Predicting As contamination risk in Red River Delta, Vietnam using machine learning methods

Abstract

Naturally-occurring As in groundwater is a worldwide health problem. In Red River Delta in Vietnam, several million people are at risk of chronic As poisoning. The main mechanism for the release of As into groundwater is through the microbially- driven reductive dissolution of Fe (III) oxides. Since majority of the drinking water source in Red River is groundwater from private tube wells, there is a need for a predictive model to assess the As risk of the studied wells. The model may be used to initiate early mitigation measures to prevent the consumption of groundwater sourced from wells with high As concentration. Four machine learning models were used to predict the As probability of study sites in Red River Delta, Vietnam. The GBM was the best performing model with accuracy, precision, sensitivity, and specificity of 98.7%, 100%, 95.2%, and 100%, respectively. It also had the highest AUC of 92% and 96% for the PRC and ROC curves, respectively. The most important variables were Eh and Fe. The partial dependence plot of As concentration on the model parameters showed that the As probability is high at shallower well depths, negative Eh, low SO4 concentrations, high DOC, high PO4 concentrations, and high NH4 concentrations, conditions that trigger the reductive dissolution of iron phases, releasing As into the groundwater.