Bartels, 2005; Petry et al., 2007). The pressure drop reduction in using a single-stage rather than a two-stage system was reported to result in a 2.5% lower power requirement (Wilf and Bartels, 2005). More recently, further reduction in RO desalination cost has been shown to occur from optimal process configuration and control schemes. Theoretical cost minimization framework have been developed and experimentally implemented using a controller to quantify the effect of energy cost with respect to membrane cost, brine management cost, energy recovery, and feed salinity fluctuation (Zhu et al., 2009b, 2010). A control system utilizing real-time sensor data and userdefined permeate flow requirements has been implemented to compute in real-time the energy-optimal set-points for controlling concentrate valve position and feed flow rate (Bartman et al., 2009, 2010). Implementation of the control system demonstrated the ability to achieve energy-optimal operation of the RO system close to the theoretically predicted energy consumption curves. When stringent water quality requirements mandate the use of multi-pass RO (Sauvet-Goichon, 2007), the overall power consumption of the RO system can be lowered if a portion of the first pass permeate is pumped to the second pass (Zhu et al., 2009). Since the permeate produced from the front-end elements is lower in salinity than the permeate produced at the back-end elements, lower feed pressure is required for the second pass when the front-end permeate is utilized as feed to the second pass. In a multi-pass system, the lowest energy consumption is obtained when membranes with the highest salt rejection is used in the first-pass (Zhu et al., 2009a). In another study, various mixing operations between feed, concentrate, and permeate streams were evaluated to assess their potential on energy usage (Zhu et al., 2010a). It was determined that various mixing approaches may provide certain operational or system design advantages but they do not provide an advantage from an energy usage perspective. A novel design modification to reduce pressure drop across membrane elements is the use of a pressure vessel with a center port design (van Paassen et al., 2005). In this innovative configuration, feed water enters the pressure vessel through two feed ports on each end of the pressure vessel in the first stage. The concentrate is collected through a middle port and flows to a similar port on the pressure vessels in the second stage. Thus, the flow path is reduced by half and although the membrane unit has eight elements per pressure vessel, the flow path length is reduced to four elements per stage, creating a lower pressure drop that lowers the feed pressure. A 15% reduction in the feed pressure has been reported using the center port design when compared to a conventional side port design (Wilf and Hudkins, 2010). The disadvantage of the center port design is the potential for scaling due to excessive concentration polarization. Thus, pilot testing and long-term operational data are recommended before considering implementation of the center port design in order to determine the influence of water quality variations on feed water recovery. Reduction in energy consumption for RO systems treating high salinity feed water has also been achieved by using a twostage hybrid system with concentrate staging (Veerapaneni

1909

et al., 2005). The first stage consists of high rejection brackish water membrane elements (or) high permeability seawater membrane elements. The second stage consists of standard seawater elements. Using a two-stage system with brackish (or) low-pressure seawater membranes in the first stage lowers feed pressure requirements due to lower membrane resistance (Veerapaneni et al., 2007) As most of the permeate is produced in the first stage with the high permeability membranes, the pressure of only a small fraction of the remaining flow is boosted, resulting in significant energy savings. A two-pass nanofiltration (NF) membrane system also substantially reduces the energy consumption (Long, 2008). The power consumption of a two-pass NF process was estimated to be 2.06 kWh/m3 compared to 2.32 kWh/m3 for a two-stage hybrid brackish and seawater membrane system (Long, 2008). A 5% and 12% reduction in energy consumption was obtained when using a hybrid brackish/seawater or two-pass NF element system, respectively (Long, 2008). Energy consumption is also reduced by minimizing the pressure drop across membrane elements (Macedonio and Drioli, 2010). An approach by which to reduce the axial pressure drop in membrane elements involves the use of a novel feed spacer design that reduces the hydraulic pressure drop in the RO elements (Subramani et al., 2006; Guillen and Hoek, 2009). The feed spacer pattern used in most spiral wound membrane elements causes a variation in the flow path of the feed water resulting in a higher axial pressure drop than flow in an open channel (Guillen and Hoek, 2009). Although feed spacer geometry was found to have a marginal impact on mass transfer, thinner spacer filaments spread apart substantially reduced hydraulic pressure losses. In addition, certain non-circular spacer filament shapes produced lower hydraulic losses when compared to conventional circular spacer filament shapes (Guillen and Hoek, 2009). Although various feed spacer geometries have been shown to reduce hydraulic pressure loss in RO elements, actual data from pilot-scale and full-scale operation are still minimal since spiral wound elements with novel feed spacer configurations are not readily available. Commercialization of feed spacers that reduce the axial pressure drop across membrane elements could potentially reduce the feed pressure requirements during RO seawater desalination. A plant design approach for improving the economics of desalination and at the same time reduce the impact on environment due to brine discharge is the co-location of membrane desalination plants with existing coastal power generation stations (Voutchkov, 2004). In this approach, overall desalination power demand and associated costs of water production are reduced as a result of the use of warmer source water. The cooling water discharged from the condensers in a power plant is 5e15  C warmer that the source ocean water. When this water is used by the RO plant, 5e8% lower feed pressure is required to desalinate the water when compared to desalination of colder source ocean water. This approach also has the advantage of sharing a common intake facility. In the Middle East, RO and thermal-based technologies are combined to provide a hybrid design (Cardona and Piacentino, 2004). Such hybrid designs not only result in capital savings by sharing a common intake and outfall facility but also have a 40e50% increase in water production related to pre-heating of feed water to the RO plant.

1910

2.2.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 0 7 e1 9 2 0

High efficiency pumping

With respect to pumping, energy is predominantly consumed from operation of primary feed pumps, second pass feed pumps (as required), pretreatment pumps, product water transfer pumps, chemical feed pumps, and water distribution pumps. The distribution of power usage in a two-stage seawater RO system is shown in Fig. 1. More than 80% of the power is required for the operation of the primary feed pumps (Wilf and Bartels, 2005). Although the flow and head of a pumping system are determined by the design specifications of the RO system, the selection and operation of pumps and other elements of a pumping system play an important role in reducing overall energy usage in the plant. To achieve the highest possible pumping efficiency, several procedures are performed including: (1) verifying energy efficient operation of the pumping system, (2) utilizing a premium efficiency motor, and (3) utilizing a variable frequency drive (Manth et al., 2003). To achieve an energy efficient operation, a pump’s speed must fall within a specified range for optimal efficiency or the best efficiency point (Veerapaneni et al., 2007). The use of high speed and high flow pumps at lower total dynamic head provides the optimal speed needed for highest efficiency. To accommodate the variability of feed pressure with time (due to salinity and temperature fluctuations) without the necessity to throttle high pressure pumps or energy recovery devices, a variable frequency drive is often incorporated into the electric motor unit that drives the high pressure pump (Torre, 2008). All of the above mentioned pumping methods have been demonstrated to significantly improve efficiency and reduce energy requirements at full scale.

2.3.

Energy recovery

Energy consumption for RO desalination processes is reduced by using energy recovery devices (ERD) that have been shown to recover energy from the RO concentrate (Andrews and Laker, 2001; Wang et al., 2004). Before the concentrate stream is sent for disposal, pressure from the stream is recovered by passing it through an ERD. The fraction of power recovered depends on the type and efficiency of the equipment used. Two broad classes exist for ERDs (Wang et al., 2004). Class I devices use hydraulic power to cause a positive displacement within the recovery device, and the hydraulic energy is directly transferred in one step (Greenlee et al., 2009). Class II devices use the hydraulic energy of the RO concentrate in a two-step

Fig. 1 e Distribution of power usage in a two-stage seawater RO system (Wilf and Bartels, 2005).

process that first converts the energy to centrifugal mechanical energy and then back to hydraulic energy. Most of the seawater desalination plants in operation today use a Class I type of ERD (Greenlee et al., 2009). When an ERD is used, a fraction of the feed must bypass the primary high-pressure pump and a booster pump is used to account for pressure losses in the RO membrane modules, piping, and ERD (Greenlee et al., 2009). The pressure or work exchanger (PWE) is a Class I type of ERD. The pelton wheel, reverse running turbine pump, and turbo charger are examples of a Class II type of ERD. Efficiency greater than 95% can be achieved using a Class I type of ERD (Greenlee et al., 2009). The PWE transfers the hydraulic energy of the pressurized RO concentrate stream to the RO feed water stream (Avlonitis et al., 2003; Stover, 2007). PWE systems can be categorized as two types: those that provide a physical barrier (piston) between the RO concentrate stream and feed side of the system, such as a Dual Work Exchanger Energy Recovery (DWEER), and those without a physical barrier such as a Pressure Exchanger (PX) (Cameron and Clemente, 2008; Mirza, 2008). In the case of a DWEER, the system is based on moving pistons in cylinders which is well suited for a wide range of water viscosities and densities, but results in a large foot print (Mirza, 2008). A PX device has higher efficiency since no transformational losses occur in the device, but individual PX’s have limited flow rates and a higher capacity must be achieved by arranging several devices in series. PX devices are also associated with very high noise levels requiring a sound abatement enclosure (Mirza, 2008). Another disadvantage of a PX device is the degree of mixing that occurs between the feed water and concentrate stream. A feed salinity increase of 1.5%e3.0% caused by such mixing will increase the required feed pressure for the RO system (Wang et al., 2004, 2005). Class 2 Centrifugal ERDs (such as the pelton wheel and turbo charger) are limited in capacity and are usually optimized for narrow flow and pressure operating conditions (Stover, 2004, 2007). The turbo charger is typically used in smaller capacity RO installations (Oklejas et al., 2005). The reverse running turbine pump is not suitable for a low flow range due to poor efficiency (Mirza, 2008). The efficiency of commercial pelton wheels can reach 90% (Stover, 2007). The overall efficiency of the mechanically coupled reverse running turbine pump is in the 75%e85% range. For the submersible generator type, the overall efficiency is in the 62%e75% range (Mirza, 2008). The efficiency of the turbo charger ranges from 55% to 60%.

2.4.

Advanced membrane materials

Significant improvements in the salt rejection capacity and permeability of RO membranes for treating high salinity feed waters have been achieved in recent years. In 1980, seawater RO systems consumed more than 26 kWh/m3. Today, seawater RO systems consume on average only 3.4 kWh/m3 (Chang et al., 2008). The minimum theoretical energy use (50% recovery) is about 1.08 kWh/m3 for seawater desalination (Voutchkov, 2010). Thus, there are further avenues for improving the permeability of RO membranes using novel membrane materials such that the energy consumption is minimized. But, the new generation membranes must provide at least double the permeability of current generation RO membranes. This is

1911

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 0 7 e1 9 2 0

based on a recent approach to determine the minimization of energy costs by improving membrane permeability (Zhu et al., 2009c). A dimensionless factor was used to reflect the impact of feed water osmotic pressure, salt rejection requirement, membrane permeability, and purchase price of electrical energy and membrane module. It was estimated that unless the permeability of the RO membrane is doubled and the capital cost of pressure vessels directly impacted by a lower membrane area requirement, further improvements in seawater RO membrane permeability is less likely to significantly reduce the cost of desalination. New generation RO membrane which show promise in providing more than double the permeability of currently available RO membranes are discussed below. New generation RO membranes offer reduced feed pressure requirements while maintaining rejection. Today’s high productivity membrane elements are designed with two features that include more fresh water per membrane element and higher surface area and denser membrane packing (Voutchkov, 2007). New generation RO membranes can be broadly classified as nanocomposite, nanotube, and biomimetic membranes. A comparison of these advanced membranes is described in Table 1.

2.4.1.

Nanocomposite membranes

Thin film nanocomposite RO membranes are made by combining zeolite nanoparticles dispersed within a traditional polymide thin film (Jeong et al., 2007; Hoek and Ghosh, 2009). The zeolite nanoparticles are dispersed in one or more of the monomer solutions used to create the membrane by an interfacial polymerization process. Incorporation of zeolite nanoparticles in the polymer matrix of seawater RO membranes has enhanced flux to more than double that of a commercial product with 99.7% salt rejection. Utilization of nanocompositebased RO membranes can result in 20% lower energy consumption (NanoH2O, 2010). Although RO membranes using zeolite nanoparticles have been reported to show substantial reduction in the feed pressure requirement, long-term

operational data are still unavailable. Chemical stability of the incorporated zeolite nanoparticles within the polyamide matrix also needs to be demonstrated. If the nanocomposite membranes are not chemically compatible in a wide pH range, their applicability would be limited. Also, rejection of the nanocomposite membrane has been reported only for sodium chloride, magnesium sulfate, and polyethylene glycol (Jeong et al., 2007). Rejection of specific constituents, such as boron, in seawater by RO membranes has become a concern recently due to stringent discharge limits (Greenlee et al., 2009). The use of nanocomposite membranes for seawater desalination would require that more rejection data for specific constituents in seawater be available.

2.4.2.

Nanotube membranes

The use of carbon nanotubes have also been shown to consume lower energy than conventional seawater RO desalination systems (Truskett, 2003; Holt and Park, 2006; Sholl and Johnson, 2006; Corry, 2008; Jia et al., 2010). The transport of water and ions occurs through membranes formed using carbon nanotubes ˚ . Membranes incorporating ranging in diameter from 6 to 11 A carbon nanotubes are promising candidates for water desalination using RO since the size and uniformity of the tubes should achieve the desired salt rejection. A ten-fold permeability increase is expected using a carbon nanotube RO membrane which should result in a 30e50% savings in energy usage. Simulations have shown that boron nitride nanotubes have superior water flow properties to carbon nanotubes and they can achieve 100% salt rejection (Hilder et al., 2009). The use ˚ can also be used to functionalize of a nanotube radius of 4.14 A the membrane to become cation-selective. When a nanotube ˚ is used, the membrane can be functionalized to radius of 5.52 A become anion-selective (Hilder et al., 2009). Similar to the nanocomposite membrane, long-term operational data, scaleup information, chemical compatibility specifications and tensile strength of the nanotube-based RO membrane is lacking. Information available on the rejection property and water

Table 1 e Comparison of advanced material based reverse osmosis membranes. Membrane type

Principle

Energy consumption

Advantages

Drawbacks

Nanocomposite

Zeolite nanoparticles incorporated in polyamide matrix creating enhanced transport of water molecules.

20% lower energy consumption than conventional seawater RO (NanoH2O, 2010).

More than double the flux of currently available seawater RO membranes (NanoH2O, 2010).

Nanotube

Transport of water molecules through structured carbon and boron nitride nanotubes. Aquaporins used to regulate transport of water molecules.

30e50% lower energy consumption than conventional seawater RO (Hilder et al., 2009).

Ten e fold higher flux than currently available seawater RO membranes (Hilder et al., 2009). Hundred times permeable than currently available seawater RO membranes (AquaZ, 2010).

Chemical compatibility and structural stability is not known. Rejection of specific contaminants is not known. Long-term operational data not available. Only modeling results available. Rejection of specific contaminants is not known.

Biomimetic

Energy consumption is not known.

Inability to withstand high operating pressures. Rejection of specific contaminants is not known. Long-term operational data not available.

1912

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 0 7 e1 9 2 0

flux of the nanotube membrane is solely based on modeling results and a significant amount of research still needs to be conducted to determine if these nanotubes can be polymerized and casted to achieve the results demonstrated through modeling.

2.4.3.

Biomimetic membranes

New developments have also occurred with the use of biomimetic membranes for desalination (Bowen, 2006). Biomimetic membranes are designed to mimic the highly selective transport of water across cell membranes. Natural proteins known as aquaporins are used to regulate the flow of water providing increased permeability and high solute rejection. Aquaporins act as water channels which selectively allow water molecules to pass through while the transport of ions is restricted by an electrostatic tuning mechanism of the channel interior. This results in only water molecules being transported through the aquaporin channels and charged ions being rejected (Sui et al., 2001; Gong et al., 2007). Aquaporin membranes are considered to be a hundred times more permeable than commercial RO membranes with anticipated specific power consumption savings of 70% of specific power consumption (AquaZ, 2010). Highly permeable and selective membranes based on the incorporation of the functional water channel protein Aquaporin Z into a novel triblock copolymer has been shown to have significantly higher water transport than existing RO membranes (Kumar et al., 2007). A particular difficulty with biomimetic membranes that needs to be overcome is their inability to withstand high operating pressures.

2.5.

Innovative technologies

Innovative technologies utilizing the principles of separation technology with membranes, osmosis and electric fields have been introduced in recent years. These technologies have the potential to offer a substantial reduction in energy consumption for desalination. A comparison of these innovative

technologies is described in Table 2. The technologies are discussed below.

2.5.1.

Forward osmosis

In the forward osmosis process, instead of using hydraulic pressure similar to a conventional RO desalination process, a concentrated draw solution is used to generate high osmotic pressure, which pulls water across a semipermeable membrane from the feed solution (McCutcheon et al., 2005; Chou et al., 2010). The draw solutes are then separated from the diluted draw solution to recycle the solutes and to produce clean product water. A mixture of ammonia and carbon dioxide gas has been used as the predominant draw solution (McCutcheon et al., 2006). Other draw solutions utilized are salt solutions and magnetic nanoparticles (Yang et al., 2009). Forward osmosis is a combination of membrane and thermal processes. Specific energy consumption of less than 0.25 kWh/ m3 has been reported for the membrane portion of the process (Cath et al., 2006; McGinnis and Elimelech, 2007). In a recent study, a combination of forward osmosis and reverse osmosis was found to produce a higher flux than the forward osmosis process alone, thus reducing the specific energy consumption (Choi et al., 2009). But, the results were valid only for certain operating conditions and effect of membrane material on energy efficiency of the process is unknown. The regeneration of the draw solution requires significant amount of energy and unless waste heat is available for regeneration of the draw solution, the forward osmosis process is not considered more efficient than an RO process (McGinnis et al., 2007; Semiat, 2008). The process also has the advantage of a lower fouling propensity than RO as a result of the absence of hydraulic pressure and application of novel thin film composite membranes (Mi and Elimelech, 2010). A particular drawback of the forward osmosis process is the utilization of an appropriate membrane to reduce internal concentration polarization and increase efficiency (McGinnis and Elimelech, 2007). Although the forward osmosis process shows promise in terms of better performance with respect to fouling and scaling on the

Table 2 e Comparison of innovative desalination technologies with reverse osmosis. Technology

Principle

Energy consumption

Forward osmosis

Utilizes natural osmosis to dilute seawater feed stream using a draw solution with higher osmotic pressure than the seawater feed.

0.25e0.84 kWh/m3 (Cath et al., 2006; McGinnis and Elimelech, 2007)

Ion concentration polarization

Nanofluidics in combination with ion concentration polarization utilized to desalinate seawater. Ions electrosorbed by polarization of electrode (carbon aerogels) by a direct current power source.

3.5 kWh/m3 (Kim et al., 2010).

Capacitive deionization

1.37e1.67 kWh/m3 (brackish water) (Welgemoed, 2005). Energy consumption of seawater is not known.

Advantages Lower energy consumption than RO (McGinnis and Elimelech, 2007). Lower fouling potential than RO due to absence of transmembrane pressure (Mi and Elimelech, 2010). Lower energy consumption than RO. Absence of membranes and applied pressure (Kim et al., 2010). Lower energy consumption than RO for brackish water treatment (Oren, 2010). Absence of membrane and applied pressure.

Drawbacks More applicable than RO only when waste heat source is available (McGinnis and Elimelech, 2007). Full-scale operation data is not available. Process suited for small and medium e scale systems. Fullscale operational data is not available. Low feed water recovery (Oren, 2010). Full-scale operational data not available.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 0 7 e1 9 2 0

membrane surface, the rejection capability of the membranes for specific contaminants such as boron, which is a design limiting factor for seawater desalination, is unknown. Studies have been conducted to improve the membrane property used for forward osmosis (Yang et al., 2009) but long-term operational data for the forward osmosis process is unavailable and only bench-scale results have been published.

achievable is very low (Oren, 2008). Capacitive deionization has primarily been used for brackish water desalination but with improvements in electrode material and better process control strategies, the technology holds promise for seawater desalination (Oren, 2008; Anderson et al., 2010).

3. 2.5.2.

Renewable energy utilization

Ion concentration polarization

Ion concentration polarization has been utilized to desalinate seawater using an energy efficient process (Kim et al., 2010). In this process, micro and nanofluidics in combination with ion concentration polarization are used to desalinate seawater. Ion concentration polarization is a fundamental transport mechanism that occurs when an ionic current is passed through an ion-selective membrane. But, in the newly developed process, no membranes are utilized. An electrical potential is used to create a repulsion zone that acts as a membrane in separating charged ions, bacteria, viruses, and microbes from seawater flowing through a 500 mm  100 mm microchannel. Water flows through the microchannel tangential to a nanochannel where the voltage is applied. The resulting force creates a repulsion zone and the stream splits into two smaller channels at a nanojunction. The two streams created are the treated water and concentrate. More than 99% salt rejection and 50% recovery has been reported using this process (Kim et al., 2010). The ion concentration polarization process has been reported to consume approximately 3.5 kWh/ m3, which is comparable to seawater desalination using RO membranes (Kim et al., 2010). The advantage of ion concentration is that the process is fouling free since membranes are not used. The process is best suited for small to medium scale systems with the possibility of battery-powered operation. The process is not efficient in the removal of neutrally charged organics and needs to be combined with other processes to achieve treatment goals. Results from the ion concentration polarization are not available for large-scale systems and only modeling and bench-scale results have been reported.

2.5.3.

1913

Capacitive deionization

Capacitive deionization technology is not a recent development, but several challenges with the identification of an optimum material for the manufacturing of the associated electrodes have delayed commercialization (Farmar et al., 1997; Dermentzis and Ouzounis, 2008; Lee et al., 2009). This technology was developed as a non-polluting, energy efficient and cost effective alternative to desalination technologies such as RO (Welgemoed, 2005). In this technology, a saline solution flows through an unrestricted capacitor type module consisting of numerous pairs of high-surface area electrodes. The electrode material, typically a carbon aerogel, contains a high specific surface area (400e1100 m2 per g) and a very low electrical resistivity. Anions and cations in solution are electrosorbed by the electric field upon polarization of each electrode pair by direct current power sources. For desalination of brackish water, energy consumption of 1.37e1.67 kWh/m3 has been reported using this technology (Welgemoed, 2005). Energy consumption for high salinity waters (such as seawater) is not readily available in the literature. The main drawback with capacitive deionization is that the feed water recovery

Renewable energy resources are the best energy supply option for stand-alone desalination systems in remote regions where energy supply from an electricity grid is not readily accessible (Schafer et al., 2007; Gude et al., 2010). In urban regions, renewable energy may provide treatment cost reductions due to the implementation of a diversified portfolio of energy sources and reduces GHG emissions. For an RO system used to desalinate seawater with traditional fossil fuel based energy source, CO2 emissions of 1.78 kg/m3 of desalted water and NOx emissions of 4.05 g/m3 of desalted water have been reported (Raluy et al., 2005). When RO was operated with electricity generated from wind or solar energy, GHG emissions were substantially lower. For RO integrated with wind energy resource, CO2 emissions were 0.1 kg/m3 and NOx emissions were 0.4 g/m3 (Raluy et al., 2005). For RO integrated with solar photovoltaic energy resource, CO2 emissions were 0.6e0.9 kg/ m3 and NOx emissions were 1.8e2.1 g/m3 (Raluy et al., 2005). Thus, an important avenue for reducing GHG emissions is the utilization of renewable energy sources in place of fossil fuels (NAS, 2010). A comparison of renewable energy resources is shown in Table 3 and is discussed in detail below.

3.1.

Solar energy

Solar energy is one of the most promising sources of renewable energy. The quantity of solar energy received by earth is a function of the season, with the highest quantity of incoming solar energy received during the summer months (Kiehl and Trenberth, 1997). Desalination using solar energy can be categorized as thermal and electromechanical processes. Thermal processes use solar thermal energy whereas electromechanical processes use photovoltaic cells.

3.1.1.

Solar thermal processes

Solar thermal desalination processes are characterized as either direct or indirect processes (Qiblawey and Banat, 2008). Direct processes consist of all parts integrated into one system whereas indirect processes refer to heat coming from a separate solar collecting device such as solar collectors or solar ponds. The predominant solar thermal processes that are integrated with or used as desalination systems are solar stills and solar ponds. Utilization of solar energy for desalination is described in detail below.

3.1.1.1. Solar stills. A solar still is a simple device that is used to convert saline water into drinking water (Qiblawey and Banat, 2008). Solar stills are used for direct solar desalination which is mainly suited for small production systems and regions where the freshwater demand is less than 200 m3 per day (Rodriguez, 2002). The solar still consists of a transparent cover (glass or plastic) which encloses saline water. The principle of operation

1914

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 0 7 e1 9 2 0

is evaporation and condensation. The solar still cover traps solar energy within the enclosure. The trapped solar energy heats the saline water causing evaporation and condensation on the inner surface of the sloped transparent cover. As the saline water is heated, its vapor pressure increases. The resultant water vapor is condensed on the underside of the roof cover and runs down into troughs, which collect the distillate. The distillate obtained is of high quality with most salts, organic and inorganic components removed. The solar still requires frequent flushing to prevent salt precipitation. Design problems encountered with solar stills are brine depth, vapor tightness of the enclosure, distillate leakage, methods of thermal insulation, and cover slope, shape, and material (Eibling and Talbert, 1971). In practice, heat losses will occur through a still. Currently available stateof-the-art single-effect solar stills have an efficiency of approximately 30e40% (Rodriguez, 2002). The solar still of the basin type is the oldest method and improvements in its design have been made to increase its efficiency (Naim et al., 2003). Modifications using passive methods include basin stills, wick stills, diffusion stills, stills integrated with greenhouses, and other configurations (Ahsan et al., 2010; Tabrizi et al., 2010; Murugavel and Srithar, 2011). A major drawback with the solar still is the energy loss in the form of latent heat of condensation. In order to solve this problem, a humidification-dehumidification (HD) principle has been developed (Mathioulakis et al., 2007). Solar desalination based on the HD principle results in an increase in the overall efficiency of the desalination plant and is therefore considered a promising technique for small capacity, solar-driven desalination plants (Mathioulakis et al., 2007). The basic principle of the HD process is the evaporation of high salinity water and condensation of water vapor from the humid air at ambient pressure. The HD process is based on the fact that air can be mixed with significant quantities of vapor. The vapor carrying

capability of air increases with temperature. Fresh water is produced by condensing out the water vapor, which results in dehumidification of the air. A significant advantage of the HD technology is that it provides a means for low pressure, low temperature desalination that operates off of waste heat and is potentially very cost competitive (Parekh et al., 2004). Since the heat transfer coefficient of the condensing vapor from air is much lower than for pure water, the heat transfer area needed is enormously high and a disadvantage for the process (Semiat, 2008).

3.1.1.2. Solar ponds. Solar ponds utilize a saline gradient to reduce the heat loss that normally occurs when the less dense water heated by the sun rises to the surface of a pond and loses energy to the atmosphere by convection and radiation (Kalagirou, 2005). The objective of the solar pond is to utilize a saline gradient to combat the thermal gradient and create a stagnant and insulating zone in the upper part of the pond that traps the hot fluid in the lower section of the pond. New technologies combining solar thermal energy and desalination have been shown to utilize up to 80% less energy than conventional desalination technologies (Saltworks, 2010). In this thermoionic desalination technology, the energy consumption is reduced by harnessing low temperature heat to overcome the energy barrier for desalination. Salt water is evaporated to produce a concentrated salt solution. The concentrated gradient energy from the concentrated salt solution is then used to operate a desalination system. Using desalination brine for solar ponds not only provides a preferable alternative to environmental disposal, but also a convenient and inexpensive source of solar pond salinity. 3.1.1.3. Concentrated solar power. A new study has estimated that 25% of the world’s electricity could come from concentrated

Table 3 e Comparison of renewable energy resources for desalination. Renewable energy resource Solar

Wind

Geothermal

Application

Advantages

Disadvantages

Solar still: Direct conversion of saline to potable water. Solar pond: Utilization of salinity gradient to store heat and produce steam for electricity generation. Concentrated solar power: Hot fluid used in turbine generator for producing electricity. Photovoltaic cell: Conversion of sunlight directly into electricity to power RO desalination. Wind turbine: Wind energy used to generate electricity to power RO desalination. Geothermal steam to generate electricity to power RO desalination.

Simple process. Inexpensive material of construction can be utilized (Qiblawey and Banat, 2008). Beneficial use of desalination brine (Qiblawey, 2008). Same equipment used in conventional power plants can be used for concentrated solar power plants (DOE, 2010). Hybrid designs with other (wind) renewable energy sources are easily achievable. Well suited for desalination plants requiring electrical power (Eltawil et al., 2009). Well suited for desalination plants requiring electrical power (Eltawil et al., 2009). Continuous power output, predictable resource, thermal storage is not necessary (Barbier, 2002; EGEC, 2010).

Energy loss in the form of latent heat of condensation (Mathioulakis et al., 2007). Large land area requirement (Kalagirou, 2005). Capital cost intensive. Output is intermittent (Trieb et al., 2009). Large land area requirement. Capital cost intensive. Output is intermittent (Kalagirou, 2005).

Output is intermittent. Resource is location dependant and unpredictable (Kalagirou, 2005). Resource is limited to certain locations (Kalagirou, 2005).

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 0 7 e1 9 2 0

solar power (CSP) by the year 2050 (Trieb et al., 2009). Most commercial CSP facilities use a system of curved mirrors to collect the sun’s energy to heat a fluid flowing through tubes. The hot fluid is then used to boil water in a conventional steamturbine generator to produce electricity. Concentrating solar power typically uses a Dish/Sterling system. Other methods to concentrate solar energy utilize a parabolic trough, solar tower, or linear Fresnel (Trieb et al., 2009). Large mirror fields concentrate the sunlight to produce high temperature steam for power generation that can be used for seawater desalination. Part of the harvested solar thermal energy is used during the day and conventional electricity is used during the night for continuous operation.

3.1.2.

1915

Kalagirou, 2005). When electrical loads require an alternating voltage, an inverter is used to transform direct current into alternating current. The batteries allow operation at constant flow and pressure. They are sized to stabilize the power supply to the RO unit on a daily basis, as well as to account for fluctuations in solar energy and water demand. Battery less PV-powered RO systems have been tested before (Thomson and Infield, 2003) but certain disadvantages such as longer operation in ‘stand by’ mode needs to be overcome. It is also common practice to connect PV systems to the local electricity grid. During the day, the energy generated from the PV systems is used directly from the grid and power is utilized from the electricity grid at night. Thus, the grid acts as an energy storage system.

Solar electromechanical process

Desalination using an electromechanical process involves the application of photovoltaic (PV) cells. The PV process converts sunlight directly into electricity. A PV cell consists of two or more thin layers of semiconducting material (mostly silicon). When the semiconducting material is exposed to sunlight, electrical charges are generated and are conducted away by metal contacts as direct current (DC). The PV sector has been growing at 20% per annum or more for several years and is now a multi-billion dollar business in Europe (Infield, 2009). A total of around 5.95 GW of capacity has been installed worldwide. Photovoltaic cells are either monocrystalline silicon cells, muticrystalline silicon cells, or amorphous cells. Monocrystalline cells are made of very pure monocrystalline silicon whereas multicrystalline cells are produced using numerous grains of monocrystalline cells (Kalagirou, 2005). The PV panel is the principle building block of a PV system and any number of panels can be connected together to give the desired electrical output. The choice of photovoltaic active material has important effects on system design and performance. Both the composition and atomic structure of the cell is an important consideration for assessing performance. Materials other than silicon which are used for photovoltaic cells are silver, cadmium telluride, and copper indium selenide (Kalagirou, 2005; Jacobson and Delucchi, 2009). The advantage of using cadmium-based and copper-based material for PV cells is that they are manufactured by relatively inexpensive industrial processes when compared to crystalline silicon technologies (Kalagirou, 2005). Limited supplies of tellurium and indium could reduce the prospects for some types of thinfilm solar cells and large-scale production could be restricted by the availability of silver (Jacobson and Delucchi, 2009). The most popular combination of PV cells is with RO. PV is considered a proper solution for small applications in sunny areas (less than 50 m3/d) (Mathioulakis et al., 2007). The feasibility of PV-powered RO for desalination at remote sites is a proven combination. Since both technologies are fairly mature, PV-powered RO requires minimum maintenance (Bayod-Rujula and Martinez-Gracia, 2009). Stand-alone PV systems are used in areas that are not easily accessible to electricity. A stand-alone system is independent of the electricity grid, with the energy produced being stored in batteries (Bayod-Rujula and Martinez-Gracia, 2009). Typically, a standalone PV-powered RO system will consist of a module, batteries, and charge controller (Bouguecha et al., 2005;

3.2.

Wind energy

Wind has re-emerged as one of the most important and fastest growing sustainable energy resources since wind turbines were first commercialized in the 1970s (Garcı´a-Rodrı´guez, 2002; Ackermann and Soder, 2002). Wind turbines are mature technologies and are commercially available. Windpowered desalination represents one of the most promising renewable energy options for desalination, especially for coastal areas with high availability of wind energy resources (Zejli et al., 2004; Forstmeier et al., 2007). After solar energy, wind energy is the most widely used renewable energy source for small capacity desalination plants (Kalagirou, 2005). The two common approaches for wind-powered desalination systems include connecting both the wind turbines and the desalination system to a grid, or direct coupling of the wind turbines with the desalination system (Ackermann and Soder, 2002). The latter option is likely to be a stand-alone system at remote locations which have no electricity grid. In this case, the desalination system may be affected by power variations and interruptions caused by the power source (wind). Hence, the stand-alone wind desalination systems are often hybrid systems, combined with another type of renewable energy source (for instance solar), or a back-up system such as batteries or diesel generators (Mathioulakis et al., 2007). For stand-alone wind energydriven desalination units, the reported cost of fresh water produced ranged from $1.35 per m3 to $6.7 per m3 when compared to RO cost of about $1.0 per m3 (Karagiannis and Soldatos, 2008; Mezher et al., 2011). The primary concern with the use of wind energy for desalination is that wind speed is highly variable. Wind speed varies both geographically and temporally and varies over a multitude of temporal and spatial time scales (NREL, 2006). In terms of using a wind turbine to generate power for desalination, this variation is amplified by the fact that the available energy in the wind varies as the cube of the wind speed (NREL, 2006). Thus, the choice of location of the wind farm is critical for the exploitation of wind resources for power generation to ensure superior economic performance. The theoretical maximum aerodynamic conversion efficiency from wind to mechanical power for wind turbines is 59% (Kalagirou, 2005). The need to economize on blade costs tends to lead to the construction of slender bladed, fast running wind turbines with peak efficiencies close to 45%.

1916

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 0 7 e1 9 2 0

Strategies for improving wind turbine performance include the utilization of lighter turbine blades and improving power storage systems (Spang, 2006; Thumthae and Chitsomboon, 2009). Wind turbines with blades that are approximately 40% lighter than standard turbine blades could reduce capital costs by up to 20e25% (Fairley, 2002). Lighter wind turbine design with the blades placed on hinges allows flexibility for reduced drag during high winds (Spang, 2006; Ameku et al., 2008). Utilization of two blades per turbine rather than three blades leads to lighter turbine design, but, the three-blade wind turbines have a lower noise level than two-blade wind turbines and the rotor moment of inertia is easier to handle. Wind generators with vertical axis turbines convert wind energy into electrical energy at a greater efficiency than horizontal axis turbines and result in approximately 5% increase in energy production and significantly reduce the investment cost per kWh (Spang, 2006). With respect to the materials available for constructing wind turbines, enough concrete and steel exist and both these commodities are fully recyclable (Jacobson and Delucchi, 2009). Increased production of wind turbines in the future could be slowed due to the availability of rare-earth metals such as neodymium, but low cost sources available in China could be utilized to compensate for a shortage of essential metals for manufacturing wind turbines (Jacobson and Delucchi, 2009).

3.3.

Geothermal energy

Geothermal energy sources are classified in terms of the measured temperature as low (<100  C), medium (100e150  C) and high temperature (>150  C). Geothermal energy is usually extracted with ground heat exchangers (Kalagirou, 2005). The primary advantage of geothermal energy compared to solar and wind, is that it is both continuous and predictable; as such, thermal storage is unnecessary (Barbier, 2002). Geothermal energy is being evaluated for desalination in Queensland, Australia (Queensland Geothermal Energy Center of Excellence, 2010). This energy center has estimated that a geothermal plant in the 1000e100,000 m3/d capacity can easily provide the entire fresh water needs for an outback city at the cost of around $0.73e1.46/m3. The first installation of geothermal energypowered desalination plants was reported by the Bureau of Reclamation of the United States Department of the Interior in the 1970s (Awerbuch et al., 1976). Geothermal energy has also been investigated for desalination recently on the Baja California peninsula (Hiriart, 2008). Geothermal energy presents a mature technology which can be used to provide energy for desalination systems. High temperature geothermal fluids generate electricity to power RO plants and are used directly as shaft power for mechanically driven desalination. The main advantage of using geothermal energy for desalination is that it is a stable and reliable heat supply 24 h a day, 365 days a year. Also, desalination using geothermal energy source is environmentally friendly with no emission of greenhouse gases (EGEC, 2010).

3.4.

Hybrid systems

Combinations of wind and solar energy have been used for driving desalination systems (Koutroulis and Kolokotsa, 2010;

Karellas et al., 2011). The purpose of using hybrid wind-solar systems for desalination is based on the fact that in certain locations, wind and solar time profiles do not coincide (Mathioulakis et al., 2007). The complementary features of wind and solar resources make use of hybrid wind-solar systems to drive desalinations systems (Charcosset, 2009). Hybrid windsolar PV systems have been implemented in the Sultanate of Oman, Israel, Mexico, Germany, and Italy (Petersen et al., 1981; Al Malki et al., 1998; Weiner et al., 2001; Pretner and Iannelli, 2002). Two RO desalination plants supplied by a 6 kW wind energy converter and a 2.5 kW solar generator have been designed for remote areas (Petersen et al., 1981). Stand-alone systems for seawater desalination using hybrid wind-PV system have also been designed (Mohamed and Papadakis, 2004). Using wind and solar conditions in Eritrea, East Africa, the hourly water production was determined to be 35 m3/d with a specific energy consumption of about 2.33 kWh/m3 (Gilau and Small, 2008). Although several studies have been performed using hybrid renewable energy desalination systems, none of them represent large-scale applications.

3.5. Design and implementation of renewable energy systems Renewable energy desalination systems need to be designed using an iterative approach (Voivontas et al., 2001). The first step of the approach involves the definition of a list of alternative technologies that satisfy the water demand. A second step focuses on a detailed design analysis of each candidate option made to determine the plant capacity, the structure of the power unit and the operational characteristics. The final step involves a financial analysis of the investment associated with the selected renewable energy-desalination combination. The most challenging issue associated with the implementation of renewable energy-desalination technology is the optimum matching of the intermittent renewable energy power output with the steady energy demand of the desalination process. Power supply management and demand-side management are considered as the two options available to address this problem (Voivontas et al., 2001). In the first case, an appropriately controlled hybrid renewable energy resource unit that is capable of providing a steady energy output is used. This unit is sized at the nominal power demand of the desalination process. In the demand-side management option, the desalination process only operates when the energy output of the renewable energy resource unit is able to cover the energy demand. Other options available to address the issue of intermittent renewable energy power output are different types of energy storage such as electro-mechanical, virtual (through process modification), and grid energy (Kalogirou, 1997). Compressed air energy storage plants have also been used when energy produced from a wind turbine exceeds grid load capacity (BINE, 2010). For limited periods, the compressed air stores cover the short-term reserve requirement, which are needed due to the unpredictable forecasts of wind power feeding the grid. In this case, wind turbines do not have to deactivate in the event of a grid overload, and if there is excess supply of electrical energy, the storage technology refines base-load electricity, converting it to peak-load electricity (BINE, 2010).

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 0 7 e1 9 2 0

The selection of the appropriate renewable energy resource depends on several factors including plant size, feed water salinity, remoteness, availability of grid electricity, technical infrastructure, and the type and potential of the local renewable energy resource and storage options. In addition, socio-economic factors and policy need to be considered as a driver for renewable energy resource implementation. The applicability of renewable energy resources for desalination strongly depends on the local availability of renewable energy and the quality of water required after treatment. In addition, some combinations of resources are better suited for large size plants, whereas some others are better suited for small scale applications. Other important factors that need to be considered are the capital cost of the equipment and the land area required for the equipment installation. When considering resource availability, solar thermal energy and photovoltaics are considered to be a better choice over wind and geothermal energy which are locationdependant. When considering the continuity and predictability of power output, geothermal energy is the most reliable resource as the output is intermittent and less predictable for solar thermal, photovoltaic, and wind energy. Renewable energy resources provide various advantages but their application for desalination has been limited. The reasons for their limited application are technology, cost, and availability. Although desalination technologies are mature, technologies for the storage of renewable energy are not completely mature and avenues for design improvements still exist. Prices for renewable energy technologies are decreasing but the capital costs still prohibit their commercialization at a large-scale. Renewable energy is also not sufficiently available in certain locales and for regions with adequate supplies, a lack of adequate storage strategies has sometimes impeded development due to supply intermittency.

4.

1917

With respect to integrating renewable energy resources with desalination systems in order to reduce GHG emissions, research is still needed to bring down the cost of these alternative energy supplies, establish appropriate storage systems to smooth out resource intermittency, and incorporate renewable sources into the electric grid for utilization by desalination facilities in regions with poor renewable energy resources.

5.

Conclusions

Minimization of energy utilization by RO desalination can be achieved in several ways. One promising avenue for reducing desalination energy is through the use of advanced membrane materials and application of innovative technologies. Innovative technologies such as forward osmosis and ion concentration polarization show promise but long-term operational data are lacking. Integration of renewable energy resources is needed to reduce GHG emissions and eliminate desalination’s dependency on fossil fuels and ultimately reduce the cost of energy. The selection of appropriate renewable energy resources depends on factors such as plant size, feed water salinity, plant location, availability of grid electricity, technical infrastructure, and the availability of local renewable energy resources, and storage options. Technology advances are important, but economic and political factors are also critical to large-scale deployment of renewable energy.

Acknowledgements The authors would like to thank the WateReuse Research Foundation (WRRF) and the California Energy Commission (CEC) for project funding (Project # WRRF-08 13).

Research needs

Minimization of the energy required for seawater RO desalination through utilization of efficient system design, high efficiency pumping, and energy recovery devices has been studied extensively and near optimal performance characteristics have already been tested and achieved. Further design improvements in these categories will only provide marginal reduction in energy consumption further. Research avenues that show the most promise for reducing energy usage lie in the development and testing of advanced membrane materials which can enhance the performance of the membrane, in terms of flux and rejection, and reduce feed pressure requirements. The development of nanocomposite, nanotube, and biomimetic membranes show promise but much more data are necessary in order to validate the application of these membranes under normal operation and chemical cleaning conditions. Innovative technologies, such as forward osmosis, require a more efficient recovery of the draw solution and methods to reduce internal concentration polarization, inorganic scaling, and fouling of the membrane. Similarly, the application of ion concentration polarization and capacitive deionization technologies requires further study in order to enhance feed water recovery to the point where this technology might be economically feasible.

references

Ackermann, T., Soder, L., 2002. An overview of wind energystatus. Renewable and Sustainable Energy Reviews 6, 67e128. Ahsan, A., Islam, K.M., Fukuhara, T., Ghazali, A.H., 2010. Experimental study on evaporation, condensation and production of a new tubular solar still. Desalination 260 (1e3), 172e179. Al Malki, A., Al Amri, M., Al Jabri, H., 1998. Experimental study of using renewable energy in the rural areas of Oman. Renewable Energy 14, 319e324. Ameku, K., Nagai, B.M., Roy, J.N., 2008. Design of a 3 kW wind turbine generator with think airfoil blades. Experimental Thermal and Fluid Science 32 (8), 1723e1730. Anderson, M.A., Cudero, A.L., Palma, J., 2010. Capacitive deionization as an electrochemical means of saving energy and delivering clean water. Comparison to present desalination practices: will it compete? Electrochemica Acta 55, 3845e3856. Andrews, W.T., Laker, D.S., 2001. A twelve-year history of large scale application of work exchanger energy recovery technology. Desalination 138 (1e3), 201e206. AquaZ, 2010. http://www.danfoss-aquaz.com/files/billeder/ Aquaporin_membrane.JPG (accessed 09.30.10).

1918

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 0 7 e1 9 2 0

Avlonitis, S.A., Kouroumbas, K., Vlachakis, N., 2003. Energy consumption and membrane replacement cost for seawater RO desalination plants. Desalination 157, 151e158. Awerbuch, L., Lindemuth, T.E., May, S.C., Rogers, A.N., 1976. Geothermal energy recovery process. Desalination 19, 325e336. BINE, 2010. Compressed air energy storage power plants. Available from: http://www.bine.info/fileadmin/content/ Publikationen/Englische_Infos/projekt_0507_engl_internetx. pdf (accessed 09.30.10). Barbier, E., 2002. Geothermal energy technology and current status: an overview. Renewable and Sustainable Energy Reviews 6, 3e65. Bartman, A.R., Christofides, P.D., Cohen, Y., 2009. Non-linear model-based control of an experimental reverse-osmosis water desalination system. Industrial and Engineering Chemistry Research 48, 6126e6136. Bartman, A.R., Zhu, A., Christofides, P.D., Cohen, Y., 2010. Minimizing energy consumption in reverse osmosis membrane desalination using optimization-based control. Journal of Process Control 20, 1261e1269. Bayod-Rujula, A.A., Martinez-Gracia, A., 2009. Photovoltaic system for brackish water desalination by electrodialysis and electricity generation. Desalination and Water Treatment 7, 142e151. Bouguecha, S., Hamrouni, B., Dhahbi, M., 2005. Small scale desalination pilots powered by renewable energy sources: case studies. Desalination 183, 151e165. Bowen, R., 2006. Biomimetic separations e learning from the early development of biological membranes. Desalination 199, 225e227. Cameron, I.B., Clemente, R.B., 2008. SWRO with ERI’s PX pressure exchanger device-a global survey. Desalination 221, 136e142. Cardona, E., Piacentino, A., 2004. Optimal design of cogeneration plants for seawater desalination. Desalination 166, 411e426. Cath, Z., Childress, A., Elimelech, M., 2006. Forward osmosis: principles, applications, and recent developments. Journal of Membrane Science 281 (1e2), 70e87. Chang, Y., Reardon, D.J., Kwan, P., Boyd, G., Brant, J., Rakness, K., Furukawa, D., 2008. Evaluation of dynamic energy consumption of advanced water and wastewater treatment technologies. AWWARF Final Report, ISBN 978-1-60573-033-2. Charcosset, C., 2009. A review of membrane processes and renewable energies for desalination. Desalination 245, 214e231. Choi, Y., Choi, J., Oh, H., Lee, S., Yang, D., Kim, J., 2009. Toward a combined system of forward osmosis and reverse osmosis for seawater desalination. Desalination 247 (1e3), 239e246. Chou, S., Shi, L., Wang, R., Tang, C.Y., Qiu, C., Fane, A.G., 2010. Characteristics and potential applications of a novel forward osmosis hollow fiber membrane. Desalination 261 (3), 365e372. Corry, B., 2008. Designing carbon nanotube membranes for efficient water desalination. The Journal of Physical Chemistry B 112 (5), 1427e1434. Dermentzis, K., Ouzounis, K., 2008. Continuous capacitive deionization e electrodialysis reversal through electrostatic shielding for desalination and deionization of water. Electrochimica Acta 53 (24), 7123e7130. Eibling, J.A., Talbert, S.G., 1971. Solar stills for community use e digest of technology. Solar Energy 13, 263e276. Eltawil, M.A., Zhengming, Z., Yuan, L., 2009. A review of renewable energy technologies integrated with desalination systems. Renewable and Sustainable Energy Reviews 13, 2245e2262. European Geothermal Energy Council (EGEC), 2010. Geothermal desalination. Available from: http://www.egec.org/target/ Brochure%20DESALINATION.pdf (accessed 07.20.10). Fairley, P., 2002. Wind Power for Pennies. MIT Technology Review, Cambridge, MA.

Farmar, J.C., Tran, T.D., Richardson, J.H., Fix, D.V., May, S.C., Thomson, S.L., 1997. The application of carbon aerogel electrodes to desalinate and waste treatment. Report No. 231717. Lawrence Livermore National Laboratory. Forstmeier, M., Mannerheim, F., D’Amato, F., Shah, M., Liu, Y., Baldea, M., Stella, A., 2007. Feasibility study on wind-powered desalination. Desalination 203, 463e470. Garcı´a-Rodrı´guez, L., 2002. Seawater desalination driven by renewable energies: a review. Desalination 143, 103e113. Gilau, A.M., Small, M.J., 2008. Designing cost-effective seawater reverse osmosis system under optimal energy options. Renewable Energy 33, 617e630. Gleick, P.H., 2006. The World’s Water 2006e2007, The Biennial Report on Freshwater Resources. Island Press, Chicago. Global Water Intelligence (GWI), 2010. Desalination Markets, 2010. Gong, X., Li, J., Lu, H., Wan, R., Li, J., Hu, J., Fang, H., 2007. A chargedriven molecular water pump. Nature Nanotechnology 2, 709e712. Greenlee, L.F., Lawler, D.F., Freeman, B.D., Marrot, B., Moulin, P., 2009. Reverse osmosis desalination: water sources, technology, and today’s challenges. Water Research 43, 2317e2348. Gude, V.G., Nirmalakhandan, N., Deng, S., 2010. Renewable and sustainable approaches for desalination. Renewable and Sustainable Energy Reviews 14 (9), 2641e2654. Guillen, G., Hoek, E.M.V., 2009. Modeling the impacts of feed spacer geometry on reverse osmosis and nanofiltration processes. Chemical Engineering Journal 149 (1e3), 221e231. Hilder, T.A., Gordon, D., Chung, S., 2009. Salt rejection and water transport through boron nitride nanotubes. Small 5 (19), 2183e2190. Hiriart, G., 2008. Geothermal energy for desalination seawater. International Geological Congress, Oslo. Hoek, E.M.V., Ghosh, A., 2009. Nanotechnology-based membranes for water purification. Nanotechnology Applications for Clean Water, 47e58. Holt, J., Park, H.G., 2006. Fast mass transport through sub 2nanometer carbon nanotubes. Science 312 (5766), 1034e1037. Infield, D.G., 2009. An overview of renewable energy technologies with a view to stand alone power generation and water provision. Desalination 248, 494e499. Jacobson, M.Z., Delucchi, M.A., 2009. A Path to Sustainable Energy. Scientific American, pp. 58e65. Jeong,B.H.,Hoek,E.M.V.,Yan,Y.,Huang,X.,Subramani,A.,Hurwitz,G., Ghosh, A.K., Jawor, A., 2007. Interfacial polymerization of thin film nanocomposites: a new concept for reverse osmosis membranes. Journal of Membrane Science 294, 1e7. Jia, Y., Li, H., Wang, M., Wu, L., Hu, Y., 2010. Carbon nanotube: possible candidate for forward osmosis. Separation and Purification Technology 75 (1), 55e60. Kalagirou, S.A., 2005. Seawater desalination using renewable energy sources. Progress in Energy and Combustion Science 31, 242e281. Kalogirou, S.A., 1997. Survey of solar desalination systems and system selection. Energy e The International Journal 22, 69e81. Karagiannis, I.C., Soldatos, P.G., 2008. Water desalination cost literature: review and assessment. Desalination 223, 448e456. Karellas, S., Terzis, K., Manolakos, D., 2011. Investigation of an autonomous hybrid solar thermal ORC e PV RO desalination system. The Chalki island case. Renewable Energy 36 (2), 583e590. Kiehl, J.T., Trenberth, K.E., 1997. Earth’s annual global mean energy budget. Bulletin of the American Meteorological Association 78, 197e208. Kim, S.J., Ko, S.H., Kang, K.H., Han, J., 2010. Direct seawater desalination by ion concentration polarization. Nature Nanotechnology 5, 297e301. Koutroulis, E., Kolokotsa, D., 2010. Design optimization of desalination systems power-supplied by PV and W/G energy sources. Desalination 258 (1e3), 171e181.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 0 7 e1 9 2 0

Kumar, M., Grzelakowski, M., Zilles, J., Clark, M., Meier, W., 2007. Highly permeable polymeric membranes based on the incorporation of the functional water channel protein Aquaporin Z. Proceedings of the National Academy of Sciences 104 (52), 20719e20724. Lee, J., Park, K., Yoon, S., Park, P., Park, K., Lee, C., 2009. Desalination performance of a carbon-based composite electrode. Desalination 237 (1e3), 155e161. Loeb, S., Sourirajan, S., 1963. Seawater demineralization by means of an osmotic membrane. Advances in Chemistry Series 38, 117e132. Long, B., 2008. Optimization of desalination for low energy. Presentation at the Singapore International Water Week, Singapore. Macedonio, F., Drioli, E., 2010. An exergetic analysis of a membrane desalination system. Desalination 261 (3), 293e299. Manth, T., Gabor, M., Oklejas, E., 2003. Minimizing RO energy consumption under variable conditions of operation. Desalination 157 (1e3), 9e21. Mathioulakis, E., Belessiotis, V., Delyannis, E., 2007. Desalination by using alternative energy: review and state-of-the-art. Desalination 203, 346e365. McCutcheon, J., McGinnis, R.L., Elimelech, M., 2005. A novel ammoniaecarbon dioxide forward (direct) osmosis desalination process. Desalination 174 (1), 1e11. McCutcheon, J., McGinnis, R.L., Elimelech, M., 2006. Desalination by ammoniaecarbon dioxide forward osmosis: influence of draw and feed solution concentrations on performance on process performance. Journal of Membrane Science 278 (1e2), 114e123. McGinnis, R.L., Elimelech, M., 2007. Energy requirements of ammoniaecarbon dioxide forward osmosis desalination. Desalination 207, 370e382. McGinnis, R.L., McCutcheon, J., Elimelech, M., 2007. Forward Osmosis Energy Use: Comparisons to RO, MSF, MED. Presentation at North American membrane Society (NAMS) annual Conference, April 14e17, Orlando, Florida. Mezher, T., Fath, H., Abbas, Z., Khaled, A., 2011. Techno-economic assessment and environmental impacts of desalination technologies. Desalination 266 (1e3), 263e273. Mi, B., Elimelech, M., 2010. Organic fouling of forward osmosis membranes: fouling reversibility and cleaning without chemical reagents. Journal of Membrane Science 348 (1e2), 337e345. Mirza, S., 2008. Reduction of energy consumption in process plants using nanofiltration and reverse osmosis. Desalination 224, 132e142. Mohamed, E., Papadakis, G., 2004. Design, simulation and economic analysis of a stand-alone reverse osmosis desalination unit powered by wind turbines and photovoltaics. Desalination 164, 87e97. Murugavel, K.K., Srithar, K., 2011. Performance study on basin type double solar still with different wick materials and minimum mass of water. Renewable Energy 36 (2), 612e620. Naim, M., Mervat, A., El-Kawi, A., 2003. Non-conventional solar stills. Part I: non-conventional solar stills with charcoal particles as absorber medium. Desalination 153, 55e64. NanoH2O, 2010. Nanotechnology advances reverse osmosis membrane performance. Available from: http://www. nanoh2o.com/Technology.php5?category¼Economics (accessed 09.30.10). National Academy of Sciences (NAS), 2010. Electricity from renewable resources: status, prospects, and impediments. The National Academy Press, Washington D.C. National Renewable Energy Laboratory (NREL), 2006. Integrated wind energy/desalination system. Final Report SR500e39485.

1919

Oklejas, M., Stidham, K., Weidmann, M., 2005. Improve energy recovery in gas processing plant using an HPT. Hydrocarbon Process, 43e46. Oren, Y., 2008. Capacitive deionization (CDI) for desalination and water treatment e past, present and future (a review). Desalination 228 (1e3), 10e29. Parekh, S., Farid, M., Selman, J., Al-Hallaj, S., 2004. Solar desalination with a humidificationedehumidification technique e a comprehensive technical review. Desalination 160, 167e186. Petersen, G., Fries, S., Mohn, J., Muller, A., 1981. Wind and solar powered reverse osmosis desalination units e design, start up, operating experiences. Desalination 39, 125e135. Petry, M., Sanz, M.A., Langlais, C., Bonnelye, V., Durand, J., Guevara, D., Nardes, W.M., Saemi, C.H., 2007. The El Coloso (Chile) reverse osmosis plant. Desalination 203, 141e152. Pretner, A., Iannelli, M., 2002. Feasibility study and assessment of the technical, administrative and financial viability of the Voltano desalination plant (Agrigento, Sicily). Desalination 153, 313e320. Qiblawey, H.M., Banat, F., 2008. Solar thermal desalination technologies. Desalination 220, 633e644. Queensland Geothermal Energy Center of Excellence, 2010. Available from: http://www.uq.edu.au/news/?article¼20223 (accessed 09.30.10). Raluy, R.G., Serra, L., Uche, J., 2005. Life cycle assessment of desalination technologies integrated with renewable energies. Desalination 183, 81e93. Rodriguez, L.G., 2002. Seawater desalination driven by renewable energies: a review. Desalination 143, 103e113. Saltworks, 2010. Thermo-ionic energy conversion. Available from: http://www.saltworkstech.com/technology.php (accessed 09.01.10). Sauvet-Goichon, B., 2007. Ashkelon desalination plant e a successful challenge. Desalination 203, 75e81. Schafer, A.I., Broeckmann, A., Richards, B.S., 2007. Renewable energy powered membrane technology. 1. Development and characterization of a photovoltaic hybrid membrane system. Environmental Science and Technology 41 (3), 998e1003. Semiat, R., 2008. Energy issues in desalination processes. Environmental Science and Technology 42 (22), 8193e8201. Service, R.F., 2006. Desalination freshens up. Science 313, 1088e1090. Sholl, D.S., Johnson, J.K., 2006. Making high-flux membranes with carbon nanotubes. Science 312 (5776), 1003e1004. Souari, L., Hassairi, M., 2007. Seawater desalination by reverse osmosis: the true needs for energy. Desalination 206 (1e3), 465e473. Spang, E., 2006. The potential for wind-powered desalination in water-scarce countries. Masters Thesis, The Fletcher School, Tufts University, Massachusetts, United States. Stover, R., 2004. Development of a fourth generation energy recovery device. A ‘CTO’s notebook. Desalination 165, 313e321. Stover, R., 2007. Seawater reverse osmosis with isobaric energy recovery devices. Desalination 203, 168e175. Subramani, A., Kim, S., Hoek, E.M.V., 2006. Pressure, flow, and concentration profiles in open and spacer-filled membrane channels. Journal of Membrane Science 277 (1e2), 7e17. Sui, H., Han, B.G., Lee, J.K., Walian, P., Jap, B.K., 2001. Structural basis of water specific transport through the AQP1 water channel. Nature 414, 872e878. Tabrizi, F.F., Dashtban, M., Moghaddam, H., Razzaghi, K., 2010. Effect of water flow rate on internal heat and mass transfer and daily productivity of a weir-type cascade solar still. Desalination 260 (1e3), 239e247. Thomson, M., Infield, D., 2003. A photovoltaic e powered seawater reverse e osmosis system without batteries. Desalination 153 (1e3), 1e8.

1920

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 0 7 e1 9 2 0

Thumthae, C., Chitsomboon, T., 2009. Optimal angle of attack for untwisted blade wind turbine. Renewable Energy 34 (5), 1279e1284. Torre, A., 2008. Efficiency optimization in SWRO plant: high efficiency and low maintenance pumps. Desalination 221 (1e3), 151e157. Trieb, F., Muller-Steinhagen, H., Kern, J., Scharfe, J., Kabariti, M., Al Taher, A., 2009. Technologies for large scale seawater desalination using concentrated solar radiation. Desalination 235 (1e3), 33e43. Truskett, T.M., 2003. Subtleties of water in small spaces. Proceedings of the National Academy of Science 100 (18), 10139e10140. van Paassen, J., van der Meer, W., Post, J., 2005. Optiflux: from innovation to realization. Desalination 178, 325e331. Veerapaneni, S., Jordan, B., Leitner, G., Freeman, S., Madhavan, J., 2005. Optimization of RO desalination process energy consumption. International Desalination Association World Congress, Singapore. Veerapaneni, S., Long, B., Freeman, S., Bond, R., 2007. Reducing energy consumption for desalination. Journal of the American Water Works Association 99 (6), 95e106. Voivontas, D., Misirlis, K., Manoli, E., Arampatzis, G., Assimacopoulos, D., Zervos, A., 2001. A tool for the design of desalination plants powered by renewable energies. Desalination 133, 175e198. Voutchkov, N., 2004. Seawater desalination costs cut through power plant co-location. Filtration and Separation 41 (7), 24e26. Voutchkov, N., 2007. Advances in seawater desalination technology. Water Conditioning and Purification Available from: http://www.wcponline.com/pdf/0709Voutchkov.pdf (accessed 09.07.10). Voutchkov, N., 2010. Membrane Seawater Desalination Overview and Recent Trends. Desalination: An energy solution. Presentation at International Desalination Association Conference, November 2e3, Huntington Beach, California. Wang, Y., Wang, S.C., Xu, S.C., 2004. Experimental studies on dynamic process of energy recovery device for RO desalination plants. Desalination 160 (2), 187e193. Wang, Y., Wang, S., Xu, S., 2005. Investigations on characteristics and efficiency of a positive displacement energy recovery unit. Desalination 177 (1e3), 179e185.

Weiner, D., Fisher, D., Moses, E.J., Katz, B., Meron, G., 2001. Operation experience of a solar- and wind-powered desalination demonstration plant. Desalination 137, 7e13. Welgemoed, T.J., 2005. Capacitive Deionization Technology: development and evaluation of an industrial prototype system. Ph.D. Dissertation, University of Pretoria. Wilf, M., Bartels, C., 2005. Optimization of seawater RO systems design. Desalination 173, 1e12. Wilf, M., Hudkins, J., 2010. Energy Efficient Configuration of RO Desalination Units. Proceedings of Water Environment Federation Membrane Applications Conference, Anaheim, California. Yang, Q., Wang, K.Y., Chung, T.S., 2009. Dual-layer hollow fibers with enhanced flux as novel forward osmosis membranes for water production. Environmental Science and Technology 43 (8), 2800e2805. Zejli, D., Benchrifa, R., Bennouna, A., Zazi, K., 2004. Economic analysis of wind-powered desalination in the south of Morocco. Desalination 165, 219e230. Zhu, A., Christofides, P.D., Cohen, Y., 2009. Effect of thermodynamic restriction on energy cost optimization of RO membrane water desalination. Industrial and Engineering Chemistry Research 48, 6010e6021. Zhu, A., Christofides, P.D., Cohen, Y., 2009a. Minimization of energy consumption for a two-pass membrane desalination: effect of energy recovery, membrane rejection and retentate recycling. Journal of Membrane Science 339, 126e137. Zhu, A., Christofides, P.D., Cohen, Y., 2009b. Energy consumption optimization of reverse osmosis membrane water desalination subject to feed salinity fluctuation. Industrial and Engineering Chemistry Research 48, 9581e9589. Zhu, A., Christofides, P.D., Cohen, Y., 2009c. On RO membrane and energy costs and associated incentives for future enhancements of membrane permeability. Journal of Membrane Science 344 (1e2), 1e5. Zhu, A., Rahardianto, A., Christofides, P.D., Cohen, Y., 2010. Reverse osmosis desalination with high permeability membranes e cost optimization and research needs. Desalination and Water Treatment 15, 256e266. Zhu, A., Christofides, P.D., Cohen, Y., 2010a. Effect of stream mixing on RO energy cost minimization. Desalination 261 (3), 232e239.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 2 1 e1 9 3 2

Available at www.sciencedirect.com

journal homepage: www.elsevier.com/locate/watres

Laccase-catalyzed oxidation of oxybenzone in municipal wastewater primary effluent Hector A. Garcia, Catherine M. Hoffman, Kerry A. Kinney*, Desmond F. Lawler Department of Civil, Architectural, and Environmental Engineering, Environmental and Water Resources Engineering Program, The University of Texas at Austin, 1 University Station C1786, Austin, TX 78712, USA

article info

abstract

Article history:

Pharmaceuticals and personal care products (PPCPs) are now routinely detected in raw and

Received 9 August 2010

treated municipal wastewater. Since conventional wastewater treatment processes are not

Received in revised form

particularly effective for PPCP removal, treated wastewater discharges are the main entry

20 December 2010

points for PPCPs into the environment, and eventually into our drinking water. This study

Accepted 22 December 2010

investigates the use of laccase-catalyzed oxidation for removing low concentrations of

Available online 28 December 2010

PPCPs from municipal wastewater primary effluent. Oxybenzone was selected as a representative PPCP. Like many other PPCPs, it is not recognized directly by the laccase enzyme.

Keywords:

Therefore, mediators were used to expand the oxidative range of laccase, and the efficacy

Laccase

of this laccaseemediator system in primary effluent was evaluated. Eight potential medi-

Oxybenzone

ators were investigated, and 2,20 -Azino-bis(3-ethylbenzthiazoline-6sulphonic acid) dia-

Pharmaceuticals

mmonium salt (ABTS), a synthetic mediator, and acetosyringone (ACE), a natural mediator,

Personal care products

provided the greatest oxybenzone removal efficiencies. An environmentally relevant

Wastewater

concentration of oxybenzone (43.8 nM, 10 mg/L) in primary effluent was completely

Primary effluent

removed (below the detection limit) after two hours of treatment with ABTS, and 95% was removed after two hours of treatment with ACE. Several mediator/oxybenzone molar ratios were investigated at two different initial oxybenzone concentrations. Higher mediator/oxybenzone molar ratios were required at the lower (environmentally relevant) oxybenzone concentration, and ACE required higher molar ratios than ABTS to achieve comparable oxybenzone removal. Oxybenzone oxidation byproducts generated by the laccaseemediator system were characterized and compared to those generated during ozonation. Enzymatic treatment generated byproducts with higher mass to charge (m/z) ratios, likely due to oxidative coupling reactions. The results of this study suggest that, with further development, the laccaseemediator system has the potential to extend the treatment range of laccase to PPCPs not directly recognized by the enzyme, even in a primary effluent matrix. ª 2011 Elsevier Ltd. All rights reserved.

1.

Introduction

The objective of this research was to investigate the possible application of laccase-catalyzed oxidation to the removal of pharmaceuticals and personal care products (PPCPs) from

municipal wastewater primary effluent. This study was performed using oxybenzone, a representative PPCP that, like many other PPCPs, is not oxidized directly by the laccase enzyme. Therefore, mediators were used to expand the oxidative range of laccase, and the efficacy of this laccaseemediator

* Corresponding author. Tel.: þ1 512 232 1740; fax: þ1 512 471 5870. E-mail address: [email protected] (K.A. Kinney). 0043-1354/$ e see front matter ª 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.watres.2010.12.027

1922

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 2 1 e1 9 3 2

system in primary effluent was evaluated. This investigation was conducted using primary effluent because, if eventually implemented in municipal wastewater treatment plants, placing enzymatic treatment after the primary clarifier would allow for potential removal of byproducts in subsequent biological treatment. PPCPs are now ubiquitous in the environment, as they are routinely detected in wastewater, natural water systems, and even in drinking water. The primary means by which these microconstituents enter ecosystems is municipal wastewater that has been contaminated (Kolpin et al., 2002; Barber et al., 2006) via excretion, flushing of unused medications, and our daily use of personal care products. Conventional wastewater treatment processes are not particularly effective for PPCP removal (Auriol et al., 2006), resulting in many of these microconstituents being discharged to surface waters. Consequently, freshwater sources of drinking water now contain many of these recalcitrant compounds (Kolpin et al., 2002). PPCPs have the potential to cause adverse physiological and developmental effects at trace levels (Khetan and Collins, 2007) and have already been found to cause physiological changes in some aquatic organisms (Vajda et al., 2008). Although certain advanced treatment processes (e.g., ozonation, advanced oxidation processes (AOPs), reverse osmosis, and adsorption) can effectively remove PPCPs from wastewater (Snyder et al., 2007), these technologies have certain disadvantages. Ozonation, AOPs, and reverse osmosis are quite energy intensive. Reverse osmosis and adsorption concentrate pollutants and change their phase, respectively, rather than destroying them (Auriol et al., 2006). Ozonation and AOPs can form undesirable byproducts due to reactions with both target PPCPs and natural organic matter (Ikehata et al., 2006). These byproducts usually undergo no further treatment before release into the environment, because such processes can only be implemented efficiently after secondary clarification to avoid competition for the oxidant by other constituents in the wastewater. Nevertheless, ozonation and AOPs have proven to be among the most viable methods for oxidizing PPCPs in wastewater. The lack of specificity, due to the formation of hydroxyl radicals, provides highly effective removal of a wide spectrum of PPCPs (Ikehata et al., 2008), despite their diverse physical/chemical properties (Snyder et al., 2007). Enzymatic treatment is potentially an attractive alternative for PPCP removal. Enzymatic systems have low energy requirements, easy process control, and can operate over a wide range of pH values, temperatures, and ionic strengths (Cabana et al., 2007a). Laccase, an oxidoreductase enzyme, catalyzes the oxidation of certain aromatic compounds, particularly phenolic compounds, using molecular oxygen as the terminal electron acceptor. Although laccase exhibits substrate specificity for some PPCPs, it cannot directly oxidize many others (Can˜as and Camarero, 2010). For example, oxybenzone, despite its phenolic moiety, is not directly oxidized by laccase. However, the presence of low molecular weight mediators (usually low molecular weight phenolic compounds) has been reported to enhance/expand laccase’s oxidative ability (Can˜as and Camarero, 2010; Kunamneni et al., 2008). Laccase is able to oxidize these mediators to free radicals, which react non-specifically, providing highly effective removal of a wide spectrum of compounds. The enzymatic

treatment that we are investigating utilizes this laccaseemediator system, which is expected to be capable of oxidizing a broader spectrum of PPCPs than previously reported enzymatic treatment. Laccase-catalyzed oxidation has been investigated for the removal of organic contaminants from industrial wastewaters. Laccase’s direct oxidation of phenols in effluents from olive oil mills, petroleum refineries, pulp and paper mills, and wine distilleries has been demonstrated (Berrio et al., 2007; Ko and Fan, 2010; Steevensz et al., 2009; Strong and Burgess, 2007). Moreover, laccase’s direct oxidation of several emerging organic contaminants (including several PPCPs) in buffered ultrapure water has been investigated, including acetaminophen (Lu et al., 2009), diclofenac, estradiol, estrone, ethynilestradiol (Lloret et al., 2010), triclosan (Kim and Nicell, 2006a; Cabana et al., 2007b), bisphenol A, and nonylphenol (Cabana et al., 2007b). Most of these compounds contain phenolic moieties, which is why they were directly oxidized by laccase. Emerging organic contaminants that were not directly oxidized by laccase, but were effectively oxidized by the laccaseemediator system in buffered ultrapure water include carbamazepine (Hata et al., 2010), naproxen (Lloret et al., 2010), anthracene (Sei et al., 2008), benzo[a]pyrene, phenanthrene (Camarero et al., 2008), and pyrene (Johannes and Majcherczyk, 2000). Some of these studies were performed in the context of PPCP removal from municipal wastewater, but the experiments were conducted in buffer, and most were at relatively high PPCP concentrations. The only published study to date in which laccase-catalyzed oxidation was performed with low PPCP concentrations in a municipal wastewater matrix is by Auriol et al. (2007). They evaluated direct laccase oxidation of trace concentrations of natural and synthetic hormones in secondary effluent. Although a mediator was used to increase oxidation rates, mediators were not required to oxidize the target hormones, as was previously demonstrated by Lloret et al. (2010). To our knowledge, our study is the first to evaluate a laccaseemediator system for removing a PPCP not directly oxidized by laccase from primary effluent. Laccase-catalyzed oxidation (like ozonation and AOPs) transforms the target compounds, generating oxidation byproducts. As part of this study, byproduct formation was evaluated and compared to that of ozonation. If enzymatic treatment is implemented after primary clarification and before the activated sludge process in a conventional wastewater treatment plant, oxidation byproducts could potentially be removed in subsequent treatment processes. However, primary effluent contains relatively high concentrations of organic and inorganic constituents (in comparison to secondary effluent and buffered ultrapure water), and these constituents could affect the efficacy of the laccaseemediator system. In particular, a primary effluent matrix would be expected to demand higher mediator and/or enzyme concentrations. These issues were explored in this research.

2.

Materials and methods

2.1.

Research design

To investigate the performance of the laccaseemediator system for removing PPCPs from primary effluent, this study

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 2 1 e1 9 3 2

was performed with oxybenzone (a representative PPCP not directly oxidized by laccase) in the following stages: (i) mediator screening experiments to evaluate the need for a mediator and determine which mediators performed best; (ii) initial experiments in primary effluent to determine if the laccaseemediator system could effectively remove oxybenzone from a primary effluent matrix (despite the presence of other organic and inorganic constituents); (iii) further experiments in primary effluent investigating the effect of the mediator/oxybenzone molar ratio; (iv) evaluation of the laccaseemediator system at environmentally relevant oxybenzone concentrations in primary effluent; and (v) characterization of the enzyme oxidation byproducts and comparison of these byproducts with those generated by ozonation. Oxybenzone ((2-Hydroxy-4-methoxyphenyl)-phenylmethanone) absorbs and dissipates UV radiation, and is therefore commonly used in sunscreens and cosmetic products. Oxybenzone was selected as a representative PPCP for three primary reasons. First, despite being a phenolic compound, it is not directly oxidized by laccase and so was a good candidate for study in the laccaseemediator system. Second, it demonstrates estrogen-like activity in vitro and in vivo (Schlumpf et al., 2004; Calafat et al., 2008). Although adverse health effects in humans have not been reported, dermal and oral administration of oxybenzone to rats and mice has shown alterations in liver, kidney, and reproductive organs (Calafat et al., 2008). Third, measurable oxybenzone concentrations in conventional wastewater treatment plant effluents have been reported (Snyder et al., 2007; Kim et al., 2007). Snyder et al. (2007) reported relatively high concentrations ranging from 37 to 3810 ng/L in raw municipal wastewater and primary effluent.

2.2.

Chemicals

Laccase derived from Trametes versicolor (CAS 80498-15-3), oxybenzone (CAS 131-57-7), ferulic acid (FA) (CAS 1135-24-6), and p-coumaric acid (PCA) (CAS 501-98-4), were purchased from SigmaeAldrich (St. Louis, MO, USA). 2,20 -Azino-bis (3-ethylbenzthiazoline-6sulphonic acid) diammonium salt (ABTS) (CAS 30931-67-0) was purchased from EMD Bioscience (Gibbstown, NJ, USA). Acetosyringone (ACE) (CAS 2478-38-8) was purchased from Indofine Chemical Company (Hillsborough, NJ, USA). 4-Hydroxybenzoic acid (HBA) (CAS 114-63-6) was purchased from Spectrum Chemical (Garden, CA, USA). 1-Hydroxy-benzotirazole (HBT) (CAS 134-96-3) was purchased from ScienceLab.com Inc. (Houston, TX, USA). Sinapinic acid (SIN) (CAS 530-59-6) and syringaldehyde (SYR) (CAS 134-96-3) were purchased from Alfa Aesar (Ward Hill, MA, USA). LC/MS grade methanol and water were purchased from JT Baker (Phillipsburg, NJ, USA). Ultrapure water was produced by filtering distilled water through a Milli-Q UV Plus water purification system (Millipore, Billerica, MA, USA).

2.3.

Enzyme activity assay

Laccase activity was determined following the methods described by Auriol et al. (2007) with some minor modifications. Enzyme activity was measured by determining the oxidation rate of a substrate, ABTS, to its final product, ABTSþ. Enzyme solutions were diluted to obtain enzyme activities of

1923

approximately 1.0  103 units/mL in assay mixtures; one unit (U) of enzyme activity is defined as the amount of enzyme that catalyzes the oxidation of 1 mmol of ABTS per minute at 37  C. In addition to an enzyme solution, an assay mixture contained 5.0 mM ABTS in 0.1 M sodium acetate buffer at pH 5 at 37  C. Assay mixtures were placed in a 96-well polypropylene round bottom microplate (Greiner Bio-One, Monroe, NC, USA) with a well volume of 300 mL. The microplate was incubated in an orbital shaker (New Brunswick Scientific, Edison, NJ, USA) at 37  C, and the absorbance was measured every 1.5 min for 15 min using a microplate reader (Biotek, Winooski, VT, USA).

2.4. Liquid chromatography and mass spectrometry analysis The quantitative analyses of oxybenzone, ABTS, and ACE were performed using liquid chromatography/tandem mass spectrometry (LC/MS/MS), as in Vanderford et al. (2003). The analytes were isolated using a Shimadzu 150  4.6 mm C18 column with a particle size of 5 mm and a binary gradient of methanol and water. A Finnigan Surveyor autosampler, a Finnigan Surveyor mass spectrometer pump, and a TSQuantum mass spectrometer (Thermo Electron Corporation, Waltham, MA, USA) were used. Electrospray ionization in the positive mode was the ionization source. Different methods were employed to measure the analytes, as described in Table 1. The relative standard deviations of the methods used for determining oxybenzone, ABTS, and ACE concentrations at the different experimental conditions are also reported in Table 1. Note that the relative standard deviation of the measurements for all four compounds ranges from 5 to 9%, suggesting excellent reproducibility in these measurements. The characterization of oxybenzone oxidation byproducts was performed using the same instrument setup described above, but a Finnigan Surveyor photodiode array (PDA) detector was incorporated. Also, the mass spectrometer was used in the full scan mode rather than in the tandem MS/MS mode. The scan range selected for the mass spectrometer full scan was between m/z ratios of 30 and 1500. The PDA detector was programmed to scan in the ultraviolet-visible range between 200 and 800 nm.

2.5.

Experimental procedures

2.5.1.

Mediator screening experiments

Mediator screening experiments were performed in 50 mL amber glass batch reactors to investigate the oxidative capacity of laccase acting alone on oxybenzone, as well as in the presence of several different mediators. The reaction mixture had an initial oxybenzone concentration of 4.38 mM (1000 mg/L) in 0.1 M sodium phosphate buffer at pH 7 in ultrapure water (Auriol et al., 2007). The initial laccase activity was 1 U/mL. Eight mediators (ABTS, ACE, FA, HBA, HBT, SIN, SYR, and PCA) were evaluated at a concentration of 1 mM. The reactors were placed in a constant temperature orbital shaking water bath (Boekel Grant, Feasterville, PA, USA) at 23  C. The reactors were sampled at 0 (before laccase addition), 2, 6, and 24 h to measure the concentration of oxybenzone remaining at each time point. The samples (3 mL) were placed in 10 mL test tubes and acidified with 40 mL of 5 N

1924

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 2 1 e1 9 3 2

Table 1 e LC/MS/MS analytical parameters for oxybenzone, ABTS, and ACE. Oxybenzone higher range

Oxybenzone lower range

ABTS

ACE

Liquid chromatography Injection volume (mL) 10 20 20 20 Flow rate (mL/min) 700 350 700 700 Gradient Oxybenzone higher range: 5% methanol held constant for 3 min, increased linearly to reach 80% at 10 min, held constant at 80% for 10 min, stepped up to 100% and held constant for 11 min. At the end of each run, the methanol was stepped down to 5% and held constant for 4 min. Oxybenzone lower range, ABTS, and ACE: 5% methanol held constant for 6 min, increased linearly to reach 80% at 18 min, held constant at 80% for 4 min, stepped up to 100% and held constant for 9 min. At the end of each run, the methanol was stepped down to 5% and held constant for 4 min. Mass spectrometry Collision energy (V) Collision gas pressure (mTorr) Ion spray voltage (V) Ion source temp. ( C) Precursor-product ion mass/charge (m/z) ratio

20 1.0 4200 400 229.0e150.9

Method detection limits (MDLs), recoveries, and relative standard deviations (RSDs) Standard curve range (mg/L) 0e1000 0.9914e0.9997 Standard curve R2 values MDL (mg/L) 15 Mean recovery by SPE (%) 81 (in buffer); 79 (in MWPE) RSD (%) 6

hydrochloric acid to a final pH of approximately 1.5 to inactivate the enzyme (Auriol et al., 2007). The analytes in the acidified samples were extracted using 3 cm3 60 mg Oasis HLB solid phase extraction (SPE) cartridges (Waters, Milford, MA, USA). SPE was used in these experiments as a purification step, not as a concentration method. The SPE cartridges were conditioned with 3 mL of LC/MS grade methanol followed by 3 mL of LC/MS grade water. After conditioning, 2 mL of the samples were passed through the cartridges, which were subsequently rinsed with 3 mL of LC/MS grade water three times. After allowing the cartridges to dry, they were eluted with two 0.7 mL and one 0.6 mL aliquots of LC/MS grade methanol. Purified samples were analyzed using LC/ MS/MS. Samples were also collected at 0 and 48 h to measure enzyme activity.

2.5.2.

Initial experiments in primary effluent

The two mediators with the highest oxybenzone removal efficiencies (ABTS and ACE) were selected for further experiments in primary effluent. These experiments were nearly identical to the mediator screening experiments, but they were conducted in filtered primary effluent rather than in phosphate buffer. The primary effluent was collected weekly from the Walnut Creek wastewater treatment facility in Austin, Texas and was stored at 4  C between collection and use. It was filtered using grade 934AH glass microfiber filters with a 1.5 mm pore size (Whatman, Piscataway, NJ, USA). The following measurements were typical of the collected primary effluent: alkalinity ¼ 273 mg/L as CaCO3, BOD ¼ 145 mg/L, soluble COD ¼ 112 mg/L, total COD ¼ 194 mg/L, total NHþ 4 ¼ 31 mg/L as N, pH ¼ 7.4, TSS ¼ 29 mg/L, pH after filtration ¼ 7.67. In one set of experiments, the pH of the primary effluent was not adjusted, and in a second set, the primary

29 1.0 4000 400 229.0e150.9

40 1.5 2700 400 514.7e230.0

20 1.5 4000 350 197.2e140.0

0e200 0.9937e0.9980 0.25 44

0e90,000 0.9932e0.9917 132 53

0e15,000 0.9906e0.9989 6 58

5

5

9

effluent was adjusted to an initial pH of 6 with 5 N hydrochloric acid. In addition to sampling and measurement as described in Section 2.5.1, pH measurements were taken at 0 and 48 h using a pH meter (Thermo Electron Corporation, Waltham, MA, USA).

2.5.3. Effect of mediator/oxybenzone molar ratios in primary effluent The initial experiments in primary effluent were performed at a high mediator/oxybenzone molar ratio of approximately 230. The next set of experiments focused on determining the minimum mediator/oxybenzone molar ratios in primary effluent at which the laccaseemediator system was still effective for oxidizing oxybenzone. These experiments were performed similarly to those described in Section 2.5.1, but they were conducted in filtered primary effluent adjusted to an initial pH of 6 rather than in phosphate buffer. The ABTS and ACE mediator concentrations were 87.6, 26.3, 8.76, and 2.63 mM to achieve mediator/oxybenzone molar ratios of 20, 6, 2, and 0.6, respectively. The reactors were sampled at 0, 0.25, 1, and 6 h to determine the oxybenzone concentration remaining. Enzyme activity and pH measurements were taken at 0 and 6 h. Several control experiments were also performed by omitting one or more of the following three constituents: oxybenzone, laccase, and ACE. Two controls were performed for oxybenzone oxidation: without laccase and without both laccase and mediator. Three controls were performed for enzyme activity: without mediator, without oxybenzone, and without both mediator and oxybenzone. Samples from all of the control reactors were taken at 0 and 6 h and measured for relevant constituents (oxybenzone and/or laccase activity).

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 2 1 e1 9 3 2

2.5.4.

Experiments at environmentally relevant concentration

Although the previous experiments at an initial oxybenzone concentration of 4.38 mM (1000 mg/L) simplified the analytical procedures, oxybenzone and other PPCPs are typically found in municipal wastewater at much lower concentrations. Thus, a set of experiments was conducted to evaluate enzyme oxidation at a more environmentally relevant oxybenzone concentration. Experiments were conducted in 250 mL amber glass batch reactors containing 100 mL of reaction mixture. The reaction mixture had an initial oxybenzone concentration of 43.8 nM (10 mg/L) in filtered primary effluent adjusted to an initial pH of 6. In one set of experiments, ACE mediator was added to yield mediator/oxybenzone molar ratios of 2000, 200, and 20, corresponding to ACE concentrations of 87.6, 8.76, and 0.876 mM, respectively. In a separate experiment, ABTS mediator was added to yield a mediator/oxybenzone molar ratio of 200, corresponding to an ABTS concentration of 8.76 mM. The initial laccase activity was 1 U/mL. Two controls without laccase were performed with this set of experiments, one with oxybenzone and ACE, and one with oxybenzone and ABTS. The reactors were placed in a constant temperature orbital shaking water bath at 23  C. Twenty mL samples were collected at 0, 0.25, and 2 h, placed in 50 mL amber glass vials, and immediately acidified with 300 mL of 5 N hydrochloric acid to a final pH of approximately 1.5 to inactivate the enzyme. The controls were sampled at 0 and 2 h. The acidified samples were passed through SPE cartridges. The SPE cartridges were conditioned prior to passing samples and rinsed after passing samples, as described in Section 2.5.1. After drying, the cartridges were eluted with two 2 mL and one 1 mL aliquots of LC/MS grade methanol. The eluates were collected in 5 mL conic vials and evaporated to less than 1 mL, but not to dryness, using a Rapidvac nitrogen evaporation system (Labconco, Kansas City, MO, USA). The samples were reconstituted to 1 mL with LC/MS grade methanol and analyzed for oxybenzone concentration using LC/MS/MS. Samples from selected experiments were also analyzed for mediator concentration. Enzyme activity and pH measurements were taken at 0 and 2 h.

2.5.5.

Characterization of enzyme oxidation byproducts

Two experiments were performed to investigate the formation of enzyme oxidation byproducts in a reaction mixture in a 0.1 M sodium phosphate buffer at pH 7. When ABTS was used as the mediator, the initial oxybenzone concentration was 43.8 mM (10 mg/L), and the ABTS concentration was 0.1 mM (ABTS/oxybenzone molar ratio of 2.3). When ACE was used as the mediator, the initial oxybenzone concentration was 218 mM (50 mg/L), and the ACE concentration was 4.38 mM (ACE/oxybenzone molar ratio of 20). The initial laccase activity was 1 U/ mL. The reactors were placed in a constant temperature orbital shaking water bath at 23  C. Three mL samples collected from the ABTS and ACE reactors were acidified and purified as described in Section 2.5.1. Samples were analyzed using LC/MS in the full scan mode to detect both the disappearance of the parent compound and the formation of byproducts.

2.5.6.

Characterization of ozonation byproducts

An experiment was performed to characterize the byproducts generated when oxidizing oxybenzone with dissolved ozone. An ozone generator (Yanco Industries Ltd., Burton, BC, Canada)

1925

was used to produce a 60 mg/L ozone stock solution. Ten different aliquots of the ozone stock solution were added to 50 mL vials containing 3 mL of a 100 mg/L oxybenzone solution in ultrapure water. Table 2 describes the ten resulting solutions. Samples were analyzed as in Section 2.5.5.

3.

Results and discussion

3.1.

Mediator screening experiments

The objective of these experiments was to evaluate the effects of several possible mediators on laccase oxidation of oxybenzone. As shown in Fig. 1, experiments performed without a mediator yielded no removal of oxybenzone. In the presence of ABTS or ACE mediators, oxybenzone was completely oxidized (below the instrument detection limit) within 24 h of treatment. With SIN mediator, a small fraction of oxybenzone (approximately 6%) remained after 24 h of treatment (Fig. 1), and oxybenzone was completely removed after 48 h of treatment (data not shown). The other mediators evaluated (FA, HBA, HBT, PCA, and SYR) did not perform as well. For instance, HBA and PCA mediators did not yield any removal of oxybenzone after 48 h of treatment. Although most of the mediators tested exhibited some oxidation of oxybenzone, ABTS, ACE, and SIN were the most effective. The main disadvantages of using a synthetic mediator, such as ABTS, are high costs and potential toxicity. On the other hand, many phenolic compounds (e.g., ACE) are naturally occurring and environmentally friendly mediators. Such phenolic mediators can be obtained at low cost due to their abundance in nature, as well as their presence in several industrial wastes (Kunamneni et al., 2008). The oxidation of oxybenzone generally caused some loss of enzyme activity, but the relationship was quite different for different mediators. When no mediator was added, no oxidation occurred, and no enzyme activity was lost in 48 h. When a mediator was present and no oxybenzone was removed, as was the case in the PCA and HBA experiments, the loss of enzyme activity was very small (between 5% and 15%) after 48 h of treatment. When greater oxidation of oxybenzone was observed (experiments with ACE, HBT, SIN, and SYR mediators), the loss of enzyme activity was more pronounced, ranging from 34% to 47%. When ABTS was used as the mediator, the loss of enzyme activity was even more significant at approximately 64%. In contrast, FA mediator achieved a relatively high removal of oxybenzone, but only a small fraction of the initial enzyme activity was lost (5%). According to Cabana et al. (2007a), enzyme inactivation might be caused by free radicals that are produced by the laccase oxidation of the mediator.

3.2.

Initial experiments in primary effluent

The two mediators that exhibited the greatest oxybenzone removal in the mediator screening experiments (ABTS and ACE) were selected for experiments in primary effluent. The initial pH of the primary effluent was 7.67. When this pH was not adjusted, it took 24 h for complete oxybenzone removal (below the instrument detection limit) in the presence of ABTS

1926

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 2 1 e1 9 3 2

Table 2 e Initial concentrations of ozone and oxybenzone as a function of the volume of ozone stock solution added. Ozone stock solution added (mL)

Total volume of solution (mL)

Initial ozone concentration in solution (mg/L)

Initial oxybenzone concentration in solution (mg/L)

Initial oxybenzone concentration in solution (mM)

Initial ozone/oxybenzone molar ratio

3 4 5 6 7 8 9 10 11 12 13

0 15 24 30 34 38 40 42 44 45 46

100 75 60 50 43 38 33 30 27 25 23

0.44 0.33 0.26 0.22 0.19 0.16 0.15 0.13 0.12 0.11 0.10

0 0.95 1.91 2.86 3.82 4.77 5.73 6.68 7.63 8.59 9.54

0 1 2 3 4 5 6 7 8 9 10

mediator (Fig. 2). In the presence of ACE mediator, no removal of oxybenzone was observed after 24 h of treatment (Fig. 2). These results contrast with the high removals achieved in the mediator screening experiments (Fig. 1). One possible reason for this difference is that the pH of the reaction mixture was buffered at 7 in the mediator screening experiments, whereas the pH increased from 7.67 to over 8.5 during the 48 h primary effluent experiments with ABTS and ACE. Thus, the lower performance of the laccaseemediator system in these experiments might not be due to matrix effects, but rather to the higher pH of the primary effluent. Kim and Nicell (2006a) found that the optimal pH range for laccase’s direct oxidation of triclosan was between 4 and 6, and that above pH 7, oxidation decreased dramatically. Auriol et al. (2007) reported that the optimum pH for laccase’s direct oxidation of several estrogens was approximately 6 in synthetic water (a buffered solution in ultrapure water), and they also observed a significant decrease in oxidation above pH 7. In all reported cases known to us, the greatest direct oxidation by laccase occurred at a pH between 4 and 6, suggesting that pH affects laccase activity (Cabana et al., 2007a). If laccase activity is reduced when using the laccaseemediator system, mediator radical generation slows down, thereby slowing down oxybenzone oxidation.

To investigate the pH effect further, these experiments were repeated after adjusting the initial pH of the primary effluent to 6.0. As shown in Fig. 2, much more rapid oxybenzone removal was achieved for both ABTS and ACE mediators; no oxybenzone was detected after two hours of treatment. The pH still increased over the course of the experiments, and the pH values after 48 h of treatment were 8.02 and 7.36 for ABTS and ACE, respectively. With ACE mediator, the oxybenzone removal observed with the initial pH of 6 in primary effluent was even better than the removal observed in the mediator screening experiment, performed in a buffered pH 7 solution. Auriol et al. (2007) reported that a municipal wastewater secondary effluent matrix did not have a significant impact on direct laccase oxidation of several estrogens, as compared to synthetic water. Our experimental results in this laccaseemediator system confirm that, despite the higher concentration of organics in primary effluent, the matrix had little effect on oxybenzone removal. As discussed further in following sections, it seems likely that a wastewater matrix would demand higher mediator/ oxybenzone molar ratios than would be required in ultrapure water, due to competition for oxidized radicals of the mediator by other organics. At the high mediator/oxybenzone molar

Oxybenzone Remaining (%)

ABTS

ACE

SIN

100 80 60 40

Oxybenzone Remaining (%)

120

No Mediator

0

0 2 6 24

0 2 6 24 0 2 6 24 Time (hr)

0 2 6 24

Fig. 1 e Selected mediator screening experiments performed in sodium phosphate buffer at 23  C, with an initial oxybenzone concentration of 4.38 mM, a mediator/ oxybenzone molar ratio of 230, and an initial laccase activity of 1 U/mL. nd means not detected.

pH 6 ABTS

pH 6 ACE

80 60 40 20 0

nd

pH 7.67 ACE

100

20 nd nd nd

pH 7.67 ABTS

nd

nd nd nd

nd nd nd

0 2 6 24

0 2 6 24 0 2 6 24 Time (hr)

0 2 6 24

Fig. 2 e Initial experiments in primary effluent at initial pH values of 7.67 and 6 with ABTS mediator and ACE mediator. Experiments were performed in filtered primary effluent at 23  C, with an initial oxybenzone concentration of 4.38 mM, a mediator/oxybenzone molar ratio of 230, and an initial laccase activity of 1 U/mL.

1927

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 2 1 e1 9 3 2

3.3. Effect of mediator/oxybenzone molar ratios in primary effluent After determining that the laccaseeABTS and laccaseeACE systems could oxidize oxybenzone in primary effluent at a high mediator/oxybenzone molar ratio of 230, experiments were performed to determine the lowest molar ratios at which the oxidation would still be effective. As can be seen in Fig. 3, the ABTS mediator was more effective at lower molar ratios than the ACE mediator. The laccaseeABTS system achieved significant oxidation of oxybenzone at ABTS/oxybenzone molar ratios as low as 2, at which only 6% of the oxybenzone remained after six hours of treatment; complete and very rapid oxidation occurred at higher molar ratios. Substantial oxybenzone removal (85%) was achieved at an ACE/oxybenzone

a Oxybenzone Remaining (%)

20

ABTS/Oxybenzone Molar Ratio 6 2

0.6

100 80 60 40 20 nd nd nd 0

b Oxybenzone Remaining (%)

ratios used in these experiments (approximately 230), more than enough mediator was present to meet the demands of oxybenzone and other components of the matrix. Thus, in this set of experiments, the effect of initial pH was much more significant than the matrix effect introduced by the primary effluent. Austin’s wastewater has relatively high alkalinity, with an average of 273 mg/L as CaCO3 in the samples in our experiments. When a system is opened to the atmosphere, the alkalinity remains constant, but the carbonate speciation and pH change as the solution achieves equilibrium with the carbon dioxide in the atmosphere. For Austin’s wastewater, that process led to a loss of carbonic acid (and therefore total carbonate) and a consequent rise in the pH. In a separate experiment with no enzymes but constant mixing, the pH rose from 6.0 to 7.94 in a six-hour period, proving that the rise in pH in our experiments was not due to enzymatic reactions but only to gas/liquid equilibration. Such a pH rise from gas transfer would not occur in an actual wastewater treatment plant reactor, because the surface area per unit volume of fluid is dramatically lower in full-scale reactors. In addition, this result suggests that the necessity to lower the pH would be much less in a full-scale reactor. As noted above, no loss of enzyme activity was observed in the mediator screening experiments when no mediator was present. In the control experiments performed in primary effluent without mediator, no oxybenzone was oxidized, but enzyme activity losses of 35% and 16% were observed at initial pH values of 7.67 and 6, respectively. These results indicate that interaction with other constituents of the primary effluent led to some inactivation of the enzyme. For instance, laccase might have been biodegraded, or inactivated by common organic/ inorganic wastewater constituents such as sulfite, sulfide, nitrite, thiosulfate, copper, iron, cyanide, and halogen ions (Kim and Nicell, 2006b). Moreover, the different enzyme activity losses observed at the two initial primary effluent pH values suggest that higher pH conditions contribute more to enzyme inactivation. For both ABTS and ACE mediators, the enzyme activity losses observed in experiments performed in primary effluent were similar to those observed in experiments performed in phosphate buffer at pH 7. Thus, the enzyme inactivation produced by the free radical oxidation byproducts is more significant than the inactivation produced by common organic wastewater constituents.

0 .25 1 6

20

nd nd 0 .25 1 6 0 .25 1 6 Time (hr) ACE/Oxybenzone Molar Ratio 6 2

0 .25 1 6

0.6

100 80 60 40 20 0

0 .25 1 6

0 .25 1 6 0 .25 1 6 Time (hr)

0 .25 1 6

Fig. 3 e Mediator/oxybenzone molar ratio experiments conducted with (a) ABTS mediator and (b) ACE mediator. Experiments were performed in filtered primary effluent at 23  C, with an initial oxybenzone concentration of 4.38 mM, and an initial laccase activity of 1 U/mL.

molar ratio of 20 after only one hour of treatment, but no additional removal was achieved in six hours. As can be seen in the ACE mediator results in Fig. 3b, most of the oxybenzone removal generally occurred in the first 15 min of the reaction. A similar trend is evident with ABTS mediator, except that the majority of the removal occurred in the first hour. In both cases, the enzyme-mediator system apparently lost its ability to oxidize the target compound after a certain point. During this initial period, the pH of the reaction mixture was still below 7. Thus, the observed decline in oxybenzone oxidation was not likely due to high pH, but rather to consumption of the mediator, as shown below.

3.4. Experiments at environmentally relevant concentration Once it was determined that the laccaseemediator system was able to oxidize oxybenzone in primary effluent at relatively low mediator/oxybenzone molar ratios, experiments were performed at a significantly lower initial oxybenzone concentration of 43.8 nM (10 mg/L). This concentration is much closer to an oxybenzone concentration of 4 nM (900 ng/L), which would be typical in raw municipal wastewater or primary effluent (Snyder et al., 2007). Comparing the results of experiments performed at an environmentally relevant oxybenzone concentration (Fig. 4a)

1928

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 2 1 e1 9 3 2

a

Mediator/Oxybenzone Molar Ratio

Oxybenzone Remaining (%)

ABTS 200

ACE 2,000

ACE 200

ACE 20

100 80 60 40 20 nd 0

0 .25 2

Mediator Remaining (%)

b

0 .25 2 0 .25 2 Time (hr)

0 .25 2

Mediator/Oxybenzone Molar Ratio ABTS 200

ACE 2,000

100 80 60 40 20 no data 0

0

.25

2

0

.25

2

Time (hr)

Fig. 4 e Experiments at environmentally relevant oxybenzone concentration: (a) oxybenzone remaining and (b) mediator remaining in selected experiments. Experiments were performed in filtered primary effluent at 23  C, with an initial oxybenzone concentration of 43.8 nM and an initial laccase activity of 1 U/mL.

to those obtained at a high oxybenzone concentration (Fig. 3) demonstrates that, for both ABTS and ACE, much higher mediator/oxybenzone molar ratios are required at a lower oxybenzone concentration to achieve comparable oxybenzone removal (e.g., a molar ratio of 2000 versus 20 for ACE, and a molar ratio of 200 versus 2 for ABTS). There appears to be a minimum effective mediator concentration, as the same absolute mediator concentration yielded similar results at the two different oxybenzone concentrations. For instance, experiments performed at an ACE concentration of 87.6 mM achieved significant oxybenzone removal at both the 10 mg/L (43.8 nM) and the 1000 mg/L (4.38 mM) initial oxybenzone concentrations (corresponding to ACE/oxybenzone molar ratios of 2000 and 20, respectively). Similarly, at an ABTS concentration of 8.76 mM, significant removal was observed at both the 10 mg/L (43.8 nM) and the 1000 mg/L (4.38 mM) initial oxybenzone concentrations (corresponding to ABTS/oxybenzone molar ratios of 200 and 2, respectively). The requirement for higher mediator/oxybenzone molar ratios when working at low oxybenzone concentrations could be due to organic compounds in the primary effluent matrix other than oxybenzone reacting with the oxidized mediator. Although laccase is very specific for its substrate (the mediator), mediators are not necessarily as specific for their target compound (oxybenzone). Free radicals produced by laccase’s

oxidation of mediators react non-specifically with compounds through hydrogen abstraction, radicaleradical reactions, and electron transfer reactions (Kunamneni et al., 2008). Therefore, some mediator radicals would be consumed by competing organic components, and this effect would be more pronounced when less oxybenzone is present in the solution. Mediator concentration was also measured in two of these experiments (ACE/oxybenzone molar ratio of 2000 and ABTS/ oxybenzone molar ratio of 200). The results shown in Fig. 4b clearly indicate that both mediators are consumed to a significant extent in the enzyme oxidation. Most of the ACE was consumed in the first 15 min of reaction. Likewise, 74% of the ABTS was consumed in two hours (perhaps earlier, but no data are available for the sample at 15 min). Although we have not yet investigated the oxidation mechanisms and pathways of ABTS and ACE mediators in primary effluent, we hypothesize that the majority of mediator consumption is due to side reactions between mediator radicals and organic constituents present in the primary effluent. The reduction in mediator concentration over time is likely the reason that oxybenzone removal slows down. Although the laccaseemediator system was able to effectively remove oxybenzone at an environmentally relevant concentration, relatively high mediator concentrations (8.76 mM ABTS and 87.6 mM ACE) were required. The treatment implications of adding high mediator concentrations are discussed in Section 3.7.

3.5.

Characterization of enzyme oxidation byproducts

To characterize the oxidation byproducts, experiments with the laccaseeABTS and laccaseeACE systems were performed in a pH 7 buffer solution at an initial oxybenzone concentration of 10 mg/L. These experiments do not necessarily reflect the full spectrum of byproducts we would see at an environmentally relevant oxybenzone concentration in a primary effluent matrix. However, we can draw qualitative conclusions about what types of byproducts that can be expected due to mediatoreoxybenzone reactions. The results from the laccaseeABTS experiment clearly indicate the formation of an enzyme oxidation byproduct that did not exist at time zero. Fig. 5a shows the chromatograms of both the photodiode array detector (top) and the mass spectrometer detector (bottom) for a sample taken after 5 min of treatment. Fig. 5b shows the same chromatograms, but for a sample taken after 4 h of treatment. After 5 min of treatment (Fig. 5a), both chromatograms show three main peaks: ABTS, the enzyme oxidation byproduct, and oxybenzone. The enzyme oxidation byproduct did not exist at time zero, but it is clearly observed after only 5 min of treatment. After 4 h of treatment (Fig. 5b), both chromatograms show the complete disappearance of oxybenzone, an increase in the response of the enzyme oxidation byproduct, and a decrease in the ABTS concentration. Similar trends were observed for the samples taken at 15, 30, 60, and 120 min (not shown). After 30 min of treatment, the oxybenzone peak was no longer present. As the reaction progressed, the response of the oxidation byproduct increased, and the response of ABTS decreased. The mass spectrometer chromatogram indicates that the major enzyme oxidation byproduct has a mass to

1929

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 2 1 e1 9 3 2

b ABTS byproduct

oxybenzone oxybenzone

byproduct ABTS

8

Relative Abundance Relative Absorbance

9 10 11 12 13 14 15 Retention Time (min)

byproduct

ABTS byproduct

ABTS 8

9 10 11 12 13 14 15 Retention Time (min)

Fig. 5 e Enzyme oxidation byproduct formation for the laccase-ABTS system: chromatograms for samples taken (a) after 5 min of treatment and (b) after 4 h of treatment. For each sample, the top chromatogram is from the photodiode array detector, and the bottom chromatogram is from the mass spectrometer detector.

charge ratio (m/z) of 740.9. Since the molar mass of oxybenzone is 228.24 g/mol, and the molar mass of ABTS is 514.62 g/mol, it is suspected (although not confirmed in our experiments) that the main oxidation byproduct consists of one oxybenzone molecule coupled to one ABTS molecule. The results from the laccaseeACE experiment were quite different from those with ABTS and clearly indicate the formation of several enzyme oxidation byproducts that did not exist at time zero. Fig. 6 shows the chromatograms from both detectors for samples taken before the enzyme addition (Fig. 6a) and after 6 h of treatment (Fig. 6b). At time zero (Fig. 6a), both chromatograms show two main peaks: ACE and oxybenzone. After 6 h of treatment (Fig. 6b), both chromatograms show a decrease in the oxybenzone concentration, a decrease in the ACE concentration, and the formation of

3.6.

oxybenzone

ACE

oxybenzone

9

10

11 12 13 Time (min)

14

15

Relative Abundance Relative Absorbance

ACE

Characterization of ozonation byproducts

The byproducts of oxybenzone ozonation displayed in Fig. 8 provide a useful comparison with the byproducts generated

b

Relative Abundance Relative Absorbance

a

several enzyme oxidation byproducts. Similar trends were observed for the samples taken at 15, 30, 60, and 120 min. As the reaction progressed, the responses of the oxidation byproducts generally increased, and the responses of oxybenzone and ACE decreased. From the mass spectrometer chromatogram, the m/z ratios of the most abundant enzyme oxidation byproducts were determined. Fig. 7 displays the responses of these major oxidation byproducts over the course of the experiment. The instrument responses of most of the byproducts increased as the reaction progressed; however, the responses of a few byproducts did eventually start to decrease. Far more oxidation byproducts were identified in the presence of ACE mediator than in the presence of ABTS mediator. Most of these detected byproducts have m/z ratios higher than the m/z ratio of oxybenzone (229). For a wastewater treatment application, producing high molar mass byproducts would be desirable, since they should be more hydrophobic and more easily removed from the aqueous phase by precipitation followed by sedimentation or filtration (Huang et al., 2005), or by physical adsorption to other constituents in the wastewater, such as biomass. Since the molar mass of oxybenzone is 228.24 g/mol, and the molar mass of ACE is 196.20 g/mol, it is suspected (although not confirmed in our experiments) that most of these oxidation byproducts consist of oxybenzone molecules coupled to different oxidation forms of ACE molecules. The byproduct with a m/z ratio of 169 is an exception, as it seems to be an oxidized form of ACE. Enzyme-catalyzed oxidative coupling of phenolic compounds has been reported by several authors (Weber and Huang, 2003; Kupriyanovich et al., 2007; Kunamneni et al., 2008). These authors suggested that laccase oxidizes organic substrates to free radicals, which undergo oxidative coupling reactions, producing dimers, oligomers, and polymers.

5.0E+07 ACE

4.5E+07 byproducts oxybenzone

ACE

byproducts oxybenzone

Mass Spectrometer Response

Relative Abundance Relative Absorbance

a

4.0E+07 3.5E+07 3.0E+07 2.5E+07 2.0E+07 1.5E+07 1.0E+07 5.0E+06 0.0E+00

9

10

11 12 13 Time (min)

14

15

Fig. 6 e Enzyme oxidation byproduct formation for the laccase-ACE system: chromatograms for samples taken (a) at time zero and (b) after 6 h of treatment. For each sample, the top chromatogram is from the photodiode array detector, and the bottom chromatogram is from the mass spectrometer detector.

169 201 421 423 447 461 470 475 613 617 639

0

15

30 60 Time (min)

120

360

Fig. 7 e Enzyme oxidation byproduct formation for the laccase-ACE system: mass spectrometer responses of the most abundant byproducts detected as a function of time. The m/z ratios of the byproducts represented are shown in the legend, and from top to bottom, they correspond to the bars from left to right.

1930

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 2 1 e1 9 3 2

b

oxybenzone

10 12 14 16 18 20 22 24 Retention Time (min)

c oxybenzone byproducts

oxybenzone

Relative Abundance

oxybenzone

Relative Abundance Relative Absorbance

Relative Abundance Relative Absorbance

a

byproducts

10 12 14 16 18 20 22 24 Retention Time (min)

10 12 14 16 18 20 22 24 Retention Time (min)

Fig. 8 e Ozonation byproduct formation: chromatograms for samples with (a) no ozone addition and (b & c) an initial ozone/ oxybenzone molar ratio of 3.82. For each sample (a & b), the top chromatogram is from the photodiode array detector, and the bottom chromatogram is from the mass spectrometer detector. (c) shows chromatograms for specific ranges of m/z ratios, capturing the eight most abundant byproducts. From top to bottom, the m/z ratio ranges are: 192.6e193.6, 208.5e209.5, 222.5e223.5, 232.5e233.5, 244.5e245.5, 246.5e247.5, 250.5e251.5, and 290.5e291.5.

from enzyme oxidation. As shown in Fig. 8a, before the addition of ozone, only oxybenzone was present. After the addition of ozone at an initial ozone/oxybenzone molar ratio of 3.82 (Fig. 8b), both detectors clearly indicate the formation of several byproducts. Both chromatograms in Fig. 8b show a decrease in the oxybenzone concentration and the formation of ozonation byproducts. Based on the mass spectrometer chromatogram, the m/z ratios of the most abundant ozonation byproducts were determined, and the eight most representative byproducts are shown in Fig. 8c. The m/z ratio distribution of the ozonation byproducts is different from that of the enzyme oxidation byproducts. Byproducts with higher m/z ratios were produced in the enzymatic treatment, likely due to oxidative coupling reactions between free radicals generated by laccase oxidizing the mediator and oxybenzone. Byproducts from ozonation had lower m/z ratios with values very similar to the molar mass of the parent compound.

3.7.

Treatment implications

It would be desirable to install an oxidative treatment for PPCPs before biological treatment so that biodegradable byproducts could be removed in the aeration basin. However, it is not feasible to install ozonation or AOPs before biological treatment due to the concentrations of organic matter and dissolved solids present in primary effluent. These constituents compete with target pollutants for free radicals, lowering treatment efficiency and requiring cost prohibitive oxidant doses (Ikehata et al., 2006). In contrast, enzymatic treatment, as demonstrated in this research, shows potential for eventual implementation prior to biological treatment in a conventional wastewater treatment plant. Although some wastewater constituents, such as heavy metals, organic compounds, and proteolytic enzymes, have the potential to inactivate laccase, additives and enzyme immobilization procedures have been

applied with success to overcome losses in enzyme activity (Cabana et al., 2007a). In addition, recent biotechnological advances allow rapid and inexpensive production of appropriate enzymes, although further studies would be required to assess the feasibility of integrating this production with the enzyme oxidation treatment process (Cabana et al., 2007a; Rodriguez and Toca-Herrera, 2007; Kunamneni et al., 2008). One potential obstacle to implementation of this treatment process is the need for neutral or slightly acidic pH. However, we do not expect that acidification of primary effluent to pH 6, as was performed in our experiments, would be required in an actual wastewater treatment plant. As discussed in Section 3.2, acidification was needed due to the small volumes of primary effluent utilized in these laboratory experiments, with a consequent high surface area to volume ratio which promoted rapid gas transfer. A wastewater treatment plant with neutral or slightly acidic wastewater is likely to be able to use enzymatic treatment with little to no acidification. The requirement for relatively high mediator concentrations introduces an extra cost to a potential enzymatic treatment system. Although some food processing wastes could potentially be used as mediators, more research is required to determine if this would be feasible in a wastewater treatment application. Some mediators might be toxic at high concentrations. However, mediators are consumed to a great extent by the enzyme oxidation, and placing this enzymatic treatment before biological treatment in a conventional wastewater treatment plant might allow for mediator oxidation byproducts, like PPCP oxidation byproducts, to be removed in subsequent treatment processes. We hypothesize that byproducts of laccase-catalyzed oxidation can be removed in biological treatment, but this has not yet been confirmed in our laboratory. This removal could occur by biodegradation or adsorption to the biomass present in activated sludge (Nakamura and Mtui, 2003). It would be desirable for byproducts to be biodegraded. If they are adsorbed

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 2 1 e1 9 3 2

to biomass or not removed at all, they will still be present in the waste activated sludge or treated wastewater, respectively. In these cases, the fate and toxicity of byproducts would need to be examined (Kinney et al., 2006; McClellan and Halden, 2010).

4.

1931

would be cost prohibitive in primary effluent. Although laccasecatalyzed oxidation of PPCPs shows promise, additional research will be required to determine if placing enzymatic treatment before biological treatment would yield complete removal of both the target PPCPs and the oxidation byproducts.

Conclusions

 In the presence of ABTS or ACE mediators, laccase can effectively remove oxybenzone from municipal wastewater primary effluent. The mediator is oxidized by laccase to a free radical, which in turn oxidizes oxybenzone; oxybenzone is not oxidized by laccase directly. In general, ABTS, a synthetic mediator, achieves more efficient removal of oxybenzone than ACE, a natural mediator.  The laccaseemediator system can achieve complete removal of oxybenzone from primary effluent at a high initial oxybenzone concentration (4.38 mM, 1000 mg/L) and at an environmentally relevant concentration (43.8 nM, 10 mg/L).  The extent of oxybenzone oxidation in primary effluent is dependent on the mediator/oxybenzone molar ratio. The minimum required molar ratio is dependent on the initial oxybenzone concentration; higher molar ratios are required at lower oxybenzone concentrations. Our results suggest that, in primary effluent, a specific concentration of the mediator is required, independent of the oxybenzone concentration.  The enzyme activity was affected by three factors: (i) pH; (ii) the wastewater matrix; and (iii) reaction byproducts generated by the oxidation of the mediator by laccase. The importance of these factors varied under different experimental conditions. Greater enzyme activity was retained when experiments were initiated at pH 6 (as opposed to pH 7.67), when experiments were performed in ultrapure water (as opposed to primary effluent), and at lower mediator concentrations.  The laccaseemediator system does not completely mineralize oxybenzone but instead generates oxidation byproducts. In buffered ultrapure water, different mediators yielded different byproducts for the same target compound. In the presence of the ABTS mediator, one oxidation byproduct was detected. In the presence of the ACE mediator, several byproducts were detected. The m/z ratios of most of the byproducts detected were higher than that of oxybenzone, indicating that mediator free radicals and oxybenzone undergo oxidative coupling reactions.  Ozonation of oxybenzone also produces many byproducts, but with lower m/z ratios that are similar to the molar mass of the parent compound. Since the discharge of treated wastewater is the main entry point for PPCPs into the environment, and eventually into our drinking water, removing PPCPs from municipal wastewater would reduce the threats they pose to ecosystems, as well as reduce human exposure. Results from this study indicate that the laccaseemediator system can remove a PPCP that is not directly oxidized by the enzyme. With further development, laccase-catalyzed oxidation might eventually be implemented in municipal wastewater primary effluent for the removal of a wide variety of PPCPs. In contrast, alternative treatment methods, such as ozonation and advanced oxidation processes,

Acknowledgments The authors would like to acknowledge and thank Malcolm Pirnie, Inc. and the National Water Research Institute for fellowships they provided to the two students working on this research.

references

Auriol, M., Filali-Meknassi, Y., Tyagi, R., Adams, C., Surampalli, R., 2006. Endocrine disrupting compounds removal from wastewater, a new challenge. Process Biochemistry 41, 525e539. Auriol, M., Filali-Meknassi, Y., Tyagi, R., Adams, C., 2007. Laccasecatalyzed conversion of natural and synthetic hormones from a municipal wastewater. Water Research 41, 3281e3288. Barber, L., Murphy, S., Verplanck, P., Sandstrom, M., Taylor, H., Furlong, E., 2006. Chemical loading into surface water along a hydrological, biogeochemical, and land use gradient: a holistic watershed approach. Environmental Science & Technology 40, 475e486. Berrio, J., Plou, F., Ballesteros, A., Martinez, A., Martinez, M., 2007. Immobilization of Pycnoporus coccineus laccase on Eupergit C: stabilization and treatment of olive oil mill wastewaters. Biocatalysis and Biotransformation 25 (2e4), 130e134. Cabana, H., Jones, J., Agathos, S., 2007a. Elimination of endocrine disrupting chemicals using white rot fungi and their lignin modifying enzymes: a review. Engineering in Life Sciences 5, 429e456. Cabana, H., Habib, J., Rozenberg, R., Elisashvili, V., Penninckx, M., Agathos, S., Jones, P., 2007b. Elimination of endocrine disrupting chemicals nonylphenol and bisphenol A and personal care product ingredient triclosan using enzyme preparation from the white rot fungus Coriolopsis polyzona. Chemosphere 67, 770e778. Calafat, A., Wong, L., Ye, X., Reidy, J., Needham, L., 2008. Concentrations of the sunscreen agent benziohenone-3 in residents of the United States: National Health and Nutrition Examination Survey 2003e2004. Environmental Health Perspectives 116 (7), 893e897. Camarero, S., Canas, A., Nousiainen, P., Record, E., Lomascolo, A., Martinez, M., Martinez, A., 2008. p-Hydroxycinnamic acids as natural mediators for laccase oxidation of recalcitrant compounds. Environmental Science & Technology 42 (17), 6703e6709. Can˜as, A., Camarero, S., 2010. Laccases and their natural mediators: biotechnological tools for sustainable eco-friendly processes. Biotechnology Advances 28, 694e705. Hata, T., Shintate, H., Kawai, S., Okamura, H., Nishida, T., 2010. Elimination of carbamazepine by repeated treatment with laccase in the presence of 1-hydroxybenzotriazole. Journal of Hazardous Materials 181, 1175e1178. Huang, Q., Tang, J., Weber, W., 2005. Precipitation of enzymecatalyzed phenol oxidative coupling products: background ion and pH effects. Water Research 39, 3021e3027. Ikehata, K., El-Din, M., Snyder, S., 2008. Ozonation and advanced oxidation treatment of emerging organic pollutants in water and wastewater. Ozone: Science and Engineering 30, 21e26.

1932

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 2 1 e1 9 3 2

Ikehata, K., Naghashkar, N., El-Din, M., 2006. Degradation of aqueous pharmaceuticals by ozonation and advanced oxidation processes: a review. Ozone: Science and Engineering 28, 353e414. Johannes, C., Majcherczyk, A., 2000. Natural mediators in the oxidation of polycyclic aromatic hydrocarbons by laccase mediator system. Applied and Environmental Microbiology 66 (2), 524e528. Khetan, S., Collins, T., 2007. Human pharmaceuticals in the aquatic environment: a challenge to green chemistry. Chemical Reviews 107, 2319e2364. Kim, S., Cho, J., Kim, I., Vanderford, B., Snyder, S., 2007. Occurrence and removal of pharmaceuticals and endocrine disruptors in South Korean surface, drinking, and waste waters. Water Research 41, 1013e1021. Kim, Y., Nicell, J., 2006a. Laccase-catalyzed oxidation of aqueous triclosan. Journal of Chemical Technology and Biotechnology 81, 1344e1352. Kim, Y., Nicell, J., 2006b. Impact of reaction conditions on the laccase-catalyzed conversion of bisphenol A. Bioresource Technology 97, 1431e1442. Kinney, C., Furlong, E., Zaugg, S., Burkhardt, M., Werner, S., Cahill, J., Jorgensen, G., 2006. Survey of organic wastewater contaminants in biosolids destined for land application. 40, 7207e7215. Ko, C., Fan, C., 2010. Enhanced chemical oxygen demand removal and flux reduction in pulp and paper wastewater treatment using laccase-polymerized membrane filtration. Journal of Hazardous Materials 181, 763e770. Kolpin, D., Furlong, E., Meyer, M., Thurman, M., Zaugg, S., Barber, L., Buxton, H., 2002. Pharmaceuticals, hormones, and other organic wastewater contaminants in U.S. streams, 1999e2000: a national reconnaissance. Environmental Science & Technology 36 (6), 1202e1211. Kunamneni, A., Camarero, S., Garcia-Burgos, C., Plou, F., Ballesteros, A., Alcalde, M., 2008. Review engineering and applications of fungal laccases for organic synthesis. Microbial Cell Factories 7, 32. Kupriyanovich, Y., Medvedeva, S., Rokhin, V., Kanitskaya, L., 2007. Regioselectivity of ferulic acid polymerization catalyzed by oxidases. Russian Journal of Bioorganic Chemistry 33, 516e522. Lloret, L., Eibes, G., Lu-Chau, T., Moreira, M., Feijoo, G., Lema, J., 2010. Laccase-catalyzed degradation of anti-inflammatories and estrogens. Biochemical Engineering Journal 51, 124e131. Lu, J., Huang, Q., Mao, L., 2009. Removal of acetaminophen using enzyme-mediated oxidative coupling processes: I. Reaction

rates and pathways. Environmental Science & Technology 43, 7062e7067. McClellan, K., Halden, R., 2010. Pharmaceuticals and personal care products in archived U.S. biosolids from the 2001 EPA national sewage sludge survey. Water Research 44, 658e668. Nakamura, Y., Mtui, G., 2003. Biodegradation of endocrinedisrupting phenolic compounds using laccase followed by activated sludge treatment. Biotechnology and Bioprocess Engineering 8, 294e298. Rodriguez, S., Toca-Herrera, J., 2007. Laccase production at reactor scale by filamentous fungi. Biotechnology Advances 25, 558e569. Schlumpf, M., Schmid, P., Durrer, S., Conscience, M., Maerkel, K., Henseler, M., Gruetter, M., Herzog, I., Reolon, S., Ceccatelli, R., Faass, O., Stutz, E., Jarry, H., Wuttke, W., Lichtensteiger, W., 2004. Endocrine activity and developmental toxicity of cosmetic UV filters e an update. Toxicology 205, 113e122. Sei, K., Takeda, T., Soda, S., Fujita, M., Ike, M., 2008. Removal characteristics of endocrine-disrupting chemicals by laccase from white-rot fungi. Journal of Environmental Science and Health Part A 43, 53e60. Snyder, S., Adham, S., Redding, A., Cannon, F., DeCarolis, J., Oppenheimer, J., Wert, E., Yoon, Y., 2007. Role of membranes and activated carbon in the removal of endocrine disruptors and pharmaceuticals. Desalination 202, 156e181. Steevensz, A., Al-Ansari, M., Taylor, K., Bewtra, J., Biswas, N., 2009. Comparison of soybean peroxidases with laccase in the removal of phenol from synthetic and refinery wastewaters samples. Journal of Chemical Technology and Biotechnology 84, 761e769. Strong, P., Burgess, J., 2007. Bioremediation of a wine distillery wastewater using white rot fungi and the subsequent production of laccase. Water Science & Technology 56 (2), 179e186. Vajda, A., Barber, L., Gray, J., Lopez, E., Woodling, J., Norris, D., 2008. Reproductive disruption in fish downstream from an estrogenic wastewater effluent. Environmental Science & Technology 42 (9), 3407e3414. Vanderford, B., Pearson, R., Rexing, D., Snyder, S., 2003. Analysis of endocrine disruptors, pharmaceuticals, and personal care products in water using liquid chromatography/tandem mass spectrometry. Analytical Chemistry 75, 6265e6274. Weber, W., Huang, Q., 2003. Inclusion of persistent organic pollutants in humification processes: direct chemical incorporation of phenanthrene via oxidative coupling. Environmental Science & Technology 37, 4221e4227.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 3 3 e1 9 4 0

Available at www.sciencedirect.com

journal homepage: www.elsevier.com/locate/watres

As(III) removal by hybrid reactive membrane process combined with ozonation Hosik Park, Heechul Choi* School of Environmental Science and Engineering, Gwangju Institute of Science and Technology (GIST), 261 Cheomdan-gwagiro, Buk-gu, Gwangju 500-712, South Korea

article info

abstract

Article history:

The removal of arsenite (As(III)) was investigated using a combined ozonation-reactive

Received 9 July 2010

ceramic membrane incorporated with iron oxide nanoparticles (IONs). A disk-type g-Al2O3

Received in revised form

ultrafiltration membrane (CM) was covered with IONs using an annealing method. The

20 December 2010

reactive ceramic membrane (RM) was then characterized using SEM, zeta potential

Accepted 22 December 2010

measurements, and pure water permeability tests. The results showed that IONs were well

Available online 28 December 2010

attached on the RM surface. In addition, doped IONs had no significant effects on the pure water permeability and the isoelectric point (IEP) of RM. Laboratory-scale experiments were

Keywords:

subsequently conducted to investigate the impact of combined RM and ozonation

Arsenite (As(III))

processes on As(III) rejection. The experimental results revealed that As(III) rejection rate of

Reactive ceramic membrane

RM with an ozonation process (92%) significantly enhanced compared with that of CM

Ozonation

(63%). The influence of operating parameters (i.e., pH, NOM, co-existing ions and temper-

Iron oxide nanoparticles

ature) revealed that an increase of pH, a decrease of temperature and presence of NOM led

Rejection

to a higher As(III) rejection, whereas the presence of co-existing ions in the feed water significantly reduced the As(III) rejection; divalent counter-ions were the greatest inhibitors for As(III) rejection. Finally, a comparison of As(III) rejection in synthetic water and real groundwater confirmed the importance of real conditions in the hybrid reactive membrane process with continuous ozonation. ª 2010 Elsevier Ltd. All rights reserved.

1.

Introduction

Arsenic is a naturally occurring metalloid that is present in the environment in a variety of forms (organic and inorganic), oxidation states, and valances depending on both natural and anthropogenic sources (Ferguson and Gavis, 1972). Arsenic in soil and water can naturally occur from the weathering of soil, volcanic activity, or forest fires as well as from anthropogenic sources such as arsenic pesticides, disposal of fly ash, mine drainage, and geothermal discharge (Cullen and Reimer, 2002). To this end, the occurrence of arsenic in natural water is of great concern due to its toxicity and the potential for chronic exposure (Goldberg and Johnston, 2001). Currently,

potable groundwater supplies in many countries around the world contain dissolved arsenic in excess of 10 mg/L; to address this problem, the World Health Organization (WHO) has set this value as the maximum guideline concentration for arsenic in drinking water (Smedley and Kinniburgh, 2002; WHO, 1993). Under normal conditions, arsenic exists in two predominant species, arsenite (As(III)) and arsenate (As(V)), depending on the surrounding pH and redox conditions. As(V) is the major arsenic species in aerobic and oxidizing conditions, whereas As(III) is the dominant arsenic in anaerobic and moderately reducing conditions (Di Natale et al., 2008). At natural pH values, As(III) predominantly exists in solutions as

* Corresponding author. Tel.: þ82 62 715 2441; fax: þ82 62 715 2434. E-mail address: [email protected] (H. Choi). 0043-1354/$ e see front matter ª 2010 Elsevier Ltd. All rights reserved. doi:10.1016/j.watres.2010.12.024

1934

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 3 3 e1 9 4 0

H3AsO3, whereas the As(V) exists in solution as H2AsO1 4 and HAsO2 4 (Pena et al., 2005). However, since their redox reactions are relatively slow, both As(III) and As(V) are often found in soil and subsurface environments regardless of the environmental conditions (Zhang et al., 2007). Treatment technologies for arsenic removal can be categorized into four main groups: coagulation, adsorption, ion exchange, and membrane separation. Although coagulation and adsorption processes provide good arsenic removal efficiency, their major disadvantages are the requirement of multiple chemical processes, regeneration of medium, and treatment of wet solid sludge (Iqbal et al., 2007). Ion exchange has also been applied to arsenic removal (Kim and Benjamin, 2004), though it too has notable limitations due to its operational instability, resin regeneration, and the required treatment of used resin (Lin et al., 2008). Conversely, membrane processes have been receiving considerable attention due to their simple process, reduced maintenance and operational requirements, and the development of high flux membranes (Ng et al., 2004). As such, the removal of arsenic by ultrafiltration (UF), nanofiltration (NF), reverse osmosis (RO), micellar enhanced UF has been reported (Beolchini et al., 2006, 2007; Brandhuber and Amy, 2001; Gecol et al., 2004; Ning, 2002; Sato et al., 2002; Seidel et al., 2001). Specifically, Brandhuber and Amy (2001) reported arsenic rejection based on a negatively charged UF and also investigated the effect of hydraulic operating conditions (i.e., permeate flux, and cross flow velocity) and feed water composition. They observed that rejection of As(V) was higher than that of As(III) over a pH range of 2e10 (Brandhuber and Amy, 2001). In addition, Seidel et al. (2001) reported arsenic rejection through the use of a negatively charged polymeric NF system (Seidel et al., 2001). This membrane showed 60e90% rejection for As(V), and 5e28% rejection for As(III). Although a number of membrane technologies are currently available, due to the fact that As(III) exists in uncharged form at the typical pH of drinking water, the removal of As(III) is more difficult than the removal of As(V). Accordingly, the oxidation of As(III) to As(V) is necessary in order to achieve a high removal efficiency of As(III) by using membrane processes. To date, studies have reported the oxidation of As(III) using different oxidants or oxidant generating processes, including O2 and/or ozone (Kim and Nriagu, 2000), chlorine (Dodd et al., 2006), H2O2 (Pettine et al., 1999), manganese oxides (Chiu and Hering, 2000), the Fenton process (Hug and Leupin, 2003), photochemical processes (Kocar and Inskeep, 2003), and the TiO2/UV process (Bissen et al., 2001), among others. However, each of these methods faces one or more of the following limitations: a) considerable concentration of external oxidants or catalysts are needed for the effective oxidation of As(III); b) high energy input is necessary in the photocatalytic process; or c) it might be necessary to separate particulate catalysts from the treated water. Furthermore, the oxidation process itself has limitations as a pretreatment of As(III) during the membrane process due to the potential destruction of organic membranes by oxidants (Shanbhag et al., 1998). Recently, ceramic membranes (CMs) have been coupled with advanced oxidation processes to enhance membrane performance by generating hydroxyl radicals as well as

making use of the ozone resistance capacity (Karnik et al., 2005a, 2005b; Schlichter et al., 2004). In addition, iron oxide nanoparticles (IONs) have been applied in the adsorption of arsenic in aqueous solutions (Park et al., 2009) and as a catalyst in heterogeneous catalytic ozonation (Jung et al., 2007). Heterogeneous catalytic ozonation is considered a promising process for oxidizing As(III) via the use of hydroxyl radicals prior to the membrane process, assuming that disadvantages of this process, such as the separation of particulate catalysts from the treated water, can be resolved. Hydroxyl radicals generated from heterogeneous catalytic ozonation have a much higher second-order rate constant for the oxidation of As(III) than ozone (von Gunten, 2003). As such, it is conceivable that the combination of ozonation and CM processes incorporated with IONs, acting as a catalyst, will be able to simultaneously oxidize As(III) to As(V) via heterogeneous catalytic ozonation and have a high arsenic rejection capacity due to the CM. Even though, the proposed process is expected to have an excellent As(III) removal efficiency from synthetic water, following issues would be addressed for full scale application: a) disinfection by-product formation; b) optimization of ozone concentration for each target wastewater; c) recirculation of retentate which contains aqueous ozone; d) fouling problems; and e) regeneration and life time of a hybrid reactive membrane. This study is one of the first known attempts to combine heterogeneous catalytic ozonation and a CM process in a hybrid system to remove As(III). Therefore, the main objectives of this study are: a) to prepare a CM incorporated with IONs using an annealing method to enhance the effective removal of As(III); b) to evaluate the membrane’s rejection capacity of As(III) with consideration of feed water chemistry such as pH, NOM, co-existing ions, and temperature; and c) to compare As(III) rejection between synthetic water and real groundwater.

2.

Materials and methods

2.1.

Materials and analytical method

All chemicals used in this study, such as all co-ions (NaCl, NaHCO3, Na2SO4) and counter-ions (MgCl2, CaCl2) were reagent grade obtained from Sigma Aldrich Chemical Co. or Junsei Chemical Co. A powder form of NOM (1R101N) was purchased from the International Humic Substance Society (IHSS) as the NOM source. For this study, all solutions were prepared using Milli-Q water (18.2 MU cm at 25  C), and stock solutions of As(III) (1000 mg/L) and NOM (80 mg/L) were prepared using NaAsO2 (Sigma Aldrich Chemical Co., USA) and Suwannee River NOM (1R101N), respectively. The UF CM used in this study was a g-Al2O3 disk-type UF membrane having a molecular weight cut-off of 7500 D (Inopor, Germany). Prior to experiments, these membranes were cleaned with 1% NaOH for 1 h, 1% HNO3 for 2 h at 55  C, and were finally rinsed with DI water. The groundwater sample was collected from a well located at the Gwangju Institute of Science and Technology, South Korea, and filtered through a 0.45-mM filter prior to use. Note that either 1 M HCl or NaOH

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 3 3 e1 9 4 0

was used for pH adjustment, as needed. The groundwater characteristics were then analyzed by an ion chromatograph (IC; Dionex 2100, USA) equipped with an AS-16 column for anions, and an IC (Dionex DX-120, USA) equipped with a CS-15 column for cations. In addition, a total organic carbon analyzer (TOC; Seivers-820, Seivers Co., USA) was used to analyze the dissolved organic carbon (DOC) in groundwater. The separation of As(III) from the total arsenic treated by the hybrid reactive membrane process with continuous ozonation was performed with solid-phase extraction cartridge (Supelco, 3 mL LC-SAX, USA) that retained As(V). The permeate which is treated with 0.1 N Na2S2O3 to quench the remaining aqueous ozone after the hybrid reactive membrane process was analyzed for total arsenic and As(III) by inductively coupled plasma-optical spectroscopy (ICP-OES; Optima 5300DV, Perkin Elmer, USA) with a detection limit of 1 mg/L. Rejection rate (R) was calculated using the following equation: R¼

CF  CP  100 CF

where CF is the feed concentration and CP is the permeate concentration. The feed water and permeate samples were collected every 30 min to calculate the average As(III) rejection rate for water quality assessment.

2.2. Synthesis and characterization of the reactive ceramic membrane The reactive ceramic membrane (RM) was prepared according to following procedure. In brief, electrochemically synthesized IONs followed by our previous report (Park et al., 2008) were first dispersed in DI water before the solution was sonicated for 2 h. IONs were then spread on the CM surface and then dried in a vacuum drier at 100  C. This procedure was sequentially repeated until the desired amounts of IONs (5 mg and 10 mg) were spread onto the CM surface. Finally, IONsdoped CMs were annealed at 900  C for 5 h under N2 conditions. A scanning electron microscope (SEM; Hitachi S-4700, Japan) was used to observe the attachment of IONs on the CM surface, and the zeta potentials of the membrane surface were determined by the electrophoretic light scattering method

1935

using a light scattering spectrophotometer with a quartz cell (ELS-8000, Photal, Otsuka electronics, Japan). The electrolyte solution was 10 mM KCl and membrane was cleaned with DI water for one day prior to measurements. For the zeta potential measurement, the pH was varied from 2 to 10 with 1 M HCl or NaOH. A pure water permeability experiment was also conducted to compare the difference in the permeability between the cleaned reactive ceramic membrane (RM) and CM. During experiment, the transmembrane pressure (TMP) was kept constant at 80 Kpa and feed water flow rate was controlled at 0.02 L/min, respectively, and the temperature was maintained at 20  1  C.

2.3.

Equipment and filtration procedure

The schematic of the hybrid reactive membrane experimental set-up is illustrated in Fig. 1. The filtration test unit was composed of a membrane cell (3 cm  6 cm; thickness of feed spacer: 0.05 cm), a feed tank, a pump, and a temperature controller. In addition, the ozone generator (PCE-WEDECO, GL-1, USA) was equipped with a mass flow controller to supply gaseous ozone to the filtration test unit. Hybrid reactive membrane filtration experiments were then conducted at a transmembrane pressure (TMP) of 0.2 Mpa under a feed water flow rate of 0.1 L/min, in which 10 g/m3 of gaseous ozone was continuously supplied using Y inline mixer to the filtration test unit and the feed water temperature was maintained at 20  1  C by using a temperature controller. All experiments were conducted without recirculation of permeate and retentate to preserve the feed water from ozone. The effects of pH, NOM, co-existing ions, and temperature on As(III) removal were examined in series of experimental trials. In addition, comparison of As(III) rejection in synthetic water and real groundwater is also presented in order to investigate the importance of real conditions in this hybrid reactive membrane process. As(III) rejection experiments were conducted by changing the operating parameters within the following ranges: a) pH: 3e10; b) NOM: 3.2 mgC/L; c) co-ions: 0.5e10 mM; d) counterions: 0.25e2.5 mM; e) temperature: 4e30  C. Throughout these experiments, the As(III) concentration was fixed at 500 ppb

Fig. 1 e Schematic of the hybrid reactive membrane process unit.

1936

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 3 3 e1 9 4 0

and the pH was maintained at 8.2, except during the pH effect experiment.

3.

Results and discussion

3.1.

Characterization of the reactive ceramic membrane

The CM and RM morphologies were observed by SEM to confirm the attachment of the ION layer on the CM surface. Cross-sectional SEM image (Fig. 2) revealed the formation of an approximately 1.5 mm ION layer when 10 mg of IONs were spread on the CM surface (a 0.6 mm ION layer was obtained for a 5 mg doped CM, figure not shown). In addition, the ION layer was composed of approximately 50 nm nanoparticles in an aggregated form. The zeta potential of CM and RM as a function of pH was then measured to determine the isoelectric point (IEP). As shown in Fig. 3, CM and RM exhibit a net negative charge at a pH higher than the IEP of 4.5 and 5.2, respectively; the IEP for RM shifts to a little higher pH than CM and is marked between that of CM and electrochemically synthesized IONs (IEP: 8.3). Previous study reported similar observations for an iron oxide coated-silica (Xu and Axe, 2005). In addition, the zeta potentials of CM and RM were in the range of 30 mV to 32 mV and 36 mV to 43 mV, respectively. Pure water permeability tests were also conducted to investigate the effect of incorporating IONs onto the CM surface. As shown in Fig. 4, RM permeability was not significantly changed compared to that of CM, as the water permeabilities of CM, 5 mg doped RM, and 10 mg doped RM were 55.3 L/m2 bar h, 53.9 L/m2 bar h, and 53.2 L/m2 bar h, respectively. These results indicate that IONs incorporation onto CM had a negligible effect on CM performance in terms of flux. This result is in accordance with a previous report, in which 15 kD and 5 kD tubular ceramic membranes were coated with iron oxide nanoparticles using a layer-by-layer technique at different sintering temperatures (Karnik et al., 2005a). It was shown that the membrane coating had a negligible effect on its permeability.

Fig. 3 e Zeta potential of CM and RM (Amount of IONs on the membrane surface: 0.56 mg/cm2; Vacuum dried at 100  C; Annealing at 900  C for 5 h).

3.2. Effect of pH on As(III) rejection with/without continuous ozonation Generally, pH plays an important role in membrane performance, since pH affects speciation of the chemical species as well as characteristics of the membrane surface such as surface charge. In Fig. 5, the effects of pH on the average As(III) rejection rate during operation (5 h) for both membranes is depicted, in which the effects of continuous ozonation is considered. For CMs and RMs without ozonation, As(III) rejection slightly increases as pH was increased. In particular, the average As(III) rejection rate increased from 17 to 23% for CMs and from 23 to 34% for RMs in the pH range investigated (pH ¼ 3e9). These results can be explained by the membrane surface charge, as the surface charge of both membranes (Fig. 4) changed from a positive charge to a negative charge as the pH was increased, possibly causing an increase of the membrane repulsive electrostatic force, thereby making it more influential during the membrane process.

CM 5 mg IONs-RM 10mg IONs-RM

70 60

2

Permeability (L/m bar hr)

80

50 40 30 20 0

100

200

300

400

500

600

700

800

Time (min.)

Fig. 2 e Cross-sectional SEM image of RM (10 mg of ION-doped CM).

Fig. 4 e Comparison of CM and RM for pure water permeability (Membrane cell: 3 cm 3 6 cm; Thickness of feed spacer: 0.05 cm; TMP: 80 Kpa, Feed water flow rate: 0.02 L/min; Temperature: 20 ± 1  C).

