WATER RESEARCH A Journal of the International Water Association
Editor-in-Chief Mark van Loosdrecht Delft University of Technology and KWR Watercycle Research The Netherlands E-mail:
[email protected] Editors Jean-Claude Block Université H. Poincaré, Nancy I France David Dixon University of Melbourne Australia Hiroaki Furumai 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 ETH Zürich and Eawag, Swiss Federal Institute of Aquatic Science and Technology Switzerland Wolfgang Rauch University of 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 University of California, Davis USA
Associate Editors Andrew Baker University of New South Wales Australia Damien Batstone University of Queensland Australia
Guang-Hao Chen Hong Kong University of Science & Technology China Tom Curtis University of Newcastle upon Tyne UK Francis de los Reyes III North Carolina State University USA Ana Deletic Monash University USA Rob Eldridge Glen Waverley Australia Monica B. Emelko University of Waterloo Canada Rosina Girones University of Barcelona Spain Stephen Gray Victoria University Australia Kate Grudpan Chiang Mai University Thailand Edwin Herricks University of Illinois at Urbana-Champaign USA Peter Hillis AECOM Australia Hong-Ying Hu Tsinghua University China Bruce Jefferson Cranfield University UK Ulf Jeppsson Lund University Sweden Sergey Kalyuzhnyi RUSNANO Russian Federation Jae-Hong Kim Georgia Institute of Technology USA Günter Langergraber University of Natural Resources and Applied Life Sciences, Vienna (BOKU) Austria
Yoshihiko Matsui Hokkaido University Japan Armand Maul Université de Metz France Max Maurer Eawag Switzerland How Yong Ng National University of Singapore Singapore Satoshi Okabe Hokkaido University Japan Jong Moon Park Pohang University of Science & Technology Korea
Shang-Lien Lo National Taiwan University Taiwan Dionisis Mantzavinos Technical University of Crete Greece
Anastasios Zouboulis Aristotle University of Thessaloniki Greece
Susan Richardson US Environmental Protection Agency USA Miguel Salgot University of Barcelona Spain David Sedlak University of California, Berkeley USA Jean-Philippe Steyer LBE-INRA France Masahiro Takahashi Hokkaido University Japan Nathalie Tufenkji McGill University Canada Kai Udert Eawag Switzerland Vayalam Venugopalan BARC Facilities India Eduardo von Sperling Federal University of Minas Gerais Brazil Julie Zilles University of Illinois at Urbana-Champaign USA
Editorial Office 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: For enquiries relating to the submission of articles (including electronic submission) please visit this journal’s homepage at http://www.elsevier.com/ locate/watres. Contact details for questions arising after acceptance of an article, especially those relating to proofs, will be provided by the publisher. You can track accepted articles at http://www.elsevier.com/trackarticle. You can also check our Author FAQs at http://www.elsevier.com/authorFAQ and/or contact Customer Support via http://support.elsevier.com. 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 ) 5 3 4 7 e5 3 6 4
Available at www.sciencedirect.com
journal homepage: www.elsevier.com/locate/watres
Model selection, identification and validation in anaerobic digestion: A review Andres Donoso-Bravo a,*, Johan Mailier a, Cristina Martin b, Jorge Rodrı´guez c,d, Ce´sar Arturo Aceves-Lara e,f,g, Alain Vande Wouwer a a
Automatic Control Laboratory, University of Mons 31 Boulevard Dolez, B-7000 Mons, Belgium modelEAU, De´partement de ge´nie civil et genie des eaux, Universite´ Laval, 1065 av. de la Me´decine, Que´bec (QC) G1V 0A6, Canada c Department of Chemical Engineering, University of Santiago de Compostela, Spain d Masdar Institute of Science and Technology, Abu Dhabi, United Arab Emirates e Universite´ de Toulouse, INSA, UPS, INP, LISBP, 135 Avenue de Rangueil, F-31077 Toulouse, France f INRA, UMR792, Inge´nierie des Syste`mes Biologiques et des Proce´de´s, F-31400 Toulouse, France g CNRS, UMR5504, F-31400 Toulouse, France b
article info
abstract
Article history:
Anaerobic digestion enables waste (water) treatment and energy production in the form of
Received 13 June 2011
biogas. The successful implementation of this process has lead to an increasing interest
Received in revised form
worldwide. However, anaerobic digestion is a complex biological process, where hundreds
26 August 2011
of microbial populations are involved, and whose start-up and operation are delicate
Accepted 29 August 2011
issues. In order to better understand the process dynamics and to optimize the operating
Available online 3 September 2011
conditions, the availability of dynamic models is of paramount importance. Such models have to be inferred from prior knowledge and experimental data collected from real plants.
Keywords:
Modeling and parameter identification are vast subjects, offering a realm of approaches
Anaerobic digestion
and methods, which can be difficult to fully understand by scientists and engineers
Modeling
dedicated to the plant operation and improvements. This review article discusses existing
Identification
modeling frameworks and methodologies for parameter estimation and model validation
Kinetic parameters
in the field of anaerobic digestion processes. The point of view is pragmatic, intentionally
Sensitivity analysis
focusing on simple but efficient methods. ª 2011 Elsevier Ltd. All rights reserved.
Contents 1. 2. 3. 4.
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Modeling procedure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Mathematical models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Experimental information . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.1. Available measurements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.2. Experimentation mode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.2.1. Batch operation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.2.2. Continuous operation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.2.3. Fed-batch operation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
* Corresponding author. Tel./fax: þ32 065 374. 130. E-mail address:
[email protected] (A. Donoso-Bravo). 0043-1354/$ e see front matter ª 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.watres.2011.08.059
5348 5348 5349 5350 5350 5352 5352 5353 5353
5348
5. 6.
7.
8.
9.
1.
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 5 3 4 7 e5 3 6 4
Structural adequacy and parameter identifiability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Methods for parameters estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.1. Cost function selection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.2. Optimization techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.2.1. Local methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.2.2. Global methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.3. What about optimization constraints? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.4. Alternative methods: The Bayesian inference . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.5. Some considerations for parameter estimation in AD models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.5.1. Steady states analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.5.2. Mass continuity (conservation laws) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.5.3. Initial conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Parameter uncertainty estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.1. Error covariance matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.2. Confidence intervals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.3. Joint posterior distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Model validation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8.1. Direct validation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8.2. Cross validation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Introduction
Anaerobic Digestion (AD) is a chain of interconnected biological reactions, where the organic matter (in the form of carbohydrates, proteins, lipids or more complex compounds), is transformed into methane, carbon dioxide and anaerobic biomass, in an oxygen-free environment. This biological process is used to simultaneously treat waste and wastewater and to produce biogas. AD is now considered as a consolidated technology with more than 2200 high-rate reactors implemented worldwide (Van Lier, 2008). In Europe’s case, between 1995 and 2010, the number of plants installed increased from 15 to 200, which implies an installation capacity rise of nearly 6,000,000 tons per year (from 200,000 to 6,000,000 tons per year) (de Baere et al., 2010). Moreover, the number of AD reactors is expected to increase due to both climate change awareness and the significant boost in the use of renewable energy. The main characteristics of the process, such as reactor design issues, microbial aspects, inhibition phenomena, and of course, the strategic advantages of this process, are well described in the literature (Appels et al., 2008; Chen et al., 2008; Ward et al., 2008). Likewise, a thorough description of the role of the microorganisms in the bioenergy production, with a special emphasis in the anaerobic digestion process, can be found in Rittmann (2008). Mathematical models enable the representation of the main aspects of a biological system. They improve the understanding of the system, the formulation and validation of some hypothesis, the prediction of the system’s behavior under different conditions, reducing, consequently, the experimental information requirements, costs, risk and time. The proper evaluation and application of mathematical models in bioprocesses must follow several stages if the final goal of the approach is to generate useful tools to improve the
5353 5354 5354 5355 5355 5356 5356 5356 5357 5357 5357 5358 5358 5358 5358 5359 5359 5359 5359 5359 5360 5360
understanding of the process or to predict the behavior of the system. This issue has been addressed, in a more general context of environmental models, by Jakeman et al. (2006), from a in-depth theoretical point of view by Walter and Pronzato (1997) and specifically for wastewater treatment processes by Dochain and Vanrolleghem (2001). Several mathematical models of anaerobic digestion have been proposed in the last two decades and a variety of methods have been used for parameter estimation and model validation. AD process is characterized by its high complexity and non-linearity and by the difficulty to collect large amounts of informative experimental data for modeling purposes. One of the consequences of the latter is variety of approaches to modeling and parameter identification is the important variability in values reported for the kinetic parameters, even when the same operational and environmental conditions have been evaluated. This paper presents an overview of the main procedures that can be used for developing and assessing dynamic models of the anaerobic digestion process. It is structured according to a step-by-step approach of the modeling task, i.e., from model selection up to model validation.
2.
Modeling procedure
The whole modeling process (selecting a model structure, identifying the parameter values, and planning the experimental measurements) should be in coherence with the objective pursued. In general, the three most common objectives of using a model are: understanding the system’s behavior and interaction of components; quantitatively expressing or verifying our hypothesis and predicting the
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 5 3 4 7 e5 3 6 4
behavior of the system in the future or under other similar circumstances. Adequate model structures should be chosen according to four principles (Spriet, 1985): (i) simplicity, the model should be as simple as possible; (ii) causality, the model should represent the most relevant causeeeffect relationships; (iii) identifiability, the values of the unknown parameters should be identifiable from the available data; and (iv) predictive capability, the model should remain valid under future or alternative reasonable conditions. As stated by Flotats et al. (2003) model identification and parameter estimation have not been given the same attention in anaerobic digestion processes, as it does with activated sludge systems in which considerable efforts have been rightfully devoted (Ossenbruggen and Stevens, 1996; Weijers, 2003; Liwarska-Bizukojc and Biernacki, 2010). Fig. 1 shows a schematic view of the parameter estimation and model validation procedure. At first, it is of course very important to define the purpose of the modeling exercise. An explicative (mechanistic) model intended for process investigation and hydraulic/chemical/biological analysis will likely include a detailed description of specific mechanisms and phenomena, which would probably be irrelevant for a global dynamic analysis or the design of controllers, for instance (Jakeman et al., 2006). Therefore, the level of details of the description has to be selected with care depending on the targeted application of the model (physical/chemical/biological investigation, process design, dynamic simulation, optimization, control, supervision). Once an appropriate model structure has been selected (usually a system of non-linear differential equations including a number of unknown or uncertain parameters), a simulator can be implemented using a platform of choice (a programming language such as Fortran or C, or an environment such as Matlab or its open-source counterparts Octave and Scilab). Local and global parametric sensitivity analysis can then be used to assess, on the one hand, the most influential parameters, and on the other hand, the parameters
Fig. 1 e Parameter estimation procedure.
5349
with weaker influences on the measured outputs (at least in the scenario under consideration), possibly involved in correlation with other parameters. This first analysis can lead to a model reformulation or simplification, eliminating correlated parameters, and it can also lead to the reformulation of a new experimental design and data collection methods in order to have more informative data in relation with the parameters to be estimated. Next, the experimental data must be examined in terms of potential errors (outliers, missing data, etc.) and the deviation between the model prediction and the measured outputs. In this last part, special attention has to be paid to the selected cost function. Finally, the model has to be evaluated with regard to the experimental data used so far (direct validation), as well as fresh data (unseen in the identification process e cross validation). These last two steps, if unsuccessful, can lead back to model reformulation and/or experiment design and data collection.
3.
Mathematical models
Mathematical modeling of the anaerobic digestion process was motivated by the need for efficient operation of anaerobic systems in the early 70’s. The first models were relatively simple due to the limited knowledge about the process. Experimental investigation, further system analysis and the increase in computing capacity lead to the development of much more detailed models in recent years. As it is not the goal of this review to list the available models in anaerobic digestion, a brief overview is given in the next paragraphs. The first modeling approaches focused on describing the limiting step of the process, considering that anaerobic digestion is a multistep process where one slower step controls the global rate (Hill and Barth, 1977). Such limiting step can, however, be different under different operating conditions (Speece, 1996). Some authors considered methanogenesis as the limiting step or the conversion of fatty acids into biogas or the hydrolysis of suspended solids (Eastman and Ferguson, 1981). These series of models were simple and easy to use but were unable to adequately describe the process performance, especially under transient conditions. A second generation of models considered the concentration of volatile fatty acids as the key parameter, incorporating acidogenesis and acetogenesis separately (Hill, 1982). The hydrogen partial pressure, as a key regulatory parameter influencing the redox potential in the liquid phase and more bacterial groups, with differentiated acetoclastic and hydrogenotroph methanogens, was included in several models (Costello et al., 1991; Ruzicka, 1996). The redox potential (as NADH/ NAD þ ratio) is a function of the hydrogen partial pressure and determines the VFA production in this family of models. Further microbiological studies led to another generation of models (Angelidaki et al., 1993, 1999; Siegrist et al., 1993; Vavilin et al., 1994, 1995; Kalyuzhnyi and Davlyatshina, 1997; Kalyuzhnyi, 1997; Kalyuzhnyi and Fedorovich, 1998; Batstone et al., 2000; Tartakovsky et al., 2002; Keshtkar et al., 2003; Haag et al., 2003). These models incorporated additional processes and species, more detailed kinetics with inhibition and considered different substrates.
5350
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 5 3 4 7 e5 3 6 4
As a response to the need for a generic model of anaerobic digestion, the IWA Task Group for Mathematical Modeling of Anaerobic Digestion Processes developed the generic Anaerobic Digestion Model No.1 (ADM1) (Batstone et al., 2002) in order to reach a common basis for further model development and validation studies with comparable results. The ADM1 model describes the dynamics of 24 species and includes 19 bioconversion processes. The latter, makes ADM1 a model with a large number of parameters. In view of its general purpose, the ADM1 neglects some processes and species, which are related to more specific applications, in order to avoid extreme complexity. Still, the large number of parameters and identifiability difficulties are the major drawbacks of ADM1, as well as, some structural weaknesses (Kleerebezem and Van Loosdrecht, 2004). Many applications based on the ADM1 have been published in recent years. Some authors applied the model to stirred tank systems while others considered distributed parameter systems (Batstone et al., 2004a, 2005). Extensions have been developed to incorporate processes that were absent in the original model. Also, reports of the applications of the ADM1 to particular types of wastewater have been published (Batstone and Keller, 2003; Fedorovich et al., 2003; Batstone et al., 2004b; Fezzani and Ben Cheikh, 2008; Fezzani and Ben Cheikh, 2009; Derbal et al., 2009; Gali et al., 2009; Lee et al., 2009; Ramirez et al., 2009; Ozkan-Yucel and Gokcay, 2010). The framework provided by the ADM1 is useful especially for process design and dynamic simulation. Due to its fixed stoichiometry approach, its applicability would, however, require important structural modifications for some processes. Implications of structural changes in some processes of the ADM1 toward a variable stoichiometry structure have been recently analyzed (Rodriguez et al., 2006). Efforts have also been directed in recent years to simplify this model (Siegrist et al., 2002; Rodriguez et al., 2008). In order to ease the application of the ADM1 some methodologies have been developed (Zaher et al., 2004; Kleerebezem and Van Loosdrecht, 2006), as well as, some structural simplifications of the model under certain conditions (Bernard et al., 2006). Among the simplified models of the AD process, the one developed by Bernard et al. (2001) has been used in different applications. This model considers two-reactions (acidogenesis and methanogenesis) and has been widely applied for control purposes, for AD process optimization (Dalmau et al., 2010) and for mathematical analysis (Dimitrova and Krastanov 2009; Rincon et al., 2009; Sbarciog et al., 2010). However, only few applications with data from lab- or fullscale plants have been reported (Donoso-Bravo et al., 2009a; Lopez and Borzacconi, 2009). Several reviews of the existing models have been published in the last two decades. Husain (1998) made a brief review of steady state and dynamic models of the kinetics of anaerobic digestion. Later on, Gavala et al. (2003) presented a comprehensive review, describing from the simplest to the most complex models. Following these general reviews, more specific studies appeared in response to the increasing number of available models. For instance, Tomei et al. (2009) focused on models developed for the anaerobic treatment of sewage sludge. Likewise, Batstone (2006) addressed anaerobic digestion modeling in the framework of domestic sewage
systems. Other studies have focused on the type of reactor, instead. For instance, Saravanan and Sreekrishnan (2006) described the different available models for UASB (Up-flow Anaerobic Sludge Blanket), AF (Anaerobic Filter) and EGSB (Expanded Granular Sludge Blanket) reactors, in which the biomass is attached to either a support or forming granules. In Table 1, a summary of different studies where modeling and optimization have been performed is shown.
4.
Experimental information
4.1.
Available measurements
Accurate and reliable measurements of key variables of the process are very important. These data will be used for model identification and validation therefore they should contain the most relevant information at the lowest possible cost of monitoring. Counting on several measured variables will increase the possible parameters that can be reliably estimated; however, and particularly in the case of large and complex models as ADM1, identifying all the parameters and coefficients is not feasible mainly due to the extreme difficulty of separately identifying specific biomass concentrations from the maximum specific uptake rates. Methanogenic applications, however, might require only very accurate identification of a limited number of key parameters to provide good results due to the dynamics of AD in which acidogenesis and acetogenesis are much faster than methanogenesis and hydrolysis (Rodriguez et al., 2006). For any AD model in general, a proper structural identification (discussed in Section 5), for instance using sensitivity analysis methods together with knowledge of the process dynamics, can play a key role in the success of the optimization process to select the most relevant parameters. In general, two types of data can be considered, those that come from off-line or on-line measurements. Off-line sensors are those in which the sample is taken usually manually (low sample frequency) and analyzed by an operator; the data will be available after hours or days. On-line measurements are attached to the process and the analysis is automatic. The analysis is performed under a high frequency, so data produced by these sensors are considered continuous compared to processes that use a time scale. These above-mentioned aspects must be carefully taken into account since the quality of experimental data, in terms of measurement error and sampling frequency, will have a substantial influence on the parameter estimation of the model (Guisasola et al., 2006). To characterize the substrates and intermediates characterization, the organic matter content (i.e. all the organic compounds present in the solution) is usually calculated through the chemical oxygen demand (COD), which is the most common (off-line) measured variable. This measure, and in a lesser extent, along with the total organic carbon (TOC) analysis has been also used for this purpose. In order to recognize the dynamic of specific variables, off-line tests are usually used for volatile fatty acids (VFAs) and for the main macromolecular compounds such as carbohydrates, proteins and lipids. New on-line sensors have been developed which are starting to be used more often (Molina et al., 2009; Boe
Table 1 e Summary and brief description of the studies found in literature about modeling and kinetic parameters identification in AD systems (IC: Initial conditions, KP: Kinetic parameters, YC: Yield coefficients, PCC: physico-chemical constants, CF: Conversion factors). Reference Batch Batstone et al. (2009) Lopez and Borzacconi (2010) Palatsi et al. (2010)
Model
Estimates
Measurements
Estimation method
Uncertainty
ADM1 Model for complex substratesa
2 KP (hydrolysis) 7 KP
Biogasb Methaneb
Gradient search technique Multiple shooting
Confidence region Monte Carlo
ADM1
3 IC, 4 KP
Non-linear weighted square minimization
n.d.
3-reaction model
2 KP, 2 YC
Methane, Acetic, butyric, propionic acidc Biogasb
n.d.
1 IC, 2 KP, 1 YC
Biogasb
Lokshina et al. (2001)
Monod and Non-competitive model Monod and Haldane Equations
Hooke and Jeeves optimization method Least-square Non-linear weighted square minimization
1 ratio (IC/YC), 1 YC, 3 KP
Methanec
Covariance matrix-FIM
Flotats et al. (2003)
ADM1
3 KP, 2 YC
Acetate, propionate, valerate, methanec
Non-linear regression with the Marquardte Levenberg algorithm Combination of random direct search and gradient methods
Continuous Batstone et al. (2009) Bernard et al. (2001)
ADM1 2-reaction model
2 KP (hydrolysis) 4 KP, 6 YC, 1 PCC
Gradient search technique Linearization at different steady states
Confidence region n.m.
Haag et al. (2003)
3-reaction model
26 IC, 8 KP, 4 YC, 6 CF
Batstone et al. (2003)
ADM1
6 KP
Directed search method followed by a gradient-based method Secant method
Covariance matrix-FIM and confidence interval Confidence region
Lopez and Borzacconi (2009) Ghaniyari-Benis et al. (2010) Bhunia and Ghangrekar (2008) Kalfas et al. (2006)
2-reaction model
6 YC
Least-squares criterion
n.d.
1-reaction model
1 KP
Biogas,b VSSc Methane,b Carbon dioxide,c COD, VFA, Z, ICc CODt, CODs, TOC, MiS, TDE, ODE, VFAc VFA, Biogas, pH, methane contentb Biogas,b COD, VFA, gas compositionc CODc
n.m.
Monod, Grau-2nd order and Haldane equations ADM1
3 KP, 1 YC, 1 CF
COD, VSSc
Non-linear regression using least-squares criterion Linearization at different steady states
2 KP, 2 YC (mesothermophilic conditions)
Secant method using unweighted least-square criterion
Confidence region and linear confidence intervals
Koch et al. (2010)
ADM1
5 KP, 1 CF
TSS, VSS, COD, VFA, BIogas, gas composition, pHc Biogas,b gas composition, NH4, NKT, VFA, alkalinity, TSc
Evaluation of the modified Nash-Sutcliffe coefficient
n.d.
Initial rate Donoso-Bravo et al. (2011) Donoso-Bravo et al. (2009b)
1 reaction model (Monod Kinetic) 1st order, Monod, Haldane equations.
2 KP
Methanec
Covariance-FIM
5 KP
Carbohydrates, VSS, VFAc
Non-linear regression using least-squares criterion Non-linear regression using least-squares criterion
Noykova and Gyllenberg (2000) Muller et al. (2002)
Covariance matrix-FIM
n.d.
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 5 3 4 7 e5 3 6 4
n.d.
5351
n.d. not determined, n.m. determined but not mentioned the used method. a Proposed by Angelidaki et al. (1993). b on-line. c off-line.
Monte Carlo
5352
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 5 3 4 7 e5 3 6 4
et al., 2010; Ward et al., 2011; Jacobi et al., 2011). Specific species concentration of the anaerobic biomass (state variables are unknown variables due to difficulties in performing measurements of the concentration of each population. Molecular biological techniques have been executed for these purposes (Sanchez et al., 1994; Pobeheim et al., 2010)); however, these are costly and usually only qualitative information can be drawn. This issue may trigger some identification problems since some parameters cannot be determine independently (Bernard et al., 2001; Noykova et al., 2002), and may cause some inaccurate model predictions (Batstone et al., 2004b). The volatile suspended solid content (VSS/l), which is normally known, has been used to approximately estimate the total biomass concentration, i.e., the sum of all the population involved (Bernard et al., 2001; Lopez and Borzacconi, 2009). In any case, this measure has not been used for parameter fit but for model validation. A typical model application, in this context, has been the use of software sensors to estimate the concentration of each population (Bernard et al., 2000; Lopez and Borzacconi, 2009). In regards to the product of the reaction, the sum of all gas compounds (formed during the process) is the most common on-line performed measurements and consequently used widely in modeling applications. The biogas measurement has been used as the only measurement for parameters estimation in plenty of articles. Independent measurement of the different gases (CO2, H2 and H2S) is generally done off-line. Depending on the model, the biogas production can be considered as a state variable (Batstone, 2006; Keshtkar et al., 2003) or as a dependent variable (Bernard et al., 2001). The biogas cumulative volume is normally used in the case of using data from batch test and biogas flow rate in the case of continuous system. Biogas flow rate contains more information than cumulative biogas volume and the latest can be derived from the previous, by numerical integration. Instantaneous gas flow rate is typically however more costly and for many applications cumulative volume can also be used to derive the instantaneous flow rate by derivation but the accuracy will be affected by the frequency of sampling points available of accumulated volume. Among other measurements, for instance, the pH is a variable which is easily measured by online sensor. In most of the model is used as an input in the form of an inhibition function (Angelidaki et al., 1999; Batstone, 2006; Haag et al., 2003; Keshtkar et al., 2003), but also can be found as an output of the model to be used for validation (Bernard et al., 2001). So far, no models have used pH for parameter identification; firstly, because it is used as an input and secondly, because pH may present a low sensitivity in well-buffered systems.
4.2.
Experimentation mode
Understanding the nature of the experimental data is a crucial point when making good use of the modeling procedure. This section analyses the different experimental conditions in which anaerobic tests are carried out. Two main issues have to be considered: 1. culture history, and 2. the selected operation mode of the anaerobic digestion process. The culture or inoculum history, involves the specific characteristics of the anaerobic biomass used for the assay,
which in the case of anaerobic digestion systems is normally a mixed culture. The crucial factor is the manner in which the culture has been developed since it determines which species are predominantly present which also influences the physiological state. It is clearly different if the anaerobic inoculum comes from a continuous reactor (where organisms with higher affinity enzymes are favored), if it comes from a batch reactor, or if the inoculum has been exposed to specific conditions for some time (microorganisms have the ability to change their macromolecular composition as a result of physiological adaptation). Usually, the culture history cannot be easily modified and in most cases, the tests are simply carried out with the available inoculum. However, it is of prime importance to document the specific conditions of the inoculum culture since the results are likely to be valid only under these experimental culture conditions. The operation mode of the assay has a paramount influence on the information content of the collected data, and thus, on the quality of the estimated parameters. The variety of operational conditions explain the variability of the reported parameter values (Grady et al., 1996), and implicitly shows that either the studies have considered too limited experimental data information or have not properly applied the parameter estimation procedure (otherwise, more reference parameter sets would have been published and validated). Batch assays are commonly used in AD for kinetic parameter determination, even though other types of operation modes have also been employed for these purposes. This situation is rather unfortunate since it has been demonstrated that the parameters of a simple Monod law cannot be uniquely determined from a batch experiment (Baltes et al., 1994). It is therefore very unlikely that the complex kinetics of AD could be determined from batch tests only, which will become clearer after the next section’s explanation on parameter sensitivity analysis.
4.2.1.
Batch operation
Batch operation can be defined as a biological process in which there is no interchange of mass with the environment, i.e. there are no input or output flow (except for the gas stream). All substrates and nutrients are added at the beginning of the reaction cycle. Several advantages have been stated with regards to the use of batch tests (often known as the biochemical methane potential test, BMP) for kinetic parameter determination in AD, such as: (1) the possibility to easily record the time evolution of several variables (2) the relatively short time span as compared to continuous operations and (3) simplicity and popularity as reflected in a wide acceptance (Noykova et al., 2002; Flotats et al., 2003; Batstone et al., 2009; Lopez and Borzacconi, 2010). However, the main drawback of batch tests, stems from the lack of input excitation (since the only input is the initial condition) resulting in a lack of parameter sensitivity (Lokshina et al., 2001). This can be partly alleviated using different sets of initial conditions (Flotats et al., 2003, 2006) and determining a proper range of substrate/biomass (S/X ) ratio (Grady et al., 1996). The S/X factor, whose inverse is also used in some studies (X/S: inoculumesubstrate ratio or IRS, (Raposo et al., 2009), may influence the correlation between some parameters, for instance, mm and Ks of the Monod-
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 5 3 4 7 e5 3 6 4
equation, and thus, affect the results of parameter identification. Despite its importance, the value of S/X is seldom mentioned in the studies where batch tests are used for parameter estimation. Whereas BMP tests are widely used for parameter estimation, most full-scale reactors operate in continuous or semi continuous conditions which are drastically different conditions. This explains the proposal to use alternative procedures such as the initial rate reaction measurement, which has been widely applied in enzymatic processes. It is used to obtain either production or consumption rates of a specific compound involved in a reaction on a short time span (Illanes, 2008). Donoso-Bravo et al. (2009b) used the initial rate measurement of the substrate degradation (starch, glucose and VFAs) to evaluate the influence of the temperature on the main reactions of the anaerobic digestion, just as Flotats et al. (2003) estimated kinetic parameters of anaerobic degradation of gelatin using the same technique. According to DonosoBravo et al. (2011) this method can be an interesting alternative to classical batch tests, as it allows to alleviate inhibitory effects of byproducts or substrate limitation (which are more likely to occur in batch than in continuous operation). The main drawback of this technique is the lack of research and validation since only a few studies have used initial rate tests in AD applications.
4.2.2.
Continuous operation
In this operation, spent medium or digestate is continuously replaced with an equal volume of fresh medium (substrate solution) and therefore a continuous discharge of biomass also occurs. Continuous systems also offer a proper platform for kinetic analysis, as long as a series of experiments at different dilution rates (D) are carried out. Overall, it is a more time-consuming method than a batch test and, therefore, the kinetic parameter calculation is usually performed with data from continuous anaerobic reactors which are already operating. For instance, more than 3 months of a pilot up-flow fixed-bed anaerobic reactor operation were required by Bernard et al. (2001) or 1.5 year for full-scale anaerobic digester by Batstone et al. (2009) in order to perform parameter estimation. A less time consuming and appealing alternative is to evaluate the dynamic response of a continuous reactor after specific substrate pulses (Batstone et al., 2003; Kalfas et al., 2006). This method allows the estimation of the kinetic parameters of specific compounds since the pulses provide some decoupling of the biological phenomena, and in turn lower parameter correlation, and thus better identifiability. The pulse amplitude has to be selected so that the substrate concentration crosses the affinity-saturation constant (Batstone et al., 2003). The main drawback of this pulse-based methodology is that it cannot be applied at full-scale and it may be expensive at laboratory scale. New approaches in parameters estimation in this field try to combine data from both continuous and batch experiments (Girault et al., 2011).
4.2.3.
Fed-batch operation
In this process, substrate and nutrients are added continuously or intermittently into the reactor, without an output stream from it, so that the volume of the reaction media increases during the cycle of operation. Fed-batch operations
5353
are scarcely used in AD, neither at full-scale nor at a pilot/ bench/lab-scale. Hence, only a few studies have used these systems for kinetic parameter determination. Rodrigues et al. (2003) evaluated the behavior of a fed-batch reactor by fitting some apparent parameters with a simple global model of AD. Likewise, Redzwan and Banks (2004) used a simplified model in order to assess the kinetic of methane production in a fedbatch reactor. Effective volume of a fed-batch reactor always increases, whereas it remains constant in batch or continuous operation; this may make the mathematical analysis of the system more complicated. In addition, the measurement of the biogas flow has to be corrected to take this volume change into account. These issues may represent significant drawbacks for the use of fed-batch operation in kinetic parameter determination.
