Bayesian modeling approach for characterizing groundwater arsenic contamination in the Mekong River basin

Chemosphere xxx (2015) xxx–xxx

Contents lists available at ScienceDirect

Chemosphere journal homepage: www.elsevier.com/locate/chemosphere

Bayesian modeling approach for characterizing groundwater arsenic contamination in the Mekong River basin YoonKyung Cha a, Young Mo Kim b, Jae-Woo Choi c, Suthipong Sthiannopkao d,⇑, Kyung Hwa Cho e,⇑ a

Cooperative Institute for Limnology and Ecosystems Research, University of Michigan, Ann Arbor, MI 48108, United States School of Environmental Science and Engineering, Gwangju Institute of Science and Technology (GIST), 261 Cheomdan-gwagiro, Buk-gu, Gwangju 500-712, Republic of Korea c Center for Water Resource Cycle Research, Korea Institute of Science and Technology, Hwarangno 14-gil 5, Seongbuk-gu, Seoul 136-791, Republic of Korea d Department of Environmental Engineering, Dong-A University, Busan 604-714, Republic of Korea e School of Urban and Environmental Engineering, Ulsan National Institute of Science and Technology, Ulsan 689-798, Republic of Korea b

h i g h l i g h t s  The groundwater in the Mekong River basin delta contained high AsTOT and As(III).  A Bayesian change-point model identified the threshold level of Eh, 100 (±15) mV.  Below the change-point, AsTOT increased with increasing Eh.  Above the change-point, AsTOT sharply decreased as Eh increased.  AsTOT was positively related to pH over the entire range of Eh levels.

a r t i c l e

i n f o

Article history: Received 10 December 2014 Received in revised form 13 February 2015 Accepted 17 February 2015 Available online xxxx Handling Editor: I. Cousins Keywords: Arsenic (As) contamination Groundwater Mekong River basin Bayesian change-point model Linear model Drinking water source

a b s t r a c t In the Mekong River basin, groundwater from tube-wells is a major drinking water source. However, arsenic (As) contamination in groundwater resources has become a critical issue in the watershed. In this study, As species such as total As (AsTOT), As(III), and As(V), were monitored across the watershed to investigate their characteristics and inter-relationships with water quality parameters, including pH and redox potential (Eh). The data illustrated a dramatic change in the relationship between AsTOT and Eh over a specific Eh range, suggesting the importance of Eh in predicting AsTOT. Thus, a Bayesian change-point model was developed to predict AsTOT concentrations based on Eh and pH, to determine changes in the AsTOT–Eh relationship. The model captured the Eh change-point (100 ± 15 mV), which was compatible with the data. Importantly, the inclusion of this change-point in the model resulted in improved model fit and prediction accuracy; AsTOT concentrations were strongly negatively related to Eh values higher than the change-point. The process underlying this relationship was subsequently posited to be the reductive dissolution of mineral oxides and As release. Overall, AsTOT showed a weak positive relationship with Eh at a lower range, similar to those commonly observed in the Mekong River basin delta. It is expected that these results would serve as a guide for establishing public health strategies in the Mekong River Basin. Ó 2015 Elsevier Ltd. All rights reserved.

1. Introduction Depending on water availability and quality, drinking water resources can be obtained from various water bodies, including rivers, lakes, reservoirs, and groundwater. Although a large population resides alongside the Mekong River and its tributaries, groundwater resources remain the primary sources of drinking

⇑ Corresponding authors. Tel.: +82 52 217 2829; fax: +82 52 217 2819 (K.H. Cho). E-mail addresses: [email protected] (S. Sthiannopkao), [email protected] (K.H. Cho).

water, household water use, and irrigation in Southeast Asian countries. Groundwater is the preferred drinking water resource in these regions mainly due to its stable biochemical properties (Schmoll et al., 2006). However, groundwater contamination from arsenic (As) has become a critical issue in the developing world, especially in Southeast Asia. Specifically, inorganic As, the dominant form found in groundwater, is a known carcinogen, and as such poses a critical threat to public health (Bagla and Kaise, 1996; AWWA, 2001; Berg et al., 2006; Kocar and Fendorf, 2009; Sthiannopkao et al., 2010; Mondal et al., 2013). Previous studies have reported on the status of groundwater As concentrations in Asian countries, many of which remain in

http://dx.doi.org/10.1016/j.chemosphere.2015.02.045 0045-6535/Ó 2015 Elsevier Ltd. All rights reserved.