1937

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 3 3 e1 9 4 0

hydroxyl radical (k$OH ¼ 8.5  109 M1s1) is much greater by approximately 9 orders of magnitude than that with ozone (kO3 > 7 M1s1) (von Gunten, 2003). The hydroxyl radical can be efficiently produced by catalytic decomposition of ozone by IONs (Jung et al., 2007). Therefore, hydroxyl radical produced by reaction between IONs on the RM surface and ozone significantly enhance the As(III) rejection as the oxidation of As(III) to As(V).

100

60

40

3.3. Effect of NOM on As(III) rejection by RM with/without continuous ozonation

20 CM w/o O CM with O

0 2

4

6

RM w/o O RM with O

8

10

pH

Fig. 5 e Effect of pH on As(III) rejection by CM and RM with/without continuous ozonation (As(III) Conc. in the feed water: 500 ppb; Temperature: 20  C; TMP: 0.2 Mpa; Feed water flow rate: 0.1 L/min).

In addition, the rejection of As(III) was practically unaffected by pH conditions varying from 3 to 7, whereas, slightly higher As (III) rejection was observed at pH 9. This phenomenon can be attributed to the charge valance of As(III) in the feed water. According to the arsenic species stability diagram, As(III) mostly exists in the form of neutral H3AsO3 species (pH < 9), whereas monovalent anion (H2AsO 3 ) would predominate with a pH greater than 9. Therefore, the electrostatic repulsion force between the monovalent anion (H2AsO 3 ) and more negative charge of membrane surface is increased resulting in the enhancement of the As(III) rejection above pH 9. The figure also shows that the As(III) rejection of RM without ozonation was higher than that for CM, due to the higher negative zeta potential of RM than CM. The higher negative charge at the RM surface can reject As(III) more effectively due to the increase of electrostatic repulsion. As such, CM and RM with the continuous ozonation process significantly enhanced the As(III) rejection, as compared to CM and RM without ozonation (Fig. 5). This result can be attributed to the oxidation of As(III) to As(V) by ozone and/or hydroxyl radicals. In general, As(V) rejection is higher than the As(III), since As(III) exists primarily as H3AsO3 which is difficult to ionize (H3AsO3, pKa ¼ 9.13) and As(V) exists as an 2 (Goldberg and anion form such as H2AsO 4 and HAsO4 Johnston, 2001). Consequently, the hybrid reactive membrane process with continuous ozonation shows the significantly high rejection of oxidized arsenic. Solid-phase extraction (SPE) of arsenic in the permeate treated by RM with continuous ozonation revealed that 92e99% of the arsenic was retained on the LC-SAX cartridge. This means that As(V) is dominant species of arsenic in the permeate which indicates the oxidiation of As(III) to As(V) by heterogeneous catalytic ozonation in the RM with continuous ozonation process. In addition, the highest As(III) rejection of RM (92%) with continuous ozonation was much higher compared with that of CM (63%) with continuous ozonation. These results could be attributed to the second-order rate constant (k) for the oxidation of As(III) to As(V) by ozone and hydroxyl radical. The second-order rate constant (k) for the oxidation of As(III) with

Suwannee River NOM (3.2 mgC/L) was added into the feed water to observe the effects of NOM on As(III) rejection. As shown in Fig. 6, the average As(III) rejection rate during operation (5 h) with NOM slightly increased compared to the absence of NOM in the feed water; it is likely that NOM adsorbed or concentrated near the membrane surface may inhibit As(III) transport through the membrane (Kim et al., 2006). The effects of NOM on As(III) rejection were more noticeable when no ozone was supplied to the hybrid reactive membrane process. NOM can be degraded to small molecule compounds or be mineralized by ozone and/or hydroxyl radicals (Her et al., 2008). Thus, the decomposition of NOM may reduce the adsorption or accumulation of NOM on the membrane surface, thereby causing less improvement of As (III) rejection for membranes with continuous ozonation than for those without ozonation.

3.4. Effect of temperature on As(III) rejection by RM with continuous ozonation The effect of temperature on As(III) rejection was studied over a temperature range of 4e30  C. Fig. 7 shows the As(III) rejection rate of RM with continuous ozonation as a function of temperature. The removal of As(III) by RM with continuous ozonation was slightly affected by varying the temperature. As the temperature was increased, As(III) rejection decreased. This result can be explained by the fact that the diffusion of As (III) through the RM increases with increased temperature, consequently reducing As(III) rejection and resulting in 100

Arsenic Rejection (%)

Arsenic Rejection (%)

80

80

Non spiked NOM w/o O3 Spiked NOM w/o O3 Non spiked NOM with O3 Spiked NOM with O3

60

40

20

0

CM

RM

Fig. 6 e Effect of NOM on As(III) rejection by RM with/ without continuous ozonation (As(III) Conc. in the feed water: 500 ppb; NOM Conc.: 3.2 mgC/L; pH: 8.2; TMP: 0.2 Mpa; Feed water flow rate: 0.1 L/min).

1938

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 3 3 e1 9 4 0

100

Table 1 e Water quality of natural groundwater. Anions

Cations

As(III) Rejection (%)

80

60

40

o

20

4 C o 20 C o 30 C

0 0

50

100 150 Time (min.)

200

250

Fig. 7 e Effect of temperature on As(III) rejection by RM with continuous ozonation (As(III) Conc. in the feed water: 500 ppb; pH: 8.2; TMP: 0.2 Mpa; Feed water flow rate: 0.1 L/min).

a higher As(III) flux. This result is in accordance with previous studies using charged UF and NF membranes (Brandhuber and Amy, 2001; Figoli et al., 2010).

3.5. Effect of co-existing ions on As(III) rejection by RM with continuous ozonation Co- and counter-ions commonly exist in groundwater and the presence of these ions may affect As(III) rejection (Brandhuber 2 and Amy, 2001). Here, the effects of co-ions (Cl, HCO 3 , SO4 ) 2þ 2þ and counter-ions (Mg , Ca ) on As(III) rejection with continuous ozonation was investigated. The concentrations of co-ions and counter-ions were varied within a range of 0.5e10 mM and 0.25e2.5 mM, respectively. As shown in Fig. 8, the average As(III) rejection rate during operation (5 h) decreased with an increasing concentration of co-existing ions. These results can be explained that the Donnan exclusion plays an important role in RM combined with ozonation

Component

Concentration (mM)

Component

Concentration (mM)

F Cl NO 3 SO2 4

0.013 0.453 0.162 0.014

Naþ Kþ Mg2þ Ca2þ

1.244 0.170 0.054 0.015

 pH: 7.1, DOC: 0.05 mgC/L.

process. Since Donnan potential weakens with increasing the concentration of co-existing ions which increased passage of arsenic. Previous studies using negatively charged UF membrane and negatively charged NF membrane also showed that increasing the concentration of anions and cations increased passage of arsenic (Brandhuber and Amy, 2001; Seidel et al., 2001). In addition, As(III) rejection in the presence of divalent counter-ions (Ca2þ, Mg2þ) was more severely affected than for the experiments with co-ions even if the concentration of counter-ions were lower than that of co-ions. RM used in this study is negatively charged at pH 8.2. Performance of charged and porous membrane can be explained by Donnan effects. When solution containing ions is brought in contact with charged and porous membrane, counter-ions which are an opposite charge of the membrane were accumulated the membrane and co-ions are excluded from the membrane (Marcel, 1996). Accumulation of counter-ion plays an important role to locally neutralize the membrane charge. Therefore, As(III) rejection showed the negative correlation with divalent counter-ions.

3.6. Comparison of synthetic and real groundwater by RM with continuous ozonation As(III) rejection in synthetic water and real groundwater was performed to evaluate the possible effects of the real 100

100

Arsenic rejection (%)

80

60

40

20

Arsenic Rejection (%)

-

Cl HCO 2SO 2+ Mg 2+ Ca

80

60

40

Synthetic Water Real GW Real GW-NOM

20

0 30

0 -4

10

-3

-2

10 10 Concentration of co-existing ion (M)

Fig. 8 e Effect of the presence of co-existing ions for As(III) rejection by RM with continuous ozonation (As(III) Conc. in the feed water: 500 ppb; pH: 8.2; TMP: 0.2 Mpa; Feed water flow rate: 0.1 L/min).

60

90

120

Time (min.)

150

180

Fig. 9 e Comparison of As(III) rejection between As(III)spiked DI water, As(III)-spiked real groundwater, and NOMAs(III)-spiked real groundwater (As(III) Conc. in the feed water: 500 ppb; NOM conc. in the feed water: 3.2 mgC/L; pH: 8.2; TMP: 0.2 Mpa; Feed water flow rate: 0.1 L/min).

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 3 3 e1 9 4 0

groundwater condition. The water quality of the real groundwater is shown in Table 1. Note that 500 ppb of As(III) was spiked into the DI water (Synthetic Water) and real groundwater (Real GW), and that feed water was prepared by adding 3.2 mgC/L of NOM and 500 ppb of As(III) into the real groundwater (Real GW-NOM). As shown in Fig. 9, the highest As(III) rejection was 90% from the Synthetic Water, with rejection from Real GW and Real GW-NOM being 67% and 70%, respectively. As(III) rejection using Real GW decreased compared with that using Synthetic Water due to the fact that the presence of co-existing ions in the Real GW affects the reduction of As(III) rejection; As(III) rejection is not significantly different for the Real GW and Real GW-NOM. It has been reported that NOM in real groundwater plays an important role in enhancing arsenic rejection (Brandhuber and Amy, 2001); however, the hybrid reactive membrane process with continuous ozonation presented here displayed little improvement in As(III) rejection. This result is attributed to the decomposition of NOM by ozone and/or hydroxyl radicals. Due to NOM decomposition, NOM has difficulty accumulating on the membrane surface and passes through the membrane, thereby diminishing the NOM effect on As (III) rejection.

4.

Conclusions

A hybrid reactive membrane was synthesized using iron oxide nanoparticles (IONs) and a commercial ceramic membrane to combine the ozonation process for effective As(III) removal. IONs were well-doped on the reactive membrane surface and it was found that doped IONs had little effect on pure water permeability compared to the original membrane. As(III) rejection from the hybrid reactive membrane with continuous ozonation was higher than that without ozonation due to the oxidation of As(III). In addition, the hybrid ceramic membrane process was shown to be influenced by operating parameters such as pH, NOM, co-existing ions, and temperature. As a common trend, it was observed that an increase of pH and a decrease of temperature increased the As(III) rejection efficiency; whereas co-existing ions significantly reduced the As (III) rejection rate. However, NOM in the hybrid reactive membrane process with continuous ozonation was found to have little effect on As(III) rejection due to the decomposition of NOM. Furthermore, As(III) rejection using real groundwater was less than for the synthetic water due to the inorganic salts in real groundwater. On the basis of our experimental results, a hybrid membrane process combined with ozonation was found to be a suitable process for As(III) rejection, since As(III) is primarily oxidized in this process, and oxidized arsenic is more effectively rejected by this type of membrane.

Acknowledgements This work was funded by the National Research Foundation of Korea (NRF) through the National Research Lab (NRL) program to the Environmental NanoParticle Lab (ENPL) at the Gwangju Institute of Science and Technology, South Korea.

1939

references

Beolchini, F., Pagnanelli, F., De Michelis, I., Veglio`, F., 2006. Micellar enhanced ultrafiltration for Arsenic(V) removal: effect of main operating conditions and dynamic modelling. Environmental Science & Technology 40 (8), 2746e2752. Beolchini, F., Pagnanelli, F., De Michelis, I., Veglio`, F., 2007. Treatment of concentrated arsenic(V) solutions by micellar enhanced ultrafiltration with high molecular weight cut-off membrane. Journal of Hazardous Materials 148 (1e2), 116e121. Bissen, M., Vieillard-Baron, M.-M., Schindelin, A.J., Frimmel, F.H., 2001. TiO2-catalyzed photooxidation of arsenite to arsenate in aqueous samples. Chemosphere 44 (4), 751e757. Brandhuber, P., Amy, G., 2001. Arsenic removal by a charged ultrafiltration membraneeinfluences of membrane operating conditions and water quality on arsenic rejection. Desalination 140 (1), 1e14. Chiu, V.Q., Hering, J.G., 2000. Arsenic adsorption and oxidation at manganite surfaces. 1. Method for simultaneous determination of adsorbed and dissolved arsenic species. Environmental Science & Technology 34 (10), 2029e2034. Cullen, W.R., Reimer, K.J., 2002. Arsenic speciation in the environment. Chemical Reviews 89 (4), 713e764. Di Natale, F., Erto, A., Lancia, A., Musmarra, D., 2008. Experimental and modelling analysis of As(V) ions adsorption on granular activated carbon. Water Research 42 (8e9), 2007e2016. Dodd, M.C., Vu, N.D., Ammann, A., Le, V.C., Kissner, R., Pham, H. V., Cao, T.H., Berg, M., Von Gunten, U., 2006. Kinetics and mechanistic aspects of As(III) oxidation by aqueous chlorine, chloramines, and ozone: relevance to drinking water treatment. Environmental Science & Technology 40 (10), 3285e3292. Ferguson, J.F., Gavis, J., 1972. A review of the arsenic cycle in natural waters. Water Research 6 (11), 1259e1274. Figoli, A., Cassano, A., Criscuoli, A., Mozumder, M.S.I., Uddin, M.T., Islam, M.A., Drioli, E., 2010. Influence of operating parameters on the arsenic removal by nanofiltration. Water Research 44 (1), 97e104. Gecol, H., Ergican, E., Fuchs, A., 2004. Molecular level separation of arsenic (V) from water using cationic surfactant micelles and ultrafiltration membrane. Journal of Membrane Science 241 (1), 105e119. Goldberg, S., Johnston, C.T., 2001. Mechanisms of Arsenic adsorption on amorphous oxides evaluated using macroscopic measurements, vibrational spectroscopy, and surface complexation modeling. Journal of Colloid and Interface Science 234 (1), 204e216. Her, N., Amy, G., Chung, J., Yoon, J., Yoon, Y., 2008. Characterizing dissolved organic matter and evaluating associated nanofiltration membrane fouling. Chemosphere 70 (3), 495e502. Hug, S.J., Leupin, O., 2003. Iron-catalyzed oxidation of Arsenic(III) by oxygen and by hydrogen peroxide: pH-dependent formation of oxidants in the fenton reaction. Environmental Science & Technology 37 (12), 2734e2742. Iqbal, J., Kim, H.-J., Yang, J.-S., Baek, K., Yang, J.-W., 2007. Removal of arsenic from groundwater by micellar-enhanced ultrafiltration (MEUF). Chemosphere 66 (5), 970e976. Jung, H., Park, H., Kim, J., Lee, J.-H., Hur, H.-G., Myung, N.V., Choi, H., 2007. Preparation of biotic and abiotic iron oxide nanoparticles (IOnPs) and their properties and applications in heterogeneous catalytic oxidation. Environmental Science & Technology 41 (13), 4741e4747. Karnik, B.S., Davies, S.H., Baumann, M.J., Masten, S.J., 2005a. Fabrication of catalytic membranes for the treatment of drinking water using combined ozonation and ultrafiltration. Environmental Science & Technology 39 (19), 7656e7661.

1940

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 3 3 e1 9 4 0

Karnik, B.S., Davies, S.H.R., Chen, K.C., Jaglowski, D.R., Baumann, M.J., Masten, S.J., 2005b. Effects of ozonation on the permeate flux of nanocrystalline ceramic membranes. Water Research 39 (4), 728e734. Kim, D.H., Kim, K.W., Cho, J., 2006. Removal and transport mechanisms of arsenics in UF and NF membrane processes. Journal of Water and Health 4 (2), 215e223. Kim, J., Benjamin, M.M., 2004. Modeling a novel ion exchange process for arsenic and nitrate removal. Water Research 38 (8), 2053e2062. Kim, M.-J., Nriagu, J., 2000. Oxidation of arsenite in groundwater using ozone and oxygen. The Science of the Total Environment 247 (1), 71e79. Kocar, B.D., Inskeep, W.P., 2003. Photochemical oxidation of As (III) in ferrioxalate solutions. Environmental Science & Technology 37 (8), 1581e1588. Lin, C.-F., Wu, C.-H., Lai, H.-T., 2008. Dissolved organic matter and arsenic removal with coupled chitosan/UF operation. Separation and Purification Technology 60 (3), 292e298. Marcel, M., 1996. Basic Principles of Membrane Technology. Kluwer Academic Publishers. 269. Ng, K.-S., Ujang, Z., Le-Clech, P., 2004. Arsenic removal technologies for drinking water treatment. Reviews in Environmental Science and Biotechnology 3 (1), 43e53. Ning, R.Y., 2002. Arsenic removal by reverse osmosis. Desalination 143 (3), 237e241. Park, H., Ayala, P., Deshusses, M.A., Mulchandani, A., Choi, H., Myung, N.V., 2008. Electrodeposition of maghemite (g-Fe2O3) nanoparticles. Chemical Engineering Journal 139 (1), 208e212. Park, H., Myung, N., Jung, H., Choi, H., 2009. As(V) remediation using electrochemically synthesized maghemite nanoparticles. Journal of Nanoparticle Research (On-line published).

Pena, M.E., Korfiatis, G.P., Patel, M., Lippincott, L., Meng, X., 2005. Adsorption of As(V) and As(III) by nanocrystalline titanium dioxide. Water Research 39 (11), 2327e2337. Pettine, M., Campanella, L., Millero, F.J., 1999. Arsenite oxidation by H2O2 in aqueous solutions. Geochimica et Cosmochimica Acta 63 (18), 2727e2735. Sato, Y., Kang, M., Kamei, T., Magara, Y., 2002. Performance of nanofiltration for arsenic removal. Water Research 36 (13), 3371e3377. Schlichter, B., Mavrov, V., Chmiel, H., 2004. Study of a hybrid process combining ozonation and microfiltration/ ultrafiltration for drinking water production from surface water. Desalination 168, 307e317. Seidel, A., Waypa, J.J., Elimelech, M., 2001. Role of charge (Donnan) exclusion in removal of arsenic from water by a negatively charged porous nanofiltration membrane. Environmental Engineering Science 18 (2), 105e113. Shanbhag, P.V., Guha, A.K., Sirkar, K.K., 1998. Membrane-based ozonation of organic compounds. Industrial & Engineering Chemistry Research 37 (11), 4388e4398. Smedley, P.L., Kinniburgh, D.G., 2002. A review of the source, behaviour and distribution of arsenic in natural waters. Applied Geochemistry 17 (5), 517e568. von Gunten, U., 2003. Ozonation of drinking water: part I. Oxidation kinetics and product formation. Water Research 37 (7), 1443e1467. WHO, 1993. Guidelines for Drinking-water Quality. In: Recommendations, second ed., vol. 1. WHO, Geneva. Xu, Y., Axe, L., 2005. Synthesis and characterization of iron oxidecoated silica and its effect on metal adsorption. Journal of Colloid and Interface Science 282 (1), 11e19. Zhang, G., Qu, J., Liu, H., Liu, R., Wu, R., 2007. Preparation and evaluation of a novel FeeMn binary oxide adsorbent for effective arsenite removal. Water Research 41 (9), 1921e1928.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 4 1 e1 9 5 0

Available at www.sciencedirect.com

journal homepage: www.elsevier.com/locate/watres

Gravity drainage of activated sludge: New experimental method and considerations of settling velocity, specific cake resistance and cake compressibility Dominik Dominiak, Morten Christensen*, Kristian Keiding, Per Halkjær Nielsen Department of Biotechnology, Chemistry and Environmental Engineering, Aalborg University, Sohngaardsholmsvej 49, 9000 Aalborg, Denmark

article info

abstract

Article history:

A laboratory scale setup was used for characterization of gravitational drainage of waste

Received 27 May 2010

activated sludge. The aim of the study was to assess how time of drainage and cake dry

Received in revised form

matter depended on volumetric load, SS content and sludge floc properties. It was

1 December 2010

demonstrated that activated sludge forms compressible cakes, even at the low pressures

Accepted 24 December 2010

found in gravitational drainage. The values of specific cake resistance were two to three

Available online 31 December 2010

orders of magnitude lower than those obtained in pressure filtration. Despite the compressible nature of sludge, key macroscopic parameters such as time of drainage and

Keywords:

cake solid content showed simple functional dependency of the volumetric load and SS of

Activated sludge

a given sludge. This suggests that the proposed method may be applied for design purposes

Gravity drainage

without the use of extensive numerical modeling. The possibilities for application of this

Sludge compressibility

new technique are, among others, the estimation of sludge drainability prior to mechanical

Sludge settling

dewatering on a belt filter, or the application of surplus sludge on reed beds, as well as

Cake resistance

adjustments of sludge loading, concentration or sludge pre-treatment in order to optimize the drainage process. ª 2010 Elsevier Ltd. All rights reserved.

1.

Introduction

The activated sludge process is the most widespread technology of biological wastewater treatment. One of the most important challenges is the handling of surplus activated sludge, which is a side product of the process. The water content of this sludge exceeds 90% by weight, which requires dewatering it in order to make the handling and transportation of sludge physically and economically feasible. Several techniques of activated sludge dewatering exist, some employ mechanical devices like filter presses or centrifuges, others, like reed beds or belt presses, depend on gravitational water drainage. Although the gravitational drainage of sludge is an economically attractive

alternative to sludge pressure dewatering, it has received little attention from researchers so far. Most research has been conducted on pressure dewatering of activated sludge in devices such as filter presses (Novak et al., 1999). The works of Severin et al. (Severin and Grethlein, 1996; Severin et al., 1999), and of Olivier et al. (2004) lead to the development of models describing the gravity drainage of activated sludge on belt presses, aiming at improved design of these devices. However, in both cases the cake compressibility was omitted. Most of the work carried out on sludge drying reed beds has been purely empirical and based on indirect sludge quality estimation using a capillary suction time (CST) method (Nielsen, 2003). It is the claim of the present authors,

* Corresponding author. Department of Biotechnology, Chemistry and Environmental Engineering, Aalborg University, Sohngaardsholmsvej 57, DK-9000 Aalborg, Denmark. Tel.: þ45 99408464. E-mail address: [email protected] (M. Christensen). 0043-1354/$ e see front matter ª 2010 Elsevier Ltd. All rights reserved. doi:10.1016/j.watres.2010.12.029

1942

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 4 1 e1 9 5 0

that design models for reed beds are insufficient, hence a more fundamental modeling approach has been taken. In a recent report (Christensen et al., 2010) we presented a comprehensive characterization of gravity drainage of compressible organic materials with a novel technique, revealing the importance of settling velocity of particles, low pressure filtrate expression, cake compressibility and cake collapse due to capillary forces. The model system used in that study was a suspension of dextraneMnO2 particles, which had been reported as a good representative of compressible organic slurries and resembles physico-chemical properties of activated sludge (Hwang et al., 2006). A novel technique for the assessment of drainage properties of compressible organic materials was created. It was reported that pressures exceeding the critical pressure do not accelerate drainage, and load and feed concentration were identified as the key parameters deciding the drainage time and the final dry matter content of the cake. The aim of this study was to adapt this new test method to the assessment of drainability of full-scale activated sludge in an easy, fast and repetitive manner and to assess the relationship between macroscopic variables such as the volumetric load and suspended solids on the parameters characterizing the drainability of the sludge, namely the time of drainage and cake dry matter content.

2.

Theory of the drainage process

Gravity drainage process differs from constant pressure filtration in that the pressure decreases as the filtrate is expressed and that the cake is partly built up due to the settling process, so the rate of cake formation is not proportional to filtrate volume. In order to correctly describe the gravity drainage process, knowledge of settling velocity, specific cake resistance and cake compressibility is necessary. Eq. (1) (Severin and Grethlein, 1996) describes the drainage rate through a filter cake. vd ¼

DP mðau þ Rm Þ

(1)

It is important to remember that throughout this paper the term ‘specific cake resistance’ is used to describe the average specific resistance of the entire cake, and not its local values. Cake resistance values differ across a compressible cake and are the highest at the cake/filter interface. This inhomogeneity problem was solved by the introduction of the average specific cake resistance term defined as in the pressure filtration literature (Tiller and Yeh, 1987; Teoh et al., 2002). When filtering activated sludge in a constant pressure apparatus, the average specific cake resistance is denoted specific resistance to filtration (SRF) (Sørensen et al., 1996). In this study it is assumed that the average specific cake resistance is constant throughout the drainage experiment. In fact, resistance decreases during the initial part of the drainage process because the pressure difference across the cake decreases with sample level. In our previous report (Christensen et al., 2010), the compression of the MnO2edextran cake exerted by the high hydrostatic pressure present at the beginning of the drainage process was however demonstrated to be irreversible and, for this reason, changes in the average specific

cake resistance as drainage takes proceeds can be neglected (Christensen et al., 2010). The pressure difference, defined as stress on the medium surface, is given in Eq. (2). DP ¼

  Mg r þ rght þ ug ¼ cgh0 1  A rs

(2)

Settling velocity is not included in Eq. (2), but it still influences the drainage rate, because it regulates the deposited amount of solids (u), and thereby the drainage rate. In order to determine the settling velocity and the specific cake resistance in the easiest way, the drainage process was divided into three stages, i.e., cake formation, pure filtration and cake collapse (Fig. 1). During the first stage, sludge particles settle and form a cake on the filter medium surface, leaving the clear watereair interface behind (stage A). This provides information on the settling velocity. When the settling is complete (time described as t1), the sludge blanket level remains constant and the level of pure water decreases as filtrate is expressed (stage B), which reveals the data necessary to calculate the specific cake resistance as described below. At t2 e denoted as the time of drainage e all the free water above the cake disappears, the cake starts to collapse slightly (stage C). It is important to remember that Eq. (2) can only be used to describe drainage until time t2, when the drag forces not included in Eq. (2) start to rise to the cake surface. The amount of deposited cake has a direct influence on the drainage rate and can be calculated according to Eq. (3).  u¼

cðh0  ht Þ þ cvs t ch0

stage A stage B

(3)

As indicated in Eq. (3), the settling velocity influences cake buildup and therefore indirectly influences the drainage process.

2.1.

Cake formation

During the initial phase of drainage, when filtration cake develops, three phases can be distinguished, i.e., the cake itself, the sludge particle suspension and the clear liquid above the suspension. The total level of the sample in the cylinder is given as ht ¼ hw þ hs þ hc. Since the particles settle, the clear water phase above the suspension develops, and the height of this water phase at a given time depends on settling velocity (Eq. (4)). hw ¼ vs t

(4)

The height of the clear water phase, which is the difference between the total level of the sample ht, and the level of the sludge blanket hs þ hc, is measured during the experiment and can be used to determine the settling velocity by revealing the speed at which the particles move away from the liquid surface. The settling velocity turned out to be constant during stage A in every experiment performed in this study.

2.2.

Pure filtration

At the time described as t1, settling is over and all sludge particles have formed a cake, the thickness of which remains

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 4 1 e1 9 5 0

1943

Fig. 1 e The three stages in gravitational drainage of activated sludge and other compressible organic slurries: cake formation (A), pure filtration (B), and cake collapse (C).

constant during stage B. This is a highlight of irreversible compression which took place in the beginning of stage A, when the pressure was the highest. The amount and consequently the height of deposited solids depend on the feed concentration and the volume of the sample used. Assuming

that the pressure on the medium/cake interface is primarily an effect of the hydrostatic pressure component (simplified Eq. (2)), and that rht >> ch0(1  r/rs)u, it is possible to derive an equation describing the level of the sample during the entire stage of pure filtration, as in Christensen et al. (2010).

1944

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 4 1 e1 9 5 0

ht ¼ ht ðt1 Þesðtt1 Þ

(5)

where rg s¼ mðach0 þ Rm Þ

(6)

Once t1 and t2 are identified from Fig. 2, and Eq. (5) is fitted to experimental data from stage B, s can be determined from Eq. (5), providing a e the specific cake resistance e according to Eq. (6). The resistance of the filtration medium itself was neglected, since it was found that it accounted for less than 1% of total resistance in a separate experiment with pure water running through a clean filter medium (data not shown).

2.3.

Cake collapse

At the time described as t2, the air reaches the cake surface. Menisci are formed at the cake surface, which results in a drag on the cake surface, and the cake starts to collapse (Barr and White, 2006). This behavior has been documented by using dextraneMnO2 particles instead of sludge whereby it is possible to measure the porosity profile through the cake during drainage (Christensen et al., 2010). Eq. (7) has been used previously to describe the cake compression of sludge at low pressure (Curves et al., 2009).  b ps 4 ¼ 40 1 þ pa

(7)

The structure pressure ps is a function of wet cake weight and thus a function of initial solids concentration and volume of the sample. At the end of the drainage process, the structure pressure ( ps) is mainly a function of the drag at the cake surface because the cake dry weight is low compared with weight of the wet cake. For that reason, the porosity is constant through the cake as also shown experimentally (Christensen et al., 2010). The structure pressure is proportional with the wet mass of the cake because the drag at the

surface arises from the weight of the liquid within the cake. Hence for practical use Eq. (8) can be used.  b M 4 ¼ 40 1 þ Ak

The final dry matter content is expected to increase with initial solids concentration and initial load, as the wet cake weight increases with solids concentration and load.

3.

Materials and methods

3.1.

Samples of activated sludge

All the experiments were performed on samples of mixed liquor activated sludge from the aeration tank of the Aalborg East Wastewater Treatment Plant in Aalborg, Denmark. It is a Biodenipho plant with biological N and P removal. The suspended solids (SS) contents of the samples were adjusted by dilution with supernatant originating from the same batch of activated sludge, or by the removal of supernatant after 30 min of settling. The dry matter contents (SS) of activated sludge and filtration cakes were determined according to standard methods (APHA, 2005) by overnight weight loss at 104  C.

3.2.

Filter medium choice

Four different filtration media of different cut-off values were tested in drainage experiments in order to assess the influence of filtration medium on the experimental outcome. The media used were the following: Kemira cloth (Kemira, Denmark, cut-off e 200 mm), polyester fabric (cut-off e 40 mm), Whatman 41 paper (Whatman, UK, cut-off e 20e22 mm) and Whatman 40 paper (Whatman, UK, cut-off e 8 mm). In each case a 200 ml sample of activated sludge originating from the same batch was used (source e Aalborg East WWTP, SS e 4.9 g/L). Due to its high cut-off value, Kemira cloth turned out to let sludge particles through in the initial phase of the experiment (9e10 s) until the cake buildup was complete. For this reason the filtration cake was thinner and comprised larger particles. This resulted in lower resistance values as compared to other media, for which this phenomenon was not observed. The other three media behaved very similarly and yielded cakes of similar resistance values. Whatman 41 paper was chosen for further experiments, due to the fact that it is routinely used for the examination of activated sludge in SRF experiments (Christensen and Dick, 1985), and that it is capable of retaining nearly all particles, thus creating a complete cake.

3.3.

Fig. 2 e Experimental setup used in drainage experiments. Beaker collects the filtrate drained from the tube through the filtration medium. A web camera records a film illustrating the drainage and settling processes. Data from the web camera is recorded by a laptop computer.

(8)

Gravity drainage and sedimentation

The experimental setup (Fig. 2) was the same as described in Christensen et al. (2010) and consisted of a transparent glass cylinder (inner diameter of 60 mm), a filter medium closing one end of the pipe and supported by a funnel located inside the tube, and a laptop computer connected to a web camera. The filter was mounted on one end of the cylinder, which was then mounted vertically above a beaker. A sludge sample of a given

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 4 1 e1 9 5 0

volume was introduced into the cylinder and gravitationally drained of water into a beaker, while the web camera filmed the process at a specified frame rate. After completion of the experiment, the film sequence was divided into single frames, and each frame was analyzed separately with ImageJ software (http://rsbweb.nih.gov/ij). The positions of sludge blanket and liquid surface (Fig. 3) as functions of time were recorded and imported to a spreadsheet for further data handling. Evaluation of the repetitiveness of the results provided by the experimental technique was carried out by measuring the specific cake resistance (a) five times on the same sample of untreated activated sludge from Aalborg East WWTP (SS of 5.1 g/l). The mean specific cake resistance turned out to be 4.8  1010 m/kg, with a standard deviation equal to 5% of the mean value, showing that the experimental technique is repetitive and reliable.

3.4.

Analytical centrifugation

Analytical centrifugation was used to study compressibility of sludge cakes (Lumiziser 613 Dispersion Analyser from L.U.M. GmbH, Berlin, Germany). For each test, 2 ml of activated sludge were added to rectangular cuvettes (10 mm). Five samples with varying amounts of dry matter content were analyzed. The raw sludge sample was diluted by filtering the sludge e the filtrate was used to vary the dry matter content of the samples from 5 to 25 g/L. The experiment was operated at 1000 rpm for 2000 s, 4000 rpm for 2000 s and then lowering the rotation speed to 1000 rpm for 2000 s. The height of the formed cake was measured during the experiment and the structure pressure within the cake was calculated using the method described in Sobisch et al. (2006). The cake height depended on both the rotation speed and the dry matter content of sludge.

3.5. Determination of settling velocity and specific cake resistance

1945

height (hw) over the drainage process, which initially increases linearly until time t1, and then decreases until time t2 (Fig. 4). The settling velocity was determined as the slope of the line fitted to data from stage A, according to Eq. (4), and it turned out to be 1.8  105 m/s for a representative experiment. This value was identical in a comparative settling experiment on the same sample of sludge, when draining was absent, and therefore the concept of zone settling, assuming the equal settling velocity of all particles, was applied. Cake compression was irreversible, no cake swelling was observed during the drainage experiment even through the pressure declines during stage B as confirmed from the analytical centrifugation experiment. Further, two drainage experiments were performed, one with simultaneously settling and drainage, and one where all particles were settled before the drainage was started. No significant difference in the determined average specific cake resistance was found. Thus, cake compression was irreversible and the average specific cake resistance was constant during stage B. This implies that the assumption hitherto stated was validated, and changes in the average specific cake resistance as drainage takes proceeds can be neglected. Specific cake resistance is a parameter describing the difficulty in dewatering a given sample. Typical values of specific cake resistance for activated sludge, determined by constant pressure filtration with pressure of 1e2 bar, are of the order of 1012 m/kg (Rasmussen et al., 1994). The technique described in this study allows for the determination of specific cake resistance during low pressure gravity drainage. In a representative experiment, the specific cake resistance during gravity drainage was 4.2  1010 m/kg, which is much lower than those usually found in SRF experiments. The much lower values of average specific cake resistance originate from much lower pressures encountered in gravity drainage resulting in lower cake compression and lower cake permeability loss.

Calculations of the settling velocity and specific cake resistance were based upon the changes of clear water phase

Fig. 3 e Settling and draining curves for a typical drainage experiment. Sludge sampled in Aalborg East WWTP, SS e 4.8 g/L.

Fig. 4 e Example of a relationship between the time and the difference between liquid level and sludge blanket level. The slope of the ascending part provides the settling velocity of sludge, and the function fitted to the descending part provides information on the specific resistance of the filtration cake.

1946

3.6.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 4 1 e1 9 5 0

Shear experiments

Shear experiments were performed in order to investigate the influence of different treatments on the quality of activated sludge in terms of drainability, as described by the specific cake resistance. A batch of fresh activated sludge from Aalborg East WWTP was divided into four samples. One sample was used directly for measurements of specific cake resistance in fresh sludge, the second was stored anaerobically at room temperature for 24 h, the third was sheared for 6 h at the rate of 300 rpm in an aerated reactor and the fourth sample was sheared for 6 h in an identical reactor, but at the rate of 800 rpm. Shear experiments were performed according to Klausen et al. (2004).

4.

Results

4.1.

The effect of volumetric load on drainage

Five different volumes (100e400 ml) of mixed liquor activated sludge, originating from the same sample batch, were tested for their drainage properties. The settling velocity, the specific cake resistance, water content of the filtration cake and the time necessary for the liquid above the cake to drain were determined for each sample (Fig. 5). The relationship between the volumetric load of the sample and the time of drainage showed that tD depends on the volumetric load squared. (Fig. 5a). Specific cake resistance, given by the a value, increased linearly with the volume of sample drained (Fig. 5b). Settling velocity was independent of the volumetric load (Fig. 5c). The total solids content of the cake increased when the volumetric load increased, which is discussed below.

4.2.

The effect of sludge concentration on drainage

The influence of sludge SS concentration on drainage dynamics was assessed in an experiment with five different SS contents (2.7e6.7 g/L). The specific cake resistance, settling velocity, water content of the filtration cake and the time necessary for the liquid above the cake to drain were determined for each sample (Fig. 6). The relationship between the SS content and the resulting time of drainage appeared clearly linear (Fig. 6a). The values of the specific cake resistance (a) tended to remain constant as SS was changed (Fig. 6b). An interesting behavior could be noticed in the case of settling velocity values, which significantly dropped as the SS concentration was increased (Fig. 6c). This was attributed to the hindered settling effect. The Vesilind equation (Vesilind, 1968) was fitted to the relationship between the SS concentration and the settling velocity, providing the Vesilind maximum settling velocity of 1.75 m/h and the Vesilind parameter of 0.58 m3/kg.

4.3. The effect of various sludge treatments on drainage characteristics Specific cake resistance is a parameter which describes the quality of sludge with respect to dewaterability. In pressure filtration studies, the average specific cake resistance

Fig. 5 e Results of the drainage experiments with five different volumes of activated sludge (source e Aalborg East WWTP, SS e 4.7 g/L). (a) Relationship between volume of sludge sample tested and the time of drainage. (b) Relationship between the volume of sludge sample tested and the specific cake resistance. (c) Relationship between the volume of sludge sample tested and the settling velocity of sludge particles. correlates with floc size and strength. The objective of this investigation was to determine whether the same effect can be observed in gravity drainage. Some of the well-described deflocculating factors are shear and anaerobic conditions

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 4 1 e1 9 5 0

1947

(SS content e 5.1 g/L) was divided into four samples. For each of the samples three volumes were drained, which resulted in four distinct linear relationships between volumetric load and specific cake resistance, one for each sample subjected to different treatments (Fig. 7). Shearing at the rate of 300 rpm caused a mild effect, consisting in the decrease in drainability, described as the slope of load vs. specific cake resistance relationship. A stronger effect was observed by a 24 h anaerobic storage, but the most significant drop in drainage speeds resulted from vigorous shearing at 800 rpm. The most probable reason for these phenomena is that both shear and anaerobic conditions have a deflocculating effect on activated sludge, resulting in floc fragmentation and liberation of small cell aggregates and single cells (Rasmussen et al., 1994). Higher resistances observed with deflocculated sludge can be attributed to small particles clogging, or blinding, the pores inside the cake, causing slower water flow and more significant cake compression due to liquid pressure.

4.4.

Effects on cake dry matter content

The cake volume as well as the cake dry matter increases when the volumetric load and/or the suspended solid concentrations are increased. These observations might by merged according to Eq. (8). By expressing the mass of solids (M ) as the volumetric load multiplied by the suspended solids concentration, and by substituting the solid volume fraction by cake dry matter content of the cake, the results of both experiments may be presented as shown in Fig. 8. The data are fitted by the rewritten version of Eq. (8), shown as Eq. (9).  b Vload $SS DMcake ¼ DMcake;0 $ 1 þ A$k

(9)

From Fig. 8 it can be seen that the cake solids are compressed as the dry matter load is increased. It may however also be noted that the solids compression has not reached a maximum (a plateau).

Fig. 6 e Results of the drainage experiments with five samples of sludge with different SS contents (source e Aalborg East WWTP). (a) Relationship between the SS content of the sludge sample tested and the time of drainage. (b) Relationship between the SS content of the sludge sample tested and the specific cake resistance. (c) Relationship between the SS content of the sludge sample tested and the settling velocity.

(Wilen et al., 2000; Nielsen et al., 2004). In order to assess the effect of shear and anaerobic conditions on the quality of sludge in terms of drainability, which is reflected by the specific cake resistance, a series of experiments was run. A batch of fresh activated sludge from Aalborg East WWTP

Fig. 7 e Results of the drainage experiments comparing the effect of 24 h anaerobic storage of activated sludge, shearing at 300 rpm and shearing at 800 rpm to the drainage properties of activated sludge from a single sample batch.

1948

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 4 1 e1 9 5 0

Fig. 8 e Relationship between the load dry matter content and cake dry matter content for combined data from experiments with varying loads and varying concentrations. Curve is fitted according to Eq. (9).

Cake compression was further studied by using analytical centrifugation. A cake was formed at low rotation speed; hence, the compression pressure through the cake was low. The rotation speed was then increased and it was observed that the cake collapses. By lowering the rotation speed after the collapse, it was possible to measure the degree of the cake reswelling. Fig. 9 shows data from the experiment. The cake height did not increase proportional with dry matter load because the cake was compressible cf. Fig. 8. Further, reswelling of the cake was below 20% of the compression and the assumption that cake compression was irreversible is found reasonable. Hence, if a sludge cake is compressed, it is not possible to reestablish the permeability of the cake by lowering the pressure.

5.

Discussion

The present work demonstrates an experimental device for characterizing the drainage process by identifying the

Fig. 9 e Results of analytical centrifugation experiment where the cake height is measured at different rotation speed (1000 rpm and 4000 rpm) and feed concentration.

sedimentation and filtration features in one experimental run. It was found the drainage process may by described by 3 stages: A e the initial sedimentation stage, B e the filtration step and C e the consolidation stage. The interpretation of the experimental data was made according to the model devised by Christensen et al. (2010). It was found that the model, which originally was developed on MnO2edextran particles, adequately described the sludge data produced in this work (Fig. 2) An implication of this is that even at the low pressures encountered in drainage (0e2000 Pa), the compressibility of the sludge has to be included in the model. The compressible nature of sludge is reflected in the cake resistance. In the drainage experiments the specific cake resistance was found in the order of 1010 m/kg, whereas this value is in the order of 1012e1013 m/kg in pressurized filtrations of unconditioned sludge. Indeed, the lower cake resistance makes drainage less costly than pressurized dewatering in terms of energy consumption, however at the expense of the cake dry matter, which for drainage is typically 4e6% compared to 13e17% for pressure dewatering. When compressibility is a dominant feature, the dewatering is a non-uniform process rendering higher solid content in the bottom than in the top of the filter cake. This implies that a full description of the process requires a description of local properties such as flow, resistance, solid content. This in turn requires the use of numerical modeling of the process, which most often is a cumbersome affair. The authors set out to investigate how macroscopic process variables such as volumetric load, suspended solid content and sludge quality influenced the result of the drainage process expressed by time of drainage and cake dry matter. It was found that the time of drainage depended on the 2 Þ. This experimental result volumetric load square ðtD fVload was confirmed by the model of Christensen et al. (2010). As the load increases, the hydrostatic pressure on the formed cake increases, hence due to cake compression the cake resistance increases. This was confirmed by the linear increase in the specific cake resistance as a function of the volumetric load (Fig. 5b). When the sludge physical structure was deteriorated by e.g., intensive shearing or exposure to anaerobic conditions, the state of flocculation was altered. This implies an increased compressibility. The simple relations between drainage time and specific cake resistance versus volumetric load were however maintained, but the proportionality constants changed with the change of sludge structure: the more deflocculated the sludge, the higher the specific cake resistance and subsequently the longer the drainage time. As the solids content is increased, maintaining the same volumetric load, the filtration properties expressed as the specific cake resistance are not altered, which results from the same hydrostatic pressure. However, a linear dependency of the drainage time on the solid content was found. The increase in solid content will increase the cake height, which again will increase the time of drainage. The formation of the cake is an increasingly slower process as the solid content is increased. This is seen on Fig. 6c, where vs is depicted as function of solid content. As predicted by Vesilind, vs decreases as solid content is increased. The impact of settling was discussed in Christensen et al. (2010). The cake dry matter (the solid volume fraction of the filter cake) was in Fig. 8 shown to depend on the

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 4 1 e1 9 5 0

dry matter load. As the dry matter load is proportional to the structure pressure in the cake, the fitting parameters obtained in Fig. 8 correspond to those obtained by the use of Eq. (7), namely: 40, pa and b values at 0.012, 0.15 and 0.25, respectively. This corresponds very well to the findings of Curves et al. (2009), who found the parameters to be equal to 0.0095, 0.3 and 0.25, respectively. This means that the compression of activated sludge filtration cakes at low pressures can be accurately predicted by Eq. (7) and that this equation can be effectively used to describe data obtained with the experimental technique presented in this study, which in practice is easier to perform and requires much simpler equipment than the technique presented in Curves et al. (2009). It is also important to note that in contrast to Sørensen and Sørensen (1997) the compression did not reach a steady state (plateau). They studied dead end filtration at low pressures, yet still pressures vastly exceeding those encountered in this study. This implies that the sludge is not completely collapsed in the gravity drainage and thus a filter skin, as known from pressure filtration, has not been formed. This explains why average values of resistances and cake dry matter are meaningful in this context, whereas they hardly are so in pressurized dewatering of compressible sludges.

6.

Conclusions

 For gravity drainage, the specific resistance of the cake depends directly and linearly on the volumetric load of the sample, due to the compressible nature of activated sludge flocs which tend to produce more compact cakes when pressure of the liquid is higher.  The final dry matter content of the cake is a function of both feed load and concentration (Eq. (9)). It is therefore of primary importance to choose the drainage conditions (feed volumetric load and concentration) carefully, so that the effect of the gravity drainage process is satisfactory, both in terms of the total drainage time and the final cake water content.  The experimental technique is very simple, but very well capable of determining the settling velocity, average specific cake resistance and cake compressibility, and it provides a very good understanding of the process of gravitational drainage of activated sludge. Due to the compactness of the setup and the ease of operation, this technique could be applied to rapidly assess the quality of sludge in terms of drainability prior to its application to dewatering devices such as belt presses, or on sludge dewatering reed beds. Especially in the case of reed beds, the benefits would be immense, since the operational failures due to sludge overdosing could be avoided.

Nomenclature List of symbols A cross-sectional area of the cylinder (m2) c particle concentration in the feed (kg/m3) DMcake dry matter content in cake (e) DMcake,0 dry matter content in cake before compression (e)

g h0 hc hs ht hw k M P pa ps Rm SS tD Vload

1949

gravitational acceleration (m/s2) initial level of the suspension (m) height of the cake (m) distance between cake surface and suspensionewater interface (m) actual level of the suspension (m) height of the clear water phase (m) constant (m2/kg) mass of the sample (kg) pressure (Pa) fitting parameter in Eq. (7) structure pressure (Pa) media resistance (m1) suspended solids (kg/m3) time of drainage (s) volumetric load of the sample (m3)

Greek symbols a specific resistance to filtration (m/kg) b fitting parameter in Eqs. (7) and (8) r density of the filtrate (kg/m3) density of the particles (kg/m3) rs s characteristic drainage time defined in Eq. (6) (s) drainage rate (m/s) nd settling velocity (m/s) ns m filtrate viscosity (Pa s) 4 solids volume fraction solids volume fraction for ps equal to 0 40 u amount of deposited material per unit area of media (kg/m2) Parameter values A 0.002827 m2 pa 0.15 b 0.25 m 0.001 Pa s g 9.81 m/s2 r 987 kg/m3

references

APHA, 2005. Standard Methods for the Examination of Water and Wastewater. American Public Health Association, American Water Works Association, Water Environment Federation, Washington D.C. Barr, J.D., White, L.R., 2006. Centrifugal drum filtration: II. A compression rheology model of cake draining. American Institute of Chemical Engineers Journal 52, 557e564. Christensen, G.L., Dick, R.I., 1985. Specific resistance measurements: methods and procedures. Journal of Environmental Engineering 111, 258e271. Christensen, M.L., Dominiak, D.M., Nielsen, P.H., Sedin, M., Keiding, K., 2010. Gravitational drainage of compressible organic materials. American Institute of Chemical Engineers Journal 56 (12), 3099e3108. Curves, D., Saveyn, H., Scales, P.J., Van der Meeren, P., 2009. A centrifugation method for the assessment of low pressure compressibility of particulate suspensions. Chemical Engineering Journal 148, 405e413.

1950

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 4 1 e1 9 5 0

Hwang, K.J., Lyu, S.Y., Chen, F.F., 2006. The preparation and filtration characteristics of dextraneMnO2 gel particles. Powder Technology Journal 161, 41e47. Klausen, M.M., Thomsen, T.R., Nielsen, J.L., Mikkelsen, L.H., Nielsen, P.H., 2004. Variations in microcolony strength of probe-defined bacteria in activated sludge flocs. FEMS Microbiology Ecology 50, 123e132. Nielsen, P.H., Thomsen, T.R., Nielsen, J.L., 2004. Bacterial composition of activated sludge importance for floc and sludge properties. Water Science and Technology 49, 51e58. Nielsen, S., 2003. Sludge drying reed beds. Water Science and Technology 48, 101e109. Novak, J.T., Agerbæk, M.L., Sørensen, B.L., Hansen, J.A., 1999. Conditioning, filtering, and expressing waste activated sludge. Journal of Environmental Engineering 125 (9), 816e824. Olivier, J., Vaxelaire, J., Ginisty, P., 2004. Gravity drainage of activated sludge: from laboratory experiments to industrial process. Journal of Chemical Technology and Biotechnology 79, 461e467. Rasmussen, H., Bruus, J.H., Keiding, K., Nielsen, P.H., 1994. Observations on dewaterability and physical, chemical and microbiological changes in anaerobically stored activated sludge from a nutrient removal plant. Water Research 28, 417e425. Severin, B.F., Grethlein, H.E., 1996. Laboratory simulation of belt press dewatering: application of Darcy equation to gravity drainage. Water Environment Research 68, 359e369.

Severin, B.F., Nye, J.V., Kim, B.J., 1999. Model and analysis of belt drainage thickening. Journal of Environmental Engineering 125 (9), 807e815. Sobisch, T., Lerche, D., Detloff, T., Beiser, M., Erk, A., 2006. Tracing the centrifugal separation of fine-particle slurries by analytical centrifugation. Filtration 6, 313e321. Sørensen, B.L., Sørensen, P.B., 1997. Structure compression in cake filtration. Journal of Environmental Engineering 123 (4), 345e353. Sørensen, P.B., Agerbæk, M.L., Sørensen, B.L., 1996. Predicting cake filtration using specific filtration flow rate. Water Environment Research 68, 1151e1155. Teoh, S.K., Reginald, B.H.T., Tien, C., 2002. Correlation of CeO cell and filtration test data using a new test cell. Separation and Purification Technology 29, 131e139. Tiller, F.M., Yeh, C.S., 1987. The role of porosity in filtration part XI: filtration followed by expression. American Institute of Chemical Engineers Journal 33, 1241e1256. Vesilind, P.A., 1968. Theoretical considerations: design of prototype thickeners from batch settling tests. Water Sewage Works 115, 302e307. Wilen, B.-M., Nielsen, J.L., Keiding, K., Nielsen, P.H., 2000. Influence of microbial activity on the stability of activated sludge flocs. Colloids and Surfaces B: Biointerfaces 18, 145e156.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 5 1 e1 9 6 0

Available at www.sciencedirect.com

journal homepage: www.elsevier.com/locate/watres

Numerical modelling of sedimentebacteria interaction processes in surface waters Guanghai Gao a,*, Roger A. Falconer b, Binliang Lin b,c a

Key Laboratory of Pollution Processes and Environmental Criteria (Ministry of Education), College of Environmental Science and Engineering, Nankai University, Tianjin 300071, China b Cardiff School of Engineering, Cardiff University, The Parade, Cardiff CF24 3AA, UK c State Key Hydroscience and Engineering Laboratory, Tsinghua University, Beijing 100084, China

article info

abstract

Article history:

Faecal bacteria exist in both free-living and attached forms in surface waters. The depo-

Received 28 March 2010

sition of sediments can take faecal bacteria out of the water column and to the bed. The

Received in revised form

sediments can subsequently be re-suspended into the water column, which can then lead

24 December 2010

to the re-suspension of the faecal bacteria of the attached form back into the water

Accepted 27 December 2010

column, where it may desorb from the sediments. Therefore, the fate and transport of

Available online 8 January 2011

faecal bacteria is highly related to the governing sediment transport processes, particularly where these processes are significant. However, little attempt has been made to model

Keywords:

such processes in terms of predicting the impact of the sediment fluxes on faecal bacteria

Numerical modelling

levels. Details are given of the refinement of a numerical model of faecal bacteria transport,

Faecal bacteria

where the sediment transport processes are significant. This model is based on the model

Sediment transport

DIVAST (Depth Integrated Velocities And Solute Transport). Analytical solutions for steady and uniform flow conditions were derived and used to test the sedimentebacteria interaction model. After testing the sedimentebacteria interaction model favourably against known results, the model was then set up for idealised case studies to investigate the effects of sediment on bacteria concentrations in the water column. Finally the model was applied to a simplified artificial flooding study to investigate the impact of suspended sediment fluxes on the corresponding bacteria transport processes. The model predictions have proved to be encouraging, with the results being compared to field measurements. ª 2011 Elsevier Ltd. All rights reserved.

1.

Introduction

Faecal bacteria are widely used worldwide as indicators to monitor surface water quality. Faecal bacteria in surface waters can be considered to exist in two forms, either as freeliving bacteria within the water column, or attached (or adsorbed) to suspended particles. The bacteria can be transported and diffused within the water column in their freeliving form, or they can be adsorbed onto the sediments and then transported and diffused with the sediments. The

adsorbed bacteria can settle out when the suspended particles deposit on the bed and then be re-suspended with the particles into the overlying water column when the sediment particles are re-suspended. Advances in numerical modelling of hydrodynamic and water quality processes have made such tools an invaluable means of predicting faecal contamination levels under different flow conditions. Numerical models used for predicting bacterial contamination have generally treated faecal bacteria as free-living in current studies, such as Lin and

* Corresponding author. Tel.: þ86 15 022623290; fax: þ86 22 23501117. E-mail address: [email protected] (G. Gao). 0043-1354/$ e see front matter ª 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.watres.2010.12.030

1952

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 5 1 e1 9 6 0

Falconer (2001), Kashefipour et al (2002), and the deposition/ re-suspension and adsorption/desorption processes are not generally included in the model and with little attempt being made to model such processes in terms of predicting the impact of the suspended sediment fluxes on bacteria levels in the water column. However, in recent years there have been many studies undertaken to study how bacteria exist in the sediments, with these studies frequently revealing higher numbers of indicator and pathogenic bacteria in the sediments than in the overlaying water column, both in marine and fresh water systems (Fries et al., 2006; Jamieson et al., 2004; Characklis et al., 2005). Fries et al. (2006) investigated the attachment of faecal indicator bacteria to particles in the Neuse rive estuary, in eastern North Carolina, USA, and found that an overall average of 38% bacteria associated with particles. Characklis et al. (2005) found attachment ratios of typically 30e55% for enterococci in storm water. Gannon et al. (1983) showed that sedimentation was an important element in the overall faecal coliform disappearance rates in a river impoundment. Suspended sediments can contribute to the disappearance of faecal bacteria from the water column in different ways. Attached faecal bacteria are adsorbed by the sediments from the water column during low energy flow conditions (Howell et al., 1996). Sediment concentrations also affect the light penetration rate in the water column, which further affects the decay rate of faecal bacteria (Stapleton et al., 2007). Allen et al. (1987) revealed that the water quality testing criteria generally do not take account of sediments as a potential reservoir of pathogens. The higher numbers of pathogenic levels occurring in sediments creates a potential health hazard from re-suspension and subsequent ingestion from increasing usage of recreational waters. Therefore, there is a need to obtain additional information on the survival of indictor and pathogenic bacteria in sediments and the factors which contribute to their survival (Allen et al., 1987). Jamieson et al. (2005) conducted field experiments in Swan Creek, Canada, by using the bacteria tracer E. coli NAR in their studies. They found that the bacteria tracer that re-appeared in the water column coincided with increases in the total suspended solids load, which indicated that the E. coli NAR were being re-suspended with the sediment load. E. coli NAR is a form of E. coli that is: resistant to nalidixic acid, non-pathogenic, rarely found in the natural environment and possesses survival characteristics similar to other E. coli (Jamieson et al., 2004). Some faecal bacteria modelling efforts have been made to include sediment effects on bacteria. Steets and Holden (2003) included sediment effects on faecal bacteria fate and transport in a coastal lagoon by introducing a constant attachment ratio throughout the modelling period. Jamieson et al. (2005) developed a model for attached faecal bacteria for steady state flow conditions. Yang et al. (2008) and Stapleton et al. (2007) have recently developed bacteria transport models for the Severn Estuary. Both deposition and re-suspension processes were modelled for the attached bacteria. However, similar to Steets and Holden (2003), attachment ratios were assumed to be constant. In this study, details are given of the refinement of a numerical model of faecal bacteria transport to include

sediment effects on bacteria fate and transport in surface waters, where the sediment transport processes are significant. This model is based on the model DIVAST (Depth Integrated Velocities And Solute Transport), which has been successfully applied to many sites, such as for predicting hydrodynamic and faecal bacteria transport processes in Swansea Bay (Lin and Falconer, 2001), faecal coliform transport in the Ribble Estuary (Kashefipour et al., 2002) and sediment transport processes in the Humber Estuary (Lin and Falconer, 1996).

2.

Hydrodynamic modelling

For many estuarine and coastal water basins the vertical velocity component is relatively small in comparison with the horizontal velocity component. Hence the continuity and momentum equations can be integrated over the depth of flow and solved numerically to give the depth averaged velocity fields (Falconer, 1993). The depth integrated Reynolds Averaged NaviereStokes equations are shown below: vx vUH vVH þ þ ¼0 vt vx vy

(1)


2  vUH vU H vUVH vx sxw sxb  þb þ ¼ fVH þ gH þ r r vt vx vy vx      v vU v vU vV 3H 3H þ2 þ þ vx vx vy vy vx (2)
 vVH vUVH vV2 H vx syw syb  þb þ ¼ fUH þ gH þ r r vt vx vy vy      v vU vV v vV 3H 3H þ þ þ2 vx vy vx vy vy (3) where x ¼ water elevation above (or below) datum; U,V ¼ depth averaged velocity components in the x, y directions; H ¼ x þ h ¼ total water depth; h ¼ water depth below datum; b ¼ momentum correction factor; f ¼ Coriolis parameter; sxw,syw ¼ surface wind shear stress components in the x, y direction; sxb,syb ¼ bed shear stress component in the x, y directions; and 3 ¼ depth averaged eddy viscosity. The momentum correction factor, the wind and bed shear stresses, and the depth averaged viscosity are described in detail in Falconer (1993).

3.

Suspended sediment transport modelling

Sediment transport formulations for predicting suspended sediment fluxes in depth integrated two-dimensional numerical models are generally based on solving the depth integrated form of the advective-diffusion equation, which can be shown to be of the form:     vSH vSUH vSVH v vS v vS þ þ  HDx  HDy ¼ED vt vx vy vx vx vy vy

(4)

where S ¼ depth averaged suspended sediment concentration, E ¼ sediment erosion rate, D ¼ sediment deposition rate, and Dx,Dy ¼ depth averaged dispersion coefficients in the x and y directions, respectively.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 5 1 e1 9 6 0

The depth averaged net sediment flux rate for non-cohesive sediment can be expressed in the form (Li et al., 2001; Yuan, 2007): E  D ¼ gws ðSe  aSÞ

(5)

where ws ¼ particle settling velocity, g ¼ a profile factor given by the ratio of the bed concentration Sa (i.e. the concentration at an elevation ‘a’ above bed) to the depth averaged equilibrium sediment concentration, Se ¼ depth-averaged equilibrium concentration, which can be determined from an appropriate sediment transport formula such as van Rijn (1984a,1984b,1993), with this being one of the most widely used formulations incorporated into computational models and included in the study reported herein, a ¼ Se/S is a sediment concentration profile factor. The equilibrium concentration is that value which occurs when the sediment flux vertically upwards from the bed due to turbulence is in equilibrium with the net sediment flux downwards due to the fall velocity associated with gravity. For modelling cohesive sediment transport the governing depth integrated advective-diffusion Eq. (4) is used, but with the net sediment flux being rewritten in the following form (Falconer and Chen, 1996):
D¼
E¼

i h b sb  sc;d ws S 1  ssc;d 0 sb > sc;d M 0

h

sb sc;e sc;e

i

sb > sc;e sb  sc;e

(7)

where sb ¼ bed shear stress, sc,d ¼ critical shear stress for deposition, sc,e ¼ critical shear stress for erosion and M ¼ empirical erosion constant. Most of the parameters included in the above formulations are sensitive to the sediment characteristics locally. During numerical modelling studies of estuarine flows the value used must be chosen with extreme care. Typically values of the critical stress for erosion and deposition are given in Van Rijn (1993) for a range of different mud types. For the empirical erosion coefficient M, reported values used in the current study for the Severn Estuary study are typically in the range of 0.00001e0.0005 for soft natural mud (Falconer and Chen, 1996).

4.

Sedimentebacteria interaction model

4.1.

Partition of bacteria between sediments and water

The total bacteria concentration in the water column CT is given by: CT ¼ Cd þ Cp

where S ¼ the suspended solid concentration, and P ¼ the mass-specific bacteria concentration, which can be defined as follows: S¼

Ms Vwþs

(10)

P¼

CFUp Ms

(11)

where Ms ¼ solid mass of sediment, Vwþs ¼ total volume of the water and solids, and CFUp ¼ colony forming unit of attached bacteria. Chapra (1997) expressed the tendency of bacteria to attach to the particles by using a partition coefficient of the form: KD ¼

P Cd

(12)

Assuming that the rate at which bacteria are adsorbed and/ or desorbed from a particulate is fast, then a local equilibrium can be assumed to give: CT ¼ Cd þ KD $S$Cd

(8)

where Cd ¼ free-living bacteria concentration and Cp ¼ attached bacteria concentration in the water column. For a given concentration of suspended solids, the quantity of faecal bacteria on the particles is often expressed as a massspecific concentration P (cfu/unit weight of suspended solids), so the volume-specific concentration on the particles Cp can be expressed as: (9) Cp ¼ S$P

(13)

which can be solved to give: Cd ¼ fd CT

(6)

1953

(14)

and fd ¼

1 1 þ KD S

(15)

where fd is the fraction of free-living bacteria in the water column. For the attached bacteria, we have: Cp ¼ fp CT

(16)

where fp ¼

KD S 1 þ KD S

(17)

and: f p þ fd ¼ 1

(18)

4.2. Exchange of bacteria at the sedimentewater interface 4.2.1.

Bacteria settlement

One of the effects of sediment transport on adsorbed bacteria is that when the sediment settles out then the adsorbed bacteria is also taken out of the water column to the bed sediments. The flux of adsorbed bacteria from the water column to the bed sediments, Fdep, can be expressed as: Fdep ¼ qdep P

(19)

where Fdep ¼ flux of adsorbed bacteria from the water column to the bed sediments (cfu/cm2/s), qdep ¼ sediment deposition flux (kg/m2/s), and P ¼ Cp/S attached faecal bacteria concentration on the suspended sediments (cfu/0.1 g).

1954

4.2.2.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 5 1 e1 9 6 0

Bacteria re-suspension

The re-suspension of bacteria from the bed sediments to the water column Fero, can be expressed as: Fero ¼ qero Pb

(20)

where: Fero ¼ re-suspension of bacteria from the bed sediments to the water column (cfu/cm2/s), Pb ¼ bacteria concentration on the bed sediments (cfu/0.1 g), and qero ¼ sediment re-suspension flux rate (kg/m2/s). To summarise, the net bacteria flux, Fnet, due to settling and re-suspension of the sediments can be expressed as:

  Fnet ¼ max qero ; 0 Pb þ min qdep ; 0 P

4.2.3.

The concentration of bacteria on the bed sediments, Pb, varies depending on the exchange of bacteria between the water column and the bed sediments. However, another reduction also arises in the bed sediment concentration as a result of the decay of the bacteria within the bed sediments. Assuming that the deposited sediments from the water column to the bed and the bed sediments are well mixed immediately after deposition, then the exchange rate of bed bacteria concentration Pb can be expressed in the following form: (22)

where Mb ¼ mass of bed sediments per unit area, and kb ¼ faecal bacteria decay/growth rate in the bed sediments. Likewise, in Eq. (22), the mass of bed sediments per unit area, Mb, also varies temporarily as given by: dMb ¼ qdep  qero dt

(23)

4.3. Governing equations for bacteria transport processes 4.3.1.

Free-living bacteria transport

The fate and transport of free-living bacteria can be described by the following two-dimensional depth integrated advectiondiffusion equation:     vCd H vCd UH vCd VH v vCd v vCd  þ þ  HDx HDy vx vy vt vx vy vx vy ¼ Cdo þ Cdt  kCd H

4.3.2.

Attached bacteria transport

In studying the transport of attached bacteria, this part of the bacteria may be transported and diffused with the sediments. The fate and transport of attached bacteria can therefore be described by the following two-dimensional depth integrated advection-diffusion equation:     vHCp vUHCp vVHCp vCp vCp v v þ þ   HDx HDy vt vx vy vx vy vx vy p

(21)

Bacteria concentrations in the bed sediments

dPb qdep ¼ ðP  Pb Þ  kb Pb dt Mb

erosion and deposition, have not been included in the transport model.

(24)

where Cd ¼ depth averaged free-living bacteria concentration, Cdo ¼ source or sink of free-living bacteria; Cdt ¼ transformation term defining the desorption of bacteria from the sediments to the free-living form and vice versa; and k ¼ the decay rate of bacteria in the water column. Eq. (24) has been solved to predict bacteria concentration levels for most studies of bacteria transport modelling, such as Kashefipour et al. (2002) and Lin and Falconer (2001), with this representation having been proven to represent the process accurately for the case where sediment transport is not significant. However, for studies where sediment transport processes are significant, then solving these equations alone will not give accurate results since the transport of bacteria through the process of sediment transport, via

p

¼ Cpo þ Ct þ Cb  kCp H

(25)

where Cp ¼ depth averaged attached bacteria concentration in the water column, Cpo ¼ source or sink of bacteria in its attached form; Cpt ¼ transformation term defining the adsorption of free-living bacteria to the attached bacteria form or vice versa; Cpb ¼ source term defining the attached bacteria from or to the bed sediments, for sediment erosion or deposition, respectively; and k ¼ the decay rate for bacteria in the water column.

4.3.3.

Total bacteria transport

In order to predict bacteria concentrations correctly, both for free-living and attached bacteria, then the transport equation must be solved simultaneously in the numerical model for both bacterial components. However, there are difficulties in solving these equations accurately since the transformation terms are difficult to quantify. The transformation processes between the free-living and adsorbed state for the bacteria are very complex, so it is almost impossible to quantify these terms. Wu et al. (2005) pointed out that in modelling heavy metals there was a problem in using separate equations to model dissolved and particulate metals due to the complex nature of the transformation between the particulate and dissolved phase. Adding Eqs. (24) and (25) and using Cdt ¼ Cpt gives     vHCT vUHCT vVHCT v vCT v vCT þ þ   HDx HDy vt vx vy vx vy vx vy p

¼ Cdo þ Cpo þ Cb  kCT H

(26)

where CT ¼ depth averaged total faecal bacteria concentration, Cpb ¼ Fnet, which is a source term defining the attached bacteria from, or to, the bed sediments and Fnet can be calculated from using Eq. (21) as follows:

  Fnet ¼ max qero ; 0 Pb þ min qdep ; 0 P By solving the total bacteria transport equation, then the total bacterial concentration level CT is determined, wherein Eqs. (14) and (16) can then be used to determine the free-living and attached bacteria levels respectively.

5. Sedimentebacteria interaction model verification In this section analytical solutions of the sedimentebacteria interaction are detailed. Idealised cases are also set up to

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 5 1 e1 9 6 0

investigate the effects of different environmental factors and parameters. The sedimentebacteria interaction model developed is based on the original DIVAST model framework which was originally developed by Falconer (1984). The DIVAST model has been used by many researchers (Wu et al., 2005; Lin and Falconer, 1997) in the past, so only the new sedimentebacteria interaction model has been tested against the analytical solutions for re-suspension and deposition of sediment and bacteria. In this section two analytical solutions for sedimentebacteria interaction case studies have been investigated, namely for a deposition and re-suspension test. These solutions were first derived and then the numerical model was set up for test cases. The numerically predicted results were compared with the analytical results in order to test the sedimentebacteria interaction properties of the numerical model.

5.1.

Derivation of analytical solutions

5.1.1.

Case 1: deposition test

Under steady flow conditions the governing equation for total bacteria can be simplified to the following form: dCT 1 p ¼ Cb  kCT dt H

1. The sediment particle size was assumed to be uniform across the domain, so that the settling velocity ws was set to be a constant; 2. The initial sediment concentration S0was assumed to be greater than the equilibrium concentration Se, which meant that the sediment and attached bacteria settled down onto the bed and the concentration of the sediment and bacteria in water column kept on decreasing until equilibrium had been reached; 3. The bacteria decay rate was assumed to be a constant. Following on from these assumptions, the bacteria and sediment concentration in the water column and the bed sediments can be obtained analytically, with details of the solutions being given below.

where Cpb represents the reduction of bacteria due to the deposition of sediments, which is given as: p

Cb ¼ qdep P ¼ gws PðSe  aSÞ

CT0

(28)

and P ¼ Cd KD ¼

KD CT 1 þ KD S

(29)

where k is the decay rate. Hence, Eq. (27) can be expressed in the following form: dCT gws KD CT ¼ ðSe  aSÞ  kCT dt H 1 þ KD S

dS gws ðSe  aSÞ ¼ H dt

(30)

(31)

Yuan (2007) derived an analytical solution for this sediment transport Eq. (31), given as:   1 1 S ¼ Se þ S0  Se elt a a

(32)

where l¼

agws H

(33)

For the derivation of the analytical solution for Eq. (30), an operator splitting scheme proposed in Lin and Falconer (1997) is used. This equation can be treated as the combination of the following two equations: dCT gws KD CT ¼ ðSe  aSÞ dt H 1 þ KD S

(34)

dCT ¼ kCT dt

(35)

Yuan (2007) derived the analytical solution of Eq. (34) to validate his heavy metal model, together with the analytical solution for dCT =dt ¼ kCT given as CT ¼ CTekt. Hence the analytical solution of Eq. (30) can be obtained by combining these two analytical solutions to give: CT ¼

S0

(27)

The sediment concentration in this equation can be obtained by solving the sediment transport equation. For steady and uniform flow conditions, the governing equation for suspended sediment transport can be simplified to give:

This test was set up to represent the deposition processes of attached bacteria due to sedimentebacteria interactions, as illustrated in Fig. 1 and where P is the bacteria concentration on the suspended sediments in cfu/(101 g), Pb is the bacteria concentration on the bed sediments in cfu/(101 g), H is the water depth, hb is the bed sediment thickness, S0 is the initial sediment concentration in the water column in kg/m3, and which is set to a constant, and CT is the total bacteria concentration in the water column in cfu/100 ml. To simplify this problem some basic assumptions have been made including:

KD

1955



a þ KD Se þ aS0  Se elt 0 kt  CT e a 1 þ KD S0

(36)

P 5.1.2.

ws

H

U

Pb hb Bed Sediment Fig. 1 e Illustration of test case set up.

Case 2: re-suspension test

This case is based on the deposition test, except that the initial conditions are now changed so that the initial sediment concentration S0 is set lower than the equilibrium value Se and the initial bacteria concentration in the bed sediments Pb is set to be a constant other than zero. In this case the fate and transport of the total bacteria can also be expressed using Eq. (30), but here Cpb represents the source of bacteria due to sediment erosion, giving:

1956

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 5 1 e1 9 6 0

p

Cb ¼ qero Pb ¼ gws Pb ðSe  aSÞ

Total Bacteria Concentration

(37)

Assuming a first order decay for the bacteria concentration in the bed sediments, then we get: (38)

dCT gws 0 kb t e ¼ P e ðS  aSÞ  kCT dt H b

(39)

which gives the analytical solution as:    P0 ekb t  e S  aS0 1  elt ekt CT ¼ C0T þ b a

Case 1: deposition

The computational parameters were set as follows: 1. Water depth ¼ 1 m; 2. Parameters for sediment transport: a ¼ 1, g ¼ 1 and the fall velocity ws ¼ 0.001 m/s and the equilibrium sediment concentration Se was set to 1 kg/m3; 3. The partition coefficient KD was set to10 l/g, as suggested in Bai and Lung (2005) and the initial sediment and water column bacterial concentrations were set to be 2 kg/m3 and 100 cfu/100 mlrespectively. These parameter values gave an initial ratio of attached to total bacteria of about 0.952. The decay rate in the water column was set to 1 day1. 4. The initial bacteria bed concentration was assumed to be zero. The comparisons between the model calculated and analytical solutions are shown in Fig. 2. From these plots it can be seen that the model predicted results are nearly identical to the analytical solution for the bacteria and suspended sediment concentrations. Fig. 2 shows that the sediment deposition processes reach the equilibrium condition after about 3600 s, or 1 h. After reaching equilibrium conditions the sediment concentrations were kept at a constant level. The decay process meant that the total bacteria concentration kept reducing after equilibrium conditions had been achieved for the sediment transport. It can be seen that the bacteria loss due to sediment deposition occurred in a fairly short time compared to the loss due to decay.

5.2.2.

0

7200

Case 2: re-suspension

The set up of the numerical model for this case was the same as that for Case 1, except for changes in some initial conditions. For this case S0 ¼ 0, CT ¼ 0 and P0b ¼ 100 cfu/0.1 g. The comparison between the model predicted and analytical solutions are shown in Fig. 3. From these results it can be seen that the model results are again nearly identical to the analytical results, for both the bacteria and suspended sediment concentrations. Fig. 3 shows that the sediment re-suspension processes reached equilibrium conditions

14400

Model

21600

28800

36000

Time (s)

Verification of model against analytical solutions

In this section details are given of the testing of the sedimentebacteria interaction model against the analytical solutions derived in the previous section.

5.2.1.

120 100 80 60 40 20 0

(40)

Sediment Concentration Model

Analytical Se dime nt C onc entration (k g/m3)

5.2.

Tota l Fa e c a l Ba c te r ia C onc entration (c fu/100ml)

Pb ¼ P0b ekb t

Analytical

2.5 2 1.5 1 0.5 0 0

7200

14400

21600

28800

36000

Time (s) Fig. 2 e Comparison of bacteria and sediment concentrations for deposition test.

again after about 3600 s or 1 h. After equilibrium conditions had been reached the sediment concentration remained at a constant level. The bacteria concentration level decreased gradually, which was due to bacterial decay. During the resuspension process the decay processes were suppressed by the rapid increase in the bacteria level due to re-suspension.

6.

Idealised case application

The sedimentebacteria interaction model has been shown to solve the governing equations accurately and can therefore be used with some confidence in investigating the effects of sediments on the fate and transport of bacteria. In this section idealised test cases were set up to study the effects of removing bacteria from the water column and the subsequent re-suspension of bacteria from the bed.

6.1. Removal of bacteria from the water column due to sediment 6.1.1.

Effect of the partition coefficient

In order to investigate the effect of the partition coefficient on the removal of bacteria from the water column, partition coefficient KD values of 10 l/g, 1 l/g, 0.1 l/g and 0.01 l/g were used. The initial bacteria bed concentration was assumed to be zero, the initial sediment concentration was set to be 2 kg/m3 and the initial bacteria concentration in the water column was set to 100 cfu/100 ml. The decay rate in the water column was

1957

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 5 1 e1 9 6 0

Total Bacteria Concentration

Tota l B acter ia C oncentration (cfu/1 0 0 ml)

Analytical

6.1.2. Effect of different sediment sizes giving different settling velocities

Model

100 80 60 40 20 0 0

7200

14400

21600

28800

36000

Time (s) Sediment Concentration

S e diment C oncentration (k g/m3)

Analytical

Model

1.2 1 0.8 0.6 0.4 0.2 0 0

7200

14400

21600

28800

Settling velocities of ws ¼ 0.001 m/s, 0.0001 m/s, 0.00001 m/s were used respectively to consider the effects of particle sediment size on the removal of bacteria from the water column. The initial bacteria bed concentration was assumed to be zero, the initial sediment concentration was set to be 2 kg/m3 and the initial bacterial concentration in the water column was set to be 100 cfu/100 ml. The decay rate in the water column was set to 1 day1, and the decay rate in the bed sediments was assumed to be zero. The partition coefficient KD was set to10 l/g, which gave an initial ratio of attached to total bacteria of 0.952, which would decrease with a corresponding decrease in the sediment concentration within the water column. The equilibrium sediment concentration Se was set to1 kg/m3. The numerical model results are shown in Fig. 5, where it can be seen that higher settling velocities reduced the total bacteria concentration level much quicker than the lower settling velocity. The concentration of free-living bacteria was not affected by the settling velocity. The reduction in the free-living bacteria in the water column was therefore purely due to decay.

6.2.

Re-suspension of attached bacteria to water column

6.2.1.

Effect of bed bacteria concentration

36000

Time (s) Fig. 3 e Comparison of bacteria and sediment concentrations for re-suspension test.

set to 1 day1. The parameters for sediment transport were set to: a ¼ 1, g ¼ 1 and the fall velocity ws ¼ 0.001 m/s and the equilibrium sediment concentration Se was set to be 1 kg/m3. The numerical model results are shown in Fig. 4. It was observed that with a higher partition coefficient then lower total bacteria concentrations were predicted in the water column. The reason for this finding is that the higher partition coefficient gave higher ratios of attached to total bacteria for the same sediment concentrations, which meant that more bacteria were deposited on the bed and thereby giving a lower concentration in the water column.

The initial bed bacteria concentration assumed in the investigations was either 100 cfu/0.1 g, or 50 cfu/0.1 g, or10 cfu/0.1 g, with this value being used to investigate the effect of the bed concentrations on the re-suspension of the attached bacteria. The initial sediment concentration was assumed to be zero and the initial bacterial concentration in the water column was also set to zero. The decay rate in the water column was set to1 day1. Parameters for the sediment transport model components included: a ¼ 1, g ¼ 1, fall velocity ws ¼ 0.001 m/s, equilibrium sediment concentration Se ¼ 1 kg/m3 and the partition coefficient KD ¼ 10 l/g. The numerical model predictions are shown in Fig. 6. From these results it can be seen that higher bed bacteria concentrations gave rise to much higher bacteria concentrations in the water column. Higher bed bacteria concentrations means that more bacteria contribute to the water column under the same conditions for re-suspension, i.e. the bacteria reTotal bacteria concentration

Total bacteria concentration KD =1

Ws=1m m /s

KD=0.1

100 80 60 40 20 0

0

7200

14400

21600

28800

Tim e (s)

Fig. 4 e Removal of bacteria with different partition coefficients.

36000

Total bacteria concentration (cfu/100ml)

Total bacteria concentrat ion (cfu/100ml)

KD=10

Ws=0.1m m /s

Ws=0.01m m /s

100 80 60 40 20 0

0

7200

14400

21600

28800

Tim e (s)

Fig. 5 e Removal of bacteria with different settling velocities.

36000

1958

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 5 1 e1 9 6 0

Sediment Concentration at Site A

Total bacteria concentration Pb=50cfu/0.1g

Pb=10cfu/0.1g

Data

100 80 60

100 0

40

24

26

E coli Concentration at Site A

0

7200

14400

21600

28800

36000

Data E c oli (cfu/1 0 0 m L)

Tim e (s)

Fig. 6 e Re-suspension of bacteria with different bed bacteria concentrations.

10000 0 24

26

The numerical model was applied to a published artificial flooding study. Muirhead et al. (2004) conducted a study in Total bacteria concentration

Total bacteria concentration (cfu/100ml)

Water Level at Site A

KD=1

KD=0.1

100 80 60 40 20 0

0

7200

14400 21600 Tim e (s)

28800

36000

Fig. 7 e Re-suspension of bacteria with different partition coefficients.

Water Level (m)

Data

7. Model application: artificial flooding case study

28

Tim e(hr s)

Effect of partition coefficient

For this test case different partition coefficients were used to investigate the effect of the partition coefficient on the resuspension of bacteria and the values considered included: KD ¼ 10 l/g, 1 l/g, 0.1 l/g and 0.01 l/g respectively. The initial bed bacteria concentration was assumed to be 100 cfu/0.1 g, the initial sediment concentration was set to zero and the initial bacterial concentration in the water column was set to zero. The decay rate in the water column was set to 1 day1. The governing parameters for the sediment transport model were: a ¼ 1, g ¼ 1, the fall velocity ws ¼ 0.001 m/s and the equilibrium sediment concentration Se was set to 1 kg/m3. The numerical model results are shown in Fig. 7, where it can be seen that the partition coefficient does not significantly affect the total bacteria concentration in the water column in the re-suspension mode.

KD=10

Model

20000

suspended into the water column with the sediments and then re-partition into the water column.

6.2.2.

28

Tim e (hr s)

20 0

Model

200 Se diment (m g /L )

Total bact eria concentrat ion (cfu/100ml)

Pb=100cfu/0.1g

Model

1 0.5 0

22

24

26

28

30

Time(hrs)

Fig. 8 e Comparison of model results and site data at Site A.

Topehaehae Stream in the Waikato Region, New Zealand, to investigate faecal bacteria transport during floods. This study was reproduced numerically by Bai and Lung (2005). The median flow rate was 260 l/s and the average stream width was 5.8 m. The relatively straight stream was simplified to a straight and uniform river. A water supply reservoir, located at the upstream end of the river, was used as the source of water for the artificial flooding. The reservoir was the only source of water supply during the artificial flooding experiment. The artificial floods were created by opening the valve of a dam for over 30 min, holding the valve for 20 min, and then closing the valve over 10 min for three successive days and with the peak flow reaching 4300 l/s. Faecal bacteria and turbidity were sampled at sites A and B, located at 1.3 and 2.5 km downstream of the reservoir respectively. A weir equation was used at the downstream boundary. A detailed account of the artificial flooding procedure is given in Muirhead et al. (2004) and Bai and Lung (2005). The hydrodynamic modelling was carried out by using time series flow rates of the artificial flood as the upstream boundary condition and a weir equation was assigned at 3.0 km downstream of the reservoir as the downstream boundary condition. The roughness height was set to 10 mm. The initial faecal bacteria and sediment concentration levels in the water column were set to zero. The faecal bacteria concentration in the river bed was set to1  106 cfu/g and the partition coefficient was set to be 10 L/g as suggested in Bai and Lung (2005). The specific weight of sediment was assumed to be

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 5 1 e1 9 6 0

Sediment Concentration at Site B Model

Data Se d ime n t (m g /L )

300 200 100 0 24

26 Tim e(hrs)

28

E coli Concentration at Site B Model

E c o li ( c fu/100mL)

Data 20000 10000 0 24

26

28

Tim e(hr s) Water Level at Site B

Model

Water L e v e l (m )

Data 1 0.5 0

22

24

26

28

30

Tim e(hr s)

Fig. 9 e Comparison of model results and site data at Site B.

2.65. The critical shear stress for sediment re-suspension and deposition were set to 0.4 and 0.1 N/m2 respectively. The purpose of this model test study was to demonstrate the ability of the model to simulate sediment and bacteria resuspension and deposition under artificial flooding conditions. The comparison of the model predicted results and the measurements are shown in Figs. 8 and 9. In general, the model has reproduced the re-suspension and deposition patterns of the sediment and E. coli levels reasonably well. The model reproduced the re-suspension pattern of sediment and E. coli closely. The time of the rise and peak values were predicted relatively well. The differences between the two datasets can be explained by the experimental measurements including turbidity, whereas in the numerical model sediment concentrations were predicted. This was also noted by Bai and Lung (2005).

8.

Conclusions

The paper reports on the refinement of a numerical model to predict the fate and transport processes of faecal bacteria in surface waters, where sediment transport processes are significant. In this model the concentration of faecal bacteria within the water column was linked to the sediment transport processes, i.e. the faecal bacteria concentrations were affected by adsorption and desorption with sediment particles and the deposition and re-suspension of sediments.

1959

Analytical solutions were derived for steady and uniform flow conditions for sedimentebacteria interaction processes and then the model was tested against these analytical solutions. The model predictions agreed almost perfectly with the analytical solutions and a number of observations were made from the results of these solutions, which showed that: (i) the reduction in the bacteria concentrations due to sediment deposition occurred over a fairly short time compared to the reduction due to decay; and (ii) the decay processes were effectively suppressed during re-suspension of the sediments, due to the rapid increase in the bacterial inputs arising from desorption from the re-suspended sediments. A series of idealised test cases were then set up to study the effects of removing bacteria from the water column and the subsequent re-suspension of bacteria from the bed sediments. The main findings from these idealised model test cases were that: (i) for a higher partition coefficient then lower total bacteria concentrations were predicted in the water column during bacteria settlement, as more bacteria were transported to the bed with sediment deposition; (ii) higher settling velocities reduced the bacteria levels in the water column more rapidly; (iii) higher bed bacteria concentrations meant that more bacteria contributed to the water column load under the same conditions for re-suspension, i.e. the bacteria re-suspended into the water column with the sediments and then repartitioned into the water column; and (iv) the partition coefficient did not significantly affect the total bacteria concentration in the water column during re-suspension, since partitioning only affected the attached and free-living bacteria levels with a higher partition coefficient, giving a higher attached bacteria ratio and a lower free-living bacteria ratio. The model was also applied to a published artificial flooding case study. The model predictions were encouraging, with reasonably good agreement being obtained between the model predictions and the corresponding field measurements.

Acknowledgement The authors would like to acknowledge the support of this project from the Environment Agency and the European Regional Development Fund. The work was carried out while the first author was a PhD student at the Hydro-environmental Research Centre, Cardiff University. The first author would also like to acknowledge the Overseas Research Students Awards and Cardiff University Studentship (Enoch James Fund) for funding his study. This study is also supported by “the Fundamental Research Funds for the Central Universities”.

references

Allen, G., Gunnison, D., Lanza, G.R., 1987. Survival of pathogenic bacteria in various freshwater sediments. Applied and Environmental Microbiology 53 (4), 633e638. Bai, S., Lung, W., 2005. Modeling sediment impact on the transport of fecal bacteria. Water Research 39 (20), 5232e5240. Chapra, S.C., 1997. Surface Water-Quality Modelling. McGraw Hill, New York, pp. 844.

1960

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 5 1 e1 9 6 0

Characklis, G.W., et al., 2005. Microbial partitioning to settleable particles in storm water. Water Research 39 (9), 1773e1782. Falconer, R.A., 1984. A mathematical model study of the flushingcharacteristics of a shallow tidal bay. Proceedings of the Institution of Civil EngineersdPart 2 Research and Theory 77 (3), 311e332. Falconer, R.A., 1993. An introduction to nearly horizontal flows. Reference Book. In: Abbott, M.B., Price, W.A. (Eds.), Coastal Estuarial and Harbour Engineers. E and F.N. Spon Ltd, London, pp. 27e36 (Chapter 2). Falconer, R.A., Chen, Y., 1996. Modeling sediment transport and water quality processes on tidal floodplains. In: Anderson, M. G., Walling, D.E., Bates, P.D. (Eds.), Floodplain Processes. Wiley, Chichester. Fries, J.S., Characklis, G.W., Noble, R.T., 2006. Attachment of faecal indicator bacteria to particles in the Neuse River Estuary, N.C. Journal of Environmental Engineering-ASCE 132 (10), 1338e1345. Gannon, J., Busse, M.K., Schillinger, J., 1983. Faecal coliform disappearance in a river impoundment. Water Research 17, 1595e1601. Howell, J.M., et al., 1996. Effect of sediment particle size and temperature on faecal bacteria mortality rates and the fecal coliform/fecal streptococci ratio. Journal of Environmental Quality 25, 1216e1220. Jamieson, R.C., Gordon, R., Joy, D., Lee, H., 2004. Assessing microbial pollution of rural surface waters. A review of current watershed scale modeling approaches. Agricultural Water Management 70, 1e17. Jamieson, R.C., Joy, D., Lee, H., Kostaschuk, R., Gordon, R., 2005. Resuspension of sediment-associated Escherichia in a natural stream. Journal of Environmental Quality 34 (2), 581e589. Kashefipour, S.M., Lin, B., Harris, E., Falconer, R., 2002. Hydroenvironmental modelling for bathing water compliance of an esturine basin. Water Research 36 (6), 1854e1868. Li, Y., Zhao, M., Cao, Zh, 2001. Two-dimensional Fluvial Flow and Sediment, Transport Model. China Waterpower Press, Beijing, pp. 189(in Chinese).

Lin, B., Falconer, R.A., 1996. Numerical modelling of threedimensional suspended sediment for estuarine and coastal waters. Journal of Hydraulic Research 34 (4), 435e455. Lin, B., Falconer, R.A., 1997. Tidal flow and transport modelling using ULTIMAT QUICKEST scheme. Journal of Hydraulic Engineering, ASCE 123 (4), 303e314. Lin, B., Falconer, R.A., 2001. Numerical modelling of 3-d tidal currents and water quality indicators in the Bristol Channel. Water and Maritime Engineering, Proceedings of Institution of Civil Engineers 148 (3), 155e166. Muirhead, R.W., et al., 2004. Faecal bacteria yields in artificial flood events: quantifying in-stream stores. Water Research 38 (5), 1215e1224. Stapleton, C.M., Wyer, M.D., Kay, D., Bradford, M., Humphrey, N., Wilkinson, J., Lin, B., Yang, Y., Falconer, R.A., Watkins, J., Francis, C.A., Crowther, J., Paul, N.D., Jones, K., McDonald, A.T., 2007. Fate and Transport of Particles in Estuaries, vols. I, II, III, IV Environment Agency Science Report SC000002/SR1-4. Steets, B.M., Holden, P.A., 2003. A mechanistic model of runoffassociated fecal coliform fate and transport through a coastal lagoon. Water Research 37 (3), 589e608. Van Rijn, L.C., 1984a. Sediment transport, part I: bed load transport. Journal of Hydraulic Engineering, ASCE 110 (10), 1431e1457. Van Rijn, L.C., 1984b. Sediment transport, part II: suspended load transport. Journal of Hydraulic Engineering, ASCE 110 (11), 1613e1641. Van Rijn, L.C., 1993. Principles of Sediment Transport in Rivers, Estuaries and Coastal Seas. Aqua Publications, Netherlands. Wu, Y., Falconer, R.A., Lin, B., 2005. Modelling trace metal concentration distributions in estuarine waters. Estuarine, Coastal and Shelf Science 64 (4), 699e709. Yang, L., Lin, B., Falconer, R.A., 2008. Modelling enteric bacteria levels in coastal and estuarine waters. Proceedings of Institution of Civil Engineers, Engineering and Computational Mechanics 161 (EM4), 179e186. Yuan, D., 2007. Development of An Integrated Hydroenvironmental Model and its Application to a Macro-tidal Estuary. PhD Thesis. Cardiff University.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 6 1 e1 9 7 2

Available at www.sciencedirect.com

journal homepage: www.elsevier.com/locate/watres

Removal of arsenic(V) from spent ion exchange brine using a new class of starch-bridged magnetite nanoparticles Byungryul An, Qiqi Liang, Dongye Zhao* Environmental Engineering Program, Department of Civil Engineering, 238 Harbert Engineering Center, Auburn University, Auburn, AL 36849, USA

article info

abstract

Article history:

Ion exchange (IX) is considered by US EPA as one of the best available technologies for

Received 27 September 2010

removing arsenic from drinking water. However, typical IX processes will generate large

Received in revised form

volumes of arsenic-laden regenerant brine that requires costly further handling and

6 January 2011

disposal. This study aimed to develop an engineered strategy to minimize the production

Accepted 9 January 2011

and arsenic leachability of the process waste residual. We prepared and tested a new class

Available online 15 January 2011

of starch-bridged magnetite nanoparticles for removal of arsenate from spent IX brine. A low-cost, “green” starch at 0.049% (w/w) was used as a stabilizer to prevent the nano-

Keywords:

particles from agglomerating and as a bridging agent allowing the nanoparticles to floc-

Arsenic

culate and precipitate while maintaining their high arsenic sorption capacity. When

Hazardous waste

applied to a simulated spent IX brine containing 300 mg/L As and 6% (w/w) NaCl, nearly

Magnetite

100% removal of arsenic was achieved within 1 h using the starch-bridged nanoparticles at

Nanoparticle

an Fe-to-As molar ratio of 7.6, compared to only 20% removal when bare magnetite

Sorption

particles were used. Increasing NaCl in the brine from 0 to 10% (w/w) had little effect on the

Water treatment

arsenic sorption capacity. Maximum uptake was observed within a pH range of 4e6. The Langmuir capacity coefficient was determined to be 248 mg/g at pH 5.0. The final treatment sludge was able to pass the TCLP (Toxicity Characteristic Leaching Procedure) based leachability of 5 mg/L as As. ª 2011 Elsevier Ltd. All rights reserved.

1.

Introduction

Arsenic contamination of drinking water has been a worldwide challenge (An et al., 2005). Arsenic has been associated with skin, lung, bladder and kidney cancers (NRC, 2001). It was reported that 45e57 million people in Bangladesh and 13 million in the United States have been exposed to unsafe levels of arsenic (WHO, 2006). To mitigate human exposure, the US EPA implemented a revised maximum contaminant level (MCL) of 10 mg/L in 2006, representing a 5-fold decrease from its previous drinking water standard. Technologies for effective arsenic removal from drinking water have been continuously sought for decades, and the

effort has been intensified in the U.S. since the new MCL took effect. Approaches such as enhanced coagulation (Hering et al., 1997), adsorption onto iron oxide (Dixit and Hering, 2003), membrane processes (Ballinas et al., 2004), and ion exchange (An et al., 2005; Clifford et al., 2003; Cumbal and SenGupta, 2005; Ramana and SenGupta, 1992) have been extensively studied. Ion exchange is considered by US EPA as one of the best available technologies for removing arsenic from drinking water (EPA, 2000). However, IX processes produce a large volume of arsenic-laden regenerant brine, which is often categorized as hazardous process waste residual and requires costly further treatment, handling and disposal. Therefore, there is a dire need for developing an engineered strategy to

* Corresponding author. Tel.: þ1 334 844 6277; fax: þ1 334 844 6290. E-mail address: [email protected] (D. Zhao). 0043-1354/$ e see front matter ª 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.watres.2011.01.004

1962

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 6 1 e1 9 7 2

minimize the production and arsenic leachability of the process waste residual resulting from IX processes. Macroscale iron (hydr)oxides have been found effective for removing arsenic from drinking water and have been extensively investigated. Ferguson and Anderson (1974) studied adsorption characteristics of As(III) and As(V) with iron hydroxide. Pierce and Moore (1982) tested amorphous iron hydroxide and found adsorption of arsenic was highly pH-dependent. Hering et al. (1997) utilized bulk iron oxide for arsenic removal and concluded that removal of both As(III) and As(V) was affected by pH and the sorption was influenced by sulfate and dissolved organic matter. Dixit and Hering (2003) studied adsorption of As(III) and As(V) on hydrous ferric oxide (HFO), goethite, and magnetite, all of which were synthesized in the laboratory, and they found that sorption of As(V) onto HFO and goethite was more favorable than that of As(III) below pH 56, whereas, above pH 78, As(III) had a higher affinity for all of the sorbents. Other types of ironbased materials were also examined for removal of arsenic, including ferrihydrite (Coker et al., 2006; Waychunas et al., 1993), hematite (Aria et al., 2004; Gimenez et al., 2007), goethite (Dixit and Hering, 2003; Fendorf et al., 1997; Gimenez et al., 2007), and zero valent iron (ZVI) (Kanel et al., 2007; Su and Puls, 2001; Yuan and Lien, 2006). In recent years, nanoscale iron oxides have attracted growing interest in water treatment and environmental remediation (Huber, 2005; Kanel et al., 2006; Yavuz et al., 2006). Iron oxide nanoparticles offer high adsorption capacities toward a variety of important water contaminants including both As(III) and As(V) (Dixit and Hering, 2003). Magnetite (Fe3O4) is a ubiquitous magnetic iron oxide existing in the lithosphere, pedosphere, and biosphere (Cornell and Schwertmann, 2003). The structure of magnetite is a cubic inverse spinal that can be expressed as FeFe2O4, and it is a face-centered cubic (fcc) structure with Fe ions coordinated to interstitial tetrahedral sites (1/3) occupied by Fe(III) and octahedral sites (2/3) by equal amounts of Fe(III) and Fe(II) (Cornell and Schwertmann, 2003). Based on the evidence from extended X-ray absorption fine structure (EXAFS) spectroscopy and Fourier transform infrared (FTIR) spectroscopy (Fendorf et al., 1997; Sun and Doner, 1996; Waychunas et al., 1993), magnetite can adsorb arsenic through inner-sphere bidentate-binuclear complexation. Various methods have been used to prepare magnetite nanoparticles, including solegel (Tang et al., 2004), g-ray radiation (Wang et al., 1997), hydrothermal technique (Qian et al., 1994), forced hydrolysis (Compean-Jasso et al., 2008), and co-precipitation from a mixture of Fe(III) and Fe(II) salts (Ataie and Heshmati-Manesh, 2001; Ferguson and Anderson, 1974; Jolivet, 2000). The co-precipitation method was first proposed by Anderson (1957) using ferric chloride and ferrous sulfate in a concentrated ammonia solution. In this method, pH of the solution is one of the key factors for controlling the physicochemical properties, such as particle size distribution, of the resultant magnetite particles. The mean particle size of magnetite synthesized by co-precipitation has been found to decrease with increasing final pH. The co-precipitation method has been the most widely employed approach for preparing magnetite particles for its straightforward procedure and the convenience for controlling size and shape.

Typically, this method starts with a solution containing both Fe3þ and Fe2þ at a molar ratio of Fe3þ:Fe2þ ¼ 2:1. Upon addition of a base (e.g. NaOH), the Fe3þ ions are converted to FeOOH through the intermediate Fe(OH)3; and Fe2þ ions to Fe (OH)2 (Eqs. (1)e(2)). One mole of magnetite is then formed by reacting 2 mol of FeOOH and 1 mol of Fe(OH)2 as shown below: Fe3þ þ 3OH /FeðOHÞ3 /FeOOH þ H2 O

(1)

Fe2þ þ 2OH /FeðOHÞ2

(2)

2FeOOH þ FeðOHÞ2 /Fe3 O4 þ 2H2 O

(3)

In the presence of dissolved oxygen (DO), magnetite will be oxidized to g-Fe2O3 (maghemite) (Alibeigi and Vaezi, 2008), which constitutes one method for preparing maghemite. High temperatures accelerate the oxidation rate and quantity, and smaller particles of magnetite will be oxidized faster than larger ones due to their shorter diffusion length (Tang et al., 2003). Because of the high surface energy, magnetite nanoparticles tend to agglomerate into larger aggregates or flocs, which can render loss of specific surface area. Consequently, a stabilizer is often required, which can be sorbed on the nanoparticles and prevent the particles from aggregating thorough the steric and/or electrostatic stabilization mechanisms (He and Zhao, 2005, 2007; Lee et al., 2002; Nishio et al., 2004; Raveendran et al., 2003; Tamaura et al., 1986; Yokoi and Kantoh, 1993). For examples, Raveendran et al. (2003) used a water-soluble starch as a stabilizer for preparing nanoscale Ag particles in aqueous media. He and Zhao (2007) and He et al. (2007) successfully stabilized ZVI nanoparticles with starch and carboxymethyl cellulose (CMC). The use of stabilizers maintained the high specific surface area and high reactivity of the nanoparticles, and enables the nanoparticles to be used for treating contaminated subsurface. For example, CMC-stabilized ZVI nanoparticels can be delivered into subsurface soils for in-situ remediation of aquifers contaminated with chlorinated solvents (He et al., 2010). However, when used for water treatment, the particle stabilization can create a problem, i.e. it can be challenging and costly to separate the highly stable nanoparticles from the product water. Therefore, separation of the spent nanoparticles must be taken into account in preparing the nanoparticles while taking advantage of the novel features (e.g. high specific surface area and high sorption capacity) of the nanoparticles. For the desired brine treatment, it was desired that the nanoparticles can offer both high sorption capacity and easy separation, and we hypothesized that the desired features can be achieved by use of starch as a bridging agent. Fig. 1 presents a conceptualized representation of such bridged nanoparticles. The starch bridging prevents the nanoparticles from aggregating, and thus, maintains their high sorption capacity. On the other hand, the polymeric molecules serve as a flocculating agent, enabling the particles to flocculate and precipitate by gravity. The primary goal for this study was to develop and test such polymer-bridged magnetite nanoparticles for removal of arsenic from spent IX brine. The specific objectives were to: 1) prepare and characterize the starch-bridged magnetite particles, 2) determine arsenic sorption and desorption performances of the bridged nanoparticles, 3) test the pH effect,

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 6 1 e1 9 7 2

1963

the suspension was sealed and placed in dark for 48 h to allow the magnetite nanoparticles to fully grow. Then the pH was lowered back to 7.0 with HCl, and then the suspension was used within 1 h. For comparison, bare magnetite particles were also prepared in the absence of starch but under otherwise identical conditions.

2.3.

a

b

c

Fig. 1 e A conceptualized illustration of stabilizer effect on particle interactions in aqueous solution: (a) Particles aggregate without a stabilizer, (b) particles are bridged and form flocculates in the presence of proper concentrations of a polymeric stabilizer, and (c) particles remain stable in the presence of elevated concentrations of a polymeric stabilizer.

4) elucidate arsenic binding mechanisms in terms of particle surface chemistry and binding modes, and 5) test the leachability of magnetite-bound arsenate in the spent particles (waste sludge).

2.

Materials and methods

2.1.

Chemicals

The following chemicals (ACS grade or higher) were purchased from Fisher Scientific (Pittsburgh, PA): Na2SO4, NaHCO3, FeCl3, NaOH, and HCl. A hydrolyzed potato starch and Na2HAsO4$7H2O were obtained from SigmaeAldrich (Milwaukee, WI), and FeSO4$7H2O was from Acros organic (Morris Plains, NJ). All solutions were prepared with ultrapure deionized (DI) (18.2 MU cm1) water.

2.2.

Synthesis of starch-bridged magnetite nanoparticles

The desired magnetite nanoparticles were prepared by modifying the co-precipitation approach in the presence of a suitable concentration of the water-soluble starch. First, a solution of Fe3þ and Fe2þ (from FeCl3$6H2O and FeSO4$7H2O, respectively) was prepared at a ferric-to-ferrous molar ratio of 2:1 at room temperature (21  C). To avoid oxidation during the preparation process, DI water was first deoxygenated by nitrogen purging. A 0.8 wt.% starch solution was prepared by mixing 8 g of the potato starch with 1 L of DI water and heating the mixture to 100  C. Once the starch solution started boiling, heating was removed and the solution was allowed for cooling at room temperature. The cooled starch stock solution was then mixed with the Fe3þ/Fe2þ solution under N2 purging at a starch concentration ranging from 0 to 0.13 wt.%. Then, a 4 M NaOH solution was added dropwise into the solution under vigorous stirring till a pH of around 11. Note that the change in color from brown to black indicates the formation of the nanoparticles (Schwertmann and Cornell, 2000). Then, the starchmagnetite solution was nitrogen-purged for 20 min. Finally,

Physical characterizations

Transmission electron micrograph (TEM) images were taken with a Zeiss EM10 transmission electron microscope (Zeiss, Thornwood, NJ) operated at 40 Kv. The images were then processed using the ImageJ software (National Institute of Mental Health, Bethesda, MD), which gave the mean size and size distribution of the particles. First, an aliquot of a nanoparticle suspension was diluted 10 times to facilitate the identification of particles and avoid “jamming” of the particles. The diluted suspension has a magnetite concentration of 0.057 g/L as Fe. Then, a single drop of the suspension was deposited on a 300 mesh copper specimen grid and allowed to air dry for at least three days before imaging. Zeta potentials (z) of particles in the particle suspension were measured with a Zetasizer nano ZS (Malvern Instruments, UK) at 25  C. A folded-capillary cell filled with 0.75 mL of the nanoparticle suspension was used for the measurements. For suspensions of starch-bridged particles, the viscosity was measured for each suspension and then used for acquiring the corrected z values.

2.4.

Kinetic tests

Batch kinetic tests were carried out to measure arsenate sorption rate for starch-bridged and bare magnetite particles. To determine the effect of starch concentration on arsenate sorption rate, suspensions of magnetite particles were prepared at 0.57 g/L as Fe in a series of vials with various concentrations (0e0.081 wt.%) of starch. For each case, 600 mL of the particle suspension was prepared. Under this range of starch concentration, the magnetite particles can be all precipitated by gravity. Then, 500 mL of the supernatant was removed from each vial. Arsenate sorption was then initiated by mixing each of the concentrated particle precipitate with 100 mL of a stock solution of simulated IX brine containing arsenate, sulfate, bicarbonate, and NaCl to yield a 200 mL mixture containing 300 mg/L As(V), 600 mg/L SO2 4 , 305 mg/L HCO 3 , 6% (w/w) NaCl, and 1.7 g/L as Fe of magnetite particles (i.e. Fe-to-As molar ratio ¼ 7.6). The reactors were then placed on a shaker and mixed continuously. At predetermined times, selected bottles were taken off the shaker and allowed to settle for 10 min, which was sufficient to remove >98% of the particles. Then, 2 mL of each supernatant was sampled with a pipette. The samples were then further centrifuged for 15 min at 6500 rpm (5857g of RCF) to remove any remaining particles, and the supernatant was analyzed for arsenic.

2.5.

Equilibrium sorption tests

A series of batch equilibrium sorption tests were performed with the starch-bridged magnetite nanoparticles. First, the particles were prepared at 0.57 g/L as Fe and with 0.049 wt.% of starch. This recipe offered the most effective removal of As

1964

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 6 1 e1 9 7 2

based on kinetic test results. The particles were concentrated by decanting proper volumes of the supernatant, and the concentrated particles were then mixed with a stock solution of the simulated IX brine, which yielded the following initial concentrations: solution volume ¼ 230 mL, SO2 4 ¼ 600 mg/L, HCO 3 ¼ 305 mg/L, NaCl ¼ 6 wt.% (w/w), As(V) ¼ 38e617 mg/L, and magnetite ¼ 1.0 g/L as Fe. The systems were equilibrated under shaking at 200 rpm on a platform shaker (New Brunswick Scientific, NJ) for 2 days. The solution pH was kept at either 5.0 or 6.9 through intermittent adjusting using 0.1 M NaOH or 0.1 M HCl. The following mass balance equation was employed to determine the arsenate uptake: qe ¼

VðCo  Ce Þ M

(4)

where qe (mg/g) is the equilibrium mass uptake of arsenic, V (L) is the solution volume, C0 and Ce (mg/L) are the initial and final concentrations of arsenic, respectively, and M (g) is the mass of sorbent added calculated as Fe.

2.6.

FTIR analysis

FTIR was employed to determine arsenate sorption mechanisms and interactions between starch and the magnetite nanoparticles. Three representative samples, namely bare magnetite particles and starch-bridged magnetite nanoparticles before and after the arsenic sorption, were subjected to FTIR and compared. The samples were first vacuum-dried, and then ground in a mortar to fine powders, which were then mixed with KBr at a sample-to-KBr ratio of 5:95 by weight. The mixtures were pressed at 9 metric tons for 2 min. The specimens were then scanned and characterized using an IR Prestige-21 spectrometer (Shimadzu) over the wave number range from 400 to 4000 cm1.

2.7.

Leachability tests

In the treatment process, arsenic sorbed on the particles is co-precipitated with the nanoparticles. As a result, an As-laden process solid waste will be produced. To avert from being categorized as a hazardous waste, the waste must pass a certain leachability test such as TCLP of US EPA and the California waste extraction test (WET). To assess the arsenic leachability in waste sludge, TCLP was carried out according to EPA Method 1311 on air-dried waste sludge samples. The #1 extraction fluid for the TCLP was prepared by adding 5.7 mL glacial CH3COOH and 64.3 mL of 1 N NaOH to 500 mL DI water, and then adding DI water to 1 L. Solid samples were mixed with the TCLP fluid at a solid-to-liquid ratio of 1:20. The mixtures were then rotated at 30 rpm for 18 h at room temperature and then centrifuged for 30 min at 6500 rmp (5857g of RCF). The supernatant was then passed through a 0.45 mm filter and analyzed for arsenic. The leachability tests were performed in duplicate.

2.8.

Chemical analysis

Solution or suspension pH was measured using an Oakton pH meter (pH 510 Benchtop Meter, Oakton). Arsenic was analyzed using a Perkin Elmer Atomic Adsorption Spectrophotometer

(ASTM D2972-B), which has a detection limit of 3 mg/L as As. Data quality was assured by checking As mass balances and/ or duplicating the experiments. Dissolved iron was measured using a Flame Atomic Absorption Spectrometer (Varian Spectra 220 FS) (detection limit ¼ 0.05 mg/L). Chloride, sulfate and bicarbonate were analyzed using a Dionex Ion Chromatograph (Model DX-20) equipped with an AS 14 column.

3.

Results and discussion

3.1. Physical characteristics of starch-bridged magnetite nanoparticles The physical properties of the starch-bridged magnetite nanoparticles, including their morphology, mean size, and size distribution, were measured by TEM. Fig. 2 shows the TEM images of a typical starch-bridged magnetite sample, prepared at 0.57 g/L as Fe with 0.049 wt.% of starch. Based on a total particle count of 176 from this image, a mean particle diameter of 26.6 nm for the sample was obtained, with a standard deviation (SD) of 4.8 nm. The TEM image reveals that the starch-magnetite nanoparticles are nearly spherical in shape and have a relatively narrow size distribution. The histogram shows that 74% of the particles were within the range 21e30 nm. No particles smaller than 10 nm or bigger than 50 nm were observed. The TEM image also suggests that the nanoparticles are inter-bridged yet remain identifiable individual nanoparticles, i.e. the particles did not aggregate into larger solid particles because of the coating of starch on the nanoparticles. Fig. 3 indicated that starch can strongly affect the particle settleability. At relatively low concentrations of stabilizer (e.g. 0.081 wt.% or less starch for 0.57 g/L as Fe of magnetite particles), the nanoparticle surface has sufficient open sites to allow a starch molecule to bind with multiple nanoparticles. As a result, the nanoparticles were bridged and large flocs were formed that precipitated easily. For example, based on the concentration of iron in the supernatant, 99.5% and 98.2% of the particles bridged with 0.049 wt.% and 0.081 wt.% starch, respectively, were removed by gravity after 1 h of settling (Fig. 3). At elevated starch concentrations, the surface of each particle is covered with a denser layer of the stabilizer molecules, which induce a strong steric repulsion force preventing the particles from aggregating or inter-bridging (Fig. 1), leading to complete stabilization of the particles. It is noteworthy that while fully stabilized particles remained suspended for days, the bridged nanoparticles displayed even better settleability than the bare particles. For instance, the 1-h particle removal in Fig. 3 for the bare particles (without salt addition) was 95.2%. The starch stabilizer is a neutral polymer and it stabilizes nanoparticles through steric repulsion arising from the osmotic force when the starch layers overlap as the particles collide. Results from Fig. 3 revealed that the presence of 6% NaCl did not affect the settleability of the starch-bridged or starch-stabilized nanoparticles, suggesting strong steric repulsion effect between starched particles. In contrast, the salt addition increased the particle aggregation for the bare magnetite particles, which is in accord with the classical

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 6 1 e1 9 7 2

1965

Fig. 2 e TEM images and size distribution of starch-bridged magnetite nanoparticles prepared at 0.057 g/L as Fe in the presence of 0.049 wt.% starch.

double layer compression theory. More detailed discussion on salt effect is provided in Section 3.4.

3.2.

Kinetic tests

Fig. 4 shows arsenate adsorption rates and removal efficiency at a fixed dosage (1.7 g/L as Fe) of magnetite but various levels of starch ranging from 0 to 0.081 wt.%. In most cases, equilibrium was reached within 4 h with most sorption capacity filled within the first 1 h. This rather fast sorption kinetics supports the assertion that the nanoparticles were present in the form of starch-bridged individual non-porous particles (Figs. 1 and 2) rather than as aggregated micro-porous sorbents (Cornell and Schwertmann, 2003). As the starch concentration increased from 0 to 0.012, 0.024, 0.041, and 0.049 wt.%, the equilibrium arsenic removal increased from w20% for bare magnetite to 42, 82, 97, and w99%, respectively. This observation clearly revealed that the presence of starch during particle nucleation and growth prevented particle aggregation and resulted in much greater specific surface area. In the range of 0.012e0.049 wt.%, a higher starch concentration resulted in more loosely bridged particles that are more accessible to arsenic. If the starch concentration is too low, particles are more tightly connected or even aggregated, resulting in greater mass transfer resistance and loss in surface area. However, further increasing starch concentration beyond the critical value of 0.049 wt.% diminished both sorption capacity and kinetics. At elevated starch concentrations, more stabilizer molecules are coated on individual nanoparticles, resulting in steric repulsion between the particles, and thus, the nanoparticles became fully stabilized (Figs. 1 and 3). Although these well-stabilized nanoparticles may offer comparably great specific area, the dense starch layer on the particle surface hinders arsenic uptake both

thermodynamically (due to reduced sorption sites and site accessibility) and kinetically (due to increased mass transfer resistance). However, when the coated starch layer is degraded, the encapsulated sites may become accessible. Consequently, there exists an optimum starch concentration for a given magnetite concentration to facilitate optimal arsenic removal and yet allow for an easy separation of spent nanoparticles. In this case, 0.049 wt.% appeared to be optimal for preparing 0.57 g/L as Fe of the bridged magnetite nanoparticles. In this case, As concentration remaining in the brine after one batch treatment was <2.8 mg/L (i.e. 1% of its initial concentration). The residual As concentration can be further reduced to nearly zero by increasing the particle dosage if needed. The treated brine can then be reused for regeneration of saturated IX resins (An et al., 2005), which not only cuts down the regenerant cost, but also greatly minimizes the discharge of the waste brine as a liquid process waste residual. From an application standpoint, the brine treatment unit can be incorporated into a typical ion exchange process. Conceptually, it is most advantageous for the particles to be employed to remove arsenate from small volumes of, and concentrated, waste residuals such as IX brine, although it may be conceivable to use the particles for direct treatment of contaminated bulk water (e.g. in enhanced coagulation).

3.3.

Isotherm tests and FTIR analysis

Fig. 5 shows arsenate sorption isotherms for the starchbridged magnetite nanoparticles (initially prepared at 0.57 g/L as Fe and with 0.049 wt.% starch) at an equilibrium pH of 5.0  0.1 and 6.9  0.1, respectively. Again, the simulated IX brine was used in the isotherm tests. The classical Langmuir isotherm model, Eq. (5), was employed to interpret the

1966

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 6 1 e1 9 7 2

Fig. 3 e Effect of starch and NaCl on separation of magnetite particles by gravity: (A) immediately after particle preparation, (B) after 1 h of settling. Numbers under vials refer to starch concentration. For each pair of vials, the left one had no NaCl added, while the right one contained 6% NaCl.

experimental data. The best data fitting gave both maximum adsorption capacity (Q) and the Langmuir affinity constant (b). qe ¼

bQCe 1 þ bCe

(5)

The resulting Q is 248 mg/g at pH 5.0  0.1, and 198 mg/g at pH 6.9  0.1. To our knowledge, these values are much greater than those reported for other iron oxides including magnetite particles reported to date. Earlier, Yean et al. (2005) reported a Q value of 152 mmol/g (11.4 mg/g as As) at pH 4.8 and 101 mmol/g (7.6 mg/g as As(V)) at pH 6.1 for commercial aggregated magnetite nanoparticles of a mean size of 20 nm, and obtained a much higher Q value of 172.5 mg/g as As(V) using magnetite of 11 nm mean size synthesized using oleic acid as a stabilizer. A magnetic separator was needed to separate the fully stabilized particles. FTIR spectroscopy was used to identify the nature of the chemical bonding between magnetite, starch, and arsenate.

Fig. 6 shows the FTIR spectra for bare magnetite particles and starch-bridged samples before and after sorption of arsenate. The absorption band at 571 cm1 was obtained for all three types of magnetite samples tested, and it can be assigned to the FeeO stretching vibration for the Fe3O4 particles (Cornell and Schwertmann, 2003). The presence of starch coating reduced the absorption of the radiation by the core magnetite, and thus, weakened the peak intensity for the two starch-coated samples. As starch mainly consists of amylose and amylopectin, three new peaks appeared at 1020, 1151, and 1078 cm1, respectively, for starched magnetite particles. The peaks at 1151 and 1078 cm1 represent the CeO stretching vibrations in the CeOeH groups, whereas the 1151 cm1 peak corresponds to the CeO stretching vibration in the CeOeC groups, which is in agreement with previous amylose studies (Huang et al., 2006). A peak at 1628 cm1 and another at 3400 cm1 were observed for all three samples though the intensities varied. The weakest peaks were observed for the bare magnetite,

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 6 1 e1 9 7 2

Fig. 4 e Arsenic removal as a function of time using magnetite particles at various starch levels in a simulated IX brine. Experimental conditions: Fe/As molar ratio [ 7.6; starch [ 0e0.081 wt.%; Magnetite [ 1.7 g/L as Fe (initially prepared at 0.57 g/L); Solution pH was kept at 6.5.

corresponding to the OeH bond from H2O (Maity and Agrawal, 2007). The peaks were greatly intensified for the starch-bearing magnetite due to the H-bonded OH groups of amylose and amylopectin of starch, which agrees with the results by Maity and Agrawal (2007) (Huang et al., 2006). Comparing the spectra of the starch-bridged magnetite before and after arsenate sorption, a new broad band was evident between 750 and 850 cm1 for the arsenate-laden magnetite. According to Goldberg and Johnston (2001) who conducted FTIR study of As(V) sorption to amorphous iron oxide, the band could be assigned to the Fe-O-As groups. This is consistent with the results by Pena et al. (2006) who observed an FTIR band at 808 cm1 at pH 5 for As(V) sorbed on TiO2 and ascribed the band to the TieOeAs groups. Pena et al. (2006) also compared FTIR bands for sorbed and dissolved

300 250 pH: 5.0 b= 0.5 L/mg Q= 248.0 mg/g

qe, mg/g

200 pH: 6.9 b= 0.2 L/mg Q= 198.2 mg/g

150 100

1967

Fig. 6 e FTIR spectra of arsenate on bare, fresh starchbridged, and arsenate-loaded starch-bridged magnetite particles. Magnetite particles were prepared at 0.57 g/L as Fe with 0.049 wt.% of starch. Arsenate was loaded at pH 5. Arsenic loaded on nanoparticles [ 250 mg/g as As.

arsenate species, noted a marked shift in the band positions of arsenate upon sorption. For dissolved H2AsO 4 , two peaks were observed at 878 and 909 cm1 corresponding to the symmetric and asymmetric stretching vibrations of AseO bonds. Upon adsorption, the peaks were shifted to 808 and 830 cm1, respectively. According to Goldberg and Johnston (2001) and Pena et al. (2006), the shift on band positions was attributed to symmetry reduction resulting from inner-sphere complex formation (i.e. formation FeeOeAs complexes). Goldberg and Johnston (2001) also reported that for arsenate sorbed on amorphous iron oxide, there existed two distinct bands corresponding to surface-complexed and non-surface-complexed AseO groups, respectively. In our case, however, only one single band was observed and the wave number was shifted to w800 cm1, indicating that surface complexation was the predominant mechanism for arsenate sorption to the starch-bridged nanoparticles. In addition, Zhang et al. (2005) reported that the rise in MeAseO (M: metal) peak was coupled with weakening of M-OH peak when As(V) was adsorbed to FeeCe oxide, and the researchers ascribed this phenomenon to ion exchange reaction between eOH and arsenate anions. However, no MeOH bond was evident in this work, suggesting that the sorption of As(V) to the magnetite nanoparticles did not necessarily conform to the standard ion exchange stoichiometry.

50 0

3.4. 0

100

200

300

400

Ce, mg/L Fig. 5 e Arsenate sorption isotherms for starch-bridged magnetite particles at pH 5.0 and pH 6.9 (final pH) from a spent brine solution (symbols: observed data; lines: Langmuir model fits). Brine compositions: initial As L (V) [ 38e617 mg/L, SO24 [ 600 mg/L, HCO3 [ 305 mg/L, NaCl [ 6 wt.% (w/w).

Effect of salt concentration on arsenic removal

Fig. 7 shows equilibrium arsenic uptake in the presence of various concentrations of NaCl ranging from 0% to 10% but under otherwise identical conditions. Evidently, the presence of brine at concentrations as high as 10% did not show any appreciable decrease in arsenic uptake. This observation violates the principle of selectivity reversal commonly cited in standard ion exchange regeneration (Clifford, 1999). However, this supports that FTIR results that specific interactions such

1968

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 6 1 e1 9 7 2

300

3.5. Effect of pH and chemical stability of starch-bridged nanoparticles

Initial: 300 mg/L as As Final pH: 4.0 1.0 g/L of Fe

250

qe, mg/g

200 150 100 50 0

0

2

4

6

Conc. of NaCl, %

8

10

Fig. 7 e Arsenate uptake by starch-bridged magnetite nanoparticles in the presence of various concentrations of NaCl ranging from 0 to 10%. The magnetite particles were initially prepared at 0.57 g/L as Fe with 0.049 wt.% of starch.

as strong inner-sphere complexation other than ion pairing was involved between arsenic and the nanoparticles. To quantify the relative affinity, the arsenate and chloride binary separation factor aAs/Cl is calculated based on the equilibrium sorption data (Clifford, 1999): aAs=Cl ¼

qAs CCl CAs qCl

Solution pH can affect both arsenate speciation (pKa values for H3AsO4 are 2.2, 7.0 and 11.5) and surface charge of the starch-bridged nanoparticles. In addition, the nanoparticles may be partially dissolved under low pH. Therefore, it is expected that pH will affect the arsenate sorption capacity. Fig. 8 shows equilibrium arsenate uptake as a function of final solution pH for bare and starch-bridged magnetite, under otherwise identical conditions. In terms of sorption capacity, the starch-bridged nanoparticles displayed an overall much greater arsenic uptake capacity than the bare particles throughout the pH range tested. In addition, the pH sensitivity also differed markedly for the two classes of sorbents. While lower pH favored arsenate sorption for both cases, the bare particles were much more sensitive to pH. For instance, at pH <7.0, arsenic uptake for the bare particles increased sharply and nearly linearly with decreasing pH. In contrast, a maximum sorption plateau was evident in the pH range of 4e6 for the starch-bridged magnetite, indicating that the starch coating not only resulted in greater sorption capacity, but also altered the surface potential of the particles. When the 2-day and 50-day sorption data are compared, greater capacity was observed at pH below w5 for starch-bridged

(6)

where qAs and qCl are the concentrations of arsenate and chloride, respectively, in the solid phase; whereas CAs and CCl are the concentration of arsenate and sulfate in the solution, respectively. As a rule, a separation factor value of greater than one indicates greater selectivity for arsenate than chloride. The calculated aAs/Cl values were 102, 139, 147, and 290 at 2%, 4%, 6%, and 10% of NaCl, respectively, indicating the nanoparticles’ affinity for arsenate far exceeds that for chloride. The observed minimal salt effect on arsenic removal contrasts with the results reported by Shipley et al. (2009), who found that increasing ionic strength to 0.1 M with KNO3 decreased arsenate adsorption by 10% with a commercial bare magnetite of 20 nm size. The authors explained the phenomenon with the classical double layer compression theory. In the presence of high concentrations of cation (Kþ), the electrostatic double layer thickness on the particle surface is reduced, increasing the likelihood of nanoparticle aggregation and thus decreasing the available surface area during arsenate adsorption. Aggregation of bare magnetite particles was observed immediately after the particles were prepared, and the bare magnetite precipitated faster with 6% NaCl than without the salt addition (Fig. 3B). However, for starch-bridged magnetite particles, the particles were bridged yet held apart due to the interparticle steric repulsion arising from the osmotic force of the starch on the particle surface. In the presence of salt, the starched particle surface potential was almost zero throughout the pH range (see Section 3.5) and increasing ionic strength did not affect the surface charge as the ionic strength exerts little impact on the steric repulsion. Consequently, the high sorption capacity was maintained even in the presence of the extremely high concentrations of NaCl.

Fig. 8 e Arsenate uptake and dissolved Fe concentration as a function of pH: (A) bare magnetite particles after 2, 7 and 50 days, and (B) starch-bridged magnetite (prepared at 0.57 g/L and with 0.049 wt.% starch) after 2 and 50 days. Initial As concentration [ 300 mg/L, Magnetite [ 1.0 g/L as Fe.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 6 1 e1 9 7 2

particles, suggesting that the starch-coating somewhat hindered the sorption kinetics. Ma and Bruckard (Ma and Bruckard, 2010) reported that adsorption of a cornstarch to kaolinite can be strongly affected by solution pH with lower pH favoring the starch uptake. The same adsorption trend for potato and cornstarch was obtained by Husband (Husband, 1998). A microscopic study by Hirashima et al. (2005) revealed that more glucose chains were leached out from starch macromolecules as pH was decreased from 6.0 to 3.6, resulting in elevated solution viscosity. Based on these findings, the observed slower arsenic removal kinetics at pH <5 can be attributed to: 1) increased sorption of starch at lower pH, 2) elevated viscosity, and 3) sorption of the leached glucose chains and the break-down smaller segments of starch, all of which are in favor of a greater mass transfer resistance, and thus, slower sorption rate. However, it should be noted that the diminished arsenic uptake at pH below 5 can be in part attributed to dissolution of the nanoparticles. Fig. 8A and B indicate that while both bare and starch-bridged particles became partially dissolved at pH <4, the starch-bridged particles were more prone to acid dissolution than the bare particles. For example, 13% of starch-bridged nanoparticles at pH 3.2 was dissolved compared to 3.2% for the bare particles at pH 2.8. At pH 5.0, the particles are chemically robust. Fig. 9 shows the measured zeta potential for the particles as a function of solution pH and under various conditions. Fig. 9A plots z for bare magnetite particles and those prepared with, respectively, 0.025, 0.049, and 0.081 wt.% of starch over a wide range of pH. Based on the results, the pH at the point of zero charge (pHPZC) was estimated to be w6.1 for the bare particles whereas pHPZC for the starched particles varied from 3.7 to 4.4. For the bare magnetite particles, Prakash et al. (1999), Anastassakis (1999), Yean et al. (2005), and Illes and Tombacz (2003) reported a pHPZC of, respectively, 6.8, 7.0, 6.8, and 7.9 for various bare colloidal magnetite particles measured using other methods. For the starch-stabilized magnetite, we also acquired a pHPZC value of 6.1 using the classical acid-base titration method (Huang and Stumm, 1973). Causes for the discrepancies have been discussed by others (e.g. Kosmulski et al., 2002) and are beyond the scope of this work. Nonetheless, it is noteworthy that the presence of starch greatly shielded the sensitivity of z to the potential-determining ions (Hþ/OH). For example, between pH 5 and 8, the z value varied from þ20 to 40 mV for the bare particles, but only 2 to 12 mV with 0.081 and 0.049% starch. When the starched particles are compared, Fig. 9A reveals that the presence of higher starch concentrations (0.081 and 0.049%) resulted in a much less negative surface potential than the case of 0.025% of starch. Evidently, the sorbed starch, a neutral polymer with dense H-bonding (Biggs, 2006; Ravishankara et al., 1995), on the surface of magnetite serves as a strong surface buffer that diminishes the effect of Hþ/OH on the surface charge. Fig. 9B shows the effects of 6% NaCl and sorbed arsenate anions on the surface potential of starched magnetite particles. It is evident by comparing curves (a) and (c) that the presence of 6% NaCl nearly zeroed the negative z of the starched magnetite particles over the wide pH range. This is consistent with the classical double-layer compression theory and is attributed to the associated suppression effect of the

1969

Fig. 9 e (A) Zeta potential (z) as a function of pH for magnetite nanoparticles synthesized with different concentrations of starch at a fixed magnetite concentration of 1.0 g/L as Fe; (B) Effects of salt and sorbed arsenate on z of starched bridged magnetite nanoparticels (Magnetite prepared at 0.57 g/L as Fe with 0.049 wt% starch, As in solid phase [ 242 mg/g).

surface potential due to elevated surface density of Naþ ions. An examination of curves (a) vs. (b) and (c) vs. (d) indicates that sorption of arsenate anions induced a marked negative potential especially in the pH range of 5e8. This observation again suggests that sorption of arsenate ions, which are present as monodentate (pH < 7) or bidentate (pH > 7) anionic ligands, is not a simple ion pairing or ion exchange process, but rather, involves more specific interactions such as Lewis acidebase interaction in accord with the inner-sphere complexation between arsenate’s donor atoms (oxygen) and the acceptors (Fe) of the magnetite. Note that at pH > 8, although arsenate is present as bidentate ligands, the competition of hydroxyl anions becomes fiercer. As a result, arsenate uptake was actually reduced and the surface negative potential was suppressed as well (curve b).

3.6.

Leachability of arsenic from magnetite waste

TCLP has been widely employed in the U.S. to determine if a process waste may be disposed of in landfills (Phenrat et al.,

1970

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 6 1 e1 9 7 2

40

Conc.of As, mg/L

35

TCLP



30 25 20



15 10 5 0

 0

5

10

15

Fe:As Molar ratio Fig. 10 e Leachability of arsenate from spent starchbridged magnetite precipitates when subjected to TCLP. The Fe:As molar ratio indicates the amount of magnetite applied in relation to the As concentration during the As removal treatment. Data plotted as mean of duplicates, errors indicate deviation from the mean. Initial pH of TCLP Fluid [ 4.95; Final pH in all cases [ 4.92e4.97.

2008). To probe the arsenic leachability, the arsenic-laden magnetite precipitate was air-dried and subjected to the TCLP procedure. Fig. 10 shows that the TCLP leachability varied with iron dosage during the As removal treatment. Increasing the Fe/As molar ratio from 5 to 15 resulted in a 90% decrease in the leached arsenic concentration, and at an Fe/As molar ratio of 15, the precipitate satisfied the current TCLP limit of 5 mg/L As. Consequently, the spent particles may be disposed of as non-hazardous waste material in municipal landfills, although the long-term arsenic stability under real landfill conditions needs to be further investigated due to the intrinsic limitation of the TCLP method (Ghosh et al., 2004).

4.

Summary

The major findings from this study are summarized as follows:  A water-soluble potato starch was successfully used as a bridging agent to prepare a class of bridged magnetite nanoparticles. Based on TEM images, the starch-bridged nanoparticles displayed a mean particle size of 26.6  4.8 nm.  Batch kinetic tests revealed that the concentration of starch can substantially affect the arsenic sorption capacity. The optimum starch concentration for 0.57 g/L as Fe nanoparticles was found to be 0.049 wt.%. Based on the kinetic data, the starched nanoparticles removed 5 times more arsenic than bare magnetite particles under otherwise identical conditions, yet the bridged nanoparticles can be easily separated from water by gravity.  The maximum Langmuir capacity for As(V) sorption was determined to be 248 mg/g at pH 5.0 in the presence of 6% of NaCl. Because of the starch coating, the high arsenate





sorption capacity was not diminished as NaCl concentration was increased to 10%. The starch coating rendered the sorbent less sensitive to pH change and the sorbent high arsenic capacity was observed over a broader pH range, compared to bare magnetite particles. The maximum arsenate uptake was obtained in the pH range of 4e6 for the starched nanoparticles. FTIR and surface potential analyses suggested that arsenate was sorbed to the starch nanoparticels through innersphere surface complexation (Lewis acidebase interaction) between the electron-pair acceptor (Fe) of magnetite and the donor (O) of arsenate. Sorption of starch to the nanoparticles greatly altered the surface potential, resulting in an unusual near-zero surface potential in the presence of 6% NaCl over a wide pH range. The nanoparticles were chemically stable and did not dissolve at pH >5. At pH 3.2, 13% of the starched magnetite was dissolved over a 50-day period. At an Fe-to-As molar ratio of 5, the arsenic-laden precipitate was able to pass the 5 mg/L TCLP standard set by US EPA.

Compared to traditional IX operations, the brine treatment technology introduced herein offers the following key advantages: 1) It allows the regenerant to be reused, which can greatly reduce the treatment cost given that regeneration cost can account for as high as 50% of operating cost in standard IX processes, and 2) Because it concentrates As in a small volume of spent magnetite precipitate that is safe for discharging, the volume of the process waste residual and the associated disposal cost are likely substantially reduced.

Acknowledgements This research was partially supported by the Alabama Agricultural Experiment Station Hatch and Multistate Funding program, an EPA STAR Grant (GR832373) and a grant from Vulcan Materials, Inc.

references

Alibeigi, S., Vaezi, M.R., 2008. Phase transformation of iron oxide nanoparticles by varying the molar ratio of F2þ:Fe3þ. Chem. Eng. Technol. 31 (11), 1591e1596. An, B., Steinwinder, T.R., Zhao, D., 2005. Selective removal of arsenate from drinking water using a polymeric ligand exchanger. Water Res. 39, 4993e5004. Anderson, P.J., 1957. On the ion adsorption properties of synthetic magnetite. Proc. 2nd Intern. Cong. Surf. Activity 3, 67e80. Anastassakis, G.N., 1999. A study on the separation of magnetite fines by magnetic carrier methods. Colloids Surf. A Physicochem. Eng. Aspects 149, 585e593. Aria, Y., Sparks, D.L., Davis, J.A., 2004. Effects of dissolved carbonate on arsenate adsorption and surface speciation at the hematite water interface. Environ. Sci. Technol. 38, 817e824. Ataie, A., Heshmati-Manesh, S., 2001. Synthesis of ultra-fine particles of strontium hexaferrite by a modified coprecipitation method. J. Eur. Ceram. Soc. 21, 1951e1955. Ballinas, M.L., Rodriguez de San Miguel, E., De Jesus Rodriguez, M.T., Silva, O., Munoz, M., De Gyves, J., 2004. Arsenic(V)

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 6 1 e1 9 7 2

removal with polymer inclusion membranes from sulfuric acid media using DBBP as carrier. Environ. Sci. Technol. 38 (3), 886e891. Biggs, S., 2006. Polymeric flocculants. In: Somasundaran, P. (Ed.) Encyclopedia of Surface and Colloid Science, second ed., vol. 7. Taylor and Francis, New York. Clifford, D.S., Ghurye, G.L., Tripp, A.R., 2003. As removal using ion exchange with spent brine recycling. J. Am. Water Works Assoc. 95 (6), 119e130. Clifford, D.S., 1999. Ion exchange and inorganic adsorption. In: Water Quality and Treatment, a Handbook of Community Water Supplies. American Water Works Association, vol. 9. 1e9.91. McGraw Hill, New York. Coker, V.S., Gault, A.G., Pearce, C.I., Van der Laan, G., Telling, N.D., Charnock, J.M., Polya, D.A., Lloyd, J.R., 2006. XAS and XMCD evidence for species-dependent partitioning of arsenic during microbial reduction of ferrihydrite to magnetite. Environ. Sci. Technol. 40, 7745e7750. Compean-Jasso, M.E., Ruiz, F., Martinez, J.R., Herrera-Gomez, A., 2008. Magnetic properties of magnetite nanoparticles synthesized by forced hydrolysis. Mater. Lett. 62, 4248e4250. Cornell, R.M., Schwertmann, U., 2003. The Iron Oxides: Structure, Properties, Reactions, Occurrences, and Uses, second ed., Wiley-VCH, Weinheim, Germany. Cumbal, L., SenGupta, A.K., 2005. Arsenic removal using polymersupported hydrated iron (III) oxide nanoparticles: role of donnan membrane effect. Environ. Sci. Technol. 39, 6508e6515. Dixit, S., Hering, J.G., 2003. Comparison of arsenic(V) and arsenic (III) sorption onto iron oxide minerals: implications for arsenic mobility. Environ. Sci. Technol. 37, 4182e4189. EPA, 2000. Arsenic Removal from Drinking Water by Ion Exchange and Activated Alumina Plants (EPA/600/R-00/088). Office of research and development, Cincinnati, OH. Fendorf, S.E., Eick, M.J., Grossl, P., Sparks, D.L., 1997. Arsenate and chromate retention mechanisms on goethite. 1. Surface structure. Environ. Sci. Technol. 31 (2), 315e320. Ferguson, J.F., Anderson, M.A., 1974. Chemical form of arsenic in water supplies and their removal. In: Rubin, A.J. (Ed.), Chemistry of Water Supply, Treatment, and Distribution. Ann Arbor Science, Ann Arbor, MI. Gimenez, J., Martinez, M., Pablo, J., Rovira, M., Duroc, L., 2007. Arsenic sorption onto natural hematite, magnetite, and goethite. J. Hazard. Mater. 141, 575e580. Ghosh, A., Mukiibi, M., Ela, W., 2004. TCLP underestimates leaching of arsenic from solid residuals under landfill conditions. Environ. Sci. Technol. 38, 4677e4682. Goldberg, S., Johnston, C.T., 2001. Mechanisms of arsenic adsorption on amorphous oxides evaluated using macroscopic measurements, vibrational spectroscopy, and surface complexation modeling. J. Colloid Interf. Sci. 234, 204e216. He, F., Zhao, D., 2005. Preparation and characterization of a new class of starch-stabilized bimetallic nanoparticles for degradation of chlorinated hydrocarbons in water. Environ. Sci. Technol. 39, 3314e3320. He, F., Zhao, D., 2007. Manipulating the size and dispersibility of zerovalent iron nanoparticles by use of carboxymethyl cellulose stabilizers. Environ. Sci. Technol. 41, 6216e6221. He, F., Zhao, D., Paul, C., 2010. Field assessment of carboxymethyl cellulose stabilized iron nanoparticles for in situ destruction of chlorinated solvents in source zones. Water Res. 44, 2360e2370. He, F., Zhao, D., Liu, J., Roberts, C.M., 2007. Stabilization of FeePd nanoparticles with sodium carboxymethyl cellulose for enhanced transport and dechlorination of trichloroethylene in soil and groundwater. Ind. Eng. Chem. Res. 46, 29e34. Hering, J.G., Chen, P.-Y., Wilkie, J.A., Elimelech, M., 1997. Arsenic removal from drinking water during coagulation. J. Environ. Eng. ASCE 123, 800e807.

