Predicting Regional-Scale Elevated Groundwater Nitrate Contamination Risk Using Machine Learning on Natural and Human-Induced Factors
Soumyajit Sarkar, Abhijit Mukherjee, Srimanti Dutta Gupta, Soumendra N. Bhanja, Animesh Bhattacharya
Abstract
Elevated groundwater nitrate poses risk to the ecosystem and human health, and delineating the extent of elevated groundwater nitrate risk is essential for effective groundwater management and public health safety. Here, using machine learning models (Random Forest, Boosted Regression Tree, and Logistic Regression) on a large, in situ dataset, we have predicted the first nationwide extent of groundwater nitrate contamination risk (concentration >45 mg/L) across India. We also aimed to delineate the intrinsic (e.g., climate, geomorphic, hydrogeologic) and extraneous (e.g., anthropogenic input) predictors for identifying groundwater pollution risk. Of these models, Random Forest performed best and was considered to develop the final prediction map of groundwater nitrate at 1 km2 resolution. Climate variables like precipitation and aridity, and anthropogenic influence, e.g., fertilizer application and population density, were identified as the most important variables for predicting groundwater nitrate risk. Dry arid and semiarid regions in the west, south, and central parts of the country contained the majority of high-risk areas. Predictions suggested that about 37% of India’s areal extent and 380 million people were exposed to elevated nitrate. The prediction model performed satisfactorily over the validation dataset that indicates the prediction ability of the model at the local scale. The study aims to provide an effective approach for identifying elevated groundwater nitrate risk and aid in the development of awareness and strategies to uphold public health safety.