Coupling Remote Sensing and Artificial Intelligence: Mapping Hydrological Variables with GEE and Predicting Surface-water Extent with Neural Networks

Managing water resources in water-scarce regions requires rigorous monitoring and reliable forecasting. To address this, we developed RSAI (Remote-Sensing-Artificial-Intelligence), a reproducible computational workflow that couples Google Earth Engine (GEE) scripts with deep learning models. RSAI is designed to map the spatiotemporal trends of six hydrological variables and specifically predict monthly surface-water extent. The workflow was applied to the Coastal Algiers Basin (02 A) over the 2001-2022 period. Trend analysis using the Mann-Kendall test revealed a statistically significant decline in surface-water area (Sen's slope = -1.03 km² year⁻¹, equivalent to a 36% loss), while climatic drivers (precipitation, temperature) showed non-significant trends, suggesting a decoupling driven by anthropogenic factors. For the predictive component, three neural network architectures (ANN, DNN, and LSTM) were trained on the GEE-derived time series using a 70/30% train-test split. The Deep Neural Network (DNN) achieved the most robust performance on the independent test set (R = 0.89, MAE = 0.36 km², RMSE = 46.31 km²), outperforming the LSTM and ANN models. Sensitivity analysis further confirmed the DNN's robustness to sensor noise (Propagation Factor = 0.26). These results validate RSAI as a practical, transferable workflow for basin-scale hydrological monitoring and forecasting in data-limited Mediterranean regions.