5. Structural adequacy and parameter identifiability After the formulation of the modeling objectives and data collection, a double question arises: a) Is the selected model structure able to fit the data? To achieve this objective, the model has to include the necessary degrees of freedom, but not too many as there is a risk of overparametrization. This risk is linked to the other side of the question. b) Once a model structure is selected, is it possible to determine a unique optimal set of parameters based on the experimental data at hand? This double-sided analysis can lead to model simplifications or, on the contrary, to the introduction of additional terms or equations, and in turn to the elimination or the introduction of some parameters. In general, AD models are mechanistic models that synthesize extensive scientific research work dedicated to understand most of the physical, chemical and biological mechanisms of the processes involved. As consequence, most of the model parameters have some physical meaning and generally some default values are available (Batstone et al., 2002). The identifiability problem is then a delicate issue where the modeler should calibrate only those parameters necessary to explain the observed mechanisms without “overfitting” the data, i.e.: an “overcalibrated” model would reproduce the experimental data pretty well but would lose predictive or exploration capability (Reichert, 2010). These structural and parametric identifiability questions are relatively seldom addressed in the reported AD modeling studies. Taylor Series Expansion is the most reported technique. It consists of calculating the successive derivatives of the output function with respect to the unknown model parameters, so as to obtain a system of independent equations in these parameters. Flotats et al. (2003) applied this approach with four of the state variables of ADM1 (Acetate, propionate, valerate and methane) so as to design an experiment for the identification of parameters related to the anaerobic degradation of valerate and the initial biomass concentrations. Noykova et al. (2002), using a more simplified
5354
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 5 3 4 7 e5 3 6 4
model, evaluated identifiability with respect to three unknown parameters based on the measurement of the biogas production. In this case, the vector of parameters was reduced in order to decrease the computational requirement of this method. In fact, this method is mostly applicable to simple models, whereas it leads to complex systems of equations in more general cases (Muller et al., 2002). Another approach to the study of identifiability is based on local and global sensitivity analysis. The goal of sensitivity analysis is to explore the change in model output resulting from a change in model parameters (kinetic or stoichiometric coefficients, parameters, input conditions, initial conditions, etc.) (Sin et al., 2011). Most of the sensitivity analysis techniques encountered in literature, for AD process, are of local nature and use a differential analysis of outputs with respect to parameters (eq. (1)): vyj yj ðqi Þ yj ðqi þ Dqi Þ ¼ Dqi vqi
(1)
Where yj corresponds to the jth output and qi to the ith parameter. Examples of local sensitivity analysis in AD modeling can be found at Tartakovsky et al. (2008) and Noykova and Gyllenberg (2000). The main drawback of this method is that it is based on the linearization of the model equations at a given set of parameter values and therefore it only describes local model behavior at this point. Other authors attempt to get a more global picture of the sensitivity by varying the parameters (one at a time) and aggregating the relative difference observed in the outputs (Vavilin et al., 2003), either by integrating the errors in time (Bernard et al., 2001), or by estimating a weighted sum of them (Wichern et al., 2009; Lin and Wu, 2011). However, none of these methods are able to detect correlations among the parameters and aggregating the errors can lead to erroneous conclusions when compensations between negative and positive terms occur. An alternative definition of sensitivity analysis is the so called “global sensitivity analysis (GSA)” and relates to uncertainty analysis. It can be viewed as an analysis of variance (ANOVA) problem (Sobol, 2001; Helton and Davis, 2003; Saltelli et al., 2006). Hence the output variance is decomposed into fractions which are attributed to the single model inputs. Examples of such sensitivity analysis methods include Morris Screening (Morris, 1991), the spectral information of measurements to characterize parameter interactions (Tarantola et al., 2006), linear regression of Monte Carlo outputs (Helton and Davis, 2003) and variance decomposition (Saltelli et al., 2008). Global sensitivity analysis has recently been applied to biological models of plant cell cultures (Mailier et al., 2011) and to activated sludge systems (Sin et al., 2011) and it offers promising perspectives in AD models.
6.
Methods for parameters estimation
Unfortunately, AD models are not universal enough and some parameters need to be estimated for each particular case study. Traditionally, they have been calibrated by a trial and error approach. However, this method is very time consuming
and does not provide any information about the uncertainty associated to the parameter values nor any guarantee about its uniqueness. The selection of the objective function (usually also called cost function), which can play a crucial role in the result of the optimization, will be reviewed in the first part of this section and will then follow with the most used techniques for parameter estimation.
6.1.
Cost function selection
In order to find the best-fit of a model to given experimental data, an appropriate criterion for the optimal solution of the model parameter vector must be selected. Several cost functions have been used for parameter identification in AD models mostly in the form of output-error criteria, i.e., measuring the deviation between the model and real system outputs. The type of selected function may influence how the optimization procedure acts and how it adjusts the parameters (Batstone et al., 2003). The most popular cost function is the sum of least squares (OLS, eq. (2)) (Noykova and Gyllenberg, 2000; Bhunia and Ghangrekar, 2008; Batstone et al., 2009; Donoso-Bravo et al., 2010; Lopez and Borzacconi, 2010), where it is implicitly assumed that the standard deviation of the measurement errors, which can be known or unknown, is constant. In eq. (2), J is the objective function, nexp are the collected measurements, nsim are the model-predicted outputs, q represents the parameters to be determined (which can include the stoichiometry, the kinetic parameters and also the unknown initial conditions of some experiments) and N is the number of measurements. Minimizing the cost function has been acknowledged as an important issue for prediction purposes or process stability (Batstone et al., 2003). JðqÞ ¼ min
N X
2 vexp ðtÞ vsim ðt; qÞ
(2)
t¼1
When the measurement errors do not have a constant standard deviation, then it is generally required to introduce weighting factors wt into eq. (2), leading to a weighted leastsquare criterion JðqÞ ¼ min
N X
2 wt vexp ðtÞ vsim ðt; qÞ
(3a)
t¼1
in scalar form, or more generally, when vectors of measurements are considered, JðqÞ ¼ min
N X
vexp ðtÞ vsim ðt; qÞ W vexp ðtÞ vsim ðt; qÞ
(3b)
t¼1
where W is a N N weighting matrix to be selected. If the measurement errors are white and normally distributed, i.e. ε w N(0,Q), then the best choice of the weighting matrix W, in a maximum likelihood sense, is the inverse of the covariance matrix of the measurement noise, i.e. W ¼ Q1. If these assumptions do not hold or a deterministic approach is preferred, some other weighting could be used. A variety of weightings have been used in published studies (Smith et al., 1998; Lokshina et al., 2001; Noykova et al., 2002;
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 5 3 4 7 e5 3 6 4
Flotats et al., 2003; Palatsi et al., 2010). For instance, the weighting factors have been estimated by calculating a local slope of the output variation (Lokshina et al., 2001), by the difference between maximum and minimum values (Palatsi et al., 2010) or by using the deviation with respect to the mean (Flotats et al., 2003). A frequent situation corresponds to constant, possibly unknown, relative errors. In this case, eq. (3a) can simply be formulated as: JðqÞ ¼ min
N X vexp ðtÞ vsim ðt; qÞ 2 t¼1
vexp ðtÞ
(3c)
or, noting that a constant absolute error on a logarithm, is equivalent to a constant relative error on its argument: JðqÞ ¼ min
N X 2 ln vexp ðtÞ ln ðvsim ðt; qÞÞ
(3d)
t¼1
as it has been used successfully in several studies, e.g. (Batstone et al., 2003; Haag et al., 2003; Vande Wouwer et al., 2006) When a cost function has been formulated, a numerical procedure has to be used to minimize it with respect to the unknown parameters.
6.2.
Optimization techniques
In order to avoid the tedious trial and error approach, several algorithms have been developed. They are search techniques that numerically approach the optimum parameter values by optimizing an objective function. These algorithms can be divided into local algorithms and global algorithms.
6.2.1.
Local methods
The vast majority of optimization methods are local in nature, i.e. they assume the convexity of the cost function. When this condition is not fulfilled, and a local method is applied, there is a high risk that the algorithm will get trapped into a local minimum. To alleviate this problem, it is recommended to start the search from several randomly selected initial parameter values so as to explore the parameter space. This procedure, called multi-start strategy (Kocsis and Gyo¨rgy, 2009), allows the assessment of the problem multimodality and, for instance by drawing a histogram of the frequency of occurrences of the different minima, to determine the global minimum and its basin of attraction (i.e., the ball-sized region containing the initial guesses leading to the global optimum). As initialization is of paramount importance, it is advised to decompose a complex optimization problem into several simpler ones, whenever possible, and to use the solution of these intermediate problems as initial guesses for the next step. This type of procedure has been successfully applied in the identification of bioprocess models by (Hulhoven et al., 2005). Among local methods, one basically distinguishes gradient-based methods, which make use of the first- and, in some cases, the second-order derivatives of the cost function, and gradient-free methods which do not require the cost function differentiability such as the direct-search methods. Another important feature of the optimization problem is the
5355
presence of equal and inequal constraints. These methods are briefly described in the sequel with respect to their application to the estimation of parameters in AD models.
6.2.1.1. Simple unconstrained optimization: steepest descent, GausseNewton and LevenbergeMarquardt methods. The steepest descent is an iterative method which uses first-order information to move downhill in the gradient direction. A large number of iterations are often required to achieve convergence. GausseNewton method is particularly well suited to the minimization of sum of squares (and therefore to the least-squares approach) and avoids the costly evaluation of the Hessian (second-order information) by building an approximation based on the Jacobian while preserving the quadratic convergence of the original method of Newton. The LevenbergeMarquard method (LMA) (Marquardt, 1963) blends the two previous methods, steepest descent and GausseNewton. LMA usually starts using a steepest descent method and progressively becomes a GausseNewton method as it gets closer to the optimum. This way, the algorithm is more robust than GausseNewton but achieves better convergence than steepest descent. LMA has been commonly applied to parameter identification in AD models for the treatment of livestock manure (Garcia-Ochoa et al., 1999), raw industrial wine distillery vinasses (Aceves-Lara et al., 2005; Martin et al., 2002), baker’s yeast effluents (Deveci and Ciftci, 2001), low temperature acetoclastic methanogenesis (Lokshina et al., 2001) and animal wastes from calf farms (Simeonov 1999). Aceves-Lara et al. (2005) combined this algorithm with an asymptotic observer to evaluate the parameters kinetics.
6.2.1.2. Non-linear constrained optimization: sequential quadratic programming. Sequential Quadratic Programming (SQP) is one of the most successful methods for the numerical solution of constrained non-linear optimization problems. SQP (Nocedal and Wright, 2006) is an iterative method which solves, for each iteration, a quadratic problem (QP), i.e., a quadratic approximation of the objective function subject to a linearization of the constraints. If the problem is unconstrained, then the method reduces to the Newton method. For solving the QP problem under inequality constraints, a variety of methods are commonly used, including among others interior point, active set. SQP has been used for parameter estimation in AD models (Sales-Cruz and Gani, 2004; AcevesLara et al., 2005).
6.2.1.3. Multiple shooting. In the previous methods, a sequential optimization approach has always been assumed, i.e., the optimization algorithm repeatedly evaluates the cost function by a call to a time integrator which numerically solves the dynamic equations of the process model (for instance, an LMA algorithm evaluates the cost function through the solution of the model differential equations e which depends on the current values of the model parameters e using a Runge-Kutta method). There is another family of methods which rather use a simultaneous approach, i.e., discretize the model differential equations and uses the resulting set of algebraic equations as constraints to the optimization algorithm. Multiple
5356
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 5 3 4 7 e5 3 6 4
shooting (Ascher et al., 1995) discretizes the time span into time intervals (ti, tiþ1) and new optimization variables are introduced which correspond to initial conditions of the state variables on each interval. A set of boundary conditions ensures the continuity of the solution. As time integration is performed over short time intervals, the numerical stability property of the algorithm is improved. In addition, state constraints can be easily incorporated. Recently, multiple shooting has been used for parameter estimation in AD (Muller et al., 2002; Lopez and Borzacconi, 2010).
6.2.1.4. Direct-search methods. Direct-search methods (Lewis et al., 2000) are derivative-free methods, which do not even require numerical function values since the relative rank of objective values is sufficient. Several classes of methods exist, such as pattern search methods, simplex methods, and methods with adaptive sets of search directions. These methods date back to the 60’s and have since been replaced by more sophisticated techniques. However, they still have an undisputed popularity due to their simplicity of use and good performance in practical use. In engineering applications, the simplex methods have always been in wide use. The basic idea of simplex search is to construct a nondegenerate simplex in the parameter space and use the simplex to drive the search (a simplex is a set of n þ 1 points in the n dimensional space, e.g. a triangle in 2D; a nondegenerate simplex is one for which any point in the domain of the search can be constructed by taking linear combinations of the edges adjacent to any given vertex). Not only does the simplex provide a frugal design for sampling the space, it has the added feature that if one replaces a vertex by reflecting it through the centroid of the opposite face, then the result is also a simplex. It means that one can proceed parsimoniously, reflecting one vertex at a time, in the search for an optimizer. The simplex algorithm is usually less sensitive to local minima than the gradient-based methods, such as the LevenbergeMarquardt method. However, the convergence is usually slower and closer to the optimum, and the algorithm of course does not provide any sensitivity information (Jacobian) that could be used as a byproduct to estimate the Fisher Information Matrix; as will be introduced in the sequel. The simplex algorithm has been widely applied to parameter estimation in AD models (Mosche and Jordening, 1999; Simeonov 1999; Ruel et al., 2002; Haag et al., 2003; Guisasola et al., 2009; Lopez and Borzacconi, 2010)
6.2.2.
Global methods
Non-linear parameter identification problems are often characterized by the presence of various local minima. Global optimization is aimed at finding the best solution to these kinds of problems. This is a very active research area and two main families of methods have emerged: deterministic algorithms, including for instance grid search and branch and bound, and stochastic algorithms including for instance simulated annealing, tabu search, genetic algorithms, differential evolution, ant colony optimization and particle swarm optimization. Among all these methods, Simulated Annealing (SA), Genetic Algorithms (GA), Particle Swarm Optimization (PSO)
have been quite popular in engineering applications. SA is a probabilistic algorithm, whose initial idea comes from annealing in metallurgy, a technique involving heating and controlled cooling of a material to increase the size of its crystals and reduce their defects. The objective function to be minimized is analogous to the internal energy of the system. GA belong to the larger class of evolutionary algorithms (EA), which generate solutions to optimization problems using techniques inspired by natural evolution, such as inheritance, mutation, selection, and crossover. PSO are search algorithms based on the simulation of the animal social behavior in a group. The advantages of these algorithms is that they do not require the objective function to be differentiable as in classic local (gradient-based) optimization algorithms, which make few assumptions about the problem to be solved, and can explore a large space of candidate solutions. Although, these algorithms produce better solutions they do not guarantee that an optimal solution will ever be found at the price of large amounts of computation. These algorithms have found applications in the identification of AD models, for instance Simulated Annealing (Haag et al., 2003), Genetic Algorithms (Jeong et al., 2005; Abu Qdais et al., 2010; Wichern et al., 2009), and Particle Swarm Optimization (Wolf et al., 2008).
6.3.
What about optimization constraints?
Constraints in the estimated parameter values are usually employed in the optimization process as long as they are allowed by the selected technique. Simple and logic constraints are the most used ones, such as: positive values and within certain reasonable ranges (i.e., parameters between some minimum and maximum values that the experimenter could determine from past experience). Nevertheless, sometimes linear and non-linear constraints can be useful during parameters estimation since they enable the inclusion of conservations laws (i.e. yields) and avoid some mathematical uncertainties linked to the model’s structure (i.e. pH and gas transfers). In other cases, it is practical to estimate separately some parameters, e.g. volumetric coefficient of mass transfer (kLa), with iterative linear approximations in order to simplify the estimation task (Batstone, 1999). On the other hand, when using sophisticated non-linear constrained algorithms is not necessary, and simple methods such as the simplex can be used, some transformation of the parameter space can be done. For instance, positivity constraints can be imposed using a logarithmic transformation, i.e. if q ¼ ln (l) is a positive unknown parameter, then the optimization algorithm explore the full real parameter space (positive and negative) for l. This technique has been applied for instance in Vande Wouwer et al. (2006).
6.4.
Alternative methods: The Bayesian inference
The Bayesian approach opens a new calibration framework because it abandons the idea of believing that the model parameters are fixed and not known, and treats them as probabilistic (or random) variables having a probability density function. This function is called joint posterior
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 5 3 4 7 e5 3 6 4
distribution of parameters and defines subjective beliefs of parameter values by summarizing the state of knowledge about the system performance (Omlin and Reichert, 1999). Equation (4) shows a very general expression of the Bayes formula: pðqjyÞ ¼
pðyjqÞ pðqÞ pðyjqÞ pðqÞ fpðyjqÞ pðqÞ ¼Z pðyÞ pðyjqÞ pðqÞdq
(4)
q
where q is the model’s parameters vector and y represents experimental data. The Bayes theorem states that the posterior probability function, p(qjy), is proportional to the multiplication of the prior beliefs, p(q), and the likelihood function of observations, p( yjq). The prior probability distribution, p(q), expresses modeler prior beliefs about possible parameter values while the posterior, p(qjy), expresses the posterior beliefs after having evaluated the model residuals. The likelihood function p( yjq) plays a very important role in Bayes’ formula because it is the function through which the data y modifies prior knowledge of q. Finally, the probability of the observations, p( y), is the expected value of the likelihood function over the parameter space, and acts as a normalizing constant. Qian et al. (2003) explained that the most important limitation of using Bayesian methods for scientific inference was that analytical solutions of the posterior distributions are available for fairly limited combinations of model forms and probability distributions (such as the linear model leading to a normal distribution of the residuals). For most non-linear models, or models with a large number of parameters to be calibrated, the estimation of the posterior likelihood becomes intractable. Fortunately, advent of fast and inexpensive computing has promoted the implementation of numerical techniques. Particularly important are the Markov Chain Monte Carlo (MCMC) techniques that construct a Markov chain which asymptotically converges into the posterior distribution. The Bayesian techniques are especially advisable in the case of poor parameter identifiability because subjective prior knowledge about possible parameter values can be used. Omlin and Reichert (1999) stress that in environmental modeling the use of complex model structures and limited experimental data makes the Bayesian techniques very important. An interesting application of Bayesian calibration of an AD model has been recently proposed by Martin et al. (2011). In this example, a digester model (De Gracia et al., 2009) able to work under anaerobic and aerobic conditions is calibrated by using a complete so-called Integrated Monte Carlo Methodology (Martin and Ayesa, 2010).
6.5. Some considerations for parameter estimation in AD models The complexity and particularities of AD processes give rise to some considerations when trying to calibrate a model.
6.5.1.
Steady states analysis
When the selected model is simple and some convenient assumptions are considered (mainly that the system has
5357
reached a steady state condition) some mathematical modifications can be done in order to get rid of the ODE and to express the outputs of the systems as a functions of other variables. Once these expressions are obtained, simple linear regression may be used to estimate the model parameters. Bernard et al. (2001) used this approach to draw some expressions in order to estimate several kinetic parameters, stoichiometric coefficients and one mass transfer constant of a 2-reaction model in the AD of wine distillery wastewater. Likewise, Simeonov et al. (1996) employed steady state analysis to estimate some kinetic parameters of a 3-reaction model in the AD of different type of animal waste at labscale. Both studies evaluated the models with 7 steady states varying either the dilution rate or the organic matter concentration in the influent. The main drawback of this method is that reaching steady state conditions requires a highly controlled reactor, which is quite difficult to achieve in a full-scale reactor because of the flow and concentration disturbances. In this context, Bhunia and Ghangrekar (2008) assessed the application of the linearized form of the non-linear expression drawn from the steady states analysis with three simple models (Monod, Haldane and second-order functions) in the treatment of synthetic sucrose-based wastewater using labscale UASB reactors. Overall, non-linear optimization showed better performance than linearization.
6.5.2.
Mass continuity (conservation laws)
The problem of mass continuity in the context of AD has been analyzed (Banks et al., 2011; Ekama et al., 2007; De Gracia et al., 2006; Huete et al., 2006). Tracking the mass trajectories of the elemental compounds (C, H, O, N and P) can uncover hidden processes when studying experimental data or identify inadequate model structures when analyzing model results. Concerning the latter objective, De Gracia et al. (2006) proposed a mass and charge conservation check methodology to verify the consistency of AD models. On the other hand, the equations of the mass continuity can represent by themselves biochemical transformation models. This is the case of Zaher et al. (2009) where a simple AD model is presented to study the microbial activity in the treatment of dairy manure. The study of the energy balance is also crucial when analyzing the process’s performance. Banks et al. (2011) presented a complete energy study to assess the performance of the anaerobic digestion of source-segregated domestic food waste. In the modeling field, De Gracia et al. (2009) incorporated the energy balance equation to model the digestion of sludge generated in wastewater treatment plants. Lu¨bken et al. (2007) proposed a modified version of ADM1 model to simulate energy production in the digestion of cattle manure and renewable energy crops. A thermodynamic analysis of the acidogenic reaction was performed by Bastidas-Oyanedel et al. (2008), where was demonstrated that the energy transfer efficiency is influenced by some operational conditions, such as pH and hydraulic retention time. The mass continuity equations have also been used to characterize input conditions in terms of substrate characteristics (Huete et al., 2006). In the case of sludge produced by wastewater treatment plants, and when trying to characterize it in terms of the ADM1 model, Huete et al. (2006) points out
5358
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 5 3 4 7 e5 3 6 4
that “the biggest uncertainty comes from the elemental composition of the organic composites and the inert soluble and particulate components, since the elemental mass fractions depend on the specific case under study”. When approaching the same problem, Ekama et al. (2007) found that the non-biodegradable particulate organics, including both (a) those originated in influent wastewater, and (b) those generated in the activated sludge endogenous process, are also nonbiodegradable under anaerobic conditions. Grau et al. (2007a) proposed a new model building philosophy called plant wide modeling which indirectly addresses the characterization problem of sludge. It proposes to generate a list of non-redundant model components for the biochemical processes and to define them in terms of C, H, N, O, P, charge and other possible elements. This extensive description allows for a straightforward relationship between the model components and the most common analytical measurements carried out in the wastewater sludge. Using this approach De Gracia et al. (2011) has presented a new tool to characterize the wastewater sludge in terms of the model components. This estimation is carried out by minimising a cost function (Grau et al., 2007b) and taking into account the physical and chemical restrictions due to the components’ elemental composition.
6.5.3.
Initial conditions
Establishing the initial conditions of the state variables of the selected model, which in many articles is merely omitted, is one the first issues that have to be defined before the optimization process itself. As previously mentioned, several experimental methods for determining substrate characterization may be used, thus knowing the initial conditions for the different organic compounds considered in the model should not de considered a big issue. However, setting the initial concentration of all the microbial populations of the model, especially in large model such as ADM1, is quite complicated considering the difficulty of its experimental estimation. On the other hand, the type of operation mode has an influence in how accurate the initial conditions has to be established. In batch tests, the initial conditions are the sole inputs of the system, hence the initial values of the state variables will exert great influence in the model behavior. By contrast, in continuous systems the initial condition had a negligible effect especially in the case of long-term operation evaluations, as long as an initial simulation period is not taken into account, allowing convergence (Batstone et al., 2003). In other cases, initial values of the microbial population have been either arbitrarily fixed (Knobel and Lewis, 2002; Noykova et al., 2002; Nopharatana et al., 2003; Ozkan-Yucel and Gokcay, 2010) or estimated through a prior simulation evaluation (Batstone et al., 2004b). A typical strategy to achieve this is to run, for a very long time, a steady state simulation of a similar system to that from where the sludge comes from and use the biomass relative composition from the results of such simulation and the total biomass from an available measurement in the real system to be simulated. In some cases, the initial concentration of each population has been considered as an unknown value, thus it has been estimated in the optimization process by using adequate experimental information
(Haag et al., 2003; Flotats et al., 2006; Palatsi et al., 2010); however, some important identification problems have been encountered in other studies (Flotats et al., 2003; Jeong et al., 2005). Regardless, estimating the initial conditions of the biomass composition is the most recommended choice, especially in the case of batch test, since besides being valuable information, it would not be appropriate to force the bestfit by fixing these initial values.
7.
Parameter uncertainty estimation
If the model has passed direct validation, and whenever possible cross validation (this later step can be made difficult by the scarcity of experimental data), it is interesting to analyze further the accuracy of the model parameters, and to provide confidence intervals for the parameters and in turn, for the model prediction.
7.1.
Error covariance matrix
The Cramer-Rao bound (Walter and Pronzato, 1997), which corresponds to the inverse of the Fisher Information Matrix (FIM), provides an optimistic estimate of the parameter error covariance matrix, i.e., information on the parameter standard deviation and correlation. The FIM can be computed using the output parameter sensitivity matrix and the inverse of the covariance matrix of measurement noise. In AD systems, FIM has been evaluated, for instance in Noykova et al. (2002), Flotats et al. (2003) and Haag et al. (2003). In the case of simple models, the parametric sensitivity can be calculated analytically as in Lokshina et al. (2001). For more complex models, a numerical procedure is preferred, e.g. finite differences. When using a gradient-based method, e.g. LevenbergeMarquardt, it is usually possible to extract the sensitivities at the optimum (as a byproduct of the algorithm computation). It is important to keep in mind that the inverse of FIM just provides a (maybe too) optimistic estimate of the parameter error covariance matrix, as has been pointed out in the case of large and non-linear biochemical systems by Schenkendorf et al. (2009), Joshi et al. (2006) and Lopez and Borzacconi (2010).
7.2.
Confidence intervals
The covariance matrix of the parameter errors, as evaluated in the previous subsection, together with a model linearization, allows the computation of confidence intervals for the model prediction. These intervals are of course only approximations, whose quality depends on the model non-linearity in the parameters, but they provide a first view of the model’s uncertainty. More accurate, but also more computationally demanding, approaches to the estimation of confidence intervals have been proposed (Dochain and Vanrolleghem, 2001). Confidence regions have been obtained by Batstone et al. (2003), Batstone et al. (2004b), Kalfas et al. (2006) and Batstone et al. (2009), in the case of the estimation of kinetic parameters or stoichiometric coefficients of ADM1 for the description of the AD of several substrates.
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 5 3 4 7 e5 3 6 4
7.3.
Joint posterior distribution
In the Bayesian approach the model parameters are defined as random variables, and therefore, the uncertainty of the parameters is defined by the joint posterior distribution. In this case, the posterior distribution of parameters is generally derived by a Markov Chain Monte Carlo technique, i.e., using a numerical implementation. This is a great advantage in the case of dynamic non-linear models because: first, the calibration method is independent of the model structure; and second, the uncertainty of the parameters is estimated by the distribution of their most feasible values, without making any assumptions about the model structure.
8.
Model validation
Once a set of parameters has been obtained, it is necessary to question the predictive quality of the resulting model and to assess the parameter accuracy. This will determine the confidence behind the model, and tell the modeler if he needs to revise the model’s identification. The overall procedure, called model validation, consists of several steps.
8.1.
Direct validation
The first test is to check whether the model is able to reproduce the experimental data that has been used for parameter identification. Otherwise, there is obviously something wrong in the identification procedure (see details in Section 2) and it has to be adjusted and repeated. There are different ways of checking the model’s adequacy. One of the best, even if it cannot be cast into mathematical formulas, is the visual inspection: the model has to follow well the data evolution while smoothing off the noise (a model that tends to reproduce noise is overparametrized and will fail later on in cross-validation tests). Model performance by visual inspection is widely used in AD system, and has even been the only applied method in many cases. On a more mathematical basis, a good test is based on residuals analysis. If the model is predicting the data well, the residual can be directly related to the measurement error. Different types of information may be drawn from the residuals, such as the determination coefficient (R2), an estimation of the variance of the data, analysis of randomness, etc. Actually, the determination coefficient has been the sole tool used to evaluate the model fit in several studies (Redzwan and Banks, 2004; Flotats et al., 2006; Palatsi et al., 2010). This parameter was also used by Flotats et al. (2003) to evaluate the model fit quality as part of a more detailed model analysis. Other statistical tests have been used to compare several models, such as the Fisher test (Aceves-Lara et al., 2005), the sum of normalized errors (Bhunia and Ghangrekar, 2008) or the determination coefficient (Donoso-Bravo et al., 2010). Nevertheless, few studies have estimated the variance of the experimental data through the residuals (Haag et al., 2003). Analysis of randomness in the residual (another method to evaluate the fit quality) is seldom performed, even though it
5359
may represent a good criterion to determine autocorrelation or the need to change the formulation of the cost function. A particular case is the study developed by Barampouti et al. (2005) who used this type of statistical analysis to generate an empirical model to predict the biogas production and to evaluate goodness-of-fit with the simulated data.
8.2.
Cross validation
Direct validation is a necessary condition, but by no means a sufficient condition to accept a model as being one that can reproduce the behavior of the system under consideration. It may well be that the model fits the data that has been used for identification adequately, but performs poorly with new unseen data. To this end, enough data must be available and divided these into two subsets, one for parameter identification (and afterward direct validation), and the other for cross validation. Notice that cross validation can in turn imply the identification of the initial conditions of the experiments under consideration (i.e., the initial conditions of the unseen experiments are sometimes unknown, or at least, know but no well, and have to be estimated before checking that the model with the previously identified parameters fits well the new data). This procedure has been applied to check the AD model validity, especially when complex models, such as ADM1, are considered (Ozkan-Yucel and Gokcay, 2010; Fezzani and Ben Cheikh, 2009; Fezzani and Ben Cheikh, 2008; Tartakovsky et al., 2008; Siegrist et al., 2002; Lu¨bken et al., 2007). In the same context, short calibration steps may also be performed regularly during the validation of the model (Batstone et al., 2009; Bernard et al., 2001), in order to take the possible variations of the substrate characteristics as well as the changes in the anaerobic population into account especially when long-term operation data are used.
9.
Conclusion
Anaerobic digestion is a very complex process involving various bacterial populations and substrates. With the progresses in instrumentation and in computer science, the development of mathematical models, predicting the dynamic process behavior has attracted considerable attention in the last two decades. ADM1 is undoubtedly one of the milestones of this research era. However, modeling is always a goal-driven exercise, and many alternative models have been proposed in the literature, depending on the aim, e.g., process understanding, dynamic simulation, optimization, or control. Models contain unknown parameters, e.g., initial conditions, stoichiometry, and kinetic parameters which have to be estimated from experimental data. Parameter identification is a delicate task due the potentially large number of parameters and the scarcity of informative experimental data. This review attempts to summarize the efforts that have been accomplished in the parameter estimation of models of anaerobic digestion processes and highlights the critical steps of the identification procedure. In general, the literature shows a lack of systematic and clear procedure for modeling AD processes. Also, sets of parameter values are reported without
5360
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 5 3 4 7 e5 3 6 4
thorough analysis of the model’s validity and parameter accuracy, which makes it difficult to exploit all of the published information. This situation will certainly improve with more awareness of these important issues. As a general recommendation, the development of benchmarks, and the availability of data bases, as open resources on the internet, would certainly speed up these developments and consolidate knowledge in the field.
