Prediction of groundwater level changes based on machine learning technique in highly groundwater irrigated alluvial aquifers of south-central Punjab, India

Abstract

Groundwater serves as a vital resource for all living organisms. In regions extensively reliant on groundwater irrigation, hydro-climatic factors, groundwater extraction, and the flow of surface water exhibit an indirect interdependence. This study primarily aims to anticipate GWL in such highly irrigated zones using the Machine Learning (ML) approach. To achieve this, the widely employed Random Forest (RF), Bagging-Reduce Error Pruning Tree (Bagging-REPTree), and Bagging-Decision Stump Tree (Bagging-DSTree) models have been employed for the accurate forecasting of groundwater levels. The long-term pre-monsoon and post-monsoon (fourteen locations) data set of South-Central Punjab state has been applied for the model calibration/training and validation/testing. Seven statistical indices were used such as percent bias (PBIAS), root mean square error (RMSE), normalized root mean square error (nRMSE), RMSE-observation standard deviation ratio (RSR), mean absolute error (MAE), Nash Sutcliffe efficiency (NSE) and correlation coefficient (CC) for the model performance analysis. The results revealed that the RF model outperformed in pre-monsoon (testing phase) (RMSE = 0.682, NSE = 0.958) as well as the post-monsoon (testing phase) (RMSE = 0.150, NSE = 0.997) compared to the other two models in the station Ahmadapur and the similar trend is observed in all the stations. Overall, the RF model demonstrates superior performance in predicting groundwater levels during both pre-monsoon and post-monsoon seasons, particularly in highly groundwater-irrigated alluvial aquifers of the southern region.