WATER RESEARCH A Journal of the International Water Association
Editor-in-Chief Mark van Loosdrecht Delft University of Technology Department of Biochemical Engineering Julianalaan 67 2628 BC Delft The Netherlands Tel: +31 15 27 81618 E-mail:
[email protected] Editors J. Block Université H. Poincaré, Nancy I France David Dixon University of Melbourne Australia Hiroaki Furumai The University of Tokyo Japan Xiaodi Hao Beijing University of Civil Engineering and Architecture China Gregory Korshin University of Washington USA Anna Ledin Formas Sweden Eberhard Morgenroth Swiss Federal Institute of Aquatic Science and Technology (EAWAG) Switzerland W. Rauch University Innsbruck Austria Maria Reis Universidade Nova de Lisboa/FCT Portugal Hang-Shik Shin Korea Advanced Institute of Science and Technology Korea Thomas Ternes Bundesanstalt für Gewässerkunde Germany Stefan Wuertz Univ. of California, Davis USA
Associate Editors Andrew Baker University of New South Wales Australia
Damien Batstone The University of Queensland Australia G-H. Chen The Hong Kong University of Science & Technology Hong Kong China Tom Curtis Univ. of Newcastle upon Tyne UK Ana Deletic Monash University USA Francis de los Reyes III North Carolina State University USA Rob Eldridge The University of Melbourne Australia Rosina Girones University of Barcelona Spain Stephen Gray Victoria University Australia Kate Grudpan Chiang Mai University Thailand E.E. Herricks University of Illinois - Urbana USA Peter Hillis United Utilities Plc UK H-Y. Hu Tsinghua University China P.M. Huck University of Waterloo Canada Bruce Jefferson Cranfield University UK Ulf Jeppsson Lund University Sweden Sergey Kalyuzhnyi Moscow State University Russian Federation Jaehong Kim Georgia Institute of Technology USA Jes La Cour Jansen Lund Institute of Technology Sweden G. Langergraber BOKU/Univ. of Natural Res. and Applied Life Scs. Austria S-L. Lo National Taiwan University Taiwan Dionisis Mantzavinos Technical University of Crete Greece
Y. Matsui Hokkaido University Japan A. Maul Université Paul Verlaine-Metz France Max Maurer EAWAG Switzerland How Yong Ng National University of Singapore Singapore Satoshi Okabe Hokkaido University Japan S.L. Ong National University of Singapore Singapore Jong M. Park Pohang University of Science & Technology Korea Susan Richardson U.S. Environmental Protection Agency USA Miguel Salgot University of Barcelona Spain David Sedlak University of California, Berkeley USA Jean-Philippe Steyer LBE-INRA France M. Takahashi Hokkaido University Japan Kai Udert EAWAG Switzerland V.P. Venugopalan BARC Facilities India E. von Sperling Federal University of Minas Gerais Brazil Hanqing Yu University of Science & Technology of China China J. Zilles University of Illinois Urbana USA A.I. Zouboulis Aristotle University Greece
Editorial Office Mailing address Water Research Radarweg 29 1043 NX Amsterdam The Netherlands E-mail:
[email protected] Publication information: Water Research (ISSN 0043-1354). For 2011, volume 45 is scheduled for publication. Subscription prices are available upon request from the publisher or from the Elsevier Customer Service Department nearest you or from this journal’s website (http://www.elsevier.com/locate/watres). Further information is available on this journal and other Elsevier products through Elsevier’s website (http://www.elsevier.com). Subscriptions are accepted on a prepaid basis only and are entered on a calendar year basis. Issues are sent by standard mail (surface within Europe, air delivery outside Europe). Priority rates are available upon request. Claims for missing issues should be made within six months of the date of dispatch. Author Enquiries: 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 ) 2 3 5 9 e2 3 7 4
Available at www.sciencedirect.com
journal homepage: www.elsevier.com/locate/watres
Prediction of biological integrity based on environmental similarity - Revealing the scale-dependant link between study area and top environmental predictors David Bedoya a,*, Elias S. Manolakos b, Vladimir Novotny a a
Civil & Environmental Engineering Department, Northeastern University, 400 Snell Engineering Center, 360 Huntington Avenue, Boston, MA 02115, USA b Department of Informatics & Telecommunications, National and Kapodistrian University of Athens, Panepistimiopolis, Ilissia, Athens 15784, Greece
article info
abstract
Article history:
Indices of Biological integrity (IBI) are considered valid indicators of the overall health of
Received 20 July 2010
a water body because the biological community is an endpoint within natural systems.
Received in revised form
However, prediction of biological integrity using information from multi-parameter envi-
10 January 2011
ronmental observations is a challenging problem due to the hierarchical organization of
Accepted 11 January 2011
the natural environment, the existence of nonlinear inter-dependencies among variables
Available online 28 January 2011
as well as natural stochasticity and measurement noise. We present a method for predicting the Fish Index of Biological Integrity (IBI) using multiple environmental observa-
Keywords:
tions at the state-scale in Ohio. Instream (chemical and physical quality) and offstream
Environmental stressors
parameters (regional and local upstream land uses, stream fragmentation, and point
Biological integrity
source density and intensity) are used for this purpose. The IBI predictions are obtained
Geographic scale
using the environmental site-similarity concept and following a simple to implement
Environmental similarity
leave-one-out cross validation approach. An IBI prediction for a sampling site is calculated
Stressor and biological hierarchy
by averaging the observed IBI scores of observations clustered in the most similar branch of a dendrogram ea hierarchical clustering tree of environmental observations- built using the rest of the observations. The standardized Euclidean distance is used to assess dissimilarity between observations. The constructed predictive model was able to explain 61% of the IBI variability statewide. Stream fragmentation and regional land use explained 60% of the variability; the remaining 1% was explained by instream habitat quality. Metrics related to local land use, water quality, and point source density and intensity did not improve the predictive model at the state-scale. The impact of local environmental conditions was evaluated by comparing local characteristics between well- and mispredicted sites. Significant differences in local land use patterns and upstream fragmentation density explained some of the model’s over-predictions. Local land use conditions explained some of the model’s IBI under-predictions at the state-scale since none of the variables within this group were included in the best final predictive model. Under-predicted sites also had higher levels of downstream fragmentation.
* Corresponding author. Tel.: þ1617 314 7116; fax: þ1 617 3147115. E-mail addresses:
[email protected] (D. Bedoya),
[email protected] (E.S. Manolakos),
[email protected] (V. Novotny). 0043-1354/$ e see front matter ª 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.watres.2011.01.007
2360
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 3 5 9 e2 3 7 4
The proposed variables ranking and predictive modeling methodology is very well suited for the analysis of hierarchical environments, such as natural fresh water systems, with many cross-correlated environmental variables. It is computationally efficient, can be fully automated, does not make any pre-conceived assumptions on the variables interdependency structure (such as linearity), and it is able to rank variables in a database and generate IBI predictions using only non-parametric easy to implement hierarchical clustering. ª 2011 Elsevier Ltd. All rights reserved.
1.
Introduction
Integrity has been defined as the ability of a water body to maintain “a balanced, integrated, adaptive community of organisms having a species composition, diversity and functional organisms comparable to that of a natural biota of the region” (Karr et al., 1986). Biological integrity of streams is usually measured with some version of a calibrated index. One of the most widely used indices in the United States is the Fish Index of Biological Integrity developed by Karr et al. (1986). Many public agencies have adopted it as a framework for deriving their own calibrated version at the state or regional scale (Ohio EPA, 1987; Bode, 1988; Roth et al., 1998; Lyons et al., 2001; Lyons, 2006). Karr’s IBI and subsequent versions and calibrations are based on a comparison of observed fish abundances and community composition against expected values in reference sites with similar environmental characteristics. The importance of IBI lies in its sensitivity to disturbances of different nature because the biological community is an endpoint in the ecological river system (Karr et al., 1986; Novotny, 2003). However, the identification of major sources of biological degradation is challenging because the natural habitat is organized as a nested hierarchy of environmental filters with different geographic scales, to which the biological community has adapted (Pickett et al., 1989). Consequently, the geographic scale at which biological integrity is evaluated is of great importance because the stressors identified as most significant to the biological community are those at the highest level in the hierarchy of environmental filters at that particular scale (Poff, 1997). Therefore, measures to improve biological integrity need to be approached in a holistic scaleadaptive manner in order to be effective. The impact of stressors should be viewed within the context of disturbances occurring at larger scales than the study region (i.e. background quality). The ecological hierarchy in the natural river system is composed of numerous instream and offstream environmental variables which are highly inter-twined and crosscorrelated (Novotny et al., 2005). Therefore, changes in one of them will most likely have a cascade effect that may translate into changes in the instream conditions affecting the biological community. For example, land use changes in the watershed will affect, among other variables, sediment and nutrient input which will, in turn, affect physical and chemical instream water quality. If enough exposure of living organisms occurs, these will be negatively affected because they are the system’s endpoint (Novotny et al., 2005).
The river system is organized a hierarchy of environmental characteristics and habitat conditions across multiple spatial scales (Frissell et al., 1986). An aquatic habitat is suitable for specific fauna when the different natural environmental filters at different spatial scales are overcome (Poff, 1997). Man-made modifications of any of these natural filters at any scale-level are stressors that will modify the pristine biological integrity of the site (Poff, 1997; Karr et al., 1986.) Large-scale variables -or environmental gradients- are those which produce a change in biological integrity of the system through their whole range of values within the study area (i.e. spatial scale of the study). These environmental parameters are usually the best integrity predictors at the selected scale. (Bedoya et al., in press; Lannert and Allan, 1999). On the other hand, smallscale variables have also an effect on particular sections of the area of study, but not on its entirety (Bedoya et al., in press; Lanmert and Allan, 1999). Therefore, identification of variables acting as gradients in a study area should be targeted as top priority for remediation purposes. Large-scale variables actually set the background biological integrity of a region and therefore, overall improvement of its biological integrity is always conditioned by them. Because of the numerous cross-correlated variables potentially affecting IBI and the non-linear variable-to-IBI relationships, development of effective predictive modeling methodologies able to exploit a large number of multi-dimensional environmental observations is paramount. Moreover, new methods to predict biological integrity should not be constrained by any pre-imposed conditions. The methodology we present in this research meets these two key requirements. IBI prediction is performed with a two-phase approach. The first phase consists of ranking variables based on their overall impact on the biological community at the scale of our study area. The second phase consists of a step-wise IBI prediction using environmental variables from different categories (e.g. instream habitat variables). The best predicting variables from each group of variables are then progressively combined in order to obtain an overall improved IBI predictive model.
2.
Methodology
2.1.
Data and study area
The research reported here was based on 429 observations within the state of Ohio. This dataset was extracted from a larger database compiled by the Ohio Environmental Protection Agency (EPA) during its Statewide Biological and
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 3 5 9 e2 3 7 4
Water Quality Monitoring and Assessment Program (Ohio EPA 2008). The original database made available to our research team consisted of 1848 observations out of which only 429 had information for all the environmental parameters (i.e. sites with no missing data). This research is based on these 429 complete observations. One observation corresponded to one or two visits to the corresponding site with a small time difference between visits (usually one to two weeks). During these site visits, grab samples for chemical water quality analyses were collected and an evaluation of habitat and biological qualities was performed. The data were collected between 1996 and 2000 by Ohio EPA. Most of the sites had at least two observations during this period, although some of them were evaluated just once. Most of the samples were collected in summer months (July through September) with very few (less than 20) in early October. By sampling in the same time period, potential IBI annual fluctuations were avoided to the maximum extent practicable. Sampling activities focused in summertime low flow periods when stress to aquatic biological communities is believed to be greatest (Ohio EPA, 2005). The distribution of the sampling sites across the state of Ohio is presented in Fig. 1.The state of Ohio follows a very systematic sampling strategy. Site selection within the watershed is driven by a stratification of the watershed based on a sequential, systematic halving of the drainage area, such that a census of all streams within the watershed down to a prescribed drainage area size are selected for sampling (Ohio EPA, 2005). Biological -fish IBI scores- as well as instream environmental parameters -chemical and physical quality- were complemented with offstream parameters obtained with a Geographic Information System (GIS). To our knowledge, all data were collected in base-flow conditions and extreme events (e.g. a spill) were not reported at any sampling site. For each observation site, biological integrity was measured using the fish IBI. In Ohio, this is a discrete score ranging from 12 (essentially no fish) to 60 (reference conditions). The IBI is calculated as the sum of 12 different metrics (each one an integer score ranging from 1 to 5) that describe the species
Fig. 1 e Distribution of sampling sites across the state of Ohio.
2361
richness and composition, the trophic composition, and the fish abundance and condition of the fish community (Karr et al., 1986; Ohio EPA, 1987). Instream variables consisted of water quality and habitat quality metrics (Table 2). Habitat parameters consisted of metrics from the Qualitative Habitat Evaluation Index (QHEI) (Rankin, 1989). The QHEI and its metrics are discrete scores with different ranges (see Table 2). The percentage of fine sediment in the river bed (embeddedness) was also available (this variable is not used as a QHEI metric itself, but as a penalizing factor for the QHEI’s substrate and channel quality metrics). The offstream environmental variables were grouped into three main categories: upstream land use, stream fragmentation, and point source density and intensity. In order to calculate the upstream land uses, each site’s watershed was delineated using a 30-m resolution Digital Elevation Map (DEM) with ArcGIS Spatial Analyst. Subsequently, the percentages of different upstream land use was calculated at two different scales: the regional scale, which included the whole upstream contributing catchment, and the local scale, which included only 2 miles upstream from the sampling site. Land use percentages were calculated for the whole upstream area as well as the 100- and 30-m buffers at both scales. These two buffer widths were selected based on literature values. A buffer width of 30 m is considered the minimum necessary to provide some benefit to the receiving water body such as temperature amelioration (Castelle et al., 1994). Moreover, 30 m was the maximum resolution of the DEM. A buffer with of 100 m was selected because this distance is considered sufficient to perform basic functions such as sediment removal, nutrient removal, and preservation of species diversity (Castelle et al., 1994). Land use percentages were calculated using the Thematic Raster Summary function within Hawth’s Analysis Tools for ArcGIS (Beyer, 2004). Land cover categories as defined in the 2001 National Land Cover Dataset (NLCD) were used (USGS, 2008b) and listed in Table 1. The Open Water (OW) land use category was only calculated for the regional- and local-scale whole catchment areas, not for the buffers. Drainage areas (DA) for each site were also calculated. The fragmentation and point source metrics (Table 2) were calculated using information from the National Hydrography Datasets (NHD) (USGS, 2008a). The ArcGIS Utility Network Analyst was used to trace upstream or downstream a specific site. Major dams (i.e. with DA 2.59 Km2) and point sources (major and minor waste water treatment plants and major industrial dischargers) were obtained from the National Inventory of Dams (USACE, 2005) and the Permit Compliance System database (USEPA, 2008) respectively. In the fragmentation metrics, downstream metrics such as downstream dam frequency in the main channel (DW_MainDf, see Table 2) or upstream metrics such as upstream dam frequency (U_Df, see Table 2) were considered indicators of site accessibility or habitat continuity from downstream or upstream points respectively. Metrics that combined upstream and downstream segments such as average dam frequency (Avg_Df, see Table 2) were considered indicators of habitat size. Sites were not segregated a priori based on ecoregions or stream size because the model should be able to separate clusters of sites with significant environmental differences. In other words, we wanted to “let the data speak”.
2362
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 3 5 9 e2 3 7 4
Table 1 e Description, percentage quartiles, and individual IBI predicting power for the different NLCD land use categories in the Ohio database. Hay [ hay/pasture; ForD [ deciduous forest; ForM [ mixed forest; ForE [ evergreen forest; Shr [ shrub/scrub; WetH [ herbaceous wetlands; WetW [ woody wetlands; Herb [ herbaceous; Crop [ crops; Bar [ barren; DevH [ high intensity urban; DevM [ medium intensity urban; DevL [ low intensity urban; DevO [ open urban space; OW [ open water; Oth [ other land uses. Regional Drainage Area (RDA) Name
Quartiles
Regional 100-Meter Buffer (R100) R
2
Name
Quartiles
Regional 30-Meter Buffer (R30) 2
R
Regional Land Use in Contributing Area RDA_Hay 3.16e7.67e14.60 0.385a RDA_ForM 0.00e0.00e0.03 0.294b RDA_DevL 1.22e2.39e6.06 0.293a RDA_OW 0.10e0.25e0.60 0.292a RDA_ForD 5.29e9.45e18.60 0.281a RDA_WetH 0.00e0.02e0.09 0.261b RDA_Crop 40.49e60.84e75.58 0.257a RDA_DevM 0.24e0.64e1.64 0.253a RDA_ForE 0.01e0.08e0.25 0.243a RDA_Bar 0.00e0.01e0.05 0.239a RDA_Herb 0.35e0.89e1.38 0.225a RDA_DevH 0.07e0.26e0.75 0.223a RDA_WetW 0.00e0.04e0.20 0.223a RDA_Shr 0.00e0.00e0.03 0.221a RDA_DevO 5.23e6.25e9.76 0.212a RDA_Oth 0.00e0.00e0.00 0.012a
Regional Land Use in 100 m Buffer R100_Hay 2.97e8.20e13.60 0.322a R100_DevH 0.00e0.13e0.38 0.320a R100_DevO 5.26e6.84e10.33 0.300a R100_Herb 0.29e1.00e1.81 0.296a R100_ForD 8.22e16.06e30.56 0.292a R100_ForE 0.00e0.09e0.24 0.285a R100_DevM 0.10 -0.40 -0.87 0.275a R100_DevL 0.94e1.85e3.79 0.274a R100_WetW 0.00e0.13e0.58 0.270a R100_ForM 0.00e0.00e0.05 0.261a R100_WetH 0.00e0.03e0.20 0.236b R100_Crop 34.65e54.32e70.27 0.231a R100_Shr 0.00e0.00e0.05 0.223a R100_Bar 0.00e0.00e0.02 0.184b R100_Oth 0.00e0.00e0.00 0.012a
Local Drainage Area (LDA)
Local 100-Meter Buffer (L100)
Local Land Use in Contributing Area LDA_DevO 4.97e7.22e13.73 0.289a LDA_ForD 4.47e13.59e28.73 0.285b LDA_Hay 0.00 -5.51 - 12.92 0.214a LDA_OW 0.00e0.19e0.96 0.200b LDA_DevL 0.42e2.42e11.19 0.183a LDA_DevM 0.00e0.27e2.01 0.159a LDA_Crop 17.20 -44.71 -69.59 0.152a LDA_DevH 0.00e0.00e0.80 0.140b LDA_Herb 0.00e0.65e1.68 0.130a LDA_WetW 0.00e0.00e0.47 0.128a LDA_WetH 0.00e0.00e0.19 0.124b LDA_Shr 0.00e0.00e0.00 0.117a LDA_ForE 0.00e0.00e0.25 0.098a LDA_ForM 0.00e0.00e0.00 0.077a LDA_Bar 0.00e0.00e0.00 0.020a LDA_Oth 0.00e0.00e0.00 0.000
Local Land Use in 100 m Buffer L100_ForD 7.39e24.43e46.14 L100_DevL 0.30e2.06e6.99 L100_DevO 4.17e7.80e14.68 L100_Hay 0.00e3.97e10.85 L100_Crop 8.55e30.06e59.61 L100_Herb 0.00e0.34e1.84 L100_WetW 0.00e0.00e1.28 L100_DevM 0.00e0.00e1.78 L100_Shr 0.00e0.00e0.00 L100_ForE 0.00e0.00e0.15 L100_WetH 0.00e0.00e0.63 L100_DevH 0.00e0.00e0.16 L100_ForM 0.00e0.00e0.00 L100_Bar 0.00e0.00e0.00 L100_Oth 0.00e0.00e0.00
Name Regional R30_ForD R30_Hay R30_Herb R30_Crop R30_DevM R30_DevL R30_WetW R30_DevO R30_ForM R30_ForE R30_Shr R30_DevH R30_Bar R30_WetH R30_Oth
Quartiles
R2
Land Use in 30 m Buffer 9.82e20.90e38.01 0.368a 3.13e7.35e12.57 0.312a 0.23e1.00 -2.08 0.312b 30.69e50.23e64.82 0.304a 0.07e0.26 -0.62 0.259a 0.79e1.51e3.48 0.245a 0.00e0.26e0.95 0.242a 4.44e6.26e10.01 0.234a 0.00e0.00e0.04 0.232a 0.00 -0.05 -0.20 0.225a 0.00e0.00e0.02 0.208a 0.00e0.07e0.20 0.198a 0.00e0.00e0.01 0.182a 0.00e0.01e0.36 0.172b 0.00e0.00e0.00 0.012a
Local 30-Meter Buffer (L30)
0.335a 0.272b 0.208a 0.196a 0.190b 0.117b 0.113a 0.107b 0.069c 0.064a 0.051a 0.048b 0.042a 0.005b 0.000
Local Land Use in 30 m Buffer L30_ForD 7.33e29.34e53.65 L30_Crop 6.33e25.24e53.97 L30_DevL 0.00e1.20e6.59 L30_Hay 0.00e1.67e9.45 L30_DevO 3.00-6.42-15.23 L30_Herb 0.00e0.00e1.61 L30_DevM 0.00e0.00e1.12 L30_WetW 0.00e0.00e2.24 L30_WetH 0.00e0.00e2.24 L30_DevH 0.00e0.00e0.00 L30_ForE 0.00e0.00e0.00 L30_ForM 0.00e0.00e0.00 L30_Shr 0.00e0.00e0.00 L30_Bar 0.00e0.00e0.00 L30_Oth 0.00e0.00e0.00
0.334a 0.202a 0.186a 0.161a 0.158a 0.154b 0.100a 0.092a 0.087a 0.071c 0.066a 0.036c 0.032c 0.004b 0.000
a a ¼ best prediction at 423 branches. b b ¼ best prediction at 328 branches. c c ¼ best prediction at 233 branches.
2.2.
