Estimating Groundwater Levels From Satellite Data And Machine Learning: A Novel Approach For Water Management
Presenter: Hector Leopoldo Venegas Quinones1
Co-Author(s): -
Advisor(s): Dr. P.A. Ty Ferré
1Department of Hydrology and Atmospheric Sciences, University of Arizona
Groundwater is a critical resource for human and ecological systems, and its management requires accurate and timely monitoring of its levels. In this study, we developed a novel approach that combines remote sensing and machine learning to estimate groundwater levels in areas where traditional monitoring methods are not available or insufficient. We used satellite data to extract surface features, including vegetation indices, land surface temperature, and topography, which are known to influence groundwater levels. We then trained a machine learning algorithm to predict groundwater levels based on these features, using the extensive ground-based measurements from the USGS and ADWR as the ground truth. The approach can accurately estimate groundwater with a test R2 of 0.99 and an RMSE of 0.2. The three most informative sources of information for inferring groundwater level were: distance to the closest streambed, stream order, and soil moisture. The technique is fully implemented in an easy-to-use format that would allow non-experts to access water levels throughout the state and through time. We expect that this will be a valuable resource as communities face increased demands on groundwater in the coming years.