Identify suitable artificial groundwater recharge zones using hybrid deep learning models

Identifying groundwater recharge zones is crucial for sustainable water resource management in water-scarce environments, such as Kurdistan, Iran. This study evaluated four deep learning models for delineating groundwater recharge zones: Artificial Neural Network (ANN), Convolutional Neural Network (CNN), Gated Recurrent Unit (GRU), and hybrid deep learning Convolutional Neural Network-Gated Recurrent Unit (CNN-GRU). Two datasets were used in this study. The first dataset comprised 10 traditional parameters: elevation, rainfall, soil type, drainage density, land use, topographic wetness index (TWI), flow direction, stream power index (SPI), slope, and curvature. The second dataset enhanced the analysis by incorporating additional parameters related to climate patterns. In addition, two feature selection methods, namely Mutual Information (MI) and Random Forest (RF), were employed to identify the most significant factors. In the end, model performance was validated using Accuracy, Kappa score, Root Mean Square Error (RMSE), F1-score, Confusion Matrix, and Receiver Operating Characteristic curve (ROC). Results demonstrate that the hybrid CNN-GRU outperformed other methods, such as ANN, CNN, and GRU, with an Accuracy of 0.9461 and RMSE of 0.2322 on the 13-factor dataset during validation. The enhanced 13-factor dataset consistently improved outcomes across all models compared to the 10-factor dataset, showing the value of climate factors. Finally, the findings from this study reveal that the hybrid CNN-GRU model with climate factors greatly improves accuracy in identifying groundwater recharge zones.