Please cite this article in press as: Cha, Y., et al. Bayesian modeling approach for characterizing groundwater arsenic contamination in the Mekong River basin. Chemosphere (2015), http://dx.doi.org/10.1016/j.chemosphere.2015.02.045

2

Y. Cha et al. / Chemosphere xxx (2015) xxx–xxx

non-compliance with 10 lg L1, the safe World Health Organization (WHO) drinking water guideline (Berg et al., 2001, 2007; Smedley and Kinniburgh, 2002; Sun et al., 2002; Polya et al., 2003, 2005; Stanger et al., 2005; Kohnhorst, 2005; Tetsuro et al., 2006; Chauhan et al., 2008; Chiew et al., 2009; Rahman et al., 2015). To this end, Sun et al. (2002) estimated that approximately 200 million people in South Asia are currently being exposed to the toxic effects of As. In the Kandal province of Cambodia, Sthiannopkao et al. (2008) found that AsTOT concentrations ranged from a non-detectable level to approximately 900 lg L1, and that 54% of collected groundwater samples exceeded the safe level of 10 lg L1 (Allan et al., 2002). As-contaminated regions can generally be characterized by four different hydro-geological features: (1) alluvial and mild-slope topographies, (2) fast holocene sedimentation, (3) enriched degradable organic materials, and (4) stagnant groundwater flow (Chanpiwat et al., 2011). To address public health issues arising from groundwater As contamination, regulations and monitoring strategies should be either developed or further reinforced in Southeast Asian countries. However, intensive monitoring efforts could be a challenging task for many of these countries because groundwater As measurements typically require advanced equipment, highly experienced technicians, and incur a high maintenance cost. Given the limited financial and human resources available, alternative modeling approaches that characterize As patterns and distributions using water quality parameters that are straightforward to measure should be developed; these can then be used to provide a scientific basis for establishing public health strategies. Previous studies developed statistical models for modeling groundwater As concentrations using water quality parameters as input variables (Purkait et al., 2008; Chang et al., 2010; Cho et al., 2011). For example, using an Artificial Neural Network (ANN) with regression models, Purkait et al. (2008) predicted the groundwater As contamination in Eastern India. Chang et al. (2010) employed an ANN model to impute missing values in As concentration datasets for an area of Taiwan. And Cho et al. (2011) compared the performance among linear and nonlinear regression models and ANNs for predicting As concentrations in Southeast Asia. Interestingly, Cho et al. (2011) found that dividing the range of the redox potential (Eh) into two groups and separately estimating the relationship of As with Eh P 0 and Eh < 0 substantially improved the model predictability. Data sets spanning a wide range of As concentrations, measured in two Southeast Asian countries in 2008, 2010, and 2012, provided an opportunity to examine the patterns of groundwater As contamination and their variability. In this study, the spatial distributions of multiple As species as well as the relationships between As and water quality parameters were analyzed. We focused on the studies for exploring the inter-relationship As and water quality parameters, rather than considering of geological properties of the watershed. Building on the findings of Cho et al. (2011), we developed a Bayesian change-point (CP) model that captures the change in the relationship between As and Eh over a range of Eh levels. Here, instead of determining the change-point based on a subjective judgment, our model estimates the changepoint that was the most compatible with the data.

2. Materials and methods 2.1. Study area With the objective of exploring As contamination problems in the natural groundwater of the Lao People’s Democratic Republic (Lao PDR) and Cambodia, sampling campaigns were conducted in