Environmental variables ranking
Environmental variables are divided into two categories: offstream and instream. The offstream category is composed of four groups: local and regional land use -in the whole upstream area and the 30- and 100-m buffers-, stream fragmentation, and point source density and intensity. The instream category is composed of two groups: water and habitat qualities. The individual predictive power of each environmental variable is initially estimated by obtaining the coefficient of determination [r2] of the observed (measured) IBI versus a calculated IBI prediction values, generated using a leaveone-out cross-validation approach detailed below. Let us assume that we are given a database of measurements represented as a matrix X of n observations (rows) by m environmental variables (columns). One element of that
matrix, i.e. one observation of variable v, (let’s call it Xi; v without loss of generality) is isolated and will be called the test or query site. The remaining n-1 observations of the same variable, namely the measurement (½X1; v; .Xn 1; v) in the same column of the data matrix, are organized in a “dendrogram” tree structure having ½X1; v; .Xn 1; v as leaves. This is done by applying agglomerative Hierarchical Clustering (HC) using the average linkage method and the standardized Euclidean as the distance metric (Jain et al., 1999). When the resulting dendrogram is “cut” at a certain distance from the root (more details on how the cut level is decided are given in the last paragraph of this section) several tree branches are emanating from the cut. Among them we identify the Most Similar Branch to the test site i when using variable v (to be called MSBi;v ) as the branch (overall branches Bk defined by the cut) for which the standardized Euclidean distance between
2363
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 3 5 9 e2 3 7 4
Table 2 e Description and individual IBI predicting power for the water quality, habitat, point source, and stream fragmentation metrics. Name
Units
Description
R2
Name
Units
Description
Water Quality Parameters BOD
Habitat Parameters 0.12
b
Embed
0e4
Embeddedness
0.28a
Riffle Subs Pool Chan DA Rip Cover Grad Fragmentation Parameters SITE_Con*
0e8 0e20 0e12 0e20 Km2 0e10 0e20 0e10
Riffle and run quality Substrate quality Pool and glide quality Channel morphology score Drainage area Riparian and bank qualities Instream vegetal cover Gradient score
0.24a 0.23a 0.23a 0.21a 0.19c 0.17b 0.13a 0.09c
Fraction
Total connected length from observation site/ Total basin network length Downstream flooded area/Downstream main channel length Downstream dam storage/Main channel length
0.47a
0.09c 0.08c 0.07b 0.06b 0.04b 0.03a 0.02a 0.02c 0.02a
Cl
Biological Oxygen. Demand mg/L Total Kjeldahl Nitrogen mg/L Total Arsenic mg/L Ammonia as Nitrogen mg/L Nitrite as Nitrogen mg/L Total Magnesium mg/L Total Sulfate mmho/cm Conductivity mg/L Dissolved Oxygen mg/L Hardness (as CaCO3) mg/L Total Chloride
pH
S.U.
pH
0.01b Dfl_MainLen* m2/km
TSS
mg/L
0.01b Dsto_MLen*
m3/km
NO3
mg/L
Total Suspended Solids Nitrate as Nitrogen
0.01c DW_MainDf*
Km
Ca Cd Cu Fe Zn TP Pb
mg/L mg/L mg/L mg/L mg/L mg/L mg/L
Total Calcium Total Cadmium Total Copper Total Iron Total Zinc Total Phosphorus Total Lead
0.01b 0.01a 0.01c 0.01b 0.01b 0.00a 0.00a
TKN As NH4 NO2 Mg SO4 Cond DO Hard
mg/L
R2
0.02a
Avg_Df* Km U_Df Km UPS_Con Fraction Uflood_len m2/km UPS_Flooded Fraction UPS_stor_len m3/km UPS_stor_DA m3/Km2 Point Source Parameters PS_LTOT No./km PSDisch_LPS
m3/d/Km
PS_LPS
No./km
PSDisch_DA PSDisch_LT
m3/d/Km2 m3/d/Km
Flow_PS LPS-DA
% Km/Km2
0.46a 0.44b
Main channel downstream length/Number of 038a downstream dams Mean value between DW_MainDf and U_Df 0.26a Upstream network length/Number of upstream dams 0.25a Upstream connected length/Total upstream length 0.19a Upstream flooded area/Upstream network length 0.17a Upstream flooded area/Drainage area 0.17a Upstream dam storage/Upstream network length 0.16a Upstream dam storage/Drainage area 0.15a Number of upstream point sources/Upstream network length Upstream point source discharge flow/Distance from site to all upstream point sources Number of upstream point sources/Distance to all upstream point sources Upstream point source discharge flow/Drainage area Upstream point source discharge flow/Upstream network length % of upstream network carrying waste water Distance to all upstream point sources/Drainage area
0.26b 0.21b 0.21a 0.21b 0.20b 0.20b 0.18b
*Downstream parameters calculated up to the basin outlet All distances were calculated following stream network channels. a a ¼ best prediction at 423 branches. b b ¼ best prediction at 328 branches. c c ¼ best prediction at 233 branches.
the test site’s value (Xi; v) and the mean value of variable v over the branch leaves is minimized. See Eqs. (1) and (2) below for the formal definition, MSBi;v
Dki;v
n o ¼ argk min Dki;v
dist½Xi; v; averageðXj; vÞ ¼ ; j˛Bk
(1)
(2)
whereDki;v is the standardized Euclidean distance between the test-site’s value Xi,v and the average {Xj,v} over the sites residing at the leaves of branch Bk. Note that depending on where the dendrogram is cut, a resulting branch may contain one or more observations. As “calculated IBI” prediction for site i based on variable v information we will use the mean IBI value of the
observations clustered in its corresponding Most Similar Branch (see Eq. (3)) IBICi;v ¼
average IBIOj j˛MSBði; vÞ
;
(3)
where “observed IBI” (IBIOj ) is the measured IBI value for each site j of branch MSBi;v recorded in the database. The same procedure is repeated for every site i (keeping v fixed) leading to an n 1 column vector of predicted IBIs for all sites based on information for variable v alone, called IBICv . Then the same procedure is repeated for every variable v giving rise to an n m matrix of IBI predictions,IBIc, having the same structure as the data matrix X. Finally three different dendrogram tree cuts are applied (leading to 233, 328, and 423 branches respectively) to obtain
2364
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 3 5 9 e2 3 7 4
three different n 1 vectors of IBI predictions based on each variable v. Note that the distance from the root for each cut has been selected so that the number of resulting branches is approximately equal to 50%, 75%, or 100% of the total number of available complete observations in the database (see example in Fig. 2). Among the three vectors the one with maximal similarity (in terms of [r2]) to the observed (measured) IBIO v vector is considered as the “best” predictor of IBI using information of variable v alone, and this [r2] value is assigned as the score of variable v in the rank ordering of the environmental variables.
Fig. 2 e Example of a dendrogram built using an array of n-1 observations composed of v environmental variables (v is a one- or possibly multi-dimensional vector of selected environmental variables [v1,v2.,vm]). Test-sites (X1,v,X2,v,.Xi-1,v,Xi D 1,v,.Xn,v) correspond to the leaves of the tree. Three cuts (dashed lines) are determined (see text for details). Each cut generates branches (test site clusters). The Most Similar Branch (MSB) to the test site Xi,v, is determined (see text for details). The average measured IBI of the sites (leaves) belonging to the MSB is used as the predicted IBI for the test site. The same procedure is repeated for each test site and then for each cut. The predicted IBI values are compared to the measured IBI for all sites. The predictive value of variable v is assessed based on the jr2j of the fit of best predictive model (among the three models corresponding to the three different cuts).
2.3.
Step-wise IBI prediction
This step consists of obtaining progressively improved IBI predictions by combining variables from each group separately (see groups in Fig. 3). The “best” variables from each group are combined to find the “best” offstream and instream predictors following the order of variables specified in Fig. 3. Finally, the subset of best offstream and instream predictors are also combined in a similar manner to obtain the overall best set of predictors. The IBI prediction methodology remains the same as in step 2.2. However, in this case, a step-wise approach is followed. For each group, and following the group’s variable ranking obtained in Section 2.2, the best predicting variable is first selected (let’s call it v1 w.l.o.g.). Subsequently, the second best predicting variable in each group (to be called v2) is also selected to form an array of two-dimensional environmental
Final set of offstream and instream variables
Selected offstream variables
Selected LU variables
Selected regional LU variables
Selected PS variables
Selected instream variables
Selected fragmentation variables
Selected habitat variables
Selected water quality variables
Selected local LU variables
LU in regional catchment area
LU in local catchment area
LU in regional 30-meter buffer
LU in local 30-meter buffer
LU in regional 100-meter buffer
LU in local 100-meter buffer
Fig. 3 e Diagram showing the order in which the groups of variables are combined. Dark grey rectangles indicate instream variables. Light grey rectangles indicate offstream variables. White rectangles indicate final model with a mix of both types of variables.
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 3 5 9 e2 3 7 4
vectors ({X1,(v1,v2),X2,(v1,v2),.,Xn,(v1,v2)}). One at a time, a twodimensional test site is isolated from the rest n-1 two-dimensional observations (exactly as we did in Section 2.2). The testsite vector is presented again to the branches of the dendrogram built using the remaining n-1 two-dimensional vectors. The average IBI of the observations clustered in the most similar branch is selected as the test-site’s calculated IBI Eq. (3), but now with v ¼ (v1,v2). The selected dendrogram’s branch is the one that minimizes the standardized Euclidean distance between the test-site vector and the average environmental vector calculated from the observations located in the particular branch (Equations (1) and (2) but now with v ¼ (v1,v2)). If the two-variable model improves the previous prediction (increase in R2) at any one of the three selected cut levels of the dendrogram (i.e. using 233, 328, or 423 branches), the new variable is retained otherwise it is discarded. Subsequently, the third best predicting variable in each group (v3) is introduced to form an array of two ({X1,(v1,v3), or three ({X1,(v1,v2,v3),X2,(v1,v2,v3),., X2,(v1,v3),.,Xn,(v1,v3)}) Xn,(v1,v2,v3)}) dimensional observations (depending on the inclusion or exclusion of v2 in the previous step). Again, one at a time, a test site is separated from the rest of n-1 observations and associated to the most similar branch of the dendrogram calculated with the variables used in this step. Improvement in the IBI prediction relatively to the previously tested model results in the inclusion of the last variable in the best set so far, or exclusion otherwise. This procedure is repeated for each group until all variables have been considered for inclusion. At the end of this “greedy” procedure, the “best” combination of predicting variables from a particular group of variables is identified. Although this method does not guarantee to find the globally optimal IBI predictive model it does move step by step toward a model with improved performance as new variables are introduced and it is quite fast to implement. In this research, strongly cross-correlated variables were not eliminated because the model’s performance is not adversely affected when more variables are introduced. Since prediction with the environmental similarity concept is merely based on comparing site environmental vectors with the same vector elements, presence of cross-correlated variables will not affect the performance because the same variables are used for all prediction sites. Therefore, even marginal improvements can be accounted for without jeopardizing model performance. Furthermore, since variables are examined for inclusion in a sequential manner, keeping all variables “in the game” has also the advantage of not retiring prematurely a variable that although is highly correlated to a variable already added to the model may have a possible dependence to a third variable not yet examined for inclusion. Subsequently, the different groups of predictors are also progressively “merged” using only the “best” variables for each individual group resulted in the previous step. The stepwise IBI prediction methodology used when two groups of variables are combined is identical as before. Fig. 3 shows the order in which the groups of variables are merged.
predicted sites. A site was labeled as mispredicted if the calculated IBI fell beyond the 1.5 RMSE interval (where RMSE is the Root Mean Square Error of the IBI predictions for all the available observations in the dataset). Significant differences in water quality (for those sites affected by point sources), point source and fragmentation density and intensity, as well as local and regional land uses were tested using a Student t-test at the 95% confidence level.
3.
Results
3.1.
IBI predictions with offstream variables
3.1.1.
Land use
The top seven dominant land uses in our database were (in decreasing order of median percentage in the watershed [Table 1]): cropland (60.84%), deciduous forest (9.45%), hay/ pasture lands (7.67%), urban open space (6.25%), low intensity urban space (2.39%), herbaceous lands (0.89%), and medium intensity urban space (0.64%). All the remaining land uses had a median extent in the watershed smaller than 0.5%. The local land use sub-model (Local LU model in Fig. 4) was able to account for 49% of the total IBI variability. Results seemed to indicate that proximity to the stream is important because most of the selected variables in the group model were land uses within the buffer zones instead of the whole catchment area. The regional land use model (Regional LU model in Fig. 4) explained 58% of the total IBI variability. In this case, selected land uses alternated between percentages in the whole catchment and in the buffers. The first selected model variable -percentage of hay/pasture in the drainage area- had also the highest individual IBI prediction power of all local or regional land uses (Table 1). Approximately 95% of the group variability was explained with the top three group variables: hay/pasture in the drainage area and deciduous forest and urban open space in the 30- and 100-m buffers respectively. The other six model variables only accounted for the remaining 5% of the group’s variability. The subsequent merger of the regional and local land use models yielded almost identical results as the regional land use model with the exception of the last two variables (see Fig. 4). The overall land use model eAll LU model in Fig. 4accounted for 60% of the total IBI variability. The two selected local land uses (medium intensity urban lands within the local 30- and 100-m buffers) introduced marginal improvement (1.2% of the group’s variability).
3.1.2.
Impact from local stressors
Mispredicted observations were isolated and tested for statistically significant differences against the group of well-
Point source density and intensity
This sub-model ePoint Sources sub-model in Fig. 4- accounted for the smallest IBI variability of all the offstream sub-models. Upstream point source intensity (PS_LTOT) explained 26% of the overall variability and was the first and only metric selected in the step-wise algorithm.
3.1.3. 2.4.
2365
Stream fragmentation
Fragmentation density and intensity metrics explained 54% of the overall variability. River network connectivity at the basin scale (SITE_Con) explained 47% of the overall IBI variability (87% of the sub-model’s variability). This variable had the
2366
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 3 5 9 e2 3 7 4
Local LU L100_ForD LDA_DevO L100_DevL L30_Crops L30_Hay L30_Herb L100_WetW L100_DevM L30_DevM LDA_ForE L30_DevH L30_ForM LDA_Bar
R2 0.335 0.363 0.363 0.373 0.392 0.427 0.458 0.465 0.472 0.476 0.480 0.484 0.492
Regional LU RDA_Hay R30_ForD R100_DevO RDA_ForD R100_DevL R30_WetW RDA_Herb RDA_WetW R100_Other
R2 0.385 0.509 0.546 0.558 0.562 0.569 0.570 0.572 0.577
Fragmentation SITE_Con DW_MainDf Avg_Df UPS_Con
R2 0.467 0.499 0.541 0.542
Point Sources PS_LTOT
R2 0.260
Water Quality BOD NO2 Cd
R2 0.116 0.124 0.130
Habitat Embed Riffle Subs Pool DA Cover
R2 0.281 0.326 0.403 0.431 0.442 0.491
All LU RDA_Hay R30_ForD R100_DevO RDA_ForD R100_DevL R30_WetW RDA_Herb RDA_WetW L100_DevM L30_DevM
R2 0.385 0.509 0.546 0.558 0.562 0.569 0.570 0.572 0.593 0.596
Offstream variables SITE_Con RDA_Hay DW_MainDf R30_ForD R100_DevO RDA_ForD R100_DevL R30_WetW
R2 0.467 0.512 0.535 0.537 0.563 0.592 0.596 0.597
Instream variables Embed Riffle Subs Pool DA Cover
R2 0.281 0.326 0.403 0.431 0.442 0.491
Overall SITE_Con RDA_Hay DW_MainDf R30_ForD R100_DevO RDA_ForD R100_DevL R30_WetW Riffle Cover
R2 0.469 0.512 0.535 0.537 0.563 0.592 0.596 0.597 0.605 0.606
Fig. 4 e Step-wise IBI predictions. R2 indicates the variability explained after adding a new variable to the model. All results were achieved using a hierarchical tree with 423 branches.
greatest IBI prediction capability of an individual variable overall. Downstream dam frequency, average dam frequency, and percentage of upstream connected network were other selected variables and accounted for the remaining 13% of the group’s variability.
predictions by the overall land use model, accounting for 60% of the total variability. This model used only eight variables instead of ten in the land use model.
3.2.
Predictions with instream variables
3.1.4.
3.2.1.
Instream habitat variables
Combination of best offstream variables
When the best offstream variables were combined (Fig. 3) only fragmentation and regional land use variables were selected (Offstream variables sub-model in Fig. 4). None of the local land uses or point source variables were selected. The best prediction of this sub-model model marginally improved
The instream habitat sub-model (which included drainage area) explained 49% of the overall IBI variability. Six variables were selected (Habitat sub-model in Fig. 4). The top four predictors, which accounted for 88% of the group’s variability, were directly or indirectly related to habitat’s substrate
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 3 5 9 e2 3 7 4
2367
quality (i.e. embeddedness and substrate quality) or habitat variability (riffle and pool qualities). Drainage area (which was positively correlated to IBI) and instream cover explained the remaining IBI variability in the group.
3.2.2.
Water quality variables
The water quality variables group was clearly sorted in three main clusters. The first one was related to nutrient concentration, especially nitrogen (BOD, TKN, NO2, and NH4) which had the group’s highest prediction powers. Nitrate (NO3) and TP concentrations were not ranked among the top chemical IBI predictors, being TP the poorest predictor in the group. The IBI prediction with water quality parameters was the poorest of all. The group’s top two variables (BOD, NO2) were related to nutrient loading and explained 95% of the group’s variability. The remaining 5% was explained by cadmium concentrations.
3.2.3.
Combination of best instream variables
The final “Instream variables” model (Fig. 4) yielded the exact same results as the “Habitat” model. Therefore, the “Water Quality” model did not bring any new valuable information beyond the habitat variables.
3.3.
Final predictions
The final model eOverall model in Fig. 4- was composed of all the selected offstream variables and only two instream habitat parameters: riffle quality and instream vegetal cover. These two variables only accounted for 1.5% of the group’s variability. The final model accounted for 61% of the overall IBI variability, which was a very modest improvement from the “Offstream variables” model. IBI prediction plots for the “Offstream variables”, “Instream variables” and “Overall” models are presented in Fig. 5.
3.4.
Local environmental stressors
In the final “Overall” model, a total of 28 sites were above the 1.5 RMSE threshold, while 27 were below it (see Fig. 5). Among the over-predicted observations, two sites had either extremely high concentrations of copper and zinc or point source density. Since the influence of these two sites in the performance of the t-tests was evident, they were removed. The biological quality of these two sites was mostly set by their extremely degraded water quality. After removing sites with outlier local conditions, significant differences among over- and well-predicted observations were identified in the upstream river fragmentation as well as the land use-related metrics (Table 3). Over-predicted sites had more severe upstream fragmentation but also better land use at the regional scale, which was a likely cause of over-prediction. Local land use results were mixed. Over-predicted sites had larger percentages of forested areas in both, the local catchment and buffer areas but they also had larger percentages of hay and pasture lands in the local catchment area. Presence of hay pasture lands in the drainage area was identified by our model as the best IBI predictor and it is negatively correlated to IBI (Table 1). On the other hand, under-predicted sites (i.e. calculated IBI < observed IBI - 1.5 RMSE) had consistently significantly
Fig. 5 e IBI predictions with the best offstream variables (top), best instream variables (middle), and best variables overall (bottom). Dashed lines indicate perfect fit line (center) and ±1.5 3 RMSE (sides). Dot size is proportional to the number of hits that is indicated in the legend.
lower hardness and hardness-related parameter values in sites with upstream point sources, while sulfate concentration was higher. Under-predicted sites had better land use quality at the local and regional levels (more forested areas,
2368
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 3 5 9 e2 3 7 4
Table 3 e List of variables with significant differences between over-predicted sites and sites with an IBI prediction within the ±1.5 3 RMSE range. Variable Name
Uflood_len UPS_Con UPS_stor_len Ups_stor_DA UPS_Con L30M_ForD L100M_ForD L100M_DevM LDA_ForD LDA_ForE LDA_Hay R30M_ForD R100M_ForD RDA_ForD
# of over-predicted sites/# of well-predicted sites 11/19 11/19 11/19 11/19 28/374 28/374 28/374 28/374 28/374 28/374 28/374 28/374 28/374 28/374
Type of sites
NPS NPS NPS NPS UF ALL ALL ALL ALL ALL ALL ALL ALL ALL
þ þ þ þ
Value in over-predicted sites (95% conf. interval)
Value in wellpredicted sites (95% conf. interval)
p
14.2 8.8 40.6 25.7 142.5 82.2 0.115 0.078 76.6 14.6 44.2 9.1 39.8 8.0 0.24 0.17 26.5 7.1 1.1 1.2 12.6 4.8 36.5 7.5 29.0 5.7 18.1 4.6
2.2 1.3 75.6 10.2 17.2 13.4 0.021 0.018 89.4 2.5 30.6 2.72 26.4 2.4 2.5 0.58 17.8 1.8 0.4 0.1 8.2 1.1 23.2 1.8 19.3 1.5 12.9 1.1
0.000 0.003 0.000 0.003 0.011 0.009 0.003 0.041 0.012 0.014 0.034 0.000 0.001 0.019
UF UF UF UF
NPS ¼ sites without point sources; UF ¼ sites with upstream fragmentation; ALL ¼ all sites.
less urban development and less crop lands). However, the density and severity of impoundments in the downstream section was greater (Table 4). Downstream fragmentation had great impact on IBI in the final model (Fig. 4).
4.
Discussion
This methodology proved to be very versatile and time-efficient when large, multi-parameter, environmental vectors are used for prediction of a target variable. The major advantageous difference with respect to more traditional approaches lies on the fact that the presented approach is able to allow easy, unbiased assessment of large, multi-dimensional
vectors composed of data of very different nature and measurement ranges such as concentrations of chemical compounds or discrete scores in the case of habitat quality. Because all environmental parameters are standardized prior to perform any IBI predictions, large site environmental vectors composed of parameters of very different nature and measurement range can be compared. Because all vector components are standardized a priori, each of them carries the same weight in the IBI prediction. Another big advantage over some commonly used, traditional prediction techniques such as regression is that model performance and speed is not affected by presence of highly cross-correlated variables since prediction is obtained by a mere comparison of site environmental parameters. Highly correlated variables do not affect
Table 4 e List of variables with significant differences between under-predicted sites and observations with a prediction within the ±1.5 3 RMSE range. Variable Name
Hard Mg SO4 Dsto_MLen L100_ForD L100_ForE L30_ForD L30_DevL L30_ForE LDA_ForD LDA_ForE R30_Crop R30_ForD R100M_ForD RDA_ForD
# of under/ well-predicted sites
Type of sites
11/213 11/213 11/213 22/331 27/374 27/374 27/374 27/374 27/374 27/374 27/374 27/374 27/374 27/374 27/374
PS PS PS DF ALL ALL ALL ALL ALL ALL ALL ALL ALL ALL ALL
Value in under-predicted sites (95% conf. interval) 247.0 20.9 135.2 1920.8 39.5 1.8 47.4 2.0 3.8 29.9 1.4 38.1 36.3 29.4 20.0
42.1 4.7 15.5 711.9 8.5 2.0 9.9 1.7 5.8 7.8 1.3 1.3 7.2 5.9 4.7
DF ¼ sites with downstream fragmentation; PS ¼ sites with point sources; ALL ¼ all site.
Value in well-predicted sites (95% conf. interval) 313.8 28.2 64.1 1194.8 26.4 0.4 30.6 6.0 0.4 17.8 0.4 49.1 23.2 19.3 12.9
13.8 1.6 25.1 157.5 2.4 0.2 2.7 1.0 0.2 1.8 0.1 2.5 1.8 1.5 1.1
p
0.033 0.046 0.042 0.025 0.005 0.002 0.002 0.038 0.000 0.000 0.000 0.025 0.000 0.000 0.002
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 3 5 9 e2 3 7 4
the outcome of the distance function because each parameter in the query site environmental vector is compared against the same exact environmental parameter in the rest of observations. Even though this methodology was not designed to “ordinate” environmental observations, the hierarchical tree used for prediction could be seen as a clustered distribution of similar sites, which is able to depict both explanatory (e.g. land use, water quality) and response variables (i.e. IBI) simultaneously. Traditional techniques such as Polar Ordination (PO) or NonMetric Multidimensional Scaling (NMDS) do not allow a simultaneous display of both explanatory variables (stressors) and response variables (IBI) on the same two-dimensional grid. Other widely used traditional ordination techniques such as Correspondence Analysis (CoA) allow a simultaneous display of variables as well (Giraudel and Lek, 2001). The site distribution using the presented methodology is obtained with no a priori relationships between the explanatory and response variables. Some multivariate ordination techniques such as Principal Component Analysis (PCA) or CoA assume linear relationships between the explanatory and the response variables which may not hold true in many cases, leading to well-known problems such as the horseshoe effect (PCA) or the arc effect (CoA) (Giraudel and Lek, 2001). The model confirmed biological integrity is the result of many inter-twined stressors of different nature acting at different scales. Out of the five main components of biological integrity (energy sources, water quality, habitat structure, flow regime, and biotic interactions) (Karr and Kerans, 1981; Karr et al., 1986; Karr, 1991), the first four were partially or fully represented in our database. At the study scale, only two groups of stressors were necessary to approximate the best variable combination for IBI prediction: regional land use and stream fragmentation at the basin-level. The relevance of these variables for IBI prediction was consistent with the geographic scale of the study, which had many sites scattered through a wide range of watersheds and within multiple basins. The relevance of the sampling strategy and geographic scale of the study area is paramount (Allan et al., 1997). At a specific scale, relevant variables in the highest possible level of the stressorresponse hierarchy reveal as best predictors of biological integrity. It has been proved that when IBI predictions are based on a wide array of observations from different watersheds and stream orders; regional scale variables will emerge as best predictors (Roth et al., 1996). Alternatively, if the study is based on similar types of observations with little regional environmental variability (e.g. same order streams in one watershed), more local variables will emerge as the most significant because the background regional quality for the group of sites is very homogeneous (Lammert and Allan, 1999).
4.1.
