The effect of different calibration periods based on bias-adjusted EURO-CORDEX simulations over Hungary

As a result of technological progress, general circulation models and regional climate models (GCMs and RCMs, respectively) became the principal tools for climate science. It is important to note, that these model simulations are encumbered with uncertainty of various origins, leading to biases in model outputs. By using bias-adjusted datasets and evaluating several RCMs together as members of an ensemble, the above-mentioned uncertainties can be quantified and reduced. In addition, unbiased data is required for impact studies (e.g. hydrology, agriculture) and the implementation of a bias-correction procedure has become a standard step in the process of using climate model outputs. However, a reliable observational dataset is required serving as reference data for all bias-adjustment methods.

Coordinated Regional Downscaling Experiments (CORDEX) is an ongoing international initiative which provides a large number of climate model simulations for 14 domains worldwide. EURO-CORDEX is a sub-programme of CORDEX, covering the European domain and providing raw and bias-adjusted RCM outputs at a horizontal resolution of 0.11° (about 12.5 km) and 0.44° (about 50 km). In our study an ensemble of 5 RCMs (CCLM, HIRHAM, RACMO, RCA, REMO) with the finer resolution driven by 4 different GCMs are investigated for the period 1976–2099 under two radiative forcing scenarios (RCP4.5 and RCP8.5). Bias-corrected model simulations are also available from the EURO-CORDEX, which were produced using a distribution-based scaling method and the calibration period of 1989–2010 from the MESAN reanalysis data.

The goal of our research is to investigate how the choice of the reference dataset and different calibration periods affects the results of the bias-corrected simulations focusing on Hungary. For this purpose, a bias-adjustment was carried out by applying the percentile-based quantile mapping method, using the HuClim dataset as a reference, which is the most accurate, measurement-based, quality controlled gridded data for Hungary currently available for the period 1971–2022. Two calibration periods were chosen for this procedure: an earlier (1976–2005) and a more recent (1993–2022) period. Four variables are used for this study (daily minimum- and maximum temperature, mean temperature and precipitation) and the following climate indices are assessed: summer days, frost days, tropical nights, wet days, the warmest day, the coldest night, the highest daily precipitation amount and extremely wet days. The validation of the data is presented here for the selected validation period 1993–2005, which is the common part of the three different calibration periods.

According to our preliminary results, the accuracy of the bias-corrected simulations depends on the chosen calibration period and the selected climate index. The average annual number of tropical nights are overestimated by bias-adjusted simulations using MESAN and the later HuClim period, but simulations corrected by the earlier HuClim period are in a good agreement with the reference values. In the case of precipitation-related indices negligible differences can be seen for the two sets of HuClim-based bias-adjusted model outputs, while the bias-adjusted data based on the MESAN dataset generally show more pronounced underestimation.