Acknowledgments This paper presents research results of the Belgian Network DYSCO (Dynamical Systems, Control, and Optimization), funded by the Interuniversity Attraction Poles Programme, initiated by the Belgian State, Science Policy Office. The scientific responsibility rests with its author(s). This study is also supported by a grant from Belspo (Belgian Science Policy) through its Postdoc fellowships to non-EU researchers program. Johan Mailier is a research fellow supported by the FNRS (Belgian National Fund for Scientific Research).
references
Abu Qdais, H., Hani, K.B., Shatnawi, N., 2010. Modeling and optimization of biogas production from a waste digester using artificial neural network and genetic algorithm. Resources Conservation and Recycling 54 (6), 359e363. Aceves-Lara, C.A., Aguilar-Garnica, E., Alcaraz-Gonzalez, V., Gonzalez-Reynoso, O., Steyer, J.P., Dominguez-Beltran, J.L., Gonzalez-Alvarez, V., 2005. Kinetic parameters estimation in an anaerobic digestion process using successive quadratic programming. Water Science and Technology 52 (1e2), 419e426. Angelidaki, I., Ellegaard, L., Ahring, B.K., 1993. A mathematical model for dynamic simulation of anaerobic digestion of complex substrates: focusing on ammonia inhibition. Biotechnology and Bioengineering 42 (2), 159e166. Angelidaki, I., Ellegaard, L., Ahring, B.K., 1999. A comprehensive model of anaerobic bioconversion of complex substrates to biogas. Biotechnology and Bioengineering 63 (3), 363e372. Appels, L., Baeyens, J., Degreve, J., Dewil, R., 2008. Principles and potential of the anaerobic digestion of waste-activated sludge. Progress in Energy and Combustion Science 34 (6), 755e781. Ascher, U.M., Mattheij, R.M.M., Russell, R.D., 1995. Numerical Solution of Boundary Value Problems for Ordinary Differential Equations. Classics in Applied Mathematics. SIAM, Philadelphia, PA. Baltes, M., Schneider, R., Sturm, C., Reuss, M., 1994. Optimal experimental design for parameter estimation in unstructured growth models. Biotechnology Progress 10 (5), 480e488. Banks, C.J., Chesshire, M., Heaven, S., Arnold, R., 2011. Anaerobic digestion of source-segregated domestic food waste: performance assessment by mass and energy balance. Bioresource Technology 102 (2), 612e620. Barampouti, E.M.P., Mai, S.T., Vlyssides, A.G., 2005. Dynamic modeling of biogas production in an UASB reactor for potato processing wastewater treatment. Chemical Engineering Journal 106 (1), 53e58. Bastidas-Oyanedel, J.R., Aceves-Lara, C.A., Ruiz-Filippi, G., Steyer, J.P., 2008. Thermodynamic analysis of energy transfer
in acidogenic cultures. Engineering in Life Sciences 8 (5), 487e498. Batstone, D., 1999. High Rate Anaerobic Treatment of Complex Wastewater. PhD thesis, Department of Chemical Engineering, University of Queensland. Batstone, D., Keller, J., Newell, R., Newland, M., 2000. Modelling anaerobic degradation of complex wastewater. I: model development. Bioresource Technology 75, 67e74. Batstone, D.J., Keller, J., Angelidaki, I., Kalyuzhnyi, S.V., Pavlostathis, S.G., Rozzi, A., Sanders, W.T., Siegrist, H., Vavilin, V.A., 2002. Anaerobic Digestion Model No. 1. (ADM1). Iwa Task Group, IWA Task Group for Mathematical Modelling of Anaerobic Digestion Processes, IWA Publishing, London. Batstone, D.J., Pind, P.F., Angelidaki, I., 2003. Kinetics of thermophilic, anaerobic oxidation of straight and branched chain butyrate and valerate. Biotechnology and Bioengineering 84 (2), 195e204. Batstone, D.J., Keller, J., 2003. Industrial applications of the IWA anaerobic digestion model No. 1 (ADM1). Water Science and Technology 47 (12), 199e206. Batstone, D., Keller, J., Bonnet, B., 2004a. The influence of substrate kinetics on the microbial community structure in granular anaerobic biomass. Water Research 38 (6), 1390e1404. Batstone, D.J., Torrijos, M., Ruiz, C., Schmidt, J.E., 2004b. Use of an anaerobic sequencing batch reactor for parameter estimation in modelling of anaerobic digestion. Water Science and Technology 50 (10), 295e303. Batstone, D., Hernandez, J.L.A., Schmidt, J.E., 2005. Hydraulics of laboratory and full-scale upflow anaerobic sludge blanket (UASB) reactors. Biotechnology and Bioengineering 91 (3), 387e391. Batstone, D.J., 2006. Mathematical modelling of anaerobic reactors treating domestic wastewater: rational criteria for model use. Re-views in Environmental Science and Biotechnology 5 (1), 57e71. Batstone, D.J., Tait, S., Starrenburg, D., 2009. Estimation of hydrolysis parameters in full-scale anerobic digesters. Biotechnology and Bioengineering 102 (5), 1513e1520. Bernard, O., Zakaria, H., Dochain, D., 2000. Software sensors to monitor the dynamics of microbial communities: application to anaerobic digestion. Acta Biotheoretica 48, 197e205. Bernard, O., Hadj-Sadok, Z., Dochain, D., Genovesi, A., Steyer, J.P., 2001. Dynamical model development and parameter identification for an anaerobic wastewater treatment process. Biotechnology and Bioengineering 75 (4), 424e438. Bernard, O., Chachuat, B., Helias, A., Rodriguez-, J., 2006. Can we assess the model complexity for a bioprocess: theory and example of the anaerobic digestion process. Water Science and Technology 53 (1), 85e92. Bhunia, P., Ghangrekar, M.M., 2008. Analysis, evaluation, and optimization of kinetic parameters for performance appraisal and design of UASB reactors. Bioresource Technology 99 (7), 2132e2140. Boe, K., Batstone, D.J., Steyer, J.P., Angelidaki, I., 2010. State indicators for monitoring the anaerobic digestion process. Water Research 44 (20), 5973e5980. Chen, Y., Cheng, J.J., Creamer, K.S., 2008. Inhibition of anaerobic digestion process: a review. Bioresource Technology 99 (10), 4044e4064. Costello, D.J., Greenfield, P.F., Lee, P.L., 1991. Dynamic modeling of A single-Stage high-rate anaerobic reactor .1. Model derivation. Water Research 25 (7), 847e858. Dalmau, J., Comas, J., Rodriguez-Roda, I., Pagilla, K., Steyer, J.P., 2010. Model development and simulation for predicting risk of foaming in anaerobic digestion systems. Bioresource Technology 101 (12), 4306e4314. de Baere, L. A., Mattheeuws, B., Velghe, F., 2010. State of the art of anaerobic digestion in Europe. 12th World Congress on
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 5 3 4 7 e5 3 6 4
Anaerobic Digestion (AD12), Guadalajara, Mexico, October 31st to November 4th, 2010. De Gracia, M., Sancho, L., Garcı´a-Heras, J.L., Vanrolleghem, P., Ayesa, E., 2006. Mass and charge conservation check in dynamic models: application to the new ADM1 model. Water Science and Technology 53 (1), 225e240. De Gracia, M., Grau, P., Huete, E., Go´mez, J., Garcı´a-Heras, J.L., Ayesa, E., 2009. New generic mathematical model for WWTP sludge digesters operating under aerobic and anaerobic conditions: model building and experimental verification. Water Research 43 (18), 4626e4642. De Gracia, M., Huete, E., Beltra´n, S., Grau, P., Ayesa, E., 2011. Automatic characterisation of primary, secondary and mixed sludge inflow in terms of the mathematical generalised sludge digester model. Water Science and Technology 64 (3), 557e567. Derbal, K., Bencheikh-Lehocine, M., Cecchi, F., Meniai, A.-H., Pavan, P., 2009. Application of the IWA ADM1 model to simulate anaerobic co-digestion of organic waste with waste activated sludge in mesophilic condition. Bioresource Technology 100, 1539e1543. Deveci, N., Ciftci, G., 2001. A mathematical model for the anaerobic treatment of baker’s yeast effluents. Waste Management 21 (1), 99e103. Dimitrova, N., Krastanov, M., 2009. Nonlinear stabilizing control of an uncertain bioprocess model. International Journal of Applied Mathematics and Computer Science 19 (3), 441e454. Dochain, D., Vanrolleghem, P., 2001. Dynamical Modelling & Estimation in Wastewater Treatment Processes. IWA Publishing, London. Donoso-Bravo, A., Carballa, M., Ruiz-Filippi, G., Chamy, R., 2009a. Treatment of low strength sewage with high suspended organic matter content in an anaerobic sequencing batch reactor and modelling application. Electronic Journal of Biotechnology 12 (3), 13e14. Donoso-Bravo, A., Retamal, C., Carballa, M., Ruiz-Filippi, G., Chamy, R., 2009b. Influence of temperature on hydrolysis, acidogenesis and methanogenesis in anaerobic digestion: parameter identification and modeling application. Water Science and Technology 60 (1), 9e17. Donoso-Bravo, A., Perez-Elvira, S., Fdz-Polanco, F., 2010. Application of simplified models for anaerobic biodegradability tests. Evaluation of pre-treatment processes. Chemical Engineering Journal 160, 607e614. Donoso-Bravo, A., Garcia, G., Perez-Elvira, S.I., Fdz-Polanco, F., 2011. Initial rate technique as a procedure to predict the anaerobic digester operation. Biochemical Engineering Journal 53, 275e280. Eastman, J.A., Ferguson, J.F., 1981. Solubilization of particulate organic-carbon during the acid phase of anaerobic digestion. Journal of Water Pollution Control Federation 53 (3), 352e366. Ekama, G.A., So¨temann, S.W., Wentzel, M.C., 2007. Biodegradability of activated sludge organics under anaerobic conditions. Water Research 41 (1), 244e252. Fedorovich, V., Lens, P., Kalyuzhnyi, S., 2003. Extension of anaerobic digestion model No. 1 with processes of sulfate reduction. Applied Biochemistry and Biotechnology 109 (1e3), 33e45. Fezzani, B., Ben Cheikh, R., 2008. Implementation of IWA anaerobic digestion model No. 1 (ADM1) for simulating the thermophilic anaerobic co-digestion of olive mill wastewater with olive mill solid waste in a semi-continuous tubular digester. Chemical Engineering Journal 141 (1e3), 75e88. Fezzani, B., Ben Cheikh, R., 2009. Extension of the anaerobic digestion model No. 1 (ADM1) to include phenolic compounds biodegradation processes for the simulation of anaerobic co-digestion of olive mill wastes at thermophilic temperature. Journal of Hazardous Materials 162 (2e3), 1563e1570.
5361
Flotats, X., Ahring, B.K., Angelidaki, I., 2003. Parameter identification of thermophilic anaerobic degradation of valerate. Applied Biochemistry and Biotechnology 109 (1e3), 47e62. Flotats, X., Palatsi, J., Ahring, B.K., Angelidaki, I., 2006. Identifiability study of the proteins degradation model, based on ADM1, using simultaneous batch experiments. Water Science and Technology 54 (4), 31e39. Gali, A., Benabdallah, T., Astals, S., Mata-Alvarez, J., 2009. Modified version of ADM1 model for agro-waste application. Bioresource Technology 100, 2783e2790. Garcia-Ochoa, F., Santos, V.E., Naval, L., Guardiola, E., Lopez, B., 1999. Kinetic model for anaerobic digestion of livestock manure. Enzyme and Microbial Technology 25 (1e2), 55e60. Gavala, H.N., Angelidaki, I., Ahring, B.K., 2003. Kinetics and modeling of anaerobic digestion process. Advances in Biochemical Engineering/biotechnology 81, 57e93. Ghaniyari-Benis, S., Martin, A., Borja, R., 2010. Kinetic modelling and performance prediction of a hybrid anaerobic baffled reactor treating synthetic wastewater at mesophilic temperature. Process Biochemistry 45 (10), 1616e1623. Girault, R., Rousseau, P., Steyer, J., Bernet, N., Beline, F., 2011. Combination of batch experiments with continuous reactor data for ADM1 calibration: application to anaerobic digestion of pig slurry. Water Science and Technology 63 (11), 2575e2582. Grady, C.P.L., Smets, B.F., Barbeau, D.S., 1996. Variability in kinetic parameter estimates: a review of possible causes and a proposed terminology. Water Research 30 (3), 742e748. Grau, P., de Gracia, M., Vanrolleghem, P.A., Ayesa, E., 2007a. A new plant-wide modelling methodology for WWTPs. Water Research 41 (19), 4357e4372. Grau, P., Beltra´n, S., de Gracia, M., Ayesa, E., 2007b. New mathematical procedure for the automatic estimation of influent characteristics in WWTPs. Water Science and Technology 56 (8), 95e106. Guisasola, A., Baeza, J.A., Carrera, J., Sin, G., Vanrolleghem, P.A., Lafuente, J., 2006. The influence of experimental data quality and quantity on parameter estimation accuracy. Education for Chemical Engineers 1, 139e145. Guisasola, A., Sharma, K.R., Keller, J., Yuan, Z.Q., 2009. Development of a model for assessing methane formation in rising main sewers. Water Research 43 (11), 2874e2884. Haag, J.E., Vande Wouwer, A., Queinnec, I., 2003. Macroscopic modelling and identification of an anaerobic waste treatment process. Chemical Engineering Science 58, 4307e4316. Helton, J.C., Davis, F.J., 2003. Latin hypercube sampling and the propagation of uncertainty in analyses of complex systems. Reliability Engineering and System Safety 81, 23e69. Hill, D.T., Barth, C., 1977. A dynamical model for simulation of animal waste digestion. Journal of Water Pollution Control Association 10, 2129e2143. Hill, D.T., 1982. A comprehensive dynamic model for animal waste methanogenesis. Transactions of the ASAE 25 (5), 1374e1380. Huete, E., de Gracia, M., Ayesa, E., Garcia-Heras, J.L., 2006. ADM1based methodology for the characterisation of the influent sludge in anaerobic reactors. Water Science and Technology 54 (4), 157e166. Hulhoven, X., Vande Wouwer, A., Bogaerts, P., 2005. On a systematic procedure for the predetermination of macroscopic reaction schemes. Journal of Bioprocess and Biosystems Engineering 27, 283e291. Husain, A., 1998. Mathematical models of the kinetics of anaerobic digestion e a selected review. Biomass and Bioenergy 14 (5e6), 561e571. Illanes, A., 2008. Enzyme Biocataysis. Principles and Applications. Springer Science, Valparaiso.
5362
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 5 3 4 7 e5 3 6 4
Jacobi, H.F., Moschner, C.R., Hartung, E., 2011. Use of near infrared spectroscopy in online-monitoring of feeding substrate quality in anaerobic digestion. Bioresource Technology 102 (7), 4688e4696. Jakeman, A.J., Letcher, R.A., Norton, J.P., 2006. Ten iterative steps in development and evaluation of environmental models. Environmental Modelling and Software 21 (5), 602e614. Jeong, H.S., Suh, C.W., Lim, J.L., Lee, S.H., Shin, H.S., 2005. Analysis and application of ADM1 for anaerobic methane production. Bioprocess and Biosystems Engineering 27 (2), 81e89. Joshi, M., Seidel-Morgenstern, A., Kremling, A., 2006. Exploiting the bootstrap method for quantifying parameter confidence intervals in dynamical systems. Metabolic Engineering 8 (5), 447e455. Kalfas, H., Skiadas, I.V., Gavala, H.N., Stamatelatou, K., Lyberatos, G., 2006. Application of ADM1 for the simulation of anaerobic digestion of olive pulp under mesophilic and thermophilic conditions. Water Science and Technology 54 (4), 149e156. Kalyuzhnyi, S.V., 1997. Batch anaerobic digestion of glucose and its mathematical modeling .2. Description, verification and application of model. Bioresource Technology 59 (2e3), 249e258. Kalyuzhnyi, S.V., Davlyatshina, M.A., 1997. Batch anaerobic digestion of glucose and its mathematical modeling .1. Kinetic investigations. Bioresource Technology 59 (1), 73e80. Kalyuzhnyi, S.V., Fedorovich, V.V., 1998. Mathematical modelling of competition between sulphate reduction and methanogenesis in anaerobic reactors. Bioresource Technology 65 (3), 227e242. Keshtkar, A., Meyssami, B., Abolhamd, G., Ghaforian, H., Khalagi Asadi, M., 2003. Mathematical modeling of non-ideal mixing continuous flow reactors for anaerobic digestion of cattle manure. Bioresource Technology 87 (1), 113e124. Kleerebezem, R., Van Loosdrecht, M.C.M., 2004. Criticizing some concepts of ADM1. 10th IWA World Congress Anaerobic Digestion, Montreal, Canada. Kleerebezem, R., Van Loosdrecht, M.C.M., 2006. Waste characterization for implementabon in ADM1. Water Science and Technology 54 (4), 167e174. Knobel, A.N., Lewis, A.E., 2002. A mathematical model of a high sulphate wastewater anaerobic treatment system. Water Research 36 (1), 257e265. Koch, K., Lubken, M., Gehring, T., Wichern, M., Horn, H., 2010. Biogas from grass silageemeasurements and modeling with ADM1. Bioresource Technology 101 (21), 8158e8165. Kocsis, L., Gyo¨rgy, A., 2009. Efficient Multi-start Strategies for Local Search Algorithms. Lecture Notes in Computer Science, 5781/2009 doi:10.1007/978-3-642-04180-8_63705e720. Lee, M.Y., Suh, C.W., Ahn, Y.T., Shin, H.S., 2009. Variation of ADM1 by using temperature-phased anaerobic digestion (TPAD) operation. Bioresource Technology 100 (11), 2816e2822. Lewis, R.M., Torczon, V., Trosset, M.W., 2000. Direct-Search Methods: Then and Now. ICASE Report No. 2000e26. Lin, Y.H., Wu, C.L., 2011. Sensitivity analysis of phenol degradation with sulfate reduction under anaerobic conditions. Environmental Modeling & Assessment 16 (2), 213e225. Liwarska-Bizukojc, E., Biernacki, R., 2010. Identification of the most sensitive parameters in the activated sludge model implemented in BioWin software. Bioresource Technology 101 (19), 7278e7285. Lokshina, L., Vavilin, V., Kettunen, H., Rintala, J., Holliger, C., Nozhevnikova, A., 2001. Evaluation of kinetic coefficients using integrated Monod and Haldane models for lowtemperature acetoclastic methanogenesis. Water Research 35 (12), 2913e2922.
Lopez, I., Borzacconi, L., 2009. Modelling a full scale UASB reactor using a COD global balance approach and state observers. Chemical Engineering Journal 146, 1e5. Lopez, I., Borzacconi, L., 2010. Modelling of slaughterhouse solid waste anaerobic digestion: determination of parameters and continuous reactor simulation. Waste Management 30, 1813e1821. Lu¨bken, M., Wichern, M., Schlattmann, M., Gronauer, A., Horn, H., 2007. Modelling the energy balance of an anaerobic digester fed with cattle manure and renewable energy crops. Water Research 41 (18), 4085e4096. Mailier, J., Delmotte, A., Cloutier, M., Jolicoeur, M., Wouwer, A.V., 2011. Parametric sensitivity analysis and reduction of a detailed nutritional model of plant cell cultures. Biotechnology and Bioengineering 108 (5), 1108e1118. Marquardt, D.W., 1963. An algorithm for least squares estimation of nonlinear parameters. Journal of the Society for Industrial and Applied Mathematics 11, 431e441. Martin, C., Ayesa, E., 2010. An integrated Monte Carlo methodology for the calibration of water quality models. Ecological Modelling 221, 2656e2667. Martin, C., de Gracia, M., Ayesa, E., 2011. Bayesian Calibration of the Disintegration process in WWTP Sludge Digesters. 8th IWA Symposium on Systems Analysis and Integrated Assessment (WaterMatex), June 20e22, San Sebastia´n, Spain. Martin, M.A., Raposo, F., Borja, R., Martin, A., 2002. Kinetic study of the anaerobic digestion of vinasse pretreated with ozone, ozone plus ultraviolet light, and ozone plus ultraviolet light in the presence of titanium dioxide. Process Biochemistry 37 (7), 699e706. Molina, F., Castellano, M., Garcia, C., Roca, E., Lema, J., 2009. Selection of variables for on-line monitoring, diagnosis, and control of anaerobic digestion processes. Water Science and Technology 60 (3), 615e622. Morris, M.D., 1991. Factorial sampling plans for preliminary computational experiments. Technometrics 33 (2), 161e174. Mosche, M., Jordening, H.J., 1999. Comparison of different models of substrate and product inhibition in anaerobic digestion. Water Research 33 (11), 2545e2554. Muller, T.G., Noykova, N., Gyllenberg, M., Timmer, J., 2002. Parameter identification in dynamical models of anaerobic waste water treatment. Mathematical Biosciences 177, 147e160. Nocedal, J., Wright, S., 2006. Numerical Optimization. Springer Verlag. Nopharatana, A., Pullammanappallil, P.C., Clarke, W.P., 2003. A dynamic mathematical model for sequential leach bed anaerobic digestion of organic fraction of municipal solid waste. Biochemical Engineering Journal 13 (1), 21e33. Noykova, N., Gyllenberg, M., 2000. Sensitivity analysis and parameter estimation in a model of anaerobic wastewater treatment processes with substrate inhibition. Bioprocess Engineering 23, 343e349. Noykova, N., Muller, T.G., Gyllenberg, M., Timmer, J., 2002. Quantitative analyses of anaerobic wastewater treatment processes: identifiability and parameter estimation. Biotechnology and Bioengineering 78 (1), 89e103. Omlin, M., Reichert, P., 1999. A comparison of techniques for the estimation of model prediction uncertainty. Ecological Modelling 115 (1), 45e59. Ossenbruggen, P.J., Stevens, D.K., 1996. Mechanistic model identification for nitrification in activated sludge. Environmetrics 7 (6), 579e600. Ozkan-Yucel, U.G., Gokcay, C.F., 2010. Application of ADM1 model to a full-scale anaerobic digester under dynamic organic loading conditions. Environmental Technology 31 (6), 633e640. Palatsi, J., Illa, J., Prenafeta-Boldu, F.X., Laureni, M., Fernandez, B., Angelidaki, I., Flotats, X., 2010. Long-chain fatty acids
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 5 3 4 7 e5 3 6 4
inhibition and adaptation process in anaerobic thermophilic digestion: batch tests, microbial community structure and mathematical modelling. Bioresource Technology 101 (7), 2243e2251. Pobeheim, H., Munk, B., Mueller, H., Berg, G., Guebitz, G., 2010. Characterization of an anaerobic population digesting a model substrate for maize in the presence of trace metals. Chemosphere 80 (8), 829e836. Qian, S.S., Stow, C.A., Borsuk, M.E., 2003. On Monte Carlo methods for bayesian inference. Ecological Modelling 159 (2e3), 269. Ramirez, I., Mottet, A., Carrere, H., Deleris, S., Vedrenne, F., Steyer, J.P., 2009. Modified ADM1 disintegration/hydrolysis structures for modeling batch thermophilic anaerobic digestion of thermally pretreated waste activated sludge. Water Research 43, 3479e3492. Raposo, F., Borja, R., Martı´n, M.A., Martı´n, A., de la Rubia, M.A., Rinco´n, B., 2009. Influence of inoculumesubstrate ratio on the anaerobic digestion of sunflower oil cake in batch mode: process stability and kinetic evaluation. Chemical Engineering Journal 149, 70e77. Reichert, P., 2010. Environmental Systems Analysis Lecture Notes Summer Course, June 21e25, Eawag Du¨bendorf, Switzerland. Redzwan, G., Banks, Ch, 2004. The use of a specific function to estimate maximum methane production in a batch-fed anaerobic reactor. Journal of Chemical Technology and Biotechnology 79, 1174e1178. Rincon, A., Angulo, F., Olivar, G., 2009. Control of an anaerobic digester through normal form of fold bifurcation. Journal of Process Control 19 (8), 1355e1367. Rittmann, B.E., 2008. Opportunities for renewable bioenergy using microorganisms. Biotechnology and Bioengineering 100 (2), 203e212. Rodrigues, J.A.D., Ratusznei, S.M., Zaiat, M., 2003. Fed-batch and batch operating mode analysis of a stirred anaerobic sequencing reactor with self-immobilized biomass treating low-strength wastewater. Journal of Environmental Management 69 (2), 193e200. Rodriguez, J., Lema, J.M., Van Loosdrecht, M.C.M., Kleerebezem, R. , 2006. Variable stoichiometry with thermodynamic control in ADM1. Water Science and Technology 54 (4), 101e110. Rodriguez, J., Roca, E., Lema, J.M., Bernard, O., 2008. Determination of the adequate minimum model complexity required in anaerobic bioprocesses using experimental data. Journal of Chemical Technology and Biotechnology 83 (12), 1694e1702. Ruel, S.M., Comeau, Y., Ginestet, P., Heduit, A., 2002. Modeling acidogenic and sulfate-reducing processes for the determination of fermentable fractions in wastewater. Biotechnology and Bioengineering 80 (5), 525e536. Ruzicka, M., 1996. The effect of hydrogen on acidogenic glucose cleavage. Water Research 30 (10), 2447e2451. Sales-Cruz, M., Gani, R., 2004. Aspects of modelling and model identification for bioprocesses through a computer-aided modelling system. In: Barbosa-Po´voa, Ana Paula, Matos, Henrique (Eds.), Computer Aided Chemical Engineering European Symposium on Computer-Aided Process Engineering14, 37th European Symposium of the Working Party on Computer-Aided Process Engineering. Elsevier, pp. 1123e1128. Saltelli, A., Ratto, M., Tarantola, S., Campolongo, F., 2006. Sensitivity analysis practices: strategies for model-based inference. Reliability Engineering and System Safety 91, 1109e1125. Saltelli, A., Ratto, M., Andres, T., Campolongo, F., Cariboni, J., Gatelli, D., Saisana, M., Tarantola, S., 2008. Global Sensitivity Analysis. The Primer. John Wiley & Sons, West Sussex, England. Sanchez, J.M., Arijo, S., Mun˜oz, M.A., Morinigo, M.A., Borrego, J.J., 1994. Microbial colonization of different support materials
5363
used to enhance the methanogenic process. Applied Microbiology and Biotechnology 41 (4), 480e486. Saravanan, V., Sreekrishnan, T.R., 2006. Modelling anaerobic biofilm reactors e a review. Journal of Environmental Management 81 (1), 1e18. Sbarciog, M., Loccufier, M., Noldus, E., 2010. Determination of appropriate operating strategies for anaerobic digestion systems. Biochemical Engineering Journal 51 (3), 180e188. Schenkendorf, R., Kremling, A., Mangold, M., 2009. Optimal experimental design with the sigma point method. IET Systems Biology 3 (1), 10e23. Siegrist, H., Renggli, D., Gujer, W., 1993. Mathematical-Modeling of anaerobic mesophilic sewage-sludge treatment. Water Science and Technology 27 (2), 25e36. Siegrist, H., Vogt, D., Garcia-Heras, J., Gujer, W., 2002. Mathematical model for meso- and thermophilic anaerobic sewage sludge digestion. Environmental Science and Technology 36 (5), 1113e1123. Simeonov, I., Momchev, V., Grancharov, D., 1996. Dynamic modeling of mesophilic anaerobic digestion of animal waste. Water Research 30 (5), 1087e1094. Simeonov, I.S., 1999. Mathematical modeling and parameters estimation of anaerobic fermentation processes. Bioprocess Engineering 21 (4), 377e381. Sin, G., Gernaey, K.V., Neumann, M.B., Van Loosdrecht, M.C.M., Gujer, W., 2011. Global sensitivity analysis in wastewater treatment plant model applications: prioritizing sources of uncertainty. Water Research 45 (2), 639e651. Sobol, I.M., 2001. Global sensitivity indices for nonlinear mathematical models and their Monte Carlo estimates. Mathematics and Computers in Simulation 55 (1e3), 271e280. Smith, L.H., McCarty, P.L., Kitanidis, P.K., 1998. Spreadsheet method for evaluation of biochemical reaction rate coefficients and their uncertainties by weighted nonlinear least-squares analysis of the integrated monod equation. Applied and Environmental Microbiology 64 (6), 2044e2050. Speece, R.E., 1996. Anaerobic Biotechnology: For Industrial Wastewater. Archae Press. Spriet, J.A., 1985. Structure characterization e an overview. In: Barker, H.A., Young, P.C. (Eds.), Identification and System Parameter Estimation 1985 e Proceedings of the seventh IFAC/ IFORS Symposium. Pergamon Press, Oxford, pp. 749e756. Tarantola, S., Gatelli, D., Mara, T.A., 2006. Random balance designs for the estimation of first order global sensitivity indices. Reliability Engineering and System Safety 91 (6), 717e727. Tartakovsky, B., Morel, E., Steyer, J.P., Guiot, S.R., 2002. Application of a variable structure model in observation and control of an anaerobic digestor. Biotechnology Progress 18 (4), 898e903. Tartakovsky, B., Mu, S.J., Zeng, Y., Lou, S.T., Guiot, S.R., Wu, P., 2008. Anaerobic digestion model no. 1-based distributed parameter model of an anaerobic reactor: II. Model validation. Bioresource Technology 99 (9), 3676e3684. Tomei, M.C., Braguglia, C.M., Cento, G., Mininni, G., 2009. Modeling of anaerobic digestion of sludge. Critical Reviews in Environmental Science and Technology 39 (12), 1003e1051. Vande Wouwer, A., Renotte, C., Queinnec, I., Bogaerts, P., 2006. Transient analysis of a wastewater treatment biofilterdistributed parameter modeling and state estimation. Mathematical and Computer Modelling of Dynamical Systems 12, 423e440. Van Lier, J.B., 2008. High-rate anaerobic wastewater treatment: diversifying from end-of-the-pipe treatment to resourceoriented conversion techniques. Water Science and Technology 57 (8), 1137e1148.
5364
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 5 3 4 7 e5 3 6 4
Vavilin, V.A., Vasiliev, V.B., Ponomarev, A.V., Rytow, S.V., 1994. Simulation-model methane as a tool for effective biogas production during anaerobic conversion of complex organicmatter. Bioresource Technology 48 (1), 1e8. Vavilin, V.A., Rytow, S.V., Lokshina, L.Y., 1995. Modelling hydrogen partial pressure change as a result of competition between the butyric and propionic groups of acidogenic bacteria. Bioresource Technology 54 (2), 171e177. Vavilin, V., Rytov, S.V., Pavlostathis, S.G., Jokela, J., Rintala, J., 2003. A distributed model of solid waste anaerobic digestion: sensitivity analysis. Water Science and Technology 48 (4), 147e154. Walter, E., Pronzato, L., 1997. Identication of Parametric Models from Experimental Data. Springer-Verla, Berlin. Ward, A.J., Hobbs, P.J., Holliman, P.J., Jones, D.L., 2008. Optimisation of the anaerobic digestion of agricultural resources. Bioresource Technology 99 (17), 7928e7940. Ward, A.J., Hobbs, P.J., Holliman, P.J., Jones, D.L., 2011. Evaluation of near infrared spectroscopy and software sensor methods for determination of total alkalinity in anaerobic digesters. Bioresource Technology 102 (5), 4083e4090.