Land use
The model identified regional land use as one of the most important contributors to biological integrity at the statescale. Generally, the IBI prediction power of the dominant land uses was greater in the buffer zone than in the whole drainage area (Table 1). Hay/pasture and low intensity urban
2369
development were the only exceptions among dominant land uses. The effect of these on IBI was more evident in the whole drainage area, especially for hay/pasture. In Ohio, combinations of hay/pasture and deciduous forest (second and third most dominant land uses) were the most relevant to IBI. Surprisingly, the most abundant land use (i.e. cropland) was not part of the final model or the offstream variables sub-model. This was most likely due to negative, strong cross-correlations between the percentage of crops and deciduous forest. Agriculture and forest have been identified as important contributors to IBI variability (Roth et al., 1996; Wang et al., 1997; Wilson and Xenopoulos, 2008). A positive correlation between quality of fish assemblages and percentage of forested lands -which are negatively correlated to agriculture- has been reported. This correlation held true for both, the drainage area and the regional buffers (Wilson and Xenopoulos, 2008; Stewart et al., 2001). In most research efforts, different agricultural land uses such as cropland, range and pasture, orchards, or hay are usually merged into one category: agricultural lands (Anderson et al., 1976). In our research, agricultural land uses were not merged. The different sub-categories were kept as originally defined in the NLCD (USGS, 2008b). This revealed hay/pasture lands within the drainage area as a great predictor of biological integrity despite its smaller extent if compared to cropland (average cropland coverage equal to 56.1% versus 9.1% for hay/ pasture). Pasture and range lands in the drainage area have been associated with reduced vegetal cover, increased water temperature, nitrate, biomass concentrations, photosynthetic rates, and total suspended solids as well as increased fine sediment loading. A major shift in species composition of the macro-invertebrate community was also observed in areas with pasture lands (Quinn et al., 1997). The presence of rangeland is particularly harmful to aquatic fauna, especially in sites with poor riparian quality (Meador and Goldstein, 2003) and proved the most harmful to the aquatic community in the state of Ohio. Regional urban land uses played an overall smaller role on the integrity of Ohio streams. The dominant urban land uses (i.e. open space and low intensity development) were mostly relevant at the regional 100-m buffer (Fig. 4). This result agreed strongly with research negatively correlating urbanization along the stream buffers and stream integrity (Stewart et al., 2001; Wang et al., 2001; Morley and Karr, 2002). Urbanization seems to be significant at the local level as well. Medium intensity urbanization in the local buffers was the only local variable present in the final land use sub-model (see All LU model in Fig. 4). Medium intensity development was not a dominant land use in local buffers (2.21 and 1.76% in the 100and 30-m buffers respectively, versus 12.3 and 11.6%, and 5.9 and 5.51% of open space and low intensity urban lands respectively). Even though local open and low intensity developed lands were not selected in any model, this was most likely the consequence of strong correlation with their regional homologues (r ¼ 0.60 and 0.57 for open space and r ¼ 0.57 and 0.59 for low development in the 30- and 100-m buffers respectively). Nonetheless, new information introduced by the local medium intensity development could indicate that proximity of intense urbanization is an important factor to the site’s integrity (Wang et al., 2001; Morley and
2370
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 3 5 9 e2 3 7 4
Karr, 2002). Around 10e12% of connected imperviousness is considered the threshold beyond which biological quality declines rapidly in watersheds with small or no riparian buffers (Schueler, 1994; Wang et al., 2000, 2001). Selection of medium intensity development in local buffers by our model may indicate that this threshold has been reached. Another minor contributor in the offstream and final models was the presence of woody wetlands in the 30-m regional corridor. Percentage of this land use in the regional 30-m buffer was present in the final model and its extent in the drainage area was selected in the land use model (Fig. 4). Even though little new variability was explained by this land use, its presence is remarkable because of its little extent (mean percentages equal to a 0.33, 0.70, and 1.07% in the drainage area, 100- and 30-m regional buffers respectively). Woody wetlands seemed to gain importance with proximity to the stream (its individual-based predictive power ranked 12th out of 16 land uses in the drainage area, 9th out of 15 land uses in the 100-m regional buffer, and 7th out of 15 land uses in the regional 30-m buffer). A similar result was reported by Richards et al. (1996), who linked forested wetlands (mean extent equal to 10% in the drainage area) with increased presence of woody debris and other channel characteristics such as bankfull depth. Wetlands regulate surface water flow and site’s hydrology (Mitsch and Gosselink, 1986). Their presence is associated with decreased sediment input, nutrients, temperature, ionic strength, and increased resilience to disturbances (Richards et al., 1996; Detenbeck et al., 2000). Of special importance is the presence of wetlands near the receiving water body as the model indicated (30-m buffer was selected over drainage area in the final model). A decrease in wetland-stream distance has been positively correlated to reduced levels of nutrients, ions, and bacteria. Wetland extent has been correlated to decreased lead and high color in downstream lakes. This was found to be especially true in areas with highly fragmented riparian corridors (Johnston et al., 1990; Detenbeck et al., 1993, 2000). In the final model, two regional land use variables, positively-correlated to IBI were selected as final predictors when in very close proximity to the stream (i.e. in the 30-m buffer). On the other hand, negatively-correlated, regional land use variables were usually selected for the whole drainage area or for the 100-m buffer. Even though a definite conclusion may not be inferred from this fact, it may be an indication that preserving watershed-wide natural continuity along a stream’s immediate lands may help improve or maintain biological integrity when development occurs beyond these limits.
4.2.
Fragmentation
The negative effects of stream fragmentation to aquatic species have been widely studied (Reyes-Gavilan et al., 1996; Morita and Yamamoto, 2002; Morita and Yokota, 2002). Stream fragmentation and anthropogenic flow regulation affects a large percentage of streams worldwide, especially in developed countries (Dynesius and Nilsson, 1994; Nilsson et al., 2005). Stream fragmentation by dams has serious consequences for the biological community, preventing fish from reaching upstream habitats and isolating trapped upstream populations. Decreased species richness and risk of
extinction of native fauna through demographic, environmental, and genetic stochasticity are some of the consequences fragmented populations face (Morita and Yamamoto, 2002). Moreover, physical barriers are not the only consequence of dams. Usually, hydrologic changes are also associated to impoundments. Alteration of the natural flow regime affects fauna by eliminating or modifying natural habitat conditions, which may generate a shift in species composition and, therefore, biological integrity (Poff and Allan, 1995; Richter et al., 1996; Poff et al., 1997; Fischer and Kummer, 2000; Freeman et al., 2001; Gilvear et al., 2002). In our research, some fragmentation metrics had the largest individual IBI predicting power overall. This was especially true with metrics that accounted for both, the upstream and downstream habitats or the downstream habitat only. These variables were able to explain around 40% of the total IBI variability by themselves. Upstream fragmentation metrics had far less prediction power and were only relevant in some sites as shown in Table 3. A potential explanation is that most of the available observations were located well inland and far from the basin outlet (average stream distance to basin outlet ¼ 284.3 Km, minimum distance ¼ 18.35 km, maximum distance ¼ 833 Km). This could have influenced the overall model results since most of the available habitat was located in the downstream section. The fact that most of the available habitat in the available observations was located in the downstream section may have generated strong correlations between overall fragmentation metrics (i.e. metrics including both, upstream and downstream sections) and downstream-only metrics. However, and as mentioned in previous sections, model performance and speed is not negatively impacted by introducing strongly cross-correlated variables. Irrespective of this caveat, the model still selected an overall fragmentation metrics as the most powerful IBI predictor (Table 2), which is a clear indication of the paramount importance of habitat size and continuity on aquatic ecosystems. The impact of a fragmented upstream network was demonstrated when comparing fragmentation levels between well- and mispredicted sites with fragmented upstream networks (Table 3). Statistical differences were identified among these. No statistical differences in the size of these sites’ drainage areas were found.
4.3.
Point sources and instream water quality
Even though most of the nutrient-related parameters were among the best water-quality IBI predictors, nitrate and TP concentrations were not ranked among them, being TP the poorest predictor in the group. A clear relationship between nitrate concentration and IBI has not been found in Ohio. Only concentrations beyond 3e4 mg/L had consistently negative effects on IBI Rankin et al. (1999). The poor prediction power of phosphorus concentrations could be attributed to high concentrations beyond the biomass limiting-nutrient condition (Rankin et al., 1999). The second cluster of variables was composed of ionic strength-related parameters (Mg, Hard, Cl, Cond, SO4). The third and last group was composed of metal concentrations (Zn, Cd, Fe, Cu, Pb) with the exception of arsenic which had the third highest individual predictive
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 3 5 9 e2 3 7 4
power of all available chemicals. Other variables such as DO, TSS, or pH had very low predictive power. The first two variables selected in the water Quality Model (BOD and NO2, Fig. 4) indicate that nutrient input is the main water quality contributor to biological degradation at the study scale. BOD has been identified as a significant source of degradation in Ohio streams (Dyer et al., 2000; Norton et al., 2000, 2002) and is an indication of highly eutrophic conditions. The higher predicting power of BOD and several nutrient-related parameters clearly indicate that eutrophication processes have a significant impact on IBI. The most significant impact of eutrophication on aquatic fauna occurs in the ultimate or “collapse” phase in which oxygen is depleted because it is used to fuel decomposition of massive amounts of decaying algae or phytoplankton. Concentrations of DO prior to the ultimate eutrophication phase fluctuate on a daily basis based on algae photosynthetic or respiration processes (Novotny, 2003). For this reason, for eutrophic systems that have not yet reached system collapse, DO may not always be a good predictor of biological integrity as the model identified. Even though samples were collected during summer months or in early Fall (period in which environmental conditions will be more favorable for algae blooms in Ohio), it is unlikely that all sites with an excess nutrient input were in the ultimate phase of eutrophication. The third selected variable in the model was cadmium concentration, which provided marginal improvement (see Fig. 4). Metal toxicity is indeed a powerful agent of biological degradation. However, it is only able to explain a significant part of the overall IBI variability at smaller scales such as the upper or lower parts of a watershed (Dyer et al., 2000). This is most likely a consequence of its highly localized nature (i.e. coming from point sources or legacy pollution). None of the chemical variables were included in the subsequent Instream variables model. Habitat and sometimes water quality -especially if related to nutrient input- are mostly driven by local and regional land uses. Therefore, in sites with severely impaired habitats (e.g. with a high level of fine sediment due to accelerated denudation processes), the most likely cause of poor water quality is non-point source pollution (i.e. chemicals attached to flushed particles in runoff). This could explain why water quality data did not provide any new information when merged with the habitat model at the study scale. Point source density and intensity only had a significant impact at the local scale as expected. When outliers were removed, significant differences were not identified between well- and mispredicted observations with reported point sources (therefore, not included in Table 3 or Table 4). Significant differences in water quality were only found in some under-predicted sites with respect to well-predicted sites. Lower hardness levels and higher sulfate concentrations were observed in under-predicted observations (Table 4). From the results, the overall effect of point source pollution on IBI is very small compared to other more ubiquitous stressors directly or indirectly related to land use at the scale of our study. Point sources are a significant factor if they have a significant presence in the area of study. For example, point source pollution has been identified as a significant negative factor in some studies based in only one basin or a portion of it (Dyer et al., 1998a, 2000). Another study based in the whole
2371
state of Ohio using only habitat and water quality data, concluded that water quality had a significant impact on IBI in specific clusters of sites only; but confirmed that the most relevant instream parameters at the state-scale were habitatrelated (Manolakos et al., 2007).
4.4.
Instream habitat
Instream habitat and drainage area were able to explain 49% of the overall IBI variability. Substrate-related metrics (i.e. embeddedness and substrate quality), stream variability (i.e. pool and riffle quality), and vegetal cover were the most relevant QHEI metrics. Habitat quality has been identified as the main instream source of IBI variability (Hall et al., 1996; Dyer et al., 1998a; Manolakos et al., 2007). Habitat quality is strongly driven by land use changes in the drainage area and may account for land use-related water quality information such as nutrient input. Our model confirmed this point and the Habitat model selected exactly the same variables as the Instream model (Fig. 4). Riffle and Cover qualities were selected in the Overall model but with very modest contributions to the final outcome. Stream variability, substrate quality, and/or instream cover have been identified as significant contributors to biotic quality in Ohio (Dyer et al., 1998a, 1998b; Yuan and Norton, 2004; Manolakos et al., 2007) and elsewhere (Minshall, 1984; Quinn and Hickey, 1990; Richards et al., 1993; Rabeni and Smale, 1995). Drainage area was positively correlated to IBI, which strongly agreed with the findings by Dyer et al. (1998a) in Ohio.
4.5.
Mispredictions due to local conditions
Two main sources of IBI overprediction due to local environmental conditions were identified. The first source consisted of higher levels of upstream fragmentation in sites with fragmented upstream networks (Table 3). The second source, which affected all observations, was local land use patterns not included in the final prediction model. Over-predicted observations had significantly higher percentages of forested areas in the drainage area and regional buffer corridor (Table 3). This contributed to high calculated IBI scores. The extent of forested land in the local catchment and buffer zones were also significantly higher in over-predicted sites while medium intensity urbanization in the local buffer was lower. These results were counter-intuitive given the lower observed IBI scores in sites with such good ‘land use quality’. However, these sites had significantly higher percentages of hay/pasture lands in the local catchment area. This land use was identified as the most detrimental to IBI and could explain the overpredictions (i.e. sites with very good regional characteristics, which were included in the final model, but significantly higher levels of pasture lands at the local scale, which was not included in the final model). Significant differences of pasture lands at the regional scale were not present in over-predicted sites. On the other hand, under-predicted sites had significantly better ‘land use quality’ at both scales as well. Therefore, exceptional local land use quality (i.e. significantly higher levels of forested areas combined with smaller percentages of urban and crop lands, see Table 4) is the most likely cause of
2372
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 3 5 9 e2 3 7 4
IBI under-prediction. The final model didn’t include information from local land use variables.
5.
Conclusions
We presented a highly versatile, predictive modeling methodology which has many potential applications in the environmental data analysis field. The new methodology is capable of dealing with a large number of observations with many associated attributes in a very time-efficient manner. Moreover, one of the main contributions of such methodology is that it does not rely on a-priori assumption on the relationship between the environmental variables and the prediction target. Many traditional exploratory techniques in the ecological modeling field make such kind of assumptions (e.g. canonical correspondence analysis is based on linear regressions between the explanatory and response variables). These two main features are of paramount importance because natural systems are composed of many inter-twined, cross-correlated variables with a highly non-linear inter-dependencies. Furthermore, the model performs well when discrete or crudely scaled data is used because it is based on assessing environmental similarities among sites with the same attributes. As a result, it allows using variables with different scoring criteria at once. At the state-scale, regional land use and basin-level stream fragmentation are the main predictors of biotic integrity in Ohio. Habitat variables only contributed marginally to model improvement, while instream water quality and point source intensity and density were not able to improve the final model at all. Most of the information from instream water and habitat qualities is introduced into the model by regional land use, which acts as a surrogate variable. We revealed the importance of local stressors which were not accounted for in the final model. Over-predictions mainly came from a combination of higher upstream fragmentation, extreme point source density and intensity, and high levels of hay/pasture in the local catchment area. under-predictions mainly came from extraordinary local land use quality which was not accounted for in the model. If the 55 mispredicted sites eout of 429 observations- could be disregarded due to unique local conditions, the model would explain 86% of the overall IBI variability. Therefore, in our dataset local stressors accounted for an extra 25% (i.e.86%e61%) of the variability explained by land use and fragmentation metrics. The remaining 14% may be due to sampling errors, data quality issues, or natural randomness (for example, a site with BOD ¼ 24 mg/L; TKN ¼ 3.1 mg/L; TP ¼ 1.29 mg/L; Zn ¼ 180 mg/L; Cu ¼ 39 mg/L; Fe ¼ 19,700 mg/L; or NO2 ¼ 0.19 mg/L had one of the highest observed IBI scores (50)). The results showed how water quality issues from point sources have a small overall impact on the biotic integrity in Ohio. This may indicate a successful control of point sources through the EPA’s NPDES Program, which have been top priority for U.S. surface waters since the Clean Water Act of 1972. These results do not indicate that water quality problems from point sources are not relevant anymore in
Ohio, but they have shifted from being a widespread issue to a local one at the state-scale. Our model identified stream fragmentation and land use change - especially in the regional buffers- as the most important stressors to biological integrity. Habitat degradation and nutrient input are the most direct instream consequences from land use disturbances. Results suggest that in order to achieve the aimed physical, chemical, and biological integrity of the Nation’s waters, protection and enforcing policies have to refocus towards a more holistic view that goes beyond the traditional point source control. Ecosystem continuum must be kept and watershed- or basin-level- land use planning is necessary to attain such goals, especially in the most immediate lands of a water body. Stressors should be approached in a scale-down manner. This would guarantee that improvements at the local level have successful outcomes because the regional background conditions meet the minimum requirements to attain the targeted integrity.
Acknowledgement This research has been partially supported by the US EPA/NSF/ USDA STAR Watershed Program, Grant No. R83-0885-010 to Northeastern University, Boston, MA. The authors would like to express their gratitude towards Mr. Ed Rankin and Mr. Dennis Mishne from Ohio EPA and Mr. Scott Dyer and Ms. Charlotte White-Hull from The Procter & Gamble Company for their help with the environmental data and valuable advice.
references
Allan, J.D., Erickson, D.L., Fay, J., 1997. The influence of catchment land use on stream integrity across multiple spatial scales. Freshwater Biology 37, 149e161. Anderson, J.R., Harvey, E.H., Roach, J.T., Whitman, R.E., 1976. A Land Use and Land Cover Classification System for Use with Remote Sensor Data. Geological Survey Professional Paper 964. U.S. Government Printing Office, Washington D.C. Bedoya, D., Manolakos, E.S, Novotny, V., 2010. Characterization of biological responses under different environmental conditions- a hierarchical modeling approach. Ecological Modelling. Ecological Modelling 222 (3), 532e545. Beyer, H.L., 2004. Hawth’s Analysis Tools for ArcGIS. Available at. http://www.spatialecology.com/htools. Bode, R.W., 1988. Methods for Rapid Biological Assessment of Streams. New York State Department of Environmental Conservation, Albany, NY. Castelle, A.J., Johnson, A.W., Connolly, C., 1994. Wetland and stream buffer size requirements - a review. Journal of Environmental Quality 23, 878e882. Detenbeck, N.E., Batterman, S.L., Brady, V.J., Brazner, J.C., Snarski, V.M., Taylor, D.L., Thompson, J.A., Arthur, J.W., 2000. A test of watershed classification systems for ecological risk assessment. Environmental Toxicology and Chemistry 19, 1174e1181. Detenbeck, N.E., Johnston, C.A., Niemi, G.J., 1993. Wetland effects on lake water-quality in the Minneapolis St-Paul metropolitan-area. Landscape Ecology 8, 39e61.
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 3 5 9 e2 3 7 4
Dyer, S.D., White-Hull, C., Carr, G.J., Smith, E.P., Wang, X.H., 2000. Bottom-up and top-down approaches to assess multiple stressors over large geographic areas. Environmental Toxicology and Chemistry 19, 1066e1075. Dyer, S.D., White-Hull, C., Wang, X., Johnson, T.D., Carr, G.J., 1998a. Determining the influence of habitat and chemical factors on instream bioitc integrity for a Southern Ohio watershed. Journal of Aquatic Ecosystem Stress and Recovery 6, 91e110. Dyer, S.D., White-Hull, C.E., Johnson, T.D., Carr, G.J., Wang, X., 1998b. The importance of space in understanding the risk of multiple stressors on the biological integrity of receiving waters. Journal of Hazardous Materials 61, 37e41. Dynesius, M., Nilsson, C., 1994. Fragmentation and flow regulation of river systems in the Northern 3rd of the world. Science 266, 753e762. Fischer, S., Kummer, H., 2000. Effects of residual flow and habitat fragmentation on distribution and movement of bullhead (Cottus gobio L.) in an alpine stream. Hydrobiologia 422, 305e317. Freeman, M.C., Bowen, Z.H., Bovee, K.D., Irwin, E.R., 2001. Flow and habitat effects on juvenile fish abundance in natural and altered flow regimes. Ecological Applications 11, 179e190. Frissell, C.A., Liss, W.J., Warren, C.E., Hurley, M.D., 1986. A hierarchical framework for stream habitat classification: viewing streams in a watershed context. Environmental Management 10, 199e214. Gilvear, D.J., Heal, K.V., Stephen, A., 2002. Hydrology and the ecological quality of Scottish river ecosystems. Science of the Total Environment 294, 131e159. Giraudel, J.L., Lek, S., 2001. A comparison of self-organizing map algorithm and some conventional statistical methods for ecological community ordination. Ecological Modelling 146, 329e339. Hall, L.W., Scott, M.C., Killen, W.D., Anderson, R.D., 1996. The effects of land-use characteristics and acid sensitivity on the ecological status of Maryland coastal plain streams. Environmental Toxicology and Chemistry 15, 384e394. Jain, A.K., Murty, M.N., Flynn, P.J., 1999. Data clustering: a review. ACM Computer Surveys 31, 264e323. Johnston, C.A., Detenbeck, N.E., Niemi, G.J., 1990. The cumulative effect of wetlands on stream water quality and quantity. A landscape approach. Biogeochemistry 10, 105e141. Karr, J.R., 1991. Biological Integrity: a long-neglected aspect of water resource management. Ecological Applications 1, 66e84. Karr, J.R., Fausch, K.D., Angermeier, P.L., Yant, P.R., Schlosser, I.J., 1986. Assessing Biological Integrity of Running Waters: a Method and Its Rationale. Illinois Natural History Survey, Champaign, IL. Karr, J.R., Kerans, B.L., 1981. Components of biological integrity: their definition and use in development of an invertebrate IBI. 1991 MidWest Pollution Control Biologists Meeting. In: Environmental Indicators: Measurement and Assessment Endpoints. U.S. Environmental Protection Agency, Lincolnwood, IL. Lammert, M., Allan, J.D., 1999. Assessing biotic integrity of streams: effects of scale in measuring the influence of land use/cover and habitat structure on fish and macroinvertebrates. Environmental Management 23, 257e270. Lyons, J., 2006. A fish-based index of biotic integrity to assess intermittent headwater streams in Wisconsin, USA. Environmental Monitoring and Assessment 122, 239e258. Lyons, J., Piette, R.R., Niermeyer, K.W., 2001. Development, validation, and application of a fish-based index of biotic integrity for Wisconsin’s large warmwater rivers. Transactions of the American Fisheries Society 130, 1077e1094. Manolakos, E., Virani, H., Novotny, V., 2007. Extracting knowledge on the links between the water body stressors and biotic integrity. Water Research 41, 4041e4050.
2373
Meador, M.R., Goldstein, R.M., 2003. Assessing water quality at large geographic scales: Relations among land use, water physicochemistry, riparian condition, and fish community structure. Environmental Management 31, 504e517. Minshall, G.W., 1984. Aquatic-insect substratum relationships. In: Resh, V.H., Rosenberg, D.M. (Eds.), The Ecology of Aquatic Insects. Praeger Scientific, New York NY, pp. 358e400. Mitsch, W.J., Gosselink, J.G., 1986. Wetlands. Van Nostrand Reinhold, New York, NY. Morita, K., Yamamoto, S., 2002. Effects of habitat fragmentation by damming on the persistence of stream-dwelling charr populations. Conservation Biology 16, 1318e1323. Morita, K., Yokota, A., 2002. Population viability of streamresident salmonids after habitat fragmentation: a case study with white-spotted charr (Salvelinus leucomaenis) by an individual based model. Ecological Modelling 155, 85e94. Morley, S.A., Karr, J.R., 2002. Assessing and restoring the health of urban streams in the Puget Sound basin. Conservation Biology 16, 1498e1509. Nilsson, C., Reidy, C.A., Dynesius, M., Revenga, C., 2005. Fragmentation and flow regulation of the world’s large river systems. Science 308, 405e408. Norton, S.B., Cormier, S.M., Smith, M., Jones, R.C., 2000. Can biological assessments discriminate among types of stress? A case study from the Eastern Corn Belt Plains ecoregion. Environmental Toxicology and Chemistry 19, 1113e1119. Norton, S.B., Cormier, S.M., Smith, M., Jones, R.C., SchubauerBerigan, M., 2002. Predicting levels of stress from biological assessment data: Empirical models from the Eastern Corn Belt Plains, Ohio. USA. Environmental Toxicology and Chemistry 21, 1168e1175. Novotny, V., Bartŏsova´, A., O’Reilly, N., Ehlinger, T., 2005. Unlocking the relationship of biotic integrity of impaired waters to anthropogenic stresses. Water Research 39, 184e198. Novotny, V., 2003. Water Quality. Diffuse Pollution and Watershed Management. John Wiley & Sons, New York. Ohio Environmental Protection Agency, 1987. Biological Criteria for the Protetction of Aquatic Life: Volume I-III. In: Standardized Field and Laboratory Methods for Assessing Fish and Macroinvertebrate Communities. Division of Water Quality Monitoring and Assessment, Surface Water Section, Columbus, OH. Ohio Environmental Protection Agency, 2005. Surface and Groundwater Monitoring and Assessment Strategy 2005e2009. Division of Surface Water and Division of Drinking and Ground Waters. Ohio EPA Technical Bulletin EAS/2005-09-1. Ohio Environmental Protection Agency, 2008. Statewide Biological and Water Quality Monitoring & Assessment. Available at: http://www.epa.state.oh.us/dsw/bioassess/ ohstrat.html. Ohio EPA, Division of Surface Water, Columbus, OH. Pickett, S.T.A., Kolasa, J., Armesto, J.J., Collins, S.L., 1989. The ecological concept of disturbance and its expression at various hierarchical levels. OIKOS 54, 129e136. Poff, N.L., 1997. Landscape filters and species traits: towards mechanistic understanding and prediction in stream ecology. Journal of the North American Benthological Society 16, 391e409. Poff, N.L., Allan, J.D., 1995. Functional-Organization of stream fish assemblages in Relation to Hydrological variability. Ecology 76, 606e627. Poff, N.L., Allan, J.D., Bain, M.B., Karr, J.R., Prestegaard, K.L., Richter, B.D., Sparks, R.E., Stromberg, J.C., 1997. The natural flow regime. Bioscience 47, 769e784. Quinn, J.M., Cooper, A.B., Davies-Colley, R.J., Rutherford, J.C., Williamson, R.B., 1997. Land use effects on habitat, water quality, periphyton, and benthic invertebrates in Waikato, New Zealand, hill-country streams. New Zealand Journal of Marine and Freshwater Research 31, 579e597.