1971

Hirashima, M., Takahashi, R., Nishinari, K., 2005. Effects of adding acids before and after gelatinization on the viscoelasticity of cornstarch pastes. Food Hydrocolloides 19, 909e914. Huang, C.P., Stumm, W., 1973. Specific adsorption of cations on hydrous Al2O3. J. Colloid Interf. Sci. 43 (2), 409e420. Huang, M., Wang, H., Yu, J., 2006. Studies of biodegradable thermoplastic amylose/kaolin composites: fabrication, characterization, and properties. Polym. Composite, 309e314. Huber, D.L., 2005. Synthesis, properties, and applications of iron nanoparticles. Small 1 (5), 482e501. Husband, J.C., 1998. The adsorption of starch derivatives onto kaolin. Colloids Surf. A Physicochem. Eng. Aspects 131, 145e159. Illes, E., Tombacz, E., 2003. The role of variable surface charge and surface complexation in the adsorption of humic acid on magnetite. Colloids Surf. A 230, 99e109. Jolivet, J.-P., 2000. Metal Oxide Chemistry and Synthesis from Solution to Solid State. Wiley, Chichester. Kanel, S.R., Nepal, D., Manning, M., Choi, H., 2007. Transport of surface-modified iron nanoparticles in porous media and application to arsenic(III) remediation. J. Nanopart. Res. 9, 725e735. Kanel, S.R., Greneche, J.M., Choi, H., 2006. Arsenic(V) removal from groundwater using nano scale zero-valent iron as a colloidal reactive barrier material. Environ. Sci. Technol. 40, 2045e2050. Kosmulski, M., Hartikainen, J., Mau¨czka, E., Janusz, W., Rosenholm, J.B., 2002. Multiinstrument study of the electrophoretic mobility of fumed silica. Anal. Chem. 74, 253e256. Lee, K.M., Kim, S.-G., Kim, W.-S., Kim, S.S., 2002. Properties of iron oxide particles prepared in the presence of dextran. Korean J. Chem. Eng. 19 (3), 480e485. Ma, X., Bruckard, W.J., 2010. The effect of pH and ionic strength on starch-kaolinite interactins. Int. J. Miner. Process. 94, 111e114. Maity, D., Agrawal, D.C., 2007. Synthesis of iron oxide nanoparticles under oxidizing environment and their stabilization in aqueous and non-aqueous media. J. Magn. Magn. Mater. 308, 46e55. Nishio, Y., Yamada, A., Ezaki, K., Miyashita, Y., Furukawa, H., Horie, K., 2004. Preparation and magnetometric characterization of iron oxide-containing alginate/poly(vinyl alcohol) networks. Polymer 45 (21), 7129e7136. NRC, 2001. Arsenic in Drinking Water: 2001 Update. National Research Council, National Academy Press, Washington DC. Pena, M., Meng, X., Korfiatis, P.G., Jing, C., 2006. Adsorption mechanism of arsenic on nanocrystalline titanium dioxide. Environ. Sci. Technol. 40, 1257e1262. Phenrat, T., Marhaba, T.F., Rachakornkij, M., 2008. Leaching behaviors of arsenic from arsenic-iron hydroxide sludge during TCLP. J. Environ. Eng. 134 (8), 671e682. Pierce, M.L., Moore, C.B., 1982. Adsorption of arsenic and arsenate on amorphous iron hydroxide. Water Res. 16, 1247e1253. Prakash, S., Das, B., Mohanty, J.K., Venugopal, R., 1999. The recovery of fine iron minerals from quartz and corundum mixtures using selective magnetic coating. Int. J. Miner. Process. 57, 87e103. Qian, Y., Xie, Y., He, C., Li, J., Chen, Z., 1994. Hydrothermal preparation and characterization of ultrafine magnetite powders. Mater. Res. Bull. 29 (9), 953e957. Ramana, A., SenGupta, A.K., 1992. A new class of selective sorbents for arsenic and selenium oxy-anions. Environ. Eng. ASCE 118 (5), 755e775. Raveendran, P., Fu, J., Wallen, S.L., 2003. Complete “green” synthesis and stabilization of metal nanoparticles. J. Am. Chem. Soc. 125, 13940e13941. Ravishankara, S.A., Pradip, Khosla, N.K., 1995. Selective flocculation of iron oxide from its synthetic mixtures with

1972

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 6 1 e1 9 7 2

clays: a comparison of polyacrylic acid and starch polymers. Int. J. Miner. Process. 43, 235e247. Schwertmann, U., Cornell, R.M., 2000. Iron Oxide in the Laboratory. WILEY-VCH, Weinheim, Germany. Shipley, H.J., Yean, S., Kan, A.T., Tomson, M.B., 2009. Adsorption of arsenic to magnetite nanoparticles: effect of particle concentration, pH, ionic strength, and temperature. Environ. Toxicol. Chem. 28, 509e515. Su, C., Puls, R.W., 2001. Arsenate and arsenite removal by zerovalent iron: kinetics, redox transformation, and implications for in situ groundwater remediation. Environ. Sci. Technol. 35, 1487e1492. Sun, X., Doner, H.E., 1996. An investigation of arsenate and arsenite bonding structure on goethite by FTIR. Soil Sci. 161, 865e872. Tamaura, Y., Takahasi, K., Kodera, Y., Saito, Y., Inada, Y., 1986. Chemical modification of lipase with ferromagnetic modifier e a ferromagnetic-modified lipase. Biotechnol. Lett. 8, 877e880. Tang, N.J., Zhong, W., Jiang, H.Y., Wu, X.L., Liu, W., Du, Y.W., 2004. Nanostructured magnetite (Fe3O4) thin films prepared by solegel method. J. Magn. Magn. Mater. 282, 92e95. Tang, J., Mayers, M., Bosnick, K.A., Brus, L.E., 2003. Magnetite Fe3O4 nanocrystals: spectroscopic observation of aqueous oxidation kinetics. J. Phys. Chem. B. 107, 7501e7506. Wang, S., Xin, H., Qian, Y., 1997. Preparation of nanocrystalline Fe3O4 by g-ray radiation. Mater. Lett. 33, 113e116.

Waychunas, G.A., Rea, B.A., Fuller, C.C., Davis, J.A., 1993. Surface chemistry of ferrihydrite: part1. EXAFS studies of the geometry of coprecipitated and adsorbed arsenate. Geochim. Cosmochim. Acta 57 (10), 2251e2269. World Health Organization, 2006. Arsenic in the Drinking Water Fact Sheet 210. World Health Organization, Geneva, Switzerland. Yavuz, C.T., Mayo, J.T., Yu, W.W., Prakash, A., Falkner, J.C., Yean, S., Cong, L., Shipley, H.J., Kan, A., Tomson, M., Natelson, D., Colvin, V.L., 2006. Low-field magnetic separation of monodisperse Fe3O4 nanocrystals. Science 314, 964e967. Yean, S., Cong, L., Yavuz, C.T., Mayo, J.T., Yu, W.W., Kan, A.T., Colvin, V.L., Tomson, M.B., 2005. Effect of magnetite particles on adsorption and desorption of arsenite and arsenate. J.Mater. Res. 20 (12), 3255e3264. Yokoi, H., Kantoh, T., 1993. Thermal decomposition of the iron(III) hydroxide and magnetite composites of poly(vinyl alcohol). Preparation of magnetite and metallic iron particles. B. Chem. Soc. Jpn. 66, 1536e1541. Yuan, C., Lien, H., 2006. Removal of arsenate from aqueous solution using nanoscale iron particles. Water Qual. Res. J. Can. 41, 210e215. Zhang, Y., Yang, M., Dou, X.-M., He, H., Wang, D.-S., 2005. Arsenate adsorption on an FeeCe bimetal oxide adsorbent: role of surface properties. Environ. Sci. Technol. 39, 7246e7253.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 7 3 e1 9 8 3

Available at www.sciencedirect.com

journal homepage: www.elsevier.com/locate/watres

Controlling harmful cyanobacterial blooms in a hyper-eutrophic lake (Lake Taihu, China): The need for a dual nutrient (N & P) management strategy Hans W. Paerl a,*, Hai Xu b, Mark J. McCarthy c,d, Guangwei Zhu b, Boqiang Qin b, Yiping Li e, Wayne S. Gardner d a

University of North Carolina at Chapel Hill, Institute of Marine Sciences, 3431 Arendell Street, Morehead City, NC 28557, USA State Key Laboratory of Lake Science and Environment, Nanjing Institute of Geography & Limnology, Chinese Academy of Sciences, 73 East Beijing Road, Nanjing 210008, PR China c Universite´ du Que´bec a` Montre´al, De´partement des sciences biologiques, Montre´al, Que´bec H3C 3P8, Canada d University of Texas at Austin, Marine Science Institute, 750 Channel View Drive, Port Aransas, TX 78373, USA e Department of Environmental Science and Engineering, Hohai University, 1 Xikan Road, Nanjing 210098, PR China b

article info

abstract

Article history:

Harmful cyanobacterial blooms, reflecting advanced eutrophication, are spreading globally

Received 16 February 2010

and threaten the sustainability of freshwater ecosystems. Increasingly, non-nitrogen (N2)-

Received in revised form

fixing cyanobacteria (e.g., Microcystis) dominate such blooms, indicating that both excessive

26 August 2010

nitrogen (N) and phosphorus (P) loads may be responsible for their proliferation. Tradi-

Accepted 14 September 2010

tionally, watershed nutrient management efforts to control these blooms have focused on

Available online 29 September 2010

reducing P inputs. However, N loading has increased dramatically in many watersheds,

Keywords:

acteristics. We examined this proliferating water quality problem in Lake Taihu, China’s

Eutrophication

3rd largest freshwater lake. This shallow, hyper-eutrophic lake has changed from bloom-

Nitrogen

free to bloom-plagued conditions over the past 3 decades. Toxic Microcystis spp. blooms

Phosphorus

threaten the use of the lake for drinking water, fisheries and recreational purposes.

Cyanobacteria

Nutrient addition bioassays indicated that the lake shifts from P limitation in wintere

China

spring to N limitation in cyanobacteria-dominated summer and fall months. Combined N

Nutrient management

and P additions led to maximum stimulation of growth. Despite summer N limitation and P

promoting blooms of non-N2 fixers, and altering lake nutrient budgets and cycling char-

availability, non-N2 fixing blooms prevailed. Nitrogen cycling studies, combined with N input estimates, indicate that Microcystis thrives on both newly supplied and previouslyloaded N sources to maintain its dominance. Denitrification did not relieve the lake of excessive N inputs. Results point to the need to reduce both N and P inputs for long-term eutrophication and cyanobacterial bloom control in this hyper-eutrophic system. ª 2010 Elsevier Ltd. All rights reserved.

1.

Introduction

Harmful (toxic, food web-disrupting) cyanobacterial blooms (CyanoHABs) are a troubling indicator of advanced

eutrophication. These blooms are increasing worldwide and represent a serious threat to drinking water supplies, and the ecological and economic sustainability of our largest freshwater ecosystems (Reynolds, 1987; Paerl, 1988; Carmichael,

* Corresponding author. Tel.: þ1 252 726 6841; fax: þ1 252 726 2426. E-mail address: [email protected] (H.W. Paerl). 0043-1354/$ e see front matter ª 2010 Elsevier Ltd. All rights reserved. doi:10.1016/j.watres.2010.09.018

1974

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 7 3 e1 9 8 3

2001; Huisman et al., 2005). Examples include Lake Erie (North America), Lake Winnipeg, Canada, Lake Victoria, the largest of the African rift lakes, Lakes Biwa and Kasimagaura, Japan’s largest lakes, and Lake Taihu, the 3rd largest freshwater lake in China. Anthropogenic nutrient over-enrichment of these and smaller lakes and reservoirs has been linked to CyanoHAB proliferation (Paerl et al., 2001; Huisman et al., 2005). Nitrogen (N) and phosphorus (P) are the key nutrients of concern (Likens, 1972; Schindler, 1977; Paerl, 2008). Inputs of both nutrients to natural waters have increased dramatically in the post-World War II era. A central question facing water researchers and managers is; which nutrient(s) control CyanoHAB production and proliferation in impacted systems? The answer to this question is of immense ecological and economic importance, because it dictates the strategies and costs involved in mitigating this serious water quality problem. Phosphorus has been implicated traditionally as having a central role in the control of freshwater primary production (Likens, 1972) and CyanoHAB bloom formation (Paerl, 1988, 2008). This conclusion is particularly relevant to nitrogen (N2) fixing CyanoHABs, since they may satisfy their own N requirements (Smith, 1983, 1990). As a result, P input restrictions have been implemented widely since the 1960s. These reductions have slowed eutrophication rates and reduced algal bloom potentials (Schindler, 1977). However, nutrient loading dynamics have changed considerably since the 1960s. Agricultural, urban and industrial nutrient sources have accelerated rapidly (Vitousek et al., 1997; Galloway and Cowling, 2002). These sources are treated more effectively for removal of P than N before being discharged, leading to higher N than P loading to already nutrient-stressed water bodies (Paerl, 1997; Rabalais, 2002; Boyer et al., 2004). Excessive N loads have promoted non-N2 fixing CyanoHABs, including expanding blooms of the toxin producing, colonial, surfacedwelling (buoyant) cyanoHAB Microcystis, and the filamentous genera Lyngbya (non-N2 fixing strains) and Oscillatoria (Paerl, 2008, 2009). A severely impacted large lake system is Taihu (meaning “great lake” in Mandarin), with an area of 2338 km2 and a volume of 4.4 billion m3 (Pu and Yan, 1998; Qin et al., 2007, 2010). This shallow (mean depth 1.9 m) polymictic lake is located in the Yangtze River delta; the most rapidly developing region in China (Fig. 1). Approximately 40 million people live within the Taihu watershed. The Taihu Basin accounts for only 0.4% of China’s land area, but the region accounts for 11% of its Gross Domestic Product (Qin et al., 2007). The lake is a key drinking water, fishing and tourism resource for the region. However, it also serves as a waste repository for urban, agricultural and industrial segments of the local economy (Guo, 2007; Qin et al., 2007). Consequently, Taihu has experienced accelerating eutrophication over the past 3 decades (Qin et al., 2007, 2010). During this period, it has changed from a mesotrophic, diatom-dominated lake to hyper-eutrophic, cyanobacteria-dominated system, with Microcystis blooms now occurring regularly throughout much of the lake (Chen et al., 2003a,b) (Fig. 1). These blooms have caused environmental, economic and societal problems, including a threat to potable water supplies for approximately 10 million consumers (Guo, 2007).

Fig. 1 e Upper: Satellite (MODIS) image of Lake Taihu, the nearby cities of Wuxi (N), Souzhou (E), and Shanghai (E), near the mouth of the Yangtze River on 7 June, 2007, when a massive cyanobacterial (Microcystis spp.) bloom that covered almost the entire lake (image courtesy NASA). Lower: Photograph of a Microcystis bloom on the north side of the lake on 24 October, 2009 (photograph by H. Paerl).

Microcystis and other non-N-fixing cyanobacteria are effective competitors for reduced N forms, especially ammonium (NHþ 4 ) (Kappers, 1980; Blomqvist et al., 1994), which is rapidly regenerated in the water column and sediments of these shallow systems. Cyanobacteria capable of N2 fixation can also assimilate NHþ 4 when available. For example, Shelburne Pond in Vermont was dominated by N2 fixers, but characterized by low rates of N fixation (<9% of total N uptake) and NOx uptake (<5%), but high acquisition rates of NHþ 4 (82e98%) in summer (Ferber et al., 2004). Thus, the presence of N2 fixers does not ensure that N2 fixation is a significant N source; regenerated NHþ 4 may be a key N source for sustaining blooms, even among N2 fixing groups. Although Microcystis blooms dominate, heterocystous, filamentous genera capable of N2 fixation (Anabaena, Aphanizomenon) are also present in the water column and as overwintering cells in the sediments of Taihu. The N2 fixing genera increase in numeric importance toward the lake center, i.e., away from shoreline locations, although they remain subdominant to Microcystis (Chen et al., 2003a,b). While N2 fixation rates have not been measured in the lake, these observations

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 7 3 e1 9 8 3

suggest that N2 fixing conditions may increase from the lake’s shorelines to its open waters, as hypothesized by McCarthy et al. (2007). These observations also suggest that nutrient availability and limitations may vary spatially in the lake. Temporal differences may also be important because watershed N and P inputs exhibit seasonal patterns (Qin et al., 2007, 2010; Xu et al., 2010). The fact that both N2 fixing and non-N2 fixing cyanoHABs coexist in hyper-eutrophic Taihu has raised questions as to whether N, P or both N and P inputs should be reduced to control blooms. Based on research in other lakes, primary production and bloom formation may be controlled by both N and P inputs, either contemporaneously or sequentially, with different individual nutrients being limiting at different times of the year (Dodds et al., 1989; Elser et al., 2007; Havens et al., 2001; Kronvang et al., 2005; Jeppesen et al., 2007; North et al., ¨ zkan et al., 2009; Xu 2007; Lewis and Wurtsbaugh, 2008; O et al., 2010). These findings have been challenged recently, based on the premise that lake eutrophication cannot be controlled by reducing N inputs (Schindler et al., 2008). This conclusion relies on the observation that many systems exhibiting advanced eutrophication also contain significant N2 fixing cyanoHAB populations, and the assumption that N fixed by these populations can meet ecosystem N requirements. However, diverse studies show that only a fraction, usually far less than 50% of ecosystem-level N demands, is met by N2 fixation (Howarth et al., 1988; Paerl, 1990; Lewis and Wurtsbaugh, 2008). In fact, a recent re-examination of Schindler et al.’s (2008) data on P-fertilized Lake 227 indicates that N2 fixation does not meet ecosystem N demands (Scott and McCarthy, 2010). In other cases, N2 fixation rates are very low and supply little new N, even when N fixing cyanobacteria are dominant (Ferber et al., 2004). Furthermore, factors in addition to stoichiometric N:P ratios (Smith, 1983, 1990) control this energy-demanding process in aquatic ecosystems (Howarth et al., 1988; Paerl, 1990; Forbes et al., 2008). Reversing eutrophication and reducing CyanoHABs in a range of lakes have required either reduction of only P (Schindler, 1977) or both N and P inputs (Kronvang et al., 2005; Jeppesen et al., 2007). Hence, nutrient reduction strategies appear system-specific. We examined effects of individual and combined N and P additions on phytoplankton growth (based on chlorophyll a concentrations) in Taihu, using short-term (up to 6 days) nutrient addition bioassays incubated under natural light and temperature conditions. These bioassays provide a rapid assessment of nutrient limitation characteristics, i.e., immediate growth responses, rather than predicting long-term phytoplankton succession patterns (Paerl and Bowles, 1987; Piehler et al., 2009). We also evaluated NHþ 4 regeneration and potential uptake rates within the context of dominance by non-N2 fixing cyanoHABs.

2.

Methods and materials

2.1.

Location and field sites

Lake Taihu is located approximately 150 km west of Shanghai; with the lake center coordinates at 31 100 000 N, 120 90 000 E. The

1975

Taihu drainage basin is 36,500 km2 (Fig. 2). The lake has more than 30 input sources, ranging from rivers to small streams and man-made drainage canals. Water exits the southeastern corner of Lake Taihu via the Taipu River, which drains through Shanghai into the East China Sea (Figs. 1 and 2). Field monitoring and bioassay water collection sites were located in one of the northern bays, Meiliang Bay (Inner Bay, Outer Bay), and the lake proper (Main Lake) (Fig. 2). Other lake sites sampled included the least bloom-impacted eastern regions of the lake (ELT stations) (Fig. 2). Meiliang Bay was chosen as a focal point because it is the site of recurring and intensifying Microcystis spp. blooms (Chen et al., 2003a,b; Qin et al., 2007). Meiliang Bay receives freshwater inputs from the Liangxi and Zhihu Gang rivers, which drain untreated wastewater from factories, residential and agricultural areas. The named rivers in Fig. 2 contribute more than 85% of the lake’s freshwater inflow.

2.2.

Nutrient inputs to Lake Taihu

Data were obtained from the Taihu Basin Authority, Ministry of Water Resources, China, for 30 primary tributaries from 2000 to 2005, including monthly flow discharge and TN and TP concentrations, to estimate N and P loads discharged to the lake (Fig. 2). These tributaries with 30 monitoring stations accounted for approximately 85% of the total runoff input to the lake (Qin et al., 2007). TN or TP load for each month was calculated by the following formula: Fi ¼ 2:592$

30 X

Cij $Qij

(1)

j¼1

where Fi is TN or TP load for ith month (i ¼ 1e12) in metric tons, Cijis the TN or TP concentration for ith month and jth tributary ( j ¼ 1e30) in mg/L. Qij is the flow discharge for ith month and jth tributary ( j ¼ 1e30) in m3/s. 2.592 is the factor for converting g/s to ton/mo. Spring, summer, autumn and winter in this paper represent MarcheMay, JuneeAugust, SeptembereNovember, and DecembereFebruary, respectively. We estimated historic changes in atmospheric N loads to the lake based on data derived from regional atmospheric deposition models (Lelieveld and Dentener, 2000) and direct measurements (Zhai et al., 2009).

2.3.

Lake environmental measurements

Monthly samples for nutrients, chlorophyll a, and phytoplankton identification and enumeration were collected at the Meiliang Bay and Main Lake locations (Fig. 2). Water samples for nutrient addition bioassays were collected at the Inner Bay location. For logistical reasons, the bioassays were incubated at a lake location near the Taihu Laboratory for Lake Ecosystem Research (TLLER), Nanjing Institute of Geography and Limnology, Chinese Academy of Sciences (Fig. 2). Monthly physical, chemical, and biological parameters, including surface water temperature, dissolved oxygen, pH, and electrical conductivity, were measured using a Yellow Springs Instruments (YSI) 6600 multi-sensor sonde. Integrated water samples were taken using a 2-m long, 0.1-m wide plastic tube with a 1-way valve.

1976

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 7 3 e1 9 8 3

Fig. 2 e Map of Lake Taihu. The lake’s largest tributaries are named. The Taipu River is the main river discharging water from the lake. The Wangyu River periodically delivers Yangtze River water to the lake (see Fig. 1 for the location of the Yangtze River relative to Taihu). Sampling locations are indicated, as is the Taihu Laboratory for Lake Ecosystem Research (TLLER). Insert shows the location of Taihu in the PR China.

2.4.

Nutrient analyses

Water samples were analyzed for total nitrogen (TN), total dissolved nitrogen (TDN), dissolved inorganic nitrogen (DIN;   ammonium (NHþ 4 ) þ nitrate (NO3 ) þ nitrite (NO2 )), total phosphorus (TP), total dissolved phosphorus (TDP), and dissolved inorganic phosphorus (DIP). DIP was determined using the molybdenum blue method (APHA, 1995). NHþ 4 was determined  using the indophenol blue method, and NO 3 and NO2 with the cadmium reduction method (APHA, 1995). TP, TDP, TN, and TDN were determined using a combined persulphate digestion (Ebina et al., 1983), followed by spectrophotometric analysis as for DIP and NO 3 . Particulate nitrogen (PN) was obtained by subtracting TDN from TN, and particulate phosphorus (PP) by subtracting TDP from TP. Analytical errors were determined as the average % coefficients of variation of triplicates. Average errors for PP and PN were 6.3% and 5% respectively.

2.5.

Biological measurements

Phytoplankton samples were preserved with Lugol’s iodine solution (2% final conc.) and settled for 48 h. Phytoplankton

species were identified and counted according to Hu et al. (1980). Chlorophyll a (Chl a) concentrations were determined spectrophotometrically, following extraction in 90% hot ethanol (Papista et al., 2002).

2.6.

Nutrient limitation bioassay experiments

In situ nutrient addition bioassays were performed seasonally in 2008e2009 on “inner bay” water (Fig. 2) to examine nutrient limitation of the natural phytoplankton community. Water samples were collected from 0.2 m below the surface using 0.01 N HCl-washed and then lake water-rinsed 20 L polyethylene carboys. Water was screened through 200 mm mesh to remove large zooplankton and dispensed into acid (0.01 HCl) and then lake water washed 1-L polyethylene Cubitainers (Hedwin Co.). Cubitainers are chemically inert, unbreakable and transparent (80% PAR transmittance). The methodology and deployment for Cubitainer bioassays followed Paerl and Bowles (1987). At the start of each experiment (To), water samples were analyzed for Chl a and nutrients. Three treatments were conducted, in addition to a control (no nutrient additions): (1) N addition (þN) (2) P addition (þP) (3) N and P addition (þNP).

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 7 3 e1 9 8 3

Nitrogen was added as KNO3, reflecting the dominant form of inorganic N in Taihu. NH4Cl was also used as an N source in summer 2009 to compare phytoplankton growth responses to different N forms. Phosphorus was added as K2HPO4$3H2O. The final concentration of N was 2.00 mg N/L, and the final concentration of P was 0.50 mg P/L. These concentrations approximated the values of these nutrient forms in the lake during maximum discharge periods (Pu and Yan, 1998; Qin et al., 2007). All treatments were conducted in triplicate. Following nutrient additions, Cubitainers were incubated in situ near the surface for 4 or 6 days by placing them in a floating frame suspended off a pier at TLLER (Fig. 2). This approach provided natural light, temperature and surface wave action conditions. One layer of neutral density screening was placed over the frame to prevent photoinhibition during the incubations. Following deployment, Cubitainers were subsampled at 2e3 d intervals for Chl a and nutrient concentrations.

2.7. Water column NHþ 4 regeneration and potential uptake rates Water column NHþ 4 regeneration and potential uptake rates in Taihu, as described in McCarthy et al. (2007), were conducted to provide insights into consequences of unmitigated N discharges into Taihu. Previously published water column N cycling rates were determined in late summer 2002 from four sites in northern Taihu (McCarthy et al., 2007). The N and P addition bioassays described in the present study were conducted on water collected at sites corresponding to “inner bay” and “main lake” in McCarthy et al. (2007). Water column NHþ 4 recycling rates also were determined at the main lake station and four sites in East Lake Taihu (ELT) in January 2004 and all northern Taihu and ELT locations in May 2004 (Fig. 2). The ELT sites were located in the southeastern portion of the lake, near the outflow of the lake into the Taipu River. Methodological details are provided in McCarthy et al. (2007). Briefly, water samples were amended with saturating levels of 15NHþ 4 and incubated at ambient temperature and light for w24 h. Total NHþ 4 concentration and isotope ratios were determined using high performance liquid chromatography (HPLC; Gardner et al., 1995), and regeneration and potential uptake rates were calculated using the isotope dilution technique (Blackburn, 1979; Caperon et al., 1979). Since Lake Taihu is shallow and well-mixed, rates were extrapolated to the whole water column to estimate a total internal N load from water column recycling processes.

2.8.

3.

Results and discussion

3.1.

Nutrient inputs and concentrations

1977

Seasonal and annual surface water TN and TP loads to Taihu during 2000e2005 indicate that the lake received high amounts of N and P on a year-round basis, with N loads being higher, relative to P, in the winter (Fig. 3). Ratios (by weight) of TN to TP loading range from w22 to over 29, indicating relatively N enriched conditions on an annual basis. Lowest N:P loading ratios occur in summertime, whereas highest ratios occur in winterespring. The Taihu basin has, over the past 3 decades, experienced dramatic increases in population and urbanization (Qin et al., 2007; Guo, 2007). The high N:P loading reflects the combined effects of population growth, changes in land use, including increasing agricultural, and urban and rural wastewater discharge. These changes and the emphasis of P over N control in wastewater treatment (Wang and Wang, 2009) have led to increases in N:P loading from cities. In addition, reflecting a worldwide pattern (Galloway and Cowling, 2002), chemical N fertilizer use has increased in agricultural regions

Statistical analyses

Differences in the growth responses between various treatments were analyzed by one-way ANOVA. Post Hoc Multiple Comparisons of treatment means were performed by Tukey’s least significant difference procedure. Untransformed data in all cases satisfied assumptions of normality and homoscedasticity. Statistical analysis was performed using the SPSS 13.0 statistical package for personal computers, and the level of significance used was at p < 0.05 for all tests.

Fig. 3 e Upper: Seasonal total nitrogen (TN) loading to Lake Taihu from its tributaries, covering a period from 2000 to 2005, in metric tons of N. Lower: Seasonal total phosphorus (TP) loading, in metric tons of P, to Lake Taihu from its tributaries, from 2000 to 2005.

1978

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 7 3 e1 9 8 3

0.30

4.0 TP

TN 3.5

0.25

TP (mg/L)

2.5

0.15

2.0 1.5

0.10

TN (mg/L)

3.0 0.20

1.0 0.05

0.5

07

05

03

01

09 20

20

20

20

20

99 19

95

93

91

97 19

19

19

19

89

0.0

19

19

87

0.00

Year

Fig. 4 e Annual average TN and TP concentrations, as mg/L of each element, in Lake Tahu. Data are from the Main Lake station.

of the Taihu watershed, leading to N relative to P enrichment. Lastly, atmospheric deposition, which is enriched in N relative to P (Zhai et al., 2009), is an increasing source of N, accounting for approximately 30% of the external N loading to the lake. Atmospheric N inputs are estimated to have undergone a 25fold increase over the past 150 years, and this trend has accelerated during the past 3 decades (Lelieveld and Dentener, 2000) (Supplementary material 1: S-1). Recent increases in TN and TP loading can be seen as trends in TN and TP concentrations in the lake (mid-lake location) (Fig. 4). Increases were evident between the mid 1980s and 2000, a period of rapid population growth and urbanization in the Taihu watershed (Qin et al., 2007; Guo, 2007). They may have leveled off since the early 2000,

possibly a result of persistent droughts (Qin et al., 2007, 2010), diversion of sewage from the cities of Wuxi and Suzhou to waterways draining to the East China Sea, and usage of the Wangyu River canal, which exchanges water between Taihu and the Yangtze River. The decreases TN loading may have resulted from a reduction in pollutant input from the watershed in 1998. However, at the end of 1998, a pollutant emission reduction movement (the so called “zero point action”) launched by the government throughout the entire watershed, did not persist over time. Both TN and TP, as well as DIN and DIP concentrations, showed strong seasonal variation in Taihu. Maximum TN and DIN values occurred in winter and spring, whereas minimum values were observed in summer and autumn during the 2-yr period (Fig. 5). In contrast, TP and DIP values showed an inverse pattern, with winter having low values and summer high values. DIN patterns closely tracked TN over time (Fig. 5). Overall, TN values were higher in Meiliang Bay than in the lake proper, reflecting large external loads and elevated internal loading.

3.2.

Nutrient addition bioassays

The in situ nutrient addition bioassays showed stimulation of algal biomass production (as Chl a) in response to individual and combined N and P additions, indicating that nutrient enrichment enhanced algal growth and bloom potentials. Nutrient limitation showed strong repeated seasonal patterns in 2008 and 2009, although different degrees of algal biomass stimulation were observed between the two years (Fig. 6). P limitation prevailed in winter and spring, while N limitation occurred in summer and fall. In most instances, algal growth responses to N and P additions exceeded those observed with

Fig. 5 e Patterns of TN, DIN, TP and DIP, as mg/L of each element, at Inner Bay and Main Lake locations during 2008e2009.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 7 3 e1 9 8 3

Fig. 6 e Phytoplankton biomass (chlorophyll a) responses in bioassays conducted in May, July, October, and December 2008 and May, July and October 2009. Water samples for bioassays were collected from the surface at the Inner Bay location in Meiliang Bay. Initial chlorophyll a content is shown. Responses were for 3-day incubations in spring, summer, and fall, 6-day incubations in winter 2008, and 2-day incubations in spring, summer, and fall 2009. Mean values are shown. Error bars represent ±1SD of triplicate samples. Differences between treatments are shown based on ANOVA post hoc tests (a > b > c; p < 0.05).

N and P alone (Fig. 6). This “synergistic” N and P effect was most pronounced in summer and fall, when algal growth rates, and hence nutrient demands, were highest. Nutrient limitation patterns followed inverse concentration patterns of dissolved inorganic N (DIN) and P (DIP). DIN concentrations were high during winter and spring, reflecting high N inputs, while DIP concentrations were at their lowest levels (Fig. 5). As a result, DIP proved to be the limiting nutrient at this time. Then, DIN decreased rapidly as growth and bloom conditions improved in late spring and summer with increased light levels and temperatures. In contrast, DIP concentrations remained quite high and actually increased during summer bloom periods. The summer DIP increases may relate to elevated pH conditions (due to photosynthetic CO2 demand by blooms), which can enhance DIP release from the sediments (Andersen, 1975; Xu et al., 2010). Further, Microcystis can store P during sedimentary phases and assimilate N in larger proportion to P during bloom phases (Ahn et al., 2002). This uneven nutrient assimilation pattern can lead to decreases in N:P and enhance N limitation. It appears that DIN availability during this period determines the magnitudes and duration of booms. The availability of iron (Fe), a nutrient required for photosynthetic growth and N2 fixation, may have played an additional role in controlling phytoplankton production. However, when Fe (as EDTA-chelated and non-chelated Feþ2) was added alone and in combination with N and P to Inner Bay and open lake water samples during summer 2009, no significant stimulatory (or inhibitory) effects of Fe were observed (not shown).

1979

Fig. 7 e Phytoplankton biomass (chlorophyll a) responses to various nitrogen sources, each added at 2.00 mg N/L, in bioassays conducted in August 2009, using Inner Bay water from Meiliang Bay. Response is 2-day chlorophyll a average. Error bars represent ±1SD of triplicate samples. Differences between treatments are shown based on ANOVA post hoc tests (a > b > c; p < 0.05).

Parallel microscopic determinations of phytoplankton biomass agreed with chlorophyll a results (Xu et al., 2010), confirming that Chl a was a good indicator of phytoplankton biomass response in the bioassays. Additional total particulate organic carbon measurements made on the bioassays confirmed that Chl a responses reflected true increases in phytoplankton biomass. Microcystis spp. remained the dominant bloom-forming cyanobacteria during the summer-fall bloom periods of both years, despite chronic N limitation. These N limited periods should have provided optimal conditions for N2 fixing genera (i.e., Anabaena, Aphanizomenon) to become dominant (Smith, 1983, 1990; Schindler et al., 2008), but this situation did not develop, even though DIP remained plentiful (Figs. 4 and 5). Possible explanations for this result include; (1) superior ability of Microcystis spp. to compete for NHþ 4 and P from sediments (Kappers, 1980) and water column regeneration (Blomqvist et al., 1994), and (2) mutually-beneficial bacterialecyanobacterial interactions in the “phycosphere” of Microcystis spp. colonies, which can enhance nutrient cycling and growth of “host” Microcystis populations (Paerl and Pinckney, 1996). Bioassays illustrated that Microcystis spp. competed for DIN effectively, especially for NHþ 4 . Per amount of N added, ammonium stimulated significantly more algal biomass formation than nitrate (Fig. 7). The extent to which natural Microcystis populations dominated the absolute uptake of NHþ 4 was not determined, but phytoplankton biomass was dominated (>80%) by Microcystis in bioassays. This mechanism helps explain the persistence of Microcystis during periods of low DIN concentrations. In addition, Microcystis’ ability to

1980

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 7 3 e1 9 8 3

adjust its vertical position by buoyancy compensation (Reynolds, 1987) may enable exploitation of the entire water column, taking advantage of regenerated as well as externally-supplied (atmospheric, surface runoff) N sources. The results indicate that inputs of both nutrients should be reduced to control bloom formation and magnitude. Algal biomass production may be controlled by P availability in the spring, while N availability may determine the magnitude, spatial extent and duration of the bloom during summer-fall when the bloom potential is highest. Nutrient co-limitation was observed during all periods; i.e., combined enrichment with N and P led to higher magnitudes of biomass formation than either N or P alone. This result suggests that N and P supplies are closely balanced with regard to the requirements for supporting and promoting eutrophication and bloom formation.

3.3. Water column NHþ 4 regeneration and potential uptake Water column NHþ 4 regeneration and potential uptake rates showed seasonal and spatial differences (S-1). In January, light uptake rates were significantly lower ( p < 0.01; ANOVA) than those in September (main lake) and May (main lake and ELT sites). In northern Taihu, only the inner bay had a seasonal difference in regeneration rates, although the difference was large (0.89 mmol N/L h in May versus 0.19 mmol N/L h in September). In ELT, uptake and regeneration rates were lower than the Meiliang Bay sites, but this pattern was expected since ELT is dominated by submerged aquatic vegetation (SAV) rather than phytoplankton. To scale regeneration rates and to compare with external loads, sites were first regrouped based on location within the lake. Lake Taihu has a surface area of 2338 km2, but Meiliang Bay and ELT account for only 100 and 131 km2, respectively, of the total surface area. The outer bay site was located at the interface between Meiliang Bay and the main lake and is 15e20 km from river discharges. Thus, this site was grouped with the main lake station. After regrouping the sampling sites, volumetric regeneration rates (see S-1) were converted to areal rates using water depth and extrapolated based on surface area of the appropriate lake region. Extrapolated water column NHþ 4 regeneration rates suggest that 3.77  107 kg N/yr are regenerated as NHþ 4 in Meiliang Bay, where the most severe Microcystis blooms occur. Despite actual regeneration rates being an order of magnitude lower in the main lake than Meiliang Bay, the large surface area of the central basin results in 6.57  107 kg N/yr regenerated as NHþ 4. As expected, ELT plays a minor role in total N recycling in the water column (0.13  107 kg N/yr). The sum of these annual regeneration estimates is about 400% of total estimated N loading to the lake (2.5  107 kg N/yr; James et al., 2009). However, these estimates include only N regenerated as NHþ 4. No estimates of the NHþ 4 proportion of the total N load are known, but the proportion is presumably small relative to oxidized (i.e., NO 3 ) and organic N forms. Internal N cycling is important for the maintenance and species succession of cyanobacteria blooms in Taihu, especially as it pertains to Microcystis spp. (McCarthy et al., 2007). Atmospheric deposition is a significant additional source of bioavailable N in the

lake (Zhai et al., 2009) (S-2). For example, NHþ 4 and NO3concentrations of a rainwater sample collected in May 2004 were 370 and 146 mM, respectively (McCarthy and Gardner, unpublished data). Direct denitrification measurements in Meiliang Bay and the main lake in late summer 2002 (McCarthy et al., 2007) and Meiliang Bay, the main lake, and ELT from January to May 2004 (McCarthy and Gardner, unpublished data) were extrapolated to estimate a lake-wide denitrification rate. This rate ranged from 7360 kg N/km2 per year when estimated by net N2 flux to 26,700 kg N/km2 per year when estimated using 15NO 3 addition assays. Both estimates were obtained from continuous-flow incubations of intact sediment cores. Rates from 15NO 3 additions should be qualified as potential rates, whereas rates from net N2 flux would include any N2 fixing activities, which were not significant in sediments of this lake (McCarthy et al., 2007). Therefore, net N2 flux represents the best estimate of denitrification and accounts for 66.2% of external N loading. This N loss via denitrification would not account for the N recycled in the water column. In late summer, Meiliang Bay and main lake sediments are an N source to the water column (McCarthy et al., 2007). Sediments also are an N source in ELT and the main lake in January and May (McCarthy and Gardner, unpublished data). However, sediments in Meiliang Bay were a strong N sink in January and May. These patterns suggest that late summer cyanobacteria blooms rely, in part, on nutrients released from sediments (McCarthy et al., 2007). Depth averaged water column NHþ 4 regeneration rates for the shallow water column imply that water column regeneration supplies a greater amount of N (5-fold more on an areal basis) than sediments for cyanobacterial assimilation in the summer (McCarthy et al., 2007). In addition to the importance that total N loads play in determining rates of eutrophication, the supply rates and ratios of various N forms help structure microalgal communities mediating freshwater primary production (Paerl, 1988; McCarthy et al., 2007, 2009). For example, the ratio of NHþ 4 to oxidized N was related to the proportion of cyanobacteria comprising the total phytoplankton community of Lake Okeechobee, FL, USA (McCarthy et al., 2009). While non-N2 fixing cyanobacteria, such as Microcystis, compete effectively for reduced N (Blomqvist et al., 1994), N2 fixing cyanobacteria also assimilate ammonium preferentially if it is available (Ferber et al., 2004). Ammonium and other reduced N forms, such as dissolved free amino acids, are more available than oxidized N forms (nitrate and nitrite) to bacteria (Vallino et al., 1996) and cyanobacteria because less energy is required to incorporate and assimilate the former (Syrett, 1981; Gardner et al., 2004; Flores and Herrero, 2005). These issues were not addressed in recent studies suggesting that eutrophication cannot be controlled by reducing N inputs (e.g., Schindler et al., 2008; Wang and Wang, 2009). The assumption that N2 fixing genera will replace non-N2 fixing genera like Microcystis when N is limiting and P is sufficient could not be confirmed in Taihu. Furthermore, our observation that Taihu does not fit the proposed “P only” management paradigm of Schindler et al. (2008) is not unique. Numerous other lakes, reservoirs, rivers and fjords worldwide exhibit N and P co-limitation, either simultaneously or in

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 7 3 e1 9 8 3

seasonally-shifting patterns (Dodds et al., 1989; Elser et al., 2007; Forbes et al., 2008; Scott et al., 2008; North et al., 2007; Lewis and Wurtsbaugh, 2008; Conley et al., 2009; Xu et al., 2010; Abell et al., 2010).

4.

Conclusions

Nutrient loading analyses, nutrient addition bioassays and nutrient cycling studies provide the basis for recommending that N control be included, along with the previously prescribed P control (Chen et al., 2003a,b; Wang and Wang, 2009), as a nutrient management strategy for Taihu. Denitrification rates, while high relative to other lakes, are lower than estimates of N loading and therefore would not mitigate high N loads. Also, late summer cyanobacterial blooms are maintained primarily by water column N regeneration. Recycling available to non-N-fixing cyanobacterial produces NHþ 4 blooms (Microcystis), regardless of the N form discharged into the lake. The fact that Microcystis spp. were not replaced by N2 fixing cyanobacterial bloom species during N limited, but P sufficient summer periods is evidence that predictions of succession from non-N2 to N2 fixing taxa based on N:P stoichiometry (Smith, 1990; Schindler et al., 2008) may not apply to hypereutrophic lakes. Excess inputs of both N and P, combined with internal cycling of these nutrients, may overwhelm the ability of a single nutrient to control increasing eutrophication and bloom intensification in Lake Taihu and other large lakes experiencing such blooms (e.g., Lake Erie, Lake Okeechobee, Lake Victoria). P input reductions are an important component of eutrophication management in large lakes and reservoirs. However, failure to control N inputs may result in continued serious eutrophication problems caused by non-N2 fixing cyanobacterial blooms.

Acknowledgements We thank the TLLER, the Taihu Basin Authority, and the Chinese Ministry of Water Resources for providing water quality data, and Xiaodong Wang, Linlin Cai, Jingchen Xue, Lu Zhang and Longyuan Yang for assistance with sampling and chemical analyses and Guangbai Cui and Yong Pang for data collection. This research was supported by the Chinese Academy of Sciences (Contract: KXCX1-YW-14), the Ministry of Science and Technology of China (Contract: 2009ZX07101013), the Chinese National Science Foundation (Contract: 40825004, 40730529, 51009049), Fundamental Research Funds for the Central Universities (China), the US Environmental Protection Agency (Project 83335101-0), and US National Science Foundation (CBET Program) Project 0826819.

Appendix. Supplementary data Supplementary data related to this article can be found online at doi:10.1016/j.watres.2010.09.018

1981

references

¨ zkundakci, D., Hamilton, D.P., 2010. Nitrogen and Abell, J.M., O phosphorus limitation of phytoplankton growth in New Zealand lakes: implications for eutrophication control. Ecosystems. doi:10.1007/s10021-010-9367-9. Ahn, C.-Y., Chung, A.-S., Oh, H.-M., 2002. Rainfall, phycocyanin, and N:P ratios related to cyanobacterial blooms in a Korean large reservoir. Hydrobiologia 474, 117e124. Andersen, J.M., 1975. Influence of pH on release of phosphorus from lake sediments. Archiv fur Hydrobiologie 76, 411e419. APHA., 1995. Standard Methods for the Examination of Water and Wastewater, nineteenth ed. American Public Health Association, American Water Works Association, Water Environment Federation. Blackburn, T.H., 1979. Method for measuring rates of NHþ 4 turnover in anoxic marine sediments, using a 15NeNHþ 4 dilution technique. Applied & Environmental Microbiology 37, 760e765. Blomqvist, P., Pettersson, A., Hyenstrand, P., 1994. Ammoniumenitrogen: a key regulatory factor causing dominance of non-nitrogen-fixing cyanobacteria in aquatic systems. Archives of Hydrobiology 132 (2), 141e164. Boyer, E.W., Howarth, R.W., Galloway, J.N., Dentener, F.J., Cleveland, C., Asner, G.P., Greene, P., Vorosmarty, C., 2004. Current nitrogen inputs to world regions. In: Mosier, A.R., Syers, J.K., Freney, J.R. (Eds.), Agriculture and the Nitrogen Cycle. Island Press, Washington, DC, pp. 221e230. SCOPE #65. Caperon, J., Schell, D., Hirota, J., Laws, E., 1979. Ammonium excretion rates in Kaneohe Bay, Hawaii, measured by a 15N isotope dilution technique. Marine Biology 54, 33e40. Carmichael, W.W., 2001. Health effects of toxin producing cyanobacteria: the cyanoHABs. Human and Ecological Risk Assessessment 7, 1393e1407. Chen, Y.W., Qin, B.Q., Teubner, K., Dokulil, M.T., 2003a. Long-term dynamics of phytoplankton assemblages: Microcystisdomination in Lake Taihu, a large shallow lake in China. Journal of Plankton Research 25, 445e453. Chen, Y.W., Fan, C.X., Teubner, K., Dokulil, M., 2003b. Changes of nutrients and phytoplankton chlorophyll-a in a large shallow lake, Taihu, China: an 8-year investigation. Hydrobiologia 506, 273e279. Conley, D.J., Paerl, H.W., Howarth, R.W., Boesch, D.F., Seitzinger, S.P., Havens, K.E., Lancelot, C., Likens, G.E., 2009. Controlling eutrophication: nitrogen and phosphorus. Science 323, 1014e1015. Dodds, W.K., Johnson, K.R., Priscu, J.C., 1989. Simultaneous nitrogen and phosphorus deficiency in natural phytoplankton assemblages: theory, empirical evidence and implications for lake management. Lake and Reservoir Management 5, 21e26. Ebina, J., Tsutsui, T., Shirai, T., 1983. Simultaneous determination of total nitrogen and total phosphorus in water using peroxodisulfate oxidation. Water Research 17, 1721e1726. Elser, J.J., Bracken, M.E.S., Cleland, E.E., Gruner, D.S., Harpole, W.S., Hillebrand, H., Bgai, J.T., Seabloom, E.W., Shurin, J.B., Smith, J.E., 2007. Global analysis of nitrogen and phosphorus limitation of primary producers in freshwater, marine and terrestrial ecosystems. Ecololgy Letters 10, 1124e1134. Ferber, L.R., Levine, S.N., Lini, A., Livingston, G.P., 2004. Do cyanobacteria dominate in eutrophic lakes because they fix atmospheric nitrogen? Freshwater Biology 49, 690e708. Flores, E., Herrero, A., 2005. Nitrogen assimilation and nitrogen control in cyanobacteria. Biochemical Society Transactions 33 (1), 164e167. Forbes, M.G., Doyle, R.D., Scott, J.D., Stanley, J.K., Huang, H., Brooks, B.W., 2008. Physical factors control phytoplankton

1982

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 7 3 e1 9 8 3

production and nitrogen fixation in eight Texas reservoirs. Ecosystems 11, 1181e1197. Galloway, J.N., Cowling, E.B., 2002. Reactive nitrogen and the world: 200 years of change. Ambio 16 (2), 64e71. Gardner, W.S., Bootsma, H.A., Evans, C., St. John, P.A., 1995. Improved chromatographic analysis of 15N:14N ratios in ammonium or nitrate for isotopic addition experiments. Marine Chemistry 48, 271e282. Gardner, W.S., Lavrentyev, P.J., Cavaletto, J.F., McCarthy, M.J., Eadie, B.J., Johengen, T.H., Cotner, J.B., 2004. The distribution and dynamics of nitrogen and microbial plankton in southern Lake Michigan during spring transition 1999e2000. Journal of Geophysical Research 109, 1e16. Guo, L., 2007. Doing battle with the green monster of Lake Taihu. Science 317, 1166. Havens, K.E., Fukushima, T., Xie, O.P., Iwakuma, T., James, R.T., Takamura, N., Hanazato, T., Yamamoto, T., 2001. Nutrient dynamics and the eutrophication of shallow lakes Kasumigaura (Japan), Donghu (PR China), and Okeechobee (USA). Environmental Pollution 111 (2), 262e272. Howarth, R.W., Marino, R., Lane, J., Cole, J.J., 1988. Nitrogen fixation rates in freshwater, estuarine, and marine ecosystems. Limnology and Oceanography 33, 669e687. Hu, H., Li, Y., Wei, Y., Zhu, H., Shi, Z., 1980. Freshwater Algae in China. Shanghai Science and Technology Press, Shanghai (in Chinese). Huisman, J.M., Matthijs, H.C.P., Visser, P.M., 2005. Harmful Cyanobacteria. Springer Aquatic Ecology Series 3. Springer, Dordrecht, The Netherlands, p. 243. James, R.T., Havens, K., Zhu, G., Qin, B., 2009. Comparative analysis of nutrients, chlorophyll and transparency in two large shallow lakes (Lake Taihu, PR China and Lake Okeechobee, USA). Hydrobiologia 627, 211e231. Jeppesen, E., Søndergaard, M., Meerhoff, M., Lauridsen, T.L., Jensen, J.P., 2007. Shallow lake restoration by nutrient loading reductiondsome recent findings and challenges ahead. Hydrobiologia 584, 239e252. Kappers, F.I., 1980. The cyanobacterium Microcystis aeruginosa kg. and the nitrogen cycle of the hypertrophic Lake Brielle (The Netherlands), pp. 37e43. In: Barica, J., Mur, L. (Eds.), Hypertrophic Ecosystems Dr. W. Junk (The Hague, The Netherlands). Kronvang, B., Jeppesen, E., Conley, D.J., Søndergaard, M., Larsen, S.E., Ovesen, N.B., Carstensen, J., 2005. Nutrient pressures and ecological responses to nutrient loading reductions in Danish streams, lakes and coastal waters. Journal of Hydrology 304, 274e288. Lelieveld, J., Dentener, F., 2000. What controls tropospheric ozone? Journal of Geophysical Research 105, 3531e3551. Lewis Jr., W.M., Wurtsbaugh, W.A., 2008. Control of lacustrine phytoplankton by nutrients: erosion of the phosphorus paradigm. Internationale Revue der gesamten Hydrobiologie und Hydrographie 93, 446e465. Likens, G.E., 1972. In: Nutrients and Eutrophication Americ. Soc. Limnol. Oceanogr. Special Symp. 1. McCarthy, M.J., Lavrentyev, P.L., Yang, L., Zhang, L., Chen, Y., Qin, B., Gardner, W.S., 2007. Nitrogen dynamics relative to microbial food web structure in a subtropical, shallow, wellmixed, eutrophic lake (Taihu Lake, China). Hydrobiologia 581, 195e207. McCarthy, M.J., James, R.T., Chen, Y., East, T.L., Gardner, W.S., 2009. Nutrient ratios and phytoplankton community structure in the large, shallow, eutrophic, subtropical Lakes Okeechobee (Florida, USA) and Taihu (China). Limnology 10, 215e227. North, R.L., Guildford, S.J., Smith, R.E.H., Havens, S.M., Twiss, M.R., 2007. Evidence for phosphorus, nitrogen, and iron colimitation of phytoplankton communities in Lake Erie. Limnology and Oceanography 52, 315e328.

¨ zkan, K., Jeppesen, E., Johansson, L.S., Beklioglu, M., 2009. The O response of periphyton and submerged macrophytes to nitrogen and phosphorus loading in shallow warm lakes: a mesocosm experiment. Freshwater Biology. doi:10.1111/j. 1365e2427.2009.02297.x. Paerl, H.W., 1990. Physiological ecology and regulation of N2 fixation in natural waters. Advances in Microbial Ecology 11, 305e344. Paerl, H.W., 1988. Nuisance phytoplankton blooms in coastal, estuarine, and inland waters. Limnology and Oceanography 33, 823e847. Paerl, H.W., 1997. Coastal eutrophication and harmful algal blooms: importance of atmospheric deposition and groundwater as “new” nitrogen and other nutrient sources. Limnology and Oceanography 42, 1154e1165. Paerl, H.W., 2008. Nutrient and other environmental controls of harmful cyanobacterial blooms along the freshwater-marine continuum. Advances in Experimental Medicine and Biology 619, 216e241. Paerl, H.W., 2009. Controlling eutrophication along the freshwateremarine continuum: dual nutrient (N and P) reductions are essential. Estuaries and Coasts 32, 593e601. Paerl, H.W., Bowles, N.D., 1987. Dilution bioassays: their application to assessments of nutrient limitation in hypereutrophic waters. Hydrobiologia 146, 265e273. Paerl, H.W., Pinckney, J.L., 1996. Microbial consortia: their role in aquatic produc-tion and biogeochemical cycling. Microbial Ecology 31, 225e247. Paerl, H.W., Fulton, R.S., Moisander, P.H., Dyble, J., 2001. Harmful freshwater algal blooms, with an emphasis on cyanobacteria. The Scientific World 1, 76e113. Papista, E., Acs, E., Boeddi, B., 2002. Chlorophyll-a determination with ethanol e a critical test. Hydrobiologia 485, 191e198. Piehler, M.F., Dyble, J., Moisander, P.H., Chapman, A.D., Hendrickson, J., Paerl, H.W., 2009. Interactions between nitrogen dynamics and the phytoplankton community in Lake George, Florida, USA. Lake and Reservoir Management 25, 1e14. Pu, P., Yan, J., 1998. Taihu Lake e a large shallow lake in the East China plain. Journal of Lake Sciences (China) 10 (suppl), 1e12. Qin, B.Q., Xu, P.Z., Wu, Q.L., Luo, L.C., Zhang, Y.L., 2007. Environmental issues of Lake Taihu, China. Hydrobiologia 581, 3e14. Qin, B., Zhu, G., Gao, G., Zhang, Y., Li, W., Paerl, H.W., Carmichael, W.W., 2010. A drinking water crisis in Lake Taihu, China: linkage to climatic variability and lake management. Environmental Management 45, 105e112. Rabalais, N.N., 2002. Nitrogen in aquatic ecosystems. Ambio 16 (2), 102e112. Reynolds, C.S., 1987. Cyanobacterial water blooms. Advances in Botanical Research 13, 67e143. Schindler, D.W., 1977. The evolution of phosphorus limitation in lakes. Science 195, 260e262. Schindler, D.W., Hecky, R.E., Findlay, D.L., Stainton, M.P., Parker, B.R., Paterson, M., Beaty, K.G., Lyng, M., Kasian, S.E.M., 2008. Eutrophication of lakes cannot be controlled by reducing nitrogen input: results of a 37 year whole ecosystem experiment. Proceedings of the National Academy of Science USA 105, 11254e11258. Scott, J.T., Doyle, R.D., Prochnow, S.J., White, J.D., 2008. Are watershed and lacustrine controls on planktonic N2 fixation hierarchically structured? Ecological Applications 18, 805e819. Scott, J.T., McCarthy, M.J., 2010. Nitrogen fixation may not balance the nitrogen pool in lakes over timescales relevant to eutrophication management. Limnology and Oceanography 55, 1265e1270. Smith, V.H., 1983. Low nitrogen to phosphorus ratios favor dominance by blueegreen algae in lake phytoplankton. Science 221, 669e671. Smith, V.H., 1990. Nitrogen, phosphorus, and nitrogen fixation in lacustrine and estuarine ecosystems. Limnology and Oceanography 35, 1852e1859.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 7 3 e1 9 8 3

Syrett, P., 1981. Nitrogen metabolism in microalgae. Physiological bases of phytoplankton ecology. Canadian Journal of Fisheries and Aquatic Science 210, 182e210. Vallino, J.J., Hopkinson, C.S., Hobbie, J.E., 1996. Modeling bacterial utilization of dissolved organic matter: optimization replaces Monod growth kinetics. Limnology and Oceanography 41, 1591e1609. Vitousek, P.M., Mooney, H.A., Lubchenko, J., Mellilo, J.M., 1997. Human omination of earth’s ecosystem. Science 277, 494e499.

1983

Wang, H.J., Wang, H.Z., 2009. Mitigation of lake eutrophication: loosen nitrogen control and focus on phosphorus abatement. Progress in Natural Science 19, 1445e1451. Xu, H., Paerl, H.W., Qin, B.Q., Zhu, G.W., Gao, G., 2010. Nitrogen and phosphorus inputs control phytoplankton growth in eutrophic Lake Taihu, China. Limnology and Oceanography 55, 420e432. Zhai, S., Yang, L., Hu, W., 2009. Observations of atmospheric nitrogen and phosphorus deposition during the period of algal bloom formation in Northern Lake Taihu, China. Environmental Management 44, 542e551.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 8 4 e1 9 9 4

Available at www.sciencedirect.com

journal homepage: www.elsevier.com/locate/watres

Using reactive tracers to detect flow field anomalies in water treatment reactors Markus Gresch a,b, Daniel Braun b, Willi Gujer a,b,* a b

Swiss Federal Institute of Aquatic Science and Technology, Eawag, 8600 Du¨bendorf, Switzerland Institute of Environmental Engineering, ETH Zurich, 8093 Zurich, Switzerland

article info

abstract

Article history:

The hydraulics of water and wastewater treatment reactors has a major impact on their

Received 6 January 2010

performance and control. The residence time distribution as a measure for the hydraulics

Received in revised form

represents macroscopic mixing in an integrated way with no spatial information. However,

13 July 2010

with regard to optimal sensor location for process control and for process optimisation

Accepted 13 November 2010

measures, spatial information about macro-mixing is helpful. Spatially distributed

Available online 20 November 2010

measurements of reactive tracers can provide this information. In this paper we generally discuss how reactive tracers can be used to detect and characterize distinct large scale flow

Keywords:

structures. It is shown that tracer substances are particularly suited if their reaction time

Aeration tank

scale is similar to the time scale of the large scale flow structure. For nitrifying activated

Ammonium

sludge systems, ammonium is identified to be a suitable tracer. In a comprehensive

Ion-selective electrode

experimental study at a real aeration tank, two distinct large scale flow features were

Macro-mixing

identified by distributed ammonium measurements. Flow velocity measurements using

Reactive tracer

acoustic Doppler velocimetry clearly supported the nature of these flow field anomalies.

Reactor hydraulics

Ion-selective electrodes are a well suited device for ammonium measurements providing the temporal resolution that is needed for such an analysis. ª 2010 Elsevier Ltd. All rights reserved.

1.

Introduction

Reactors in water and wastewater treatment plants are typically designed as mixed tank reactors or as dispersive plug flow reactors. However, real reactors often show more complex macroscopic flow structures with negative effects on plant performance and control. For plant performance, the hydraulic characteristic of the reactor is one of the decisive parameters (Orhon et al., 1989). Depending on the relevant time scale of the reactions, different hydrodynamic time scales may be important. In water treatment, many reactions belong to the slow regime, where macro-mixing is the relevant mixing scale. The

residence time distribution (RTD) is an integrated quantity of the macroscopic flow field and the most important hydraulic parameter for the performance of such a reactor. Due to earliness of mixing effects, the knowledge of the residence time distribution is not sufficient in a strict sense (Levenspiel, 1999). But for many practical applications in water treatment, these effects are not of prime importance and the use of simplified hydraulic models aiming to capture the residence time distribution is commonly accepted. However, these simplified models like tank-in-series or advection-dispersion models often lack flexibility since their parameters essentially capture only different degrees of turbulent diffusion.

* Corresponding author. Swiss Federal Institute of Aquatic Science and Technology, Eawag, 8600 Du¨bendorf, Switzerland. Tel.: þ41 44 823 5036; fax: þ41 44 823 5389. E-mail address: [email protected] (W. Gujer). 0043-1354/$ e see front matter ª 2010 Elsevier Ltd. All rights reserved. doi:10.1016/j.watres.2010.11.017

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 8 4 e1 9 9 4

Plant control often relies on measurements of process relevant substances at specific locations in the reactor. The macroscopic flow pattern strongly influences the concentration of these substances and may lead to undesired effects in the control loop. Depending on the reactors’ geometry, infloweoutflow situation, location of separating walls, aeration systems and mixers or other flow relevant equipment very different macroscopic flow fields evolve. Typical undesired large scale flow features involve shortcuts, dead zones, large stationary vortices or flow instabilities leading to multidimensional flows. A characterization of these large scale flow features will help to understand reactor malfunctioning, to improve hydraulic models and to evaluate best sensor locations. The residence time distribution itself is a very integrated quantity of large scale flow features with no spatial information. More spatial information can be gained by using multidimensional hydrodynamic models (e.g. Computational Fluid Dynamics (CFD)), by spatially distributed monitoring of flow velocities or by using appropriate tracer techniques. Computational Fluid Dynamics is progressively used to analyze flow patterns in various water treatment reactors (Armbruster et al., 2001; Craig et al., 2002; Do-Quang et al., 2001; He et al., 2008; Saalbach and Hunze, 2008; Templeton et al., 2006). This tool has shown its general applicability to flow related problems in water technology. Although CFD is mostly based on physical principles, the simulation of a multiphase flow problem is still not straightforward. Closure terms, in particular interaction forces between the phases and the turbulence closure are still widely discussed in literature (Jakobsen et al., 2005; Tabib et al., 2008). Therefore, experimental validation of a CFD study, especially in complex flow situations, is still desired. In this paper we will primarily focus on tracer techniques to determine and characterize large scale flow features. In addition we use acoustic Doppler velocimetry for cross validation of the detected flow features. Unlike direct measurement of local flow properties like flow velocities of different phases, tracer measurements give no direct information on the flow field and need further processing and interpretation. But, they have the advantage that they work well in the harsh environment of many water treatment reactors (field conditions, very large tanks, opaque liquid, multiple phases) where most flow velocity measurements are hard to carry out. Tracer experiments with a pulse addition of a conservative tracer substance have often been used to determine the residence time distribution. They are particularly suited to give a measure for the overall dispersion. To detect large scale internal flow structures, they are only partly suited since valuable information quickly diminishes (Gresch et al., 2010). Applications to detect short circuiting and dead zones have been reported so far (Capela et al., 2009; Kjellstrand et al., 2005; Newell et al., 1998). Especially to gain more insight into large scale flow instabilities with tracer techniques, conventional tracer experiments are of little use because the time of tracer addition strongly effects the tracer distribution in the system. A single tracer experiment will therefore not be suited to characterize dynamic large scale flow structures. Different methods are required to provide insight into these features.

1985

Fig. 1 e Structure of a stationary vortex as it is often found in reactors with a flow field that is driven by the momentum of the inflowing fluid.

A very promising approach is the use of reactive tracer substances. The distribution of a reactive tracer is a result of the underlying flow field and the reaction rate of the substance. We therefore may use reactive tracer measurements to identify and characterize the large scale flow field. In this paper, we introduce this method for flow field detection. We discuss the tracer’s properties that are needed to produce valuable information about the underlying flow field and give general advice on the choice of the tracer substance and the measurement device that is suited to detect common large scale flow features. The method will be applied on an aeration tank of an activated sludge system. There, we use ammonium as tracer substance and ion-selective electrodes as a measurement device to detect internal flow structures. Two different large scale flow features could be successfully identified.

2. Concept of flow field detection with reactive tracers Typical large scale flow structures such as shortcuts, dead zones, large stationary vortices or flow instabilities produce a tracer distribution that is characteristic for the flow structure itself. Moreover, also the properties of the structure (e.g. the size of a large vortex and its rotational frequency) strongly influence the tracer distribution. For the following analysis, two typical flow structures are chosen and then analyzed with regard to their reactive tracer distribution. The first structure that is analyzed is a stationary vortex flow structure and the second flow feature is a periodic unsteadiness of the flow rate. Despite the fact that we will discuss these two distinct flow features, the method may easily be adapted to other flow situations.

2.1.

Stationary flow structure: vortex flow

The basic structure of a stationary vortex, as it is often found in the inlet region of a reactor, is shown in Fig. 1. It is

1986

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 8 4 e1 9 9 4

a simplified view, but it allows making a rigorous description from a reaction engineering point of view. Three time scales determine this system: - Mean hydraulic residence time: qh ¼ V/Q. - Turn over time for the vortex: qt ¼ 1/f ¼ V/R with f being the turn over frequency. - Reaction time scale: qr ¼ 1/k. Here, we assume to have a firstorder reaction with rate constant k. These time scales can be grouped in two dimensionless groups: - Dimensionless reaction rate: Damko¨hler number: DaI ¼ qh/qr ¼ k$qh. - Dimensionless frequency of the vortex: qh/qt ¼ f$qh. If reactive tracers are used to identify this macroscopic flow feature, the concentration of the reactive tracer is measured at different locations (e.g. cout, c1 and c2). From a mass balance for the tracer, the following relationships can be deduced: cout 1 ¼ c0 1 þ qqht ð1  expðk$qt ÞÞ
exp 14k$qt c1 ¼ c0 1 þ qqh ð1  expðk$qt ÞÞ t
exp 34k$qt c2 ¼ c0 1 þ qqh ð1  expðk$qt ÞÞ

(1)

t

Here we assumed that the two locations c1 and c2 are located within the vortex after 1/4 and 3/4 of the turn over time. The relative concentration difference between the two measurements c1 and c2 is:

Dc c1  c2 exp 14k$qt  exp 34k$qt ¼ ¼ c0 c0 1 þ qqh ð1  expðk$qt ÞÞ

(2)

t

Equations (1) and (2) can be expanded to the following relations giving the relative concentration as a function of the two dimensionless groups k$qh and f$qh:    1 k$q  3 k$qh h  exp  exp  $ Dc c1  c2 4 f $qh 4 f $qh    ¼ ¼ c0 c0 h 1 þ f $qh 1  exp  k$q f $qh cout 1    ¼ c0 h 1 þ f $qh 1  exp  k$q f $qh

Fig. 2 e Spatial concentration differences caused by a vortex flow as a function of the dimensionless reaction rate and the dimensionless rotational frequency (solid lines). For comparison, the performance of the reactor is indicated by dashed lines.

(3)

(4)

We can graphically express Equations (3) and (4) by plotting the relative concentration as a function of the two dimensionless groups k$qh and f$qh. In Fig. 2 this has been done for different reaction rates from very slow to fast reactions and a set of three dimensionless frequencies covering a typical range (solid lines). For slow and very fast reactions, the measurable concentration difference gets small since the tracer is either not degraded because of poor performance (slow reactions, conservative tracer) or since it is fully degraded (fast reactions) throughout the reactor. For both cases, no information about the vortex can be extracted from such measurements. However, there is an intermediate range

of reaction rates (0.1 < k$qh < 10) where the difference in concentration is increased and where it reacts sensitive to the rotation frequency of the vortex. In other words, the reaction rate constant of a suitable reactive tracer has to lie in this favourable range. If this is the case, the flow structure can be identified. If the reaction rate is significantly lower or higher, the flow structure is not sensitive for the reactive tracer distribution. Hence, it cannot be identified. For performance measures like the concentration in the outflow cout, the situation is partly different. This measure is indicated in Fig. 2 by dashed lines. It is sensitive to the rotation frequency for high reaction rates (k$qh > 10). This is equivalent to the fact that hydraulics generally gets more decisive the higher the reactors’ performance is. For slow reactions, performance measures are also not suited, as they react not sensitive to variations in the hydraulics.

2.2. Unsteady flow structure: periodic unsteadiness of the flow rate The second structure that is analyzed is an unsteady flow situation. It exemplarily shows how transient flow phenomena can be identified with reactive tracers. The flow situation is characterized by a periodic variation of the flow rate. This flow situation occurs in reactors with periodic inflow like for reactors with a daily or hourly load pattern but it may also occur within a reactor when flow instabilities occur. The latter will be the case in the accompanying example of a wastewater aeration tank. As a result of a periodic unsteadiness of the inflow a reactive tracer with a constant inflow concentration will exhibit concentration fluctuations in the system and also in the outflow. The amplitude of the concentration fluctuations is influenced by

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 8 4 e1 9 9 4

1987

Fig. 3 e Scheme for a reactor with variable inflow (harmonic oscillation with frequency f ). As a result of the varying flow rate the outflow concentration fluctuates.

reaction and dispersion in the reactor. Fig. 3 schematically shows a reactor that exhibits this flow feature and introduces the notation for the generalization of this flow situation. Assuming the flow enters a fully mixed compartment, the following differential equation describes the behaviour of a reactive tracer in the system:   dC QðtÞ QðtÞ ¼ $C0  k þ $C dt V V

(5)

This is written for a reactive tracer being degraded by a firstorder reaction with rate constant k. For a fully mixed reactor the outflow concentration is equal to the concentration in the system. We further assume the inflow to follow a harmonic oscillation of the form Q(t) ¼ Qm$(1þsin(2pft)). Again, the system can be described as a function of two dimensionless groups: Dimensionless reaction rate: Damko¨hler Number DaI ¼ k$V/Qm ¼ k$qh Dimensionless Frequency: f$V/Qm ¼ f$qh To calculate the mean hydraulic residence time qh, we used the mean flow rate Qm. There exists no closed form solution for Equation (5). It was therefore solved numerically and the results are reported using the two dimensionless groups k$qh and f$qh. Since the main interest is devoted to the amplitude of the concentration fluctuation, we present the results in the dimensionless functional form Amplitude/C0 ¼ g(k$qh, f$qh). Unlike the flow variation, the concentration is not following a harmonic oscillation. The amplitude of the concentration fluctuation is obtained from the main component of the Fourier transformation of the concentration time series. The choice of the tracer substance results in different reaction rates (x-axis in Fig. 4). In the slow reactions regime, the amplitude of concentration fluctuations gets small since the tracer is almost not degraded. In this regime, variable reaction times due to inflow variations only have minor effects on the concentration. For the extreme case of a conservative tracer, no fluctuations at all are found. On the opposite, in the fast reaction regime, the tracer substance is almost fully degraded. The amplitude of fluctuations is then limited by the mean performance (dashed line in Fig. 4) and diminishes for high reaction rates. For both extreme cases of very slow and very fast reactions, such kind of a flow feature could not be identified. However, there is an intermediate range of reaction rates (1 < k$qh < 10) where concentration fluctuations are strongly expressed. This is the range the reaction rate of a suitable reactive tracer has to lie in.

Fig. 4 e Amplitude of concentration fluctuations caused by varying flow rate as a function of the dimensionless reaction rate and the dimensionless frequency of the fluctuations (solid lines). For comparison, the performance of the reactor is indicated by a dashed line.

Beside the reaction rate of the substance, the frequency of the fluctuation is the other important factor for this example. In Fig. 4 frequencies in the range of the mean residence time (e.g. diurnal variations in the case of an aeration tank in activated sludge system) down to high frequency variations with time scales much lower than the mean residence time are depicted. The lower the frequency the more pronounced the concentration fluctuations get. This is due to the effects of dispersion in the system that acts as a filter for high frequency disturbances. Disturbances with time scales much lower than the mean hydraulic residence time are smoothed out.

3.

Application

In this section two examples that illustrate the use of the reactive tracer analysis are introduced. They both have been observed in an aeration tank of a wastewater treatment plant. First, the plant and the experimental methods applied are outlined and then, the flow features encountered are discussed.

3.1.

Plant description

An aeration tank of the wastewater treatment plant Werdho¨lzli, Zurich serves as an example for the complex flow features that evolve in such tanks. The aeration tank consists of an anoxic zone with a volume of 1400 m3 followed by an aerobic zone with a volume of 3500 m3. In the anoxic zone, two axial-mixers are installed that provide homogenization of the wateresludge suspension. This zone is separated from the aerated part by an under-flowed textile wall (Fig. 5B). There are no physical separations within the aerobic tank (Fig. 5A).

1988

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 8 4 e1 9 9 4

The oxygen is introduced by 1800 ceramic air diffusers, arranged in five equally sized sections with 420, 420, 360, 300 and 300 diffusers. From the second to the third aeration section, not only the number of diffusers changes but also the layout changes from a rather homogenous distribution to a more aligned one (Fig. 5B).

analyzed by ion chromatography for concentration above 2 mg N/L or by the ammonium cuvette tests LCK304 and LCK305 (Hach-Lange GmbH) for concentrations below 2 mg N/L. Calibration data were fitted to a reformulated form of the Nikolsky-equation: EE0 s

c ¼ 10

3.2.

Ammonium measurements

With ion-selective electrodes (ISE), it became feasible to measure ammonium inline with a high temporal resolution (Winkler et al., 2004). To get also a high spatial resolution, we simultaneously used several ion-selective electrodes (Nadler, Zuzwil-Switzerland) and placed them in different configurations in the aeration tank. The sensor locations for data that are shown in this paper are indicated in Fig. 5. Each configuration was measured for at least half a day (a multiple of the mean hydraulic residence time of the aeration tank being 2e4 h). Before each new configuration, all sensors were calibrated. Ammonium concentrations are often low (0e2 mg/L NH4-N) towards the back end of the tank. For these low ammonium concentrations, ion-selective electrodes often are not very accurate. To produce reliable results also in this low concentration range, high efforts were put in a multiple-point in-situ calibration procedure: All sensors were placed as close together as possible at the front end of the aerated zone (location P1.1) over a period of 3e4 h capturing the morning peak of ammonium and therefore also the whole measurement range. Grab samples were taken every 30 min, filtered at 0.45 mm and

7m

A

14m

20m

o

(6)

E is the measured electromotive force and c is the measured concentration. E0, s and o are fitting parameters. In addition to ammonium, temperature, pH and potassium were continuously measured at location P3. Variations in these parameters were generally slow during a measurement period. Their effects on the ammonium concentration were implicitly included through the in-situ calibration procedure applied. After calibration, the sensors were redistributed in the aeration tank according to the proposed spatial configuration. Response time of ammonium electrodes were checked regularly by standard addition tests in the field (ISO15839, 2003). They were in the range of 15e20 s for 63% of the final value and 35e40 s for 90% of the final value. Fig. 6 shows typical calibration curves for three of the ammonium sensors (A) and the concentration time series for the time span of calibration (B). After several months of operation, the calibration curves significantly differ in the parameter E0 although this parameter is characteristic for a particular ISE/reference pair. This is caused by variable ageing of the electrodes and shows that frequent calibration is necessary to achieve highly accurate measurements. For the time of calibration, the sensors perfectly followed each other indicating that the calibration procedure applied is adequate.

30m

40m

50m

Influent Nitrification (aerated)

Denitrification

Return sludge

B

P0

P1.1 P1.2

P2

P3

P4

P5

11 m Ptop

Opening anoxic/oxic zone

Spatially overlapping velocity measurements

Aeration pattern for section 1 and 2 Aeration pattern for section 3, 4 and 5

4.2 m

Pbottom

3D Flow velocity measurement Ammonium measurement

Fig. 5 e Top view (A) and cross sectional view (B) of the aeration tank. The sensor positions for ammonium (cross) and flow velocity measurements (circle) are shown in both views. The two different aeration patterns are shown in the cross sectional view.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 8 4 e1 9 9 4

1989

Fig. 6 e Calibration of the ammonium sensors (ion-selective electrodes). Subplot A shows three typical calibration curves for this type of sensor after several months of operation. Subplot B compares the concentration time series of the three sensors during the time span of calibration.

3.3.

Flow velocity measurements

Acoustic Doppler Velocimeters (ADV sensors of type Nortek 10 MHz) were used to directly measure 3D flow velocities at different locations in the aeration tank. Velocities are measured based on the Doppler shift between the transmitted acoustic signal and the detected signal after scattering at particles moving with the fluid. They proved to give reasonable results also in a bubbly flow environment if gas holdup is as low as encountered in aeration tanks of wastewater treatments plants (Thiersch and Valentin, 2002). Nevertheless, due to inevitable scattering of the acoustic signal at bubbles having a different vertical flow velocity than the water, this flow component has to be examined with care. We simultaneously used four ADV sensors. The sensors were fixed in vertical direction on a movable aluminium rack system stabilized at the side walls of the aeration tank (Ortmanns and Minor, 2006). Within a cross section, flow velocities were measured at 30 locations (Fig. 5B). Flow velocities were recorded with a frequency of 25 Hz (with an internal sample frequency of 250 Hz). Recorded data were filtered with a despiking algorithm of Goring and Nikora (2002). In average, 5% of the data were removed. Since the focus of this study lay rather on the mean flow velocity or on slow variation of flow velocities, data were smoothed with a moving average window of 30 s. Based on preliminary ammonium measurements (Braun and Gujer, 2008), we expected to find flow velocity oscillation with a period of 6e7 min. We therefore used measurement periods of at least 20 min capturing three periods of the expected flow variation.

3.4.

Results

In this aeration tank, two different macroscopic flow structures were identified. They will be subsequently discussed and

their effects on the ammonium and flow velocity distribution will be shown. In the inlet zone of the aeration tank, a large stable vortex is built. It is caused by the inflow from the denitrification zone which occurs at the bottom of the tank. Such vortexes are a very common feature in the inflow section of reactors. Sometimes they can capture a whole reactor and lead to a disadvantageous residence time distribution as for example often encountered in unbaffled ozone contactors (Wols et al., 2010). This flow feature produces a characteristic spatial field of ammonium concentration: We could observe different concentrations at different depths in the front part of the reactor. Ammonium concentrations are higher at the bottom than at the water surface (Fig. 7A). This characteristic only appears at the first measurement location P1.1 which is located 7 m from the inlet and is not apparent in all other locations. Due to aeration, which is by far the dominant energy input, this vortex is confined to its rather small extent. The longitudinal flow velocity is shown for two distinct locations in this cross section (Fig. 7B). For the time of measurement, velocities are, apart from turbulent fluctuations which were smoothed out in this figure, rather constant over time. No specific temporal pattern can be observed. However, these measurements also give indication for the vortex, since longitudinal velocities at the bottom and at the top are in opposite direction. Fig. 7C finally shows a contour plot of a linear interpolation of all 30 mean velocities at this cross section. Although the velocities were recorded at different times, such an aggregation is still valid, since the flow rate though the reactor was kept constant and the velocities are fairly constant over time. Again, the vortex structure is clearly visible. The second flow feature is an instability of the flow formed at the transition of the two different aeration patterns (see Fig. 5B for details). Except for the inflow region, the front part of the aerated zone (the first two aerated sections) is well mixed as it is expected when the air diffusers are homogenously

1990

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 8 4 e1 9 9 4

Stationary vortex

A

Flow alternation

D

P1.2 top (14 m) P2 top (20 m) P3 top (30 m)

P0 C0

11

P1.1 top C2

10

P1.2 bottom 9

6

NH 4 -N concentration (mg/L)

NH 4 -N concentration (mg/L)

P1.1 bottom C1

P1.2 top

8 7 6

P4 top (40 m) P5 top (50 m)

4

2

5 4

0 22:00

02:00

04:00

14:00

Velocity (m/s)

10 8

0.5 0.0

-0.5 0

10 T ime (min)

5

15

E

20

C

15:00 1.5

P3a: NH4-N

P3a: Velocity

1.0

Velocity (m/s)

12

P1.1 top

P1.1 bottom

NH 4 -N (mg/L)

Time

B

14:45

14:30 Time

14:15

0.5

0.5

0.0

0

NH 4 -N (mg/L)

00:00

-0.5 0

10

20 T ime (min)

30

F P3a

P1.1 top

P1.1 bottom 30

20

10

0

-10 -20 -30

cm/s

30 20

10

0

-10

-20 -30

cm/s

Fig. 7 e Ammonium (A, B) and longitudinal flow velocity time series (B) in the front of the aerated zone. Refer to Fig. 5 for sensor locations. Subplot C shows a linear spatial interpolation of the mean longitudinal flow velocities at the 30 measurement locations for the 7 m cross section. Measurement locations are indicated by dots. Longitudinal profile of ammonium time series (D). Evolution of regular ammonia oscillation is shown. Simultaneous ammonium and longitudinal flow velocity time series (E) at 30 m from the inlet. Snapshot in time of spatially interpolated longitudinal flow velocities at the 30 m cross section (F). Black dots indicate measurement locations. The corresponding time is shown in subplot E as a dotted vertical line. distributed. But, at the transition of the aeration pattern, two countercurrent rotating cells with axes in longitudinal direction develop. As a result of the change in the flow pattern, a discontinuity in the air void fraction and consequently in the pressure distribution is formed. This finally drives the flow to periodically alternate between the two sides of the tank.

Again, this flow feature produces a characteristic spatial field of ammonium concentration: Ammonium shows an overall degradation along the average flow direction. However, at half length of the reactor, regular fluctuations in concentration with a period of about 6 min evolve (Fig. 7D). These fluctuations are then further transported towards the end of

1991

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 8 4 e1 9 9 4

P1.1bottom (7 m)

Bromide concentration (mg/L)

6.5 min 7 min

30

20

10 Tracer addition

0 11:30

Fig. 8 e Vortex flow structure in the aeration tank example. The circle depicts the situation of this example. Areas below the limit of detection are gray shaded. The second y-axis serves for illustration of concentration values: Effluent concentration (dashed lines) and concentration differences (solid lines) are shown on this axis which is only valid for this particular example.

the tank. Due to reaction and dispersion, these fluctuations get attenuated and finally almost disappear. The flow velocity measurements at the 30 m cross section clearly support the proposed nature of the flow phenomena. The velocity in main flow direction oscillates the same way as the ammonium concentration (Fig. 7E). When comparing the concentrations at both sides of the reactor at the 30 m cross section, it appears that these fluctuations are inversely phased (not shown here). In the centre of the reactor, the fluctuations in ammonium are almost extinguished. The signals of ammonium and flow velocities are both periodic and ammonium was continuously measured at the same location during all the flow velocity measurements. Therefore, the ammonium signal can be used to synchronize the subsequent flow velocity measurements. This finally allows visualizing the flow field at this cross section. Fig. 7F shows for the 30 m cross section a contour plot of linearly interpolated velocity measurements at a distinct time. It shows a state of the flow with strong forward flow at one side of the tank, backflow in the centre and stagnation at the other side of the tank.

Table 1 e Relevant numbers for the analysis of a large vortex flow structure in the aeration tank example. Q: Inflow 24 m3/min V: Volume of vortex zone 500 m3 (10 m  12 m  4.2 m) 21 min qh: Hydraulic residence time k: First-order rate constant 0.013 min1 k$qh: Dimensionless reaction rate 0.27 0.86 (10 mg/Le9.1 mg/L)/10.5 mg/L DC/C0: Measured signal (see Fig. 3A) f$qh: Resulting frequency 3.5 6 min qt: Resulting turn over time

11:40

11:50 Time

12:00

12:10

Fig. 9 e Tracer experiment with pulse addition of sodium bromide in the front part of the aeration tank. The tracer was added into the vortex zone (close to position P1.1). Tracer recirculation with a period of 6e7 min can be identified. After three turn over periods, all information is lost.

4.

Discussion

Two distinct macroscopic flow features could be identified in this aeration tank by spatially distributed ammonium measurements. These measurements will now be discussed with the help of the general reactive tracer analysis introduced in Section 2.

4.1.

Vortex flow: situation in the aeration tank

The actual situation for the example of the aeration tank is depicted in Fig. 8 by a circle. The relevant numbers are summarized in Table 1. For the analysis the situation during the first 2 h in Fig. 7A is taken. Inflow is taken from operational data. During the time of analysis, the flow rate and the inflow concentration are fairly stable. For this aeration tank, inflow variations are slow compared to the time scale of the vortex structure. This allows a stationary analysis to be made. For the estimation of the volume, the vortex zone is assumed to capture the first 10 m of the reactor. This is in agreement with the measurements that only show different concentrations over depths at the first measurement location (Fig. 7A). At high ammonium concentrations, nitrification follows a zero-order reaction. We measured a nitrification capacity of 8.2 g/m3/h by on-site batch tests. An approximate first-order rate constant with the same mean reaction rate is 0.76 h1. This value is valid for the high concentration range encountered in the inflow section of the reactor and is used here. Since ammonium concentrations are at this high level everywhere in the region of interest for this analysis, the first-order approximation with a mean first-order rate constant is valid. With these numbers applied, the vortex is expected to transport 3.5 times the mean flow. Consequently, it has a turn over time of approximately 6 min. This finding is supported by conservative tracer experiments with pulse addition of bromide into the vortex zone. Fig. 9 shows

1992

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 8 4 e1 9 9 4

A

B

Fig. 10 e Flow situation with varying flow rate for the aeration tank example. For subplot A, a fully mixed compartment is used. The dotted line is devoted to variable mean hydraulic residence times (variable reactor size). Reaction rate constant and frequency are fixed according to the numbers in Table 2. Filled circles show the result of the simplified model. Open circles depict the actual situation in the aeration tank. Areas below the limit of detection are gray shaded. The second y-axis serves for illustration of concentration values for this example. For subplot B, a more realistic model with a cascade of three reactors is used.

the results of such an experiment. The vortex periodically returns bromide with a period of 6e7 min. Due to dispersion, the initial tracer pulse gets quickly blurred. As a consequence, valuable information can only be extracted during the first two to three turn over periods after tracer addition. A restriction of the reactive tracer method is given by the uncertainty of the measurement device. This gives a lower limit for the detection of reactive tracers and also for detection of concentration differences between two locations. The measurement errors have a systematic and a random component. Typical values for the devices used here are a random error of 1% of the concentration value and a limit of detection of 0.05 mg/L. This information is also included in Fig. 8 as gray shaded areas. For this example, the measurement uncertainty is not crucial for vortex identification. However, for larger values of f$qh, which is related to highly mixed system, the situation is different. In agitated vessels for example, turn over frequencies are high and tracer substances are almost homogenously distributed. In such a case, this method is only suited if highly sensitive measurement devices are available.

Table 2 e Relevant numbers for the analysis of a flow alternation in the aeration tank example. Qm: Mean inflow V: Volume qh: Hydraulic residence time k: First-order rate constant f: Frequency k$qh: Dimensionless reaction rate f$qh: Dimensionless frequency A/C0: Measured signal (see Fig. 3C)

12 m3/min (50% of full flow) 85 m3 (P3), 250 m3 (P4), 420 m3 (P5) 7 min (P3), 21 min (P4), 35 min (P5) 0.13 min1 0.18 min1 0.9 (P3), 2.7 (P4), 4.5 (P5) 1.3 (P3), 3.7 (P4), 6.3 (P5) 0.10 (P3), 0.03 (P4), 0.009 (P5)

4.2.

Unsteady flow rate: situation in the aeration tank

The essential result of the flow instability is an alternating inflow from the mixed front part of the reactor in the subsequent left or right part of the reactor. Focusing on either the left or the right side of the reactor, the situation is described by a periodically varying inflow. Therefore, it can be analyzed with the help of Fig. 4. The actual situation for the example of the aeration tank is depicted in Fig. 10A. The analysis includes only the section of the tank that is influenced by this flow alternation. It captures the three measurement locations P3, P4, P5 that exhibit concentration fluctuations. Inflow varies with a period of 5.5 min from zero to the total inflow taken from operational data. The relevant reactor volume is estimated to one-third of the cross section according to Fig. 7F with a variable length starting at the 25 m cross section where the change in aeration pattern occurs to the last measurement location at the 50 m cross section. In the concentration range encountered in this part of the tank, nitrification follows a mixed order reaction (between zero- and first-order). We measured the nitrification capacity to 14.2 g/m3/h by on-site batch tests. This value is higher than in the previous example due to higher wastewater temperatures during this measurement campaign. An approximate first-order rate constant with the same mean reaction rate is 7.8 h1. This value, which is only valid in the concentration range from 3 to 0.5 mg/L, is used here. The relevant numbers are summarized in Table 2. With the simplified model as outlined by Fig. 3, the amplitude of concentration variations is calculated for different hydraulic residence times according to the different sampling locations along the reactor (dotted line and full circles in Fig. 10A). These values are compared with the measurements (open circles). In principle, the observed attenuation along the reactor and the approximate value of the amplitude is captured by the model.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 8 4 e1 9 9 4

However, the model predicts a slower attenuation of the fluctuations along the reactor. The reactor is regarded as fully mixed, which is certainly not true as there is a clear gradient in concentration along the reactor shown by the measurements. An improved model with a cascade of three reactors, each denoted to one measurement location, allows a more accurate description of the effects of dispersion. Fig. 10B shows the result with this improvement. The model results are now (with exception of an offset) in line with the measurements. Further improvements need to include more details of the flow. We neglected backflow that occurs not only in the centre but to some extent also at the side of the tank. Such backflows further amplify the variations that have been detected in this section of the aeration tank which explains an offset in the amplitude between the general model and the measurements. As the example serves primarily for illustration, a more detailed description is out of the scope of this work. Due to the uncertainty of the measurement device, fluctuations with small amplitudes may not be detectable. For the device used here, we can assume that amplitude of approximately 0.04 mg/L can still be detected. This corresponds to the situation at measurement location P5. In this context, it is important to notice that the response time of the ion-selective electrodes used here causes an additional attenuation. In this example, fluctuations have a time constant of 5.5 min whereas the time constant of the response time is one order of magnitude lower. Therefore, this effect can be neglected here.

5.

Conclusions

For distinct macroscopic flow features in water and wastewater treatment reactors, we generally discussed the use of a reactive tracer for their identification. The type of analysis is general and the methodology can also be applied for other flow situations. Based on our work we conclude: - The large scale flow field in water treatment reactors can be complex. In particular, spatial inhomogeneities in major energy sources (e.g. aeration) are important drivers for unexpected flow patterns. These patterns effect plant performance and can negatively interfere with plant control. - Anomalies in the flow field result in characteristic spatial or temporal distributions of reactive tracers. Therefore, reactive tracer measurements with sufficient spatial or temporal resolution allow characterizing these large scale flow structures. In a full scale aeration tank, two distinct large scale flow structures were identified by reactive tracer measurements. - Ammonium is the tracer substance of choice in a nitrifying activated sludge system. Generally, the substance that served as a basis for the design of a system is appropriate and allows the detection of flow features that are relevant for the process. - For the detection of large dynamic flow structures like flow instabilities at least one sensor with a high temporal resolution (with respect to the flow structure) is needed. For stationary flow structures like large vortices or shortcuts at least two sensors are needed. In this case, temporal resolution is of less importance.

1993

- When using ammonium as a tracer substance, ion-selective electrodes have a sufficient accuracy and an adequate temporal resolution.

Acknowledgments We thank the operational staff of the WWTP Werdho¨lzli and Fabienne Steiner, Raphael Bru¨gger and Luzia Sturzenegger for their help with the experiments. We acknowledge the Laboratory of Hydraulics, Hydrology and Glaciology at ETH Zu¨rich for the use of their acoustic Doppler velocimetry equipment and the support by Daniel Gubser. This work was supported by the Swiss National Science Foundation Grant 200021-113298, as well as by Eawag and ETH.

references

Armbruster, M., Krebs, P., Rodi, W., 2001. Numerical modelling of dynamic sludge blanket behaviour in secondary clarifiers. Water Science and Technology 43 (11), 173e180. Braun, D., Gujer, W., 2008. Reactive tracers reveal hydraulic and control instabilities in full-scale activated sludge plant. Water Science and Technology 57 (7), 1001e1007. Capela, I., Bile, M.J., Silva, F., Nadais, H., Prates, A., Arroja, L., 2009. Hydrodynamic behaviour of a full-scale anaerobic contact reactor using residence time distribution technique. Journal of Chemical Technology and Biotechnology 84 (5), 716e724. Craig, K., De Traversay, C., Bowen, B., Essemiani, K., Levecq, C., Naylor, R., 2002. Hydraulic study and optimisation of water treatment processes using numerical simulation. Water Science and Technology: Water Supply 2 (5e6), 135e142. Do-Quang, Z., Cockx, A., Laıˆne, J., Roustan, M., 2001. Applying CFD modelling in order to enhance water treatment reactors efficiency: example of the ozonation process. Water Science and Technology: Water Supply 1 (4), 125e130. Goring, D.G., Nikora, V.I., 2002. Despiking acoustic Doppler velocimeter data. Journal of Hydraulic Engineering-Asce 128 (1), 117e126. Gresch, M., Braun, D., Gujer, W., 2010. The role of the flow pattern in wastewater aeration tanks. Water Science and Technology 61 (2), 407e414. He, C., Wood, J., Marsalek, J., Rochfort, Q., 2008. Using CFD modeling to improve the inlet hydraulics and performance of a storm-water clarifier. Journal of Environmental EngineeringAsce 134 (9), 722e730. ISO15839, 2003. Water Quality e On-line Sensors/Analysing Equipment for Water e Specifications and Performance Tests. Switzerland, Geneva. Jakobsen, H.A., Lindborg, H., Dorao, C.A., 2005. Modeling of bubble column reactors: progress and limitations. Industrial & Engineering Chemistry Research 44 (14), 5107e5151. Kjellstrand, R., Mattsson, A., Niklasson, C., Taherzadeh, M.J., 2005. Short circuiting in a denitrifying activated sludge tank. Water Science and Technology 52 (10e11), 79e87. Levenspiel, O., 1999. Chemical Reaction Engineering. Wiley, New York, etc. Newell, B., Bailey, J., Islam, A., Hopkins, L., Lant, P., 1998. Characterising bioreactor mixing with residence time distribution (RTD) tests. Water Science and Technology 37 (12), 43e47. Orhon, D., Soybay, S., Tunay, O., Artan, N., 1989. The effect of reactor hydraulics on the performance of activated-sludge

1994

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 8 4 e1 9 9 4

systems. I. The traditional modeling approach. Water Research 23 (12), 1511e1518. Ortmanns, C., Minor, H.-E., 2006. Entsander von Wasserkraftanlagen (Desilting Chambers of Hydro Power Plants). Versuchsanstalt fu¨r Wasserbau Hydrologie und Glaziologie ETH-Zentrum, Zu¨rich. in German. Saalbach, J., Hunze, M., 2008. Flow structures in MBR-tanks. Water Science and Technology 57 (5), 699e705. Tabib, M.V., Roy, S.A., Joshi, J.B., 2008. CFD simulation of bubble column e an analysis of interphase forces and turbulence models. Chemical Engineering Journal 139 (3), 589e614. Templeton, M.R., Hofmann, R., Andrews, R.C., 2006. Case study comparisons of computational fluid dynamics (CFD) modeling versus tracer testing for determining clearwell residence

times in drinking water treatment. Journal of Environmental Engineering and Science 5 (6), 529e536. Thiersch, B., Valentin, F., 2002. Stro¨mungstrukturen in Belebungsbecken und Einfluss auf den Sauerstoffeintrag (Hydrodynamic Structures of Aeration Tanks and their Influence on Aeration Efficiencies). Wasserwirtschaft 2002 (1e2), 34e38. in German. Winkler, S., Rieger, L., Saracevic, E., Pressl, A., Gruber, G., 2004. Application of ion-sensitive sensors in water quality monitoring. Water Science and Technology 50 (11), 105e114. Wols, B.A., Hofman, J.A.M.H., Uijttewaal, W.S.J., Rietveld, L.C., van Dijk, J.C., 2010. Evaluation of different disinfection calculation methods using CFD. Environmental Modelling and Software 25 (4), 573e582.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 9 5 e2 0 0 1

Available at www.sciencedirect.com

journal homepage: www.elsevier.com/locate/watres

Effect of natural organic matter on toxicity and reactivity of nano-scale zero-valent iron Jiawei Chen a,b,*, Zongming Xiu b, Gregory V. Lowry c,d, Pedro J.J. Alvarez b,* a

State Key Laboratory of Geological Processes and Mineral Resources, China University of Geosciences, Beijing 100083, China Dept. of Civil & Environmental Engrg., Rice University, Houston, TX 77005, USA c Dept. of Civil & Environmental Engrg., Carnegie Mellon University, Pittsburgh, PA 15213, USA d Center for Environmental Implications of Nanotechnology (CEINT), USA b

article info

abstract

Article history:

Nano-scale zero-valent iron (NZVI) particles are increasingly used to remediate aquifers

Received 28 July 2010

contaminated with hazardous oxidized pollutants such as trichloroethylene (TCE).

Received in revised form

However, the high reduction potential of NZVI can result in toxicity to indigenous bacteria

23 November 2010

and hinder their participation in the cleanup process. Here, we report on the mitigation of

Accepted 24 November 2010

the bactericidal activity of NZVI towards gram-negative Escherichia coli and gram-positive

Available online 3 December 2010

Bacillus subtilis in the presence of Suwannee River humic acids (SRHA), which were used as

Keywords:

E. coli in aerobic bicarbonate-buffered medium. SRHA (10 mg/L) significantly mitigated

Natural organic matter

toxicity, and survival rates after 4 h exposure increased to similar levels observed for

a model for natural organic matter (NOM). B. subtilis was more tolerant to NZVI (1 g/L) than

Electrosteric hindrance

controls not exposed to NZVI. TEM images showed that the surface of NZVI and E. coli

NZVI

was surrounded by a visible floccus. This decreased the zeta potential of NZVI from
30

Toxicity

to
45 mV and apparently exerted electrosteric hindrance to minimize direct contact with

Reactivity

bacteria, which mitigated toxicity. H2 production during anaerobic NZVI corrosion was not significantly hindered by SRHA ( p > 0.05), However, NZVI reactivity towards TCE (20 mg/L), assessed by the first-order dechlorination rate coefficient, decreased by 23%. Overall, these results suggest that the presence of NOM offers a tradeoff for NZVI-based remediation, with higher potential for concurrent or sequential bioremediation at the expense of partially inhibited abiotic reactivity with the target contaminant (TCE). ª 2010 Elsevier Ltd. All rights reserved.

1.

Introduction

Nano-scale zero-valent iron (NZVI) is a valuable material for in-situ remediation of aquifers contaminated with oxidized priority pollutants (Klimkova et al., 2008; Zhang, 2003), such as chlorinated solvents (Liu et al., 2005; Song and Carraway, 2005), hexavalent chromium (Xu and Zhao, 2007), and nitrate (Choe et al., 2000; Yang and Lee, 2005). However, as a strong reductant (Eh ¼
440 mV) with a relatively high specific

surface area, NZVI might be toxic to indigenous bacteria and hinder their participation in the cleanup process (Diao and Yao, 2009; Xiu et al., 2010b). For example, NZVI had a bactericidal effect on Escherichia coli (Lee et al., 2008; Li et al., 2010), which was not observed with other types of iron including maghemite nanoparticles, micro-scale ZVI, and Fe3þ ions (Auffan et al., 2008; Lee et al., 2008). When NZVI is injected into the subsurface for groundwater remediation (usually as a slurry at concentrations from 1.9 g/L

* Corresponding authors. Dept. of Civil & Environmental Engrg., Rice University, Houston, TX 77005, USA. Tel.: þ1 713 348 5903. E-mail addresses: [email protected] (J. Chen), [email protected] (P.J.J. Alvarez). 0043-1354/$ e see front matter ª 2010 Elsevier Ltd. All rights reserved. doi:10.1016/j.watres.2010.11.036

1996

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 9 5 e2 0 0 1

(Zhang, 2003) to 10 g/L (Henn and Waddill, 2006)), it may interact with dissolved natural organic matter (NOM), which is one of the most abundant materials on earth and exists in natural waters in the range of a few mg/L to a few hundred mg/L C (Wall and Choppin, 2003). NOM usually includes a skeleton of alkyl and aromatic units with functional groups such as carboxylic acids, phenolics, hydroxyls, and quinones. Accordingly, NOM is known to readily adsorb to colloidal particles and bind with NZVI to enhance its mobility in water or porous media (Johnson et al., 2009). NZVI coated with humic acid showed minor toxicity to E. coli (Li et al., 2010). However, the effect of NZVI in the presence or absence of NOM on gram-positive bacteria (such as Bacillus subtilis, which is ubiquitous in the environment) has not yet been addressed in the literature. Furthermore, NOM is known to shuttle electrons for iron reducing bacteria, and could also transfer electrons in abiotic reactions. Therefore, it is important to understand how NOM affects NZVI’s performance and its toxicity to bacteria that might participate in the cleanup process. Although it is known that some nanoparticles can affect biological systems at cellular, sub-cellular and protein levels (Farre et al., 2009; Klaine et al., 2008), the mechanism of NZVI toxicity towards bacteria remains unclear. A variety of toxicity mechanisms have been proposed (Nel et al., 2009), including disruption of the cell membrane integrity (Fang et al., 2007), interference with respiration (Lyon et al., 2008), and damage of DNA or enzymatic proteins caused by released metal ions from NPs (Gogoi et al., 2006). Li et al. (2010) found that humic acid associated with NZVI decreased NZVI’s toxicity to E. coli. They proposed that electrosteric repulsions afforded by the adsorbed NOM decreased NZVI adhesion to E. coli which decreased its toxicity. Aside from this study with only one type of organism, little quantitative information has been published on how NOM affects the reactivity, corrosion and toxicity of NZVI. This paper addresses the effect of Suwannee River humic acids (SRHA), a commonly used model for NOM, on the bactericidal properties of NZVI to bacteria with different cell wall composition and morphology (i.e., the gram-negative bacterium, E. coli and the gram-positive bacterium, B. subtilis). Solution pH was controlled and zeta potential was measured along with TEM imaging to investigate how NOM affects NZVI toxicity. Abiotic reactivity was also assessed by quantifying TCE dechlorination kinetics and cathodic H2 production (with known biostimulatory potential (Oh et al., 2001; Till et al., 1998; Weathers et al., 1997)) to assess potential tradeoffs associated with NOMeNZVI interactions.

prepared in an anoxic chamber. Suwannee River humic acid (SRHA II, International Humic Substances Society, Atlanta, Georgia) was used as model NOM without further pretreatment. Stock solutions were prepared by dissolving 1 g/L SRHA in sterile deionized water. A stock solution was freshly prepared for 1 month use and kept at 4  C in the dark. TCE (99%) was purchased from SigmaeAldrich (St. Louis, MO) and NaHCO3 was obtained from Fisher Scientific (Fair Lawn, NJ). Ultra-high purity gases (helium, nitrogen, hydrogen, etc.) were purchased from Matheson Tri-gas (Houston, TX). All chemicals used were reagent grade or better unless otherwise specified.

2.2.

The gram-negative E. coli (ATTC strain 10798) and the grampositive B. subtilis CB310 (courtesy of Dr. Charles Stewart, Rice University, Houston, TX) were used as model bacteria in inhibition experiments (Adams et al., 2006). E. coli (or B. subtilis) was inoculated in 30 mL of LB Broth medium (Difco Co.) and grown at 37  C overnight. The bacteria were harvested by centrifugation at 4000g for 5 min, washed three times with 2 mM bicarbonate solution (pH 8.10), and re-suspended in 30 mL bicarbonate buffer to make the bacteria stock solution.

2.3.

Materials and methods

2.1.

Chemicals

NZVI particles were obtained from Toda Kogyo Corporation, Onoda, Japan. According to the TEM images, the NZVI consisted of irregularly shaped particles ranging in size from 5 to 100 nm with a median radius of about 50 nm. The Fe0 content in these NZVI particles was about 40%. A stock solution of 10 g/ L NZVI containing 2 mM sodium bicarbonate (NaHCO3) was

Exposure of E. coli and B. subtilis to NZVI

NZVI (10 g/L) was ultrasonicated for 1 min and added to 20 mL bacteria suspension (2e3  106 colony forming units (CFU)/ mL) to make a final NZVI solution of 1 g/L under aerobic conditions. SRHA was added to achieve a final concentration of 10 mg/L. Sodium bicarbonate was chosen as a natural buffer. All reactors were shaken on a Labnet Orbit Shaker (Labnet, USA) at 200 rpm at 22  C. Bacterial controls were prepared similarly but without NZVI or SRHA. Exposure experiments were conducted in triplicate: (i) E. coli or B. subtilis with 1 g/L NZVI; (ii) E. coli or B. subtilis with 10 mg/L SRHA; and (iii) E. coli or B. subtilis with 1 g/L NZVI plus 10 mg/L SRHA.

2.4.

Toxicity assessment

Viable bacterial concentrations were determined after 0, 1, 2, and 4 h exposure by the spread plate method (Adams et al., 2006). Briefly, triplicate samples were plated on LB agar plates, incubated at 37  C for 12 h, and the colony forming units (CFU) were counted. The bacteria survival was expressed as N/N0 (%), where N and N0 are the remaining and initial numbers of live bacteria (CFU/mL), respectively.

2.5.

2.

Bacteria

TCE degradation and cathodic H2 generation

The effect of SRHA on NZVI reactivity towards TCE and cathodic H2 generation in the absence of TCE were assessed separately in 250-mL serum bottles containing 100 mL bicarbonate buffer (2 mM). Two sets of experiments were conducted simultaneously in triplicate for TCE degradation: (i) TCE (20 mg/L, 15.2 mmol) with 1 g/L NZVI alone; and (ii) TCE with 1 g/L NZVI and 10 mg/L SRHA. Buffer solutions were sparged with ultra-high purity nitrogen gas for 20 min prior to NZVI and SRHA addition, and shaken at 200 rpm at 22  C

1997

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 9 5 e2 0 0 1

throughout the experiment. A negative control without NZVI and SRHA was run to discern volatilization and adsorption losses. The concentration of TCE and its byproducts were measured by GC-FID over time as described below. Cathodic H2 generation tests were set up similarly without the addition of TCE.

2.6.

Transmission electron microscopy (TEM)

To examine the interaction of SRHA with E. coli and NZVI in bicarbonate buffer, NZVI particles and E. coli were added separately to 10 mg/L SRHA and shaken at 200 rpm at 22  C for 1 h. A 5 mL sample of the suspension was deposited on a 400mesh copper grid (Ultrathin carbon type-A, Ted Pella Inc., Redding, CA) and air-dried. Imaging was performed by TEM using a JEOL 1230 (JEOL, Tokyo, Japan) operated at 120 kV (SEA, Shared Equipment Authority, Rice University, Houston, TX).

2.8.

3.

Results and discussion

3.1.

Bacteria exposure to NZVI

are

reported

as

Analytical methods

The pH and oxidationereduction potential (ORP) of reacting solutions were measured with a Basic pH Meter (Denver Instruments, CO, USA). Zeta potentials of bacteria and NZVI were measured using a Zeta PALS particle analyzer (Brookhaven Instruments, Holtsville, NY). Measurements were performed in 2 mM bicarbonate solution at pH 8.10 in the presence and absence of 10 mg/L SRHA. The temperature was maintained at 22  C for all measurements. Headspace samples (100 mL) were withdrawn from each reactor to analyze for TCE, using a GC (HP5890, MN, USA) equipped with a flame ionized detector (FID). Separation was achieved with a packed column (6 ft.  1/8 in o. d. 60/80 carbopack B/1% SP-1000, Supelco). Analysis of H2 was conducted by direct injection of headspace samples (200 mL) into a GC (HP 6890) equipped with a thermal conductivity detector (TCD) and a packed column (H9-Q 60/80 9 ft. 2.0 mm ID 1/8 in OD, ResTek, PA, USA). The detailed analytical procedures for TCE and H2 analysis were described previously (Xiu et al., 2010b).

2.7.

confidence level. All measurements mean  one standard deviation.

Whether differences between treatments were statistically significant was determined using Student’s t-test at the 95%

Effect on pH, ORP and zeta potential

NZVI (Fe0) anaerobic corrosion in water can lead to an increase in pH:

B

100 80 60 40 20

(1)

However, the pH of the solution (1 g/L NZVI with/without SRHA) remained stable (at pH 8.10  0.10) in the 2 mM bicarbonate buffer, which is within the favorable pH domain for

B. subtilis Survival (%)

E. coli Survival (%)

3.2.

Fe0 þ 2H2 O/Fe2þ þ H2 þ 2OH


Statistical analyses

A

Although NZVI was ultrasonicated to disperse the particles, significant aggregation occurred (especially in the absence of SHRA). TEM images showed that the size of the aggregates ranged from about 50 nm to 5 mm in the presence of SRHA (Supporting Information [SI], Fig. 1A), and were larger than 5 mm without SRHA ([SI], Fig. 1B). B. subtilis was more tolerant to NZVI than E. coli (80.2  6.5% vs. 35.9  5.5% survival after 1 h incubation). The higher negative charge of B. subtilis compared to E. coli (
32.5 versus
26.9 mV) probably contributed to higher electrostatic repulsion and reduced toxicity. Whether its thicker gram-positive cell wall (composed of a relatively thick (20e80 nm) peptidoglycan layer (Madigan et al., 2006) confers additional protection remains to be determined. SRHA (10 mg/L) significantly mitigated NZVI toxicity, increasing bacterial survival to similar levels as unexposed controls after 1 h incubation (96.3  3.9% vs. 97.9  1.8% for E. coli (Fig. 3A), and 85.1  4.4% vs. 89.1  6.4% for B. subtilis (Fig. 3B)). The positive effect of SRHA was more pronounced after 4 h exposure, increasing E. coli survival from 3.7  0.5% to 84.2  3.0% and B. subtilis survival from 12.4  2.6% to 69.8  3.0%. The type and concentration of humic acids and other types of NOM can vary widely in natural systems, and accordingly mitigate NZVI toxicity to different extents than observed in these experiments.

100 80 60 40 20 0

0 0

1

2

3

Contact Time (h)

4

0

1

2 3 Contact Time (h)

4

Fig. 1 e Mitigation of NZVI toxicity to (A) E. coli and (B) B. subtilis by SRHA in 2 mM bicarbonate medium, pH [ 8.10 (Symbols: ,, control; C, 1 g/L NZVI; Δ, 10 mg/L SRHA; ;, 1 g/L NZVI D 10 mg/L SRHA).

1998

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 9 5 e2 0 0 1

Fig. 2 e Enhanced NZVI dispersion by SRHA after shaking for 1 h at 200 rpm (A) NZVI (1 g/L) alone; (B) SRHA (10 mg/L) alone; (C) NZVI (1 g/L) in the presence of SRHA (10 mg/L).

Fig. 3 e TEM images of NZVI and E. coli in the presence or absence of SRHA. (A) NZVI; (B) NZVI (1 g/L) in the presence of SRHA (10 mg/L); (C) E. coli; (D) E. coli in the presence of SRHA (10 mg/L).

1999

20

TCE Concentration (mg/L)

both E. coli and B. subtilis. The solution ORP also remained constant (around
280 mV), indicating that changes in pH and ORP were not responsible for bacterial inactivation. Bare NZVI partly aggregated and precipitated to the bottom of the reactor after 1 h incubation. In contrast, NZVI remained stable in the presence of SRHA (Fig. 2), indicating that SRHA associated with NZVI and changed its surface properties. This was corroborated by TEM observation (Fig. 3A,B). SRHA decreased the zeta potential of NZVI from around
30 to
45 mV (Fig. 4). The zeta potential of bacterial cells also decreased in the presence of SRHA, from
26.9  2.6 to
35.7  1.7 mV for E. coli, and from
32.5  1.2 to
45.6  2.1 mV for B. subtilis after 1 h exposure, due to adsorption of SRHA onto bacterial cells (Fig. 3C,D).

T C E ( m g /L )

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 9 5 e2 0 0 1

15

20 19 0

10

1

2 3 Time (h)

4

1 g/L NZVI 1 g/L NZVI + 10 mg/L SRHA

5

0 0

3.3.

21

50

100

Reactivity of NZVI in the presence of SRHA

150

200

250

300

Time (h)

Reductive dechlorination of TCE by NZVI followed first-order kinetics, and the degradation rate constant (kobs
NZVI) decreased by 23% (from 0.0178  0.0007 h
1 (R2 ¼ 0.99) to 0.0137  0.0004 h
1 (R2 ¼ 0.99)) in the presence of SRHA (Fig. 5). This was likely due to occlusion of NZVI reactive sites (Redman et al., 2002). Note that SRHA did not significantly decrease the reactivity of NZVI with TCE during the first 4 h ( p > 0.05; Fig. 5 insert), which indicates that toxicity mitigation over the 4 h exposure period was not due to a decrease in NZVI reactivity. The generation of cathodic H2, which is an indirect measure of NZVI corrosion and is also of interest due to its biostimulation potential (Till et al., 1998; Weathers et al., 1997) was also considered. H2 production was not hindered by the presence of SRHA (Fig. 6), which actually resulted in a slight, statistically indiscernible enhancement. SRHA association with the surface of NZVI would reduce its direct contact with water, which is conducive to slower generation of H2. However, this may be offset by SRHA-enhanced NZVI dispersion (Fig. 2), which is conducive to faster corrosion and H2 generation.

Fig. 5 e TCE degradation by NZVI in the presence and absence of SRHA. Degradation was first order with respect to TCE. Insert shows no effect of SRHA during the first 4 h.

3.4.

Mechanistic implications about toxicity

The mechanism by which SRHA mitigates NZVI toxicity to bacteria cannot be fully discerned because the bactericidal mechanism has not yet been elucidated (Nel et al., 2009). Nevertheless, direct contact between NZVI and cells appears to play a critical role (Auffan et al., 2008; Lee et al., 2008; Xiu et al., 2010a). In this work, SRHA hindered direct contact by coating both NZVI and bacteria (Fig. 3). Although association with SRHA increased the (negative) surface charge of E. coli and NZVI (Fig. 4), both were also negatively charged in the absence of SRHA. Thus, electrostatic repulsion alone is unlikely to be the principal mechanism for NZVI toxicity mitigation. Partial or complete encapsulation of both NZVI and cells by a visible SRHA floccus apparently served as a physical barrier hindering direct contact. Accordingly, electrosteric repulsion was likely a greater hindrance to direct

60 40

H2 Production ( mol/bottle)

Zeta Potential (mV)

1 g/L NZVI 1 g/L NZVI + 10 mg/L SRHA

20 0 -20 -40 -60 -80

400

300

200

1 g/L NZVI 1 g/L NZVI + 10 mg/L SRHA

100

0 0

1

2

3

4

Contact Time (h) Fig. 4 e Zeta potential of NZVI in the presence or absence of SRHA in 2 mM bicarbonate solution.

0

50

100

150

200

Time (h) Fig. 6 e H2 production by NZVI (1 g/L) with/without SRHA (10 mg/L) in bicarbonate solution.

2000

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 9 5 e2 0 0 1

bacterial contact than electrostatic repulsion alone, as observed previously for nZVI coated with synthetic polymers (Li et al., 2010). Coating by NOM has been observed for other nanoparticles (Baalousha et al., 2008; Diegoli et al., 2008; Hyung et al., 2007; Li et al., 2010) with an accompanying decrease in toxicity (Li et al., 2008). Therefore, electrosteric hindrance is likely the main mechanism by which humic acid mitigates NZVI toxicity (Phenrat et al., 2008, 2010). This mechanism helps explain why adding g/L concentrations of NZVI to contaminated aquifers does not significantly decrease biomass in those systems (e.g., Kirschling et al., 2010).

4.

Conclusions

SRHA associated with both NZVI and cells and mitigated bacterial toxicity dramatically, likely due to the electrosteric hindrance of direct contact. Understanding the effect of NOM on cathodic H2 evolution and TCE degradation is important because these are competing processes that consume Fe(0). Unchanged H2 evolution with hindered TCE degradation in the presence of NOM suggests that greater amount of NZVI would be needed to treat a given TCE mass. However, H2 evolution may also offer (as a positive tradeoff) a biostimulatory effect conducive to a higher potential for concurrent or sequential bioremediation of the target contaminants. Thus, further research may be warranted to optimize such biogeochemical interactions in groundwater remediation schemes.

Acknowledgments This study was sponsored by the USEPA (R833326), the Chinese government (Program for New Century Excellent Talents in University NCET-07-0769; the Fundamental Research Funds for the Central Universities 2010ZD14; National Program of Control and Treatment of Water Pollution 2009ZX07424-002) and the China Scholarship Council for Dr. Jiawei Chen. We thank Dr. Qilin Li and Xiaolei Qu for providing SRHA and helping with TOC measurement, and Dr. Michal Wong and Youlun Fang for assistant on zeta potential measurement.

Appendix. Supplementary material Supplementary data related to this article can be found online at doi:10.1016/j.watres.2010.11.036.

references

Adams, L.K., Lyon, D.Y., Alvarez, P.J.J., 2006. Comparative ecotoxicity of nanoscale TiO2, SiO2, and ZnO water suspensions. Water Research 40, 3527e3532. Auffan, M., Achouak, W., Rose, J., Roncato, M.A., Chaneac, C., Waite, D.T., Masion, A., Woicik, J.C., Wiesner, M.R., Bottero, J.Y., 2008. Relation between the redox state of iron-based

nanoparticles and their cytotoxicity toward Escherichia coli. Environmental Science and Technology 42, 6730e6735. Baalousha, M., Manciulea, A., Cumberland, S., Kendall, K., Lead, J. R., 2008. Aggregation and surface properties of iron oxide nanoparticles: influence of pH and natural organic matter. Environmental Toxicology and Chemistry 27, 1875e1882. Choe, S., Chang, Y.Y., Hwang, K.Y., Khim, J., 2000. Kinetics of reductive denitrification by nanoscale zero-valent iron. Chemosphere 41, 1307e1311. Diao, M.H., Yao, M.S., 2009. Use of zero-valent iron nanoparticles in inactivating microbes. Water Research 43, 5243e5251. Diegoli, S., Manciulea, A.L., Begum, S., Jones, I.P., Lead, J.R., Preece, J.A., 2008. Interaction between manufactured gold nanoparticles and naturally occurring organic macromolecules. Science of the Total Environment 402, 51e61. Fang, J.S., Lyon, D.Y., Wiesner, M.R., Dong, J.P., Alvarez, P.J.J., 2007. Effect of a fullerene water suspension on bacterial phospholipids and membrane phase behavior. Environmental Science and Technology 41, 2636e2642. Farre, M., Gajda-Schrantz, K., Kantiani, L., Barcelo, D., 2009. Ecotoxicity and analysis of nanomaterials in the aquatic environment. Analytical and Bioanalytical Chemistry 393, 81e95. Gogoi, S.K., Gopinath, P., Paul, A., Ramesh, A., Ghosh, S.S., Chattopadhyay, A., 2006. Green fluorescent proteinexpressing Escherichia coli as a model system for investigating the antimicrobial activities of silver nanoparticles. Langmuir 22, 9322e9328. Henn, K.W., Waddill, D.W., 2006. Utilization of nanoscale zerovalent iron for source remediation e a case study. Remediation Journal 16, 57e77. Hyung, H., Fortner, J.D., Hughes, J.B., Kim, J.H., 2007. Natural organic matter stabilizes carbon nanotubes in the aqueous phase. Environmental Science and Technology 41, 179e184. Johnson, R.L., Johnson, G.O., Nurmi, J.T., Tratnyek, P.G., 2009. Natural organic matter enhanced mobility of nano zerovalent iron. Environmental Science and Technology 43, 5455e5460. Kirschling, T.L., Gregory, K.B., Minkley, E.G., Lowry, G.V., Tilton, R.D., 2010. Impact of nanoscale zero valent iron on geochemistry and microbial populations in trichloroethylene contaminated aquifer materials. Environmental Science and Technology 44, 3474e3480. Klaine, S.J., Alvarez, P.J.J., Batley, G.E., Fernandes, T.F., Handy, R.D., Lyon, D.Y., Mahendra, S., McLaughlin, M.J., Lead, J.R., 2008. Nanomaterials in the environment: behavior, fate, bioavailability, and effects. Environmental Toxicology and Chemistry 27, 1825e1851. Klimkova, S., Cernik, M., Lacinova, L., Nosek, J., 2008. Application of nanoscale zero-valent iron for groundwater remediation: laboratory and pilot experiments. Nano 3, 287e289. Lee, C., Kim, J.Y., Il Lee, W., Nelson, K.L., Yoon, J., Sedlak, D.L., 2008. Bactericidal effect of zero-valent iron nanoparticles on Escherichia coli. Environmental Science and Technology 42, 4927e4933. Li, D., Lyon, D.Y., Li, Q., Alvarez, P.J.J., 2008. Effect of soil sorption and aquatic natural organic matter on the antibacterial activity of a fullerene water suspension. Environmental Toxicology and Chemistry 27, 1888e1894. Li, Z., Greden, K., Alvarez, P.J.J., Gregory, K.B., Lowry, G.V., 2010. Adsorbed polymer and NOM limits adhesion and toxicity of nano scale zerovalent iron to E. coli. Environmental Science and Technology 44, 3462e3467. Liu, Y.Q., Majetich, S.A., Tilton, R.D., Sholl, D.S., Lowry, G.V., 2005. TCE dechlorination rates, pathways, and efficiency of nanoscale iron particles with different properties. Environmental Science and Technology 39, 1338e1345. Lyon, D.Y., Brunet, L., Hinkal, G.W., Wiesner, M.R., Alvarez, P.J.J., 2008. Antibacterial activity of fullerene water suspensions

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 1 9 9 5 e2 0 0 1

(nC60) is not due to ROS-mediated damage. Nano Letters 8, 1539e1543. Madigan, M.T., Martinko, J.M., Brock, T.D., 2006. Brock Biology of Microorganisms, eleventh ed. Pearson Prentice Hall, Upper Saddle River, NJ. Nel, A.E., Madler, L., Velegol, D., Xia, T., Hoek, E.M.V., Somasundaran, P., Klaessig, F., Castranova, V., Thompson, M., 2009. Understanding biophysicochemical interactions at the nano-bio interface. Nature Materials 8, 543e557. Oh, B.-T., Just, C.L., Alvarez, P.J.J., 2001. Hexahydro-1,3,5-Trinitro1,3,5-Triazine (RDX) mineralization by zero-valent iron and mixed anaerobic cultures. Environmental Science and Technology 35, 4341e4346. Phenrat, T., Saleh, N., Sirk, K., Kim, H.J., Tilton, R.D., Lowry, G.V., 2008. Stabilization of aqueous nanoscale zerovalent iron dispersions by anionic polyelectrolytes: adsorbed anionic polyelectrolyte layer properties and their effect on aggregation and sedimentation. Journal of Nanoparticle Research 10, 795e814. Phenrat, T., Song, J.E., Cisneros, C.M., Schoenfelder, D.P., Tilton, R.D., Lowry, G.V., 2010. Estimating attachment of nano- and submicrometer-particles coated with organic macromolecules in porous media: development of an empirical model. Environmental Science and Technology 44, 4531e4538. Redman, A.D., Macalady, D.L., Ahmann, D., 2002. Natural organic matter affects arsenic speciation and sorption onto hematite. Environmental Science and Technology 36, 2889e2896. Song, H., Carraway, E.R., 2005. Reduction of chlorinated ethanes by nanosized zero-valent iron: kinetics, pathways, and effects

2001

of reaction conditions. Environmental Science and Technology 39, 6237e6245. Till, B.A., Weathers, L.J., Alvarez, P.J.J., 1998. Fe(0)-supported autotrophic denitrification. Environmental Science and Technology 32, 634e639. Wall, N.A., Choppin, G.R., 2003. Humic acids coagulation: influence of divalent cations. Applied Geochemistry 18, 1573e1582. Weathers, L.J., Parkin, G.F., Alvarez, P.J.J., 1997. Utilization of cathodic hydrogen as electron donor for chloroform cometabolism by a mixed methanogenic culture. Environmental Science and Technology 31, 880e885. Xiu, Z.M., Gregory, K.B., Lowry, G.V., Alvarez, P.J.J., 2010a. Effect of bare and coated nanoscale zerovalent iron on tceA and vcrA gene expression in Dehalococcoides spp. Environmental Science and Technology 44, 7647e7651. Xiu, Z.M., Jin, Z.H., Li, T.L., Mahendra, S., Lowry, G.V., Alvarez, P.J.J., 2010b. Effects of nano-scale zero-valent iron particles on a mixed culture dechlorinating trichloroethylene. Bioresource Technology 101, 1141e1146. Xu, Y.H., Zhao, D.Y., 2007. Reductive immobilization of chromate in water and soil using stabilized iron nanoparticles. Water Research 41, 2101e2108. Yang, G.C.C., Lee, H.L., 2005. Chemical reduction of nitrate by nanosized iron: kinetics and pathways. Water Research 39, 884e894. Zhang, W.X., 2003. Nanoscale iron particles for environmental remediation: an overview. Journal of Nanoparticle Research 5, 323e332.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 0 2 e2 0 1 0

Available at www.sciencedirect.com

journal homepage: www.elsevier.com/locate/watres

Formation of assimilable organic carbon during oxidation of natural waters with ozone, chlorine dioxide, chlorine, permanganate, and ferrate Maaike K. Ramseier a,b, Andreas Peter a,1, Jacqueline Traber a, Urs von Gunten a,b,* a b

Eawag, Swiss Federal Institute of Aquatic Science and Technology, Ueberlandstrasse 133, Duebendorf CH-8600, Switzerland School of Architecture, Civil and Environmental Engineering, Ecole Polytechnique Fe´de´rale de Lausanne, CH-1015 Lausanne, Switzerland

article info

abstract

Article history:

Five oxidants, ozone, chlorine dioxide, chlorine, permanganate, and ferrate were studied

Received 12 July 2010

with regard to the formation of assimilable organic carbon (AOC) and oxalate in absence

Received in revised form

and presence of cyanobacteria in lake water matrices. Ozone and ferrate formed significant

26 November 2010

amounts of AOC, i.e. more than 100 mg/L AOC were formed with 4.6 mg/L ozone and ferrate

Accepted 1 December 2010

in water with 3.8 mg/L dissolved organic carbon. In the same water samples chlorine

Available online 9 December 2010

dioxide, chlorine, and permanganate produced no or only limited AOC. When cyanobacterial cells (Aphanizomenon gracile) were added to the water, an AOC increase was

Keywords:

detected with ozone, permanganate, and ferrate, probably due to cell lysis. This was

Assimilable organic carbon (AOC)

confirmed by the increase of extracellular geosmin, a substance found in the selected

Oxalate

cyanobacterial cells. AOC formation by chlorine and chlorine dioxide was not affected by

Oxidation

the presence of the cells. The formation of oxalate upon oxidation was found to be a linear

Ozone

function of the oxidant consumption for all five oxidants. The following molar yields were

Chlorine

measured in three different water matrices based on oxidant consumed: 2.4e4.4% for

Chlorine dioxide

ozone, 1.0e2.8% for chlorine dioxide and chlorine, 1.1e1.2% for ferrate, and 11e16% for

Permanganate

permanganate. Furthermore, oxalate was formed in similar concentrations as trihalomethanes during chlorination (yield w 1% based on chlorine consumed). Oxalate formation

Ferrate

kinetics and stoichiometry did not correspond to the AOC formation. Therefore, oxalate cannot be used as a surrogate for AOC formation during oxidative water treatment. ª 2010 Elsevier Ltd. All rights reserved.

1.

Introduction

Drinking water needs to be delivered to the tap in a hygienically impeccable state. To minimize bacterial regrowth in distribution systems a chemical disinfectant is frequently added, which can lead to the formation of undesired disinfection byproducts (Krasner et al., 2006; Richardson et al., 2007). Alternatively, the amount of assimilable organic carbon (AOC) in

the water can be reduced to avoid bacterial regrowth through nutrient limitation (van der Kooij, 1992). During drinking water treatment, primary disinfection with chemical oxidants can produce AOC by transforming macromolecular dissolved organic matter (DOM) into smaller molecules, which in turn can be taken up by bacteria more easily (van der Kooij et al., 1989; Hammes et al., 2006). Therefore, oxidative water treatment, e.g. ozonation, is generally followed by a biological

* Corresponding author. Eawag, Swiss Federal Institute of Aquatic Science and Technology, Ueberlandstrasse 133, Duebendorf CH-8600, Switzerland. Tel.: þ41 44 8235270; fax: þ41 44 8235210. E-mail address: [email protected] (U. von Gunten). 1 Present address: Kantonales Labor Zurich, Fehrenstrasse 15, Zurich CH-8032, Switzerland. 0043-1354/$ e see front matter ª 2010 Elsevier Ltd. All rights reserved. doi:10.1016/j.watres.2010.12.002

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 0 2 e2 0 1 0

filtration step to remove AOC (van der Kooij et al., 1989; Lykins et al., 1994; Vahala et al., 1998). To date many studies concerning AOC production have been performed in full-scale treatment systems under highly variable treatment conditions, which do not allow comparison of different oxidants (Liu et al., 2002; Volk and Le Chevallier, 2002; Polanska et al., 2005; Chen et al., 2007). In the present study five oxidantsdozone (O3), chlorine dioxide (ClO2), chlorine (Cl2 as 2 HOCl), permanganate (MnO 4 ), and ferrate (FeO4 )dwere investigated in the same water matrices with regard to their AOC formation potential and kinetics. Ozone, chlorine dioxide, and chlorine are widely applied in drinking water treatment for oxidation and disinfection purposes, permanganate is used for iron and manganese removal, and ferrate is discussed as a novel oxidant for micropollutant and phosphate removal mainly in wastewater treatment (Lee et al., 2009). These oxidants undergo various reactions with specific functional groups of DOM and as a consequence their stability in water differs significantly. Lee and von Gunten summarized the reactivity of ozone, chlorine dioxide, chlorine, and ferrate (Lee and von Gunten, 2010) and Waldemer and Tratnyek discussed permanganate reactions (Waldemer and Tratnyek, 2006). Briefly, of the five oxidants ozone is known as the most reactive, chlorine dioxide and ferrate mainly react with phenolic compounds, chlorine or hypochlorous acid reacts fast only with amines, and permanganate reacts only slowly with organic compounds, mainly with olefines. The objective of this study was a comparison of the selected oxidants concerning the extent of their AOC formation from oxidation of DOM in water in absence and presence of bacterial cells. Furthermore, oxalate formation during oxidative treatment was studied as a possible surrogate for AOC and for trihalomethane (THM) formation.

2.

Materials and methods

2.1.

Water samples

ozone generator (CMG 3-3, Apaco AG, Switzerland) through icecooled water. The ozone concentration of the stock solution was determined spectrophotometrically with 3(258 nm) ¼ 3000 M1 cm1 (Elovitz and von Gunten, 1999) and was 1.3e1.5 mM. A chlorine dioxide stock solution was prepared by reaction of chlorite with peroxodisulfate (Hoigne´ and Bader, 1994b; Huber et al., 2005). The concentration of the stock solution was determined spectrophotometrically with 3(358 nm) ¼ 1200 M1 cm1. Sodium hypochlorite (10% active chlorine Riedel-deHaen, Germany) was diluted in micropure water and standardized spectrophotometrically at pH 8.7 with 3(292 nm) ¼ 350 M1 cm1 (supporting information section of Lee et al., 2008). Potassium ferrate (K2FeO4) was prepared by the method of Thompson and co-workers (Thompson et al., 1951) and had a purity of 88% as Fe (VI) (w/w). Ferrate stock solutions were prepared from solid potassium ferrate in micropure water and used immediately. The stock concentration was measured spectrophotometrically, with 3(510 nm) ¼ 1150 M1 cm1 (Lee et al., 2005). A 5 mM permanganate stock solution was prepared in water from solid potassium permanganate and standardized spectrophotometrically with 3(525 nm) ¼ 2430 M1 cm1 (supporting information S2).

2.3. Measurement of oxidant consumption and determination of oxidant exposure Oxidant stock solutions were added to the stirred water samples in a 250 ml Schott bottle equipped with a dispenser system (Hoigne´ and Bader, 1994a). At various time intervals, the residual oxidant concentration was determined by dispensing aliquots of the sample into a plastic sampling tube containing a buffer and a quenching agent dye that changes color upon oxidation which allowed calculation of the residual oxidant concentration (supporting information S3). For all oxidants, the exposure (ct) was calculated as the integral of the transient oxidant concentration over the reaction time (von Gunten and Hoigne´, 1994).

2.4. Waters were sampled from three different surface waters. Lake Greifensee (LG) (Switzerland) is a eutrophic lake surrounded by agricultural land. Samples were taken directly at the surface of the lake, 30 m off shore in November 2008 (LGn) and January 2009 (LGj). Lake Zuerich (LZ) (Switzerland) is a mesotrophic lake and samples were taken in April 2009, 30 m below the lake surface around 600 m off shore at the intake pipe for drinking water production. Chriesbach (CB) (Switzerland) is a little creek fed to a considerable part by wastewater treatment plant effluent. Water was sampled in April 2009 at the surface. The pH was 8.0e8.3 for all water samples and the DOC concentration varied between 1.3 and 3.8 mg C/L (supporting information S1). The water samples were filtered through a 0.2 mm filter (regenerated cellulose, Sartorius AG, Goettingen, Germany, rinsed with 1 L of micropure water (NANOpure Diamond, Barnstead)) and stored at 4  C in the dark until use (maximum 10 days).

2.2.

Oxidant stock solutions

Ozone was applied as aqueous stock solution prepared by sparging an oxygeneozone gas mixture from an oxygen-fed

2003

Measurement of oxidation products

AOC was measured as described elsewhere (Hammes and Egli, 2005). Briefly, all samples were prepared in AOC-free glassware and the water was filtered prior to the experiment (see 2.1). After the experiment the water was partitioned into two AOC-free vials. Cells from a natural microbial community were added and the sample was incubated at 30  C for three days. The added natural microbial inoculum had been prepared by mixing different water samples (oxidized and untreated LZ, CB, and LG water) and had been stored at 4  C for several days to weeks. The final cell concentration after regrowth in the actual water sample was determined by staining the cells with SYBR Green and an addition of 5 mM ethylenediaminetetraacetate and counting the cells by flow cytometry on a PASIII flow cytometer (Partec, Germany). The AOC concentration was calculated assuming a bacterial growth of 107 cells per mg assimilable organic carbon (Hammes et al., 2006). The limit of detection was 10 mg/L and the standard error of triplicate incubated samples <10%. Oxalate was measured after sample filtration through a 0.45 mm nylon syringe filter (BGB Analytik AG, Switzerland)

2004

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 0 2 e2 0 1 0

using ion chromatography (Dionex ICS-3000) as described elsewhere (Hammes et al., 2006). A Dionex AS11HC column with an OH-gradient was used for separation and the target compound was detected by conductivity measurement with a limit of quantification of 0.1 mM and the standard deviation of three repetitions of a same sample (0.2 mM) was <10%. Trihalomethanes (CHCl3, CHCl2Br, CHClBr2, CHBr3) were measured by GC-ECD (GC 8000 and ECT 80/800, Fisons). 6 ml of the sample solution were heated at 80  C before 1 ml of the headspace was injected on a Restek RTX 624 column (Gallard and von Gunten, 2002). Detection limits were 0.1 mg/L and the standard deviation of multiple measurements of the standards used for calibration was <3%, except for chloroform, where it was <15%.

2.5.

Experimental procedure

20e50 ml of the filtered LGj water were poured into AOC-free Erlenmeyer flasks with three replicates for each designated reaction time. Oxidant stock solution was added to each flask and after the appropriate reaction time the oxidant was quenched and the samples were analyzed for AOC, oxalate, and THM. The oxidant doses were: 4.6 mg/L or 0.096 mM for O3, 3 mg/L or 0.042 mM for Cl2, 3 mg/L or 0.045 mM for ClO2, 3.2 mg Mn/L or 0.058 mM for MnO 4 , and 4.6 mg Fe/L or 0.083 mM for and the quencher was added in the following concenFeO2 4 trations: 0.5 mM nitrite in the case of ozone, and 0.3 mM of freshly prepared FeCl2 solution (pH around 1) in the case of chlorine dioxide and ferrate, and 0.2 mM in the case of chlorine and permanganate (see supporting information S4). To measure the production of oxalate upon oxidation of different water matrices, additional experiments with LZ and CB water (filtered through 0.45 mm cellulose nitrate membrane filter (Sartorius AG, Goettingen, Germany)) were preformed where the oxidants were quenched with nitrite or thiosulfate. All experiments were performed at 22  C (1  C).

3.

Results and discussion

3.1. AOC formation by ozone, chlorine dioxide, chlorine, permanganate, and ferrate The reactivity of the oxidants in the waters studied followed the order ozone >> ferrate > chlorine dioxide > chlorine >> permanganate (supporting information S5). Accordingly, the most significant increase of the AOC concentration is expected during ozonation. Previous investigations support this expectation: while ozone is known to produce high amounts of AOC (e.g. Hammes et al., 2006), chlorine was found to sometimes produce moderate amounts of AOC (e.g. Polanska et al., 2005). Less is known about the extent of AOC formation from chlorine dioxide, permanganate, or ferrate. Fig. 1 shows the AOC formation for various oxidants in LGj water. Significant concentrations of AOC are only formed during ozonation. An exposure of 1 mg*min/L was achieved after 16 s for an ozone dose of 4.6 mg/L (1.3 mg O3/mg DOC) and resulted in an AOC formation of 60 mg/L. The fast formation kinetics can be explained with the high reactivity of ozone towards several organic functional groups (Lee and von Gunten, 2010). Similar findings were reported earlier (Hammes et al., 2006) and could also be observed for the production of small organic acids during ozonation of phenol, a surrogate compound for DOM (Ramseier and von Gunten, 2009). Consequently, a significant AOC production cannot be avoided when ozone is applied for disinfection and a suitable technique to control this parameter after ozonation is indispensable. For chlorine dioxide, the AOC concentration remained constant within the standard deviation of triplicate experiments for the investigated water matrix. The negligible formation of AOC as well as the low initial consumption and

2.6. Oxidation experiments in presence of Aphanizomenon gracile The A. gracile strain SAG 31.79 was grown on BG-11 medium and the culture stock solution had an optical density of 0.47 measured at 685 nm. The bacterial cells were washed by centrifuging and decanting the supernatant twice, resuspended in LZ water and then spiked into 0.2 mm-filtered LZ water (see above) in AOC-free glassware to reach cell concentrations similar to natural conditions. Three oxidant doses (0.01, 0.02, and 0.05 mM) of each oxidant were added to LZ water with and without cells with reaction times of 24 h at 12  C. The lower temperature was chosen to avoid additional stress on the cells by an increased temperature. After 24 h, ozone and ferrate were consumed completely and the remaining chlorine, chlorine dioxide, and permanganate were quenched by addition of FeCl2. All samples were filtered through a 0.22 mm polyethersulfon membrane syringe filter (Millipore, Cork, Ireland) into two AOC-free vials and inoculated to measure AOC. Additionally, the geosmin concentration in the solution after the oxidation process was measured by solid phase micro extraction (SPME) coupled with GCeMS as described elsewhere (Peter et al., 2009).

Fig. 1 e AOC formation in LGj water upon treatment with different oxidants. Inset: same data plotted versus oxidant consumption. The slopes of the linear fits forced through the origin indicate the molar yield factor for AOC formation as the ratio of mM AOC formed over mM oxidant consumed: ozone L0.16, chlorine dioxide L0.06, chlorine L0.07, and permanganate L0.27.

2005

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 0 2 e2 0 1 0

the slow kinetics of consumption of chlorine dioxide (supporting information S5) are remarkable since from known rate constants a higher reactivity of chlorine dioxide would have been expected (Hoigne´ and Bader, 1994b). Reasons for this unexpected result have not been confirmed but can be hypothesized: the results might indicate that the content of activated aromatic systems (e.g. phenols) in the water was relatively low and hence only few compounds reacted with chlorine dioxide under formation of AOC (supporting information S5). Furthermore, mechanistic considerations might explain the low AOC formation in a second way: for phenol oxidation with chlorine dioxide an electron transfer reaction followed by dimerization can be expected. This may partially increase the molecular weight of the products possibly leading to a decrease of the bioavailability. Benzoquinone is another major reaction product of chlorine dioxide with phenol and it has been shown to inhibit bacterial growth. A study where biodegradable dissolved organic carbon (BDOC) was measured after application of chlorine dioxide to natural water samples confirms the low formation of bioavailable compounds (Swietlik et al., 2009). Fig. 1 shows a slight decrease of the AOC concentration in LGj water after chlorination. The low chlorine consumption in the present water matrix (supporting information S5) certainly contributes to the absence of AOC formation. Nevertheless, this result contradicts some former studies that observed an AOC increase (e.g. 31 mg/L (Polanska et al., 2005), 100 mg/L (Miettinen et al., 1998)). However, in another study, chlorination of six different water samples led to an increase of the BDOC concentration in only three waters, in one sample no change was observed, and in two water samples the BDOC concentration decreased (Swietlik et al., 2009). Chlorination of organic compounds typically renders them less bioavailable, since halogenation of organic compounds is often implicated as a reason for microbial persistance (Rasul Chaudhry and Chapalamadugu, 1991; Dercova´ et al., 2007). However, additional experiments showed that chlorination (0.1 mg/L Cl2) of preozonated water decreased the amount of AOC by 30% (data not shown). A similar observation was reported in an earlier study when measuring AOC using strain P17 (Le Chevallier et al., 1992). Considering the low reactivity of chlorine with the oxidized state of the organic compounds after ozonation, it seems unlikely that chlorine reacts with large parts of AOC present after ozonation. Therefore, the loss of AOC is probably not attributed to chlorination of former AOC compounds.

