Evaluating hydrological forcing datasets for GRACE-based terrestrial water storage downscaling in Central Asia

Gravity Recovery and Climate Experiment (GRACE) and GRACE Follow-On (GRACE-FO) observations provide unique large-scale information on terrestrial water storage (TWS), yet their coarse spatial and temporal resolutions limit their applicability for regional and event-scale hydrological analyses. In this study, we investigate the performance of different hydrological forcing datasets in a flexible three-step downscaling framework to derive daily, 1 km terrestrial water storage change (TWSC) estimates over the Naryn–Kara Darya basins and the Fergana Valley in Central Asia. The framework integrates monthly GRACE-derived TWSCs with high-resolution precipitation, evapotranspiration, and runoff information from multiple sources, including GLDAS, FLDAS-CA, ERA5-Land, and a mixed forcing combination based on MSWEP, GLEAM, and GloFAS. Temporal downscaling is achieved by constraining daily water-balance-derived storage changes with GRACE observations, while spatial downscaling maps coarse GRACE signals onto fine-scale hydrological predictors. Model performance is assessed using multiple validation strategies, including comparison with the ITSG-Grace2018 daily solution, consistency tests, and event-based analyses, accounting for the scarcity of in situ observations in the region. Our results demonstrate that the choice of hydrological forcing dataset strongly influences the quality of downscaled TWSCs. While all forcing scenarios capture the dominant seasonal and interannual variability, substantial differences emerge in their representation of trends, variability, and short-term events. In particular, the mixed forcing dataset shows the most consistent performance across validation metrics and better reproduces both long-term TWS changes and hydrologically relevant extreme events. These findings highlight the critical role of forcing data selection in GRACE downscaling applications and demonstrate the transferability of the proposed framework to other data-sparse regions.