2374
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 3 5 9 e2 3 7 4
Quinn, J.M., Hickey, C.W., 1990. Magnitude of effects of substrate particle size, recent flooding, and watershed development on benthic invertebrates in 88 New Zealand rivers. New Zealand Journal of Marine Freshwater Resources 24, 411e428. Rabeni, C.F., Smale, M.A., 1995. Effects of Siltation on stream Fishes and the potential Mitigating role of the buffering riparian zone. Hydrobiologia 303, 211e219. Rankin, E.T., 1989. The Qualitative Habitat Evaluation Index (QHEI): Rationale, Methods, and Application. Ecological Assessment Section, Division of Water Quality, Planning, and Assessment. Ohio Environmental Protection Agency, Columbus, OH. Rankin, E.T., Miltner, B., Yoder, C.O., Mishne, D., 1999. Association between Nutrients, Habitat, and the Aquatic Biota in Ohio Rivers and Streams Ohio EPA Technical Bulletin MAS/1999-11, Columbus, OH. Reyes-Gavilan, F.G., Garrido, R., Nicieza, A.G., Toledo, M.M., Brana, F., 1996. Fish community variation along physical gradients in short streams of northern Spain and the disruptive effect of dams. Hydrobiologia 321, 155e163. Richards, C., Host, G.E., Arthur, J.W., 1993. Identification of predominant environmental-factors structuring stream macroinvertebrate communities within a large agricultural catchment. Freshwater Biology 29, 285e294. Richards, C., Johnson, L.B., Host, G.E., 1996. Landscape-scale influences on stream habitats and biota. Canadian Journal of Fisheries and Aquatic Sciences 53, 295e311. Richter, B.D., Baumgartner, J.V., Powell, J., Braun, D.P., 1996. A method for assessing hydrologic alteration within ecosystems. Conservation Biology 10, 1163e1174. Roth, N., Southerland, M., Chaillou, J., Klauda, R., Kazyak, P., Stranko, S., Weisberg, S., Hall, L., Morgan, R., 1998. Maryland biological stream survey: development of a fish index of biotic integrity. Environmental Monitoring and Assessment 51, 89e106. Roth, N.E., Allan, J.D., Erickson, D.L., 1996. Landscape influences on stream biotic integrity assessed at multiple spatial scales. Landscape Ecology 11, 141e156.
Schueler, T., 1994. The importance of imperviousness. Watershed Protection Techniques 1, 100e111. Stewart, J.S., Wang, L.Z., Lyons, J., Horwatich, J.A., Bannerman, R., 2001. Influences of watershed, riparian-corridor, and reachscale characteristics on aquatic biota in agricultural watersheds. Journal of the American Water Resources Association 37, 1475e1487. U.S. Army Corps of Engineers, 2005. National Inventory of Dams. Available at. http://crunch.tec.army.mil/nidpublic/webpages/ nid.cfm. U.S.Environmental Protection Agency, 2008. Permit Compliance System Database. Available at. http://epa.gov/enviro/html/ pcs/pcs_query_java.html. U.S. Geological Survey, 2008a. National Hydrography Dataset. Available at. U.S. Department of the Interior. http://nhd.usgs. gov/index.html. U.S. Geological Survey, 2008b. National Land Cover Database. Available at. Multi-resolution Land Characteristics Consortium U.S. Department of the Interior. http://www.mrlc. gov/index.php. Wang, L., Lyons, J., Kanehl, P., Bannerman, R., 2001. Impacts of urbanization on stream habitat and fish across multiple spatial scales. Environmental Management 28, 255e266. Wang, L.Z., Lyons, J., Kanehl, P., Bannerman, R., Emmons, E., 2000. Watershed urbanization and changes in fish communities in southeastern Wisconsin streams. Journal of the American Water Resources Association 36, 1173e1189. Wang, L.Z., Lyons, J., Kanehl, P., Gatti, R., 1997. Influences of watershed land use on habitat quality and biotic integrity in Wisconsin streams. Fisheries 22, 6e12. Wilson, H.F., Xenopoulos, M.A., 2008. Landscape influences on stream fish assemblages across spatial scales in a northern Great Plains ecoregion. Canadian Journal of Fisheries and Aquatic Sciences 65, 245e257. Yuan, L.L., Norton, S.B., 2004. Assessing the relative severity of stressors at a watershed scale. Environmental Monitoring and Assessment 98, 323e349.
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 3 7 5 e2 3 8 2
Available at www.sciencedirect.com
journal homepage: www.elsevier.com/locate/watres
Carbon cycling in a zero-discharge mariculture system Kenneth Schneider a,b,*, Yonatan Sher a,1, Jonathan Erez b, Jaap van Rijn a a
The Hebrew University of Jerusalem, Faculty of Agricultural, Food and Environmental Quality Sciences, Department of Animal Sciences, P.O. Box 12, Rehovot 76100, Israel b The Hebrew University of Jerusalem, The Institute of Earth Sciences, Edmond Safra Campus, Givat Ram, Jerusalem 91904, Israel
article info
abstract
Article history:
Interest in mariculture systems will rise in the near future due to the decreased ability of
Received 16 September 2010
the ocean to supply the increasing demand for seafood. We present a trace study using
Received in revised form
stable carbon and nitrogen isotopes and chemical profiles of a zero-discharge mariculture
8 December 2010
system stocked with the gilthead seabream (Sparus aurata). Water quality maintenance in
Accepted 25 January 2011
the system is based on two biofiltration steps. Firstly, an aerobic treatment step comprising
Available online 21 February 2011
a trickling filter in which ammonia is oxidized to nitrate. Secondly, an anaerobic step comprised of a digestion basin and a fluidized bed reactor where excess organic matter and
Keywords:
nitrate are removed. Dissolved inorganic carbon and alkalinity values were higher in the
Zero-discharge mariculture
anaerobic loop than in the aerobic loop, in agreement with the main biological processes
Carbon stable isotopes
taking place in the two treatment steps. The d13C of the dissolved inorganic carbon (d13CDIC)
Alkalinity
was depleted in 13C in the anaerobic loop as compared to the aerobic loop by 2.5e3&. This
DIC
is in agreement with the higher dissolved inorganic carbon concentrations in the anaerobic loop and the low water retention time and the chemolithotrophic activity of the aerobic loop. The d13C and d15N of organic matter in the mariculture system indicated that fish fed solely on feed pellets. Compared to feed pellets and particulate organic matter, the sludge in the digestion basin was enriched in
15
N while d13C was not significantly different. This
latter finding points to an intensive microbial degradation of the organic matter taking place in the anaerobic treatment step of the system. ª 2011 Elsevier Ltd. All rights reserved.
1.
Introduction
The oceans supply of fish is stagnant and it is expected to decrease in the near future due to overfishing and biodiversity loss (Jackson et al., 2001; Worm et al., 2006). As such, it is expected that mariculture systems will have a significant role in the global fish supply because of the growing demand for fish (Tidwell and Allan, 2001). Mariculture systems in coastal areas (such as floating cages) or coastal, land-based ponds impose environmental concerns due to their contamination of
coastal waters with organic matter and nutrients (Wu, 1995). Such contamination brings about environmental alterations such as anoxic sediments that produce toxic H2S (Holmer and Kristensen, 1992), eutrophication, and a resulting decrease in biodiversity and biomass of the benthic communities (Mazzola et al., 1999; Karakassis et al., 2000). In the present study, a zero-discharge recirculation system, first developed for freshwater fish farming (van Rijn, 1996; Shnel et al., 2002) and later converted to a system for culture of marine fish (Gelfand et al., 2003) was examined.
* Corresponding author. Present address: Department of Global Ecology, Carnegie Institution, 260 Panama street, Stanford, CA 94305, USA. 1 Current address: The Jacob Blaustein Institute for Desert research, The Ben Gurion University of the Negev, Sede Boqer Campus, 84990, Israel. 0043-1354/$ e see front matter ª 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.watres.2011.01.021
2376
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 3 7 5 e2 3 8 2
Water quality in the system is maintained by recirculating the culture water through two main treatment loops: an aerobic loop consisting of a trickling filter (TF) used for oxidation of ammonia to nitrate by nitrifying bacteria (Chen et al., 2006; Eding et al., 2006) and an anaerobic loop, consisting of a digestion basin (DB) and a fluidized bed reactor (FBR), in which, among other processes, organic matter is digested and nitrate respired to elemental nitrogen (van Rijn et al., 2006). The high organic load in the DB creates a redox gradient that facilitates conditions not only for nitrate reduction but also for sulfate reduction to harmful sulfide. It was demonstrated that while in the DB some of the produced sulfide was reoxidized to sulfate by autotrophic denitrifiers (Sher et al., 2008), sulfide oxidation by these organisms was particularly evident in the FBR, which served as a final sulfide polishing stage before water from the anaerobic treatment step was returned back to the fish basin (Cytryn et al., 2005). Stable isotopes can be used to trace natural processes on the basis of their distribution as different biological processes result in different isotopic fractionation. Processes such as photosynthesis and chemosynthesis are associated with a large discrimination against 13C resulting in low d13C values of 11 to 35& depending on the process type and the enzymes associated with them (Guy et al., 1993; Goericke and Fry, 1994; Robinson et al., 2003). Contrary to other processes, carbon stable isotopes have little fractionation in the biological food web. It was observed that with each trophic level a slight increase in d13C of about 0.8 1.1& takes place; a phenomenon known as “you are what you eat 1&” (DeNiro and Epstein, 1978). As opposed to carbon, nitrogen isotopes have a larger fractionation with each trophic level and increases in d15N by about 3 2.6& have been reported (DeNiro and Epstein, 1981). The distinctive fractionation of carbon and nitrogen stable isotopes has been used as tracers in food web studies in natural systems and in engineered systems using mixing models based on mass balance consideration (Schroeder, 1983; Fry and Sherr, 1984). In this study we present a description of the carbon cycling in a zero-discharge mariculture system based on carbon and nitrogen stable isotopes and chemical parameter in the water and sludge phases of a zero-discharge mariculture system through measurement of changes in values of d13C and d15N and changes in concentrations of NH3, NO 3 , H2S, alkalinity, dissolved inorganic carbon (DIC), redox potential and dissolved oxygen (DO).
2.
Materials and methods
2.1.
General description of the mariculture system
a rate of 10 m3 h1. In addition to the TF, the aerobic compartment comprised a foam fractionator (FF), which received water from the trickling filter basin (TFB). The particulate organic matter captured by the FF was discharged into the digestion basin (DB). This latter basin (volume: 5.4 m3) was part of the anaerobic treatment compartment. By gravitation, water from the bottom of the FB was led (0.8 m3 h1) into this latter basin. Effluent water from the DB was collected in an intermediate collection basin (ICB) before being returned, by gravitation, to the TFB. Water from the ICB was recirculated (0.8 m3 h1) through a fluidized bed reactor (FBR, volume:13 L) which effluent water was led through a swirl separator (SS) before being returned to the ICB. Sludge captured by the SS, was discharged into the DB. Previous studies on this and similar systems revealed that nitrification (aerobic treatment loop), digestion of organic matter together with nitrate and sulfate respiration (digestion basin), and microbial sulfide oxidation (FBR), were major processes affecting the overall water quality in the system (van Rijn et al., 1995).
2.2.
Sampling regime
2.2.1.
Sampling frequency and locations
The mariculture system was sampled on three separate occasions between July and December 2006. Samples were withdrawn from nine different locations within the system (Fig. 1): (1) fish basin (FB), (2) trickling filter collection basin (TFB), (3) effluent water from the trickling filter (TFout), (4) influent water to the trickling filter (TFin), (5) top layer (5 cm depth) of influent zone in the digestion basin (DBinT), (6) bottom layer (30 cm depth) of influent zone in the DB (DBinB), (7) top layer (5 cm depth) of effluent zone in the DB (DBoutT), (8) bottom layer (30 cm depth) of effluent zone in the DB (DBoutB), and (9) fluidized bed reactor (FBR).
2.2.2.
2.2.3. The intensive fish mariculture system was an enlarged version of the system previously described by Gelfand et al. (2003). The system (Fig. 1) comprised a fish basin (5 m3) stocked with the gilthead seabream (Sparus aurata) from which water was recirculated through aerobic and anaerobic treatment compartments. The aerobic compartment consisted of a trickling filter (TF) with a volume of 8 m3 and a surface area of 1920 m2. Surface water from the fish basin (FB) was recirculated through the aerobic compartment at
Sampling procedure
Water from each sampling point was collected in a 1.5 L plastic bottle and was initially filtered through cotton gauze with a mesh size of several mm for removal of agglomerated, floating sludge particles. Water for chemical and isotopic analysis was further filtered with a GF/F or GF/C before storage as described below. Water for DB chemical profiling of the DB was collected in 50 ml vials as described below (treatment of sludge samples). The vials were centrifuged immediately and the water and sludge where separated for different analysis. Water for measuring sulfide was transfer to 10 ml vial under nitrogen environment and the sulfide was fixed immediately with zinc acetate (Strocchi et al., 1992).
Treatment of water samples
Immediate after collection the pH, redox and inorganic 2 nitrogen species (NO 3 , NO2 and NH3) and sulfide (S ) were analyzed as described below. Samples for alkalinity, dissolved inorganic carbon (DIC) and carbon stable isotopes of DIC (d13CDIC) were stored in 60 mL brown glass bottles with gastight screws and refrigerated (4 C) until measurements. DIC and d13CDIC samples were poisoned with 0.6 mL (1% v:v) of saturated HgCl2 solution immediate after sampling. Particulate organic matter (POM) from a known water volume was
2377
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 3 7 5 e2 3 8 2
A
Aerobic loop
Anaerobic loop
TFin
FB
DB
FF
FBR
TFout
ICB
S
TF
TFB
-
DBinT DBinB
Digestion Basin (DB)
Pump
B
DBin Inlet
DBout Outlet DBoutT DBoutB
Center
Sludge
Fig. 1 e Schematic design of the zero-discharge fish mariculture system (A). Detailed diagram of the digesting basin (B). Abbreviations are as follows: FB [ fish basin, TF [ trickling filter, TFB [ trickling filter basin, TFin [ influent water to the trickling filter, TFout [ effluent water from the trickling filter, FF [ foam fractionator, DB [ digestion basin, DBinT [ top layer (5 cm depth) of influent zone in the digestion basin, DBinB [ bottom layer (30 cm depth) of influent zone in the DB, DBoutT [ top layer (5 cm depth) of effluent zone in the DB, DBoutB [ bottom layer (30 cm depth) of effluent zone in the DB, FBR [ fluidized bed reactor, S [ swirl separator and ICB [ intermediate collection basin.
collected on combusted (450 C for 2 h) GF/F filters and dried at 30 C for about 48 h.
2.2.4.
Treatment of sludge samples
Sludge from the DB was sampled from three places (inlet, center and outlet) at 2e4 depths (12 cm apart) depending on the sludge depth. The samples were collected from different depths using a sampling device consisting of a 50 ml vial attached to a scaled pole. Samples were collected by lowering the closed vial to the desired depth and lifting the lid of the vials for a few seconds to allow filling of the vial. When possible, sludge was collected from the wall of the TFB. About 2e3 g of wet sludge was dried at 30 C for about 48 h.
2.3.
Chemical analysis
NH3 and NHþ 4 , referred to as total ammonia nitrogen (TAN), were determined by oxidation with salicylate-hypochlorite
method (Bower and Holm-Hansen, 1980). Nitrite was determined by reaction with sulfanilamide (Strickland and Parsons, 1972). Nitrate was measured according to the light absorption at two wave lengths 220 and 275 nm (APHA, 1998). When sulfide was present in the samples, samples were diluted with an HCl (0.1 N) solution in 1:1 proportion and nitrate was measured at an additional wave length of 250 nm accounting for the HS remaining in the solution (Sher et al., 2008). Sulfide was determined by the methylene blue method (Cline, 1969). Total alkalinity was determined by titration with hydrochloric acid and calculated according to the Gran titration method (Grasshoff et al., 1983). Measurements of pH were conducted with a Radiometer Copenhagen pH meter (PHM92 Research pH Meter) and a Radiometer Copenhagen combination electrode (GK2401c). The electrode was calibrated using NBS scale standard buffers (Radiometer analytical) of 7.000 and 10.012. DO and temperature measurements at different sites at the mariculture system were conducted by means of an oxygen
2378
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 3 7 5 e2 3 8 2
electrode combined with a temperature probe (OxyGuard, H01c Handy Gamma). Salinity was monitored with a refractometer (model: S-10E, Atago, Tokyo, Japan).
significance was set at a level of 5% or less. Statistica 6 software (StatSoft) was used for statistical calculations.
2.3.1.
3.
Stable isotopes
Stable isotopes are measured relative to an international standard (eq. (1)) in per mil (&). d13 C or d15 Nð&Þ ¼
Rsample 1 1000 Rstandard
3.1. In situ measurements within the mariculture system
(1)
Where d13C and d15N are the values measured for C and N, respectively and R is the ratio of the heavier isotope to the light isotope (13C/12C and 15N/14N). For each element, we used a commonly used international standard. For C, this standard is PDB (Pee Dee Belemnite-marine limestone) and for N, atmospheric air is used as the international standard.
2.3.2.
Results
3.1.1.
The inorganic carbon system
The DIC, alkalinity and the d13CDIC showed distinctive differences between the aerobic (TF-FB) and anaerobic (DB-FBR-FB) loops (Fig. 2). DIC in the aerobic loop was in the range of 2900e3900 mmol L1 with an average of 3400 385 mmol L1, and was lower than in the anaerobic loop which was in the range of 4200e5200 mmol L1 with an average of
d13CDIC and DIC measurements
1-mL samples were injected into a 10-mL vial with a gas-tight screw filled with He gas at atmospheric pressure. Ten drops of H2PO4 (85%) were added and the vials were left for 24 h to equilibrate at 25 C in a temperature controlled sample tray (Finnigangasbench, Thermo Electro cooperation, USA). In each sample d13CDIC was measured eight times using an autosampler gas bench system connected online to an isotope ration mass spectrometer (IRMS; 252 mat, Finnegan) and the d13CDIC was averaged. The DIC was estimated from the first measured signal peak (mV) of each sample according to a calibration curve calculated from samples with a known DIC concentration, freshly prepared at each day of analysis.
2.3.3. POM, fish tissue, feed pellets and sludge d13C and d15N measurements
A
B
Dried sludge, freeze dried fish tissue and feed pellets were grained and samples weighing between 150 and 1100 mg, depending on their organic matter content, were analyzed. The filter upon POM sample were collected (as described above) was divided into quarters. The grounded material or a 1/4 of filter were wrapped in tin cups and measured using an element analyzer connected online to an IRMS (252 mat, Finnegan).
2.4.
Calculations
Carbon concentration in the POM (POMC) was calculated according to the following equation, POMC ¼
WCs 4 Vs
C
(2)
Where POMC is expressed in mg L1, WCs is the carbon weight fraction of the organic matter on the filter measured by the mass spectrometer in mg and multiplied by 4 to account for using only 1/4 of the filtered material in the analysis, Vs is the water volume filtered. Nitrogen concentration in the POM (POMN) was calculated in the same manner as POMC by replacing WCs with WNs the weight of N in the filter as measured by the mass spectrometer in mg.
2.4.1.
Statistical analysis
Data are expressed as mean SD. A comparison between treatments was performed using the ANOVA test. Statistical
Fig. 2 e DIC (A), d13CDIC (B) and alkalinity (C) measured at different locations of the mariculture system, abbreviations as in Fig. 1. The data is presented as mean ± SD.
2379
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 3 7 5 e2 3 8 2
4630 296 mmol L1. In the FBR, DIC was higher by about 1000 mmol L1 (Fig. 2a). The high variation in the DIC is mainly due to a variation between sampling dates. Within each sampling session, the variation averaged 138 51 and 184 99 mmol L1 for the aerobic and the anaerobic loops, respectively. The average DIC difference between the aerobic and anaerobic loops was statistically significant (ANOVA, p < 0.002). The d13CDIC (Fig. 2b, Table 1) showed 13C enriched values in the aerobic loop of 8.03 0.25& compared to values of 10.8 0.37& in the anaerobic loop. The average difference between the two loops was statistically significant (ANOVA, p < 0.005). Alkalinity in the aerobic loop ranged from 4080 to 4260 mmol L1 with an average value of 4160 80 mmol L1, significantly lower than the anaerobic loop, which fluctuated between 4900 and 5600 mmol L1 with an average of 5300 350 mmol L1 (Fig. 2c).
3.1.2.
A
POM
d13CPOM in all the 9 sampling sites of the system was very similar (Table 1) with an average of 23.05 1.08&, 23.29 1.43& and 23.16 1.28& for the aerobic, anaerobic and the all the stations combined, respectively. Carbon concentrations in POM (POMC) averaged 2.5 0.5 and 4.1 0.4 mg C L1 in the aerobic loop and anaerobic loop, respectively. This difference was statistically significant (ANOVA, p < 0.03). d15NPOM in the TF was enriched in 15N with values of around 9& compared to the other sampling sites in the mariculture system where d15NPOM values were around 7.5& (Table 1). Nitrogen concentrations in the POM (POMN) were lower in the aerobic loop than in the anaerobic loop, 0.76 0.26 and 1.58 0.23 mg N L1, respectively. This difference was statistically significant (ANOVA, p < 0.03). POM C/N ratios were significantly (ANOVA, p < 0.002) higher in the aerobic loop than in the anaerobic loop, 6 0.24 and 4.9 0.21, respectively.
3.1.3.
B
Digestion basin profiles
DO in the surface water of the DB decreased from the DBin to the DBout. A decrease in oxygen was also observed with depth (Fig. 3; a representative profile from the DB center measured on 6/11/2006). It was found that oxygen was totally consumed within the upper 5 cm of the sludge layer. These findings were
Fig. 3 e (A) A typical depth profile of DO (,) and Redox (C) in the DB center. (B) A typical depth profile of NOL 3 (B), S2L (:) and TAN (,) measured in the DB center.
Table 1 e The POM isotopic values (&) and the C and N concentrations (mg/L) measured in the water at the mariculture sampling stations, abbreviation as in Fig. 1. The results are expressed as average ± SD. Parameter d13CPOM SD d15NPOM SD CPOM SD NPOM SD C/NPOM SD
FB
TFB
TFout
TFin
DBinT
DBinB
DBoutT
DBoutB
FBR
23.03 1.52 7.63 2.23 2.31 1.53 0.52 0.34 5.97 1.69
22.96 1.16 9.03 2.32 2.71 2.39 0.60 0.52 6.50 2.35
23.18 1.03 8.58 2.50 2.72 2.13 0.52 0.45 5.84 1.21
23.03 0.83 9.20 2.49 2.32 1.85 0.51 0.38 5.77 1.02
23.35 1.04 7.42 3.64 3.90 1.97 0.95 0.25 4.88 0.38
23.16 0.92 7.47 3.13 4.24 2.63 0.79 0.55 4.94 0.34
23.10 2.01 7.51 4.33 3.68 2.26 0.87 0.35 4.82 0.17
23.59 2.09 7.23 3.52 4.58 3.35 1.14 0.59 5.00 0.39
23.11 2.01 7.57 3.31 3.87 3.17 0.93 0.57 4.93 0.24
2380
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 3 7 5 e2 3 8 2
further validated by redox potential measurements showing a similar profile (Fig. 3A). A representative nutrient profile of the center of the DB (Fig. 3B) showed a nitrate decrease from the surface to the bottom with depletion at about 15 cm into the sludge. TAN and sulfide profiles showed an opposite trend as compared to nitrate.
