Multiscale Prediction and Verification of Water Fluxes and States over Large River Basins

Abstract:

Developing the ability to predict streamflow at various scales is a key step for improving our understanding of the water balance and the skill of hydrologic models (HM). To cope with this grand challenge, we postulate: 1) validating a HM only against an integral basin response such as streamflow is a necessary but not a sufficient condition to warranty the proper partitioning of incoming precipitation and net-radiation into different water storage components and fluxes, 2 proper partitioning can be ensured by using additional multi-scale observations during parameter estimation (PE), and 3) HM should be evaluated at locations and scales different from those used for PE to ensure transferability.

Water fluxes and states variables are estimated over the Pan-EU with the mesoscale hydrologic model (mHM). Its main features are the treatment of the sub-grid variability of input variables and model parameters and the possibility to estimate fluxes in nested-scales and/or in multiple basins keeping its regionalization coefficients unaltered across scales and basins. These essencial features allow to simulate fluxes and states and to evaluate them with disparate sources of information such as satellite, streamflow, and eddy covariance data at their native resolutions.

mHM was setup over more than 340 Pan-European river basins. This model was forced with the gridded EOBS data set (1/4º, ECA&D) for the period 1950-2012. Morphological data was derived from the FAO soil map (1:5000000), the SRTM DEM (500 m), CORINE land cover (500 m), and MODIS LAI. The multi-scale evaluation was carried out using latent heat (LH) from FLUXNET stations and LandFlux-EVAL (1/2º), runoff from GRDC gauging stations, soil moisture from ESA-CCI (1/4º), total water storage (TWS) anomalies from GRACE (1º) and groundwater (GW) stage stations. Results lead to the conclusion that mHM water fluxes are robust since less than 25% of river basins exhibit Nash-Sutcliffe efficiencies of 0.5 or less. Simulated TWS anomalies and LH exhibit a large spatial and temporal correlation with GRACE and point measurements respectively. The first postulate, however, is clearly confirmed in extreme situations. Deficient performance in streamflows is mainly attributed to either heavily regulated river basins or regions with a poor rainfall gauge network.