2008, 2010, and 2012. Because the source of the high As concentration was thought to be its release from river sediment under reducing conditions (Rawlings et al., 1998; Polizzotto et al., 2008; Sthiannopkao et al., 2008; Kocar et al., 2008), we selected the provinces located along the Mekong River as study sites (Fig. 1). Sites in the Lao PDR included Vientiane, Bolikhamxai, Savannakhet, Saravane, Champasak, and Attapeu. The provinces in Cambodia were Prey Veng and Kandal. All selected tube-wells were cement tube-wells equipped with a covered lid and hand-pumping equipment, all of which were developed and maintained by local communities and households. 2.2. Sample collection As a result of the three sampling campaigns, a total of 112 and 80 groundwater samples were collected from the Lao PDR and Cambodia, respectively. Table S1 summarizes the samples and parameters studied in each sampling campaign. The common water quality parameters were temperature, pH, electrical conductivity, Eh, total dissolved solids (TDS), and AsTOT (total arsenic). However, it should be noted that the full spectrum of As species (As(III), As(V), and particulate As) and well depth information was only collected in 2012. For groundwater collection, standing well water was first pumped out for about 10 min, and groundwater samples were then collected for AsTOT, soluble As, and As(III) concentration analyses. All samples were collected in clean polypropylene bottles previously soaked with concentrated nitric acid (HNO3) and washed with groundwater water drawn from the sampling sites. For AsTOT, groundwater samples were directly collected with no filtration or treatment. On the other hand, filtered water samples (using a 0.45 lm pore-sized membrane filter) were collected and analyzed for dissolved As (As(III) and As(V) concentrations). Moreover, As(III) were collected using a combination of a 0.45 lm pore-sized membrane filter and As speciation cartridge. After collection, all samples were preserved using concentrated HNO3, kept at 4 °C, and delivered to the laboratory at Gwangju Institute of Science and Technology (GIST), Korea for analysis. At the time samples were collected As concentrations and its speciation, water temperature, pH, and Eh were measured by a Horiba D-54 meter (Horiba, Kyoto, Japan). Electrical conductivity and TDS were determined by an Orion 2-Star meter (Thermo Scientific, Waltham, MA) and a Consort C533 multi-parameter analyzer (Montreal Biotech Inc., Canada), respectively. 2.3. Sample analysis Without pretreatment, all samples for AsTOT, dissolved As, and As(III) concentrations in the groundwater were determined using inductively coupled plasma mass spectrometry (ICP-MS; Agilent 7500ce, Agilent Technologies, Japan) equipped with an autosampler. The detection limits of As was 0.05 lg L1. For the solvent to be used in solution preparation, and analytical procedures, 2% (percent by volume) HNO3 was prepared from 18.2 MX cm1 deionized water obtained from a Millipore Milli-Q water purification system (Millipore Corp., USA). Working standard solutions were prepared in an ICP-MS range of 0 lg L1, 0.1 lg L1, 1 lg L1, 5 lg L1, 10 lg L1, 20 lg L1, 50 lg L1, and 100 lg L1. A correlation coefficient (r) of the linear regression (concentrations of working standard solutions vs. concentrations measured by ICP) of P0.998 was used for the standard calibration curves. Quality control and quality assurance for all instrumental analyses were conducted for each batch of 10 samples using an analytical blank sample (2% HNO3), an external standard (a standard concentration prepared from a stock solution obtained from Agilent, USA), and a standard reference material (SRM 1640: trace metals in natural

Please cite this article in press as: Cha, Y., et al. Bayesian modeling approach for characterizing groundwater arsenic contamination in the Mekong River basin. Chemosphere (2015), http://dx.doi.org/10.1016/j.chemosphere.2015.02.045

3

Y. Cha et al. / Chemosphere xxx (2015) xxx–xxx

Fig. 1. A map showing positions of provinces studied located along the Mekong River in Lao PDR and Cambodia.

water, obtained from the National Institute of Standards and Technology-NIST, USA). The performance of ICP-MS analyses for all target elements was within the range of 97.3–104.6% recovery of the calibration standard and 88.7–113.6% recovery of the SRM 1640.

Given the assumption that log-transformed AsTOT concentrations follow a normal distribution, the CP model equation is expressed as follows:

logðAsTOTi Þ  Nðb0 þ b1;j½i  ðEhi  pÞ þ fb2 þ b3;j½i  ðEhi  pÞg  pHi ; r2j½i Þ

2.4. Model development Based on a model that showed noticeable relationships with AsTOT in previous studies and our data set, we developed a Bayesian regression model to predict AsTOT concentrations using Eh and pH (Buschmann et al., 2007; Amini et al., 2008; Chanpiwat et al., 2011; Cho et al., 2011). Also, to address the potential change in the relationship between AsTOT and Eh levels, the model incorporated the change-point at which the change of the AsTOT relationship with Eh occurs. When this change-point is recognized, the CP model estimates the location of the changepoint as a probability distribution based on the data (Qian, 2014).

 for

j ¼ 1 if j ¼ 2 if

Ehi < p Ehi P p

ð1Þ

where the subscript i is the index of individual samples (i = 1, . . ., 512) and p represents the change-point. In addition, b0, b1, b2, and b3 indicate the intercept, slope of Eh, slope of pH, and slope of EhpH, respectively, while r denotes the standard deviation. For a comparison with the CP model, we further developed a multiple linear regression model (LM), which also uses Eh and pH as predictors but does not incorporate the change-point in the structure. Note that if no change-point is observed in the data,

Please cite this article in press as: Cha, Y., et al. Bayesian modeling approach for characterizing groundwater arsenic contamination in the Mekong River basin. Chemosphere (2015), http://dx.doi.org/10.1016/j.chemosphere.2015.02.045

4

Y. Cha et al. / Chemosphere xxx (2015) xxx–xxx

the CP model reduces to the LM model; the model is then expressed as follows:

logðAsTOTi Þ  Nðb0 þ b1  Ehi þ fb2 þ b3  Ehi g  pHi ; r2 Þ:

ð2Þ

Diffuse prior distributions were used for all model parameters. A weakly constrained prior distribution was defined at the changepoint (p), which was uniformly distributed over the range of observed Eh levels. Posterior distributions were obtained using Markov chain Monte Carlo (MCMC) simulation procedures in WinBUGS software (Lunn et al., 2012). Model convergence was decided when the potential scale reduction parameter for all parameters equaled to 1 (Gelman and Hill, 2007). A comparison between the CP and LM models was made based on several summary measures of model fit. The coefficient of determination (R2) was defined as 1-(SSRES/SSTOT), where SSRES and SSTOT represent the residual sum of squares and total sum of squares, respectively. R2 was used to estimate the model goodness of fit. Deviance was used to measure the prediction error; the lower the deviance, the better a model fits the data. The deviance information criterion (DIC), the compromise between model fit and complexity, provides a basis for Bayesian model selection. A model with the lowest DIC is preferred to models having a higher DIC. As a measure of both model fit and complexity, DIC can be expressed as follow, (Spiegelhalter et al., 2002)

DIC ¼ lD þ 2pD

ð3Þ

where lD is mean deviance over a number of simulations and pD is the effective number of parameters. The DIC provides a basis for Bayesian model selection, with the lowest score indicating a preferred model. In addition, the fit of the CP and LM models was evaluated by comparing the data replicated (yrep) from the posterior predictive distribution obtained using the observed data (y) (Gelman et al.,

2013). To perform the posterior predictive checking, a set of model parameters, h, was drawn from their joint posterior distribution at the lth MCMC simulation (l = 1, . . ., 1000), and this process was iterated for a number of simulations. From a log-normal distribution for each set of parameters, 512 samples representing yrep were drawn. We defined a test quantity, T(y, h), the aspect of data being examined, as the median of the data. Of the 1000 simulations, the Bayesian p-value at the lth draw was defined as the probability of the simulated test quantities, T(yrep l, hl), being equal to or greater than the observed test quantities, T(y, hl). An extreme Bayesian p-value close to either 0 or 1 implies a poor model fit.

3. Results and discussion 3.1. Comparison of As conditions between Cambodia and Lao PDR Despite their geographical proximity, Cambodia and the Lao PDR groundwater exhibited contrasting As conditions (Table S1, Fig. 2). During the study period, the mean AsTOT concentration in Cambodia, 345.6 (±28.9) lg L1, was higher than the mean AsTOT in Lao PDR, 25.2 (±3.6) lg L1 (±denotes one standard error). Consistent with the AsTOT pattern, the Cambodian groundwater contained substantially higher mean concentrations of As(III) (411.4 (±34.3) lg L1) and As(V) (42.0 (±5.8) lg L1 vs. 7.1 (±1.3) lg L1) than in the Laotian groundwater. In addition to As, differences in several physicochemical variables were also quite distinct between the two countries (Table S1). Over the study years, a comparison of the mean Eh levels (163 (±14.9) mV vs. 40.8 (±3.7) mV) and mean pH (7.0 (±0.0) vs. 6.6 (±0.1)) in Cambodia and the Lao PDR indicated that the Cambodian groundwater had a more reducing, less acidic environment. The mean depth of groundwater wells in Cambodia (35.3 (±0.8) m), was also significantly deeper than the mean depth in the Lao PDR (17.0 (±1.1) m).

Fig. 2. Relationships among As species. (a) log(AsTOT) (ppb) and log(AsIII) (ppb) conentrations, (b) log(AsTOT) and log(AsV) (ppb), (c) log(AsV) and log(AsIII), (d) log(AsTOT) and As (III) proportion, (e) log(AsTOT) and As (V) proportion, and (f) As (V) proportion and As (III) proportion. Solid red and empty blue circles represent As sampled in Cambodia and Lao PDR, respectively. The vertical dashed line indicates overall mean As (III) (panel a) and As (V) (panel b) concentration in the logarithmic scale. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

Please cite this article in press as: Cha, Y., et al. Bayesian modeling approach for characterizing groundwater arsenic contamination in the Mekong River basin. Chemosphere (2015), http://dx.doi.org/10.1016/j.chemosphere.2015.02.045