AOC (µg C/L)

O3

ClO2

Furthermore, oxalate, a non-chlorinated compound that is known to contribute to AOC (van der Kooij et al., 1989; Hammes et al., 2006), was consumed less efficiently in chlorinated than in ozonated water during AOC measurements (supporting information S6). These two observations might indicate that not only newly chlorinated organic compounds are less assimilable and therefore responsible for the lower carbon bioavailabilty, but also chlorination of the water matrix might inhibit consumption of non-chlorinated compounds. Consequently, a lower bacterial regrowth could be expected in chlorinated waters even without residual disinfectant. Permanganate is the least reactive of the five oxidants investigated and its consumption by the water matrix is only small (supporting information S5). The small decrease in concentration results in high permanganate exposures after short reaction times. The low reactivity is responsible for the very low AOC formation kinetics shown in Fig. 1. The inset in Fig. 1 shows the AOC formation plotted versus the oxidant consumption. The slopes of the linear fits give the AOC formation yield in mM AOC/mM oxidant consumed. Even though the absolute amount of AOC formed is small, permanganate has an even higher AOC formation yield than ozone (note the gap and difference in the scale on the x-axis). Therefore, AOC formation could be significant in cases where the water matrix consumes more permanganate than in this study. When ferrate was added to water and left to react without quenching, the AOC concentration increased considerably as is shown in Fig. 2 for LZ water and in the supporting information for LG water (Fig. S3). This indicates that the reaction of ferrate with DOM increases the AOC concentration. However, even when ferrate was quenched with iron(II) chloride immediately upon addition to the water, the AOC concentration increased with increasing ferrate dose (data not shown). This increase cannot be explained by reactions of ferrate (VI) with DOM. The AOC increase was clearly too high to stem from organic contaminations in the ferrate. Additional experiments furthermore showed that (1) the biological growth in the investigated water was not iron limited, (2) hydrogen peroxide that can be formed during ferrate application (Lee et al., in preparation), was not responsible for the AOC increase, (3) a Fenton type reaction with iron(II) and H2O2 to increase the AOC concentration could be excluded due to the high pH, and (4) iron(II) chloride alone did not lead to increased AOC results when used in water treated with ferrate (supporting information S4). Cl2

-

MnO4

2-

FeO4

300

300

200

200

100

100

0

0 10 20 50

0 10 20 50

0 10 20 50 0 10 20 50 oxidant dose (µM)

0 10 20 50

0

2L Fig. 2 e AOC concentration after 24 h of treatment of LZ water with varying oxidant doses of O3, ClO2, Cl2, MnOL in 4 , FeO4 absence and presence of Aphanizomenon gracile. Black bars: AOC in water without A. gracile, grey bars: additional AOC in water containing A. gracile, i.e. the sum (black D grey bars) is the AOC concentration in water samples containing cells. Mean values of two AOC measurements are shown.

2006

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 0 2 e2 0 1 0

Although this study cannot explain the parameters influencing the AOC concentration during ferrate application, it can be concluded that ferrate has the potential to increase the AOC concentrationein some experiments larger yields than during ozonation were measured. This is remarkable because a significant amount of ferrate is consumed by self-decomposition and therefore not available for reaction with the matrix (supporting information S7). To our knowledge no systematic study concerning AOC and ferrate has been performed before. To better understand the AOC formation by ferrate further research is needed, especially because after this study it seems plausible that in water treatment with ferrate measures to control the AOC formation might be necessary.

and co-workers (Hammes et al., 2007). Tung and co-workers (Tung et al., 2004) showed that as well ozone as chlorine and permanganate were able to damage cyanobacteria to such an extent that intra-cellular 2-methylisoborneol leached into the water. Permanganate, ferrate, and ozone seem to react according to mechanism (i). This mechanism offers an explanation why only small oxidant doses are necessary to produce large AOC concentrations. Apparently, already low oxidant doses are sufficient to damage the cells. However, in the case of ozone the AOC concentration increased with increasing ozone doses. This can be explained by reactions with non-assimilable cellular organic compounds according to mechanism (ii). This can be expected, since ozone has generally the highest reactivity with organic compounds. Besides AOC, the taste and odor compound geosmin was measured. Geosmin is a compound produced by A. gracile (Durrer et al., 1999; Suffet et al., 1999) (intracellular concentration was 2 ng/L). Geosmin is not reactive towards the applied oxidants except OH radicals formed during ozonation (Peter and von Gunten, 2007). When 10 mM chlorine, chlorine dioxide, permanganate, or ferrate was added to the water containing cells, the dissolved geosmin concentration increased by 50% and remained stable at higher oxidant doses (data not shown). This indicates that chlorine and chlorine dioxide are in fact able to damage the cells, too. However, even though organic compounds are expected to leach from the cells, the measured AOC concentration in the solution did not increase (Fig. 2). This result is in line with the hypothesis discussed in Section 3.1 stating that a water matrix treated with chlorine makes organic compounds less bioavailable. Two previous studies report similar results: moderate chlorination of picoplankton (Okuda et al., 2006) and of intra- and extracellular (Kim et al., 2006) material increased the AOC concentration initially, indicating, that chlorine is able to damage the cells and even react with cellular material. However, a further increase of the chlorine dose led to a decrease of the AOC concentration in both cases. Chlorination of extracellular

3.2. AOC production by oxidants in presence of cyanobacterial cells The results discussed so far were obtained from experiments performed in 0.2 mm filtered lake water. Fig. 2 shows AOC concentrations of LZ water that was spiked with cells of the cyanobacterium A. gracile before oxidation with three doses of the selected oxidants. Again, chlorine and chlorine dioxide did not result in the formation of a considerable amount of AOC, in presence or absence of A. gracile. In contrast, the treatment of water containing cyanobacteria with ozone, ferrate, and permanganate yielded a significantly higher AOC concentration than treatment of water without cells. In all three cases an oxidant dose of 10 mM was sufficient to increase the AOC concentration considerably but only in the case of ozone a further increase of the oxidant dose resulted in a further increase of the AOC concentration. At least two mechanisms can be suggested to explain the AOC production by oxidation of cells: (i) damage of the cell membranes by the oxidant and subsequent leaching of easily bioavailable cell constituents and (ii) oxidation of initially nonassimilable cellular material to better assimilable compounds. The leakage of AOC compounds from algal cytoplasm of Scenedesmus vacuolatus upon ozonation was proposed by Hammes

O

OH O3

A

O

O O3

OH

O 3 / HO

H

HO

HO

OH

OH

O O O

O

B

6 HOCl

H3C CH3

8 OH

HO

-

OH

O

C

H

H H

+

2 CHCl3

O

O

H

O

MnO4

-

OH Mn

O H H

MnO4

O H H

-

HO H H

OH H H

MnO4

O

-

MnO4

H

H O

O

-

HO

OH O

Scheme 1 e Oxalate formation mechanisms by different oxidants as proposed in former studies. Formation by (A) ozone (Gould and Weber, 1976; Ramseier and von Gunten, 2009), (B) chlorine (Deborde and von Gunten, 2008), and (C) permanganate (Yan and Schwartz, 2000).

2007

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 0 2 e2 0 1 0

organic material with chlorine doses of >5 mg/L even decreased the AOC concentration to a lower level than before chlorination (Kim et al., 2006).

3.3.

Formation of a specific oxidation product: oxalate

Oxalate is a known oxidation product from oxidation of DOM by most oxidants. For example oxidation of phenol-like structures or olefins by ozone following the classical Criegee mechanism

a

5.0

5

O

oxalate (µM)

4.0

oxalate (µM)

4

3

3 2

ClO 2

1

HOCl

0

3.0

0

100

Mn(VII)

O 3

Fe(VI)*

time (min)

200

2.0

Mn(VII) 1.0

Fe(VI)*

ClO , HOCl 2 0.0 0

25

50

75

100

oxidant consumed (µM)

b

0.8

0.6

oxalate (µM)

LGj

CB

0.4

LGn LZ 0.2

0.0

0

10

20

30

and further oxidation can lead to the formation of oxalate, Scheme 1A (Gould and Weber, 1976; Ramseier and von Gunten, 2009). Chlorination of diketones is known to yield trihalomethanes and oxalate, Scheme 1B (Deborde and von Gunten, 2008, and references therein). Substituted phenolic compounds undergo ring opening when reacting with ClO2 and result in diand tri- carboxylic acids and glyoxal as stable end products (RavAcha, 1984). Permanganate forms diols from olefines and is able to oxidize primary alcohols to aldehydes and subsequently to carboxylic acids, Scheme 1C (Smith and March, 2001). Therefore, oxalate was found after treatment of phenolic compounds with permanganate (Mohindra Chawla et al., 1989). Hu and coworkers propose a reaction mechanism for olefin oxidation with ferrate similar to the permanganate reaction forming a cyclic adduct intermediate (Hu et al., 2009). Hence, similar reaction products including oxalate as in the case of permanganate reaction can be expected. Furthermore, it has been shown previously that oxalate concentrations increase in parallel to AOC during ozonation in a full-scale water treatment plant (Ramseier and von Gunten, 2009). Fig. 3 shows the oxalate formation as a function of the oxidant consumption in a) LG water for ozone, chlorine dioxide, chlorine, and permanganate and in CB water for ferrate and b) in LGj, LGn, CB and LZ waters for chlorine dioxide representing all oxidants. A linear increase of oxalate as a function of the consumed oxidants was observed for all conditions. Therefore, the oxalate yield defined by the molar ratio of oxalate produced over oxidant consumed was constant over time. It is also remarkable that the oxalate yields normalized to the DOC concentrations given in Table 1 were very similar in the three water samples with a similar DOC concentration (LGn, LGj, and CB water) even though the origin of the LG and CB water is different. In the water sample with a lower DOC concentration (LZ water) more oxalate per DOC was formed. The linearity and independence of oxalate formation from the DOM source suggest that oxalate is a very general oxidation product and not selectively formed from concentration-limited precursors. The inset in Fig. 3a documents an initial fast increase of the oxalate concentration as a function of time and a subsequent decrease of the formation rate. Similar formation kinetics were also observed in an earlier study for the formation of trihalomethanes (THMs) during chlorination (Gallard and von Gunten, 2002). Interestingly, the comparison of the formation of THM and oxalate upon chlorination results in a linear correlation

40

chlorine dioxide consumed (µM) Fig. 3 e Oxalate formation (a) upon treatment of LGj water with different oxidants (initial doses were 44e97 mM (supporting information S5)), inset: same data plotted vs. time (*note the ferrate data were collected in CB water) and (b) during oxidation of different water samples treated with chlorine dioxide representative for all five oxidants (for water characteristics cf. supporting information S1). Slopes and R-Square values of the linear fits are (a) ozone: 0.044, 0.999; chlorine dioxide: 0.028, 0.989; chlorine: 0.028, 0.995; permanganate: 0.163, 0.991; ferrate: 0.012, 0.998; (b) LGn: 0.018, 0.990; LGj: 0.028, 0.989; LZ: 0.009, 0.994; CB: 0.016, 0.957.

Table 1 e Oxalate yields in % (mol oxalate/mol oxidant consumed) normalized to the DOC of the water. For DOC concentrations refer to Table S1 in the supporting information.

ozone chlorine dioxide chlorine permanganate ferrate

LGn

LGj

1.0 0.5a 0.3

a

1.1 0.7a 0.7a 4.2a

LZ

CB

1.4 0.7a 1.3 11 0.9

1.0 0.5a 0.5 3.6 0.4a

All R2 values for the determination of the yield were >0.9. a Data for the yield determination are shown in Fig. 3.

2008

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 0 2 e2 0 1 0

dependency on oxidant consumption shows a potential for further elucidation of disinfection by-product formation during chlorination processes. For AOC a slight correlation between the formation of oxalate and AOC could only be observed in the case of ozone and permanganate (supporting information S8). Oxalate only represents a part of the AOC. The weak correlation suggests that not all parts of newly produced AOC follow the same formation kinetics as oxalate. Furthermore, oxalate does obviously not represent AOC in the cases of chlorine and chlorine dioxide application as discussed in Section 3.1. Therefore, oxalate cannot be considered as a general AOC surrogate independent of the oxidant used. Nevertheless, the linear formation of oxalate upon oxidation with all oxidants in all water samples tested makes it a useful tracer compound for oxidation processes.

total THM

0.4

0.3

THM (µM)

0.90 CHCl3 0.64

0.2

CHBrCl2

0.1

0.21

CHBr2Cl

0.05

0.0 0.0

0.2

0.4

oxalate (µM)

Fig. 4 e Correlation between the formation of oxalate and THM during chlorination of LGn water. Initial chlorine dose was 4 mg/L. Slopes and R-square values of the linear fits are for TTHM: 0.896, 0.972; CHCl3: 0.638, 0.972; CHBrCl2: 0.206, 0.958; CHBr2Cl: 0.049, 0.966.

4.

 AOC was formed fast upon ozonation; 60 mg/L of AOC were formed at an ozone exposure of 1 mg*min/L. Chlorine dioxide and chlorine produced no AOC. AOC formation from permanganate was slow, however, the AOC yield (AOC formed/ permanganate consumed) was higher than for ozone. Ferrate application led to elevated AOC concentrations. These results confirm the necessity of adequate measures to reduce AOC after ozone and ferrate application and that AOC formation during chlorination is only of minor concern.  In a water containing A. gracile, increased AOC formation was observed after ozone, permanganate, and ferrate application. No significant AOC formation was found after application of chlorine dioxide and chlorine, despite the ability of all oxidants of damaging the cells, as was shown by an increase of the extracellular geosmin concentration.  Oxalate was produced linearly as a function of the oxidant consumption from all oxidants in all water matrices, suggesting it to be a general oxidation product and not to be formed from concentration-limited precursors.

(Fig. 4) with a THM:oxalate formation ratio of roughly 1:1. Scheme 2 shows a possible reaction mechanism for oxalate and trihalomethane formation from humic acids. 2,3,3,5,5, 5-hexachloro-4-hydroxy-pentanoic acid (compound 1) was found during chlorination of 3,5-dihydroxybenzoic acid and it was the most abundant chloroform precursor after chlorination of humic acids (de Leer et al., 1985). de Leer et al. (1985) proposed reactions (I)e(III) in Scheme 2 leading to the formation of one equivalent of trichloromethane. Reaction (VII) was proposed by Bartlett (Bartlett, 1934) and forms a second equivalent of trichloromethane. Hence, the reaction proposed in Scheme 2 does not reflect the stoichiometry found experimentally in this study. However, it demonstrates the possibility of simultaneous oxalate and THM formation during chlorination of DOM. Finally, the similar kinetics of two different oxidation products found in this study and their linear

O

OH OH Cl2

or Humic Acid

Cl

OH

OH

Cl

- CO2 - HCl

OH Cl

O

O

Cl

H OH

Cl 3C

slow (II)

Cl

OH-/H+ Cl

Cl

O

O

H

HO

(IV) Cl

Cl

Cl2 (oxidation)

Cl

O

+ H2O - CHCl3 4 fast (III)

1

H

HO

O

H

Cl 3C

(I)

Cl HO

O

Conclusion

(V)

H H

HO

Cl Cl

OH

O

H H

- HCl (VI)

HO

Cl O

2 O

+ H2O + 2 HOCl (VII)

OH

HO

+

CHCl3

O

3

4

Scheme 2 e Proposed reaction mechanism of chlorination of humic acids leading to 2,3,3,5,5,5-hexachloro-4-hydroxypentanoic acid 1, 3-chloro-2-oxo-propionic acid 2, and oxalate 3 and trichloromethane 4 according to de Leer (de Leer et al., 1985) and Bartlett (Bartlett, 1934).

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 0 2 e2 0 1 0

Furthermore, oxalate was found to be formed in roughly the same concentrations as trihalomethanes during chlorination. Due to the differences in formation kinetics and yields between oxalate and AOC, oxalate cannot be considered as a general surrogate parameter for AOC formation during oxidation processes.

Acknowledgement Frederik Hammes is acknowledged for his support with AOC measurements and for fruitful discussions. We thank Yunho Lee for assistance in the lab and for sharing general knowledge, especially concerning ferrate application. Elisabeth Salhi is acknowledged for laboratory support, WVZ (the water works of Zurich) for providing the A. gracile culture and Thomas Egli for helpful discussions.

Appendix. Supplementary material Supplementary data related to this article can be found online at doi:10.1016/j.watres.2010.12.002.

references

Bartlett, P.D., 1934. Enolization as directed by acid and basic catalysts. II. Enolic mechanism of the haloform reaction. Journal of the American Chemical Society 56 (4), 967e969. Chen, C., Zhang, X.J., He, W.J., Lu, W., Han, H.D., 2007. Comparison of seven kinds of drinking water treatment processes to enhance organic material removal: a pilot test. Science of the Total Environment 382 (1), 93e102. de Leer, E.W.B., Sinninghe Damste, J.S., Erkelens, C., de Galan, L., 1985. Identification of intermediates leading to chloroform and C-4 diacids in the chlorination of humic acid. Environmental Science and Technology 19 (6), 512e522. Deborde, M., von Gunten, U., 2008. Reactions of chlorine with inorganic and organic compounds during water treatmentekinetics and mechanisms: a critical review. Water Research 42 (1e2), 13e51. Dercova´, K., Seja´kova´, Z., Skokanova´, M., Barancı´kova´, G., Makovnı´kova´, J., 2007. Bioremediation of soil contaminated with pentachlorophenol (PCP) using humic acids bound on zeolite. Chemosphere 66 (5), 783e790. Durrer, M., Zimmermann, U., Juttner, F., 1999. Dissolved and particle-bound geosmin in a mesotrophic lake (Lake Zurich): spatial and seasonal distribution and the effect of grazers. Water Research 33 (17), 3628e3636. Elovitz, M.S., von Gunten, U., 1999. Hydroxyl radical ozone ratios during ozonation processes. I-The R-ct concept. OzoneScience and Engineering 21 (3), 239e260. Gallard, H., von Gunten, U., 2002. Chlorination of natural organic matter: kinetics of chlorination and of THM formation. Water Research 36 (1), 65e74. Gould, J.P., Weber, W.J., 1976. Oxidation of phenols by ozone. Journal Water Pollution Control Federation 48 (1), 47e60. Hammes, F.A., Egli, T., 2005. New method for assimilable organic carbon determination using flow-cytometric enumeration and a natural microbial consortium as inoculum. Environmental Science and Technology 39 (9), 3289e3294. Hammes, F., Salhi, E., Koster, O., Kaiser, H.P., Egli, T., von Gunten, U., 2006. Mechanistic and kinetic evaluation of

2009

organic disinfection by-product and assimilable organic carbon (AOC) formation during the ozonation of drinking water. Water Research 40 (12), 2275e2286. Hammes, F., Meylan, S., Salhi, E., Koster, O., Egli, T., von Gunten, U., 2007. Formation of assimilable organic carbon (AOC) and specific natural organic matter (NOM) fractions during ozonation of phytoplankton. Water Research 41 (7), 1447e1454. Hoigne´, J., Bader, H., 1994a. Characterization of water-quality criteria for ozonation processes .2. Lifetime of added ozone. Ozone-Science and Engineering 16 (2), 121e134. Hoigne´, J., Bader, H., 1994b. Kinetics of reactions of chlorine dioxide (OClO) in water .1. Rate constants for inorganic and organic compounds. Water Research 28 (1), 45e55. Hu, L., Martin, H.M., Arce-Bulted, O., Sugihara, M.N., Keating, K.A., Strathmann, T.J., 2009. Oxidation of carbamazepine by Mn(VII) and Fe(VI): reaction kinetics and mechanism. Environmental Science and Technology 43 (2), 509e515. Huber, M.M., Korhonen, S., Ternes, T.A., von Gunten, U., 2005. Oxidation of pharmaceuticals during water treatment with chlorine dioxide. Water Research 39 (15), 3607e3617. Kim, J., Lee, S., Xu, S., Akiba, M., Nomura, M., Chiba, N., Nakano, K., Nishimura, O., 2006. Assimilable organic carbon formation from algogenic organic matter and its variation by chlorination. Water Science and Technology: Water Supply 6 (2), 239e244. Krasner, S.W., Weinberg, H.S., Richardson, S.D., Pastor, S.J., Chinn, R., Sclimenti, M.J., Onstad, G.D., Thruston, A.D., 2006. Occurrence of a new generation of disinfection byproducts. Environmental Science and Technology 40 (23), 7175e7185. Le Chevallier, M.W., Becker, W.C., Schorr, P., Lee, R.G., 1992. Evaluating the performance of biologically-active rapid filters. Journal American Water Works Association 84 (4), 136e146. Lee, Y., von Gunten, U., 2010. Oxidative transformation of micropollutants during municipal wastewater treatment: comparison of kinetic aspects of selective (chlorine, chlorine dioxide, ferrate(VI), and ozone) and non-selective oxidants (hydroxyl radical). Water Research 44 (2), 555e566. Lee, Y., Yoon, J., von Gunten, U., 2005. Spectrophotometric determination of ferrate (Fe(Vl)) in water by ABTS. Water Research 39 (10), 1946e1953. Lee, Y., Escher, B.I., von Gunten, U., 2008. Efficient removal of estrogenic activity during oxidative treatment of waters containing steroid estrogens. Environmental Science and Technology 42 (17), 6333e6339. Lee, Y., Zimmermann, S.G., Kieu, A.T., von Gunten, U., 2009. Ferrate (Fe(VI)) application for municipal wastewater treatment: a novel process for simultaneous micropollutant oxidation and phosphate removal. Environmental Science and Technology 43 (10), 3831e3838. Lee, Y., Kieu, A.T., Kissner, R., von Gunten, U., in preparation. Self-decomposition of ferrate (Fe(VI)) in water: kinetics, mechanism, and formation of hydrogen peroxide. Liu, W., Wu, H., Wang, Z., Ong, S.L., Hu, J.Y., Ng, W.J., 2002. Investigation of assimilable organic carbon (AOC) and bacterial regrowth in drinking water distribution system. Water Research 36 (4), 891e898. Lykins, B.W., Koffskey, W.E., Patterson, K.S., 1994. Alternative disinfectants for drinking water treatment. Journal of Environmental EngineeringeASCE 120 (4), 745e758. Miettinen, I.T., Vartiainen, T., Nissinen, T., Tuhkanen, T., Martikainen, P.J., 1998. Microbial growth in drinking waters treated with ozone, ozone hydrogen peroxide or chlorine. Ozone-Science and Engineering 20 (4), 303e315. Mohindra Chawla, H., Johny, C.J., Mittal, R.S., 1989. New phenolic components from Dalbergia Volubilis. Bulletin de la Societe Chimique de France 126 (1), 82e87.

2010

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 0 2 e2 0 1 0

Okuda, T., Nishijima, W., Okada, M., 2006. Assimilable organic carbon (AOC) originating from picophytoplankton in drinking water. Water Science and Technology: Water Supply 6 (2), 169e176. Peter, A., von Gunten, U., 2007. Oxidation kinetics of selected taste and odor compounds during ozonation of drinking water. Environmental Science and Technology 41 (2), 626e631. Peter, A., Koster, O., Schildknecht, A., von Gunten, U., 2009. Occurrence of dissolved and particle-bound taste and odor compounds in Swiss lake waters. Water Research 43 (8), 2191e2200. Polanska, M., Huysman, K., van Keer, C., 2005. Investigation of assimilable organic carbon (AOC) in flemish drinking water. Water Research 39 (11), 2259e2266. Ramseier, M.K., von Gunten, U., 2009. Mechanisms of phenol ozonationeKinetics of formation of primary and Secondary reaction products. Ozone: Science and Engineering 31 (3), 201e215. Rasul Chaudhry, G., Chapalamadugu, S., 1991. Biodegradation of halogenated organic compounds. Microbiological Reviews 55 (1), 59e79. Rav-Acha, C., 1984. The reactions of chlorine dioxide with aquatic organic materials and their health-effects. Water Research 18 (11), 1329e1341. Richardson, S.D., Plewa, M.J., Wagner, E.D., Schoeny, R., DeMarini, D.M., 2007. Occurrence, genotoxicity, and carcinogenicity of regulated and emerging disinfection byproducts in drinking water: a review and roadmap for research. Mutation Research-Reviews in Mutation Research 636 (1e3), 178e242. Smith, M.B., March, J., 2001. March’s Advanced Organic Chemistry. Reactions, Mechanisms, and Structure, fifth ed. Wiley-Interscience, John Wiley & Sons, Inc., New York, p. 917, 1049, 1515.

Suffet, I.H., Khiari, D., Bruchet, A., 1999. The drinking water taste and odor wheel for the millennium: beyond geosmin and 2methylisoborneol. Water Science and Technology 40 (6), 1e13. Swietlik, J., Raczyk-Stanislawiak, U., Nawrocki, J., 2009. The influence of disinfection on aquatic biodegradable organic carbon formation. Water Research 43 (2), 463e473. Thompson, G.W., Ockerman, L.T., Schreyer, J.M., 1951. Preparation and purification of potassium ferrate .6. Journal of the American Chemical Society 73 (3), 1379e1381. Tung, S.C., Lin, T.F., Liu, C.L., Lai, S.D., 2004. The effect of oxidants on 2-MIB concentration with the presence of cyanobacteria. Water Science and Technology 49 (9), 281e288. Vahala, R., Ala-Peijari, T., Rintala, J., Laukkanen, R., 1998. Evaluating ozone dose for AOC removal in two-step GAC filters. Water Science and Technology 37 (9), 113e120. van der Kooij, D., 1992. Assimilable organic carbon as an indicator of bacterial regrowth. Journal American Water Works Association 84 (2), 57e65. van der Kooij, D., Hijnen, W.A.M., Kruithof, J.C., 1989. The effects of ozonation, biological filtration and distribution on the concentration of easily assimilable organic-carbon (AOC) in drinking water. Ozone-Science and Engineering 11 (3), 297e311. Volk, C.J., Le Chevallier, M.W., 2002. Effects of conventional treatment on AOC and BDOC levels. Journal American Water Works Association 94 (6), 112e123. von Gunten, U., Hoigne´, J., 1994. Bromate formation during ozonation of bromide-containing waters - Interaction of ozone and hydroxyl radical reactions. Environmental Science and Technology 28 (7), 1234e1242. Waldemer, R.H., Tratnyek, P.G., 2006. Kinetics of contaminant degradation by permanganate. Environmental Science and Technology 40 (3), 1055e1061. Yan, Y.E., Schwartz, F.W., 2000. Kinetics and mechanisms for TCE oxidation by permanganate. Environmental Science and Technology 34 (12), 2535e2541.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 1 1 e2 0 1 9

Available at www.sciencedirect.com

journal homepage: www.elsevier.com/locate/watres

Microbial UV fluence-response assessment using a novel UV-LED collimated beam system Colleen Bowker a,1, Amanda Sain b,2, Max Shatalov c,3, Joel Ducoste b,* a

Hazen and Sawyer, P.C., 4944 Parkway Plaza Blvd., Suite 375, Charlotte, NC 28217, USA Department of Civil, Construction, and Environmental Engineering, North Carolina State University, 208 Mann Hall, Box 7908, Raleigh, NC 27695, USA c Sensor Electronic Technology, Inc., 1195 Atlas Road, Columbia, SC 29209, USA b

article info

abstract

Article history:

A research study has been performed to determine the ultraviolet (UV) fluence-response of

Received 10 September 2010

several target non-pathogenic microorganisms to UV light emitting diodes (UV-LEDs) by

Received in revised form

performing collimated beam tests. UV-LEDs do not contain toxic mercury, offer design

22 November 2010

flexibility due to their small size, and have a longer operational life than mercury lamps.

Accepted 4 December 2010

Comsol Multiphysics was utilized to create an optimal UV-LED collimated beam design

Available online 13 December 2010

based on number and spacing of UV-LEDs and distance of the sample from the light source while minimizing the overall cost. The optimized UV-LED collimated beam apparatus and

Keywords:

a low-pressure mercury lamp collimated beam apparatus were used to determine the UV

Disinfection

fluence-response of three surrogate microorganisms (Escherichia coli, MS-2, T7) to 255 nm

UV

UV-LEDs, 275 nm UV-LEDs, and 254 nm low-pressure mercury lamps. Irradiation by low-

LED

pressure mercury lamps produced greater E. coli and MS-2 inactivation than 255 nm and

Water treatment

275 nm UV-LEDs and similar T7 inactivation to irradiation by 275 nm UV-LEDs. The 275 nm

CFD

UV-LEDs produced more efficient T7 and E. coli inactivation than 255 nm UV-LEDs while both 255 nm and 275 nm UV-LEDs produced comparable microbial inactivation for MS-2. Differences may have been caused by a departure from the time-dose reciprocity law due to microbial repair mechanisms. ª 2010 Elsevier Ltd. All rights reserved.

1.

Introduction

The use of ultraviolet (UV) light in drinking water disinfection applications has become a growing alternative to chemical disinfectants since its determination as an effective method for inactivating chlorine resistant pathogenic organisms without the production of any known disinfection by-products (Bohrerova et al., 2006). The majority of UV disinfection

systems currently use low- or medium-pressure mercury lamps, which are toxic, require significant amounts of energy, and have a short lifetime. Other alternative UV light emission sources have been developed (Wang et al., 2005; Bohrerova et al., 2008). A possible alternative to UV mercury lamps or these alternative UV emission sources for point of use applications is the use of UV light emitting diodes (UV-LEDs). UVLEDs contain no known toxic elements that can be released

* Corresponding author. Tel.: þ1 919 515 8150; fax: þ1 919 515 7908. E-mail addresses: [email protected] (C. Bowker), [email protected] (A. Sain), [email protected] (M. Shatalov), jducoste@ncsu. edu (J. Ducoste). 1 Tel.: þ1 704 906 2749. 2 Tel.: þ1 980 621 8620. 3 Tel.: þ1 803 647 9757; fax: þ1 803 647 9770. 0043-1354/$ e see front matter ª 2010 Elsevier Ltd. All rights reserved. doi:10.1016/j.watres.2010.12.005

2012

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 1 1 e2 0 1 9

upon breakage and have a much longer lifetime. In addition, UV-LEDs provide design flexibility due to their small size (Shur and Gaska, 2008). Recent advances in technology have allowed the production of UV-LEDs with emission wavelengths in the germicidal range, with the shortest wavelength at 237 nm (Shur and Gaska, 2008). Light emitting diodes (LEDs) are created by connecting p-type and n-type semiconductors that move electrons into positively charged holes between these two materials. Light is generated when the electrons and holes recombine at a junction. The wavelength of light will depend on the type of material used for the two semiconductors (i.e., indium gallium nitride for light in the visible range and aluminium gallium nitride and aluminium nitride for ultraviolet light) (Dume, 2006). Recently, deep UV-LEDs with peak emission wavelength from 250 to 340 nm have been manufactured for many potential applications including microbial disinfection (Shur and Gaska, 2008, www.s-et.com). While mercury lamps only emit light at one wavelength (i.e., low-pressure) or over a broad range of wavelengths simultaneously (i.e., medium-pressure), UV-LEDs are capable of emitting light at multiple individual wavelengths. Therefore, UV-LED peak emission wavelength can be tuned within the germicidal UV range by adjusting the composition of aluminium gallium nitride during its manufacturing. Several studies have analyzed the microbial response of different strains of E. coli to UV-LED irradiation. These studies suggested that UV-LEDs may inactivate microorganisms efficiently only over long exposure times (Crawford et al., 2005; Mori et al., 2007; Vilhunen et al., 2009). The extensive exposure times required in these studies were likely due to low power outputs from the UV-LEDs. UV-LED power inefficiency was listed as a constraint for UV-LED applicability in Shur and Gaska (2008), but it was also mentioned that efficiency may soon be improved by an order of magnitude or more since the technology is still in its infancy. An issue with the previously mentioned studies using UV-LEDs for microbial inactivation was the lack of applying a standardized measurement of UV fluence delivered to the microorganisms; making the results difficult to compare to UV fluence-response studies in the literature for low- or medium-pressure mercury lamp systems. In order to determine the UV fluence-response of different microorganisms to UV-LEDs that are comparable to those in the literature for low- and medium-pressure mercury lamps, it is necessary to build a UV-LED collimated beam apparatus based on the parameters listed in Bolton and Linden (2003) and use standardized correction factors during irradiation experiments. UV-LEDs in the germicidal range currently have low power outputs that are a function of the emitting wavelength and the applied current. For example, the maximum optical power output of a UVTOP UV-LED is around 0.5 mW at 260 nm and 0.75 mW at 280 nm for a current value of 30 mA (SET, Inc., 2008). The power output will decrease if a lower current value is provided. The UV absorption curve for DNA displays a peak around 260e265 nm, indicating this range as the most effective for germicidal inactivation (Kalisvaart, 2004). However, the DNA UV absorption curve still displays significant absorption at higher wavelengths (e.g. 270e280 nm). Linden et al. (2001) suggested that the use of a UV source emitting between 240

and 280 nm is reasonable for water disinfection purposes. Therefore, it may be valuable to investigate if the higher UV-LED power output at wavelengths larger than 260 nm compensates for the reduction in the germicidal efficiency. The objective of this study was to determine the UV doseresponse of target non-pathogenic microorganisms to germicidal UV-LEDs by performing detailed collimated beam tests on MS-2 coliphage (commonly used surrogate for Cryptosporidium in biodosimetry tests), T7 coliphage (possible alternative surrogate that mimics Cryptosporidium UV response kinetics (Fallon et al., 2007)), and E. coli, which has a lower resistance to UV irradiation than the other two study microorganisms. The collimated beam tests were completed using low-pressure mercury lamps and UV-LEDs with emissions at 255 nm and 275 nm to determine whether the microbial UV dose-response from UV-LEDs at 255 nm is similar to that of low-pressure mercury lamps emitting at 254 nm and to study if the higher power output of 275 nm compensates for its likely lower germicidal effectiveness.

2.

Methods

2.1.

UV-LED collimated beam design

Collimated beam experiments were performed with UV-LEDs to determine the UV response kinetics of the challenge microorganisms (MS-2, T7, and E. coli). UVTOP UV-LEDs with an AlInGaN semiconductor material were obtained from Sensor Electronic Technologies in Columbia, SC. The UV-LEDs used in this study had an overall chamber diameter of 0.9 cm, a height of 0.35 cm, and a flat quartz crystal window diameter of 0.6 cm. Light emission was allowed only through the flat quartz window. Experimental tests were performed with UV-LEDs emitting two unique wavelengths (255 and 275 nm). The two wavelengths represent UV-LEDs with significant differences in power outputs that may impact the UV response kinetics of the microorganisms. For a manufacturer recommended current value of 20 mA, the 255 nm and 275 nm UVTOP UV-LEDs produce a power output of approximately 0.3 mW and 0.5 mW, respectively. Comsol Multiphysics, a numerical modeling and CFD software (Comsol Inc., Burlington, MA) was utilized to find an optimal collimated beam apparatus design with design parameters including the number of UV-LEDs, length of collimator tube, and height of samples relative to the light source. The incident fluence rates on a microbial sample in the UV-LED collimated beam apparatus were predicted using a simple point source summation model (Bolton, 2000). In a collimated beam apparatus, the irradiance is equivalent to fluence rate. As the objective of the model was to find the incident irradiance on the sample, transmittance and absorbance of the sample was not included in model simulations. In this case, the irradiance over a surrounding spherical surface area for a non-absorbing medium at a distance r from the point source can be described by Eq. (1) (Bolton, 2000): I¼

P 4pr2

(1)

where I ¼ irradiance and P ¼ radiant power from the light source.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 1 1 e2 0 1 9

2013

However, the UVTOP UV-LEDs used in this study have a 60 viewing angle (SET, Inc., 2008), meaning that the power from the point source is only distributed over a certain fraction of the surrounding spherical surface area. This narrow distribution of power was corrected in the irradiance calculation assuming all the power was projected onto a spherical cap at a distance r from the UV-LED shown in Eq. (2): I¼

P 2pr2 ð1  cosðaÞÞ

(2)

where a is 60 . The total incident irradiance at each point on the sample surface is equal to the sum of the irradiances from all individual LEDs, given by the distance r from each LED and Eq. (2). In order to determine the optimal LED configuration for both sets of wavelengths, several spatial arrangements for each set were analyzed. Fig. 1 and Fig. 2 display examples of simulated arrangements explored in this study. The 255 and 275 nm arrays consisted of 8 UV-LEDs and 4 UV-LEDs, respectively, because the power output of 275 nm UV-LEDs was almost twice that of the 255 nm UV-LEDs. For each arrangement, the light intensity distribution over the plane representing the sample surface area was analyzed for a range of distances from the light source. The Petri Factor and the average irradiance values were used to determine the optimal configuration for each UV-LED wavelength setup. An ideal configuration would allow a Petri Factor of approximately 0.9 (Bolton and Linden, 2003) while still maintaining a high enough irradiance to have reasonable exposure times for a desired UV dose. A collimated beam apparatus was constructed to allow for the UV-LEDs at 255 nm and 275 nm and their electrical equipment to be interchangeable. A resistance in series was used to limit this current flow for each UV-LED. Pairs of UV-LED and resistor were connected in parallel to the DC power source. The apparatus consisted of a 56 cm  56 cm box with the LEDs centered at the top in either an X-Array or Spaced Original configuration. The array configurations were based on the Comsol simulations that will be discussed in the results section. Both arrays of LEDs were placed in a 7.6 cm  7.6 cm metal holder that allowed for heat dissipation. The collimating tube was approximately 3.3 cm long with a diameter of 10.2 cm.

Fig. 1 e Spatial arrangements for 255 nm UV-LEDs (Set of 8 LEDs).

Fig. 2 e Spatial arrangements for 275 nm UV-LEDs (Set of 4 LEDs).

2.2.

UV-LED collimated beam experiments

Bench scale experiments were completed using a collimated beam apparatus with either low-pressure mercury lamps (four 15 W T8 UV lamps), 255 nm UV-LEDs, or 275 nm UV-LEDs as the light source to determine the UV response kinetics of MS-2, T7, and E. coli. The absorbance of samples prior to irradiation was found with a HACH DR 5000 Spectrophotometer at the wavelength corresponding to the UV-LED or low-pressure mercury lamp wavelength output (depending on the light source used for the particular experiment) in order to determine the water factor (Bolton and Linden, 2003). The UV-LEDs used in this study emit at a small range of wavelengths, with the majority centered at the peak value of either 255 nm or 275 nm. The absorbance was also taken at the maximum and minimum wavelengths emitted by each UVLED and the difference in absorbance from peak wavelength to the outside range was negligible. The protocol described for low-pressure mercury lamps in Bolton and Linden (2003) was used to find the average irradiance for the UV-LED and lowpressure mercury lamp experiments. The incident irradiance at the surface of the liquid sample was found for the UV-LED experiments using a Stellarnet EPP2000C-100 Spectroradiometer with an attached fiber optic probe to capture fine planar variations in light intensity calibrated at both 255 and 275 nm. For the low-pressure mercury lamp experiments, a UVX Digital radiometer (UVP, Inc.) calibrated at 254 nm was used to find the incident irradiance. Both radiometers were checked with a potassium iodide actinometry method (Rahn, 1997; Rahn et al., 2003). The average UV fluence was calculated as the exposure time multiplied by the average irradiance. The range of average UV fluences paralleled the UV fluence ranges in Sommer et al. (1998) and Bohrerova et al. (2008), which contain the UV dose-response curve for MS-2, E. coli, and T7 irradiated by a UV low-pressure mercury lamp (0e60 mJ/cm2 for MS-2, 0e20 mJ/cm2 for T7, and 0e12 mJ/cm2 for E. coli). Table 1 displays the incident irradiance and exposure time ranges for each microorganism.

2014

2.3.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 1 1 e2 0 1 9

Microorganisms

Table 2 e Microorganism specifications.

MS-2 coliphage (ATCC 15597-B1) and T7 coliphage received from Clancy Environmental Consultants were propagated and enumerated according to the methods described in Bohrerova et al. (2006) and Bohrerova et al. (2008), respectively. For enumeration, the appropriate E. coli host (Table 2) was cultivated using the appropriate broth (Table 2) in an incubator shaker at 37  C for 4e6 h. Tryptone-based top agar was placed in 4 mL centrifuge tubes in a water bath at 45  C. 1 mL of E. coli host and 0.1 mL of the appropriate bacteriophage (MS-2 or T7) dilution were added to a 4 mL centrifuge tube of top agar and poured onto tryptone-based bottom agar. Samples were plated in triplicate. After the agar solidified, plates were incubated at 37  C in the inverted position for 18e24 h for MS-2 or 5 h for T7. Plates that contained between 20 and 200 plaques were counted. The method was similar for propagation, except for after incubation, where 5 mL of the appropriate sterile solution (Table 2) was poured onto the plates and left to sit for 30 min. The top agar was then scraped off the plates and centrifuged for 15 min at 5000 rpm and 4  C. The supernatant was filtered through a 0.2 m cellulose acetate membrane syringe filter and stored in 45 mL light sensitive centrifuge tubes at 4  C. E. coli (ATCC 11229) was propagated as specified in the ATCC material data sheet. For each experiment, 0.2 mL of E. coli (ATCC 11229) stock was added to 50 mL nutrient broth and placed in an incubator shaker at 37  C for 18e24 h. After incubation, the E. coli cells were washed in phosphate buffered saline (PBS) by centrifugation and re-suspension in PBS twice to produce a concentration of approximately 1 108 cfu/mL for the irradiation experiments. Unlike the bacteriophages, E. coli 11229 was kept at room temperature throughout experiments and during centrifugation. After irradiation, the necessary dilutions were spread on nutrient agar plates in 0.1 mL volumes and allowed to dry. The plates were incubated upside down at 37  C for 18e24 h. Each sample was plated in triplicate and plates yielding 0e200 colonies were counted.

3.

Results and discussion

3.1.

Comsol model

Microorganism

E. coli host

Broth

Agar

MS-2 (15597-B1)

ATCC 15597 ATCC 11303 N/A

Tryptonebased TSB and 0.5% NaCl Nutrient

Tryptonebased Tryptonebased Nutrient

T7 E. coli (11229)

Solution Salinecalcium PBS PBS

The analyzed configurations for the 255 nm setup were labeled the X-Array, 4  2, 8 Evenly Spaced, 3 Rows, and 2  2 with Corners (Fig. 1). The 275 nm setup consisted of 4 UV-LEDs. The analyzed configurations for the 275 nm setup, shown in Fig. 2, were labeled the Original Array, 4 Corners, 4 Lines Arrays, 4 Staggered Lines, and 4 Diagonals. Fig. 3 displays the predicted Petri factor and the average irradiance for the compared arrays at a range of distances from the 255 nm UV-LEDs and 275 nm UV-LEDs. The results in Fig. 3a show only minor differences in the Petri factor between arrays analyzed for the 255 nm configuration. The 8 Evenly Spaced and the X-Array produced a slightly higher Petri factor than the rest of the arrays for the majority of distance ranges analyzed. Also, both the 8 Evenly Spaced and X-Array configurations reached the optimal minimum Petri factor of 0.9 at 4 cm away from the LEDs. The Comsol model output for the irradiance values (Fig. 3b) shows that X-Array produced the highest average irradiance values for all analyzed distances from the LEDs and the 8 Evenly Spaced configurations produced the lowest. Therefore, the X-Array was selected as the optimal configuration for the 255 nm UV-LEDs due to its higher Petri factor and irradiance values in the simulated Comsol collimated beam apparatus. The X-Array configuration consisted of a 1.5 cm distance between each of the center UVLEDs and a 2 cm diagonal distance from each center UV-LED to the UV-LED at its respective corner. The Comsol output for the analyzed 275 nm configurations (Fig. 3c) shows that the simulated Four Corners array had the highest Petri factor for the entire range of analyzed distances while the other 275 nm arrays produced similar Petri factors for the same range. The 275 nm configuration output (Fig. 3d) also shows that the Original Array had much higher average irradiance values than the other arrays for close distances to the simulated UV-LEDs. For distances beyond 4 cm, the Original Array average irradiance results started to converge with the results from the other arrays. The optimal configuration was determined to be a hybrid between the Original and Four Corners arrays formed by increasing the distance between the LEDs in the Original array and will now be referred as the

Comsol Multiphysics was utilized to determine the optimal LED configurations while minimizing the overall cost for the 255 nm and 275 nm LEDs by predicting the Petri factor and average irradiance value for each configuration. For the 255 nm setup, 8 UV-LEDs were used to account for a low power output.

Table 1 e Irradiance and exposure time ranges. Incident irradiance range (mW/cm2)

T7 MS-2 E. coli

Exposure time range (sec)

LP

275

255

LP

275

255

0.32e0.33 0.32 0.34

0.090e0.10 0.089e0.10 0.094e0.11

0.048e0.057 0.046e0.057 0.049e0.060

0e50 0e164 0e33

0e310 0e859 0e158

0e634 0e1577 0e292

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 1 1 e2 0 1 9

2015

Fig. 3 e Comsol output: a) Petri Factor for 255 nm, b) Irradiance values for 255 nm configurations, c) Petri Factor for 275 nm configuration and d) Irradiance values for 275 nm configuration.

Spaced Original configuration. The Spaced Original configuration consisted of a 1.5 cm distance between each UV-LED and provided higher Petri factor values than the Original Array while still maintaining relatively high irradiance values. After the UV-LED collimated beam was constructed, irradiance measurements for the 255 nm and 275 nm UV-LED configurations were performed with a Stellarnet EPP2000C Spectrometer and compared to Comsol model output at a range of distances from the UV-LEDs. A comparison of the experimental and model Petri factors and average irradiance for both configurations at a range of distances is presented in Fig. 4. Fig. 4b shows that for the 255 nm UV-LED configuration, the model over predicted the Petri factor at a close range, but approached the experimental value at distances greater than 4 cm from the light source. The opposite occurred with the 275 nm configuration, where the model under predicted Petri factors at small distances from the UV-LEDs and converged with the experimental values at distances greater than 4 cm. In Fig. 4c and d, the results indicate good agreement between the model and experimental average irradiance values for the 255 nm and 275 nm configurations over the entire range of distances analyzed. Fig. 4c displays that at the extreme distances (i.e. very close and very far away) the 275 nm UV-LED model predicted slightly higher values than was shown in the experimental data. The 255 nm UV-LED model predicted average irradiance values very similar to the experimental data at a close distance to the UV-LEDs and predicted slightly lower irradiance values at farther distances. All 255 nm and 275 nm UV-LED collimated beam experiments were completed at a 4 cm distance from the UV source. Additional analysis was completed at this distance to determine the difference in model and experimental irradiance

values at each individual point over the entire sample area. Fig. 5a displays the fractional difference between model and experimental irradiance values for the 255 nm UV-LED configuration, showing that the majority of the sample area contained less than a 20% variation between model and experimental data. A small section in the bottom-left corner of the sample contained a 20e40% variation between the model and experimental irradiance values for the 255 nm UV-LEDs. Fig. 5b displays the fractional difference between model and experimental irradiance values over the sample area at a 4 cm distance for the 275 nm UV-LED setup. Most of the sample area had less than a 10% difference in irradiance between model output and experimental data. The majority of the remaining area had less than a 20% discrepancy between experimental and model irradiance values for the 275 nm UV-LEDs. The results of these model and experimental UV irradiance comparisons suggest that the model did not include all of the physics inside the collimated beam apparatus. It is hypothesized that the absence of the collimator tube from the model caused a slight shadow zone from the outer edge to some of the UV-LEDs. Consequently, the model would over-predict the contribution of the UV irradiance from these partially blocked UV-LEDs in these regions. In addition, the presence of the collimator tube was designed to eliminate the divergence of the UV light rays from the emission source. Since the divergence was allowed to occur from the modified point source in Eq. (2), then the model would tend to under predict the UV irradiance received at the sample surface.

3.1.1.

E. coli 11229

Bench scale collimated beam experiments were completed on E. coli 11229 using 255 nm UV-LEDs, 275 nm UV-LEDs, and

2016

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 1 1 e2 0 1 9

Fig. 4 e a) Model and Experimental Petri Factor comparison for 275 nm and 255 nm, b) Model Petri Factor e Experimental Petri Factor for 275 nm and 255 nm, c) Model and Experimental average irradiance value comparison for 275 nm and 255 nm, d) Model irradiance e Experimental Irradiance for 275 nm and 255 nm.

low-pressure mercury lamps as the UV light sources. Fig. 6 displays the results of these experiments in the form of E. coli UV fluence-response curves for each UV light source. The curves for all three UV light sources have similar shapes, with a shoulder at the lower UV fluences. The 255 nm UV-LED results show significantly lower log inactivations than the lowpressure mercury lamp experiments. Assuming that the germicidal effectiveness was only dependent on the adsorption spectra of the target microorganism, this lower log inactivation result was unexpected since the 255 nm UV-LEDs and low-pressure mercury lamps emit very similar wavelengths (Kalisvaart, 2004). However, it has been previously proven that a UV fluence produced with a low irradiance and a long exposure time may result in lower E. coli inactivation rates when compared to a UV fluence produced with a high irradiance and short exposure time (Sommer et al., 1998; Harm, 1980). The 255 nm UV-LEDs provide a much lower power output than the low-pressure mercury lamps, resulting in

Fig. 5 e a) Comparison of Model and Experimental irradiance values for 255 nm UV-LEDs (4 cm distance) and b) Comparison of Model and Experimental irradiance values for 275 nm UV-LEDs (4 cm distance).

Fig. 6 e UV fluence-response curves for E. coli due to irradiation by 255 nm UV-LEDS, 275 nm UV-LEDs and lowpressure mercury lamps.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 1 1 e2 0 1 9

lower incident irradiance values (Table 1). Therefore, the lower irradiance values may be causing the lower inactivation rates produced in the 255 nm LED experiments compared to the lowpressure mercury lamp experiments. Sommer et al. (1998) mentioned that higher inactivation rates from high irradiance values combined with low exposure times might be due to repair enzymes that are more susceptible to high UV intensities. However, for a similar range of lower irradiance values and the same UV doses, this study utilizing 255 nm UV-LEDs measured much lower E. coli inactivations than Sommer et al. (1998), who used low-pressure mercury lamps. For the higher UV intensity experiments where both Sommer et al. (1998) and this study used lowpressure mercury lamps, the E. coli UV dose-response kinetics for both studies were very similar. A decrease in UV irradiance from low-pressure mercury lamps in Sommer et al. (1998) at a UV fluence of 7 mJ/cm2 resulted in a slight E. coli log inactivation decrease from 3.97 to 3.65. A similar decrease in UV irradiance values from the low-pressure mercury lamps to the 255 nm LED experiments in the present study resulted in a log inactivation decrease from 3.76 to 1.68. Therefore, the change in UV light intensity may not be the only factor in the difference in UV response kinetics between the low-pressure and 255 nm LED experiments. If the photon absorption is different in the UV-LED experiments than the low-pressure mercury lamp experiments, this difference in photon absorption could also partially explain the difference in the UV dose-response kinetics between the two UV light sources. Wayne and Wayne (1999) hypothesized that higher microbial inactivations resulting from high irradiance values and low exposure times could also be due to the higher UV intensities causing simultaneous photon absorption in a shorter exposure time as opposed to lower UV intensities with higher exposure times causing sequential photon absorption. The 275 nm UV-LED experiments produced higher E. coli 11229 inactivation rates than the 255 nm experiments (Fig. 6), despite the UV absorption curve for DNA showing a lower absorption for 275 nm (Kalisvaart, 2004). The larger E. coli inactivation in the 275 nm UV-LED experiments compared to the 255 nm UV-LED experiments may be due to the higher power output of the 275 nm UV-LEDs, resulting in higher irradiance values and shorter exposure times to reach the same UV fluence as discussed in the results for the 255 nm UV-LEDs. Another contributing factor for the 275 nm LED irradiation providing higher inactivation rates than 255 nm UV-LEDs may be the absorption spectra of proteins reaching a peak around 280 nm, meaning that repair enzymes are more prone to damage from UV irradiation at wavelengths near 280 nm (Kalisvaart, 2004).

3.1.2.

2017

Fig. 7 e UV fluence-response curves for MS-2 Bacteriophage due to irradiation by 255 nm UV-LEDS, 275 nm UV-LEDs and low-pressure mercury lamps.

low intensity-high exposure time for the same UV fluence. Therefore, the decrease in inactivation may not be a result of the decrease in UV intensity from low-pressure mercury lamps to UV-LEDs emitting at 255 nm. Fig. 7 shows that the 255 nm LED results are almost within the expected range for MS-2 UV fluence-response kinetics due to low-pressure mercury lamp irradiation as specified by the US EPA UV Disinfection Guidance Manual (USEPA, 2006). Therefore, it may also be argued that the difference in the MS-2 UV fluence-response kinetics between the low-pressure mercury lamp and 255 nm LED experiments is not significant. The experiments with UV-LEDs emitting at 275 nm resulted in very similar UV dose-response kinetics to the 255 nm LED experiments with slightly lower inactivations at higher UV doses. The spectral sensitivity of MS-2 displays a peak around 260 nm, which may explain the slightly higher inactivation for 255 nm compared to 275 nm LEDs (MamaneGravetz et al., 2005).

3.1.3.

T7 bacteriophage

T7 bacteriophage experiments resulted in slightly lower inactivation values for the same UV fluences with irradiation by 255 nm UV-LEDs compared to low-pressure mercury lamp

MS-2 bacteriophage

The MS-2 UV fluence-response curves for all UV light sources resulted in a log-linear relationship. Like the results for E. coli 11229, MS-2 bacteriophage had lower reduction rates for the 255 nm UV-LED experiments than for the low-pressure mercury lamp experiments as is displayed in Fig. 6. However, the decrease in inactivation rates from low-pressure to 255 nm UV-LEDs was much smaller than with the E. coli experiments. Sommer et al. (1998) saw no significant change in MS-2 response kinetics from high intensity-low exposure time to

Fig. 8 e UV fluence-response curves for T7 Bacteriophage due to irradiation by 255 nm UV-LEDS, 275 nm UV-LEDs and low-pressure mercury lamps.

2018

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 1 1 e2 0 1 9

irradiation (Fig. 8). Time-dose reciprocity experiments for T7 are not readily available in the literature so it is not clear whether the lower inactivation rates for 255 nm UV-LEDs are due to the lower UV intensity from the UV-LEDs compared to the low-pressure mercury lamps or to some other phenomenon. However, since T7 and MS-2 are both bacteriophages, it is possible that these microorganisms would behave similarly in terms of following the time-dose reciprocity law, meaning that the lower irradiance values would not produce lower inactivation rates. Unlike MS-2, the T7 275 nm LED experiments resulted in very similar UV fluence-response kinetics to the low-pressure mercury lamp results and a slightly higher log inactivation for each UV fluence when compared to the 255 nm UV-LED experiments (Fig. 8). The action spectrum of T7 displays a small peak around 270 nm, which may explain the increased inactivation by the 275 nm UV-LEDs compared to the 255 nm UV-LEDs (Ronto et al., 1992).

4.

Conclusion

This study compared the UV fluence-response of MS-2, T7, and E. coli 11229 when exposed to low-pressure mercury lamps emitting at 254 nm and UV-LEDs emitting at 255 nm and 275 nm. E. coli 11229 inactivation was most efficient in the low-pressure mercury lamp experiments and the least efficient for the 255 nm UV-LED experiments. MS-2 had very similar UV fluence-response kinetics for all three UV sources, with slightly higher inactivation rates corresponding to lowpressure mercury lamp irradiation. T7 also had similar UV fluence-response kinetics for all UV sources, but responded with slightly lower inactivation rates to irradiation by 255 nm UV-LEDs compared to the inactivation rates for irradiation by low-pressure mercury lamps and 275 nm UV-LEDs. This research indicates that 275 nm UV-LEDs may produce more efficient microbial inactivation than 255 nm UV-LEDs for T7 and E. coli and almost identical microbial inactivation for MS-2. Typically, wavelengths around the peak of DNA absorption (260 nm) are considered to produce the highest disinfection efficiency (Kalisvaart, 2004). Therefore, further investigation should be performed to investigate the UV-LED disinfection trends found in this study and whether they are applicable for a variety of UV irradiances. The results of this study indicate that UV-LEDs may be suitable for UV disinfection applications as long as steps are taken to determine the UV fluence-response of target surrogate microorganisms. However, the low UV-LED power output makes very long exposure times necessary to induce significant microbial inactivation. Consequently, until the UV-LED technology is improved to make the power output more efficient, UV-LEDs may initially be appropriate for point-of-use, low flow disinfection applications.

Acknowledgements This research was sponsored by a grant from the National Science Foundation (BES-0932116). The authors would like to

thank the assistance of Ms. Susan Dunn on some of the preliminary collimated beam modeling work, Mr. Jake Rhoads for the construction of the collimated beam apparatus, and Mr. Karthik Sundaramoorthy for the electronics work on the UV-LEDs.

references

Bohrerova, Z., Mamane, H., Ducoste, J., Linden, K.G., 2006. Comparative inactivation of Bacillus subtilis spores and MS-2 coliphage in a UV reactor: implications for validation. Journal of Environmental Engineering ASCE 132, 1554e1561. Bohrerova, Z., Shemer, H., Lantis, R., Impellitteri, C.A., Linden, K.G., 2008. Comparative disinfection efficiency of pulse and continuous-wave UV irradiation technologies. Water Research 42, 2975e2982. Bolton, J.R., 2000. Calculation of ultraviolet fluence rate distributions in an annular reactor: significance of refraction and reflection. Water Research 34 (13), 3315e3324. Bolton, J.R., Linden, K.G., 2003. Standardization of methods for fluence (UV dose) determination in bench-scale UV experiments. Journal of Environmental Engineering 129 (3), 2009e2215. Crawford, M., Banas, M., Ross, M., Ruby, D., Nelson, J., Boucher, R., Allerman, A., 2005. Final LDRD Report; Ultraviolet Water Purification Systems for Rural Environments and Mobile Applications. Sandia Report. October 26, 2008. Dume, B., 2006. LEDs Move into the Ultraviolet. Physics World, Bristol, UK.. http://physicsworld.com/cws/article/news/24926 Retrieved Mar. 25, 2010. Fallon, K.S., Hargy, T.M., Mackey, E.D., Wright, H.B., 2007. Development and characterization of nonpathogenic surrogates for UV reactor validation. Journal AWWA 99 (3), 73e82. Harm, W., 1980. In Biological Effects of Ultraviolet Radiation. Cambridge University Press, New York. Kalisvaart, B.F., 2004. Re-use of wastewater: preventing the recovery of pathogens by using medium-pressure UV lamp technology. Water Science and Technology 50 (6), 337e344. Linden, K.G., Shin, G., Sobsey, M.D., 2001. Comparative effectiveness of UV wavelengths for the inactivation of Cryptosporidium parvum oocysts in water. Water Science and Technology 43 (12), 171e174. Mamane-Gravetz, H., Linden, K.G., Cabaj, A., Sommer, R., 2005. Spectral sensitivity of Bacillus subtilis spores and MS2 Coliphage for validation testing of ultraviolet reactors for water disinfection. Environmental Science and Technology 39, 7845e7852. Mori, M., Hamamoto, A., Takahashi, A., Nakano, M., 2007. Development of a new water sterilization device with a 365 nm UV-LED. Medical and Biological Engineering and Computing 45, 1237e1241. Rahn, R.O., 1997. Potassium Iodide as a chemical actinometer for 254 nm radiation: use of iodate as an electron scavenger. Photochemistry and Photobiology 66 (4), 450e455. Rahn, R.O., Stefan, M.I., Bolton, J.R., Goren, E., 2003. Quantum yield of the IodideeIodate chemical actinometer: dependence on wavelength and concentration. Photochemistry and Photobiology 78 (2), 146e152. Ronto, G., Gaspar, S., Berces, A., 1992. Phages T7 in biological UV dose measurement. Journal of Photochemistry and Photobiology B: Biology 12 (3), 285e294.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 1 1 e2 0 1 9

SET, Inc., 2008. UVTOP Technical Data. Sensor Electronic Technology (SET, Inc.), Columbia, SC. http://www.s-et.com/ Retrieved Sep. 25, 2009. Shur, M.S., Gaska, R., 2008. In: III-nitride Based Deep Ultraviolet Light Sources. Proc., vol. 6894. SPIE, San Jose, CA, USA. Sommer, R., Haider, T., Cabaj, A., Pribil, W., Lhotsky, M., 1998. Time dose reciprocity in UV disinfection of water. Water Science and Technology 38 (12), 145e150. U.S. Environmental Protection Agency (USEPA), 2006. Ultraviolet Disinfection Guidance Manual.
2019

safewater/lt2/pdfs/guide_lt2_uvguidance_draft.pdf> October 21, 2008, Washington, D.C. Vilhunen, S.H., Sarkka, H., Sillanpaa, M., 2009. Ultraviolet lightemitting diodes in water disinfection. Environmental Science & Pollution Research 16, 439e442. Wang, T., MacGregor, S.J., Anderson, J.G., Woolsey, G.A., 2005. Pulsed ultra-violet inactivation spectrum of Escherichia coli. Water Research 39, 2921e2925. Wayne, C.E., Wayne, R.P., 1999. Photochemistry, second ed. Oxford University Press.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 2 0 e2 0 2 8

Available at www.sciencedirect.com

journal homepage: www.elsevier.com/locate/watres

Water ingestion during water recreation5 Samuel Dorevitch a,b,*, Suraj Panthi a, Yue Huang c, Hong Li b, Angela M. Michalek d, Preethi Pratap a, Meredith Wroblewski a, Li Liu b, Peter A. Scheff a, An Li a a

University of Illinois at Chicago School of Public Health, Division of Environmental and Occupational Health Sciences, 2121 W. Taylor, M/C 922, Chicago, IL 60612, USA b University of Illinois at Chicago School of Public Health, Division of Epidemiology and Biostatistics, USA c University of Illinois at Chicago College of Pharmacy, USA d University of Michigan School of Natural Resources & Environment, USA

article info

abstract

Article history:

Quantitative risk assessments have estimated health risks of water recreation. One input to

Received 12 October 2010

risk assessment models is the rate of water ingestion. One published study estimated rates of

Received in revised form

water ingestion during swimming, but estimates of water ingestion are not available for

26 November 2010

common limited contact water recreation activities such as canoeing, fishing, kayaking,

Accepted 6 December 2010

motor boating, and rowing. In the summer of 2009 two related studies were conducted to

Available online 13 December 2010

estimate water ingestion during these activities. First, at Chicago area surface waters, survey research methods were utilized to characterize self-reported estimates of water ingestion

Keywords:

during canoeing, kayaking, and fishing among 2705 people. Second, at outdoor swimming

Risk assessment

pools, survey research methods and the analysis of cyanuric acid, a tracer of swimming pool

Water ingestion

water, were used to characterize water ingestion among 662 people who engaged in a variety

Water recreation

of full-contact and limited-contact recreational activities. Data from the swimming study

Gastrointestinal illness

was used to derive translation factors that quantify the volume of self-reported estimates. At surface waters, less than 2% of canoers and kayakers reported swallowing a teaspoon or more and 0.5% reported swallowing a mouthful or more. Swimmers in a pool were about 25e50 times more likely to report swallowing a teaspoon of water compared to those who participate in limited-contact recreational activities on surface waters. Mean and upper confidence estimates of water ingestion during limited-contact recreation on surface waters are about 3e4 mL and 10e15 mL, respectively. These estimates of water ingestion rates may be useful in modeling the health risks of water recreation. ª 2010 Elsevier Ltd. All rights reserved.

1.

Introduction

Water recreation has been associated with outbreaks of acute gastrointestinal illness caused by viruses, bacteria, and protozoa (Dziuban et al., 2006; Yoder et al., 2004). Higher rates of acute gastrointestinal illness (AGI) have been reported

5

among swimmers compared to non-swimmers (USEPA, 1983, 1984; Seyfried et al., 1985; McBride et al., 1998; Wade et al., 2008; Colford et al., 2007). Controlled immersion trials have demonstrated higher rates of AGI among those randomized to perform head immersion compared to a non-immersion group (Kay et al., 1994; Wiedenmann et al., 2006; Fleisher et al., 2010).

The Metropolitan Water Reclamation District of Greater Chicago, The Water Environment Research Foundation. * Corresponding author. University of Illinois at Chicago School of Public Health, Division of Environmental and Occupational Health Sciences, 2121 W. Taylor, M/C 922, Chicago, IL 60612, USA. Tel.: þ1 312355 3629; fax: þ1 312413 9898. E-mail address: [email protected] (S. Dorevitch). 0043-1354/$ e see front matter ª 2010 Elsevier Ltd. All rights reserved. doi:10.1016/j.watres.2010.12.006

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 2 0 e2 0 2 8

Epidemiologic studies require substantial resources and time, and may generate results that have limited generalizability to settings with different types of recreational use or pollutant sources. Quantitative microbial risk assessment (QMRA) offers an alternative method for evaluating health risks for a particular water quality measure or for a range of water quality measures (Wong et al., 2009; Ashbolt et al., 2010). Conversely, reverse QMRA has can performed to identify an acceptable level of water quality for a given targeted risk level and has recently been utilized to estimate the contribution of specific pathogens to known rates of recreational waterborne illness (Soller et al., 2010). Limited-contact recreation activities are popular, particularly at inland waters. The number of people who engage in limited-contact activities per year in the US (in millions) are: fishing (71), motor boating (52), canoeing (20.7), rowing (9.4), and kayaking per year (6.4) (Cordell et al., 2004). Despite the large number of people exposed through limited-contact recreation, very little is known about the health risks of these activities. Recently site-specific standards for limited (or “secondary”) contact recreation have been explored in the United States for waters that do not support full-contact recreation. These include efforts in Idaho (Idaho Department of Environmental Quality, 2004), Illinois (Illinois Pollution Control Board, 2010), Kansas (USEPA, 2003), Missouri (Missouri Coalition for the Environment, 2010), Texas (Texas Commission on Environmental Quality, 2009) and Utah (Utah Department of Environmental Quality, 2008). Estimated rates of water ingestion could serve as inputs in risk assessments that model the health risks of limited contact water recreation. The most recent version of USEPA’s Exposure Factor Handbook (USEPA, 2009), available in draft form, does not address the volume of water ingested during recreational activities other than swimming. The ingestion rate during swimming was estimated by Dufour et al. (2006), who utilized cyanuric acid (CYA), routinely added to outdoor swimming pools to stabilize chlorine, as a tracer of swimming pool water. In this work we report the volume of water ingested during a variety of water recreational activities using survey research methods and through the measurement of CYA.

2.

Materials and methods

2.1.

Overview

Two groups of participants were enrolled in the spring and summer of 2009. The first group of surface water participants was enrolled in a prospective cohort study of limited contact water recreation, the Chicago Health, Environmental Exposure, and Recreation Study (CHEERS). Participants in the surface water study were recruited at Chicago area locations including piers, harbors, and beaches of Lake Michigan; the Chicago area waterways system (CAWS); several inland lakes; and rivers (excluding the CAWS). Most of the flow in the CAWS is wastewater effluent that has undergone secondary treatment (activated sludge with aeration) but has not been disinfected. A second study group of participants was enrolled into the swimming pool study at participating outdoor public pools. Participants in both studies (surface water and

2021

swimming pool) completed post-recreation interviews that addressed water ingestion during recreation. Those in the swimming pool study collected urine samples for 24 h following recreation, to be analyzed for CYA. The surface water study was an observational design; the swimming pool study was a controlled exposure design.

2.2.

Participants

Data analysis was limited to those age six years or greater. In the swimming pool study, people who had used a swimming pool in the prior four days were excluded, as were those with underlying renal disease, which may influence the elimination half-life of CYA.

2.3.

Recreational activities

Surface water study participants engaged in one of five activities: canoeing, fishing, kayaking, motor boating, or rowing. There were no constraints on the duration of recreation for the surface waters group. Swimming pool participants engaged in canoeing, kayaking, simulated fishing, wading/splashing, head immersion, or swimming. Head immersion involved standing in the water and immersing one’s head three times over a 10 min interval. The duration of all other swimming pool activities was 60 min. Canoes and kayaks were placed in a large swimming pool, and participants were asked to paddle around the pool. During specific study sessions, canoe and kayak instructors were present so that interested participants could practice “rolling” to recover from capsize. Additionally, capsize occurred accidentally and, at times, teenagers and young adults intentionally capsized each other’s kayaks. Simulated fishing involved casting a fishing rod with a toy fish on a plastic hook, reeling in the toy fish, removing it from the hook, replacing it, and repeating the procedure every 5 min. Wading/splashing took place in shallow swimming pools (“splash pools”), some of which had fountains. Participants were instructed to walk around, play, or splash in the water, but not to swim. Swimming took place in large swimming pools and, like the study by Dufour et al. (2006), was limited to lap swimming. In the surface waters study, participants were recruited at boat launches, piers, harbors and, in the case of sea kayakers, at beaches. Neither the choice of recreational activity nor the duration of recreation was assigned to surface waters participants.

2.4.

Data collection

Following recreational activity, participants in the surface water and swimming pool studies were interviewed about water ingestion. Those who reported water ingestion were asked whether they swallowed “a drop or two”, “a teaspoon”, or “one or more mouthfuls”. All interviews were conducted using computer-assisted personal interview methods on laptop computers running BLAISE software (Westat, Inc). The programming of the questionnaires and the consolidation of survey datasets were performed by the UIC Survey Research Laboratory. Following their field interview, participants were given 2 L amber bottles and instructed to collect their urine for

2022

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 2 0 e2 0 2 8

the next 24 h. Upon completion of urine collection, urine samples were transferred from participants’ homes to the UIC laboratory by a courier service.

2.5.

Measurement of CYA

Previously described methods for measuring CYA in swimming pool water by Cantu et al. (2001a, 2001b) and in urine by Patel and Jones (2007) were the basis for a series of method optimization studies. The method optimization studies and a detailed description of the CYA analysis methods are provided in the Supplementary material. Briefly, Chem-Elut cartridges and the solvent MTBE were used in urine sample preparation. The major analytical instrument was an Agilent Model 1100 HPLC system equipped with a vacuum degasser, a quaternary pump, a thermostated column compartment and a diode-array UV detector. A porous graphite carbon (PGC) column, 110 mm  3.0 mm, 5 mm particle size (Hypercarb, Thermo-Fisher) was used. The HPLC was operated in an isocratic mode with a flow rate of 0.8 mL/min. The wavelength of the UV detector was set at 213 nm. A subset of urine samples were analyzed by liquid chromatography tandem mass spectrometry (LC/MS/MS), with the intention of confirming the results obtained using the HPLC/DAD method. The procedure of urine sample pretreatment for LC/MS/MS analysis is similar to that described above for HPLC/DAD, except that 100 ng/mL 13C3 15N3 labeled CYA was spiked in each urine sample as the first step. The Waters MicroMass Quattro LCZ LC/MS/MS system equipped with a SeQuant ZIC HILIC column (2.1  150 mm PEEK, 5 micron) was used. The operational conditions were developed based on those previously reported by Smoker and Krynitsky (2008). For each sample, a 20 mL full-loop injection was made at 8 mL/ min. The electron spray ionization MS was operated in negative mode with multiple reaction mechanism (MRM). The native CYA was monitored with primary transition m/z 128/42 and secondary transition m/z 128/85. The labeled CYA was monitored with transition m/z 134/44. Quantitation of CYA was based on integrated peak area of the primary transition.

2.6.

Data analysis

2.6.1. Water ingestion estimated from tracer chemical methods Estimated water volumes were calculated using the following mass-balance approach, as described by Dufour et al. (2006):

Volume ingestedðmLÞ ¼

2.6.2.

Statistical methods

Individuals were permitted to participate in the study more than once, but not within four days of prior participation. For those who participated more than once it was necessary to determine whether each round of participation could be considered an independent event. If repeat participation events were independent, relatively simple statistical methods such as logistic regression would be used. If, on the other hand, repeat participation events were not independent, mixedeffects regression models would be necessary to account for correlations between repeated events of individuals. The SAS NLMIXED procedure (SAS Institute, Cary, NC) was used to evaluate the independence of repeated measurements within individuals. In this analysis we employed logistic (for dichotomous and ordinal outcomes) and linear regression models with a random individual effect. A variance parameter for the random individual effect was estimated to account for the covariance between repeated measurements from the same individual. A likelihood ratio test on the significance of this individual variation parameter reveals whether repeated events from the same participant can be assumed to be independent. The outcomes of interest included (1) any selfreported water ingestion (vs. none), (2) the estimated volume of water ingested reported (four ordinal categories), and (3) the calculated volume of ingestion based on results of CYA measurement. Age, gender, recreational activity, and the time of enrollment were included as fixed-effect covariates for all models. For the dichotomous outcome (any water ingestion), model selection was performed using likelihood ratio tests to compare between a random-intercept logistic model and random-trend logistic model. For ordinal outcomes (estimated volume of water ingested), the proportional odds assumption was checked by the likelihood ratio test. If the proportional odds assumption was shown to be violated, non-proportional odds ordinal logistic model would be compared to nonproportional odds ordinal logistic model with random intercepts to check whether significant correlations existed between repeated measurements. For the continuous outcome, volume of water ingested, log transformation was performed first to normalize the data. Likelihood ratio tests were employed to choose between mixed-effects regression models with different random effects covariates (random intercept vs. random-trend), as well as appropriate error variance-covariance structures. Model fits were evaluated using Akaike’s information criterion (AIC) and Bayesian information criterion (BIC). Using the selected model with best fit, Wald tests of the random-effect parameters were performed to evaluate independence of the repeated measurements.

Urine CYA conc:ðmg=mLÞ  Urine VolumeðmLÞ Pool CYA conc:ðmg=mLÞ

If less than 20% of the expected 24-urine output was provided, the urine sample volume was considered insufficient. These thresholds for sufficient volume were 150 mL for those age 10 and younger, 200 mL for those age 11e16, and 300 mL for those age 17 and older based on estimates of urine volume being 60% of daily weight-based fluid requirements (Greenbaum, 2007).

Translation factors: As described in the “Results” section, reliable measures of cyanuric acid were only available for a subset of urine samples. Using the calculated ingestion volumes from those samples, as well as the self-reported ingestion estimates from the participants who provided those samples, translation factors were calculated. Two types of translation factors were calculated. The mean estimate was

2023

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 2 0 e2 0 2 8

the mean calculated volume of water ingested for a given level of self-reported ingestion. For each level of self-reported ingestion an upper confidence estimate was calculated as: Upper confidence estimate ¼ Mean estimate þ ð1:96Þðstandard deviationÞ: Logistic regression models were used to evaluate variables associated with dichotomous categories of water ingestion. The three outcomes (evaluated in three separate models) were (1) any water ingestion (vs. none), (2) a teaspoon or more ingested (vs. none or a drop), and (3) a mouthful or more ingested (vs. none, a drop, or a teaspoon). Multivariate models included gender and age category (less than age 18 years vs. 18 years or more) as predictor variables. For recreational activities that took place in both the swimming pool study and the surface water study, models included a variable to evaluate whether estimates of ingestion were different in the pool than in the surface water study. For surface water study participants, water ingestion was also evaluated as a function of whether recreation took place in the CAWS vs. other area waters. All data analyses were conducted using SAS version 9.1 (SAS Institute, Cary, NC).

2.7.

Human subject research

This research was approved by the Institutional Review Board of the University of Illinois at Chicago. Adult participants provided written documentation of informed consent. Parents or guardians provided written documentation of informed consent for participation of their children. Children above the age 8 years provided written assent to participate.

3.

Results

3.1.

Study participants

Of the 662 swimming pool study participants, 410 (62%) participated more than once; 9% participated four or more times. An analysis of random effects in models of survey and CYA-based estimates of water ingestion determined that the covariance of observations from same individual are not significant, and hence, each repeated event from the same participant can be considered independent. Only 6% of participants in the surface water study enrolled more than once, too infrequent for within-individual covariance to be a concern. Therefore the 3367 sets of water ingestion data were treated as though they came from 3367 different individuals. Subject demographics and recreational activities are summarized in Table 1. A comparable proportion of participants were children in the surface water and swimming pool studies.

3.2.

Self-reported water ingestion

Self-reported water ingestion is summarized in Table 2, by recreational activity and study setting, as frequencies, and as relative frequencies. Pool study fishers and participants who walked around the swimming pool reported no water ingestion. No more than 5% of participants in any given activity on surface waters reported ingesting water. Those who canoed and kayaked in swimming pools tended to report water ingestion more frequently than those who engaged in these same activities in surface waters. Swimmers more frequently reported water ingestion than those who performed head immersion.

Table 1 e Recreational activities and demographics of study participants. Surface waters Adult M

Swimming Pool

Child F

M

Total F

Comparison activity Walking Limited contact Canoeing Boating Fishing Kayaking Rowing Wade/splash Full contact Immersion Swimming Total Percent Grand total M: male; F: female.

243 129 190 346 75

983 36.3 3367

410 140 236 372 101

1259 46.5

51 26 87 31 23

218 8.1

62 21 87 52 23

245 9.1

766 316 600 801 222

2705 100.0

Adult

Child M

Total

M

F

F

12

11

29

37

3

7

78 31

26 47

6 4

11 22

59

35

6

12

76 0 121 104 0 112

52 49 310 46.8

45 46 247 37.3

9 11 39 5.9

6 8 66 10.0

112 114 662 100.0

23

2024

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 2 0 e2 0 2 8

Table 2 e Self-reported water ingestion of 3,367 recreators, by activity, and study setting. Surface water study Activity Canoeing Fishing Immersion Kayaking Motor boating Rowing Swimming Wade/splash Walking

Pool study

n

Drop

Tsp

Mouthful

Relative ingestion

n

Drop

Tsp

Mouthful

Relative ingestion

766 600

5 0.3

1.4 0.3

0.5 0.3

1.5 0.4

801 316 222

4.2 2.2 4.5

1.1 0.9 1.8

0.2 0.3 0.5

1.2 1.0 1.9

76 121 112 104

7.9 0 20.5 15.4

6.6 0 9.8 11.5

3.9 0 0.9 6.7

6.9 0 10.3 12.1

114 112 23

53.5 2.7 0

50.9 2.7 0

27.2 0 0

53.6 2.8 0

Numbers in the "Drop" "Tsp" and "Mouthful" columns refer to the percent of participants reporting that volume of ingestion. Relative refers to the proportion of participants in study-activity category, who ingested at least a teaspoon, relative to motor boaters.

3.3.

Predictors of self-reported water ingestion

Multivariate logistic regression models were used to identify and evaluate predictors of ingesting a teaspoon or more of water, by self-report among surface water recreators (Table 3). After adjusting for activity and age category, males tended to report water ingestion more frequently than females. Age category was not associated with self-reported water ingestion. The odds of ingesting a teaspoon or more of water were substantially higher among swimmers than among those who performed head immersion (in the pool study) or limitedcontact activities on surface waters. Odds ratios and confidence intervals similar to those reported in Table 3 were noted for swallowing any water and for swallowing at least a mouthful of water (data not presented). Capsizing occurred among 41/766 (5.4%) surface water and 21/76 (27.6%) of swimming pool canoers. Capsizing occurred among 27/801 (3.4%) surface water and 40/104 (38.5%) swimming pool kayakers. One of 222 surface water rowers (0.5%) reported capsizing as well. Swallowing water among canoers, kayakers, and rowers was evaluated using a multivariate logistic model with capsizing, age category, gender, activity, and study setting (CAWS vs. other surface waters). The only significant predictor of water ingestion was capsizing. Compared to those who did not capsize, the odds (95% CI) of ingesting any, a teaspoon or more, or a mouthful or more among those who did capsize were 4.83 (2.42, 9.64), 24.57 (9.32, 64.75), and 247.22 (17.45, 3501.85), respectively.

Table 3 e Multivariate odds ratios for ingesting at a teaspoon or more of water, by self-report. Variable Child Male Swimming Swimming Swimming Swimming Swimming Swimming Swimming

Reference category

Odds ratio (95% CI), swallowed  teaspoon

Adult Female Motor boating Canoeing Fishing Immersion Kayaking Rowing Wade/splashing

0.81(0.43, 1.50) 2.0 (1.3, 3.2) 125.0 (35.7, 333.3) 62.5 (33.3, 125.0) 333.3 (90.9, 1000.0) 10.0 (4.8, 20.8) 50.0 (27.0, 90.9) 62.5 (21.7, 200.0) 38.5 (11.5, 125.0)

Given the strong association between capsizing and ingesting water, a logistic regression model was run to identify predictors of capsizing. The only significant predicator was that capsizing was much more likely to occur on general use waters compared to CAWS, as presented in Table 4.

3.4.

CYA in pool water and urine samples

A total of 130 pool water samples were analyzed using the HPLC/DAD method. As detailed in the Supplementary material, quality-monitoring data indicate excellent performance of the method for pool samples. Urine samples were collected from 665 of the 685 study participants (97.1%). Excellent agreement between self-reported water ingestion and the estimates obtained from CYA measurement using LC/MS/MS was observed for the subset (N ¼ 27) of the urine samples collected from participants who enrolled in the study on four dates in July, 2009. These participants used a total of four pools, and represented five of the six recreational activities (none fished). The mean age of these participants was comparable to those whose samples were not analyzed by MS (34.8 vs. 34.3 years, p ¼ 0.85). The means  standard deviations are 1.4  0.8, 9.4  11.0, and 26  37 mL for the participant groups with self-reported injection volumes of none, drop to teaspoon, and mouthful, respectively, showing strong agreement between the self-reported results and those based on CYA measurement using LC/MS/MS. However, quality control data for HPLC/DAD analysis indicated a strong matrix effect, particularly when the CYA concentration level in the swimming pool water was low. The correlation between CYA concentrations measured by the two instrument systems was very poor (r2 ¼ 0.1). In addition, the estimates of ingestion

Table 4 e Multivariate predictors of capsize in surface waters. Variable

Reference category

Male Child Kayaking Rowing Other surface waters

Female Adult Canoeing Canoeing CAWS

Odds ratio

95% confidence interval

1.34 0.64 0.75 0.20 9.26

(0.80,2.25) (0.29,1.43) (0.45,1.24) (0.03,1.50) (3.32, 25.64)

2025

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 2 0 e2 0 2 8

Table 5 e Translation factors for self-reported ingestion (mL) based on LC/MS/MS measures of CYA.

None Drop/teaspoon Mouthful

n

Mean

Standard deviation

Median

Minimum

Maximum

UCL

15 6 6

3.5 10.8 20.3

3.6 10.5 30.1

2.0 7.9 11.1

0.3 0.7 1.2

12.7 27.6 80.9

10.6 31.4 79.3

UCL: 95% confidence limit.

from measured CYA in urine using HPLC/DAD were inconsistent with the self-reported information (refer to the online Supplement for details). Therefore, the results of HPLC/DAD analysis were not explored further.

3.5.

Translation factors for self-reported water ingestion

Among participants whose urine samples were analyzed by LC/MS/MS, because few reported swallowing a drop or a teaspoon of water, these two categories were collapsed into a single category. Log10-transformed values of MS-based calculations of ingestion were associated with the ordinal levels of self-reported ingestion (none, drop-teaspoon, mouthful), with an r2 of 0.24, p ¼ 0.009. Table 5 summarizes the values of calculated ingestion for each level of selfreported ingestion. The translation factors (Table 5) were used to estimate the volume of water ingested for recreational activities. The absolute estimated ingestion volumes for the mean and upper confidence level by activity are summarized in Table 6. The table also summarizes the mean ingestion volume relative to that observed during surface water fishing and rowing. Generally, the mean estimates are about 50% greater than the median estimates. This occurs because the distribution of mean ingestion volumes is skewed in a positive direction, with most of the values found near the minimum value (0.3 mL, corresponding to “no water ingestion” in Table 5), consistent with a lognormal distribution. Three categories of recreational activities are apparent based on ingestion

estimates for the 95th percentile (the upper confidence limit). The low ingestion category is comprised of rowing, motor boating, fishing, wading/splashing, and non-capsizing kayaking and canoeing. These activities have an upper confidence estimate of about 10e12 mL/h. Those who capsized during canoeing or kayaking comprise a middle ingestion category, with an upper confidence limit estimate of about 17e20 mL/h. Swimmers were the high ingestion category, with an estimated upper confidence estimate of ingesting about 35 mL/h.

4.

Discussion

In this first study designed to evaluate water ingestion during limited-contact recreational activities, less than 5% of limitedcontact recreators on surface waters reported swallowing any water, compared to more than 50% of swimmers in a pool. Compared to those who canoed or kayaked in a swimming pool, swimmers in a pool were about four to seven times more likely to report swallowing at least a teaspoon of water. Compared to those who canoed or kayaked on surface waters, swimmers in a pool were more than 50 times as likely to report swallowing a teaspoon of water. Because the vast majority of limited-contact recreators on surface waters denied swallowing any water, less dramatic differences were observed in the estimated mean volume of water ingested: canoeing, 3.9 mL; fishing 3.6 mL; kayaking, 3.8 mL; motor boating, 3.7 mL; and rowing 3.9 mL. These are all about

Table 6 e Estimated water ingestion in mL, by activity, study, and capsize status. Activity

Boating Canoeing Canoeing Canoeing Fishing Immersion Kayaking Kayaking Kayaking Rowing Rowing Rowing Swimming Wading/ splashing Walking

Capsize

No No Yes All No NA No Yes All No Yes All NA NA NA

Surface water study

Swimming pool study

Median

Mean

UCL

Relative to surface water fishing and rowing mean

2.1 2.2 3.6 2.3 2.0

3.7 3.8 6 3.9 3.6

11.2 11.4 19.9 11.8 10.8

1.0 1.1 1.7 1.1 1.0

2.2 2.9 2.3 2.3 2.0 2.3

3.8 5 3.8 3.9 3.5 3.9

11.4 16.5 11.6 11.8 10.6 11.8

1.1 1.4 1.1 1.1 1.0 1.1

Median

Mean

UCL

Relative to surface water fishing and rowing mean

2.1 3.9 2.6 2.0 3.2 2.1 4.8 3.1

3.6 6.6 4.4 3.5 5.1 3.6 7.9 5.2

11 22.4 14.1 10.6 15.3 10.9 26.8 17

1.0 1.8 1.2 1.0 1.4 1.0 2.2 1.4

6.0 2.2

10 3.7

34.8 11.2

2.8 1.0

3.5

10.6

1

2

2026

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 2 0 e2 0 2 8

35e40% of the 10.0 mL estimated to occur during swimming in a pool. The mean volume of water ingested during canoeing or kayaking among those who capsized was about 60% higher than among those who did not capsize. Canoers and kayakers were about 5e10 times more likely to capsize in a swimming pool than on surface waters. These observations may have been due to a perception of safety in a pool with lifeguards and kayaking instructors in attendance. Among surface waters, study participants, capsizing was nine times more common on general use waters than on the CAWS. Again, this may have been due to the recognition of unique physical hazards and the presence of wastewater at the CAWS, which are communicated to the public at CAWS boat launches via posted signs. The fact that capsizing is less common on the CAWS suggests that efforts to communicate health and safety hazards to the public have resulted in risk reduction. Work by Fleisher and Kay (2006) suggests that risk perception influences the occurrence of self-reported gastrointestinal illness; our findings suggest that this may occur despite the ingestion of smaller volumes of water. The LC/MS/MS method resulted in calculated volumes of ingestion that were consistent with self-reported data, supporting the validity of the survey instrument for evaluating water ingestion.

4.1.

Significance of the results

QMRA analyses are probabilistic, utilizing a distribution of values for each of the following key inputs: (1) the volume of water ingested during recreation, (2) the density of microbes in the water, (3) the relationship between the dose of microbes ingested and probability of illness. The volume of water ingested can be calculated as a function of the rate of ingestion (volume per unit time) and the duration of recreation. As a consequence of this research, estimates of water ingestion during a variety of water recreational activities are now available, which should reduce the uncertainty in quantitative estimates of health risk due to water recreation. The measurement of CYA in urine (but not in swimming pool water) was problematic using HPLC/DAD, and future efforts to quantify CYA in urine should rely on LC/MS/MS. Several epidemiologic studies have identified associations between self-reported water exposure during full-contact water recreation and the development of illness. Exposure has previously been categorized based on the occurrence of head immersion (Seyfried et al., 1985; McBride et al., 1998; Wade et al., 2006) or body immersion (Wade et al., 2008). A study of windsurfers evaluated the influence of the number of times participants fell into the water (Dewailly et al., 1986). Other studies analyzed the occurrence of illness as function of selfreported water ingestion for controlled exposure (Wiedenmann et al., 2006) and for observational studies of limited-contact (Fewtrell et al., 1994; Lee et al., 1997) and fullcontact (Colford et al., 2007) recreation. In general, health risks are higher in association with increased exposure, though not for all outcomes and not for all definitions of exposure. Two broad categories of study designs - randomized trials of controlled exposure and observational studies - have been employed to evaluate the health risks of water recreation. It has not been possible to directly compare results of the two approaches because of differences in the definition of

exposure used in each of the approaches. In the randomized trials exposure has been defined as three head immersions during a 10-min interval (Kay et al., 1994; Wiedenmann et al., 2006; Fleisher et al., 2010). In the observational studies (USEPA, 1983, 1984; Wade et al., 2006, 2008) exposure was not determined by the investigators, but rather, participants swam or played in the water for as long and in whatever ways they chose. We found that head immersion results in about half the mean volume of ingestion and about 1/5th the likelihood of swallowing at least a teaspoon of water compared to swimming. This information should be useful in synthesizing the findings generated by the two study designs. It should be noted, however, that we evaluated head immersion in a swimming pool while the other studies evaluated immersion in surface water. It is possible that those in swimming pool may have perceived a lesser risk of ingesting pool water and may not have avoided water ingestion as they may have in surface waters, particularly in marine waters. Our findings of relative rates of water ingestion (as mL/ hour or percent that swallow a teaspoon or more) are likely more meaningful than the absolute estimates of ingestion. The mean estimate of water ingestion during limited-contact recreation is about one third of that during swimming (in a pool). Given the mean volume ingestion during swimming relative to limited-contact recreation, rates of illness attributable to limited-contact activities should be about one third of that reported during swimming on the same water, assuming a linear relationship between ingested pathogen dose and illness risk. The fact that self-reported ingestion of a teaspoon or more of water is relatively uncommon (less than 5% of surface water limited-contact recreators) suggests that a small percentage of recreators constitute an at-risk group. We found that capsizing is a strong determinant of ingestion, and should be discouraged in water where water quality is relatively poor. While studies of water recreation have generally focused on infectious hazards, health risk assessments for chemical exposure during water recreation have also been reported (Hussain et al., 1998; Dor et al., 2003). The findings of the present study may also be useful in risk assessments of recreational exposure to mercury, polychlorinated biphenyls, and of recent interest, oil dispersants.

4.2.

Our findings in context

Two prior studies have attempted to quantify the volume of water ingested during water recreation. Water ingestion among surfers on the Oregon coast was investigated by Stone et al. (2008). In that study participants were also asked to estimate the volume of water swallowed while surfing with response options of a few drops, 1e3 teaspoons, the amount in a shot glass (2 ounces), or the amount in a small juice glass (4 ounces). Based on the self-reported estimates of ingestion volume and ingestion frequency, the authors estimated a median daily ingestion of 34.4 mL, and an arithmetic mean of 170.6 mL. Water ingestion among swimmers was estimated by Dufour and colleagues Dufour et al. (2006) at USEPA using CYA as a tracer of swimming pool water. Over the course of a 45-min swim, the 12 adults in the study swallowed an estimated 16 mL of pool water while the 41 children swallowed an average of 37 mL. The authors followed up with a larger scale

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 2 0 e2 0 2 8

study with involving 549 swimmers, the results of which have been presented at a conference by Evans et al. (2006). In that study children were found to swallow more water than adults (47 mL vs. 24 mL), and males swallowed more than females (37 mL vs. 27 mL), differences that were highly significant statistically. Twenty-five percent of the swimmers swallowed 85 mL or more, and some swallowed up to 280 mL. The Dufour estimate of 16 mL/45 min (or 21 mL/h) is intermediate between our mean estimate of 10 mL and the upper confidence estimate of 35 mL.

Appendix. Supplementary material

4.3.

references

Limitations

As originally designed, this research sought to generate a calculated volume of water ingestion for each participant, using HPLC measures of CYA in pool water and urine samples. We did not rely on the HPLC measures of CYA in urine because of their limited precision and the frequency of false positives and false negatives. We based our estimates of ingestion volume on the 27 (5%) of participants who provided selfreported estimates of ingestion, as well as urine samples measured by LC/MS. Because that subset of participants was relatively small, the point estimates of the translation factors (Table 5) are limited in their precision. Additionally, because of the small number of participants whose urine samples were analyzed with the gold standard method, we were unable to generate adult-specific and child-specific volumes of ingestion (based on CYA measures) for each level of self-reported ingestion. Another limitation is that the highest category of self-reported water ingestion was “a mouthful or more”. Some participants who reported this level of ingestion may have ingested “a mouthful” while others ingested more than that. The collection of more detailed data about the number of mouthfuls ingested may have generated better agreement between self-reported and CYA-based estimates of ingestion. Because <1% of surface water participants reported ingesting “a mouthful or more” any improvement in categorizing ingestion among that small subset would have minimal impact on the overall findings. Again, because of the relatively small number of participants for whom high quality urine CYA data was available, we were unable to compare the volume of water ingested by adults and by children. Future research could focus on this question. Although it is not known how representative study participants were of surface water recreators in general, there is no reason to believe that participants systematically over- or under-estimated the volume of water that they ingested.

5.

Conclusions

The mean volume of water ingested during limited-contact recreation activities, about 3.5e4 mL is about 35e40% of that observed during swimming (about 10 mL). The frequency of swallowing at least teaspoon of water during limited-contact recreation (about 1% of study participants) is about 1/50th the frequency observed during swimming in a pool (51% of participants). In surface waters with high concentrations of chemical or microbial hazards, avoiding capsizing should significantly reduce the percentage of paddlers who swallow

2027

a teaspoon or more of water. LC/MS/MS is preferred over HPLC for analyzing CYA in urine samples.

Supplementary data related to this article can be found online at doi:10.1016/j.watres.2010.12.006.

Ashbolt, N.J., Schoen, M.E., Soller, J.A., Roser, D.J., 2010. Predicting pathogen risks to aid beach management: the real value of quantitative microbial risk assessment (QMRA). Water Res. 44 (16), 4692e4703. Cantu, R., Evans, O., Kawahara, F.K., Wymer, L.J., Dufour, A.P., 2001a. HPLC determination of cyanuric acid in swimming pool waters using phenyl and confirmatory porous graphitic carbon columns. Anal. Chem. 73 (14), 3358e3364. Cantu, R., Evans, O., Magnuson, M.L., 2001b. Rapid analysis of cyanuric acid in swimming pool water by high performance liquid chromatography using porous graphite carbon. Chromatographia 53, 454e456. Colford Jr., J.M., Wade, T.J., Schiff, K.C., Wright, C.C., Griffith, J.F., Sandhu, S.K., Burns, S., Sobsey, M., Lovelace, G., Weisberg, S.B., 2007. Water quality indicators and the risk of illness at beaches with nonpoint sources of fecal contamination. Epidemiology 18 (1), 27e35. Cordell, H.K., Betz, C.J., Green, G.T., Mou, S., Leeworthy, V.R., Wiley, P.C., Barry, J.J., Hellerstein, D., 2004. Outdoor Recreation for 21st Century America: A Report to the Nation: The National Survey on Recreation and the Environment. Venture Publishing, Inc., State College, PA. Dewailly, E., Poirier, C., Meyer, F.M., 1986. Health hazards associated with windsurfing on polluted water. Am. J. Public Health 76 (6), 690e691. Dor, F., Bonnard, R., Gourier-Frery, C., Cicolella, A., Dujardin, R., Zmirou, D., 2003. Health risk assessment after decontamination of the beaches polluted by the wrecked ERIKA tanker. Risk Anal. Int. J. 23 (6), 1199e1208. Dufour, A.P., Evans, O., Behymer, T.D., Cantu, R., 2006. Water ingestion during swimming activities in a pool: a pilot study. J. Water Health 4 (4), 425e430. Dziuban, E.J., Liang, J.L., Craun, G.F., Hill, V., Yu, P.A., Painter, J., Moore, M.R., Calderon, R.L., Roy, S.L., Beach, M.J., 2006. Surveillance for waterborne disease and outbreaks associated with recreational watereUnited States, 2003e2004. MMWR Surveill. Summ. 55 (12), 1e30. Evans, O.M., Wymer, L.J., Behymer, T.D., Dufour, A.P., 2006. An Observational Study Determination of the Volume of Water Ingested During Recreational Swimming Activities. National Beaches Conference, Niagra Falls, NY. Fewtrell, L., Kay, D., Salmon, R.L., Wyer, M., Newman, G., Bowering, G., 1994. The Health Effects of Low-contact Water Activities in Fresh and Estuarine Waters. IWEM, pp. 97e101. Fleisher, J.M., Kay, D., 2006. Risk perception bias, self-reporting of illness, and the validity of reported results in an epidemiologic study of recreational water associated illnesses. Mar. Pollut. Bull. 52 (3), 264e268. Fleisher, J.M., Fleming, L.E., Solo-Gabriele, H.M., Kish, J.K., Sinigalliano, C.D., Plano, L., Elmir, S.M., Wang, J.D., Withum, K., Shibata, T., Gidley, M.L., Abdelzaher, A., He, G., Ortega, C., Zhu, X., Wright, M., Hollenbeck, J., Backer, L.C., 2010. The BEACHES study: health effects and exposures from non-point

2028

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 2 0 e2 0 2 8

source microbial contaminants in subtropical recreational marine waters. Int. J. Epidemiol. 64 (1), 16e21. Greenbaum, L., 2007. In: Kliegman, R.M., Behrman, R.E., Jenson, H. B., Stanton, B. (Eds.), Nelson Textbook of Pediatrics. Saunders, An Imprint of Elsivier. Hussain, M., Rae, J., Gilman, A., Kauss, P., 1998. Lifetime health risk assessment from exposure of recreational users to polycyclic aromatic hydrocarbons. Arch. Environ. Contam. Toxicol. 35 (3), 527e531. Idaho Department of Environmental Quality, 2004. Regional and National Overview of Use Attainability Analysis. http://www. deq.idaho.gov/water/assist_business/workshops/uaa_ regional_national_overview_uaa_workshop_handout.pdf (Accessed 24.11.10). Illinois Pollution Control Board, 2010. In the Matter Of: Water Quality Standards and Effluent Limitations for the Chicago Area Waterway System (CAWS) and the Lower Des Plaines River: Proposed Amendments to 35 Ill. Adm. Code 301, 302, 303 and 304. http://www.ipcb.state.il.us/COOL/External/CaseView. aspx?case¼13363 (Accessed 24.11.10). Kay, D., Fleisher, J.M., Salmon, R.L., Jones, F., Wyer, M.D., Godfree, A.F., Zelenauch-Jacquotte, Z., Shore, R., 1994. Predicting likelihood of gastroenteritis from sea bathing: results from randomised exposure. Lancet 344 (8927), 905e909. Lee, J.V., Dawson, S.R., Ward, S., Surman, S.B., Neal, K.R., 1997. Bacteriophages are a better indicator of illness rates than bacteria amongst users of a white water course fed by a lowland river. Water Sci. Technol. 35 (11e12), 165e170. McBride, G.B., Salmond, C.E., Bandaranayake, D.R., Turner, S.J., Lewis, G.D., Till, D.G., 1998. Health effects of marine bathing in New Zealand. Int. J. Environ. Health Res. 8, 173e189. Missouri Coalition for the Environment, 2010. Issues and Actions: UAAs. http://www.moenviron.org/uaa.asp (Accessed 24.11.10). Patel, K., Jones, K., 2007. Analytical method for the quantitative determination of cyanuric acid as the degradation product of sodium dichloroisocyanurate in urine by liquid chromatography mass spectrometry. J. Chromatogr. B. Analyt. Technol. Biomed. Life Sci. 853 (1e2), 360e363. Seyfried, P.L., Tobin, R.S., Brown, N.E., Ness, P.F., 1985. A prospective study of swimming-related illness. I. Swimmingassociated health risk. Am. J. Public Health 75 (9), 1068e1070. Smoker, M., Krynitsky, A.J., 2008. Interim Method for Determination of Melamine and Cyanuric Acid Residues in Foods Using LC-MS/MS: Version 1.0. Soller, J.A., Bartrand, T., Ashbolt, N.J., Ravenscroft, J., Wade, T.J., 2010. Estimating the primary etiologic agents in recreational freshwaters impacted by human sources of faecal contamination. Water Res. 44 (16), 4736e4747.

Stone, D.L., Harding, A.K., Hope, B.K., Slaughter-Mason, S., 2008. Exposure assessment and risk of gastrointestinal illness among surfers. J. Toxicol. Environ. Health A 71 (24), 1603e1615. Texas Commission on Environmental Quality, 2009. Overview of Proposed Standards Revisions. http://www.tceq.state.tx.us/ assets/public/permitting/waterquality/attachments/ stakeholders/overstand_jan09.pdf (Accessed 24.11.10). USEPA, 1983. Health Effects Criteria for Marine Recreational Waters. http://www.epa.gov/microbes/mrcprt1.pdf (Accessed 24.11.10). USEPA, 1984. Health Effects Criteria for Fresh Recreational Waters. http://www.epa.gov/nerlcwww/frc.pdf (Accessed 24.11.10). USEPA, 2003. Water Quality Standards for Kansas, Federal Register. http://www.federalregister.gov/articles/2003/07/07/03-16924/ water-quality-standards-for-kansas (Accessed 24.11.10). USEPA, 2009. Exposure Factor Handbook (External Review). http:// cfpub.epa.gov/ncea/cfm/recordisplay.cfm?deid¼209866 (Accessed 23.09.10). Utah Department of Environmental Quality, 2008. Use Attainability Analysis: Great Salt Lake. http://www. waterquality.utah.gov/WQS/20080715_UAA_GSL.pdf (Accessed 24.11.10). Wade, T.J., Calderon, R.L., Sams, E., Beach, M., Brenner, K.P., Williams, A.H., Dufour, A.P., 2006. Rapidly measured indicators of recreational water quality are predictive of swimming-associated gastrointestinal illness. Environ. Health Perspect. 114 (1), 24e28. Wade, T.J., Calderon, R.L., Brenner, K.P., Sams, E., Beach, M., Haugland, R., Wymer, L., Dufour, A.P., 2008. High sensitivity of children to swimming-associated gastrointestinal illness: results using a rapid assay of recreational water quality. Epidemiology 19 (3), 375e383. Wiedenmann, A., Kruger, P., Dietz, K., Lopez-Pila, J.M., Szewzyk, R., Botzenhart, K., 2006. A randomized controlled trial assessing infectious disease risks from bathing in fresh recreational waters in relation to the concentration of Escherichia coli, intestinal enterococci, Clostridium perfringens, and somatic coliphages. Environ. Health Perspect. 114 (2), 228e236. Wong, M., Kumar, L., Jenkins, T.M., Xagoraraki, I., Phanikumar, M.S., Rose, J.B., 2009. Evaluation of public health risks at recreational beaches in Lake Michigan via detection of enteric viruses and a human-specific bacteriological marker. Water Res. 43 (4), 1137e1149. Yoder, J.S., Blackburn, B.G., Craun, G.F., Hill, V., Levy, D.A., Chen, N., Lee, S.H., Calderon, R.L., Beach, M.J., 2004. Surveillance for waterborne-disease outbreaks associated with recreational watereUnited States, 2001e2002. MMWR Surveill. Summ. 53 (8), 1e22.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 2 9 e2 0 3 7

Available at www.sciencedirect.com

journal homepage: www.elsevier.com/locate/watres

Behavior of inorganic elements during sludge ozonation and their effects on sludge solubilization Pengzhe Sui a,*, Fumitake Nishimura a, Hideaki Nagare b, Taira Hidaka a, Yuko Nakagawa a, Hiroshi Tsuno a a

Department of Urban and Environmental Engineering, Kyoto University, Kyoto-Daigaku-Katsura, Nishikyo-ku, Kyoto 615-8540, Japan Division of Sustainability of Resources, Graduate School of Environmental Science, Okayama University, 3-1-1 Tsushima-naka, Okayama 700-8530, Japan b

article info

abstract

Article history:

The behavior of inorganic elements (including phosphorus, nitrogen, and metals) during

Received 12 August 2010

sludge ozonation was investigated using batch tests and the effects of metals on sludge

Received in revised form

solubilization were elucidated. A decrease of w50% in the ratio of sludge solubilization was

2 December 2010

found to relate to a high iron content 80e120 mgFe/gSS than that of 4.7e7.4 mgFe/gSS.

Accepted 8 December 2010

During sludge ozonation, the pH decreased from 7 to 5, which resulted in the dissolution of

Available online 16 December 2010

chemically precipitated metals and phosphorus. Based on experimental results and thermodynamic calculation, phosphate precipitated by iron and aluminum was more difficult to

Keywords:

release while that by calcium released with decrease in pH. The release of barium, manga-

Excess sludge reduction

nese, and chrome did not exceed 10% and was much lower than COD solubilization;

Sludge ozonation

however, that of nickel, copper, and zinc was similar to COD solubilization. The ratio of

Metals

nitrogen solubilization was 1.2 times higher than that of COD solubilization (R2 ¼ 0.85). Of the

Phosphorus

total nitrogen solubilized, 80% was organic nitrogen. Because of their high accumulation

Nitrogen

potential and negative effect on sludge solubilization, high levels of iron and aluminum in

Thermodynamic calculation

both sewage and sludge should be considered carefully for the application of the advanced sewage treatment process with sludge ozonation and phosphorus crystallization. ª 2010 Elsevier Ltd. All rights reserved.

1.

Introduction

Biological wastewater treatment processes have been employed to treat a wide variety of wastewater especially municipal wastewater. The disposal challenge of excess sludge generated in all types of biological treatment processes has been a considerable impetus to the development of excess sludge reduction technologies (Chu et al., 2009). The increasing amount of sludge generated and its high cost of treatment and disposal that accounts for more than 20%, even up to 60%, of the total plant operating costs (Foladori et al., 2010; Ginestet, 2007; LeBlanc et al., 2008; Perez-Elvira et al.,

2006; Spellman, 1997) have both necessitated the development of excess sludge reduction technologies. One of the sludge reduction strategies is based on the lysis-cryptic growth in which the biomass generated in the biological treatment process is first solubilized and then utilized again by microorganisms to increase the mineralization ratio of organics in raw wastewater. Among the approaches to sludge solubilization, ozonation is effective and has been demonstrated to attain high ratios of sludge solubilization (Deleris et al., 2002; Dytczak et al., 2007; Kamiya and Hirotsuji, 1998; Lee et al., 2005; Sakai et al., 1997; Yasui et al., 1996; Yasui and Shibata, 1994).

* Corresponding author. Tel.: þ81 75 3833352; fax: þ81 75 3833351. E-mail address: [email protected] (P. Sui). 0043-1354/$ e see front matter ª 2010 Elsevier Ltd. All rights reserved. doi:10.1016/j.watres.2010.12.011

2030

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 2 9 e2 0 3 7

After combining sludge ozonation with biological wastewater treatment, a decrease in the efficiency of phosphorus removal even failure was reported because phosphorus could not be taken out of the system with excess sludge (Sakai et al., 1997; Yasui et al., 1996). Phosphorus is a vital element for humans, animals, and plants. Modern agricultural production mainly depends on the input of mineral phosphorus fertilizer. However, phosphate rock mining can only be sustained for about 100 years (Steen, 1998). Phosphorus in sewage and industrial wastewater is not only a type of nutrient pollutant responsible for serious entrophication in enclosed water bodies (besides nitrogen) but also a phosphorus resource that should be recovered and recycled. In order to reduce the excess sludge generated and recover phosphorus from sewage, a new advanced sewage treatment process has been developed. This process incorporates sludge ozonation and phosphorous crystallization into A/O process (Nagare et al., 2008; Saktaywin et al., 2005; Saktaywin et al., 2006; Tsuno et al., 2008) or A/A/O process (Tsuno et al., 2008). In the sludge ozonation unit of this advanced sewage treatment process (Fig. S1), the inorganic elements such as nitrogen, phosphorous, and metals that are present in the excess sludge are released as a soluble fraction along with sludge solubilization. This release of phosphorus and nitrogen implies an increase in their loading rates and has the possibility to affect their removal ratios in A/O and/or A/A/O process. Another issue caused by the solubilization process with ozone is the accumulation of metals and other inert particles, which results from the decrease of the withdrawal of excess sludge and could be one of the obstacles for the sustainable long-term operation (Chu et al., 2009). The behavior of metals during sludge ozonation is one of the important factors affecting the accumulation of metals in the system. It was also found that more ozone was required to accomplish the same ratio of sludge solubilization of the excess sludge containing a high metal content (Saktaywin, 2005); however, no detailed investigation was reported. In this study, in order to discuss the relationship between the soluble organic release and the behavior of inorganic elements (such as phosphorus, nitrogen, and metals) during sludge ozonation, sludge solubilization by ozone was investigated by batch tests using different sources of excess sludge, which had different metal content. The effects of metal content on sludge solubilization and phosphorus release were elucidated.

2.

Material and methods

2.1.

Sludge sources

In this study, four types of sludge were used. Two of which came from actual sewage treatment plants (STP-H and STP-L) while the other two came from lab-scale sequencing batch reactor (SBR) systems (LAB-H and LAB-L), which were run as an A/O process for enhanced biological phosphorus removal (EBPR) by using the same types of reactors and operation modes as reported by Saktaywin et al. (2006).

The sludge from STP-H (A/O process with a sludge retention time (SRT) of 8.2 days) contained relatively high iron content, because iron salt was employed to enhance the phosphorus removal efficiency. The sludge from STP-L (conventional activated sludge process with SRT of 15 days) had relatively low metal content. The return sludge was taken as source sludge in both cases. The sludge from LAB-H had relatively high metal content as well as more metal species, whereas that from LAB-L contained a relatively low metal content and fewer metal species because the composition of artificial sewage that was used as influent differed (Table 1). For the sludge from LAB-H, metal content in the influent was similar to that in typical municipal wastewater (Asai et al., 2005). The SRT was about 10 days for both lab-scale SBR systems.

2.2.

Sludge ozonation apparatus

The ozone contact reactor was a cylindrical glass reactor with an effective volume of 6 L (inner diameter 10 cm, height 100 cm) and was operated in a semi-batch mode (Fig. S2) with a continuous supply of ozone gas and a batch feed of sludge. Ozone gas generated by a PSA ozonizer (SGA-01A-PSA4, SUMITOMO SEIMITSU Co., Ltd, Japan) was pretreated using ultra-purified water and a dehumidifier (DH-106-1, KOMATSU ELECTRONICS INC., Japan) to remove nitrogen oxide and moisture and was then continuously introduced into the contactor by using a diffuser equipped at the bottom. Ozone concentrations and flow rate were measured using two UV ozone monitors (Model-600, EBARA JITSUGYO Co., Ltd., Japan) and a flow meter (KOFLOC RK1350V, KOJIMA INSTRUMENTS INC., Japan), respectively. The excess and residual ozone gases were treated using two ozone decomposers (OZ-1000EB, EBARA JITSUGYO Co., Ltd., Japan) before releasing them into the atmosphere.

Table 1 e Compositions of artificial sewage with SBR-B and SBR-S. Components SBR-B SBR-S Elements SBR-B SBR-S [mg/L] [mg/L] [mg/L] [mg/L] 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18

D(þ)-Glucose Polypepton CH3COONa (NH4)2SO4 KH2PO4 NaHCO3 NaCl KCl CaCl2$2H2O MgCl2$6H2O FeCl2$4H2O AlCl3$6H2O ZnCl2 MnCl2$4H2O BaCl2 NiCl2$6H2O CuCl2$2H2O CrCl3$6H2O

167 83 58 56 22 194 128 13 84 84 6.4 14 1.5 2.5 0.91 1.6 0.64 0.61

111 56 39 56 22 83 56 8.4 28 34 1.4 e e e e e e e

COD TN TP

300 21 5.0

200 18 5.0

Na K Ca Mg Fe Al Zn Mn Ba Ni Cu Cr

120 14a 40a 12a 1.8 1.6 0.72 0.68 0.60 0.40 0.24 0.12

56 12a 25a 6.6a 0.43a 0.044a e e e e e e

a Note: Taking tap water used for dilution into consideration.

2031

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 2 9 e2 0 3 7

2.3.

Experimental conditions and analytical methods

The samples were periodically collected during sludge ozonation and nitrogen gas was introduced into the reactor to purge the residual ozone gas after ozonation. Experimental conditions of the sludge ozonation in batch tests and the sampling periods are described in Table 2. After sampling, all soluble samples were obtained by immediate filtration through a 1.0 mm glass filter (Whatman GF/B 1821-47, England) except for samples of soluble metals, which were pretreated by filtration through a 0.45 mm glass filter (DISMIC 25AS045AS, ADVANTEC group, Japan). Analytical items included chemical oxygen demand (COD), phosphate, soluble total phosphorus (S-TP), total phosphorus (TP), ammonia, nitrite, nitrate, soluble total nitrogen (S-TN), total nitrogen (TN), suspended solid (SS), volatile suspended solid (VSS), soluble metal (S-metal), and total concentration with non-filtrated samples of each metal (T-metal). All analytical methods were based on Standard Methods (APHA et al., 2005). S-metal and T-metal were measured using the HP-4500 ICP-MS apparatus (APHA et al., 2005; Asai et al., 2005; Stewart and Olesik, 1998a,b; Toda et al., 2000). The specific ozone consumption (mgO3/gSS) was calculated using equation (1) based on the initial SS and ozone concentrations of the inlet and outlet, which were recorded every 10 s by using a data logger system. The integral equation (1) was calculated using the step length of 1 min. 8 t 9, < Z  =   CO3in  CO3out  QO3 $dt  VH  CO3out RO3 consumption ¼ : ; 0

ðSSinitial $VÞ

ð1Þ

where: RO3 consumption e specific ozone consumption [mgO3/ gSS]; CO3 in e inlet ozone concentration [mgO3/L]; CO3 out e outlet ozone concentration [mgO3/L]; QO3 e ozone gas flow rate [L/min]; SSinitial e initial SS concentration [gSS/L]; V e sludge volume [L]; VH e volume of head space of ozone contactor [L] (VH ¼ 1.5 L); t e ozonation time [min]. The sludge solubilization degree was calculated on the basis of the decrease of each particulate matter (SS, COD, nitrogen, phosphorus, and each metal) as defined in equation (2).

Solubilization degree   Initial particulate conc:  Current particulate conc: ¼ ð2Þ Initial particulate conc: The phosphorus fractionation method was employed to study the interaction between metals and phosphorus. The basic procedures of the phosphorus fractionation method (STS method) include cold perchloride acid (PCA) extraction, ethanol extraction, hot acid extraction, and hot alkali extraction (Mino et al., 1983a,b). In this study, only the cold PCA extraction was employed for the phosphorus fractionation as reported by de Haas et al. (2000).

2.4.

Normalization method of sludge solubilization

In order to compare the solubilization ratios (based on particulate COD) of different sources of sludge, normalization was carried out according to the following procedures. First, ratios of sludge solubilization at the specific ozone consumptions of 20, 30, 40, 50, and 60 mgO3/gSS were calculated using the regression curve of the experimental results. Second, the solubilization ratio of the sludge from LAB-L was used as reference (100%) and the others were expressed as its percentages at each point of specific ozone consumption. Third, the sludge solubilization degree was compared by using average value of above percentages of different types of sludge.

3.

Results and discussion

3.1.

Total metals content in the sludge

T-metal and S-metal of four types of source sludge before ozonation were measured. Particulate metal (P-metal, the difference between T-metal and S-metal) content in sludge (mgP-Metal/gSS) is shown in Fig. 1. Extremely high iron content (about 80 mgFe/gSS) was found in the STP-H sludge compared to the other three types of sludge that contained about 10 mgFe/gSS. The metal content in the sludge from STP-L was lower than that in the sludge from STP-H, while the metal content in the sludge from LAB-H was higher than that in the sludge from LAB-L except for magnesium and potassium,

Table 2 e Experimental conditions of sludge ozonation in batch tests. Items Initial sludge concentration [gSS/L] MLVSS/MLSS [%] Initial sludge volume [L] Inlet ozone concentration [mgO3/L] Ozone gas flow rate [L/min] Ozone gas flux in contactor [m3/(m2 s)] Mixer speed [rpm] Circulation pump flow ratea [ml/min] Circulation direction Final ozone consumption rate [mgO3/gSS] Ozonation time [h] Sampling period a Note: Measured by tap water.

SBR-S 1.87 78.2% 6.0 30 0.4 850 200 420 From bottom to top 120 5 Every hour

SBR-B 8.53 81.2% 6.0 30 0.4 850 200 420 125 24 Every 3 h

STP-I 4.11 84.5% 6.0 80 0.2 425 200 420 From top to bottom 150 5.5 Every hour

STP-H 3.97 67.7% 6.0 80 0.2 425 200 420 175 7 Every hour

2032

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 2 9 e2 0 3 7

LAB-L

LAB-H

STP-L

LAB-L

STP-H

P-Metal content i n sludge [mg/gSS] ]

P-M etal & P content in sludge [mg/gSS] ]

LAB-H

STP-L

STP-H

5

90 80 70 60 50 40 30 20 10

4.5 4 3.5 3 2.5 2 1.5 1 0.5 0

0 P

K

Mg

Ca

Al

Cr

Fe

Mn

Ni

Cu

Zn

Ba

Fig. 1 e Particulate metals and P content in source sludge before ozonation.

which are associated with the ingestion and release as well as accumulation of phosphorus in PAOs (Somiya et al., 1987). Metals present in the sludge in high level (ranging 5e20 mgM/ gSS) were calcium, aluminum, and iron except for magnesium and potassium. The metal content of the sludge from LAB-H and the STP-L was similar. The sludge from LAB-L had the highest phosphorus content of 55.8 mgP/gSS. In the lab-scale A/O process (LAB-L and LAB-H), iron, aluminum, manganese, zinc, and barium tended to be highly accumulated (Fig. S3).

a

LAB-L

Variations in the ratios of sludge solubilization (%, based on particulate COD) with specific ozone consumption (mgO3/gSS) are shown in Fig. 2(a). The sludge was solubilized with ozone consumption, but the solubilization efficiency (the slope of the curve) decreased more in the sludge from LAB-H, STP-L, and STP-H than that in the sludge from LAB-L. About 20% sludge solubilization was accomplished with specific ozone consumption w55 and w75 mgO3/gSS for the sludge from LAB-L and LAB-H, respectively, which were cultured with artificial sewage. However, it was w90 and w120 mgO3/gSS for the sludge from STP-L and STP-H, respectively, which came from actual sewage treatment plant. Using the sludge from actual STP and a semi-batch operational mode, Sievers et al. (2004) indicated that 9.2e13.3% of the sludge was solubilized with a specific ozone consumption of 30e60 mgO3/gSS, while the research conducted by Bougrier et al. (2006) observed a ratio of sludge solubilization of 20e25% with an ozone dosage of 100e160 mgO3/gSS. Using the sludge cultured by artificial sewage and a semi-batch operational mode, Saktaywin et al. (2005) obtained a ratio of sludge solubilization of 30% with a specific ozone consumption of 30e40 mgO3/gSS. Manterola et al. (2008) investigated sludge solubilization with ozone in a continuous operational mode by using sludge from actual STP and reported that 20e30% of the sludge was solubilized with an ozone dosage of 25e35 mgO3/ gSS. The results obtained in this study were in the range shown in the above literature. With certain specific ozone consumption (mgO3/gSS), sludge solubilization is only related to the composition of sludge (organic, metals, etc.), which could be oxidized by ozone. The

Sludge solubi lization rate ] [mgC OD/mgCOD ]

Sludge solubilization

b

LAB-H

STP-L

STP-H

40% 35% 30% 25% 20% 15% 10% 5% 0% 0

50 100 150 200 Specific ozone consumption [mgO3/gSS]

Cultured in this study Cultured in former study 1.2 Normal ized sludge] solubili zation rate ]

3.2.

other properties of sludge, such as floc size and floc strength, will affect reaction rate, but not the ratio of sludge solubilization at the specific ozone consumption. Furthermore, metal accumulation in the advanced sewage treatment process was a concern in this study. Thus, the correlation between the normalized sludge solubilization and metal content in the sludge was studied. In our previous study (Saktaywin, 2005), sludge solubilization by ozone was investigated using four types

Actual STP in this study Actual STP in former study

1 0.8 0.6 0.4 0.2 0 0

50

100

150

Particulate iron content in sludge [mgFe/gSS]

Fig. 2 e Sludge solubilization during ozonation (based on particulate COD). (a) Variation in ratios of sludge solubilization with specific ozone consumption. (b) Variation in normalized ratio of sludge solubilization with particulate iron content in sludge.

2033

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 2 9 e2 0 3 7

a

of sludge from actual STP as well as sludge cultivated by a labscale SBR using the same artificial sewage as used in LAB-L. Their data were normalized using the sludge from the lab-scale SBR as the standard and following the same procedures and were combined with that obtained in this study. Statistical analysis of the relationship between sludge solubilization and content of each metal in the sludge showed that iron had the highest correlation coefficient of 0.80. Iron consumed ozone for the oxidation and production of radicals and was present in high content in the sludge. The relationship between iron content in the sludge and the normalized ratios of sludge solubilization is shown in Fig. 2(b). Sludge solubilization was found to become more difficult when iron content in the sludge was higher. The normalized ratio of sludge solubilization decreased to about 50% with an iron content in the sludge of 80e120 mgFe/gSS compared to that of 4.7e7.4 mgFe/gSS. Based on this result, the iron content in the actual sewage and sludge should be considered carefully for the design and operation of an advanced sewage treatment process.

LAB-H

STP-L

STP-H

1:1

Soluble Organic N [%]]

35% 30%

Inorganic N

25% 20% 15% 10%

y = 0.8323x

5%

R = 0.9821

Organic N

2

0% 0%

10%

20%

30%

40%

50%

Released S-TN [%]

b

LAB-L

LAB-H

STP-L

STP-H

Solubil ization ratio ] based on P-N [%] ]

40%

3.3. Behavior of nitrogen and phosphorus during ozonation Nitrogen behavior (expressed as percentages of TN) during sludge ozonation is shown in Fig. 3. Nitrogen was found to be solubilized and 20e30% of TN was released with a specific ozone consumption of 100 mgO3/gSS. As shown in Fig. 3(a), organic nitrogen accounted for about 80% of S-TN for different types of sludge similar to that reported by Zhao et al. (2007), while soluble inorganic nitrogen mainly existed as nitrate (Fig. S4). The ratio of nitrogen solubilization was 1.2 times higher (R2 ¼ 0.85) than that of COD as shown in Fig. 3(b). However, released soluble COD (S-COD, mgCOD/L) was more than 12 times higher than released S-TN (mgN/L), which could contribute to nitrogen removal for the denitrification limited by the carbon source in the influent. It was reported that the denitrification rate improved up to 20% due to additional carbon release by ozonation at an ozone dosage of 0.08 gO3/ gTSS (Dytczak et al., 2007). The behavior of phosphorus during sludge ozonation is shown in Fig. 4. The variation patterns of phosphorus solubilization were different among the different sources of sludge as shown in Fig. 4(a). Phosphorus was released continuously and reached w25% with consumption of ozone for the sludge from LAB-L and STP-L, which contained the low metal content. For the sludge from LAB-H, the soluble phosphorus concentration decreased at the beginning of ozonation, which might have resulted from chemical precipitation, and then started to increase and finally reached the same release ratio as that of LAB-L and STP-L. The sludge from STP-H, which contained extremely high iron content, had a very small ratio of phosphorus solubilization and even less than 10% at the end of ozonation. The phosphate ratio in the released phosphorus, which was easily recovered as crystals, was nearly 68% (Fig. 4(b)). The relationship between the solubilization ratios of phosphorus and COD is shown in Fig. 4(c). A linear relationship for the sludge from LAB-L and STP-L was found with a slope of 0.85. The sludge from LAB-H followed this regression line at the final stage of ozonation. However, the corresponding slope for the sludge from STP-H was as low as 0.49.

LAB-L 40%

30%

y = 1.175x 2 R = 0.8462

20% 10% 0% 0%

10%

20%

30%

40%

Solubilization ratio based on P-COD [%]

Fig. 3 e Nitrogen behavior during sludge ozonation. (a) Organic nitrogen and inorganic nitrogen in released soluble nitrogen. (b) Relationship between solubilization ratios of particulate COD and nitrogen.

The variation in the behavior of phosphorus release among different sources of sludge might be due to the chemical precipitation of metals such as Fe, Al, and Ca. In order to elucidate the effects of metals on the phosphate release during sludge ozonation, phosphorus fractionation was carried out in the next two tests, i.e., ozonation of the sludge from STP-L and STP-H; moreover, pH was measured because it has the greatest influence on chemical precipitation. Results of the fractionation (Fig. S6) clearly showed that part of the soluble phosphate came from chemically precipitated phosphorus. At the same time, pH decreased from w7.0 to w5.0 during sludge ozonation similar to that reported by Deleris et al. (2002). Although the detailed mechanisms of pH decrease are still unclear, this may be one of the reasons that led to the dissolution of precipitated phosphate. The effect of chemical precipitation on the phosphorus release and the interaction between the metals and phosphorus will be discussed in detail later.

3.4.

Behavior of metals during sludge ozonation

The comparison of the solubilization ratio based on each particulate metal and COD is shown in Fig. 5. There were four groups of metals based on their behavior. In the first group, all soluble fractions of potassium and magnesium increased during ozonation. They were associated with the release and uptake of phosphate in PAOs (Somiya et al., 1987) and their release might relate to the phosphorus release along with cells disintegration. The

2034

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 2 9 e2 0 3 7

a Par ticulate P decrease rate [%]

LAB-L

LAB-H

STP-L

STP-H

30% 25% 20% 15% 10% 5% 0% -5% -10%

b

0

50 100 150 200 Specific ozone consumption [mgO3 /gSS]

Phosphate ratio in TP [%] ]

LAB-L

LAB-H

STP-L

STP-H

1:1

40% 35% Non-phosphate

30% 25% 20% 15% 10% 5% 0% 0%

Phosphate y = 0.6777x 2 R = 0.9415 10%

20%

30%

40%

50%

S-TP ratio in TP [%]

c

LAB-L

LAB-H

STP-L

STP-H

a decrease in the beginning followed by an increase for the sludge from LAB-L, LAB-H, and STP-L; however, it showed slight initial decrease for the sludge from STP-H, which had high iron content. After the increase of the soluble calcium concentration, almost the same slope was observed in the linear portion of the curve except for LAB-L, which could have been due to a decrease in pH and the dissolution of chemically precipitated metals. The solubilization of iron, aluminum, and calcium was affected by the interaction with phosphate. Although the initial content of other metals in the sludge was relatively low, their behavior was different. The release of soluble barium, manganese, and chrome (the third group) did not exceed 10% and was much lower than COD solubilization while that of nickel, copper, and zinc (the fourth group) was similar to COD solubilization. These results indicate that barium, chrome, and manganese also have high accumulation potential in an advanced sewage treatment process, although their levels in both sewage and sludge were relatively low. Similar to magnesium, the release of soluble nickel and copper was lower than that of COD for the sludge from STP-H while was higher than that for the sludge from STP-L and LAB-H. The corresponding metal content in the sludge was also similar to that of magnesium as shown in Fig. 1.

Solubili zation ratio ] based on P-P [%] ]

40% 30%

y = 0.8478x 2 R = 0.8402

3.5.

20% 10% y = 0.4909x 2 R = 0.9913

0% -10% 0%

10%

20%

30%

40%

Solubilization ratio based on P-COD [%]

Fig. 4 e Phosphorus behavior during sludge ozonation. (a) Variation of particulate phosphorus release with specific ozone consumption. (b) Phosphate and non-phosphate in soluble phosphorus. (c) Relationship between solubilization ratios of particulate COD and phosphorus (linear regression result y [ 0.8478x was based on the data of LAB-L and STP-L, while y [ 0.4909x for STP-H only).

release ratio of potassium was 3.3 times higher than that of COD. Magnesium was also solubilized at a ratio that was 2.6 times higher than that of COD for the sludge from LAB-H and STP-L but at a low ratio of 0.82 for the sludge from LAB-L and STP-H. The difference in behavior was thought to relate to the difference in magnesium content in sludge (Fig. 1), which was lower for the sludge from LAB-H and STP-L than that from LAB-L and STP-H. In the second group, the solubilization ratios of iron, aluminum, and calcium were much lower than that of COD while their content in sludge was relatively high (Fig. 1). This resulted in the highest accumulation potential in the advanced sewage treatment process. The soluble iron release did not exceed 5% except for the sludge from STP-L, which had the lowest iron content. Aluminum was not released until the P-COD solubilization ratio reached 0.15 after which it gradually released in accordance with COD solubilization. The release of soluble calcium showed the similar behavior with

Thermodynamic calculation

In order to elucidate the interaction between the metals and phosphate during sludge ozonation, Visual MINTEQ (Allison et al., 1991), a free software for thermodynamic calculation, was employed to calculate the percentages of precipitated metals and phosphorus under equilibrium conditions in the pH range 5e7 by using measured metal concentrations. The results are shown in Fig. 6. The calculation results indicated that almost all the aluminum, iron, and manganese could be precipitated in the pH range of calculation after attaining chemical equilibrium. However, calcium precipitated at pH 7 but dissolved completely at pH 5. Similar to calcium, precipitated chrome, copper, and zinc could dissolve again but their precipitated percentages varied for the different types of sludge depending on their content in sludge. No barium, nickel, potassium, and magnesium precipitated according to the calculation results. The above results correlated well with measured results for iron, aluminum, and calcium, which were the main contributors to the chemical precipitation of phosphorus because their content in both sewage and sludge was much higher than that of other metals as shown in Table 1 and Fig. 1. For the metals with a relatively low content in the sludge, calculated behavior of manganese, nickel, zinc, and copper was in accordance with that measured; however, for barium and chrome, the calculated behavior did not agree with that measured. Low measured release ratios of barium and chrome were supposed to result from bio-adsorption effects of the activated sludge. During sludge ozonation, calcium phosphate could be precipitated at the beginning without a decrease in pH and dissolved again with a decrease in pH. The initial sharp decrease and final similar release ratio of phosphorus for the sludge from LAB-H as shown in Fig. 4(a) could be explained by the relatively high initial concentrations of soluble phosphate and calcium and without the decrease in pH at the beginning.

LAB-L 1.4 1.2 1 0.8 0.6 0.4 0.2 0

LAB-H

STP-L

b

STP-H

Mg

y = 3.2885x 2 R = 0.6016

LAB-L

0.1

LAB-L

Fe

0.2

0.3

0.4 0.3 0.2

y = 0.8179x 2 R = 0.9328

0.1

0.4

LAB-H

STP-L

0

d

STP-H

0.2 1:1

0.15 0.1 0.05

LAB-L 0.15

0.05

LAB-H

STP-L

0.4

0

f

STP-H

1:1

0 -0.2 -0.4 -0.6 0

0.1

0.2

0.3

0.4

LAB-H 0.08 0.06 0.04 0.02 0 -0.02 -0.04 -0.06

LAB-H

STP-L

h

STP-H

1:1

0.1 0.05 0 -0.05 0

0.1

0.2

0.3

LAB-H

Solubilization rate ] based on P-Ni [-] ]

0.7 0.6 0.5 0.4 0.3 0.2 0.1 0

STP-H

y = 1.8613x 2 R = 0.8865 y = 0.6884x 2 R = 0.9873

STP-H

0.1 1:1

0.08 0.06 0.04 0.02 0

0.1

0.2

0.3

0.4

Solubilization rate based on P-COD [-]

j

LAB-H

Cu 0.6

STP-L

STP-H

y = 1.8167x 2 R = 0.929

0.5 0.4 0.3 0.2

y = 0.8871x 2 R = 0.9593

0.1 0

0

0.1 0.2 0.3 0.4 Solubilization rate based on P-COD [-] LAB-H

Zn Solubilization rate] based on P-Zn [-] ]

STP-L

STP-L

0.12

0

Solubilization rate ] based on P-Cu [-] ]

LAB-H

Ni

STP-H

0.1 0.2 0.3 0.4 Solubilization rate based on P-COD [-]

Cr

0.4

STP-L

1:1

0

Solubili zation rate ] based on P-C r [-] ]

Solubili zation rate ] based on P-Mn [-] ]

Mn 0.15

0.1 0.2 0.3 0.4 Solublization rate based on P-COD [-]

Ba Solubilization rate ] based on P-Ba [-] ]

LAB-L 0.2

Solubi lization rate ] based on P-C a [-] ]

1:1

Solubilization rate based on P-COD [-]

k

STP-H

0.10

Solubilization rate based on P-COD [-]

i

STP-L

0.00 0.1 0.2 0.3 Solublization rate based on P-COD [-]

Ca

-0.1

LAB-H

0.20

0 0

g

0.1 0.2 0.3 0.4 Solubilization rate based on P-COD [-]

Al Solubilization rate ] based on P-Al [-] ]

Solubilization rate ] based on P-Fe [-] ]

0.25

-0.8

STP-H

y = 2.6208x 2 R = 0.9259

0.5

Solubilization rate based on P-COD [-]

e

STP-L

0 0

c

LAB-H

0.6

Solubil ization rate ] based on P-Mg [-] ]

K Solubilization rate ] based on P-K [-] ]

a

STP-L

0

0.1 0.2 0.3 0.4 Solubilization rate based on P-COD [-]

STP-H

0.4 y = 0.9459x 2 R = 0.6768

0.3 0.2 0.1 0 0

0.1 0.2 0.3 0.4 Solubilization rate based on P-COD [-]

Fig. 5 e Comparison of solubilization ratios based on particulate metals and COD.

2036

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 2 9 e2 0 3 7

Al Fe

LAB-L

Ca Mn

Cr P

Cu Zn

120 80

120 Fe

Precipitated percentage [%]

Precipitated percentage [%]

100

Ca Al

60 40

Mn

P

20 0 4.5

Zn Cu Cr 5.5

6 pH [-] Ca Mn

6.5

7

Cr P

Cu Zn

Zn Ca

60

Cu

40

P

20 4.5

5

5.5

Al Fe

6 pH [-] Ca Mn

6.5

7

Cr P

7.5

Cu Zn

120 Precipitated percentage [%]

Precipitated percentage [%]

Cu Zn

Cr

80

STP-H

100 Al Fe Mn

Cu Ca

60

P

40 20 0 4.5

Cr P

Al Fe Mn

100

7.5

120 80

Ca Mn

0 5

Al Fe

STP-L

Al Fe

LAB-H

5.5

6 pH [-]

6.5

7

80

P

Al Fe Mn Cr

40

7.5

0 4.5

Ca

Cu

60 20

Cr Zn 5

100

Zn 5

5.5

6 pH [-]

6.5

7

7.5

Fig. 6 e Precipitated metals change with pH according to thermodynamic calculation.

Phosphorus precipitated by iron and aluminum was more difficult to release. For the sludge from STP-H, phosphate was precipitated by iron mainly and almost no release of iron phosphate partly led to a relatively low phosphate release. The chemically precipitated phosphate should be mainly released from precipitated calcium phosphate (Fig. S6). High concentration of iron or aluminum in the actual sewage should be taken into account, not only because of the high accumulation potential in the sludge but also because of their effects on phosphorus recovery in the advanced sewage treatment process. Introducing additional iron or aluminum into the biological treatment process to further enhance phosphorus removal should be avoided in the advanced sewage treatment process because of its effect on phosphorus recovery efficiency.

4.

was more difficult to release while that by calcium released with a decrease in pH. 5) A high content of iron and aluminum was present in the sludge. Their release during ozonation was much lower than COD solubilization, which led to a high accumulation potential. 6) For metals having relatively low content in the sludge, the release of barium, manganese, and chrome did not exceed 10% and was much lower than COD solubilization. However, the release of nickel, copper, and zinc was similar to COD solubilization. From these results, it was concluded that the high content of iron and aluminum in sewage and sludge should be taken into consideration carefully for the actual application of the new advanced sewage treatment process with sludge ozonation and phosphorus crystallization.

Conclusions

The behavior of inorganic elements during sludge ozonation and their effects on sludge solubilization were investigated in this study. The main conclusions are as follows: 1) A low ratio of sludge solubilization was found to relate to a high iron content, which decreased to about 50% for the sludge having an iron content of 80e120 mgFe/gSS compared to that of 4.7e7.4 mgFe/gSS. 2) The ratio of nitrogen solubilization was 1.2 times higher than that of COD (R2 ¼ 0.85). Of the solubilized nitrogen, 80% was organic nitrogen. 3) The pH decreased from 7 to 5 during sludge ozonation, which resulted in the dissolution of chemically precipitated metals and phosphate. 4) Based on the experimental results and the thermodynamic calculation, phosphate precipitated by iron and aluminum

Appendix. Supplementary data Supplementary data associated with this article can be found in the online version, at doi:10.1016/j.watres.2010.12.011.

references

Allison, J.D., Brown, D.S., Novo-Gradac, K.J., 1991. MINTEQA2/ PRODEFA2, A Geochemical Assessment Model for Environmental Systems: Version 3.0 User’s Manual. U.S. EPA, Athens, Georgia, USA. APHA, AWWA, WEF, 2005. Standard Methods for the Examination of Water & Wastewater, 21st ed. American Public Health Association, Washington, DC.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 2 9 e2 0 3 7

Asai, T., Kaneko, T., Sega, K., Kato, M., 2005. The fate of trace elements at sewage treatment plants in Nagoya city. Journal of Japan Sewage Works Association 42, 85e95 (in Japanese). Bougrier, C., Albasi, C., Delgenes, J.P., Carrere, H., 2006. Effect of ultrasonic, thermal and ozone pre-treatments on waste activated sludge solubilisation and anaerobic biodegradability. Chemical Engineering and Processing 45, 711e718. Chu, L.B., Yan, S.T., Xing, X.H., Sun, X.L., Jurcik, B., 2009. Progress and perspectives of sludge ozonation as a powerful pretreatment method for minimization of excess sludge production. Water Research 43, 1811e1822. de Haas, D.W., Wentzel, M.C., Ekama, G.A., 2000. The use of simultaneous chemical precipitation in modified activated sludge systems exhibiting biological excess phosphate removal. Part 2: method development for fractionation of phosphate compounds in activated sludge. Water SA 26, 453e466. Deleris, S., Geaugey, V., Camacho, P., Debellefontaine, H., Paul, E., 2002. Minimization of sludge production in biological processes: an alternative solution for the problem of sludge disposal. Water Science and Technology 46, 63e70. Dytczak, M.A., Londry, K.L., Siegrist, H., Oleszkiewicz, J.A., 2007. Ozonation reduces sludge production and improves denitrification. Water Research 41, 543e550. Foladori, P., Andreottola, G., Ziglio, G., 2010. Sludge Reduction Technologies in Wastewater Treatment Plants. IWA Publishing, London, UK. Ginestet, P., 2007. Comparative Evaluation of Sludge Reduction Routes. IWA Publishing, London, UK. Kamiya, T., Hirotsuji, J., 1998. New combined system of biological process and intermittent ozonation for advanced wastewater treatment. Water Science and Technology 38, 145e153. LeBlanc, R.J., Matthews, P., Richard, R.P., 2008. Global Atlas of Excreta, Wastewater Sludge, and Biosolids Management: Moving Forward the Sustainable and Welcome Uses of a Global Resource. United Nations Human Settlements Programme (UN-HABITAT), Nairobi, Kenya, p. 608. Lee, J.W., Cha, H.Y., Park, K.Y., Song, K.G., Ahn, K.H., 2005. Operational strategies for an activated sludge process in conjunction with ozone oxidation for zero excess sludge production during winter season. Water Research 39, 1199e1204. Manterola, G., Uriarte, I., Sancho, L., 2008. The effect of operational parameters of the process of sludge ozonation on the solubilisation of organic and nitrogenous compounds. Water Research 42, 3191e3197. Mino, T., Matsuo, T., Kawakami, T., 1983a. Studies on phosphorus composition and phosphorus metabolism in activated sludge (1). Journal of Japan Sewage Works Association 20, 28e36 (in Japanese). Mino, T., Matsuo, T., Kawakami, T., 1983b. Studies on phosphorus composition and phosphorus metabolism in activated sludge (2). Journal of Japan Sewage Works Association 20, 22e29 (in Japanese). Nagare, H., Tsuno, H., Saktaywin, W., Soyama, T., 2008. Sludge ozonation and its application to a new advanced wastewater

2037

treatment process with sludge disintegration. Ozone: Science & Engineering 30, 136e144. Perez-Elvira, S.I., Nieto Diez, P., Fdz-Polanco, F., 2006. Sludge minimisation technologies. Reviews in Environmental Science and Biotechnology 5, 375e398. Sakai, Y., Fukase, T., Yasui, H., Shibata, M., 1997. An activated sludge process without excess sludge production. Water Science and Technology 36, 163e170. Saktaywin, W., 2005. Development of advanced sewage treatment process with excess sludge reduction and phosphorus recovery. Ph. D Thesis, Kyoto University, Kyoto, p. 213. Saktaywin, W., Tsuno, H., Nagare, H., Soyama, T., Weerapakkaroon, J., 2005. Advanced sewage treatment process with excess sludge reduction and phosphorus recovery. Water Research 39, 902e910. Saktaywin, W., Tsuno, H., Nagare, H., Soyarna, T., 2006. Operation of a new sewage treatment process with technologies of excess sludge reduction and phosphorus recovery. Water Science and Technology 53, 217e227. Sievers, M., Ried, A., Koll, R., 2004. Sludge treatment by ozonation e evaluation of full-scale results. Water Science and Technology 49, 247e253. Somiya, I., Tsuno, H., Nishikawa, M., 1987. Behavior of phosphorus and metals in the anaerobic-oxic activated sludge process. In: Advances in Water Pollution Control (IAWPRC): Biological Phosphorus Removal from Wastewater, Rome, Italy, pp. 321e324. Spellman, F., 1997. Wastewater Biosolids to Compost. Technomic Publishing Company, Lancaster, PA, USA. Steen, I., 1998. Phosphorus availability in the 21st century: management of a non-renewable resource. Phosphorus and Potassium 217, 25. Stewart II, , Olesik, J.W., 1998a. Steady state acid effects in ICP-MS. Journal of Analytical Atomic Spectrometry 13, 1313e1320. Stewart II, , Olesik, J.W., 1998b. Transient acid effects in inductively coupled plasma optical emission spectrometry and inductively coupled plasma mass spectrometry. Journal of Analytical Atomic Spectrometry 13, 843e854. Toda, M., Kaneko, T., Takeda, S., Saiki, T., 2000. Elemental analysis of sewage and sewage sludge samples using inductively coupled plasma mass spectrometry. Journal of Japan Sewage Works Association 37, 155e165 (in Japanese). Tsuno, H., Arakawa, K., Kato, Y., Nagare, H., 2008. Advanced sewage treatment with ozone under excess sludge reduction, disinfection and removal of EDCs. Ozone: Science & Engineering 30, 238e245. Yasui, H., Nakamura, K., Sakuma, S., Iwasaki, M., Sakai, Y., 1996. A full-scale operation of a novel activated sludge process without excess sludge production. Water Science and Technology 34, 395e404. Yasui, H., Shibata, M., 1994. An innovative approach to reduce excess sludge production in the activated-sludge process. Water Science and Technology 30, 11e20. Zhao, Y.X., Yin, J., Yu, H.L., Han, N., Tian, F.J., 2007. Observations on ozone treatment of excess sludge. Water Science and Technology 56, 167e175.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 3 8 e2 0 4 8

Available at www.sciencedirect.com

journal homepage: www.elsevier.com/locate/watres

Oxidation of 2,4-dichlorophenol and 3,4-dichlorophenol by means of Fe(III)-homogeneous photocatalysis and algal toxicity assessment of the treated solutions Roberto Andreozzi a, Ilaria Di Somma a, Raffaele Marotta a,*, Gabriele Pinto b, Antonino Pollio b, Danilo Spasiano a a

Universita` degli Studi di Napoli “Federico II”, Facolta` di Ingegneria, Dipartimento di Ingegneria Chimica, p.le V. Tecchio, 80, 80125 Napoli, Italy b Universita` degli Studi di Napoli “Federico II”, Facolta` di Scienze, Dipartimento di Biologia, 80125 Napoli, Italy

article info

abstract

Article history:

Chlorophenols are used worldwide as broad-spectrum biocides and fungicides. They have

Received 6 September 2010

half-life times in water from 0.6 to 550 h and in sediments up to 1700 h and, due to their

Received in revised form

numerous origins, they can be found in wastewaters, groundwaters or soils. Moreover,

14 December 2010

chlorophenols are not readily biodegradable.