Weijers, S.R., 2003. Model reduction of activated sludge model no. 1 and bioprocess models for identification and control. Biotechnology for the environment: wastewater treatment and modeling. Waste Gas Handling 3C, 11e63. Wichern, M., Gehring, T., Fischer, K., Andrade, D., Lubken, M., Koch, K., Gronauer, A., Horn, H., 2009. Monofermentation of grass silage under mesophilic conditions: measurements and mathematical modeling with ADM 1. Bioresource Technology 100 (4), 1675e1681. Wolf, C., McLoone, S., Bongards, M., 2008. Plant optimization using genetic algorithms and particle swarm optimization. In: Signals and Systems Conference, 208. (ISSC 2008). IET Irish. Zaher, U., Vanrolleghem, P., Rodriguez, J., Franco, A., 2004. Conceptual approach for ADM1 applications. In: Ujang, Z., Henze, M. (Eds.), Environmental Biotechnology: Advancement in Water and Wastewater Applications in the Tropics. IWA Publishing. Zaher, U., Pandey, P., Chen, S., 2009. A simple elemental continuity based model application to study the anaerobic microbial activity for the treatment of dairy manure. Applied Mathematical Modelling 33 (9), 3553e3564.
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 5 3 6 5 e5 3 7 4
Available at www.sciencedirect.com
journal homepage: www.elsevier.com/locate/watres
Evaluation of a laboratory-scale bioreactive in situ sediment cap for the treatment of organic contaminants David W. Himmelheber a,*, Kurt D. Pennell b, Joseph B. Hughes a,c a
School of Civil and Environmental Engineering, Georgia Institute of Technology, Atlanta, GA 30332, USA Department of Civil and Environmental Engineering, Tufts University, Medford, MA, USA c School of Material Science and Engineering, Georgia Institute of Technology, Atlanta, GA 30332, USA b
article info
abstract
Article history:
The development of bioreactive sediment caps, in which microorganisms capable of
Received 13 December 2010
contaminant transformation are placed within an in situ cap, provides a potential remedial
Received in revised form
design that can sustainably treat sediment and groundwater contaminants. The goal of
7 April 2011
this study was to evaluate the ability and limitations of a mixed, anaerobic dechlorinating
Accepted 17 June 2011
consortium to treat chlorinated ethenes within a sand-based cap. Results of batch exper-
Available online 30 June 2011
iments demonstrate that a tetrachloroethene (PCE)-to-ethene mixed consortium was able to completely dechlorinate dissolved-phase PCE to ethene when supplied only with sedi-
Keywords:
ment porewater obtained from a sediment column. To simulate a bioreactive cap,
Sediment remediation
laboratory-scale sand columns inoculated with the mixed culture were placed in series
In situ capping
with an upflow sediment column and directly supplied sediment effluent and dissolved-
Microbial processes
phase chlorinated ethenes. The mixed consortium was not able to sustain dechlorina-
Bioremediation
tion activity at a retention time of 0.5 days without delivery of amendments to the sediment effluent, evidenced by the loss of cis-1,2-dichloroethene (cis-DCE) dechlorination to vinyl chloride. When soluble electron donor was supplied to the sediment effluent, complete dechlorination of cis-DCE to ethene was observed at retention times of 0.5 days, suggesting that sediment effluent lacked sufficient electron donor to maintain active dechlorination within the sediment cap. Introduction of elevated contaminant concentrations also limited biotransformation performance of the dechlorinating consortium within the cap. These findings indicate that in situ bioreactive capping can be a feasible remedial approach, provided that residence times are adequate and that appropriate levels of electron donor and contaminant exist within the cap. ª 2011 Elsevier Ltd. All rights reserved.
1.
Introduction
The management and remediation of contaminated aquatic sediments pose major technical and economic challenges. Treatment of contaminated sediment sites with in situ caps has become an established practice that can provide advantages over alternative methods in certain settings
(Reible et al., 2003). Clean sand has traditionally been employed as capping material, and remains a large component of many field-scale capping applications. Sand-based caps have the potential to delay contaminant breakthrough when diffusive transport dominates (Go et al., 2009; Thoma et al., 1993), but eventual contaminant breakthrough remains a source of concern. Additionally, traditional sand
* Corresponding author. Geosyntec Consultants, 10220 Old Columbia Road, Suite A, Columbia, MD 21046, USA. Tel.: þ1 410 381 4333; fax: þ1 410 381 4499. E-mail address:
[email protected] (D.W. Himmelheber). 0043-1354/$ e see front matter ª 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.watres.2011.06.022
5366
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 5 3 6 5 e5 3 7 4
caps are less effective at sites where groundwater seepage or mobile contaminants (i.e., low Koc) are present (Go et al., 2009). Research studies have focused on in situ sequestration (Cho et al., 2007; Zimmerman et al., 2004), in situ transformation (Krumins et al., 2009; Lowry and Johnson, 2004), and the development of active caps which incorporate reactive and/or sorptive constituents designed to reduce contaminant and bioavailability (Choi et al., 2009; Hyun et al., 2006; Jacobs and Fo¨rstner, 1999; McDonough et al., 2007; Murphy et al., 2006; Reible et al., 2007). Ideally, active caps eliminate the risk of contaminant breakthrough into the overlying water column, and can potentially be implemented at sediment sites with groundwater seeps and relatively mobile contaminants. The employment of physicochemical-based active caps appears promising, but possible limitations (e.g., high material costs, sorption and reaction capacities) have stimulated the consideration of in situ bioreactive caps, in which contaminant biotransformations are designed to occur within the cap matrix to produce environmentally-acceptable reaction products. Enhanced in situ bioremediation, through biostimulation and bioaugmentation, has proven to be a successful groundwater remediation technology for a diverse range of contaminants (Lo¨ffler and Edwards, 2006). Adaptation of these principles to subaqueous sediment remediation has not been demonstrated, prompting the recent identification of in situ bioremediation as a priority research and development need (SERDP/ESTCP, 2008). Biologically-based active caps have the potential to maintain reactivity over long periods of time and could serve as a sustainable remedial option if microorganisms capable of biotransformation are present and necessary metabolic requirements are met. Previous studies that investigated the activity of microbial populations within a sediment cap demonstrated that microorganisms indigenous to underlying sediment, including organisms capable of contaminant biotransformation, are able to colonize the overlying cap and possibly participate in contaminant bioattenuation processes (Himmelheber et al., 2009). Bioaugmentation of microorganisms within a cap, as opposed to intrinsic colonization (defined here as the natural redistribution of microorganisms native to the sediment into the cap matrix), could provide enhanced degradation capacity and minimize the potential for contaminant release to benthic and aqueous receptors. Such a bioaugmentation strategy was recently evaluated by the US Geological Survey (USGS) as a means to reductively dechlorinate a mixture of chlorinated ethenes, ethanes, and methanes present in a groundwater seep discharging into a tidal wetland (Majcher et al., 2007). A mixed, anaerobic culture was enriched from the site (Lorah et al., 2008) and incorporated into an organic-based matrix that was placed at the sediment-water interface. This bioreactive mat successfully treated the chlorinated contaminants prior to discharge (Majcher et al., 2009). Although the bioreactive mat was constructed on the banks of a tidal wetland (i.e., not completely subaqueous) and the design is not immediately suitable for submergence (e.g., buoyancy restrictions, delivery of bioaugmentation culture), the success of the approach supports the concept of bioreactive capping as an in situ remedial technique.
The USGS bioreactive mat was designed in part because the chlorinated organics present in the groundwater were undergoing only partial dechlorination in the sediment prior to discharge, a phenomenon commonly reported at sediment sites (Abe et al., 2009; Conant et al., 2004; Hamonts et al., 2009; Himmelheber et al., 2007; Lendvay et al., 1998; Lorah and Voytek, 2004; Majcher et al., 2007). Additionally, recent studies have demonstrated that anaerobic conditions develop within sediment caps subject to diffusive and upflow conditions (i.e., groundwater seeps) (Himmelheber et al., 2008, 2009). It is therefore expected that contaminated groundwater seeps will carry partially-degraded contaminants into the overlying anaerobic cap, thereby providing an opportunity for treatment by reductive biotransformations. Detailed assessment of bioreactive in situ sediment caps has not been previously undertaken and little is currently known about the feasibility of bioreactive caps, particularly their limitations and maintenance requirements. The objective of this work was to establish an actively dechlorinating microbial consortium within a simulated overlying cap and to determine how contaminant mass flux and electron donor amendments influenced bioreactive cap performance. More specifically, the bioreactive cap experiments were designed to determine whether or not amendments are necessary to sustain complete reductive dechlorination by an active microbial community. Chlorinated ethenes were utilized as the contaminants due to their frequent occurrence as groundwater contaminants, their presence in groundwater seeps, and their greater mobility relative to other sediment contaminants (e.g., chlorinated benzenes, polychlorinated biphenyls). Batch reactor and bioaugmented column studies were conducted to assess bioreactive cap performance over a range of electron donor and contaminant conditions.
2.
Materials and methods
2.1.
Chemicals
PCE (99þ%, SigmaeAldrich, St. Louis, MO), TCE (99.5%, SigmaeAldrich), cis-DCE (97%, Acros Organics, Morris Plains, NJ), trans-DCE (99.7%, Acros Organics), and 1,1-DCE (99.9%, Acros Organics) were obtained in neat liquid form. Vinyl chloride (8%/N2 balance), ethene (99.5%), ethane (99.5%), and methane (99%) were obtained from Matheson Tri-Gas (Parsippany, NJ). Sodium bicarbonate, potassium chloride, magnesium chloride, and calcium chloride were used in the preparation of simulated groundwater and were purchased from Fisher Scientific (Pittsburgh, PA). Sodium lactate syrup (60% vol/vol, Fisher Scientific) was used during the preparation of stock lactate solutions. Potassium bromide, calcium sulfate, and potassium phosphate were purchased from Fisher Scientific and used for IC standards and non-reactive tracer studies.
2.2.
Batch reactors
Batch reactors were established in triplicate and consisted of Anacostia River (Washington, D.C., USA) sediment porewater, dissolved-phase PCE, and a mixed PCE-to-ethene dechlorinating consortium. A PCE-to-ethene dechlorinating mixed
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 5 3 6 5 e5 3 7 4
consortia referred to as OW served as the inoculum. The OW consortia, which is capable of complete reductive dechlorination of PCE to ethene, has been described previously (Daprato et al., 2006). The OW culture has been found to contain multiple dechlorinating microorganisms, including Dehalococcoides species, and known reductive dehalogenases including tceA, vcrA, and bvcA (Daprato et al., 2006). Three 25 mL aliquots of OW culture were transferred to 70 mL serum bottles pre-capped with Teflon-faced butyl septa and sparged with N2 gas for 15 min to remove oxygen from the empty bottles. The collected OW aliquots were then sparged with N2 gas for 15 min in attempt to remove residual chlorinated ethenes, methanol, and volatile fatty acids from the batch reactors. Sediment effluent was collected from a sediment column that was supplied only with simulated groundwater and dissolved-phase PCE (Himmelheber et al., 2007). The composition of simulated groundwater was slightly modified from that described by Dries et al. (2004) and consisted of 3.5 mM NaHCO3, 0.1 mM KCl, 0.25 mM MgCl2, 0.75 mM CaCl2, and resazurin as a redox indicator. Sediment effluent was collected under anoxic conditions and 25 mL of effluent were added directly to batch reactors containing the OW consortium. Dissolved-phase PCE was obtained from a saturated stock solution containing neat PCE in contact with anaerobic, sterilized simulated groundwater. Stock PCE concentrations were quantified immediately prior to injection into the batch reactors. Approximately 16 mmol of dissolvedphase PCE was added to each microcosm using a 10 mL Hamilton glass syringe. All reactors were wrapped in foil and incubated at 20 C on an orbital shaker operated at 150 rpm. Chlorinated ethenes, ethene, ethane, and methane concentrations were determined from headspace samples of the microcosms.
2.3.
Bioreactive cap operation
Two one-dimensional (1-D) columns (designated herein as Bioreactive Cap A and Bioreactive Cap B) were constructed using 2.5 cm inside diameter (I.D.) glass chromatography columns 30 cm in length (Spectrum Chromatography, Houston, TX) and equipped with custom-built stainless steel end plates (Dutton & Hall, Atlanta, GA). A 2.5 cm diameter disc of 80 mesh stainless steel (Small Parts, Inc., Miami Lakes, FL) was placed on the column end plates to retain sand grains within the column. A fabricated glass reservoir (15 mL) fitted with a stopcock was placed at the column effluent to allow for aqueous effluent sampling. The columns were packed with ASTM C-33 grade concrete sand (U.S. Silica, Mauricetown, NJ). This particular sand was selected because it is representative of the solids used for submerged sediment caps and was utilized in the Anacostia River Capping Demonstration Project (Reible, 2005). An elemental analysis of the sand was performed at the University of Georgia Laboratory for Environmental Analysis (see Supplementary Information, Table S.1). The dry, autoclaved sand was packed into the bioreactive columns under aerobic conditions in 5-cm increments with vibration along the outside wall of the column. Three pore volumes of N2-sparged, autoclaved simulated groundwater were flushed through the columns to check for leakage and to ensure anaerobic conditions. The columns were then
5367
inoculated by flushing the columns with three pore volumes of the OW culture suspension. Following inoculation, the two sand cap columns were wrapped in foil to avoid exposure to light then connected in series with an upflow column packed with Anacostia River sediment as depicted in Fig. 1. The sediment column effluent, which was provided only with simulated groundwater and dissolved-phase PCE, served as the influent for the bioreactive sand columns. Therefore, the influent for the sand columns consisted of sediment effluent and a mixture of partial PCEdechlorination products, similar to the conditions that would be anticipated in a submerged sediment capping scenario subject to a PCE-contaminated groundwater seep. Table 1 provides a summary of experimental conditions employed for each bioreactive sand column. Chlorinated ethene and ethene concentrations in the effluent of Bioreactive Caps A and B were normalized on a molar basis to total chlorinated ethenes and ethene eluted per sample to reduce scatter in concentration data and to monitor product distribution.
2.3.1.
Bioreactive Cap A
Bioreactive Cap A was designed to assess the ability of sediment effluent to maintain an external dechlorinating community in a cap, simulating a bioreactive cap inoculated with a mixed dechlorinating consortia and operating under reducing conditions. Prior to inoculating Bioreactive Cap A, a tracer test was conducted with a pulse injection of 100 mg L1 (1.25 mM) bromide obtained from an autoclaved, sparged stock solution of potassium bromide in simulated groundwater. A total of 1.2 pore volumes were flushed through the column, collected with a fraction collector, and analyzed via ion chromatography. Three pore volumes of simulated groundwater were then flushed through the column following the tracer test to remove residual bromide prior to inoculation. A 200 mL aliquot of aqueous OW culture was obtained for inoculation and stored in a 160 mL serum bottle that had previously been capped with a Teflonfaced butyl septum and sparged with N2 for 15 min to remove oxygen. The 200 mL aliquot was tested for its dechlorination ability in batch conditions by spiking with PCE and methanol. After successfully dechlorinating PCE to ethene (Supplementary Information, Fig. S.1A), 1.5 pore volumes of the OW culture were supplied to the column at a flow rate of 2.2 mL h1 (1-day residence time). Following a 24-h attachment period during which there was no flow, Bioreactive Cap A was connected in series with the sediment column from 67 to 83 sediment pore volumes. The unamended sediment column effluent served as the influent for the duration of the Bioreactive Cap A experiment.
2.3.2.
Bioreactive Cap B
The Bioreactive Cap B experiment was designed to simulate a dechlorinating bioreactive cap operating under reducing conditions, but differed from Cap A in that the influent for this experiment was supplied at various flow rates and periodically spiked with amendments. Thus, Bioreactive Cap B demonstrates the impact of contaminant influx and the presence of reducing equivalents on the capacity of sediment column effluent to maintain an external dechlorinating
5368
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 5 3 6 5 e5 3 7 4
Gas purge following sample collection
Anacostia Sediment
30 cm
Sand incoluated with PCF, to ethene mixed consortia
(C) (A) Anoxic Simulated Groundwater + Dissolved PCE
(B)
Upflow
30 cm
Effluent Samples
Upflow
Effluent Samples
Amendments and Flow Rate Control to Bioreactive Cap B
Captured Sediment Effluent for Bioreactive Cap B
Fig. 1 e Conceptual schematic of laboratory simulation of a bioreactive sand cap placed in series with an anaerobic sediment bed subject to a PCE-contaminated groundwater seep. (A) Sediment effluent was directly supplied to Bioreactive Cap A. (B) Sediment effluent was initially captured via a syringe for Bioreactive Cap B, spiked with amendments, then (C) supplied to Bioreactive Cap B at select flow rates.
community. An aliquot of OW culture was retrieved and sparged with nitrogen prior to inoculation as described for Bioreactive Cap A. The aliquot of OW culture again demonstrated the ability to completely dechlorinate PCE to ethene in batch culture (Supplementary Information, Fig. S.1B). A total of 1.7 pore volumes of OW culture was then supplied to the column at a flow rate of 2.6 mL h1 (1-day residence time), followed by a no-flow attachment period of one day. Unlike Bioreactive Cap A, Bioreactive Cap B was not immediately connected to the sediment column effluent, but rather positive-control experiments were conducted to ensure the inoculated column could completely dechlorinate cis-DCE to ethene when provided DCB-1 media, Wolin vitamins, and 5 mM lactate as an electron donor and carbon source. Following this demonstration of complete dechlorination in the cap under optimal conditions (Supplementary Information, Fig. S.2), one pore volume of anaerobic simulated groundwater was flushed through the column to remove these constituents from the system prior to the introduction of sediment effluent. Sediment column effluent was obtained
from 146 to 180 sediment pore volumes to serve as Bioreactive Cap B influent. For Bioreactive Cap B, sediment column effluent was captured under anoxic conditions by connecting an empty, gas-tight syringe to sediment effluent tubing and allowing the aqueous flow to gradually fill the syringe at the same rate of sediment column influent (5.5 mL h1). Once the effluent syringe had been filled, it was immediately transferred to a separate syringe pump and introduced into the sand column as the influent. This method allowed for manipulation of flow rates within the sand column and for addition of electron donor and acceptor to the influent prior to connection with the sand column. The electron donor used for this study was lactate, which was obtained from a 100 mM stock solution in autoclaved, sparged simulated groundwater. Lactate was supplied to the bioreactive sand column (Cap B) at a concentration of 5 mM from 0 to 13.3 pore volumes (Table 1). The experimental conditions employed for Bioreactive Cap B were designed to gradually decrease aqueous residence times, as well as electron donor concentrations, to determine
5369
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 5 3 6 5 e5 3 7 4
Pore volume (PV)a (mL) Porosity (n)a (cm3 void (cm3 total)1) Connected in series to sediment column (sediment pore volumes) Experimental flow rate (Q) (mL h1) Porewater velocity (v) (cm day1) (Darcy velocity (cm day1)) Peclet number (Pe)b (dimensionless) Alterations to influent
Influent chloroethene concentration (mM total chlorinated ethenes)
Bioreactive Cap A
Bioreactive Cap B
62.72 0.41
61.82 0.41
67.0e83.2
146.0e180.0
5.46
1.29; 2.58; 5.46
62.67 (25.88)
14.99; 29.98; 63.59 (6.10); (12.20); (25.88) N/Ac
80.5 None
16.19 11.06d
Addition of Lactate Addition of cisDCE Decrease of flow rate 0e3.44 PV: 200 42d 3.44 PV to end: 34 3.6d
a Estimated from mass difference between dry and wet packed columns. b Obtained with the CFITM3 breakthrough curve fitting program under equilibrium constraints. c Tracer test not performed. d Average one standard deviation.
limitations on dechlorination (Table 1). The influent flow rate for Bioreactive Cap B was increased step-wise from 1.3 mL h1 (2-day retention time), to 2.6 mL h1 (1-day retention time) to 5.5 mL h1 (0.47-day retention time). From 0 to 3.4 sand pore volumes, additional cis-DCE was provided to the influent to ensure chlorinated ethenes were present due to complete dechlorination of PCE to ethene in the sediment column effluent prior to connecting Bioreactive Cap B. cis-DCE was chosen assuming partial, intrinsic PCE dechlorination would occur in sediment beds, based on prior research findings (Himmelheber et al., 2007). The cis-DCE was obtained from a saturated stock solution of cis-DCE (i.e., NAPL present) in autoclaved, sparged simulated groundwater and supplied to the influent at a concentration of 200 42 mM. After 3.4 pore volumes, however, the only source of chlorinated ethenes to Bioreactive Cap B was the sediment effluent. Lactate (5 mM) was provided from 0 to 13.3 pore volumes, at which point it was removed from the influent and no electron donor was provided for the remainder of the experiment.
2.4.
Analytical methods
PCE, TCE, DCE isomers, VC, ethene, ethane, and methane concentrations were determined from the headspace of 5 mL aqueous effluent samples, which were analyzed using an Agilent 6890 gas chromatograph (GC) equipped with a flame ionization detector (FID), as described previously (Carr and Hughes, 1998). Bromide was measured using a Dionex DX-
3.
Results and discussion
3.1.
Batch reactors
The OW culture successfully dechlorinated PCE to ethene when provided only sediment effluent and dissolved-phase PCE (Fig. 2). Complete PCE dechlorination to ethene was achieved after 19 days of incubation. Chlorinated ethene mass balance was within 10% for each time point except day 12, when chloroethene mole totals were 124% of the initial dissolved-phase PCE introduced to the batch reactors. This discrepancy arose because one of the triplicate reactors recorded unusually high concentrations of VC, despite balanced ethene concentrations at the end of the experiment. This is reflected in the relatively high standard deviation of VC at day 12. Duplicate analysis at the same time point yielded similar results. Regardless of this isolated analytical discrepancy, the presence of VC and ethene indicates that dechlorinating species within the OW consortium, specifically Dehalococcoides, remained active for at least one dechlorination cycle when provided only sediment effluent and a dissolved-phase electron acceptor (PCE). Thus, the sources of carbon, electron donor, and micronutrients were provided by the sediment effluent or from microbial biomass (Adamson and Newell, 2009). Methane concentrations rose steadily during the dechlorination of PCE (Fig. 2), indicating that methanogenic populations were also able to remain active when provided only sediment effluent. These data suggest that dechlorinating species within a bioreactive cap inoculated with a methanogenic mixed consortia may have to 20
120 Methane PCE TCE DCE VC ETH
16
12
100
80
60
8 40
4
Total Methane (mM)
Parameter
100 ion chromatograph with a Dionex AG4A IonPac guard column and Dionex AS4A IonPac column at a flow rate of 1.5 mL/min and an ED40 electrochemical detector.
Total Chloroethene (µmoles)
Table 1 e Summary of experimental conditions for sand column experiments.
20
0
0
0
3
6
9 12 Time (days)
15
18
Fig. 2 e Batch microcosm results of OW culture provided only PCE and sediment effluent. Chlorinated ethenes are reported as the sum of aqueous and gas phases within the microcosms. Error bars represent one standard deviation calculated from triplicate reactors. The shaded background area corresponds to methane production (mM) at each time point and is referenced to the right vertical axis. Total methane is calculated as the sum of aqueous and gas phase methane.
5370
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 5 3 6 5 e5 3 7 4
compete with methanogens for electron donors, which could result in reduced dechlorination efficiency over time. Methanogens and other microbial populations indigenous to the sediment are also expected to populate the cap material (Himmelheber et al., 2009) and may, therefore, compete for electron donor and other nutrients.
3.2.
Bioreactive Cap A
The pore volume of Bioreactive Cap A was estimated to be 62.7 mL from differences between wet and dry column mass and assuming complete water saturation (Table 1). The nonreactive tracer test conducted at the onset of Bioreactive Cap A operation yielded a symmetrical breakthrough curve, indicative of the absence of immobile regions of water (see Supplementary Information, Fig. 2). The measured tracer BTC, expressed as the relative concentration versus number of dimensionless pore volumes applied, was fit to an analytical solution of the one-dimensional advectivedispersive reactive (ADR) transport equation using the CXTFIT model (van Genuchten, 1981). As anticipated, the fitted retardation factor (RF) obtained from the tracer BTC was approximately equal to 1.0, indicating no detectable interactions between the solid phase and tracer during transport through the sand column. The fitted Peclet number (Pe) was approximately 81, yielding a hydrodynamic dispersion coefficient (DH) of 2.8 108 m2 s1 and a hydrodynamic dispersivity (aD) of 0.37 cm. These data are consistent with values reported for similar water-saturated columns packed with graded sands, and indicate that advective flow and transport through the column was normal and not subject to physical nonequilibrium. Chlorinated ethene effluent product distributions, normalized to moles of chlorinated ethenes and ethene eluted, are shown for Bioreactive Cap A (Fig. 3B). The applied influent flow rate of 5.5 mL h1 corresponded to a column residence time of 0.47 days. When Bioreactive Cap A was connected in series to the sediment column, cis-DCE was the predominant chlorinated ethene present in influent solution. The bioreactive sand column was initially able to dechlorinate cis-DCE to VC, but ethene was not detected (Fig. 3B). This dechlorination activity disappeared prior to 5 pore volumes, and eventually only 5% of the cis-DCE was dechlorinated to VC, indicating that Dehalococcoides activity was impaired. Methane data collected during the Bioreactive Cap A experiment reveal that microbes other than dechlorinators also lost activity, suggesting microbial impairment in the system as a whole and not just for the dechlorinating population (Fig. 3C). Based on data presented in Fig. 3B, the sediment column effluent was not able to sustain the dechlorinating consortium OW without additional amendments. Data were not collected to determine if non-contaminant stressors (e.g., ammonia) were present in the sediment, which could suppress microbial activity. However, previous research (Himmelheber et al., 2007) has demonstrated that microorganisms, specifically Dehalococcoides strains, can be stimulated in the Anacostia sediment with the addition of electron donor, suggesting that non-contaminant stressors were not a major concern in the system. Previous research (Himmelheber et al., 2007) has also demonstrated that
microbial activity in the sediment column was limited by electron donor availability. It was therefore hypothesized that the levels of electron donor eluting from the sediment column effluent prevented the dechlorinating community in the sand cap from maintaining sufficient activity to achieve complete reductive dechlorination of the cis-DCE introduced to the sand cap column. A second possibility is that the relatively high flow rates through the sand cap column did not provide sufficient contact time between the contaminants and the dechlorinating community to achieve complete reductive dechlorination.
3.3.
Bioreactive Cap B
To address the hypotheses raised above, the second sand column, Bioreactive Cap B, was operated at three different flow rates with and without the addition of lactate as an electron donor and cis-DCE as an electron acceptor (Table 1). The experimental conditions associated with Bioreactive Cap B are presented in Fig. 4A, while normalized chloroethene product distributions are shown in Fig. 4B. The pore volume for Bioreactive Cap B was estimated to be 61.8 mL from differences between wet and dry column mass and assuming complete water saturation (Table 1). Prior to supplying Bioreactive Cap B with sediment effluent, the inoculated column was able to completely dechlorinate cis-DCE to ethene when provided electron donor, carbon sources, vitamins, and reduced media; confirming the ability of the OW culture to achieve complete dechlorination within the column (Supplementary Information, Fig. S.2). After applying one pore volume of simulated groundwater, sediment effluent was supplied to Bioreactive Cap B, indicated as pore volume 0 in Fig. 4A and B. The influent for Bioreactive Cap B was the sediment column effluent from 146 to 180 sediment pore volumes, which contained a mixture of cis-DCE, VC, and ethene. The influent solution provided to Bioreactive Cap B was initially augmented with cis-DCE to yield a total influent chloroethene concentration of approximately 200 mM and 5 mM lactate, operated at a residence time of 2 days (flow rate ¼ 1.29 mL h1) (Table 1). Incomplete dechlorination was observed during this period, with a mix of VC and ethene in the sand column effluent. From 3.4 to 5.7 pore volumes, only lactate was provided to the influent porewater (i.e., no cis-DCE was added) and the sediment effluent served as the sole source of chlorinated ethenes (ca. 34 mM). The sand column successfully achieved complete reductive dechlorination of the applied chlorinated ethenes to ethene during this period, demonstrating that with lactate addition and a residence time of 2 days the sand cap could detoxify the flux of chlorinated ethenes exiting the sediment column. These data, coupled with the lack of complete dechlorination during the previous condition (0e3.4 sand pore volumes) when additional cis-DCE was provided to the influent, suggests that high chloroethene concentrations entering the sand column limited the extent of dechlorination. The results obtained from Bioreactive Cap B indicate that electron donor concentrations and contaminant residence times within the cap can impact dechlorination activity. Complete dechlorination was observed between 8.0 and 13.3 pore volumes when lactate was provided to the sand column
5371
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 5 3 6 5 e5 3 7 4
A
Bioreactive Cap Experimental Conditions Sand Cap connected in series with mud effluent No lactate provided R = 0.47 days 0
B
5
10
15
20
25
30
35
Product Distribution Normalized to Total Chloroethenes + Ethenes Eluted
Bioreactive Sand Cap Effluent Product Distribution 1.0
1.0
0.8
0.8
0.6
0.6
0.4
0.4
0.2
0.2
0.0
DCE VC ETH Ethane
0.0 0
5
10
15
20
25
30
35
Pore Volumes Eluted from Cap
C
Bioreactive Sand Cap Effluent Methane Production
Cummulative Methane (mg/L)
4
4
3
3
2
2
1
1
0
0 0
5
10
15
20
25
30
35
Pore Volumes Eluted from Cap Fig. 3 e AeB. (A) Operating conditions for Bioreactive Cap A. The effluent of the sediment column served as the influent of the sand column and was not amended with exogenous electron donors, electron acceptors, carbon sources, minerals, nor vitamins. (B) Effluent product distribution of Bioreactive Cap A inoculated with a PCE-to-ethene dechlorinating mixed consortia and connected in series with sediment column effluent between 68 and 83 sediment pore volumes. (C) Cumulative aqueous methane concentration in samples collected from Bioreactive Cap A effluent.
despite relatively fast flow rates (0.47 residence time). When lactate was removed from the influent at 13.3 pore volumes, however, a mixture of chlorinated ethenes was observed in the effluent, indicating the importance of exogenous reducing equivalents to the sand column. Delivery of external electron donor is a common technique used to stimulate and enhance reductive dechlorination in groundwater aquifers (Anderson et al., 2003; Haas and Trego, 2001; Lendvay et al., 2003; Scow and Hicks, 2005) and may be necessary for bioreactive caps employing anaerobic biotransformations. Contaminant mass entering the cap also dictated performance, as noted above, since incomplete dechlorination was observed when additional cis-DCE was supplemented into the influent (0e3.4 PV) while complete dechlorination was observed when the cap was only treating sediment effluent (5e10 PV).