5

Y. Cha et al. / Chemosphere xxx (2015) xxx–xxx

3.2. Relationships among As species Positive correlations were found between AsTOT and As(III) (q = 0.99), and AsTOT and As(V) (q = 0.73) (Fig. 2a and b). Since As(III) generally constituted a higher proportion (mean proportion = 0.57) of AsTOT than As(V) (mean proportion = 0.29), As(III) was deemed more critical for determining As levels. As(III) and As(V) concentrations were also positively correlated (q = 0.69) (Fig. 2c). Interestingly, when As(V) as the proportion of AsTOT increased, there was a corresponding decrease in the AsTOT concentration (Fig. 2d), because the increase in As(V) proportion was coincident with a decrease in the proportion of As(III) (Fig. 2f), which existed at higher concentrations (overall mean of 193.6 lg L1 in natural metrics) relative to As(V) (mean = 22.8 lg L1) (Fig. 2a–c). In contrast, the As(III) proportion was positively related to the AsTOT concentrations (Fig. 2e). Organic As is known to be less toxic than inorganic As. Mainly ingested from seafood, organic As is easily excreted in the urine, and does not accumulate in the body (Navas-Acien et al., 2011). Inorganic As generally exists in the environment as As(III) or As(V), with As(III) being considered more toxic than As(V). When absorbed, As(III) is stored in the liver, kidney, heart, and lungs, with a lower amount of As(III) observed in the muscles and neuronal tissues. The accumulated As(III) in the body can cause a variety of disorders, including cancer, diabetes, hepatotoxicity, neurotoxicity, and cardiac dysfunction. Compared to the Lao PDR, the groundwater in Cambodia contained a substantially higher As(III) concentration, not only in terms of absolute content but also as a proportion of AsTOT (Fig. 2a and d), thus posing a greater threat to public health in the area (Adriano, 2001; Chanpiwat et al., 2011; Kim et al., 2011). 3.3. Bayesian change-point model The observed log(AsTOT) displayed a slightly positive correlation with low Eh, up to a specific level, but then sharply decreased with a further increase in Eh (Fig. 3a). The AsTOT concentrations exhibited moderate, positive relationships with pH (q = 0.43) and well depth (q = 0.49) across the two countries (Fig. 3b and c). The results from the CP model demonstrated that log(AsTOT) increased with increasing Eh, and showed substantial decreases with increasing Eh at Eh levels greater than the change-point (100 ± 15 mV) (±denotes one standard deviation) (Table 1, Fig. 4a). The increase in pH led to an increase in log(AsTOT), both below and above the change-point (Table 1, Fig. 4b). The noticeable correlation of AsTOT at an Eh higher than the change-point (100 ± 15 mV) indicates a reductive dissolution of mineral oxides and As release in the upstream of the watershed (Berg et al., 2007;

Table 1 Parameter estimates of the CP and the LM summarized by mean and one standard deviation (in parenthesis). Parameter

CP

Parameter

LM

b0 b1,1 b1,2 b2 b3,1 b3,2

2.971 (2.049) 0.011 (0.013) 0.034 (0.011) 1.258 (0.283) 0.005 (0.001) 0.006 (0.002) 2.031 (0.113) 0.634 (0.053) 99.97 (15.10)

b0 b1

8.290 (1.328) 0.032 (0.009)

b2 b3

1.663 (0.192) 0.006 (0.001)

r

1.831 (0.082)

r1 r2 p

Buschmann et al., 2007). These conditions are reflected in most observations in the Lao PDR and a portion of the Cambodian observations (Fig. 3a). In contrast, an Eh lower than the change-point showed a weak positive correlation with AsTOT concentrations, most of which were collected from Kandal Province in Cambodia (downstream). These findings are consistent with previous reports, in which As was found to be less responsive to variations in Eh in young alluvial soil, such as that in Kandal Province (Berg et al., 2007). The well depths of Cambodia were deeper than those of the Lao PDR, with little overlap; the depths of groundwater wells ranged from 25 m to 56 m in Cambodia and from 1 m to 32 m in the Lao PDR (Fig. 3c). Well depth is associated with the depth of aquifers. The depth of an aquifer tends to be deeper in the downstream areas of a watershed, whereas the upstream in a watershed is greatly affected by groundwater discharge from shallow aquifers. This substantial difference in the well depths may thus result in sharp contrasts in Eh, pH, and As concentrations between Cambodia and the Lao PDR. However, well depth could not be used as a predictor in our models because the well depth information was available only for the 2012 samples. Nevertheless, given the causal relationship and observed high correlation among well depth and Eh and pH, it would be reasonable to use Eh and pH as surrogates for well depth. Compared to the prediction uncertainty above the changepoint, uncertainty below the change-point was lower mainly due to the smaller model error variance in that range (Table 1, Fig. 4a). The LM model prediction suggested that log(AsTOT) linearly decreased as Eh levels increased (Table 1, Fig. 4c); in contrast, log(AsTOT) concentrations were positively related to pH (Table 1, Fig. 4d). All summary measures of model fit were in favor of using the CP model over the LM model. The CP model provided a better fit, with