Accepted 16 December 2010 Available online 22 December 2010

Recently, classic Advanced Oxidation Processes (AOP) have been proposed for their abatement in an aqueous solution. This paper investigates the oxidation of 2,4-dichlorophenol and 3,4-dichlorophenol, at starting concentrations of 6.1 $ 10
5 mol L
1, in aqueous solutions

Keywords:

through Fe(III)/O2 homogeneous photocatalysis under UV light (303 O 366 nm). The Fe(III)/O2

Chlorophenols

homogeneous photocatalysis is less expensive than using H2O2 due to the capability of Fe(III) to

Fe(III)/O2 homogeneous

produce OH radicals, if irradiated with an UVA radiation, and of oxygen to re-oxidize ferrous

photocatalysis

ions to ferric ones when dissolved in solution. The results show that the best working

Kinetics

conditions, for both compounds, are found for pH ¼ 3.0 and initial Fe(III) concentration equal to

Algal toxicity

1.5$10
4 mol L
1 although the investigated oxidizing system can be used even at pH close to 4.0 but with slower abatement kinetics. Toxicity assessment on algae indicates that treated solutions of 2,4-dichlorophenol are less toxic on algae Pseudokirchneriella subcapitata if compared to not treated solutions whereas in the case of 3,4-dichlorophenol only the samples collected during the runs at 20 and 60 min are capable of inhibiting the growth of the adopted organism. The values of the kinetic constant for the photochemical re-oxidation of iron (II) to iron (III) and for HO attack to intermediates are evaluated by a mathematical model for pH range of 2.0e3.0 and initial Fe(III) concentrations range of 1.5 $ 10
5e5.2 $ 10
4 mol L
1. ª 2010 Elsevier Ltd. All rights reserved.

1.

Introduction

The use of chlorinated phenols in chemical industries as solvents and intermediates for pesticides and dyes manufacture is reported (Ahlborg et al., 1980; Freiter, 1979) in the literature along with their presence in river, groundwater and seawater

(Puig et al., 1997; Abrahamsson and Xie, 1983; Wegman and Van der Broek, 1983). Due to their toxicity many of the species belonging to this class are considered of great health and environmental concern and are listed by US EPA as priority pollutants (Hayward, 1999; Keith and Telliard, 1979). In the present paper the attention is focused on the adoption of iron

* Corresponding author. Tel.: þ390817682968; fax: þ390815936936. E-mail address: [email protected] (R. Marotta). 0043-1354/$ e see front matter ª 2010 Elsevier Ltd. All rights reserved. doi:10.1016/j.watres.2010.12.016

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 3 8 e2 0 4 8

homogeneous photocatalysis e an innovative AOP, less expensive than ozonation and H2O2 photolysis, which is based on the use of oxygen e even that contained in an air stream e as oxidant and Fe(III) as homogeneous catalyst e for the removal from aqueous solutions of two dichlorophenols. Previous investigations have demonstrated that the use of solar radiation (or proper lamps simulating it), Fe2þ or Fe3þ ions e at concentrations as low as those of treated waters (98/83/EC) e and oxygen makes possible the removal of organic pollutants from water for pH lower than 4.0 (Andreozzi et al., 2006). In particular the present research is devoted to the oxidative removal from water of 2,4-(2,4-DCP) and 3,4-dichlorophenols (3,4-DCP). 2,4-DCP is used in the manufacturing of the herbicide 2,4dichlorophenoxyacetic acid (2,4-D) (EP0509518A1). It has been reported that after the application of the latter in the agricultural sites, 2,4-DCP is formed in the environment as the result of the abiotic and biotic transformations of 2,4-D. The presence of 2,4-DCP has been also documented in chlorinated water (Canosa et al., 2005) and in flue gases from inceneration plants (Blumenstock et al., 2001; Heeb et al., 1995). Moreover 2,4-DCP has been recognized as a species capable of causing some changes in the endocrine system of fish and rats (Handan et al., 2005; Zhang et al., 2005). No indications were found in the literature about the production and possible uses of 3,4-DCP. However, its presence has been reported in river water by Merriman and coworkers (Merriman et al., 1991) thus raising the problem of its environmental fate and impact. Moreover, toxicity studies on reproductive effects revealed that 3,4-DCP disrupted the sperm acrosome (Seyler et al., 1984). Due to their low biodegradability, some works with an application of AOP to the removal of chlorophenols from water are found in the specialised literature. For example, some authors (Scott and Ollis, 1995) have described AOPbased treatments of chlorophenols under a variety of conditions. Pera-Titus et al. (Pera-Titus et al., 2004) reviewed the oxidation of chlorophenols by means of different AOP. Classic AOP which make use of expensive oxidants (O3, H2O2, O3/UV, H2O2/UV, Fe(II)/H2O2 and Fe(III)/H2O2/UV) have been reported also by others to be capable of destroying chlorophenols (Chu et al., 2005; Al Momani et al., 2004a; Pera-Titus et al., 2004; Benitez et al., 2001; Hautaniemi et al., 1998) as an alternative to the adoption of traditional methods of removal from water as GAC adsorption (Pirbazari et al., 1991). However, in most of the studies the attention was focused on the monochlorine derivatives as the para isomer (4-chlorophenol) (Catalkaya et al., 2003; Untea et al., 2006; Kuo and Wu, 2010), ortho isomer (2-chlorophenol) (Xu et al., 2003; Poulopoulos and Philippopoulos, 2004; Hong et al., 2008), on a species such as 2,4-DCP (Al Momani et al., 2004b, Pandiyan et al., 2006; Terashima et al., 2006; Bayarri et al., 2007, Li et al., 2007), 2,4,6-TCP (Benitez et al., 2000; Chu and Law, 2003; Kansal et al., 2007; Vijayan et al., 2009). Only two papers that investigate the ozonation of 3,4-DCP e along that of other dichlorophenols e had been found (Qiu et al., 2004; Emery et al., 2003). The effect of iron concentration and pH on the reactivity of studied chlorophenols at the adopted experimental conditions and on their mineralization and dechlorination degree is evaluated in the present work.

2039

Algae form the basis for all food webs in aquatic ecosystems. Because of their ecological importance, algal measurements are fundamental components of water quality monitoring programs. The toxicity towards algae of the samples taken from the reactor during an iron homogeneous photocatalytic treatment of chlorophenol-containing aqueous solutions is also assessed. A description of the system behaviour is also attempted through the use of a model already proposed by the authors for the photo-oxidation of some organic substances (Andreozzi et al., 2006; Andreozzi and Marotta, 2004).

2.

Experimental

All the experiments with UV lamp were carried out at 298 K in a batch annular glass jacketed reactor with an outer diameter of 6.5 cm and a height of 40 cm wrapped with an aluminium foil and filled with 0.280 L of solution. At the top, the reactor had two inlets for feeding reactants and an outlet for withdrawing samples. The reactor was equipped with a 125 W (power input) highpressure lamp (by Helios Italquartz), mainly emitting at 305, 313 and 366 nm (manufacturer’s data), enclosed in a glass sleeve, immersed in the solution in the center of the reactor. The reactor was open to air, without bubbling, and the solution was mixed with a magnetic stirrer placed at the bottom. Each run was carried out by firstly switching on the lamp without solutions in the reactor and, after the achievement of the maximum lamp radiant power (10 min), by rapidly feeding the solution into the reactor. The photon flow of the lamp at 305 nm was 1.47 $ 10
6 einstein s
1 (I (305), determined through hydrogen peroxide photolytic experiments (Kuhn et al., 2004)), at 313 nm was 1.56 $ 10
6 einstein s
1 (I (313), determined by valerophenone actinometry (Zepp et al., 1998)) and at 366 nm was 4.10 $ 10
6 einstein s
1 (I (366), measured using a UV radiometer Delta Ohm HD 9021). The solutions containing iron salts, mainly Fe(ClO4)3, were freshly prepared in dark for each run by dissolving the ferric salt in a proper volume of doubly distilled water at a desired pH. Since literature data indicated that EC50 for both the substrates were in the order of concentrations of 10e25 mg/l (Yoshioka and Sato, 1986; Elpabarawy et al., 1988; Seibert et al., 2002; Kunihiro et al., 2004), being interested to understand if the treatment of a water containing one or both of them by means of the proposed system could ensure a safe reduction of toxicity, starting concentrations of about 8e10 mg/l were fixed for the experimental runs. All glassware and reactor (photolysis tubes) were cleaned with hydrochloric acid and washed several times with bidistilled water before the use. Ferric solutions were immediately used to avoid the formation of high molecular weight hydrolysis products (Knight and Sylva, 1975). The pH was controlled with perchloric acid and sodium hydroxide addition; all measurements were performed by means of a pH-meter (Orion 420Aþ). The concentrations of 2,4-DCP and 3,4-DCP solutions at different reaction times were evaluated by HPLC analysis. For this purpose, the HPLC apparatus (Agilent 1100) was equipped with a UVeVIS detector (l ¼ 210 nm) and a Synergi Max-RP 80A

2040

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 3 8 e2 0 4 8

Table 1 e Effect of untreated and treated 2,4-DCP and 3,4DCP mixtures (EC50) and confidence intervals (C.I.) on P. subcapitata. EC50 (95%C.I.) (10
3 mol L
1) 2,4-DCP Untreated 20 min 60 min 120 min

0.006 0.022 0.610 0.048

3,4-DCP Untreated 20 min 60 min 120 min

>0.06 0.038 (0.003e0.0043) 0.022 (0.0208e0.0229) >0.06

1.00

0.80

0.60

0.40

0.20

0.00 0

3.

Results and discussion

3.1.

Investigations on 2,4-dichlorophenol

In Fig. 1 are shown the results of preliminary runs of the photo-oxidation of 2,4-DCP collected by means of the system Fe(III)/hn/O2 in which different ferric salts are used to prepare

(0.0056e0.0073) (0.0205e0.0229) (0.5670e0.6441) (0.0445e0.0638)

the reacting solutions. The diagrams, reported in the Fig. 1, indicate that no appreciable degradation of the substrate was observed during the direct photolysis (without ferric ions addition). Moreover, they also indicate that ferric perchlorate is capable of ensuring the highest reactivity to the Fe(III)/hn/O2 system. The influence of the counterion could be explained if one considers that inorganic anions, i.e., nitrates, sulfates and chlorides, may exert some complexing effect on Fe(III) with
the formation of complexes such as FeCl2þ, FeClþ 2 , Fe(SO4)2 , 2þ etc., thermodynamically more favoured than Fe(OH) , thus reducing the concentration of photolyzable aquocomplex Fe (OH)2þ (Millero et al., 1995). The effect of initial Fe(III) concentration on the photodegradation of 2,4-DCP is shown in Fig. 2 when all the solutions are prepared by dissolving in water proper amounts of ferric perchlorate.

[2,4-DCP]/[2,4-DCP]o

column, using a mobile phase (50/50 acetonitrile/buffered aqueous solution), flowing at 1.0 $ 10
3 L min
1. The buffered aqueous eluent was prepared with 1.0 $ 10
2 L phosphoric acid (85% by weight), 2.5 $ 10
2 L methanol in 1 L HPLC water. Each run was repeated two times and single average values were reported in the diagrams. Percentage standard deviations of about 3.5% (not reported in the figure) were obtained for each point in the diagrams. Total organic carbon (TOC) was monitored by a TOC analyzer (Shimadzu 5000 A). 2,4-DCP, 3,4-DCP, iron salts (perchlorate, sulfate, nitrate and chloride), sodium hydroxide, perchloric acid and tert-butanol were purchased from SigmaeAldrich and used as received. The 2,4-DCP and 3,4-DCP toxicity towards Pseudokirchneriella subcapitata were assessed using either EPA medium or BBM (Bold Basal Medium). The results obtained (data not shown) indicate that salt concentration in the medium does not influence the toxicity of the selected compounds and that in 96-h end-point experiments algae grew at a faster growth rate with BBM. For this reason all the experiments were performed with the latter. For the algal bioassays, each solution of the selected compounds was prepared by dissolving a known quantity of the selected compound in water to have a final concentration of 6.0 $ 10
5 mol L
1. The toxicity test was based on the measurement of the growth inhibition of the green unicellular alga P. subcapitata, strain UTEX 1648. Algal inocula corresponding to 10,000 cells/ml from laboratory cultures in mid exponential phase were grown in cell culture Cluster (12 well plate) containing each well 5.0 $ 10
3 L of BBM and the tested compound (2,4-DCP or 3,4-DCP) at different concentrations 3.0 $ 10
5 mol L
1, 1.5 $ 10
6 mol L
1, (6.0 $ 10
5 mol L
1, 7.0 $ 10
7 mol L
1 and 3.0 $ 10
7 mol L
1). The cell culture clusters were incubated at 297 K under continuous illumination at a light intensity of 90 m einstein s
1 m
2. The tests were carried out in triplicate and in axenic conditions. A series of controls containing only BBM and the algal inocula, were also prepared. Although chlorides are contained in the medium used for algal growth, some blank experiments were done to assess the effect of chlorides formed during each run on the toxicity assessments. The results of these runs indicated that in the range of chloride concentrations observed in the present work no effects are evidenced on the algal growth. The algal growth was followed after 72 h from the addition of the compounds by measuring the in vivo chlorophyll fluorescence using a fluorometer (Turner, model Aquaflor, Turner Designs CA, USA). The concentrations that cause 50% of effect (EC50) were determined by using the linear interpolation method. Toxicity tests were carried out in triplicate and repeated two times. To test the degree of normality of the data reported in Table 1, the arithmetic mean and the confidence intervals were computed for both experiments.

20

40

60

80

100

Time (min)

Fig. 1 e Influence of the nature of ferric salts upon the photodegradation of 2,4-DCP in presence of air (no bubbling) at pH [ 3.0 and T [ 25  C. [2,4-DCP]o [ 6.1 $ 10L5 mol LL1, [Fe(III)]o [ 1.0 $ 10L5 mol LL1. Only UV without Fe (III) (C); iron sulfate (-); iron perchlorate (D); iron nitrate (:); iron chloride (A).

2041

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 3 8 e2 0 4 8

1.00

1.00

0.80

[2,4-DCP]/[2,4-DCP]o

[2,4-DCP]/[2,4-DCP]o

0.80

0.60

0.40

0.60

0.40

0.20 0.20

0.00 0

0.00

2

4

6

8

10

12

Time (min) 5

10

15

20

25

30

Time (min)

Fig. 2 e Photodegradation of 2,4-DCP at varying initial concentration of [Fe(ClO4)3], in presence of air (no bubbling). [2,4-DCP]o [ 6.1 $ 10L5 mol LL1. pH [ 3.0. T [ 25  C. [Fe(III)]o: 3.2 $ 10L5 mol LL1 (-); 5.0 $ 10L5 mol LL1 (C); 1.1 $ 10L4 mol LL1 (:); 1.5 $ 10L4 mol LL1 (A), 2.0 $ 10L4 mol LL1 (D).

As expected, based on the results obtained in previous investigations (Andreozzi et al., 2006), the system reactivity increases with increasing initial iron (III) concentration. For example, for an iron concentration increasing from 3.2 $ 10
5 mol L
1 to 2.0 $ 10
4 mol L
1 the time required to gain a complete conversion of 2,4-DCP reduces from 90 to 15 min nearly. Due to the fact that in the applications chlorophenols may be found in contaminated water in which other pollutants may be present, some of them being volatile (Abrahamsson and Ekdahl, 1996; Deinzer et al., 1978), an investigation has been undertaken on the behaviour of the system when oxygen is not present in the solution. To this purpose, some photolytic runs on aqueous solutions, preliminarily purged with a helium flow, were carried out (Fig. 3). It is noteworthy to observe that also when no oxygen is present in the reacting solution, it is possible to achieve an almost complete removal of the substrate by properly increasing the initial iron (III) concentration. These results are easily explained if one considers that for the proposed oxidative system the main oxidant species is represented by OH radicals forming directly from aquocomplex Fe(OH)2þphotolysis, the role of oxygen being only that of re-oxidizing Fe2þ ions. Following this approach, it is clear that if an excess of iron(III) is used in a single run a complete removal of the substrate is observed. The results collected under UV lamp irradiation at varying pH values with contaminant (2,4-DCP) and Fe(III) starting concentrations, respectively, of 6.0 $ 10
5 mol L
1 and 3.0 $ 10
5 mol L
1 are shown in Fig. 4. A change in the reactivity was observed in the pH range of 2.0e4.0 with a maximum for pH ¼ 3.0. This result may be easily justified considering that in the investigated pH range the concentration of photolyzable

Fig. 3 e Photodegradation of 2,4-DCP at varying initial concentration of [Fe(ClO4)3], in presence of a helium flow. [2,4-DCP]o [ 6.1 $ 10L5 mol LL1. pH [ 3.0. T [ 25  C. [Fe(III)]o: 8.8 $ 10L5 mol LL1 (:), 5.2 $ 10L4 mol LL1 (-).

aquocomplex Fe(OH)2þ shows a maximum approximately at a value of 3.0 with the rate of OH radical generation following the same trend. However, other causes may be responsible for the very low value of reactivity of the system at pH ¼ 4.0. In fact, a simple calculation indicates that for the adopted iron (III) concentration, at pH ¼ 4.0, the product [Fe(III)]$[OH
]3

1.00

0.80

[2,4-DCP]/[2,4-DCP]o

0

0.60

0.40

0.20

0.00 0

20

40

60

80

100

Time (min)

Fig. 4 e Influence of pH upon the photodegradation of 2,4DCP, in presence of air (no bubbling). [2,4-DCP]o [ 6.0 $ 10L5 mol LL1. T [ 25  C. [Fe(III)]o [ 3.0 $ 10L5 mol LL1. pH: 2.0 (:), 3.0 (C), 3.5 (-), 4.0 (A). Note: At pH ¼ 4.0, after the mixture preparation, the solution became yellow-coloured; no precipitation of solid particles was observed during the run.

2042

3.2.

Investigations on 3,4-dichlorophenol

Very similar results were collected by submitting to the oxidation process aqueous solutions containing 3,4-DCP. Some of these results are shown in Fig. 6. A mineralization degree of about 90% was observed for long treatment times as those adopted for 2,4-DCP, whereas the percentage of initial

[TOC]/[TOC]o or [Cl]/[Cl]o (%)

100

80

60

40

1.0

100

0.8

80

0.6

60

0.4

40

0.2

20

0.0

[TOC]/[TOC]o or [Cl]/[Cl]o (%)

exceeds the value reported in the literature for the solubility constant Kps ¼ 1.1 $ 10
36 (mol L
1)3 (Liu and Liptak, 1997). That is, for the adopted experimental conditions the precipitation of iron (III) hydroxide was expected at pH ¼ 4.0, although after the preparation, the solution just appeared as yellow-coloured and e at the least for the duration of the photolytic run e no presence of solid particles was observed. A possible explanation for this behaviour seems to be the slow formation of polynuclear hydroxocomplexes, as intermediates during the transition of ferric ions from soluble species to solid precipitates, which changes the speciation of iron (III) considered at the other pHs (Stumm and Morgan, 1996). TOC (Total Organic Carbon) measurements were also performed in order to evaluate the degree of mineralization of contaminant load achieved during the runs. Measured TOC values and the degrees of dechlorination are reported in Fig. 5 for an experimental run ([Fe(III)]o ¼ 2.0 $ 10
4 mol L
1) in presence of UV irradiation. For a reaction time of 15 min, when a complete conversion of the substrate was observed (Fig. 2), about 50% of initial organic carbon was mineralized. However, for prolonged oxidation times a degree of mineralization close to 90% was reached. Moreover, for prolonged reaction times dechlorination degrees lower than 40% were measured. These results suggest that some unidentified highly chlorinated chemical intermediates, refractory to mineralization also for long treatment times, are formed in the course of the photolytic process.

[ 3,4- DCP]/[3,4-DCP] o

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 3 8 e2 0 4 8

0 50

100

150

200

250

Time (min)

Fig. 6 e Photodegradation of 3,4-DCP in presence of air (no bubbling). [3,4-DCP]o [ 6.0 $ 10L5 mol LL1. [Fe(III)]o [ 2.0 $ 10L4 mol LL1. T [ 25  C. pH [ 3.0. 3,4-DCP (C); TOC removal (:) and dechlorination percentage (A).

chlorine present in the molecule which was converted to chloride resulted equal to about 45%. The obtained results undoubtedly indicate that, also in the case of 3,4-DCP, the presence of highly chlorinated organic species, refractory to mineralization, among the reaction intermediates and products is expected.

3.3.

Toxicity assessments on algae

The EC50 was determined using the ICPIN program (NorbergKing, 1993) which calculates the EC values by linear interpolation and 95% confidence intervals by the bootstrap method. 2,4-DCP and 3,4-DCP effects on P. subcapitata growth are reported in Table 1. As can be seen, 2,4-DCP showed a toxicity higher than 3,4-DCP, which did not inhibit the algal growth also at the highest concentration tested in the present study (6.0 $ 10
5 mol L
1). In the case of 2,4-DCP, treated samples were less inhibitory than the untreated one, and the highest reduction of toxicity (about two orders of magnitude with respect to the control) was observed after a 60-min treatment. On the other hand, for 3,4-DCP, the samples collected at 20 and 60 min caused higher inhibitory effects on the algal growth, compared to the untreated compound whereas only after 120 min of treatment the solution was less toxic than the starting sample.

20

4. 0 0

50

100

150

200

250

Time (min) Fig. 5 e TOC removal (:) and dechlorination percentage (-) in presence of air (no bubbling). [2,4DCP]o [ 6.0 $ 10L5 mol LL1. [Fe(III)]o [ 2.0 $ 10L4 mol LL1. T [ 25  C. pH [ 3.0.

Model development

An attempt to model the behaviour of both the species during the photo-oxidation runs was done for pH values: 2.0, 3.0 and 3.5. A set of differential material balance equations was thus written, based on the reactions shown in Table 2, for each of the species participating to the process. A complete description of the adopted mathematical model is reported elsewhere (Andreozzi et al., 2006).

2043

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 3 8 e2 0 4 8

Table 2 e List of reactions considered in the proposed model. hv

FeOH2þ /Fe2þ þ HO hv Fe2þ þ O2 /Fe3þ þ O
 2 
þ HO2 4H þ O þ 2 H 2þ 3þ Fe þ HO2 /Fe þ H2 O2 2Hþ

3þ Fe2þ þ O
 þ H2 O2 2 / Fe Si þ HO /intermediates



(r1) (r2) (r3) (r4) (r5)

k1 KO2
=HO2 ¼ 10
4:8 mol L
1 kFeðIIÞ=HO2 ¼ 1:2  106 L mol
1 s
1 kFeðIIÞ=O
2 ¼ 1:0  107 L mol
1 s
1

(r6)

kSi/OH

Intermediates þ HO /

(r7)

2þ þ O2 Fe3þ þ O
 2 /Fe Fe3þ þ HO2 /Fe2þ þ O2 þ Hþ Fe2þ þ H2 O2 /Fe3þ þ HO: þ HO
HO þ H2 O2 /H2 O þ HO2 Fe3þ þ HO
4FeOH2þ FeOH2þ þ HO
4FeðOHÞþ 2 Fe2þ þ HO /Fe3þ þ OH


(r8) (r9) (r10) (r11) (r12) (r13) (r14)

i
III 
3þ 
 h Fe ¼ Fe þ FeOH2þ þ FeðOHÞ2þ

k2;4
DCP=HO ¼ 6:0  109 L mol
1 s
1

k3;4
DCP=HO ¼ 6:0  109 L mol
1 s
1 k2i k2/2,4-DCP k2/3,4-DCP kFeðIIIÞ=O
2 ¼ 1:5  108 L mol
1 s
1 kFeðIIIÞ=HO2 ¼ 1:0  103 L mol
1 s
1 kFeðIIÞ=H2 O2 ¼ 76 L mol
1 s
1 kH2 O2 =HO ¼ 2:7  107 L mol
1 s
1 KFeðIIIÞ=HO
¼ 1011:8 L mol
1 KFeðOHÞ2þ=HO
¼ 1010:49 L mol
1 kFe(II)/HO ¼ 5.0  108 L mol
1

(1)

d½Fe2þ  dt

h

(4)



2þ 
2þ   + + ¼
Fþ kFeðIIÞ=O2 , Fe2þ ,½O2  þ k FeðIIÞ=H2 O2 , Fe ,½H2 O2  þ kFeðIIÞ=HO , Fe ,½HO þ
2þ 
KHO+  , Hþ2 , Fe , HO+2 þ þ kFeðIIÞ=HO+2 þ kFeðIIÞ=O
+ 2 ½    KHO+

 2 + , Fe3þ , HO+2
k , þ
kFeðIIIÞ=O
+ FeðIIIÞ=HO þ 2 2 ½H 

(5)




   ¼ þF
kFeðIIÞ=O2 , Fe2þ ,½O2 
kFeðIIÞ=H2 O2 , Fe2þ ,½H2 O2 
kFeðIIÞ=HO+ , Fe2þ ,½HO+ þ KHO+

 , Hþ2 , Fe2þ , HO+2 þ
kFeðIIÞ=HO+2 þ kFeðIIÞ=O
+ 2 ½    KHO+

 2 + , Fe3þ , HO+2 þ k , þ kFeðIIIÞ=O
+ FeðIIIÞ=HO þ 2 2 ½H 

(6)

i 2 gFe3þ ,g2HO

3þ 

þ FeðOHÞþ , Fe , HO
2 ¼ KFeðIIIÞ=ðHO Þ ,KFeðOHÞ2 =HO
, gðFeOHÞþ2 (2)

 
g 3þ ,gHO

3þ 
Fe , HO
FeOH2þ ¼ KFeðIIIÞ=ðHO
Þ , Fe gFeOH2þ



Zimbron and Reardon (2005), Benitez et al., (2001) Assumed in this work In this work In this work Rush and Bielski (1985) Rush and Bielski (1985) Xie et al. (2000) Buxton et al. (1988) Flynn (1984) Flynn (1984) Zhang and Bartlett (1999)

The disappearance of the substrate (2,4-DCP or 3,4-DCP) was accounted for through the integration of the following equation: d½Si  ¼
kSi =HO ,½Si ,½HO  dt

d½FeðIIIÞ dt

In this work Martell et al. (1993) Rush and Bielski (1985) Rush and Bielski (1985)

½HO SS ¼

(3)


 d½H2 O2  ¼
kFeðIIÞ=H2 O2 , Fe2þ ,½H2 O2 
kH2 O2 =HO+ ,½H2 O2 ,½HO+ þ dt   
 KHO+
þ kFeðIIÞ=HO+2 þ kFeðIIÞ=O
+ ,
þ2 , Fe2þ , HO+2 (7) 2 H The concentrations of HO and HO2 radicals were calculated assuming a steady-state hypothesis:

!
 F þ kFeðIIÞ=H2 O2 , Fe2þ ,½H2 O2   
 kSi =HO ,½Si  þ kH2 O2 =HO ,½H2 O2  þ k2i , ½Si 0
½Si  þ kFeðIIÞ=OH , Fe2þ

1 0
2þ    ,½H O ,½HO  þk þ2 k , Fe  ,½O
 H2 O2 =HO 2 2 2 Fe =O2 ss C B HO2 ss ¼ @ KHO =O
 



    A kFe2þ =HO , Fe2þ þ kFe3þ =HO , Fe3þ þ H2þ 2 kFe2þ =O
 , Fe2þ þ kFe3þ =O
 , Fe3þ 2 2 2 2 ½ 

(8)

(9)

2044

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 3 8 e2 0 4 8

where

F¼

3  h   i 1 X ðFeHO2þ Þ ðFeOH2þ Þ
I F , 1
exp
2; 3,L,3li , FeOH2þ , V i¼1 li li

(10) accounts for the radiation absorption by the photolyzable aquocomplex Fe(OH)2þ. The model does not take into account any effect exerted by chloride ions formed during the runs. That is, it can be supposed that during the experiments the complexing capability of formed chlorides can only negligibly affect the reactivity of the system. This can be explained by considering that, for the dechlorination degree achieved during the process

(about 35%) and the starting concentrations of ferric ions and of the substrates used in each run, the ratio between the concentration of iron and chlorides was generally higher and quite different than that occurring when ferric chlorides salt was used as a starting material. All kinetic and equilibrium constants necessary to run the model were known with the exception of k1 for the photo-reoxidation of Fe2þ to Fe3þ (reaction r2) and k2i for the HO radical attack to unknown intermediates. For the kinetic constant of attack of HO radicals to 3,4-DCP, although no data were found from the literature, the same value, as the one adopted for the 2,4-dichloro isomer, was used. On the other hand, in an UV/H2O2 photolytic run, in which both the substrates were present, a similar reactivity was recorded (data not shown).

0.12

0.25

b

a 0.20

(10 mol L-1 )

-3

-1

(10 -3 mol L )

0.08

0.15

0.10

0.04

0.05

0.00

0.00 0

2

4

6

8

0

10

4

8

12

16

Time (min)

Time (min) 0.08

0.06

c

d 0.06

-3

(10 mol L-1)

(10 -3 mol L-1 )

0.04

0.04

0.02

0.02

0.00

0.00 0

10

20

30

Time (min)

40

50

0

20

40

60

80

100

Time (min)

Fig. 7 e Comparison between experimental and calculated concentrations for some photodegradation runs of 2,4-DCP at pH [ 3.0 and T [ 25  C. Values concentration of 2,4-DCP: experimental (C), calculated (d). Calculated values concentration of Fe (III)T (—) and Fe(II)T ($d$). (a) [2,4-DCP]o [ 6.4 $ 10L5 mol LL1; [Fe(III)]o [ 2.0 $ 10L4 mol LL1; [O2]o [ 2.1 $ 10L4 mol LL1. (b) [2,4DCP]o [ 6.8 $ 10L5 mol LL1; [Fe(III)]o [ 1.0 $ 10L4 mol LL1; [O2]o [ 2.1 $ 10L4 mol LL1. (c) [2,4-DCP]o [ 5.7 $ 10L5 mol LL1; [Fe (III)]o [ 5.0 $ 10L5 mol LL1; [O2]o [ 2.1 $ 10L4 mol LL1. (d) [2,4-DCP]o [ 6.8 $ 10L5 mol LL1; [Fe(III)]o [ 3.2 $ 10L5 mol LL1; [O2]o [ 2.1 $ 10L4 mol LL1.

2045

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 3 8 e2 0 4 8

0.25

0.12

a

b

0.20

(10 -3mol L )

-1

-1

(10 -3 mol L )

0.08 0.15

0.10

0.04 0.05

0.00

0.00 0

10

20

30

0

4

8

Time (min)

12

16

20

Time (min)

0.06

0.05

c

d

0.5

-1

-3

-3

-3

-1

(10 mol L )

0.03

0.02

0.3

0.02 0.2

0.01

0.00

0.1

0.00 0

10

20

30

-1

[3,4-DCP] (10 mol L )

0.4 0.04

[Fe(III)] [Fe(II)] (10 mol L )

0.04

0.0 0

2

Time (min)

4

6

Time (min)

Fig. 8 e Comparison between experimental and calculated concentrations for some photodegradation runs of 3,4-DCP at pH [ 3.0 and T [ 25  C. Values concentration of 3,4-DCP: experimental (C), calculated (d). Calculated values concentration of Fe (III)T (—) and Fe(II)T ($d$). (a) [3,4-DCP]o [ 4.9 $ 10L5 mol LL1; [Fe(III)]o [ 2.0 $ 10L4 mol LL1; [O2]o [ 2.1 $ 10L4 mol LL1. (b) [3,4DCP]o [ 6.1 $ 10L5 mol LL1; [Fe(III)]o [ 1.0 $ 10L4 mol LL1; Helium bubbling. (c) [3,4-DCP]o [ 5.6 $ 10L5 mol LL1; [Fe (III)]o [ 5.5 $ 10L5 mol LL1; Helium bubbling. (d) [3,4-DCP]o [ 4.7 $ 10L5 mol LL1; [Fe(III)]o [ 5.2 $ 10L4 mol LL1; Helium bubbling.

Table 3 e Values of unknown parameters estimated at pH [ 3.0 along with statistical indices. [Fe(III)]o (mol L
1)
4

2.00 $ 10 1.00 $ 10
4 5.00 $ 10
5 3.20 $ 10
5 5.00 $ 10
4

[Fe(III)]o (mol L
1)
4

2.00 $ 10 1.00 $ 10
4 5.20 $ 10
4 1.00 $ 10
4 5.50 $ 10
5

[2,4-DCP]o (mol L
1)
5

[O2]a (mol L
1)
4

k1 (L mol
1 s
1)

k2/2,4-DCP (L mol
1 s
1) 9

7

6.40 $ 10 6.80 $ 10
5 5.70 $ 10
5 6.80 $ 10
5 6.50 $ 10
5

2.6 $ 10 2.6 $ 10
4 2.6 $ 10
4 2.6 $ 10
4 0

0.96  0.04

1.14 $ 10  9.70 $ 10

[3,4-DCP]o (mol L
1)

[O2]a(mol L
1)

k1 (L mol
1 s
1)

k2/3,4-DCP (L mol
1 s
1)


5

4.90 $ 10 5.10 $ 10
5 4.70 $ 10
5 6.10 $ 10
5 5.60 $ 10
5


4

2.1 $ 10 2.1 $ 10
4 0 0 0

0.88  0.15

a Calculated according to the Henry’s law (H ¼ 4.38 $ 104 atm. (Perry and Green, 1997)).

9

1.90 $ 10  1.40 $ 10

8

s% 0.19 1.10 1.61 0.67 1.45

s% 1.08 2.30 1.03 1.08 1.26

2046

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 3 8 e2 0 4 8

Table 4 e Values of unknown parameters estimated at pH [ 2.0 and 3.5 along with statistical indices. [Fe(III)]o (mol L
1)
4

1.00 $ 10 1.50 $ 10
4 3.00 $ 10
5 1.80 $ 10
5

[Fe(III)]o (mol L
1)
4

1.00 $ 10 1.50 $ 10
4 3.00 $ 10
5 1.50 $ 10
5

[2,4-DCP]o (mol L
1)
5

pH

[O2]a (mol L
1)

k1 (L mol
1 s
1)


4

k2/2,4-DCP (L mol
1 s
1) 9

8

5.38 $ 10 6.20 $ 10
5 5.51 $ 10
5 5.46 $ 10
5

2.0 2.0 3.5 3.5

2.6 $ 10 2.6 $ 10
4 2.6 $ 10
4 2.6 $ 10
4

0.82  0.13

1.91 $ 10  3.51 $ 10

0.42  0.04

4.62 $ 109  3.51 $ 108

[3,4-DCP]o (mol L
1)

pH

[O2]a (mol L
1)

k1 (L mol
1 s
1)

k2/3,4-DCP (L mol
1 s
1)


5

5.87 $ 10 5.30 $ 10
5 6.04 $ 10
5 5.99 $ 10
5

2.0 2.0 3.5 3.5


4

2.6 $ 10 2.6 $ 10
4 2.6 $ 10
4 2.6 $ 10
4

9

8

0.83  0.26

5.74 $ 10  2.55 $ 10

0.39  0.07

2.43 $ 109  4.50 $ 108

s% 0.32 2.87 0.14 0.37

s% 0.80 0.94 0.61 0.91

a Calculated according to the Henry’s law (H ¼ 4.38 $ 104 atm. (Perry and Green, 1997))

The latter (k2,4-DCP/HO ¼ 6.0 $ 109 L mol
1 s
1) was obtained as an average among kinetic constant values reported in the literature: 5.1 $ 109 L mol
1 s
1 (Benitez et al., 2001) and 7.1 $ 109 L mol
1 s
1 (Zimbron and Reardon, 2005). The values ðFeðHOÞ2þ Þ ðFeðHOÞ2þ Þ and Fl were taken from Benkelberg and of 3l Warneck (1995). The model was thus used to identify, for both the substrates, the values of k1 and k2i constants. The identification of the values for these two unknown parameters was performed by using simultaneously in a least squares optimization procedure. All the data were collected in a set of runs with different initial Fe(III) concentrations at same pH under UV lamp irradiation. To determine the goodness of the fitting, the percentage standard deviation was also calculated. A comparison of the decay of the substrate predicted concentrations by the model (solid lines) and the experimental data (symbols) was shown in Figs. 7 and 8, for the photolytic runs of 2,4-DCP and 3,4-DCP respectively, at pH ¼ 3.0 and different experimental conditions. For both species, acceptable results are obtained thus indicating a good suitability of the adopted model to predict the behaviour of the investigated compounds during the studied oxidation process. The best estimates for k1 and k2i values are reported in Table 3 along with percentage standard deviations values (s%) for 2,4-DCP and 3,4-DCP for experiments at pH ¼ 3.0. For all the runs, low values of s% indicate a good capability of the model to well predict the system behaviour. It is interesting to observe that estimated values of k1 for both the substrates are in good agreement between them although they are a little lower (by a factor 2e3) than those reported by some of the authors (Andreozzi et al., 2006) for treatment of 1,2-dichlorobenzene in a similar experimental apparatus. However, since the re-oxidation of iron (II) to iron (III) ions is obtained through a photochemical reaction, the values found for the constant k1 are strictly dependent on the radiation power at different wavelengths of the lamp adopted. The values obtained for k2 are slightly different. This difference can be easily explained if one considers that k2i constant refers to a reaction in which HO radicals attack a pseudo-component representing all the intermediates and products present in the solution and that different species e which do not have the same reactivity e may form from the oxidation of the two investigated substrates. The same model was successfully used with the data collected in the runs at pH ¼ 2.0 and 3.5. Table 4 shows the

values estimated at these pHs for k1 and k2i for both the substrates. Also at these pHs good results were obtained both in terms of uncertainties on the estimated parameters and of percentage standard deviations which confirm the suitability of the proposed model to correctly predict the system behaviour. As it is evident from the data reported in Tables 3 And 4, for both the substrates, a reduction of the value of the constant k1 is observed increasing the pH from 3.0 to 3.5 whereas no changes are recorded in the range 2.0e3.0. On the other hand no clear dependence upon pH are shown by constant k2i for the investigated species which changes, in the range 2.0e3.5, up to four times and three times respectively for 2,4-DCP and 3,4-DCP.

5.

Conclusion

In the present work the oxidation of 2,4-DCP and 3,4-DCP through the system Fe(III)/O2/UV was studied in the pH range 2.0e4.0. The results of the present investigation showed that even in absence of oxygen in the solution, due to the production of HO radicals by the photolysis of Fe(HO)2þ complex, it is possible to achieve a complete removal of both the substrates by increasing the initial iron (III) concentration. TOC measurements indicated that for prolonged oxidation times, for the adopted experimental conditions, high degrees of mineralization were obtained whereas the percentage of chlorine converted into chloride was lower than 45% for both the species. The results of toxicity assessments towards algae on samples of treated aqueous solutions containing 2,4-DCP or 3,4-DCP showed that the application of the studied process to the removal of these species results into different behaviours. Starting solutions of 2,4-DCP at a concentration of 6.0 $ 10
5 mol L
1 are toxic towards the algal organism used in the present investigation, the toxicity decreasing during the treatment process. On the other hand, aqueous solutions of 3,4-DCP, at concentration of 6.0 $ 10
5 mol L
1, are not toxic towards the algae but treated samples (at 20 and 60 min) show the capability to inhibit the growth of the adopted organism. Only after 120 min of treatment the solution is newly less toxic than the starting sample. The adoption of a model previously proposed by some of the authors to simulate the behaviour of other substrates when submitted to the same oxidative system gave satisfactory results and enabled to best estimate the values of the

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 3 8 e2 0 4 8

kinetic constants k1 (for the reaction of re-oxidation of iron (II) to iron (III)) and k2i (for the attack of HO radical to the intermediates of oxidation).

Acknowledgements This research was financially supported by Italian Ministry of Scientific Research in the framework of Italian Projects of National Relevance Programme (PRIN 2008).

Nomenclature

[FeIII] [Fe3þ] H IðlÞ Ki ki L V

Total concentration for iron(III) containing species Concentration for ferric ions Henry’s law constant for O2, 4.38 $ 104 atm. Incident photon flow at wavelength l, einstein s
1 (measured) Equilibrium constant (see Table 2) Kinetic constant, (see Table 2) Reactor optical length, 1.1 cm (measured) Irradiated reactor volume, 0.28 L (measured)

Greek symbols Activity coefficient (dimensionless) gi ðFeðHOÞ2þ Þ Molar absorption coefficient of Fe(HO)2þ at 3l wavelength l (L mol
1 cm
1) ðFeðHOÞ2þ Þ Quantum yield of Fe(HO)2þ photolysis at Fl wavelength l (mol einstein
1) s percentage standard deviation (%)

references

Abrahamsson, K., Xie, T.M., 1983. Direct determination of trace amounts of chlorophenols in fresh water, wastewater and seawater. Journal of Chromatography 279 (25), 199e208. Abrahamsson, K., Ekdahl, A., 1996. Volatile halogenated compounds and chlorophenols in the Skagerrak. Journal of Sea Research 35 (1e3), 73e79. Ahlborg, U.G., Thunberg, T.M., Spencer, H.C., 1980. Chlorinated phenols: occurrence, toxicity, metabolism and environmental impact. Critical Reviews in Toxicology 7 (1), 1e35. Al Momani, F., Sans, C., Esplugas, S., 2004a. A comparative study of the advanced oxidation of 2,4-dichlorophenol. Journal of Hazardous Materials B107, 123e129. Al Momani, F., Gonzalez, O., Sans, C., Esplugas, S., 2004b. Combining photo-Fenton process with biological sequencing batch reactor for 2,4-dichlorophenol degradation. Water Science and Technology 49 (4), 293e298. Andreozzi, R., Canterino, M., Marotta, R., 2006. Fe(III) homogeneous photocatalysis for the removal of 1,2dichlorobenzene in aqueous solution by means UV lamp and solar light. Water Research 40 (20), 3785e3792. Andreozzi, R., Marotta, R., 2004. Removal of benzoic acid in aqueous solution by Fe(III) homogeneous photocatalysis. Water Research 38 (5), 1225e1236. Bayarri, B., Gonzalez, O., Maldonado, M.I., Gimenez, J., Esplugas, S., 2007. Comparative study of 2,4-dichlorophenol

2047

degradation with different advanced oxidation processes. Journal of Solar Energy Engineering 129 (1), 60e67. Benkelberg, H.J., Warneck, P., 1995. Photodecomposition of iron (II) hydroxo and sulfato complexes in aqueous solution: wavelength dependence of OH and SO4
quantum yields. Journal of Physical Chemistry 99 (14), 5214e5221. Benitez, F.J., Beltran-Heredia, J., Acero, J.L., Rubio, F.J., 2001. Oxidation of several chlorophenolic derivatives by UV irradiation and hydroxyl radicals. Journal of Chemical Technology 76, 312e320. Benitez, F.J., Beltran-Heredia, J., Acero, J.L., Rubio, F.J., 2000. Contribution of free radicals to chlorophenols decomposition by several advanced oxidation processes. Chemosphere 41 (8), 1271e1277. Blumenstock, M., Zimmermann, R., Schramm, K.W., Kettrup, A., 2001. Identification of surrogate compounds for the emission of PCDD/F (I-TEQ value) and evaluation of their on-line realtime detectability in flue gases of waste incineration plants by REMPIeTOFMS mass spectrometry. Chemosphere 43 (5e7), 507e518. Buxton, G.V., Greenstock, C.L., Helman, W.P., Ross, A.B., 1988. Critical review of rate constants for reactions of hydrated electrons, hydrogen atoms and hydroxyl radicals (OH/O
) in aqueous solution. Journal of Physical and Chemical Reference Data 17 (2), 513e886. Canosa, P., Morales, S., Rodrı´guez, I., Rubı´, E., Cela, R., Go´mez, M., 2005. Aquatic degradation of triclosan and formation of toxic chlorophenols in presence of low concentrations of free chlorine. Analytical and Bioanalytical Chemistry 383, 1119e1126. Catalkaya, E.C., Bali, U., Sengul, F., 2003. Photochemical degradation and mineralization of 4-chlorophenol. Environmental Science and Pollution Research International 10 (2), 113e120. Chu, W., Law, C.K., 2003. Treatment of trichlorophenol by catalytic oxidation process. Water Research 37 (10), 2339e2346. Chu, W., Kwan, C.Y., Chan, K.H., Kam, S.K., 2005. A study of kinetic modelling and reaction pathway of 2,4-dichlorophenol transformation by photo-fenton-like oxidation. Journal of Hazardous Materials B121, 119e126. Deinzer, M., Schaumburg, F., Klein, E., 1978. Environmental health sciences center task force review on halogenated organics in drinking water. Environmental Health Perspectives 24, 209e239. Elpabarawy, M.T., Robidean, R.R., Beach, S.A., 1988. Comparison of three rapid toxicity test procedures: microtoxs, politox and activated sludge respiration inhibition. Toxicity Assess 3, 361e363. EP0509518A1, Process for the preparation of 2,4dichlorophenoxyacetic acid. Date 16/04/1992. Emery, R.J., Papadaki, M., Mantzavinos, D., 2003. Sonochemical degradation of phenolic pollutants in aqueous solutions. Environmental Technology 24 (12), 1491e1500. Flynn Jr., C.M., 1984. Hydrolysis of inorganic iron(III) salts. Chemical Reviews 84 (1), 31e41. Freiter, E.R., 1979. Chlorophenols. In: KirkeOthmer Encyclopedia of Chemical Technology, third ed., vol. 5. John Wiley & Sons, New York. 864e872. Handan, A., Nurullah, O., Nuray, U., 2005. Investigation of the accumulation of 2,4-dichlorophenoxyacetic acid (2,4-D) in rat kidneys. Forensic Science International 153, 53e57. Hayward, K., 1999. Drinking water contaminant hit list for US EPA. Water 21 SeptembereOctober, 4. Hautaniemi, M., Kallas, J., Munter, R., Trapido, M., Laari, A., 1998. Modelling of chlorophenol treatment in aqueous solutions. 2. Ozonation under basic conditions. Ozone: Science and Engineering 20, 283e302.

2048

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 3 8 e2 0 4 8

Heeb, N.V., Dolezal, I.S., Buhrer, T., Mattrel, P., Wolfensberger, M., 1995. Distribution of halogenated phenols including mixed brominated and chlorinated phenols in municipal water incineration flue-gas. Chemosphere 31, 3033e3041. Hong, S.-H., Kwon, B.-H., Lee, J.-K., Kim, I.-K., 2008. Degradation of 2-chlorophenol by Fenton and photo-Fenton processes. Korean Journal of Chemical Engineering 25 (1), 46e52. Kansal, S.K., Singh, M., Sud, D., 2007. Optimization of photocatalytic process parameters for the degradation of 2,4,6-trichlorophenol in aqueous solutions. Chemical Engineering Communications 194 (6), 787e802. Keith, L.H., Telliard, W.A., 1979. Priority pollutants: a prospective view. Environmental Science and Technology 13, 416e424. Knight, R.J., Sylva, R.N., 1975. Spectrophotometric investigation of iron(III) hydrolysis in light and heavy water at 25  C. Journal of Inorganic Nuclear Chemistry 37 (3), 779e783. Kuhn, H.J., Braslavsky, S.E., Schmidt, R., 2004. Chemical actinometry (IUPAC technical report). Pure and Applied Chemistry 76, 2105e2146. Kunihiro, T., Ichikawa, K., Hu, H.Y., Fujie, K., 2004. Development of ecotoxicity assay based on inhibition of respiring activity in microbial community using XTT reduction. Journal of General and Applied Microbiology 50 (2), 91e96. Kuo, W.S., Wu, L.N., 2010. Fenton degradation of 4-chlorophenol contaminated water promoted by solar irradiation. Solar Energy 84 (1), 59e65. Liu, D.H.F., Liptak, B.G., 1997. Environmental Engineers’ Handbook, second ed. CRC Press, Boca Raton, p. 975. Li, X.Z., Zhao, B.X., Wang, P., 2007. Degradation of 2,4dichlorophenol in aqueous solution by a hybrid process. Journal of Hazardous Materials 147, 281e287. Martell, A.E., Smith, R.M., Motekaitis, R.J., 1993. NIST Critical Stability Constants of Metal Complex Database. NIST Standard Reference Data, Gaithersburg, MD, U.S.A. Merriman, J.C., Anthony, D.H.J., Kraft, J.A., Wilkinson, R.J., 1991. Rainy river water quality in the vicinity of bleached kraft mills. Chemosphere 23 (11e12), 1605e1615. Millero, F.J., Yao, W., Aicher, J., 1995. The speciation of Fe(II) and Fe(III) in natural waters. Marine Chemistry 50, 21e39. Norberg-King, T.J., 1993. A linear interpolation method for sublethal toxicity: the inhibition concentration (ICp) approach. Version 2.0. National Effluent Toxicity Assessment Center Technical Report 03-93, Environmental Research Laboratory, Duluth, MN 55804. Pandiyan, T., Carrillo, M.A., Flores, H.M., Santoyo, M.E.R., De Bazua, C.D., 2006. Photochemical oxidation of chlorinated phenols in comparison with electro-oxidation. Toxicological and Environmental Chemistry 88 (1), 23e33. Pera-Titus, M., Garcia-Molina, V., Banos, M.A., Gimenez, J., Esplugas, S., 2004. Degradation of chlorophenols by means of advanced oxidation processes: a general review. Applied Catalysis B: Environmental 47 (4), 219e256. Perry, R.H., Green, D.W., 1997. Perry’s Chemical Engineers’ Handbook, seventh ed. McGraw-Hill, New York, p. 3e103. Pirbazari, M., Badriyha, B.N., Miltner, R.J., 1991. GAC adsorber design for removal of chlorinated pesticides. Journal of Environmental Engineering 117 (1), 80e100. Poulopoulos, S.G., Philippopoulos, C.J., 2004. Photo-assisted oxidation of chlorophenols in aqueous solutions using hydrogen peroxide and titanium dioxide. Journal of Environmental Science and Health, Part A: Toxic/Hazardous Substances and Environmental Engineering A39 (6), 1385e1397. Puig, D., Silgoner, I., Grasserbauer, M., Barcelo´, D., 1997. Part-pertrillion level determination of priority methyl-, nitro-, and chlorophenols in river water samples by automated on-line liquid/solid extraction followed by liquid chromatography/

mass spectrometry using atmospheric pressure chemical ionization and ion spray interfaces. Analytical Chemistry 69 (14), 2756e2761. Qiu, Y., Kuo, C.-H., Zappi, M.E., Fleming, E.C., 2004. Ozonation of 2,6-, 3,4-, and 3,5-dichlorophenol isomers within aqueous solutions. Journal of Environmental Engineering 130 (4), 408e416. Rush, J.D., Bielski, B.H.J., 1985. Pulse radiolytic studies of the reactions of HO2/O
2 with Fe(II)/Fe(III) ions. The reactivity of HO2/O
2 with ferric ions and its implication on the occurrence of the HabereWeiss reaction. Journal of Physical Chemistry 89, 5062e5066. Scott, J.P., Ollis, D.F., 1995. Integration of chemical and biological oxidation processes for water treatment: review and recommendations. Environmental Progress 14, 88e103. Seibert, H., Morchel, S., Gulden, M., 2002. Factors influencing nominal effective concentrations of chemical compounds in vitro: medium protein concentration. Toxicology in Vitro 16, 289e297. Seyler, D.E., East, J.M., Condie, L.W., Borzelleca, J.F., 1984. The use of in vitro methods for assessing reproductive toxicity: dichlorophenols. Toxicology Letters 20 (3), 309e315. Stumm, W., Morgan, J.J., 1996. Aquatic Chemistry, Chemical Equilibria and Rates in Natural Waters, third ed. John Wiley & Sons, Inc., New York, p. 264. Terashima, Y., Ozaki, H., Giri, R.R., Tano, T., Nakatsuji, S., Takanami, R., Taniguchi, S., 2006. Photocatalytic oxidation of low concentration 2,4-D solution with new TiO2 fiber catalyst in a continuous flow reactor. Water Science and Technology 54 (8), 55e63. Untea, I., Orbeci, C., Tudorache, E., 2006. Oxidative degradation of 4-chlorophenol from aqueous solution by photo-Fenton advanced oxidation process. Environmental Engineering and Management Journal 5 (4), 661e674. Vijayan, P., Mahendiran, C., Suresh, C., Shanthi, K., 2009. Photocatalytic activity of iron doped nanocrystalline titania for the oxidative degradation of 2,4,6-trichlorophenol. Catalysis Today 141 (1e2), 220e224. Wegman, R.C.C., Van der Broek, H.H., 1983. Chlorophenols in river sediments in the Netherlands. Water Research 17, 227e230. Xie, Y., Chen, F., He, J., Zhao, J., Wang, H., 2000. Photoassisted degradation of dyes in the presence of Fe3þ and H2O2 under visible irradiation. Journal of Photochemistry and Photobiology A: Chemistry 136 (3), 235e240. Xu, X.-H., Zhao, W.-R., Huang, Y.-Q., Wang, D.-H., 2003. Chlorophenol oxidation kinetic by photo-assisted Fenton process. Journal of Environmental Sciences 15 (4), 475e481. Yoshioka, Y., Sato, T., 1986. Evaluation of the test method activated sludge respiration inhibition test proposed by the OECD. Ecotoxicology and Environmental Safety 12, 206e209. Zepp, R.G., Gumz, M.M., Miller, W.L., Gao, H., 1998. Photoreaction of valerophenone in aqueous solution. Journal of Physical Chemistry, A 102 (28), 5716e5723. Zhang, F., Liu, H., Sunb, Y.Y., Wang, X.R., Wu, J.C., Xue, Y.Q., 2005. Responses of the antioxidant defenses of the goldfish carassius exposed to 2,4-dichlophenol. Environmental Toxicology and Pharmacology 19, 185e190. Zhang, H., Bartlett, R.J., 1999. Light-induced oxidation of aqueous chromium(III) in the presence of iron(III). Environmental Science and Technology 33 (4), 588e594. Zimbron, J.A., Reardon, K.F., 2005. Hydroxyl free radical reactivity toward aqueous chlorinated phenols. Water Research 39, 865e869. 98/83/EC. Official Journal of the European Communities (1998). Council Directive 98/83/EC on the quality of water intended for human consumption, December 5. pp. L330/32-L330/50.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 4 9 e2 0 5 9

Available at www.sciencedirect.com

journal homepage: www.elsevier.com/locate/watres

The effect of roofing material on the quality of harvested rainwater Carolina B. Mendez a, J. Brandon Klenzendorf a,1, Brigit R. Afshar a,2, Mark T. Simmons b, Michael E. Barrett a, Kerry A. Kinney a, Mary Jo Kirisits a,* a

Department of Civil, Architectural, and Environmental Engineering, The University of Texas at Austin, 1 University Station C1786, Austin, TX 78712, USA b Lady Bird Johnson Wildflower Center, The University of Texas at Austin, 4801 La Crosse Avenue, Austin, TX 78739, USA

article info

abstract

Article history:

Due to decreases in the availability and quality of traditional water resources, harvested

Received 7 November 2010

rainwater is increasingly used for potable and non-potable purposes. In this study, we

Received in revised form

examined the effect of conventional roofing materials (i.e., asphalt fiberglass shingle,

15 December 2010

Galvalume
metal, and concrete tile) and alternative roofing materials (i.e., cool and green) on

Accepted 16 December 2010

the quality of harvested rainwater. Results from pilot-scale and full-scale roofs demonstrated

Available online 22 December 2010

that rainwater harvested from any of these roofing materials would require treatment if the consumer wanted to meet United States Environmental Protection Agency primary and

Keywords:

secondary drinking water standards or non-potable water reuse guidelines; at a minimum,

Rainwater harvesting

first-flush diversion, filtration, and disinfection are recommended. Metal roofs are commonly

Roof-runoff

recommended for rainwater harvesting applications, and this study showed that rainwater

Water quality

harvested from metal roofs tends to have lower concentrations of fecal indicator bacteria as

First-flush

compared to other roofing materials. However, concrete tile and cool roofs produced harvested

Roofing material

rainwater quality similar to that from the metal roofs, indicating that these roofing materials

Potable

also are suitable for rainwater harvesting applications. Although the shingle and green roofs produced water quality comparable in many respects to that from the other roofing materials, their dissolved organic carbon concentrations were very high (approximately one order of magnitude higher than what is typical for a finished drinking water in the United States), which might lead to high concentrations of disinfection byproducts after chlorination. Furthermore the concentrations of some metals (e.g., arsenic) in rainwater harvested from the green roof suggest that the quality of commercial growing media should be carefully examined if the harvested rainwater is being considered for domestic use. Hence, roofing material is an important consideration when designing a rainwater catchment. ª 2011 Elsevier Ltd. All rights reserved.

1.

Introduction

Roof-based rainwater harvesting is the capture of rainwater from a roof for potable or non-potable use. Rainwater

harvesting is undergoing a surge in popularity in the United States, particularly in locations where traditional, high-quality freshwater supplies are absent or where consumers desire to contribute to sustainability. The United States Environmental

* Corresponding author. Tel.: þ1 512 232 7120; fax: þ1 512 471 0592. E-mail address: [email protected] (M.J. Kirisits). 1 Present address: Geosyntec Consultants, 3600 Bee Caves Road, Suite 101, Austin, TX 78746, USA. 2 Present address: CH2M Hill, 12377 Merit Drive, Ste. 1000, Dallas, TX 75251, USA. 0043-1354/$ e see front matter ª 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.watres.2010.12.015

2050

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 4 9 e2 0 5 9

Nomenclature cm CFU

centimeter colony forming unit  degree angle  C degrees Celsius DOC dissolved organic carbon FC fecal coliform h hour km kilometer L liter MCL maximum contaminant level mg/L microgram per liter mS/cm microsiemens per centimeter mg-N/L milligram of nitrogen per liter mg/L milligram per liter

Protection Agency (USEPA) does not regulate the water quality of residential rainwater harvesting systems in the U.S., but some state and local agencies do issue voluntary water quality guidelines for residential rainwater harvesting systems (e.g., Texas Rainwater Harvesting Evaluation Committee, 2006). The quality of harvested rainwater is of great importance because it is increasingly used for domestic purposes. Several types of chemical contaminants have been found in harvested rainwater including heavy metals (Fo¨rster, 1999; Lee et al., 2010), polycyclic aromatic hydrocarbons (PAHs) (Fo¨rster, 1998, 1999), pesticides (Zobrist et al., 2000), and herbicides (Bucheli et al., 1998). Microorganisms also are present in roofrunoff, and fecal indicator bacteria and potentially pathogenic bacteria and protozoa have been detected (Ahmed et al., 2008). The type of roofing material used for the catchment can affect the quality of harvested rainwater. Nicholson et al. (2009) compared harvested rainwater quality among six roof types: galvanized metal, cedar shake, asphalt shingle, two types of treated wood, and green (i.e., vegetated). The galvanized metal, asphalt shingle, and green roofs neutralized the acidic rainwater to a greater extent than did the other roofing materials. The treated woods yielded the highest copper concentrations (mg/L range), and the galvanized metal yielded the highest zinc concentrations (mg/L range), as compared to the mg/L concentrations of these metals from the other roofing materials. Van Metre and Mahler (2003) found galvanized metal roofs to be a source of particulate zinc and cadmium and asphalt shingle roofs to be a source of particulate lead and potentially mercury. Kingett Mitchell Ltd. (2003) found higher zinc concentrations in rainwater harvested from painted galvanized iron roofs that showed evidence of weathering as compared to those in excellent condition. Despins et al. (2009) found that harvested rainwater quality from steel roofs was superior to that from asphalt shingle roofs, particularly with respect to turbidity, total organic carbon, and color. Most studies to date have focused on examining conventional roofing materials, such as galvanized metal, for rainwater harvesting. Green roofs are increasingly being installed for their stormwater retention characteristics, and both green and cool roofs (made of reflective roofing material) are being installed for their energy conservation potential; however,

mL mm NTU % PAH PVC SMCL SVOC a m2 TC TDS TSS USEPA VOC

milliliter millimeter nephelometric turbidity unit percent polycyclic aromatic hydrocarbon polyvinyl chloride secondary maximum contaminant level semi-volatile organic compound significance level square meter total coliform total dissolved solids total suspended solids United States Environmental Protection Agency volatile organic compound

neither has been widely considered for rainwater harvesting. To our knowledge, only one study has analyzed harvested rainwater quality from a green roof (Nicholson et al., 2009), and it was examined in the context of non-potable use. The objective of the current study was to thoroughly examine the effect of roofing material on the quality of harvested rainwater (untreated, except for the use of a first-flush diverter) from the standpoint of suitability for domestic use. Five co-located pilot-scale roofs were used to compare harvested rainwater quality among conventional roofing materials (i.e., asphalt fiberglass shingle, Galvalume
metal, and concrete tile) and alternative roofing materials (i.e., cool and green). This is the first study to extensively compare harvested rainwater quality (20 water quality parameters) among these conventional and alternative roofing materials. In addition, the quality of the rainwater harvested from the asphalt fiberglass shingle and Galvalume
roofing materials was compared between the pilot-scale roofs and full-scale residential roofs.

2.

Materials and methods

2.1.

Study sites

This study was conducted in Austin, Texas, where the mean annual rainfall is 856 mm (National Oceanic and Atmospheric Administration, 2010). The pilot-scale roofs were constructed at the Lady Bird Johnson Wildflower Center (Austin, TX), and the full-scale residential roofs were located within 42 km of the pilot site.

2.2.

Pilot-scale roofs

The conventional pilot-scale roofs, constructed in 2009, were made of asphalt fiberglass shingle (GAF-Elk, Wayne, NJ), Galvalume
(55% aluminum-zinc alloy coated steel, United States Steel Corporation, Pittsburgh, PA), or concrete tile (MonierLifetile, Irvine, CA) (Fig. 1A). They were angled 18.4 from the horizontal, with a footprint area of 2.8 m2. The alternative pilot-scale roofs, constructed in 2007 as described by Simmons et al. (2008), were an unfertilized (type E) green roof (Fig. 1B) and an acrylic-surfaced, 2-ply atactic polypropylene

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 4 9 e2 0 5 9

2051

Fig. 1 e Pilot-scale roofs (A) from left to right: asphalt fiberglass shingle, metal, concrete tile, (B) green, and (C) cool.

modified bituminous membrane cool roof (Fig. 1C); these lowslope roofs were angled 1.2 from the horizontal, with a footprint area of 3.4 m2. All of the pilot-scale roofs faced north.

2.3.

Full-scale residential roofs

Three full-scale residential roofs were sampled: a 12-year-old Galvalume
roof (22 from the horizontal, footprint area of 4.3 m2), a 5-year-old asphalt fiberglass shingle roof (23 from the horizontal, footprint area of 4.3 m2) and another 5-year-old asphalt fiberglass shingle roof (18 from the horizontal, footprint area of 5.3 m2). All of the full-scale roofs faced north.

2.4.

Rainwater sampling devices

A sampling insert (7.6-cm inner-diameter potable-quality polyvinyl chloride (PVC) pipe cut lengthwise in half) was placed in the gutters of the conventional pilot-scale and full-scale roofs. The sampling insert drained to a passive collection system consisting of a 2-L glass bottle to collect the “first-flush” and two 10-L polypropylene tanks in series (Fig. 2). The firstflush volume was designed according to the Texas Water Development Board guidelines, which state that the first-flush should divert a minimum of 38 L for every 93 m2 of collection area (Texas Water Development Board, 2005). For the alternative pilot-scale roofs, the drainage system described in Simmons et al. (2008) emptied to the aforementioned passive collection system. The first-flush bottle and tanks in the passive collection system filled sequentially during a rain event, with excess rainfall exiting the system through an overflow spout (Fig. 2). A 46-cm inner-diameter polyethylene funnel attached to a 10-L polypropylene tank was used as an ambient sampler to collect rainwater without roof exposure; the cover on the funnel was removed up to 24 h prior to a rain event, so some atmospheric deposition was possible. Between rain events, the sampling devices were washed and rinsed, and the collection bottle and tanks were autoclaved. Rainfall was measured with a rain gauge, and the sampled events had rainfalls between 25 mm and 38 mm at the pilot-scale and full-scale sites.

2.5.

sampler, first-flush bottle, and first and second tanks were analyzed in triplicate for pH, conductivity, total coliform (TC), fecal coliform (FC), turbidity, total suspended solids (TSS), dissolved organic carbon (DOC), and nine metals (dissolved þ particulate): aluminum, arsenic, cadmium, chromium, copper, iron, lead, selenium, and zinc. Nitrate and nitrite were measured once for each sample. For the fourth rain event, samples from the ambient sampler and first-flush bottle were analyzed for pesticides and PAHs. Rainwater harvested from the full-scale roofs was sampled during three rain events in 2009: February 9, February 11, and March 11. For all three rain events, the ambient sampler, firstflush bottle, and first and second tanks were analyzed in triplicate for TC, FC, TSS, DOC, lead, and zinc. The pH, conductivity, turbidity, nitrate, and nitrite were measured once for each sample. For the first rain event, samples from the ambient sampler and first-flush bottle were analyzed for a suite of volatile and semi-volatile organic compounds (VOCs and SVOCs). At the end of each rain event, samples were transported to the laboratory for analysis. The pH, conductivity, TC, FC, turbidity, and TSS were analyzed immediately upon arrival at the laboratory. The pH was measured with a combination electrode and meter (Model 920A, Orion Research Inc., Boston,

Sampling and analytical methods

Rainwater harvested from the pilot-scale roofs was sampled during four rain events in 2009: April 18, June 11, July 23, and September 11. For the first three rain events, the ambient

Fig. 2 e Schematic of the sampling device.

2052

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 4 9 e2 0 5 9

MA). Conductivity was measured with a conductivity meter (Model CDM 230, Radiometer Analytical, Lyon, France). TC was measured according to method 9222B (American Public Health Association, 1998) with m-Endo broth (Millipore, Billerica, MA), and FC was measured according to method 9222D (American Public Health Association, 1998) with m-FC agar (Difco, Detroit, MI). Turbidity was measured with a turbidimeter (Model 2100A, Hach, Loveland, CO). TSS was measured according to method 2540B (American Public Health Association, 1998). Samples for the remaining analytes were preserved for later analysis. Samples for nitrate and nitrite analysis were stored at 4  C and analyzed within 48 h with Test 0 N Tube Reactor/Cuvette Tubes (Hach, Loveland, CO); nitrate was measured with the NitraVer
X reagent set, and nitrite was measured with the NitriVer
3 reagent set according to the manufacturer’s instructions. Samples for DOC analysis were stored at 4  C after addition of reagent grade phosphoric acid; DOC was measured with an Apollo 9000 Combustion Analyzer (Tekmar-Dohrmann, Cincinnati, OH). Samples for metal analysis were preserved with trace metal grade nitric acid. For the pilot-scale study, metal concentrations were measured by inductively coupled plasma-mass spectrometry (Model 7500ce, Agilent Technologies, Inc., Tokyo, Japan). For the full-scale study, metal concentrations were measured by graphite furnace atomic absorption spectroscopy (Model AAnalyst 600, PerkineElmer, Waltham, MA). At the end of the February 9 and September 11, 2009 rain events, samples were delivered to DHL Analytical (Round Rock, TX) for the analysis of organic compounds. Pesticides were measured according to method 8081A (USEPA, 1996a); SVOCs and PAHs were measured according to method 8270C (USEPA, 1996b); VOCs were measured according to method 8260B (USEPA, 1996c).

2.6.

Statistical analysis

The nonparametric Mann-Whitney test was used to make comparisons among the data because of the small number of rainfall events and the inability to show that the water quality

data could be characterized by a normal distribution. A significance level of a ¼ 0.1 was used. Since the consumer often diverts the first-flush water from use, the discussion in this study generally focuses on the quality of the rainwater harvested after the first-flush (i.e., tanks 1 and 2 in Fig. 2). The quality of the rainwater harvested after the first-flush was compared to the quality of the firstflush, to the quality of rainwater captured by the ambient sampler, to the U.S. drinking water standards, and to the U.S. non-potable urban water reuse guidelines. Finally, the quality of the rainwater harvested after the first-flush was compared among the roofing materials.

3.

Results and discussion

3.1.

pH and conductivity e pilot-scale roofs

The average pH of rainfall collected in the ambient sampler was 6.1, and all of the harvested rainwater samples from the pilot-scale roofs were in the near-neutral range (Fig. 3A). For all of the pilot-scale roofs except the metal roof, the pH of the rainwater harvested after the first-flush was significantly higher than that in the ambient sampler ( p-values < 0.048). Similarly, other studies have reported an increase in pH from ambient rainwater to harvested rainwater (Kingett Mitchell Ltd., 2003). The pH of the rainwater harvested after the firstflush for the tile roof was significantly higher than those of the other pilot-scale roofs ( p-values < 0.021). This was expected due to the alkaline nature of concrete and is consistent with other studies (Kingett Mitchell Ltd., 2003). The average pH of the rainwater harvested after the first-flush for all pilot-scale roofs met the USEPA secondary drinking water standard range of 6.5e8.5 (Fig. 3A) and the non-potable urban water reuse guideline range of 6.0e9.0 (USEPA, 2004). The average conductivity of rainfall collected in the ambient sampler was 34 mS/cm (Fig. 3B), which is in the typical range for ambient rainwater (Yaziz et al., 1989; Lee et al., 2010). For the green roof only, the conductivity of the rainwater

Fig. 3 e Average pH (panel A) and conductivity (panel B) for the pilot-scale events: ( ) Quality of the first-flush, ( ) Quality after the first-flush (average of tank 1 and tank 2),  USEPA secondary drinking water standard range for pH (6.5e8.5), - - - Ambient sampler. One standard deviation is shown.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 4 9 e2 0 5 9

harvested after the first-flush was significantly higher than that in the ambient sampler ( p-value ¼ 0.012). The conductivity of the rainwater harvested after the first-flush for the green roof also was significantly higher than those for the other pilot-scale roofs ( p-values < 0.013); this was expected due to mineral dissolution as the rainwater passed through the growing media of the green roof. For the shingle, metal, and cool pilot-scale roofs, the conductivity of the first-flush was significantly higher than that of the rainwater harvested after the first-flush ( p-values < 0.024). Conductivity and total dissolved solids (TDS) are correlated, although a specific correlation was not prepared in this study. Using the correlation of Singh and Kalra (1975) between conductivity and TDS in mountain streams, the average, estimated TDS of rainwater harvested after the first-flush for all pilot-scale roofs met the USEPA secondary maximum contaminant level (SMCL) for TDS, which is 500 mg/L.

3.2.

TC and FC e pilot-scale roofs

The harvested rainwater was assessed using TC (Fig. 4A) and FC (Fig. 4B) as indicators of microbial quality. For the shingle, tile, and cool pilot-scale roofs, the TC concentration of the first-flush was significantly higher than that of the rainwater harvested after the first-flush ( pvalues < 0.024), but the TC concentration from the metal and green pilot-scale roofs did not change significantly from the first-flush to the subsequent tanks ( p-values ¼ 0.131 and 0.179, respectively). For all of the pilot-scale roofs, the TC concentration of the rainwater harvested after the first-flush was statistically indistinguishable from that in the ambient sampler ( p-values > 0.131). According to the USEPA primary drinking water standards, no more than five percent of samples in a month are allowed to be positive for TC; all of the pilot-scale roofs violate this standard because they showed measurable TC in the first-flush and subsequent tanks for all rain events. For the shingle, metal, tile, and cool pilot-scale roofs, the FC concentration of the first-flush was significantly higher than that of the rainwater harvested after the first-flush ( p-values < 0.083). For all of the pilot-scale roofs except the metal roof, the FC concentration of the rainwater harvested

2053

after the first-flush was statistically indistinguishable from that in the ambient sampler ( p-values > 0.179). According to the USEPA primary drinking water standards, if two consecutive samples have TC and at least one of them has FC (or Escherichia coli), then the system has an acute maximum contaminant level (MCL) violation; all of the pilot-scale roofs violate this standard. Only the rainwater harvested after the first-flush from the metal roof meets the non-potable urban water reuse guidelines (USEPA, 2004), which state that FC should not exceed 14 colony forming units (CFU)/100 mL in any sample. Our data are consistent with the literature, which shows detectable TC and FC in harvested rainwater (Yaziz et al., 1989; Simmons et al., 2001). The use of these traditional indicator bacteria might cause the underestimation of the risk of waterborne disease to the rainwater consumer (Lye, 2002; Ahmed et al., 2008). Nonetheless, based on the TC and FC concentrations in the rainwater harvested after the first-flush, most of the pilot-scale roofs produced water that did not meet potable and non-potable guidelines. Thus treatment is needed if the guidelines are to be met. Due to filtration of the rainwater through the substrate layer, we expected the pilot-scale green roof to consistently yield the lowest concentrations of indicator bacteria in harvested rainwater, but this was not always the case. For instance, while the first two rain events showed no detectable FC in the rainwater harvested from the green roof, the third rain event showed a spike in FC (330 CFU/100 mL) in the rainwater harvested after the first-flush. The third rain event also showed a spike in TC concentrations from the green roof but no spikes in TSS or turbidity (data not shown). The other pilot-scale roofs did not have similar contaminant spikes in the third rain event; thus, the reason for the spike in indicator bacteria concentrations in this event from the green roof is unclear at this time. On a per-event basis, the green roof most often showed the best water quality with respect to indicator bacteria among the roofing materials, but the potential for spikes in indicator bacteria concentrations from the green roof should be investigated further. For the pilot-scale metal roof, the FC concentration of the rainwater harvested after the first-flush was significantly

Fig. 4 e Geometric mean of TC (panel A) and FC (panel B) for the pilot-scale events: ( ) Quality of the first-flush, ( ) Quality after the first-flush (average of tank 1 and tank 2), - - - Ambient sampler. One geometric standard deviation is shown.

2054

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 4 9 e2 0 5 9

lower than that in the ambient sampler ( p-value ¼ 0.083). The TC and FC concentrations of the first-flush from the metal roof were significantly lower than those for the cool, shingle, and tile roofs ( p-values < 0.001). Similarly, Yaziz et al. (1989) showed that the concentration of indicator bacteria in rainwater harvested from metal roofs was lower than that from concrete tile roofs. One possible reason for this trend is that low emissivity materials, like metals, have higher surface temperatures in sunlight (Bretz et al., 1998), which might have inactivated a fraction of the indicator bacteria. Another possible reason for this trend is that heavy metal exposure can decrease cellular viability (Teitzel and Parsek, 2003), so contact with the metal roofing material might have inactivated a fraction of the indicator bacteria.

3.3.

Turbidity and TSS concentrations e pilot-scale roofs

Turbidity (Fig. 5A) and TSS concentrations (Fig. 5B) were measured for the pilot-scale roofs. For all of the pilot-scale roofs, the turbidity of the first-flush was significantly higher than that of the rainwater harvested after the first-flush ( p-values < 0.083). The turbidity of the first-flush from the metal and cool pilot-scale roofs was significantly higher than those for the shingle, tile, and green pilot-scale roofing materials ( p-values < 0.001). This was expected because the metal and cool roofs visually had the smoothest surfaces, and Egodawatta et al. (2009) noted that particulate matter washes off from smoother surfaces more rapidly than it does from rougher surfaces. For the tile, cool, and green pilot-scale roofs, the turbidity of the rainwater harvested after the first-flush was statistically indistinguishable from that in the ambient sampler ( p-values > 0.131). The turbidity of the rainwater harvested after the first-flush for the green roof was lower than that for the shingle, metal, and tile pilot-scale roofs but indistinguishable from that of the cool pilot-scale roof ( p-values < 0.078); thus, the green roof produced the highest quality rainwater with respect to turbidity. According to the USEPA primary drinking water standards, for systems using conventional or direct filtration, turbidity

must never be above 1 nephelometric turbidity unit (NTU), and 95% of samples in one month must be less than or equal to 0.3 NTU. According to the non-potable urban water reuse guidelines (USEPA, 2004), the average turbidity in a 24-h period should be less than or equal to 2 NTU and turbidity should not exceed 5 NTU at any time. All samples of rainwater harvested after the first-flush from the pilot-scale roofs violate these standards. The TSS data showed very similar trends to the turbidity data. For example, the TSS of the rainwater harvested after the first-flush for the green roof was lower than those for the other pilot-scale roofs ( p-values < 0.047); thus, the green roof produced the highest quality rainwater with respect to TSS. No primary or secondary drinking water standards exist for TSS. The non-potable urban water reuse guidelines state that TSS should not exceed 5 mg/L (USEPA, 2004); the rainwater harvested after the first-flush for all pilot-scale roofs often exceeded this standard. Based on the turbidity and TSS concentrations in the rainwater harvested after the first-flush, all of the pilot-scale roofs produced water that did not meet potable and nonpotable guidelines. Thus, treatment is needed if the guidelines are to be met.

3.4.

Nitrate and nitrite concentrations e pilot-scale roofs

Nitrate (Fig. 6A) and nitrite (Fig. 6B) concentrations were measured for the pilot-scale roofs. For the shingle, metal, and tile pilot-scale roofs, the average nitrate concentration of the first-flush was significantly higher than that of the rainwater harvested after the first-flush ( p-values < 0.082). For all of the pilot-scale roofs, the average nitrate concentration of the rainwater harvested after the firstflush was statistically indistinguishable from that in the ambient sampler ( p-values > 0.25). Elevated nitrate levels were not expected in the rainwater harvested from the green roof because no fertilizers were applied to the green roof, and plants and soil biota can consume nitrate. All samples of the rainwater harvested after the first-flush for all pilot-scale roofs met the USEPA MCL for nitrate (10 mg-N/L), consistent with the

Fig. 5 e Average turbidity (panel A) and TSS concentrations (panel B) for the pilot-scale events: ( ) Quality of the first-flush, ( ) Quality after the first-flush (average of tank 1 and tank 2),  USEPA filtered system guideline (1 NTU), - - - Ambient sampler. One standard deviation is shown.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 4 9 e2 0 5 9

2055

Fig. 6 e Average nitrate (panel A) and nitrite (panel B) concentrations for the pilot-scale events: ( ) Quality of the first-flush, ( ) Quality after the first-flush (average of tank 1 and tank 2),  USEPA MCLs for nitrate (10 mg-N/L) and nitrite (1 mg-N/L), - - - Ambient sampler. One standard deviation is shown. findings of Nicholson et al. (2009) for asphalt fiberglass shingle, metal, and green roofs. The average concentration of nitrate in the first-flush from all of the pilot-scale roofs was greater than the average nitrate concentration in the ambient sampler (Fig. 6A), which is likely due to dry deposition of nitrate onto the roof surfaces (Fo¨rster, 1998). The nitrate concentrations in the first-flush increased from 4 to 8 antecedent dry days and then either reached a plateau or continued to increase from 8 to 14 antecedent dry days (Fig. 7), supporting the hypothesis that dry deposition contributes to nitrate levels in harvested rainwater. For all of the pilot-scale roofs, the average nitrite concentration of the first-flush was significantly higher than that of the rainwater harvested after the first-flush ( p-values < 0.083). For the metal, cool, and green pilot-scale roofs, the average nitrite concentration of the rainwater harvested after the firstflush was statistically indistinguishable from that in the ambient sampler ( p-values > 0.25), and the shingle and tile pilot-scale roofs produced nitrite concentrations in the rainwater harvested after the first-flush that were just slightly higher than ambient ( p-values < 0.083). All samples of the rainwater harvested after the first-flush for all pilot-scale roofs met the USEPA MCL for nitrite (1 mg-N/L).

3.5.

concentrations in the rainwater harvested after the firstflush as compared to all of the other pilot-scale roofs ( pvalues < 0.047). Given the nature of the roofing materials, the shingle and green pilot-scale roofs were expected to be sources of organic matter. Asphalt fiberglass shingles are noted for producing colored water (Despins et al., 2009). Berndtsson et al. (2009) hypothesized that the elevated DOC concentrations observed in green roof-runoff come from the soil organic matter and from decay of the vegetation. DOC also might be produced by soil bacteria colonizing the green roof. The DOC concentrations in the rainwater harvested after the first-flush for the shingle and green roofs were elevated as compared to the DOC concentration of a typical, finished drinking water in the U.S. (w2.5e3.5 mg/L). Although DOC does not have an MCL, it is a precursor for regulated disinfection byproducts (e.g., trihalomethanes and haloacetic acids). To reduce disinfection byproduct formation, the Stage 1 Disinfectants/Disinfection Byproducts Rule mandates that water systems that treat with disinfectants other than ultraviolet light must reduce the total organic carbon of the water by 15e50%, where the percentage is based

DOC concentrations e pilot-scale roofs

DOC concentrations were measured for the pilot-scale roofs (Fig. 8). For the metal and cool roofs, the average DOC concentration of the first-flush was significantly higher than that of the rainwater harvested after the first-flush ( p-values < 0.048). For the metal, tile, and cool pilot-scale roofs, the average DOC concentration of the rainwater harvested after the first-flush was statistically indistinguishable from that in the ambient sampler ( p-values > 0.33). Conversely, for the shingle and green pilot-scale roofs, the average DOC concentration in the rainwater harvested after the first-flush was significantly higher than that in the ambient sampler ( p-values < 0.012). In fact, the shingle pilot-scale roof showed a significant increase in DOC concentration from rainwater in the first-flush tank to the rainwater harvested after the first-flush ( p-value < 0.012). The green roof showed statistically significant higher DOC

Fig. 7 e Effect of antecedent dry days on average nitrate concentrations in first-flush samples from the pilot-scale roofs: dCd Shingle, dBd Metal, d;d Tile, d6d Cool, d-d Green,  USEPA MCL for nitrate (10 mg-N/L).

2056

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 4 9 e2 0 5 9

Fig. 8 e Average DOC concentrations for the pilot-scale events: ( ) Quality of the first-flush, ( ) Quality after the first-flush (average of tank 1 and tank 2), - - - Ambient sampler. One standard deviation is shown. on the total alkalinity and source water total organic carbon concentration. Thus, systems with high source water total organic carbon concentrations (>8 mg/L) must decrease the total organic carbon by 30e50%. Given the particularly high DOC concentrations in the rainwater harvested after the firstflush from the shingle and green roofs, it is quite possible that the concentrations of disinfection byproducts formed in these waters would violate the primary drinking water standards. If the water is used for non-potable purposes, the consumer also might come into contact with these disinfection byproducts by inhalation, since many of them are volatile (e.g., chloroform).

3.6.

Metal concentrations e pilot-scale roofs

The average concentrations of cadmium, chromium, and selenium in the rainwater harvested after the first-flush for each pilot-scale roof were at least an order of magnitude below the corresponding primary drinking water standard (data not shown); these metals are not discussed further. Data for the remaining metals (aluminum, arsenic, copper, iron, lead, and zinc) are shown in Fig. 9. In general, the average concentration of each metal in the first-flush was significantly higher than that in the rainwater harvested after the first-flush at a 10% significance level. The notable exception to this is the green roof, where the metal concentrations (except for lead) in the first-flush were statistically indistinguishable from those in the rainwater harvested after the first-flush at a 10% significance level. In general, the average concentration of each metal in the rainwater harvested after the first-flush was significantly less than or statistically indistinguishable from that in the ambient sampler at a 10% significance level, with three notable exceptions. First, the rainwater harvested after the first-flush from the tile and green roofs had arsenic and lead concentrations that were significantly higher than those in the ambient sampler ( p-values < 0.083). Second, the rainwater harvested after the first-flush from the metal, tile, and green roofs had zinc concentrations that were significantly higher than those in the ambient sampler ( p-values < 0.083). The elevated arsenic, lead, and zinc found in the rainwater

harvested from the green roof are hypothesized to have come from the growing media, which suggests that the quality of commercial growing media must be carefully examined if the harvested rainwater is being considered for domestic use. Third, the rainwater harvested after the first-flush from the shingle roof had copper concentrations that were significantly higher than that in the ambient sampler ( p-value ¼ 0.012) and those from all of the other pilot-scale roofs ( p-values < 0.001), suggesting that the shingle roof is a source of copper. This is consistent with the study of Nicholson et al. (2009), which showed that rainwater harvested from asphalt fiberglass shingle roofs contained more copper than did rainwater harvested from galvanized metal or green roofs. In other studies, elevated copper in roof-runoff was attributed to atmospheric deposition due to vehicles (Van Metre and Mahler, 2003) and copper fittings (Fo¨rster, 1999), but those sources were not present at our pilot-scale site. The concentration of each metal in the rainwater harvested after the first-flush was compared to the appropriate USEPA primary and secondary drinking water standards or action levels (Fig. 9). None of the pilot-scale roofs violated the primary standards or action levels for the rainwater harvested after the first-flush (i.e., arsenic, copper, and lead). However, aluminum (Fig. 9A) and iron (Fig. 9D) concentrations in the rainwater harvested after the first-flush often exceeded the secondary standards. All of the pilot-scale roofs violated the aluminum SMCL of 200 mg/L in the rainwater harvested after the first-flush. Only the green pilot-scale roof did not violate the iron SMCL of 300 mg/L in the rainwater harvested after the first-flush. Zinc concentrations in the rainwater harvested after the first-flush from all roofing materials were at least one order of magnitude below the zinc secondary drinking water standard (Fig. 9F). Previous studies have shown that galvanized metal roofs (Kingett Mitchell Ltd., 2003; Nicholson et al., 2009) or painted galvanized metal roofs in less-than-excellent condition (Kingett Mitchell Ltd., 2003) yield high concentrations of zinc (mg/L) in harvested rainwater. Conversely, painted galvanized metal roofs in excellent condition and Zincalume
(55% aluminum-zinc alloy coated steel, which is the same composition as Galvalume
) roofs yield lower concentrations of zinc (mg/L) in harvested rainwater (Kingett Mitchell Ltd., 2003), similar to our data from the Galvalume
roof. This is consistent with the work of Sullivan and Worsley (2002), who found that the presence of aluminum in alloy-coated steel roofs increases the corrosion resistance as compared to uncoated galvanized roofs, thereby decreasing the zinc concentration in roof-runoff. In terms of primary standards and action levels for metals in drinking water, all of the pilot-scale roofs produced acceptable quality water. The green pilot-scale roof produced the best quality water in terms of secondary standards, as its aluminum and iron concentrations in the rainwater harvested after the first-flush were the lowest as compared to the other pilot-scale roofs.

3.7. Water quality ranges e comparing full-scale and pilot-scale roofs Previous studies have shown that variation in roof-runoff quality is attributable to the age of the roof, presence of overhanging vegetation, wind direction, and proximity to local

2057

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 4 9 e2 0 5 9

Fig. 9 e Average total aluminum (panel A), arsenic (panel B), copper (panel C), iron (panel D), lead (panel E), and zinc (panel F) concentrations for the pilot-scale events: ( ) Quality of the first-flush, ( ) Quality after the first-flush (average of tank 1 and tank 2),  USEPA primary or secondary drinking water standards or action levels: aluminum (200 mg/L), arsenic (10 mg/L), copper (1300 mg/L), iron (300 mg/L), lead (15 mg/L), and zinc (5000 mg/L). - - - Ambient sampler. One standard deviation is shown.

sources (Fo¨rster, 1999; Kingett Mitchell Ltd., 2003; Evans et al., 2006). In contrast to the pilot-scale roofs, our full-scale roofs were older, had overhanging vegetation, and were located closer to local emission sources (e.g., chimneys and traffic). Despite these differences, the qualities of the rainwater harvested after the first-flush from the pilot-scale and fullscale roofs were remarkably similar (Table 1). Consistent with the pilot data, the rainwater harvested after the first-flush from the full-scale metal roof had average TC and FC concentrations that were significantly lower than those from both full-scale shingle roofs ( p-values < 0.09). Also consistent with the pilot data, the TC, FC, turbidity, and TSS in the rainwater harvested after the first-flush for the full-scale roofs do not meet the USEPA primary drinking water standards nor the non-potable urban water reuse guidelines. Similar to the pilot data, the average nitrate and nitrite concentrations in the rainwater harvested after the first-flush for the full-scale roofs were well below the USEPA MCLs.

3.8. VOCs, SVOCs, PAHs, and pesticides e pilot- and full-scale roofs The first-flush samples and ambient rain samples from one rain event were analyzed for PAHs and pesticides for the pilotscale roofs and for a larger suite of VOCs and SVOCs for the full-scale roofs. In the pilot-scale study, no pesticides or PAHs were detected. In the full-scale study, 2,4-dinitrophenol was detected in the first-flush of the metal roof and one of the

shingle roofs at a concentration of 3 mg/L; benzyl alcohol also was detected in the first-flush from the same shingle roof at a concentration of 0.20 mg/L. These compounds are both unregulated in drinking water. Our lack of detection of PAHs in harvested rainwater is consistent with Van Metre and Mahler (2003), who observed that metal and shingle roofing materials are not a source of PAHs but that PAHs come from atmospheric deposition. Our lack of detection of pesticides in

Table 1 e Water quality parameters (minimumemaximum) of the rainwater harvested after the first-flush for the pilot-scale and full-scale roofs. Parameter

Metal

Shingle

Pilot-scale Full-scale Pilot-scale Full-scale pH Conductivity (mS/cm) TC (CFU/100 mL) FC (CFU/100 mL) Turbidity (NTU) TSS (mg/L) Nitrate (mg-N/L) Nitrite (mg-N/L) DOC (mg/L) Lead (mg/L) Zinc (mg/L)

6.0e6.8 9e56

5.4e6.3 18e60

6.7e6.9 18e57

5.8e6.5 20e102

117e770 <1e8 7e30 20e87 0.0e2.0 0.01e0.03 2e11 0.3e2.3 77e362

64e173 37e127 5e35 10e50 0.4e4.1 0.01e0.05 4e13 2.1e5.8 18e23

177e1367 9e87 8e24 12e54 0.0e1.8 0.01e0.04 10e15 0.4e1.2 8e85

102e353 73e253 6e23 20e150 0.3e4.7 0.01e0.06 5e31 0.7e8.6 1e15

2058

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 4 9 e2 0 5 9

harvested rainwater is likely related to surrounding land-use, where major pesticide applications are absent.

4.

Conclusions

Conventional roofing materials (i.e., asphalt fiberglass shingle, Galvalume
metal, and concrete tile) and alternative roofing materials (i.e., cool and green) were examined for their suitability to harvest rainwater for domestic use. pH, conductivity, TC, FC, turbidity, TSS, nitrite, nitrate, DOC, metals, VOCs, and SVOCs were measured in harvested rainwater. The following conclusions were drawn:  Rainwater harvested from any of the tested roofing materials would require treatment if the consumer wanted to meet the USEPA primary and secondary drinking water standards or the USEPA non-potable water reuse guidelines. In particular, the rainwater harvested after a first-flush consisting of a minimum of 38 L for every 93 m2 of collection area from the asphalt fiberglass shingle, metal, concrete tile, and cool roofs would need treatment for TC, FC, turbidity, aluminum, and iron to meet the drinking water standards. The rainwater harvested after the first-flush from the green roof would need treatment for TC, FC, turbidity, and aluminum. Thus, at a minimum, first-flush diversion, filtration, and disinfection are recommended to meet current USEPA standards/guidelines. However, since the quality of ambient rainwater varies by location, we caution that rainwater harvested in other locations might exceed standards/guidelines for different water quality parameters than those observed in this study.  While metal roofs are commonly recommended for rainwater harvesting applications, the metal roofs examined in this study did not produce clearly superior harvested rainwater quality as compared to the other roofing materials. Concrete tile and cool roofs also appear to be good candidates for rainwater harvesting catchments, given the overall similarity in harvested rainwater quality among those three materials.  The DOC concentrations in rainwater harvested from the shingle and green roofs are very high (i.e., approximately one order of magnitude higher than that of a typical, finished drinking water in the U.S.) and could lead to high concentrations of disinfection byproducts. The formation of trihalomethanes and haloacetic acids, two classes of regulated disinfection byproducts in drinking water, should be assessed after disinfecting harvested rainwater with chlorine. The elevated DOC concentrations observed in the rainwater harvested from the shingle and green roofs in our short-term study suggest that other roofing materials (e.g., metal, tile, and cool) might be preferable for rainwater harvesting applications when chlorine is the disinfectant of choice.  The rainwater harvested from the metal roof showed lower concentrations of fecal indicator bacteria than did the other roofing materials. This might be due to the low emissivity of metal, resulting in higher surface temperatures on the roof.  The concentrations of some metals (e.g., arsenic) in rainwater harvested from the green roof suggest that the quality

of commercial growing media should be carefully examined if the harvested rainwater is being considered for domestic use.  Concentrations of most of the tested water quality parameters decreased as a result of diverting a first-flush of at least 38 L for every 93 m2 of collection area. While the depth of rainfall to be diverted is a subject of debate, a first-flush device can provide some improvement in water quality.

Acknowledgments We are grateful to several sources of funding for this work: the Texas Water Development Board (contract number 0804830855), Texas AgriLife Research (subcontract number 570464), the National Science Foundation (Graduate Research Fellowship to C.B. Mendez), and the American Water Works Association and CH2M Hill (Holly A. Cornell Scholarships to B.R. Afshar and C.B. Mendez). We thank Dr. Sanjeev Kalaswad and Mr. Jorge Arroyo from the Texas Water Development Board for their input on this project and Mr. David Phillips (JaMar Roofing) for donating the materials and labor for construction of the pilot-scale roofs. We also thank Mr. Charles Perego, Dr. Steve Windhager, Dr. Cindy Menches, Dan Hemme, Jeffrey Stump, Brett Buff, and David Stump for their assistance on the project.

references

Ahmed, W., Huygens, F., Goonetilleke, A., Gardner, T., 2008. Real-time PCR detection of pathogenic microorganisms in roof-harvested rainwater in southeast Queensland, Australia. Applied and Environmental Microbiology 74 (17), 5490e5496. American Public Health Association, 1998. Standard Methods for the Examination of Water and Wastewater, twentieth ed. Washington, D.C. Berndtsson, J.C., Bengtsson, L., Jinno, K., 2009. Runoff water quality from intensive and extensive vegetated roofs. Ecological Engineering 35 (3), 369e380. Bretz, S., Akbari, H., Rosenfeld, A., 1998. Practical issues for using solar-reflective materials to mitigate urban heat islands. Atmospheric Environment 32 (1), 95e101. Bucheli, T.D., Mu¨ller, S.R., Voegelin, A., Schwarzenbach, R.P., 1998. Bituminous roof sealing membranes as major sources of the herbicide (R,S )-mecoprop in roof runoff waters: potential contamination of groundwater and surface waters. Environmental Science and Technology 32 (22), 3465e3471. Despins, C., Farahbakhsh, K., Leidl, C., 2009. Assessment of rainwater quality from rainwater harvesting systems in Ontario, Canada. Journal of Water Supply: Research and Technology-AQUA 58 (2), 117e134. Egodawatta, P., Thomas, E., Goonetilleke, A., 2009. Understanding the physical processes of pollutant build-up and wash-off on roof surfaces. Science of the Total Environment 407 (6), 1834e1841. Evans, C.A., Coombes, P.J., Dunstan, R.H., 2006. Wind, rain, and bacteria: the effect of weather on the microbial composition of roof-harvested rainwater. Water Research 40 (1), 37e44.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 4 9 e2 0 5 9

Fo¨rster, J., 1998. The influence of location and season on the concentrations of macroions and organic trace pollutants in roof runoff. Water Science and Technology 38 (10), 83e90. Fo¨rster, J., 1999. Variability of roof runoff quality. Water Science and Technology 39 (5), 137e144. Kingett Mitchell Ltd., 2003. A Study of Roof Runoff Quality in Auckland New Zealand: Implications for Stormwater Management. Auckland Regional Council, Auckland, New Zealand. Lee, J.Y., Yang, J.S., Han, M., Choi, J., 2010. Comparison of the microbiological and chemical characterization of harvested rainwater and reservoir water as alternative water resources. Science of the Total Environment 408 (4), 896e905. Lye, D.J., 2002. Health risks associated with consumption of untreated water from household roof catchment systems. Journal of the American Water Resources Association 38 (5), 1301e1306. National Oceanic and Atmospheric Administration, 2010. Austin Climate Records e NWS Austin/San Antonio, National Weather Service. National Oceanic and Atmospheric Administration. http://www.srh.noaa.gov/ewx/?n¼ausclidata. htm (accessed July 2010). Nicholson, N., Clark, S.E., Long, B.V., Spicher, J., Steele, K.A., 2009. Rainwater harvesting for non-potable use in gardens: a comparison of runoff water quality from green vs. traditional roofs. In: Proceedings of World Environmental and Water Resources Congress 2009 e Great Rivers Kansas City, Missouri. Simmons, G., Hope, V., Lewis, G., Whitmore, J., Gao, W., 2001. Contamination of potable roof-collected rainwater in Auckland, New Zealand. Water Research 35 (6), 1518e1524. Simmons, M.T., Gardiner, B., Windhager, S., Tinsley, J., 2008. Green roofs are not created equal: the hydrologic and thermal performance of six different extensive green roofs and reflective and non-reflective roofs in a sub-tropical climate. Urban Ecosystems 11 (4), 339e348.

2059

Singh, T., Kalra, Y.P., 1975. Specific conductance method for in situ estimation of total dissolved solids. Journal of American Water Works Association 67 (2), 99e100. Sullivan, J.H., Worsley, D.A., 2002. Zinc runoff from galvanised steel materials exposed in industrial/marine environment. British Corrosion Journal 37 (4), 282e288. Teitzel, G.M., Parsek, M.R., 2003. Heavy metal resistance of biofilm and planktonic Pseudomonas aeruginosa. Applied and Environmental Microbiology 69 (4), 2313e2320. Texas Rainwater Harvesting Evaluation Committee, 2006. Rainwater Harvesting Potential and Guidelines for Texas. Texas Water Development Board, Austin, Texas. The Texas Manual on Rainwater Harvesting, third ed., 2005. Texas Water Development Board, Austin, Texas. USEPA, 1996a. Organochlorine Pesticides by Gas Chromatography SW-846 Ch. 4.3.1, Washington, D.C. USEPA, 1996b. Semivolatile Organic Compounds by Gas Chromatography/Mass Spectrometry SW-846 Chapter 4.3.2, Washington, D.C. USEPA, 1996c. Volatile Organic Compounds by Gas Chromatography/Mass Spectrometry SW-846 Chapter 4.3.2, Washington, D.C. USEPA, 2004. Guidelines for Water Reuse Washington, D.C. EPA/ 625/R-04/108. Van Metre, P.C., Mahler, B.J., 2003. The contribution of particles washed from rooftops to contaminant loading to urban streams. Chemosphere 52 (10), 1727e1741. Yaziz, M.I., Gunting, H., Sapari, N., Ghazali, A.W., 1989. Variations in rainwater quality from roof catchments. Water Research 23 (6), 761e765. Zobrist, J., Mu¨ller, S.R., Ammann, A., Bucheli, T.D., Mottier, V., Ochs, M., Schoenenberger, R., Eugster, J., Boller, M., 2000. Quality of roof runoff for groundwater infiltration. Water Research 34 (5), 1455e1462.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 6 0 e2 0 7 2

Available at www.sciencedirect.com

journal homepage: www.elsevier.com/locate/watres

Engineering of an MBR supernatant fouling layer by fine particles addition: A possible way to control cake compressibility Benoıˆt Teychene a,*, Christelle Guigui b, Corinne Cabassud c a

Universite´ de Toulouse, INSA, UPS, INP, LISBP, 135 Avenue de Rangueil, F-31077 Toulouse, France INRA, UMR792 Inge´nierie des Syste`mes biologiques et des Proce´de´s, F-31400 Toulouse, France c CNRS, UMR5504, F31400 Toulouse, France b

article info

abstract

Article history:

For membrane bioreactors (MBR) applied to wastewater treatment membrane fouling is

Received 19 October 2010

still the prevalent issue. The main limiting phenomena related to fouling is a sudden jump

Received in revised form

of the transmembrane pressure (TMP) often attributed to the collapse of the fouling layer.

16 December 2010

Among existing techniques to avoid or to delay this collapse, the addition of active particles

Accepted 19 December 2010

membrane fouling reducers (polymer, resins, powdered activated carbon (PAC), zeolithe.)

Available online 24 December 2010

showed promising results.

Keywords:

of inert particles (500 nm and inert compared to other fouling reducers) and which is the

Wastewater

impact on filtration performances of the structuring of the fouling. Those particles were

Membrane bioreactor

chosen for their different surface properties and their capability to form well structured layer.

Thus the main objective of this work is to determine if fouling can be reduced by inclusion

Membrane fouling reducers Fouling layer engineering

Results, obtained at constant pressure in dead end mode, show that the presence of particles changes foulant deposition and induces non-compressible fouling (in the range of 0.5e1 bar) and higher rejection values compared to filtration done on supernatant alone. Indeed dead end filtration tests show that whatever interactions between biofluid and particles, the addition of particles leads to better filtration performances (in terms of rejection, and fouling layer compressibility). Moreover results confirm the important role played by macromolecular compounds, during supernatant filtration, creating highly compressible and reversible fouling. In conclusion, this study done at lab-scale suggests the potential benefit to engineer fouling structure to control or to delay the collapse of the fouling layer. Finally this study offers the opportunities to enlarge the choice of membrane fouling reducers by taking into consideration their ability to form more consistent fouling (i.e. rigid, structured fouling). ª 2010 Elsevier Ltd. All rights reserved.

1.

Introduction

Membrane bioreactors (MBRs) are now widely applied for municipal wastewater treatment. As reported by Hanft (2006), Lesjean and Huisjes (2007) and Lesjean and Huisjes (2008),

since 2005 the MBR market has shown a significant increase of about 11% per year. According to Lesjean and Huisjes (2008), in the coming years at least 70 new MBRs plants are expected to be constructed each year. This large development was impulsed by the growing interest for immersed membranes which were

* Corresponding author. Tel.: þ33 5 61 49 38 46. E-mail addresses: [email protected] (B. Teychene), [email protected] (C. Guigui), Corinne.Cabassud@ insa-toulouse.fr (C. Cabassud). 0043-1354/$ e see front matter ª 2010 Elsevier Ltd. All rights reserved. doi:10.1016/j.watres.2010.12.018

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 6 0 e2 0 7 2

initially proposed by Yamamoto et al. (1989) and are today the predominant system, because of their reduced energy demand in comparison with previous membrane systems. It is also nowadays supported by strict environmental regulations and the decrease of membrane costs (Judd, 2008). However the widespread application of MBRs processes still suffers of membrane fouling caused by the activated sludge compounds. During the last few years MBR fouling was extensively studied by focussing on the different possible factors influencing fouling (sludge characteristics, operating conditions (Chang et al., 2002), hydraulic conditions, (Le-Clech et al., 2006), aeration, (Pollet et al., 2009) membrane materials (Meng et al., 2009), etc). Some process choices or operating procedures are known as key ways to mitigate fouling: adapted module design, aeration conditions, backwashing and intermittent filtration (relaxation). Despite numerous studies, knowledge on the fouling mechanisms and on the fouling compounds is still a key issue. Indeed an activated sludge (AS) is a complex and a highly variable suspension containing an array of microorganisms, which are present in the wastewater, and metabolites such as extracellular polymeric substances (EPS) produced by the microorganisms inside the membrane bioreactor. EPS compounds are the construction materials for microorganism aggregates (such as flocs, biofilm, etc.) and consist of a large array of organic polymers (proteins, polysaccharides, nucleic acids and humic substances.) (Flemming and Wingender, 2001). These compounds can be found in different forms: soluble, colloidal and particulate materials. Each of these fractions might affect membrane performances by different fouling mechanisms such as pore blocking, cake formation and adsorption. The resulting fouling and the most prevalent associated issue are a sudden jump of the transmembrane pressure (TMP). This TMP jump was attributed to the change in local fluxes (Cho and Fane, 2002), or to a sudden change in biomass or cake layer structure by compressibility effects (Masse´ et al., 2006; Zhang et al., 2006). A recent study showed that the TMP jump could be due to an increase of EPS at the bottom of the cake layer, probably due to the death of the bacteria (Hwang et al., 2008). For more than a decade many studies tried to investigate the respective role of each fraction of the biofluid in terms of filtration resistance. The possible role and contribution of EPS is still controversed but for numerous studies the most problematic compounds for fouling are the biomacromolecular compounds and are identified as protein substances and/or polysaccharides leading to a highly hydrated gel (Okamura et al., 2009; Teychene et al., 2008; Rosenberger et al., 2006; Bouhabila et al., 2001; Lesjean et al. 2007). However as recently reviewed by Drews (2010) results are still controversed due to the large variety of experimental methods (lab or full scale experiments) and to the definition of fouling relevant fraction (biopolymers, soluble or extracellular microbial products, transparent exopolymer particles). Nevertheless, much important research work focuses on the reduction of macromolecular compounds in the supernatant (or interstitial liquid) because of their serious hindrance to membrane separation processes. Upon these techniques the most interesting results were obtained by the addition, inside the biofluid, of chemicals compounds that might be: salts,

2061

polymers (Koseoglu et al., 2008; Zhang et al., 2008; Hwang et al., 2007; Yoon et al., 2005), adsorbent and resins (Remy et al., 2010; Ying and Ping, 2006, Fang et al., 2006, Lesage et al., 2005; Kim and Lee, 2003; Lee et al., 2001) and nanoparticles (Chae et al., 2009). Yoon et al., (2005) have reported that the addition of cationic polymer in a sludge was able to entrap macromolecular compounds within the flocs and to enhance the permeate flux. In a same way Lee et al. (2007), Hwang et al. (2007), Ji et al. (2008) and Iversen et al. (2008, 2009a,b) have demonstrated that the addition of cationic polymer or coagulant led to bigger and more porous flocs. Moreover as demonstrated by Lesage et al. (2005), Guo et al. (2006) and Ying and Ping (2006), the addition of adsorbent fine particles (activated carbon, zeolithe) in MBR allows to enhance filtration performances and to delay fouling (Remy et al., 2009; Kim and Lee, 2003; Fang et al., 2006; Park et al., 1999). Recently Chae et al. (2009) have demonstrated the potentiality of using colloidal suspension to control biofouling in MBR. Indeed they have demonstrated that microfiltration membrane coated with nanoparticles (fullerene C60) could reduce the microbial attachment and the respiratory activity during short-term filtration experiment. Different mechanisms were suggested as an explanation for the improvement of the filtration performances : (1) adsorption of foulants to the PAC and thus protection of the membrane surface from foulants deposition (Lesage et al., 2005; Ng et al., 2008), (2) Scouring effect removing foulants deposited at the membrane surface (Park et al., 1999; Fang et al., 2006), (3) improvement of colloids and macromolecules flocculation leading to stronger floc strength (Li et al., 2005), (4) bacterial inhibition by colloidal aggregates (Fabrega et al., 2009; Chae et al., 2009; Sondi and Salopek-Sondi, 2004) and (5) formation of a more porous and less compressible cake layer (Pirbazari et al., 1996; Lesage et al., 2005). The objective of this paper is to focus on this last assumption and to investigate how a voluntary inclusion of fine inert particles formed during real biofluid filtration (without or very slight coagulation, flocculation or adsorption effect) can impact on the filtration performances of the structuring of the fouling layer. The aim is to check if it is possible to engineer the structure of the fouling layer in an MBR to control or to limit fouling. Thus the present paper introduces an experimental study of the artificial structuring of a cake layer formed from a biofluid (supernatant of a sludge sampled in an MBR) using submicron synthetic particles of different properties. The particle properties were chosen with the objective to determine which parameter between structuring and adsorption is the most responsible in the filterability improvement of an AS supernatant in presence of particles. For this it was proposed here to use fine monodispersed particles having the same spherical shape and size but significantly different surface properties (surface charge, hydrophilicity.). Filtration experiments aimed to characterise the impact of the addition of the fine particles in MBR supernatant on the flux decline, on fouling layer properties (specific resistance, compressibility, flow potential) in relation with the quality of the biological fluid. For that filtration tests were performed in dead end at constant pressure with suspension of particles (in pure water, solutions of supernatant without particles and after

2062

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 6 0 e2 0 7 2

mixing with particles). In order to investigate the possible interactions between the added particles and the organics in the supernatant, the interactions were characterised by monitoring different parameters such as the electrostatic surface charges (by streaming potential) and the particle size distribution. The modification of the organic matrix of the supernatant was characterised using HPLC-SEC measurements with a fluorescence detector.

2.

Materials and methods

2.1.

Particle suspensions and biofluids

2.1.1.

Biofluids

In order to determine the role of fine particles in the formation of the fouling layer induced by colloidal and soluble compounds, AS was sampled from an MBR pilot plant situated on site on a wastewater treatment plant (Labe`ge, France, Sludge Retention Time (SRT) ¼ 40 days; Mixed Liquor Suspended Solids (MLSS) ¼ 8 g l1). Indeed a decision was made to work with an AS fed with a real wastewater, more representative of the biomass complexity than a synthetic one. For the fractionation of particular, colloidal and soluble compounds, AS was centrifuged at 4000 g during 10 min at room temperature (Teychene, 2008). As used by Yamato et al. (2006) this fractionation step could remove all suspended solids from the liquid phase. The supernatant contains only colloids and soluble compounds. No particle was detected during particle size measurements which were performed on the supernatant. Thus it was considered that no particle remains in the supernatant and that colloids are poorly present. The organic carbon content (DOC) of the supernatant was around 7.6  0.1 mg l1. The conductivity of the supernatant was equal to 736 mS cm1. In addition, the supernatant organic matters are negatively charged and have an average zeta potential (z) of about 9.5  1 mV.

2.1.2.

Suspensions of sub-micron particles

When the sub-micron particles are put in contact with the supernatant different interactions are expected. Latex and melamine sub-micron particles were selected for this study. Indeed as reported by a study of Jons et al. (1999) the use of small synthetic particles could form well-structured particles array at the membrane with a broad flexibility in surface chemistry. Those particles (deposited at the membrane) were extensively studied in terms of processing and properties (Jons et al., 1999; Strathmann, 1990; Porter, 1990). However those particles were not investigated during the filtration of a biofluid.

In order to control the interactions between the organic matters from the supernatant and the particles, two different synthetic particles (From Duke scientific Corporation, USA) of the same average size (z500 nm, given by the manufacturer) have been chosen. Each kind of sub-micron particle was used at a constant concentration of about 112 mg l1. The choice of this concentration will be more detailed later in this paper (in part 2.2.2). The first suspension consisted of (i) polystyrene latex particles which are negatively charged with a hydrophobic surface (LogKow > 0); the second one consisted of (ii) particles based on melamine resin, and thus positively charged with a hydrophilic surface (LogKow < 0). These particles were firstly chosen for MBR applications due to their small particle diameter. Indeed as reported by Jons et al. (1999) such particles could pack, during filtration of stabilised colloids suspension, at the membrane in regular crystalline arrays (i.e. hexagonal and/or cubic closest packing). Therefore these particle layers have specific porosity (depending on the packing) and specific pore size or interstitial spaces of narrow distribution (depending on particle diameter). Characteristics of each kind of particles are given in Table 1. Log Kow represents the octanolewater partition coefficient and is used to characterise the hydrophobic character of materials. As shown on Table 1, Melamine is more prone to hydrogen bonding than polystyrene latex. Particle size measurement showed that particles are monodispersed with a coefficient of variation lower than 10%. In addition, measurement of zeta potential in the same condition showed that latex particles are negatively charged, with a charge of around 27  1 mV, while melamine particles are positively charged (11  1 mV).

2.2.

Membranes and filtration tests

2.2.1.

Membranes

Microfiltration membranes (from Alfa laval, FR), made of polysulfone, with a nominal pore size of 0.1 mm (Lpo ¼ 250 l h1 m2 bar1 at 20  C) were used in this study. Polysulfone polymer gives a certain hydrophobic character to the membrane and slight electrostatic charges (1.6  0.2 mV, obtained by streaming potential measurements on virgin membrane). A new membrane sample was used for each filtration test. All membranes were soaked in ultra pure water over night to maintain pores wetted. They were then rinsed with ultra pure water at 1 bar during 10 min prior to filtration test. All filtration tests were carried out in an Amicon cell, with a membrane surface of about 12.5 cm2, at constant transmembrane

Table 1 e Characteristics of the sub-micron particles.

Polystyrene latex Melamine

Log Kowb

Mean diameter (nm)a

Densityb (g/cm3)

H-Bond donor/acceptorb

z (mV)a

3 1.4

542  30 498  20

1.05 1.5

0/0 3/6

27  1 11  1

a Measurements done in the lab in 1  103 mol l1 of KCl (145 mS cm1) and particle concentration set at 112 mg l1. b Value obtained from U.S National centre for biotechnology Information [http://pubchem.ncbi.nlm.nih.gov/].

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 6 0 e2 0 7 2

pressure (TMP). Transmembrane pressure was provided by pressurised nitrogen.

2.2.2.

Filtration of supernatant and particles separately

In order to investigate the role of the sub-micron particles at low concentration on the filtration performances of a biofluid, each suspension has to be filtered separately. In each case (supernatant or particles suspension separately) filtration was performed in the same operating conditions. As said before concentration of particles is set at a constant value of about 112 mg l1. In addition, according to simple geometrical considerations, particle diameters (Table 2) were chosen to achieve a pore size of around 0.1 mm, that is to say closed to the average membrane pore size. Indeed a theoretical calculation showed that whatever the particle-packing mode (hexagonal or cubic) the theoretical pore size of the particle layer was centred around 0.1 mm (0.07 mm and 0.11 mm for hexagonal and cubic packing, respectively). More details on calculation of particle arrays pore size are given by Jons et al. (1999). A preliminary filtration test, performed at the constant pressure of 0.5 bar, (concentration of about 112 mg l1 in ultra pure water, results not shown here) showed that the particle arrays induced a permeate flux decline from 5 to 27% for a deposited mass of particles varying from 5 to 22 g m2, respectively. Thus a decision was made to use a deposited mass of particles of about 11 g m2 (inducing a reasonable flux decline of 15%). As all suspensions filtered in this study are stable (no settling were observed by turbidity measurements), each filtration test was performed without stirring. The TMP was set in the range of 0e1 bar. A large vessel (5 l) was connected to the filtration cell to allow longer filtration times. Indeed whatever the suspension (supernatant alone, or particles suspensions alone), filtration tests were stopped when the filtered volume reached 0.1 m3 m2. The permeate flux was monitored every 20 s, with an electronic balance. Experiments were performed at room temperature (20  C). After filtration the membrane was turned face down and was backwashed with 50 mL of ultra pure water

Table 2 e Summarised results, obtained after 1 h, when the latex and melamine particles are added to an AS supernatant. Latex HPLCeSEC

Global analysis

Macromolecular compounds Area of peak 1 (fluorescence signal loss %) Low molecular weight compounds Area of peak 2 (UV 254 loss %) PSD (nm) Zeta potential (mV) DOC Loss (%)

15

18

1230 8 Variation in the range of analytical error

Melamine 84

28

1800 2 8

2063

at 1 bar, in order to investigate flux recovery by permeability measurement. Moreover all filtrations done at 0.5 bar were repeated three times to test the reproducibility of filtration test and to investigate fouling layer properties (by streaming potential and scanning electron microscopy).

2.2.3. Filtration of a mixture of MBR supernatant plus submicron particles When synthetic particles were mixed in suspension into the supernatant their concentration was equal to 112 mg l1. This particle concentration was chosen in order to study the impact on filtration performances of a biofluid of a really small quantity of particles (in comparison to PAC concentration often reported in literature, Remy et al., 2009). The stabilisation time (with slight stirring 140 rpm at 20  C) was about 1 h, which is sufficient for the stabilisation of particle interactions for small particles. After this time the stabilised suspensions were directly filtered at the chosen pressure with the same operating condition as said before. Therefore at the end of the filtration test (for a filtered volume of 0.1 m3 m2) the deposited mass of sub-micron particles was equal to 11 g m2. Once again, in order to investigate filtration test reproducibility and fouling layer characteristics, each filtration test at 0.5 bar was repeated three times.

2.3. Determination of filtration performances and of fouling layer properties The filtration performances of supernatant (with or without particles), during each dead end filtration test, were studied in terms of flux decline trends and fouling rate using filtration laws. First of all the membrane resistance was determined according to Eq. (1). J¼

DP hðRm þ Rc Þ

(1)

With J is the permeate flux (l h1 m2), Rm is the membrane resistance (m1) and Rc is the fouling layer resistance (m1), DP is the transmembrane pressure (bar) and h is the fluid viscosity at 20  C (Pa s). The fouling rate dRc/dV, calculated as the variation of Rc against cumulative filtration volume, was used as an indicator of the fouling ability of the filtrated suspension. After membrane backwash the flux recovery was calculated as given by (Eq. (2)).   Lpbw Recovery ¼ 100 Lpo

(2)

Where Lpbw is the membrane permeability after backwash (l h1 m2 bar1) and Lpo is the initial membrane permeability using ultra pure water (l h1 m2 bar1) at 20  C. Finally the apparent organic matter rejection was determined by the following equation (Eq. (3)).   Cp Rejection ¼ 100 1  Cb

(3)

Cb and Cp are the concentrations (in mg l1) of dissolved organic carbon of the bulk and permeate suspensions, respectively. Cb and Cp were obtained by DOC measurements.

2064

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 6 0 e2 0 7 2

The properties of each fouling layer were investigated in terms of specific cake resistance calculated from the filtration law for filtercake filtration at constant pressure (Eq. (4)). t haC hRm Vþ ¼ SDP V 2S2 DP

(4)

Where a is the cake specific resistance (m kg1); h is the fluid viscosity at 20  C (103 Pa s); S is the membrane surface (m2); V is the permeate cumulated volume (m3); t is the filtration time (s) and C is the deposited mass per permeated volume (kg m3). Depending on the considered suspension the deposited mass could be calculated from the particle concentration or from the supernatant organic matter rejection. For cake layers made of particles only (without supernatant) specific resistances were calculated from the particles deposited mass (11 g m2). When fouling layers are made of organic matter from supernatant and particles, the calculation of the deposited mass could be calculated from (i) the global deposited mass (particles þ organic matter), (ii) deposited mass of organic matter only or (iii) deposited particles only. However, it is important to note that this study focuses only on how sub-micron particles could influence the deposition of organic matter during filtration or the fouling layer properties. Consequently fouling layers made of supernatant (with or without particles) have to be compared on the same criteria. Thus it was chosen to calculate the specific resistance on the basis of the deposited organic matter only. Therefore when fouling layers involve supernatant, with or without particles, the specific resistance was calculated using the deposited mass (C) obtained on the basis of a mass balance from the DOC measurements (Eq. (5)). VC ¼ mb  mp

(5)

Where V is the permeate volume, mb is the DOC content (in mg) sent to the membrane and mp is the DOC content (in mg) in the permeate. As widely reported in literature one of the most important issue of MBR processes is the TMP jump assumed to be due to the fouling layer compaction (Masse´, 2004). Thus an important characteristic of a fouling layer in MBR processes, is its compressibility against transmembrane pressure. Consequently all fouling layers investigated in this study were compared in terms of compressibility calculated from Eq. (6). a ¼ a0 DPn

(6)

Where a is the specific resistance (m kg1), n the compressibility factor and a0 is a constant obtained at y-intercept of the logelog plot of a versus DP (m kg1). Moreover the average cake porosity (3) of the membranes fouled by particles is determined from the value of cake specific resistance thanks to the CarmaneKozeny equation (Eq. (7)). a¼

180ð1  3Þd2p rp 33

organic matter (from supernatant). Indeed CarmaneKozeny equation is hardly usable for cake layers made of organic matter with or without particles, due to the uncertainty on the particle diameter inside the cake layers.

(7)

With a is the specific resistance of the particle cake layer (m kg1), dp is the mean diameter of the particle suspension (m) and rp is the particle density (kg m3) and 3 is the cake porosity. Note that the CarmaneKozeny equation was used here to characterise layers made of particles only, in the absence of

2.4.

Analytical methods

2.4.1. Particle size distribution (PSD) and zeta potential measurements The hydrodynamic diameter and electrophoretic mobility of sub-micron particles (latex and melamine) were measured over the range of suspensions created in this study (with or without supernatant) by dynamic light scattering using a Nanosizer ZS (Malvern SA, FR). This apparatus allowed to measure particle sizes ranged from 0.6 nm to 6 mm. Electrophoretic mobility was measured at 25  1  C. The electrophoretic mobility was converted to zeta potential using the Smoluchowski principle (Von Smoluchowski, 1903). Each measurement was repeated at least three times on all samples.

2.4.2.

Dissolved organic carbon (DOC measurement)

DOC measurements were performed with a Total Organic Carbon analyser (Shimadzu, FR). Before the analyses each sample was prefiltered at 0.45 mm, in order to remove particles. All prefilters used were rinsed with ultra pure water to remove any organic contamination. Each DOC measurement was performed three times and the standard deviation is about 0.1 mg l1.

2.4.3. High performance liquid chromatography e size exclusion chromatography (HPLCeSEC) As reported by previous studies, the supernatant of activated sludges sampled from an MBR contains polysaccharides, proteins like substances and humic like substances with various molecular weight. It is generally accepted that macromolecular compounds consist of polysaccharides and protein-like substances, thus only the protein-like structure of those macromolecules were investigated here. The aim of this study was to state on the behaviour of protein substances during filtration of an AS supernatant. As this study focuses only on the protein-like substances, therefore different spectrophotometers were used (UV and fluorescence) associated to the HPLCeSEC line to determine the presence of protein-like substances. The use of fluorescence spectroscopy could only bring information on protein substances fate. Therefore no information was brought in this study on polysaccharide compounds and their fate with particles. The UV wavelength 254 nm was chosen in order to characterise double bonds (carbon sp2). The UV detection was used as a global indicator of organic matter (with double bonds) in each fluid. A fluorescence detector was added (Varian, USA) at the end of the HPLCeSEC line. Emission/excitation wavelengths were set at 280/350 nm, as reported by Her et al. (2004) such wavelengths are more specific and are used for the detection of protein-like substances (Teychene et al., 2008; Her et al., 2004; Amy, 2008; Galinha et al., 2008). Those wavelengths were chosen by these authors on the basis of an excitation emission matrix done on various samples and reference materials.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 6 0 e2 0 7 2

Each peak obtained by these detectors was used as quantitative analysis by integration of peaks. HPLCeSEC apparatus was an Akta Purifer (GE Healthcare, USA). The column Protein Kw804, (Shodex, JP), was a silica-based column (particle size 7 mm, 8.0  300 mm) with an exclusion limit of 1 MDa. Column calibration was done with PEG standards (Sigma Aldrich, FR) ranged from 1 MDa to 1 kDa. The buffer solution for the mobile phase was prepared with 25 mM Na2SO4 and a phosphate buffer (2.4 mM NaH2PO4 and 1.6 mM Na2HPO4 at pH 6.8). Ionic strength of a sample was adjusted to that of the mobile phase, with the intention of minimising the interactions between the stationary phase and sample components. The flow rate was set at 1 mL min1 and the injection volume was equal to 1 mL; prior to analysis, each sample was prefiltered at 0.45 mm.

2.4.4.

Fouled membrane analysis

2.4.4.1. Streaming potential measurements. In order to investigate electrostatic properties of virgin or fouled membranes, streaming potential (SP) measurements were performed using a ZETACAD (CAD Instrument, FR). Each flat sheet membrane (clean or fouled) was placed into the measuring cell. Two reversible electrodes made of Ag/AgCl placed on each side of the membrane were used to measure the voltage difference (DE ) induced by various discreet pressure gradients (DP). Each measurement was performed with a KCl electrolyte at 1  103 mol l1. Before each measurement electrostatic charges were stabilised during 1 h by flowing the KCl electrolyte through the membrane at 0.5 bar TMP. As reported by various studies, streaming potential is measured by applying different pressure steps (in descending order) ranged from 0 to 0.5 bar across the membrane. The pressure is decreased only when a stable potential is reached (Nystro¨m et al., 1994; Wang et al., 2006; Teychene et al., submitted for publication). Each measurement was repeated twice. The zeta potential (z) of the membrane sample can be obtained by applying the Helmholtz principle (Eq. (8)). SP ¼

DE 3r 30 z ¼ DP hl0

(8)

With SP, the streaming potential (mV Pa1), DE, the voltage difference (mV), DP, pressure gradient (Pa), 3r and 30 are the relative static water permitivity and the void permitivity (3r30 ¼ 6.95  1010) respectively. h is the water viscosity at 20  C (103 Pa s) and l0 the electrolyte conductivity (145 mS cm1).

2.4.4.2. Scanning electron microscopy. Once dried at room temperature during 24 h in dust free atmosphere, each fouled membrane surface was thin-coated with carbon. A scanning electron microscope working at primary void (JEOL 5410 LV Instrumentation) was used to obtain micrographs of the top surface of the fouled membrane samples.

3.

Results and discussion

Results will be discussed on the basis of characterisation of the particles and supernatant interactions and of filtration performances for the different suspensions. Finally it will allow to propose a global and critical comparison to state on

2065

the role of adsorption or fouling layer structuring on membrane filtration performances.

3.1. Characterisation of the interactions between OM from MBR supernatant and particles After 1 h of contact time, all suspensions of particles in AS supernatants were characterised by various analytical tools. PSD measurements, done on suspensions consisting of particles and supernatant after 1 h of contact time, have shown that melamine mixture had a mean diameter centred on 1800  77 nm (498  20 nm alone in suspension) whereas the mean particle size of the latex mixture was equal to 1230  60 nm (542  20 nm alone in suspension). Therefore the synthetic particles and OM from supernatant were aggregated and the presence of the melamine particles has induced bigger aggregates than the presence of the latex particles. Electrophoretic measurements have shown an average zeta potential value of 2 mV and 8 mV for the melamine and the latex aggregates respectively. Organics from supernatant have completely changed the melamine surface charge (initially 11 mV) and have partially cancelled the latex particles surface charges (initially 27 mV). Different phenomena could occur during the aggregation process, the cancellation of the latex particles charges could be due to the presence of cations (such as calcium) or the OM could have totally coated the surface of the particles (resulting in a global charge closed to the OM alone in suspension around 9 mV). During the stabilisation time, a small amount of suspensions was sampled (every 15 min) and analysed by HPLCeSEC and DOC measurement. DOC analyses done on melamine mixture showed a loss of 8% of DOC after 30 min and afterwards values were constant, whereas values obtained for the latex mixture were constant during 1 h. Thus, considering the measurement error (0.1 mg l1, around 2% of the initial DOC value), it can be considered that particle addition induces only a very slight adsorption of organic matter (from supernatant). As reported on Fig. 1a, the HPLCeSEC analysis on the supernatant shows a typical “fingerprint” of an MBR supernatant. Two peaks are observed. The first peak (peak 1) (with a molecular weight of about 250 kDa and a small response to fluorescence signal) is attributed to macromolecular proteins like substances. The second peak (peak 2), with a molecular weight lower than 10 kDa, is attributed to small protein-like substances due to a high fluorescence signal of around 150 mV. For both suspensions HPLCeSEC results after 1 h contact time are given in Fig. 1b and c. As it can be seen, the area of the macromolecular protein peak has decreased (in fluorescence intensity) by 84% and 15% for suspensions in comparison to the supernatant alone (peak 1 on Fig. 1a), for melamine and latex particles respectively. Therefore the melamine particles (Fig. 1b) adsorb more macromolecular proteins than the latex particles (Fig. 1c). Furthermore, with particles the peak characteristics of the small compounds (<10 kDa) have changed compared to the one obtained for the supernatant (Fig. 1a). Indeed UV254 peak (peak 2) area is 28% and 18% lower than in the supernatant, for a melamine mixture and a latex mixture respectively. In conclusion these results, summarised in Table 2, indicate that organic matter from supernatant are clearly aggregated with synthetic particles.

2066

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 6 0 e2 0 7 2

Fig. 1 e HPLCeSEC analysis of supernatant alone (a) mixed with melamine prefiltred at 0.45 mm (b) and latex prefiltered at 0.45 mm (c) after 1 h. Proteins detection by fluorescence (Ex 280/Em 350).

The macromolecular compounds that are present in an AS supernatant show more affinity with melamine particles. It is widely accepted that organic matter is negatively charged, thus organic matter could link to melamine particles by electrostatic attractions and hydrogen bonding. In addition, organic compounds are less aggregated with latex particles due to electrostatic repulsion or hydrophobic forces. Therefore these results show that the supernatant matrix is modified when fine particles are added. It is also demonstrated that macromolecular proteins adsorb more on melamine particles and lead to bigger and less charged aggregates than with the latex particles. Thus during filtration tests, it can be expected that a melamine/supernatant mixture will induce less fouling and a more porous fouling layer than a latex/supernatant mixture. This point will be checked in the following of this paper.

3.2.

Filtration tests

Previous results show that the organic matrix of the supernatant is modified by the presence of particles in suspension. It was demonstrated that different interactions occur depending on the particles properties. Thus, in order to determine the impact of the added particles on the filtration performances for a supernatant,

it was firstly necessary to determine separately the filtration performances of particle suspensions and of supernatant. Then the filtration performances of the supernatant were investigated in presence of particles mixed in suspension.

3.2.1.

Filtration of sub-micron particles in pure water

Fig 2a represents the permeate flux decline trends versus the permeate volume during the filtration of the sub-micron particle suspensions. Whatever the particles (latex or melamine) or the applied transmembrane pressure no difference in terms of flux decline was observed. Indeed at the same deposited mass (11 g m2) both particle type induced the same flux decline of around 15%. Membrane showed a total rejection of the sub-micron particles, checked by turbidity and PSD measurements done on permeate. Furthermore TMP has no effect on the specific cake resistances: fine particles deposits are non compressible. The specific resistances are around 13  1012 and 19  1012 m kg1 for latex and melamine particles respectively. The flux recovery after membrane backwash for both particles exhibits an average value of about 98%. The CarmaneKozeny equation (Eq.(7)) leads to an average cake porosity of 32% and 27% for latex and melamine particles respectively. These results and the SEM picture (Fig. 2c) show

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 6 0 e2 0 7 2

2067

Fig. 2 e Filterability of particle suspensions (a), Streaming potential of membrane fouled by particles (at 0.5 bar) and (c) SEM picture of latex cake layer (obtained at 0.5 bar).

that particle cake deposits are close to a hexagonal compact packing (highlighted on Fig. 2c by black hexagon). As reported by Jons et al. (1999), the pore size of such particle layers is primarily governed by the particle size. Thus the pore size of the particles array, defined as the interstitial space between particles, can be calculated from simple geometry and is of about 0.08 mm in both cases. Such pore size is similar to the virgin membrane pore size (0.1 mm). However beyond geometrical consideration the electrostatic properties are other important parameters in membrane fouled by particles cake deposit. Therefore, streaming potential measurement (Fig. 2b) performed on membrane fouled with melamine or latex cake layer leads to a zeta potential (Using Eq. (8)) of about 0.60  0.08 mV and 3.90  0.40 mV, respectively. Note that streaming potential measurement by pressure steps could be performed because of the non-compressibility of the particle cake layers. Compared to the virgin membrane (1.60  0.20 mV), one can see that cake deposit made of latex particles is negatively charged whereas the one made of melamine particles is positively charged (zeta potential is lower in absolute value than the virgin membrane’s zeta potential). Those cake layers

electrostatic properties are linked to the electrostatic properties of each particles suspension (as given in Table 1). In conclusion, those results show that particles cake deposits (without organic matter from supernatant) are non compressible, well organised (compact packing) and show different electrostatic properties depending on the particle electrostatic charges.

3.3.

Filtration of supernatant without added particles

As shown on Fig. 3a and on Table 3 flux decline trends and fouling rate during supernatant filtration are similar whatever is the TMP. However DOC rejection values decrease as the TMP increases (or filtration period decreases). Rejection, in terms of DOC, is about 13% at 0.5 bar and 5% at 1 bar (Table 3). It should be noted that these values are apparent rejections throughout the filtration period. Whatever the transmembrane pressure used for the filtration test, flux recovery after membrane backwash is higher than 90% meaning that the fouling induced by organic matter is easily reversible by backwashing. The compressibility factor of the fouling is about 1.5, thus in this pressure range the fouling is defined as highly compressible and is closed to the

2068

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 6 0 e2 0 7 2

Fig. 3 e Filterability of supernatant alone on a virgin membrane, TMP [ 0.5; 0.8; 1 bar (25  C) (a). HPLC-SEC analysis of the permeate after supernatant filtration on a virgin membrane (TMP [ 0.5 bar, T [ 20  C) (b).

compressibility factor found from filtration of a biological fluid (Foley, 2006; Chellam and Xu, 2006). Therefore due to this high compressibility factor, no streaming potential measurement by pressure step could be performed on such a fouling. Consequently a new method to measure at constant pressure streaming potential of highly compressible fouling layer was developed (Teychene et al., submitted for publication). Fig. 3b demonstrates that no macromolecular protein remains in the permeate, as peak 1 is absent (after supernatant filtration at 0.5 bar). Therefore these compounds are totally rejected by the membrane. Moreover chromatogram emphasises that the area of the peak 2 corresponding to small proteinlike substances (UV 254 or fluorescence signal) remains unchanged after filtration, compared to one obtained by analysis of the supernatant (Fig. 1a). Therefore, during filtration of an MBR supernatant, main foulant species are shown to be macromolecular compounds leading to highly compressible but easily removable fouling, so called biofouling.

3.3.1. Filtration behaviour of a mixture of MBR supernatant and particles At this point, properties of a particle cake deposit or of a fouling obtained with organic matter alone are identified.

Thus that information will be helpful to state on the influence of particles during the filtration of a biofluids in presence of sub-micron particles. As shown on Fig. 4a, the flux decline (at 0.5 bar) is much lower for the melamine þ supernatant mixture than for the latex þ supernatant mixture and for the supernatant. With the TMP (Table 3), the same behaviour was observed: the fouling rates induced by the melamine/supernatant mixture are much lower than for the latex/supernatant mixture or supernatant alone. Moreover as reported on Table 3, the fouling rates for the latex/supernatant mixture are similar to the values obtained for the supernatant alone (Table 3) and are constant with the pressure. Rejection values increase with TMP and are equal to 14% at 0.5 bar and 24% at 1 bar for the latex mixture. In the case of the melamine mixture, the DOC rejection shows the same behaviour but is lower than for the latex mixture (from 10% at 0.5 bar to 18% at 1 bar). Rejection values are higher for particle mixtures than for the supernatant alone. Fig. 4b represents chromatogram of permeate obtained after the filtration at 0.5 bar of latex/supernatant mixture. Chromatogram of permeate of supernatant/ melamine shows similar results and is not shown here. Fig. 4b

Table 3 e Filtration performances of the supernatant alone and the mixed suspensions filtrated on a virgin membrane. dRc/dV  1015 DOC rejection Flux recovery a  1015 m Compressibility DP J/J0 3 2 m1 m3 (%)a (%) 6% kg1 factor (bar) (%) 0.1 m m Supernatant

Melamine particles þ supernatant Latex particles þ supernatant

0.5 0.8 1 0.5 0.8 1 0.5 0.8 1

65 58 56 85 74 71 59 55 54

8 9 10 2 5 5 8 10 10

13 9 5 10 13 18 14 20 24

93 99 92 95 96 97 98 99 98

9.1 14.6 28.8 3.3 2.5 3.5 7.0 6.8 6.3

a Rejection values based on DOC measurement done on the initial mixture prefiltered at 0.45 mm (Cb) and on the permeate (Cp).

