Accurate forecasting of groundwater level fluctuations plays a crucial role in the sustainable management of water resources, particularly in arid and semi-arid regions. The aim of this study is to develop and evaluate a data-driven, deep-learning-based framework for groundwater level prediction in the Birjand aquifer using multi-temporal and multi-band satellite remote sensing data. A comprehensive set of remote sensing indicators—including vegetation indices, land surface temperature (LST), evapotranspiration, and various hydro-climatic parameters—was extracted and, after applying advanced feature engineering procedures, used as input to a suite of machine learning and deep learning models. A key innovation of this research is that, unlike many previous studies, no direct groundwater-level lag variables were used in the modeling process.

Model evaluation results showed that the proposed DNN-Enhanced model achieved superior performance compared to the other models, with a coefficient of determination of R² = 0.9807 and an RMSE of 4.723 meters. Variable importance analysis revealed that vegetation-related indices, particularly NDVI, as well as temperature-based indices, especially extreme temperature variables, contributed most significantly to groundwater level prediction. Furthermore, uncertainty analysis demonstrated that the proposed model exhibits stable and reliable performance, successfully covering the full range of 95% confidence intervals.

Overall, the findings indicate that integrating multi-source remote sensing data with enhanced deep learning architectures—without relying on direct groundwater lag inputs—can yield accurate and reliable groundwater level predictions. The proposed framework can serve as an effective tool for groundwater monitoring and management, drought assessment, and aquifer operation planning.