A generalized regression model of arsenic variations in the shallow groundwater of Bangladesh.

Mohammad Shamsudduha,Richard E Chandler,Richard G Taylor

doi:10.1002/2013wr014572

Mohammad Shamsudduha, Richard E Chandler + Show 1 more

Open Access

https://doi.org/10.1002/2013wr014572

Copy DOI

Abstract

Localized studies of arsenic (As) in Bangladesh have reached disparate conclusions regarding the impact of irrigation‐induced recharge on As concentrations in shallow (≤50 m below ground level) groundwater. We construct generalized regression models (GRMs) to describe observed spatial variations in As concentrations in shallow groundwater both (i) nationally, and (ii) regionally within Holocene deposits where As concentrations in groundwater are generally high (>10 μg L−1). At these scales, the GRMs reveal statistically significant inverse associations between observed As concentrations and two covariates: (1) hydraulic conductivity of the shallow aquifer and (2) net increase in mean recharge between predeveloped and developed groundwater‐fed irrigation periods. Further, the GRMs show that the spatial variation of groundwater As concentrations is well explained by not only surface geology but also statistical interactions (i.e., combined effects) between surface geology and mean groundwater recharge, thickness of surficial silt and clay, and well depth. Net increases in recharge result from intensive groundwater abstraction for irrigation, which induces additional recharge where it is enabled by a permeable surface geology. Collectively, these statistical associations indicate that irrigation‐induced recharge serves to flush mobile As from shallow groundwater.

Highlights

Biogeochemical controls on aqueous arsenic (As) concentrations in very shallow ( 50 m below ground level) groundwater in the Bengal Basin have been studied extensively over the last two decades [Nickson et al, 1998; McArthur et al, 2004; Zheng et al, 2005; Harvey et al, 2006; Mukherjee et al, 2008; Chowdhury et al, 2012]
Complete results from the national-scale, final generalized regression models (GRMs) are provided in supporting information. bModel coefficients, standard errors, and P values for categorical surface geology covariate listed in the supporting information. cModel coefficients are associated with the respective covariate only and not including their statistical interactions. dNet changes in mean recharge between predeveloped groundwater-fed irrigation (PGI) (197521980) and developed groundwater-fed irrigation (DGI) (199521999) periods
We demonstrate the application of generalized regression models (GRMs) to explain the spatial variation in groundwater As data set in Bangladesh that features (1) a highly skewed distribution, (2) a substantial number censored or nondetect observations, and (3) correlations between sites from neighboring spatial locations

Summary

Introduction

Biogeochemical controls on aqueous arsenic (As) concentrations in very shallow ( 50 m below ground level) groundwater in the Bengal Basin have been studied extensively over the last two decades [Nickson et al, 1998; McArthur et al, 2004; Zheng et al, 2005; Harvey et al, 2006; Mukherjee et al, 2008; Chowdhury et al, 2012]. Using observational data and groundwater flow modeling in a localized study area, Neumann et al [2010, 2011] show that groundwater recharge from anthropogenic ponds, rich in biologically available OC, produces groundwater elevated in arsenic, whereas recharge derived from rice-field irrigation return flows gives rise to groundwater low in arsenic The latter mechanism is consistent with the assertion that recharge induced by groundwater pumping serves primarily to flush mobile As from alluvial aquifers in the Bengal Basin [van Geen et al, 2003; Stute et al, 2007; Datta et al, 2011; McArthur et al, 2011a; Reich, 2011]

Objectives

Methods

Results

Discussion

Conclusion