3.1.4.
d13C and d15N values in sludge
13
d Csludge and d15Nsludge values were 22.5 0.9& and 9.2 1.3&, respectively. d13Csludge and d15Nsludge did not show any significant difference between the different areas within the DB and the organic matter derived from the wall of the TFB. The average values of the d13Csludge were very similar to that of the fish feed 22.2 3& (ANOVA, p < 0.67), while those of the d15Nsludge were 15N enriched compared to fish feed 6.2 0.5& (ANOVA, p < 0.0001). The C/N ratio of the sludge was slightly lower than that of fish feed 6 0.31 and 6.64 0.26, respectively (ANOVA, p < 0.01).
4.
Discussion
Significant differences in the inorganic carbonate system were found between the aerobic and anaerobic loops. Higher DIC and alkalinity, and 13C depleted d13CDIC values were measured in the anaerobic loop than in the aerobic loop (Fig. 2). Considering the main processes taking place in the treatment systems, we suggest that in the anaerobic loop 13C depletion in the DIC is caused by a CO2 release through respiratory processes and by an alkalinity increase through denitrification (van Rijn et al., 2006). In the aerobic loop, alkalinity is consumed due to nitrification (Chen et al., 2006; Eding et al., 2006) by chemolithoautotrophic microorganisms that assimilate 13C depleted CO2 from the water (Foesel et al., 2008; Sakata et al., 2008). This CO2 assimilation and, in addition, the intensive CO2 degassing taking place as a result of the low water retention time and the specific configuration of the trickling filters (Eding et al., 2006), may provide an explanation for the 13C enrichment in the aerobic treatment compartment. It seems, therefore, that the difference in alkalinity, CO2 uptake/release and the consequent difference in buffering capacity as well as the differences in filter configuration and retention time are responsible for the difference in DIC values between the two loops. Lower DIC values were measured in the TF components than in the fish basin. Within the aerobic loop, lowest DIC values were measured in the TFout. This is consistent with the chemolithoautotrophic utilization of CO2 and the consumption of alkalinity by the nitrification process within the TF. Based on the methodology used in this study as well as results from previous studies, which demonstrated oxidation of TAN to nitrate within the filter (van Rijn and Rivera, 1990; Gelfand et al., 2003), it might be concluded that autotrophic nitrification is a major process within this filter. Organic matter degradation in the DB is rapid. It was estimated that during one growth season (around one year) about 90% of the total organic matter added to the system and not utilized by fish, is digested (Fine, unpublished data; van Rijn et al., 1995; Gelfand et al., 2003; Neori et al., 2007). The d13CDIC in the anaerobic loop is 13C depleted compared to the aerobic loop by 2.5e3&. This finding is consistent with
the relative 13C depleted organic matter respired and mineralized in the DB. A bigger difference would be expected but, as previously noted, the difference in DIC between the aerobic and anaerobic loops is controlled by alkalinity and the water retention time in each of the loops. Both factors directly affect the efficiency of CO2 degassing, which is a dynamic process. Based on thermodynamic considerations, a significant kinetic discrimination against 13C is expected due to this process. The relative small difference between the loops may be explained by 13C depleted CO2 gas escaping from the DB and degassing during the water return to the aerobic loop due to the lower alkalinity and low water retention time there. It should be emphasized that, despite its static plug-flow mode of operation, large quantities of CO2 are released into the atmosphere in the DB since CO2 generation in this reactor is high. CO2 generation in the DB is high as feces and uneaten feed are all diverted to this basin. How much CO2 is released by digestion of the feces and uneaten feed can roughly be estimated by assuming that fish utilize around 50% of the carbon supplied with the feed (Neori et al., 2007). In this particular system, feed loads were as high as 4 kg per day thus, without accounting for uneaten feed, at least 1 kg of carbon was daily added to the DB. In the DB around 90% of the carbon is digested (Neori et al., 2007) which means that 0.9 kg carbon in the form of CO2 is produced daily. This daily added amount of carbon to the DB is equal to about 3e4 times the average amount of DIC in the DB, which can be calculated from DIC concentrations (Fig. 2) and the water volume of the DB. If no CO2 gas is released from the DB, it would be added to the DIC in the water. In such a case, the only exchange of DIC in the DB would be by water exchange. Because of the long water retention time in the DB (4.5 h), the d13CDIC would equilibrate to values similar to that of the organic material added to the DB i.e. w22&, as opposed to measured values of w10&. The carbonate system is probably in equilibrium between the DIC concentration determined by its alkalinity and the pCO2 in the atmosphere due to a relative long retention time of the DB. Based on these findings it seems likely, therefore, that the difference in d13CDIC between the aerobic and the anaerobic loops is controlled by the alkalinity differences. The fish in the mariculture are fed with pellets (48% protein, 20% fat) as the single external source of organic matter and feed, therefore, it is the source of most of the new organic matter in the system. The only other new source of carbon in the system is CO2 fixed by autotrophic bacteria, which in this particular system are mainly represented by nitrifying bacteria in the TF. Most of the organic matter produced in the TF is removed by the foam fractionator and is disposed in the DB, but quantitatively this source is highly insignificant. Feed pellets d13C and d15N values were 22.2 3& and 6.2 0.5&, respectively, while in the fish tissue these values were 20.4 1.2& and 10.8 0.6&, respectively, thus showing an increase in the isotopic values of about 2& and 4& in the d13C and d15N, respectively. These findings are consistent with previous studies, which point to an increase in d13C and d15N with each trophic level (DeNiro and Epstein, 1978; Fry and Sherr, 1984; Minagawa and Wada, 1984) and confirming that the fish in this system use feed pellets as their sole source of nutrition.
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 3 7 5 e2 3 8 2
The input of organic matter from the fish basin to the DB as POM consists of undigested feed pellets and fish feces. This organic matter settles in the DB and serves as substrate for microbial respiratory processes mainly using nitrate and sulfate as electron acceptors as can be seen from the chemical and redox potential profiles in the DB (Fig. 3). The DB average d13Csludge and d15Nsludge values were 22.5 0.9& and 9.2 1.3&, respectively and the d13CPOM and d15NPOM were 23.3 1.4&, 7.4 3.3&, respectively. The carbon isotopes values in the DB sludge and POM were not significantly different from that of the pellets, while the nitrogen isotopes were 15N enriched compared to the pellets. The enrichment in the nitrogen isotopes may be a result of microbial decomposition (Fellerhoff et al., 2003), probably releasing 15N depleted ammonia. This possibility is further substantiated by the high TAN (Fig. 3B) in the bottom layers of the sludge column in agreement with previous work on a smaller scale system (Gelfand et al., 2003). It was shown by Cytryn et al. (2003), using DNA sequence analysis from the DB sludge, that a number of dominant microorganisms were affiliated with fermentative bacteria, Fusibacteria, Dethiosulfovibrio and members of the Bacteroidetes phylum. These fermentative bacteria are involved in the degradation of macromolecular compounds whereby secondary metabolites such as volatile fatty acids (VFA) are liberated (van Rijn et al., 1995). Under these conditions, liberated VFA were shown to undergo a rapid oxidation by bacterial respiration with mainly sulfate and nitrate as electron acceptors (Aboutboul et al., 1995; van Rijn et al., 1996, 1995).
5.
Conclusions
In this study, an approach, based on integrating stable isotopes with additional chemical analysis, was used to trace carbon in a zero-discharge mariculture system. It was shown that alkalinity values provide a clear indication for the main microbiological processes taking place in each of the system components. The carbon (DIC and POM) and nitrogen (POM) values show a consistent difference between the aerobic and anaerobic loops caused by a combination of differences in microbial processes and water retention time in these loops. Further studies are required to determine how d13C profiles, POM formation and different respiratory pathways affect the isotopic values in the system. Once such links are established, the technique of stable isotope tracing has the potential to diagnose changes in systems, such as examined in this study, by means of a few relatively simple measurements.
references
Aboutboul, Y., Arviv, R., van Rijn, J., 1995. Anaerobic treatment of intensive fish culture effluents: volatile fatty acid mediated denitrification. Aquaculture 133 (1), 21e32. APHA, 1998. Standard Methods for the Examination of Water and Wastewater. American Public Health Association, Washington, DC. Bower, C.E., Holm-Hansen, T., 1980. A salicylate-hypochlorite method for determining ammonia in seawater. Canadian Journal of Fisheries and Aquatic Sciences 37, 794e798.
2381
Chen, S., Ling, J., Blancheton, J.-P., 2006. Nitrification kinetics of biofilm as affected by water quality factors. Aquacultural Engineering 34 (3), 179e197. Cline, J.D., 1969. Spectrophotometric determination of hydrogen sulfide in natural waters. Limnological Oceanography 14, 454e458. Cytryn, E., Gelfand, I., Barak, Y., van Rijn, J., Minz, D., 2003. Diversity of microbial communities correlated to physiochemical parameters in a digestion basin of a zerodischarge mariculture system. Environmental Microbiology 5 (1), 55e63. Cytryn, E., Minz, D., Gelfand, I., Neori, A., Gieseke, A., De Beer, D., Van Rijn, J., 2005. Sulfide-oxidizing activity and bacterial community structure in a fluidized bed reactor from a zerodischarge mariculture system. Environmental Science and Technology 39 (6), 1802e1810. DeNiro, M.J., Epstein, S., 1978. Influence of diet on the distribution of carbon isotopes in animals. Geochimica et Cosmochimica Acta 42 (5), 495e506. DeNiro, M.J., Epstein, S., 1981. Influence of diet on the distribution of nitrogen isotopes in animals. Geochimica et Cosmochimica Acta 45 (3), 341e351. Eding, E.H., Kamstra, A., Verreth, J., Huisman, E.A., Klapwijk, A., 2006. Design and operation of nitrifying trickling filters in recirculating aquaculture: a review. Aquacultural Engineering 34 (3), 234e260. Fellerhoff, C., Voss, M., Wantzen, K.M., 2003. Stable carbon and nitrogen isotope signatures of decomposing tropical macrophytes. Aquatic Ecology 37 (4), 361e375. Foesel, B.U., Gieseke, A., Schwermer, C., Stief, P., Koch, L., Cytryn, E., de la Torre´, J.R., van Rijn, J., Minz, D., Drake, H.L., Schramm, A., 2008. Nitrosomonas Nm143-like ammonia oxidizers and Nitrospira marina-like nitrite oxidizers dominate the nitrifier community in a marine aquaculture biofilm. FEMS Microbiology Ecology 63 (2), 192e204. Fry, B., Sherr, E.B., 1984. d13C measurements as indicators of carbon flow in marine and freshwater ecosystems. Contributions in Marine Sciences 27, 13e47. Gelfand, I., Barak, Y., Even-Chen, Z., Cytryn, E., van Rijn, J., Krom, M.D., Neori, A., 2003. A novel zero discharge intensive seawater recirculating system for the culture of marine fish. Journal of the World Aquaculture Society 34 (3), 344e358. Goericke, R., Fry, B., 1994. Variations of marine plankton d13C with Latitude, temperature, and dissolved CO2 in the world ocean. Global Biogeochemical Cycles 8 (1), 85e90. Grasshoff, K., Ehrhardt, M., Kremling, K., 1983. Methods of Seawater Analysis. Verlag Chemie GmbH, Weinheim, 419 pp. Guy, R.D., Fogel, M.L., Berry, J.A., 1993. Photosynthetic fractionation of the stable isotopes of oxygen and carbon. Plant Physiology 101 (1), 37e47. Holmer, M., Kristensen, E., 1992. Impact of marine fish cage farming on metabolism and sulfate reduction of underlying sediments. Marine Ecological Progress Series 80 (2e3), 191e201. Jackson, J.B.C., Kirby, M.X., Berger, W.H., Bjorndal, K.A., Botsford, L. W., Bourque, B.J., Bradbury, R.H., Cooke, R., Erlandson, J., Estes, J.A., Hughes, T.P., Kidwell, S., Lange, C.B., Lenihan, H.S., Pandolfi, J.M., Peterson, C.H., Steneck, R.S., Tegner, M.J., Warner, R.R., 2001. Historical overfishing and the recent collapse of coastal ecosystems. Science 293 (5530), 629e637. Karakassis, I., Tsapakis, M., Hatziyanni, E., Papadopoulou, K.N., Plaiti, W., 2000. Impact of cage farming of fish on the seabed in three Mediterranean coastal areas. ICES Journal of Marine Science 57 (5), 1462e1471. Mazzola, A., Mirto, S., Danovaro, R., 1999. Initial fish-farm impact on meiofaunal assemblages in coastal sediments of the Western Mediterranean. Marine Pollution Bulletin 38 (12), 1126e1133.
2382
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 3 7 5 e2 3 8 2
Minagawa, M., Wada, E., 1984. Stepwise enrichment of 15N along food chains: further evidence and the relation between d15N and animal age. Geochimica et Cosmochimica Acta 48 (5), 1135e1140. Neori, A., Krom, M.D., van Rijn, J., 2007. Biogeochemical processes in intensive zero-effluent marine fish culture with recirculating aerobic and anaerobic biofilters. Journal of Experimental Marine Biology and Ecology 349 (2), 235e247. Robinson, J.J., Scott, K.M., Swanson, S.T., O’Leary, M.H., Horken, K., Tabita, F.R., Cavanaugh, C.M., 2003. Kinetic isotope effect and characterization of form II RubisCO from the chemoautotrophic endosymbionts of the hydrothermal vent tubeworm Riftia pachyptila. Limnological Oceanography 48 (1), 48e54. Sakata, S., Hayes, J.M., Rohmer, M., Hooper, A.B., Seemann, M., 2008. Stable carbon-isotopic compositions of lipids isolated from the ammonia-oxidizing chemoautotroph Nitrosomonas europaea. Organic Geochemistry 39 (12), 1725e1734. Schroeder, G.L., 1983. Sources of fish and prawn growth in polyculture ponds as indicated by dC analysis. Aquaculture 35, 29e42. Sher, Y., Schneider, K., Schwermer, C.U., van Rijn, J., 2008. Sulfideinduced nitrate reduction in the sludge of an anaerobic digester of a zero-discharge recirculating mariculture system. Water Research 42 (16), 4386e4392. Shnel, N., Barak, Y., Ezer, T., Dafni, Z., van Rijn, J., 2002. Design and performance of a zero-discharge tilapia recirculating system. Aquacultural Engineering 26 (3), 191e203. Strickland, J.D.H., Parsons, T.R., 1972. A practical handbook of seawater analysis. Fisheries Research Board of Canada, Ottowa. Strocchi, A., Furne, J.K., Levitt, M.D., 1992. A modification of the methylene blue method to measure bacterial sulfide production in feces. Journal of Microbiological Methods 15 (2), 75e82.
Tidwell, J.H., Allan, G.L., 2001. Fish as food: aquaculture’s contribution e ecological and economic impacts and contributions of fish farming and capture fisheries. EMBO Reports 2 (11), 958e963. van Rijn, J., 1996. The potential for integrated biological treatment systems in recirculating fish culture e a review. Aquaculture 139 (3e4), 181e201. van Rijn, J., Fonarev, N., Berkowitz, B., 1995. Anaerobic treatment of intensive fish culture effluents: digestion of fish feed and release of volatile fatty acids. Aquaculture 133 (1), 9e20. van Rijn, J., Rivera, G., 1990. Aerobic and anaerobic biofiltration in an aquaculture uniteNitrite accumulation as a result of nitrification and denitrification. Aquacultural Engineering 9 (4), 217e234. van Rijn, J., Tal, Y., Barak, Y., 1996. Influence of volatile fatty acids on nitrite accumulation by a Pseudomonas stutzeri strain isolated from a denitrifying fluidized bed reactor. Applied Environmental Microbiology 62 (7), 2615e2620. van Rijn, J., Tal, Y., Schreier, H.J., 2006. Denitrification in recirculating systems: theory and applications. Aquacultural Engineering 34 (3), 364e376. Worm, B., Barbier, E.B., Beaumont, N., Duffy, J.E., Folke, C., Halpern, B.S., Jackson, J.B.C., Lotze, H.K., Micheli, F., Palumbi, S.R., Sala, E., Selkoe, K.A., Stachowicz, J.J., Watson, R., 2006. Impacts of biodiversity loss on ocean ecosystem services. Science 314, 787e790. Wu, R.S.S., 1995. The environmental impact of marine fish culture: towards a sustainable future. Marine Pollution Bulletin 31 (4e12), 159e166.
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 3 8 3 e2 3 9 1
Available at www.sciencedirect.com
journal homepage: www.elsevier.com/locate/watres
Dechlorination of endocrine disrupting chemicals using Mg0/ZnCl2 bimetallic system Asfiya Begum, Sumit Kumar Gautam* The Energy and Resources Institute (TERI), 4th Main, II Cross, Domlur II Stage, Bangalore 560071, Karnataka, India
article info
abstract
Article history:
In the present study, Mg0/ZnCl2 bimetallic system was evaluated for its efficiency to
Received 15 September 2010
dechlorinate endosulfan and lindane in aqueous phase. Presence of acetone in the reaction
Received in revised form
mixture played an important role by increasing the solubilities of both pesticides and
18 January 2011
thereby accelerating its mass transfer. Water acetone ratio of 2:1 and 1:1 (v/v) was found
Accepted 21 January 2011
optimum for the dechlorination of endosulfan and lindane respectively. Presence of Hþ
Available online 31 January 2011
ions in the reaction mixture (50 ml ml1 of glacial acetic acid) accelerated the degradation efficiency of 30 ppm initial concentration of endosulfan (96% removal) and lindane (98%
Keywords:
removal) at Mg0/ZnCl2 dose of 5/1 mg ml1 within 30 min of reaction. Dechlorination
Endocrine
kinetics for endosulfan and lindane (10, 30 and 50 ppm initial concentration of each
disrupting
chemicals
(EDCs)
pesticide) with varying Mg0/ZnCl2 doses and the time course profiles of each pesticide were
Endosulfan
well fitted into the first order dechlorination reaction. The optimum observed rate constant (kobs’) values for endosulfan (0.2168, 0.1209 and 0.1614 min1 for 10, 30 and 50 ppm initial
Lindane Magnesium (Mg )
concentration respectively) and lindane (0.1746, 0.1968 and 0.2253 min1 for 10, 30 and
Zinc chloride (ZnCl2)
50 ppm initial concentration respectively) dechlorination were obtained when the reac-
0
tions were conducted with doses of 7.5/1 mg ml1 and 5/1 mg ml1 Mg0/ZnCl2 respectively. Endosulfan and lindane were completely dechlorinated into their hydrocarbon skeletons namely, Bicyclo [2,2,1] hepta 2-5 diene and Benzene respectively as revealed by GCMS analysis. ª 2011 Elsevier Ltd. All rights reserved.
1.
Introduction
Endocrine disrupting chemicals (EDCs) are considered to be emerging contaminants, which means that they are either still unregulated or they are in the process of regularization. EDCs have been defined by the Organisation of Economic and Cooperative Development (OECD) as “an exogenous substance or mixture that alters the functions of the endocrine systems and consequently causes adverse health effects in an intact organism, or its progeny or (sub) populations” (Esplugas et al., 2007; McKinlay et al., 2008). A wide range of chemical compounds have been found to be capable of disrupting the
endocrine system. The EDCs include chlorinated pesticides (e.g. DDT, vinclozolin, TBT, atrazine, lindane, and endosulfan), persistent organochlorines and organohalogens (e.g. PCBs, dioxins, furans, and brominated fire retardants), alkyl phenols (e.g. nonylphenol and octylphenol), and heavy metals like cadmium, lead, mercury etc (Esplugas et al., 2007; Mediratta et al., 2008; Mertens, 2006; Usha and Harikrishnan, 2005; Weber et al., 2009). The presence of EDCs affect the environment and have been reported that they result in (1) pre-mature breakage of eggs of birds, fishes and turtles, (2) sex reversal like feminization of male fish, and (3) reproductive abnormalities in fishes, reptiles,
* Corresponding author. Tel.: þ91 80 25356590; fax: þ91 80 25356589. E-mail address:
[email protected] (S.K. Gautam). 0043-1354/$ e see front matter ª 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.watres.2011.01.017
2384
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 3 8 3 e2 3 9 1
birds and mammals. These effects may also lead to extinction of these species (Esplugas et al., 2007; McKinlay et al., 2008). Two of the EDCs, Endosulfan and Lindane have been extensively used on a wide variety of crops, in warehouses, in public health (to control vector-borne diseases), and as wood preservatives. These compounds have been chemicals of choice as a result of their low cost, easy availability, applicability and more importantly their stability in the environment (Andreozzi et al., 1999; Gogate and Pandit, 2004). On the other hand, these pollutants are characterised by high chemical stability, lipophilic nature, hydrophobicity, accumulation in various compartments of earth (air, soil, and water), carcinogenic property, presence of chlorine atoms and have half-lives of many years that makes them recalcitrant and difficult to be completely mineralized by biological treatments (Andreozzi et al., 1999). In the last decade, extensive research has been carried out on the application of bimetallic systems to degrade chlorinated organic compounds. Bimetallic systems make use of two metals, one in the zero-valent form (with a high reduction potential like Mgþ2/Mg0, Feþ2/Fe0, etc.) to produce nascent hydrogen by anodic corrosion, and the other metal with a relatively high (positive) reduction potential (such as Agþ1/ Ag0, Pdþ4/Pd0) as a catalyst. Nascent hydrogen produced thus is intercalated by the catalyst to form a metal hydride (M H), which reacts with the target substrate to reduce it into less recalcitrant compounds. The main advantage of bimetallic systems is the ability to conduct the reaction at ambient temperature and pressure without exclusion of atmospheric oxygen (Gautam and Suresh, 2006; Lin et al., 2004; Patel and Suresh, 2008; Simagina and Stoyanova, 2001). There are four major factors that influence the rates and extent of dechlorination by zero-valent metal systems: (i) ionization potential and E0 of the zero-valent metal; (ii) solubility of the metal hydroxide formed following corrosion of metal; (iii) availability of protons; and (iv) solubility of the target compound (Patel and Suresh, 2008; Wang et al., 2009). Based on the factors mentioned above, Mg0 offers distinct advantage as it has the high reduction potential (Mgþ2/Mg0 ¼ - 2.2 V) and it works in the presence of oxygen as compared to generally used iron. Further, the solubility of magnesium hydroxide is relatively high, which accelerates the corrosion of the metal. While the role of Zinc is to enhance the corrosion of primary metal and is a moderately strong reductant (0.76 V). It is being used because of its low cost and easy availability as compared to Ni, Pd, Cu etc. Based on the above mentioned distinct advantages of magnesium and zinc, Mg0/ZnCl2 bimetallic system was chosen to dechlorinate endosulfan and lindane in the present study. The specific objectives of the study were: 1. To study the effect of Hþ ion concentration on the dechlorination reactions 2. To optimize the solvent ratio (water: acetone) for the dechlorination reactions. 3. To study the dechlorination kinetics of Endosulfan and Lindane; and 4. To identify the intermediates/end products of dechlorination reaction and elucidate the dechlorination pathways for endosulfan and lindane dechlorination.
2.
Materials and methods
2.1.
Source of chemicals
Magnesium (Mg0) granules (w200 mesh), Endosulfan (6,7,8,9,10,10-hexachloro-1,5,5a,6,9,9a-hexahydro-6,9-methano2,4,3-benzodioxathiepine-3-oxide), Lindane (1r,2R,3S,4r,5R,6S )1,2,3,4,5,6-hexachlorocyclohexane) were purchased from SigmaeAldrich Chemical Company (USA) and were >98% pure. Acetone, zinc chloride (ZnCl2), hydrochloric acid and glacial acetic acid were purchased from Merck Ltd. (India) and cyclohexane was purchased from Fisher Scientific (India). All the chemicals were of analytical grade. No pre-treatment was performed with the chemicals and was used as received. All the glasswares used were of “A” grade.
