Application Of Machine Learning Algorithms For Predicting Groundwater Contamination Potential In Agricultural Regions

Abstract

The use of fertilizers, pesticides and the mishandling of wastes causes groundwater contamination which is a growing environmental and human health issue in agriculture areas. Although the traditionally used monitoring methods are quite effective, they usually have temporal and spatial limitations, which means that early detection and wide-ranging predictions are also difficult to achieve. The present study examines the prospects of machine learning (ML) models in forecasting the possible risk of groundwater contamination over agricultural land-scapes through a combination of hydrochemical measures and land-use patterns in combination with forecasted climatic features. Three agriculturally intensive districts namely Ludhiana (Punjab), Bhopal (Madhya Pradesh) and Thrissur (Kerala) were chosen and the samples of water were analysed on the basis of key physicochemical parameters such as nitrate (NO 3 - ), phosphate (PO 4 3 - ), pH, total dissolved solids (TDS) and heavy metals. Remotely sensed land cover indices coupled with soil type data (and these parameters) were utilized as input features in training the models. The implementation of the Random Forest (RF) algorithm, Support Vector Machine (SVM) algorithm and Artificial Neural Network (ANN) algorithm was done and the performance of the model tested by the 10-fold cross-validation method. Random Forest model resulted in the highest classification accuracy (92%) and ROC-AUC (0.94) values allowing accurate differentiation of high, moderate, and low risk zones in terms of the contamination potential. It was found that areas of high fertilizer capacity and shallow water tables were found to be areas of high contamination through the formation of spatial prediction maps using the GIS. We show that when coupled with geospatial analysis, ML-based methods serve as a scalable and cost-effective solution to formulating problems in the assessment of groundwater contamination risks in agricultural areas, allowing data-driven policymaking and to the creation of specialized mitigation measures to address these risks.