Mapping groundwater vulnerability to nitrate contamination using machine learning techniques

Abstract

The evaluation of groundwater vulnerability to contamination in the eastern Mitidja aquifer has become very important for water resources control and preservation. This study aims to model the spatial groundwater vulnerability to nitrate based on the maximum acceptable concentration in drinking water (50 mg/L) by using 10 influencing parameters, which are rainfall, vadose zone, depth to groundwater, slope, permeability, distance to river, drainage density, land use, NDVI and TWI. The dataset was randomly divided between training (70%) and validation (30%). We compared between the results of Random Forest and AdaBoost machine learning models, based on the Receiver Operating Characteristic (ROC) curve, Area Under Curve (AUC) equals 86% and 94%, respectively. In addition, both ML models revealed that rainfall, permeability, and depth to groundwater are the key factors determining groundwater vulnerability to nitrate (NO3) in the eastern Mitidja and it also predicted indexes for each parameter based on their importance. As a result, the groundwater vulnerability map was elaborated.