1.5

0

0

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 6 0 e2 0 7 2

2069

Fig. 4 e Flux decline trends versus filtered volume of an AS supernatant mixed with melamine or latex particles (a) (TMP [ 0.5 bar; T [ 20  C); HPLC-SEC measurement of permeate of the latex/OM mixture (b).

shows that no macromolecular compounds in permeate (peak 1 is absent). This result shows that at 0.5 bar small organic matter compounds (peak 2) are unchanged (compared to Fig. 1c) and is identical to peak 2 on the analysis done on permeate of the supernatant (Fig. 3b). Thus according to rejection values at this pressure (0.5 bar, Table 3) same rejection is observed compare to rejection obtained during the filtration of supernatant alone. Thus as no macromolecular compounds remain in permeate at this lowest pressure (0.5 bar), the increase of OM rejection for higher TMP is assumed to be due to a better rejection of those small compounds (probably due to sieving effect induced by a possible change of the cake porosity). As said before, proteins adsorb more on melamine than on latex particles leading to bigger melamine aggregates. Moreover in the latex mixture macromolecular proteins are slightly adsorbed. As shown on Table 3, in both cases the specific resistance of fouling layer is constant with the TMP, and consequently the compressibility factor is equal to 0. Therefore whatever the interactions between particles and OM, the presence of

small particles modifies the comportment of organic matter fouling to pressure effects and makes the fouling becoming non-compressible. However, it is demonstrated that when the macromolecular proteins slightly interact with particles (latex particles) the flux decline is more severe. Indeed on the basis of Table 3 and specific resistances obtained for particle deposit, organic matters alone induce a more resistant fouling compared to the particle cake built with particles alone. The particles provide a lower contribution to fouling resistance compared to organic matter, thus specific resistances are calculated by using the amount of deposited organic matter (determined by DOC measurements). After membrane backwash the flux recovery is, in all cases, higher than 90%. The presence of sub-micron particles does not influence the high reversibility of the OM fouling. Finally as shown on Fig. 5a, the fouling formed by OM from supernatant/latex particles aggregates leads to a global zeta potential of the fouling structure (i.e. membrane þ fouling) (using Eq. (8)) of about e 6.40  0.80 mV. These results show that the electrostatic repulsions inside the fouling are

Fig. 5 e Streaming potential measurements of the fouled membranes (a). SEM picture of the fouling layer obtained form the filtration of the latex/OM mixture at 0.5 bar (b).

2070

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 6 0 e2 0 7 2

important, confirming that the macromolecular compounds are slightly adsorbed on the latex particles. In the other hand, the streaming potential measurement done on the membrane fouled with OM from supernatant/melamine particles aggregates leads to a global zeta potential of the fouling of about 1.84  0.30 mV. This value is closed to the zeta potential of the virgin membrane and consequently confirms that these aggregates are slightly charged. Thus in conclusion these results showed that if macromolecular compounds adsorb on particles the permeate flux is increased, inducing slight electrostatic charges inside the fouling. Moreover it is also shown that whatever the interaction between fouling species and a sub-micron particle, the organic matter rejection is improved by the presence of particles and the fouling becomes non compressible. As shown on SEM picture, done on a fouling layer made of latex/ OM aggregates (Fig. 5b), OM are embedded with the latex particles. Consequently here the non-compressibility of the fouling seems to be provided by the particles, working as a frame for the fouling. In addition the SEM picture (Fig. 5b) shows that latex/OM aggregates seems to be randomly packed at the membrane compare to a particles cake deposit (Fig. 2c).

4.

Conclusion

This study emphasised that the addition of sub-micron particles in suspension during biofluid filtration changes the fouling properties. Moreover results confirm the important role played by the macromolecular compounds during the filtration of an AS supernatant done at constant pressure at lab-scale. Those compounds, defined as the construction materials of biological flocs, lead to a highly compressible and reversible fouling. One of the major results of this study is that the fouling becomes non-compressible when particles are added in suspension with an MBR supernatant. Therefore beyond the different effects induced by the presence of particles, the present paper demonstrates that the formation of a noncompressible fouling could also be an important effect in the improvement of filtration performances. Indeed the formation of more consistent (i.e. rigid, structured) fouling layer could avoid the cake compressibility due to a possible fouling layer collapse. As a consequence the frequency of cleaning procedure would be reduced. In conclusion these results show that if the fouling layer structuring is controlled and taken into consideration the amount of chemical compounds added into an MBR could be strongly reduced. Add to these results show that whatever the particles properties their presence induces a better organic matter rejection compared to the one observed during the filtration of supernatant without particles. These observations have to be linked with MBR processes applied to wastewater treatment and more precisely when membrane-fouling reducers are added (PAC, polymers, etc.). Indeed as reported by various studies, different mechanisms could explain the positive effect of membrane fouling reducers (coagulation, adsorption, stronger sludge flocs.). It is demonstrated here that the addition in really low concentration of slightly adsorbent and smaller particles (compared to powder activated carbon commonly used in

membrane processes) could bring excellent filtration performances (non-compressible fouling layer and better OM rejection). Indeed results show that even if sub-micron particles exhibit only slight affinities regarding to fouling species (leading to slight adsorption or coagulation of those compounds), the filtration performances are still improved. This positive effect was attributed to the structuring of the fouling in presence of particles. However, as all filtration tests were performed with the supernatant of an AS at constant pressure in dead end mode, the impact of the biological activity or the hydrodynamic conditions on the cake architecture was not discussed. Thus, those results have been confirmed during a real MBR operation. Finally this study offers the opportunities to enlarge the choice of membrane fouling reducers by taking into consideration their ability to form more rigid fouling.

Acknowledgements This study was supported in part by the European Commission under the Sixth Framework Programme (EUROMBRA project).

references

Amy, G., 2008. Fundamental understanding of organic matter fouling of membranes. Desalination 231, 44e51. Bouhabila, E.H., Ben Aı¨m, R., Buisson, H., 2001. Fouling characterisation in membrane bioreactors. Separation and Purification Technology 22e23, 123e132. Chae, S., Wang, S., Hendren, Z.D., Wiesner, M.R., Watanabe, Y., Gunsch, C.K., 2009. Effects of fullerene nanoparticles on Escherichia coli K12 respiratory activity in aqueous suspension and potential use for membrane biofouling control. Journal of Membrane Science 329 (1e2), 68e74. Chang, S., Fane, A.G., Vigeswaran, S., 2002. Modeling and optimizing submerged hollow fiber membrane modules. Journal AIChE 48 (10), 2203e2212. Chellam, S., Xu, W., 2006. Blocking laws analysis of dead-end constant flux microfiltration of compressible cakes. Journal of Colloid and Interface Science 301 (1), 248e257. Cho, B.D., Fane, A.G., 2002. Fouling transients in nominally subcritical flux operation of a membrane bioreactor. Journal of Membrane Science 209 (2), 391e403. Drews, A., 2010. Membrane fouling in membrane bioreactorsCharacterisation, contradictions, cause and cures. Journal of Membrane Science 363 (1e2), 1e28. Fabrega, J., Renshaw, J.C., Lead, J.R., 2009. Interactions of silver nanoparticles with Pseudomonas putida biofilms. Environmental Science and Technology 43 (23), 9004e9009. Fang, H.H., Shi, X., Zhang, T., 2006. Effect of activated carbon on fouling of activated sludge filtration. Desalination 189 (1e3), 193e199. Flemming, H.C., Wingender, J., 2001. Relevance of microbial extracellular polymeric substances (EPSs). Part I. Structural and ecological aspects. Water Science and Technology 43 (6), 1e8. Foley, G., 2006. A review of factors affecting filter cake properties in dead-end microfiltration of microbial suspensions. Journal of Membrane Science 274 (1e2), 38e46. Galinha, C., Portugal, C., Carvalho, G., Gugliemi, G., Chiarani, D., Andreattalo, G., 2008. Modelling of membrane bioreactor for wastewater treatment incorporating 2D-fluorescence

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 6 0 e2 0 7 2

monitoring. In: Proceeding of membranes in drinking water production and wastewater treatment conference in France, Toulouse. Guo, W.S., Vigneswaran, S., Ngo, H.H., Van Nguyen, T.B., Ben Aim, R., 2006. Influence of bioreactor on a long term operation of a submerged membrane adsorption hybrid system. Desalination 191, 92e99. Hanft, S., 2006. Membrane Bioreactors in the Changing World Water Market. In: Business Communications Company Inc. Report C-240. Her, N., Amy, G., Park, H., Song, M., 2004. Characterizing algogenic organic matter (AOM) and evaluating associated NF membrane fouling. Water Research 38, 1427e1438. Hwang, B., Lee, W., Park, P., Lee, C., Chang, I., 2007. Effect of membrane fouling reducer on cake structure and membrane permeability in membrane bioreactor. Journal of Membrane Science 288 (1e2), 149e156. Hwang, B.K., Lee, W.N., Yeon, K.M., Park, P.K., Lee, C.H., Chang, I. S., Drews, A., Kraume, M., 2008. Correlating TMP increases with microbial characteristics in the bio-cake on the membrane surface in a membrane bioreactor. Environmental Science and Technology 42 (11), 3963e3968. Iversen, V., Mohaupt, J., Drews, A., Lesjean, B., Kraume, M., 2008. Side effects of flux enhancing chemicals in membrane bioreactors (MBRs): study on their biological toxicity and their residual fouling propensity. Water Science and Technology 57 (1), 117e123. Iversen, V., Mehrez, R., Horng, R., Chen, C., Meng, F., Drews, A., Lesjean, B., et al., 2009a. Fouling mitigation through flocculants and adsorbents addition in membrane bioreactors: comparing lab and pilot studies. Journal of Membrane Science 345 (1e2), 21e30. Iversen, V., Koseoglu, H., Yigit, N.O., Drews, A., Kitis, M., Lesjean, B., Kraume, M., 2009b. Impacts of membrane flux enhancers on activated sludge respiration and nutrient removal in MBRs. Water Research 43 (3), 822e830. Ji, J., Qiu, J., Wong, F.S., Li, Y., 2008. Enhancement of filterability in MBR achieved by improvement of supernatant and floc characteristics via filter aids addition. Water Research 42, 3611e3622. Jons, S., Ries, P., McDonald, C.J., 1999. Porous latex composite membranes: fabrication and properties. Journal of Membrane Science 155 (1), 79e99. Judd, S., 2008. The status of membrane bioreactor technology. Trends in Biotechnology 26 (2), 109e116. Kim, J.-S., Lee, C.-H., 2003. Effect of powdered activated carbon on the performance of an aerobic membrane bioreactor: comparison between cross flow and submerged membrane bioreactor (MBR) mixed liquor. Water Environment Research 75 (4), 300e307. Koseoglu, H., Yigit, N., Iversen, V., Drews, A., Kitis, M., Lesjean, B., Kraume, M., 2008. Effects of several different flux enhancing chemicals on filterability and fouling reduction of membrane bioreactor (MBR) mixed liquors. Journal of Membrane Science 320 (1e2), 57e64. Le-Clech, P., Chen, V., Fane, A.G., 2006. Fouling in membrane bioreactors used in wastewater treatment. Journal of Membrane Science 284 (1e2), 17e53. Lee, W.N., Chang, I.-S., Hwang, B.Y., Park, P.K., Lee, C.H., Huang, X., 2007. Changes in biofilm architecture with addition of membrane fouling reducer in a membrane bioreactor. Process Biochemistry 42 (4), 655e661. Lee, J.C., Kim, J.S., Kang, I.J., Cho, M.H., Park, P.K., Lee, C.H., 2001. Potential and limitations of alum or zeolite addition to improve the performance of a submerged previous membrane bioreactor. Water Science and Technology 43 (11), 59e66. Lesage, N., Spe´randio, M., Cabassud, C., 2005. Performances of a hybrid adsorption/submerged membrane biological process for toxic waste removal. Water Science and Technology 51 (6e7), 173e180.

2071

Lesjean, B., Huisjes, E., 2007. IWA 4th International Membrane Technologies Conference, 15e17 May 2007, Harrogate, UK. Lesjean, B., Huisjes, E., 2008. Survey of the European MBR market: trends and perspectives. Desalination 231 (1e3), 71e81. Lesjean, B., Rosenberger, S., Laabs, C., Jekel, M., Gnirss, R., Amy, G., 2005. Correlation between membrane fouling and soluble/ colloidal organic substances in membrane bioreactors for municipal wastewater treatment. Water Science and Technology 51 (6e7), 1e8. Li, Y., He, Y., Liu, Y., Yang, S., Zhang, G., 2005. Comparison of the filtration characteristics between biological powdered activated carbon sludge and activated sludge in submerged membrane bioreactors. Desalination 174 (3), 305e314. Masse´ A., 2004. Biore´acteur a` membranes immerge´es pour le traitement des eaux re´siduaires urbaines: spe´cificite´s physicochimiques du milieu biologique et colmatage. The`se Insa Toulouse. Non peer reviewed. Masse´, A., Spe´randio, M., Cabassud, C., 2006. Comparison of sludge characteristics and performance of a submerged membrane bioreactor and an activated sludge process at high solids retention time. Water Research 40 (12), 2405e2415. Meng, F., Chae, S.R., Drews, A., Kraume, M., Shin, H.-S., Yang, F., 2009. Recent advances in membrane bioreactors (MBRs): membrane fouling and membrane material. Water Research 43 (6), 1489e1512. Ng, C.A., Sun, D., Fane, A.G., 2008. Operation of membrane bioreactor with powdered activated carbon addition. Separation Science and Technology 41 (7), 1447e1466. Nystro¨m, M., Pihlajama¨ki, A., Ehsani, N., 1994. Characterization of ultrafiltration membranes by simultaneous streaming potential and flux measurements. Journal of Membrane Science 87 (3), 245e256. Okamura, D., Mori, Y., Hashimoto, T., Hori, K., 2009. Identification of biofoulant of membrane bioreactors in soluble microbial products. Water Research 43 (17), 4356e4362. Park, H., Choo, K.-H., Lee, C.-H., 1999. Flux enhancement with powdered activated carbon addition in the membrane anaerobic bioreactor. Separation Science and Technology 34 (14), 2781e2792. Pirbazari, M., Ravindran, V., Badriyha, B.N., Kim, S., 1996. Hybrid membrane filtration process for leachate treatment. Water Research 30 (11), 2691e2706. Pollet, S., Guigui, C., Cabassud, C., 2009. Influence of intermittent aeration and relaxation on a side-stream membrane bioreactor for municipal wastewater treatment. Desalination 6 (1e3), 108e118. Porter, M.C., 1990. Microfiltration. In: Porter, M.C. (Ed.), Handbook of Industrial Membrane Technology. Noyes Publications, Park Ridge, NJ, pp. 61e135. Remy, M., van der Marel, P., Zwijnenburg, A., Rulkens, W., Temmink, H., 2009. Low dose powdered activated carbon addition at high sludge retention times to reduce fouling in membrane bioreactors. Water Research 43 (2), 345e350. Remy, M., Potier, V., Temmink, H., Rulkens, W., 2010. Why low powdered activated carbon addition reduces membrane fouling in MBRs. Water Research 44 (3), 861e867. Rosenberger, S., Laabs, C., Lesjean, B., Gnirss, R., Amy, G., Jekel, M., Schrotter, J.C., 2006. Impact of colloidal and soluble organic material on membrane performance in membrane bioreactors for municipal wastewater treatment. Water Research 40 (4), 710e720. Sondi, I., Salopek-Sondi, B., 2004. Silver nanoparticles as antimicrobial agent: a case study on E. coli as a model for Gram-negative bacteria. Journal of Colloid and Interface Science 275 (1), 177e182. Strathmann, H., 1990. Synthetic membranes and their preparation. In: Porter, M.C. (Ed.), Handbook of Industrial Membrane Technology. Noyes Publications, Park Ridge, NJ, pp. 1e60.

2072

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 6 0 e2 0 7 2

Teychene, B., 2008. Roˆle des nanoparticules organiques dans le colmatage membranaire. Application au traitement d’eaux use´es urbaines par biore´acteur a` membranes. The`se Insa Toulouse. Non peer reviewed. Teychene, B., Guigui, C., Cabassud, C., Amy, G., 2008. Toward a better identification of foulant species in MBR processes. Desalination 231 (1e3), 27e34. Teychene, B., Loulergue, P., Guigui, C., Cabassud, C., 2010. Development and use of a novel method for in-line characterisation of fouling layers electrokinetics properties and for fouling monitoring. Journal of Membrane Science, submitted for publication. Von Smoluchowski, M., 1903. Bull. Int. Academic. Science, Cracovie, 184 pp. Wang, Z., Zhao, Y., Ye, N., Wang, J., Zhao, Z., Wang, S., 2006. Evaluation of four measurement operation modes of streaming potential for microfiltration and ultrafiltration membranes. Chinese Journal of Chemical Engineering 14 (4), 456e463. Yamamoto, K., Hiasa, M., Mahmood, T., Matsuo, T., 1989. Direct solid-liquid separation using hollow fiber membrane in an

activated sludge aeration tank. Water Science and Technology 21 (4e5), 43e54. Yamato, N., Kimura, K., Miyoshi, T., Watanabe, Y., 2006. Difference in membrane fouling in membrane bioreactors (MBRs) caused by membrane polymer materials. Journal of Membrane Science 280 (1e2), 911e919. Ying, Z., Ping, G., 2006. Effect of powered activated carbon dosage on retarding membrane fouling in MBR. Separation and Purification Technology 52 (1), 154e160. Yoon, S.H., Collins, J.H., Musale, D., Sundararajan, S., Tsai, S.P., Hallsby, G.A., Kong, J.F., Koppes, J., Cachia, P., 2005. Effects of flux enhancing polymer on the characteristics of sludge in membrane bioreactor process. Water Science and Technology 51 (6e7), 151e157. Zhang, J., Chua, H.C., Zhou, J., Fane, A.G., 2006. Factors affecting the membrane performance in submerged membrane bioreactors. Journal of Membrane Science 284 (1e2), 54e66. Zhang, H., Sun, B., Zhao, X., Gao, Z., 2008. Effect of ferric chloride on fouling in membrane bioreactor. Separation and Purification Technology 63 (2), 341e347.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 7 3 e2 0 8 1

Available at www.sciencedirect.com

journal homepage: www.elsevier.com/locate/watres

Life cycle assessment of vertical and horizontal flow constructed wetlands for wastewater treatment considering nitrogen and carbon greenhouse gas emissions Valerie J. Fuchs a,*, James R. Mihelcic b, John S. Gierke c a

Department of Chemical and Environmental Engineering, Yale University, New Haven, CT 06501, USA Department of Civil & Environmental Engineering, University of South Florida, Tampa, FL 33620, USA c Department of Geological and Mining Engineering and Science, Michigan Technological University, Houghton, MI 49931, USA b

article info

abstract

Article history:

Life cycle assessment (LCA) is used to compare the environmental impacts of vertical flow

Received 16 September 2010

constructed wetlands (VFCW) and horizontal flow constructed wetlands (HFCW). The LCAs

Received in revised form

include greenhouse gas (N2O, CO2 and CH4) emissions. Baseline constructed wetland

17 December 2010

designs are compared to different treatment performance scenarios and to conventional

Accepted 20 December 2010

wastewater treatment at the materials acquisition, assembly and operation life stages. The

Available online 4 January 2011

LCAs suggest that constructed wetlands have less environmental impact, in terms of resource consumption and greenhouse gas emissions. The VFCW is a less impactful

Keywords:

configuration for removing total nitrogen from domestic wastewater. Both wetland designs

Carbon dioxide

have negligible impacts on respiratory organics, radiation and ozone. Gaseous emissions,

Constructed wetlands

often not included in wastewater LCAs because of lack of data or lack of agreement on

Life cycle assessment

impacts, have the largest impact on climate change. Nitrous oxide accounts for the

Methane

increase in impact on respiratory inorganic, and the combined acidification/eutrophication

Nitrogen cycle

category. The LCAs were used to assess the importance of nitrogen removal and recycling,

Nitrous oxide

and the potential for optimizing nitrogen removal in constructed wetlands.

Vertical flow

ª 2011 Elsevier Ltd. All rights reserved.

Wastewater treatment

1.

Introduction and motivation

According to the National Academy of Engineering (NAE, 2008), one of the 14 Grand Challenges for Engineering for the 21st century is managing the nitrogen cycle. The NAE cites the needs to combat anthropogenic nitrogen fixation and subsequent water pollution, smog and acid rain, global warming, and associated environmental and human impacts, as the motivation for finding “countermeasures for nitrogen cycle problems”. The challenge for engineers is to improve the effectiveness of

human uses of nitrogen, including the biochemistry occurring within wastewater treatment plants. In accordance with the recommendations of NAE, a wastewater treatment system should be designed for the least life cycle environmental impacts, including nitrogen emissions, and with the most potential for denitrification. Life cycle comparisons of conventional wastewater treatment (e.g., activated sludge technology) to constructed wetlands have shown that constructed wetlands cause less environmental impacts. However, these studies have not included gaseous emissions in

* Corresponding author. Department of Chemical and Environmental Engineering, Yale University, 300A Mason Lab, 9 Hillhouse Avenue, New Haven CT 06520, USA. E-mail address: [email protected] (V.J. Fuchs). 0043-1354/$ e see front matter ª 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.watres.2010.12.021

2074

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 7 3 e2 0 8 1

their assessments nor have they specified constructed wetland designs that differ significantly in size or treatment capacity. LCA is used herein to assess the environmental impact differences in constructed wetland design by comparing a hypothetical horizontal flow system to vertical flow. Other findings from the LCA include how to account for wastewater treatment gaseous emissions; whether constructed wetlands emit less greenhouse gases than conventional treatment; which constructed wetland life stages have the greatest impact; and which design, operational and management variables could be adjusted to reduce environmental impacts. Furthermore, the analysis allows for considering potential tradeoffs in environmental impact between global and local scales. Although these results are comparable to other reported wastewater LCAs, this work quantifies the impacts specifically related to nitrogen emissions from constructed wetlands.

2.

Background

2.1.

Wetland greenhouse gas emissions

Vertical flow constructed wetlands (VFCW) are efficient at converting wastewater ammonium to nitrate and also effective in denitrification (Fuchs, 2009a). Because of the efficient oxygen transport of a VFCW, its physical footprint can be much smaller than a horizontal flow constructed wetland (HFCW) designed for the same effluent quality (Brix and Arias, 2005). Constructed wetlands have also been shown to produce less environmental impacts over the life cycle compared to conventional wastewater treatment plants (Dixon et al., 2003; Machado et al., 2007). Wetland researchers (listed in Table 1) have measured greenhouse gas emissions of CO2, CH4, and N2O from both VFCWs and HFCWs. The quantity and impact of all these gases are important because CH4 has 25 times the global warming potential of CO2 and N2O has 298 times the global warming potential of CO2 (IPCC, 2007). Table 1 lists reported effluent and

areal emissions pertaining to different influent parameters. The range of reported emissions is large (emissions are related to seasonal and other environmental conditions) and the areal emissions from VFCWs are generally higher. Sovik et al. (2006) found that VFCWs had significantly higher areal gaseous emissions than HFCWs, and gas emissions were correlated to temperature, substrate supply (influent N and C concentrations), and degree of oxidation in the wetland.

2.2.

Life cycle assessment

LCA is a comprehensive and transparent tool for estimating potential environmental impacts of products and processes (UNEP, 2009). LCA can be used in the design phase for choosing between technologies with similar performance by accounting for the impacts caused by each technology over its life cycle. For example, researchers have compared conventional activated sludge wastewater treatment to constructed wetlands using LCA. Dixon et al. (2003) found that constructed wetlands have less global warming potential (CO2 emissions) and less energy use than conventional treatment. Machado et al. (2007) found that wetlands also reduced aquatic toxicity and eutrophication compared to conventional activated sludge wastewater treatment. LCA can also be used to identify which life stage carries the most significant environmental impacts for particular designs or classes of designs. LCA on wastewater treatment indicates that the life cycle impacts of the operation phase are much greater than the construction phase for conventional activated sludge treatment systems (Tillman et al., 1998; Lundin et al., 2000; Dixon et al., 2003; Emmerson et al., 1995; Ortiz et al., 2007; Renou et al., 2008). For constructed wetlands, however, the construction phase dominates the life cycle impacts due to the amount of materials transported for construction as well as the reduction of energy use during operation (Dixon et al., 2003; Machado et al., 2007). Only two LCAs considered nitrogen emissions in the wastewater treatment life cycle, but the emissions were only counted from fuel use, not from wastewater nitrogen transformation

Table 1 e Influent and effluent water quality and gaseous emissions of N2O, CO2, and CH4 reported for vertical flow and horizontal flow wetlands. Influent (mg/L)

Vertical flow constructed wetlands Zhou et al., 2008 Inamori et al., 2007 Inamori et al., 2007 Inamori et al., 2007 Sovik et al., 2006 (Koo VFCW) Horizontal flow constructed wetlands Maltais-Landry et al., 2009 Fey et al., 1999 Sovik et al., 2006 (Kodijarve wetland) Sovik et al., 2006 (Koo wetland)

TNeN

NHþ 4 eN

9.8 18.4 36.7 50.9

8.1 10.2 19.3 35.7

21.7 96.5 43.1

BOD

38 60 163 142

0.18 83.9 31.7

Emissions (mg/m2/d)

Effluent (mg/L)

500 125 62.8

NHþ 4 eN

NO 3 eN

N2O

31.7

1.7

1.4 to 188 <0.24 <0.48 <1.44 10.8a

0.2 (g/m2d)

0.05 (g/m2d)

36.2 34

5.9 1.2

Yearly average ¼ [summer average (#samples) þ winter average (#samples)]/total #samples a Yearly average calculated from reported summer and winter data.

3 3.2 4.5a 2.9a

CO2

CH4

5200a

<72 <240 <480 77.4a

1400

5

2490a 1301a

182a 96.5a

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 7 3 e2 0 8 1

(Lundin and Morrison, 2002; Hospido et al., 2004). The production of nitrogen and carbon emissions from wastewater microbial processes has typically been excluded from LCAs because of lack of data. The Intergovernmental Panel on Climate Change (IPCC) reported estimates between 16 and 96 mg/m2/d N2O from municipal wastewater treatment (IPCC, 2001). Czepiel et al. (1995) measured N2O emissions from wastewater treatment plants and found 20e1800 mg/m2/d for aerated processes and 10e40.8 mg/ m2/d for processes without aeration. Sumer et al. (1995) found a range of 0e77 mg/m2/d N2O from activated sludge operations, while the range of measurements of Benckiser et al. (1996) were much larger, from 53 to 4903 mg/m2/d. In general, all the emissions rates reported for conventional processes are significantly higher than emissions from either vertical flow or horizontal flow constructed wetlands (Table 1).

3.

Methods

The goals of the LCA are to determine which constructed wetland flow regime (horizontal or vertical) has the least environmental impact over its life cycle and whether there may be tradeoffs in impacts (on air versus water quality, for example). The scope considers primary treatment with a septic system and secondary treatment by a wetland including land use, soil, vegetation, liner, and wastewater collection and conveyance systems. The scope of the life cycle assessment is shown in Fig. 1. The functional unit is treatment of 400 person-equivalents (p.e.) of wastewater for a system lifetime of 50 years, with an effluent discharge requirement less than 5 mg/L NHþ 4 eN. The life cycle includes material assembly, construction and septic tank and wetland operation (assuming that the wetlands would be decommissioned but remain in place at end of life). One p.e. is assumed to produce 150 L/day wastewater containing 60 g BOD5, 13 g N and 2.5 g P (Brix and Arias, 2005). The primary design assumptions are that: 1) all influent N to the wetland is in the form of NHþ 4 and 2) there are no safety factors included in the designs.

Fig. 1 e Life cycle scope for VFCW and HFCW.

3.1.

2075

Treatment system design

A community septic tank is included in the design as a pretreatment requirement for the wetlands in order to remove solids (by settling) and allow for conversion of organic nitrogen to ammonium. The septic tank design follows guidelines of Crites and Tchobanoglous (1998) for a steel-reinforced concrete tank with a volume of 328.5 m3, meeting the recommended size of about five times the average flow of 60,000 L/d. The septic tank has 2 longitudinal cells of width (w) 4.1 m, length (l ) ¼ 8 m, and depth (d ) ¼ 4.1 m. The maintenance requirement is pumping out septage every five years, or 10 times over the 50year life cycle of the whole treatment system. Sludge disposal is outside the scope of this LCA. The septic tank is followed by a pump to transfer wastewater to the wetland. The horizontal flow wetland was sized using the Kadlec and Knight (1996) model for constructed wetlands as described in Fuchs (2009b). The area required to treat 60,000 L/ 2 d of wastewater to <5 mg/L NHþ 4 eN is 5049 m (single cell with w ¼ 100.5 m, l ¼ 50.2 m, and d ¼ 1.3 m). The main filter media is coarse gravel with a porosity of 0.38, d50 of 64 mm, and hydraulic conductivity of 10,000 m/d. The vertical flow wetland was sized using the Danish guidelines for vertical flow constructed wetlands (Brix and Arias, 2005). The guidelines offer a rule-of-thumb areal requirement proportional to the population served (3.2 m2/p.e.) along with depth and distribution/collection guidelines, which is currently the best vertical flow constructed wetland design process available. The vertical flow wetland is 1280 m2 (two cells with w ¼ 20 m, l ¼ 32 m, and d ¼ 1.4 m), which treats 60,000 L/d to NHþ 4 eN <5 mg/L. The main filter media is coarse sand with d50 of 1 mm. See Fuchs (2009a) for specific design calculations for the wastewater treatment systems (septic tank, HFCW, and VFCW).

3.2.

LCA inventory and impact assessment

For inventory analysis, a number of material and process databases have been compiled (based on data for Europe, the US, and the whole world for different databases) and are incorporated into tools that categorize and aggregate the impacts. The LCA software Simapro (Pre´, 2008) contains a range of databases and impact assessment options. The inventory for materials and processes for the life cycle of the constructed wetlands are provided in Fuchs (2009a). Inventory data was arranged into assembly (construction phase) and life cycle (use phase) for each wetland. In this LCA, gaseous emissions from the wastewater treatment processes are also included. Because there is such a large range of values reported for N2O, CO2 and CH4 emissions from constructed wetlands (Table 1), the values from a single comprehensive study are used for this analysis for consistency. Sovik et al. (2006) reported influent BOD and nitrogen; effluent BOD, ammonium and nitrate; and gaseous emissions of N2O, CO2 and CH4 for both vertical and horizontal flow wetlands. The fraction of influent total nitrogen emitted as N2O and the fraction of influent BOD emitted as CO2 and CH4 from Sovik et al. (2006) are used as emission factors for this LCA, and they are converted to influent TN and BOD for 1 p.e. as defined by Brix and Arias (2005). Gaseous emissions are reported in

2076

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 7 3 e2 0 8 1

Table 2 e Influent and effluent water quality and gaseous emissions (mg/L equivalents) from Sovik et al. (2006) used in this study. Influent (mg/L)

VFCW Simapro input HFCW Simapro input

TNeN

NHþ 4 eN

BOD

86.6 86.6

86.6 86.6

400.0 400.0

Effluent (mg/L)

Emissions (mg/L)

P

NHþ 4 eN

NO 3 eN

BOD

P

N2O

CO2

CH4

16.6 16.6

5.0 5.0

35.0 35.0

1.0 1.0

4.0 4.0

0.4 0.5

312.5 679.3

4.6 49.9

mg/m2/d but are entered in the life cycle assessment in mg/L and thus are proportional to the wetland surface area and daily flow rate. Table 2 lists the aqueous and gaseous emissions used to compare the VFCW and HFCW. The assessment was conducted using Simapro 7.0. Two assessment methods were used: Ecoindicator 99 and CML 2 Baseline 2000. Both methods have been used in other wastewater treatment life cycle assessments (Hospido et al., 2004; Machado et al., 2007; Lassaux et al., 2007; Ortiz et al., 2007; Hospido et al., 2008). Renou et al. (2008) showed that the overall difference between assessment methods in Simapro is small, so one may choose any impact assessment method based on the impact/damage categories it considers or how the impact/damages are calculated. The Ecoindicator method calculates life cycle impacts in eleven impact categories and also calculates endpoint damages to human health, ecosystem quality and resources. The Ecoindicator 99 category for acidification and eutrophication, however, only considers emissions to air but not water. Because wastewater treatment can significantly impact eutrophication potential through emissions to water, the CML 2 Baseline 2000 impact assessment was used to calculate eutrophication potential for various wetland treatment scenarios. Ecoindicator 99 calculates an overall indicator for endpoint damages to human health, ecosystem quality and resources. Human health is affected by impacts from carcinogens, respiratory organics and inorganics, and climate change. Ecosystem quality is impacted by radiation, ozone layer destruction, ecotoxicity, acidification/eutrophication and land use. Loss of resources is due to the use of minerals and fossil fuels. The impacts and damages are weighted according to an egalitarian (E ) approach where long-term ecosystem quality

is viewed as equally beneficial to human health, a mid-term hierarchical (H ) scheme where human health is somewhat more important than the environment, or a short-term individualist (I ) viewpoint where human health is of primary concern over ecosystem quality. The hierarchical weighting scheme is applied for this LCA.

4.

Results

The impact assessments using Ecoindicator 99 (H ) for the VFCW and HFCW, both considering and excluding gaseous emissions, are shown with impact point values listed in Fig. 2. The life cycle impacts are negligible for respiratory organics, radiation and ozone layer and therefore not shown. In all other impact categories, the VFCW impacts are significantly less than the HFCW impacts. N2O, CO2, and CH4 influence the respiratory inorganics, climate change, and acidification/ eutrophication categories with the largest influence being in climate change. The influence of each of the three gaseous emissions is demonstrated in Fig. 3a for the VFCW and Fig. 3b for the HFCW, which show the impacts of each individual emission beyond the baseline of not considering gaseous emissions. For example, for the climate change impact category, the baseline impact for a VFCW is 196 damage points, but 572 damage points with the consideration of just CH4, 772 considering the addition of just N2O, and 1599 considering the baseline plus CO2. For a VFCW, the baseline climate change impact plus consideration of all three gaseous emissions during operation is 2631 (as shown in Fig. 2), an order of magnitude larger than the baseline impact without consideration of gaseous emissions. For respiratory inorganics and

Fig. 2 e Environmental impacts assessed using Ecoindicator 99 (H ) for VFCW and HFCW with and without gaseous emissions. Carc. [ Carcinogen; R.I. [ Respiratory Inorganics; C.C. [ Climate Change; Ecotox. [ Ecotoxicity; A./E. [ Acidification/Eutrophication; F.F. [ Fossil Fuels.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 7 3 e2 0 8 1

Fig. 3 e Environmental impacts of different gaseous emissions in (a) VFCW and (b) HFCW (Ecoindicator 99-H ). Horizontal bars for CH4, N2O and CO2 are for the additional impact of the individual gas. Carc. [ Carcinogen; R.I. [ Respiratory Inorganics; C.C. [ Climate Change; Ecotox. [ Ecotoxicity; A./E. [ Acidification/Eutrophication; F.F. [ Fossil Fuels.

acidification/eutrophication, the increase in impact from the baseline is due to N2O, while CO2 and CH4 only influence climate change. CO2 is the greatest climate change factor for VFCWs, while CH4 has the larger climate change impact for HFCWs due to greater global warming potential (gaseous emissions listed in Table 2 and impacts shown in Fig. 3). LCA was also used to test the impacts of various treatment performance levels for the VFCW and HFCW on the eutrophication impact category using the CML 2 Baseline 2000 impact assessment. Gaseous emissions from wetland treatment processes account for only 0.3% or 0.4% of eutrophication impacts for a VFCW and HFCW, respectively. Major eutrophication impacts are phosphorus (75% of impact) and nitrate (21%) emissions to water. The baseline LCA considered that the wetlands meet the following treatment levels: BOD ¼ 1 mg/L, NHþ 4 eN ¼ 5 mg/L, and P ¼ 4 mg/L. The scenarios for eutrophication were no phosphorus treatment (high P ¼ 16.6 mg/L effluent phosphorus), low nitrification (effluent  NHþ 4 eN ¼ 25 mg/L and NO3 eN ¼ 15 mg/L), poor treatment of biochemical oxygen demand (high BOD ¼ 50 mg/L effluent BOD), and complete nitrification and denitrification (effluent N ¼ 0 mg/L). Though unrealistic for normal operation, these scenarios were used to explore life cycle impacts in various cases of treatment, failure or overload. Since the two wetlands were designed to meet the same standards, there is no

2077

difference between the VFCW and HFCW for each scenario; the test showed how treatment scenarios changed the baseline eutrophication. Eutrophication is most impacted by phosphorus, for which complete lack of treatment would increase the eutrophication potential 200% from the baseline LCA. Poor treatment of BOD or N increases eutrophication by only 5 and 10%, respectively. Potential for total N removal through denitrification would reduce the eutrophication impact by about 30% (Fig. 4 ). Endpoint damages to human health, ecosystem quality and resources are also calculated using Ecoindicator 99 (H ). Human health damage points are normalized Disability Adjusted Life Years (DALYs) caused by carcinogens, respiratory inorganics, respiratory organics, climate change, radiation, and ozone layer reduction. Ecosystem quality damage points are normalized by the Potentially Disappeared Fraction (PDF) of species per unit of land area per year for ecotoxicity, acidification/eutrophication and land use (which includes the consideration of constructed wetland as habitat). Resource damage points are normalized surplus of minerals and fossil fuels measured in megajoules (MJ). As shown in Fig. 5, the VFCW has significantly less impact to human health (40% of HFCW), ecosystem (50% of HFCW) and resources (50% of HFCW). The importance of including gaseous emissions in the LCA is demonstrated as they are nearly half of the human health impacts for VFCW and more than half for HFCW. The impact of gaseous emissions on ecosystem quality is small and is due to N2O factored into acidification/eutrophication. The life cycles of VFCW and HFCW can also be parsed into life stage contributions, as shown in Fig. 6. Over the entire life cycle, VFCWs have much less environmental impact, due to requiring less materials and construction equipment in assembly as well as lower gaseous emissions during the use phase (wastewater treatment). For VFCWs, other research has shown that the life stage with the greatest overall impact is the assembly or construction phase (Dixon et al., 2003), which here have about the same level of impact as the use phase. The construction impacts could be significantly reduced by using local materials to avoid transporting wetland media. The availability of sand for VFCW filter media or gravel for HFCW filter media may be site dependent. Fuchs (2009a) showed that a nitrifying VFCW wetland could be much shallower than guidelines call for, so the sand volume could be reduced by potentially 60%.

Fig. 4 e Eutrophication impacts of various scenarios on baseline VFCW and HFCW life cycles (CML 2 baseline 2000).

2078

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 7 3 e2 0 8 1

Fig. 5 e Endpoint damages of VFCW and HFCW with and without gaseous emissions assessed with Ecoindicator 99-H.

5.

Discussion

The life cycle assessment was conducted to answer the hypothesis that a vertical flow constructed wetland will have less environmental impact than a horizontal flow constructed wetland due to its smaller footprint and better nitrogen removal. Despite the fact that nitrogen and carbon emissions are higher per unit area for VFCW than for HFCW (Table 1), the overall impacts are higher for HFCW because the wetland must be larger for equivalent levels of treatment. For both midpoint and endpoint damages, VFCW impacts are half or less of HFCW impacts, depending on the impact category (Figs. 2 and 5). The life cycle assessment shows that VFCW have less environmental impacts through the life cycle than HFCW due to better treatment efficiency (nitrogen removal in a smaller wetland) and because it requires less materials and construction equipment and produces less greenhouse gas emissions. The results of this LCA are significant for designing community wastewater treatment systems. Beyond other reported LCAs, which show that constructed wetlands have more environmental benefit than conventional wastewater treatment technologies, this LCA demonstrates that wetland hydraulic regime choice is important for reducing environmental impacts. Not only VFCWs are more efficient in land use and nitrification, but they also cause considerably less environmental impact over

the entire life cycle than HFCW designed to meet the same effluent standards. This LCA also provides new information because it shows the necessity of including gaseous emissions from the wastewater treatment process. Figs. 2 and 5 demonstrate the difference in LCA with and without gaseous emissions, particularly pertinent to the impact categories of respiratory inorganics, climate change and eutrophication, and the endpoint categories human health and ecosystem quality. While the emissions included in this LCA were based on limited data, and LCAs have typically excluded gaseous emissions due to lack of agreement in available data; this LCA shows that even mid-range measurements have a significant impact in the constructed wetland life cycle. N2O emission rates for wetlands are much smaller than for conventional treatment. However, conventional systems concentrate the treatment in a smaller area, so it is not clear if constructed wetlands truly provide a benefit over conventional treatment in terms of gaseous emission reduction. HFCW N2O emissions have been reported in the range of 1.3e13 g/yr per capita, while a VFCW had per capita emissions of 2.3 g/yr (Sovik et al., 2006). In comparison, the Water Environment Research Foundation reports that septic tanks emit 1.8 g/yr per capita (WERF, 2010). For conventional systems, Czepiel et al. (1995) measured 3.2 g/yr per capita and Kimochi et al. (1998) found a range of 0e1.9 g/yr per capita from

Fig. 6 e Breakdown of impacts over life stages of HFCW (top) and VFCW (bottom). (Impact categories with less than 1% of the total impact were removed: respiratory organics, radiation, ozone layer and minerals). While mowing and coal for electricity for pumping are shown as separate activities from septage transport and wastewater treatment (WWT) in the use phase, their effects are negligible. Carc. [ Carcinogen; R.I. [ Respiratory Inorganics; C.C. [ Climate Change; Ecotox. [ Ecotoxicity; A./E. [ Acidification/Eutrophication; F.F. [ Fossil Fuels.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 7 3 e2 0 8 1

activated sludge treatment plants. Recently, Ahn et al. (2010) reported a large range of N2O emission factors from 0.28 to 140 g/yr per capita for activated sludge plants, depending on the biological nutrient removal process. Conventional treatment may reduce gaseous emissions per unit of treatment volume due to controlled optimal conditions for nitrification and denitrification, but other researchers (Dixon et al., 2003; Machado et al., 2007) have shown that over the life cycle, constructed wetlands are environmentally superior to an extent that may outweigh the difference in gas emissions between conventional technologies and wetlands. Several design recommendations grow out of these results. As shown in Fig. 3, gaseous emissions make up the majority of climate change impacts for a constructed wetland life cycle. Nitrous oxide and other nitrogen oxides are formed during nitrification and denitrification processes at suboptimal conditions (low oxygen during nitrification, high oxygen or low C/N ratio during denitrification) and are directly related to temperature and influent nitrogen concentration. The formation of nitrogen oxides can be avoided with a high BOD/N ratio (Osada et al., 1995; Young Park et al., 2000; Tallec et al., 2006), low O2/NOx ratio for denitrification (Osada et al., 1995), long denitrification detention time, and avoiding simultaneous ammonium oxidation and nitrite reduction (not the same as anaerobic ammonium oxidation where ammonium and nitrite are converted directly to nitrogen gas) (Gjelsbjerg et al., 1998). Optimizing the design for oxygen transfer, nitrification and denitrification processes will reduce nitrous oxide emissions, therefore reducing impacts from respiratory organics, climate change and acidification/eutrophication. A VFCW may be optimized for nitrification by using a downward flow regime and for denitrification by using an upflow saturated regime with available carbon (Fuchs, 2009a). Foley et al. (2007) suggested that there may be global versus local tradeoffs in design and operational decisions for wastewater treatment. The locally valued impact categories considered here include eutrophication and land use, while global impact categories are climate change and fossil fuel consumption (other impact categories either have a very small impact in this study and are excluded or are not specifically globally or locally valuable). In this case, there is no tradeoff between global and local impacts: VFCWs have less impact for all categories than HFCWs. Optimizing the design for nitrification and denitrification would improve both global and local impact categories by reducing nitrous oxide, ammonium and nitrate emissions. Though this study shows that constructed wetlands can be environmentally superior to conventional treatment, it does not address the broader issues of needing to recover water, energy and nutrients. In this LCA, the scope included only the wastewater treatment. If the system boundaries are extended to include water treatment or distribution, it is pertinent to consider closing the water resource loop through reuse, recycle, or even keeping water and waste separate. Water reuse and recycle could be feasible with constructed wetland effluent, particularly from an optimized design that removes a high level of nutrients. Separating water and waste would make a constructed wetland infeasible since it requires water to move the waste to it and distribute waste over/through it.

2079

Nutrient recovery is another loop to close, which requires the consideration of agriculture in LCA. Researchers are beginning to consider source separation of urine and feces, which could tie directly into water recovery and reuse technologies, energy recovery, and returning nitrogen and phosphorus to agriculture without toxic contaminants (Maurer et al., 2003; Remy and Jekel, 2008; Larsen et al., 2009). This is one reason that recovery of water, energy, and nutrients is a key consideration in discussion of what make a particular wastewater treatment technology sustainable (Guest et al., 2009). LCA of constructed wetlands opens up policy questions, such as where wastewater decentralization may be appropriate, how to manage decentralized systems, and how to facilitate technology transfer and adoption. Muga and Mihelcic (2008) determined that decentralized technologies, such as land-based treatment, may be more sustainable than mechanical treatment systems based on a set of sustainability indicators. The Water Environment Research Foundation (WERF) is beginning to examine how to transfer decentralized technologies and overcome barriers to technology adoption (Etnier et al., 2007). WERF recommends increasing financial incentives associated with decentralized technology, enhancing knowledge of decentralized systems, improving favorability of regulators toward decentralization, and increasing systems thinking. These improvements, along with research into management of decentralized systems, may enable the resource conservation discussed above by minimizing the contamination of resources.

6.

Conclusions

VFCWs are an efficient and low-energy technology for wastewater nitrification and have excellent potential for denitrification. They require significantly less land area than HFCWs and achieve water quality standards at much lower environmental impact. The consideration of resource conservation and reduction of environmental impacts is becoming a priority in engineering design. Wastewater treatment technology and management needs to consider water, energy, and nutrients as resources to recycle rather than wastes to separate. Constructed wetlands may be an appropriate solution for resource recovery and reducing environmental impacts.

Acknowledgments This material is based upon work supported by the National Science Foundation under Grant Nos. (DGE 0333401 and 0806569) and their Graduate Research Fellowship Program, the Water Environment Research Foundation (Project DEC11U06), the DeVlieg Foundation, and the Michigan Water Environment Association, and technical assistance from Matt Seib.

references

Ahn, J.H., Kim, S., Park, H., Rahm, B., Pagilla, K., Chandran, K., 2010. N2O emissions from activated sludge processes, 2008e2009: results of a national monitoring survey in the

2080

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 7 3 e2 0 8 1

United States. Environmental Science & Technology 44, 4505e4511. Benckiser, G., Eilt, E., Linn, A., Lorch, H.J., Sumer, E., Weiske, A., Wenzhofer, F., 1996. N2O emissions from different cropping systems and from aerated, nitrifying and denitrifying tanks of a municipal wastewater treatment plant. Biology & Fertility of Soils 23, 257e265. Brix, H., Arias, C.A., 2005. The use of vertical flow constructed wetlands for on-site treatment of domestic wastewater: new Danish guidelines. Ecological Engineering 25, 491e500. Crites, R., Tchobanoglous, G., 1998. Small and Decentralized Wastewater Management Systems. McGraw-Hill. Czepiel, P., Crill, P., Harris, R., 1995. Nitrous oxide emissions from municipal wastewater treatment. Environmental Science & Technology 29 (9), 2352e2356. Dixon, A., Simon, M., Burkitt, T., 2003. Assessing the environmental impact of two options for small-scale wastewater treatment: comparing a reedbed and an aerated biological filter using a life cycle approach. Ecological Engineering 20, 297e308. Emmerson, R.H.C., Morse, G.K., Lester, J.N., Edge, D.R., 1995. The life-cycle analysis of small-scale sewage-treatment processes. Journal of the Chartered Institution of Water and Environmental Management 9, 317e325. Etnier, C., Pinkham, R., Crites, R., Johnstone, D.S., Clark, M., Macrellis, A., 2007. Research Digest: Promoting Equitable Consideration of Decentralized Wastewater Options. WERF Final Report, Project 04-DEC-2a. Water Environment Research Federation/International Water Association Publishing, Alexandria, VA/London, UK. Fey, A., Benckiser, G., Ottow, J.C.G., 1999. Emissions of nitrous oxide from a constructed wetland using a groundfilter and macrophytes in waste-water purification of a dairy farm. Biology & Fertility of Soils 29, 354e359. Foley, J., de Haas, D., Hartley, K., Lant, P., 2007. Life cycle assessment of biological nutrient removal wastewater treatment plants. In: Proceedings of the 3rd International Conference on Life Cycle Management, Zurich, Switzerland, Aug. 27e29. Fuchs, V.J. (2009a). Nitrogen Removal and Sustainability in Vertical Flow Constructed Wetlands for Small Scale Wastewater Treatment. PhD dissertation, Michigan Technological University, Houghton, MI. Fuchs, V.J., 2009b. Constructed wetlands and evapotranspiration beds. In: Mihelcic, J.R., Fry, L.M., Myre, E.A., Phillips, L.D., Barkdoll, B.D. (Eds.), Guide in Environmental Engineering for Development Workers: Water Supply, Sanitation Systems, and Indoor Air. American Society of Civil Engineers (ASCE) Press, Reston, Virginia, p. 550 (Chapter 23 in Field). Gjelsbjerg, B., Frette, L., Westermann, P., 1998. Dynamics of N2O production from activated sludge. Water Research 32 (7), 2113e2121. Guest, J.S., Skerlos, S.J., Barnard, J.L., Beck, M.B., Daigger, G.T., Hilger, H., Jackson, S.J., Karvazy, K., Kelly, L., Macpherson, L., Mihelcic, J.R., Pramanik, A., Raskin, L., van Loosdrecht, M.C.M., Yeh, D., Love, N.G., 2009. A new planning and design paradigm to achieve sustainable resource recovery from wastewater. Environmental Science & Technology 43 (16), 6126e6130. Hospido, A., Moreira, M.T., Fernandez-Couto, M., Feijoo, G., 2004. Environmental performance of a municipal wastewater treatment plant. International Journal of Life Cycle Assessment 9 (4), 261e271. Hospido, A., Moreira, M.T., Feijoo, G., 2008. A comparison of municipal wastewater treatment plants for big centres of population in Galicia (Spain). International Journal of Life Cycle Assessment 13 (1), 57e64. IPCC, 2001. CH4 and N2O emissions from waste water handling. In: Background Paper for Good Practice Guidance and Uncertainty Management in National Greenhouse Gas

Inventories. Intergovernmental Panel on Climate Change Taskforce on National Greenhouse Gas Inventories. IPCC, 2007. In: Pachauri, R.K., Reisinger, A. (Eds.), Intergovernmental Panel on Climate Change Fourth Assessment Report (AR4). Intergovernmental Panel on Climate Change, Geneva, Switzerland, p. 104. Inamori, R., Gui, P., Dass, P., Matsumura, M., Xu, K.Q., Kondo, T., Ebie, Y., Inamori, Y., 2007. Investigating CH4 and N2O emissions from eco-engineering wastewater treatment processes using constructed wetland microcosms. Process Biochemistry 42, 363e373. Kadlec, R., Knight, R., 1996. Treatment Wetlands. CRC Press, Boca Rato, Florida, 928 pp. Kimochi, Y., Inamori, Y., Mizuochi, M., Xu, K.Q., Matsumura, M., 1998. Nitrogen removal and N2O emission in a full-scale domestic wastewater treatment plant with intermittent aeration. Journal of Fermentation and Bioengineering 86 (2), 202e206. Larsen, T.A., Alder, A.C., Eggen, R.I., Maurer, M., Leinert, J., 2009. Source separation: will we see a paradigm shift in wastewater handling? Environmental Science & Technology 43 (16), 6121e6125. Lassaux, S., Renzoni, R., Germain, A., 2007. Life cycle assessment of water from the pumping station to the wastewater treatment plant. International Journal of Life Cycle Assessment 12 (2), 118e126. Lundin, M., Morrison, G.M., 2002. A life cycle assessment based procedure for development of environmental sustainability indicators for urban water systems. Urban Water 4, 145e152. Lundin, M., Bengtsson, M., Molander, S., 2000. Life cycle assessment of wastewater systems: influence of system boundaries and scale on calculated environmental loads. Environmental Science & Technology 34 (1), 180e186. Machado, A.P., Urbano, L., Brito, A.G., Janknecht, P., Salas, J.J., Nogueira, R., 2007. Life cycle assessment of wastewater treatment options for small and decentralized communities: energy-saving systems versus activated sludge. Water Science and Technology 56 (3), 15e22. Maltais-Landry, G., Maranger, R., Brisson, J., 2009. Effect of artificial aeration and macrophyte species on nitrogen cycling and gas flux in constructed wetlands. Ecological Engineering 35, 221e229. Maurer, M., Schwegler, P., Larsen, T.A., 2003. Nutrients in urine: energetic aspects of removal and recovery. Water Science and Technology 48 (1), 37e46. Muga, H.E., Mihelcic, J.R., 2008. Sustainability of wastewater treatment technologies. Journal of Environmental Management 88, 437e447. NAE, 2008. Grand Challenges for Engineering. http://www. engineeringchallenges.org/ (accessed 10.11.09.). Ortiz, M., Raluy, R.G., Serra, L., Uche, J., 2007. Life cycle assessment of water treatment technologies: wastewater and water-reuse in a small town. Desalination 204, 121e131. Osada, T., Kuroda, K., Yonaga, M., 1995. Reducing nitrous oxide gas emissions from fill-and-draw type activated sludge process. Water Research 29 (6), 1607e1608. Pre´, 2008. Simapro 7.1.9.16, Multi-user, Beta Version. Pre´ Product Ecology Consultants, the Netherlands. Remy, C., Jekel, M., 2008. Sustainable wastewater management: life cycle assessment of conventional and source-separating urban sanitation systems. Water Science and Technology 58 (8), 1555e1562. Renou, S., Thomas, J.S., Aoustin, E., Pons, M.N., 2008. Influence of impact assessment methods in wastewater treatment LCA. Journal of Cleaner Production 16, 1098e1105. Sovik, A.K., Augustin, J., Heikkinen, K., Huttunen, J.K., Necki, J.M., Karjalainen, S.M., Klove, B., Liikanen, A., Mander, U., Puustinen, M., Teiter, S., Wachniew, P., 2006. Emission of the greenhouse gases

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 7 3 e2 0 8 1

nitrous oxide and methane from constructed wetlands in Europe. Journal of Environmental Quality 35, 2360e2373. Sumer, E., Weiske, A., Benckiser, G., Ottow, J.C.G., 1995. Influence of environmental conditions on the amount of N2O released from activated sludge in a domestic wastewater treatment plant. Experientia 51, 419e422. Tallec, G., Garnier, J., Gousailles, M., 2006. Nitrogen removal in a wastewater treatment plant through biofilters: nitrous oxide emissions during nitrification and denitrification. Bioprocess and Biosystems Engineering 29, 323e333. Tillman, A.M., Svingby, M., Lundstrom, H., 1998. Life cycle assessment of municipal wastewater systems. International Journal of Life Cycle Assessment 3 (3), 145e157.

2081

UNEP, 2009. The Life Cycle Initiative. http://lcinitiative.unep.fr/ (last accessed 20.10.09.). WERF, 2010. Evaluation of Greenhouse Gas Emissions from Septic Systems; Final Report from Project DEC1R09. Water Environment Research Foundation, Alexandria, VA. Young Park, K., Inamori, Y., Mizuochi, M., Hong Ahn, K., 2000. Emission and control of nitrous oxide from a biological wastewater treatment system with intermittent aeration. Journal of Bioscience and Bioengineering 90 (3), 247e252. Zhou, S., Hou, H., Hosomi, M., 2008. Nitrogen removal, N2O emission and NH3 volatilization under different water levels in a vertical flow treatment system. Water, Air, and Soil Pollution 191, 171e182.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 8 2 e2 0 9 4

Available at www.sciencedirect.com

journal homepage: www.elsevier.com/locate/watres

CFD investigation of turbulence models for mechanical agitation of non-Newtonian fluids in anaerobic digesters Binxin Wu Philadelphia Mixing Solutions Ltd., Palmyra, PA 17078, USA

article info

abstract

Article history:

This study evaluates six turbulence models for mechanical agitation of non-Newtonian

Received 19 July 2010

fluids in a lab-scale anaerobic digestion tank with a pitched blade turbine (PBT) impeller.

Received in revised form

The models studied are: (1) the standard ke3 model, (2) the RNG ke3 model, (3) the real-

15 December 2010

izable ke3 model, (4) the standard keu model, (5) the SST keu model, and (6) the Reynolds

Accepted 20 December 2010

stress model. Through comparing power and flow numbers for the PBT impeller obtained

Available online 24 December 2010

from computational fluid dynamics (CFD) with those from the lab specifications, the realizable ke3 and the standard keu models are found to be more appropriate than the

Keywords:

other turbulence models. An alternative method to calculate the Reynolds number for

Anaerobic digester

the moving zone that characterizes the impeller rotation is proposed to judge the flow

Computational fluid dynamics

regime. To check the effect of the model setup on the predictive accuracy, both dis-

Turbulence model

cretization scheme and numerical approach are investigated. The model validation is

Non-Newtonian fluid

conducted by comparing the simulated velocities with experimental data in a lab-scale

Mechanical mixing

digester from literature. Moreover, CFD simulation of mixing in a full-scale digester with two side-entry impellers is performed to optimize the installation. ª 2010 Elsevier Ltd. All rights reserved.

1.

Introduction

Anaerobically digesting organic waste is an economical solution to the pressing concerns of the environment and utilizing sustainable energy. Mixing is an important operation that homogenizes anaerobic bacteria, nutrients, and temperature throughout the digester to maximize biogas production. The common mixing methods involve the use of gas mixers, mechanical stirring, and mechanical pumping, among which the mechanical stirring has proven to be the most efficient method in terms of mixing intensity per unit power consumption (Wu, 2009, 2010b). With the exception of highly viscous fluids mixed at a low impeller speed, mixing in the digesters always creates turbulence. When using computational fluid dynamics (CFD), choosing an appropriate turbulence model is critical to characterize the flow fields. Generally, the approaches involved in modeling turbulence are direct numerical simulation (DNS), large eddy simulation (LES), and the eddy

viscosity models. Both the DNS and LES are too computationally expensive for most engineering applications despite the fact that the DNS provides the best solution to turbulent flow and the LES shows a high accuracy in capturing large-scale chaotic structures. By contrast, an economic approach is to solve an eddy viscosity model that is based on the Reynolds-averaged NaviereStokes (RANS) equations with a turbulence closure. Among a large family of turbulence closures (zero-, one-, and two-equation, etc.), the standard ke3 model has been the most popular one used to simulate mixing (Sahu et al., 1999; Alexopoulos et al., 2002; Chapple et al., 2002; Pruvost et al., 2004; Kukukova et al., 2005; Mostek et al., 2005; Deglon and Meyer, 2006; Vakili and Nasr Esfahany, 2009). Sahu et al. (1999) introduced a zonal modeling method to predict mixing by five different axial-flow impellers in a tank. They claimed that predictions of the turbulent kinetic energy (k) closely match the laser Doppler anemometry (LDA) measurements, and proposed a new method to estimate the turbulent energy

E-mail address: [email protected]. 0043-1354/$ e see front matter ª 2010 Elsevier Ltd. All rights reserved. doi:10.1016/j.watres.2010.12.020

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 8 2 e2 0 9 4

Nomenclature B C CFD CPU d D DNS ! F G H k k Ks LDA LES MRF n N NP NQ p P PBT Q

baffle width, m impeller clearance, m computational fluid dynamics central processing unit diameter of moving zone, m impeller diameter, m direct numerical simulation body force, N generation for k or e liquid height, m consistency coefficient, Pa sn turbulent kinetic energy, m2/s2 empirical data laser Doppler anemometry large eddy simulation multiple reference frame power-law index rotating speed, rpm power number, dimensionless flow number, dimensionless pressure, Pa power, hp or kW pitched blade turbine flow rate, m3/s

dissipation rate (e). Alexopoulos et al. (2002) developed a twocompartment model to simulate the turbulent flow in a pilot plant reactor by varying vessel size, impeller diameter, agitation rate, and viscosity. The model validation was conducted in a non-homogeneous liquideliquid dispersion process, and an excellent agreement was obtained between predicted and measured values on the droplet size distributions over a wide range of experimental conditions. Chapple et al. (2002) reported that the power number is independent of the blade thickness and stays constant for Re > 2
104 while mixing with a pitched blade turbine (PBT) impeller via the LDA validation. Pruvost et al. (2004) assessed the standard keu model for a marine impeller in a torus reactor by comparing the CFD predictions with the LDA data. Kukukova et al. (2005) and Mostek et al. (2005) simulated the flow fields and homogenization in cylindrical vessels with multiple impellers on a central shaft to check the velocity profiles, power and pumping numbers, in which the PBT and standard Rushton turbine (RT) impellers were used. The simulated results were shown to closely agree with the experiments from the literature. Deglon and Meyer (2006) numerically investigated mixing by a RT impeller in a 15 cm diameter tank. Their studies showed that the standard ke3 model solved with the multiple reference frame (MRF) method can accurately predict the turbulent kinetic energy, provided very fine grids (nearly 2 million control volumes for half of the tank) are coupled with a higher-order discretization scheme. However, the results indicated that the flow field and mean fluid velocity predictions are not strongly influenced by either the grid resolution or the discretization scheme. Vakili and Nasr Esfahany (2009) studied the effects of agitator speed, impeller diameter, baffle width and impeller clearance

RANS Re Reg RSM RT S t T TS ! u UN v Y

2083

Reynolds-averaged NaviereStokes Reynolds number, dimensionless generalized Reynolds number Reynolds stress model Rushton turbine source term for k or e time, s tank diameter, m total solids concentration, g/l absolute velocity, m/s average velocity, m/s average velocity, m/s dissipation for k or e

Greek symbols b left (or right) inclined angle, deg g up (or down) inclined angle, deg G effective diffusivity g_ shear rate, s1 d error, dimensionless e dissipation rate of turbulent kinetic energy, m2/s3 h non-Newtonian viscosity, Pa s q spacing between two impellers, deg r density, kg/m3 s viscous stress, N/m2 u specific dissipation rate, 1/s

on turbulent flow field in the tank with a two-blade impeller and four baffles. The model calculations were validated against the specifications in an unbaffled tank reported by Alexopoulos et al. (2002). Despite the wide and intense utilization of the standard ke3 model, the model has deficiencies such as poorly simulating non-equilibrium boundary layers. Thus, examination of the other turbulence models remains an active topic of CFD research (Jaworski et al., 1998; Jaworski and Zakrzewska, 2002; Aubin et al., 2004; Murthy and Joshi, 2008). Jaworski et al. (1998) applied the RNG (renormalization group) ke3 model to simulate mixing by a hydrofoil impeller in a cylindrical tank, and obtained a good agreement of numerical predictions with the LDA velocity data. Later, Jaworski and Zakrzewska (2002) checked six turbulence models involving the standard ke3, the RNG ke3, the realizable ke3, the CheneKim ke3, the optimized CheneKim ke3, and the Reynolds stress model (RSM) in a flat-bottomed tank with a PBT impeller and four baffles. They compared the simulated tangential and axial mean velocity components as well as the turbulence kinetic energy with LDA data for the wall jet region in the tank, and concluded that (1) the tangential velocity was irrespective of the turbulence model, (2) the axial velocity was well predicted using the standard ke3 and the optimized CheneKim ke3 models, and (3) the turbulent kinetic energy was significantly under-predicted by all the turbulence models. Aubin et al. (2004) studied the effects of the standard ke3 and the RNG ke3 models on the numerical solution in a tank stirred by a PBT impeller, and showed that these two models underpredict the k value in the discharge jet of the impeller through the comparison of simulated and LDA results. Murthy and

2084

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 8 2 e2 0 9 4

Joshi (2008) conducted an extensive review of the LES, and evaluated the standard ke3 model, RSM and LES for five different impellers. The validation of the mean axial, radial and tangential velocities along with the turbulent kinetic energy revealed that the LES performs well for predicting all the flow variables. The research cited above has been done with the assumption of Newtonian fluids. Thus far, only a few studies have been published on non-Newtonian fluid mixing (Cumby, 1990; Kelly and Gigas, 2003; Dular et al., 2006; Bakker et al., 2009; Wu, 2009, 2010a,b,c). Cumby (1990) studied the effects of non-Newtonian characteristics of agricultural slurries on the impeller performance and compared various expressions for scale-up of impellers. Kelly and Gigas (2003) used the LDA to verify a CFD flow model for a shear-thinning fluid near an impeller in a tank, and presented the power number as the function of the Reynolds number defined by Metzner and Otto (1957) and the power-law index under laminar flow conditions. Dular et al. (2006) evaluated the capability of numerical simulation to predict the laminar flow of the carboxymethyl cellulose (CMC) power-law fluid stirred by a six-bladed vane impeller and checked the flow characteristics against the LDA measurements. Bakker et al. (2009) investigated agitation of HerscheleBulkey fluids by a PBT impeller, and declared that the shearestress transport (SST) keu model is more suitable than the standard keu model to predict turbulent mixing. Wu (2009, 2010a) performed CFD simulations of mechanical mixing in anaerobic digesters employing the realizable ke3 model, in which the liquid manure was assumed to be water or a nonNewtonian fluid that is dependent on the total solids (TS) concentrations. The predicted power and flow numbers of an impeller were validated against the lab specifications. A key issue that arises here is whether the realizable ke3 model is the most appropriate for simulating anaerobic digestion mixing by mechanical impellers. Subsequently, Wu (2010b,c) examined twelve turbulence models for single phase and two-phase fluid flow in a pipe, and reported that the SST keu model and the standard keu model could be used to simulate gas mixing and mechanical pumping in anaerobic digesters, respectively. Until now, no research is available to evaluate turbulence models for mechanical agitation of non-Newtonian fluids in the digesters.

2.

Objectives

The focus of this study is on examining six RANS-based twoequation turbulence models for anaerobic digestion mixing by mechanical agitators. The specific objectives of this study are to: 1. Develop a numerical model that describes mechanically stirring non-Newtonian fluids in anaerobic digesters; 2. Evaluate each turbulence model by comparing power and flow numbers from CFD with those from the lab specifications; 3. Predict the Reynolds number for the moving zone that characterizes the impeller rotation to judge the flow regime; 4. Check the effects of the discretization scheme and the numerical approach on predicting the power and flow numbers; 5. Validate the simulated velocities against the experimental data from the literature in a lab-scale digester; and

6. Optimize the mixing design in a full-scale digester with two side-entry impellers.

3.

Model development

The mathematical model that describes mechanical agitation of non-Newtonian fluids in anaerobic digesters was developed based on the following assumptions:  The digestion temperature is constant at 35  C, and water or a non-Newtonian fluid is isothermal and incompressible;  The manure slurry exhibits non-Newtonian pseudo-plastic fluid behavior when TS  2.5%;  The model is single phase, in which gas bubbleeliquid phase-interaction due to the biogas production is negligible.

3.1.

Governing equations

The governing equations consisting of continuity, momentum, turbulence transport (standard keu model), and rheological equations can be summarized as: vr þ V$ðr! uÞ ¼ 0 vt

(1)

v ! ! ðr u Þ þ V$ðr! u! u Þ ¼ Vp þ V$s þ F vt

(2)

v ðrkÞ þ V$ðr! u kÞ ¼ V$ðGk VkÞ þ Gk  Yk þ Sk vt

(3)

v ðruÞ þ V$ðr! u uÞ ¼ V$ðGu VuÞ þ Gu  Yu þ Su vt

(4)

h ¼ kg_ n1

(5)

r ¼ 0:0367TS3  2:38TS2 þ 14:6TS þ 1000

(6)

where r is liquid density, t is time, ! u is the absolute velocity ! vector, p is static pressure, s is viscous stress, F is body force, k in equation (3) and u are turbulence kinetic energy and specific dissipation rate, Gk and Gu are the effective diffusivity of k and u, Gk and Gu are the generation of k and u, Yk and Yu are the dissipation of k and u, Sk and Su are source terms, h is viscosity, k in equation (5) is the consistency coefficient, g_ is the shear rate, n is the power-law index, and TS is the percentage of total solids by weight in liquid manure.

3.2.

Definition of dimensionless number

When mechanically agitating non-Newtonian fluids, the Reynolds number (Re) is expressed as (Chen, 1981): Re ¼

rND2 h

(7)

where N is the rotating speed of the impeller, D is the impeller diameter, and h is determined by equation (5) in which the shear rate can be expressed as (Metzner and Otto, 1957): g_ ¼ Ks N where Ks is a constant value determined experimentally.

(8)

2085

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 8 2 e2 0 9 4

Similar to a pipe flow, the Reynolds number for fluid flow through the moving zone can be expressed as (Metzner and Reed, 1955): Reg ¼

n rU2n N d kð0:75 þ 0:25=nÞn 8n1

(9)

where UN is the average velocity of the fluid at the exit of the moving zone, and d is the diameter of the moving zone. The power number (NP) and flow number (NQ) in a stirred vessel are expressed as (Paul et al., 2004): NP ¼

P rN3 D5

(10)

NQ ¼

Q ND3

(11)

where P is the power input, and Q is the flow rate through the moving zone.

3.3.

CFD methodology

The commercial CFD package, Fluent 12.0, was used in this study (ANSYS-Fluent Inc., 2008). The implementation of a CFD modeling involves pre-processing, model setup, iterative calculation, and post-processing of the results. In the preprocessing, the baffles and impeller blades were treated as zero-thickness walls to reduce the mesh quantities. The whole computational domain was decomposed into several subdomains, in which only the sub-domains with the impellers (moving zones) were modeled using the tetrahedral grids while all the others were modeled using the hexahedral grids. In addition, the size functions were used to control the mesh growth. The model setup can be briefly described as follows:  Define a 3-D, steady, implicit, and pressure-based solver.  Define a turbulence model from the viscous model panel.  Enable the non-Newtonian fluid properties with turbulent flow conditions using the text command: define/models/ viscous/turbulence-expert/turb, and then define materials.  Define the operating conditions by turning on gravity.  Define the boundary conditions.  Set the rotation origin and direction, and specify each moving zone with MRF at a rotational speed.

Table 1 e Effect of mesh refinement. Mesh cells

NP dP (%) NQ dQ (%) d (%) CPU time (h)

96,460

157,254

2,420,395

1.29 20.4 0.81 5.8 13.1 1

1.27 21.6 0.82 4.7 13.15 2

1.24 23.5 0.83 3.4 13.45 36

 Under the same rotation origin and direction for each moving zone, set the impeller blades and the shaft inside the moving zone at zero rotational speed relative to the adjacent cell zone, and set the shaft outside the moving zone at the absolute rotational speed of the impeller.  Specify zero normal gradients of all variables at the liquid surface (symmetry boundary).  Specify the solution controls by setting the under-relaxation factors and choosing the discretization schemes.  Initialize the flow fields.  Enable the residual monitors and a surface monitor for the moment of the impeller blades (torque).  Solve the steady-state flow fields until convergence.  Switch the solver from steady to unsteady.  Switch the motion type from MRF to sliding mesh, and use the converged flow fields solved by the MRF approach as the initial conditions for solving the sliding mesh.  Set the time step size, number of time steps, and maximum iterations per time step.  Run the transient calculation to achieve a steady-state solution. In this study, all simulations were executed on a HewlettPackard xw6200 workstation with 12 GB of RAM using five parallel processors for Fluent 12.0 in windows. The primary convergence criterion for each solution was the impeller torque approaching a constant value.

4.

Results and discussions

The RANS-based turbulence models consist of one-equation (k), two-equation (ke3 or keu), the RSM, etc. It should be

Fig. 1 e Two axial impellers for CFD simulation.

2086

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 8 2 e2 0 9 4

Table 2 e Rheology of manure slurry at six TS levels (Wu and Chen, 2008). TS (%) 0 2.5 5.4 7.5 9.1 12.1

k (Pa sn)

n

0.001 0.042 0.192 0.525 1.052 5.885

1 0.710 0.562 0.533 0.467 0.367

g_ (s1)

hmin (Pa s)

hmax (Pa s)

r (kg/m3)

e 226e702 50e702 11e399 11e156 3e149

e 0.006 0.01 0.03 0.07 0.25

e 0.008 0.03 0.17 0.29 2.93

998.2 1000.36 1000.78 1001.00 1001.31 1001.73

mentioned that there are many low-Reynolds-number ke3 models in which the transport equations are solved through the boundary layer using a fine near-wall mesh. In the author’s previous research (Wu, 2010b), it has been shown that using a low-Reynolds-number ke3 model has a high computing cost even though it is superior to the other twoequation models in anaerobic digestion mixing. Accordingly, this study focused on checking six turbulence models except for the low-Reynolds-number ke3 models. The candidate models included: (1) the standard ke3 model, (2) the RNG ke3 model, (3) the realizable ke3 model, (4) the standard keu model with shear flow corrections, (5) the SST keu model, and (6) the RSM. The power number of an impeller in a mixing tank is analogous to the friction factor in a pipe flow or the drag coefficient of a solid body immersed in a flowing stream (Wang et al., 2005). It has been established that the power number is inversely proportional to the Reynolds number under laminar flow conditions, while it is constant and independent of the Reynolds number under turbulent flow conditions. In the turbulence model tests, a PBT 4/45 impeller as shown in Fig. 1(a) was used to agitate non-Newtonian fluids in a cylindrical tank with a flat bottom and four equally spaced baffles. The power and flow numbers of this impeller were measured as 1.62 and 0.86 from the lab experiments at Philadelphia Mixing Solutions Ltd. (PMSL). Provided that D ¼ 0.3 m for the impeller diameter, the other dimensions such as tank diameter (T ), liquid height (H ), impeller clearance (C ), and width of each baffle (B) can be

determined with T ¼ 3D, H ¼ T, C ¼ D, and B ¼ T/12. In order to quantitatively evaluate the performance of a turbulence model, three error indicators were proposed as: NP  NP dP ¼ N



lab


100%

P lab

NQ  NQ dQ ¼ N

(12)


100%

lab

Q lab

(13)

d ¼ 0:5ðdP þ dQ Þ

(14)

where dP and dQ are errors for predicted power and flow numbers, NP_lab and NQ_lab are obtained from the lab specifications, and d is the weighted mean error.

4.1.

Grid-independent test

The grid-independent test is used to determine the minimum grid resolution that achieves a solution independent of the mesh quantities. In this test three different mesh sizes were evaluated: 96,460 (coarse mesh), 157,254 (fine mesh), and 2,420,395 (very fine mesh). In the model setup, the standard discretization scheme was used for the pressure equation and the first-order upwind discretization scheme was used for all other equations, the pressureevelocity coupling was solved using the SIMPLE (semi-implicit method for pressure-linked equations) algorithm, and the impeller rotational motion was characterized by the MRF approach. This model setup would apply to all the simulations described hereafter unless noted. Table 1 shows the predicted power and flow numbers for water (TS ¼ 0) being stirred at N ¼ 60 rpm, in which the standard ke3 model was used to characterize turbulence. This table shows that the predicted power number decreases while the predicted flow number increases with the increase in mesh cells. In general, the weighted mean error fluctuates insignificantly from coarse mesh to very fine mesh. Obviously, using a very fine mesh does not improve the model prediction because it yields a high discretization error. Moreover, the central processing unit (CPU) time increases exponentially

Table 3 e Reynolds number calculated by two methods. TS (%)

0

N (rpm) Re SKE RNG RKE SKO SST RSM

Reg Error Reg Error Reg Error Reg Error Reg Error Reg Error

(%) (%) (%) (%) (%) (%)

2.5

5.4

7.5

9.1

12.1

60 89,838

60 3496

150 3667

250 3060

350 3143

800 3061

e e e e e e e e e e e e

3364 3.8 3601 3.0 3531 1.0 3603 3.1 3501 3.2 3380 3.3

3422 6.7 3652 0.4 3638 0.8 3431 6.4 3431 6.4 3502 4.5

2572 16.0 2625 14.2 2670 12.7 2474 19.1 2415 21.1 2622 14.3

2585 17.8 2630 16.3 2760 12.2 2472 21.4 2411 23.3 2717 13.6

2384 22.1 2446 20.1 2554 16.6 2289 25.2 2216 27.6 2527 17.4

Note: SKE d standard ke3, RNG d RNG ke3, RKE d realizable ke3, SKO d standard keu, SST d SST keu, RSM d Reynolds stress model. The error calculations were based on the Reynolds number obtained from equation (7) with Ks ¼ 5.4.

2087

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 8 2 e2 0 9 4

Table 4 e Power number versus turbulence model and TS. TS (%) 0

SKE RNG RKE SKO SST RSM

2.5

5.4

7.5

12.1

NP

dP (%)

NP

dP (%)

NP

dP (%)

NP

dP (%)

NP

dP (%)

NP

dP (%)

1.29 1.28 1.26 1.37 1.28 1.26

20.4 21.0 22.2 15.4 21.0 22.2

1.26 1.24 1.24 1.31 1.23 1.26

22.2 23.5 23.5 19.1 24.1 22.2

1.23 1.21 1.23 1.23 1.17 1.24

24.1 25.3 24.1 24.1 27.8 23.5

1.20 1.18 1.18 1.22 1.17 1.17

25.9 27.2 27.2 24.7 27.8 27.8

1.20 1.18 1.19 1.23 1.18 1.19

25.9 27.2 26.5 24.1 27.2 26.5

1.23 1.21 1.22 1.23 1.18 1.21

24.1 25.3 24.7 24.1 27.2 25.3

with the number of grids. To reduce CPU time meshing with the lowest number of mesh cells (96,460) was selected after the grid-independent test since the solution accuracy would not be sacrificed.

4.2.

9.1

with reasonable CPU time and computational stability, (2) the standard keu model shows excellent predictions at TS ¼ 0 and 2.5%, (3) the realizable ke3 model have solutions with the lowest errors at TS ¼ 5.4%, 9.1%, and 12.1%, (4) the standard ke3, the RNG ke3, and the realizable ke3 models perform marginally better than the other models at TS ¼ 7.5%, (5) the standard ke3 model converges relatively fast, and (6) the RSM is computationally expensive compared to the other turbulence models. In flows with strong swirl, a modeling strategy starting with the standard ke3 model, then switching to the RNG ke3 or the realizable ke3 model, and eventually using the RSM (if desired) is recommended.

Evaluation of turbulence models

Given the rheology of manure slurry at six TS levels (Table 2), the rotating speed of the impeller can be adjusted accordingly to perform turbulent mixing (Re > 2000) as shown in the 2nd row of Table 3. With the introduction of Ks ¼ 5.4 for the PBT 4/45 impeller (Skelland, 1983), the Reynolds number with respect to TS can be determined with equations (7) and (8) as shown in the 3rd row of Table 3. Based on these model inputs, comprehensive simulations of mechanically agitating manure slurry at six TS levels in the tank were carried out for each turbulence model. Tables 4 and 5 show the power and flow numbers predicted by equations (10) and (11), in which the torque and pumping capacity of the impeller for each simulation were obtained through post-processing. The results show that CFD performs much better in predicting the flow number than the power number. Table 6 shows the weighted mean error with respect to the turbulence model and TS where the power and flow numbers are assumed to have equal weights. If the errors arising from both the numerical calculations and lab measurements are taken into consideration, d  10% could be regarded as a very good prediction. Due to unavailability of quantitative criteria for the evaluation of turbulence models, d  30% was set as the criterion for selecting a useful model. According to the error analysis along with the observation of convergence history for each run, it came out that (1) all the models are capable of predicting mechanical agitation of non-Newtonian fluids at six TS levels

4.3.

Characterization of turbulent mixing

Upon obtaining the converged flow fields, an image of the flow patterns can be processed. Taking the case of N ¼ 150 rpm and TS ¼ 5.4% as an example, Fig. 2(a) shows the global velocity contours in the vertical plane midway between two baffles, a characteristic pattern of a PBT impeller pumping down. When the impeller rotates clockwise, the impeller blades drive the liquid down to the bottom of the vessel, where the stream spreads radially in all directions toward the side wall, flows upward along the wall, and returns axially to the top of the impeller. The contour bar below the figure represents the velocity scales ranging from zero (blue area) to the maximum value (red area) in which the velocity magnitudes are greater than (or equal to) the maximum value specified (For interpretation of the references to color in this sentence, the reader is referred to the web version of this article.). These representations apply to all the contour figures in this study. Fig. 2(b) shows the local velocity vectors through the moving zone, from which the average velocity magnitude that is normal to the exit

Table 5 e Flow number versus turbulence model and TS. TS (%) 0

SKE RNG RKE SKO SST RSM

2.5

5.4

7.5

9.1

12.1

NQ

dQ (%)

NQ

dQ (%)

NQ

dQ (%)

NQ

dQ (%)

NQ

dQ (%)

NQ

dQ (%)

0.81 0.85 0.82 0.86 0.84 0.81

5.8 1.2 4.7 0 2.3 5.8

0.80 0.82 0.81 0.82 0.82 0.80

7.0 4.7 5.8 4.7 4.7 7.0

0.79 0.81 0.81 0.76 0.77 0.80

8.1 5.8 5.8 11.6 10.5 7.0

0.76 0.77 0.77 0.73 0.72 0.76

11.6 10.5 10.5 15.1 16.3 11.6

0.76 0.76 0.78 0.72 0.72 0.77

11.6 11.6 9.3 16.3 16.3 10.5

0.75 0.76 0.77 0.72 0.72 0.76

12.8 11.6 10.5 16.3 16.3 11.6

2088

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 8 2 e2 0 9 4

Table 6 e Weighted mean error for power and flow numbers (%).

Table 7 e Effect of the discretization scheme. TS (%)

TS (%)

SKE RNG RKE SKO SST RSM

0

2.5

5.4

7.5

9.1

12.1

13.1 11.1 13.45 7.7 11.65 14.0

14.6 14.1 14.65 11.9 14.4 14.6

16.1 15.55 14.95 17.85 19.15 15.25

18.75 18.85 18.85 19.9 22.05 19.7

18.75 19.4 17.9 20.2 21.75 18.5

18.45 18.45 17.6 20.2 21.75 18.45

plane of the moving zone (UN) can be obtained using the function of “area-weighted average” in the post-processing. Given an empirical value of Ks, the Reynolds number for mechanical agitation of non-Newtonian fluids can be predicted by equations (7) and (8) to determine the flow regime (laminar, transitional, or turbulent flow). For some commonly used impellers such as PBT and RT, the experimental data on Ks has been extensively reported in the literature. If Ks is unknown to an impeller, however, the flow information in the moving zone can be used to calculate the Reynolds number (Reg) defined by equation (9). For example, Reg can be predicted as 3431 if d ¼ 0.4 m and UN ¼ 0.59 m/s at TS ¼ 5.4% using the standard keu model. Following this procedure, all other Reynolds numbers can be obtained as shown in the rows labeled Reg in Table 3. Comparison of the Reynolds numbers calculated by two different methods demonstrates that the difference between Reg and Re is insignificant at TS ¼ 2.5% and 5.4%, whereas Reg is less than Re for TS  7.5%. Overall, the Reg calculated from the moving zone is a conservative value compared to the Re value.

4.4.

Effect of the discretization scheme

To check the effect of the discretization scheme on the power and flow numbers, the second-order upwind discretization

0

2.5

5.4

7.5

9.1

12.1

RKE

NP dP (%) NQ dQ (%) d (%)

1.25 22.8 0.85 1.2 12

1.22 24.7 0.84 2.3 13.5

1.21 25.3 0.83 3.4 14.35

1.17 27.8 0.79 8.1 17.95

1.18 27.2 0.79 8.1 17.65

1.19 26.5 0.80 7.0 16.75

SKO

NP dP (%) NQ dQ (%) d (%)

1.33 17.9 0.86 0 8.95

1.25 22.8 0.84 2.3 12.55

1.26 22.2 0.83 3.4 12.8

1.20 25.9 0.75 12.8 19.35

1.18 27.2 0.76 11.6 19.4

1.21 25.3 0.74 13.9 19.6

scheme was used for the momentum equations while all other schemes were kept unchanged. Based on the turbulence model analysis, the realizable ke3 and the standard keu models have proved to be more appropriate than the other turbulence models, discussions from here on were conducted using these two models. To save computing time iterations were started with the first-order scheme and then switched to the secondorder scheme. The results presented in Table 7 are an improvement in accuracy for the flow number, however, the accuracy of the power number declined. Hence, the secondorder upwind scheme can be adopted if the flow number is of primary interest.

4.5.

Effect of the numerical approach

In order to examine the effect of the numerical approach on the power and flow numbers, simulations with the sliding mesh approach were performed based on the converged flow fields solved by the MRF approach. This modeling strategy requires the conformal interfaces between the moving and stationary zones be switched to the non-conformal ones. Since the sliding

Fig. 2 e Flow patterns in a tank with PBT 4/45 impeller at N [ 150 rpm and TS [ 5.4%.

2089

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 8 2 e2 0 9 4

Table 8 e Effect of the numerical approach. TS (%) 0

2.5

5.4

7.5

9.1

12.1

RKE

NP dP (%) NQ dQ (%) d (%)

1.27 21.6 0.83 3.4 12.5

1.27 21.6 0.82 4.7 13.15

1.24 23.5 0.82 4.7 14.1

1.19 26.5 0.78 9.3 17.9

1.20 25.9 0.79 8.1 17.0

1.22 24.7 0.78 9.3 17.0

SKO

NP dP (%) NQ dQ (%) d (%)

1.38 14.8 0.86 0 7.4

1.32 18.5 0.82 4.7 11.6

1.25 22.8 0.77 10.5 16.65

1.23 24.1 0.73 15.1 19.6

1.24 23.5 0.72 16.3 19.9

1.26 22.2 0.72 16.3 19.25

mesh approach solves the transient flow fields in which the grid surrounding the impeller physically moves during the solution, the plot of impeller torque with respect to time exhibits a sine-like curve with periodic cycles in which the maximum point occurs at the moment the blade tips face the baffles while the minimum point occurs when the blade tips arrive at the midway point between two baffles. In this study, the torque value on the maximum point was used to calculate the power number because four blade tips were set to face four baffles in the pre-processing of Gambit. Table 8 shows the power and flow numbers predicted by the sliding mesh approach with remaining the model setup unchanged as indicated in Section 4.1. Compared to the MRF approach, the sliding mesh approach generally improves the predictions but it takes much longer CPU time to obtain one solution. From an engineering standpoint, it is recommended that the MRF approach be used because of its acceptable accuracy and low computing cost.

4.6.

Model validation

Model validation was conducted in a lab-scale digester reported by Hoffmann et al. (2008). In their experiment, mechanical agitation of manure slurry (TS ¼ 5%) was carried out in a cylindrical tank with a 25 slope angle hopper bottom (T ¼ 0.152 m and H ¼ 0.23 m) equipped with a top-entry Lightnin A310 impeller (D ¼ 0.062 m and C ¼ 0.05 m). The operating temperature was maintained at 34  1  C, and the impeller was operated at N ¼ 250 and 500 rpm with down-pumping mode. A non-invasive technique that involves computer automated radioactive particle tracking and computed tomography was used to measure velocities. Due to unavailability of the A310 impeller and the rheological properties of manure slurry at TS ¼ 5%, an alternative impeller made in PMSL (3LS39) was

Fig. 4 e Comparison of azimuthally averaged axial velocities from CFD and measurements.

used to agitate the manure slurry at TS ¼ 5.4% in the CFD simulations. Although these two impellers have different geometrical configurations as shown in Fig. 3, both are axial impellers with three blades having the identical power number (NP ¼ 0.3) and flow number (NQ ¼ 0.56) which are available from the literature (Spicer et al., 1996) and the lab specifications at PMSL. Because of the similar flow patterns for N ¼ 250 and 500 rpm, model validation was only conduced at N ¼ 500 rpm in this section. Fig. 4 shows comparison of the simulated and measured axial velocities with respect to the radial position at z ¼ 0.13 m. If the replacement of impeller, the use of approximate values for rheological properties, the measurement inaccuracies, and the numerical errors are taken into consideration, the CFD predictions are in reasonable agreement with the measurements. Moreover, checking the predicted power and flow numbers shows that the weighted mean errors are within 30% (Table 9), demonstrating that the simulated results are acceptable.

4.7.

Model application

In this section, mechanically stirring manure slurry with two identical side-entry impellers in a full-scale anaerobic digester constructed from a cylindrical tank with a conical bottom was investigated. The major design parameters were:    

Tank diameter ¼ 12 m. Cylindrical height ¼ 6.7 m. Conical height ¼ 0.9 m. Impeller diameter ¼ 0.69 m.

Table 9 e Predicted power and flow number for PMSL 3LS39 impeller.

Fig. 3 e Geometrical configurations of two axial impellers for model validation.

RKE SKO

NP

dP (%)

NQ

dQ (%)

d (%)

0.234 0.233

21.8 22.1

0.423 0.422

24.5 24.6

23.15 23.35

2090

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 8 2 e2 0 9 4

Fig. 5 e Definition of installation angles for side-entry impellers.

 Shaft length ¼ 1 m.  Impeller mounted height ¼ 1.6 m. Fig. 1(b) shows the focused flow impeller which has been reported in the author’s previous study (Wu, 2009). As shown

in Fig. 5, placing impellers involves the spacing between two impellers (q), left (or right) inclined angle for each impeller (b), and up (or down) inclined angle for each impeller (g). Note that the positive b and g represent left and up inclined angles, while the negative b and g represent right and down inclined angles. For the purpose of quantitative analysis, the mixing intensity was quantified by the average velocity throughout the whole digester tank ðvÞ. In order to determine the optimum installation parameters (q, b, and g), CFD simulations of mixing by two focused flow impellers without draft tubes were carried out at N ¼ 300 rpm and TS ¼ 5.4% using the standard keu model. In the post-processing, all the velocity contours were produced from the horizontal plane at the impeller mounted height. First, the effect of the spacing between two impellers on the flow fields was checked by q ¼ 45 , 90 , 135 , and 180 , provided that b ¼ 10 and g ¼ 5 . Fig. 6 shows that the mixing intensity decreases with an increase in q, and the decline in intensity for q > 90 is significant. This phenomenon may be attributed to the enhancement of two

Fig. 6 e Velocity contours versus spacing of two impellers at b [ 10 , g [ L5 , N [ 300 rpm, and TS [ 5.4%.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 8 2 e2 0 9 4

flow streams along the same direction when q  90 but the attenuation of two flow streams in opposite directions when q > 90 . Second, sensitivity analysis of the left inclined angle for each impeller (b) was conducted with the assumption of q ¼ 15 and g ¼ 5 . As shown in Fig. 7, the average velocity at b ¼ 30 is greater than those at b ¼ 15 , 20 , and 25 . This can be explained by the large inclined angle of each impeller toward the tank wall producing high tangential velocities leading to a strongly swirling flow. Third, mixing respectively with down-pumping (g ¼ 5 ), horizontal-pumping (g ¼ 0 ), and up-pumping (g ¼ 5 ) for each impeller was analyzed, in which the parameters of q and b were fixed. Fig. 8 illustrates that the overall flow patterns along the vertical plane for three pumping modes are fairly similar. However, the pumping down is helpful to prevent solid particles from settling evidenced by the change of the color map near the tank bottom from green at g ¼ 5 to blue at g ¼ 5 (For interpretation of the references to color in this sentence, the reader is

2091

referred to the web version of this article.). Consequently, q ¼ 15 and b ¼ 30 as well as g ¼ 5 were selected as the optimal installation parameters for this mechanically stirring system. Finally, the effect of rheology on the mixing performance was investigated based on the optimum installation of the impellers. Figs. 9 and 10(a) show the velocity contours at five TS levels, from which it can be shown that the average velocity deceases with an increase in TS, and that v at TS ¼ 12.1% is much less than those at other TS levels. Because of the lack of a Ks value for the focused flow impeller, the Reynolds number (Reg) defined by equation (9) can be calculated as 97,717, 31,190, 12,176, 7133, and 1602 for TS ¼ 2.5%, 5.4%, 7.5%, 9.1%, and 12.1%, respectively. Given the fact that the calculated Reg is a conservative value as compared to the Re value as indicated before, turbulent mixing can be achieved at TS  2.5% in this digester with two side-entry impellers rotating at N ¼ 300 rpm. To gain an insight into the flow patterns, taking the case of TS ¼ 0 as an example, the flow

Fig. 7 e Velocity contours versus left inclined angle for each impeller at q [ 15 , g [ L5 , N [ 300 rpm, and TS [ 5.4%.

2092

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 8 2 e2 0 9 4

Fig. 8 e Velocity contours versus vertically inclined angle for each impeller at q [ 15 , b [ 30 , N [ 300 rpm, and TS [ 5.4%.

Fig. 9 e Velocity contours versus TS at q [ 15 , b [ 30 , g [ L5 , and N [ 300 rpm.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 8 2 e2 0 9 4

2093

Fig. 10 e Flow patterns at q [ 15 , b [ 30 , g [ L5 , N [ 300 rpm, and TS [ 0: (a) velocity contours, and (b) pathlines colored by velocity magnitude.

pathlines colored by the velocity magnitude released from two impellers are depicted in Fig. 10(b). The 3-D visualization shows a high velocity along each impeller discharging direction while a low velocity near the central zone, which corresponds to the velocity contours shown in Fig. 10(a). It should be indicated that this section aims at developing a CFD methodology to perform optimum installation of impellers for a specific case rather than providing a universal solution to mixing by side-entry impellers.

5.

Conclusions

The following conclusions are based on the results obtained from this study: 1. Of the six turbulence models (standard ke3, RNG ke3, realizable k-e, standard keu, SST keu, and Reynolds stress model) used to predict mechanical agitation of non-Newtonian fluids at six TS levels, the standard keu and the realizable ke3 models are highly recommended. 2. If the value of Ks for agitating non-Newtonian fluids by an impeller is unavailable, the Reynolds number calculated from the moving zone is a conservative quantity for judging the flow regime. 3. The second-order discretization scheme for the momentum equations improves the prediction of the flow number but adversely affects the accuracy of the power number. The sliding mesh approach generally performs somewhat better in predicting the power and flow numbers than the MRF approach, however, it requires a much longer computing time. From an engineering standpoint, the MRF approach is recommended. 4. Simulation of mixing in a full-scale digester with two sideentry impellers using the standard keu model and the MRF approach indicates that 15 -spacing between two impellers with each one aimed 30 left and 5 down yields the most

efficient mixing, and that the mixing intensity decreases with an increase in TS.

references

Alexopoulos, A.H., Maggioris, D., Kiparissides, C., 2002. CFD analysis of turbulence non-homogeneity in mixing vessels: a two-compartment model. Chemical Engineering Science 57, 1735e1752. ANSYS-Fluent Inc., 2008. Fluent 12.0. Lebanon, NH. Aubin, J., Fletcher, D.F., Xuereb, C., 2004. Modeling turbulent flow in stirred tanks with CFD: the influence of the modeling approach, turbulence model and numerical scheme. Experimental Thermal and Fluid Science 28, 431e445. Bakker, C.W., Meyer, C.J., Deglon, D.A., 2009. Numerical modeling of non-Newtonian slurry in a mechanical flotation cell. Minerals Engineering 22, 944e950. Chapple, D., Kresta, S.M., Wall, A., Afacan, A., 2002. The effect of impeller and tank geometry on power number for a pitched blade turbine. Trans IChemE 80 (Part A), 364e372. Chen, Y.R., 1981. Impeller power consumption in mixing livestock manure slurries. Transactions of the ASAE 24 (1), 187e192. Cumby, T.R., 1990. Slurry mixing with impellers: part 1, theory and previous research. Journal of Agricultural Engineering Research 45, 157e173. Deglon, D.A., Meyer, C.J., 2006. CFD modeling of stirred tanks: numerical consideration. Minerals Engineering 19, 1059e1068. Dular, M., Bajcar, T., Slemenik-Perse, L., Zumer, M., Sirok, B., 2006. Numerical simulation and experimental study of nonNewtonian mixing flow with a free surface. Brazilian Journal of Chemical Engineering 23 (4), 473e486. Hoffmann, R.A., Garcia, M.L., Veskivar, M., Karim, K., Al-Dahhan, M.H., Angenent, L.T., 2008. Effect of shear on performance and microbial ecology of continuously stirred anaerobic digesters treating animal manure. Biotechnology and Bioengineering 100 (1), 38e48. Jaworski, Z., Dyster, K.N., Mishra, V.P., Nienow, A.W., Wyszynski, M.L., 1998. A study of an up- and a down-pumping

2094

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 8 2 e2 0 9 4

wide-blade hydrofoil impeller: part II. CFD analysis. The Canadian Journal of Chemical Engineering 76, 866e876. Jaworski, Z., Zakrzewska, B., 2002. Modeling of the turbulent wall jet generated by a pitched blade turbine impeller: the effect of turbulence model. Trans IChemE 80 (Part A), 846e854. Kelly, W., Gigas, B., 2003. Using CFD to predict the behavior of power law liquids near axial-flow impellers operating in the transitional flow regime. Chemical Engineering Science 58, 2141e2152. Kukukova, A., Mostek, M., Jhoda, M., Machon, V., 2005. CFD prediction of flow and homogenization in a stirred vessel: part I vessel with one and two impellers. Chemical Engineering & Technology 28 (10), 1125e1133. Metzner, A.B., Otto, R.E., 1957. Agitation of non-Newtonian fluids. AIChE Journal 3 (1), 3e10. Metzner, A.B., Reed, J.C., 1955. Flow of non-Newtonian fluidscorrelation of laminar, transition and turbulent-flow regions. AIChE Journal 1, 434e440. Mostek, M., Kukukova, A., Jhoda, M., Machon, V., 2005. CFD prediction of flow and homogenization in a stirred vessel: part II vessel with three and four impellers. Chemical Engineering & Technology 28 (10), 1134e1143. Murthy, B.N., Joshi, J.B., 2008. Assessment of standard ke3, RSM and LES turbulence models in a baffled stirred vessel agitated by various impeller designs. Chemical Engineering Science 63, 5468e5495. Paul, E.L., Atiemo-Obeng, V.A., Kresta, S.M., 2004. Handbook of Industrial Mixing. John Wiley & Sons, Inc, New Jersey. Pruvost, J., Legrand, J., Legentilhomme, P., Rosant, J.M., 2004. Numerical investigation of bend and torus flow e part II: flow

simulation in torus reactor. Chemical Engineering Science 59, 3359e3370. Sahu, A.K., Kumar, P., Patwardhan, A.W., Joshi, J.B., 1999. CFD modeling and mixing in stirred tanks. Chemical Engineering Science 54, 2285e2293. Skelland, A.H.P., 1983. Mixing and agitation of non-Newtonian liquids. In: Cheremisionoff, N.P. (Ed.), Handbook of Fluids in Motion. Ann Arbor Science, Michigan, pp. 179e209. Spicer, P.T., Keller, W., Pratsinis, S.E., 1996. The effect of impeller type on floc size and structure during shear-induced flocculation. Journal of Colloid and Interface Science 184, 112e122. Vakili, M.H., Nasr Esfahany, M., 2009. CFD analysis of turbulence in a baffled stirred tank, a three-compartment model. Chemical Engineering Science 64, 351e362. Wang, L.K., Hung, Y.T., Shammas, N.K., 2005. Physicochemical Treatment Process. Humana Press, Totowa, New Jersey. Wu, B., 2009. CFD analysis of mechanical mixing in anaerobic digesters. Transactions of the ASABE 52 (4), 1371e1382. Wu, B., 2010a. CFD simulation of mixing in egg-shaped anaerobic digesters. Water Research 44 (5), 1507e1519. Wu, B., 2010b. CFD simulation of gas and non-Newtonian fluid two-phase flow in anaerobic digesters. Water Research 44 (13), 3861e3874. Wu, B., 2010c. Computational fluid dynamics investigation of turbulence models for non-Newtonian fluid flow in anaerobic digesters. Environmental Science & Technology 44 (23), 8989e8995. Wu, B., Chen, S., 2008. CFD simulation of non-Newtonian fluid flow in anaerobic digesters. Biotechnology and Bioengineering 99 (3), 700e711.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 9 5 e2 1 0 3

Available at www.sciencedirect.com

journal homepage: www.elsevier.com/locate/watres

Silver-modified mesoporous TiO2 photocatalyst for water purification Zhigang Xiong a, Jizhen Ma a, Wun Jern Ng b, T. David Waite c, X.S. Zhao a,* a

Department of Chemical and Biomolecular Engineering, National University of Singapore, 4 Engineering Drive 4, Singapore 117576, Singapore b Nanyang Environment & Water Research Institute, Nanyang Technological University, Nanyang Avenue, Singapore 639798, Singapore c School of Civil and Environmental Engineering, The University of New South Wales, Sydney, NSW 2052, Australia

article info

abstract

Article history:

Mesoporous anatase (TiO2) was modified with silver (Ag) nanoparticles using a photore-

Received 22 July 2010

duction method. Performance of the resulting TiO2eAg nanocomposites for water purifi-

Received in revised form

cation was evaluated using degradation of Rhodamine B (RhB) and disinfection of

5 December 2010

Escherichia coli (E. coli) under ultraviolet (UV) irradiation. The composites with different Ag

Accepted 20 December 2010

loadings were characterized using physical adsorption of nitrogen, X-ray diffraction, X-ray

Available online 24 December 2010

photoelectron spectroscopy and UVeVisible diffuse reflectance spectroscopic techniques. The results showed that metallic Ag nanoparticles were firmly immobilized on the TiO2

Keywords:

surface, which improved electron-hole separation by forming the Schottky barrier at the

Silver nanoparticles

TiO2eAg interface. Photocatalytic degradation of RhB and inactivation of E. coli effectively

Mesoporous titanium dioxide

occurred in an analogical trend. The deposited Ag slightly decreased adsorption of target

Photocatalysis

pollutants, but greatly increased adsorption of molecular oxygen with the latter enhancing

Disinfection

production of reactive oxygen species (ROSs) with concomitant increase in contaminant photodegradation. The optimal Ag loadings for RhB degradation and E. coli disinfection were 0.25 wt% and 2.0 wt%, respectively. The composite photocatalysts were stable and could be used repeatedly under UV irradiation. ª 2011 Elsevier Ltd. All rights reserved.

1.

Introduction

The titanium dioxide (TiO2) photocatalyst is gaining growing research interest (Chen and Mao, 2007; Han et al., 2009; Hoffmann et al., 1995; Linsebigler et al., 1995; Xu et al., 2007; Xu and Langford, 1995; Yu et al., 2003). Upon UV light irradiation, TiO2 generates highly reactive oxygen species (ROSs) and valence holes, which can degrade most recalcitrant pollutants, including non-biodegradable dyes (Han et al., 2009), phenols and their derivatives (Xu et al., 2007), and inactivate various microorganisms, such as bacteria, viruses, and tumor cells (Sunada et al., 2003b). However, the lifetime of ROSs, such as

O
 2 , HO , HOO and H2O2, is relatively short, particularly under

ultraviolet (UV) irradiation. This has often led to incomplete mineralization of organic pollutants (Tachikawa et al., 2007). To effectively improve the mineralization process, a means is needed to enhance adsorption of organic pollutants or dissolved O2 onto the surface of TiO2 (Kang et al., 2008; Mogyorosi et al., 2002). However, due to its poor affinity towards organic pollutants and the low surface area of TiO2 nanoparticles, the adsorption of organic pollutants on TiO2 surface is relatively low, resulting in slow photocatalytic degradation rates. To overcome this shortcoming, TiO2 nanoparticles have been supported on various porous solids (Hidaka et al., 1990; Kang et al., 2008; Mogyorosi et al., 2002; Xu and Langford, 1995). Using advanced porous materials for water treatments has

* Corresponding author. Tel.: þ65 65164727; fax: þ65 67791936. E-mail address: [email protected] (X.S. Zhao). 0043-1354/$ e see front matter ª 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.watres.2010.12.019

2096

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 9 5 e2 1 0 3

been demonstrated to be a very promising approach to water purification (Pan et al., 2010; Zhao et al., 2010). The materials can be synthesized via template (Tian et al., 2002; Yu et al., 2003) and solegel methods (Pinna and Niederberger, 2008; Raveendran et al., 2008). Mesoporous TiO2 photocatalysts have been observed to display an improved performance in the degradation of chlorophenols and dyes under UV irradiation (Beyers et al., 2005; Schattka et al., 2002). The porous structure of mesoporous TiO2 may also provide an adequate environment for the adsorption of dissolved O2 on the oxygen vacancies (Epling et al., 1998). It is recognized that reduction of the adsorbed O2 is a rate-limiting step in the photocatalytic degradation of organic pollutants (Hoffmann et al., 1995; Sadeghi et al., 1996), and so increase in O2 adsorption on TiO2 surfaces would enhance charge separation, and so generate more ROSs. Another factor that greatly restricts the photocatalytic activities of TiO2 is the low quantum yield of excitons due to the fast electron-hole (eehþ) recombination. Methods to effectively improve the photocatalytic activity of TiO2 have been widely explored over the past decade. Direct doping of an element, optimization of morphology and facets, and adding additional scavengers are amongst the most effective approaches (Chen and Mao, 2007; Hoffmann et al., 1995; Xu et al., 2007; Xu and Langford, 1995). In addition, deposition of metal nanoparticles with a large work function, such as Ag (Tran et al., 2006), Pt (Kowalska et al., 2008), and Au (Bannat et al., 2009; Subramanian et al., 2001; Wang et al., 2008), onto TiO2 surface has been found to efficiently retard the eehþ recombination because of the Schottky barrier formed at the metalesemiconductor interface. Here, the metal nanoparticles act as a mediator in storing and shuttling photogenerated electrons from the TiO2 surface to an acceptor. Iliev et al. (Iliev et al., 2006) demonstrated that deposition of Pt or Ag on the surface of TiO2 greatly enhanced the photocatalytic degradation of oxalic acid due to the increased separation of eehþ and higher rate of O2 reduction. Similarly, Li et al. (2007) homogenously embedded gold particles onto the TiO2 framework, which significantly improved photocatalytic degradation of phenol and chromium in solution. Other organic pollutants, like methanol (Ismail et al., 2009), methylene blue (Wang et al., 2008) and methyl orange (Arabatzis et al., 2003), could also be more efficiently degraded by the metaleTiO2 nanocomposite. Reactive oxygen species, and particularly hydroxyl radicals (HO
∙), produced by the irradiated metaleTiO2 has been considered to be the dominant species contributing to the degradation of various organic pollutants (Arabatzis et al., 2003). The metal-modified TiO2 could also be used for the inactivation of microorganisms. Sunada et al. (Sunada et al., 2003a) compared the photocatalytic disinfection of Escherichia coli in pure TiO2 and Cu-loaded TiO2 with the latter displaying an enhanced bactericidal activity. Silver modification of TiO2 has been shown to enhance the bactericidal activity of UV-irradiated TiO2 by about seventy folds as a result of both improved microorganism adsorption to the particle surface and suppression of eehþ recombination (Zhang et al., 2010). The formed ROSs was found to lead to perturbation of various cellular processes and eventually to bacterial death (Chen et al., 2009b). In the present study, Ag-modified mesoporous anatase TiO2 was synthesized via a photoreduction method. A non-

biodegradable dye, Rhodamine B (RhB), and a gram-negative bacterium, E. coli, were used as probes to investigate the photocatalytic detoxification and disinfection properties of the composite materials under UV irradiation. The effect of Ag loadings on catalyst performance and the role of Ag nanoparticles in the photocatalytic degradations were investigated.

2.

Materials and methods

2.1.

Materials

Titanium isopropoxide, silver nitrate, RhB, ethanol, and phosphate buffer saline of analytical grade were used without further purification. Pluronic surfactant poly(ethylene glycol)block-poly(propylene glycol)-block-poly(ethylene glycol) (EO20 PO70EO20, P123) was purchased from Aldrich. Deionized water was used in this work.

2.2.

Preparation and modification of mesoporous TiO2

The synthesis of mesoporous anatase TiO2 was similar to that reported previously (Yu et al., 2003). 6.0 g of P123 and 9.0 mL of titanium isopropoxide were dissolved in 60 mL of ethanol at ambient temperature. After the suspension was stirred for 1 h, 30 mL of H2O was added and stirred for another 1 h. After complete evaporation of ethanol and water, white solids were obtained. The solids were calcined at 350  C for 5 h in air with a heating rate of 0.5  C/min. The resultant product was denoted as m-TiO2. Modification of mesoporous TiO2 with Ag nanoparticles was conducted using a photoreduction method (Tran et al., 2006). 1.0 g of m-TiO2 was suspended in 300 mL of water, followed by sonication for 10 min to completely disperse the solid. Then, AgNO3 solution was added to the suspension. After stirring for 20 min in the dark, the mixture was irradiated with a 125 W high-pressure mercury lamp (Philips, Belgium) for 4.0 h. The resulting pale-gray solids were collected, washed with water and dried at ambient temperature. No Ag ion was detected in the filtrate indicating that all Ag had been completely loaded onto the m-TiO2. Different volumes of AgNO3 solution with a concentration of 6.27  102 M were used to vary the loading of Ag. The samples thus obtained are denoted as m-TiO2eAgx%, where x% indicates the weight percentage of Ag on the composite particles.

2.3.

Characterization

X-ray diffraction (XRD) patterns were collected on a Shimadzu XRD-6000 X-ray diffractometer using Cu Ka irradiation operated at 40 kV and 30 mA. The average crystallite size was estimated using the Scherrer equation based on the (101) reflection of anatase. Adsorptionedesorption isotherms of N2 were measured on a Micromeritics ASAP2020 system at 77 K. The BrunauereEmmetteTeller surface area (SBET) was calculated in the relative pressure range of P/P0 ¼ 0.05  0.2. The pore volume was estimated at P/P0 ¼ 0.97. Diffuse-reflectance spectra were recorded on a Shimadzu 3100 UVeVis-NIR spectrometer equipped with an integrating sphere ISR-3100 using BaSO4 as the reference. High-resolution transformation

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 9 5 e2 1 0 3

2097

electron microscope (HRTEM) images were obtained on a PerkineElmer JEM-2010F operated at 200 kV. X-ray photoelectron spectra (XPS) were collected on an AXIS HIS (Kratos Analytical Ltd., UK) instrument with an Al Ka irradiation source (1486.7 eV) operated at 15 kV and 10 mA. Photoluminescence spectra were collected at room temperature on a PTI luminescence spectrometer with a solid accessory.

2.4.

Photocatalytic degradation of RhB

Photocatalytic degradation of RhB was carried out in a semibatch swirl flow monolithic-type photoreactor (Zhou et al., 2006). Before UV irradiation, a suspension containing 300 mL of 8.6  103 mM RhB solution and 0.15 g of solid photocatalyst was sonicated for 5 min and stirred for 20 min in the dark to allow sorption equilibrium. The mixture was then irradiated with a 125 W high-pressure mercury lamp. At a given time interval of irradiation, 5 mL aliquots were withdrawn and filtered through a 0.45 mm membrane. The concentration of RhB in the filtrate was analyzed using a Shimadzu 1601 PC UVeVis spectrophotometer.

2.5.

Photocatalytic disinfection of E. coli

The antibacterial experiments were carried out in a sterilized 100 mL glass conical flask containing 50 mL 7.7  108 CFU/mL E. coli and 0.5 g/L solid photocatalyst under room temperature. The suspension was irradiated by two 8 W Black light Blue (BLB) lamps (200 mW/cm2). After a given time interval, 50 mL of the suspension was sampled, and diluted with sterile phosphate buffer saline (PBS, 0.03 mol/L, pH 7.2) to yield appropriate starting concentrations (ensuring cells were well suspended by vortex mixing before transfer). The suspensions were then spread on Petri Dishes which had been previously coated with nutrient agar. After being incubated at 37  C for 18 h, the formed colonies were counted and used to calculate the amount of active cells. Each data point was determined three times and was used as an average. Control experiments without the photocatalyst were also conducted for comparison.

2.6.

Stability of photocatalyst

Stability of the photocatalyst was evaluated as follows: 0.15 g of the photocatalyst m-TiO2eAg 0.25% was suspended in 300 mL of 8.0  103 mM RhB solution. After a given time interval of UV irradiation, 5 mL aliquots were withdrawn, filtered and analyzed on the UVeVis spectrophotometer. After RhB was completely degraded, the suspension was further irradiated for 20 min. Then, 15 mL of stock solution (0.16 mM) of RhB was supplied to restore the initial concentration of RhB (8.0  103 mM). The suspension was equilibrated again for 20 min in the dark before again being irradiated with UV light.

3.

Results and discussion

3.1.

Characterization

Fig. 1A shows the adsorptionedesorption isotherms of TiO2 and Ag-deposited TiO2 photocatalysts. It is seen that all samples

Fig. 1 e (A) Nitrogen adsorption isotherms and (B) pore size distribution curves of mesoporous TiO2 modified with Ag nanoparticles of loadings of (a) 0%, (b) 0.1%, (c) 0.25%, (d) 0.5%, (e) 1.0%, (f) 2.0%, and (g) 3.0%.

displayed a type IV isotherm with an H3 hysteresis loop, indicating that mesoporous materials are involved (Sing et al., 1985). The relatively narrow pore-size distribution curves indicate a single pore-size system (Fig. 1B). Table 1 summarizes the pore parameters of the samples. The BET surface area of m-TiO2 was calculated to be about 166 m2/g, higher than that reported previously (Yu et al., 2003). Modification with Ag gradually decreased the surface area with increasing loading of Ag - from 0.1% to 2.0%. Pore volume also decreased - from 0.43 to 0.33 cm3/g on increasing Ag loading from 0.1% to 2.0%. The above results indicate that Ag had been loaded onto the m-TiO2 sample. As the amount of silver increased, more Ag particles formed on the surface of mesoporous TiO2. The Ag particles covered some sorption sites and lowered the pore volume of the TiO2. However, they did not remarkably affect the pore structure (Fig. 1B) (Liu et al., 2008). For sample m-TiO2eAg 3.0%, an obvious decrease in pore size can be seen suggesting that the Ag particles blocked some of the pores. Fig. 2 shows the XRD patterns of the samples. All resolved peaks can be indexed as the (101), (004), (200) and (211) reflection of pure anatase (JCPDS No. 21-1272) (Chen et al., 2009a; Yu et al., 2003). The crystal sizes were calculated from the (101) plane and are given in Table 1. It can be seen that within the Ag loading range of 0.1%e3.0%, the crystal size of the mesoporous

2098

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 9 5 e2 1 0 3

Table 1 e Physical parameters of TiO2 and Ag-deposited TiO2.

m-TiO2 m-TiO2eAg m-TiO2eAg m-TiO2eAg m-TiO2eAg m-TiO2eAg m-TiO2eAg

0.1% 0.25% 0.5% 1.0% 2.0% 3.0%

Crystallite size (nm)a

SBET (m2/g)

Pore volume (cm3/g)b

Average pore size (nm)c

6.7 7.1 6.7 6.9 6.9 6.6 6.6

166 162 157 149 139 128 125

0.43 0.38 0.38 0.38 0.36 0.33 0.29

11.3 10.8 11.5 11.5 11.6 11.0 10.1

a Calculated using Scherrer equation. b Pore volume at P/P0 ¼ 0.97. c Estimated from the desorption branches using the BJH model.

anatase was not altered. No reflection peak owing to metallic Ag can be seen in Fig. 2, presumably due to the small Ag particle size. The TEM image (Fig. S1) confirmed that the size of the Ag particles in sample m-TiO2eAg 3.0% with the highest Ag loading was smaller than 1.5 nm. The size of the Ag particles in other samples was too small to be distinguished from the TiO2 background using the TEM technique. The Ag 3d XPS spectra of Ag-modified mesoporous TiO2 samples (Fig. S2) showed two peaks, at 368 and 374 eV, assigned to Ag 3d5/2 and 3d3/2, respectively (Guin et al., 2007). No peak due to AgD is seen. Considering that no Ag ion was detected in the filtrate after the photoreduction process, it thus can be concluded that all AgD ions added into the solution had been completely reduced to metallic Ag nanoparticles. The UVeVis diffuse reflectance spectra of all samples are shown in Fig. 3. The m-TiO2 samples exhibited strong absorption in the UV region with a sharp absorption edge located at about 400 nm (Zhang et al., 2009). Upon loading different amounts of Ag, no change in the absorption edge is seen. A typical surface plasmon absorption of Ag nanoparticles at around 400e550 nm gradually emerged and

Fig. 3 e UVeVisible diffuse reflectance spectra of mesoporous TiO2 modified with Ag nanoparticles of loadings ranging from 0 (bottom) to 3.0% (top). The label on top of each curve is the wavelength of the highest absorption.

increased with Ag loading, indicating the deposition of Ag nanoparticles on the TiO2 surface (Linsebigler et al., 1995). However, the maximum wavelength of the surface plasmon absorption varied from one sample to another. As Ag loading increased from 0.1% to 3.0%, the wavelength of the highest surface plasmon absorption gradually shifted from 468 nm to 498 nm, accompanied with the broadening of the absorption peaks. It is known that surface plasmon resonance (SPR) strongly depends on the size and uniformity of the nanoparticles (He et al., 2002) with the SPR peak becoming broader and shifting to longer wavelengths as the particle size increases, and/or when the particle size distribution becomes broad. The results in Fig. 3 suggest that as Ag loading gradually increased, more Ag nanoparticles aggregated on the TiO2 surface. Such an aggregation resulted in a decrease in the pore volume as evidenced by the data in Table 1.

3.2. Photocatalytic detoxification and disinfection under UV light irradiation

Fig. 2 e XRD patterns of mesoporous TiO2 before and after modification with Ag nanoparticles of different loadings. The standard XRD pattern of anatase is also included for reference.

The photocatalytic properties of the mesoporous anatase TiO2 and Ag-deposited TiO2 photocatalysts were evaluated by examination of RhB dye degradation and Gram-negative bacterium E. coli inactivation under UV irradiation. Before commencement of light irradiation, the mixtures of RhB or E. coli solution and solid photocatalyst were stirred for 20 min in the dark to achieve complete sorption equilibrium. As shown in Fig. 4A, minor removal of RhB occurred with catalyst m-TiO2eAg in the absence of UV irradiation. This removal was due to adsorption of the dye on the solid particle surface. In the absence of a photocatalyst but under UV irradiation, however, a slow degradation was observed. In the presence of m-TiO2, the concentration of RhB in suspension decreased rapidly with irradiation time. The data could be satisfactorily described by the first-order kinetic equation, ln (C/C0) ¼ kappt, where kapp is the apparent rate constant, C and

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 9 5 e2 1 0 3

Fig. 4 e Photocatalytic degradation of (A) RhB and (B) E. coli in different system. The initial concentration of RhB and E. coli were 8.6 3 10L3 mM and 7.7 3 108 CFU/mL, respectively.

et al., 2009; Sharma et al., 2009). When the suspension was irradiated with UV in the presence of m-TiO2eAg 3%, the inactivation of E. coli was greatly enhanced. Analyzing using first-order kinetics (Chen et al., 2009c), the total rate constants decreased in the order m-TiO2eAg 3% (0.102 min1) > m-TiO2 (0.055 min1) > m-TiO2eAg 3% (dark, 0.040 min1) > UV (0.024 min1) > m-TiO2 (dark, 0.0073 min1). The contribution of Ag inactivation to the total rate constant could be estimated to be about 0.033 min1.After eliminating the Ag inactivation effect, the photocatalytic disinfection effect can be assessed as exhibiting a net rate constant of k ¼ 0.062 min1, larger than that of m-TiO2. This trend is similar to that observed for the photocatalytic degradation of RhB. The time-course of inactivation of E. coli however is substantially more complicated than the degradation of RhB, possibly as a result of the combined effects of the antimicrobial activity of Ag itself and the photocatalytic disinfection effect of the TiO2eAg composite with the synergic effect of the two processes leading to a significant improvement of antibacterial activity (Liu et al., 2008). Silver is known for its antimicrobial properties with some investigators proposing that Ag nanoparticles may attach to the surface of the cell membrane and so disturb permeability and respiration functions of the cell (Kvitek et al., 2008). Notwithstanding this, the bactericidal mechanism of Ag nanoparticles remains to be fully understood (Sharma et al., 2009). The photocatalytic disinfection process has also been widely investigated. Chen and coworkers (Chen et al., 2009c) compared the photocatalytic degradation of formaldehyde and inactivation of E. coli with TiO2 and concluded that both processes depended on the ROS production rate. This is also apparent in the present system since the generation of ROSs is an important phenomenon in the TiO2eAg suspension. Disinfection performance of the photocatalyst can be evaluated by the analogy method based on dye degradation under UV irradiation since the latter is easier to measure quantitatively.

3.3. C0 are the concentrations of RhB at time t ¼ t and t ¼ 0, respectively. The rate constant of the photocatalytic degradation with m-TiO2 was calculated to be about 0.044 min1. When m-TiO2eAg 0.25% was used, the RhB degradation was enhanced exhibiting a rate constant of 0.069 min1, comparable to 0.073 min1 of P25. The absorption spectra of RhB were gradually decreased with irradiation time with no shift of the main absorption peak, suggesting complete cleavage of RhB chromophores. Although the UV energy is also adequate for RhB excitation, the photosensitized degradation, as evidenced by a concomitant wavelength shift of the absorption band to shorter wavelengths (Sung-Suh et al., 2004; Wu et al., 1998), can be neglected compared to the photocatalytic degradation. The antibacterial properties of the catalysts were studied using E. coli as a model strain. In the presence of UV irradiation and without a photocatalyst (Fig. 4B) or in the presence of m-TiO2 without UV light, the number of viable E. coli cells decreased slowly. With m-TiO2eAg 3% and without UV irradiation, the viable E. coli number decreased sharply in the first 60 min, then became stable. This behavior is different to RhB degradation and may be partially attributed to the intrinsic disinfection properties of the deposited Ag nanoparticles (Lv

2099

Effect of Ag loading

Deposition of Ag nanoparticles on TiO2 surface not only enhances photocatalytic reactions (Sung-Suh et al., 2004) but

Fig. 5 e Effect of Ag loading on the photocatalytic degradation of RhB (C) and E. coli (B), and on the photoluminescent intensity ratio between Ag-deposited TiO2 and TiO2 (¤).

2100

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 9 5 e2 1 0 3

Fig. 6 e Effect of Ag loading on the adsorption of RhB and the surface area using O2 as an adsorptive. The initial concentration of RhB was 8.6 3 10L3 mM.

also provides additional disinfection properties (Liu et al., 2008), however, both activities greatly depend on the amount of Ag loaded. As shown in Fig. 5, the photocatalytic degradation of RhB and disinfection of E. coli were changed by the loading of Ag nanoparticles with the rate constants of both processes higher at low loadings and lower at high loadings. The optimal loadings of Ag were different with about 0.25 wt% found to be optimal for RhB degradation and about 2.0 wt% optimal for E. coli disinfection. This is perhaps not surprising given that the photocatalyst itself exhibits bactericidal properties with the extent of inhibition (in the dark) increasing with increasing Ag loading while the degradation of RhB is strictly light dependent. Further discussion of the photocatalytically-mediated degradation of RhB is provided below. Fig. S3 shows the photoluminescence spectra of TiO2 with different Ag loadings. TiO2 has a strong luminescence spectrum with the luminescence intensity decreasing progressively on increasing the Ag loading from 0 to 2.0% (Fig. 5). This effect may be attributed to Schottky barrier formation at the TiO2emetal interface, indicating a strong inhibition of eehþ recombination as reported for other noble metals (Bannat et al., 2009; Sadeghi et al., 1996; Subramanian et al., 2001; Tada et al., 2009; Xin et al., 2005), eehþ recombination was decreased at the low loadings and enhanced at the high loadings with an optimal loading of Ag at about 0.10%. However, the optimal Ag loading for the

photocatalytic degradation of RhB (about 0.25 wt%) was much larger than that for maximal eehþ separation. The results suggest that the deposition of Ag nanoparticles on TiO2 might have other roles other than to simply depress eehþ recombination. Considering the deposited Ag particles covered the surface of TiO2, the adsorption affinity of the solid catalyst towards both dissolved oxygen and RhB would most likely be altered. As shown in Fig. 6, upon Ag deposition, RhB adsorption decreased with increasing Ag loading. Since RhB molecules mainly adsorbed on TiO2 surface with the diethylamino group (Wu et al., 1998) or the carboxyl group (Zhao et al., 2008), deposition of Ag particles on TiO2 would inevitably covered some sorption sites, resulting in a decrease of the RhB adsorption (Fig. 7). Control experiments (Fig. S4A) showed that the adsorption of RhB on m-TiO2eAg 0.25% surface followed a Langmuir model with the maximum absorption capacity at about 5.3  103 mmol g1. The degradation rate increased almost linearly with the extent of adsorption of RhB (Fig. S4B) and could be adequately described by the LangmuireHinshelwood (LeH) equation, r ¼ kLHKCe/(1 D KCe), yielding an LeH rate constant of kLH, ¼ 7.9  104 mM min1 and Langmuir adsorption constant of K ¼ 34.3 mM1. The decrease in RhB adsorption with Ag loading would inevitably be expected to lower the rate of photocatalytic degradation. Significantly, RhB degradation was much more rapid in an O2 equilibrated suspension than in an air saturated solution (Table S1) and was greatly inhibited in an atmosphere of pure N2. Since photocatalytic degradation greatly depends on the amount of ROSs, which were produced by electron transfer between excited TiO2 and the adsorbed O2 (Hoffmann et al., 1995; Sun and Xu, 2009), increasing the concentration of O2 presumably greatly increases the rate of production of ROSs, thus enhancing the degradation of RhB. O2 adsorptionedesorption isotherms obtained through solid-gas phase studies (Fig. S5) indicate that as Ag loading increased from 0 to 3.0%, the BET surface area gradually increased from 134 to 156 m2/g (Fig. 6) and the pore volume increased by about 11%. Apparently, the increased affinity of the catalyst to O2 could be attributed to the deposition of Ag nanoparticles, which increased the adsorption of oxygen on the photocatalyst surface (Fig. 7) (Xin et al., 2005). As such, two factors may be considered to account for the promotion of RhB degradation. First, electrons are likely to be transferred from the TiO2 surface to the Ag particles due to formation of the Schottky barrier at the interface of TiO2 and Ag (Fig. S3) with the adsorption of O2 on the

Fig. 7 e Schematic illustration of the effect of Ag on the adsorption of RhB and O2.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 9 5 e2 1 0 3

4.

2101

Conclusions

Mesoporous anatase TiO2 materials modified with Ag nanoparticles displayed significant capacity for the degradation of RhB and disinfection of E. coli under UV light irradiation. Both dye degradation and E. coli inactivation increased at low Ag loadings and decreased at high loadings with optimal Ag loadings of about 0.25 and 2.0 wt% for RhB degradation and E. coli disinfection, respectively. Besides formation of the Schottky barrier at the TiO2emetal interface, the deposited Ag nanoparticles slightly decreased the extent of adsorption of RhB, but increased the affinity of the surface to oxygen, which seemed to play a much more important role in enhancing the photocatalytic reactions. After four cycles of reuse, the composite catalyst still showed significant capacity for dye degradation. Fig. 8 e Reuse experiment for RhB degradation over m-TiO2eAg 0.25%. The hollow symbols represent the concentration of RhB before adsorption.

Ag surface facilitating the electron transfer process. Second, since the reduced O2 will transform to various ROSs, the enhancement of O2 adsorption would generate more ROSs in solution, leading to enhanced RhB degradation. Thus, the impact of the Ag particles on the photocatalytic activity can be summarized as follows: (1) the Ag particles created a Schottky barrier that depressed the eehþ recombination and promoted the utilization of excitons, (2) the Ag particles decreased the adsorption of RhB on the catalyst surface, resulting in a decline of the photoactivity, (3) the Ag particles increased the adsorption affinity of the catalyst towards dissolved oxygen, leading to an enhanced production of ROSs and the RhB degradation rate therefore. However, compared to the decrease in RhB adsorption with increase in Ag loadings, the incremental increase in extent of O2 adsorption appeared to play a much more important role in the photocatalytic reactions. Thus, the optimal Ag loading further increased to 0.25 wt% as compared to sample mTiO2eAg 0.10% with the lowest eehþ recombination.

3.4.

Stability of photocatalyst

The recyclability of the catalyst m-TiOeAg 0.25% is shown in Fig. 8. Since the photocatalytic inactivation of E. coli follows a similar trend as RhB degradation (Fig. 4), stability of the catalyst was evaluated with RhB degradation. It can be seen that the photocatalyst was recyclable for continuous adsorption and removal of RhB under UV irradiation, although the photoactivity and sorption capacity gradually decreased from one run to another. The results can be attributed to two reasons. First, any intermediates formed would be expected to compete for the adsorption sites and reactive oxygen species thereby diminishing oxidation effectiveness. Second, the catalyst concentration was decreased from one run to another because of the removal of 5 mL aliquots for analysis purposes. It is supported by the results shown in Fig. S6. It is seen that the rate constants of both RhB degradation and E. coli inactivation increased with the catalyst loading under the otherwise identical conditions to that of Fig. 4.

Acknowledgment This work was supported by Environment and Water Industry Development Council (EWI) of Singapore under project number MEWR 651/06/161.

Appendix. Supporting information. TEM images, Ag 3d XPS spectra, and PL spectra of m-TiO2eAg nanocomposite; Sorption isotherms of RhB and O2, and the effect of catalyst loading on the photoactivity.

Appendix. Supplementary data Supplementary data associated with this article can be found in the online version at doi: doi:10.1016/j.watres.2010.12.019.

references

Arabatzis, I.M., Stergiopoulos, T., Bernard, M.C., Labou, D., Neophytides, S.G., Falaras, P., 2003. Silver-modified titanium dioxide thin films for efficient photodegradation of methyl orange. Appl. Catal. B. 42, 187e201. Bannat, I., Wessels, K., Oekermann, T., Rathousky, J., Bahnemann, D., Wark, M., 2009. Improving the photocatalytic performance of mesoporous titania films by modification with gold nanostructures. Chem. Mater. 21, 1645e1653. Beyers, E., Cool, P., Vansant, E.F., 2005. Anatase formation during the synthesis of mesoporous titania and its photocatalytic effect. J. Phys. Chem. B. 109, 10081e10086. Chen, D.H., Huang, F.Z., Cheng, Y.B., Caruso, R.A., 2009a. Mesoporous anatase TiO2 beads with high surface areas and controllable pore sizes: a superior candidate for high-performance dye-sensitized solar cells. Adv. Mater. 21, 2206e2207. Chen, F.N., Yang, X.D., Wu, Q., 2009b. Photocatalytic oxidation of Escherischia coli, aspergillus niger, and formaldehyde under different ultraviolet irradiation conditions. Environ. Sci. Technol. 43, 4606e4611. Chen, F.N., Yang, X.D., Xu, F.F., Wu, Q., Zhang, Y.P., 2009c. Correlation of photocatalytic bactericidal effect and organic matter degradation of TiO2 part I: observation of phenomena. Environ. Sci. Technol. 43, 1180e1184.

2102

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 9 5 e2 1 0 3

Chen, X., Mao, S.S., 2007. Titanium dioxide nanomaterials: synthesis, properties, modifications, and applications. Chem. Rev. 107, 2891e2959. Epling, W.S., Peden, C.H.F., Henderson, M.A., Diebold, U., 1998. Evidence for oxygen adatoms on TiO2(110) resulting from O-2 dissociation at vacancy sites. Surf. Sci. 412e13, 333e343. Guin, D., Manorama, S.V., Latha, J.N.L., Singh, S., 2007. Photoreduction of silver on bare and colloidal TiO2 nanoparticles/nanotubes: synthesis, characterization, and tested for antibacterial outcome. J. Phys. Chem. C 111, 13393e13397. Han, F., Kambala, V.S.R., Srinivasan, M., Rajarathnam, D., Naidu, R., 2009. Tailored titanium dioxide photocatalysts for the degradation of organic dyes in wastewater treatment: a review. Appl. Catal. A 359, 25e40. He, J.H., Ichinose, I., Kunitake, T., Nakao, A., 2002. In situ synthesis of noble metal nanoparticles in ultrathin TiO2-gel films by a combination of ion-exchange and reduction processes. Langmuir 18, 10005e10010. Hidaka, H., Yamada, S., Suenaga, S., Zhao, J., Serpone, N., Pelizzetti, E., 1990. Photodegradation of surfactants. 6. Complete photocatalytic degradation of anionic, cationic and nonionic surfactants in aqueous semiconductor dispersions. J. Mol. Catal. 59, 279e290. Hoffmann, M.R., Martin, S.T., Choi, W.Y., Bahnemann, D.W., 1995. Environmental applications of semiconductor photocatalysis. Chem. Rev. 95, 69e96. Iliev, V., Tomova, D., Bilyarska, L., Eliyas, A., Petrov, L., 2006. Photocatalytic properties of TiO2 modified with platinum and silver nanoparticles in the degradation of oxalic acid in aqueous solution. Appl. Catal. B. 63, 266e271. Ismail, A.A., Bahnemann, D.W., Bannat, I., Wark, M., 2009. Gold nanoparticles on mesoporous interparticle networks of titanium dioxide nanocrystals for enhanced photonic efficiencies. J. Phys. Chem. C 113, 7429e7435. Kang, C., Jing, L., Guo, T., Cui, H., Zhou, J., Fu, H., 2008. Mesoporous SiO2-modified nanocrystalline TiO2 with high anatase thermal stability and large surface area as efficient photocatalyst. J. Phys. Chem. C 113, 1006e1013. Kowalska, E., Remita, H., Colbeau-Justin, C., Hupka, J., Belloni, J., 2008. Modification of titanium dioxide with platinum ions and clusters: application in photocatalysis. J. Phys. Chem. C 112, 1124e1131. Kvitek, L., Panacek, A., Soukupova, J., Kolar, M., Vecerova, R., Prucek, R., Holecova, M., Zboril, R., 2008. Effect of surfactants and polymers on stability and antibacterial activity of silver nanoparticles (NPs). J. Phys. Chem. C 112, 5825e5834. Li, H.X., Bian, Z.F., Zhu, J., Huo, Y.N., Li, H., Lu, Y.F., 2007. Mesoporous Au/TiO2 nanocomposites with enhanced photocatalytic activity. J. Am. Chem. Soc. 129, 4538e4539. Linsebigler, A.L., Lu, G.Q., Yates, J.T., 1995. Photocatalysis on TiO2 surfaces-principles, mechanisms, and selected results. Chem. Rev. 95, 735e758. Liu, Y., Wang, X.L., Yang, F., Yang, X.R., 2008. Excellent antimicrobial properties of mesoporous anatase TiO2 and Ag/ TiO2 composite films. Micropor. Mesopor. Mater. 114, 431e439. Lv, L., Lu, Y.Q., Ng, W.J., Zhao, X.S., 2009. Bactericidal activity of silver nanoparticles supported on microporous titanosilicate ETS-10. Micropor. Mesopor. Mater. 120, 304e309. Mogyorosi, K., Farkas, A., Dekany, I., Ilisz, I., Dombi, A., 2002. TiO2based photocatalytic degradation of 2-chlorophenol adsorbed on hydrophobic clay. Environ. Sci. Technol. 36, 3618e3624. Pan, J.H., Dou, H.Q., Xiong, Z.G., Xu, C., Ma, J.Z., Zhao, X.S., 2010. Porous photocatalysts for advanced water purifications. J. Mater. Chem. 20, 4512e4528. Pinna, N., Niederberger, M., 2008. Surfactant-free nonaqueous synthesis of metal oxide nanostructures. Angew. Chem. Int. Ed. Engl. 47, 5292e5304.

Raveendran, P., Eswaramoorthy, M., Bindu, U., Chatterjee, M., Hakuta, Y., Kawanami, H., Mizukami, F., 2008. Template-free formation of meso-structured anatase TiO2 with spherical morphology. J. Phys. Chem. C 112, 20007e20011. Sadeghi, M., Liu, W., Zhang, T.G., Stavropoulos, P., Levy, B., 1996. Role of photoinduced charge carrier separation distance in heterogeneous photocatalysis: oxidative degradation of CH3OH vapor in contact with Pt/TiO2 and cofumed TiO2eFe2O3. J. Phys. Chem. 100, 19466e19474. Schattka, J.H., Shchukin, D.G., Jia, J.G., Antonietti, M., Caruso, R.A., 2002. Photocatalytic activities of porous titania and titania/ zirconia structures formed by using a polymer gel templating. Chem. Mater. 14, 5103e5108. Sharma, V.K., Yngard, R.A., Lin, Y., 2009. Silver nanoparticles: green synthesis and their antimicrobial activities. Adv. Colloid Interfac. 145, 83e96. Sing, K.S.W., Everett, D.H., Haul, R.A.W., Moscou, L., Pierotti, R.A., Rouquerol, J., Siemieniewska, T., 1985. Reporting physisorption data for gas solid systems with special reference to the determination of surface-area and porosity (recommendations 1984). Pure Appl. Chem. 57, 603e619. Subramanian, V., Wolf, E., Kamat, P.V., 2001. Semiconductormetal composite nanostructures. To what extent do metal nanoparticles improve the photocatalytic activity of TiO2 films? J. Phys. Chem. B. 105, 11439e11446. Sun, Q., Xu, Y.M., 2009. Sensitization of TiO2 with aluminum phthalocyanine: factors influencing the efficiency for chlorophenol degradation in water under visible light. J. Phys. Chem. C 113, 12387e12394. Sunada, K., Watanabe, T., Hashimoto, K., 2003a. Bactericidal activity of copper-deposited TiO2 thin film under weak UV light illumination. Environ. Sci. Technol. 37, 4785e4789. Sunada, K., Watanabe, T., Hashimoto, K., 2003b. Studies on photokilling of bacteria on TiO2 thin film. J. Photochem. Photobiol. A 156, 227e233. Sung-Suh, H.M., Choi, J.R., Hah, H.J., Koo, S.M., Bae, Y.C., 2004. Comparison of Ag deposition effects on the photocatalytic activity of nanoparticulate TiO2 under visible and UV light irradiation. J. Photochem. Photobiol. A 163, 37e44. Tachikawa, T., Fujitsuka, M., Majima, T., 2007. Mechanistic insight into the TiO2 photocatalytic reactions: design of new photocatalysts. J. Phys. Chem. C 111, 5259e5275. Tada, H., Kiyonaga, T., Naya, S., 2009. Rational design and applications of highly efficient reaction systems photocatalyzed by noble metal nanoparticle-loaded titanium (IV) dioxide. Chem. Soc. Rev. 38, 1849e1858. Tian, B.Z., Yang, H.F., Liu, X.Y., Xie, S.H., Yu, C.Z., Fan, J., Tu, B., Zhao, D.Y., 2002. Fast preparation of highly ordered nonsiliceous mesoporous materials via mixed inorganic precursors. Chem. Commun., 1824e1825. Tran, H., Scott, J., Chiang, K., Amal, R., 2006. Clarifying the role of silver deposits on titania for the photocatalytic mineralisation of organic compounds. J. Photochem. Photobiol. A 183, 41e52. Wang, X., Mitchell, D.R.G., Prince, K., Atanacio, A.J., Caruso, R.A., 2008. Gold nanoparticle incorporation into porous titania networks using an agarose gel templating technique for photocatalytic applications. Chem. Mater. 20, 3917e3926. Wu, T.X., Liu, G.M., Zhao, J.C., Hidaka, H., Serpone, N., 1998. Photoassisted degradation of dye pollutants. V. Selfphotosensitized oxidative transformation of rhodamine B under visible light irradiation in aqueous TiO2 dispersions. J. Phys. Chem. B. 102, 5845e5851. Xin, B., Jing, L., Ren, Z., Wang, B., Fu, H., 2005. Effects of simultaneously doped and deposited Ag on the photocatalytic activity and surface states of TiO2. J. Phys. Chem. B. 109, 2805e2809.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 0 9 5 e2 1 0 3

Xu, Y., Lv, K., Xiong, Z., Leng, W., Du, W., Liu, D., Xue, X., 2007. Rate enhancement and rate inhibition of phenol degradation over irradiated anatase and rutile TiO2 on the addition of NaF: new insight into the mechanism. J. Phys. Chem. C 111, 19024e19032. Xu, Y.M., Langford, C.H., 1995. Enhanced photoactivity of a titanium(IV) oxide-supported on ZSM5 and Zeolite-A at lowcoverage. J. Phys. Chem. 99, 11501e11507. Yu, J.C., Zhang, L.Z., Zheng, Z., Zhao, J.C., 2003. Synthesis and characterization of phosphated mesoporous titanium dioxide with high photocatalytic activity. Chem. Mater. 15, 2280e2286. Zhang, D.Q., Li, G.S., Yu, J.C., 2010. Inorganic materials for photocatalytic water disinfection. J. Mater. Chem. 20, 4529e4536.

