Sensitivity of Earth Model Selection on Terrestrial Water Storage Inversion From GNSS Vertical Displacements

Abstract

The global navigation satellite system (GNSS)-inverted terrestrial water storage (TWS) approach commonly adopts the PREM Earth model while neglecting crustal heterogeneity. This study systematically evaluates the impacts of different Earth models, including three spherically symmetric models (PREM, STW105, and AK135F) and two global crustal models (CRUST and TEA12), on GNSS-inverted TWS through synthetic experiments over part of mainland China (100°E–115°E and 25°N–40°N). Synthetic checkerboard experiments show that, relative to PREM-derived equivalent water height (EWH, 250.79 mm), inversion errors are 11.49% for AK135F, 1.85% for STW105, and ∼25% for crustal models. Using CRUST-derived EWH (234.90 mm) as reference, the relative error of TEA12 is 7.36%. For the synthetic smooth model (mean EWH ∼565 mm), the inversion errors are below 2% among spherically symmetric Earth models, within 4% among crustal models, and ∼11% between symmetric Earth models and crustal models. Application to GNSS observations in Yunnan Province (China) shows that EWH annual amplitudes from different Earth models exhibit spatial patterns generally consistent with GLDAS and GRACE, but substantial magnitude differences, with the maximum mean absolute difference reaching 41.82 mm (PREM versus CRUST). The discrepancies are spatially correlated with GNSS station density, and decrease as station coverage increases. At larger spatial scales (>1000 km), the regional mean EWH shows limited sensitivity to Earth model selection, whereas at smaller scales (<50 km), spherically symmetric models are recommended for Mojiang County. These results highlight Earth model selection as a key factor affecting GNSS-inverted TWS accuracy, and provide guidance for high-precision hydrological monitoring.