3.4.
Implications for capping
The combined results from the batch study and the two sand columns suggest that the sediment effluent alone could not sustain complete dechlorination in a bioreactive cap over the range of residence times (0.5e2 days) examined in this study. Batch results showed complete dechlorination occurred after 19 days when the OW consortium was provided only sediment effluent, much longer than the 2 day retention times utilized for these column studies. However, at sites where diffusive conditions exist, or where groundwater seepage rates are significantly slower than those employed here, complete dechlorination could be achieved. For instance, at the USGS biomat pilot test described by Majcher et al. (2009), a bioreactive layer successfully dechlorinated a range of
5372
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 5 3 6 5 e5 3 7 4
0
B
5
5 mM Lactate Rt = 0.47 day
5 mM Lactate Rt = 1 day
R = 2 days
5 mM Lactate Rt = 2 days
5 mM Lactate
DCE to influent
Bioreactive Cap Experimental Conditions
A
10
No Lactate R = 0.47 day
15
20
25
30
35
Product Distribution Normalized to Total Chloroethenes + Ethene Eluted
Bioreactive Sand Cap Effluent Product Distribution 1.0
1.0
0.8
0.8
0.6
0.6
0.4
0.4
0.2
0.2
0.0
DCE VC ETH Ethane
0.0 0
5
10
15
20
25
30
35
Pore Volumes Eluted from Cap
C Cummulative Methane (mg/L)
Bioreactive Sand Cap Effluent Cummulative Methane 80
80
60
60
40
40
20
20
0
0 0
5
10
15
20
25
30
35
Pore Volumes Eluted from Cap Fig. 4 e AeB. (A) Operating conditions for Bioreactive Cap B. The effluent of the sediment column served as the influent of the Bioreactive Cap and was not amended with minerals nor vitamins. Exogenous electron donor (lactate), carbon sources (lactate), and electron acceptor (cis-DCE) were added where indicated. (B) Effluent product distribution of Bioreactive Cap B inoculated with a PCE-to-ethene mixed dechlorinating consortia and connected in series with sediment column effluent between 146 and 185 sediment pore volumes. (C) Cumulative aqueous methane concentration in samples collected from Bioreactive Cap B effluent.
chlorinated aliphatics at a site where average hydraulic residence times in the reactive mat were assumed to be 8e14 days. Thissystem also included an organic layer composed of a mixture of peat, compost, and chitin to provide long-term electron donor.
4.
Conclusions
Based on the results presented herein, Engineered controls may be needed to maintain microbial dechlorination activity, reduce contaminant flux, or increase contaminant residence time for bioreactive caps to achieve
complete reductive dechlorination of dissolved chlorinated ethenes to ethene. Incorporation of electron donor was required to stimulate and sustain long-term contaminant biotransformations in a bioreactive cap under the conditions tested. At sites with lower seepage velocities, allowing for greater residence time in the cap, complete dechlorination without electron donor may be possible. The need for electron donor delivery in bioactive design could support greater cell numbers of degrading populations, resulting in greater degradation rates and possible deployment at sites with reasonably high contaminant flux (e.g, high concentrations, high flow rates). This is supported
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 5 3 6 5 e5 3 7 4
by data in this study during Bioreactive Cap B, where the addition of electron donor, albeit at relatively high concentrations, supported complete dechlorination under relative short residence times (1e2 days). Careful attention should be provided to accurately characterize seepage rates and contaminant concentrations at sites where contaminated groundwater seeps are present. In summary, this study examined the conditions governing the implementation of novel subaqueous bioreactive in situ caps. Experimental results suggest that the process is feasible provided that sufficient electron donor and contaminant mass fluxes exist in the bioactive cap.
Acknowledgments Funding for this research was provided by the Hazardous Substance Research Center-South and Southwest, the National Institute of Environmental Health Sciences, and a fellowship to D.W.H from the Georgia Institute of Technology.
Appendix. Supplementary information Supplementary data associated with this article can be found, in the online version, at doi:10.1016/j.watres.2011.06.022.
references
Abe, Y., Aravena, R., Zopfi, J., Parker, B., Hunkeler, D., 2009. Evaluating the fate of chlorinated ethenes in streambed sediments by combining stable isotope, geochemical and microbial methods. Journal of Contaminant Hydrology 107 (1e2), 10e21. Adamson, D.T., Newell, C.J., 2009. Support of source zone bioremediation through endogenous biomass decay and electron donor recycling. Bioremediation Journal 13 (1), 29e40. Anderson, R.T., Vrionis, H.A., Ortiz-Bernad, I., Resch, C.T., Long, P.E., Dayvault, R., Karp, K., Marutzky, S., Metzler, D.R., Peacock, A., White, D.C., Lowe, M., Lovley, D.R., 2003. Stimulating the in situ activity of Geobacter species to remove uranium from the groundwater of a uranium-contaminated aquifer. Applied and Environmental Microbiology 69 (10), 5884e5891. Carr, C.S., Hughes, J.B., 1998. Enrichment of high rate PCE dechlorination and comparative study of lactate, methanol, and hydrogen as electron donors to sustain activity. Environmental Science & Technology 32 (12), 1817e1824. Cho, Y.-M., Smithenry, D.W., Ghosh, U., Kennedy, A.J., Millwarde, R.N., Bridges, T.S., Luthy, R.G., 2007. Field methods for amending marine sediment with activated carbon and assessing treatment effectiveness. Marine Environmental Research 64 (5), 541e555. Choi, H., Agarwal, S., Al-Abed, S.R., 2009. Adsorption and simultaneous dechlorination of PCBs by GAC impregnated with ZVI/Pd bimetallic particles: mechanistic aspects and reactive capping barrier concept. Environmental Science & Technology 43 (2), 488e493. Conant, B., Cherry, J.A., Gillham, R.W., 2004. A PCE groundwater plume discharging to a river: influence of the streambed and
5373
near-river zone on contaminant distributions. Journal of Contaminant Hydrology 73 (1e4), 249e279. Daprato, R.C., Lo¨ffler, F.E., Hughes, J.B., 2006. Comparative analysis of three tetrachloroethene to ethene halorespiring consortia suggests functional redundancy. Environmental Science & Technology 41 (7), 2261e2269. Dries, J., Bastiaens, L., Springael, D., Agathos, S.N., Diels, L., 2004. Competition for sorption and degradation of chlorinated ethenes in batch zero-valent iron systems. Environmental Science & Technology 38 (10), 2879e2884. Go, J.L., Lampert, D.J., Stegemann, J.A., Reible, D.D., 2009. Predicting contaminant fate and transport in sediment caps: mathematical modeling approaches. Applied Geochemistry. 24 (7), 1347e1353. Haas, J.E., Trego, D.A., 2001. A field application of hydrogenreleasing compound (HRC) for the enhanced bioremediation of methyl tertiary butyl ether (MTBE). Soil & Sediment Contamination 10 (5), 555e575. Hamonts, K., Kuhn, T., Maesen, M., Bronders, J., Lookman, R., Kalka, H., Diels, L., Meckenstock, R.U., Springael, D.D., Dejonghe, W., 2009. Factors determining the attenuation of chlorinated aliphatic hydrocarbons in eutrophic river sediment impacted by discharging polluted groundwater. Environmental Science & Technology 43 (14), 5270e5275. Himmelheber, D.W., Pennell, K.D., Hughes, J.B., 2007. Natural attenuation processes during in situ capping. Environmental Science & Technology 41 (15), 5306e5313. Himmelheber, D.W., Taillefert, M., Pennell, K.D., Hughes, J.B., 2008. Spatial and temporal evolution of biogeochemical processes following in situ capping of contaminated sediments. Environmental Science & Technology 42, 4113e4120. Himmelheber, D.W., Thomas, S.H., Lo¨ffler, F.E., Taillefert, M.T., Hughes, J.B., 2009. Microbial colonization of an in situ sediment cap and correlation to stratified redox zones. Environmental Science & Technology 43, 66e74. Hyun, S., Jafvert, C.T., Lee, L.S., Rao, P.S.C., 2006. Laboratory studies to characterize the efficacy of sand capping a coal tarcontaminated sediment. Chemosphere 63 (10), 1621e1631. Jacobs, P.H., Fo¨rstner, U., 1999. Concept of subaqueous capping of contaminated sediments with active barrier systems (ABS) using natural and modified zeolites. Water Research 33 (9), 2083e2087. Krumins, V., Park, J.-W., Son, E.-K., Rodenburg, L.A., Kerkhof, L.J., Ha¨ggblom, M.M., Fennell, D.E., 2009. PCB dechlorination enhancement in Anacostia River sediment microcosms. Water Research 43 (18), 4549e4558. Lendvay, J.M., Dean, S.M., Adriaens, P., 1998. Temporal and spatial trends in biogeochemical conditions at a groundwatersurface water interface: implications for natural bioattenuation. Environmental Science & Technology 32 (22), 3472e3478. Lendvay, J.M., Lo¨ffler, F.E., Dollhopf, M., Aiello, M.R., Daniels, G., Fathepure, B.Z., Gebhard, M., Heine, R., Helton, R., Shi, J., Krajmalnik-Brown, R., Major, C.L., Barcelona, M.J., Petrovskis, E., Hickey, R., Tiedje, J.M., Adriaens, P., 2003. Bioreactive barriers: a comparison of bioaugmentation and biostimulation for chlorinated solvent remediation. Environmental Science & Technology 37 (7), 1422e1431. Lo¨ffler, F.E., Edwards, E.A., 2006. Harnessing microbial activities for environmental cleanup. Current Opinion in Biotechnology 17 (3), 274e284. Lorah, M.M., Voytek, M.A., 2004. Degradation of 1,1,2,2tetrachloroethane and accumulation of vinyl chloride in wetland sediment microcosms and in situ porewater: biogeochemical controls and associations with microbial communities. Journal of Contaminant Hydrology 70 (1e2), 117e145. Lorah, M.M., Majcher, E.H., Jones, E.J., Voytek, M.A., 2008. Microbial Consortia Development and Microcosm and Column
5374
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 5 3 6 5 e5 3 7 4
Experiments for Enhanced Bioremediation of Chlorinated Volatile Organic Compounds, West Branch Canal Creek Wetland Area, Aberdeen Proving Ground, Maryland. USGS Scientific Investigations Report. 2007-5165, pp. 1e79. Lowry, G.V., Johnson, K.M., 2004. Congener specific PCB dechlorination by microscale and nanoscale zero-valent iron in a methanol/water solution. Environmental Science & Technology 38 (19), 5208e5216. Majcher, E.H., Lorah, M.M., Phelan, D.J., McGinty, A.L., 2009. Design and performance of an enhanced bioremediation pilot test in a tidal wetland seep, West Branch Canal Creek, Aberdeen Proving Ground, Maryland. USGS Scientific Investigations Report 2009-5112, 1e70. Majcher, E.H., Phelan, D.J., Lorah, M.M., McGinty, A.L., 2007. Characterization of Preferential Ground-water Seepage from a Chlorinated Hydrocarbon-contaminated Aquifer to West Branch Canal Creek, Aberdeen Proving Ground, Maryland, 20022004. USGS Scientific Investigations Report. 2006-5233, pp. 1e193. McDonough, K.M., Murphy, P., Olsta, J., Zhu, Y.W., Reible, D., Lowry, G.V., 2007. Development and placement of a sorbentamended thin layer sediment cap in the Anacostia River. Soil & Sediment Contamination 16 (3), 313e322. Murphy, P., Marquette, A., Reible, D., Lowry, G.V., 2006. Predicting the performance of activated carbon-, coke-, and soilamended thin layer sediment caps. Journal of Environmental Engineering e ASCE 132 (7), 787e794. Reible, D., 2005. Personal communication.
Reible, D., Hayes, D., Lue-Hing, C., Patterson, J., Bhowmik, N., Johnson, M., Teal, J., 2003. Comparison of the long-term risks of removal and in situ management of contaminated sediments in the Fox River. Soil & Sediment Contamination 12 (3), 325e344. Reible, D.D., Lampert, D., Constant, D.W., Mutch, R.D., Zhu, Y., 2007. Active capping demonstration in the Anacostia River, Washington, DC. Remediation 17 (1), 39e53. Scow, K.M., Hicks, K.A., 2005. Natural attenuation and enhanced bioremediation of organic contaminants in groundwater. Current Opinion in Biotechnology 16 (3), 246e253. SERDP and ESTCP, 2008. Expert Panel Workshop on Research and Development Needs for Understanding and Assessing the Bioavailability of Contaminants in Soils and Sediments. Thoma, G.J., Reible, D.D., Valsaraj, K.T., Thibodeaux, L.J., 1993. Efficiency of capping contaminated sediments in situ. 2. Mathematics of diffusion adsorption in the capping layer. Environmental Science & Technology 27 (12), 2412e2419. van Genuchten, M., 1981. Non-equilibrium Transport Parameters from Miscible Displacement Experiments. Research Report 119, U.S. Salinity Lab, USDA, pp. 1e94. Zimmerman, J.R., Ghosh, U., Millward, R.N., Bridges, T.S., Luthy, R.G., 2004. Addition of carbon sorbents to reduce PCB and PAH bioavailability in marine sediments: physicochemical tests. Environmental Science & Technology 38 (20), 5458e5464.
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 5 3 7 5 e5 3 8 0
Available at www.sciencedirect.com
journal homepage: www.elsevier.com/locate/watres
Preparation of ion exchanger layered electrodes for advanced membrane capacitive deionization (MCDI) Ju-Young Lee, Seok-Jun Seo, Sung-Hyun Yun, Seung-Hyeon Moon* School of Environmental Science and Engineering, Gwangju Institute of Science and Technology (GIST), 261 Cheomdan-gwagiro, Buk-gu, Gwangju 500-712, Republic of Korea
article info
abstract
Article history:
A noble electrode for capacitive deionization (CDI) was prepared by embedding ion
Received 18 February 2011
exchanger onto the surface of a carbon electrode to practice membrane capacitive deion-
Received in revised form
ization (MCDI). Bromomethylated poly (2, 6-dimethyl-1, 4-phenylene oxide) (BPPO) was
30 April 2011
sprayed on carbon cloth followed by sulfonation and amination to form cation exchange
Accepted 22 June 2011
and anion exchange layers, respectively. The ion exchange layers were examined by
Available online 3 July 2011
Scanning electron microscopy (SEM) and Fourier transform infrared spectrometer (FT-IR). The SEM image showed that the woven carbon cloth was well coated and connected with
Keywords:
BPPO. The FT-IR spectrum revealed that sulfonic and amine functional groups were
Capacitive deionization (CDI)
attached on the cationexchange and anionexchange electrodes, respectively. The advan-
BPPO
tages of the developed carbon electrodes have been successively demonstrated in a batch
Carbon cloth
and a continuous mode CDI operations without ion exchange membranes for salt removal
Spraying
using 100 mg/L NaCl solution. ª 2011 Elsevier Ltd. All rights reserved.
1.
Introduction
Recently, capacitive deionization (CDI) has been used as a water treatment technology due to the simple principle and low operating potential, without the need for chemicals (Ito et al., 2007). Therefore, it is considered as an environmentally-friendly and economical system (Foo and Hameed, 2009). A CDI system operation consists of adsorption and desorption periods for obtaining purified water and concentrated water, respectively (Lee et al., 2010). When an electric potential is applied to CDI cells, charged ions in contaminant water are adsorbed onto the surface of charged electrodes, and formed an electric double layer due to the charged electrode and adsorbed ions, producing purified water. After the adsorption of ions, the saturated electrode undergoes regeneration by desorption of the adsorbed ions under zero electrical potential or reversed electric field (Seo
et al., 2010). Hereby utilization of the adsorption ability of an electrode is the key parameter for the CDI operation. In order to maintain acceptable operation efficiencies, the complete adsorption and desorption of charged ions should be accomplished within appropriate periods. Practically, however, when a potential is applied to a CDI cell, counter ions are attracted onto the electrode surface, simultaneously co-ions expelled from the counter electrode (Kim and Choi, 2010). It leads to a higher energy consumption and a lower operation efficiency due to mobility of unwanted ions. To avoid this phenomenon, a membrane-CDI (MCDI) is employed with the help of ion selective membranes in the CDI cell. A MCDI has two types of ion exchange membranes, i.e. anion exchange and cation exchange membranes (AEM & CEM, respectively). The AEM and CEM are positioned in front of the positively and negatively charged electrodes, respectively (Lee et al., 2006). The ion exchange membrane has the
* Corresponding author. Fax: þ82 62 715 2434. E-mail address:
[email protected] (S.-H. Moon). 0043-1354/$ e see front matter ª 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.watres.2011.06.028
5376
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 5 3 7 5 e5 3 8 0
ability to selectively permeate ions, i.e., a CEM permit the passage of cations only, while an AEM allow the passage of anions only. The selectivity of ion exchange membranes prevent reverse adsorption and prohibit the mobility of unwanted ions. However, a MCDI requires strong physical pressure for smooth contact between the membrane and electrode material. Also the diffusion layer on the membrane surface will become thick when the concentration of contaminants in the feed water is low (Dlugolecki et al., 2010). Accordingly, this phenomenon induces a decreased mobility of wanted ions and a high interfacial resistance of the membrane. In this study to solve the problems of contact resistance and the diffusion layer, an advanced-MCDI (A-MCDI) was developed by adhering the ion exchanger onto the carbon electrode surface as a thin layer, which reduces the contact resistance between the ion exchanger and electrode of a MCDI. Therefore, an A-MCDI is expected to exhibit a high removal efficiency and a low current consumption compared to a conventional MCDI.
The amination of the BPPO embedded electrode was conducted by immersion of the electrode in 25% TMA for 20 min. The embedded electrode was maintained in distilled water (Tang et al., 2005).
2.3.
Characterization of the embedded electrodes
The surface morphology of the BPPO embedded electrodes were characterized using Scanning electron microscopy (SEM, JEOL, Japan). Both the bare and embedded carbon electrodes were scanned to compare the morphologies before and after coating with the BPPO slurry, at magnifications of 2,000 and 10,000 times. The sulfonated and aminated electrodes were examined using Fourier transform infrared spectrometry (FT-IR, 460 Plus, Jasco Japan) within the wavelength range 400e4000 cm1. FT-IR is used to confirm the functional groups via the bond vibration and stretching energies between atoms. The electrode with the base polymer, sulfonated and aminated electrodes were analyzed to compare the peaks corresponding the functionalization.
2.
Experimental
2.4.
Salt removal test
2.1.
Materials
2.4.1.
Comparison of the CDI, MCDI and A-MCDI
Bromomethylated poly (2, 6-dimethyl-1, 4-phenylene oxide) (BPPO) was used as the base polymer, and N-methyl-2pyrrolidone (NMP, Fluka, Japan) was purchased as the solvent to dissolve BPPO. Carbon cloth (Kuraray, Japan), 3 cm 8.5 cm in size, was employed as the carbon electrode material. The BPPO embedded electrode was sulfonated using sulfuric acid (H2SO4 99%, DC chemical, Korea), while amination was performed using trimethylamine (TMA, 25wt % in D.I.water, Aldrich). In the salt removal test, sodium chloride (NaCl, Dongyang Chemical, Korea) was used to prepare the feed water.
2.2. Preparation of the ion exchanger embedded electrodes Embedding the ion exchanger onto the electrode surface was carried out in two steps; coating the carbon electrode with a base polymer and then attaching functional groups onto the polymer. BPPO (donated by laboratory of fundamental membranes in USTC) slurry was prepared by mixing 1 g of BPPO with 5 ml of NMP at room temperature for 1 day. The BPPO slurry was then sprayed onto the surface of one side of the carbon cloth using an air brush. In order to evaporate the NMP, the electrode was dried in an oven at 40 C for 12 h, forming the BPPO coated electrodes for sulfonation and amination. The sulfonation of the BPPO embedded electrode was conducted using four different solution concentrations, i.e. 99, 80, 50 and 30% sulfuric acid solutions. The BPPO embedded carbon cloth was initially immersed in 99% sulfuric acid solution for 20 min, and subsequently moved to each of the 80, 50 and 30% sulfuric acid solutions for 1 min. The embedded electrode was finally maintained in distilled water (Liu et al., 2006).
To compare the salt removal efficiency of the A-MCDI, experiments were performed using 3 cm 8.5 cm CDI, MCDI and A-MCDI unit cells, with simultaneous recording of the ion conductivity and current in a batch system (Brose´us et al., 2009). Fifty mililiters of feed water was continuously circulated by a pump (Masterflex Cole-parmer, USA). The initial conductivity of feed water, 100 mg/L sodium chloride, was 190 mS/cm. The variation in the ion conductivity was measured every 30 s in the reservoir using a TDS conductivity meter (OAKTON, Japan). A potential of 1.8 V was applied to the CDI cells using an Agilent 6613C (Agilent, USA) power supply. Feed water was continuously circulated at a flow rate of 4 ml/min, which was determined with respect to the surface area of the electrode. Fig. 1 shows schematic diagrams of the assembled CDI, MCDI and A-MCDI cells. In the structure of the CDI cell, the electrodes were located between the current collectors; whereas, in case of the MCDI cell, the positively and negatively charged electrodes were located behind the anion exchange (AMX, Tokuyama, Japan) and cation exchange membranes (CMX, Tokuyama, Japan), respectively. In the A-MCDI cell, the sulfonated BPPO embedded electrode was positioned in front of negatively charged current collector, while the aminated BPPO embedded electrode was positioned in front of positively charged current collector. The arrangement of the ion exchangers in A-MCDI is the same as MCDI, because even though the ion exchangers have the ion exchange capacity for the selected ions, the system still requires electrical potential as a driving force between the cathode and anode. The three systems were compared in terms of their removal efficiencies and energy consumptions.
2.4.2.
Cyclic testing of the A-MCDI
Cyclic testing of the A-MCDI was performed to demonstrate its ability of repeated operation in a continuous mode (Dai et al.,
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 5 3 7 5 e5 3 8 0
5377
Fig. 1 e Schematics of the CDI, MCDI and A-MCDI cell structures.
2005). The experiment was carried out under the same conditions as the previous salt removal test and subjected to repeated periods of adsorption and desorption. The adsorption and desorption periods were determined to be 5 and 5 min, respectively, to allow sufficient time for adsorption and desorption.
3.
Results and discussion
3.1.
Morphologies of the electrodes
Fig. 2 shows the SEM images of the surfaces of the bare electrode and ion exchanger embedded electrodes. The left and right sides of the figure are the bare carbon cloth and ion exchanger embedded electrode surfaces, respectively. Each sample was magnified by 2,000 and 10,000 times, respectively. Originally the woven carbon cloth material was complicatedly interlinked (Ahn et al., 2007). The bare carbon
electrode surface is observed to have clean surface, with only a tiny amount of dust, whereas, in case of the ion exchanger embedded electrode, the polymer is filled within the carbon cloth and interconnects the individual fibers. In a higher magnification, the bare electrode shows porous carbon cloth surface. However, the ion exchanger embedded electrode shows the pores blocked by the base polymer. Consequently, it was found that the ion exchanger layer was well formed on the surface of the carbon cloth, so that the ion-conducting surface of the embedded electrode may provide good ion selectivity and enhanced conductivity.
3.2.
FT-IR analyses
To impart ion exchange capabilities, the sulfonic and amine functional groups should be attached to the base polymer. The sulfonic and amine groups have the abilities to transport cations and anions, respectively in accordance with the Grotthuss mechanism (Agmon, 1995) and vehicle mechanism
Fig. 2 e SEM images of (a) bare carbon cloth surface at 2,000 times, (b) embedded carbon electrode surface at 2,000 times, (c) bare carbon cloth surface at 10,000 times, (d) embedded carbon cloth surface at 10,000 times.
5378
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 5 3 7 5 e5 3 8 0
Fig. 3 e FT-IR analyses of the untreated, sulfonated and aminated electrode surfaces.
(Schuster et al., 2008). In order to examine the sulfonic and amine groups on the BPPO surface, three types of electrodes prepared with the base polymer, sulfonated and aminated polymers were analyzed using FT-IR. Fig. 3 shows the existence of functional groups, indicated by the unique peaks in the sulfonated and aminated BPPO. When the sulfonated BPPO was compared with the untreated BPPO, the sulfonic group peak appeared around 1165e1120 cm1 (Panicker et al., 2006). Similarly, when the aminated BPPO was compared with the untreated BPPO, the amine group peak appeared around 850e750 cm1 (Volkov et al., 1980). The results show that the functional groups were successfully attached to the BPPO.
3.3.
Salt removal test
3.3.1.
Comparison of the CDI, MCDI and A-MCDI
CDI, MCDI, and A-MCDI tests were performed to observe the initial adsorption capability of each system. The results
Fig. 4 e Variations in the ion conductivity of the CDI, membrane-CDI (MCDI) and advanced-MCDI (A-MCDI) during their operation.
obtained for the CDI, MCDI and A-MCDI were compared in terms of salt removal and current consumption, as shown in Figs. 4 and 5, respectively, at an applied potential of 1.8 V and flow rate of 4 ml/min over a 30 min period. As shown in Fig. 4, the CDI and A-MCDI exhibited higher degrees of removal than the MCDI. It means that the removal efficiency of MCDI was lower than the CDI and A-MCDI at the same voltage, because the MCDI suffered problems, such as contact resistance and membrane resistance. Fig. 5 shows that the MCDI and A-MCDI allowed lower currents than that of the CDI. This phenomenon is due to the fact that the current depends on the total ionic flux. In other words, the ion selectivity increases the ion mobility resistance at a constant voltage; therefore, a high resistance will decrease the current in accordance with Ohm’s law (I ¼ V/R) (Kim and Choi, 2010). The operation parameters, the ion removal efficiency and power consumption of each system, are listed in Table 1. These results were obtained based on the measured voltages and currents at 30 min after starting the operation.
Fig. 5 e Operating currents of the CDI, MCDI and A-MCDI.
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 5 3 7 5 e5 3 8 0
Table 1 e Numerical results of the ion removal efficiencies, currents and power consumptions during the operation of each system.
Applied electric potential Initial conductivity Conductivity at 30 min Removal efficiency Current at 30 min Power consumption
CDI
MCDI
A-MCDI
1.8 V 190.8 mS/cm 39.9 mS/cm 79.1% 235.7 mA 212.3 mWh
1.8 V 185.1 mS/cm 168 mS/cm 9.23% 2.575 mA 2.318 mWh
1.8 V 188.9 mS/cm 31.5 mS/cm 83.4% 25.64 mA 23.07 mWh
The variation in the feed water conductivity was measured to get the removal efficiency, h which was determined using the following equation (Dermentzis and Ouzounis, 2008): hð%Þ ¼
Ci C 100 Ci
(1)
where Ci is the initial conductivity [mS/cm] of the feed water and C is the conductivity of the diluted feed water [mS/cm] at 30 min after starting the operation. According to Eq. (1), the removal efficiencies of the CDI, MCDI and A-MCDI were approximately 79, 9.2 and 83%, respectively. Obviously, the removal efficiency of A-MCDI was the highest. The power consumption of each system was calculated based on the current measured at the same applied potential according to the following equation (Masiuk, 1999): E ¼ VIt
(2)
where E is the power consumption [mWh], V the applied potential [V], I the current [A] and t the time [h]. As a result, the power consumptions of the CDI, MCDI and A-MCDI were approximately 210, 2.3 and 23 mWh, respectively. Considering the removal efficiency and energy consumption, the A-MCDI showed better performance among the systems tested.
3.3.2.
5379
concentration of the charged ions in the effluent was reduced. When the saturated electrode underwent regeneration with zero electrical potential, the adsorbed ions were desorbed. Accordingly, the charged ions in the effluent increased again. The results showed that the repeated adsorption and desorption occurred in a regular pattern of a CDI system. This implies that the A-MCDI system would have stable performance over repeated operation; thus, demonstrated the practical applicability as a water treatment system.
4.
Conclusion
An A-MCDI has been developed by introducing an ion exchanger layer on electrode surface for a CDI system. The electrodes were prepared by coating a base polymer on the carbon cloth followed by functionalization of sulfonic and amine groups on the polymer structure. The noble electrode enables to overcome the drawbacks of a membrane-CDI system by reducing the interfacial resistance between the ion exchanger layer and the carbon electrode. Practically the A-MCDI is operated without ion exchange membranes while the system performs better than a conventional MCDI in terms of salt removal efficiency. This research may contribute significantly in application of CDI in various water treatment systems such as desalination, hardness removal, and drinking water treatment.
Acknowledgment This research was supported by a grant (07seaheroB02-02-01) from the Plant Technology Advancement Program, funded by the Ministry of Land, Transport and Maritime Affairs.
Cyclic testing of the A-MCDI
A cyclic test was performed for a continuous operation of AMCDI. Fig. 6 shows the variation of the ion conductivity in the effluent stream. During the initial adsorption period, charged ions in feed water were adsorbed onto the electrode; therefore, the
Fig. 6 e Continuous mode operation of the A-MCDI system by repeated cycles of adsorption and desorption.
references
Agmon, N., 1995. The Grotthuss mechanism. Chemical Physics Letters 244, 456e462. Ahn, H.-J., Lee, J.-H., Jeong, Y., Lee, J.-H., Chi, C.-S., Oh, H.-J., 2007. Nanostructured carbon cloth electrode for desalination from aqueous solutions. Materials Science and Engineering: A 449e451, 841e845. Brose´us, R., Cigana, J., Barbeau, B., Daines-Martinez, C., Suty, H., 2009. Removal of total dissolved solids, nitrates and ammonium ions from drinking water using charge-barrier capacitive deionisation. Desalination 249, 217e223. Dai, K., Shi, L., Fang, J., Zhang, D., Yu, B., 2005. NaCl adsorption in multi-walled carbon nanotubes. Materials Letters 59, 1989e1992. Dermentzis, K., Ouzounis, K., 2008. Continuous capacitive deionization-electrodialysis reversal through electrostatic shielding for desalination and deionization of water. Electrochimica Acta 53, 7123e7130. Dlugolecki, P., Anet, B., Metz, S.J., Nijmeijer, K., Wessling, M., 2010. Transport limitations in ion exchange membranes at low salt concentrations. Journal of Membrane Science 346, 163e171. Foo, K.Y., Hameed, B.H., 2009. A short review of activated carbon assisted electrosorption process: an overview, current stage and future prospects. Journal of Hazardous Materials 170, 552e559.
