Calibration of a global hydrological model while simultaneously assimilating satellite-derived total water storage anomalies and in-situ streamflow observations

Global hydrological models simulate water storages and fluxes of the water cycle, which is important for e.g. drought and flood predictions. However, model simulations are underlying uncertainties due to the inputs (e.g. climate forcing data), parameters, and model assumptions, resulting in disagreements with observations. To reduce these uncertainties, models are often calibrated against in-situ streamflow observations or compared against total water storage anomalies (TWSA) derived from the Gravity Recovery And Climate Experiment (GRACE) satellite mission. In recent years, TWSA data are integrated into some models via data assimilation.

In this study, we jointly assimilated TWSA and streamflow observations into the WaterGAP Global Hydrology Model (WGHM) applying an Ensemble Kalman Filter. Simultaneously, model parameters are calibrated via state vector augmentation. Our simultaneous calibration and assimilation (CDA) approach was tested within the Mississippi River Basin from 2003 to 2016.

First, we evaluated how the spatial resolution and study set up impact our CDA approach. Our results suggest that applying the CDA approach sequentially to all subbasins works better than applying the approach once to the entire Mississippi River Basin. Second, we compared the results of our CDA approach against uncalibrated model simulations as well as the results of the WGHM standard calibration. The CDA approach led to higher Nash-Sutcliffe efficiency (NSE) and lower root mean square error (RMSE) values (and thus a better agreement with the observations) regarding TWSA and streamflow than the uncalibrated WGHM simulations, which is in line with our expectations. In addition, it also resulted in higher NSE and lower RMSE values than the WGHM standard calibration in most subbasins. This was expected for the metrics regarding TWSA. Our expectations regarding the streamflow results were more complex: On one hand, our findings were surprising since the WGHM standard calibration approach is based on streamflow observations only and takes significantly more streamflow stations into account than the CDA approach. On the other hand, the results reflected that less parameters are calibrated and only the long-term averages of the streamflow observations are considered in the WGHM standard calibration approach.