Assessment of the application of machine learning in groundwater prospecting for siting of boreholes

Abstract

This report presents an evaluation of the role of machine learning in predictive studies focusing on groundwater. Machine learning algorithms such and Artificial Neural Networks (ANN) and Extreme Gradient Boost(XGB) to offer solutions in groundwater prospecting through data-driven modeling, requiring less labor, cost, and time. The main objective of this study was to evaluate the applicability of two machine learning algorithms, ANN and XGB in the prediction of static water level and total yield for new borehole sites within the Awoja catchment. The study utilized borehole data for the existing boreholes, including all the borehole information together with groundwater explanatory variables, soil, lithology, landuse and precipitation, and remote sensed data to train a multivariate ANN and XGB models for predicting Static Water Level and total yield. Both models were evaluated using the coefficient of correlation (Ri2), MAE, and MSE and the models performed well with the XGB model better than the ANN for the prediction of both the static water level and total yield having scores of 0.901, and 0.019 for R2 and MAE. It has been revealed that ML techniques cannot replace the existing methods of prospecting but are very important when used alongside each other. The results of this study aim to inform the practicability of data driven modeling in groundwater prospecting for the siting of boreholes as a way of improving the efficiency of the process.