5380
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 5 3 7 5 e5 3 8 0
Ito, E., Mozia, S., Okuda, M., Nakano, T., Toyoda, M., Inagaki, M., 2007. Nanoporous carbons from cypress II. Application to electric double layer capacitors. New Carbon Materials 22, 321e326. Kim, Y.-J., Choi, J.-H., 2010. Enhanced desalination efficiency in capacitive deionization with an ion-selective membrane. Separation and Purification Technology 71, 70e75. Lee, J.-B., Park, K.-K., Eum, H.-M., Lee, C.-W., 2006. Desalination of a thermal power plant wastewater by membrane capacitive deionization. Desalination 196, 125e134. Lee, J.Y., Seo, S.J., Park, J.W., Moon, S.H., 2010. A study on the cell structure for capacitive deionization system. Korean Chemical Engineering Research 48, 791e794. Liu, J., Xu, T., Han, X., Fu, Y., 2006. Synthesis and characterizations of a novel zwitterionic hybrid copolymer containing both sulfonic and carboxylic groups via sulfonation and zwitterionic process. European Polymer Journal 42, 2755e2764. Masiuk, S., 1999. Power consumption measurements in a liquid vessel that is mixed using a vibratory agitator. Chemical Engineering Journal 75, 161e165.
Panicker, C.Y., Varghese, H.T., Philip, D., Nogueira, H.I.S., 2006. FT-IR, FT-Raman and SERS spectra of pyridine-3-sulfonic acid. Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy 64, 744e747. Schuster, M., Kreuer, K.-D., Steininger, H., Maier, J., 2008. Proton conductivity and diffusion study of molten phosphonic acid H3PO3. Solid State Ionics 179, 523e528. Seo, S.-J., Jeon, H., Lee, J.K., Kim, G.-Y., Park, D., Nojima, H., Lee, J., Moon, S.-H., 2010. Investigation on removal of hardness ions by capacitive deionization (CDI) for water softening applications. Water Research 44, 2267e2275. Tang, B., Xu, T., Gong, M., Yang, W., 2005. A novel positively charged asymmetry membranes from poly(2,6-dimethyl-1,4phenylene oxide) by benzyl bromination and in situ amination: membrane preparation and characterization. Journal of Membrane Science 248, 119e125. Volkov, A., Tourillon, G., Lacaze, P.-C., Dubois, J.-E., 1980. Electrochemical polymerization of aromatic amines: IR, XPS and PMT study of thin film formation on a Pt electrode. Journal of Electroanalytical Chemistry 115, 279e291.
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 5 3 8 1 e5 3 8 8
Available at www.sciencedirect.com
journal homepage: www.elsevier.com/locate/watres
Electrochemical sulfide oxidation from domestic wastewater using mixed metal-coated titanium electrodes Ilje Pikaar, Rene´ A. Rozendal, Zhiguo Yuan, Ju¨rg Keller, Korneel Rabaey* The University of Queensland, Advanced Water Management Centre (AWMC), Brisbane, QLD 4072, Australia
article info
abstract
Article history:
Hydrogen sulfide generation is a major issue in sewer management. A novel method based
Received 8 April 2011
on electrochemical sulfide oxidation was recently shown to be highly effective for sulfide
Received in revised form
removal from synthetic and real sewage. Here, we compare the performance of five
11 July 2011
different mixed metal oxide (MMO) coated titanium electrode materials for the electro-
Accepted 25 July 2011
chemical removal of sulfide from domestic wastewater. All electrode materials performed
Available online 6 August 2011
similarly in terms of sulfide removal, removing 78 5%, 77 1%, 85 4%, 84 1%, and 83 2% at a current density of 10 mA/cm2 using Ta/Ir, Ru/Ir, Pt/Ir, SnO2 and PbO2,
Keywords:
respectively. Elevated chloride concentrations, often observed in coastal areas, did not
Electrochemical oxidation
entail any significant difference in performance. Independent of the electrode material
Oxygen generation
used, sulfide oxidation by in situ generated oxygen was the predominant reaction mech-
Sulfide oxidation
anism. Passivation of the electrode surface by deposition of elemental sulfur did not occur.
Sewer
However, scaling was observed in the cathode compartment. This study shows that all the MMO coated titanium electrode materials studied are suitable anodic materials for sulfide removal from wastewater. Ta/Ir and Pt/Ir coated titanium electrodes seem the most suitable electrodes since they possess the lowest overpotential for oxygen evolution, are stable at low chloride concentration and are already used in full scale applications. ª 2011 Elsevier Ltd. All rights reserved.
1.
Introduction
Hydrogen sulfide is ubiquitously present in industrial and domestic waste streams, and is of special concern in sewer systems since it is responsible for odor issues in urban areas, is toxic to sewer workers, and is the main cause for sewer pipe corrosion (Zhang et al., 2008). Repair and/or replacement of corroded sewer pipes results in considerable costs (Sydney et al., 1996; Vincke et al., 2002; Kaempfer and Berndt, 1998), and therefore measures for mitigating sulfide production and emission are required. Current strategies to prevent sewer corrosion come with substantial costs due to both chemical consumption and system maintenance (Zhang et al., 2008; Hvitved-Jacobsen, 2001).
Recently, we described the electrochemical oxidation of aqueous sulfide using Ta/Ir coated titanium electrodes from domestic wastewater (Pikaar et al., 2011). At the used current densities of 5 mA/cm2 sulfide could be oxidized, producing elemental sulfur, thiosulfate and sulfate as the final products. Indirect oxidation of sulfide with in situ generated oxygen rather than the direct oxidation of sulfide at the electrode was shown to be the predominant reaction mechanism due to the low sulfide concentrations (i.e. w10 mg/L) normally observed in sewers. In comparison to sulfide control with conventional oxygen injection, the method does not require any transport and storage of oxygen, whereas in situ generated oxygen is expected to have a much higher efficiency due to the fine dispersion (i.e. 1e30 mm) of the generated oxygen (Chen, 2004).
* Corresponding author. Tel.: þ61 7 3365 7519; fax: þ61 7 3365 4726. E-mail address:
[email protected] (K. Rabaey). 0043-1354/$ e see front matter ª 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.watres.2011.07.033
5382
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 5 3 8 1 e5 3 8 8
Electrode materials possess different selectivity and catalytic activities, and may hence produce different intermediates such as oxygen, chlorine/hypochlorous acid, hydroxyl radicals and other reactive oxygen species (Chen, 2004). These intermediates may have a different product spectrum for sulfide oxidation, and may increase/decrease overall oxidation efficiency. In this study, the feasibility of anodic sulfide oxidation in domestic wastewater using titanium electrodes coated with 5 different types of electrocatalyst materials, namely Ta/Ir, Pt/Ir, Ru/Ir, PbO2 and SnO2, was investigated. Ta/Ir coated titanium electrodes are known for their low overpotential for oxygen and their stability. They are therefore commonly used as oxygen evolving electrodes (Chen, 2004). Ru/Ir electrodes find widespread use in the chloro-alkali industry for the in situ oxidation of chloride to chlorine/ hypochlorite (Feng and Li, 2003; Takasu et al., 2010). Both Ru/Ir and Pt/Ir electrodes also possess a low overpotential for oxygen evolution, and therefore especially at low chloride concentrations oxygen evolution may become a predominant reaction mechanism. It is important to note that these electrodes have a low overpotential for the production of chlorine from chloride. In coastal areas (e.g., Queensland, Australia) the chloride concentrations in domestic wastewater can be a high as >1 g/L due to marine infiltrations into the sewer system (Taylor and Gardner, 2007). Hence, in situ generation of hypochlorous acid/hypochlorite may become an important reaction mechanism for sulfide oxidation from domestic wastewater. Hypochlorous acid/hypochlorite can oxidize sulfide to elemental sulfur at pH 7.5, values normally observed in sewer systems. Disadvantage of in situ chlorine generation is the possible formation of toxic organochlorine derivatives (Sun et al., 2009), which needs to be prevented. Contrarily to the three aforementioned materials, PbO2 and SnO2 have a high overpotential for oxygen evolution. Just like boron doped diamond (BDD), the electrocatalysts PbO2 and SnO2 are known to generate hydroxyl radicals from the oxidation of water (Panizza and Cerisola, 2008; Zhu et al., 2008; Panizza et al., 2008). However, contrarily to BDD, PbO2 and SnO2 are made of inexpensive materials which are readily available in practical mesh geometries and at scale, and have a low electrical resistivity. This makes them suitable materials for applications on large scale (i.e. industrial applications and sewer systems). Considering the above, the main aims of this study are to assess the impact of electrode coating on electrochemical sulfide oxidation in wastewater, to assess the differences in energy requirement for the different electrode materials and to assess the impact of chloride concentrations on the sulfide oxidation process.
2.
Materials and methods
2.1.
Electrochemical cell and operation
Fig. 1 gives a schematic diagram of the electrochemical cell. The two-chambered electrochemical cell consisted of two parallel Perspex frames (internal dimensions 20 4.8 1.2 cm) separated by a cation exchange membrane (Ultrex CM17000, Membranes International Inc., USA) to create an anode and
cathode compartment each with a volume of about 100 mL. In the anode chamber, mesh shaped Ta/Ir (TaO2/IrO2: 0.35/0.65), Ru/Ir (RuO2/IrO2: 0.70/0.30), Pt/Ir (PtO2/IrO2: 0.70/0.30), PbO2 and SnO2 coated titanium electrodes (diameter: 240 mm; thickness: 1 mm; specific surface area: 1.0 cm2/cm2) were used (Magneto Anodes BV, The Netherlands). Stainless steel fine mesh (24 cm2) with a stainless steel current collector (6 mm mesh size, 0.8 mm wire connected via a 6 mm stainless steel rod) was used as electrode material in the cathode chamber. Both the anode and cathode had a projected electrode surface area of 24 cm2. In all experiments, an Ag/AgCl (RE-5B, Bio Analytical, USA) was used as the reference electrode (þ197 mV versus SHE). The anode liquid medium was constantly recirculated over a 5 L vessel, allowing a total anode liquid volume of 5 L. The influent flow rate through the anode chamber was maintained at 3.6 L/h using a peristaltic pump (Watson Marlow, UK). An additional recirculation flow in the anode chamber, which was kept at 22 L/h using a peristaltic pump (Watson Marlow, UK), to obtain a good mixing rate in the anode chamber was used. PVC tubing with an internal diameter of 8 mm was used for the feeding and recirculation lines. The off-gas coming from the external buffer vessel was captured in a gas collection bag (TKC Tedlar bags, Air-Met Scientific Pty Ltd, USA). An external buffer flask of 2 L was used in the recirculation of the cathode chamber. A 0.10 M NaCl solution in the cathode chamber was used in all experiments. The recirculation flow of the cathode solution was kept at 22 L/h using a peristaltic pump (Watson Marlow, UK). The anode liquid medium, domestic wastewater, was collected weekly from a local pumping station and stored at 4 C. Prior to use, 5 L of the domestic wastewater was heated up to ambient temperatures (24.3 0.5 C) and fed to the influent buffer tank. pH in the influent was measured using a pH probe (Ionode Pty Ltd., AU) and maintained at 7.5 during the experiment through a PLC controlled dosage of a 0.5 M NaOH solution. The wastewater was continuously recirculated. This fed batch system was used to minimize the amount of domestic wastewater needed (continuous feeding would have required significant amounts of wastewater per day). A concentrated sodium sulfide (Na2S$9H2O) stock solution (5.5 g/L sulfide-S), prepared according to Dutta et al. (2008), was continuously supplied to the incoming line of the anode chamber via a syringe pump (NE-1600, New Era Pump Systems, Inc., USA) at a dosing rate of w36 mg sulfide-S/h, which was sufficient to give an anode influent concentration of w10.0 mg S/L by assuming that the recirculated wastewater contained no sulfide.
2.2.
Measurements and calculations
Galvanostatic measurements and control were performed using a Wenking potentiostat/galvanostat (KP07, Bank Elektronik, gmbH, Germany). The anode, cathode potentials and the current were recorded every 60 s using an Agilent 34970A data acquisition unit. The amount of sulfide dosed to the reactor can be expressed as a charge quantity expressed in Coulomb: Q ¼ nFcadded =M
(1)
With F the Faraday constant (96,485 C/mol, n the number of electrons involved (i.e. 8 electrons for the oxidation of 1 mol
5383
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 5 3 8 1 e5 3 8 8
13 Power supply
14
14
7 12 15
10 2
+
H
+
Na K
3
11
8
+
1 5
Sulfide solution
9
6
4
Fig. 1 e Schematic overview experimental setup. 1. Influent buffer (5 L); 2. Influent anodic compartment; 3. Recirculation flow anodic compartment; 4. Sulfide feeding line; 5. Anode compartment; 6. Cathode compartment; 7. Effluent anode compartment; 8. Influent cathode compartment; 9. Cathode buffer (2 L); 10. Effluent cathode compartment; 11. Cathode water-lock; 12. Cathode vent gas (H2); 13. Potentiostat/galvanostat; 14. Sampling points; 15. Anode vent gas (H2S, O2, CO2).
sulfide to sulfate), c the amount of sulfide added (g) and M the molar weight of sulfide (i.e. 32 g/mol). Hence, by determining the coulombic efficiency (CE) based on the conversion of sulfide to sulfate the coulombic efficiency can be calculated as follows: CE ¼
2.3.
nFcremoved =M cremoved ¼ cadded nFcadded =M
(2)
Chemical analyses
Sulfide, sulfite, thiosulfate and sulfate concentrations were measured by Ion Chromatography (IC), using a Dionex 2010i system, according to Keller-Lehmann et al. (2006). Samples were immediately filtered using a 0.22 mm syringe filter (Millipore, USA) and preserved in previously prepared Sulfide Antioxidant Buffer (SAOB) solution prior to ion chromatography analysis. SAOB solution was prepared using Helium purged MilliQ (18MU) water, 3.2 g/L NaOH and 2.8 g/L aascorbic acid. After preparation, the solution was kept refrigerated, shielded from light and not used beyond 24 h. Elemental sulfur was assumed to be the difference between the total sulfide added and the soluble sulfur species (i.e. sulfide, sulfite, thiosulfate and sulfate). COD (range 25e1500 mg/L) and free chlorine (range 0e4 mg/L) concentrations were determined by means of cuvette tests (Merck, Germany). COD concentrations were corrected for the soluble sulfur species in the solution. The conductivity was measured using a hand-held meter
(Cyberscan PC 300, Eutech Instruments). The produced gas (i.e. in situ oxygen generation) was analyzed using a Gas Chromatography (Shimadzu, Japan).
2.4.
Experimental procedures
Experiments were divided into 2 different sets. The first set of experiments compared the performance of five different mixed metal-coated titanium electrode materials (Ta/Ir, Ru/Ir, Pt/Ir, PbO2, SnO2,). Key focal points were (a) the kinetics of the sulfide oxidation, (b) possible oxidation of organics (i.e. COD), (c) the amount of excess in situ generated oxygen and (d) the required energy input (i.e., based on the obtained cell potential) of the different electrode materials. The second set of experiments investigated the influence of chloride concentrations on (a) the kinetics of the sulfide oxidation, (b) possible oxidation of organics (i.e. COD) and (c) the required energy input (i.e. obtained cell potential) using Ru/Ir and Ta/Ir coated titanium as electrode material. Each time, the performance was assessed during 6-h experimental runs using galvanostatic control at a fixed current density of 10 mA/cm2. This current density level was enough to oxidize all sulfide added to the system to sulfate. All experiments were performed in triplicate. Prior to each experiment, the headspace of the influent buffer vessel was flushed with nitrogen for at least 5 min to obtain a headspace that consisted of 100% nitrogen. In this way, any excess in situ
5384
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 5 3 8 1 e5 3 8 8
generated oxygen could be measured using Gas Chromatography after every experiment.
3.
Results and discussion
3.1.
Influence of electrode material on sulfide oxidation
The influence of the electrode material on the kinetics of anodic sulfide oxidation was investigated during 6-h experiments at a fixed current density of 10 mA/cm2. During the experimental runs, sulfide concentrations increased to approximately 7e10 mg/L due to the recirculatory mode of operation. The typical profiles of the sulfide concentration are presented in the supplementary information S1. In Table 1 an overview of the results using Ta/Ir, Ru/Ir, Pt/Ir, PbO2 and SnO2 as electrode material is presented. The obtained removal and Coulombic efficiencies using Ta/Ir, Ru/Ir, Pt/Ir, PbO2 and SnO2 were 78 5, 77 1, 85 4, 83 2 and 84 1%, respectively. This is equal to sulfide removal rates of 10.8 0.2, 11.7 1.1, 1 12.4 0.4, 12.4 0.4 and 12.9 0.8 g S m2 electrode surface h . The obtained Coulombic efficiencies were calculated based on the oxidation of sulfide to sulfate (see Equation (2)). The values observed for the sulfide removal rates expressed in mg S/L wastewater h1 7.5 0.3 (Ta/Ir) to 7.8 0.6 (SnO2) are in agreement with the chemical sulfide oxidation rates found under high dissolved oxygen concentrations in domestic wastewater (Sharma and Yuan, 2010). Furthermore, GC analysis of the produced gas at the end of every experiment showed that in all experiments similar amount of excess oxygen was generated and transferred to the headspace. The obtained excess oxygen generation using Ta/Ir, Ru/Ir, Pt/Ir, PbO2 and SnO2 was 65 13, 60 17, 69 8, 66 17 and 47 8 mg (Table 1). In a previous study, we showed that direct
sulfide oxidation at the electrode surface was negligible under the given operational conditions (Pikaar et al., 2011). If reactive oxygen species were primarily responsible for sulfide oxidation, we would have expected sulfate being the primary product of oxidation, contrarily to the results (see Table 1). If oxygen would be the predominant reaction mechanism, a mixture of sulfur species would be expected. The results suggest that in situ oxygen generation is significant and could be primarily responsible for the sulfide oxidation observed. The materials used are known to have different overpotentials for oxygen evolution. This however does not necessarily mean that they will produce different amounts of oxygen. The formation of reactive oxygen species such as OH radicals is intermediate products during the oxidation of water to oxygen. In the first step, adsorbed OH radicals are formed on the electrode surface. In the second step, the adsorbed OH interacts with the oxygen already present in the oxide anode to form physisorbed or chemisorbed ‘active oxygen’. In absence of any oxidizable pollutant this ‘active oxygen’ subsequently produces oxygen (Comninellis, 1994). Thus, the efficiency of the formation of reactive oxygen species does not only depend on the electrode material but also on the concentration of the pollutant and its reactivity toward oxidation. Especially at low pollutant concentrations, low Coulombic efficiencies for the formation of reactive oxygen species have been observed (Martinez-Huitle and Brillas, 2009), which means that high Coulombic efficiencies for oxygen evolution can be expected. Considering the low sulfide (i.e. w10 mg/L) and organics concentrations (i.e. 380 140 mg/L) during the experiments, this could explain the similar sulfide removal and excess oxygen production rates for all electrode materials. It should be noted that in this study a laboratory reactor was used rather than a real rising main system. Hence, our
Table 1 e Sulfide oxidation (n [ 3) from domestic wastewater using MMO coated titanium electrodes at a current density of 10 mA/cm2. Parameter a
Coulombic efficiency Removal efficiency Removal rate Removal rate Total S added (mg) Final sulfide conc. S0 produced S2O2 3 produced SO2 3 produced SO2 4 produced COD removed COD removal rate O2 produced Temperature Conductivity Chloride concentration pHb Average anode potential Average cell voltage
Unit
Ta/Ir
Ru/Ir
Pt/Ir
SnO2
PbO2
% % 1 g S m2 electrode surface h mg S L1 h1 mg mg/L mg mg mg mg mg mg COD h1 mg C mS/cm mg/L e V V
78 5 78 5 10.8 0.2 7.5 0.3 199 9 7.9 2.6 100 9 63 9 0.9 0.2 10 5 199 43 60 17 24.5 0.5 1.11 0.01 114 9 7.5 1.41 0.08 5.3 0.4
77 1 77 1 11.7 1.1 7.7 0.5 219 21 10.0 1.2 97 4 64 13 1.5 0.5 14 15 256 33 65 13 24.1 0.4 1.15 0.01 117 9 7.5 1.42 0.06 5.2 0.4
85 4 85 4 12.4 0.4 7.7 0.2 210 18 6.2 2.3 108 19 52 6 1.1 0.7 34 16 204 34 69 8 24.6 0.4 1.17 0.06 115 9 7.5 1.53 0.12 5.0 0.2
84 1 84 1 12.9 0.8 7.8 0.6 220 12 6.6 0.6 136 25 43 10 0.8 0.5 28 8 274 46 47 8 25.0 0.6 1.16 0.04 109 10 7.5 2.2 0.11 9.0 0.4
83 2 83 2 12.4 0.1 7.8 0.7 214 6 6.9 1.8 101 36 50 8 1.1 0.7 30 17 117 26 66 17 24.3 0.4 1.09 0.09 114 6 7.5 1.8 0.04 5.0 0.2
a Coulombic efficiency is based on the oxidation of sulfide to sulfate. b pH is controlled online.
5385
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 5 3 8 1 e5 3 8 8
system differed from a real sewer system as such that in our system (a) a headspace was present and (b) in situ generated oxygen was dissolved into only 5 L. Hence, we produced much more oxygen per unit wastewater. The above caused that oxygen was transferred to the headspace instead of remaining in the water phase. The anode potentials remained constant during the course of all the experiments. The anode potentials for Ta/Ir, Ru/Ir and Pt/Ir (low overpotential for oxygen evolution) were 1.41 0.08, 1.42 0.06 and 1.53 0.12 V vs. SHE, whereas the anode potentials for PbO2 and SnO2 were 1.80 0.04 and 2.22 0.11 V vs. SHE, respectively. The overall cell voltage for all electrode materials, except for SnO2 (9.0 0.4 V), was approximately 5 V (Table 1). As yet, we have no explanation for the unexpectedly high cell voltage for SnO2, the use of this electrode apparently caused a higher ohmic resistance or higher cathodic overpotential. The mechanism of the generation of oxygen active species is very complex and can involve the generation of several radical species containing oxygen and/or halogen atom (e.g. C C OC 2 , HO2 , HClO ) other than hydroxyl radicals which can occur at lower potentials. Hence, reactive oxygen species other than hydroxyl radicals might have been formed but did not have a significant impact on the sulfide removal process. Fig. 2 shows the electron distribution among the different electron sinks (i.e. sulfur, thiosulfate, sulfite, sulfate, excess oxygen and COD (i.e. organics)) during the oxidation of sulfide using the different electrode materials. A significant part of the charge supplied to the system was used for the oxidation of organics. The charge used for the oxidation of organics was 46%, 60%, 47%, 64% and 36% for Ta/Ir, Ru/Ir, Pt/Ir, PbO2 and SnO2, respectively. Concentrations of the different dissolved sulfur species (i.e. sulfide, sulfite, thiosulfate and sulfate) indicate that 43 2%,
39 7%, 47 8%, 42 11% and 57 10% of the total sulfide added was oxidized to elemental sulfur using Ta/Ir, Ru/Ir, Pt/ Ir, PbO2 and SnO2, respectively. This is further supported by the reasonably small gap in the electron balances of the experiments (see Fig. 2). No elemental sulfur was visually observed on the electrode surface. This suggests that the elemental sulfur was formed in the bulk by means of indirect oxidation with oxygen. In summary, the results highlight that under the applied operational conditions the sulfide removal process in terms of removal efficiency, excess oxygen generation, overall cell potential (except SnO2) and electron distribution is not significantly influenced by the electrode material used.
3.2. Influence of chloride concentration on sulfide oxidation The impact of elevated chloride concentration was investigated using Ta/Ir and Ru/Ir coated titanium electrodes; electrodes with a low and high reported catalytic activity toward chlorine generation, respectively. Table 2 shows that similar sulfide removal efficiencies and removal rates as in the experiments at low chloride concentrations were obtained (i.e. 7.7 0.5 versus 7.0 1.0 and 7.5 0.3 versus 7.7 0.3 mg S/L h). Fig. 3 shows the electron distribution among the different electron sinks (i.e. sulfur, thiosulfate, sulfite, sulfate, excess oxygen and organics) during the oxidation of sulfide using of Ta/Ir and Ru/Ir electrodes, respectively. Similar to the experiments at low chloride concentrations a large part of the charge supplied to the system was used for the oxidation of organics (i.e., 50 10, 50 8% for Ta/Ir and Ru/Ir, respectively). Ru/Ir coated titanium electrodes are well-known for their low overpotential for chlorine generation. Hence, at high chloride concentrations in situ chlorine generation may have
organics excess oxygen sulfate sulfite thiosulfate sulfur
electron distribution (%)
100
80
60
40
20
0 Ta/Ir
Ru/Ir
Pt/Ir
SnO2
PbO2
Fig. 2 e Electron distribution (%) among the different electron sinks (i.e. sulfur, thiosulfate, sulfite, sulfate, excess oxygen and organics) during the oxidation of sulfide using Ta/Ir, Ru/Ir, Pt/Ir, PbO2 and SnO2 electrodes at a current density of 10 mA/ cm2 (n [ 1).
5386
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 5 3 8 1 e5 3 8 8
h i JL ¼ nFkm Cl
(3)
where JL is the mass transfer limited current density (mA/ cm2), n is the number of electrons involved (i.e. 1 for the oxidation of chloride to chlorine), F the Faraday constant (96,485.3 C/mol), km the mass transport rate coefficient (m/s) and Cl the chloride concentration (mg/L). Assuming a km of 0.8 105 m/s (Szpyrkowicz et al., 2005) and an average chloride concentration of 114 mg/L (i.e. the chloride concentration present in the sewage wastewater used) the mass transfer limited current is only 0.25 mA/cm2. In addition, a migration current due to the migration in the chloride transport can be expected as result of the low conductivity of the wastewater (Bergmann and Koparal, 2005). However, at the applied current density (10 mA/cm2), the chloride concentration and the applied electrical field the relative importance of the migration current density is small. The applied current density in all experiments is 10 mA/cm2 and thus sulfide oxidation by means of in situ chlorine production is expected to be small. Hence, it is expected that at low chloride concentrations Ru/Ir coated titanium electrodes mainly will result in the in situ generation of oxygen and hence similar kinetics for the anodic sulfide oxidation are expected. However, at high chloride concentrations up to 1100 mg/L, which are often observed in sewers in many coastal areas,
Table 2 e Sulfide oxidation (n [ 3) from domestic wastewater at elevated chloride concentrations using Ta/ Ir and Ru/Ir coated titanium electrodes at a current density of 10 mA/cm2. Parameter Coulombic efficiencya Removal rate Removal rate Total S added (mg) Final sulfide conc. S0 produced S2O2 3 produced SO2 3 produced SO2 4 produced COD removed COD removal rate O2 produced Temperature Conductivity Chloride concentration pH Average anode potential Average cell voltage a n ¼ 2.
Unit %
Ta/Ir
Ru/Ir
87 4
86 5
1 g S m2 electrode surface h mg S L1 h1 mg mg/L mg mg mg mg mg mg COD h1 mg C mS/cm mg/L
12.9 1.0 12.2 1.2 7.7 0.3 7.0 1.0 213 8 203 8 5.4 1.3 5.7 1.8 139 0 103 20 29 14 33 5 0.1 0.1 0.3 0.1 20 12 39 14 213 16 216 33 36 6 37 7 77 19 65 10 24.0 0.5 24 0.4 3.73 0.05 3.73 0.05 1117 37 1119 25
e V
7.5 7.5 1.51 0.18a 1.43 0.03
V
4.21 0.9a
4.64 0.2
70
60
elecetron distribution (%)
an impact on the sulfide oxidation process and might also affect the removal of organics. The maximum chlorine generation under the mass transfer limiting conditions at a planar electrode (under optimum mixing conditions) can be described according to:
50
40
30
20
10
0 sulfur
thiosulfate
sulfite
sulfate
oxygen
COD
Fig. 3 e Electron distribution (in %) among the different electron sinks (i.e. sulfur, thiosulfate, sulfite, sulfate, excess oxygen and COD (i.e. organics)) during the oxidation of sulfide using Ta/Ir and Ru/Ir, electrodes at a current density of 10 mA/cm2 at high chloride concentrations (n [ 3).
chlorine generation can play a significant role in the anodic sulfide removal process. Depending on the pH, sulfide can be oxidized by chlorine either to elemental sulfur or sulfate. At pH values 7.5, sulfide is oxidized by chlorine to elemental sulfur (Chwirka and Satchell, 1990). Thus, under the conditions normally observed in sewer systems (i.e. pH 7.5) the addition of chlorine would result in the oxidation of sulfide (HS) to elemental sulfur. The amount of sulfide dosed in the experiments was equivalent to 2.5 mA/cm2 (when oxidized to sulfur), whereas the mass transport limited current for the generation of chlorine at chloride concentrations of 1100 mg/L is approximately 2.4 mA/cm2. Thus, under elevated chloride levels sufficient chlorine could have been produced to oxidize the sulfide added while this might not have been the case at low chloride concentrations. The results showed that at elevated chloride concentrations similar sulfide and organic removal rates were obtained (Table 2). Moreover, similar electron distribution among the electron sinks was observed. Hence, it appears that elevated chloride concentrations did not entail any significant differences in sulfide removal rate as well as in the organic removal. Analysis of the free chlorine concentration revealed that in all experiments no free chlorine or other reactive oxygen species (which are also detected with the used method) were present. It cannot be excluded that chlorine was formed and instantly reacted with the sulfide and/or organics present. This would be in line with several studies earlier reported (Chen, 2004; Bergmann and Koparal, 2005).
3.3.
