Downscaling GRACE Data for Improved Groundwater Forecasting Using Artificial Neural Networks

This study introduces a dual-phase approach utilizing Artificial Neural Networks (ANNs) to overcome the challenges of groundwater monitoring at regional scales. Traditional well-based methods provide limited spatial coverage, while GRACE satellite data, despite its value for large-scale hydrological analysis, suffers from low spatial resolution (~300 km), limiting its application for local-scale assessments. Existing downscaling methods such as geographically weighted regression and Random Forests are computationally intensive and often lack adaptability to complex groundwater systems. In this study, Phase 1 refines GRACE data using ANNs to achieve a 4×4 km spatial resolution, addressing the resolution challenge for regional applications. Phase 2 integrates the downscaled GRACE data with groundwater well observations and climatic factors to predict groundwater levels with high accuracy (R² = 0.9885). This dual-phase framework demonstrates significant improvements over existing methods, providing an efficient and scalable solution for groundwater monitoring in hydrologically complex regions. The findings highlight the potential of machine learning to enhance groundwater resource management, particularly in addressing water scarcity and climate variability challenges. Doi: 10.28991/CEJ-2025-011-02-01 Full Text: PDF

Downscaling GRACE Data for Improved Groundwater Forecasting Using Artificial Neural Networks Keywords:

Artificial Neural Networks , GRACE Data , Downscaling , Groundwater Level.