Fig. 3. Relationships among AsTOT and related variables. (a) log(AsTOT) (ppb) and Eh (mV), (b) log(AsTOT) and pH, and (c) log(AsTOT) and well depth. Solid red and empty blue circles represent As sampled in Cambodia and Lao PDR, respectively. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

Please cite this article in press as: Cha, Y., et al. Bayesian modeling approach for characterizing groundwater arsenic contamination in the Mekong River basin. Chemosphere (2015), http://dx.doi.org/10.1016/j.chemosphere.2015.02.045

6

Y. Cha et al. / Chemosphere xxx (2015) xxx–xxx

Fig. 4. Predicted log(AsTOT) from (a) Eh (with pH fixed at its mean), and (b) pH using the CP, and log(AsTOT) predicted from (c) Eh (with pH fixed at its mean), and (d) pH (with Eh fixed at its mean) using the LM. In panels (a) and (b), the solid violet line and the solid green line denote predicted mean log(AsTOT) at Eh < estimated p and Eh P p, respectively. Vertical solid gray line indicates the predicted mean p. In panel (b) to estimate log (AsTOT) at Eh < p Eh was fixed at the mean of Eh < p, while Eh was fixed at the mean of Eh P p to estimate log(AsTOT) at Eh P p. In panels (c) and (d), the solid black line represents predicted mean log(AsTOT). All dashed lines denote 95% intervals. Solid violet circles and empty green circles indicate observations at Eh < change-point and Eh P p, respectively. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

Table 2 Summary measures of model fit for the CP and the LM. Summary

CP

LM

R2 Deviance DIC Bayesian p-value

0.48 884.8 892.6 0.38

0.40 1014.4 1019.3 0.01

a higher R2 and lower deviance (Table 2). Indeed, the lower DIC for the CP model suggested that the improved fit more than offset the effect of added complexity of using a greater number of parameters (Table 2). Also, posterior predictive checks on the CP and LM models were performed on the test quantity, the median AsTOT concentration (Fig. 5). The comparison between the observed test quantity and the simulated test quantity revealed a substantial discrepancy between the observed median AsTOT and predictive median AsTOT

Fig. 5. Posterior predictive checks for the median of AsTOT concentrations. Vertical bars indicate the observed median of AsTOT and histograms represent the median of 1000 draws from the (a) CP and (b) LM.

Please cite this article in press as: Cha, Y., et al. Bayesian modeling approach for characterizing groundwater arsenic contamination in the Mekong River basin. Chemosphere (2015), http://dx.doi.org/10.1016/j.chemosphere.2015.02.045

Y. Cha et al. / Chemosphere xxx (2015) xxx–xxx

from the LM model (Table 2, Fig. 5b), while the observed median was consistent with the medians expected under the CP model (Table 2, Fig. 5(a)). 4. Conclusions The groundwater data we collected in Cambodia and the Lao PDR spanned a wide gradient of As, Eh, pH, and well depth conditions, covering mountainous upstream and downstream delta areas along the Mekong River basin. The collected data were then used to characterize the site-specific relationships among As species, Eh, pH, and well depth. The relationships among these variables involved sudden, dramatic changes, indicating the existence of a change-point. Thus, we developed a CP model for predicting AsTOT contamination using the Eh and pH, in order to determine the AsTOT–Eh relationship change occurring at an Eh threshold value. The CP model successfully identified the change-point of Eh (100 ± 15 mV), which was supported by observations. Also, a relaxation of the fixed relationship between AsTOT and Eh improved the model fit and prediction accuracy. It is expected that the CP model would produce reliable predictions regardless of Eh levels, thereby providing a helpful tool to immediately detect As contamination in the groundwater in Southeast Asian countries. However, it should be noted that the groundwater in Cambodia exhibits extremely low Eh levels, which are rarely observed in other regions. Therefore, future monitoring efforts and investigations are required in order to confirm whether the existence and range of change-points in the AsTOT–Eh relationship is specific to Cambodia or whether it is common to other As polluted regions having low Eh levels. Moreover, enhanced monitoring of As(III), the form more negatively affecting human health than AsTOT or As(V), will enable researchers to further optimize the CP model for predicting As(III) concentrations. Acknowledgements This study was funded by the Toyota Foundation (Grant number D11-R-81) and was also supported by a grant from the Korea Institute of Science and Technology (KIST) Institutional Program (Project No. 2E24560). Appendix A. Supplementary material Supplementary data associated with this article can be found, in the online version, at http://dx.doi.org/10.1016/j.chemosphere. 2015.02.045. References Adriano, D.C., 2001. Trace Elements in Terrestrial Environments Biogeochemistry, Bioavailability, and Risks of Metals, second ed. Springer-Verlag Inc., New Tork, NY, USA. Allan, H.S., Peggy, A.L., Michael, N.B., Craig, M.S., 2002. Arsenic epidemiology and drinking water standards. Science 296, 2145–2146. American Water Works Association (AWWA), Arsenic Rule, Mainstream, 45, 2001. Amini, M., Mueller, K., Abbaspour, K.C., Rosenberg, T., Afyuni, M., Møller, K.N., Sarr, M., Johnson, C.A., 2008. Statistical modeling of global geogenic fluoride contamination in groundwaters. Environ. Sci. Technol. 42 (10), 3662–3668. Bagla, P., Kaise, J., 1996. India’s spreading health crisis draws global arsenic experts. Science 274 (5285), 174–175. Berg, M., Luzi, S., Kim, P.T., Viet, P.H., Giger, W., Stuben, D., 2006. Arsenic removal from groundwater by household sand filters: comparative field study, model calculations, and health benefits. Environ. Sci. Technol. 40 (17), 5567–5573. Berg, M., Tran, H.C., Nguyen, T.C., Pham, H.V., Schertenleib, R., Giger, W., 2001. Arsenic contamination of groundwater and drinking water in Vietnam: a human health threat. Environ. Sci. Technol. 35 (13), 2621–2626. Berg, M., Tran, H.C., Nguyen, T.C., Pham, H.V., Schertenleib, R., Giger, W., 2007. Magnitude of arsenic pollution in the Mekong and Red River Deltas—Cambodia and Vietnam. Sci. Total. Environ. 372 (2–3), 413–425.