2.1.1.
Dechlorination reaction protocol
Separate experiments for the degradation of endosulfan and lindane were conducted in water: acetone (4 ml, 1:1, v/v) reaction phase in the absence and in the presence of acid (either 0.1 N hydrochloric acid or glacial acetic acid). An aliquot of ZnCl2 stock solution (0.2 g ml1) was added into reaction phase to attain the required final concentrations (1e5 mg ml1 for endosulfan dechlorination and about 1e2 mg ml1 for lindane dechlorination). An initial concentration of 30 ppm of endosulfan or lindane was added into the respective reaction phase from a 1000 ppm stock solution of each pesticide prepared separately in acetone. Reactions were initiated by the addition of Mg0 granules (1e25 mg ml1 for endosulfan experiment and 5e10 mg ml1for lindane experiment). Separate control experiments were carried out with either only Mg0 or ZnCl2 to evaluate the significance of each in the overall degradation of both the pesticides. Table 1 depict the contents of reaction mixtures for endosulfan and lindane dechlorination experiments. All the reactions were conducted in quadruplicate under atmospheric pressure with continuous shaking in a water bath maintained at 130 rpm at 30 C. Initial pH of the reaction mixture was recorded and pH was not maintained during the reaction course. No precautions were taken to exclude oxygen or reduce redox potential of the reaction phase. The entire reaction mixtures were sacrificed after 30 min of reaction, extracted twice using cyclohexane (total 8 ml) and 0.2 ml volume of the pooled hexane extracts were injected for GC-ECD analyses.
2.2. Solvent ratio optimisation for the dechlorination of endosulfan and lindane The solvent (water: acetone) ratio for the dechlorination of endosulfan was optimised using a reaction system consisting of ZnCl2 (2 mg ml1), Mg0 (10 mg ml1), and 30 ppm initial concentration of endosulfan while in the case of lindane the water: acetone ratio was optimized using a reaction system containing ZnCl2 (1 mg ml1), Mg0 (5 mg ml1), and 30 ppm initial concentration of lindane. In both the cases, the water: acetone ratios were varied from 1:1 to 19:1 (v/v) and a reaction time of 30 min was chosen to analyse the extent of disappearance of each pesticide.
2385
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 3 8 3 e2 3 9 1
Table 1 e Dechlorination of endosulfan and lindane with/without hydrochloric acid or glacial acetic acid by varying concentrations of Mg0 and ZnCl2 (Reaction conditions: Reaction Time: 30 min; Water: Acetone; 1:1v/v). Mg0 (mg ml1)
ZnCl2 (mg ml1)
Glacial Acetic Acid (50 mL ml1)
Hydrochloric Acid (0.1N, 50 mL m1)
% Disappearance of pesticides
Initial Endosulfan Conc ¼ 30 ppm 1 10 2 10 3 10 4 25 7 5 8 5 9 5
2 2 2 5 1 1 1
* added * * * added *
* * added * * * added
89.54 99.3 93.86 96.75 57.06 95.89 59.56
Initial Lindane Conc ¼ 30 ppm 1 10 2 10 3 10 4 10 5 10 6 10 7 7.5 8 7.5 9 7.5 10 7.5 11 7.5 12 7.5 13 5 14 5 15 5 16 5 17 5 18 5
1 1 1 2 2 2 1 1 1 2 2 2 1 1 1 2 2 2
* added * * added * * added * * added * * added * * added *
* * added * * added * * added * * added * * added * * added
1.84 99.7 38.7 62.7 99.8 78.3 47.2 99.5 59.4 74.5 99.5 81.6 13.8 98.3 32.2 62.3 98.7 76.6
SNo
*- no acid added.
2.3. Kinetic study of endosulfan and lindane degradation using Mg0/ZnCl2 bimetallic system
2.4. GC-ECD (Gas ChromatographyeElectron Capture Detection) analyses
In the case of endosulfan, kinetic studies were conducted in water:acetone (2:1, v/v) phase to determine the rates and extent of dechlorination of 10, 30 and 50 ppm initial concentrations of endosulfan each as a function of: a) varying Mg0 (5.0, 7.5 and 10 mg ml1) concentrations at a fixed ZnCl2 concentration (1 mg ml1) and b) varying ZnCl2 concentrations (0.5, 1 and 1.5 mg ml1) at a fixed concentration (7.5 mg ml1) of Mg0 to establish the order of reaction and rate constant (kobs) values. In the case of lindane, the reactions were conducted at: a) varying Mg0 (1, 2.5 and 5 mg ml1) concentrations at a fixed ZnCl2 concentration (1 mg ml1) and b) varying ZnCl2 concentrations (0.5, 1 and 1.5 mg ml1) at a fixed concentration (5 mg ml1) of Mg0 with 10, 30 and 50 ppm initial concentrations of lindane. The corresponding control experiments were also conducted to determine the extent of degradation of the above mentioned pesticides, if any, using ZnCl2 or Mg0 alone under same conditions as the test samples. The entire reaction mixtures were sacrificed at chosen time points, extracted twice using cyclohexane (total 8 ml) and 0.2 ml or 1.0 ml volume of the pooled hexane extracts were analysed for residual pesticides, intermediates and end products using GC-ECD and GC-MS.
Analyses of extracted samples were done using a gas chromatograph equipped (Agilent, model no. 6890 N) with Ni63 electron capture detector (ECD). The column used was HP-5 capillary column of 0.32 mm ID, 0.25 mm film thickness and 30 m length. Injections were made in splitless mode using nitrogen as the carrier gas. The following temperature programming was used: initial oven temperature of 150 0C with hold time for 2 min, then ramped to 200 C at 6 C min1 with hold time of 2 min, again ramped to 250 C at 10 0C min1 with hold time of 2 min. Injector and detector temperatures were set at 200 and 290 0C respectively. The residual concentrations of endosulfan and lindane, partially chlorinated intermediates and end products were quantified from peak areas obtained through automated integration and also by comparison with known concentrations of the pure standard compounds.
2.5. GC-MS (gas chromatographyemass spectroscopy) analyses GC-MS analyses were carried out using TRACE GC ULTRA (Thermo make) equipped with MS (model DSQII). The column used for GC-MS analysis was TR-5 column of 0.25 mm I.D., 0.25 mm film thickness and 30 m length. 1 mL volume of
2386
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 3 8 3 e2 3 9 1
samples was injected for analyses. He (Helium) was used as carrier gas. The temperature programming used was: Initial oven temperature of 40 C with the hold time of 1 min, ramped to 53 C at 1 C/min with hold time of 1 min, again ramped 53e60 C at 10 C/min with hold time of 1 min and finally ramped from 60 to 250 C with hold time of 1 min. Injector and detector temperatures were set at 200 and 300 C, respectively. The mass spectral data coupled with the systematic reduction in the retention times of the dechlorinated products (due to loss of chlorine atoms) allowed identification of the intermediates and end products with reasonable certainty. Wiley Registry (8e Mass Spectral Library) was used to identify the intermediate and end products.
In our study it was observed that the degradation efficiency was better when acetic acid (weak acid, pH range of 3.35e3.63) was added into the solution as against HCl (strong acid, pH range of 2.14e2.22), this may be due to very fast dissolution of Mg0 granules in the presence of strong acids which leads to immediate formation of nascent hydrogen which provides lesser time to Zn to capture it. While in case of weak acid, the dissolution of Mg granules is comparatively slow which provides sufficient time to catalyst Zn to capture produced nascent hydrogen to form hydrides. Based on the results discussed above, Mg0/ZnCl2 dose of 5/1 mg ml1 with 50 mL ml1 glacial acetic acid was chosen as optimum to conduct the experiments further.
3.
3.2. Solvent optimisation for the degradation of endosulfan and Lindane
Results and discussion
3.1. Catalyst and Hþ ion optimisation for the dechlorination of endosulfan and Lindane Table 1 compares extent of endosulfan and lindane (30 ppm initial concentration) dechlorination by varying Mg0/ZnCl2 concentrations in the presence and absence of acid (50 mL ml1 HCl or glacial acetic acid). 30 min of reaction time was chosen to carryout the optimisation studies in water: acetone (1:1 v/v) reaction phase. It can be observed from Table 1 that 90% disappearance of endosulfan at Mg0/ZnCl2 dose of 10/2 mg ml1 was achieved without the addition of any acid. The addition of 0.1 N HCl had a degradation efficiency of 94% while the addition of glacial acetic acid resulted in 99.3% removal of endosulfan. However on the reduction of Mg0/ZnCl2 dose to 5/1 mg ml1, 57% of endosulfan was degraded without the addition of acid. On the addition of 0.1 N HCl 60% removal of endosulfan was achieved. The addition of glacial acetic acid increased the degradation efficiency of endosulfan to 96%. In case of lindane, at Mg0/ZnCl2 dose to 10/1 mg ml1 about 2% removal of lindane was observed, which on the addition of 0.1 N HCl resulted in 39% degradation and the addition of glacial acetic acid had a degradation efficiency of 99.7%. However on reducing the Mg0/ZnCl2 dose to 5/1 mg ml1 about 14% lindane degradation efficiency was observed. While on the addition of 0.1 N HCl, 32% degradation of lindane was observed and about 98% degradation efficiency was observed on the addition of glacial acetic acid. Similar observations were also reported in the earlier studies conducted by Gautam and Suresh (2007); Mu et al., 2004; Patel and Suresh, 2008; Wang et al., 2009 and Xinhua et al., 2009 while studying the degradation of DDT using Mg0/Pd4þ; nitrobenzene using zero-valent metallic iron; pentachlorophenol using Mg0/K2PdCl6; hexachlorocyclohexane using zero-valent metallic iron and nitrochlorobenzene using Ni/Fe respectively where in all cases the presence of acid improved the dechlorination efficiency of metallic systems. This is attributed to the fact that addition of acid enhanced the rate of dechlorination of target compounds by: (a) facilitating fast corrosion of Mg0 and reduction of ZnCl2 (b) providing protons to generate nascent hydrogen and (c) delaying the creation of alkaline conditions in the reaction phase that may prevent passivation of metals which inturn may retard the dechlorination process.
Table 2 shows the effect of various water: solvent ratios on the degradation of 30 ppm initial concentration of each endosulfan and lindane in the presence of acetic acid. It is depicted in Table 2 that the presence of acetone had a positive influence on the extent of degradation. At water: acetone ratio of 1:1 (v/v), about 90% of endosulfan and about 98% of lindane was degraded within 30 min of reaction time. However at water: acetone ratio of 2:1, about 81% of endosulfan and 83% of lindane was degraded. Further, 73% of endosulfan and 78% lindane were degraded at water: acetone ratio of 4: 1. While at 9:1 water: acetone ratio, only 51% endosulfan and 73% of lindane could be degraded. However at 19:1 water: acetone ratio pesticides were not completely dissolved and a slightly milky solution appeared which was indicative that presence of acetone in the reaction mixture was very crucial. Overall the extent of endosulfan and lindane degradation decreases with the decreasing acetone ratio in the reaction mixture. The results are in accordance with the similar study conducted Gautam and Suresh, 2007 to dechlorinate DDT using Mg0/ K2PdCl6 bimetallic system where highest loss of DDT (84%) was obtained at 1:1 water: acetone ratio. The increase in water: acetone ratio led to a lowering of dechlorination efficiency. At 19:1 water: acetone ratio, least (40%) dechlorination efficiency was observed. In another study conducted by Patel and Suresh (2008) to dechlorinate pentachlorophenol (PCP) using Mg/Ag and Mg/Pd bimetallic systems, the effect of different co-solvents (acetone, methanol, ethanol, 1-propanol and no-solvent) on dechlorination efficiency of PCP was studied, and it was observed that maximum removal of PCP was achieved in the presence of acetone.
Table 2 e Effect of varying Water: Acetone ratio on the degradation of endosulfan and lindane (Initial Concentration of endosulfan/lindane: 30 ppm; Reaction Time: 30 min; Water: Acetone as indicated in figure). Water: Acetone Ratio (1:1) (2:1) (4:1) (9:1)
% degradation of endosulfan
% degradation of lindane
90.05 82.64 72.62 51.15
98.25 81.12 77.74 72.63
2387
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 3 8 3 e2 3 9 1
30 Residua l Endo sulfa n co nc (ppm )
Comparison of these results with the present investigation reveals that decrease in degradation efficiency on reduction of acetone is because of the in-solubility of the target compound in solvent as it is a critical parameter which influences the mass transfer and hence the extent of degradation of target compounds. Thus water: acetone ratio of 2:1 was chosen to study the degradation kinetics of endosulfan and 1:1 ratio was chosen to carryout degradation kinetics of lindane.
3.3. Kinetic studies for the degradation of endosulfan and lindane using Mg0/ZnCl2 bimetallic system
MgZn: 10/1 mg ml-1 -0.1209x
y = 30e 20 15
2
R = 0.9142 10
R = 0.9663
2
R = 0.9241
5
-0.2168x
MgZn: 7.5/0.5 mg ml-1
y = 10e
MgZn: 7.5/1.5 mg ml-1
2
R = 0.9911
6
-0.1432x
y = 10e 2
R = 0.8565 -0.0797x y = 10e
4
2
R = 0.9409 2
10
20
30
Reaction time (min)
40
50
-0.1235x
-0.0909x
y = 10e 2
R = 0.9492
0
60
MgZn: 5/1 mg ml-1
y = 50e 2
MgZn: 10/1 mg ml-1
8
-0.0829x
y = 30e
Fig. 4. As shown in Fig. 4 Mg0/ZnCl2 dose of 7.5/1 mg ml 1 showed higher k obs’ values (0.2168, 0.1209, 0.1614 min1 for 10, 30 and 50 ppm initial concentration of endosulfan respectively) when compared with other Mg0/ZnCl2 doses (Fig. 4). The k obs’ values calculated at Mg0/ZnCl2 dose of 5/1 mg ml1 were 0.155, 0.1273 and 0.1235 min1 for 10, 30 and 50 ppm initial endosulfan concentration respectively indicating that the reduction of Mg0 dose had a diminishing effect on the rate of the reaction. Further, on increasing the Mg0 dose to 10 mg ml1; k obs’ values decreased to 0.1432, 0.0976 and 0.076 min1 for 10, 30 and 50 ppm initial endosulfan concentration respectively which indicated that excess of Mg0 in the reaction mixture also hindered the reaction rate. These rate constants could also be normalized by loading of the second metal (i.e., catalytic metal) under an assumption that the reaction rate is highly affected by the extent of primary metal surface coverage with a secondary metal (Choi and Kim, 2009). In our study, the reaction rate constants were normalized by the loading of the second metal (ZnCl2), assuming that the reactivity depends on the mass of the second metal. It was observed that when ZnCl2
Residual Endosulfan conc (ppm)
Residua l Endo uslfa n co nc (ppm )
R = 0.9711
-0.0371x
y = 30e 2
50
MgZn: 7.5/1 mg ml-1
2
MgZn: 7.5/1.5 mg ml-1 -0.0976x
y = 30e
Fig. 2 e Kinetic profile for the degradation of 30 ppm initial endosulfan concentration (Mg0 dose: 5e10 mg mlL1 and ZnCl2 dose: 0.5e1.5 mg mlL1 with water: acetone ratio of 2:1v/v and 50 mL mlL1 of glacial acetic acid).
MgZn: 5/1 mg ml-1
-0.155x
MgZn: 7.5/0.5 mg ml-1
2
R = 0.894
0
y = 10e
MgZn: 7.5/1 mg ml-1
2
R = 0.9676
25
0
Figs. 1, 2 and 3 shows the time course dechlorination profiles of 10, 30 and 50 ppm initial endosulfan concentration respectively as a function of time by various Mg0/ZnCl2 doses (5/1, 7.5/1, 10/1, 7.5/0.5, 7.5/1.5 mg ml1) in the presence of glacial acetic acid (50 mL ml1). As shown in Fig. 1, Mg0/ZnCl2 dose of 5/1 mg ml1 degraded 89% of 10 ppm initial concentration of endosulfan within 60 min of reaction. At 30 ppm and 50 ppm initial concentrations of endosulfan (Figs. 2 and 3 respectively), about 92% and 94% degradation was achieved within 60 min respectively. At Mg0/ZnCl2 dose of 7.5/ 1 mg ml1, 95%, 96%, and 97% degradation of 10, 30 and 50 ppm of initial endosulfan concentrations respectively was achieved within 60 min of reaction (Figs. 1e3). While 97%, 94% and 90% degradation of 10, 30, 50 ppm initial concentrations of endosulfan was recorded within 60 min of reaction time by Mg0/ZnCl2 dose of 10/1 mg ml1 (Figs. 1e3). As shown in Figs. 1e3, 10, 30 and 50 ppm initial concentrations of endosulfan at Mg0/ZnCl2 dose of 7.5/0.5 mg ml1 were degraded with the efficiency of 87%, 89% and 92% respectively. While at Mg0/ ZnCl2 dose of 7.5/1.5 mg ml1 about 90%, 92% and 96% of 10, 30 and 50 ppm initial endosulfan concentrations were degraded respectively within 60 min of reaction. The set of data presented in Figs. 1e3 could be fitted into the first order kinetics of the reaction and indicates that the rate of the reaction is dependent upon the initial concentration of endosulfan.. The observed rate constant (k obs’) values for the set of kinetics performed were calculated and are presented in
10
MgZn: 5/1 mg ml-1
-0.1273x
y = 30e
MgZn: 7.5/1 mg ml-1
R = 0.9154
MgZn: 10/1 mg ml-1
-0.1614x
40
y = 50e
MgZn: 7.5/0.5 mg ml-1
2
R = 0.9621
MgZn: 7.5/1.5 mg ml-1
-0.076x
30
y = 50e 2
R = 0.9696 -0.184x
y = 50e
20
2
R = 0.9117
-0.141x
y = 50e 2
10
R = 0.9675
0 0
5
10
15
20
25
30
Reaction time (min)
Fig. 1 e Kinetic profile for the degradation of 10 ppm initial endosulfan concentration (Mg0 dose: 5e10 mg mlL1 and ZnCl2 dose: 0.5e1.5 mg mlL1 with water: acetone ratio of 2:1v/v and 50 mL mlL1 of glacial acetic acid).
0
5
10 15 20 Reaction time (min)
25
30
Fig. 3 e Kinetic profile for the degradation of 50 ppm initial endosulfan concentration (Mg0 dose: 5e10 mg mlL1 and ZnCl2 dose: 0.5e1.5 mg mlL1 with water: acetone ratio of 2:1v/v and 50 mL mlL1 of glacial acetic acid).
2388
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 3 8 3 e2 3 9 1
10 ppm
0.25
30 ppm
10
50 ppm
MgZn: 1/1 mg ml-1
-0.0252x
y = 10e
MgZn: 2.5/1 mg ml-1
2
R = 0.9678 8
0.15 0.1 0.05 0 (5/1)
(7.5/1)
(10/1)
(7.5/0.5)
(7.5/1.5)
Mg 0/ ZnCl2 (mg ml-1)
Fig. 4 e Observed rate constants for the degradation of endosulfan by varying Mg0/ZnCl2 dose.
Residua l Linda ne co nc (ppm )
k o bs' (m in -1)
0.2
MgZn: 5/1 mg ml-1 MgZn: 5/0.5 mg ml-1
-0.0821x
y = 10e
MgZn: 5/1.5 mg ml-1
2
R = 0.9651 6
-0.1746x
y = 10e 2
R = 0.9637 -0.0838x
y = 10e 2
4
R = 0.9042
-0.0364x
y = 10e 2
R = 0.936 2
0 0
dose was decreased to 0.5 mg ml1; lower k obs’ values were observed (0.0797, 0.0371 and 0.184 min1 for 10, 30 and 50 ppm initial endosulfan concentration respectively) while with the increase in ZnCl2 dose to 1.5 mg ml1; k obs’ values calculated were 0.0909, 0.0829 and 0.141min1 for 10, 30 and 50 ppm initial endosulfan concentration respectively. Thus from Fig. 4 it can be inferred that optimum degradation and rate constant values were achieved at Mg0/ZnCl2 dose of 7.5/1 mg ml1, thus this dose was chosen as optimum. No partially dechlorinated intermediates/end products of endosulfan degradation appeared in GC-ECD profiles. Hence the samples were analysed using GC-MS to identify the intermediates formed, if any and end products of endosulfan degradation. GC-MS analysis was carried out for the 1000 ppm initial endosulfan degradation using Mg0/ZnCl2 dose of 7.5/ 1 mg ml1. Water: acetone ratio of 2:1 was chosen and 50 mL ml1 of glacial acetic acid was added to the reaction mixture. Reaction mixtures were sacrificed at 10 min and 30 min reaction time and analysed using GC-MS. Elution profile of 10 min reaction time showed an abundant peak at 5.17 min. The molecular ion fragmentation of this peak matched with Bicyclo [2,2,1] hepta 2-5 diene using Wiley library. The structure of this compound was similar to dechlorinated endosulfan and this might have formed by the removal of all six chlorine atoms, two carbon atoms, three oxygen atoms and one sulphur atom during dechlorination reaction. The absence of any other partially dechlorinated intermediates suggests that nascent hydrogen attacks on all the six chlorine atoms simultaneously. Based on the GC MS analysis, the proposed mechanism of endosulfan degradation by Mg0/ZnCl2 system is elucidated in Fig. 5.
Cl
10
20
30
40
Figs. 6, 7 and 8 shows dechlorination kinetics of 10, 30 and 50 ppm initial lindane concentration respectively as a function of time and time course profile for lindane removal by various Mg0/ZnCl2 doses (1/1, 2.5/1, 5/1, 5/0.5, 5/1.5 mg ml1) in the presence of glacial acetic acid (50 mL ml1). As shown in Figs. 5e7, Mg0/ZnCl2 dose of 1/1 mg ml1 degraded 95%, 77% and 62% of 10, 30, and 50 ppm initial concentration of lindane respectively within 60 min of reaction. At Mg0/ZnCl2 dose of 2.5/1 mg ml1 about 72% of 10 ppm of initial lindane concentration was degraded within 60 min of reaction while 92% and 93% degradation of 30 and 50 ppm initial lindane concentrations were degraded in 60 min of reaction time. About 99% degradation of 10, 30, and 50 ppm initial concentrations of lindane was observed within 60 min of reaction by 5/1 mg ml1 dose of Mg0/ZnCl2. About 99% removal of 10 ppm initial concentration and w96% removal of 30 and 50 ppm initial concentration of lindane was achieved by Mg0/ZnCl2 dose of 5/0.5 mg ml1 (Figs. 6e8). At Mg0/ZnCl2 dose of 5/1 mg ml1, about 88% of 10 ppm lindane was degraded. However about 97% and 99% degradation was observed at 30 and 50 ppm of initial lindane concentration respectively. The set of data presented in Figs. 6e8 could be fitted into exponential curves thereby suggesting that lindane dechlorination reaction follows first order kinetics and indicates that the rate of the reaction is dependent upon the initial
Cl O
Cl
Cl
Endosulfan
Mg 0 / ZnCl2 S
O
H+
60
Fig. 6 e Kinetic profile for the degradation of 10 ppm initial lindane concentration (Mg0 dose: 1e5 mg mlL1 and ZnCl2 dose: 0.5e1.5 mg mlL1 water: acetone ratio of 1:1v/v and 50 mL mlL1 of glacial acetic acid).
Cl
Cl
50
Reaction time (min)
+
6 Cl - + 3 O + S+ 2 C
O
Bicyclo (2,2,1) hepta (2,5) diene
Fig. 5 e Proposed endosulfan degradation pathway using Mg0/ZnCl2 system.
2389
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 3 8 3 e2 3 9 1
30
MgZn: 1/1 mg ml-1
-0.0479x
y = 30e
R = 0.9803
MgZn: 5/0.5 mg ml-1
y = 30e 2
MgZn: 5/1.5 mg ml-1
R = 0.9189 -0.1968x
y = 30e 2
R = 0.952
15
k o bs' ( m in-1)
-0.1004x
Residua l Linda ne co nc (ppm )
30 ppm
50 ppm
0.2
MgZn: 5/1 mg ml-1
25
20
10 ppm
0.25
MgZn: 2.5/1 mg ml-1
2
0.15 0.1 0.05
-0.1335x
y = 30e
0
2
10
R = 0.9926 -0.1294x
(1/1)
y = 30e
(2.5/1)
5
(5/1)
(5/0.5)
(5/1.5)
Mg 0/ ZnCl2 dose (mg ml-1)
2
R = 0.9091
Fig. 9 e Observed Rate Constant for the degradation of lindane by varying Mg0/ZnCl2 concentration.
