Scale-dependent model–observation inconsistencies in global terrestrial water storage models

Accurate representation of terrestrial water storage is essential for water resource management and climate adaptation, yet model‑observation inconsistencies hinder global water cycle assessment. Here we show, using a unified multi‑scale framework, that seven model families exhibit scale‑dependent performance against Gravity Recovery and Climate Experiment observations across global, climate-zone, and multiple basin scales (large, medium, small). Globally, hydrological models show the highest temporal correlation (0.94) but underperform spatially (0.32), whereas the satellite observation‑constrained assimilation system balances both dimensions with higher spatial agreement (0.73). Across climate zones, the assimilation system maintains robust performance, while physical models degrade in polar regions (-0.40). From large to small basin scales, performance deteriorates systematically: correlations decline from 0.67 to 0.33, while the assimilation system maintains consistency and higher directional accuracy (80.65%). These findings demonstrate that the satellite observation-constrained assimilation system dramatically reduces inconsistencies, providing essential guidance for climate adaptation and water management strategies.