Abstract:
The accuracy of terrestrial water storage (TWS) estimates from a land surface model is, in large part, limited by a lack of observations for model evaluation and by inherent biases and uncertainties present in the land surface model. In order to enhance model-based estimates of TWS and its constituent components, a multi-variate, multi-sensor data assimilation (DA) framework is presented that simultaneously assimilates low-frequency (10-36 GHz) passive microwave brightness temperature (Tb) observations over snow-covered terrain and gravimetric retrievals of TWS from the Gravity Recovery and Climate Experiment (GRACE). The framework uses the NASA Catchment land surface model (Catchment) and an ensemble Kalman filter (EnKF). A synthetic case study is presented for the Volga River Basin in Russia that compares model results with DA and without assimilation (open loop; OL) against synthetic observations of hydrologic states and fluxes. The results show that the DA procedure has more accurate estimates of TWS, TWS anomalies, SWE, and subsurface storage (i.e., soil moisture plus groundwater) compared to the OL. Dual assimilation of GRACE and AMSR-E typically led to more accurate estimates of all four hydrologic states in terms of smaller unbiased root mean square error (ubRMSE) compared to the assimilation of only AMSR-E or only GRACE observations. For subsurface storage estimates, GRACE-only DA had a smaller ubRMSE (1.3 cm) than AMSR-E only DA (ubRMSE = 1.4 cm) whereas dual assimilation yielded ubRMSE = 1.1 cm, which was reduced from the OL (ubRMSE = 1.5 cm). In terms of SWE estimates, GRACE-only DA had a bigger ubRMSE (1.2 cm) than AMSR-E only DA (ubRMSE = 1.1 cm) whereas dual assimilation yielded ubRMSE = 1.0 cm, which was reduced from the OL (ubRMSE = 1.3 cm). TWS and TWS anomaly estimates for GRACE-only and AMSR-E only assimilation were similar. The results demonstrate that GRACE and AMSE-E can be jointly assimilated to further improve TWS constituent estimates.