Implications for practice
The results could indicate that sulfide removal by in situ generated oxygen was the predominant reaction mechanism, independent of the electrode material used. Other reaction mechanisms including the formation of radical species
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 5 3 8 1 e5 3 8 8
C containing oxygen and/or halogen atom (e.g. OC 2 , HO2 , HClOC) cannot be excluded but did not have a significant impact on the sulfide removal process. In perfectly mixed conditions, such as in a rotating disc setup, the different mechanisms for the electrodes would be observed to a higher extent. However, in the reactors here, which are more amenable to wastewater treatment, limitations exist on the mixing intensity. Oxygen injection is presently considered as an attractive option for sulfide abatement in sewer systems. It is less expensive than most other chemicals, and can target rising mains where the SRB activity is the highest (Hvitved-Jacobsen, 2001). However, transport and storage of pure oxygen carries serious safety issues and precise control of dosing is not straightforward. By generating oxygen in situ the requirement for transport and storage are avoided, thus mitigating safety concerns. Other advantages of in situ oxygen generation compared to traditional methods for oxygen supply are the fine dispersion, high controllability and the ease to monitor. The disadvantage is the cost of the oxygen per unit weight. All electrode materials performed similarly in terms of sulfide removal. Therefore, Ta/Ir, Ru/Ir and Pt/Ir coated titanium electrodes seem the most suitable electrodes since they posses the lowest overpotential for oxygen evolution and they are already used in full scale applications. However, the life time of Ru/Ir electrodes for the oxygen evolution reaction, which is the predominant reaction at low chloride concentrations, is low (Hine et al., 1979). Hence, Ta/Ir and Pt/Ir coated titanium electrodes appear the most suitable electrodes for sulfide oxidation from domestic wastewater in sewer systems. Based on a cell potential of 5 V, a Coulombic efficiency of 95% (for oxygen generation) and a cost of $0.06 per kWh, the estimated delivery cost is $1.06 per kg, relative to a delivery cost of $0.54e0.82 per kg for standard oxygen purchases (de Haas et al., 2008). However, standard oxygen injection in sewer systems has low efficiency (i.e. w20 to 40%) (de Haas et al., 2008) since oxygen is often dosed in an inefficient way (i.e., coarse bubbles) which results in a significant loss of undissolved gas from air in gas release valves downstream. The latter is avoided when oxygen is generated in situ due to the high transfer efficiency and fine dispersion of in situ generated oxygen. The deployment of a sulfide or dissolved oxygen sensor before or after the electrochemical system respectively would also allow further fine-tuning of the oxygen dosing. While entailing a higher cost, sustainable electricity solutions such as photovoltaic power would allow total independence of the dosing system of transport or utility requirements. An in-depth life cycle analysis would provide more insight into the sustainability of the existing dosing methods. In all experiments excess oxygen was produced, which in a practical situation would be used for chemical oxidation or by the biofilm in the sewer pipes for biological sulfide oxidation or the removal of organic matter. Nonetheless, in this study, the impact of electrode material and chloride concentration on the kinetics of sulfide oxidation from real domestic wastewater was successfully investigated. Over the course of the experiments, none of the electrode materials deteriorated or demonstrated changing potential over time. Earlier studies at lower current densities indicated
5387
electrode fouling with sulfur (Dutta et al., 2009; Ateya et al., 2003; Rabaey et al., 2006), it appears that at the higher current densities used here sulfur either flakes off or is further oxidized to its soluble forms. However, in realistic conditions of the sewer, other forms of fouling such as ragging, particle settling and scaling may occur. Long term trials on site are presently underway to investigate these possible impacts, which were not observed here. Scaling of the membrane and the electrode surface, on the other hand, was observed in the cathode chamber. This is caused by transport of bivalent cations such as calcium through the cation exchange membrane causing precipitation of inorganics such as calcium hydroxide or calcium carbonate (Jeremiasse et al., 2010). To overcome problems with scaling the cathode needs to be cleaned periodically. This can be done either by chemical cleaning (i.e. the addition of citric or hydrochloric acid) or by periodic switching of the polarity of the electrodes (anode becomes cathode/cathode becomes anode), the latter option is attractive from an operational perspective but does restrict the types of electrodes that can be used.
4.
Conclusions
In this study, we investigated the kinetics of sulfide oxidation from real domestic wastewater using five different types of mixed metal-coated titanium electrodes (Ta/Ir, Ru/Ir, Pt/Ir, SnO2 and PbO2) at a current density of 10 mA/cm2. The obtained sulfide removal efficiencies were not significantly influenced by the electrode material used. The results indicate that, independent of electrode material used, sulfide was removed by means of chemical oxidation with in situ generated oxygen. Ta/Ir and Pt/Ir appear the most suitable electrodes since they have a low overpotential for oxygen evolution and are known for their stability even at low chloride concentrations. The obtained sulfide removal rates were in the same order of chemical rates found under high oxygen concentrations whereas in all experiments excess oxygen was produced. Elevated chloride concentrations did not entail any significant difference in sulfide removal rate. Analysis revealed that no free chlorine was present even at high chloride concentrations.
Acknowledgments Ilje Pikaar, Rene´ Rozendal and Korneel Rabaey thank the University of Queensland for scholarship (University of Queensland Research Scholarship) and fellowship support (RR UQ Postdoctoral Fellowship). KR is supported by the Australian Research Council (APD, DP0879245). This work was funded by the Australian Research Council (ARC Linkage project: LP0882016 “Optimal Management of Corrosion and Odor Problems in Sewer Systems”). The authors also want to acknowledge Dr. Beatrice Keller-Lehmann and Ms. Susan Cooke for their helpful collaboration with the chemical analyses.
5388
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 5 3 8 1 e5 3 8 8
Appendix. Supplementary material Supplementary data associated with this article can be found, in the online version, at doi:10.1016/j.watres.2011.07.033.
references
Ateya, B.G., AlKharafi, F.M., Al-Azab, A.S., 2003. Electrodeposition of sulfur from sulfide contaminated brines. Electrochemical and Solid-State Letters 6 (9), C137eC140. Bergmann, M.E.H., Koparal, A.S., 2005. Studies on electrochemical disinfectant production using anodes containing RuO2. Journal of Applied Electrochemistry 35 (12), 1321e1329. Chen, G., 2004. Electrochemical technologies in wastewater treatment. Separation and Purification Technology 38 (1), 11e41. Chwirka, J.D., Satchell, T.T., 1990. A 1990 guide for treating hydrogen sulfide in sewers. Water Engineering & Management 137, 32e35. Comninellis, C., 1994. Electrocatalysis in the electrochemical conversion/combustion of organic pollutants for waste water treatment. Electrochimica Acta 39, 1857. de Haas, D.W., Corrie, S., O’Halloran, K., Keller, J., Yuan, Z., 2008. Odour control by chemical dosing: a review. Journal of the Australian Water Association 35 (02), 138e143. Dutta, P.K., Rabaey, K., Yuan, Z., Keller, J., 2008. Spontaneous electrochemical removal of aqueous sulfide. Water Research 42 (20), 4965e4975. Dutta, P.K., Rozendal, R.A., Yuan, Z., Rabaey, K., Keller, J., 2009. Electrochemical regeneration of sulfur loaded electrodes. Electrochemistry Communications 11 (7), 1437e1440. Feng, Y.J., Li, X.Y., 2003. Electro-catalytic oxidation of phenol on several metal-oxide electrodes in aqueous solution. Water Research 37 (10), 2399e2407. Hine, F., Yasuda, M., Noda, T., Yoshida, T., Okuda, J., 1979. Electrochemical behavior of the oxide-coated metal anodes. Journal of the Electrochemical Society 126 (9), 1439e1445. Hvitved-Jacobsen, T., 2001. Sewer processes: microbial and chemical process engineering of sewer networks. xi þ 237 pp. Jeremiasse, A.W., Hamelers, H.V.M., Buisman, C.J.N., 2010. Microbial electrolysis cell with a microbial biocathode. Bioelectrochemistry 78 (1), 39e43. Kaempfer, W., Berndt, M., 1998. Polymer modified mortar with high resistance to acid to corrosion by biogenic sulfuric acid. In: Proceedings of the IX ICPIC Congress, Bologna, Italy, pp. 681e687. Keller-Lehmann, B., Corrie, S., Ravn, R., Yuan, Z., Keller, J., 2006. Preservation and simultaneous analysis of relevant soluble sulfur species in sewage samples. In: 2nd International IWA Conference on Sewer Operation and Maintenance, Vienna, Austria.
Martinez-Huitle, C.A., Brillas, E., 2009. Decontamination of wastewaters containing synthetic organic dyes by electrochemical methods: a general review. Applied Catalysis B: Environmental 87 (3e4), 105e145. Panizza, M., Cerisola, G., 2008. Electrochemical degradation of methyl red using BDD and PbO2 anodes. Industrial and Engineering Chemistry Research 47 (18), 6816e6820. Panizza, M., Kapalka, A., Comninellis, C., 2008. Oxidation of organic pollutants on BDD anodes using modulated current electrolysis. Electrochimica Acta 53 (5), 2289e2295. Pikaar, I., Rozendal, R.A., Yuan, Z., Keller, J., Rabaey, K., 2011. Electrochemical sulfide removal from synthetic and real domestic wastewater at high current densities. Water Research 45 (6), 2281e2289. Rabaey, K., Van de Sompel, K., Maignien, L., Boon, N., Aelterman, P., Clauwaert, P., De Schamphelaire, L., Pham, H.T. , Vermeulen, J., Verhaege, M., Lens, P., Verstraete, W., 2006. Microbial fuel cells for sulfide removal. Environmental Science & Technology 40 (17), 5218e5224. Sharma, K., Yuan, Z., 2010. Kinetics of chemical sulfide oxidation under high dissolved oxygen levels. Submitted for Oral Presentation, 6th International Conference on Sewer Processes and Networks, 7e10 November 2010. Sun, Y.-X., Wu, Q.-Y., Hu, H.-Y., Tian, J., 2009. Effects of operating conditions on THMs and HAAs formation during wastewater chlorination. Journal of Hazardous Materials 168 (2e3), 1290e1295. Sydney, R., Esfandi, E., Surapaneni, S., 1996. Control concrete sewer corrosion via the crown spray process. Water Environment Research 68, 338e347. Szpyrkowicz, L., Kaul, S.N., Neti, R.N., Satyanarayan, S., 2005. Influence of anode material on electrochemical oxidation for the treatment of tannery wastewater. Water Research 39 (8), 1601e1613. Takasu, Y., Sugimoto, W., Nishiki, Y., Nakamatsu, S., 2010. Structural analyses of RuO2eTiO2/Ti and IrO2eRuO2eTiO2/Ti anodes used in industrial chlor-alkali membrane processes. Journal of Applied Electrochemistry, 1e7. Taylor, B., Gardner, T., 2007. Southeast Queensland recycled water aspects and soil impacts, Sunshine Coast, Australia. Vincke, E., Wanseele, E.V., Monteny, J., Beeldens, A., Belie, N.D., Taerwe, L., Gemert, D.V., Verstraete, W., 2002. Influence of polymer addition on biogenic sulfuric acid attack of concrete. International Biodeterioration & Biodegradation 49 (4), 283e292. Zhang, L., De Schryver, P., De Gusseme, B., De Muynck, W., Boon, N., Verstraete, W., 2008. Chemical and biological technologies for hydrogen sulfide emission control in sewer systems: a review. Water Research 42 (1e2), 1e12. Zhu, X., Tong, M., Shi, S., Zhao, H., Ni, J., 2008. Essential explanation of the strong mineralization performance of boron-doped diamond electrodes. Environmental Science and Technology 42 (13), 4914e4920.
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 5 3 8 9 e5 3 9 8
Available at www.sciencedirect.com
journal homepage: www.elsevier.com/locate/watres
Biological iron oxidation by Gallionella spp. in drinking water production under fully aerated conditions W.W.J.M. de Vet a,c,d,*, I.J.T. Dinkla b,1, L.C. Rietveld d, M.C.M. van Loosdrecht c,a a
Oasen Drinking Water Company, PO Box 122, 2800 AC Gouda, The Netherlands Bioclear bv. Rozenburglaan 13, 9727 DL Groningen, The Netherlands c Department of Biotechnology, Delft University of Technology, Julianalaan 67, 2628 BC Delft, The Netherlands d Department of Water Management, Delft University of Technology, Stevinweg 1, 2628 CN Delft, The Netherlands b
article info
abstract
Article history:
Iron oxidation under neutral conditions (pH 6.5e8) may be a homo- or heterogeneous
Received 7 December 2010
chemically- or a biologically-mediated process. The chemical oxidation is supposed to
Received in revised form
outpace the biological process under slightly alkaline conditions (pH 7e8). The iron
8 July 2011
oxidation kinetics and growth of Gallionella spp. e obligatory chemolithotrophic iron
Accepted 25 July 2011
oxidizers e were assessed in natural, organic carbon-containing water, in continuous lab-
Available online 29 July 2011
scale reactors and full-scale groundwater trickling filters in the Netherlands. From Gallionella cell numbers determined by qPCR, balances were made for all systems. The homo-
Keywords:
geneous chemical iron oxidation occurred in accordance with the literature, but was
qPCR
retarded by a low water temperature (13 C). The contribution of the heterogeneous
Gallionella spp.
chemical oxidation was, despite the presence of freshly formed iron oxyhydroxides, much
Groundwater trickling filtration
lower than in previous studies in ultrapure water. This could be caused by the adsorption
Biological and chemical
of natural organic matter (NOM) on the iron oxide surfaces. In the oxygen-saturated
iron oxidation
natural water with a pH ranging from 6.5 to 7.7, Gallionella spp. grew uninhibited and biological iron oxidation was an important, and probably the dominant, process. Gallionella growth was not even inhibited in a full-scale filter after plate aeration. From this we conclude that Gallionella spp. can grow under neutral pH and fully aerated conditions when the chemical iron oxidation is retarded by low water temperature and inhibition of the autocatalytic iron oxidation. ª 2011 Elsevier Ltd. All rights reserved.
1.
Introduction
The existence and relevance of iron-oxidizing bacteria (FeOB) in drinking water treatment has been well established from the very beginning of central water supply. Berger and Berger
(1928) mentioned that only five years after start-up, all Berlin Water Works were forced to switch from groundwater to surface water in 1882 due to a so-called ‘Eisenkalamita¨t’ (iron calamity). Biological essays demonstrated that Crenothrix polyspora and probably also Leptothrix ochracea caused
Abbreviations: FeOB, iron-oxidizing bacteria; NOM, natural organic matter; qPCR, (quantitative) real-time polymerase chain reaction; WTP, water treatment plant. * Corresponding author. Oasen Drinking Water Company, PO Box 122, 2800 AC Gouda, The Netherlands. Tel.: þ31 610927947; fax: þ31 152782355. E-mail addresses:
[email protected] (W.W.J.M. de Vet),
[email protected] (I.J.T. Dinkla),
[email protected] (L.C. Rietveld),
[email protected] (M.C.M. van Loosdrecht). 1 Tel.: þ31 505718455; fax: þ31 505717920. 0043-1354/$ e see front matter ª 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.watres.2011.07.028
5390
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 5 3 8 9 e5 3 9 8
pollution of the unfiltered, distributed water with ‘ocheryellow, dirty brownish up to coffee brown flocky deposits’ (Ibid.). Similar problems with iron-containing groundwater occurred in Rotterdam, The Netherlands (de Vries, 1890). Iron-containing groundwater could only be used for drinking water production once properly working de-ironing filters were developed. Since the establishment of de-ironing filters, there has been an ongoing discussion with regard to the importance of chemical versus biological iron oxidation (Czekalla et al., 1985; Sharma et al., 2005, and references therein). In drinking water filters in northern Germany, at least four species of FeOB have been reported (Gallionella sp., L. ochracea, Toxothrix trichogenes and an unknown bacterium), next to five species of manganese-oxidizing bacteria. From these observations it was concluded that iron and manganese removal was a bacterial process (Czekalla et al., 1985). Søgaard et al. (2000) studied precipitates from backwash sludge from three water treatment plants (WTP) in Denmark. They suggested low oxygen content of the raw water, poor aeration and relatively low pH as the determining prerequisites for biological iron oxidation; however, they did not provide consistent data from the WTPs to substantiate these presumptions. The presence of ferrous iron in combination with low dissolved oxygen and/ or slightly acidic pH is also regarded by other researchers as prerequisites for growth of FeOB (Hallbeck and Pedersen, 1990; Emerson and Floyd, 2005). The distinction between heterogeneous chemical and biological iron oxidation is, however, hard to make (Sharma et al., 2005; Tekerlekopoulou and Vayenas, 2008). Only in some cases, the distinguishable characteristic forms of iron deposits e like the twisted stalks formed by Gallionella spp. e indicate biological action, but in many other cases, particulate amorphous iron oxyhydroxides, very similar to chemical precipitates, are shown to be of biological origin as well (Emerson and Weiss, 2004). In recent studies the catalysis of iron oxidation by excreted RedOx-enzymes like flavins (Degre´mont, 2007) or exopolymers (Søgaard et al., 2000) has been reported, however this chemical process does not yield energy for bacterial growth. The chemo-lithotrophy of some FeOB is still under dispute (Spring and Ka¨mpfer, 2005). Gallionella spp. are, however, generally regarded as strictly lithotrophic, unable to catabolize organic matter (Lu¨tters-Czekalla, 1990), so the growth of Gallionella spp. can be seen as direct proof of biological iron oxidation. For this reason, this paper focuses on Gallionella spp., even though other FeOB such as Leptothrix spp. were found to be growing in the studied systems as well (data not shown). New molecular techniques provide powerful tools to assess and quantify the role of FeOB in full-scale treatment systems. For this paper, the kinetics of iron oxidation and the growth of the iron-oxidizing Gallionella bacteria were assessed in continuous lab- and full-scale reactors and trickling filters. The results of these studies were used to discuss the competition of biological iron oxidation with chemical iron oxidation at different pH’s in groundwater filtration. We hypothesize that Gallionella spp. can also grow under fully aerated and slightly alkaline pH conditions when chemical iron oxidation is retarded.
2.
Methods and materials
2.1.
Lab-scale experiments
The oxidation and removal of iron were investigated in two lab-scale setups at WTP Lekkerkerk of the Oasen drinking water company in The Netherlands. The lab-scale research consisted of oxidation column and filtration column experiments, which are described separately in the next two sections. Both experimental setups were fed with drinking water locally produced from riverbank groundwater. This water is moderately hard (Ca2þ w2 mM), well buffered (HCO 3 w3.0 mM), has a constant temperature of 13 C, a pH of 7.8 0.1. The dissolved oxygen content of the feed water was 9.9 0.6 mg L1 (74 data points in the period April 2008eAugust 2009 by potentiometric measurement in accordance with NEN-ISO 5814, NEN, 1993). Ferrous iron (FeSO47H20, Merck 103965 5000) was added to the feed water of all but the reference filter columns in a concentration of 3.3 mg L1 Fe, resembling the groundwater quality at WTP Lekkerkerk. A nutrient solution, containing phosphorus (0.6 mM PO4-P), nitrogen (3.8 mM NH4-N) and trace elements (Zn, Co, Cu and Mo), was added to prevent bacterial growth limitation. All water and chemical flows were controlled by tube pumps and all flow rates were checked weekly by mass measurements. All columns had an internal diameter of 0.089 m resulting in a water velocity of about 2.2 m h1, similar to the full-scale filters. While the desired flow rate was 14.0 L h1, the realized flow rates for the oxidation and filter columns were 13.9 0.4 and 13.5 0.8 L h1, respectively. The flow direction was upwards for the oxidation columns and downwards for the filter columns. Additional information on the columns’ setup, including schemes and pictures, is given in the Supplementary Material A.
2.1.1.
Oxidation columns’ setup
The chemical iron oxidation strongly depends on pH (Sung, 1980; Tamura et al., 1976), it is therefore supposed that, with a decreasing pH, biological oxidation might outcompete chemical oxidation processes. In order to determine which rates of chemical oxidation still allow simultaneous biological oxidation by Gallionella spp., different pH conditions allowing different rates of chemical oxidation were applied. The influence of feed water pH on the oxidation rate of the ferrous iron and the growth of Gallionella spp. in the natural water of WTP Lekkerkerk was studied in six oxidation columns. With an overflow level of 0.58 m, the residence time calculated from mass balances was 16 min. Determination of the residence time by NaCl spiking and corresponding conductivity measurements (not presented) showed no short-circuiting, indicating mainly plug flow conditions in the oxidation columns. To set the pH, HCl or NaOH was added to the column influent. The required doses of HCl and NaOH were determined in triplicate by titration of the feeding drinking water and checked by offline potentiometric pH measurement in conformity with the Dutch NEN-ISO 10523 protocol (NEN, 2009) of the columns’ influent, also in triplicate (Figure B.1 of Supplementary Material B). Although the titrations and control measurements were executed at 20 C, operational pH
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 5 3 8 9 e5 3 9 8
values at 13 C will not have differed much because of the good buffering of the feed water. The deviation of the control measurements at the more extreme pH values was probably caused by gas exchange during sampling and offline measurements and calcium carbonate precipitation during storage. As the values determined by the titrations best represented the actual system during the experiments, these pH values will be used in the results section, with the uncertainty range calculated from mass measurements of the acid and base dosing. The realized iron dose determined by mass balances and total iron analysis (see Section ‘Iron analyses’) was 3.3 0.7 mg L1 Fe and is shown per column in Figure B.2 of the Supplementary Material B. In total, 58 4 g Fe was dosed over 7 weeks. The oxidation columns were run for seven weeks from July 8 to August 26, 2009. The concentrations of ferrous and ferric iron in the column effluents were determined weekly. The Gallionella spp. cell numbers were determined after three and seven weeks in the column influents and effluents and after seven weeks in the accumulated sludge in the oxidation columns. A picture of the oxidation columns’ setup at the end of the experiment is given as Figure A.2 in the Supplementary Material A.
2.1.2.
Filter columns’ setup
To model groundwater trickling filtration, iron removal was studied in a lab-scale filter columns’ setup, consisting of seven trickling filter columns in duplicate, in total 14 columns. All columns were filled with standard filter sand (1.7e2.5 mm). The pH of the feed water was lowered by HCl to resemble the groundwater before filtration (7.25 0.15). One column was used as a reference with no iron removal. In three columns, ferrous iron was dosed just before the filter top; in three other columns, ferrous iron passed a pre-oxidation column before filtration. These columns are referred to as “without preoxidation” and “with pre-oxidation", respectively. All trickling filter columns had forced ventilation which raised the pH in the filter effluents to 7.67 0.07. Each filter column was automatically backwashed every 24 h with a fixed volume of 30 L drinking water and under expansion (fluidization) of the filter bed. The filtration columns were run for six months from April 2 to October 8, 2008. The Gallionella spp. cell numbers were determined after six months in the influents and effluents, the backwash water and the filter material of seven columns (one of each duplicate).
2.2.
Full-scale groundwater trickling filters
The growth of Gallionella spp. and their role in iron oxidation was verified in three full-scale trickling filters at two Oasen WTPs. All three full-scale filters treated moderately hard and well-buffered anoxic groundwater. The filters were backwashed automatically after a filter runtime of 48 h, to prevent clogging by removal of inorganic precipitates and excess biomass. At WTP Lekkerkerk, the filter material of a trickling filter was externally washed and the filter performance and growth of Gallionella spp. were monitored extensively for nine months after restart of the filters from December 12, 2007 to September 19, 2008. At WTP De Hooge Boom, the growth of Gallionella spp. in two groundwater trickling filters was
5391
assessed in a quick scan on March 1, 2010. In one of these filters, the anoxic groundwater was sprayed directly on the trickling filter, while in the other the groundwater was intensively aerated on a plate aerator prior to spraying on top of the trickling filter. Both plate aeration and trickling filtration raised the dissolved oxygen content to nearly a saturated level and the pH by the stripping of carbon dioxide (see Table 1). All iron present in the groundwater was virtually completely removed in the filters (>95%). Table 1 gives an overview of the groundwater and filtrate qualities as well as the characteristics of the three studied filters.
2.3.
DNA extraction
Samples for detection and identification of Gallionella spp. were taken in sterilized glass bottles at different points in the Oasen WTP Lekkerkerk. All samples were stored at 4 C. Gallionella spp. cell numbers were determined by qPCR. Groundwater, influent and effluent water and backwash water samples were filtered over 0.2 mm polycarbonate membranes to concentrate the cells prior to DNA extraction. DNA was extracted from a volume of 100e150 ml water per sample. The filter was subsequently subjected to DNA extraction by bead beating in a Fast DNA spin kit for soil (MP Biomedicals, Soton, Ohio, United States). The DNA was purified using a silica-based column and eluted in 100 ml TE. DNA from approximately 10 g of filter sand was extracted as described by de Vet et al. (2009). In all cases, an internal control was used to determine the extraction efficiency.
2.4.
Quantification of Gallionella spp.
In order to quantify the number of Gallionella spp. cells in the systems, a specific PCR was developed to detect these bacteria, including the Gallionella spp. sequence that was previously found in the drinking water filters (Ibid.). PCR primers were developed for the detection of the 16S rRNA gene from Gallionella spp. using ARB software. One forward primer, GALFER0218-F 50 -GCTTTCGGAGTGGCCGATA-30 -, and one reverse primer, GALFER1408-R 50 - CAGATTCCACTCCCATGGTG -30 were designed. Amplification was performed by initial denaturation for 3 min at 94 C, followed by 35 cycles of amplification (30 s denaturation at 94 C; 30 s annealing at 62 C; 1 min elongation at 72 C), and 5 min at 72 C to complete elongation. Quantification was based on a comparison of the sample Ct value to the Ct value of a calibration curve using standard numbers of 16S rDNA fragments of Gallionella (see Supplementary Material C). It was assumed that Gallionella cells contain 1 16S gene copy per cell. An internal control was added to all samples to correct for the efficiency of the PCR reaction. The specificity of the qPCR method was checked through the construction and sequencing of clone libraries of the PCR products from filtrate and backwash water samples of the WTP Lekkerkerk full-scale filter (de Vet, 2011). From the qPCR enumeration results, balances for the labscale columns and the full-scale filters were calculated. To assess the role of biological iron oxidation, the following assumption was made: when the cell numbers entering and leaving the filters are constant in time, no net accumulation occurs and the net washout measured by qPCR balances the
5392
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 5 3 8 9 e5 3 9 8
Table 1 e Groundwater and filtrate quality and filter characteristics of the full-scale trickling filters at the Oasen WTPs. WTP Lekkerkerk
WTP De Hooge Boom
Direct trickling filtration Filter sand d10-d90, mm Filter bed area, m2 Average production flow, m3 h1 Filter runtime, h Groundwater quality Temperature, C pH, 1 HCO 3 , mg L Total iron, mg L1 TOC, mg L1 C Effluent plate aerator pH, O2, mg L1 Filtrate quality pH, O2, mg L1 Total iron removed per filter runtime, kg Fe
Direct trickling filtration
1.7e2.5 18.0 37 48 Average St. Dev. Jan.eSept. 2008 11.6 0.3 7.33 0.04 216 7 5.5 0.6 2.2 0.1
Plate aeration and trickling filtration
2.0e3.15 28.0 64 48 Average St. Dev. Jan. 2008eMar. 2010 11.5 0.2 7.10 0.05 387 12 8.5 0.8 8.3 0.2
e e
7.7a 10.1a
e e 7.8a 9.7a
7.69 0.10 9.7 0.6 (10 data points) 9.6 1.1
7.80 0.03 10.2 0.4 (3 data points) 26.1 2.3
a Indicative local measurement on March 1, 2010.
growth of Gallionella spp., during one filter run. For every water flow entering or leaving a system, the totalized values for the cell numbers were calculated by multiplying the measured concentration with the flow rate and duration of the phase. For the full-scale filter at WTP Lekkerkerk, the groundwater was sampled in duplicate; the filtrate in duplicate five and nine months after external washing with four samples per filter runtime of 48 h; the backwash water was sampled three, six and nine months after external washing; control backwash samples were taken in quintuplet 15e16 months after external washing. At WTP De Hooge Boom, the influent and backwash water of each filter and the effluent of the plate aerator were sampled once; the filtrate water of each filter was sampled twice, at the beginning and at the end of the filter runtime of 48 h. The filtrate of the filter columns was sampled two hours after backwash. Filter column sand samples were taken from the top half of the bed during expansion backwashing.
2.5.
Iron analyses
Samples for iron analysis were taken directly into acid containing bottles to set the pH below 2. Nitric and hydrochloric acid were used to stabilize the samples for total and ferrous iron analysis, respectively. All samples were stored cool and analyzed within 24 h after sampling. Total iron in water samples was determined by inductively coupled plasma mass spectrometry (ICP-MS). Ferrous iron was determined by the 1,10-phenanthroline method according to the Dutch NEN 6482 protocol, based on Standard methods (1975). The iron concentration in the sludge was measured by atomic emission spectroscopy after sample destruction in a microwave. The mass of the filter coating was determined by measurements of the dry mass before and after acidification with 4 M
hydrochloric acid and oxalic acid. The iron concentration in the decanted acid solution was measured by ICP-MS.
3.
Results
3.1. pH effect on growth of Gallionella and iron oxidation rate in oxidation columns The pH dependency of Gallionella growth and iron oxidation was examined in six lab-scale oxidation columns. During the first week after start-up of the experiment, the degree of oxidation was lower than the average for the following six weeks for all oxidation columns except the one with pH 8.25 (Fig. 1). Apart from the first week, the oxidation degree of iron in the columns’ effluent e calculated from the ferrous iron concentrations in filter effluent (by 1,10-phenanthroline method; Figure B.3 of Supplementary Material B) and added iron concentrations (from mass balances) e was constant in time. During the first week after start-up, virtually no iron oxyhydroxides or FeOB had formed in the columns yet, and the measured iron oxidation was accounted for mainly by the homogeneous chemical process. After the start-up period, both iron oxyhydroxides and FeOB accumulated in the columns and influenced the oxidation kinetics. The numbers of Gallionella cells determined by qPCR in the influents and effluents of the columns after 3 and 7 weeks and the total iron and Gallionella cell numbers accumulated as sludge in the oxidation columns after 7 weeks are shown in Fig. 2. The cell concentrations in the influent, effluent and sludge show that Gallionella grew and accumulated in all oxidation columns. Statistical analysis (Supplementary Material D) shows that Gallionella grew equally fast in all columns with a pH between 7.0 and 7.73, and slightly (but not significantly)
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 5 3 8 9 e5 3 9 8
5393
present in the effluent of the columns in dissolved ferrous or colloidal ferric form. Growth of Gallionella spp. was confirmed based on the morphology of the deposits. Fig. 3 shows a phase contrast picture of deposits in the oxidation column with pH 7.73. In the oxidation column experiment, the specific growth rate m of Gallionella spp. can be approached by Equation (1). ln ðDtY0Þm ¼
Fig. 1 e Oxidation degree of iron after 16 min’ passage through oxidation columns; green solid bars, averages and standard deviations for 6e7 data points for whole period except first week after start-up, calculated from the ferrous iron concentrations in filter effluent (by 1,10phenanthroline method) and total iron concentrations (from mass balances); blue striped bars, oxidation degree of iron during first week after start-up; calculated with [OHL]0.6, calculated with [OHL]2 (see Discussion section for explanation). (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
faster at pH 6.5. The increasing rate of chemical iron oxidation did not inhibit the growth of Gallionella up to a pH of 7.73. Only in the column with pH 8.25 was Gallionella growth significantly slower. The total iron accumulated in the oxidation columns had a maximum at pH 7.00, and was for all columns between 1.8 and 3.6 g (3e6 102 mol). This was between 3 and 6% of the iron loading. The majority of the iron, therefore, was
Fig. 2 e Gallionella spp. concentrations (left axis, in cells mLL1) in the influent and the effluents of the oxidation columns operated at different pH values after 3 weeks (yellow solid bars) and after 7 weeks (green striped bars); total Gallionella spp. numbers (red dotted bars on left axis, in cells) and total iron (white bars on right axis, in g Fe) accumulated in the oxidation column after 7 weeks; error bars show uncertainty of qPCR method (between 0.5*N and 2*N). (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
Xt;column þ Q Dt xt;effluent Xt;column Dt
(1)
where, Xt,column ¼ total cells in column (cells), Q ¼ flow rate (m3 h1), [x]t,effluent ¼ cell concentration washed out of column (cells m3), t ¼ experimental time (h). At the end of the oxidation column experiment, m was 0.08 0.06 h1, corresponding to a doubling time of 8.4 h on average. Hallbeck and Pedersen (1990) found a generation time of 8.3 h in vitro at the optimal temperature of 20 C. This is comparable to the growth rate observed in our experiments at 13 C, which suggests slightly more favorable growth conditions in situ.