0 0
10
20
30
40
50
60
Reaction time (min)
Fig. 7 e Kinetic profile for the degradation of 30 ppm initial lindane concentration (Mg0 dose: 1e5 mg mlL1 and ZnCl2 dose: 0.5e1.5 mg mlL1 water: acetone ratio of 1:1v/v and 50 mL mlL1 of glacial acetic acid).
concentration of lindane. The observed rate constant values (kobs’) for the set of kinetics performed were calculated and are presented in Fig. 9. It was observed that at Mg0/ZnCl2 dose of 1/1 mg ml 1 the kobs’ values were calculated to be 0.0252, 0.0479 and 0.0692 min1 for 10, 30 and 50 ppm initial lindane concentration. At Mg0/ZnCl2 dose of 2.5/1 mg ml 1 the kobs’ values for 10, 30 and 50 ppm initial lindane concentration were 0.0821, 0.1004 and 0.131 min1 respectively.. Further increase in Mg0 dose to 5 mg ml1 resulted in higher kobs’ values of 0.1746, 0.1968 and 0.2253 min1 for 10, 30 and 50 ppm initial lindane concentrations (Mg0/ZnCl2 dose of 5/1 mg ml 1). At Mg0/ZnCl2 dose of 5/0.5 mg ml 1 the kobs’ values were calculated to be 0.0838, 0.1335 and 0.0896 min1 for 10, 30 and 50 ppm initial lindane concentrations respectively. However, on increasing
50
-0.0692x
MgZn: 1/1 mg ml-1
y = 50e 2
MgZn: 2.5/1 mg ml-1
R = 0.8342
MgZn: 5/1 mg ml-1 -0.131x
Residua l Linda ne co nc (ppm )
40
y = 50e
MgZn: 5/0.5 mg ml-1
2
R = 0.926
MgZn: 5/1.5 mg ml-1 -0.2253x
y = 50e
30
2
R = 0.9112 -0.0896x
y = 50e 20
2
R = 0.9558 -0.2141x
y = 50e 2
R = 0.9887
the ZnCl2 dose (1.5 mg ml 1) decrease in kobs’ values (0.0364, 0.1294 and 0.2141 min1 was observed for 10, 30 and 50 ppm initial lindane concentrations respectively. Based on the above observations Mg0/ZnCl2 dose of 5/1 mg ml1 is proposed as optimum for the lindane dechlorination. The GC-ECD profiles of kinetics of lindane degradation did not show appearance of any other peak. Hence samples were analysed using GC MS to identify the intermediates formed, if any and end products formed of lindane dechlorination. GCMS analysis was carried out with initial lindane concentration of 1000 ppm, Mg0/ZnCl2 dose of 5/1 mg ml1, in the presence of glacial acetic acid (50 mL ml1) in reaction mixture comprising water: acetone (1:1 v/v). Reaction mixtures were sacrificed at 10 min and 30 min reaction time and analysed. Elution profile after 10 min reaction time showed the appearance of an abundant peak at 2.94 min. Based on the molecular ion fragmentation of this peak, it was identified as mono chlorobenzene probably formed by the removal of 5 chlorine atoms from lindane molecule. Further, the elution profile of 30 min reaction time did not show any peak at 2.94 min indicating the complete disappearance of mono chlorobenzene. However a new peak emerged at 2.64 min which was identified as benzene based on its molecular ion fragmentation profile. The formation of benzene and absence of any partially dechlorinated intermediate suggest the complete dechlorination of lindane into its hydrocarbon skeleton, benzene. Thus Fig. 10 elucidates the proposed mechanism of lindane degradation by Mg0/ZnCl2 system. Similar results have been also observed by Gautam and Suresh (2007) and Patel and Suresh, 2008 for the dechlorination of DDT and pentachlorophenol using Mg0/Pd4þ and Mg/Ag bimetallic systems respectively wherein second order kinetics was observed. However, results obtained by Lin et al. (2004);
10 Cl Cl
Cl
Mg0/ ZnCl2
0 0
5
10
15
20
25
30
Reaction time (min)
Fig. 8 e Kinetic profile for the degradation of 50 ppm initial lindane concentration (Mg0 dose: 1e5 mg mlL1 and ZnCl2 dose: 0.5e1.5 mg mlL1 water: acetone ratio of 1:1v/v and 50 mL mlL1 of glacial acetic acid).
Cl Cl
Lindane
Cl
Cl
H+
Mg0/ ZnCl2 H+
Monochlorobenzene
Benzene
Fig. 10 e Proposed lindane degradation pathway using Mg0/ZnCl2 system.
2390
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 3 8 3 e2 3 9 1
Shih et al., 2009 and Zhang et al., 2009 reported pseudo-first order kinetics for the dechlorination of trichloroethylene, hexa chlorobenzene and 2,4- dichlorophenol respectively. First order kinetic reaction was observed in work carried out by Ghauch and Tuqan (2009) for the reductive destruction of dichlorophen using Pd, Ru and Ag. In the study conducted by Patel and Suresh (2007) dechlorination of pentachlorophenol was sequential with the formation of tri chlorophenol and tetra chlorophenol and finally phenol was identified as the end product. Xinhua et al., 2009 dechlorinated p-nitrochlorobenzene using Ni/Fe bimetallic systems and p-chloroaniline and aniline were identified as intermediate and end product respectively. In another study conducted by Shih et al. (2009) for dechlorination of hexachlorobenzene using Pd/Fe bimetallic particles, the intermediates identified were pentachlorobenzene, two isomers of tetra chlorobenzene and trichlorobenzene. Zhang et al., 2009 degraded 2,4-dichlorophenol using Ni/Fe nanoparticles and o-chlorophenol, p-chlorophenol were formed as intermediates followed by formation of phenol as end product. However, in consonance to the present study by Gautam and Suresh (2007) for the dechlorination of DDT using Mg0/Pd4þ bimetallic system no partially dechlorinated intermediates were reported and formation of its hydrocarbon skeleton, diphenyl ethane was observed in a single step.
4.
Conclusion
The following salient points emerged from the present study: 1. Mg0/ZnCl2 system was found to be an efficient system for dechlorination of endosulfan and lindane solubilized by the addition of acetone in the aqueous phase. 2. Water: acetone ratios of 2:1 and 1:1 were needed to completely dissolve endosulfan and lindane respectively. 3. Acetic acid being a weak acid facilitated slower dissolution of Mg0 granules which provided sufficient time to catalyst ZnCl2 to capture produced nascent hydrogen to form hydrides as against HCl which is a stronger acid. 4. The dechlorination reactions of endosulfan and lindane follow first order kinetics and the rate of reaction depends upon the initial concentration of target compound. 5. GC MS analyses reveals that Mg0/ZnCl2 system is efficient in complete dechlorination of endosulfan and lindane converting them into their hydrocarbon skeletons namely Bicyclo [2,2,1] hepta 2-5 diene and Benzene respectively. Authors conclude that Mg0/ZnCl2 bimetallic is a promising technology for the hydrodechlorination of environmentally problematic compounds viz. endosulfan and lindane. Also it would be worthwhile to evaluate Mg0/ZnCl2 reactive system for designing indigenous permeable barriers or reactors for contaminated water, ground water and wastewater effluent sites. However, the authors propose that before on field scaling up of Mg/Zn bimetallic system to remediate contaminated sites, a detailed study on the actual mechanism of catalysis is highly required.Surface properties of Mg0 should be charecterised as it plays the pivotal role in degradation
process. In-situ techniques such as EXAFS (EXAFS (Etended Xray Absorption Fine Structure) spectroscopy could be used study the chemical and structural nature of ZnCl2 on the Mg0 particles, before, during, and after reaction to understand the mechanism of catalysis. To the best of the author’s knowledge, no other study could be cited from open literature on the degradation of endosulfan and/or lindane using Mg0/ZnCl2 bimetallic systems.
Acknowledgements The authors would like to thank Department of Science and Technology (DST), Government of India for providing financial support to conduct this study. The authors would also like to thank Spectroscopy and Analytical Test Facility Lab, Indian Institute of Science (IISC), Bangalore, India for allowing us to use their GCeMS facility. Authors would also wish to extend their heartfelt thanks to Mr. K Johnson and Mr. Prakhar Agnihotri for their assistance extended during the project tenure.
references
Andreozzi, R., Caprio, V., Insola, A., Marotta, R., 1999. Advanced oxidation processes (AOP) for water purification and recovery. Catalysis Today 53, 51e59. Choi, J.H., Kim, Y.H., 2009. Reduction of 2,4,6-trichlorophenol with zero-valent zinc and catalyzed zinc. Journal of Hazardous Materials 166, 84e991. Esplugas, S., Bila, D.M., Krause, L.T., Dezotti, M., 2007. Ozonation and advanced oxidation technologies to remove endocrine disrupting chemicals (EDCs) and pharmaceuticals and personal care products (PPCPs) in water effluents. Journal of Hazardous Materials 149, 631e664. Gautam, S.K., Suresh, S., 2006. Dechlorination of DDT, DDD and DDE in soil (slurry) phase using magnesium/palladium system. Journal of Colloid and Interface Science 304, 144e151. Gautam, S.K., Suresh, S., 2007. Studies on dechlorination of DDT (1,1,1-trichloro-2,2-bis(4-chlorophenyl)ethane) using magnesium/palladium bimetallic system. Journal of Hazardous Materials B 139, 146e153. Ghauch, A., Tuqan, A., 2009. Reductive destruction and decontamination of aqueous solutions of chlorinated antimicrobial agent using bimetallic systems. Journal of Hazardous Materials 164, 665e674. Gogate, P.R., Pandit, A.B., 2004. A review of imperative technologies for wastewater treatment I: oxidation technologies at ambient conditions. Advances in Environmental Research 8, 501e551. Lin, C.J., Lo, S.L., Liou, Y.H., 2004. Dechlorination of trichloroethylene in aqueous solution by noble metalmodified iron. Journal of Hazardous Materials B116, 219e228. McKinlay, R., Plant, J.A., Voulvoulis, N., 2008. Endocrine disrupting pesticides: Implications for risk assessment. Environment International 34, 168e183. Mediratta, P.K., Tanwar, K., Reeta, K.H., Mathur, R., Banerjee, B.D., Singh, S., Sharma, K., 2008. Attenuation of the effect of lindane on immune responses and oxidative stress by Ocimum sanctum seed oil (OSSO) in rats. Indian Journal of Physiology & Pharmacology 52, 171e177. Mertens, I.R., 2006. Microbial monitoring and degradation of lindane in soil. Journal of Hazardous Materials 175, 680e687.
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 3 8 3 e2 3 9 1
Mu, Y., Yu, H.Q., Zheng, J., Zhang, S., Sheng, G., 2004. Reductive degradation of nitrobenzene in aqueous solution by zerovalent iron. Chemosphere 54, 789e794. Patel, U.D., Suresh, S., 2007. Dechlorination of chlorophenols using magnesiumepalladium bimetallic system. Journal of Hazardous Materials 147, 431e438. Patel, U.D., Suresh, S., 2008. Effects of solvent, pH, salts and resin fatty acids on the dechlorination of pentachlorophenol using magnesium-silver and magnesium-palladium bimetallic systems. Journal of Hazardous Materials 156, 308e316. Shih, Y.H., Chen, Y.C., Chen, M.Y., Tai, Y.T., Tso, C.P., 2009. Dechlorination of hexachlorobenzene using nanoscale Fe and nanoscale Pd/Fe bimetallic particles. Colloids and Surfaces A: Physicochemical and Engineering Aspects 332, 84e89. Simagina, V.I., Stoyanova, I.S., 2001. Hydrodechlorination of polychlorinated benzenes in the presence of a bimetallic catalyst in combination with a phase-transfer catalyst. Mandeleev Communications 11, 38e39.
2391
Usha, S., Harikrishnan, V.R., 2005. Endosulfan- Fact Sheet and Answers to Common Questions. IPEN Pesticide Working group secretariat. Wang, Z., Peng, P., Huang, W., 2009. Dechlorination of g- hexachlorocyclohexane by zero-valent metallic iron. Journal of Hazardous Materials 166, 992e997. Weber, J., Halsall, C.J., Muir, D., Teixeira, C., Small, J., Solomon, K., 2009. Endosulfan, a global pesticide: a review of its fate in the environment and occurrence in the Arctic. Science for Total Environment 408, 2966e2984. Xinhua, X.U., Jingjing, W., Jinghui, Z., Yanjum, W., Yong, L., 2009. Catalytic dechlorination of p-NCB in water by nanoscale Ni/Fe. Desalanisation 242, 346e354. Zhang, Z., Cissoko, N., Wo, J., Xu, X., 2009. Factors influencing the dechlorination of 2,4-dichlorophenol by 0Ni-Fe nanoparticles in the presence of humic acid. Journal of Hazardous Materials, 78e86.
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 3 9 2 e2 4 0 0
Available at www.sciencedirect.com
journal homepage: www.elsevier.com/locate/watres
Exposure assessment for swimmers in bathing waters and swimming pools Franciska M. Schets*, Jack F. Schijven, Ana Maria de Roda Husman National Institute for Public Health and the Environment, Laboratory for Zoonoses and Environmental Microbiology, PO Box 1, 3720 BA Bilthoven, The Netherlands
article info
abstract
Article history:
Bathing water compliant with bathing water legislation may nevertheless contain patho-
Received 13 April 2010
gens to such a level that they pose unacceptable health risks for swimmers. Quantitative
Received in revised form
Microbiological Risk Assessment (QMRA) can provide a proper basis for protective
12 October 2010
measures, but the required data on swimmer exposure are currently limited or lacking. The
Accepted 29 January 2011
objective of this study was to collect exposure data for swimmers in fresh water, seawater
Available online 1 March 2011
and swimming pools, i.e. volume of water swallowed and frequency and duration of swimming events.
Keywords:
The study related to swimming in 2007, but since the summer of 2007 was wet and this
Recreational water
might have biased the results regarding surface water exposure, the study was repeated
Bathing water
relating to swimming in 2009, which had a dry and sunny summer. Exposure data were
Swimming pool
collected through questionnaires administered to approximately 19 000 persons repre-
Exposure
senting the general Dutch population.
QMRA
Questionnaires were completed by 8000 adults of whom 1924 additionally answered the questions for their eldest child (15 years) 2007 number Total responders 4000 Swimmers 2149 Male 1068 Female 1081 Age 15e79 Non-swimmers 1851 Male 941 Female 910 Age 15e80 General health Swimmers Excellent 435 Good 1448 Mediocre 241 Poor 25 Non-swimmers Excellent 271 Good 1177 Mediocre 361 Poor 42 Water type used Swimming pool 1919 Fresh water 560 Seawater 605 Official bathing site Fresh water 386 Seawater 380 Head submersion Swimming pool 1053 Fresh water 256 Seawater 264 Health complaints after bathing 176 Swimming pool 118 Fresh water 39 Seawater 19
%
54 50 50 46 51 49
2009 number 4000 1974 969 1005 15e81 2026 1002 1024 15e98
Children (98.5% (SigmaeAldrich, St. Louis, MO, USA) and sodium dodecyl benzene sulfonate (SDBS) with a purity of >88% (Acros Organics), a non-ionic surfactant Polyoxyethylene (20) sorbitan monooleate (Tween 80) (Aldrich) and a cationic surfactant hexadecyltrimethylammonium bromide (HDTMA) with a purity of w99% (Sigma). Table 1 provides relevant properties of these surfactants. Two soils were tested in this study. A potting soil (HYPONEX, OH, USA) purchased from a local Wal-Mart store (Auburn, AL, USA) was used to represent soils of relatively high organic content, whereas a top (0.4 m) loam soil obtained from a local farm (E.V. Smith Farm, Auburn, AL, USA) was used to represent soils lean of organic matter. The Smith Farm soil is designated as Lynchburg fine sandy loam (siliceous, semiactive, thermic Aeric Paleaquults). Before use, the soils were sieved through a standard sieve of 2 mm openings, and then washed with tap water to remove fine colloids and watersoluble compositions, which can adsorb significant amounts of TCE but hardly separable from water. The washed soils can be completely separated from water through centrifugation at 400 g-force. Finally the soils were air-dried at room temperature and stored for use. Soil analyses were performed by the
Table 1 e Selected properties of surfactants used in this study. Surfactants
Ionic property
Molecular formula
Critical micellar concentration (cmc), mM
Sodium dodecyl sulfate (SDS)
Anionic
NaC12H25SO4
8.2 (Filippi et al., 1999)
Sodium dodecyl benzene sulfonate (SDBS)
Anionic
C12H25C6H4SO3Na
1.5 (Zhang et al., 2006)
Polyoxyethylene (20) sorbitan monooleate (Tween 80)
Neutral
C64H124O26
0.012 (Yeom et al., 1995)
Hexadecyltrimethyle ammonium bromide (HDTMA)
Cationic
C16H33N(Br)(CH3)3
0.9 (Karapanagioti et al., 2005; Li, 2004)
Molecular structure
2404
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 4 0 1 e2 4 1 4
Soil Testing Laboratory at Auburn University. The soil textural analysis was conducted following the bouyoucos hydrometer method. Soil pH was measured via the Reference Soil Test Methods for the Southern Region of the United States: Southern Cooperative Series Bulletin 289 (1983). The total organic carbon and sulfur were analyzed with an Elementar Vario Macro CNS Analyzer (Elementar, Hanau, Germany) at 140 F. The metal contents of the soils were analyzed with an Inductively Coupled Plasma Emission Spectroscopy (Varian Vista-MPX Axial Spectrometer, Varian, Walnut Creek, CA, USA) after acid digestion following EPA method 3230. Soil potassium, calcium, magnesium and sodium were determined by a Varian Vista-MPX Radial Spectrometer following the Melich I extraction. The pH at point of zero salt effect (PZSE) of the soil was determined following the potentiometric titration method (Marcano-Martinez and McBride, 1989). Table 2 gives salient physical and chemical properties of the soils. The hydrodynamic particle size and zeta potential of ZVI nanoparticles were determined by dynamic light scattering (DLS) (Zetasizer Nano ZS, Malvern Instruments Ltd, Malvern, Southborough, MA, USA) at 25 C. The resultant intensity data were then converted to the volume-weighted hydrodynamic diameter.
2.2.
TCE sorption tests
TCE sorption to the two soils was tested through batch isotherm tests. A series of TCE solutions at concentrations of 50, 100, 200, 300, 500 and 600 mg/L, respectively, were prepared by adding a known mass of TCE, delivered in a small volume of methanol, into deionized (DI) water. Total methanol content in the final solution was below 0.02% (v/v). To inhibit any possible biological activities during the sorption tests, 0.2 g/L of NaN3 was included in the solutions. Sorption tests were then initiated by adding 12 g of each of the soils in w63 mL of the respective TCE solution in 67 mL screw-capped glass vials sealed with PTFE-lined septa. Nearly zero-head space was maintained in the vials to avoid volatilization loss of TCE, and the mixtures were mixed on a rotator placed in an incubator at 21 1 C. Based on separate sorption kinetic tests, the mixtures were equilibrated for 1 week for the potting soil and 2 weeks for the Smith Farm soil to assure equilibrium. Upon equilibrium, the vials were centrifuged with a Fisher Marathon 21K/R Centrifuge (Fisher Scientific) at 400 g-force for 10 min. Then, 100 mL of the supernatant was withdrawn using a 100 mL gastight glass-syringe and transferred to 1 mL of hexane in a 2-mL GC vial. Upon phase separation, TCE in hexane was analyzed using an HP 6890GC (Hewlett Packard, Palo Alto, CA, USA) equipped with an electron capture detector (ECD) following the method by He and Zhao (2005).
Mass balance results showed that the overall recovery of TCE was always within 90e105%.
2.3.
Effects of surfactants on TCE desorption
To examine the physical availability of soil-sorbed TCE, desorption kinetic tests were carried out with the same batch reactors as in the isotherm tests and the soils that were preequilibrated with TCE. Upon centrifuging, about 93% of the supernatant was pipetted out and replaced with soil-amended water, which was prepared by mixing DI water and a TCE-free soil at the same soil to water ratio as in the sorption tests. The amendment ensures that the background compositions (e.g. dissolved SOM) during sorption and desorption remain identical. Again, 0.2 g/L of NaN3 was maintained to minimize biological activity. The vials were resealed and mixed on the rotator at 21 1 C. At selected times, the suspension was centrifuged and the supernatant was extracted by hexane and analyzed with GC-ECD following the same method as aforedescribed. The amount desorbed from the soils was obtained via mass balance calculations. To test effect of surfactants on the desorption rate, desorption kinetic tests were carried out in the presence of the four surfactants at initial concentrations of 1cmc and 5cmc values.
2.4.
Effects of surfactants on TCE degradation in water
CMC-stabilized ZVI nanoparticles were prepared at 0.1 g/L as Fe following the approach of aqueous phase reduction with borohydride as described in our previous study (He and Zhao, 2007). Trace amounts (0.1 wt% or 0.3 wt% of Fe) of Pd catalyst was added to the fresh ZVI particles by adding a known amount of Na2PdCl4 into the nanoparticle suspension. The addition of Pd was able to greatly enhance the dechlorination rate of TCE (He and Zhao, 2005, 2008). Batch degradation tests were carried out using 43 mL amber glass vials with open-top screw caps and PTFE-lined septa. To test the effect of each surfactant, 1 mL of a surfactant stock solution was added into the FeePd nanoparticle suspension to yield a desired concentration level. TCE degradation was then initiated by injecting 25 mL of a TCE stock solution, resulting in an initial TCE concentration of 10 mg/L for all cases. The mixture was then mixed on a rotator (50 rpm) at room temperature. At selected times, 100 mL of aqueous samples were taken, extracted with hexane, and analyzed via GC-ECD for TCE. Parallel control experiments were conducted with 0.2% CMC solution but without the nanoparticles. Mass balance analyses of TCE in the control tests indicated that the mass loss was 1 kDa, with high residual COD representing the compounds 1 kDa) (Fig. 4e and f). The remaining DON consisted mainly of organics with MW < 3 kDa (Fig. 4g) after the longest UV/H2O2 oxidation time, whereas the residual DOC represented mostly molecules
smaller than 1 kDa (Fig. 4h). The main part of the initial DON in the BD ROC was comprised of molecules with MW larger than 3 kDa. Less than 10% removal observed for this fraction (i.e. 5 kDa) and >20% (>0.5 kDa) 80% (>3 kDa) and 60% (3 kDa), 1 kDa), 5 kDa)
400 mg L1 H2O2; 6 kWh m3 80% or higher (>1 kDa)
LP AOP BD MIEX
BD AOP
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 4 1 5 e2 4 2 7
4.
Discussion
4.1.
ROC characterisation
In general, the main characteristics of the two streams were similar, apart from specific aspects related to the catchment. There was an additional non-coloured organic compound in BD ROC, possibly due to its inland status. The location may affect the feed water characterisation of the BD AWTP as the feed contains less salinity and more complex NOM from the vegetation life cycle. The BD plant also had a higher recovery ratio which leads to higher concentration of rejected organics in the ROC stream. The higher level of ammonia measured in the BD ROC is consistent with chloramination which is applied for bio-fouling control prior to RO treatment in the BD AWTP.
80% (>0.5 kDa) and 40% (all sizes); except 5% (0.5e1 kDa) 5% (0.5e1 kDa), 50% (other sizes) 40% or higher (>1 kDa), and 10% (1 kDa), lowest (5 kDa), 20% (10 kDa) Fractionation
BD Alum LP Alum Source/ Treatment
Table 4 e Impact of treatment on each size fraction.
LP Iron
BD Iron
LP MIEX
4.2.
