Potential of machine learning algorithms in groundwater level prediction using temporal gravity data

Abstract

Prediction of groundwater levels with precision and dependability is crucial for effective water resource development and management. This study was carried out to establish the relationship between groundwater level (GWL) and temporal gravity variation through five nonlinear machine learning (ML) models: Polynomial Regression (PR), Random Forest, XG- Boost, K-Nearest Neighbourhood (KNN), and support vector machine - radial basic function (SVM-RBF). These models were employed to predict GWL at a specific well located at the Hydrology Department, IIT Roorkee, India. The models were trained and tested using a dataset that includes gravity, time, and relevant hydro-meteorological factors like precipitation (P), temperature (T), evaporation (E), relative humidity (RH), and wind speed (WS). The study is organized into three groups of model runs: the first group considers gravity as the sole input parameter, the second group includes both gravity and time, and the third incorporates all hydro-meteorological parameters. Comparative evaluation of the models was done using four different evaluation metrics, i.e., coefficient of determination (R2), Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and Nash-Sutcliffe Efficiency (NSE). Results highlight XG-Boost as the most efficient model for predicting groundwater levels, demonstrating exceptional performance, particularly when gravity and time are the input parameters, yielding a minimum MAE of 0.11 and a maximum R2 of 0.97.