3.2. Effect of pre-oxidation on Gallionella growth in trickling filters The effect of pre-oxidation on the number of Gallionella spp. in trickling filters was studied in the combined oxidation and filtration column experiment. The pre-oxidation caused an oxidation degree of 29 6% before the water entered the trickling filters. At the applied pH (7.25 0.15) this corresponded with the oxidation degree measured in the oxidation column experiments (Fig. 1). The total numbers of Gallionella spp. for the water flows cumulated over one filter runtime (24 h) and for the filter beds at the end of the test period of 6 months are shown in Fig. 4. This figure clearly shows the growth of Gallionella spp. in all filter columns spiked with ferrous iron, but none in the reference column. No significant difference was found in the water samples from filter columns without and with pre-oxidation and only marginally more
Fig. 3 e Phase contrast microscopic picture (4003 magnification) of a sludge sample from the bottom of the oxidation column with pH set to 7.73 after three weeks of operation.
5394
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 5 3 8 9 e5 3 9 8
Fig. 4 e Iron deposition in filter coating and cumulative numbers of Gallionella spp. by qPCR determined for a reference column, three columns without and three with pre-oxidation in the filter columns’ setup at the end of 6-months’ trial; numbers in water flows (solid and striped bars) cumulated over one filter run of 24 h, yellow solid bar in feed water, green vertically striped bar in filtrate water; and blue horizontally striped bars backwash water; total present in filter material (dotted bars); iron deposition in filter coating ( after 91 days, after 187 days, on the right axis); error bars show uncertainty of qPCR method for the outflow measurements of the reference and standard deviation for the other measurements. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
Gallionella spp. in the filter samples from filter columns without pre-oxidation (shown in Supplementary Material D). Fig. 4 also shows iron deposition in the filter coating. After 91 days, the amount of iron deposited in the filter coatings was comparable regardless of pre-oxidation or not, and on average 60% of the loaded iron (Supplemental Material C) was encapsulated in the filter coating. After 187 days, however, the amount of iron in the filter coatings of columns with pre-oxidation had not increased, while it had in the columns without preoxidation. This suggests that the growth of attached FeOB in the filters may enhance the formation of iron coating. In the filter column experiments, the absence of preoxidation resulted, on average, in slightly higher cell numbers attached to the filter material, but due to error margins it is not possible to judge if this is significant. Although this is consistent with the higher ferrous iron loading in the columns without pre-oxidation, there was no significant difference in Gallionella spp. numbers in the water flows from columns with and without pre-oxidation. With the approach according to Equation (1), m was calculated as 0.01 0.005 and 0.03 0.01 h1 for the filter columns without and with pre-oxidation, respectively (with equal distribution of the cells washed out during backwash periods over the filter runtime). This suggests that the Gallionella cells in the columns without pre-oxidation were better attached, more encapsulated in the iron oxyhydroxide filter coating, and less active than in the columns with pre-oxidation.
3.3.
Full-scale groundwater trickling filters
In order to determine the potential role of biological oxidation in the groundwater trickling filters, the abundance and growth of the iron-oxidizing Gallionella species were assessed in the three full-scale filters by qPCR. The balances for Gallionella spp.
in duplicate calculated over one filter run of 48 h from the qPCR cell numbers, water flows, and time are shown in Fig. 5. Repeated measurements over the trial period of nine months at WTP Lekkerkerk (see Supplemental Material C) showed no trend in cell numbers, indicating a stable population. The measurements show that significant numbers of Gallionella cells were found in all three full-scale filters despite the fact that these filters were very well aerated and the oxygen content of the filtrate water was close to saturation level. This condition is usually associated with chemical iron oxidation (Sharma et al., 2005). The cell numbers leaving the filter through the filtrate and backwash water were much higher than in the groundwater feeding the filter. This indicates a strong growth of Gallionella spp. in these full-scale trickling filters. The plate aeration prior to the filtration at WTP De Hooge Boom did not inhibit the growth of Gallionella spp. in the filter, despite the oxygen saturation and elevated pH of the effluent water. Gallionella spp. started to grow in the plate aerator, were filtered off in the trickling filter and continued growing there.
4.
Discussion
4.1. Chemical versus biological iron oxidation in groundwater filtration Iron oxidation under aerobic, neutral pH conditions may be homogeneous, heterogeneous or biologically mediated. At the start of the oxidation column experiment, a negligible amount of iron oxyhydroxides and FeOB was present, and the iron oxidation was predominantly homogeneous. The general kinetic equation for homogeneous iron oxidation is given by Equation (2): n m (2) dFe=dt ¼ k OH PO2 Fe2þ
5395
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 5 3 8 9 e5 3 9 8
Fig. 5 e Gallionella spp. balances for the WTP Lekkerkerk trickling filter (multiple measurements, graph A, red solid bars) and for the WTP De Hooge Boom trickling filters (singular measurements, graph B, green square bars, trickling filtration after plate aeration; blue line bars, direct trickling filtration); cumulative cell numbers inoculated directly from groundwater or via plate aerator, washed out to effluent and to backwash are calculated for one filter runtime of 48 h; Influent filter values are for effluent plate aerator if present and equal to groundwater for the other filters; error bars show standard deviation (graph A) and uncertainty of qPCR method (graph B, between 0.5*N en 2*N). (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
where m ¼ 1 and n ¼ 2, as found by Sung (1980); only for the first week measurements did the model reasonably match the measured data (blue squares in Fig. 1; 13 C, ionic strength 0.01 M) with a rate constant of k ¼ 4$1012 M2 atm1 min1. This is approximately 10 times lower than the rate reported by Sung (Ibid.) in water with similar salinity but at 25 C. The temperature difference between 25 C and 13 C, explains this difference for the larger part. The lower temperature reduces the rate constant by a factor of 7, not because of changes in the activation energy (almost zero), but by the decline in Kw and thus [OH] activity (Stumm and Lee, 1961). This strongly indicates that the iron oxidation was mostly a homogeneous chemical reaction in the first week of the oxidation column experiment. The measurements during the rest of the experimental period can only be fitted to the model by reducing the order of [OH], i.e. n, to 0.6 (green triangles in Fig. 1). Tamura et al. (1976) found that rate of heterogeneous chemical iron oxygenation was proportional to the first order of the reciprocal [Hþ]. The general kinetic equation for heterogeneous chemical iron oxidation is given by Equation (3): dFe= ¼ k1 þ k2 Fe3þ Fe2þ dt
(3)
where 2 k1 ¼ khom OH PO2 Homogeneous Oxidation rate Constant; (3a)
1 Heterogenous oxidation rate constant; k2 ¼ ks;O ½O2 K Hþ (3b) 1
ks;O ¼ 4380 M1 min K ¼ 104:85
Surface rate ;
Adsorption constant of Fe2þ on FeOOH;
(3c) (3d)
The iron sludge that accumulated at the end of the experimental period of seven weeks in the oxidation columns was assumed to be equally distributed in the oxidation columns, leading to a ferric iron concentration of 9e18 mM (see Fig. 2; sludge volume per column was 4.0 0.1 L). Under these conditions, the heterogeneous oxidation rate constant k2 Fe3þ according to Equation (3) would be in the range 1e35 min1 (for pH 6.5 up to 8.25, respectively). As this means an oxidation half-life (t½) of less than 1 min, it would implicate a nearly complete chemical oxidation of iron after the average residence time of 16 min in all the oxidation columns, which was not the case in our experiments. The reason for this reduced heterogeneous oxidation rate cannot be deduced from our experiments, but is probably related to the composition of the natural water. In many studies on chemical iron oxidation (Stumm and Lee, 1961; Sung, 1980; Tamura et al., 1976) the oxidation rates were determined with ultrapure water. Some studies determined the effects of natural organic matter (NOM) on the chemical iron oxidation. Davison and Seed (1983) and Liang et al. (1993)
5396
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 5 3 8 9 e5 3 9 8
found that the rate constant for homogeneous iron oxidation in natural freshwater under oxygen-saturated conditions was comparable to the one in synthetic water, as we did. Other researchers found a significant effect of NOM complexation on iron removal and oxidation, but that effect could be either accelerating (Ninh Pham et al., 2004) or inhibiting (Theis and Singer, 1974). All these studies were confined to homogeneous chemical iron oxidation. Sung (1980) stated that catalytic iron oxidation was only noticeable at pH 7 and above because of the slow surface formation at a lower pH. Our research with natural water indicates that the heterogeneous iron oxidation rate was strongly reduced even at a higher pH. This reduced heterogeneous chemical iron oxidation rate may be caused by surface complexation of inorganic and (natural) organic compounds in water. Complexation of inorganic ions had little influence on the adsorption capacity of ferrous iron (Sharma, 2001). Tipping (1981) showed that the surface charge and adsorption capacity of iron oxyhydroxides could be influenced by the complexation of humic substances. Gallionella growth may have contributed to the surface complexation and stabilization of the iron (oxy)hydroxides by the excretion of polysaccharides, comparable to the stalk formation described by Chan et al. (2011). Analysis of the growth of Gallionella spp. by qPCR demonstrates the significance of bacterial iron oxidation in the fullscale filter and laboratory filter columns. The direct enumeration of Gallionella spp. by this method combined with the biomass yield on iron oxidation makes it possible to quantify the share of biological iron oxidation. The maximum biomass yield reported in the literature is low (0.006 g DW g1 Fe oxidized (Lu¨tters and Hanert, 1989) and 0.013 g DW g1 Fe oxidized (Neubauer et al., 2002). Thermodynamically, a maximum theoretical yield of 0.012 g DW g1 Fe can be expected, based on the anabolic reaction energy of 3500 kJ C mol1 biomass (Heijnen and Van Dijken, 1992) and the catabolic reaction energy (Hanselmann, 1991): Fe2þ þ ¼ O2 þ 1½ H2O / FeOOH þ 2Hþ with ΔGr ¼ 83.8 kJ mol1 Fe at pH 7.73 and 1 mM Fe2þ. During one filter runtime, 0.9 g of iron was removed in the laboratory filter system (Supplementary Material B) and 9.6 1.1 kg in the full-scale filter at WTP Lekkerkerk. During one runtime, in total 4.7 2.9 1011 and 3.4 2.9 1015 Gallionella cells were washed out of these filters, respectively. When biomass accumulation, maintenance and decay were not considered, the observed yield was 5.1 3.3 1011 and 3.6 3.4 1011 Gallionella cells g1 Fe oxidized, respectively.
This assumes that the iron oxidation was completely biological and exclusively by Gallionella spp.. These maximum Gallionella cell yields can be related to dry weight (DW) by using the cell dimensions (mean volume of 0.4 mm3) determined by Hallbeck and Pedersen (1991). With a specific cell DW of 1.2 1013 g, the yield equals 0.062 0.040 and 0.043 0.041 g DW g1 Fe oxidized, for the filter columns and the full-scale filter, respectively (Table 2). Although the standard deviations are large, the average yield was higher than reported in the literature and the theoretical maximum. The high cell yields found suggest that biological iron oxidation by Gallionella spp. played a dominant role in both the full-scale filter and in the filter columns.
4.2.
Growth conditions of Gallionella spp.
The results reported in this manuscript show that Gallionella spp. may grow under broader conditions than generally assumed. No growth inhibition was found in the natural water under fully aerated conditions and at a pH ranging from 6.5 to 7.73. This finding contrasts with the general perception of Gallionella being strictly microaerophilic (Emerson, 2000). The oxygen concentration at the microsites where Gallionella grew in our experiments may have been reduced compared to the bulk water. We did not measure the oxygen concentration in those microsites. When iron oxidation is the dominant oxygen-consuming process, however, only a limited reduction in oxygen concentration may be expected from the bulk water into microsites, based on the stoichiometry of the iron oxidation, the measured oxygen and ferrous iron concentrations in the bulk water and the diffusion coefficients for both compounds from the literature (Broecker and Peng, 1974; Li and Gregory, 1974). According to Degre´mont (2007), biological iron oxidation will only prevail under conditions where physico-chemical iron oxidation is not possible: oxygen concentration between 0.2 and 0.5 mg L1, pH 6.3, oxidation reduction potential þ100 mV and rH2 between 14 and 20 (whereas rH2 ¼ log ( pH2) ¼ Eh/0.0296 V þ 2 pH). Under rH2 of 14, the biological oxidation should be inhibited, while over 20, the bacteria would lose the competition with the physico-chemical iron precipitation. At pH 7.73, the upper limit of rH2 indicates a maximum redox potential of 135 mV and an oxidation degree of less than 98%. It was stated that the boundaries are not strictly defined and can shift e.g. by chelation. Hanert (2006) listed the broad array of the environments where Gallionella spp. have been
Table 2 e Overview of iron conversion, net Gallionella cells washout and calculated yield for filter column experiment and full-scale trickling filter at WTP Lekkerkerk. Parameters Iron removed per filter runtime Gallionella cells washed out per filter runtime Cell yield Biomass yielda
Unit g Fe cells cells g1 Fe g DW g1 Fe
Filter columns
Full-scale trickling filter
0.92 0.04 4.7 2.9 1011 5.1 3.3 1011 0.062 0.040
9.6 1.1 103 3.4 2.9 1015 3.6 3.4 1011 0.043 0.041
a Calculated with 1.2 1013 g DW cell1, mean cell volume 0.4 mm3, Hallbeck and Pedersen (1991).
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 5 3 8 9 e5 3 9 8
found and concluded that the stability of ferrous iron in combination with oxygen is crucial for their existence, more than mere pH or Eh. This paper substantiates this claim by showing the growth of Gallionella spp. on ferrous iron under fully aerated and slightly alkaline circumstances, when the chemical iron oxidation is slow. In the oxidation column experiment with a fixed pH ranging from 6.5 to 7.73 and oxygen-saturated natural water, the initial iron oxidation was homogeneous with rates consistent with the literature. After the start-up period, FeOB and iron oxyhydroxydes accumulated in the columns but the oxidation rate increased less than theoretically expected from heterogeneous chemical oxidation. Heterogeneous chemical iron oxidation may be seriously hampered in natural water compared to synthetic water by complexation of natural organic matter on iron oxyhydroxide surfaces. The specific growth of Gallionella spp. was in accordance with the values found in culture experiments. The comparable Gallionella cell growth and the increase in iron oxidation degree indicate that, for pH ranging from 6.5 to 7.73, the increased iron oxidation rate had to be attributed to the growth and activity of FeOB, rather than to chemical catalysis. Yield calculations for the biological iron oxidation by Gallionella spp. in lab- and fullscale trickling filters, indicate that the dominant iron oxidation mechanism in groundwater filtration is biological under wider process conditions (pH and oxygen content) than previously thought.
5.
Conclusions
The quantitative PCR approach targeting the 16S rRNA of Gallionella spp. was successfully used to determine the significance of biological versus chemical oxidation in full-scale groundwater trickling filters and lab-scale column experiments. Gallionella spp. grew in fully aerated full-scale groundwater trickling filters and lab-scale oxidation columns and trickling filters at neutral pH (up to pH 7.7) and at a moderate temperature of 13 C. Biological oxidation by Gallionella spp. was the dominant process for iron oxidation in this type of groundwater, and heterogeneous chemical iron oxidation in natural water was substantially reduced, compared to experimental results from the literature for synthetic water.
Acknowledgements The authors gratefully acknowledge the contribution of Peter Dijkstra for assistance in the full-scale research, Petra Lafeber for support in the column experiments and Sabine Doddema and Paul van der Wielen for the DNA-isolation for the qPCR.
Appendix. Supplementary data Supplementary data related to this article can be found online at doi:10.1016/j.watres.2011.07.028.
5397
references
Berger, H., Berger, E., 1928. Biologie der Trink- und Brauchwasseranlagen. Verlag von Gustav Fischer, Jena. BerlinDahlem, Germany. Broecker, W.S., Peng, T.H., 1974. Gas exchange rates between air and sea. Tellus 26 (1e2), 21e35. Chan, C.S., Fakra, S.C., Emerson, D., Fleming, E.J., Edwards, K.J., 2011. Lithotrophic iron-oxidizing bacteria produce organic stalks to control mineral growth: implications for biosignature formation. ISME Journal 5 (4), 717e727. Czekalla, C., Mevius, W., Hanert, H., 1985. Quantitative removal of iron and manganese by microorganisms in rapid sand filters (in situ investigations). Water Supply 3 (1), 111e123. Davison, W., Seed, G., 1983. The kinetics of the oxidation of ferrous iron in synthetic and natural waters. Geochimica et Cosmochimica Acta 47 (1), 67e79. Degre´mont (Ed.), 2007. Water Treatment Handbook. Lavoisier SAS. Emerson, D., Floyd, M.M., 2005. Enrichment and isolation of ironoxidizing bacteria at neutral pH. Methods in Enzymology 397, 112e123. Emerson, D., Weiss, J.V., 2004. Bacterial iron oxidation in circumneutral freshwater habitats: findings from the field and the laboratory. Geomicrobiology Journal 21 (6), 405e414. Emerson, D., 2000. Environmental microbe-metal interactions. In: Lovley, D.R. (Ed.). ASM, Washington, pp. 31e52. Hallbeck, L., Pedersen, K., 1990. Culture parameters regulating stalk formation and growth rate of Gallionella ferruginea. Journal of General Microbiology 136 (9), 1675e1680. Hallbeck, L., Pedersen, K., 1991. Autotrophic and mixotrophic growth of Gallionella ferruginea. Journal of General Microbiology 137 (11), 2657e2661. Hanert, H.H., 2006. The Prokaryotes: a Handbook on the Biology of Bacteria. Springer, pp. 990e995. Hanselmann, K.W., 1991. Microbial energetics applied to waste repositories. Experientia 47 (7), 645e687. Heijnen, J.J., Van Dijken, J.P., 1992. In search of thermodynamic description of biomass yields for the chemotrophic growth of microorganisms. Biotechnology and Bioengineering 39 (8), 833e858. Li, Y.-H., Gregory, S., 1974. Diffusion of ions in sea water and in deep-sea sediments. Geochimica et Cosmochimica Acta 38 (5), 703e714. Liang, L., Andrew McNabb, J., Paulk, J.M., Gu, B., McCarthy, J.F., 1993. Kinetics off Fe(II) oxygenation at low partial pressure of oxygen in the presence of natural organic matter. Environmental Science and Technology 27 (9), 1864e1870. Lu¨tters, S., Hanert, H.H., 1989. The ultrastructure of chemolithoautotrophic Gallionella ferruginea and Thiobacillus ferrooxidans as revealed by chemical fixation and freezeetching. Archives of Microbiology 151 (3), 245e251. Lu¨tters-Czekalla, S., 1990. Lithoautotrophic growth of the iron bacterium Gallionella ferruginea with thiosulfate or sulfide as energy source. Archives of Microbiology 154 (5), 417e421. Nederlands Normalisatie instituut (NEN), 1993. NEN-ISO 5814: 1993 en, Water e Bepaling van het gehalte aan opgeloste zuurstof e Elektrochemische methode. http://www.nen.nl/ web/Normshop/Norm/NENISO-58141993-en.htm. Nederlands Normalisatie instituut (NEN), 2009. NEN-ISO 10523: 2008 en, Water e Bepaling van de pH. http://www.nen.nl/web/ Normshop/Norm/NENISO-105232008-en.htm. Neubauer, S.C., Emerson, D., Megonigal, J.P., 2002. Life at the energetic edge: kinetics of circumneutral iron oxidation by lithotrophic iron-oxidizing bacteria isolated from the wetlandplant rhizosphere. Applied and Environmental Microbiology 68 (8), 3988e3995.
5398
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 5 3 8 9 e5 3 9 8
Ninh Pham, A., Rose, A.L., Feitz, A.J., Waite, T.D., 2004. The Effect of Dissolved Natural Organic Matter on the Rate of Removal of Ferrous Iron in Fresh Waters, pp. 213e219. Sharma, S.K., Petrusevski, B., Schippers, J.C., 2005. Biological iron removal from groundwater: a review. Journal of Water Supply: Research and Technology e AQUA 54 (4), 239e247. Sharma, S.K. (2001) Adsorptive iron removal from groundwater. Dissertation Wageningen University/International Institute for Infrastructural, Hydraulic and Environmental Engineering, Delft, the Netherlands. Søgaard, E.G., Medenwaldt, R., Abraham-Peskir, J.V., 2000. Conditions and rates of biotic and abiotic iron precipitation in selected Danish freshwater plants and microscopic analysis of precipitate morphology. Water Research 34 (10), 2675e2682. Spring, S., Ka¨mpfer, P., 2005. Bergey’s Manual of Systematic Bacteriology, pp. 740e746. Standard methods for the Examination of Water and Waste Water, fourteenth ed., 1975 American Public Health Association, Washington, pp. 208e213. Stumm, W., Lee, G.F., 1961. Oxygenation of Ferrous Iron, pp. 143e146. Sung, W., 1980. Kinetics and product of ferrous iron oxygenation in aqueous systems. Environmental Science and Technology 14 (5), 561e568.
Tamura, H., Goto, K., Nagayama, M., 1976. Effect of ferric hydroxide on the oxygenation of ferrous ions in neutral solutions. Corrosion Science 16 (4), 197e207. Tekerlekopoulou, A.G., Vayenas, D.V., 2008. Simultaneous biological removal of ammonia, iron and manganese from potable water using a trickling filter. Biochemical Engineering Journal 39 (1), 215e220. Theis, T.L., Singer, P.C., 1974. Complexation of iron(II) by organic matter and its effect on iron(II) oxygenation. Environmental Science and Technology 8 (6), 569e573. Tipping, E., 1981. The adsorption of aquatic humic substances by iron oxides. Geochimica et Cosmochimica Acta 45 (2), 191e199. de Vet, W.W.J.M., Dinkla, I.J.T., Muyzer, G., Rietveld, L.C., van Loosdrecht, M.C.M., 2009. Molecular characterization of microbial populations in groundwater sources and sand filters for drinking water production. Water Research 43 (1), 182e194. de Vet, W.W.J.M. (2011). Biological drinking water treatment of anaerobic groundwater in trickling filters. Dissertation. Technical University Delft, The Netherlands. http://repository. tudelft.nl. de Vries, H., 1890. Die Pflanzen und Thiere in den dunklen Ra¨umen der Rotterdamer Wasserleitung; Bericht u¨ber die biologischen Untersuchungen der Crenothrix-Commission zu Rotterdam, vom Jahre 1887. Verlag von Gustav Fischer, Jena.
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 5 3 9 9 e5 4 1 1
Available at www.sciencedirect.com
journal homepage: www.elsevier.com/locate/watres
Temporal variability of pharmaceuticals and illicit drugs in wastewater and the effects of a major sporting event Daniel Gerrity a,b,*, Rebecca A. Trenholm b, Shane A. Snyder b,c a
Trussell Technologies, Inc., 6540 Lusk Blvd., Suite C274, San Diego, CA 92121, United States 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 c Department of Chemical and Environmental Engineering, University of Arizona, 1133 E. James E. Rogers Way, Harshbarger 108, Tucson, AZ 85721-0011, United States b
article info
abstract
Article history:
Diurnal variations in wastewater flows are common phenomena related to peak water use
Received 23 April 2011
periods. However, few studies have examined high-resolution temporal variability in trace
Received in revised form
organic contaminant (TOrC) concentrations and loadings. Even fewer have assessed the
23 June 2011
impacts of a special event or holiday. This study characterizes the temporal variability
Accepted 17 July 2011
associated with a major sporting event using flow data and corresponding mass loadings of
Available online 23 July 2011
a suite of prescription pharmaceuticals, potential endocrine disrupting compounds (EDCs), and illicit drugs. Wastewater influent and finished effluent samples were collected during
Keywords:
the National Football League’s Super Bowl, which is a significant weekend for tourism in
Pharmaceutical
the study area. Data from a baseline weekend is also provided to illustrate flows and TOrC
Endocrine disrupting
loadings during “normal” operational conditions. Some compounds exhibited interesting
compound (EDC)
temporal variations (e.g., atenolol), and several compounds demonstrated different loading
Illicit drug
profiles during the Super Bowl and baseline weekends (e.g., the primary cocaine metabolite
Temporal variation
benzoylecgonine). Interestingly, the influent mass loadings of prescription pharmaceuti-
Loading
cals were generally similar in magnitude to those of the illicit drugs and their metabolites.
Reuse
However, conventional wastewater treatment was more effective in removing the illicit
Wastewater
drugs and their metabolites. Total influent and effluent mass loadings are also provided to summarize treatment efficacy and environmental discharges. ª 2011 Elsevier Ltd. All rights reserved.
1.
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 (Snyder et al., 2001, 2004; Lange et al., 2009), potential human health effects (Snyder et al., 2008; Schriks et al., 2010; Stanford et al., 2010), and increased media coverage (Donn et al., 2008), which ultimately led to increased public awareness. This increased interest was coupled with the development of extremely sensitive analytical methods such as liquid chromatographyetandem mass spectrometry (LCeMS/MS) that allowed researchers to approach parts-per-quadrillion (sub-ng/L) detection limits for
* Corresponding author. Trussell Technologies, Inc., 6540 Lusk Blvd., Suite C274, San Diego, CA 92121, United States. Tel.: þ1 858 458 1030; fax: þ1 626 486 0571. 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.2011.07.020
5400
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 5 3 9 9 e5 4 1 1
a variety of trace organic contaminants (TOrCs) (Ternes et al., 2002; Snyder et al., 2003; Vanderford and Snyder, 2006; Postigo et al., 2011). Each of these factors led to more thorough scientific investigations into the presence, fate, and transport of TOrCs in natural and engineered systems. With respect to organic compounds intended for human consumption, contamination of water supplies stems 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, 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 (Nelson et al., 2011). To aid in this effort, one design factor that must be studied in greater detail is the temporal variability of TOrC occurrence in wastewater. Diurnal variations in wastewater flows are common phenomena related to peak water use periods (Nelson et al., 2011). Sewers and wastewater treatment plants must be designed to account for the maximum and minimum flows and associated loadings each day (Ort et al., 2010). However, few studies have examined temporal fluctuations in TOrC concentrations and mass loadings (Joss et al., 2005; Takao et al., 2008; Nelson et al., 2011; Plosz et al., 2010; Postigo et al., 2011). In the existing studies, the temporal resolution was typically limited to sampling intervals of 8 h or longer. However, one recent study evaluated finished effluent samples for a suite of TOrCs on an hourly basis (Nelson et al., 2011). These high-resolution finished effluent samples characterize the temporal variability of environmental discharges, but they do not indicate the temporal variability of the influent mass loadings due to attenuation during treatment. Another recent study indicated that influent TOrC concentrations may vary on extremely short time scalesdeven as short as 2 mindand this may bias many of the recent wastewater monitoring studies (Ort et al., 2010). The authors emphasized that influent wastewater is “composed of a number of intermittently discharged, individual wastewater packets from household appliances, industries, or subcatchments” (Ort et al., 2010). The authors demonstrated that such temporal variation exists by monitoring TOrCs at time scales that could capture a single toilet flush. As noted in their study, such resolution is often limited by the costly, labor-intensive analytical methods necessary to detect trace concentrations of organic contaminants in wastewater. In fact, the authors indicated that only two previous studies had reported TOrC concentrations with sufficient temporal resolution (Ort et al., 2005; Ort and Gujer, 2006). The extent of temporal variation is dependent on the characteristics of the target compounds and the size of the service area for a particular wastewater treatment plant, thereby accounting for the number of toilet flushes containing the compounds of interest (Ort et al., 2010). In a small
catchment, contrast media used for magnetic resonance imaging will demonstrate much more temporal variability than compounds with more widespread human consumption (Joss et al., 2005; Ort et al., 2010; Nelson et al., 2011), such as non-steroidal anti-inflammatory drugs (Ternes, 1998; Joss et al., 2005). In particular, X-ray contrast media are more prevalent on weekdays when most scheduled medical appointments occur (Nelson et al., 2011). Ort et al. (2010) also emphasized that sampling uncertainty will even be a factor for large systems. Therefore, additional studies are necessary to characterize the temporal variation that is generally lost in large composite samples. Days of the week, seasons, and even special events may warrant design or operational considerations given their potential for unusual flow patterns and contaminant loadings. Nelson et al. (2011) reported significant concentration spikes for the insect repellant N,N-diethyl-meta-toluamide (DEET) during the warmer months when mosquitoes are most prevalent. With respect to special events, a recent article documented spikes in wastewater flows caused by different stages of a major auto race in Speedway, Indiana, USA (Enfinger and Stevens, 2011). Another study evaluated days of the week, seasons, and winter holidays for their effects on illicit drug concentrations in Spanish surface water (Huerta-Fontela et al., 2008). The authors observed higher illicit drug concentrations, including amphetamine-type stimulants, cocaine, and cocaine metabolites, on weekends compared to weekdays, and the authors also observed the highest concentrations in the winter, particularly after the Christmas and New Year holidays (Huerta-Fontela et al., 2008). These weekly fluctuations and holiday-specific spikes are supported by other studies of illicit drug use in Canada and Spain (Metcalfe et al., 2010; Postigo et al., 2011). The current study addresses some of these issues, including high-resolution temporal variability and the effects of a special event. This study presents flow data and corresponding influent and effluent mass loadings of a suite of prescription pharmaceuticals, potential EDCs, and illicit drugs at a wastewater treatment plant in a major metropolitan area in the United States (U.S.). Samples were collected during the National Football League’s Super Bowl weekend in addition to a baseline weekend to compare flows and mass loadings during “special event” versus “normal” operational conditions. Although the game was not held in the study area, the Super Bowl causes a tremendous spike in tourism and associated wastewater flows. The additional wastewater flows pose potential operational issues for local wastewater treatment plants, including fluctuations in the loadings of prescription and illicit drugs. This study characterizes these issues and provides further evidence of the importance of sample collection strategies in accurately characterizing TOrC concentrations in wastewater, as emphasized in Ort et al. (2010).
2.
Materials and methods
2.1.
Sampling location
Samples were collected from a municipal wastewater treatment plant with an average daily flow of approximately
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 5 3 9 9 e5 4 1 1
380,000 m3/day (100 million gallons per day (MGD)). The service area of the wastewater treatment plant is approximately 505 km2 (195 mi2) with a total population of approximately 1 million people. More than 99% of the flow is delivered to the treatment plant by 3206 km (1992 mi) of gravity sewer lines, while the remaining portion is delivered by continuously operated lift stations and 64 km (40 mi) of pressurized lines. The flow rates at the lift stations remain relatively constant throughout the day. The principal treatment train consists of bar screens; grit removal; primary clarification with ferric chloride addition; activated sludge with full nitrification (NH3,eff
