Improving drought simulations within the Murray-Darling Basin by combined calibration/assimilation of GRACE data into the WaterGAP Global Hydrology Model

Simulating hydrological processes within the (semi -)arid region of the Murray-Darling Basin (MDB), Australia, is very challenging specially during droughts. In this study, we investigate whether integrating remotely sensed terrestrial water storage changes (TWSC) from the Gravity Recovery And Climate Experiment (GRACE) mission into a global water resources and use model enables a more realistic representation of the basin hydrology during droughts. For our study, the WaterGAP Global Hydrology Model (WGHM), which simulates the impact of human water abstractions on surface water and groundwater storage, has been chosen for simulating compartmental water storages and river discharge during the so-called 'Millennium Drought' (2001-2009). In particular, we test the ability of a parameter calibration and data assimilation (C/DA) approach to introduce long-term trends into WGHM, which are poorly represented due to errors in forcing, model structure and calibration. For the first time, the impact of the parameter equifinality problem on the C/DA results is evaluated. We also investigate the influence of selecting a specific GRACE data product and filtering method on the final C/DA results. Integrating GRACE data into WGHM does not only improve simulation of seasonality and trend of TWSC, but also it improves the simulation of individual water storage components. For example, after the C/DA, correlations between simulated groundwater storage changes and independent in-situ well data increase (up to 0.82) in three out of four sub-basins. Declining groundwater storage trends - found mainly in the south, i.e. Murray Basin, at in-situ wells - have been introduced while simulated soil water and surface water storage do not show trends, which is in agreement with existing literature. Although GRACE C/DA in MDB does not improve river discharge simulations, the correlation between river storage simulations and gauge-based river levels increases significantly from 0.15 to 0.52. By adapting the C/DA settings to the basin-specific characteristics and reducing the number of calibration parameters, their convergence is improved and their uncertainty is reduced. The time-variable parameter values resulting from C/DA allow WGHM to better react to the very wet Australian summer 2009/10. Using solutions from different GRACE data providers produces slightly different C/DA results. We conclude that a rigorous evaluation of GRACE errors is required to realistically account for the spread of the differences in the results.