Uncertainty in simulated groundwater and terrestrial water storage as revealed through comparison with GRACE data and GRACE data assimilation

Numerical simulation is often needed when assessing climate impacts on global groundwater and terrestrial water storage (TWS) given the scarcity of in situ and satellite observations of these states at the spatial and temporal scales of climate variability. Studies have shown that model simulations of long-term TWS and groundwater storage variability may exhibit significant discrepancies due to inconsistencies in the atmospheric data used to force the models and differences in the parameterization schemes employed by the models. Biases in re-analysis-based atmospheric forcing fields such as precipitation and temperature can be corrected using in situ and satellite observations, but their temporal variability relies solely on the underlying re-analysis data. Further, large-scale hydrological models typically are only calibrated or tuned using mean annual fluxes (streamflow or ET) which provide no constraints on the temporal variability of simulated states and fluxes. In this study, we evaluate groundwater and TWS estimates from several global-scale models using GRACE and GRACE FO data as well as output from our global-scale GRACE and GRACE FO data assimilation to assess uncertainty in their simulated inter-annual variability, and also investigate the sources of uncertainty. The results have implications for understanding uncertainty in climate model predictions of terrestrial water storage changes. We also present our latest global-scale GRACE/GRACE FO data assimilation results.