Exploring AI approaches for predicting groundwater levels in coastal agro-climatic zones: a case study in Cuttack District, Odisha

Abstract Groundwater level (GWL) prediction across various time scales is essential for efficient management and governance of water resources especially in regions characterized by arid and semi-arid climates, and it holds great significance. Within certain coastal regions, agro-climatic zones give rise to challenges like water scarcity in summer and waterlogging during the rainy season, resulting in reduced GWL during scarcity periods and saltwater intrusion that contaminates groundwater. This study emphasizes on application of diverse AI methodologies, encompassing Artificial Neural Networks (ANN), Adaptive Neuro-Fuzzy Inference Systems (ANFIS), Support Vector Regression (SVR), Long Short-Term Memory (LSTM), and Wavelet Transform-based ANN (W-ANN), ANFIS (W-ANFIS), SVR (W-SVR), and LSTM (W-LSTM) models for quantitative assessment of groundwater in Odisha's Cuttack District, aiming to comprehend GWL fluctuations across the region. The investigation leverages historical groundwater data from monitoring wells, incorporating monthly datasets of rainfall, temperature, relative humidity, and GWLs. Through comparative assessment using statistical methods namely Pearson’s R (R), co-efficient of determination (R2), Root Mean Squared Error (RMSE), and Sum of Squared Error (SSE), the most precise and robust AI approach for groundwater estimation in the area is identified. The W-LSTM (R2-0.78196, RMSE- 0.09254, R-0.88428 and SSE-2.66357) and W-ANFIS (R2-0.74068, RMSE-0.08229, R-0.86063 and SSE-2.10596) hybrid algorithms consistently achieved the most accurate predictions for GWLs followed by W-SVR, W-ANN hybrid models and LSTM and ANN for all stations. Overall, this study demonstrated promising outcomes, offering a dependable foundation for water resources planners to guide future investigations into groundwater resources.