Near-Real-Time Monitoring of Global Terrestrial Water Storage Anomalies and Hydrological Droughts

Abstract

Global terrestrial water storage anomaly (TWSA) products from the Gravity Recovery and Climate Experiment (GRACE) and its Follow-On mission (GRACE/FO) have an approximately three-month latency, significantly limiting their operational use in water management and drought monitoring. To address this challenge, we develop a Bayesian convolutional neural network (BCNN) to predict TWSA fields with uncertainty estimates during the latency period. The results demonstrate that BCNN provides near-real-time TWSA estimates that closely match GRACE/FO observations, with median correlation coefficients of 0.92–0.95, Nash-Sutcliffe efficiencies of 0.81–0.89, and root mean squared errors of 1.79–2.26 cm for one- to three-month ahead predictions. More importantly, the model advances global hydrological drought monitoring by enabling detection up to three months before GRACE/FO data availability, with median characterization mismatches below 16.4%. This breakthrough in early warning capability addresses a fundamental constraint in satellite-based hydrological monitoring and offers water resource managers critical lead time to implement drought mitigation strategies.