2103

Zhang, F.L., Zheng, Y.H., Cao, Y.N., Chen, C.Q., Zhan, Y.Y., Lin, X. Y., Zheng, Q., Wei, K.M., Zhu, J.F., 2009. Ordered mesoporous AgeTiO2-KIT-6 heterostructure: synthesis, characterization and photocatalysis. J. Mater. Chem. 19, 2771e2777. Zhao, D., Chen, C., Wang, Y., Ma, W., Zhao, J., Rajh, T., Zang, L., 2008. Enhanced photocatalytic degradation of dye pollutants under visible irradiation on Al(III)-modified TiO2: structure, interaction, and interfacial electron transfer. Environ. Sci. Technol. 42, 308e314. Zhao, D.Y., Jaroniec, M., Hsiao, B.S., 2010. Editorial for themed issue on “advanced materials in water treatments”. J. Mater. Chem. 20, 4476e4477. Zhou, J.K., Zhang, Y.X., Zhao, X.S., Ray, A.K., 2006. Photodegradation of benzoic acid over metal-doped TiO2. Ind. Eng. Chem. Res. 45, 3503e3511.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 1 0 4 e2 1 1 0

Available at www.sciencedirect.com

journal homepage: www.elsevier.com/locate/watres

Inactivation and surface interactions of MS-2 bacteriophage in a TiO2 photoelectrocatalytic reactor Min Cho, Ezra L. Cates, Jae-Hong Kim* School of Civil and Environmental Engineering, Georgia Institute of Technology, 200 Bobby Dodd Way, Atlanta, GA 30332-0373, USA

article info

abstract

Article history:

Inactivation of MS-2 bacteriophage in a TiO2 photoelectrocatalytic system was evaluated,

Received 21 July 2010

wherein TiO2 particles were coated onto an indium tin oxide (ITO) electrode and an electrical

Received in revised form

potential was applied under black light blue (BLB) irradiation. MS-2 phage inactivation was

8 December 2010

greatly enhanced by anodic potential, whereas cathodic potential completely inhibited

Accepted 20 December 2010

inactivation. Experiments performed with radical scavengers showed that inactivation was

Available online 25 December 2010

primarily caused by hydroxyl radicals, both in the bulk phase and on the TiO2 surface. Application of positive potential to the electrode was found to result in two distinct bene-

Keywords:

ficial effects: (i) electrostatic attraction between the negatively charged viral capsid and

TiO2

catalyst surface, causing improved usage of surface-bound hydroxyl radical, in comparison

MS-2 bacteriophage

to conventional TiO2 photocatalytic disinfection; and (ii) higher reactive oxygen species

Inactivation

production. Results also suggest that inactivation of various microorganisms including

Photoelectrocatalysis

Escherichia coli, Staphylococcus aureus, Klebsiella pneumoniae, Bacillus subtilis spores and

Photocatalysis

Cryptosporidium parvum oocyst was enhanced via positive potential induction to TiO2. ª 2010 Elsevier Ltd. All rights reserved.

Potential

1.

Introduction

Photocatalysis has been intensively studied since the 1970’s in the context of sustainable, alternative methods for waterborne pollutant degradation, with application of TiO2 comprising a large majority of past efforts (Choi and Hoffmann, 1995; Ireland et al., 1993; Li et al., 2008; Matsunaga et al., 1985; Prairie et al., 1993; Sunada et al., 1998). TiO2 is a semiconductor with a band gap of 3.2 eV in the anatase phase, allowing effective formation of conduction band electrons ðe cb Þ and valence band holes ðhþ vb Þ upon absorption of UV irradiation below 380 nm. The subsequent photo-induced heterogeneous reactions occurring at the TiO2 surface result in generation of various reactive oxygen species (ROS) such as OH, O2, and H2O2, some of which are effective for oxidative degradation of organic pollutants (Choi and Hoffmann, 1995; Sclafani et al., 1991). However, use of ROS for microbial inactivation has

received much less attention than pollutant degradation (Cho et al., 2004; Ireland et al., 1993), as practical application of photocatalytic disinfection has been hindered by relatively long inactivation times, compared to conventional chemical disinfection techniques such as chlorination. While some studies have focused on material engineering for improved quantum yields under visible light irradiation, the science behind the fundamental interactions involved in TiO2 photocatalytic disinfection is still evolving. Previous works have revealed several mechanistic nuances involving TiO2 photochemistry and microbial inactivation, wherein the biophysical structure of microorganisms influence the resultant interactions with the catalyst surface and ROS (Baram et al., 2009; Cho et al., 2004, 2005; Ireland et al., 1993; Sjogren and Sierka, 1994; Sunada et al., 1998). These studies indicate that Escherichia coli, for example, is inactivated to some extent by all forms of ROS produced by irradiated

* Corresponding author. Tel.: þ1 404 894 2216; fax: þ1 404 385 7087. E-mail address: [email protected] (J.-H. Kim). 0043-1354/$ e see front matter ª 2010 Elsevier Ltd. All rights reserved. doi:10.1016/j.watres.2010.12.017

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 1 0 4 e2 1 1 0

TiO2, while MS-2 bacteriophage demonstrates notably higher resistance to photocatalytic disinfection (Cho et al., 2005). This difference has been attributed to the simpler structure of MS-2 bacteriophage, such as greater integrity of the viral capsid and no reliance of viability on chemically fragile enzymes, compared to the bacterial metabolism and cell wall, thus permitting susceptibility only to OH and not to the less reactive O2 and H2O2 (Cho et al., 2005; Sjogren and Sierka, 1994). Furthermore, two different types of radical photoproducts exist, including free OH in bulk phase and surface OH on the TiO2 particles (Cho et al., 2005; Kim and Choi, 2002), which arise when chemisorbed water undergoes homolytic cleavage (Sakai et al., 2003). Electrostatic repulsion between the negatively charged catalyst and like-charged microbe surfaces is believed to encourage a spatial separation resulting in minimal utilization of surface OH for inactivation and a significant overall contribution to poor efficacy. Still of importance are steric and hydrophobic motifs on microbial surfaces which further confound interfacial forces and influence adhesion (Dickson and Koohmaraie, 1989; Penrod et al., 1996). Cho et al. (2005) thus found MS-2 inactivation kinetics to be unreliant on surface OH, while this radical species did in fact partially contribute to inactivation of E. coli (ATCC 8739). In the present work, we utilized a photoelectrocatalytic system to gain insight into the effects of surface interactions involving MS-2 bacteriophage, while demonstrating a possible strategy for improving photoreactor design. In the photoelectrocatalytic TiO2 reactor, the surface charge of the titania particles is artificially reversed by applying a positive bias. It has been reported that anodic potential evokes electrophoretic movement of microorganisms toward the electrode surface (Hong et al., 2008; Poortinga et al., 2000) and, in the case of TiO2, þ concomitant trapping of e cb =hvb pairs and decreased recombination rates (Mandelbaum et al., 1999; Wang et al., 2001). While electrophoretic movement through the bulk solution can be considered largely insignificant in a mechanically stirred system, static forces may, however, affect contact dynamics at the interface. We hypothesized that in a photoelectrocatalytic system, the electric attractive force and decreased recombination rates lead to increased exposure of microorganisms to surface OH and increased radical formation, respectively, ultimately resulting in improved inactivation kinetics. Prior demonstration of this technique was reported by Baram et al. (2009), but only with E. coli (CN13); thus, the goal of this work is to assess inactivation of MS-2 phage (indicator for human enteric virus) e which has presented particular challenges in TiO2 photocatalytic systems e with further comparisons to Cryptosporidium parvum oocysts (pathogenic protozoa), Bacillus subtilis pores (spore-forming gram-positive bacteria, indicator for protozoa and Bacillus anthracis), E. coli (indicator for gram-negative bacteria), Staphylococcus aureus (pathogenic gram-positive bacteria), and Klebsiella pneumonia (pathogenic gram-negative bacteria).

2.

2105

was washed with deionized water and sterilized by autoclaving at 121  C for 15 min. The pH was maintained at 7.1 by phosphate buffer (10e60 mM: KH2PO4 þ NaOH).

2.1.

Preparation and analysis of microorganisms

MS-2 bacteriophage (ATCC 15597) was cultured by inoculating E. coli C3000 incubated at 37  C in exponential to early stationary phase with MS-2 phage and quantified by the soft-agar overlay (double-agar layer) plaque assay method (Cho et al., 2005). MS-2 phage stock was prepared from overlay agar plates of confluent lysis as described elsewhere (Debartolomeis and Cabelli, 1991). E. coli host was grown and assayed in media containing 10 g/L tryptone, 1 g/L glucose, 1 g/L yeast extract, 8 g/L NaCl, 0.8 g/L CaCl2. The media for the top and bottom agars used in the plaque assays contained 7 g/L and 15 g/L agar, respectively. The soft-agar overlay was suspended by the addition of 150 mM phosphate buffered saline (pH 7.2) and centrifuged for 10 min at 3000  g. The supernatants, containing non-submerged MS-2, were then collected for experimentation. E. coli (ATCC 8739) was inoculated in Tryptic Soy Broth and harvested by centrifugation. Suspensions of B. subtilis spores (ATCC 6633) were prepared following the procedure by Nakayama et al. (1996), except for minor modifications such as using an 1/10 diluted nutrient agar and providing an extended incubation period of 5e7 days at 37  C (Cho et al., 2003). S. aureus (ATCC 29213) and K. pneumoniae (ATCC 700603), obtained from Centers for Disease Control and Prevention (CDC), were cultivated in Luria-Bertani broth and nutrient broth, respectively. The viabilities of these four bacteria (i.e. E. coli, B. subtilis, S. aureus and K pneumoniae) were assessed by a spread plate method using their specific agar following serial, 10-fold dilution of the samples. For B. subtilis spores, the diluted or treated spore suspensions were heated to 80  C for 15 min directly prior to each experiment to inactivate vegetative cells. C. parvum oocysts were purchased from Waterborne Co. (Lousiana, USA). The viable oocysts were washed before the experiments and the samples were analyzed using the in vitro excystation method as described previously (Jenkins et al., 1998).

2.2.

Preparation of electrode materials

To uniformly apply electrical current on the TiO2 coated surface, indium tin oxide (ITO), a widely used, transparent conductive material, was coated onto a 4  4 cm quartz plate. The 200 nm coating was obtained using radio frequency sputter (300 W) in ultrapure argon gas for 8 min TiO2 (P25, Degussa Co.) nano-particles were coated onto the ITO/quartz plates using PEG (polyethylene glycol: #20,000) as a binder (Kim and Kwak, 2009). Briefly, a viscous suspension of TiO2/ PEG was spread onto an ITO coated quartz plate and then calcined at 450  C for 30 min to remove the organic component. The coated plate was used as the working electrode for this study, with a stainless steel 316 counter electrode.

Materials and methods 2.3.

All solutions were prepared using ultrapure water (Milli-Q water purification system; Millipore Co., USA) and reagent grade chemicals (Aldrich Co. and Sigma Co., USA). Glassware

Experimental procedures

Disinfection experiments were conducted using the reactor portrayed in Fig. 1, with the working and the counter

2106

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 1 0 4 e2 1 1 0

0

Potentiostat

-1

Working electrode (Quartz plate with ITO coating)

Counter electrode (stainless steel plate)

TiO2 coating

Magnetic bar

Log (N/N0)

BLB Lamp

-2 Applied voltage (V) - 2.0 0.0 + 0.5 + 1.0 + 1.5 + 2.0

-3 Stirrer Fig. 1 e Photoelectrocatalytic TiO2 disinfection reactor configuration.

electrodes placed in parallel and separated by 1 cm. The top of the electrodes was connected to a potentiostat (PARSTAT 263, Princeton Applied Research Co.). Buffer suspension containing MS-2 phage (approximately 2  105 pfu/mL; pfu ¼ plaque forming unit) was placed in the reactor and intensively mixed via magnetic stirring. Three black-light-blue lamps (BLB; 4 W, Philips Co.) which emit in the 300e420 nm wavelength range (Cho et al., 2004) were positioned outside the reactor, as shown in Fig. 1, to minimize any direct microbial inactivation by UVA irradiation. UV intensity as 315 nm on the coated plate was controlled at 10 mW/cm2, measured using a UVX radiometer with UV 315 detector (UVP Co.). In control test, no inactivation of microorganisms was observed in the absence of either BLB light irradiation or TiO2 coating within the present experimental time scale, nor was there any inactivation by applying electrical current and stirring without irradiation. In general, 4 e 9 samples were collected during the experimental period to measure the degree of microbial inactivation. All disinfection studies were repeated three times and their averaged values with statistical deviations were used for data analysis. Selected experiments were performed with addition of tertbutanol (t-BuOH) and methanol (MeOH) as scavengers for bulk OH and surface OH, respectively, and superoxide dismutase (SOD) as a O2 scavenger (Cho et al., 2005; Kim and Choi, 2002). The concentrations of para-chlorobenzoic acid ( pCBA) and 3-bis(2-methoxy-4-nitro-5-sulfophehyl)-2H-tetrazolium5-carboxanilide inner salt (XTT sodium salt, Sigma Co.) were measured using high performance liquid chromatography (HPLC 1000, Agilent Co.) for quantification of produced OH and O2, respectively (Cho et al., 2004; Lee et al., 2007).

3.

Results and discussion

3.1. Enhanced MS-2 phage inactivation with positively charged TiO2 Photochemical inactivation kinetics of MS-2 phage in 20 mM phosphate buffer at pH 7.1 and 20  C with negative, positive, and zero potential applied to the TiO2 electrode are shown in

-4 0

20

40

60

80

100

120

Time (min) Fig. 2 e Effect of the potential applied to TiO2 electrode on inactivation of MS-2 phage (20 mM phosphate buffer at pH 7.1, 20 ± 1  C).

Fig. 2. With unaltered TiO2 electrode charge (i.e. no applied potential), MS-2 phage showed 1 log (90%) and 2 log (99%) inactivation after 71 and 109 min, respectively, consistent with the kinetics reported in literature (Cho et al., 2005). When negative (cathodic) potential of 2.0 V was applied, MS-2 phage inactivation was negligible within the time scale of experiments. In marked contrast, when positive (anodic) potential was applied up to þ2.0 V, inactivation was gradually enhanced. Fig. 3 shows the results of MS-2 phage inactivation experiments performed with various scavengers added to the reaction mixture during þ2.0 V conditions. Addition of excess SOD (30 mM) did not affect MS-2 phage inactivation, revealing a negligible role of O2 in phage inactivation. Addition of excess MeOH (30 mM), which effectively quenches both surface and free OH, completely inhibited MS-2 phage inactivation. In contrast, MS-2 phage inactivation was only partially inhibited by excess t-BuOH (a complete scavenger of free OH and partial scavenger of surface OH) and required 60% more time for a 1 log reduction. This result is different from previous observations made using an electrically unmodified TiO2 photocatalytic reactor operated under the similar conditions (Cho et al., 2005); the past study in fact demonstrated a negligible role of surface OH, as excess t-BuOH completely inhibited MS-2 phage inactivation. Zeta potential of MS-2 phage used in this study was measured at 55.4  5.75 mV in 20 mM phosphate buffer (Ionic strength; I ¼ 0.036 M) using a Zetasizer Nano zetapotentiometer (Malvern Instruments Co.). Therefore, electrostatic repulsive interactions will occur with a negatively charged electrode, since the zeta potential of TiO2 is also negative, even without application of a negative bias (19 mV at pH 7.1 (Sentein et al., 2009)). Furthermore, Penrod et al. (1996) suggested that the adsorption of MS-2 phage onto solid surfaces is

2107

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 1 0 4 e2 1 1 0

0

0

a -1

Ln ([pCBA]/[pCBA0])

Log (N/N0)

-1

-2

without Scavenger (Figure 2) SOD (30 mM) t-BuOH (30 mM) MeOH (30 mM)

-3

-4 0

20

40

60

kexp = 0.0133 s-1 -2 -1

kexp = 0.0152 s

-3 0.0 V + 2.0 V

-4

80

100

0

120

1

2

3

4

Time (min)

Time (min) Fig. 3 e The role of ROS on photoelectrocatalytic MS-2 phage inactivation (D2.0 V applied, 20 mM phosphate buffer at pH 7.1, 20 ± 1  C).

0.0

b

also hindered by steric repulsion originating from hydrophilic polypeptide loops on the outside of the viral capsid. This suggests that the MS-2 phage inactivation would occur primarily in the bulk phase if positive potential is not induced. When TiO2 electrode is positively charged, however, MS-2 phages are electrostatically attracted to the surface (Hong et al., 2008; Poortinga et al., 2000) such that surface OH is allowed to contribute to inactivation, which otherwise would be not efficiently used for MS-2 phage inactivation.

3.2.

Ln ([XTT]/[XTT0])

-0.1

kexp = 0.0088 min-1

-0.2

kexp = 0.0111 min-1

-0.3

-0.4 0.0 V + 2.0 V

Enhanced ROS production

Application of positive voltage to TiO2 electrodes has been shown to enhance ROS production (Mandelbaum et al., 1999; Wang et al., 2001). Results shown in Fig. 4 were obtained from the experiments performed without adding any microorganisms to the photoelectrocatalytic reactor. Instead, 1.92 mM of pCBA (OH probe, k(OH þ pCBA) ¼ 5  109 M1 s1 (Cho et al., 2004)) and 0.15 mM of XXT (O2 probe, k (O2 þ XTT) ¼ 8.6  0.8  104 M1 s1 (Sutherland and Learmonth, 1997)) were added to monitor the production of ROS as positive potential was applied to TiO2 electrode (Cho et al., 2009). Both the rates of pCBA and XXT degradation were increased by 14 and 26% with positive potential, compared to 0 V condition, which would correspond to increase of OH and O2 steady-state concentrations by 14 and 26%. This is consistent with past studies which suggested þ trapping of e cb =hvb pairs and decreased recombination rates when positive potential was applied to TiO2 (Mandelbaum et al., 1999; Wang et al., 2001). Note that the effect of inducing positive charge on inactivation is disproportionate in comparison to the relatively small increase in ROS production. Increase of 14% in OH concentration in the bulk phase, for

-0.5 0

10

20

30

40

Time (min) Fig. 4 e Effect of anodic potential on (a) OH measured using pCBA and (b) O2L measured using XXT (20 mM phosphate buffer at pH 7.1, 20 ± 1  C).

example, would contribute only to an approximate 14% increase in microbial inactivation rate (Cho and Yoon, 2008). Therefore, it is believed that the MS-2 phage inactivation was enhanced primarily by electro-attractive interactions at the interface in the photoelectrocatalytic reactor.

3.3.

Effect of current density

Table 1 summarizes the time required to achieve 1 and 2 log inactivation of MS-2 phage in varying buffer concentrations from 10 to 60 mM at different applied potentials from 0 to

2108

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 1 0 4 e2 1 1 0

Table 1 e Required time for 1 log or 2 log MS-2 phage inactivation. Buffer (mM) Voltage (V) I (mA/cm2) Required time

Buffer (mM) Voltage (V) I (mA/cm2) Required time

1 log 2 log

10 0.0 0.000 71 109

10 0.5 0.004 68 107

10 1.0 0.006 64 100

10 1.5 0.116 51 72

10 2.0 1.72 39 57

10 2.5 15 38 55

15 1.0 0.008 61 92

15 1.5 0.137 48 69

15 2.0 1.98 39 56

20 1.0 0.016 57 84

20 1.5 0.232 46 64

1 log 2 log

20 2.0 2.86 40 56

30 1.0 0.320 44 61

30 1.5 1.57 39 57

30 2.0 7.69 38 56

40 1.0 14.0 37 56

40 1.5 21.7 40 58

40 2.0 29.2 42 61

40 2.5 32.8 44 62

60 1.5 65.4 48 69

60 2.0 92.4 51 72

60 2.5 114 53 75

2.5 V. As the potential was increased at lower buffer concentration (up to 30 mM), the time required to achieve 1 or 2 log in activation of MS-2 phage decreased (i.e. inactivation was enhanced). In contrast, when the potential was increased at higher buffer concentration, the opposite trend was observed. Note that as the buffer concentration was increased, the current density across the electrode also increased, e.g., 1.72 mA/cm2 at 10 mM with þ2.0 V, to 92.4 mA/cm2 at 60 mM with the same applied potential. The data in Table 1 were plotted in Fig. 5 which presents the relationship between the required time to achieve 1 and 2 log inactivation versus current density. At current densities under approximately 10 mA/cm2, increasing positive field strength leads to increased current and lower inactivation time. Beyond this point, increasing voltage (hence current) no longer resulted in improved inactivation. This two-phase phenomenon can be explained by rising electro-osmotic forces under high current density, which has been previously elucidated in studies on electrophoretic movement of colloidal particles (Trau et al., 1996, 1997). Under high voltage, redox reactions produce an ionic concentration gradient at the interface that induces osmotic flow of water

molecules perpendicular to the electrode surface, eventually reaching a dynamic equilibrium. For electro-sorbed particles (or microorganisms), this fluid movement causes random migration parallel to the surface and increases detachment in the presence of shearing forces (Hong et al., 2008; Trau et al., 1997). Baram et al. (2009) have demonstrated a twophase inactivation/current density relationship in photoelectrocatalytic inactivation of E. coli; however, in the second phase that they observed, increasing current simply resulted in no further improvement of inactivation. Since E. coli is susceptible to surface and free radicals, we speculate that the electro-osmotic repulsion is overshadowed by higher ROS concentrations in general, causing increased voltage to eventually evoke no further response. On the other hand, we have demonstrated that MS-2 phage inactivation is primarily caused by surface OH under anodic potential; therefore, any detachment resulting from the combination of electroosmotic and stirring forces would show a decrease in inactivation once detachment is promoted, as opposed to leveling off, as shown in Fig. 5.

3.4. Photoelectrocatalytic inactivation of bacteria and protozoa

120

Time required for inactivation (min)

No current : 109 min for 2 log inactivation 100

No current : 71 min for 1 log inactivation

80

60

40 Phase II

Phase I 20 1 log inactivation 2 log inactivation 0 -3

-2

-1

0

1

2

Log ( I (mA/cm2)) Fig. 5 e Relationship between inactivation efficiency (as expressed in time required to achieve a target inactivation level) versus current density at TiO2 electrode.

Fig. 6 shows inactivation kinetics of other microorganisms with and without applied potential. Photochemical inactivation kinetics without external potential were consistent with previous findings, with C. parvum oocysts showing greatest resistance and gram-negative bacteria (E. coli and K. pneumoniae) being the most susceptible (Pal et al., 2007). As reported previously (Cho et al., 2005; Cho and Yoon, 2008), MS-2 phage was more resistant than E. coli but less resistant than B. subtilis spore or C. parvum oocyst. Upon induction of positive potential, inactivation kinetics were enhanced with all the microorganisms; but the extent of enhancement, as quantified by the decrease in the time required to achieve 1 log inactivation, varied depending on the microorganisms and was in the order of K. pneumoniae > E. coli > MS-2 > S. aureus > B. subtilis spore > C. parvum oocyst. The substantial decrease in inactivation time of the gram-negative bacteria is likely attributed to the greater susceptibility toward OH, surface OH in particular, in the presence of anodic potential. C. parvum oocysts have the least enhancement, as it is most resistant in general against OH attack (Cho et al., 2005; Cho and Yoon, 2008). Regardless, the results suggest that inducing the positive charge to the TiO2 electrode enhanced inactivation of various microorganisms in addition to MS-2 phage.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 1 0 4 e2 1 1 0

0

a S. aureus S. aureus (+ charged) K. pneumoniae K. pnuemoniae (+ charged) E. coli E. coli (+ charged)

Log (N/N0)

-1

-2

2109

have been shown to occur in photoelectrocatalytic degradation of charged chemical species (Kim and Anderson, 1994, 1996), thus the charge interactions between a photocatalyst and the target of radical attack are highly influential on the efficacy of the system.

Acknowledgements The study was supported by the Korea Research Foundation Grant funded by the Korean Government (KRF-2008-357D00142)

-3

references -4 0

60

120

180

240

Time (min) 0

b

Log (N/N0)

-1

-2

C. parvum C. parvum (+ charged) B. subtilis spore B. subtilis spore (+ charged)

-3

-4 0

120

240

360

480

600

720

Time (min) Fig. 6 e Effect of the positive surface potential on inactivation of various microorganisms: (a) S aureus, K. pneudomonia and E. coli and (b) C. parvum oocyst and B. subtilis spore (No or D 2.0 V applied, 20 mM phosphate buffer at pH 7.1, 20 ± 1  C).

3.5.

Significance of findings

Applying a positive potential to an immobilized TiO2 surface was found to alleviate mechanisms that would normally result in poor inactivation of MS-2 bacteriophage in a photocatalytic disinfection reactor. Application of positive potential up to þ2.0 V to the TiO2 electrode results in two distinct beneficial effects: (i) electrostatic attraction between the negatively charged viral capsid and catalyst surface, causing improved usage of surface-bound OH; and (ii) higher ROS production under applied anodic potential. Similar effects

Baram, N., Starosvetsky, D., Starosvetsky, J., Epshtein, M., Armon, R., Ein-Eli, Y., 2009. Enhanced inactivation of E. coli bacteria using immobilized porous TiO2 photoelectrocatalysis. Electrochimica Acta 54 (12), 3381e3386. Cho, M., Chung, H., Choi, W., Yoon, J., 2004. Linear correlation between inactivation of E. coli and OH radical concentration in TiO2 photocatalytic disinfection. Water Research 38 (4), 1069e1077. Cho, M., Chung, H.M., Choi, W.Y., Yoon, J.Y., 2005. Different inactivation behaviors of MS-2 phage and Escherichia coli in TiO2 photocatalytic disinfection. Applied and Environmental Microbiology 71 (1), 270e275. Cho, M., Chung, H.M., Yoon, J., 2003. Disinfection of water containing natural organic matter by using ozone-initiated radical reactions. Applied and Environmental Microbiology 69 (4), 2284e2291. Cho, M., Fortner, J.D., Hughes, J.B., Kim, J.H., 2009. Escherichia coli inactivation by water-soluble, ozonated C60 derivative: kinetics and mechanisms. Environmental Science & Technology 43 (19), 7410e7415. Cho, M., Yoon, J., 2008. Measurement of OH radical CT for inactivating Cryptosporidium parvum using photo/ferrioxalate and photo/TiO2 systems. Journal of Applied Microbiology 104 (3), 759e766. Choi, W.Y., Hoffmann, M.R., 1995. Photoreductive mechanism of CCl4 Degradation on TiO2 particles and effects of electron donors. Environmental Science & Technology 29 (6), 1646e1654. Debartolomeis, J., Cabelli, V.J., 1991. Evaluation of an Escherichia coli Host Strain for enumeration of F-male-specific Bacteriophages. Applied and Environmental Microbiology 57 (5), 1301e1305. Dickson, J.S., Koohmaraie, M., 1989. Cell-surface charge characteristics and their relationship to bacterial attachment to meat surfaces. Applied and Environmental Microbiology 55 (4), 832e836. Hong, S.H., Jeong, J., Shim, S., Kang, H., Kwon, S., Ahn, K.H., Yoon, J., 2008. Effect of electric currents on bacterial detachment and inactivation. Biotechnology and Bioengineering 100 (2), 379e386. Ireland, J.C., Klostermann, P., Rice, E.W., Clark, R.M., 1993. Inactivation of Escherichia coli by titanium dioxide photocatalytic oxidation. Applied and Environmental Microbiology 59 (5), 1668e1670. Jenkins, M.B., Bowman, D.D., Ghiorse, W.C., 1998. Inactivation of Cryptosporidium parvum oocysts by ammonia. Applied and Environmental Microbiology 64 (2), 784e788. Kim, D.H., Anderson, M.A., 1994. Photoelectrocatalytic degradation of formic acid using a porous TiO2 thin film

2110

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 1 0 4 e2 1 1 0

electrode. Environmental Science & Technology 28 (3), 479e483. Kim, D.H., Anderson, M.A., 1996. Solution factors affecting the photocatalytic and photoelectrocatalytic degradation of formic acid using supported TiO2 thin films. Journal of Photochemistry and Photobiology A Chemistry 94 (2e3), 221e229. Kim, D.S., Kwak, S.Y., 2009. Photocatalytic inactivation of E. coli with a mesoporous TiO2 coated film using the film adhesion method. Environmental Science & Technology 43 (1), 148e151. Kim, S., Choi, W., 2002. Kinetics and mechanisms of photocatalytic degradation of (CH3)nNHþ 4-n (0  n  4) in TiO2 suspension: the role of OH radicals. Environmental Science & Technology 36 (9), 2019e2025. Lee, J., Fortner, J.D., Hughes, J.B., Kim, J.H., 2007. Photochemical production of reactive oxygen species by C60 in the aqueous phase during UV irradiation. Environmental Science & Technology 41 (7), 2529e2535. Li, Q.L., Mahendra, S., Lyon, D.Y., Brunet, L., Liga, M.V., Li, D., Alvarez, P.J.J., 2008. Antimicrobial nanomaterials for water disinfection and microbial control: potential applications and implications. Water Research 42 (18), 4591e4602. Mandelbaum, P.A., Regazzoni, A.E., Blesa, M.A., Bilmes, S.A., 1999. Photo-electro-oxidation of alcohols on titanium dioxide thin film electrodes. Journal of Physical Chemistry B 103 (26), 5505e5511. Matsunaga, T., Tomoda, R., Nakajima, T., Wake, H., 1985. Photoelectrochemical Sterilization of microbial cells by semiconductor powders. Fems Microbiology Letters 29 (1e2), 211e214. Nakayama, A., Yano, Y., Kobayashi, S., Ishikawa, M., Sakai, K., 1996. Comparison of pressure resistances of spores of six Bacillus strains with their heat resistances. Applied and Environmental Microbiology 62 (10), 3897e3900. Pal, A., Pehkonen, S.O., Yu, L.E., Ray, M.B., 2007. Photocatalytic inactivation of gram-positive and gram-negative bacteria using fluorescent light. Journal of Photochemistry and Photobiology A Chemistry 186 (2e3), 335e341. Penrod, S.L., Olson, T.M., Grant, S.B., 1996. Deposition kinetics of two viruses in packed beds of quartz granular media. Langmuir 12 (23), 5576e5587.

Poortinga, A.T., Bos, R., Busscher, H.J., 2000. Controlled electrophoretic deposition of bacteria to surfaces for the design of biofilms. Biotechnology and Bioengineering 67 (1), 117e120. Prairie, M.R., Evans, L.R., Stange, B.M., Martinez, S.L., 1993. An Investigation of TiO2 photocatalysis for the treatment of water contaminated with metals and organic chemicals. Environmental Science & Technology 27 (9), 1776e1782. Sakai, N., Fujishima, A., Watanabe, T., Hashimoto, K., 2003. Quantitative evaluation of the photoinduced hydrophilic conversion properties of TiO2 thin film surfaces by the reciprocal of contact angle. Journal of Physical Chemistry B 107 (4), 1028e1035. Sclafani, A., Palmisano, L., Davi, E., 1991. Photocatalytic degradation of phenol in aqueous polycrystalline TiO2 Dispersions e the Influence of Fe3þ, Fe2þ and Agþ on the reaction rate. Journal of Photochemistry and Photobiology A Chemistry 56 (1), 113e123. Sentein, C., Guizard, B., Giraud, S., Ye, C., Tenegal, F., 2009. Dispersion and stability of TiO2 nanoparticles synthesized by laser pyrolysis in aqueous suspensions. Journal of Physics: Conference Series 012013. Sjogren, J.C., Sierka, R.A., 1994. Inactivation of Phage MS2 by iron-aided titanium dioxide photocatalysis. Applied and Environmental Microbiology 60 (1), 344e347. Sunada, K., Kikuchi, Y., Hashimoto, K., Fujishima, A., 1998. Bactericidal and detoxification effects of TiO2 thin film photocatalysts. Environmental Science & Technology 32 (5), 726e728. Sutherland, M.W., Learmonth, B.A., 1997. The tetrazolium dyes MTS and XTT provide new quantitative assays for superoxide and superoxide dismutase. Free Radical Research 27 (3), 283e289. Trau, M., Saville, D.A., Aksay, I.A., 1996. Field-induced layering of colloidal crystals. Science 272 (5262), 706e709. Trau, M., Saville, D.A., Aksay, I.A., 1997. Assembly of colloidal crystals at electrode interfaces. Langmuir 13 (24), 6375e6381. Wang, H.L., He, J.J., Boschloo, G., Lindstrom, H., Hagfeldt, A., Lindquist, S.E., 2001. Electrochemical investigation of traps in a nanostructured TiO2 film. Journal of Physical Chemistry B 105 (13), 2529e2533.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 1 1 1 e2 1 2 1

Available at www.sciencedirect.com

journal homepage: www.elsevier.com/locate/watres

Characterization of organic membrane foulants in a submerged membrane bioreactor with pre-ozonation using three-dimensional excitationeemission matrix fluorescence spectroscopy Ting Liu, Zhong-lin Chen*, Wen-zheng Yu, Shi-jie You State Key Laboratory of Urban Water Resources and Environments (SKLUWRE), School of Municipal and Environmental Engineering, Harbin Institute of Technology, No. 73 Huanghe Road, Nangang District, Harbin 150090, PR China

article info

abstract

Article history:

This study focuses on organic membrane foulants in a submerged membrane bioreactor (MBR)

Received 19 July 2010

process with pre-ozonation compared to an individual MBR using three-dimensional excita-

Received in revised form

tioneemission matrix (EEM) fluorescence spectroscopy. While the influent was continuously

25 October 2010

ozonated at a normal dosage, preferable organic matter removal was achieved in subsequent

Accepted 22 December 2010

MBR, and trans-membrane pressure increased at a much lower rate than that of the individual

Available online 1 January 2011

MBR. EEM fluorescence spectroscopy was employed to characterize the dissolved organic matter (DOM) samples, extracellular polymeric substance (EPS) samples and membrane fou-

Keywords:

lants. Four main peaks could be identified from the EEM fluorescence spectra of the DOM

Organic membrane foulants

samples in both MBRs. Two peaks were associated with the protein-like fluorophores, and the

Fluorescence spectroscopy

other ones were related to the humic-like fluorophores. The results indicated that pre-ozon-

Pre-ozonation

ation decreased fluorescence intensities of all peaks in the EEM spectra of influent DOM

Membrane bioreactor

especially for protein-like substances and caused red shifts of all fluorescence peaks to different extents. The peak intensities of the protein-like substances represented by Peak T1 and T2 in EPS spectra were obviously decreased as a result of pre-ozonation. Both external and internal fouling could be effectively mitigated by the pre-ozonation. The most primary component of external foulants was humic acid-like substance (Peak C ) in the MBR with preozonation and protein-like substance (Peak T1) in the individual MBR, respectively. The content decrease of protein-like substances and structural change of humic-like substances were observed in external foulants from EEM fluorescence spectra due to pre-ozonation. However, it could be seen that ozonation resulted in significant reduction of intensities but little location shift of all peaks in EEM fluorescence spectra of internal foulants. ª 2010 Elsevier Ltd. All rights reserved.

1.

Introduction

Low-pressure, hollow-fiber membrane filtration has been generally accepted as the most promising technology for surface water purification in recent years (Huang et al., 2007;

Gray et al., 2008). As the filtration process continues, a submerged membrane filtration system becomes a membrane bioreactor (MBR) system because of accumulation of microorganism and organic substances in raw water. Thus, a submerged MBR system, which combines membrane

* Corresponding author. Tel./fax: þ86 451 86283028. E-mail address: [email protected] (Z.-l. Chen). 0043-1354/$ e see front matter ª 2010 Elsevier Ltd. All rights reserved. doi:10.1016/j.watres.2010.12.023

2112

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 1 1 1 e2 1 2 1

rejection with microorganism biodegradation in a single tank, was introduced for treatment of drinking water by Li and Chu (2003). Afterward, much work has been done on the use of MBR or MBR-coupled technologies for drinking water treatment (Tian et al., 2008; Sagbo et al., 2008). Recently, an increasing attention has been paid to MBR process with pre-ozonation used for treating surface water. It was reported that MBR, especially containing powdered activated carbon, could be effectively used to remove total aldehydes, assimilable organic carbon (AOC) and biodegradable dissolved organic carbon (BDOC) from pre-ozonated water (Williams and Pirbazari, 2007; Treguer et al., 2010). Although the removal of contaminants can be improved by ozone oxidation, a better understanding of the influence of preozonation on membrane fouling process is still needed. The relative contribution of dissolved organic matter (DOM) to membrane fouling has been proved in the range of 26e52% in MBRs (Bouhabila et al., 2001; Lee et al., 2003). Moreover, extracellular polymeric substances (EPS), which are produced by bacteria, are reported as one of the factors causing membrane fouling in MBRs (Chang et al., 2001; Drews et al., 2006). Therefore, an insight into the impact of pre-ozonation on DOM and EPS is helpful to develop efficient strategies for membrane fouling control. Three-dimensional excitationeemission matrix (EEM) fluorescence spectroscopy has been successfully utilized to obtain the structural information of organic substances at  a relatively low concentration (Chen et al., 2003; Swietlik and Sikorska, 2004b; Gone et al., 2009). It can be used as a simple and sensitive technique to capture specific fluorescence features which correspond to humic- and protein-like materials in a single matrix in terms of fluorescence intensities (Hudson et al., 2007). Therefore, EEM fluorescence spectroscopy was employed to investigate the componential differences of DOM and EPS between the anoxic and oxic phases of an MBR process (Wang et al., 2009). EEM fluorescence spectroscopy was also applied to monitor the performance of pretreatment stages of membrane systems (Peiris et al., 2010b) and identify the oxidation-induced structural changes of DOM fractions of a filtered river water (Zhang et al., 2008). The great potential of EEM fluorescence spectroscopy is noticed for analysis of membrane foulants. Wang et al. (2009) found that the dominant fluorescence substances in gel layer (mainly caused by soluble microbial byproduct, colloids, solutes, etc.) on membrane surface of MBR were protein-like substances that might be due to the retention of proteins by the fine pores of the membrane. In addition, Kimura et al. (2009) demonstrated that EEM fluorescence spectra could be an effective analytical tool for the investigation of physically irreversible foulants in MBRs under different solid retention times. Peiris et al. (2010a) combined principal component analysis and fluorescence EEM measurements to characterize three membrane foulant fractions in the loosely attached foulants and chemically extracted foulants during UF of natural river water. According to a review by Meng et al. (2010), membrane fouling mainly results from the accumulation of retained substances on the membrane surface (i.e., external fouling) and the deposition of substances in membrane pores (i.e., internal fouling). Some studies reported that external fouling or foulant layer formation is the major cause of

membrane fouling in MBRs (Lee et al., 2001; Meng et al., 2007). On the other hand, the internal fouling or pore-blocking can lead to the formation of irreversible fouling, which is harmful for the long-term operation of MBRs (Meng et al., 2010). Thus, if external and internal foulants are both taken into consideration and analyzed using EEM fluorescence spectroscopy, it will be of great significance to give an insight into the fouling behavior in membrane-based water treatment processes. The aim of the present work is to characterize the organic membrane foulants in a submerged MBR with pre-ozonation compared to an individual MBR by EEM fluorescence spectroscope. The DOM and EPS samples, which are closely related with membrane fouling, were also analyzed by the EEM fluorescence technology. The external and internal foulants in both MBRs were identified and the comparison between them was conducted to contribute to a better understanding of membrane fouling in MBR processes.

2.

Materials and methods

2.1.

Experimental set-up

An individual MBR process without pre-ozonation (denoted as MBR-A) and an identical MBR process with pre-ozonation (denoted as MBR-B) were operated in parallel in this study. A schematic illustration of MBR-B is shown in Fig. 1. The hollowfiber UF membrane modules (Litree, China) were made of polyvinyl chloride (PVC) with a nominal pore size of 0.01 mm and a total surface area of 0.1 m2. The raw water was fed into a constant-level tank to manipulate the water head for the subsequent units. Ozone gas generated from an ozone generator (DHX-1, Jiujiu, China) was continuously bubbled into the water through a porous glass plate in an ozone contact column. A gas-phase ozone monitor was connected to a side stream from the generator to measure the ozone concentration. Preozonated feedwater was then supplied to MBR-B from a retention column for further reaction of residual ozone to prevent its impact on microorganism in MBR (Li et al., 2006). The effluent was collected directly from the membrane module by a suction pump, and a manometer was fixed between the membrane module and the suction pump to monitor the trans-membrane pressure (TMP). To supply the oxygen for microbial respiration and turbulence for membrane surface cleaning, continuous aeration was provided at the bottom of bioreactor. The experimental set-up of MBR-A was the same as that of MBR-B except for the absence of ozonation unit.

2.2.

Simulated raw water supply

The raw water was prepared in a way similar to that used by Tian et al. (2008). Domestic sewage was added to the local tap water (Harbin, China) of a volumetric ratio of 1:30 to simulate the surface water supply slightly polluted by sewage discharge. 1 mg/L of humic acid (Jufeng, Shanghai, China) was also added to the raw water. The synthesized raw water was then stabilized for 24 h at room temperature before use. During the experiment, the raw water was kept at a temperature in the range of 15.5e18.3  C and the pH in the range of 7.1e7.4; other water quality parameters are listed in Table 1.

2113

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 1 1 1 e2 1 2 1

water line ozone gas line air line

Fig. 1 e Schematic diagram of MBR-B (1-feed pump; 2-high level tank; 3-constant level tank; 4-ozone contact column; 5-retention column; 6-bioreactor; 7-membrane module; 8-manometer; 9-effluent pump; 10-backwash pump; 11-ozone generator; 12-gas phase ozone analyzer; 13-ozone gas flowmeter; 14-ozone destruction unit; 15-air blower; 16-air flowmeter; 17-air diffuser; 18-sludge discharge valve).

2.3.

Operating conditions

2.4.

MBRs with each effective volume of 1 L were conducted in a dead-end filtration mode at constant flux. Membrane flux was predetermined at a relatively low value of 10 L/(m2h), which corresponded to a hydraulic retention time (HRT) of 1.0 h. A 30-min operational cycle for suction followed by 1-min backwashing with the effluent was controlled by a timer. The ratio of air to influent was kept at 20:1 (V:V), and sludge retention time (SRT) was maintained at 40 d. The two MBRs fed with raw water had been operated stably for three months, so that the biomass was accumulated in the reactors for acclimation to the water. Then the mixed liquid of the two bioreactors was mixed and shared between them to have the same initial condition. The MBRs were operated continuously again with new membranes and pre-ozonation process was applied to MBR-B at ozone dosage of 1.5 mg/L-raw water. The HRT of the ozone contact column and the retention column was 15 min and 20 min respectively. The experiments were carried out under normal operating conditions (e.g., ozone dosage and reaction time) commonly adopted in water plants.

Extraction of EPS from the mixed liquid

EPS were extracted from the mixed liquid in the MBR according to the thermal treatment method (Chang and Lee, 1998). The mixed liquid was centrifuged for 30 min at 3200 rpm to remove the bulk solution. After the supernatant was discarded, the remaining pellet was washed and resuspended with saline water (0.9% NaCl solution). The mixed liquid was then subjected to heat treatment (100  C, 1 h) and centrifuged again under the same operating conditions. The centrifuged supernatant was EPS solution, which was filtered through a 0.45 mm acetate fiber membrane and used for EEM fluorescence analysis.

2.5.

Extraction of foulants

At the end of operation, the fouled membrane modules were taken out from the reactors when the TMP exceeded 35 kPa. The external foulants on membrane surface were carefully scraped off with a plastic sheet (Deli, China) and simultaneously flushed with deionized (DI) water. The collected

Table 1 e Pollutants removal efficiencies of MBR-A and MBR-B. Water quality indexs

Raw water

Turbidity (NTU) CODMn (mg/L) DOC (mg/L) UV254 (cm1)

2.31 4.25 6.020 0.077

   

1.00 0.27 0.784 0.003

After pre-ozonation

2.09 3.15 5.497 0.046

   

0.60 0.22 0.572 0.002

MBR-A Effluent 0.07  2.30  4.336  0.059 

0.01 0.28 0.603 0.002

MBR-B

Total removal (%) 97.0  45.9  28.0  23.4 

1.3 3.7 6.3 2.8

Effluent 0.07 1.60 3.542 0.036

   

0.01 0.20 0.568 0.002

Total removal(%) 97.0 62.4 41.2 53.2

   

1.2 3.3 6.2 3.0

2114

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 1 1 1 e2 1 2 1

sample was fully mixed using a magnetic blender (JB-2, Leici, China) at 200 rpm for 1 h, it was then filtered through a 0.45 mm acetate fiber membrane for EEM fluorescence analysis (Wang et al., 2009). After the membrane surface was wiped with a sponge, 0.01 mol/L NaOH was used for extraction of internal foulants and the fibers were soaked for 24 h at 20  C in the alkaline solution according to the method described by Kimura et al. (2009).

2.6.

Analytical methods

2.6.1.

Water quality analysis

Water quality analysis was conducted following the standard methods (APHA, 1998). Turbidity was monitored by a turbidimeter (Turbo550, WTW, Germany). The dissolved organic carbon (DOC) was measured by a total organic carbon (TOC) analyzer (TOC-VCPH, Shimadzu, Japan). CODMn was analyzed by the potassium permanganate oxidation methods. UV absorbance at 254 nm (UV254) was determined by using a spectrometer (T6, Puxi, China).

2.6.2. Three-dimensional excitationeemission matrix (EEM) fluorescence spectroscopy Fluorescence measurements were conducted using a spectrofluorometer (FP-6500, Jasco, Japan) equipped with a 150 W xenon lamp at ambient temperature of 24  C. A 1-cm quartz cuvette with four optical windows was used for the analyses. Emission scans were performed from 220 to 550 nm at 5 nm steps, with excitation wavelengths from 220 to 450 nm at 5 nm intervals. The detector was set to high sensitivity, and the scanning speed was maintained at 2000 nm/min in this study; the slit widths for excitation and emission were 5 nm and 3 nm respectively. Under the same conditions, fluorescence spectra for Milli-Q water were subtracted from all the spectra to eliminate water Raman scattering and to reduce other background noise. During the course of fluorescence analysis, the Raman scattering peak intensity for Milli-Q water (excitation at 350 nm, emission at 400 nm) was recorded as a standard to verify the instrument stability. Mean intensity of the Raman peak was 36.10 units and the differences were less than 2%, confirming that there were no significant fluctuations in the performance of the spectrofluorometer throughout the experimental period. The EEM spectra are plotted as the elliptical shape of contours. The X-axis represents the emission spectra while the Y-axis represents the excitation wavelength, and the third dimension, i.e., the contour line, is given to express the fluorescence intensity.

was ascribed to pre-ozonation implementation in MBR-B. Especially, an approximately 30% higher UV254 decrease was achieved in MBR-B (53.2  3.0%) compared to that in MBR-A. Three functions were identified for the contributions to removal of organic contaminants in MBR-B: partial degradation or complete mineralization by ozone oxidation, physical retention by UF membrane, and biodegradation by active biomass within the reactor. It should be noted that the organic matter removal was attributed to the synergetic effect of these three functions. It can be seen from Fig. 2 that TMP increased with operation time while the membrane flux was maintained constant at about 10 L/(m2h) during the experiment before TMP exceeded 30 kpa. As a two-step fouling phenomenon, the TMP variations exhibited a slow increase followed by a rapid increase. The TMP gradually increased with time from the initial 5 kPa for both systems with a similar trend within the beginning phase (0e8 days). However, there was a distinct gap of 1.5 kPa between them on Day 16. The permeate flux obviously declined when the TMP of MBR-A reached 35 kPa, and this operation process came to an end for a further analysis of membrane foulants. The final TMP of MBR-B only increased to 23.5 kPa, which was 12.5 kPa lower than that of MBR-A (36 kPa). It was thus believed that ozone pre-oxidation was an effective pre-treatment strategy to reduce the increasing rate of TMP to lower energy requirement for the membrane filtration process at a constant flux. In order to identify the proportions of external and internal fouling resistances, the membranes were taken out from the reactors at the end of experiment and the external foulants were removed from membrane surface. The membrane modules were reinstalled into the bioreactors and internal fouling resistances were evaluated. The TMP in MBR-A and MBR-B were then decreased to 12 kPa and 9 kPa respectively. Compared to those of 77.4% and 22.6% in MBR-A, the contributions of external and internal fouling resistances to TMP development were 78.4% and 21.6% in MBR-B. Therefore, it can be concluded that the proportions of external and internal fouling resistances were similar in both MBRs, suggesting that ozone pre-oxidation was able to alleviate both of the two kinds of membrane fouling.

12

50 45

10

40

3.1.

Results and discussion Process performance

As shown in Table 1, the turbidity was reduced from 2.31  1.00 NTU to a level as low as 0.07  0.01 NTU for the two MBRs, which indicated an excellent performance of particle removal. The organic matter removal efficiencies in terms of CODMn, DOC and UV254 in both MBR-A and MBR-B processes are summarized in Table 1. The results showed that the remarkably improved performance in organic matter removal

TMP (MBR-A) TMP (MBR-B) Flux (MBR-A) Flux (MBR-B)

30 25

8

6

20 4

15 10

Flux ( L/( m2h) )

3.

TMP (kpa)

35

2

5 0

0 0

10

20

30

40

50

Time (day) Fig. 2 e Comparison of TMP and membrane flux variations in MBR-A and MBR-B.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 1 1 1 e2 1 2 1

3.2.

EEM fluorescence spectra of DOM in the two systems

Three-dimensional EEM fluorescence spectroscopy has been successfully utilized to identify the chemical composition of DOM because of its ability to distinguish among certain classes of organic matter in natural waters (Saadi et al., 2006). In UF application, the major membrane foulant is natural organic matter, which contains a complex mixture of humic and fulvic acids, proteins, carbohydrates of various molecular size and functional groups (Her et al., 2003; Saravia et al., 2006). There are five key fluorescence peaks referred to as fluorophores A, C, T1, T2 and B commonly observed in freshwater samples (Coble, 1996; Baker, 2001). Peak A and C are related to humic-like substance derived from the breakdown of plant material (Lee et al., 2008); protein-like fluorophores including tryptophan-like (Peak T ) and tyrosine-like (Peak B) materials, are usually detected at enhanced levels in water impacted by domestic sewage (Baker et al., 2003). As shown in Fig. 3, Peak B has relatively lower fluorescence intensity for the DOM samples, so the other four peaks including Peak A, C, T1 and T2 which are distinctly identified were investigated in this section. The fluorescence parameters including peak locations, fluorescence intensity, and different peak intensity

2115

ratios were extracted from EEM fluorescence spectra and summarized in Table 2, which could be employed for quantitative analysis. Generally, intensity reduction of the fluorescence peak between raw water and treated water is an indication for degradation of fluorescing material. It can be seen from Table 2 that ozonation approximately decreased the fluorescence intensities of Peak A and C by 30e40% and those of Peak T1 and T2 by 60e70% for DOM in raw water. Consequently, the fluorescence intensities of Peak A and C in EEM spectra of MBR-B effluent were nearly the same percentage lower than those of MBR-A effluent. Meanwhile, the intensities of Peak T1 and T2 in EEM spectra of MBR-B effluent were 40% and 53% lower than those of MBR-A effluent. The peak intensity ratios are shown as ratios to Peak C, as this component is considered to be present in a wide range of water environments (Henderson et al., 2009). Presented in this way, the data reflect the differences in composition rather than the considerable differences in concentration. Peak T1 and C in EEM spectra of DOM samples, which indicate protein- and humic acid-like substance respectively, can be referred to as biodegradable and nonbiodegradable DOM (Reynolds, 2002; Wang et al., 2009). Since the feedwater synthesized to simulate the

Fig. 3 e EEM fluorescence spectra of (a) the influent (raw water) and (b) the effluent DOM of MBR-A, (c) the influent (after preozonation) and (d) the effluent DOM of MBR-B.

2116

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 1 1 1 e2 1 2 1

Table 2 e Fluorescence spectral identifications of DOM samples in MBR-A and MBR-B. Process

MBR-A MBR-B

Samples

Influent (Raw) Effluent Influent (Ozonated) Effluent

Peak A

Peak C

Peak T1

Peak T2

Peak Int. ratio

Ex/Em

Int.

Ex/Em

Int.

Ex/Em

Int.

Ex/Em

Int.

A/C

T1/C

C/T2

240/405 245/420 250/420 255/425

311.7 304.5 230.5 213.7

310/420 315/415 325/415 330/425

298.3 199.0 170.7 154.3

280/340 270/340 280/345 275/345

283.2 119.5 115.0 71.2

225/335 230/345 230/340 230/345

535.7 206.9 154.2 96.5

1.04 1.53 1.35 1.38

0.95 0.60 0.67 0.46

0.56 0.96 1.11 1.60

Int.: intensity.

surface water polluted by sewage discharge had a relatively higher biodegradability, some biodegradable DOM was mineralized during the ozonation process. The intensity ratio of Peak T1/Peak C decreased in both bioreactors as shown in Table 2, suggesting that the biodegradable DOM with fluorescence was gradually metabolized by microorganism. The intensity ratio of Peak T1/Peak C of MBR-B effluent was 0.46 compared to 0.60 of MBR-A effluent. MBR-B process thus displayed a greater capacity for biodegradable DOM removal and this is beneficial to biological stability of treated water and restraint of bacterial regrowth in distribution system. The differences in peak intensity ratios of EEM fluorescence spectra in the two MBR systems imply that ozone oxidation is responsible for the compositional variation of the fluorescent compounds in DOM samples. The location shift of fluorescence peak provides spectral information on the chemical structure changes of DOM samples. After ozonation, the locations of the four fluorescence peaks shifted toward longer wavelength (red shift) to different extents on the emission and/or excitation scale, and this observation is in line with Chen et al. (2002). As reported  by Swietlik et al. (2004a), the ozonation of hydrophobic acid (HOA) and hydrophilic neutral (HIN) produces carboxylic acids, aldehydes and ketones, and this may cause the formation of oxidation byproducts with double bond-containing substituents. A red shift is related to the increase of carbonyl, hydroxyl, alkoxyl, amino, and carboxyl groups in the structures of fluorophores (Chen et al., 2002; Uyguner and Bekbolet, 2005), while a blue shift is ascribed to the elimination of particular functional groups such as carbonyl, hydroxyl and amine, a reduction in the degree of p-electron systems, and the decrease in the number of aromatic rings and conju gated bonds in a chain structure (Swietlik et al., 2004a). In the two systems, the locations of Peak A and T2 in EEM spectra of effluent DOM were all red-shifted (5e15 nm) to longer wavelengths than those of influent DOM, while Peak T1 of effluent DOM showed a blue shift (5e10 nm) on the excitation scale compared to that of influent DOM. The location of Peak C showed different shift trends in the two systems with respect to the emission axis. Wang et al. (2009) observed that Peak T1 and T2 respectively demonstrated a blue and red shift of the effluent DOM compared to those of the influent DOM in a MBR for wastewater treatment, which is in good agreement with the results of both MBRs in this study. Furthermore, they described that the location of Peak C of the effluent DOM was red-shifted along the excitation axis and blue-shifted along the emission axis. This finding agrees with the experimental results of MBR-A but disagrees with those of MBR-B obtained during this study. It was therefore assumed that this

difference between the two MBRs was a consequence of structural changes of the humic-like substances in the raw water during ozonation process. As a matter of fact, it is likely that some of the EEM peak shifts result from the changes in concentration of one of the several overlapping components (Stedmon et al., 2003; Peiris et al., 2010a). For the purpose of identifying structural changes, the EEM fluorescence intensities were normalized with respect to the highest peak intensity. Fig. 4 shows the normalized intensities of EEM fluorescence spectra of DOM in raw water and pre-ozonated water on the excitation scale (emission at 420 nm). When the intensity of Peak C reduced significantly due to the ozonation process, the visual location of Peak A would have a blue shift to a lesser degree, which was dependent on concentration changes. Nevertheless, Peak A showed a red shift of 10 nm on the contrary, which means that the peak shift was necessarily caused by the structural changes of this kind of humic-like substances (Peak A). The peak locations in EEM spectra of the effluent DOM of MBR-B also demonstrated some differences in comparison with those of the effluent DOM of MBR-A.

3.3. EEM fluorescence spectra of EPS in mixed liquid of the two MBRs As shown in Fig. 5, the intensity of Peak B in the EEM spectra of EPS was significantly enhanced in both MBRs, suggesting that the organic substances indicated by Peak B were closely related with metabolic activity of microorganism. A new peak,

Fig. 4 e Normalized intensities of EEM fluorescence spectra of DOM in raw water and pre-ozonated water on the excitation scale (emission at 420 nm).

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 1 1 1 e2 1 2 1

a

b

extensively present in EPS samples extracted from various origins. The locations of the two peaks in EPS spectra in this study, which located at the Ex/Em of 280/350 nm and 340/435 nm, were similar to those reported by Sheng and Yu (2006) but different from those observed by Wang et al. (2009). The differences might be attributed to the fact that the EPS samples were extracted from different origins and thus the components in EPS were chemically different. The intensity of Peak B did not show any decline in EPS spectra of MBR-B (Fig. 5b) due to pre-ozonation compared to that of MBR-A (Fig. 5a). It may indicate that the protein-like substances represented by Peak B were excreted by microorganism in MBRs, and pre-ozonation had little effect on the metabolic level related to this compound. The peak intensities of protein-like substances represented by Peak T1 and T2 in EPS spectra of MBR-B were decreased significantly resulted from pre-ozonation. Cho et al. (2005) established a functional equation in which the specific cake resistance was proportional to the EPS concentration. Ahmed et al. (2007) also observed that as EPS concentration rose, the specific cake resistance increased, and this consequently resulted in the rise of TMP. These investigations showed that there is a close relationship between EPS and the resistance of cake layer on membrane surface. It can therefore be concluded that the protein-like substances represented by Peak T1 and T2 in EPS might contribute more to external fouling.

3.4.

Fig. 5 e EEM fluorescence spectra of EPS extracted from (a) MBR-A and (b) MBR-B.

i.e., Peak D, was present at the excitation/emission wavelengths (Ex/Em) of 270e280/300e310 nm in EEM spectra of EPS. According to the five regions of EEM divided by Chen et al. (2003), Peak D indicates soluble microbial byproduct (SMP)like substances (Region IV). It could be seen that protein- and SMP-like substances are dominant among fluorescent organic matters in EPS from both MBRs. Compared with the EEM spectra of DOM samples (Fig. 3), intensity of Peak T2 in EEM spectra of EPS was much weaker than that of raw water, which means that the fluorescent DOM represented by Peak T2 in effluents were originated from raw water rather than EPS. The locations of Peak A and C in EEM spectra of EPS were both red-shifted by 15e35 nm along the two axes compared to those in Table 2, indicating that the structure and components of humic-like substances in EPS were different from those in DOM samples. Sheng and Yu (2006) found three fluorescence peaks including Peak T1, T2 and C were present in EPS spectra from a conventional activated sludge system. Wang et al. (2010) identified Peak T1, C, together with a new peak associated with humic acid-like substances at the Ex/Em of 415e420/ 470e475 nm in EPS spectra of an MBR. In this study, three main peaks and three lesser peaks were observed in EPS spectra. The organic matters indicated by Peak T1 and C are

2117

EEM fluorescence spectra of membrane foulants

The amount and composition of organic membrane foulants were related to the interaction between organic substance and membrane. The UF membrane used in this study is made of Polyvinyl chloride (PVC), which is a hydrophobic material. According to the manufacturer, doping technology is used to improve its hydrophilicity for higher flux and some other physical properties. The contact angle reflects the hydrophobic/hydrophilic character of membranes. The PVC membrane has an average contact angle of 68  2 (provided by the manufacturer). On the one hand, ozone oxidation is able to increase the polarity and hydrophilicity of the substances to make hydrophobic membranes less susceptible to adsorptive fouling. In this study, membrane fouling was reduced by pre-ozonation, which means that the PVC membrane might still exhibit hydrophobic property. On the other hand, ozonation reduced the amount of humic substances because of their breakdown to lower molecular weight (MW) compounds. Therefore, although the size of some DOM molecules before or after pre-ozonation was smaller than the nominal pore size of PVC membrane (0.01 mm), biodegradation of these low-MW ozonation products caused less accumulation of foulants on membrane surface and/or in membrane pores. Unlike conventional methods such as the ratio of carbohydrate to protein (C/P) which is incompetent to fully characterize membrane foulants, EEM fluorescence analysis is able to provide more useful information on the characteristics of organic membrane foulants (Kimura et al., 2009). Hence, both external and internal foulants were extracted from the fouled membranes and analyzed by using EEM fluorescence spectroscopy at the end of operation.

2118

3.4.1.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 1 1 1 e2 1 2 1

External foulants

Fig. 6 shows EEM fluorescence spectra measured for external foulants extracted from the fouled membranes of the two MBRs. Peak T1 predominated the EEM fluorescence spectra of external foulants in MBR-A, while Peak C was dominant in the EEM fluorescence spectra of external foulants in MBR-B. It was demonstrated that the external foulants were composed of the protein-like substances represented by Peak T1 and the humic acid-like substances represented by Peak C as the most primary components in MBR-A and MBR-B, respectively. The comparison in TMP development of both MBRs showed that membrane fouling of MBR-A was more serious than that of MBR-B. As stated in “Section 3.1”, the contribution of external fouling resistance to TMP development reached 75e80% in the two systems. Therefore, it can be concluded that protein-like substances rather than humic acid-like substances contributed more to the external fouling resistance. This is consistent with the findings of Hong et al. (2007) and Drews et al. (2007), who reported that proteins could induce severe membrane fouling as one of the major components in membrane foulants. The appearance of dominant protein-like peak in external foulants in MBR-A indicated the accumulation of protein-like substances on the membrane surface, whereas the intensity of protein-like peak was weakened to a significant degree in MBR-B which could be attributed to ozonation.

a

b

Thus, it was reasonable to infer that the reduced accumulation of protein-like substances into gel layer may play an important role against TMP increase in MBR-B at constant flux. The locations of peak T1 (Ex/Em ¼ 280/345) and peak T2 (Ex/Em ¼ 230/335) in the EEM fluorescence spectra of external foulants from MBR-B were similar to those of external foulants from MBR-A as shown in Fig. 6. However, the location of Peak A in the EEM fluorescence spectra of external foulants of MBR-A was red-shifted by 10 nm along the excitation axis and blue-shifted by as much as 25 nm along the emission axis compared to that of external foulants of MBR-B. The location of Peak C of MBR-A external foulants was blue-shifted to shorter wavelengths than that of MBR-B external foulants. These observations implied that the structures of humic-like substances represented by Peak A and C in the external foulants of the two MBRs differed from each other. In conclusion, the content decrease of protein-like substances and the structural change of humic-like substances were observed in external foulants from EEM fluorescence spectra due to pre-ozonation. The study carried out by Schlichter et al. (2003) indicated that continuous addition of ozone caused a drastic reduction in adsorptioninduced membrane fouling during the UF of humic acid solution. Nevertheless, Her et al. (2007) reported that nanofiltration (NF) membrane fouling increased mainly due to the adhesive EPS released by algae upon ozonation. It may actually be attributed to the average size of NF membrane pores which is smaller than that of UF membrane pores. Moreover, it may also be attributed to the presence of abundant algae in the raw water used for their study in that season, which indicated the unsuitability for application of pre-ozonation. They also found that ozonation showed opposite results for humic- and protein-like substances as for UV absorbance ratio index (UVA210/UVA254), which provides information on the relative proportions between UV-absorbing functional groups and unsaturated compounds in NOM. The results of our study coincide with the results of their investigations in this respect. It is possible for ozonation to reduce the number of unsaturated groups and form new groups to cause the structural changes of humic substances with large molecular size. Ozonation may also destroy the particular functional groups of protein-like substances to thereby result in a relatively lower level of their characteristics.

3.4.2.

Fig. 6 e EEM fluorescence spectra of external foulants extracted from (a) MBR-A and (b) MBR-B.

Internal foulants

Membrane fibers of the two MBRs with the same quantity were chemically treated to extract the internal pore foulants for investigation. In order to identify the difference of DOM characteristics of internal foulants caused by ozone preoxidation, analysis of EEM fluorescence spectra of internal foulants was carried out and the results are shown in Fig. 7. Peak C at Ex/Em ¼ 440e445/275 nm predominated in both EEM fluorescence spectra, indicating that the humic acid-like substances represented by Peak C were the dominant components of the internal foulants. The characteristics of EEM fluorescence spectra of internal foulants obviously differed from those of external foulants. The fluorescence intensities of the four peaks of internal foulants from MBR-B were much weaker than those of internal foulants from

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 1 1 1 e2 1 2 1

a

b

4.

2119

Conclusions

Two identical submerged MBRs with or without pre-ozonation were comparatively operated to investigate performance of the processes and characterize organic membrane foulants using EEM fluorescence spectroscopy. It can be seen that preferable organic matter removal was achieved in the MBR process with pre-ozonation, and its TMP increased at a rate much lower than that of the individual MBR. EEM fluorescence spectroscopy was employed to characterize the DOM samples, EPS samples and membrane foulants of both MBRs. The results indicated that pre-ozonation decreased fluorescence intensities of the four main peaks in the influent DOM spectra especially for protein-like substances and caused red shifts of all fluorescence peaks to different extents. The peak intensities of the protein-like substances represented by Peak T1 and T2 in EPS spectra were obviously decreased as a result of preozonation. Both external and internal fouling could be effectively mitigated by the pre-ozonation. The most primary components of external foulants were humic acid-like substance (Peak C ) in the MBR with pre-ozonation and protein-like substance (Peak T1) in the individual MBR, respectively. The content decrease of protein-like substances and structural change of humic-like substances were observed in external foulants from EEM fluorescence spectra due to pre-ozonation. However, it could be seen that ozonation resulted in significant reduction of intensities but little change of locations of all peaks in EEM fluorescence spectra of internal foulants. Further work is required to assess the impact of pre-ozonation on other kinds of DOM, such as polysaccharide substances, to extend the knowledge of fouling control of MBR processes.

Fig. 7 e EEM fluorescence spectra of internal foulants extracted from (a) MBR-A and (b) MBR-B.

Acknowledgements

MBR-A, which was in line with the tendency of internal fouling resistances indicated by TMP increase. By comparison of the EEM fluorescence spectra of internal foulants from the two MBRs, it could be seen that the four main peaks were similarly located with a difference of no more than 5 nm along the two axes. The results indicated that the pre-ozonation process had the potential to effectively mineralize some NOM with small molecular size in raw water. For the NOM with relatively higher molecular size, pre-ozonation could cleave unsaturated bonds in aromatic moieties and reduce the molecular size/weight of the substances to make them more amenable to microbial degradation and utilization. The less quantity of organic matter tended to deposit or adsorb into membrane pores which induced internal fouling, suggesting that ozonation conducted well as a pre-treatment process for the biodegradation by microorganism in MBR-B. It was likely that the structures of organic substances in internal foulants changed slightly, and almost the same proportion of the organic matter content decreased as a result of ozone oxidation.

This research was funded by National High Technology Research and Development Program of China (2007AA06Z339) and State Key Laboratory of Urban Water Resource and Environment (HIT, Grant No. 2010DX12).

references

Ahmed, Z., Cho, J., Lim, B.-R., Song, K.-G., Ahn, K.-H., 2007. Effects of sludge retention time on membrane fouling and microbial community structure in a membrane bioreactor. Journal of Membrane Science 287 (2), 211e218. APHA, 1998. Standard Methods for the Examination of Water and Wastewater, twentieth ed. American Public Health Association/American Water Works Association/Water Environment Federation, Washington DC, USA. Baker, A., 2001. Fluorescence excitationeemission matrix characterization of some sewage-impacted rivers. Environmental Science and Technology 35 (5), 948e953. Baker, A., Inverarity, R., Charlton, M., Richmond, S., 2003. Detecting river pollution using fluorescence spectrophotometry: case studies from the Ouseburn, NE England. Environmental Pollution 124 (1), 57e70.

2120

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 1 1 1 e2 1 2 1

Bouhabila, E.H., Aim, R.B., Buisson, H., 2001. Fouling characterisation in membrane bioreactors. Separation and Purification Technology 22e23 (1e3), 123e132. Chang, I.S., Lee, C.H., 1998. Membrane filtration characteristics in membrane-coupled activated sludge system-the effect of physiological states of activated sludge on membrane fouling. Desalination 120 (3), 221e233. Chang, I.S., Bag, S.Q., Lee, C.H., 2001. Effects of membrane fouling on solute rejection during membrane filtration of activated sludge. Process Biochemistry 36 (8e9), 855e860. Chen, J., Gu, B., LeBoeuf, E.J., Pan, H., Dai, S., 2002. Spectroscopic characterization of the structural and functional properties of natural organic matter fractions. Chemosphere 48, 59e68. Chen, J., LeBoeuf, E.J., Dai, S., Gu, B., 2003. Fluorescence spectroscopic studies of natural organic matter fractions. Chemosphere 50 (5), 639e647. Cho, J., Song, K.G., Yun, H., Ahn, K.H., Kim, J.Y., Chung, T.H., 2005. Quantitative analysis of biological effect on membrane fouling in submerged membrane bioreactor. Water Science and Technology 51 (6e7), 9e18. Coble, P.G., 1996. Characterization of marine and terrestrial DOM in seawater using excitationeemission matrix spectroscopy. Marine Chemistry 51 (4), 325e346. Drews, A., Lee, C.H., Kraume, M., 2006. Membrane fouling a review on the role of EPS. Desalination 200 (1e3), 186e188. Drews, A., Mante, J., Iversen, V., Vocks, M., Lesjean, B., Kraume, M., 2007. Impact of ambient conditions on SMP elimination and rejection in MBRs. Water Research 41 (17), 3850e3858. Gone, D.L., Seidel, J.-L., Batiot, C., Bamory, K., Ligban, R., Biemi, J., 2009. Using fluorescence spectroscopy EEM to evaluate the efficiency of organic matter removal during coagulationflocculation of a tropical surface water (Agbo reservoir). Journal of Hazardous Materials 172 (2e3), 693e699. Gray, S.R., Ritchie, C.B., Tran, T., Bolto, B.A., Greenwood, P., Busetti, F., Allpike, B., 2008. Effect of membrane character and solution chemistry on microfiltration performance. Water Research 42 (3), 743e753. Henderson, R.K., Baker, A., Murphy, K.R., Hambly, A., Stuetz, R.M., Khan, S.J., 2009. Fluorescence as a potential monitoring tool for recycled water systems: a review. Water Research 43 (4), 863e881. Her, N., Amy, G., McKnight, D., Sohn, J., Yoon, Y., 2003. Characterization of DOM as a function of MW by fluorescence EEM and HPLC-SEC using UVA, DOC, and fluorescence detection. Water Research 37 (17), 4295e4303. Her, N., Amy, G., Plottu-Pecheux, A., Yoon, Y., 2007. Identification of nanofiltration membrane foulants. Water Research 41 (17), 3936e3947. Hong, S.H., Lee, W.N., Oh, H.S., Yeon, K.M., Hwang, B.K., Lee, C.H., Chang, I.S., Lee, S., 2007. The effects of intermittent aeration on the characteristics of bio-cake layers in a membrane bioreactor. Environmental Science and Technology 41 (17), 6270e6276. Huang, H.O., Lee, N.H., Young, T., Gary, A., Lozier, J.C., Jacangelo, J.G., 2007. Natural organic matter fouling of lowpressure, hollow-fiber membranes: effects of NOM source and hydrodynamic conditions. Water Research 41 (17), 3823e3832. Hudson, N., Baker, A., Reynolds, D., 2007. Fluorescence analysis of dissolved organic matter in natural, waste and polluted watersa review. River Research and Applications 23 (6), 631e649. Kimura, K., Naruse, T., Watanabe, Y., 2009. Changes in characteristics of soluble microbial products in membrane bioreactors associated with different solid retention times: relation to membrane fouling. Water Research 43 (4), 1033e1039. Lee, J.M., Ahn, W.Y., Lee, C.H., 2001. Comparison of the filtration characteristics between attached and suspended growth microorganisms in submerged membrane bioreactor. Water Research 35 (10), 2435e2445.

Lee, W., Kang, S., Shin, H., 2003. Sludge characteristics and their contribution to microfiltration in submerged membrane bioreactors. Journal of Membrane Science 216 (1e2), 217e227. Lee, E.K., Chen, V., Fane, A.G., 2008. Natural organic matter (NOM) fouling in low pressure membrane filtration-effect of membranes and operation modes. Desalination 218 (1e3), 257e270. Li, X.Y., Chu, H.P., 2003. Membrane bioreactor for drinking water treatment of polluted surface water supplies. Water Research 37 (19), 4781e4791. Li, L.S., Zhu, W.P., Zhang, P.Y., Zhang, Q.Y., Zhang, Z.L., 2006. AC/ O3-BAC processes for removing refractory and hazardous pollutants in raw water. Journal of Hazardous Materials 135 (1e3), 129e133. Meng, F.G., Zhang, H.M., Yang, F.L., Liu, L.F., 2007. Characterization of cake layer in submerged membrane bioreactor. Environmental Science and Technology 41 (11), 4065e4070. Meng, F.G., Liao, B.Q., Liang, S., Yang, F.L., Zhang, H.M., Song, L.F., 2010. Morphological visualization, componential characterization and microbiological identification of membrane fouling in membrane bioreactors (MBRs). Journal of Membrane Science 361 (1e2), 1e14. Peiris, R.H., Budman, H., Moresoli, C., Legge, R.L., 2010a. Understanding fouling behaviour of ultrafiltration membrane processes and natural water using principal component analysis of fluorescence excitation-emission matrices. Journal of Membrane Science 357 (1e2), 62e72. Peiris, R.H., Halle´, C., Budman, H., Moresoli, C., Peldszus, S., Huck, P.M., Legge, R.L., 2010b. Identifying fouling events in a membrane-based drinking water treatment process using principal component analysis of fluorescence excitationemission matrices. Water Research 44 (1), 185e194. Reynolds, D.M., 2002. The differentiation of biodegradable and non-biodegradable dissolved organic matter in wastewaters using fluorescence spectroscopy. Journal of Chemical Technology and Biotechnology 77 (8), 965e972. Saadi, I., Borisover, M., Armon, R., Laor, Y., 2006. Monitoring of effluent DOM biodegradation using fluorescence, UV and DOC measurements. Chemosphere 63 (3), 530e539. Sagbo, O., Sun, Y.X., Hao, A.L., Gu, P., 2008. Effect of PAC addition on MBR process for drinking water treatment. Separation and Purification Technology 58 (3), 320e327. Saravia, F., Zwiener, C., Frimmel, F.H., 2006. Interactions between membrane surface, dissolved organic substances and ions in submerged membrane filtration. Desalination 192 (1e3), 280e287. Schlichter, B., Mavrov, V., Chmiel, H., 2003. Study of a hybrid process combining ozonation and membrane filtrationfiltration of model solutions. Desalination 156 (1e3), 257e265. Sheng, G.P., Yu, H.Q., 2006. Characterization of extracellular polymeric substances of aerobic and anaerobic sludge using three-dimensional excitation and emission matrix fluorescence spectroscopy. Water Research 40 (6), 1233e1239. Stedmon, C.A., Markager, S., Bro, R., 2003. Tracing dissolved organic matter in aquatic environments using a new approach to fluorescence spectroscopy. Marine Chemistry 82 (3e4), 239e254.  Swietlik, J., Sikorska, E., 2004b. Application of fluorescence spectroscopy in the studies of natural organic matter fractions reactivity with chlorine dioxide and ozone. Water Research 38 (17), 3791e3799.  Swietlik, J., Dabrowska, A., Raczyk-Stanis1awiak, U., Nawrocki, J., 2004a. Reactivity of natural organic matter fractions with chlorine dioxide and ozone. Water Research 38 (3), 547e558. Tian, J.Y., Liang, H., Li, X., You, S.J., Tian, S., Li, G.B., 2008. Membrane coagulation bioreactor (MCBR) for drinking water treatment. Water Research 42 (14), 3910e3920.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 1 1 1 e2 1 2 1

Treguer, R., Tatin, R., Couvert, A., Wolbert, D., Tazi-Pain, A., 2010. Ozonation effect on natural organic matter adsorption and biodegradation-application to a membrane bioreactor containing activated carbon for drinking water production. Water Research 44 (3), 781e788. Uyguner, C.S., Bekbolet, M., 2005. Evaluation of humic acid photocatalytic degradation by UV-vis and fluorescence spectroscopy. Catalysis Today 101 (3e4), 267e274. Wang, Z.W., Wu, Z.C., Tang, S.J., 2009. Characterization of dissolved organic matter in a submerged membrane bioreactor by using three-dimensional excitation and emission matrix fluorescence spectroscopy. Water Research 43 (6), 1533e1540.

2121

Wang, Z.W., Tang, S.J., Zhu, Y.F., Wu, Z.C., Zhou, Q., Yang, D.H., 2010. Fluorescent dissolved organic matter variations in a submerged membrane bioreactor under different sludge retention times. Journal of Membrane Science 355 (1e2), 151e157. Williams, M.D., Pirbazari, M., 2007. Membrane bioreactor process for removing biodegradable organic matter from water. Water Research 41 (17), 3880e3893. Zhang, T., Lu, J.F., Ma, J., Qiang, Z.M., 2008. Fluorescence spectroscopic characterization of DOM fractions isolated from a filtered river water after ozonation and catalytic ozonation. Chemosphere 71 (5), 911e921.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 1 2 2 e2 1 3 0

Available at www.sciencedirect.com

journal homepage: www.elsevier.com/locate/watres

Advanced oxidation processes coupled with electrocoagulation for the exhaustive abatement of Cr-EDTA Christian Durante, Marco Cuscov, Abdirisak Ahmed Isse, Giancarlo Sandona`, Armando Gennaro* Department of Chemical Sciences, University of Padova, Via Marzolo 1, 35131 Padova, Italy

article info

abstract

Article history:

Using Cr-EDTA as a model system, a two-step method has been investigated for the abate-

Received 14 September 2010

ment of persistent chromium complexes in water. The treatment consists of an oxidative

Received in revised form

decomposition of the organic ligands by means of ozonization or electrochemical oxidation

15 December 2010

at a boron doped diamond (BDD) electrode, followed by removal of the metal via electro-

Accepted 22 December 2010

chemical coagulation. In the designed synthetic waste, EDTA has been used both as

Available online 28 December 2010

a chelating agent and as a mimic of the organic content of a typical wastewater provided by a purification leather plant. A crucial point evaluated is the influence of the oxidative

Keywords:

pretreatment on the chemical modification of the synthetic waste and hence on the elec-

Electrocoagulation

trocoagulation efficacy.

Cr abatement

Because of the great stability of Cr complexes, such as Cr-EDTA, the classical coagulation

Wastewater

methods, based on ligand exchange between Cr(III) and Fe(II) or Fe(III), are ineffective toward

Ozonization

Cr abatement in the presence of organic substances. On the contrary, when advanced oxidation processes (AOPs), such as ozonization or electrooxidation at a BDD anode are applied in series with electrocoagulation (EC), complete abatement of the recalcitrant Cr fraction can be achieved. ECs have been carried out by using Fe sacrificial anodes, with alternating polarization and complete Cr abatement (over 99%) has been obtained with modest charge consumption. It has been found that Cr(III) is first oxidized to Cr(VI) in the AOP preceding EC. Then, during EC, Cr(VI) is mainly reduced back to Cr(III) by electrogenerated Fe(II). Thus, Cr is mainly eliminated as Cr(III). However, a small fraction of Cr(VI) goes with the precipitate as confirmed by XPS analysis of the sludge. ª 2010 Elsevier Ltd. All rights reserved.

1.

Introduction

Chromium compounds are used in various industrial processes such as manufacture of dyes and paints, chrome plating and leather tanning. Although the vast majority of effluents coming from such processes contains Cr(III), which is much less toxic than Cr(VI) (Katz and Salem, 1993; Lamson and Plaza, 2002;

Shanker et al., 2005), some concern is being expressed about the possible hazards arising from the formation of mutagenic and carcinogenic Cr(VI), as a consequence of chemical oxidations of Cr(III) or natural bio-transformations in the environment (Apte et al., 2006; Dai et al., 2009). Driven by this concern, there is a general tendency to reduce the use of Cr in industrial processes and/or to limit its discharge in rivers and main

* Corresponding author. Tel.: þ39 049 8275132; fax: þ39 049 8275239. E-mail address: [email protected] (A. Gennaro). 0043-1354/$ e see front matter ª 2010 Elsevier Ltd. All rights reserved. doi:10.1016/j.watres.2010.12.022

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 1 2 2 e2 1 3 0

sewers to very low levels. For example, the most recent European Commission Directive on chromium requires that emissions of the metal in rivers and sewers be progressively reduced down to zero by 2020 (European Commission, 2000). However, Cr employment in some industrial sectors, including leather and dyeing industries, will be hardly abandoned for the lack of new processes capable of competing with those based on Cr chemistry, both in terms of costs and product quality. So, as long as Cr is being used in the transformation of raw materials, new abatement techniques and the improvement of those already existing are required (Hu et al., 2010). The main drawback associated with Cr(III) abatement is the formation of stable complexes with organic substrates present in the wastewater, which enhances the solubility of the metal and hence reduces the efficacy of the abatement process (Puzon et al., 2005; Remoundaki et al., 2007). Organic compounds are often present simultaneously with chromium in the wastewater under treatment; these have mainly three possible origins: (i) organic compounds, like leather residuals or tanning mask agents, co-existing with Cr in the effluent; (ii) excess of nutrients added in the bioreactor to support microbial growth during treatment; (iii) organic compounds metabolically produced by the microbial biomass in the bioreactor environment. Considering that complete depletion of the organic substrates is almost impossible even by prolonged mineralization processes, abatement of the recalcitrant Cr(III) fractions remains a demanding challenge. Electrochemical methods, and electrocoagulation (EC) in particular, have shown to be promising technologies for the removal of Cr(III), and heavy metals more in general (Yamaguchi et al., 2006; Lakshmanan et al., 2010), from a vast range of effluents, being more efficient than other conventional techniques such as chemical coagulation or adsorption (Golder et al., 2007a,b). Furthermore, EC has been proven to be effective also in the abatement of some heavy metal complexes (Kabdas‚lı et al., 2009). Although there are few examples in the literature reporting the successful abatement of Cr(III) complexes by means of EC treatment alone, the efficiency of the process may indeed be considerably enhanced by combining it with chemical or electrochemical pretreatments (Linares-Herna´ndez et al., 2010; Pociecha and Lestan, 2010). In fact, we have recently shown (Durante et al., 2010) that complete removal of recalcitrant Cr fractions from a real wastewater can be successfully achieved by an ozonization pretreatment followed by electrocoagulation performed at either FeeFe or FeeAl electrodes. Despite the interesting results obtained from an applicative point of view for the abatement of recalcitrant Cr by a combined ozonization-EC treatment (Durante et al., 2010), some fundamental aspects of the process, such as the role of organic ligands, inorganic anions and the oxidation state of chromium, have not been fully explored. Detailed investigation of these aspects will be of help not only in the full comprehension of the process but also in the optimization of the relevant operating parameters. This paper reports the results of a study aimed to cover some of the most important issues in the abatement of recalcitrant Cr by the two-step method. Since, however, a real wastewater is highly variable in composition and is too complex for a systematic investigation of the effects of each step in the overall treatment as well as the role of the chemical

2123

composition, we have chosen a simplified synthetic wastewater in which Cr is present as a metal complex with ethylenediaminetetraacetic acid (EDTA), which is also the unique source of the measured total organic carbon (TOC). EDTA has been chosen because it is commonly employed in many industrial processes and is one of the organic compounds present at the highest concentrations in many surface and drinking waters. Furthermore, EDTA is not degraded by conventional biological and physicochemical methods for the treatment of wastewaters and the purification of drinking water (Hinck et al., 1997; No¨rtemann, 1999) and, although it is not toxic to mammals at the concentrations found in the aquatic environment, there has been some concern about its potentiality to remobilize heavy metals out of river sediments and sewage sludges. Therefore, it is also of interest to understand the degradation of EDTA in order to develop new approaches for the abatement of heavy metals, especially recalcitrant Cr fractions.

2.

Experimental

All chemicals used in the study were at least of reagent grade purity. Cr-EDTA was synthesized by a literature method (Weyh and Hamm, 1968) and recrystallized twice in ethanol. All solutions were prepared in deionized water and pH correction was made by dripping concentrated NaOH or HCl. The synthetic wastewater was prepared to mimic a typical wastewater coming from a purification plant after the biological treatment. Therefore, it presents a pH of 7.8 and high 1 of concentrations of Naþ, Cl (1.8 g L1 of NaCl), SO2 4 (1.5 g L Na2SO4) and Cr(III) 0.4 mg L1. Chromium is introduced as an EDTA complex and is to be considered recalcitrant because it is the residual Cr(III) concentration after conventional treatments. The total organic carbon (TOC), which is entirely due to EDTA, was around 40 mg L1. A typical experimental setup consists in the preparation of 1 L of synthetic wastewater on which an advanced oxidation process (AOP) was carried out. Samples of the wastewater were periodically withdrawn during the treatment for TOC analysis (Shimadzu TOC-5050 analyzer) and HPLC measurements (JASCO 2075; Prevail organic acid column, phosphate buffer at pH ¼ 2.5). The electrocoagulation process was then carried out on the pretreated wastewater, after pH correction to 7.8. Again samples of the solution were periodically withdrawn for analysis to determine the concentration of Cr by means of inductively coupled plasma (ICP, Spectro-Ametek GENESIS ICP Spectrometer). Cyclic voltammetric measurements were performed by an EG&G PARC Model 173A potentiostat. The experiments were carried out in aqueous 0.5 M Na2SO4 by using a three-electrode cell system with a glassy carbon disk as working electrode. The counter and the reference electrodes were a Pt wire and a saturated calomel electrode (SCE), respectively (see supplementary material S1 for more details). Electrocoagulations (ECs) were carried out under galvanostatic conditions by using a Galvanostat/Potentiostat Amel 553. Two Fe rods (steel S185 of 15 cm2 area), acting as alternating sacrificial anodes, were used as working electrodes. The polarizations of the electrodes were alternated with

2124

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 1 2 2 e2 1 3 0

a frequency of 0.13 Hz by means of a square wave generator Amel 565, so that each electrode worked for 7.7 s alternatively as a cathode or as an anode. The electrochemical oxidations at the BDD electrode were carried out under galvanostatic control by means of a Galvanostat/Potentiostat Amel 7060, equipped with a Slave Unit 7061 for applying currents over 1 A. Ozonization experiments were performed at room temperature in a 2 L Pyrex glass bottle. Ozone was produced from oxygen (Air Liquide, Alphagaz 2) using a Fischer Model 502 corona generator and was continuously introduced through a porous distributor in the reaction vessel. The pH of the reaction mixture was monitored throughout the reaction course. The BDD electrode used in this work was about 1 mm thick film purchased from CSEM (Switzerland). A plate of about 25 cm2 was used and, prior to each experiment, the electrode surface was cleaned in 1 M H2SO4 by applying an anodic current density of 30 mA cm2 for 5 min. The experimental setup includes BDD as the anode and a platinized titanium (Ti/Pt) grid, with an apparent surface of 4.7 cm2, as the cathode.

3.

Results and discussions

3.1.

Cyclic voltammetry

The electrochemical behaviors of EDTA, Cr-EDTA and Cr(III) were investigated by cyclic voltammetry at a GC electrode in 0.5 M Na2SO4. All three compounds show irreversible oxidation peaks (Fig. S1, supplementary material); the peak potentials measured at v ¼ 0.2 Vs1 are 0.87, 1.11 and 1.57 V vs. SCE for EDTA, Cr(III) and Cr-EDTA, respectively. Thus, the effect of Cr(III) chelation by EDTA is to increase the oxidation potential of the metal by at least 400 mV, in virtue of the stabilization induced by the formation of the complex. This is also true for EDTA, which becomes much harder to oxidize when coordinating Cr(III). The voltammetric analyses show that oxidation of Cr(III) and Cr-EDTA leads to the same common products (see the supplementary material for more information).

3.2.

EDTA oxidation

In order to investigate the oxidative destruction of EDTA more in detail, a series of degradation experiments by means of ozonization or electrochemical oxidation at BDD has been carried out. These two techniques are among the most common advanced oxidation processes (AOPs) employed for wastewater purification. The AOPs have been carried out on a solution at pH ¼ 7.8 containing EDTA for a TOC value of 40 mg L1 and the typical Naþ, Cl and SO2 4 concentrations of a wastewater coming from a purification plant. Because EDTA degradation leads to the formation of many byproducts, which can influence Cr(III) solubility, we chose to follow the degradation reaction by two different techniques, namely HPLC and TOC analysis. Fig. 1a shows the effect of ozonization treatment on the TOC values at different O2/O3 flow rates. The TOC trends follow an exponential decay in accord with a pseudo-first-order dependence on the O3 concentration (Beltran et al., 2009). At flow rates higher than 30 L h1 (not shown), the TOC

Fig. 1 e Advanced oxidation treatments of a synthetic wastewater containing EDTA (97 mg LL1), NaCl (1.8 g LL1) and Na2SO4 (1.5 g LL1) at pH [ 7.8. TOC abatement: (a) ozonization at a flow rate of 4 (>), 7 (A), 12 (B), 16 (;) and 30 L hL1 (7); (b) electrooxidation at BDD at j [ 40 (-), 80 (,), 160 (:), 240 (C) and 320 mA cmL2 (6). Lines are intended to show trends.

abatements show superimposable trends, meaning that O3 saturation has been reached. When a flow rate of 30 L h1 is used, the TOC abatement increases with time up to a maximum of ca 80%, meaning that a recalcitrant fraction has been produced. This is a very common phenomenon when O3 is used as an oxidizing agent; in fact, oxidation by ozone often leads to the formation of low molecular weight carboxylic acids that are hardly oxidizable. Degradation of EDTA by electrochemical oxidation at the BDD electrode is shown in Fig. 1b. Particularly high current densities ( j ), covering a wide range from 40 to 320 mA cm2, have been used in order to attain TOC degradation rates that are comparable to those of the ozonization process. As expected, the rate of TOC decay shows a marked dependence on current density: the higher the current density, the faster the degradation kinetics. When a high current density ( j ¼ 320 mA cm2) is used, an EDTA mineralization above 99.5% is reached after 50 min of treatment and the TOC abatement follows an exponential decay. However, at lower current densities (e.g., j ¼ 40 mA cm2), the TOC abatement shows a sigmoidal decay, meaning that an initial induction period is present. To better explain this effect the contribution of Cl oxidation has been taken into account. When Cl ions are present in solution, electrooxidation of Cl to Cl2 occurs in competition with that of EDTA. This is followed by dismutation of Cl2 to hypochlorite, which can oxidize EDTA to CO2 in solution (Eqs. (1)e(3)). The hypochlorite ion is in equilibrium with hypochlorous acid and their distribution strongly

2125

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 1 2 2 e2 1 3 0

Fig. 2 e Advanced oxidation treatments of a synthetic wastewater containing EDTA (97 mg LL1) and Na2SO4 (1.5 g LL1) at pH [ 7.8. Effect of NaCl on TOC abatement by (a) electrooxidation at BDD (A [ 25 cm2, j [ 40 mA cmL2) or (b) ozonization at a flow rate of 30 L hL1; [NaCl] [ 0 g LL1 (-), 0.45 g LL1 (B), 0.9 g LL1 (C), 1.4 g LL1 (6), 1.8 g LL1 (:). Lines are intended to show trends.

depends on solution pH. Neutral or acid conditions favor the formation of the acid, whereas hypochlorite is the predominating species at high pH values. In our experimental conditions, the solution pH increases from 7.8 to 12 during electrolysis, thus creating conditions favorable for the formation of ClO rather than HClO.

2Cl / Cl2 þ 2e

(1)

Cl2 þ 2OH / ClO þ Cl þ H2O

(2)

17ClO þ C10H16N2O8 / 17Clþ10CO2 þ 2NH3 þ 5H2O

(3)

The effect of Cl concentration on EDTA degradation at the BDD electrode at 40 mA cm2 is reported in Fig. 2a. As shown, the TOC degradation rate increases with increasing Ccl . A similar effect has been observed by other Authors (Regina Costa and Olivi, 2009) on an IreSnO2 electrode. It is worth noting that the complete absence of Cl is detrimental to EDTA mineralization, causing a decrease of the abatement efficiency from 75 to 25%. These results point out that ClO plays a key role in the degradation not only of EDTA (Eq. (3)) but also of its oxidation products. At the beginning of the electrochemical treatment, a fraction of the charge is expended in Cl oxidation, resulting in low efficiency of EDTA degradation. However, as soon as hypochlorite becomes available in the bulk, the synergic effect with the electrodic heterogeneous oxidation leads to a fast TOC decay. This explains the sigmoidal shape of the degradation trend, which becomes more evident upon increasing Cl concentration. Just for comparison, the effect of Cl concentration on the ozonization process is reported in Fig. 2b, where four almost superimposable trends of TOC decay are shown for experiments performed at different Cl concentrations. This is a clear evidence that reaction (3) does not play any role in the mineralization of EDTA by ozonization, possibly because O3 is not capable of oxidizing Cl to Cl2. 2 Besides Cl, tannery effluents often contain SO2 4 and CO3 , which might be oxidized to persulfate and percarbonate, respectively (Martinez-Huitle and Brillas, 2009). Both these species are known to be extraordinary oxidizing and disinfectant agents, so their intermediate generation may ignite an efficient catalytic cycle. Therefore, the effect of these ions on the abatement of EDTA has been considered in both AOPs. 2 However, as reported in Table 1, neither SO2 4 nor CO3 has an

Table 1 e Effect of SO2L and CO2L concentrations on TOC abatement in a synthetic wastewater.a 4 3 Electrolyte

Na2SO4

NaHCO3

a b c d e f g

C/(g L1)

0 1.5 3 6 12 0 0.5 1 2

O3c

BDDb TOC0d

TOC100e

DTOC(%)f

TOC0d

TOC90g

DTOC(%)f

38.1 37.4 37.9 37.5 37.5 38.1 38.4 37.8 37.5

10.6 9.7 9.8 13.0 11.7 10.6 9.8 9.0 5.5

72 74 74 65 69 72 74 76 85

41.1 39.8 e 39.3 e 41.1 e e 41.4

10.6 8.9 e 9.8 e 10.6 e e 5.7

74 78 e 75 e 74 e e 86

Wastewater composition: Cr z 0.4 mg L1; Cl ¼ 3.1  102 M (1.8 g L1 NaCl). j ¼ 40 mA cm2. Flow rate ¼ 30 L h1. Starting TOC (EDTA). TOC after 100 min of treatment. TOC abatement after the treatment. TOC after 90 min of treatment.

2126

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 1 2 2 e2 1 3 0

appreciable effect on TOC degradation, even when the ions are present at high concentrations (2e6 g L1). This points out 2 that neither SO2 4 nor CO3 can undergo oxidation under the experimental conditions of the two AOPs, which is not surprising as the oxidation potentials of both Cl and EDTA are much lower than those required for the oxidation of these ions.

3.3.

Cr-EDTA degradation

Generally, EDTA can form stable complexes with any free cation present in the wastewater, the efficiency of complexation depending on both the thermodynamics and kinetics of the reaction. Although Cr-EDTA formation is thermodynamically favorable (log K ¼ 23.4 at r.t.), the reaction is kinetically very sluggish, complete complexation requiring some days (Pettine et al., 2008). Therefore, to be sure of the presence of Cr-EDTA in solution, the complex was synthesized and added to the synthetic waste. The Cr-EDTA concentration was chosen so as to obtain a Cr concentration in solution of 0.4 mg L1, which is the typical value found as a recalcitrant fraction in real wastewaters. After this, an excess of free EDTA was added to the solution in order to reach a TOC value of 40 mg L1. Comparing this wastewater with wastewaters of the same TOC value but prepared either with EDTA alone or with a mixture of EDTA and Cr(III), added as Cr(NO3)3$9H2O, has shown only slight differences in TOC degradation trends. This means that the effect of Cr on the TOC abatement cannot be appreciated under these experimental conditions. This is easily conceivable because at a Cr concentration of 0.4 mg L1 the TOC contribution of the complex amounts to less than

Fig. 3 e (a) TOC abatement by ozonization of (C) EDTA or (B) Cr-EDTA at a flow rate of 30 L hL1; (b) decrease of (6) EDTA and (:) Cr-EDTA concentrations during degradation by ozonization at a flow rate of 30 L hL1; (c) TOC abatement by electrooxidation of (-) EDTA or (,) Cr-EDTA at BDD (A [ 25 cm2, j [ 40 mA cmL2); d) decrease of (>) EDTA and (A) Cr-EDTA concentrations during degradation by electrooxidation at BDD (A [ 25 cm2, j [ 40 mA cmL2). Lines are intended to show trends.

2.5%, which makes insignificant the role of Cr-EDTA on TOC degradation. Therefore, we found it useful to investigate the degradation of Cr-EDTA at a concentration of 3.3  104 M, which corresponds to a TOC concentration of 40 mg L1. Fig. 3 reports the results of TOC degradation experiments on wastewaters containing 1.8 g L1 of NaCl together with either free EDTA or Cr-EDTA, with a TOC concentration of about 40 mg L1 in each case. In Fig. 3a TOC depletions during ozonization are reported and what clearly appears is that free EDTA has a faster degradation kinetics (2.5 times faster) than Cr-EDTA, although both degradation trends reach a similar final TOC value, corresponding to the recalcitrant fraction. Both TOC degradation trends show an exponential decay and it is likely that the process follows a pseudo-first-order decay. HPLC analysis reveals that, when EDTA is free, complete disappearance, by degradation to lower molecular weight compounds, is achieved after 20 min of treatment (Fig. 3b), which corresponds, more or less, to a TOC abatement of 50%. On the contrary, formation of free EDTA or lower molecular weight compounds has not been detected during the treatment of Cr-EDTA, which is completely depleted after 80 min of treatment (Fig. 3b, full triangles), resulting in a TOC abatement of 68%. Clearly, the two substrates behave differently; degradation of free EDTA involves the formation of intermediate products, probably by sequential loss of the acetic groups, which are more difficult to oxidize than EDTA, whereas degradation of Cr-EDTA leads directly to the recalcitrant fraction. The same experiments were carried out by electrochemical oxidation of EDTA and Cr-EDTA at the BDD electrode; in this case, complete TOC mineralization was reached for both of them after 60 min (Fig. 3c). HPLC analysis of the electrochemically oxidized sample (Fig. 3d) reveals that the concentration of both free EDTA and Cr-EDTA decreases to zero after 48 min of treatment, corresponding to a TOC abatement of 95% and 88%, respectively. Therefore, in this case, the degradation of free and complexed EDTA proceeds without a significant accumulation of intermediates. In other words, once an EDTA or a Cr-EDTA molecule is oxidized, the

Fig. 4 e UVeVis spectra recorded during Cr-EDTA degradation by (a) ozonization or (b) electrooxidation at BDD.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 1 2 2 e2 1 3 0

degradation proceeds until complete mineralization is achieved. It is to be noted, however, that direct oxidation of EDTA at the electrode surface is likely to involve a sequential loss of carboxylic groups, which implies the formation of intermediate products having higher oxidation potentials than EDTA. While these intermediates can be oxidized at BDD in the case of Cr-EDTA because of the very high oxidation potential of the complex, they are expected to accumulate in the case of the wastewater containing free EDTA. Indeed, besides direct oxidation of EDTA and its intermediates at the electrode, homogeneous oxidation by ClO must be taken into account (Eq. (3)). This homogeneous oxidation process appears to contribute significantly to the EDTA mineralization, particularly in the case of the free ligand. In fact, HPLC analyses (not shown) reveal that, when electrooxidation at BDD is carried out in the absence of Cl, the EDTA peak completely disappears after 24 min, but the TOC concentration decreases by 35% only, clearly pointing out that EDTA has been mainly converted to low molecular weight organic molecules. The same experiment performed on a solution containing Cr-EDTA shows after 84 min a complete disappearance of the complex with a TOC decrement of 75%. In this case, both EDTA and its intermediates are oxidized at the electrode at the oxidation potential of Cr-EDTA, leading to almost total mineralization of the organic substances.

Monitoring the degradation of Cr-EDTA by UVeVis spectroscopy (Fig. 4) reveals that the oxidation of the complex leads to the formation of Cr(VI). In fact, at neutral or slightly basic pH, Cr-EDTA shows two well defined absorption bands at 393 and 541 nm that evolve during the oxidative treatment to a new set of bands attributable to Cr(VI) species. A comparison of spectroscopic features observed after the AOPs with those of different Cr(VI) species shows that Cr(III) is oxidized to Cr2 O2 7 and CrO2 by ozonization and electrooxidation at BDD, 4 respectively. It is known that the UVeVis spectroscopic pattern of Cr(VI) species such as chromates and dichromates depends on pH. Ozonization involves the formation of low molecular weight carboxylic acids that lower the pH to a slightly acidic value (pH ¼ 5) and, as a consequence, the most stable anion at such a pH value is Cr2 O2 7 , which gives characteristic absorption bands at 252, 351 and 417 nm (Fig. 4a). On the contrary, during electrooxidation at the BDD electrode performed in an undivided cell, the pH increases to values around 12 as a result of the cathodic reduction of H2O to H2 at the Ti/Pt electrode (Eq. (4)). Of course, part of the OH ions produced at the cathode is neutralized by the anodic process. Indeed, although this process mainly involves oxidation of EDTA and Cl, it has some contribution from oxidation of H2O and/or OH to Hþ and H2O, respectively. Nevertheless, the overall effect of water electrolysis at the two electrodes is to increase the pH to high values at which CrO2 4 is the most stable Cr(VI) species in solution as confirmed by the two sharp bands at 272 and 373 nm, characteristic of chromate solutions (Fig. 4b) (Sena et al., 2000). 2H2O þ 2e / H2 þ 2OH

3.4.

Fig. 5 e Cr abatement by (a) electrochemical and (b) chemical coagulations performed at different pretreatment conditions. a) EC at j [ 64 mA cmL2: (-) not pretreated, (6) not pretreated, Cr(VI) in presence of EDTA (TOC [ 40 mg LL1), (B) pretreated by ozonization at a flow rate of 30 L hL1 for 90 min, (C) pretreated by electrooxidation at BDD (A [ 25 cm2, j [ 320 mA cmL2) for 90 min. (b) Chemical coagulation (after ozonization at a flow rate of 30 L hL1 for 90 min) by addition of (A) FeSO4 or (,) Fe2(SO4)3.

2127

(4)

Electrocoagulation on Cr-EDTA

Fig. 5a reports the results of chromium abatement via EC performed after different pretreatment conditions on a couple of Fe electrodes with alternating electrode polarization (0.13 Hz) at a current density of 64 mA cm2. When EC was carried out on a wastewater without pretreatment, a maximum abatement of 18% was reached after 30 s of treatment (Fig. 5a, solid squares), corresponding to a charge of 30 C, after which the concentration remains unchanged. This result could not be improved by either prolonging the treatment or varying the current density. Besides the modest Cr depletion, a very poor TOC abatement (4%) was observed. The low efficacy of the EC process must therefore be ascribed to the incapability of Fe(II) and Fe(III) to dislodge Cr(III) from the EDTA complex to produce the insoluble Cr(OH)3. Fig. 5a also reports the abatement of Cr(VI) (open triangles) in the presence of excess EDTA, which gives a TOC of 40 mg L1. In this case, Cr is no longer complexed by EDTA and a sensible improvement of its abatement is observable; in fact, now a maximum Cr abatement of 90% is obtained after furnishing 250 C. Experiments carried out in the presence of a lower concentration of EDTA (stoichiometric 1:1 ratio with Cr(VI)), which gives a TOC of 1 mg L1 show a further improvement of the abatement efficacy to over 99% after only 30 C. This is a clear indication that removal of the complexing agents is necessary in order to enhance the abatement of the recalcitrant fractions of Cr present in the wastewater.

2128

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 1 2 2 e2 1 3 0

The results reported in section 3.3 suggest that the oxidative treatment on Cr-EDTA produces Cr decomplexation not only by EDTA mineralization but also by oxidation of Cr(III) to Cr(VI). Therefore, we have subjected the synthetic waste to oxidative pretreatment by ozonization or by electrooxidation at BDD and then, after pH correction to 7.8, Fe(II) was electrochemically added. The pretreatment conditions were appropriately chosen in order to obtain similar TOC abatement kinetics (details are reported in the figure caption) for the two methods, whereas the EC experiments were carried out as described above. The results reported in Fig. 5a show that, although Cr(VI) is definitely more soluble than its precursor, complete Cr abatement (over 99%) is accomplished for both ECs carried out on the electrolyzed (solid circles) and ozonized (open circles) wastewaters, after a charge consumption of only 60 C for the former and 12 C for the latter. This is a clear evidence that oxidation of Cr(III) to Cr(VI) is crucial for the abatement of Cr by EC. Oxidation of Cr(III) to Cr(VI) is, however, a necessary but not sufficient condition to achieve complete Cr abatement; in fact, the presence of EDTA significantly reduces the efficiency of the EC process, even if Cr(VI) is not complexed by EDTA. Therefore, substantial depletion of TOC is also a prerequisite for a quantitative abatement of Cr by EC. Interestingly, both of the above conditions are fulfilled by the pretreatments considered in this study. Besides EC, the possibility of Cr abatement by chemical coagulation was investigated. To this end, Fe(II) or Fe(III) was added by dripping a concentrated solution of the metal into a wastewater pretreated by ozonization and the results are shown in Fig. 5b. Addition of Fe(III) as Fe2(SO4)3 into the pretreated wastewater leads to a maximum Cr depletion of 45%, which is reached after addition of 0.14 g L1 of Fe(III). On the other hand, addition of Fe(II) as FeSO4 results in a fast and complete abatement of Cr; in this case almost complete Cr abatement is reached after addition of 0.01 g L1 of Fe(II), which is equivalent to the consumption of 35 C in an EC process. The efficiency of this chemical coagulation is therefore comparable with that observed for the EC process. A further insight into the abatement mechanism is given by XPS examination (see supplementary material) of sludges coming from ECs on pretreated samples. The features of the deconvoluted spectra (Fig. S2, supplementary material) reveal that Cr is present in the sludge not only in the trivalent form (80%), but also as Cr(VI) (20%). On the other hand, it has been established by WAXD analysis of the sludge samples (Fig. S3, supplementary material) that the d values match well those of FeO(OH) (JCPDS file no. 44-1415). Unfortunately, it was not possible to identify the Cr phases in the solid, probably because of the very low concentration of the metal in the sludge. Therefore, it was not possible to unambiguously identify the Cr species in the sludge. However, in previous works reporting ECs on Cr wastewater, Cr(VI) has been detected in the sludge as NH4Fe(CrO4)2 (Durante et al., 2010), whereas the principal component of Cr(III) species has been found to be chromite (FeCr2O4) (Kongsricharoern and Polprasert, 1995). On the basis of XPS and WAXD data, reduction of Cr(VI) to Cr(III) during the EC process appears to be crucial for a successful abatement of Cr. This is corroborated by the results of the chemical coagulation experiments. Indeed,

Fig. 6 e Proposed mechanism for the two-step Cr abatement; the numbers on the arrows identify the different reaction pathways.

efficient Cr abatement was observed only when Fe(II) was added to the pretreated wastewater. This suggests that Fe(II) reduces Cr(VI) to Cr(III), which then precipitates together with the hydroxide complexes of Fe(II) and Fe(III). As a matter of fact, some Cr depletion is observed also with Fe(III) (Fig. 5a, solid squares), but this is likely due to entrapping of Cr inside hydroxide flocks and the formation of insoluble bimetallic complexes such as NH4Fe(CrO4)2.

3.5.

Mechanistic considerations

We have shown that, unless an oxidative pretreatment is applied to the wastewater, EC is ineffective toward Cr-EDTA abatement. This is because Cr-EDTA is very stable and hence ligand displacement reactions with Fe(II) or Fe(III) are very unlikely. On the other hand, EDTA forms very stable and soluble complexes with Cr(III), Fe(II) and Fe(III), so also a TOC decrease is very unlikely after EC treatments. We have also shown that complete Cr abatement can be achieved only if the metal is present as Cr(VI) and the TOC value of the wastewater is not high. Both these prerequisites can be fulfilled by either ozonization or electrochemical oxidation at BDD. A mechanism that takes into account all these considerations is schematically reported in Fig. 6, where ligand (EDTA indicated as L) displacement reactions have been neglected as they are deemed unfavorable. In the oxidative pretreatment step, Cr-EDTA is oxidized to chromate or dichromate, while EDTA is oxidatively degraded to lower molecular weight molecules. In the second stage, Fe(II) ions are added into the solution by anodic dissolution of Fe. These may undergo either oxidation to Fe(III) by dissolved oxygen (followed by precipitation as hydroxide complexes) or complexation by residual EDTA or its degradation products present in solution. In the latter case, the efficiency of electrocoagulation is reduced until all free ligands form Fe(II) complexes. The experimental results point out that Cr(VI) formed during the pretreatment is mainly reduced to Cr(III) by Fe(II) according to, for example, Eq. (5):

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 1 2 2 e2 1 3 0

2þ CrO2 þ 8Hþ /Cr3þ þ 3Fe3þ þ 4H2 O DE< ¼ þ0:64 4 þ 3Fe

(5)

2129

references

 E< . where DE< ¼ E< Fe3þ =Fe2þ CrO2 =Cr3þ 4

Once Cr is reduced back to the trivalent form, it can either precipitate as Cr(OH)3 (Fig. 6, P1) or co-precipitate with Fe(III) ions via hydroxide or polyhydroxide formation (Fig. 6, P2). Another option is the formation of an insoluble Cr(III)-Fe(II) oxide such as FeCr2O4 (Fig. 6, P3). All these precipitation processes shift Eq. (5) to the right. Last, Cr may also precipitate as a Cr(VI) salt by formation of insoluble compounds such as NH4Fe(CrO4)2 (Fig. 6, P4).

4.

Conclusions

It has been shown that complete removal of recalcitrant fractions of Cr, present as a Cr-EDTA complex in a synthetic wastewater, can be successfully achieved by an ozonization or electrooxidation pretreatment followed by EC performed at two Fe electrodes with an alternating polarization of 7.7 s. The results show that the pretreatment is fundamental for a complete abatement of Cr. In fact, both ozonization and electrochemical oxidation at BDD allow Cr decomplexation via oxidation to Cr(VI) as well as mineralization of the organic ligand. Investigation of the effects of anions such as Cl, SO2 4 and CO2 3 , which are commonly present in wastewaters, on the pretreatments indicated that only Cl deeply affects Cr-EDTA oxidation at the BDD electrode, whereas the ozonization process is not affected at all and hence is more suitable for the treatment of complex systems. The maximum efficacy of the treatment, with almost complete depletion of Cr, can be achieved with only few seconds of EC. This method allows the consumption of a really low charge and the production of a small amount of sludge and hence provides an easy and cheap process with low costs not only for chromium removal but also for sludge disposal. It has been shown that during EC, Cr(VI) is mainly reduced to Cr(III) by Fe(II) ions electrochemically produced at the sacrificial Fe anode, but, as confirmed by XPS analysis, a small fraction of Cr(VI) is still present in the sludge. The results reported here indicate that electrocoagulation with ozonization pretreatment would be a suitable means of decontamination of dilute effluents at the end of an industrial depuration line.

Acknowledgments This work was financially supported by the University of Padova (Italy). We are indebted to Prof. Gaetano Granozzi for XPS analysis and to Prof. Carla Marega for WAXD spectra interpretations.

Appendix. Supplementary material The supplementary data associated with this article can be found in the on-line version at doi:10.1016/j.watres.2010.12.022.

Apte, A.D., Tare, V., Bose, P., 2006. Extent of oxidation of Cr(III) to Cr(VI) under various conditions pertaining to natural environment. J. Hazard. Mater. B128, 164e174. Beltran, F.J., Aguinaco, A., Garcia-Araya, J.F., 2009. Mechanism and kinetics of sulfamethoxazole photocatalytic ozonation in water. Water Res. 43, 1359e1369. Dai, R., Liu, J., Yu, C., Sun, R., Lan, Y., Maob, J.D., 2009. A comparative study of oxidation of Cr(III) in aqueous ions, complex ions and insoluble compounds by manganese-bearing mineral (birnessite). Chemosphere 76, 536e541. Durante, C., Isse, A.A., Sandona`, G., Gennaro, A., 2010. Exhaustive depletion of recalcitrant chromium fractions in a real wastewater. Chemosphere 78, 620e625. European Commission, 2000. EC water framework directive (2000/60/EC). Off. J. Eur. Communities L327 (43), 1e72. Golder, A.K., Samanta, A.N., Ray, S., 2007a. Removal of trivalent chromium by electrocoagulation. Sep. Purif. Technol. 53, 33e41. Golder, A.K., Chanda, A.K., Samanta, A.N., Ray, S., 2007b. Removal of Cr(VI) from aqueous solution: Electrocoagulation vs chemical coagulation. Sep. Sci. Technol. 42, 2177e2193. Hinck, M.L., Ferguson, J., Puhaakka, J., 1997. Resistance of EDTA and DTPA to aerobic biodegradation. Wat. Sci. Tech. 35 (2e3), 25e31. Hu, C.Y., Lo, S.L., Liou, Y.H., Hsu, Y.W., Shih, K., Lin, C.J., 2010. Hexavalent chromium removal from near natural water by coppereiron bimetallic particles. Water Res. 44, 3101e3108. ¨ lmez-Hancı, T., Arslan-Alaton, I., Kabdas‚lı, I., Arslan, T., O Tu¨nay, O., 2009. Complexing agent and heavy metal removals from metal plating effluent by electrocoagulation with stainless steel electrodes. J. Hazard. Mater. 165, 838e845. Katz, S.A., Salem, H., 1993. The toxicology of chromium with respect to its chemical speciation: a review. J. Appl. Toxicol. 13, 217e224. Kongsricharoern, N., Polprasert, C., 1995. Electrochemical precipitation of chromium (Cr6þ) from an electroplating wastewater. Water Sci. Technol. 31 (9), 109e117. Lakshmanan, D., Clifford, D.A., Samanta, G., 2010. Comparative study of arsenic removal by iron using electrocoagulation and chemical coagulation. Water Res 44, 5641e5652. Lamson, D.W., Plaza, S.M., 2002. The safety and efficacy of high-dose chromium. Altern. Med. Rev. 7 (3), 218e235. Linares-Herna´ndez, I., Barrera-Dı´az, C., Bilyeu, B., Jua´rez-Garcı´aRojas, P., Campos-Medina, E., 2010. A combined electrocoagulationeelectrooxidation treatment for industrial wastewater. J. Hazard. Mater. 175, 688e694. Martinez-Huitle, C.A., Brillas, E., 2009. Decontamination of wastewaters containing synthetic organic dyes by electrochemical methods: a general review. Appl. Catal. B. Environ. 87, 105e145. No¨rtemann, B., 1999. Biodegradation of EDTA. Appl. Microbiol. Biot. 51, 751e759. Pettine, M., Gennari, F., Campanella, L., Millero, F.J., 2008. The effect of organic compounds in the oxidation kinetics of Cr(III) by H2O2. Geochim. Cosmochim. Ac. 72, 5692e5707. Pociecha, M., Lestan, D., 2010. Using electrocoagulation for metal and chelant separation from washing solution after EDTA leaching of Pb, Zn and Cd contaminated soil. J. Hazard. Mater. 174, 670e678. Puzon, G.J., Roberts, A.G., Kramer, D.M., Xun, L., 2005. Formation of soluble organo-chromium(III) complexes after chromate reduction in the presence of cellular organics. Environ. Sci. Technol. 39, 2811e2817.

2130

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 1 2 2 e2 1 3 0

Regina Costa, C., Olivi, P., 2009. Effect of chloride concentration on the electrochemical treatment of a synthetic tannery wastewater. Electrochim. Acta 54, 2046e2052. Remoundaki, E., Hatzikioseyian, A., Tsezos, M., 2007. A systematic study of chromium solubility in the presence of organic matter: consequences for the treatment of chromiumcontaining wastewater. J. Chem. Technol. Biot 82, 802e808. Sena, M.M., Scarminio, I.S., Collins, K.E., Collins, C.H., 2000. Speciation of aqueous chromium (VI) solution with the aid of Q-mode factor analysis followed by oblique projection. Talanta 53, 453e461.

Shanker, A.K., Cervantes, C., Loza-Taverac, H., Avudainayagamd, S., 2005. Chromium toxicity in plants. Environ. Int. 31, 739e753. Weyh, J.A., Hamm, R.E., 1968. Iminodiacetato, methyliminodiacetato, and 1,3- propanediaminetetraacetato complexes of chromium(III). Inorg. Chem. 7, 2431e2435. Yamaguchi, Y., Yamanaka, Y., Miyamoto, M., Fujishima, A., Honda, K., 2006. Hybrid electrochemical treatment for persistent metal complex at conductive diamond electrodes and clarification of its reaction route. J. Electrochem. Soc. 153, J123eJ132.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 1 3 1 e2 1 4 5

Available at www.sciencedirect.com

journal homepage: www.elsevier.com/locate/watres

A two-class population balance equation yielding bimodal flocculation of marine or estuarine sediments Byung Joon Lee a,*, Erik Toorman a, Fred J. Molz b, Jian Wang a a

Hydraulics Laboratory, Department of Civil Engineering, Katholieke University of Leuven, Kasteelpark Arenberg 40, B-3001 Heverlee, Belgium b Department of Environmental Engineering & Earth Sciences, Clemson University, 342 Computer Court, Anderson, SC 29625, USA

article info

abstract

Article history:

Bimodal flocculation of marine and estuarine sediments describes the aggregation and

Received 16 June 2010

breakage process in which dense microflocs and floppy macroflocs change their relative

Received in revised form

mass fraction and develop a bimodal floc size distribution. To simulate bimodal floccula-

23 December 2010

tion of such sediments, a Two-Class Population Balance Equation (TCPBE), which includes

Accepted 23 December 2010

both size-fixed microflocs and size-varying macroflocs, was developed. The new TCPBE

Available online 31 December 2010

was tested by a model-data fitting analysis with experimental data from 1-D column tests, in comparison with the simple Single-Class PBE (SCPBE) and the elaborate Multi-Class PBE

Keywords:

(MCPBE). Results showed that the TCPBE was the simplest model that is capable of simu-

Sediment

lating the major aspects of the bimodal flocculation of marine and estuarine sediments.

Flocculation

Therefore, the TCPBE can be implemented in a large-scale multi-dimensional flocculation

Population balance equation

model with least computational cost and used as a prototypic model for researchers to

Bimodal

investigate complicated cohesive sediment transport in marine and estuarine environ-

Microfloc

ments. Incorporating additional biological and physicochemical aspects into the TCPBE

Macrofloc

flocculation process is straight-forward also. ª 2011 Elsevier Ltd. All rights reserved.

1.

Introduction

Bimodal flocculation describes the aggregation and breakage process in which dense microflocs and floppy macroflocs change their relative mass fraction and develop a bimodal floc size distribution (FSD) with two peaks in the mass or volumetric size distribution of sediment flocs (Manning et al., 2007a; Mietta et al., in press; Verney et al., 2009; van Leussen, 1994). Bimodal flocculation has often been observed in marine and estuarine environments, especially in turbidity maximum zones which involve dynamic fronts between fresh and brackish water (Burd and Jackson, 2002; Chen et al., 2005; Curran et al., 2002, 2004, 2007; Hill et al., 2000; Jackson, 1995; Jackson et al., 1995; Li et al., 1993, 1999; Manning et al., 2006, 2007a, b; Manning and Bass, 2006; Mietta et al., in press;

Mikkelsen et al., 2006; Mikkelsen and Pejrup, 2001; van Leussen, 1994; Yuan et al., 2009). For example, a mixing process between microflocs supplied from an upstream river and macroflocs matured in an estuary was proposed as a cause of an observed bimodal FSD (Orange et al., 2005). Also, floc erosion and re-suspension from the bottom layer can enhance a bimodal FSD (Yuan et al., 2009). In this paper, however, we are interested mainly in internal causes of bimodal flocculation, rather than that due to simple mixing of different size classes of flocs. Internally, bimodal flocculation can occur due to primary and secondary particle/floc binding mechanisms. The primary binding mechanism is characterized by direct contact between clay particles (e.g. the face-to-face or face-to-edge bonding between clay particles), whereas the secondary

* Corresponding author. Tel.: þ32 16 321672; fax: þ32 16 321989. E-mail address: [email protected] (B.J. Lee). 0043-1354/$ e see front matter ª 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.watres.2010.12.028

2132

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 1 3 1 e2 1 4 5

binding mechanism is by loose agglomeration between microflocs including the effects of heterogeneous inorganic or organic materials (e.g. polymeric bridging between microflocs) (van Leussen, 1994; Winterwerp and van Kesteren, 2004). The primary binding mechanism is strong but size-limited, whereas the secondary is weak but size-extending. Thus, the primary and secondary binding mechanisms play the distinct roles of packing less-porous and hard microflocs and agglomerating highly-porous and floppy macroflocs, respectively. As long as the flocculation process of a cohesive sediment is governed by both the primary and secondary binding mechanisms in a fluid shear field, the cohesive sediment will soon develop a bimodal FSD due to a mixture of resistant microflocs and fragile macroflocs. When the effects of a typical tidal cycle are included, the relative mass fractions of microflocs and macroflocs will continuously change while maintaining a bimodal FSD (Manning et al., 2006, 2007b; Manning and Bass, 2006; Winterwerp, 2002). Marine and estuarine sediments are composed of heterogeneous particles with varying size, shape, mineralogy and so on. Such variability has been reported to enhance bimodal flocculation. For example, clay and silt have different particle binding capabilities and thus may be involved in bimodal flocculation. Considering that silt has the smaller contact area and higher density for a unit volume than clay flocs, it must have a lower binding capability, and thus may fail to build macroflocs above a certain size range (Li et al., 1993, 1999; Manning et al., 2007a). Similarly, heterogeneous mineral content can also limit floc size. For instance, smectite was found to remain in fragmented microflocs rather than in aggregated macroflocs because it has a smaller particle binding capability than other minerals (Li et al., 1999). Natural organic matter can also enhance bimodal flocculation by modifying the primary and secondary binding mechanisms. The “gluing capacity” of such materials can enhance the building of macroflocs from constituent microflocs. Among various natural organic matter, linear polymeric organic materials, such as polysaccharides produced by benthic organisms, are known for enhancing inter-particle polymeric bridges (Chen et al., 2005; Manning et al., 2006, 2007a,b; Manning and Bass, 2006; Mietta et al., in press; Mikkelsen et al., 2006; van Leussen, 1994; Verney et al., 2009). The resulting macroflocs composed of polymeric organic matter and inorganic microflocs have large and floppy structures, and are often called “marine snow” to differentiate them from more compact flocs (Droppo et al., 2005). Floc sizes of marine and estuarine sediments were found to span from hundreds to thousands of micrometers in the organicenriched condition, but only from tens to a few hundred micrometers in the organic-free condition (Chen et al., 2005; Manning et al., 2006, 2007a,b; Manning and Bass, 2006; Mietta et al., 2009, in press; Mikkelsen et al., 2006; van Leussen, 1994; Verney et al., 2009). Clearly, bimodal flocculation mechanisms are common and important to understand (Chen et al., 2005; Manning et al., 2006, 2007b; Manning and Bass, 2006; Mikkelsen et al., 2006). Irrespective of the common occurrence of bimodal flocculation in estuarine and marine environments, most of the contemporary flocculation models simply assume that flocs have a single spatial-averaged size, with an underlying

unimodal FSD. Such output may result from a single-class population balance equation (SCPBE) or a concentrationdependent empirical equation (settling velocity versus solid concentration) (van Rijn, 1984, 2007; van Leussen, 1994; Winterwerp, 2002; Winterwerp and van Kesteren, 2004; Perianez, 2005; Son and Hsu, 2008, 2009; Maggi, 2009). However, it is obvious that single size class flocculation models have a fundamental limitation in approximating a bimodal FSD. For example, a single-class flocculation model cannot estimate the collecting capability of matured macroflocs in a marine or estuarine system for fresh microflocs supplied from an upstream river (Winterwerp, 2002; Winterwerp and van Kesteren, 2004). From both engineering and ecological viewpoints, the particle collecting capability of a marine or estuarine system is very important for such activities as determining a dredging schedule for a navigation channel or evaluating the environmental and ecological impact of siltation. Effort to overcome the limitation of the contemporary single-class flocculation models recently started by developing a size class-based model and a distribution-based model, which can simulate a floc distribution of distinct size classes and an underlying continuous distribution of an average radius, respectively (Verney et al., in press; Maerz et al., in press). Beside those state-of-the-art flocculation models, a simplified two-class population balance equation (TCPBE) was developed and tested to overcome the drawbacks of the single size class flocculation models while maintaining simplicity. This TCPBE consists of two particle classes, a sizefixed microfloc and size-varying macrofloc, a minimum requirement to approximate bimodal flocculation. The sizefixed microflocs decrease in number concentration as they combine with macroflocs, but by definition they do not change size with time. The macroflocs change size with time, but again by definition the TCPBE yields only the average macrofloc size as a function of time. The two resulting floc sizes at a given time are used to approximate the two modal or average values of a true bimodal FSD. Based on a generic equation used in the crystallization process, the TCPBE was further modified for marine and estuarine sediments by discarding the nucleation process (forming solid nuclei with dissolved molecules) while maintaining the shear-induced breakage process (Jeong and Choi, 2003, 2004, 2005; Megaridis and Dobbins, 1990; Mueller et al., 2009a,b). The validity and applicability of the new TCPBE was tested with experimental data obtained from 1-D settling column tests (van Leussen, 1994), and compared with results from the simple SCPBE and the more elaborate multi-class PBE (MCPBE), which, at considerable computational expense, allows both floc size and number density to change with space and time so that an actual time-dependent bimodal floc size distribution results.

2.

Model description and numerical methods

One of the most realistic ways to simulate flocculation and non-homogeneous turbulent settling in a multi-dimensional space is by applying Population Balance Equations (PBE) within a Computational Fluid Dynamics (CFD) framework

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 1 3 1 e2 1 4 5

(Krishnappan and Marsalek, 2002; Fox, 2003; Prat and Ducoste, 2006). Following Prat and Ducoste (2006), a generic and general mathematical model for the PBEs in a multi-dimensional fluid field may be written as:
 vni ðIÞ vt
  v v v þ ðIIÞ ðux ni Þ þ uy ni þ ðuz ni Þ vz  vy   
vx (1) v vni v vni v vni þ þ ðIIIÞ Dtx Dty Dtz  vx  vy  vy vz vx vz v ws;i ni ðIVÞ ¼ ðAi þ Bi Þ  vz In Equation (1), ni ¼ n(x,y,z,Di,t) ¼ number concentration of the ith particle size class with a particle diameter of Di(i ¼ 1, 2, .imax; D1  Di  Dmax; for all D, n is called the population density function), x, y, z, t ¼ position and time, ux, uy, uz ¼ mean fluid velocities in the x, y and z directions, Dtx, Dtx, Dtz ¼ turbulent dispersion coefficients in the x, y and z directions, Ai and Bi ¼ growth and decay kinetics of ni by aggregation and breakage, respectively, and ws,i ¼ settling velocity of the ith particle class due to gravity. On the left-hand side of Equation (1), the respective terms in brackets represent the storage change (I), the particle mean advection (II), and the turbulent diffusion of flocs (III), while on the right-hand side, the source/sink terms (IV) represent the net effects of floc aggregation, breakage and settling due to gravity. The quantities depending on fluid variables (u and Dt) couple Equations (1) to the turbulent fluid dynamics equations. The aggregation and breakage kinetics, to be discussed later, form the core of the multi-dimensional PBEs because they largely determine floc size and settling velocity. In a 0-dimensional case, the aggregation and breakage kinetics (Ai þ Bi) can stand alone without the other space-dependent terms. The problem with the general form of this model is that the number (i) of discrete floc sizes can be huge, numbering in the thousands or millions. This problem is usually simplified by defining groups or classes of flocs that contain a range of discrete floc sizes.

2.1.

2133

required in order to quantify the many parameters involved and realize the full capability of the MCPBE formulation. The differential equations of the 1-dimensional version of the MCPBE without fluid advection are formulated as Equation (2) and tested with experimental data from a 1-dimensional settling column test (van Leussen, 1994):     v ws;i ni dni v vni  ¼ ðAi þ Bi Þ  Dtz vz dt vz vz Ai ¼ ni1

i2 X j¼1

ni

i1 X 1 2jiþ1 abi1;j nj þ abi1;i1 n2i1  ni 2ji abi;j nj 2 j¼1

ðmax XiÞ

abi;j nj

j¼i

Bi ¼ ai ni þ

ðmax XiÞ

bi;j aj nj

(2)

j¼iþ1

A multi-class population balance equation

Following the multi-class discretization scheme proposed by Hounslow et al. (1988) and Spicer and Pratsinis (1996), a set of floc size classes are defined such that each class contains all discrete flocs up to a maximum floc size that is two times the maximum floc size contained in the previous smaller class. Thus if d-sized floc monomers were linearly organized in this way, class 1 would contain flocs of size “d”, class 2 would contain flocs of size “2d”, class 3 would contain sizes “3d” and “4d”, class 4 would contain “5d” through “8d”, class 5 would contain “9d” through “16d”, and so on. Since the maximum floc size in class “i” increases as 2(i1), 30 mean classes will contain floc sizes varying from “d” to “229d”, which represents a growth factor of more than 537 million. However, the MCPBE still requires about 30 floc size classes and differential equations for spherical and porous flocs to cover from several micrometers to millimeters (according to the floc packing strategy of the fractal theory, Di ¼ d$(2i1)1/nf; 1 < nf < 3 for spherical and porous flocs) (Spicer and Pratsinis, 1996). Thus, the MCPBE will still have computational difficulties in higher dimensional problems. Also, much expensive experimental study is

Fig. 1 e Flocculation strategies of the SCPBE and TCPBE, representing the time- and space-dependent change of the floc size distributions (FSDs). t and x represent time and spatial coordinate and change from 0 to 1 in a flocculation process. NP and NF are the number concentration of microflocs and macroflocs in suspension, respectively, DF is the diameter of macroflocs, and NC represents the number of microflocs bound in a macrofloc as a floc size index.

2134

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 1 3 1 e2 1 4 5

Table 1 e Aggregation and breakage processes of the TCPBE, shown in the Peterson matrix. Process ( j ) Y

Component (i) /

Description

NF

NT ( ¼ NF  NC)

  1 NC 2 NC  1

  1 NC 2 NC  1

abPPNPNP

þ1

abPFNPNF

NP

(1)

Collision Microflocs

(2)

Collision Micro & Macroflocs

(3)

Collision Macroflocs

+



+

1

+

12

f (4)

  1 NC 2 NC  1

Process rate rj (/m3s1)

abFFNFNF

1-f þf$NC

Breakage Macroflocs

þ1

-f$NC

aNF

Nomenclature: NP ¼ Number concentration of microflocs in suspension (/m3); NF ¼ Number concentration of macroflocs in suspension(/m3); NT ¼ Number concentration of microflocs in macroflocs (/m3); NC ¼ Number of microflocs in a macrofloc (); a ¼ Collision efficiency efficiency; b ¼ Collision frequency frequency; aF ¼ Breakage kinetic function; f ¼ Fraction of microflocs generated by breakage.

In Equation (2), Ai and Bi represent aggregation and breakage kinetics of ni, respectively. Several empirical or theoretical factors or functions (a, b, a, and b) are incorporated into the aggregation and breakage kinetics. In the aggregation kinetics, the collision efficiency factor or the fraction of collisions that result in aggregation (0  a  1) describes the physicochemical properties of solid and liquid to cause inter-particle attachments, while the collision frequency factor or the rate at which particles of volumes Vi and Vj colloid (b) represents the mechanical fluid properties that induce inter-particle collisions. In experimental and modeling applications, the collision efficiency factor (a) is generally used as an application-specific fitting parameter and the collision frequency factor (b) is applied as a fixed theoretical function correlated with Brownian motion, shear rate, and differential settling. In the breakage kinetics, the breakage kinetic parameter (a) represents the speed of the breakage process and is generally applied as a shear- and size-dependent function. The breakage distribution function (b) represents the volume fraction of the fragments of size i coming from j-sized flocs. The binary breakage function, describing birth of two equallysized daughter fragments from one parent floc, is given by Equation (3) (Kusters, 1991; Jackson, 1995; Spicer and Pratsinis, 1996; Ding et al., 2006). ðmax XiÞ

bi;j aj nj ¼ bi;iþ1 aiþ1 niþ1 ¼ 2aiþ1 niþ1

(3)

j¼iþ1

In this research, an additional breakage parameter ( f ) was incorporated in the binary breakage model to calculate the mass fraction of residual microflocs generated by breakage of

macroflocs. The modified binary breakage model is given by Equation (4). ðmax P iÞ j1 ðmax XiÞ ¼ 2a2 n2 þ 2 2fajþ1 njþ1 for i ¼ 1 bi;j aj nj (4) j¼2 j¼iþ1 for i ¼ 2 to maxi ¼ 2ð1  f Þaiþ1 niþ1

Table 2 e Aggregation and breakage kinetic kernels and parameters of the SCPBE, MCPBE, and TCPBE. PBE

Aggregation kernel

SCPBE

b ¼ 16ð2DF Þ3 G

MCPBE

bij ¼ bBR;ij þ bSH;ij þ bDS;ij  1 1 bBR;ij ¼ 2kT 3m Di þ Dj ðDi þ Dj Þ

TCPBE

bSH;ij ¼ 16ðDi þ Dj Þ3 G

Breakage kernel

 P P ai ¼ Eb G DiDD Pq  mG  Fy =D2i

Kinetic constants a, Eb a, Eb, f

a, Eb, f

bDS;ij ¼ p4ðDi þ Dj Þ2 jwi  wj j Nomenclature: Di ¼ Diameter of a particle size class i (MCPBE); DP, DF 1=n ¼ Diameter of a microfloc/macrofloc (SCPBE, TCPBE); DF ¼ NC f DP (Fractal Theory); NC ¼ Number of microflocs bound in a macrofloc; nf ¼ Fractal Dimension; a ¼ Collision efficiency factor; bi,j ¼ Collision frequency function between size classes i and j; bBR ¼ Collision frequency function by Brownian motion; bSH ¼ Collision frequency function by fluid shear; bDS ¼ Collision frequency function by discrete settling; k ¼ Boltzmann’s constant; p, q ¼ Empirical parameters determined by experiments; T ¼ Absolute Temperature (K); m ¼ Absolute viscosity of the fluid; G ¼ (3/n)0.5 ¼ Shear rate (/s); 3 ¼ Kinametic energy dissipation rate; n ¼ Kinametic viscosity; wi ¼ Sedimentation velocity of particle size class i; ai ¼ Breakage kinetic function (Maggi, 2005); Eb ¼ Efficiency for the breakage process (Maggi, 2005); Fy ¼ Yield strength of flocs (1010 Pa) (Maggi, 2005); f ¼ Fraction of microflocs generated by breakage.

2135

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 1 3 1 e2 1 4 5

Table 3 e Parameters and initial conditions used in the best-quality simulation. Classification

Symbol

Kinetic parameters and physicochemical constants Agg/Brk Kinetics a Eb Fy DC f pa q Fluid Turbulence s 3b Dtzb Gb Sediment Property c NPTc rP DP nf a Initial Conditions (at t ¼ 0) Seeded Macroflocs MCPBE

SCPBE TCPBE

DF,0 FracF,0 n1,0 n4,0 ni,0 NF,0 NP,0 NF,0 NT,0

Value

Description

0.20 f 0.30 f 0.10 1.0e-4 1.0e-10 450 0.10 1.0 3.0 - nf 0.100 5.36e-5 1.07e-4 7.31 1.1 2.25eþ11 1600 18.0 2.0 4.0

Collision efficiency factor [-] (MCPBE f SCPBE f TCPBE) Efficiency factor for breakage [s0.5/m] Yield strength of flocs [Pa] Critical diameter for floc growth [mm] Fraction of microflocs by breakage [-] Empirical parameter of breakage kinetics Empirical parameter of breakage kinetics Frequency of the oscillating grid [/s] Kinametic energy dissipation rate [m2/s3] Vertical dispersion coefficient [s0.5/m] Shear rate [/s] Mass conc. of the tested sediment [g/L] No. conc. of microflocs [/m3] Density of microflocs [kg/m3] Diameter of microflocs [mm] Fractal dimension of macroflocs [-] Exponent of RichardsoneZaki eqn [Pa]

50 0.001 (1- FracF,0) NPT 23 (FracF,0) NPT 0 1.0 NPT (1- FracF,0) NPT (DF,0/DP)inf (FracF,0) (FracF,0) NPT

Diameter of seeded macroflocs [mm] Mass fraction of seeded macroflocs [e] No. conc. of 1st size class [/m3] No. conc. of seeded macroflocs [/m3] No. conc. of other size classes [/m3] No. conc. of primary flocs [/m3] No. conc. of microflocs [/m3] No. conc. of macroflocs [/m3] Total no. conc. of microflocs [/m3]

N PT

a p was set as 1.0 to narrow FSDs, instead of 0.5 of Winterwerp and van Kesteren (2004). pffiffiffiffiffiffiffi b 3 ¼ 127a30 s3 , Dz ¼ 0:19a20 s, and G ¼ 3=n (van Leussen, 1994), where, a0 ¼ 0.075 ¼ amplitude oscillating grid [m], n ¼ 1.10e-6 ¼ kinematic 2 viscosity [m /s]. c NPT ¼ c=ð0:167pD3P Þ=rS , assuming that a primary microfloc is spherical.

2.2.

A single-class population balance equation

The 1-D form of the SCPBE selected for study is formulated as Equations (5) and (6), (Winterwerp, 2002; Winterwerp and van Kesteren, 2004; Son and Hsu, 2008, 2009). The single-class floc size (DF) varies at different time and spatial points, because of transport and flocculation, but it is a single size value without any other floc size classes (Fig. 1). Thus, at a fixed time DF may be viewed as the mean floc size of the floc size distribution that is actually present at each point. For the 1-dimensional simulation of a settling column test, the sediment transport equation (vc/vt) and the SCPBE (vNF/vt) are solved in a coupled way, and the particle/floc diameter (DF) is updated with Equation (6) (Winterwerp and van Kesteren, 2004). The SCPBE is easy to solve and results in a robust numerical simulation, but it cannot even approximate bimodal flocculation and differential settling-induced flocculation because it has a single floc size class at each point. Also, only fluid shearinduced flocculation is considered.   dNF v vNF vðws;F NF Þ  ¼ ðAF þ BF Þ  Dtz dt vz vz vz ðAF þ BF Þ ¼ 12abFF N2F þ aNF

(5)

1=nf  r 3n DF ¼ fs s DP f NF c

(6)

In Equations (5) and (6), NF ¼ number concentration of flocs, a ¼ collision efficiency factor, b ¼ collision freguency factor, a ¼ breakage kinetic constant, c ¼ mass concentration of flocs, DF ¼ floc size, DP ¼ primary particle size, fs ¼ shape factor (for spherical particles, fs ¼ p/6), rs ¼ solid density.

2.3.

