Remote-Sensing-Driven Soft Computing for Linking Vegetation Dynamics and Groundwater Storage Variations in the Eastern Urmia Lake Basin, Iran

Groundwater depletion and vegetation degradation are closely related environmental challenges in the Eastern Urmia Lake basin, where prolonged drought and intensive water use have gradually altered the region’s hydrological and ecological conditions. Although satellite-derived vegetation indices are widely used to monitor drought and land-cover change, their potential for explaining regional groundwater variability has received less attention, particularly in semi-arid basins with limited and discontinuous field observations. This study develops a remote-sensing-driven soft computing framework in which vegetation dynamics are considered an indirect ecohydrological indicator of groundwater storage variability. Landsat imagery acquired between 2005 and 2025 was processed to derive vegetation indicators, including the Normalized Difference Vegetation Index (NDVI) and Soil-Adjusted Vegetation Index (SAVI), and to classify vegetation cover using a Fuzzy Logic-based approach. Regional groundwater storage variations were estimated from GRACE/GRACE-FO terrestrial water storage anomalies and combined with climatic and hydrological variables. These remotely sensed and hydroclimatic variables were used as inputs to Fuzzy Logic, Feedforward Neural Network (FFNN), and Adaptive Neuro-Fuzzy Inference System (ANFIS) models. Model performance was assessed using a temporally independent hold-out subset of the GRACE/GRACE-FO-derived groundwater storage series that was not used during model training. Therefore, the reported RMSE, MAE, and R² values represent out-of-sample predictive performance rather than validation against independent in situ groundwater measurements. ANFIS provided the most consistent results, with an RMSE of 0.42 m, MAE of 0.31 m, and R² of 0.86. FFNN achieved an RMSE of 0.52 m, MAE of 0.39 m, and R² of 0.79, whereas the Fuzzy Logic model obtained an RMSE of 0.66 m, MAE of 0.50 m, and R² of 0.68. The results indicate that ANFIS was more effective in representing the nonlinear relationships among vegetation conditions, climatic forcing, and GRACE-derived groundwater storage variations. The proposed framework provides a practical approach for regional groundwater assessment in drought-affected areas where continuous ground observations are limited.