Optimal treatment conditions
As coagulation is pH dependent, the pH needs to be maintained at a specific optimum value allowing the coagulant to work effectively. Coagulant addition caused a large drop of pH from its original value, e.g. alum addition of 1.5 mM caused a pH of ROC dropped from pH 7.8 to around pH 4. Under the correct conditions of dosage and pH the coagulation process is optimal in terms of obtaining good flocs and effective organics removal (Jarvis et al., 2005). Coagulation is most effective when monomers formation like Al (OH)3 and Fe(OH)3 are maximised at optimal pH (Dominguez et al., 2007; Duan and Gregory, 2003). Such monomers are more effective to settle than other formations. Coagulation for both concentrates provided substantial, but not high removal of colour and organics (as COD, DOC, and DON), consistent with previous study indicating pH 5e6 for alum and more acidic pH 4.5e5.5 for iron as optimum point (Sharp et al., 2006b). In general, higher resin dose allows higher removal of colour and organic compounds. However, doses above 15 mL L1 did not significantly enhance the treatment performance (according to t-tests within the error bars). Increased resin dose shows a higher impact than prolonging the contact time. After 10e20 min of contact time organic compounds reached adsorption equilibrium, lowering its further removal. This optimum contact time was similar with previous study (Humbert et al., 2005). They reported that the process of DOC removal by MIEX reached maximum equilibrium at 10e15 min of contact time. Optimisation of advanced oxidation utilising UV/H2O2 includes both chemical dose and power consumption. The oxidant consumed by UV light to produce free radicals has an optimal point which yields high kinetics rate of oxidation. Further addition of oxidant above that point would slow down the oxidation kinetics as less reactive radicals would be formed. Knowing depletion of H2O2 during the experiment is essential to evaluate the efficiency of the process (Wang et al., 2000), including the need of an efficient power (energy) consumption.
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 4 1 5 e2 4 2 7
2423
Fig. 4 e Fractions of treated organics and colour in LP ROC (left) and BD ROC (right); (a) 4.5 mM alum, (b) 1.5 mM alum, (c and d) 1.48 mM iron, (e) 10 mL LL1 MIEX at 20 min contact time, (f) 10 mL LL1 MIEX at 40 min contact time, (g and h) 40 mg LL1 H2O2 at 5.9 kWh mL3. The error bars represent ninety-five percent confidence interval of three replicate samples.
2424
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 4 1 5 e2 4 2 7
Fig. 5 e EEM characterisation on the fractions of treated ROC; (a) >10 kDa treated LP ROC by 4.5 mM alum at pH 5, (b) 5e10 kDa treated BD ROC by 10 mL LL1 MIEX at 40 min contact time, (c) 5e10 kDa treated LP ROC by 1.48 mM iron at pH 5, and (d) 3 kDa) that contributed to the initial DON were largely eliminated in the MIEX and AOP treatments for both ROC samples. Smaller MW compounds of the initial DON were more difficult to remove. DON was poorly removed in MIEX and AOP because it consists mainly in smaller MW organics, which could also be neutral or positively charged compounds and minimally reactive to AOP. Moreover, MIEX had a low affinity to form agglomerates with smaller MW organic as there were more competitive large MW organics in the ROC. In the UV/H2O2 oxidation process, the residual organics were mainly in the smaller MW ranges and were probably the breakdown by-product of larger molecules, creating more biodegradable compounds (Dwyer and Lant, 2008; Kerc et al., 2004). This process appeared simultaneously with the inorganics formation and some gas production due to the oxidation to bicarbonate and nitrogen gas as mentioned in Dwyer et al. (2008). Complete decolourisation appeared for all molecular size fractions except for the organics with MW < 0.5 kDa in the BD ROC (which around 90% removal of colour). This also confirms that the smaller MW organic compounds are the most difficult to treat. Most humic acids and SMPs were well removed by the AOP, as can be seen from the EEM contour plots (Fig. 5). Alum had the highest efficiency in removing the organic fraction of the LP ROC with MW > 10 kDa. This is consistent with literature observation (Shon et al., 2006), which found that large molecules are treated most efficiently using coagulant methods. Alum was ineffective for the BD ROC with larger MW organics (i.e. 3 kDa in both ROC samples. However, iron coagulation was capable of removing a broader size range of organic compounds than the alum coagulation. The remaining organics were mainly in the lower MW range ( 0.86 (Fig. 4b), though affected by 1 scattering by inorganic particles. Multiplication of ðR1 665 R708 Þ
Fig. 3 e Spectra of root mean square error (RMSE) of the linear relationship between Chl-a concentration and 3band NIR-red model (Eq. (1)): (a) l2 is the wavelength where the reflectance is minimally sensitive to absorption by Chla and l3 is the wavelength where the reflectance is minimally sensitive to absorption by all suspended and dissolved constituents in water; (b) l1 is the wavelength where the reflectance is maximally sensitive to absorption by Chl-a. Shaded areas delineate the wavebands which correspond (from left to right) to MERIS bands 7, 9 and 10.
by the reflectance at 753 nm, as in the three-band model (Eq. (1)), significantly minimized the effect of scattering by inorganic particles and resulted in a closer relationship with Chl-a concentration, with r2 > 0.93 (Fig. 4c). A two-band model (Eq. (3)), widely used for Chl-a estimation (Gitelson, 1992), was also accurate (r2 > 0.94) (Fig. 4d). Thus, both the three-band (Eq. (1), Fig. 4c) and the two-band (Eq. (3), Fig. 4d) models had close relationships with Chl-a concentration, with high correlation coefficients. In all cases a non-linear regression produced a better fit than a linear regression, but the latter still yielded high values of correlation coefficient. The two-band MODIS NIR-red model (Eq. (5)), which was based on the approach of Stumpf and Tyler (1988), yielded a moderate correlation with Chl-a concentration (r2 ¼ 0.52, Fig. 5a). Due to the merely moderate correlation of the twoband MODIS NIR-red model with Chl-a concentrations and its general inability to accurately estimate low-to-moderate Chla concentrations (Moses et al., 2009a), no attempt was made to calibrate this model or do further assessment of its accuracy in estimating Chl-a concentration. The three-band (Eq. (6)) and two-band (Eq. (7)) NIR-red models with simulated MERIS bands yielded very high correlations with Chl-a concentration, with quadratic polynomial
2432
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 4 2 8 e2 4 3 6
Fig. 4 e Models with narrow spectral bands plotted versus Chl-a concentration: (a) Reciprocal reflectance at 665 nm, (b) The difference of reciprocal reflectances at 665 and 708 nm; (c) Three-band model based on the reflectance at 665, 708 and 753 nm; (d) Two-band model (Eq. (3)) with wavelengths 708 nm and 665 nm.
best fit functions (Fig. 5b, c). The linear relationships between the model values and Chl-a concentrations were as follows: Chl-a ¼ 80.167 (3-band model) þ 17.105
(8)
Chl-a ¼ 41.127 (2-band model) 23.484
(9)
The MERIS-based three-band and two-band NIR-red models were previously parameterized and calibrated in small and shallow lakes in Nebraska, USA, where Chl-a concentrations ranged from 2 to 200 mg m3 (Gitelson et al., 2009). For Chl-a concentrations not surpassing 25 mg m3, the relationships between the models and Chl-a concentration for the Nebraska lakes dataset were as follows: Chl ¼ 142.27 (3-band model) þ 19.516
(10)
Chl ¼ 45.535 (2-band model) 25.895
(11)
As can be seen, the coefficients of Eq. (9) for the two-band NIR-red model is similar to the model calibrated using data from Nebraska lakes, Eq. (11). We applied the two-band and three-band algorithms calibrated using Nebraska lakes data (Eqs. (10) and (11)) to the data collected in Lake Kinneret. The results of this validation test are presented in Fig. 6. For the two-band model, the relationship between the estimated (Chlest) and measured Chl (Chlmeas) concentrations was:
Chlest ¼ 0.985Chlmeas þ 0.6814
(12)
with the RMSE of the estimated Chl-a concentration less than 1.25 mg m3 and the mean normalized bias below 5.5% (Fig. 6a). For the three-band model, the relationship was Chlpred ¼ 1.386Chlmeas 5.0202
(13)
with an RMSE of 2.61 mg m3 and the mean normalized bias below 46% (Fig. 6b). While the two-band algorithm calibrated in Nebraska was very accurate in estimating Chl-a concentrations over the entire range in Lake Kinneret (Fig. 6c), the three-band algorithm was accurate only for Chl-a concentrations above 10 mg m3 and exhibited a significant underestimation at lower Chl-a concentrations (Fig. 6d).
4.
Discussion
4.1.
Optical effect of water constituents
Seldom are pigment features as clearly evident in reflectance spectra as seen in our study. That is particularly the case in productive waters, where non-pigmented particles and CDOM mask the pigment signature in the blue range of the
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 4 2 8 e2 4 3 6
2433
Fig. 5 e NIR-Red models with simulated bands of the space-borne sensors plotted versus Chl-a concentration: (a) two-band MODIS NIR-red model (Eq. (5)); (b) two-band MERIS NIR-red model (Eq. (7)); (c) three-band MERIS NIR-red MERIS model (Eq. (6)).
electromagnetic spectrum (e.g., Dekker, 1993; Schalles, 2006). The troughs at 440 nm and 670 nm are formed by Chla absorption, and the prominent peak around 560 nm is an outcome of minimum absorption by all pigments. Although the optical signature of Chl-a around 670 nm is clearly identified in all but the most oligotrophic waters (Schalles, 2006), the appearance of a trough around 440 nm, is not common in spectra acquired in productive coastal and inland waters. The optical activity of detritus and CDOM declines exponentially from 400 nm towards longer wavelengths, but in productive water is often high enough to mask pigment absorption (Gege and Albert, 2006). The effect of the absorption of CDOM on reflectance is often a major factor that renders the blue range of the electromagnetic spectrum ineffective for use in estimating Chl-a concentration in productive waters. But, CDOM concentration in Lake Kinneret during the time period of the reported study was extremely small (absorption coefficient of filtrate passing through a 0.45 mm filter at 440 nm, was 0.93). However, in addition to the ratio R708/R665, Gons’ model involves reflectance in NIR region at 778 nm, which is in the region of very high water absorption and, thus, very low reflectance. Hence, while performing well for data taken with field spectrometers at water surface level, this model is quite susceptible to effects arising from low signal-to-noise ratio in the detector and residual effects from
2434
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 4 2 8 e2 4 3 6
Fig. 6 e Comparison of Chl-a concentration determined in laboratory and Chl-a concentration estimated using NIR-red models calibrated in Fremont Lakes, NE: (a) the two-band NIR-red model (Eq. 11), and (b) the three-band NIR-red model (Eq. 10), with the 1-by-1 line (dashed line) and the best fit function (solid line). The relative errors of Chl-a estimation by the (c) two-band and (d) three-band models are also shown.
atmospheric correction when applied to satellite data, whereas the two-band model (Eq. (3)) is not affected by these factors to the same degree. While the MERIS-based two-band NIR-red model gave consistently highly accurate estimates over the entire range of Chl-a concentrations in our dataset, the three-band NIR-red model yielded less accurate estimates for Chl-a concentrations less than 10 mg m3. The three-band model, which involves the use of information acquired at l3 to remove the effects of particulate backscattering, is theoretically robust. However, it relies on the assumption of spectral uniformity of backscattering coefficient over the entire range of wavelengths (l1el3), thus, between 660 and 750 nm. Such an assumption may be invalid for inland waters (Gons, 1999; Oki and Yasuoka, 2002). Moreover, there are also documented instances of non-uniform backscattering in the visible and NIR regions and evidences that the pattern of this non-uniformity might vary across water bodies (Herlevi, 2002; Kutser et al., 2009; Aas et al., 2005). The effects of such non-uniformity in backscattering coefficient will have a higher impact on the accuracy of the three-band model at low Chl-a concentrations. Moreover, with the two-band model spanning a lower range of wavelengths than the three-band model, the effects of the spectral non-uniformity of backscattering will be less pronounced in the two-band model. We, therefore, postulate that the spectral non-uniformity of backscattering coefficient is a primary factor that caused the MERIS-based three-band NIR-red model to be less accurate than the MERIS-based
two-band NIR-red model at Chl-a concentrations lower than 10 mg m3. Thus, the MERIS-based two-band NIR-red model seems to be the most optimal model for estimating low-tomoderate Chl-a concentrations in turbid productive waters such as Lake Kinneret.
5.
Conclusions
The reflectance spectra in the dataset were relatively uniform, i.e., the location of peaks and troughs showed only minor shifts with changes in Chl-a concentration. The MERIS-based NIR-red models had a consistent close correlation with Chl-a concentration. The rationale behind the waveband choice for the construction of NIR-red algorithms for turbid productive waters have been outlined in a recent review (Gitelson et al., 2011), and tested in several campaigns in different water bodies (Gitelson et al., 2008, 2009; Moses et al., 2009a). The MERIS-based NIR-red algorithms developed, parameterized, and calibrated for lakes in Nebraska, were reliable for estimating Chl-a concentration in Lake Kinneret. Similar results were obtained when the algorithms calibrated for lakes in Nebraska were used to estimate Chl-a concentrations in the Azov Sea, Russia, using actual MERIS data (Moses et al., 2009b) and when they were applied to reflectances simulated by the radiative transfer model, Hydrolight (Gilerson et al., 2010). This strongly suggests that the MERIS-based NIR-red algorithms, especially the two-band NIR-red algorithm, do not
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 4 2 8 e2 4 3 6
need to be re-parameterized for waters with varying biophysical characteristics, and have a strong potential for being applied universally for turbid productive waters around the globe. Our study shows the robustness of the MERIS-based NIR-red algorithms at low-to-moderate Chl-a concentrations, which are typical for mesotrophic waters around the globe. However, further tests need to be done to validate the universal applicability of these algorithms for inland and coastal waters.
Acknowledgements We would like to thank skipper Moti Diamant for his reliable partnership in the execution of our lake missions. We gratefully acknowledge the use of radiometers provided by the Center for Advanced Land Management Information Technologies (CALMIT), University of Nebraska-Lincoln. We acknowledge the contribution of three anonymous reviewers who provided constructive criticisms that helped to improve the clarity and quality of the presentation. This work was partially supported by the Lake Kinneret Extended Monitoring Program, funded by the Israeli Water Authority.
references
Aas, E., Høkedal, J., Sørensen, K., 2005. Spectral backscattering coefficient in coastal waters. Int. J. Remote Sens 26, 331e343. Bidigare, R.R., Ondrusek, M.E., Morrow, J.H., Kiefer, D.A., 1990. In vivo absorption properties of algal pigments. SPIE 1302, 290e302. Ocean Optics X. Dall’Olmo, G., Gitelson, A.A., 2005. Effect of bio-optical parameter variability on the remote estimation of chlorophylla concentration in turbid productive waters: experimental results. Appl. Opt. 44, 412e422. see also Erratum, Appl. Opt., 44, 3342. Dall’Olmo, G., Gitelson, A.A., Rundquist, D.C., 2003. Towards a unified approach for remote estimation of chlorophyll- a in both terrestrial vegetation and turbid productive waters. Geoph. Res. Lett. 30, 1938. doi:10.1029/2003GL018065. Dekker, A., 1993. Detection of the optical water quality parameters for eutrophic waters by high resolution remote sensing. Ph.D. thesis, Free University, Amsterdam, The Netherlands. Eckert, W., Parparov, A., 2006. Feasibility Study for Monitoring Dissolved and Particulate Carbon in Lake Kinneret, IOLR Report T15/06. Israel Oceanographic and Limnological Research, Tabgha. Field, C.B., Behrenfeld, M.J., Randerson, J.T., Falkowski, P., 1998. Primary production of the biosphere: integrating terrestrial and oceanic components. Science 281, 237e240. Gege, P., Albert, A., 2006. A tool for inverse modeling of spectral measurements in deep and shallow waters. In: Richardson, L., LeDrew, E. (Eds.), Remote Sensing of Aquatic Coastal Ecosystem Processes: Science and Management Applications. Springer Verlag, New York, pp. 81e109. Gilerson, A.A., Gitelson, A.A., Zhou, J., Gurlin, D., Moses, W.J., Ioannou, I., Ahmed, S.A., 2010. Algorithms for remote estimation of chlorophyll-a in coastal and inland waters using red and near-infrared bands. Optics Express 18, 24109e24125. Gitelson, A.A., 1992. The peak near 700 nm on reflectance spectra of algae and water: relationships of its magnitude and position
2435
with chlorophyll concentration. Int. J. Remote Sens. 13, 3367e3373. Gitelson, A.A., Dall’Olmo, G., Moses, W., Rundquist, D.C., Barrow, T., Fisher, T.R., Gurlin, D., Holz, J., 2008. A simple semi-analytical model for remote estimation of chlorophylla in turbid waters: validation. Remote Sens. Environ. 112, 3582e3593. Gitelson, A.A., Gurlin, D., Moses, W.J., Barrow, T., 2009. A biooptical algorithm for the remote estimation of the chlorophyll-a concentration in case 2 waters. Environ. Res. Lett. 4. doi:10.1088/1748-9326/4/4/045003. Gitelson, A.A., Gurlin, D., Moses, W.J., Yacobi, Y.Z., 2011. Remote estimation of chlorophyll-a concentration in inland, estuarine, and coastal waters. Chapter 18. In: Weng, Q. (Ed.), Advances in Environmental Remote Sensing: Sensors, Algorithms and Applications. CRC Press, Boca Raton, FL, USA, pp. 449e478. Gons, H.J., 1999. Optical teledetection of chlorophyll a in turbid inland waters. Environ.Sci. Technol. 33, 1127e1132. Gons, H.J., Rijkeboer, M., Ruddick, K.G., 2002. A chlorophyllretrieval algorithm for satellite imagery (medium resolution imaging spectrometer) of inland and coastal waters. J. Plankton Res. 24, 947e951. Gons, H.J., Rijkeboer, M., Ruddick, K.G., 2005. Effect of a waveband shift on chlorophyll retrieval from MERIS imagery of inland and coastal waters. J. Plankton Res. 27, 125e127. Gordon, H., Morel, A., 1983. Remote Assessment of Ocean Color for Interpretation of Satellite Visible Imagery. A Review. In: Lecture Notes on Coastal and Estuarine Studies 4. SpringerVerlag. Gordon, H.R., Brown, O.B., Jacobs, M.M., 1975. Computed relationships between the inherent and apparent optical properties of a flat homogeneous ocean. Appl. Opt. 14, 417e427. Han, L., Rundquist, D., 1997. Comparison of NIR/RED ratio and first derivative of reflectance in estimating algal-chlorophyll concentration: a case study in a turbid reservoir. Remote Sens. Environ. 62, 253e261. Herlevi, A., 2002. A study of scattering, backscattering and a hyperspectral reflectance model for boreal waters. Geophysica 38, 113e132. Holm-Hansen, O., Lorenzen, C.J., Holmes, R.W., Strickland, J.D.H., 1965. Fluorometric determination of chlorophyll. J. Cons. Cons. Int. Explor. Mer. 30, 3e15. Hoge, E.F., Wright, C.W., Swift, R.N., 1987. Radiance ratio algorithm wavelengths for remote oceanic chlorophyll determination. Appl. Opt. 26, 2082e2094. Joint, I., Groom, S.B., 2000. Estimation of phytoplankton production from space: current status and future potential of satellite remote sensing. J. Exp. Mar. Biol. Ecol. 250, 233e255. Kutser, T., Hiire, M., Metsamaa, L., Vahtma¨e, E., Paavel, B., Aps, R., 2009. Field measurements of spectral backscattering coefficient of the Baltic Sea and boreal lakes. Boreal Environ. Res. 14, 305e312. Le, C., Li, Y., Zha, Y., Sun, D., Huang, C., Lu, H., 2009. A four-band semi-analytical model for estimating chlorophyll a in highly turbid lakes: the case of Taihu Lake, China. Remote Sens. Environ. 113, 1175e1182. Moses, W.J., Gitelson, A.A., Berdnikov, S., Povazhnyy, V., 2009a. Estimation of chlorophyll-a concentration in case II waters using MODIS and MERIS data-successes and challenges. Environ. Res. Lett. 4. doi:10.1088/1748-9326/4/4/045005. Moses, W.J., Gitelson, A.A., Berdnikov, S., Povazhnyy, V., 2009b. Satellite estimation of chlorophyll- a concentration using the red and NIR bands of MERIS e the Azov Sea case study. IEEE Geosci. Remote Sens. Lett. 6, 845e849. Ohde, T., Siegel, H., 2003. Derivation of immersion factors for the hyperspectral Trios radiance sensor. J. Opt. A Pure Appl. Opt. 5, 12e14.
2436
w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 2 4 2 8 e2 4 3 6
Oki, K., Yasuoka, Y., 2002. Estimation of chlorophyll concentration in lakes and inland seas with a field spectroradiometer above the water surface. Appl. Opt. 41, 6463e6469. Pierson, D., Strombeck, N., 2000. A modeling approach to evaluate preliminary remote sensing algorithms: use of water quality data from Swedish great lakes. Geophysica 36, 177e202. Schalles, J.F., 2006. Optical remote sensing techniques to estimate phytoplankton chlorophyll a concentrations in coastal waters with varying suspended matter and CDOM concentrations. In: Richardson, L., LeDrew, E. (Eds.), Remote Sensing of Aquatic Coastal Ecosystem Processes: Science and Management Applications. Springer Verlag, New York, pp. 27e79.
Stumpf, R.P., Tyler, M.A., 1988. Satellite detection of bloom and pigment distributions in estuaries. Remote Sens. Environ. 24, 385e404. Yacobi, Y.Z., 2006. Temporal and vertical variation of chlorophyll a concentration, phytoplankton photosynthetic activity and light attenuation in Lake Kinneret: possibilities and limitations for simulation by remote-sensing. J. Plankton Res. 28, 725e736. Yacobi, Y.Z., Gitelson, A.A., Mayo, M., 1995. Remote sensing of chlorophyll in Lake Kinneret using high spectral resolution radiometer and Landsat TM: spectral features of reflectance and algorithm development. J. Plankton Res. 17, 2155e2173.
water research 45 (2011) 2437
Available at www.sciencedirect.com
journal homepage: www.elsevier.com/locate/watres
Comment
Comment on “Adsorption mechanism of selenate and selenite on the binary oxide systems” by Y.T. Chan et al. [Water Research 43 (2009) 4412-4420] Yuh-Shan Ho* Department of Environmental Sciences, College of Environmental Science and Engineering, Peking University, Beijing 100871, People’s Republic of China
article info Article history: Received 4 November 2009 Accepted 14 April 2010 Available online 24 April 2010
Recently, Chan et al. (2009) published the paper entitled above. In section of 2.4. Adsorption kinetics, the authors noticed “the initial sorption rate” with Eq. (4) without any citations. In fact, Ho has presented a definition for the initial adsorption rate from the parameters of pseudo-second-order model (Ho, 1995; Ho et al., 1996). A modified initial adsorption rate equation has also been made in the following years because a mistake was included in the paper published in 1996 (Ho and McKay, 1998; Ho, 2006). In addition, authors cited Saha et al. (2004) for pseudo-second-order kinetic model. In fact there is nothing related the model in the reference. Accuracy of citations and quotations are very important for the transmission of scientific knowledge. I suggest that Chan et al. cite the original or the most frequently cited papers for the initial adsorption rate to have more accuracy and details of information about its expression.
references
Chan, Y.T., Kuan, W.H., Chen, T.Y., Wang, M.K., 2009. Adsorption mechanism of selenate and selenite on the binary oxide systems. Water Res. 43, 4412e4420. Ho, Y.S., 1995. Adsorption of heavy metals from waste streams by peat. Ph.D. thesis, The University of Birmingham, Birmingham, U.K. Ho, Y.S., 2006. Review of second-order models for adsorption systems. J. Hazard. Mater. 136, 681e689. Ho, Y.S., McKay, G., 1998. Sorption of dye from aqueous solution by peat. Chem. Eng. J. 70, 115e124. Ho, Y.S., Wase, D.A.J., Forster, C.F., 1996. Kinetic studies of competitive heavy metal adsorption by sphagnum moss peat. Environ. Technol. 17, 71e77. Saha, U.K., Liu, C., Kozak, L.M., Huang, P.M., 2004. Kinetics of selenite adsorption on hydroxyaluminum- and hydroxyaluminosilicateemontmorillonite complexes. Soil Sci. Soc. Am. J. 68 (4), 1197e1209.
DOI of original article: 10.1016/j.watres.2009.06.056. * Tel./fax: þ86 10 62751923. E-mail address:
[email protected] 0043-1354/$ e see front matter ª 2010 Elsevier Ltd. All rights reserved. doi:10.1016/j.watres.2010.04.013
