**5. Conclusions**

In this paper, the influence of bias corrections on K-G climate classification was investigated. Climate classification was calculated by eQM-corrected precipitation and temperature, by a combination of gpQM-corrected precipitation and eQM of temperature, by a combination of power transformation of precipitation and variance scaling of temperature, or by a combination of LOCI for precipitation and variance scaling for temperature. These bias correction methods were applied in five 25 km resolution ENSEMBLE RCMs in the historical time period of 1961−1990 and their results were compared with climate classification based on E-OBS-observed precipitation and temperature values to study their performance. The corrections were tested by a split-sample test, where the 1961−1980 period was training, and the corrections were validated in the 1981−2000 period. Subsequently, the climate classification was evaluated in eight individual subdomains: the Alps, the British Isles, Eastern Europe, France, the Iberian Peninsula, the Mediterranean, Mid-Europe and Scandinavia, defined according to the methodology devised for the PRUDENCE project.

When assessing the performance of the bias correction methods, we found similar results for eQMand gpQM-corrected K-G classifications when daily data were used during the whole 30-year time period (not shown). Both of them were strongly dependent on the RCM, as the simulated climate zones varied between these RCMs. Moreover, the simulated climate zones significantly differed from the observed ones. These differences stemmed from the large bias in the seasonal precipitation amount. The 90-day moving window improved these correction methods. In comparison, a combination of LOCI and power transformation for precipitation with variance scaling of temperature, respectively, properly reproduced the climate zones by each of the RCMs in each region in the historical period. Furthermore, their test run contained the smallest differences from the observed K-G zones in most regions.

Our results suggest that the eQM and gpQM methods manifest a strong dependence on the spatial distribution of parameters, and this dependence causes a limitation in climate classification considering the large domain. Conversely, power transformation–and local intensity scaling of precipitation and variance scaling of temperature corrections–also generated a smaller bias between the simulated and observed parameters, except in HIRHAM in JJA, but their combination produced better results in climate classification for the whole European domain. This can be explained by the fact that they are distribution-free approaches.

This study is valid for Europe as a whole, since it was based on the E-OBS dataset with a resolution that may be coarser than that of some small regions studied in the quoted papers, where dense national datasets could be used. In the latter case, the statistical properties of the points reflect the smaller area and the results of the method evaluations could be different. It was beyond the scope of this study to devote itself to the several high-resolution gridded datasets that exist in Europe, but this will be the topic of future investigation using the next generation EURO-CORDEX regional climate model simulations.

**Author Contributions:** Writing–original draft preparation, B.S.-T.; review and editing, P.S., A.F. and J.M.

**Funding:** This research was funded by Ministry of Education, Youth and Sports of CR within the National Sustainability Program I (NPU I), grant number LO1415.

**Acknowledgments:** We are thankful for the E-OBS data set from the EU-FP6 project ENSEMBLES (http://www. ensembles-eu.org) and the data provided by the ECA&D project (http://www.eca.knmi.nl).We would like to thank Sixto Herrera García for his help in bias correction and MeteoLab application. We also thank the anonymous reviewer(s) for their comments.

**Conflicts of Interest:** The authors declare no conflict of interest. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or in the decision to publish the results.
