Precipitation or evapotranspiration? Bayesian analysis of potential error sources in the simulation of sub-basin discharges in the Czech Elbe River basin

A global change assessment required detailed simulation of water availability in the Elbe River basin in Central Europe (148,268?km2). Using the spatially semi-distributed, eco-hydrological model SWIM, spatial calibration was applied. For 225 sub-areas covering the model domain (134,890?km2), evapotranspiration and groundwater dynamics were individually adjusted. The calibration aimed at good correspondences with long-term run-off contributions and the hydrographs for two extreme years. Measured run-off was revised from water management effects to produce quasi-natural discharges for calibration. At some gauges, there were large volume differences between these reference data and the simulations of the spatially uncalibrated model. Most affected were some sub-basins in the Czech part of the basin where the density of available climate stations was much lower than the German part. Thus, both erroneous precipitation data and systematic flaws in the evapotranspiration module of SWIM could have caused the differences. In order to identify the major error source and to validate the choice of spatial calibration parameters (evapotranspiration and groundwater dynamic corrections), MCMC analyses were made for three Czech areas. Optional precipitation correction had been considered by a third calibration parameter in the MCMC assessment. In two of the three cases, it can be shown that evapotranspiration corrections are preferable as precipitation errors are negligible. In the third case, where the analyses indicate a substantial error in precipitation data, an interpolation problem of the climate data at the edge of the model domain could be found. Hence, the applied method shows its potential to identify specific sources of uncertainty in hydrological modelling.