Uncertainties in GRACE-derived terrestrial water storage changes over mainland China based on a generalized three-cornered hat method

It remains a challenging problem to assess the uncertainties in GRACE-derived terrestrial water storage (TWS) changes due to insufficient observations in many areas globally. Particularly, China has a complicated terrain covering a range of climate settings, aquifers and levels of human interventions. Therefore, it is very important to evaluate the performance of GRACE observations from different processing centers for GRACE applications in China. Due to the absence of ground truth observations, this study analyzed the relative uncertainties in GRACE-derived TWS changes from five solutions over mainland China based on a generalized three-cornered hat (TCH) method, including Stokes coefficients from the Center of Space Research (CSR), the German Research Center for Geoscience (GFZ), the Groupe de Recherches de geodesie spatiale (GRGS) and the Huazhong University of Science and Technology (HUST) as well as the Jet Propulsion Laboratory (JPL) GRACE mascon solution. The results showed that the averaged uncertainties ( in terms of root-mean-square, RMS) of derived monthly TWS changes over mainland China for CSR, GFZ, JPL, HUST and GRGS were 14.4 mm, 26.3 mm, 25.3 mm, 26.6 mm and 56.1 mm, respectively; compared to the monthly scale, the uncertainties of each solution were lower at seasonal and annual scales for both the original and nonseasonal terms; the inverted TWS changes were more consistent at the nonseasonal term (after removing the trend and seasonal cycles from the original signals). At the basin scale, except for the Yangtze River basin, CSR showed the lowest uncertainties for the 13 river basins over mainland China, while GRGS showed relative large uncertainties. In addition, GRGS-based TWS showed larger variability than other GRACE solutions and two hydrological models (Global Land Data Assimilation System, GLDAS and WaterGAP Global Hydrology Model, WGHM) in the temperate continental climate region; CSR and JPL were less affected by the surrounding hydrological conditions, climate settings, size and geometry of the basins, with the uncertainties varying from 2.3 similar to 17.1 mm and 5.6 similar to 22.5 mm, respectively. Whereas GFZ and HUST were influenced more by these factors, with the ranges of the uncertainties were 5.5 similar to 35.1 mm and 4.0 similar to 40.6 mm, respectively. It should be noted that GRGS-derived TWS changes were not restored because of the difficulty in quantifying signal loss resulted from regularization for end-users. High uncertainty in GRGS suggests serious signal loss in regularized GRACE solution in the study region. Therefore, more independent data or models are necessary for validation before using the GRGS solution. This study provides a new way to assess the uncertainty in GRACE products and will be helpful for choosing proper models for specific studies.