Comparison and Analysis of Four Terrestrial Water Storage Monitoring Models: A Case Study of the Loess Plateau

Accurate estimation of terrestrial water storage change (TWSC) remains challenging in regions where hydrological variability interacts with complex geological conditions and intensive human activities. Taking the Loess Plateau (LP) in the middle Yellow River region as a case study, this work integrates GLDAS simulations, GRACE observations, GNSS vertical-displacement records, a joint GNSS–GRACE inversion, and meteorological data for 2013–2024 to investigate regional TWS variability and model-dependent discrepancies. The results show that GLDAS, GRACE, GNSS, and the joint solution exhibit distinct temporal trends and spatial patterns. GRACE indicates a stronger long-term depletion signal, whereas GNSS-derived equivalent water height (EWH), which relies on the assumption of elastic surface loading, shows a weaker trend but stronger seasonal variability. This discrepancy suggests that GNSS-based inversion over the LP may be affected by non-elastic or non-loading deformation processes, such as wetting-induced loess collapse, aquifer compaction, mining-related subsidence, and other near-surface effects. In contrast, GRACE may include non-TWS mass redistribution associated with soil erosion and mineral exploitation. The joint solution is more consistent with the GLDAS-derived hydrological model benchmark than either single geodetic estimate, but this agreement should not be interpreted as direct proof of higher accuracy or complete removal of non-hydrological effects. Overall, this study highlights the need to diagnose model-dependent discrepancies, effective spatial resolution, and non-loading deformation when applying GRACE- and GNSS-based approaches to TWSC estimation in geologically and anthropogenically complex regions.