Determining the probability of arsenic in groundwater using a parsimonious model

Spatial distributions of groundwater quality are commonly heterogeneous, varying with depths and locations, which is important in assessing the health and ecological risks. Owing to time and cost constraints, it is not practical or economical to measure arsenic everywhere. A predictive model is necessary to estimate the distribution of a specific pollutant in groundwater. This study developed a logistic regression (LR) model to predict the residential well water quality in the Lanyang plain. Six hydrochemical parameters, pH, NO3- -N, NO2- -N, NH+ -N, Fe, and Mn, and a regional variable (binary type) were used to evaluate the probability of arsenic concentrations exceeding 10 microg/L in groundwater. The developed parsimonious LR model indicates that four parameters in the Lanyang plain aquifer, (pH, NH4+, Fe(aq), and a component to account for regional heterogeneity) can accurately predict probability of arsenic concentration > or =1 microg/Lin groundwater. These parameters provide an explanation for release of arsenic by reductive dissolution of As-rich FeOOH in NH4+ containing groundwater. A comparison of LR and indicator kriging (IK) show similar results in modeling the distributions of arsenic. LR can be applied to assess the probability of groundwater arsenic at sampled sites without arsenic concentration data apriori. However, arsenic sampling is still needed and required in arsenic-assessment stages in other areas, and the need for long-term monitoring and maintenance is not precluded.