Simulation of summer temperature extremes over the Czech Republic in regional climate models

Three regional climate models (RCMs) are examined in view of the reproduction of high summer temperatures in control climate runs and scenarios of changes for the late 21(st) century over the Czech Republic. The 'peaks-over-threshold' analysis is applied for modelling probabilities of extremes and the estimation of high quantiles, and particular attention is paid to the choice of the threshold values and the minimum separation time between exceedances. The model validation shows that rather large differences between control climate runs and observations exist for average summer temperatures and mean annual maxima, mainly in HadRM, and the biases become more pronounced for high quantiles of temperature distributions (reaching 8.3 degrees C for 50-yr return levels in HadRM if averaged over the area, and 12.6 degrees C in individual grid boxes). The particular formulation of the RCMs plays a more important role in influencing the spatial patterns and biases of summer temperatures than within-ensemble variability. A strong dependence of the bias in the temperature characteristics on altitude in HadRM and HIRHAM is related to errors in simulating the vertical gradient of cloud cover, suggesting deficiencies in the parameterization of convection. Scenarios of changes for the late 21(st) century yield large increases in mean annual temperature maxima as well as the 20-yr and 50-yr return values, which appear to be related to the projected summer drying over central and south-eastern Europe. As the magnitudes of the increases largely depend on the RCM, the scenarios of changes and/or daily time series used for a future time horizon in climate change impact studies should be based on an ensemble of outputs from different RCMs and thus reflect the uncertainty due to the RCM formulation.