A two-class population balance equation

In contrast to the single-class PBE, the selected two-class PBE is able to approximate bimodal flocculation i.e. the fate of residual microflocs and aggregating macroflocs, each in a single-class size sense. Therefore, the 1-D TCPBE, formulated as Equations (7), tracks the number concentration of microflocs and macroflocs and the size of macroflocs as the time- and space-dependent variables (NP (t,z), NF (t,z), and DF (t,z)). This two-class PBE includes size-fixed microflocs and size-varying macroflocs, which describe primary building blocks and secondary agglomerates, respectively. The average size of macroflocs (DF (t,z)) varies at different times and spatial points because of transport and flocculation, but no other macrofloc size classes are allowed. However, it is important to note that the size of the microflocs (DP) in this model remain constant irrespective of time and space. The number of microflocs bound in a macrofloc (NC) is used as a floc size index in the TCPBE. Because this new size index becomes one for microflocs (NC ¼ 1) and an integer for macroflocs (NC ¼ i),

2136

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 1 3 1 e2 1 4 5

instead of a real number of linear or volumetric size indices (DP and DF), it yields simplicity in mathematical formulation and computation to the TCPBE. As shown in Fig. 1, three variables e the number concentrations of microflocs and macroflocs in suspension (NP and NF) and the size index of macroflocs (NC) e are unknown in time and space. Therefore, the TCPBE incorporates three coupled differential equations describing the time rate of change of: (1) the number concentration of microflocs in suspension (dNP/dt), (2) the number concentration of macroflocs in suspension (dNF/dt), and (3) the number concentration of microflocs bound in macroflocs (dNT/dt) (NT ¼ NC  NF) (Equation (7)). A more generic TCPBE was also derived by the moment conservation law of a floc size distribution (FSD) and explained in depth by Jeong and Choi (2003, 2004, 2005).   dNP v vNP vðws;P NP Þ  ¼ ðAP þ BP Þ  Dtz dt vz vz vz  C  abPF NP NF þ fNC aNF ðAP þ BP Þ ¼ 12abPP NP NP NNC 1   dNF v vNF vðws;F NF Þ  ¼ ðAF þ BF Þ  Dtz dt vz vz vz   1 1 1 ðAF þ BF Þ ¼ þ abPP NP NP  abFF NF NF þ aNF 2  NC  1 2  dNT v vNT vðws;T NT Þ  ¼ ðAT þ BT Þ  Dtz dt vz vz vz  1 C þ abPF NP NF  fNC aNF ðAT þ BT Þ ¼ þ2abPP NP NP NNC 1

The aggregation and breakage kinetic equations of the TCPBE are shown in the Peterson matrix for better understanding (Table 1) (Peterson, 1965). Each aggregation and breakage kinetic equation of a component (AP þ BP, AF þ BF, or AT þ BT) can be formulated by multiplying stoichiometric coefficients (nij) with a process rate (ri) and summing the P multiplied terms ðAi þ Bi ¼ nij rj Þ. The stoichiometric coeffij cients are shown in the third w fifth columns of Table 1 and the reaction rates are in the last column. The TCPBE includes four aggregation or breakage kinetic processes: (1) aggregation between microflocs, (2) aggregation between microflocs and macroflocs, (3) aggregation between macroflocs, and (4)

(7)

In Equation (7), the subscript P and F represent microfloc and macrofloc, respectively. f represents fraction of microflocs generated by breakage of macroflocs, whereas 1  f is fraction of smaller macroflocs by breakage of larger macroflocs. a ¼ collision efficiency factor, b ¼ collision freguency factor, and a ¼ breakage kinetic constant. The settling velocity of NT (ws,T) is equal to the settling velocity of NF (ws,F).

Fig. 2 e Plots of the sum of residual errors (SREs) versus collision efficiency factor (a) for the MCPBE (Triangles), SCPBE (Black Circles), and TCPBE (Gray Diamond). The SREs were calculated based on the difference between the measured and simulated concentration profiles while changing collision efficiency factor (a).

Fig. 3 e (a) Measured (symbols) and simulated solid concentration profiles of the best-quality simulations with the MCPBE, SCPBE, and TCPBE (lines) at time [ 450, 900, 1350, 1800, and 3600 s. (b) The floc size distributions (FSDs) of the best-quality simulations with the MCPBE, SCPBE, and TCPBE. They were captured at a different water height of 1, 2, 3, or 4 m at t [ 1800 s, and are plotted in the four separate figures.

2137

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 1 3 1 e2 1 4 5

Fig. 4 e Measured floc size distributions (FSDs) (shaded area), the two peaks averaged from the measured FSD (triangle), and simulated FSDs of the best-quality simulations with the MCPBE (dashed line), SCPBE (diamond), and TCPBE (dark circle) at time [ 180, 250, 490, 800, 1280, and 2180 s. FSDs were measured for settling flocs right above the concentrated bottom layer, to avoid the effect of floc deposition (van Leussen, 1994). Similarly, Simulated FSDs were obtained at 300 mm above the bottom of the settling column.

breakage of macroflocs. Again, an additional breakage parameter ( f ) was incorporated into the breakage process to calculate the mass fraction of microflocs generated by breakage of macroflocs. The TCPBE obtains higher computational efficiency than MCPBEs by omitting many floc size classes. In contrast to the SCPBE, the TCPBE is able to simulate interactions between microflocs and macroflocs. Thus, the TCPBE maintains much simplicity compared to the MCPBE but significantly more capability than the SCPBE.

capable of simulating flocculation in highly concentrated suspension in the hindered regime. In detail, the shearinduced breakage could not limit the infinite size growth of macroflocs in the highly concentrated bottom layer of an estuary or a settling column. Thus, an additional empirical parameter (DC; critical diameter) was introduced to prevent unrealistic floc size growth. Above the critical diameter (DC), the breakage rate was set sufficiently high to break all flocs.

2.5. 2.4. Aggregation and breakage kernels and kinetic parameters Table 2 summarizes the aggregation and breakage kernels and kinetic parameters of the MCPBE, SCPBE, and TCPBE. The aggregation and breakage kernels which are commonly used in marine and estuarine environments were adopted again in this research (van Leussen, 1994; Jackson, 1995; Winterwerp, 2002; Winterwerp and van Kesteren, 2004; Maggi, 2005). Noteworthy is that the MCPBE and TCPBE have the aggregation kernels by Brownian motion, differential settling, and fluid shear. However, the SCPBE has only the aggregation kernel by fluid shear (Winterwerp, 2002; Winterwerp and van Kesteren, 2004; Son and Hsu, 2008, 2009). For the breakage kernel, the shear-induced breakage kinetic function was adopted for all the PBEs (Winterwerp, 2002; Winterwerp and van Kesteren, 2004; Maggi, 2005). However, the contemporary PBEs and their kinetic kernels and parameters are not

Floc settling equations

The modified Stokes equation, including the use of fractal theory for floc packing and shaping and Schiller’s equation for a particle drag effect, was used to calculate the floc settling velocity (ws,i) (Schiller, 1932). The RichardsoneZaki equation was used to calculate the correction factor (FHS) for hindered settling occurring in the highly concentrated bottom layer (Equations (8) and (9)) (Richardson and Zaki, 1954; Toorman, 1999; Winterwerp, 2002; Winterwerp and van Kesteren, 2004). The empirical parameters, the fractal dimension (nf) and the exponent of the RichardsoneZaki equation (a), were reported to be 1.7e2.3 and 2.5e5.5, respectively, for estuarine and marine sediments, so they were fixed at 2.0 and 4.0 in this research (Winterwerp and van Kesteren, 2004).

n 1

ws;i ¼ FHS

Di f 1 ðrs  rw Þg 3nf DP 18 m 1 þ 0:15Re0:687 i

! (8)

2138

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 1 3 1 e2 1 4 5

FHS ¼ ð1  fÞa

(9)

In Equations (8) and (9), rs ¼ particle density, rw ¼ fluid density, g ¼ gravitational acceleration, m ¼ fluid viscosity, FHS ¼ correction factor for hindered settling, F ¼ volumetric concentration of flocs (m3 Flocs/m3) calculated by multiplying the volumetric size of a floc (m3) and the number concentration of flocs (/m3), and Rei ¼ Reynolds number of a particle or floc.

2.6.

Fig. 5 e (a) Sensitivity of the solid concentration profiles of the TCPBE to the collision and breakage efficiency factors (a and Eb). Symbols represent measured solid concentration profiles at time [ 450, 900, 1800, and 3600 s. Lines represent simulated profiles with different collision and breakage efficiency factors (a and Eb). (b) sensitivity of the size and mass fraction of macroflocs to the collision efficiency factor (a) and (c) sensitivity to the breakage

Experimental and numerical methods

The 1-dimensional settling column test (van Leussen, 1994) provided both of the experimental indices, (1) solid concentration profiles and (2) FSDs (i.e. floc size and mass faction), which were used for the comparative study between the PBEs and the sensitivity analyses of the TCPBE. The settling column had a 0.29 m diameter and a height of 4.25 m and was placed in a thermostatically controlled water bath. The homogeneous and isotropic intensity of a turbulence field was generated by means of axial oscillations of a grid installed inside the settling column (see van Leussen, 1994 for details). The intensity of a turbulence field was controlled by the alternating frequency of the oscillating grid (van Leussen, 1994; Maggi, 2005). To initiate the experiment, the settling column was filled with a mixture consisting of mud from the Ems estuary in the northern part of the Netherlands and salt water. This water/sediment mixture was adjusted to have a solids concentration of 1000 mg/L and a salinity of 32. The mixture was then homogenized at a very high oscillating speed of the inner grids (s ¼ 4 Hz, G ¼ 1848/s) until all the aggregates were destroyed down to the size of the initial microflocs. At the beginning of the test, the turbulent shear rate (G) was set at 7.31/s by adjusting the oscillating speed of the inner grids (s) to 0.1 Hz (Table 3), and the experimental indices e solid concentration profiles and FSDs e were measured as a function of time. FSDs were measured with a Malvern 2600 particle sizer (Fraunhofer diffraction) for flocs collected near the bottom of the settling column. Detailed experimental methods and techniques of the 1-D settling column test may be found in van Leussen (1994). The operator splitting algorithm and Gauss-Seidel iteration were applied to solve the nonlinear partial differential equations of the 1-dimensional PBEs (Equations (2), (5), (7)). According to the operator splitting algorithm, the transport (advectionediffusion) and source/sink (reaction-settling) operators were decoupled and sequentially solved in each time step, to cope with the complexity and nonlinearity of the PBEs (Aro et al., 1999; Winterwerp, 2002; Winterwerp and van Kesteren, 2004). This decoupling strategy separating the complex reaction-settling operator may be useful for solving a large-scale multi-dimensional flocculation system. Each operator was solved using Gauss-Seidel iteration, in which the dependent variables are updated iteratively until satisfying the convergence limit in each time step. Then the converged

efficiency factor (Eb). Each symbol represents the size and mass fraction of macroflocs, which were measured (the cross-hair symbol) and simulated (the filled symbol) at time [ 180, 250, 490, 800, 1280, and 2180 s.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 1 3 1 e2 1 4 5

Fig. 6 e Plots of floc diameter versus settling velocity. The diamond symbols represent the measured diameter and settling velocity of a floc in settling column tests (van Leussen, 1994) and the lines represent the simulated data with the modified Stokes equation (Winterwerp, 2002). The schematic diagrams illustrate the floc packing strategies of microflocs and macroflocs with individual clay particles. Microflocs form by direct contact between clay minerals whereas macroflocs form by loose agglomeration between microflocs and organic matters (after van Leussen (1994)). rp represents the density of microflocs and both nf and nf,macro are the fractal dimension of macroflocs. The first and second numbers in the parentheses are the size and density of a floc, respectively.

values are used as the seeding values of the next time step (Other iteration methods should also be applicable.). All the simulations were done with the standard values of the kinetic and physicochemical factors and the initial conditions. The standard values of the kinetic and physicochemical factors were obtained from values found in previous studies (Jackson, 1995; Maggi, 2005; Winterwerp and van Kesteren, 2004), and the initial conditions were those measured in the settling column test (van Leussen, 1994) (Table 3). To obtain optimum simulations of the column data, the collision efficiency factor (a) was used as the only adjustable fitting parameter in the model-data fitting analyses, while the other parameters were simply fixed at standard values found in the literature. This was done to reduce the complexity and uncertainty caused by many highly interactive parameters. The simulation quality was quantified by the sum of residual errors (SRE) between simulated and measured sediment concentrations (Csim and Cexp) (Equation (10)) (Berthouex and Brown, 1994), with the minimum SRE defining the optimum simulation for the selected parameters. SRE ¼

n n X X  2 ðerrori Þ2 ¼ Cexp;i  Csim;i i¼1

(10)

2139

3.

Results and discussion

3.1.

Comparison between the MCPBE, SCPBE, and TCPBE

Solutions of all PBEs produced U-shaped SRE curves with changes in the collision efficiency factor (a) (Fig. 2). Generally, as the fitting parameter (a) increases, the solid concentration profiles move downward, become the best fit curve by definition at the minimum SRE (Fig. 3), and further collapse downward, because the higher collision efficiency factor (a) increased the down-gradient flux of sediment by increasing floc size and settling velocity. The SRE of the TCPBE was slightly lower than the SRE of the MCPBE. However, by comparing the best fit solid concentration profiles the difference between the simulated concentration profiles of the TCPBE and MCPBE is shown to be relatively small (Fig. 3). Considering the sensitivity and uncertainty of contemporary flocculation models and experimental techniques (Fettweis, 2008), both the TCPBE and MCPBE are equally capable of simulating flocculation in the system tested using the data fitting procedure described. However, the SCPBE was obviously less capable by obtaining about 2e3 times higher SREs than the MCPBE and TCPBE. This represents the low-quality simulations of the SCPBE, which might be caused by the incapability of the SCPBE for simulating interactions between microflocs and macroflocs. The weaknesses of the SCPBE will be discussed further in the following paragraphs.

3.1.1.

Solid concentration profiles

Fig. 3(a) shows the simulated (best-quality) and measured solid concentration profiles in the settling column test. Measured solid concentration profiles had low concentrations near the water surface, approached a concentration just above 1 g/L along the water depth and then a significant increase in solids concentration well above 1 g/L near the water bottom. Compared to the similar results of the MC and TC PBEs, the SCPBE curve was more above the data at early times and below the data at large times. This illustrates that the SCPBE was less able to fit the data compared to the MCPBE and TCPBE, possible due to the single size class limitation of the SCPBE. Fig. 3(b) illustrates the potential superiority of the MCPBE in that bimodal floc size classes are actually calculated. However, the TCPBE could simulate two size peaks and the mass change of micro- and macro-flocs along the water depth as well as the MCPBE as it is currently formulated (Fig. 3). Considering that the TCPBE requires only three differential equations, its simplicity and capability are promising for large-scale simulation of a complicated marine and estuarine system. In contrast, the SCPBE with two differential equations could neither simulate well the early or late concentration data nor the bimodal mass changes because it simply tracks a single floc size class. One would expect this limitation to be even more prominent for a large-scale marine and estuarine system with a massive influx of fresh microflocs because the fate of fresh microflocs could not be tracked.

i¼1

After the optimum a was selected, the sensitivity of the TCPBE to the kinetic and physicochemical parameters was evaluated by varying these parameters about their standard values.

3.1.2.

Floc size distribution details

Both the MC and TC PBEs could better simulate the size growth of flocs (DF) in the initial growth phase (t  490 s) and the final

2140

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 1 3 1 e2 1 4 5

bimodal FSD in the steady state (t  800 s) than the SCPBE (Fig. 4). (However, only the MCPBE generated a true bimodal size distribution.) For example, the TCPBE could track the size and mass fraction of microflocs and macroflocs while approximating the bimodal FSD in the initial growth phase and the steady state; the SCPBE approximates only a unimodal FSD. In Fig. 4, the “mean” floc sizes and mass fractions simulated with the TCPBE (dark sphere) could well follow the two peaks averaged from the measured FSD (triangle). Different to the MCPBE developing two peaks in a continuous manner, the TCPBE developed two sharp peaks of micro- and macroflocs. However, the SCPBE maintained a single sharp peak at a certain floc size in the initial growth phase and the steady state. Furthermore, the SCPBE could not correctly estimate the settling flux of the bimodal FSD, which is the key index for estimating sediment deposition and transport in a marine or P estuarine system. The settling flux ( ws;i  Ci g/m2/sec) of the measured bimodal FSD was estimated to be 0.912 for 1 g/L solid concentration in the steady state (t ¼ 2180 s). The simulated settling fluxes were estimated to be 0.938, 1.278, and 0.953 for the MCPBE, SCPBE, and TCPBE respectively, Noteworthy is that the simulated settling flux of the SCPBE had 40% error against the measured settling flux while the settling fluxes of the MCPBE and TCPBE had errors less than 5%,,The SCPBE could not simulate 20% mass fraction of microflocs and consequently generated 40% error of the settling flux. This significant error propagation might be caused by the difference between the settling velocities of microflocs and macroflocs (z0.106 and 1.14 mm/s, respectively) (Fettweis, 2008).

3.2. Model sensitivity to kinetic and physicochemical factors After determining the “best fit to data” parameter values, the model sensitivity to kinetic and physicochemical parameter variations around the “best values” was investigated. Among numerous parameters shown in Table 3, (1) flocculation kinetic parameters e the aggregation and breakage efficiency factors (a and Eb), (2) floc structural parameters e the density of microflocs (rp) and the fractal dimension of macroflocs (nf), (3) fate of broken macroflocs e the mass fraction of microflocs generated by breakage of macroflocs ( f ), (4) initial conditions of seeded macroflocs for sweep flocculation e the size and mass fraction of seeded macroflocs (DF,0 and FracF,0) were selected and model sensitivity tested. This was done by calculating the degree of deviation of simulations from the measured data e solid concentration profiles (e.g. Fig. 5(a)) and the mean size and mass fraction of macroflocs (e.g. Fig. 5(b) and (c)) which were calculated with the measured FSDs (Fig. 4). Fig. 7 e (a) Sensitivity of the solid concentration profiles of the TCPBE to the density of microflocs (rp) and the fractal dimension of macroflocs (nf). Symbols represent measured solid concentration profiles at time [ 450, 900, 1800, and 3600 s. Lines represent simulated profiles with different density of microflocs (rp) and fractal dimension of macroflocs (nf). (b) sensitivity the size and mass fraction of

macroflocs to the density of microflocs (rp) and (c) sensitivity to the fractal dimension of macroflocs (nf). Each symbol represents the size and mass fraction of macroflocs, which were measured (the cross-hair symbol) and simulated (the filled symbol) at time [ 180, 250, 490, 800, 1280, and 2180 s.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 1 3 1 e2 1 4 5

a

profiles upward (Fig. 5(b) and (c)). The interdependency between the collision and breakage efficiency factors in determining the overall flocculation kinetics agrees with the finding of Verney et al. (in press) and Maerz et al. (in press). However, over time the respective changes of the collision and breakage efficiency factors (a and Eb) determined the size and mass fraction of macroflocs in a different manner. A change of the collision efficiency factor (a) differentiated the size and mass fraction of aggregating macroflocs in the initial growth phase (t  490 s) (see the circle, triangle, and square symbols), but made a relatively small effect on the size and mass fraction of macroflocs at steady state (t  800 s) (see the diamond, triangle, and hexagonal symbols in Fig. 5(b)). In contrast, the breakage efficiency factor (Eb) could differentiate the size and mass fraction of macroflocs in the steady state (t  800 s). Thus, the collision efficiency factor (a) of the TCPBE determined the size and mass fraction of macroflocs mainly in the initial growth phase as a floc growth accelerator, while the breakage efficiency factor (Eb) was most effective near steady state as a floc-growth limiter (Equation (7) and Table 1).

b

3.2.2.

Fig. 8 e (a) Sensitivity of the solid concentration profiles of the TCPBE to the fraction of microflocs generated by breakage of macroflocs ( f ). Symbols represent measured solid concentration profiles at time [ 450, 900, 1800, and 3600 s. Lines represent simulated profiles with different fraction of microflocs generated by breakage of macroflocs ( f ). (b) Sensitivity of the size and mass fraction of macroflocs to the fraction of microflocs generated by breakage of macroflocs ( f ). Each symbol represents the size and mass fraction of macroflocs, which were measured (the cross-hair symbol) and simulated (the filled symbol) at time [ 180, 250, 490, 800, 1280, and 2180 s.

3.2.1.

2141

Flocculation kinetic parameters

The collision and breakage efficiency factors (a and Eb) cause in increase and decrease in floc size growth, respectively. For example, a higher collision efficiency factor (a ¼ 0.15) or a lower breakage efficiency factor (Eb ¼ 0.7  104) both moved the solid concentration profiles downward (Fig. 5(a)). Either an increase in a or a decrease in Eb increased the size and mass fraction of macroflocs (Fig. 5(b) and (c)) which in turn increased the floc settling velocity. In contrast, a collision efficiency factor decrease (a ¼ 0.05) or a breakage efficiency factor increase (Eb ¼ 2.5  104) moved the solid concentration

Floc structural parameters

Unlike unimodal flocculation, bimodal flocculation involves microflocs and macroflocs. Thus, the TCPBE requires two distinct structures of less-porous and hard microflocs and highly-porous and floppy macroflocs (Fig. 6) (van Leussen, 1994; Winterwerp and van Kesteren, 2004). The structure of primary microflocs is characterized by their size and density (Dp and rp), while the structure of aggregating macroflocs is determined by microflocs density (rp) and macrofloc fractal dimension (nf) (Table 2). Therefore, microfloc density (rp) and macrofloc fractal dimension (nf) were selected as the floc structural parameters for sensitivity analysis. Standard values for rp and nf (rp ¼ 1600 kg/m3 and nf ¼ 2.0, respectively) were selected from the literatures (Winterwerp and van Kesteren, 2004; Maggi, 2005). The standard values and the modified settling equations (Equations 8, 9) proved their validity by showing that simulated settling velocities were reasonably matched to measured velocities for different floc sizes (Fig. 6). A higher microfloc density (rp ¼ 1800 kg/m3) or a higher macrofloc fractal dimension (nf ¼ 2.10) increased both the density of macroflocs (rF) and the aggregation kinetics, and so moved the solid concentration profiles downward by increasing the size and settling velocity of macroflocs (Fig. 7 (a)). In contrast, a lower density for microflocs (rp ¼ 1450 kg/ m3) or a lower fractal dimension of macroflocs (nf ¼ 1.90) moved the solid concentration profiles upward by decreasing the density of macroflocs and the aggregation kinetics. However, the respective changes of microfloc density (rp) and macrofloc fractal dimension (nf) occurred in different ways (Fig. 7(b) and (c)). For example, the effect of the fractal dimension of macroflocs (nf) was more weighted on larger macroflocs while approaching steady state (t  490 s) (Fig. 7 (b)). Considering that the fractal theory is based on an exponential relation between the sizes of microflocs and macroflocs (DP and DF Table 2 and Equation 11), the effect of the fractal dimension of macroflocs (nf) should be exponential while increasing the macrofloc size. However, the density of microflocs (rp) made rather a consistent effect on the size and mass fraction of macroflocs in the entire size range (Fig. 7(c))

2142

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 1 3 1 e2 1 4 5

because of the linear relation between the densities of microflocs and macroflocs (rp and rF) (Equation 11). The sensitivity to the floc structural parameters was not as significant as the sensitivity in simulation of Maerz et al. (in press), in which a 0.2 variation of fractal dimension doubled an average floc size. The smaller sensitivity to the floc structural parameters might occur in the steady flow condition of the 1-D settling column than in the unsteady flow condition of the shear-varying reactor (Maerz et al., in press).
3nf DP (11) rF ¼ rw þ ðrP þ rw Þ DF

3.2.3.

Distribution of fragmented flocs

Sediments in the settling column still had a 20% mass fraction of microflocs while approaching the steady state (t  800 s Fig. 4). This residual mass fraction of microflocs was hypothesized to occur by fragmentation of macroflocs to microflocs. The fragmentation process was formulated with the additional parameter of the fraction of microflocs generated by breakage of macroflocs ( f ), and incorporated into the TCPBE as shown in Equation (7) and Table 1. The fragmentation process of the TCPBE has a similar theoretical background and mathematical formula to the one of the size class-based flocculation model (Verney et al., in press). The TCPBE with the fragmentation process ( f ¼ 0.1) could better simulate the long tail of the solid concentration profile at the last measuring time (t ¼ 3600 s Fig. 8(a)) and the residual mass fraction of microflocs (¼ 1  {the mass fraction of macroflocs}) in the steady state (t  800 s Figs. 4 and 8(b)). However, without a fraction of microflocs generated by breakage of macroflocs ( f ¼ 0), the long solid tail at t ¼ 3600 s disappeared near the water surface, and the mass fractions of microflocs and macroflocs became zero and one, respectively, in the steady state, because all microflocs agglomerated to macroflocs. In contrast, a higher fraction of microflocs generated by breakage of macroflocs ( f ¼ 0.3) reduced the down-gradient flux of solid concentration profiles by decreasing the mass fraction of macroflocs but increasing the mass fraction of microflocs (Fig. 8). In fact, the fraction of microflocs generated by breakage of macroflocs ( f ) could determine both the solid concentration profiles and the size and mass fraction of macroflocs. Considering that the breakage and fragmentation processes are highly dependent on the fluid shear rate (G Table 2), the fraction of microflocs generated by breakage of macroflocs ( f ) should be more important for predicting the fate of sediments (e.g. solid concentration profiles and FSDs) under the varying shear rate in a marine and estuarine system (e.g. Verney et al., in press) than in the well-controlled settling column test studied herein.

3.2.4.

Initial conditions of seeded macroflocs

Sweep flocculation describes flocculation in which seeded macroflocs enmesh surrounding particles or microflocs in Fig. 9 e (a) Sensitivity of the solid concentration profiles of the TCPBE to the size and mass fraction of seeded macroflocs (DF,0 and FracF,0). Symbols represent measured solid concentration profiles at time [ 450, 900, 1800, and 3600 s. Lines represent simulated profiles with different size and mass fraction of seeded macroflocs (DF,0 and FracF,0). (b) Sensitivity of the size and mass fraction of

macroflocs to the size of seeded macroflocs (DF,0) and (c) sensitivity to the mass fraction of seeded macroflocs (FracF,0). Each symbol represents the size and mass fraction of macroflocs, which were measured (the cross-hair symbol) and simulated (the filled symbol) at time [ 180, 250, 490, 800, 1280, and 2180 s.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 1 3 1 e2 1 4 5

a fluid shear field and enhance the floc growth and the bimodality of an FSD. Sweep flocculation is commonly applied in water treatment, and enhances flocculation by metalhydroxide macroflocs collecting small impure particles or microflocs (Gregory, 2006). In a marine or estuarine system, sweep flocculation can be similarly defined as flocculation of sticky organic-bound macroflocs colleting nearby microflocs. Considering the sudden appearance of a substantial amount of large macroflocs without a sequential floc growth in the initial growth phase (Fig. 4), sweep flocculation might also occur in the settling column test. A very small amount of macroflocs surviving the preliminary breaking and homogenizing step was hypothesized to function as seeded macroflocs for sweep flocculation. Therefore, different sizes (DF,0 ¼ 20, 50 and 100 mm) and mass fractions (FracF,0 ¼ 0.001, 0.1 and 10%) of seeded macroflocs were tested as initial conditions in simulations for investigating sweep flocculation. The solid concentration profiles remained almost consistent, irrespective of the change of the initial conditions of seeded macroflocs (DF,0 and FracF,0) (Fig. 9(a)). However, the size and mass fraction of macroflocs were changed significantly by the initial size of seeded macroflocs (DF,0), especially in the initial growth phase (t  490 s) but not in the steady state (t  800 s). For example, macroflocs slowly grew up to the final size of macroflocs of the steady state with the smaller seeded macroflocs (DF,0 ¼ 20 mm), but they rapidly grew with the larger seeded macroflocs (DF,0 ¼ 100 mm) (Fig. 9(b)). However, these different initial macrofloc-growth patterns had a minor effect on the solid concentration profiles. Sedimentation continued even after the macrofloc size became constant at steady state (t  800 s), and thus was more dependent on the size and settling velocity of macroflocs. Therefore, as long as the size and mass fraction of macroflocs remained relatively constant a steady state, like the case shown in Fig. 9(b) and (c), the solid concentration profiles were close each other due to similar rates of sedimentation. However, a very small amount of seeded macroflocs (e.g. FracF,0 ¼ 0.001% and DF,0 ¼ 50 mm) is necessary to correctly simulate the sweep flocculation and bimodal FSDs of aggregating macroflocs in the TCPBE.

4.

Conclusion and recommendation

The two-class PBE was shown to be the simplest model that is capable of approximating bimodal flocculation of marine and estuarine sediments. In contrast to the SCPBE, the TCPBE can simulate bimodal interactions between micro- and macroflocs and thus estimate the collector capability of a marine or estuarine system for fresh microflocs supplied by an upstream river. Compared to the MCPBE, the TCPBE incorporates simplicity and computational efficiency at the expense of producing non-detailed floc size distributions. In an average floc size sense, however, it can be used to simulate large-scale sediment transport in marine and estuarine environments in a practical manner. Against the recent class-based and distribution-based models (Verney et al., in press and Maerz et al., in press), the TCPBE has the simplicity to require less computational cost and the capability to simulate bimodal flocculation, respectively, and is easy for marine engineers or scientists to adopt it without numerical and mathematical

2143

background. For example, the TCPBE can be easily solved by commercial or in-house differential equation solvers and thus free users from effort for programming. Furthermore, the TCPBE allows one to include the effect of additional biological and physicochemical processes on flocculation, such as interactions between micro-organisms and inorganic flocs or adsorption of natural organic matter with ease and flexibility (Maggi, 2009). Therefore, the TCPBE may be used as a practical model for investigation of the highly complicated fate of marine and estuarine sediments by incorporating other coupled equations of micro-organisms, organic matter, and so on. While the MCPBE may be ultimately superior in simulating these coupled processes and producing realistic floc size distributions, the experimental and computational work required to realize such superiority is significant. The two-class PBE as well as the other PBEs still require fine adjustment for their aggregation and breakage kinetics. Wellcontrolled flocculation experiments may be required to find realistic kinetic and physicochemical parameters. Only in the last decade, have various investigators began using PBEs for simulating flocculation in a marine and estuarine systems. Thus, experimental data, which can be used for estimating kinetic and physicochemical parameters of the PBEs, are still limited. A serious bias may sometimes occur by extrapolating the PBEs out of the boundary of experimental data. In fact, intensive investigation on the aggregation and breakage kinetics will be required for improved application of the PBEs in marine and estuarine systems.

Acknowledgment The authors would like to acknowledge the Flemish Science Foundation (FWO Vlaanderen) for funding the FWO project no. G.0263.08. Wim van Leussen kindly allowed the authors to use the experimental data of his PhD dissertation.

references

Aro, C., Rodrigue, G., Rotman, D., 1999. A high performance chemical kinetics algorithm for 3-D atmospheric models. The International Journal of High Performance Computing Applications 13 (1), 3e15. Berthouex, P., Brown, L., 1994. Statistics for Environmental Engineers. Lewis Publishers, Boca Raton, FL. Burd, A., Jackson, G., 2002. Modeling steady-state particle size spectra. Environmental Science and Technology 36, 323e327. Chen, M., Wartel, S., Temmerman, S., 2005. Seasonal variation of floc characteristics on tidal flats, the Scheldt estuary. Hydrobiologia 540, 181e195. Curran, K., Hill, P., Milligan, T., 2002. Fine-grained suspended sediment dynamics in the Eel River flood plume. Continental Shelf Research 22, 2537e2550. Curran, K., Hill, P., Milligan, T., Cowan, E., Syvitski, J., Konings, S., 2004. Fine-grained sediment flocculation below the Hubbard Glacier meltwater plume, Disenchantment Bay, Alaska. Marine Geology 203, 83e94. Curran, K., Hill, P., Milligan, T., Mikkelsen, O., Law, B., Durrieu de Madron, X., et al., 2007. Settling velocity, effective density, and mass composition of suspended sediment in a coastal bottom

2144

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 1 3 1 e2 1 4 5

boundary layer, Gulf of Lions, France. Continental Shelf Research 27, 1408e1421. Ding, A., Hounslow, M., Biggs, C., 2006. Population balance modelling of activated sludge flocculation: investigating the size dependence of aggregation, breakage and collision efficiency. Chemical Engineering Science 61, 63e74. Droppo, I., Leppard, G., Liss, S., Milligan, T., 2005. Flocculation in Natural and Engineered Environmental Systems. CRC Press, Boca Raton, FL, USA. Li, Y., Wolanski, E., Xie, Q., 1993. Coagulation and settling of suspended sediment in the Jiaojiang river estuary, China. Journal of Coastal Research 9 (2), 390e402. Li, B., Eisma, D., Xie, Q., Kalf, J., Li, Y., Xia, X., 1999. Concentration, clay mineral composition and Coulter counter size distribution of suspended sediment in the turbidity maximum of the Jiaojiang river estuary, Zhejiang, China. Journal of Sea Research 42, 105e116. Fettweis, M., 2008. Uncertainty of excess density and settling velocity of mud flocs derived from in situ measurements. Estuarine. Coastal and Shelf Science 78 (2), 426e436. Fox, R., 2003. Computational Models for Turbulent Reacting Flows. Cambridge University Press, Cambridge UK. Gregory, J., 2006. Particles in Water: Properties and Processes. CRC Press, Boca Raton, FL, USA. Hill, P., Milligan, T., Rockwell Geyer, W., 2000. Controls on effective settling velocity of suspended sediment in the Eel River flood plume. Continental Shelf Research 20, 2095e2111. Hounslow, M., Ryall, R., Marshall, V., 1988. A discretized population balance for nucleation, growth, and aggregation. AiChE Journal 34 (1), 1821e1832. Jackson, G., 1995. Comparing observed changes in particle size spectra with those predicted using coagulation theory. DeepSea Research II 42 (1), 159e184. Jackson, G., Logan, B., Alldredge, A., Dam, A., 1995. Combining particle size spectra from a mesocosm experiment measured using photographic and aperture impedance (Coulter and Elzone) technique. Deep-Sea Research 42 (1), 139e157. Jeong, J., Choi, M., 2003. A simple bimodal model for the evolution of non-spherical particles undergoing nucleation, coagulation and coalescence. Journal of Aerosol Science 34, 965e976. Jeong, J., Choi, M., 2004. A bimodal moment model for the simulation of particle growth. Journal of Aerosol Science 35, 1071e1090. Jeong, J., Choi, M., 2005. A bimodal particle dynamics model considering coagulation, coalescence and surface growth, and its application to the growth of titania aggregates. Journal of Colloid and Interface Science 281, 351e359. Krishnappan, B., Marsalek, J., 2002. Modelling of flocculation and transport of cohesive sediment from an on-stream stormwater detention pond. Water Research 36, 3849e3859. Kusters, K., 1991. The Influence of Turbulence on Aggregation of Small Particles in Agitated Vessels. Eindhoven University of Technology, the Netherlands. van Leussen, W., 1994. Estuarine Macroflocs: Their Role in Finegrained Sediment Transport. Universiteit van Utrecht, the Netherlands. Maerz, J., Verney, R., Wirtz, K., Feudel, U. Modeling flocculation processes: intercomparison of a size class-based model and a distribution-based model. Continental Shelf Research, in press, corrected proof. doi:10.1016/j.csr.2010.05.011. Maggi, F., 2005. Flocculation Dynamics of Cohesive Sediment. Technishce Universiteit Delft, the Netherlands. Maggi, F., 2009. Biological flocculation of suspended particles in nutrient-rich aqueous ecosystems. Journal of Hydrology 376, 116e125. Manning, A., Bass, S., 2006. Variability in cohesive sediment settling fluxes: observations under different estuarine tidal conditions. Marine Geology 235, 177e192.

Manning, A., Bass, S., Dyer, K., 2006. Floc properties in the turbidity maximum of a mesotidal estuary during neap and spring tidal conditions. Marine Geology 235, 193e211. Manning, A., Friend, P., Prowse, N., Amos, C., 2007a. Estuarine mud flocculation properties determined using an annular mini-flume and the LabSFLOC system. Continental Shelf Research 27, 1080e1095. Manning, A., Martens, C., de Mulder, T., Vanlede, J., Winterwerp, J., Ganderton, P., et al., 2007b. Mud floc observations in the turbidity maximum zone of the Scheldt estuary during neap tides. Journal of Coastal Research SI 50, 832e836. Megaridis, C., Dobbins, R., 1990. A bimodal integral solution of the dynamic equation for an aerosol undergoing simultaneous particle inception and coagulation. Aerosol Science and Technology 12, 240e255. Mietta, F., Chassagne, C., Winterwerp, J., 2009. Shear-induced flocculation of a suspension of kaolinite as function of pH and salt concentration. Journal of Colloid and Interface Science 336, 134e141. Mietta, F., Chassagne, C., Manning, A., Winterwerp, J. Influence of shear rate, organic matter content, pH and salinity on mud flocculation. Ocean Dynamics, PECS 2008 Special Issue, in press. Mikkelsen, O., Pejrup, M., 2001. The use of a LISST-100 laser particle sizer for in-situ estimates of floc size, density and settling velocity. Geo-Marine Letters 20, 187e195. Mikkelsen, O., Hill, P., Milligan, T., 2006. Single-grain, microfloc and macrofloc volume variations observed with a LISST-100 and a digital floc camera. Journal of Sea Research 55, 87e102. Mueller, M., Blanquart, G., Pitsch, H., 2009a. A joint volumesurface model of soot aggregation with the method of moments. Proceedings of the Combustion Institute 32, 785e792. Mueller, M., Blanquart, G., Pitsch, H., 2009b. Hybrid Method of Moments for modeling soot formation and growth. Combustion and Flame 156, 1143e1155. Orange, D., Garcia-Garcia, A., Lorenson, T., Nittrouer, C., Milligan, T., Miserocchi, S., et al., 2005. Shallow gas and flood deposition on the Po Delta. Marine Geology 222-223, 159e177. Perianez, R., 2005. Modelling the transport of suspended particulate matter by the Rhone River plume (France). Implications for pollutant dispersion. Environmental Pollution 133, 351e364. Peterson, E., 1965. Chemical Reaction Analysis. Prentice Hall, Englewood Cliffs, NJ, US. Prat, O., Ducoste, J., 2006. Modeling spatial distribution of floc size in turbulent processes using the quadrature method of moment and computational fluid dynamics. Chemical Engineering Science 59, 685e697. Richardson, J., Zaki, W., 1954. Sedimentation and fluidisation, Part I. Transactions of the American Institute of Chemical Engineers 2, 35e53. van Rijn, L., 1984. Sediment transport. Part II: suspended laod transport. Journal of Hydraulic Engineering 110 (11), 1613e1641. van Rijn, L., 2007. Unified View of sediment transport by Currents and Waves.II: suspended transport. Journal of Hydraulic Engineering 133 (6), 668e689. Schiller, L., 1932. Fallversuche mit kugeln und scheiben in Handbuch der experimental-Physik. Akademische Verlagsgesellschaft, Leipzig. Son, M., Hsu, T., 2008. Flocculation model of cohesive sediment using variable fractal dimension. Environmental Fluid Mechanics 8, 55e71. Son, M., Hsu, T., 2009. The effect of variable yield strength and variable fractal dimension on flocculation of cohesive sediment. Water Research 43, 3582e3592.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 1 3 1 e2 1 4 5

Spicer, P., Pratsinis, S., 1996. Coagulation-fragmentation: universal steady state particle size distribution. AiChE Journal 42, 1612. Toorman, E., 1999. Sedimentation and self-weight consolidation: constitutive equations and numerical modelling. Geotechnique 49 (6), 709e726. Verney, R., Lafite, R., Brun-Cottan, J., 2009. Flocculation potential of estuarine particles: the importance of environmental factors and of the spatial and seasonal variability of suspended particulate matter. Estuarines and Coasts 32, 678e693. Verney, R., Lafite, R., Claude Brun-Cottan, J., Le Hir, P. Behaviour of a floc population during a tidal cycle:

2145

laboratory experiments and numerical modelling. Continental Shelf Research, in press, corrected proof. doi:10. 1016/j.csr.2010.02.005. Winterwerp, J., 2002. On the flocculation and settling velocity of estuarine mud. Continental Shelf Research 22, 1339e1360. Winterwerp, J., van Kesteren, W., 2004. Introduction to the Physics of Cohesive Sediment in the Marine Environment. Elsevier B.V, Amsterdam, The Netherlands. Yuan, Y., Wei, H., Zhao, L., Cao, Y., 2009. Implications of intermittent turbulent bursts for sediment resuspension in a coastal bottom boundary layer: a field study in the western Yellow Sea, China. Marine Geology 263, 87e96.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 1 4 6 e2 1 5 4

Available at www.sciencedirect.com

journal homepage: www.elsevier.com/locate/watres

An examination of the mechanisms for stable foam formation in activated sludge systems Steve Petrovski*, Zoe A. Dyson, Eben S. Quill, Simon J. McIlroy, Daniel Tillett, Robert J. Seviour Biotechnology Research Centre, La Trobe University, PO Box 199, Bendigo, Victoria 3552, Australia

article info

abstract

Article history:

Screening pure cultures of 65 mycolic acid producing bacteria (Mycolata) isolated mainly

Received 26 August 2010

from activated sludge with a laboratory based foaming test revealed that not all foamed

Received in revised form

under the conditions used. However, for most, the data were generally consistent with the

21 December 2010

flotation theory as an explanation for foaming. Thus a stable foam required three

Accepted 23 December 2010

components, air bubbles, surfactants and hydrophobic cells. With non-hydrophobic cells,

Available online 28 December 2010

an unstable foam was generated, and in the absence of surfactants, cells formed a greasy surface scum. Addition of surfactant converted a scumming population into one forming

Keywords:

a stable foam. The ability to generate a foam depended on a threshold cell number, which

Activated sludge

varied between individual isolates and reduced markedly in the presence of surfactant.

Bacillus subtilis

Consequently, the concept of a universal threshold applicable to all foaming Mycolata is

Foaming

not supported by these data. The role of surfactants in foaming is poorly understood, but

Mycolata

evidence is presented for the first time that surfactin synthesised by Bacillus subtilis may be

Surface scum

important. ª 2010 Elsevier Ltd. All rights reserved.

Surfactants

1.

Introduction

The generation of stable foam on the surface of aerated reactors is a common feature of activated sludge systems around the world (de los Reyes, 2010; Soddell, 1999; Soddell and Seviour, 1990). In attempts to seek strategies for its control, considerable effort has been directed at better understanding the microbial ecology of foaming (de los Reyes, 2010; Kragelund et al., 2007; Mu¨ller et al., 2007; Seviour et al., 2008). Microscopic examination reveals that most foams contain either long unbranched Gram positive filaments of Candidatus ‘Microthrix parvicella’ or short Gram positive branched filaments of mycolic acid producing bacteria (Mycolata) (de los Reyes, 2010; Kragelund et al., 2007; Seviour et al., 2008). The latter may fragment during their life cycles into coccoid unicells, which are often reported in foams (de los Reyes, 2010).

In many of the early microbiological foam surveys, the molecular techniques now available to allow their unequivocal identification were not applied (de los Reyes, 2010; Nielsen et al., 2009; Seviour et al., 2008), and so attempts to relate foaming incidents to specific operational conditions were compromised by the methodology used. It is now clear from these molecular approaches that a single bacterial morphotype may include several phylogenetically unrelated organisms, differing markedly often in their physiology and ecology (Nielsen et al., 2009; Seviour et al., 2008; Soddell, 1999). For example, the Gordonia amarae like organisms (GALO) morphotype of right-angled branching Gram positive filaments is shared by members of several other genera in the Mycolata (de los Reyes, 2010; Seviour et al., 2008; Soddell, 1999). On the other hand the branching frequencies and angles of the pine tree like organisms (PTLO) can vary considerably in activated sludge.

* Corresponding author. Tel.: þ61 3 5444 7868; fax: þ61 3 5444 7476. E-mail address: [email protected] (S. Petrovski). 0043-1354/$ e see front matter ª 2010 Elsevier Ltd. All rights reserved. doi:10.1016/j.watres.2010.12.026

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 1 4 6 e2 1 5 4

Yet on isolation all these differing morphotypes belong to a single bacterial species Skermania piniformis (Soddell and Seviour, 1998). Other phylogenetically unrelated GALO and PTLO probably exist in activated sludge communities (Kragelund et al., 2007), emphasising that our present understanding of the microbial ecology of foams is incomplete. The current view is that these stable foams are generated by a selective enrichment of hydrophobic bacteria in them by a process of flotation (de los Reyes, 2010; Soddell and Seviour, 1990). Flotation requires three components: gas bubbles surrounded by liquid films, generated by the aeration system; surfactants which reduce the surface tension and thus prevent liquid drainage from the gas bubble walls; and small hydrophobic particles (the bacterial cells), responsible for the long term stabilisation of such foams (Blackall and Marshall, 1989; Soddell and Seviour, 1990). According to this flotation model, all hydrophobic bacteria, regardless of their morphology have the potential to stabilise foams, and so populations other than those discussed above are probably involved in stable foaming incidents, even if not as major contributors (Lemmer et al., 2005). Equally, in the absence of hydrophobic particles, any foam that might develop on reactors would be expected to be ephemeral, and rapidly collapse. Yet whether this is the case remains unknown. Fluorescence in situ hybridisation (FISH) has been used to elucidate possible relationships between Mycolata population sizes in mixed liquor samples and the onset of foaming incidents (Davenport et al., 2000; de los Reyes and Raskin, 2002). Two foaming threshold values have been proposed for Gordonia spp by de los Reyes and Raskin (2002). The formation threshold was thought to correspond to the foaming potential of the mixed liquor, while the stability threshold was considered to be the population level at which a stable foam developed. Davenport et al. (2000) estimated their foaming thresholds with a FISH probe targeting all the known Mycolata, and expressed it as number of cells ml1 mixed liquor. So it is not possible to compare their values with those of de los Reyes and Raskin (2002), who based theirs on whole filament lengths. Davenport et al. (2008) extended this approach to suggest that their threshold was a universal value applicable to all foaming activated sludge plants. This proposal was based on several untested assumptions. Thus, all Mycolata cells were assumed to possess the same cell surface hydrophobicity and propensity to foam regardless of plant operating conditions. Yet many foaming plants contain more than a single foam stabilising Mycolata population. Furthermore, whether their foaming threshold value might be affected by the presence of surfactants, which will probably vary in type and concentration within and among individual plants, was not considered. FISH probes are not available for all foaming Mycolata and will only allow metabolically active cells to be quantified (Amann and Ludwig, 2000). Any dead or moribund Mycolata likely to be present (Kragelund et al., 2007), which may retain their hydrophobicity and hence foam stabilising properties, will not be included in any FISH based enumeration. Therefore this important question of whether such a universal stability threshold value exists for Mycolata is not yet resolved conclusively and needs to be examined further.

2147

Equally only a small selection of the Mycolata isolated from foams have been used in standardised pure culture experiments in attempts to understand what factors might determine their foaming potentials (de los Reyes and Raskin, 2002). So whether this flotation theory applies to all Mycolata is unknown, as is the role if any of surfactants in foam generation and foaming capacity. Where these experiments have been carried out, usually with small numbers of ‘Nocardia’ (probably Gordonia) and Rhodococcus foam isolates, the data are often confusing and contradictory (Ho and Jenkins, 1991; Stratton et al., 1998, 2003). Consequently our study set out to examine the foaming behaviour under standardised laboratory conditions of pure cultures of most of the known non-pathogenic Mycolata, especially those isolated from foams in attempts to see whether flotation explains their foaming behaviour. It also examined closely whether the concept of a universal threshold value for Mycolata foaming stabilisation has any credibility. Particular attention in this study was paid to the possible role of surfactants in stable foaming incidents, and their influence on Mycolata foaming behaviour.

2.

Materials and methods

2.1.

Isolates

Sixty five Mycolata strains listed in Table 1 were used. Their identity was confirmed in this study by partial sequencing (>500 bp) of their 16S rRNA genes, which in all cases were 100% identical to those generated previously for them. All were grown either in peptone yeast extract (PYCa; 0.5 g/L peptone, 0.3 g/L yeast extract, 0.1 g/L glucose and 0.1 g/L calcium chloride) broth or in an activated sludge mixed liquor medium at 30  C for two to seven days. This medium used mixed liquor from the Carrum plant (Victoria, Australia) and was first filter sterilised through a 0.22 mm nitrocellulose filter (Millipore). The filtered effluent was supplemented with several different carbon sources as detailed later.

2.2. Assessment of foaming ability of the Mycolata strains A foaming apparatus described by Stratton et al. (2002) with a sintered glass disc fitted to its base was connected to a rotameter. A 20 ml aliquot of each Mycolata broth culture (A600 adjusted to 1.0) was added to the cylinder and aerated at 100 ml/min for 1 min. Adjusting culture absorbance was achieved by diluting each culture in its own supernatant to ensure that any exocellular surfactants present there are not removed. Foaming abilities were assessed using modified criteria of Blackall and Marshall (1989), and distinctions between foam and scum formation made as described later (Table 2). Foaming thresholds were determined by assessing foaming abilities of strains over a range of different cell densities as determined by their A600. The lowest A600 supporting stable foam formation (category 3 in Table 2) and the corresponding cfu/ml on the appropriate medium (see above) were then determined.

2148

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 1 4 6 e2 1 5 4

Table 1 e Strains used in this study.

Table 1 (continued).

Name of organism

Name of organism

Culture collection numbers and other synonymsa

Gordonia sp. G. aichiensis G. alkanivorous G. amarae G. amarae G. amictica G. australis G. defluvii G. desulfuricans G. hydrophobica G. malaquae G. malaquae G. polysoprenororaus G. rubropertincta G. sputi G. sputi G. terrae G. terrae G. terrae G. terrae G. terrae

Raic22T, DSMZ 43978 Ben606 Gama44T, DSMZ 43392 Gama9, UQCC2810 Ben607 18F3M J4T, DSMZ 44981 213ET, NCIMB 40816 N1123T, DSMZ 44015 A554T, ATCC 35215 A448 Ben605 Grub48T, DSMZ 43197 Gspu49T, ATCC 29627 Gspu48, ATCC 336609 Gter34T, DSMZ 43249 Ben601 Ben602 Ben603 Ben604

Rhodococcus sp. R. coprohilus R. coprohilus R. equi R. equi R. erythropolis R. erythropolis R. globerulus R. luteus R. obuensis R. rhodnii R. rhodochrous R. rhodochrous R. rhodochrous R. ruber R. tritomae

Rcop41T, DSMZ 43347 Rcop18, UQCC 1259 Requ10, UQCC 702 Requ28T, UQCC20307, DSMZ 20307 Rery19, UQCC 379 Rery29T, DSMZ 43066 Rglo35T, DSMZ 43954 IMV 385T, AUCNM A-594 ATCC 33610T Rrho46T, DSMZ 43336 Rrho3, UQCC 2807 Rrho39T, DSMZ 43241 Rrho11S, UQCC 2808 Rrub33T, DSMZ 43338 DSM 44892T

Nocardia sp. N. asteroides N. asteroides N. brasiliensis N. carnea N. nova N. otitdidiscariarum N. otitdidiscariarum N. otitdidiscariarum N. transvalensis

Nast23T, DSMZ 43757 Noast4, UQCC 131 Nbra42T, DSMZ 43758 Ncar30T, DSMZ 43397 Nnov47T, ATCC 33726 Noti14, AMMRL 19.11 Noti25T, DSMZ 43242 Noti15, AMMRL 19.12 Ntra40T, DSMZ 43405

Tsukamurella sp. T. inchonensis T. inchonensis T. paurometabola T. paurometabola T. paurometabola T. paurometabola T. paurometabola T. paurometabola T. paurometabola T. pseudospumae T. pulmonis T. spumae T. spumae T. tyrosinosolvens

DSMZ 44067T NCTC 10741 Tpau37T, ATCC25938 NCTC107411 IMRU1283 DSMZ 20162 IMRU 1520, M334, DSMZ 44119 IMRU 1312, M343 IMRU 1505, M337 N1176T, DSMZ 44118 DSM44142 N1171T, DSMZ 44113, NCIMB 139647 JC85 DSMZ 44234T

Culture collection numbers and other synonymsa

Mycobacterium sp. M. chlorophenolicus M. smegmatis M. fortuitum

Mchl24T, DSMZ 43826 Msme1, UQCC 120 Mfor21, UQCC 422

Other Dietzia maris Streptomyces griseus Millisia brevis

Dmar27T, DSMZ 43672 Sgri05 J82T, DSMZ 44463

T

Type strain. All other culture numbers have been derived from the La Trobe University Bendigo culture collection. a Culture collection numbers have been obtained from the following organisations; DSMZ ¼ German collection of microorganisms and cell cultures. ATCC ¼ American type culture collection. UQCC ¼ Australian collection of microorganisms (now ACM). AMMRL ¼ Australian national reference laboratory in medical mycology. NCIMB ¼ National collection of industrial bacteria.

2.3. Determination of cell surface hydrophobicity and surface tension Cell surface hydrophobicities were determined using the microbial adherence to hydrocarbon (MATH) assay with n-hexadecane (Rosenberg et al., 1980) as the solvent. The percentage cell hydrophobicities were calculated by

Table 2 e Foaming assay scale. Rating 0 1

1aa

2

3

4

5

6

Description As for pure water; No foam 1.0e3.0 cm of foam with fragile ill formed bubbles. Insufficient stability to form films. Immediate collapse on cessation of aeration. Flotation of clumped bacterial cells to the surface of the airewater interface. Clumped cells remain afloat upon cessation of aeration producing a scum layer. Intermitted films sufficiently stable. Usually generated from a fragile foam structure of limited height. Films unstable on cessation of aeration. Substantial foaming (i.e., bubbles about 10 cm diameter) to 3e8 cm height. Infrequent or regular film formation, with both film and foam semi-stable on cessation of aeration. Initially 8e15 cm of foam (about 1 cm diameter bubbles) with stable films being formed at regular intervals. Body of the foam and films stable for 3e5 min once aeration ceases. Stable foam 5e10 cm in height in 2 min, after which collapse to 3e5 cm height. Foam is stable when aeration ceases. No films. Stable foam 15e30 cm in height with no films. Bubble size about 0.5 cm during aeration and increases to 2.0e3.0 cm diam. in 3e5 min from the time aeration ceases.

a This key has been added to represent the ‘scummers’.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 1 4 6 e2 1 5 4

determining the A600 before and after n-hexadecane addition. Surface tensions of broth culture media were obtained by the Wihelmy plate method using an Analyte surface tension metre Model 2141 (Mc Van Instrument Pty Ltd, Australia).

2.4.

16S rRNA gene sequencing

Amplification of 16S rDNA genes was performed by colony PCR using universal primers 27f (AGAGTTTGATCMTGGCTCAG) and 1525r (AAGGAGGTGWTCCARCC) (Lane, 1991). The reaction mixture (final volume 30 ml) contained the following components: w100 ng of template DNA, 0.2 mM dNTPs, 1 PCR reaction buffer, 1.5 mM MgCl2, 1 mg of each oligonucleotide (Geneworks), 2.5 U Taq DNA polymerase and 10% (v/v) DMSO. The mixture was subjected to 36 thermal cycles, as follows: 92  C, 3 min (first cycle only); 92  C, 1 min; 52  C, 70 s; 72  C, 2 min; 72  C, 5 min (last cycle only). The presence of appropriate reaction product was assessed by agarose gel electrophoresis. PCR products were cleaned up using a Wizard SV gel and PCR clean-up kit (Promega) and sequenced at Australian Genome Research Facility (AGRF, University of Queensland, Australia).

2.5.

Fluorescence in situ hybridisation (FISH)

Samples were fixed in 50% [v/v] ethanol overnight at 4  C and stored at 20  C. Fluorescence in situ hybridisation (FISH) was performed using the protocol of Daims et al. (2005). The LGC354B FISH probe (50 -CGGAAGATTCCCTACTGC-30 ), targeting some members of the Firmicutes (Meier et al., 1999) was used to screen foam samples for bacteria of interest (see later). Pre-treatment of the biomass with achromopeptidase, lysozyme and mild acid hydrolysis as detailed by Kragelund et al. (2007), together with extended FISH hybridisation times of 16 h, did not make an appreciable difference to the FISH signal fluorescence intensities.

2.6.

High performance liquid chromatography (HPLC)

Surfactin concentrations in mixed liquors and foams were determined using a modification of the method described by Gong et al. (2009). Analyses by HPLC used a Shimadzu LC-10Ai fitted with a SCL-10Avp PDA detector and a Synergy Hydro-RP column (150  4.6 mm I.D., 4 mm, Phenomenex) with a mobile phase of 3.8 mM trifluoroacetic acid (20%) and acetonitrile (80%) at a flow rate of 1 ml/min. All samples (20 mL) were run after a 20 min column equilibration period. UV spectra of surfactin were collected during analysis with the PDA detector, revealing a lmax of 205 nm, which was used for surfactin quantification. Surfactin standards (purchased from SigmaeAldrich) were prepared from a stock solution of 0.0016 g surfactin in 10 ml analytical grade methanol and sufficient Milli-Q water added to bring the final solution volume to 100 ml.

2.7.

Nucleotide sequence accession number

The nucleotide sequence for the 16S rRNA gene of B. subtilis has been deposited in GenBank under accession number HQ190905.

3.

2149

Results

3.1. Foaming ability of Mycolata cells and insights into the flotation theory in activated sludge foaming Sixty five independently isolated Mycolata strains mainly from activated sludge systems behaved differently in their abilities to form stable foams under the conditions chosen here. While the majority of strains produced foams, the amount produced and its stability varied considerably (Table 3). With some strains e.g. Gordonia malaquae, Rhodococcus coprophilus and Tsukamurella spumae, no foam was generated, but instead a greasy scum layer was formed at the airewater interface after aeration, and these are described here as scummers. The production of this scum layer seemed to occur exclusively only with those cultures growing as large aggregates or clumps, and where unicells in the liquid medium were rare (ie broth turbidity was very low). The foaming capacity scoring system of Blackall and Marshall (1989) does not allow for such scum producing organisms, therefore we have amended this table to incorporate the scummers (Table 2). Surface tension measurements were used as indicators of surfactant production levels of individual cultures. Most broth cultures of strains producing a scum or a foam at the low arbitrary foaming level of 1 had surface tensions >60 nm/M (Table 3). Those with foaming capacities of 2 gave generally lower surface tension values, although there were exceptions to this generalisation, as with Rhodococcus equi (Requ28). This culture gave a very persistent foam (level 6), but no correspondingly large drop in broth surface tension. Cell surface hydrophobicity determinations on all these Mycolata strains revealed that they were all hydrophobic, although the CSH values, reflecting the percentages of hydrophobic cells in each suspension, varied considerably among them (Table 3). When the cells of N. otitididiscaviarum (Noti25), R. equi (Requ28) and G. amarae (Gama44) were rendered no longer viable after autoclaving and subjected to the foaming test, they were still able to produce foam. Even allowing for the limitations known to exist with using the MATH assay (Rosenberg, 2006; Stratton et al., 2002), these results raised questions as to whether a universal threshold value for stable foaming incidents caused by these Mycolata is valid. Consequently, attempts were made to clarify this.

3.2. exist?

Does a universal Mycolata foaming threshold value

Whether the same number of cells is required to obtain stable foams with pure culture of all foaming Mycolata strains was examined. The data presented here clearly show that foaming threshold values varied with individual strain under the controlled conditions used in this study, as the data shown in Table 3 demonstrate. Although it appears that stable foams are only generated at high cell numbers, these vary from 2.5  106 cfu ml1 with G. sputi (Gspu48) to 1.5  109 cfu ml1 with G. terrae (Ben601). G. amarae the most commonly recorded Mycolata in activated sludge foams (de los Reyes, 2010) requires 1.5  108 cfu ml1 to produce a stable foam under these conditions, which is in the higher range when compared to the other Mycolata strains (Table 3).

2150

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 1 4 6 e2 1 5 4

Table 3 e Foaming capacity, hydrophobicity, surface tension and threshold results of Mycolata cultures tested. Straina

Foaming capacityb

Hydrophobicity (%)c

Surface tension (nm/M)

Foaming threshold (cfu/ml)d

Watere Brothe G. aichiensis (Raic22) G. alkanivorous (Ben606) G. amarae (Gama44) G. amarae (Gama9) G. amictica (Ben607) G. australis (18F3M) G. defluvii (J4) G. desulfuricans (213E) G. hydrophobica (N1123) G. malaquae (A554) G. malaquae (A448) G. polysoprenororaus (Ben605) G. rubropertincta (Grub48) G. sputi (Gspu49) G. sputi (Gspu48) G. terrae (Gter34) G. terrae (Ben601) G. terrae (Ben602) G. terrae (Ben603) G. terrae (Ben604)

0 0 6 1 3 1a 1 1a 4 2 3 1a 1a 3 1 1a 2 2 2 2 1a 1

e e þþþþ þþ þþþ þþþ þþþþ þþþþ þþþþ þþþ þþþþ þþþþ þþþþ þþ þþþ þþþþ þþþþ þþþ þþþ þþþ þþþ þþþ

75.0 66.6 50e55 60e65 55e60 55e60 65e70 60e65 65e70 60e65 55e60 60e65 60e65 55e60 65e70 >65 55e60 60e65 55e60 65e70 65e70 65e70

NA NA 2.0  108 5.5  108 1.5  108 NA 4.0  108 NA 4.1  107 7.3  106 1.0  108 NA NA 5.1  108 2.6  107 NA 2.5  106 4.2  108 1.5  109 4.0  108 NA 1.7  108

R. coprophilus (Rcop41) R. coprophilus (Rcop18) R. equi (Requ10) R. equi (Requ28) R. erythropolis (Rery19) R. erythropolis (Rery29) R. globerulus (Rglo35) R. luteus (IMV 385) R. obuensis (ATCC 33610) R. rhodni (Rrho46) R. rhodochrous (Rrho3) R. rhodochorus (Rrho39) R. rhodochorus (Rrho11S) R. ruber (Rrub33) R. tritomae (DSM44892)

1a 1a 2 6 4 6 5 5 2 2 2 5 2 2 3

þþþþ þþ þþþ þþ þþþ þþþ þþ þþ þþþ þþþ þþ þþþþ þþ þþ þþþþ

>70 65e70 60e65 55e60 60e65 50e55 55e60 50e55 60e65 55e60 55e60 55e60 55e60 55e60 60e65

NA NA 3.4  108 1.0  107 4.0  108 2.0  108 1.0  108 3.9  108 1.0  108 1.3  108 1.8  108 1.0  107 2.0  108 2.1  108 8.0  107

N. asteroides (Nast23) N. asteroides (Noast4) N. brasiliensis (Nbra42) N. carnea (Ncar30) N. nova (Nnov47) N. otitididiscaviarum (Noti14) N. otitididiscaviarum (Noti25) N. otitididiscaviarum (Noti15) N. transvalensis (Ntra40)

4 6 1a 3 1a 6 5 1a 1a

þþþ þþþþ þþþ þþ þþ þþþþ þþþþ þþþþ þþ

55e60 60e65 60e65 55e60 60e65 55e60 60e65 65e70 55e60

2.0  108 1.0  107 NA 8.5  107 NA 2.0  107 1.0  108 NA NA

Tsukamurella inchonensis (DSMZ 44067) T. paurometabola (Tpau37) T. paurometabola (DSM20162) T. paurometabola (IMRU1520) T. paurometabola (IMRU1312) T. paurometabola (IMRU1505) T. paurometabola (NCTC107411) T. pseudospumae (N1176) T. pulmonis (DSM44142) T. spumae (N1171) T. spumae (JC85) T. tyrosinosolvens (DSMZ 44234)

1a 3 3 4 1a 1a 1a 1a 3 1a 1a 1a

þþþþ þþþ þþþ þþþ þþþ þþþ þþþ þþþþ þþþþ þþþþ þþþþ þþ

65e70 45e50 35e40 55e60 65e70 65e70 65e70 65e70 55e60 65e70 60e65 55e60

NA 5.1  107 1.7  108 5.0  107 NA NA NA NA 5.0  107 NA NA NA

M. chlorophenolicus (Mchl24) M. smegmatis (Msme1)

3 1a

þþ þþþþ

55e60 60e65

5.0  107 NA

2151

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 1 4 6 e2 1 5 4

Table 3 (continued). Straina

Foaming capacityb

Hydrophobicity (%)c

Surface tension (nm/M)

Foaming threshold (cfu/ml)d

M. fortuitum (Mfor21)

1a

þþþþ

65e70

NA

D. maris (Dmar27) S. griseus (Sgri05) M. brevis (J82)

1 1 1a

þþ þþþ þþþþ

55e60 55e60 60e65

4.0  108 3.1  108 NA

NA ¼ Not applicable. a Strains used in this study. b Foaming capacity determined by the foaming assay described by Blackall and Marshall (1989). The scale is represented in Table 2. c Cell hydrophobicity. % cell hydrophobicity; e ¼ 0%; þ ¼ 1%e30%; þþ ¼ 31%e60%; þþþ ¼ 61%e80%; þþþþ ¼ 80%e100%. d The average minimum number of cells required to form a stable foam. e Controls that contain no bacteria.

Of course this pure culture behaviour may not represent the situation in activated sludge systems, where mixtures of substrates are available to these organisms, and so each of the cultures listed in Table 1 was also grown in activated sludge mixed liquor. Carbon and nitrogen sources (glucose and yeast extract) were added to this clarified mixed liquor to allow the cells to grow and cell numbers to increase to the levels needed for stable foam formation. These cultures grown in mixed liquor gave similar threshold values to those obtained in artificial media (e.g. G. amarae (Gama44) ¼ 2.5  108 cfu ml1; G. aichiensis (Raic22) ¼ 1.5  108 cfu ml1, N. asteroides (Nast23) ¼ 2.0  108 cfu ml1 and T. paurometabola (Tpau37) ¼ 3.5  107 cfu ml1). The presence of any biosurfactants, likely to vary in both their composition and concentrations between plants would be expected to affect substantially these individual foaming threshold values, as shown below.

3.3. Investigation of biosurfactant producing organisms in foaming WWTP Addition of a surfactant (Triton-X 100) to pure cultures of all of the scumming Mycolata seen in this study converted all of them to stable foam formers. For example after the addition of Triton-X 100 to the scum producing bacterial cultures the foaming capacity increased from 1a to 3 for G. terrae (Ben603), N. brasiliensis (Nbra42) 4, T. spumae (N1171) and increased to 4 for R. coprophilus (Rcop18) and M. smegmatis (Msme1). Thus, it is clear that biosurfactants impact considerably on the foaming capacities of these Mycolata, and the flotation model incorporates an essential role for them. Some will enter the plants in the raw influent, yet comparatively little attention has been given previously to which surface active agents might be produced by the existing activated sludge communities (de los Reyes, 2010). A sample from a biodiesel producing activated sludge plant in northern Victoria with a very serious foaming problem was sent to this laboratory for microscopic examination. The foam, overflowing from the reactor and responsible for severe loss of biomass appeared greyish in colour but had collapsed completely within three days of collection. It contained no Mycolata, but instead was dominated by Gram positive rod shaped bacteria growing in chains. Streaking this sample onto PYCa agar yielded three different colony types, which were

repeatedly subcultured until a pure culture of each was obtained, as assessed by microscopy. Colony PCR was used to generate partial sequences of each of their 16S rRNA genes, identifying tentatively one of the isolates as Bacillus subtilis. When pure cultures of this B. subtilis were examined by the laboratory foaming test, this organism produced considerable amounts of foam, far in excess of the foams formed by any of the Mycolata cultures examined earlier. Yet this foam rapidly collapsed. The MATH assay showed that these B. subtilis cells were not hydrophobic, but the surface tension of its culture medium decreased sharply after 3 days incubation from 66.6 nm/M to 29.5 nm/M (Table 4), suggesting it was producing surfactant material. HPLC analysis of the culture medium identified this surfactant as surfactin, known to be synthesised by B. subtilis strains especially when grown with glucose as carbon source (Besson and Michel, 1992).

3.4. How common is B. subtilis in activated sludge foaming plants? A survey of foaming activated sludge plants in Australia suggested that B. subtilis is commonly found in foams. Thus, when foams from 12 plants in New South Wales, Victoria and South Australia were streaked onto PYCa agar, a variety of colony morphologies were seen. Each was screened for its foaming ability using the foaming assay. All isolates producing foams in excess of 10 cm were identified after 16S rRNA sequencing, and their ability to synthesise surfactants assessed by surface tension changes in their culture medium after incubation (Table 4). Isolates of B. subtilis sharing >99% similarity in their 16S rRNA sequences and producing an unstable foam were obtained from six of the 12 plants. Analysis of these foam samples suggested that each was dominated by a different hydrophobic bacterial population (see Table 5). In situ analysis of these was then carried out. It was not possible to design a probe specific for the 16S rRNA sequence of our B. subtilis isolates, or an encompassing phylotype of closely related sequences, because of their high levels of sequence similarity shared with other related organisms. Differentiation of individual Bacillus species is problematic because of the highly conserved nature of their 16S rRNA genes (Rooney et al., 2009). Subsequent FISH analysis applying the LGC354B FISH probe targeting some of the members of the Firmicutes, including the B. subtilis strains

2152

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 1 4 6 e2 1 5 4

Table 4 e Bacterial isolated from foam samples producing surfactant. Source WWTP

Strain Foam Surface number height tension (cm) (mn/M)

Putative identity based on 16S rRNA sequence

Biodiesel

BioD2

17

29.5

Bacillus subtilis

Carrum Carrum Carrum

ETP1 ETP2 ETP4

16 10 20

32.3 70.9 46.8

Bacillus subtilis Pseudomonas nitroreducens Bacillus subtilis

Pakenham

Pak3

16

51.5

Bacillus subtilis

Liverpool Liverpool

Liv1 Liv3

16 10

40.1 77.0

Bacillus subtilis Pseudomonas alcaligenes

Somers

Som4

13

72.5

Klebsiella sp.

Heatherton Heatherton Heatherton Heatherton

Hea1 Hea2 Hea3 Hea4

15 16 14 12

71.4 32.9 23.3 55.8

Acinetobacter xiamenensis Bacillus subtilis Bacillus pumilus Rhodococcus erythropolis

Mt Martha

MtMA3

16

31.1

Bacillus subtilis

Boneo

BonC3

16

37.3

Bacillus subtilis

isolated from these foam samples, detected small numbers of hybridised rod shaped bacteria in 3 of the foam samples (Mt Martha, Heatherton and Liverpool) indicating that their presence in these foams was low. However, surfactin could be detected in Mt Martha samples as well as those from the Biodiesel plant samples at low concentration (<0.1 mg/l), and laboratory foaming tests with commercial surfactin suggest that this biosurfactant enhances foaming abilities at levels below those detectable by HPLC analyses (<0.1 mg/l) (data not shown).

Table 5 e Summary of the WWTP surveyed and their dominant foam associated filaments. Plant

State

Biodiesel Carrum ETP Pakenham (aeration tank)

Vic Vic Vic

Liverpool Glenfield

NSW NSW

Somers (aeration tank) Heatherton

Vic Vic

Mt Martha (aeration tank)

Vic

Mt Martha (Digester) Boneo (Clarifier)

Vic Vic

Boneo (Bioreactor)

Vic

Whyalla (SBR1)

SA

Whyalla (SBR2)

SA

Dominant filamenta Gram þve rods in chains GALO N. limicola II (Gram þve) 0041/0675 GALO GALO Haliscomenobacter hydrossis Microthrix parvicella 0041/0675 Cocci/rods related to GALO Microthrix parvicella N. limicola II (Gram eve) Microthrix parvicella Microthrix parvicella 0041/0675 1863 (Acinetobacter) Microthrix parvicella 0041/0675 GALO 0041/0675 GALO 0041/0675

a GALO ¼ Gordonia amarae like organism.

3.5. Are foaming thresholds affected by the presence of surfactants? When PYCa broth was supplemented with Triton-X 100 at a concentration of 0.00075% (v/v), unstable foam could be generated when aerated in the foaming test. The addition of this small quantity of Trition-X 100 to Mycolata cultures had a dramatic affect on the extent and stability of foaming with the laboratory foaming assay system. In particular, the threshold cell number required for stable foam formation decreased by at least 10 fold in all cases. The foaming threshold of the precultured Mycolata strains was also determined in the presence of the B. subtilis culture supernatant containing surfactin. Again the foaming threshold was reduced by at least a 10 fold decrease in cell numbers and in some instances up to a 100 fold decrease. Such outcomes confirm that in the presence of surfactant, substantially fewer cells are needed to form a stable foam.

3.6. Effect of hydrophobic substrates on foaming abilities of Mycolata When selected Mycolata were grown in media containing the hydrophobic substrate olive oil instead of glucose these organisms grew as clumps which stuck to the glass surfaces of the culture vessels. This suggests that their cell hydrophobicities increase markedly when in the presence of such a hydrophobic carbon source, giving them an increased access to it as a carbon and energy source. Unfortunately the MATH and foaming assays could not be performed with these cultures because the cells did not disperse within the culture medium, but instead adhered strongly to the glass surface of the culture vessel as a film, thus preventing performing any such analyses.

4.

Discussion

The data presented in this study confirm that the flotation theory helps to explain most of what is known about the formation of stable foams by members of the Mycolata on activated sludge reactors. However, not all strains tested form stable foams under the conditions used in this study, suggesting that foaming is not necessarily an inherent feature of these organisms, but a complex end result of the interaction of many environmental conditions (Stratton et al., 2002, 2003). Furthermore, in some strains, foaming ability did not always correspond to a high cell percentage MATH hydrophobicity value or substantial culture surface tension reduction thought to reflect biosurfactant production. Similar inconsistencies have been recorded in previous studies (Stratton et al., 2002), and suggest that these measurements are not always appropriate for such foaming experiments (Eikelboom, 1991). Nevertheless, the results clearly show that stable foam production is an event which requires the presence of three essential components; air bubbles, surfactants and hydrophobic bacterial cells. With insufficient hydrophobic cells, but in the presence of the other two, large amounts of an unstable foam will be generated, and explains why these are commonly seen during the start up of plants (Jenkins et al., 2004), when the population sizes of such organisms would be expected to

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 1 4 6 e2 1 5 4

be low. In some activated sludge plants the biosolids form thin greasy layers or scums over the surface of the mixed liquor (Lemmer et al., 2005). How their formation relates to foaming incidents has been unclear, and the terms foaming and scumming have often been used to describe the same operational problem (Lemmer et al., 2005). However, these data suggest that unlike foams, scums only form in the presence of sufficient levels of hydrophobic bacteria but insufficient surfactant levels for foaming. Consequently these terms should not be used synonymously. The data presented here also show that stable foams will form with non-viable Mycolata cells, which from MATH assay results appear to retain their hydrophobicity. This is an important observation, since it questions the validity of using FISH, where detection of cells relies on high ribosome levels, for determining bacterial cell numbers in activated sludge foams (Davenport et al., 2008; de los Reyes and Raskin, 2002). Thus, the current concept of a universal foaming threshold based on this approach should remain speculative, especially since the results presented here clearly show that the stable foaming threshold number varies with individual Mycolata, and falls substantially in the presence of surfactants. Production of surfactants by Mycolata is well documented (de los Reyes, 2010; Lechevalier, 1975), although the level of their synthesis can vary with the carbon source used. Here we show that the nonhydrophobic B. subtilis may be an important contributor to stable foam formation in activated sludge plants. Its frequent isolation from foams, and demonstrated ability to generate large amounts of an unstable foam in pure culture, together with production of the powerful surfactant surfactin are consistent with such an involvement. Surfactin is a complex molecule, unlikely to be degraded rapidly in activated sludge, and likely to persist there long enough to encourage foaming. In situ analysis with FISH in activated sludge foams indicated that the detectable presence of B. subtilis was low. However, as these cells are not hydrophobic by the MATH assay (Table 4), they may occur more commonly in the mixed liquor and end up in foams incidentally, where they may then sporulate under the nutrient deficient conditions likely to exist there (Eales et al., 2005), thus becoming important but indirect participants in stable foaming events. Even in the mixed liquor their abundance need not be high for a possible important contribution to foaming, given the low levels of surfactin demonstrated as needed for foam stabilisation (<0.1 mg/l) and the high level of its production by these strains. This is the first report where a possible important role for B. subtilis and surfactin production has been demonstrated in foaming, and further work directed at understanding better their influence on its formation in activated sludge plants is required. This population is unlikely to be alone in affecting foaming by in situ surfactant production, and with the availability of a simple assay for detecting bacterial biosurfactant formation (Burch et al., 2010), a search for other populations, especially in unstable activated sludge foams may be rewarding.

5.

Conclusion

This study presents data from 65 foaming Mycolata that the flotation theory can be applied to most of these to explain

2153

their role in stable activated sludge foam formation. Not all conformed to this theory under the experimental conditions used in the study, and with some isolates, the bases for their foaming ability remains uncertain, which suggests that foaming is a complex phenomenon but requires both hydrophobic cells and surfactants as required by the flotation theory. The data also demonstrate a clear distinction between the properties of foaming and scumming bacteria, and proposes that these terms should not be used synonymously. When surfactants are absent but hydrophobic Mycolata are present, a scum is produced. On the other hand with no hydrophobic particles but a surfactant present, an unstable foam is generated. The concept of a universal threshold for foaming applicable to all plants is not supported by these data, where the numbers of cells required depends on the bacterial strain and the presence of surfactant. The study shows for the first time that B. subtilis, shown here to be commonly culturable from foams may be an important participant, by its production of the powerful surfactant surfactin.

Acknowledgements We wish to thank Dr Michael Angove for his advice on the analysis of the HPLC data. The research was supported by the Australian Research Council Linkage Grant (LP0774913) in collaboration with Melbourne Water (David Gregory) and South East Water (Graham Short) who are thanked for their financial support. Z.A. Dyson was the recipient of an Australian Postgraduate Award PhD Scholarships. S. Petrovski was funded by the ARC linkage grant and La Trobe University.

references

Amann, R., Ludwig, W., 2000. Ribosomal RNA-targeted nucleic acid probes for studies in microbial ecology. FEMS Microbiology Reviews 24 (5), 555e565. Besson, F., Michel, G., 1992. Biosynthesis of iturin and surfactin by Bacillus subtilis. Biotechnology Letters 14 (11), 1013e1018. Blackall, L.L., Marshall, K.C., 1989. The mechanism of stabilization of actinomycete foams and the prevention of foaming under laboratory conditions. Journal of Industrial Microbiology 4 (3), 181e188. Burch, A.Y., Shimada, B.K., Browne, P.J., Lindow, S.E., 2010. Novel high-throughput detection method to assess bacterial surfactant production. Applied and Environmental Microbiology 76 (16), 5363e5372. Daims, H., Stoecker, K., Wagner, M., 2005. Fluorescence in situ hybridization for the detection of prokaryotes. In: Osborn, A. M., Smith, C.J. (Eds.), Molecular Microbial Ecology. Taylor & Francis Group, New York, pp. 213e240. Davenport, R.J., Curtis, T.P., Goodfellow, M., Stainsby, F.M., Bingley, M., 2000. Quantitative use of fluorescent in situ hybridization to examine relationships between mycolic acid containing actinomycetes and foaming in activated sludge plants. Applied and Environmental Microbiology 66 (3), 1158e1166. Davenport, R.J., Pickering, R.L., Goodhead, A.K., Curtis, T.P., 2008. A universal threshold concept for hydrophobic mycolata in activated sludge foaming. Water Research 42 (13), 3446e3454.

2154

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 1 4 6 e2 1 5 4

de los Reyes, F.L., Raskin, L., 2002. Role of filamentous microorganisms in activated sludge foaming: relationship of Mycolata levels to foaming initiation and stability. Water Research 36 (2), 445e459. de los Reyes, F.L., 2010. Foaming. In: Seviour, R., Nielsen, P.H. (Eds.), Microbial Ecology of Activated Sludge. IWA Publishing, London, UK, pp. 215e258. Eales, K.L., Nielsen, J.L., Kragelund, C., Seviour, R., Nielsen, P.H., 2005. The in situ physiology of pine tree like organisms (PTLO) in activated sludge foams. Acta Hydrochimica et Hydrobiologica 33, 203e209. Eikelboom, D.H., 1991. Scuim-en driflaagvorming op zuiveringsinrichtingen. Report. TNO Milieu en Energie, Delft, Netherlands. Gong, G., Zheng, Z., Chen, H., Yuan, C., Wang, P., Yao, L., Yu, Z., 2009. Enhanced production of surfactin by Bacillus subtilis E8 mutant obtained by ion beam implantation. Food Technology and Biotechnology 47 (1), 27e31. Ho, C., Jenkins, D., 1991. The effect of surfactants on nocardia foaming in activated sludge. Water Science and Technoogy 23 (4e6), 879e887. Jenkins, D., Richard, M.G., Daigger, G.T., 2004. Manual on the Causes and Control of Activated Sludge Bulking, Foaming and Other Solids Separation Problems. IWA Publishing, London. Kragelund, C., Remesova, Z., Nielsen, J.L., Thomsen, T.R., Eales, K., Seviour, R., Wanner, J., Nielsen, P.H., 2007. Ecophysiology of mycolic acid-containing Actinobacteria (Mycolata) in activated sludge foams. FEMS Micriobiology Ecology 61 (1), 174e184. Lane, D.J., 1991. 16S/23S rRNA sequencing. In: Goodfellow, M., Stackebrandt, E. (Eds.), Nucleic Acid Techniques in Bacterial Systematics. John Wiley & Sons, Chichester, UK, pp. 115e175. Lechevalier, H., 1975. Actinomycetes of sewage treatment plants. In: U.S. Dept. Of Commerce National Technical Information Services Report. Lemmer, H., Lind, G., Muller, E., Schade, M., 2005. Non-famous scum bacteria: biological characterization and troubleshooting. Acta Hydrochemica et Hydrobiologica 33 (3), 197e202. Meier, H., Amann, R., Ludwig, W., Schleifer, K.H., 1999. Specific oligonucleotide probes for in situ detection of a major group of gram-positive bacteria with low DNA G þ C content. Systematic and Applied Microbiology 22 (2), 186e196. Mu¨ller, E., Schade, M., Lemmer, H., 2007. Filamentous scum bacteria in activated sludge plants: detection and

identification quality by conventional activated sludge microscopy versus flouresence in situ hybridization. Water Environment Research 79 (11), 2274e2286. Nielsen, P.H., Daims, H., Lemmer, H., 2009. FISH Handbook for Biological Wastewater Treatment. IWA Publishing, London. Rooney, A.P., Price, N.P.J., Ehrhardt, C., Swezey, J.L., Bannan, J.D., 2009. Phylogeny and molecular taxonomy of the Bacillus subtilis species complex and description of Bacillus subtilis subsp. inaquosorum subsp. nov. International Journal of Systematic and Evolutionary Microbiology 59, 2429e2436. Rosenberg, M., 2006. Microbial adhesion to hydrocarbons:twentyfive years of doing MATH. FEMS Microbiology Letters 262, 129e164. Rosenberg, M., Cutnick, D., Rosemberg, E., 1980. Adherence of bacteria to hydrocarbons. A simple method for measuring cell surface hydrophobicity. FEMS Microbiology Letters 9, 29e33. Seviour, R.J., Kragelund, C., Kong, Y., Eales, K.L., Nielsen, J.L., Nielsen, P.H., 2008. Ecophysiology of the Actinobacteria in activated sludge systems. Antonie van Leeuwenhoek 94 (1), 21e33. Soddell, J.A., 1999. Foaming. In: Seviour, R.J., Blackall, L.L. (Eds.), Microbiology of Activated Sludge. Kluwer, Dordrecht, pp. 161e202. Soddell, J.A., Seviour, R.J., 1990. Microbiology of foaming in activated sludge plants e a review. Journal of Applied Bacteriology 69 (2), 145e176. Soddell, J.A., Seviour, J.R., 1998. Numerical taxonomy of Skermania piniformis and related isolates from activated sludge. Journal of Applied Microbiology 84 (2), 272e284. Stratton, H., Seviour, B., Brooks, P., 1998. Activated sludge foaming: what causes hydrophobicity and can it be manipulated to control foaming? Water Science and Technology 37 (4e5), 503e509. Stratton, H.M., Brooks, P.R., Griffiths, P.C., Seviour, R.J., 2002. Cell surface hydrophobicity and mycolic acid composition of Rhodococcus strains isolated from activated sludge foams. Journal of Industrial Microbiology and Biotechnology 28 (5), 264e267. Stratton, H.M., Brooks, P.R., Carr, E.L., Seviour, R.J., 2003. Effects of culture conditions on the mycolic acid composition of isolates of Rhodoccous spp. From activated sludge foams. Systematic and Applied Microbiology 26 (2), 165e171.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 1 5 5 e2 1 6 5

Available at www.sciencedirect.com

journal homepage: www.elsevier.com/locate/watres

Pilot-scale evaluation of ozone and biological activated carbon for trace organic contaminant mitigation and disinfection Daniel Gerrity a,*, Sujanie Gamage a, Janie C. Holady a, Douglas B. Mawhinney a, Oscar Quin˜ones a, Rebecca A. Trenholm a, Shane A. Snyder b a

Applied Research and Development Center, Southern Nevada Water Authority, River Mountain Water Treatment Facility, P.O. Box 99954, Las Vegas, NV 89193-9954, United States b Department of Chemical and Environmental Engineering, University of Arizona, 1133 E. James E. Rogers Way, Harshbarger 108, Tucson, AZ 85721-0011, United States

article info

abstract

Article history:

In an effort to validate the use of ozone for contaminant oxidation and disinfection in

Received 22 September 2010

water reclamation, extensive pilot testing was performed with ozone/H2O2 and biological

Received in revised form

activated carbon (BAC) at the Reno-Stead Water Reclamation Facility in Reno, Nevada.

22 December 2010

Three sets of samples were collected over a five-month period of continuous operation,

Accepted 29 December 2010

and these samples were analyzed for a suite of trace organic contaminants (TOrCs), total

Available online 8 January 2011

estrogenicity, and several microbial surrogates, including the bacteriophage MS2, total and fecal coliforms, and Bacillus spores. Based on the high degree of microbial inactivation and

Keywords:

contaminant destruction, this treatment train appears to be a viable alternative to the

Trace organic contaminant (TOrC)

standard indirect potable reuse (IPR) configuration (i.e., membrane filtration, reverse

Pharmaceutical

osmosis, UV/H2O2, and aquifer injection), particularly for inland applications where brine

Endocrine disrupting compound

disposal is an issue. Several issues, including regrowth of coliform bacteria in the BAC

(EDC)

process, must be addressed prior to full-scale implementation.

Disinfection

ª 2011 Elsevier Ltd. All rights reserved.

Pilot-scale Ozone Biological activated carbon (BAC) Wastewater Indirect potable reuse (IPR)

Abbreviations: AOP, advanced oxidation process; BAC, biological activated carbon; BHA, butylated hydroxyanisole; CDPH, California Department of Public Health; CFU, colony-forming unit; CT, Concentration (Disinfectant Residual)
Time (Exposure Time); DEET, N,N-diethyl-meta-toluamide; EBCT, empty bed contact time; EDC, endocrine disrupting compound; EEq, estradiol equivalent; EEM, excitation-emission matrix; EfOM, effluent organic matter; FI, fluorescence index; IPR, indirect potable reuse; KOW, octanolewater partition coefficient; LCeMS/MS, liquid chromatography tandem mass spectrometry; MCL, maximum contaminant level; MPN, most probable number; MRL, method reporting limit; N/A, not applicable/not available; NDMA, N-nitrosodimethylamine; OH, hydroxyl radical; PFU, plaque-forming unit; PPCPs, pharmaceuticals and personal care products; RSWRF, Reno-Stead Water Reclamation Facility; TCEP, tris-(2-chloroethyl)-phosphate; TCPP, tris-(2-chloroisopropyl)-phosphate; TI, treatment index; TOC, total organic carbon; TOrC, trace organic contaminant; TSA, tryptic soy agar; TSB, tryptic soy broth; USEPA, United States Environmental Protection Agency; UV, ultraviolet; YES, yeast estrogen screen. * Corresponding author. Tel.: þ1 702 856 3666; fax: þ1 702 856 3647. E-mail address: [email protected] (D. Gerrity). 0043-1354/$ e see front matter ª 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.watres.2010.12.031

$

2156

1.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 1 5 5 e2 1 6 5

Introduction

Pharmaceuticals and personal care products (PPCPs) and endocrine disrupting compounds (EDCs) are often considered “emerging contaminants,” but researchers have been aware of their presence in water for decades. However, the occurrence of PPCPs and EDCs in water did not become a mainstream research topic until the late 1990s and early 2000s. The spike in scientific interest stemmed from demonstrated impacts on aquatic ecosystems, potential human health effects, and increased media coverage, all of which led to increased public awareness (Snyder et al., 2003b). This was coupled with the development of extremely sensitive analytical methods that allowed researchers to approach parts-per-quadrillion (subng/L) detection limits for a variety of trace organic contaminants (TOrCs) (Snyder et al., 2003a; Vanderford and Snyder, 2006). Each of these factors increased the number and scope of scientific investigations into the presence, fate, and transport of TOrCs in natural and engineered systems. Despite this tremendous knowledge base, there are still research areas related to PPCPs and EDCs in water and wastewater that warrant further investigation. With respect to TOrCs in water, the occurrence of the various parent compounds and metabolites results from their release during manufacturing, excretion after personal use, and public disposal of unused quantities (Daughton and Ternes, 1999). Considering that each of these routes directly impacts wastewater, it is reasonable to assume that discharged wastewater is a major source of these contaminants in environmental waters. In the past, wastewater treatment trains were generally not designed for TOrC removal. However, the prevalence of indirect potable reuse (IPR), whether “planned” or “unplanned,” and demonstrated impacts on aquatic ecosystems now justify some consideration of TOrCs in the design process. In fact, expansion and optimization of wastewater treatment processes may be the most efficient strategy to mitigate the potential effects of these contaminants. TOrC mitigation studies span the continua of biological treatment (e.g., activated sludge), physicochemical treatment (e.g., media or membrane filtration), conventional oxidation (e.g., chlorine and ozone), and advanced oxidation processes (e.g., UV/H2O2) in drinking water and wastewater (Huber et al., 2003; Kim et al., 2007; Snyder et al., 2006, 2007; Ternes et al., 2002; Westerhoff et al., 2005). For example, high-pressure reverse osmosis or nanofiltration membranes can be very effective for TOrC removal, but the disposal of concentrated brines limits their widespread applicability. The ultravioletbased advanced oxidation process (i.e., UV/H2O2) is another viable alternative, but the general necessity for upstream pretreatment, the relatively high consumption of H2O2, and the potential need for downstream H2O2 quenching can result in a cost-prohibitive process. Ozone is a unique option because its efficacy is generally similar to that of UV/H2O2 but with significantly reduced energy and chemical requirements (Rosenfeldt et al., 2006). Ozone alone has the ability to generate hydroxyl radicals ($OH) when applied to wastewater due to side reactions with electron-rich moieties, such as amines, phenols, and alkoxylated aromatics (No¨the et al., 2009; Pocostales et al.,

2010). The advanced oxidation process (AOP) can also be optimized with the addition of H2O2 (Buffle et al., 2006). The combination of molecular ozone and the more powerful, nonselective $OH allows for the degradation of more recalcitrant compounds and acceleration of the overall treatment process. Additionally, ozone is an effective disinfectant for wastewater applications, which is particularly important for regulatory compliance (e.g., the California Department of Public Health (CDPH) Title 22 requirements for recycled water). Despite the effectiveness of ozone, issues such as the formation of bromate, N-nitrosodimethylamine (NDMA) (Stalter et al., 2010b), and other potentially toxic oxidation byproducts (Wert et al., 2007) must be considered. With respect to oxidation byproducts, recent studies suggest that post-ozone biological filtration (i.e., sand filtration or biological activated carbon (BAC)) is sufficient to eliminate the toxicity that has been attributed to ozonation of effluent organic matter (EfOM) (Stalter et al., 2010a,b). The standard treatment train for IPR generally includes membrane filtration, reverse osmosis, UV/H2O2, and aquifer injection. However, the aforementioned limitations associated with reverse osmosis and UV/H2O2 have prompted the development of alternative treatment strategies for IPR. In an effort to establish water quality criteria for aquifer injection of reclaimed water in Nevada, the City of Reno Public Works Department commissioned a comprehensive evaluation of secondary biological treatment, ultrafiltration/ sand filtration, ozone/H2O2, and BAC at the Reno-Stead Water Reclamation Facility (RSWRF). The objective of the RSWRF study was to determine whether this treatment train constituted a viable alternative to the standard IPR configuration, particularly for inland applications where brine disposal is an issue. Similar treatment trains have already demonstrated promise in pilot- and full-scale installations in Europe and Australia (Reungoat et al., 2010; Stalter et al., 2010b). The following sections summarize the methodology and results from three sampling events over the course of five months of continuous operation.

2.

Materials and methods

2.1.

Sampling location and operational conditions

Secondary effluent (solids retention time of 25 days) from the RSWRF (Reno, Nevada, USA) was fed into a 40 L/min pilotscale treatment train (Fig. 1). The secondary treatment achieved complete nitrification and partial denitrification (total nitrogen z 4 mg/L) with no detectable nitrite. The pH of the secondary effluent was 6.9, the alkalinity was approximately 90 mg/L as CaCO3, and the bromide concentration was 110 mg/L. The study was separated into two phases. The first phase, which was previously completed by ECO:LOGIC Engineering (Rocklin, CA), evaluated full-scale secondary treatment, pilot-scale ultrafiltration (WesTech Engineering, Salt Lake City, UT), pilot-scale ozone/H2O2 (HiPOx, APTwater, Pleasant Hill, CA), and pilot-scale BAC (WesTech Engineering) (Sundaram et al., 2009). The second phase, which is the focus of the current study, evaluated full-scale secondary

2157

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 1 5 5 e2 1 6 5

Phase 1 O3/H2O2

BAC

UF Preliminary

1°

CAS

2°

Phase 2 Sand

O3/H2O2

BAC

Fig. 1 e Experimental IPR treatment train at the Reno-Stead Water Reclamation Facility. Phase 1 testing was performed previously by ECO:LOGIC Engineering and consisted of full-scale secondary treatment, pilot-scale ultrafiltration, pilot-scale ozone/H2O2, and pilot-scale BAC. The current study summarizes the data from Phase 2, which consisted of full-scale secondary treatment, full-scale sand filtration, pilot-scale ozone/H2O2, and pilot-scale BAC.

treatment, full-scale sand filtration, pilot-scale ozone/H2O2, and pilot-scale BAC. During both phases, ozone was dosed at an ozone to total organic carbon (O3:TOC) ratio of approximately 0.8 (5 mg/L), and H2O2 was added at a molar ratio of 1.0 (3.5 mg/L). These dosing conditions were selected based on previous testing of TOrC oxidation and bromate mitigation (Sundaram et al., 2009). Briefly, Sundaram et al. (2009) evaluated ozone doses of 3, 5, and 7 mg/L with molar H2O2:O3 ratios ranging from 0 to 1.5. TOrC oxidation proved to be extremely effective even at the lowest ozone dose, but higher ozone doses were necessary to achieve significant oxidation of the most recalcitrant target compounds (e.g., meprobamate and N,N-diethyl-meta-toluamide (DEET)). However, bromate formation exceeded 10 mg/L at an ozone dose of 7 mg/L, regardless of H2O2 dose. Therefore, 5 mg/L of applied ozone and 3.5 mg/L of H2O2 were selected to achieve significant oxidation of all target compounds while maintaining bromate levels below 10 mg/L. The H2O2 was added immediately prior to the ozone and both were directly injected into the flow stream. Although the residence time in the HiPOx reactor was approximately 5 min, the oxidation reactions were complete in a matter of seconds due to the rapid conversion of molecular ozone to $OH. The BAC column was 1 m in diameter and contained approximately 570 kg of Filtrasorb F-400 carbon (Calgon Carbon, Pittsburgh, PA) at a bed depth of 1.4 m. The BAC was operated with an empty bed contact time (EBCT) of 30 minutes. During the ultrafiltration phase (Phase 1), BAC backwashing was performed once every 14 days, but due to the higher solids loading during the sand filtration phase (Phase 2), backwashing frequency was increased to once every 7 days. The ozone/H2O2 and BAC pilots were operated continuously from September 2008 to May 2010, but the data presented below only pertains to the operational period from January 2010 to May 2010. Continuous operation was necessary to achieve a stable microbial community in the BAC process, as characterized by periodic phospholipid fatty acid analyses (Sundaram et al., 2009). For the current study, samples were collected from the secondary clarifier, sand filter, ozone/H2O2 pilot, and BAC column in February, April, and May of 2010. The samples were analyzed for a suite of TOrCs, total estrogenicity, bulk organic parameters, bromate formation, and three microbial surrogates, as described below.

2.2. Analysis of trace and bulk organics and total estrogenicity In order to focus the scope of the research, a representative subset of the TOrC universe was selected for evaluation (Table S1). Thirty-one indicator compounds were selected based on several factors, including structural and chemical properties (e.g., functional groups, polarity, aromaticity, etc.), use classes (e.g., antibiotic, fragrance, hormone, etc.), high frequency of environmental occurrence (Benotti et al., 2009; Kolpin et al., 2002; Snyder et al., 2008), resistance to natural (e.g., biodegradation, photolysis, etc.) and engineered treatment processes (e.g., adsorption, oxidation, etc.) (Huber et al., 2003; Ternes et al., 2002; Westerhoff et al., 2005), and amenability to existing analytical methods. Although these compounds have generated considerable interest in the research, treatment, and regulatory arenas, only atrazine is currently regulated by the United States Environmental Protection Agency (USEPA) at a maximum contaminant level (MCL) of 3 mg/L. All samples were collected in amber glass bottles preserved with sodium azide (1 g/L) and ascorbic acid (50 mg/L). The samples were processed with solid phase extraction and analyzed with liquid chromatography tandem mass spectrometry (LCeMS/MS) with isotope dilution according to previously published methods (Trenholm et al., 2006; Vanderford et al., 2003; Vanderford and Snyder, 2006). Reporting limits for the target compounds ranged from 0.25 to 2000 ng/L, as indicated in the supplementary information (Tables S2eS4). In addition to concentrations of the target contaminants, the total estrogenicity of each sample was quantified with the yeast estrogen screen (YES) assay according to previously published methods (Routledge and Sumpter, 1996). Excitation-emission matrices (EEMs) were created using a QuantaMaster UVeVis QM4 Steady State Spectrofluorometer (Photon Technology International, Inc., Birmingham, NJ). The spectrofluorometer included a 75-W, short-arc xenon lamp with an excitation range from 240 to 1200 nm. Data processing included corrections for Raman scattering, blank response, the spectral sensitivity of the lamp, and the inner filter effect (i.e., the absorbance of the matrix). The inner filter correction was applied using equations from the literature (MacDonald et al., 1997) and the absorbance spectra of the sample

2158

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 1 5 5 e2 1 6 5

matrices. The light was assumed to illuminate a small volume at the center of the cell, and the excitation and emission pathlengths were assumed to be 0.5 cm (Westerhoff et al., 2001). The width of the excitation beam was assumed to be 0.1 cm, and the width of the emission was assumed to be 1 cm.

2.3.

Microbial assays

The disinfection component of the study targeted three groups of indicator/surrogate microorganisms: total/fecal coliforms, Bacillus spores as a surrogate for protozoan parasites (e.g., Cryptosporidium oocysts and Giardia cysts), and MS2 bacteriophage as a surrogate for human viruses (e.g., poliovirus, coxsackievirus, and echovirus). Indigenous populations of total/fecal coliforms and Bacillus spores were evaluated during the study, but due to the low influent concentrations of MS2 (i.e., <1 plaque-forming unit (PFU) per mL), a spiking experiment was necessary to verify compliance with the CDPH Title 22 virus requirements for water recycling. Total and fecal coliforms were assayed with the 24-h Colilert (Idexx, Westbrook, ME) method using the Quanti-Tray 2000 quantification protocol. The Colilert method, which is a USEPA-approved method for total and fecal coliform quantification in wastewater, reports coliform levels as most probable number (MPN) per 100 mL. MS2 was assayed with the double agar layer method (Adams, 1959) using antibioticresistant Escherichia coli 700891 (ATCC 700891, Manassas, VA, USA) as the bacterial host. With the double agar layer method, the MS2 levels are reported as PFU per mL. The growth media for the MS2 assay included tryptic soy broth (TSB) supplemented with 0.7% agar for the soft overlay and tryptic soy agar (TSA) as the solid substrate. The plating media and the TSB used for log-phase propagation of the bacterial host were all supplemented with ampicillin and streptomycin (both at final concentrations of 150 mg/L) to reduce indigenous bacterial growth during the assay. For the Bacillus spore assay, samples were heat-shocked at 80  C and 50 rpm for 12 min to inactivate indigenous vegetative bacteria. Samples with higher spore counts (i.e., >20 colony-forming units (CFU) per mL) were then quantified using the pour plate method (i.e., 1 mL of neat or diluted sample) with molten nutrient broth supplemented with 1.0% agar and tryptan blue. Lower spore counts (i.e., <20 CFU/mL) were quantified with membrane filtration (i.e., 10e200 mL of neat sample) using 0.45-mm filters and nutrient agar plates supplemented with tryptan blue. After adjusting for dilution factors and plating volumes, the final Bacillus spore levels were reported as CFU/100 mL.

3.

Results and discussion

3.1.

TOrCs and total estrogenicity

Table 1 summarizes the efficacy of the various full- and pilotscale treatment processes in removing or oxidizing the target contaminants. Unless indicated, the secondary effluent concentrations in Table 1 are averages of the three sample events. The percent removals for each sample event were calculated independently based on their respective secondary effluent concentrations, and the overall averages are reported

in Table 1. As indicated by the footnotes, at least one sample for many of the contaminant/treatment combinations produced a concentration below the method reporting limit (MRL), which limited the possible range of removal or oxidation. The actual concentrations at each treatment stage are reported in the supplementary information (Tables S2eS4). Of the 31 target contaminants, 27 were detected at reportable concentrations in at least one sample event, and 4 were below the MRL in all three sample events. Several contaminants demonstrated significant variability over the five-month study period. For example, the anticonvulsant phenytoin ranged from 1300 ng/L in February to less than 300 ng/L in April and May. Since phenytoin is dosed throughout the day in most patients, the concentrations should have been relatively stable in each sample event. However, phenytoin is not readily biodegraded in secondary treatment so its effluent concentration is more susceptible to temporal fluctuations, which are exacerbated by grab sampling as opposed to composite sampling (Ort et al., 2010). Estrone, which ranged from 15 to 140 ng/L, is another example of a compound exhibiting significant variability. This variability was also reflected in the estradiol equivalent (EEq) values, which are a measure of total estrogenicity from the YES assay, ranging from 12 to 120 ng/L. Excluding these contaminants, the concentrations in the secondary effluent were generally similar between the three sample events and ranged from <MRL for iopromide, testosterone, progesterone, and ethynylestradiol to over 1 mg/L for atenolol and tris-(2-chloroisopropyl)-phosphate (TCPP). A majority of the target compounds were present at concentrations <100 ng/L in the secondary effluent. The field blanks for each sample event were <MRL for all of the target compounds. The percent removals for sand filtration were extremely sporadic, as indicated by the high standard deviations for many of the target compounds. Some of the highest removals were experienced by compounds with log KOW values > 3 (e.g., gemfibrozil, bisphenol A, and estrone), which would indicate particle-assisted removal by the filtration process. However, several compounds with high log KOW values also experienced low removals during the filtration process (e.g., diclofenac and naproxen). Therefore, physicochemical properties are not always a reliable indicator of treatment efficacy. Some of the observed filter removals for the target compounds were consistent with the literature (e.g., bisphenol A, carbamazepine, DEET, naproxen, tris-(2-chloroethyl)-phosphate (TCEP), and trimethoprim), whereas others demonstrated opposite trends (e.g., atenolol, gemfibrozil, ibuprofen, musk ketone, triclosan, and TCPP) (Stevens-Garmon et al., in press). Despite the relatively consistent secondary effluent concentrations for most of the target compounds, some of the removal variability can possibly be attributed to temporal variability in TOrC concentrations due to grab sampling. In general, sand filtration was neither sufficiently effective nor consistent to classify it as a recommended TOrC mitigation strategy, but it was relatively successful in removing steroid hormones and other estrogenic target compounds (e.g., octylphenol and bisphenol A). With respect to oxidation, it is important to understand the differences between molecular ozone and ozone/H2O2. As shown in Table S1, the second-order ozone and $OH rate constants vary significantly depending on the contaminant of

2159

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 1 5 5 e2 1 6 5

Table 1 e Average secondary effluent concentrations and percent removals of the 31 target TOrCs and EEq during each treatment process. The error values represent ±1standard deviation based on variability over the three sample events. Contaminant

Average secondary effluent concentrations (ng/L)

Average % removal after sand filtration (%)

Average % removal after O3/H2O2 (%)

Average % removal after BAC (%)

46 (11) 6 (7) 5 (5) 16 (11) 4 (20) >72c,d 89 (116) 4 (3) 5 (4) 9 (12) 9c 5 (12) 55 (11) 49b N/Aa 4 (4) 3c 10 (12) >42c,d 33 (40) 0 (0) 3 (14) 2 (2) 5c 3 (5) 30 (5)

99 (0.7) >99 (0.2)d 69 (8) >56 (14)d >98 (0.4)d >78c,d >74 (19)d >99 (0.0)d 90 (3) >90 (3)d >99c,d >99 (0.2)d >99 (0.0)d 83b N/Aa 80 (7) >37c,d >99 (0.2)d >42c,d 97 (1) 92 (3) 98 (0.2) 13 (7) 26c >99 (0.5)d >99 (0.0)d

>99 (0.0)d >99 (0.2)d >92 (2)d >57 (16)d >98 (0.4)d >78c,d >74 (19)d >99 (0.0)d 98 (0.8) >90 (3)d >99c,d >99 (0.1)d >99 (0.0)d >92b,d N/Aa 97 (1) >47c,d >99 (0.2)d >42c,d >99 (0.2)d 99 (0.4) >99 (0.0)d 95 (0.8) >96c,d >99 (0.2)d >99 (0.0)d

71b 66 (18) N/Aa N/Aa N/Aa 67 (17)

>83b,d >98 (2)d N/Aa N/Aa N/Aa >98 (2)d

>83b,d >99 (0.7)d N/Aa N/Aa N/Aa >98 (2)d

Pharmaceuticals and personal care products Atenolol 1267 (58) Atorvastatin 58 (12) Atrazine 3 (0.6) Benzophenone 233 (61) BHA 62 (14) Bisphenol A 43c Caffeine 43 (50) Carbamazepine 180 (10) DEET 106 (56) Diazepam 3 (0.7) Diclofenac 165c Fluoxetine 44 (5) Gemfibrozil 890 (61) Ibuprofen 13b Iopromide <10a Meprobamate 603 (91) Musk Ketone 48c Naproxen 72 (25) Octylphenol 50c Phenytoin 600 (606) Primidone 183 (32) Sulfamethoxazole 443 (172) TCEP 543 (72) TCPP 3750c Triclosan 203 (70) Trimethoprim 620 (89) Steroid hormones Estradiol Estrone Ethynylestradiol Progesterone Testosterone EEq a b c d

3b 69 (64) <1.0a <0.5a <0.5a 51 (60)

All sample locations < MRL in 3 of 3 sample events. All sample locations < MRL in 2 of 3 sample events. All sample locations < MRL in 1 of 3 sample events. Percent removal limited by MRL in at least one sample event.

interest. Some compounds are susceptible to both ozone and $OH (e.g., naproxen and carbamazepine), some are only susceptible to $OH (e.g., atrazine and meprobamate), and some are resistant to both forms of oxidation (e.g., TCEP and TCPP). Although dissolved ozone is effective in oxidizing most TOrCs, the combination of high applied ozone doses and H2O2 allows for rapid conversion to $OH, which broadens the range of susceptible contaminants and reduces the reaction time to a matter of seconds. On the other hand, high doses of molecular ozone (e.g., O3:TOC ratios greater than 1.5) may require more than 20 min of contact time before the residual is depleted. With respect to molecular ozone, this translates into larger process footprints in full-scale applications. It is important to reiterate that molecular ozone naturally decomposes into $OH via reactions with EfOM. With sufficient reaction time, ozone and ozone/H2O2 generally provide similar overall $OH exposure in wastewater (Pocostales et al., 2010). Therefore, H2O2 addition is not always necessary in

wastewater applications unless reaction time is a critical design parameter. This is not necessarily the case in drinking water applications where there is less dissolved organic matter to promote ozone decomposition (Pocostales et al., 2010). There are also significant limitations associated with the addition of H2O2. For example, the additional costs and complexities associated with chemical storage, handling, and injection may limit its attractiveness, and the residual H2O2 must also be quenched in some applications prior to environmental discharge. With respect to the RSWRF, catalytic decomposition in the activated carbon should be sufficient to quench any residual H2O2. Although $OH is capable of inactivating microbes, the addition of H2O2 generally reduces the level of inactivation in comparison to ozone alone and limits the applicability of the CT (disinfectant concentration
exposure time) disinfection framework. Furthermore, $OH is fast-reacting and non-selective so it is more susceptible to scavengers,

2160

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 1 5 5 e2 1 6 5

such as EfOM and alkalinity. For target parameters that are susceptible to ozone alone, such as absorbance, fluorescence, and some TOrCs, there may be a slight reduction in treatment efficacy with the addition of H2O2. Finally, H2O2 provides some degree of bromate mitigation, but both ozone and ozone/H2O2 generate bromate, NDMA, and other potentially toxic oxidation byproducts that may pose problems for utilities. In this study, the ozone/H2O2 process demonstrated significant reductions in nearly all of the target contaminants. Only 10 compounds failed to achieve their respective MRLs in at least one sample event. Of those 10 compounds, meprobamate, atrazine, and ibuprofen still experienced removals of 60e90%, and only those compounds specifically designed to resist oxidation (i.e., TCPP and TCEP) experienced removals less than 30%. The remaining target contaminants were removed by more than 90% or to the extent of quantification. The compounds that experienced the highest levels of oxidation were characterized by high ozone and $OH rate constants (Table S1) corresponding to their electron-rich moieties (e.g., phenols, anilines, olefins, and activated aromatics). Huber et al. (2003, 2005) illustrate the sites that are most susceptible to oxidation for many of the target compounds in this study. Except for the May sample event in which estrone was detected at 7.6 ng/L, the steroid hormones were always <MRL after secondary clarification or ozone. Although temporal variability

may have affected the secondary effluent and filter effluent concentrations, the ozone/H2O2 process overcame the effects of sampling error with extensive oxidation. In other words, the consistency and effectiveness of the ozone process correlates to a high degree of confidence in the effluent concentrations and calculated removal percentages. The degree of oxidation in the current study can also be compared to the differential UV254 absorbance model developed by Wert et al. (2009). Based on the approximately 50% reduction in UV254 absorbance observed in the current study (Table 2), Wert et al. (2009) proposed concomitant reductions in atenolol (98%), carbamazepine (>99%), meprobamate (61%), phenytoin (74%), and primidone (72%). However, the current study generally achieved higher levels of oxidation: atenolol ¼ 99%, carbamazepine ¼ >99%, meprobamate ¼ 80%, phenytoin ¼ 97%, and primidone ¼ 92%. Therefore, Wert et al. (2009) provide a conservative estimate of contaminant oxidation for the RSWRF. Nanaboina and Korshin (2010) also developed a model to describe the relationship between differential UV absorbance and contaminant oxidation. Based on a 50% reduction in UV254 absorbance, their study predicted >99% reductions in atenolol, carbamazepine, diclofenac, ibuprofen, iopromide, gemfibrozil, naproxen, and trimethoprim. This coincides with the levels of oxidation in the current study, except for iopromide (<MRL in all samples) and ibuprofen (83%;

Table 2 e TOC, UV254 absorbance, bromate, fluorescence, and treatment indices for each component of the IPR treatment train. For reference, the bromide concentration of the wastewater was approximately 110 mg/L. Sample location

Parameter

February 2010

April 2010

May 2010

Average

Secondary effluent

TOC (mg/L) UV254 (cm1) Bromate (mg/L) FI TI

6.8 0.15 <1 1.34 1.00

6.9 0.13 <1 1.42 1.00

6.8 0.15 <1 1.45 1.00

6.8 0.14 <1 1.40 1.00

Filter effluent

TOC (mg/L) UV254 (cm1) Bromate (mg/L) FI TI

6.5 0.14 <1 1.33 0.97

6.2 0.13 <1 1.37 0.93

7.2 0.14 <1 1.49 0.90

6.6 0.14 <1 1.40 0.93

Ozone/H2O2 effluent

TOC (mg/L) UV254 (cm1) Bromate (mg/L) FI TI

7.6 0.08 2.3 1.37 0.19

7.1 0.07 3.6 1.35 0.17

7.3 0.07 4.6 1.39 0.15

7.3 0.07 3.5 1.37 0.17

BAC e 0.15 m

TOC (mg/L) Bromate (mg/L)

7.4 2.1

7.3 3.5

7.2 4.6

7.3 3.4

BAC e 0.46 m

TOC (mg/L) Bromate (mg/L)

7.7 1.8

7.2 3.4

7.3 4.6

7.4 3.3

BAC e 0.76 m

TOC (mg/L) Bromate (mg/L)

5.4 1.8

5.8 3.4

6.2 4.8

5.8 3.3

BAC e 1.1 m

TOC (mg/L) Bromate (mg/L)

5.1 1.9

5.0 3.5

5.0 4.2

5.0 3.2

BAC Effluent e 1.4 m

TOC (mg/L) UV254 (cm1) Bromate (mg/L) FI TI

4.8 0.05 1.9 1.31 0.08

5.0 0.05 3.6 1.30 0.08

4.8 0.05 4.4 1.31 0.08

4.9 0.05 3.3 1.31 0.08

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 1 5 5 e2 1 6 5

only quantified in April sample event). It is unclear how the model and pilot results would compare at lower levels of oxidation, and it is also unclear how the addition of H2O2 in the pilot affects the applicability of the model, considering the complete dependence on ∙OH oxidation. Regardless, these models provide a potential online surrogate measure of oxidation efficacy that can be used to monitor process performance in large-scale AOP applications. Finally, the BAC process achieved the limit of quantification or at least 95% removal for every target contaminant except benzophenone. Benzophenone removal exceeded the limit of quantification in the May sample event, but the samples from February (130 ng/L) and April (130 ng/L) achieved only 50% removals by the BAC process. Meprobamate (11, 17, and 29 ng/L), TCEP (34, 21, and 35 ng/L), and TCPP (250, <2000, and 100 ng/L) were the only other target contaminants with effluent concentrations exceeding 10 ng/L. DEET, ibuprofen, primidone, and sulfamethoxazole were detected at less than 2 ng/L in the BAC effluent. The efficacy of activated carbon adsorption for TOrC mitigation is also supported by Nowotny et al. (2007), which provides isotherm parameters for several of the target compounds in the current study. In their study, iopromide was the only problematic compound that required excessive powdered activated carbon dosages to achieve effluent concentrations of less than 100 ng/L. In addition to the individual steroid hormones, Table 1 also provides estimates of the total estrogenicity of the various samples. The high degree of variability in the secondary effluent EEq (51  60 ng/L) was likely related to the variability observed for the individual estrogenic compounds: bisphenol A (13, <100, 72 ng/L), estradiol (<0.5, <0.5, 3 ng/L), estrone (51, 15, 140 ng/L), and octylphenol (<26, 69, 31 ng/L). There were likely other compounds contributing to the total estrogenicity of the wastewater, however. Similar to the individual estrogenic target compounds, the filtration process effectively reduced the total estrogenicity of the samples (67  17%), and the subsequent ozone/H2O2 process reduced the total estrogenicity to the reporting limit of the assay in each sample event (i.e., <0.5 ng/L).

3.2. Bulk organic characterization and bromate formation Table 2 summarizes the bulk organic parameters and bromate concentrations measured during the pilot study. Data are provided for the four major sample locations (i.e., secondary effluent, sand filter effluent, ozone/H2O2 effluent, and BAC effluent) in addition to multiple locations within the BAC column. On average, the sand filter provided a small reduction in TOC and no reduction in UV254 absorbance. Although the ozone/H2O2 process reduced the UV254 absorbance by approximately 50%, the TOC actually increased after oxidation. This is likely attributable to biological growth inside the HiPOx reactor. The lack of a dissolved ozone residual, the high bacterial counts in the HiPOx reactor, and the conversion of complex organic matter to assimilable organic carbon created a suitable environment for biofilm growth, which was visually apparent along the walls of the reactor. This biofilm may have been sloughing off over time, which would explain the increase in TOC. Finally, the BAC

2161

effluent achieved a 33% reduction in TOC and a slight reduction in UV254 absorbance. The TOC values for the various depths within the BAC contactor indicate that the carbon was approaching its adsorption capacity toward the end of the study. The first two depths within the contactor (i.e., 0.15 and 0.46 m) provided no reduction in TOC, but significant TOC removal was observed at the final three depths (i.e., 0.76, 1.1, and 1.4 m). However, the 0.76-m sample showed a steady increase in TOC over the three sample events, which may have indicated reduced capacity at that depth over time. Since the TOC concentrations remained relatively constant at depths of 0.15 and 0.46 m, it can be assumed that both adsorption and biodegradation of bulk organic matter were minimal. However, there may have been biodegradation of trace oxidation byproducts, such as carboxylic acids and aldehydes, that would not have been apparent at the mg/L level (Sundaram et al., 2009; Wert et al., 2007). As mentioned earlier, the ozone and H2O2 doses were originally selected based on preliminary evaluations of contaminant oxidation and bromate mitigation (Sundaram et al., 2009). The bromate data in Table 2 indicate that the peroxide addition successfully controlled the formation of bromate (<5 mg/L). However, there was no biological reduction of bromate within the BAC process despite the presence of anaerobic microbes (Sundaram et al., 2009). It is unclear why the bromate level in the ozone/H2O2 effluent increased from 2.3 to 4.6 mg/L over the three sample events. In addition to TOC and UV254 absorbance, 3D EEMs were developed for each process in the treatment train. Fig. 2 provides an example EEM for the February sample event, and the EEMs for the April and May sample events are provided in Figs. S1 and S2, respectively, in the supplementary information. The EEMs provide an illustrative representation of the fluorescence associated with the EfOM in each sample. Similar to UV254 absorbance, decreases in fluorescence are often associated with improvements in water quality. Sand filtration provided minimal reductions in fluorescence, but despite the lack of organic mineralization, there was a distinct transformation of the EfOM during ozonation, as indicated by the dramatic decrease in intensity in the ozone/H2O2 portion of Fig. 2. Although minimal, the BAC process reduced the overall fluorescence even further. EEMs also provide an interesting representation of the EfOM that cannot be captured by bulk organic parameters such as TOC and UV254 absorbance. For example, EEMs can be divided into regions associated with specific organic fractions (Chen et al., 2003), as indicated in the first image of Fig. 2. Region 1 represents the fluorescence associated with soluble microbial products, region 2 represents fulvic acids, and region 3 represents humic acids. The ozone/H2O2 process preferentially transformed the organic matter associated with soluble microbial products and fulvic acids, and although the humic acids experienced significant transformation as well, the process was more gradual. The fluorescence of the final effluent was primarily associated with fulvic acids and humic acids. In addition to the qualitative nature of the EEMs, the underlying excitation-emission data allows for more quantitative evaluations as well. One such parameter is known as the fluorescence index (FI). The FI is the ratio of the

2162

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 1 5 5 e2 1 6 5

Fig. 2 e Excitation-emission matrices for samples from February 2010. The numbered regions in the first image represent the fluorescence associated with specific organic fractions: (1) soluble microbial products, (2) fulvic acids, and (3) humic acids. Excitation-emission matrices for April 2010 and May 2010 are provided in the supplementary information.

fluorescence emission at 450 nm to that of 500 nm within a single EEM when excited by a wavelength of 370 nm (i.e., Ex370Em450/Ex370Em500). The FI has been used to differentiate terrestrially derived organic matter with lower FIs (e.g., surface water from a forested watershed) from microbially derived organic matter with higher FIs (e.g., wastewater) (McKnight et al., 2001). On average, the FI decreased at each stage of the RSWRF treatment train. Overall reductions in fluorescence can also be used to characterize the transformation of EfOM. In wastewaterdominated EEMs, the maximum fluorescence generally occurs with an excitation wavelength of approximately 254 nm and an emission wavelength of approximately 450 nm. Changes in the treatment index (TI), which is defined as the ratio of fluorescence in a treated versus a baseline sample at this location (i.e., Ex254Em450,Treated/Ex254Em450,Secondary), provide another quantitative measure of treatment efficacy. For example, the RSWRF treatment train achieved a 92% reduction in fluorescence according to the TI. In general, these 3D EEMs and the associated indices illustrate the evolution of the water matrix as it loses its wastewater “identity.” This type of reuse treatment train effectively reduces concentrations of trace organic contaminants and bulk organic parameters, thereby transforming the identity of the water matrix so that it more closely resembles an unimpacted water source.

3.3.

Disinfection

According to the CDPH Title 22 requirements, water supplies intended for IPR must demonstrate a 7-day median total coliform level of <2.2 MPN/100 mL, and no more than one sample can exceed 23 MPN/100 mL over a 30-day period (CDPH, 2009). Also, no sample can exceed a total coliform concentration of 240 MPN/100 mL. In addition to complying with the coliform guidelines, water intended for IPR must also satisfy one of the following disinfection requirements: (1) a free chlorine CT value of at least 450 mg-min/L or (2) an alternative treatment certified by the State of California to achieve at least 5-log inactivation of poliovirus or an acceptable surrogate. Previous validation testing in the literature has correlated 6.5-log MS2 inactivation with 5-log poliovirus inactivation (Ishida et al., 2008). Currently, the HiPOx system by APTwater, Inc., which was used at the RSWRF, is the only ozone-based technology certified under Title 22. In order to evaluate the pilot system with respect to the Title 22 requirements, samples were analyzed for indigenous total and fecal coliforms during the sand filtration phase. Samples were also analyzed for Bacillus spores as a surrogate for protozoan parasites (e.g., Cryptosporidium oocysts). The total coliform, fecal coliform, and Bacillus spore data for the February sample event are illustrated in Fig. 3, and the April

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 1 5 5 e2 1 6 5

and May sample events are illustrated in the supplementary information (Figs. S3 and S4, respectively). In contrast to ultrafiltration (data not shown), sand filtration provided limited reductions in both vegetative and spore-forming bacteria. The subsequent ozone/H2O2 process achieved nearly 3-log inactivation of total coliforms and 4-log inactivation of fecal coliforms in the February sample event. The April and May sample events achieved 2-log inactivation of total coliforms and 3-log inactivation of fecal coliforms. However, the ozone/H2O2 conditions were insufficient to comply with the Title 22 requirements due to effluent total coliform levels ranging from 284 to 1069 MPN/100 mL. Fecal coliform levels ranged from 3.1 to 37 MPN/100 mL in the ozone/H2O2 effluent. As expected, the ozone/H2O2 treatment provided limited inactivation of the disinfectant-resistant Bacillus spores. Advanced oxidation has the potential to convert recalcitrant organic matter to more bioavailable forms (e.g., carboxylic acids). This is important for BAC considering that the carbon provides a substrate for vegetative bacteria to attach and develop biofilm communities. Coupled with the continuous supply of biodegradable organic carbon, the contactor provides an ideal environment for bacteria to thrive. This is beneficial for reductions in residual trace organic contaminants and oxidation byproducts, but it provides a regrowth opportunity for indicator bacteria, such as total and fecal coliforms, and possibly opportunistic pathogens. Although attenuated in the April and May sample events, this theory is supported by the 1-log increase in total (1990 MPN/100 mL) and fecal (22.2 MPN/100 mL) coliforms in the BAC effluent. The Bacillus spores are unable to multiply due to their dormant life stage so the BAC basically acts as a simple filtration process, as indicated by the slight decrease in spores in the BAC effluent. With respect to the coliform data, significantly higher ozone doses would have to be used to achieve the 2.2 MPN/100 mL threshold in the sand filtration configuration. Alternatively, the use of a microfiltration or ultrafiltration membrane as a pretreatment process would presumably be sufficient to satisfy the Title 22 coliform requirement after filtration and ozonation. However, it would still be possible for the

MPN/100 mL or CFU/100 mL

1.0E+06

1.0E+05

1.0E+04

2163

Fig. 4 e Results from the MS2 spiking study based on samples collected from two different sample ports on the pilot-scale ozone/H2O2 reactor. The left side of the graph illustrates the consistent inactivation of spiked MS2 in five independent samples. Error bars (±1standard deviation) represent variability from triplicate plates. The right side of the graph illustrates the simultaneous inactivation of indigenous total and fecal coliforms in a single sample. The dashed lines represent the limits of quantification for each assay.

subsequent BAC process to trigger coliform violations due to regrowth so a final disinfection process (e.g., UV disinfection) may still be warranted. Although the sand filtration configuration was insufficient to comply with the Title 22 coliform requirements, the MS2 spiking study (Fig. 4) indicated that the ozone/H2O2 conditions were sufficient to achieve >6.5-log inactivation of MS2. Five replicate samples were collected and analyzed to identify any potential temporal variability in oxidation and inactivation. Data are also provided for two different sampling ports in the HiPOx reactor: Sample Port 4 refers to samples with approximately 3 s of reaction time while Sample Port 6 refers to samples with approximately 5 s of reaction time. As indicated in the graph, the reactions are nearly complete after only 3 s (i.e., Sample Port 4) due to the addition of H2O2, which caused nearly instantaneous decomposition of ozone into $OH. In all cases, 5 mg/L of ozone and 3.5 mg/L of H2O2 were sufficient to achieve >6.5-log inactivation of MS2 in tertiary-treated wastewater effluent. During the MS2 spiking study, the inactivation of total and fecal coliforms was comparable to that of the monthly sample events.

1.0E+03 1.0E+02

4.

Conclusion

1.0E+01

1.0E+00 Total Coliform Secondary Clarifier

Fecal Coliform Sand Filter

Ozone/H2O2

Spores BAC

Fig. 3 e Coliform and spore removal/inactivation in samples from February 2010. Error bars (±1standard deviation) are provided for samples with replicate analyses or multiple dilutions with valid counts. Data for April 2010 and May 2010 are provided in Figs. S3 and S4 in the supplementary information.

Based on the results from five months of continuous pilot-scale testing at the RSWRF, the use of sand filtration, ozone/H2O2, and biological activated carbon may be a viable alternative to the standard IPR configuration (i.e., membrane filtration, reverse osmosis, UV/H2O2, and aquifer injection). Despite sporadic removals during the sand filtration process, the ozone/H2O2 and BAC processes were extremely effective in reducing the concentrations of a suite of TOrCs. The BAC was particularly effective for the oxidation-resistant compounds, such as TCPP and TCEP. With respect to total estrogenicity, the

2164

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 1 5 5 e2 1 6 5

ozone/H2O2 process was able to reduce the EEq level below the MRL for each sample event. With the exception of estrone in the May sample event, the individual steroid hormones were also <MRL after the ozone/H2O2 process. An evaluation of bulk organic parameters and 3D fluorescence indicated that minimal organic removal occurred prior to the BAC, but significant transformation of the EfOM occurred during the ozone/H2O2 process, thereby causing the water matrix to lose some of its wastewater “identity.” With the exception of bromate, oxidation byproducts and toxicity were not evaluated as part of this study, but the literature suggests that post-ozone biological filtration with sand or activated carbon is an effective mitigation strategy for transformation products. Although it was not monitored during this study, direct NDMA formation during ozonation should also be considered in full-scale designs. Finally, the BAC process achieved 33% TOC removal, but the system appeared to be approaching exhaustion based on samples from multiple depths within the column. With respect to disinfection, the pilot-scale treatment train achieved significant reductions in total (>2-log) and fecal (>3-log) coliforms after oxidation, but the ozone/H2O2 conditions were insufficient to satisfy the CDPH Title 22 requirements. Due to insufficient inactivation and regrowth in the BAC process, effluent total coliform levels routinely exceeded the 2.2 MPN/100 mL threshold. Bacillus spore removal/inactivation was also minimal through the treatment train (i.e., <1-log overall), but the ozone/H2O2 process consistently achieved >6.5-log viral inactivation, which satisfies the additional Title 22 disinfection criterion. Although the treatment train proved to be effective for TOrC mitigation and MS2 inactivation, there are still several issues that need to be addressed to determine whether this IPR system is superior to the standard configuration. For example, a more detailed cost analysis would have to be performed to determine whether ultrafiltration or sand filtration should be used as a pretreatment step. Ultrafiltration reduces the backwashing requirements for the subsequent BAC process and would significantly reduce downstream disinfection requirements in comparison to sand filtration. However, the additional capital and operational costs associated with ultrafiltration would have to be considered. If sand filtration is used, the issue of coliform inactivation would have to be addressed with higher ozone doses or a supplementary downstream disinfection process (e.g., UV disinfection). In either the ultrafiltration or sand filtration configurations, the issue of microbial regrowth in the BAC contactors would have to be monitored to determine whether a downstream disinfection process would be necessary. Despite these concerns, filtration, ozone or ozone/H2O2, and BAC likely provide a viable alternative to membrane filtration, reverse osmosis, UV/H2O2, and aquifer injection for IPR, particularly for inland applications where brine disposal is a concern.

Acknowledgements The pilot study was primarily funded and operated by the City of Reno Public Works Department. A portion of the study was

also funded by WateReuse Research Foundation projects WRF-08-05 (Use of ozone in water reclamation for contaminant oxidation) and WRF-09-10 (Use of UV and fluorescence spectra as surrogate measures for contaminant oxidation and disinfection in the ozone/H2O2 advanced oxidation process). The comments and views detailed herein may not necessarily reflect the views of the City of Reno or the WateReuse Research Foundation. This study would not have been possible without the assistance of Stantec (formerly ECO:LOGIC Engineering; Vijay Sundaram and Robert Emerick), the Reno-Stead Water Reclamation Facility (Stan Shumaker, Michael Drinkwater, and Scott Nelson), EAWAG (Yunho Lee and Urs von Gunten), WesTech Engineering, Inc., and APTwater, Inc. Finally, the authors would like to thank researchers and staff at the Applied Research and Development Center at the Southern Nevada Water Authority (Brett Vanderford, Susanna Blunt, Josephine Chu, Shannon Ferguson, Jasmine Koster, and Christy Meza) for all of their efforts during this project.

Appendix A. Supplementary information Target compound properties (Table S1), TOrC concentrations (February, April, and May; Tables S2eS4), 3D excitationemission matrices (April and May; Figs. S1 and S2), and additional microbiology summaries (April and May; Figs. S3 and S4) are available free of charge via the Internet. Supplementary data associated with this article can be found, in the online version, at doi:10.1016/j.watres.2010. 12.031.

references

Adams, M., 1959. Bacteriophages. Interscience Publishers. Benotti, M.J., Trenholm, R.A., Vanderford, B.J., Holady, J.C., Stanford, B.D., Snyder, S.A., 2009. Pharmaceuticals and endocrine disrupting compounds in U.S. drinking water. Environ. Sci. Technol. 43, 597e603. Buffle, M.O., Schumacher, J., Meylan, S., Jekel, M., von Gunten, U., 2006. Ozonation and advanced oxidation of wastewater: effect of O3 dose, pH, DOM and HO scavengers on ozone decomposition and HO generation. Ozone Sci. Eng. 28, 247e259. CDPH, 2009. Regulations related to recycled water. California Code of Regulations. Chen, W., Westerhoff, P., Leenheer, J.A., Booksh, K., 2003. Fluorescence excitation-emission matrix regional integration to quantify spectra for dissolved organic matter. Environ. Sci. Technol. 37, 5701e5710. Daughton, C.G., Ternes, T.A., 1999. Pharmaceuticals and personal care products in the environment: agents of subtle change? Environ. Health Perspect. 107, 907e938. Huber, M.M., Canonica, S., Park, G.Y., von Gunten, U., 2003. Oxidation of pharmaceuticals during ozonation and advanced oxidation processes. Environ. Sci. Technol. 37, 1016e1024. Huber, M.M., Goebel, A., Joss, A., Hermann, N., Loeffler, D., McArdell, C.S., Ried, A., Siegrist, H., Ternes, T.A., von Gunten, U., 2005. Oxidation of pharmaceuticals during ozonation of municipal wastewater effluents: a pilot study. Environ. Sci. Technol. 39, 4290e4299. Ishida, C., Salveson, A., Robinson, K., Bowman, R., Snyder, S., 2008. Ozone disinfection with the HiPOx reactor: streamlining

$

$

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 1 5 5 e2 1 6 5

an “old technology” for wastewater reuse. Water Sci. Technol. 58, 1765e1773. Kim, S.D., Cho, J., Kim, I.S., Vanderford, B.J., Snyder, S.A., 2007. Occurrence and removal of pharmaceuticals and endocrine disruptors in South Korean surface, drinking, and waste waters. Water Res. 41, 1013e1021. Kolpin, D.W., Furlong, E.T., Meyer, M.T., Thurman, E.M., Zaugg, S.D., Barber, L.B., Buxton, H.T., 2002. Pharmaceuticals, hormones, and other organic wastewater contaminants in U.S. streams, 1999e2000: a national reconnaissance. Environ. Sci. Technol. 36, 1202e1211. MacDonald, B.C., Lvin, S.J., Patterson, H., 1997. Correction of fluorescence inner filter effects and the partitioning of pyrene to dissolved organic carbon. Anal. Chim. Acta 338, 155e162. McKnight, D.M., Boyer, E.W., Westerhoff, P.K., Doran, P.T., Kulbe, D. T., Andersen, D.T., 2001. Spectrofluorometric characterization of dissolved organic matter for indication of precursor organic material and aromaticity. Limnol. Oceanogr. 46, 38e48. Nanaboina, V., Korshin, G.V., 2010. Evolution of absorbance spectra of ozonated wastewater and its relationship with the degradation of trace-level organic species. Environ. Sci. Technol. 44, 6130e6137. No¨the, T., Fahlenkamp, H., von Sonntag, C., 2009. Ozonation of wastewater: rate of ozone consumption and hydroxyl radical yield. Environ. Sci. Technol. 43, 5990e5995. Nowotny, N., Epp, B., von Sonntag, C., Fahlenkamp, H., 2007. Quantification and modeling of the elimination behavior of ecologically problematic wastewater micropollutants by adsorption on powdered and granulated activated carbon. Environ. Sci. Technol. 41, 2050e2055. Ort, C., Lawrence, M.G., Reungoat, J., Mueller, J.F., 2010. Sampling for PPCPs in wastewater systems: comparison of different sampling modes and optimization strategies. Environ. Sci. Technol. 44, 6289e6296. Pocostales, J.P., Sein, M.M., Knolle, W., von Sonntag, C., Schmidt, T.C., 2010. Degradation of ozone-refractory organic phosphates in wastewater by ozone and ozone/hydrogen peroxide (peroxone): the role of ozone consumption by dissolved organic matter. Environ. Sci. Technol. 44, 8248e8253. Reungoat, J., Macova, M., Escher, B.I., Carswell, S., Mueller, J.F., Keller, J., 2010. Removal of micropollutants and reduction of biological activity in a full scale reclamation plant using ozonation and activated carbon filtration. Water Res. 44, 625e637. Rosenfeldt, E.J., Linden, K.G., Canonica, S., von Gunten, U., 2006. Comparison of the efficiency of OH radical formation during ozonation and the advanced oxidation processes O3/H2O2 and UV/H2O2. Water Res. 40, 3695e3704. Routledge, E.J., Sumpter, J.P., 1996. Estrogenic activity of surfactants and some of their degradation products assessed using a recombinant yeast screen. Environ. Toxicol. Chem. 15, 241e248. Snyder, S., Vanderford, B., Pearson, R., Quinones, O., Yoon, Y., 2003a. Analytical methods used to measure endocrine disrupting compounds in water. Pract. Period. Haz. Tox. Radioactive Waste Manage. 7, 224e234. Snyder, S.A., Westerhoff, P., Yoon, Y., Sedlak, D.L., 2003b. Pharmaceuticals, personal care products, and endocrine

$

2165

disruptors in water: implications for the water industry. Environ. Eng. Sci. 20, 449e469. Snyder, S.A., Wert, E.C., Rexing, D.J., Zegers, R.E., Drury, D.D., 2006. Ozone oxidation of endocrine disruptors and pharmaceuticals in surface water and wastewater. Ozone Sci. Eng. 28, 445e460. Snyder, S.A., Wert, E.C., Lei, H., Westerhoff, P., Yoon, Y., 2007. Removal of EDCs and Pharmaceuticals in Drinking and Reuse Treatment Processes. AWWA Research Foundation, AWWA, IWA Publishing. Snyder, S.A., Trenholm, R.A., Snyder, E.M., Bruce, G.M., Pleus, R.C., Hemming, J.D.C., 2008. Toxicological Relevance of EDCs and Pharmaceuticals in Drinking Water. AWWA Research Foundation, AWWA, IWA Publishing. Stalter, D., Magdeburg, A., Oehlmann, J., 2010a. Comparative toxicity assessment of ozone and activated carbon treated sewage effluents using an in vivo test battery. Water Res. 44, 439e448. Stalter, D., Magdeburg, A., Weil, M., Knacker, T., Oehlmann, J., 2010b. Toxication or detoxication? In vivo toxicity assessment of ozonation as advanced wastewater treatment with the rainbow trout. Water Res. 44, 439e448. Stevens-Garmon, J., Drewes, J., Khan, S., McDonald, J., Dickenson, E., 2011. Sorption of emerging trace organic compounds onto wastewater sludge solids. Water Res., in press. Sundaram, V., Emerick, R.W., Shumaker, S.E., 2009. Field evaluation of MF-Ozone-BAC process train for the removal of microconstituents from wastewater effluent. 24th Annual WateReuse Symposium. Ternes, T.A., Meisenheimer, M., McDowell, D., Sacher, F., Brauch, H.J., Haist-Gulde, B., Preuss, G., Wilme, U., Zulei-Seibert, N., 2002. Removal of pharmaceuticals during drinking water treatment. Environ. Sci. Technol. 36, 3855e3863. Trenholm, R.A., Vanderford, B.J., Holady, J.C., Rexing, D.J., Snyder, S.A., 2006. Broad range analysis of endocrine disruptors and pharmaceuticals using gas chromatography and liquid chromatography tandem mass spectrometry. Chemosphere 65, 1990e1998. Vanderford, B.J., Snyder, S.A., 2006. Analysis of pharmaceuticals in water by isotope dilution liquid chromatography/tandem mass spectrometry. Environ. Sci. Technol. 40, 7312e7320. Vanderford, B.J., Pearson, R.A., Rexing, D.J., Snyder, S.A., 2003. Analysis of endocrine disruptors, pharmaceuticals, and personal care products in water using liquid chromatography/ tandem mass spectrometry. Anal. Chem. 75, 6265e6274. Wert, E.C., Rosario-Ortiz, F.L., Drury, D.D., Snyder, S.A., 2007. Formation of oxidation byproducts from ozonation of wastewater. Water Res. 41, 1481e1490. Wert, E.C., Rosario-Ortiz, F.L., Snyder, S.A., 2009. Using ultraviolet absorbance and color to assess pharmaceutical oxidation during ozonation of wastewater. Environ. Sci. Technol. 43, 4858e4863. Westerhoff, P., Chen, W., Esparza, M., 2001. Fluorescence analysis of a standard fulvic acid and tertiary treated wastewater. J. Environ. Qual. 30, 2037e2046. Westerhoff, P., Yoon, Y., Snyder, S., Wert, E., 2005. Fate of endocrine-disruptor, pharmaceutical, and personal care product chemicals during simulated drinking water treatment processes. Environ. Sci. Technol. 39, 6649e6663.