7

Buschmann, J., Berg, M., Stengel, C., Sampson, M.L., 2007. Arsenic and manganese contamination of drinking water resources in Cambodia: coincidence of risk areas with low relief topography. Environ. Sci. Technol. 41 (7), 2146–2152. Chang, F.J., Kao, L.S., Kuo, Y.M., Liu, C.W., 2010. Artificial neural networks for estimating regional arsenic concentrations in a blackfoot disease area in Taiwan. J. Hydrol. 388 (1–2), 65–76. Chanpiwat, P., Sthiannopkao, S., Cho, K.H., Kim, K.W., San, V., Suvanthong, B., Vongthavady, C., 2011. Contamination by arsenic and other trace elements of tube-well water along the Mekong River in Lao RDR. Environ. Pollut. 159 (2), 567–576. Chauhan, V.S., Nickson, R.T., Chauhan, D., Iyengar, L., Sankararamakrishnan, N., 2008. Ground water geochemistry of Ballia district, Uttar Pradesh, India and mechanism of arsenic release. Chemosphere 75 (1), 83–91. Chiew, H., Sampson, M.L., Huch, S., Ken, S., Bostick, B.C., 2009. Effect of groundwater iron and phosphate on the efficacy of arsenic removal by iron-amended biosand filters. Environ. Sci. Technol. 43 (16), 6295–6300. Cho, K.H., Sthiannopkao, S., Pachepsky, Y.A., Kim, K.W., Kim, J.H., 2011. Prediction of contamination potential of groundwater arsenic in Cambodia, Laos, and Thailand using artificial neural network. Water Res. 45 (17), 5535–5544. Gelman, A., Carlin, J.B., Stern, H.S., Dunson, D.B., Vehtari, A., Rubin, D.B., 2013. Bayesian Data Analysis. CRC Press. Gelman, A., Hill, J., 2007. Data Analysis Using Regression and Multilevel/ Hierarchical Models. Cambridge University Press. Kim, K.W., Chanpiwat, P., Hanh, H.T., Phan, K., Sthiannopkao, S., 2011. Arsenic geochemistry of groundwater in Southeast Asia. Front Med. 5 (4), 420–433. Kocar, B.D., Fendorf, C., 2009. Thermodynamic constraints on reductive reactions influencing the biogeochemical of Arsenic in soils and sediments. Environ. Sci. Technol. 43 (13), 4871–4877. Kohnhorst, A., 2005. Arsenic in groundwater in selected countries in South and Southeast Asia: a review. J. Tropical Med. Parasitol. 28 (2), 73–82. Lunn, D., Jackson, C., Spiegelhalter, D.J., Best, N., Thomas, A., 2012. The BUGS Book: A Practical Introduction to Bayesian Analysis. CRC Press. Mondal, P., Bhowmick, S., Chatterjee, D., Figoli, A., Van der Bruggen, B., 2013. Remediation of inorganic arsenic in groundwater for safe water supply: a critical assessment of technological solutions. Chemosphere 92 (2), 157–170. Navas-Acien, A., Francesconi, K.A., Silbergeld, E.K., Guallar, E., 2011. Seafood intake and urine concentrations of total arsenic, dimethylarsinate and arsenobetaine in the US population. Environ. Res. 111 (1), 110–118. Polizzotto, M.L., Kocar, B.D., Benner, S.G., Sampson, M., Fendorf, S., 2008. Nearsurface wetland sediments as a source of arsenic release to ground water in Asia. Nature 454 (7203), 505–508. Polya, D.A., Gault, A.G., Bourne, N.J., Lythgoe, P.R., Cooke, D.A., 2003. Coupled HPLC– ICP–MS analysis indicates highly hazardous concentrations of dissolved arsenic species in Cambodian groundwaters. In: