Bridging GRACE/GRACE-FO data gaps with ESTConvS2S for enhanced water storage monitoring

The GRACE and GRACE-FO satellite missions have revolutionized the monitoring of terrestrial water storage (TWS), providing critical insights into groundwater variability—particularly in regions facing water scarcity and excessive extraction. However, the transition between these missions led to an 11-month data gap (July 2017–May 2018), introducing biases and uncertainties into groundwater assessments. To address this challenge, we proposed the Enhanced Spatiotemporal Convolutional Sequence-to-Sequence Network (ESTConvS2S) model, which incorporates a spatiotemporal framework and integrates attention mechanisms to reconstruct data gaps during the transition period. Applied to southwestern China—a region characterized by intricate karst topography, complex hydrological dynamics, and diverse climatic regimes—the model effectively bridged the gap between GRACE and GRACE-FO. Results indicate strong agreement between ESTConvS2S predictions and GRACE observations (maximum correlation coefficient r = 0.81, p ＜ 0.05), further validated against in situ measurements. The derived groundwater storage anomaly (GWSA) exhibited high accuracy, underscoring the model's potential for advancing groundwater monitoring under climate change. Moreover, the model revealed spatially heterogeneous groundwater depletion rates across different karst lithologies, highlighting its capacity to resolve detailed hydrological patterns crucial for informed water resource management in geologically diverse settings.