Impact of statistical bias correction on the climate change signal of extreme climate indices from convection-permitting climate simulations over central Europe

Systematic biases are still inherent in the newest generation of regional climate models at convection-permitting scale. This complicates the direct application of such simulation results for impact studies with vegetation models. Here, we investigate the impact of a statistical bias correction method (quantile mapping) after Piani et al. (2010) on the climate change signal (CCS) of extreme climate indices in time and space-distribution from convection-permitting climate simulations based on the Representative Concentration Pathway (RCP) 8.5. In the frame of the Multisectoral analysis of climate and land use change impacts on pollinators, plant diversity and crops yields (MAPPY), transient regional climate model simulations are performed with COSMO-CLM (v5.16) at a spatial horizontal resolution of 3 km over central Europe from 1980 to 2070. CCSs are computed from the ETCCDI set of climate extreme indices for the “near” (2021-2050) and the “far” (2041-2070) future relative to the reference period 1981-2010.

We find that model biases influence the spatial distribution of climate extremes, even though the mean properties are not heavily changed. However important differences are observed for the total precipitation amount and for heavy precipitation indices. Bias-corrected precipitation data show an increase of 3.5% for the “far” future in the annual total precipitation relative to the reference period. Non bias-corrected data would instead suggest a lower increase of 0.7%. The frequency of heavy precipitation days is also enhanced in the bias-corrected data. For example the amount of rainfall which exceeds the 95 and 99 percentiles for the “far” future is 12.7% more than the reference period. The projections from the non bias-corrected data would instead predict an increase of 9.4% and 9.2% respectively.

The bias-corrected simulation data for the temperature parameters suggest generally warmer winters for both the “near” and “far” future periods with a dampening of the extreme temperatures. As an example, the maximum values of the daily maximum temperatures in the “far” future are in average 1.6 °C warmer relative to the reference period. The non bias-corrected data would instead return an higher value of about 1.1 °C (i.e. 2.7 °C). Vice versa the minimum values of the daily minimum temperatures in the “far” future are in average 2.2 °C warmer relative to the reference period, whereas the non bias-corrected data give a lower increase of 1.8 °C. The dampening of extreme temperatures is also consistent with other observations such as the percentage of warm days, where the maximum temperature is above the 90 percentile or the number of frost days, where the minimum temperature is below 0 °C. In both latter cases the bias-corrected data give lower values with respect to the non bias-corrected data with a relative difference of about 30%.

We conclude that systematic biases in regional climate models can have a significant impact on climate change signals both in space distribution and absolute values. Yet the statistical robustness of our results, the seasonal variability of some extremes as well as the dependency on the resolution scale is currently under investigation.