Annual maximum daily rainfalls will change in the future because of climate change, according to climate projections provided by EURO-CORDEX. This study aims at understanding how the expected changes in precipitation extremes will affect the flood behaviour in the future. The expected changes in precipitation extremes cannot be transformed directly into changes in runoff, as for a given rainfall event, the flood magnitude depends on the initial moisture content in the catchment, which in turn also depends on precipitation and temperature in the days before its occurrence. Therefore, hydrological modelling is required to characterise the rainfall-runoff process adequately in a changing climate to estimate flood changes.
Precipitation and temperature projections given by climate models in the control period usually do not fit exactly the observations in the same period from a statistical point of view. To correct such errors, bias correction methods are used. This paper aims at finding the most adequate bias correction method for both temperature and precipitation projections, minimising the errors between observed and simulated precipitation and flood frequency curves.
Four catchments located in central western Spain have been selected as case studies. The HBV hydrological model has been calibrated, using the observed precipitation, temperature and streamflow data available at a daily scale. Daily rainfall and temperature projections for RCP 4.5 and 8.5 provided by EURO-CORDEX have been used.
The results have shown that the correct calibration of some parameters of the HBV model is essential to obtain coherent results, mainly those related to surface runoff generation. In addition, soil moisture content at the beginning of flood events affects flood magnitudes. Consequently, expected changes in precipitation extremes are usually smoothed by the reduction of soil moisture content due to expected increases in temperatures and decreases in mean annual precipitation. Because of this the rainfall is the most signifcant imput to the model and the best bias correction is quantille mapping polynomial.