Assessment of inter-product uncertainty in GRACE/GRACE-FO-derived terrestrial water storage

Terrestrial water storage anomalies (TWSA) derived from the Gravity Recovery and Climate Experiment (GRACE) and its successor, GRACE Follow-On (GRACE-FO), provide essential information for assessing regional and global water storage changes. Several processing centers produce GRACE(-FO) TWSA products intended for use by non-geodesy experts, particularly for validating and calibrating large-scale hydrological models. However, discrepancies among these TWSA products remain poorly understood within the hydrology community. In this study, we quantify differences among six GRACE(-FO) TWSA products—three leakage-corrected spherical harmonic solutions (HydroSat, COST-G, and GFZ-GravIS) and three mascon solutions (CSR-M, GSFC-M, and JPL-M)—for the period April 2002 to December 2024 across 15 large river basins (>210,000 km²). We compare linear trends and seasonal amplitudes and apply the Generalized Three-Cornered Hat (GTCH) method to identify which products deviate most and least from the others. Our results show substantial variability in both trends and seasonal amplitudes across products and basins. For example, in the Uruguay Basin, the trend estimated by HydroSat is 1.77 mm/yr, whereas GFZ-GravIS reports -5.08 mm/yr. In the Churchill Basin, the seasonal amplitudes of the detrended time series range from 84 mm (GFZ-GravIS) to 45 mm (JPL-M), with a median of 57 mm. In another case, the mean range of TWSA across products in the Bravo Basin is 1.7 times larger than the median seasonal amplitude across the same products. According to the GTCH results, GFZ-GravIS shows the largest disagreements with the other products across all 15 basins, whereas GSFC-M and COST-G achieve the highest agreement in 7 and 5 basins, respectively. These results demonstrate that reliance on a single GRACE(-FO) product can mislead hydrological model evaluation. We therefore recommend using ensemble-based approaches and explicitly accounting for GRACE TWSA uncertainty when employing these products as reference data.