Enhancing Basin-Scale Hydrological Insights in Greece by Integrating Machine Learning and Satellite Gravimetry.

This study examines the potential of satellite gravimetry for monitoring basin-scale hydrological variability across Greece by downscaling coarse-resolution terrestrial water storage anomalies derived from the Gravity Recovery and Climate Experiment (GRACE) and its successor mission, GRACE Follow-On. Monthly Liquid Water Equivalent (LWE) anomalies from the Jet Propulsion Laboratory’s mascon solutions (~1° resolution) are refined to 0.1° (~10 km) using a supervised machine learning approach. A random forest regression model is trained on a suite of physically relevant environmental predictors, including precipitation, evapotranspiration, runoff, near-surface and land surface temperatures, relative humidity, and vegetation indices, aggregated to monthly scales and spatially aligned with the GRACE grid.

The resulting high-resolution product represents a data-driven reconstruction of GRACE-based water storage anomalies, whose hydrological validity is assessed through cross-comparisons with independent satellite datasets. First, correlations with surface soil moisture time series evaluate the coherence of near-surface and total water storage variability. Second, multi-mission radar altimetry data over Lake Kremasta and Lake Polyfytou are analyzed to determine consistency between lake level fluctuations and GRACE-derived patterns. These comparisons serve as an indirect validation of the downscaled product’s hydrological relevance.

By integrating satellite gravimetry, environmental indicators, and machine learning techniques, this research offers a scalable framework for enhancing the spatial resolution of terrestrial water monitoring in data-scarce regions. It contributes to understanding the strengths and limitations of data-driven GRACE downscaling for hydrological applications.