Holland, J.G., Tanner, S.D. (Eds.), Plasma Source Mass Spectrometry: Applications and Emerging Technologies. Royal Society of Chemistry, Cambridge, pp. 27–140. Polya, D.A., Gault, A.G., Diebe, N., Feldman, P., Rosenboom, J.W., Gilligan, E., Fredericks, D., Milton, A.H., Sampson, M., Rowland, H.A.L., Lythgoe, P.R., Middleton, C., Cooke, D.A., 2005. Arsenic hazard in shallow Cambodian groundwaters. Mineral Mag. 69 (5), 807–823. Purkait, B., Kadam, S.S., Das, S.K., 2008. Application of artificial neural network model to study arsenic contamination in groundwater of Malda district, Eastern India. J. Environ. Inform. 12 (2), 140–149. Rahman, M., Mamun, A.A., Karim, M.R., Islam, K., Amin, H.A., Hossain, S., Hossain, M.I., Saud, Z.A., Noman, A.S., Miyataka, H., Himeno, S., Hossain, K., 2015. Associations of total arsenic in drinking water, hair and nails with serum vascular endothelial growth factor in arsenic-endemic individuals in Bangladesh. Chemosphere 120, 336–342. Rawlings, J.O., Pantula, S.G., Dickey, D.A., 1998. Applied Regression Analysis: A Research Tool. Springer, New York. Schmoll, O., Howard, G., Chilton, J., Chorus, I., 2006. Protecting Groundwater For Health: Managing the Quality of Drinking-water Sources. World Health Organization. IWA Publishing, London, UK. Smedley, P.L., Kinniburgh, D.G., 2002. A review of the source, behaviour and distribution of arsenic in natural waters. Appl. Geochem. 17 (5), 517–568. Spiegelhalter, D.J., Best, N.G., Carlin, B.P., van der Linde, A., 2002. Bayesian measures of model complexity and fit (with discussion). J. R. Stat. Soc. B 64, 583–639. Stanger, G., Truong, T.V., My Ngoc, K.S.L.T., Luyen, T.V., Tuyen, T.T., 2005. Arsenic in groundwaters of the Lower Mekong. Environ. Geochem. Health 27 (4), 341–357. Sthiannopkao, S., Kim, K.W., Cho, K.H., Wantala, K., Sotham, S., Sokuntheara, C., Kim, J.H., 2010. Arsenic levels in human hair, Kandal Province, Cambodia: the influences of groundwater arsenic, consumption period, age and gender. Appl. Geochem. 25 (1), 81–90. Sthiannopkao, S., Kim, K.W., Sotham, S., Chop, S., 2008. Arsenic and manganese in tube well waters of Prey Veng and Kandal Provinces, Cambodia. Appl. Geochem. 23 (5), 1086–1093. Sun, G., Liu, J., Luong, T.V., Sun, D., Wang, L., 2002. Endemic arsenicosis: a clinical diagnostic with photo illustrations. UNICEF East Asia and Pacific Regional Office, Bangkok, Thailand (ISBN: 974-685-040-7). Tetsuro, A., Takashi, K., Junko, F., Reiji, K., Tu, B.M., Pham, T.K.T., Hisato, I., Annamalai, S., Pham, H.V., Shinsuke, T., 2006. Contamination by arsenic and other trace elements in tube-well water and its risk assessment to humans in Hanoi, Vietnam. Environ. Pollut. 139 (1), 95–106. Qian, S.S., 2014. Ecological threshold and environmental management: a note on statistical methods for detecting thresholds. Ecol. Indic. 38, 192–197.

Please cite this article in press as: Cha, Y., et al. Bayesian modeling approach for characterizing groundwater arsenic contamination in the Mekong River basin. Chemosphere (2015), http://dx.doi.org/10.1016/j.chemosphere.2015.02.045