*2.2. Datasets and Bias Corrections*

For the analysis of bias correction influence on K-G zone distribution in Europe, we used simulations of five regional climate models from the ENSEMBLES project, as summarized in Table 2. The large scale forcing for two RCMs was taken from driving ARPÉGE GCM, and three of them were driven by ECHAM5-r3 GCM. The E-OBS version 10 gridded dataset of daily station observations with a spatial resolution of 0.25º in longitude and latitude was used as a reference dataset for validation and bias correction in the period from 1961 to 1990. Before the direct comparison of models and observations, the RCMs were interpolated from their native grids to the E-OBS 0.25 º regular grid by the nearest neighbour remapping method.

**Table 2.** The institute, global climate models (GCMs), regional climate models (RCMs) and resolution of chosen models from the ENSEMBLES EU project.


The simulated climate zones were analysed in the Alps (AL), the British Isles (BI), Eastern Europe (EA), France (FR), the Iberian Peninsula (IP), the Mediterranean (MD), Mid-Europe (ME) and Scandinavian (SC) regions (Figure 1) specified in the framework of the PRUDENCE project [38].

**Figure 1.** Subdomains based on the Prudence project: the Alps (AL), the British Isles (BI), Eastern Europe (EA), France (FR), the Iberian Peninsula (IP), the Mediterranean (MD), Mid-Europe (ME) and Scandinavia (SC).

Figure 2 demonstrates the simulated K-G zones without bias correction. Large differences can be seen between the simulated zones and between the simulated and observed zones. The distribution of K-G zones varied from the K-G zones based on the observed parameters in the case of ARPÈGE driven RCMs. HIRHAM RCM produced dryer climate zones in each region owing to the underestimated precipitation. It produced Csa and Csb zones instead of Cfb in France, Mid-Europe, Eastern Europe and in the Mediterranean and Dsb instead of Dfb in the Alps. Furthermore, the extension of BSk was extremely large in the Iberian Peninsula and in Eastern Europe. Both HIRHAM and ALADIN overestimated the Tundra climate (ET) zone in Scandinavia. In ALADIN simulation the Cfb zone was overestimated in the Iberian Peninsula and in the Italian Peninsula, while it was underestimated in Eastern Europe, in the Mediterranean and on the Western coast of France. The ECHAM5-r3 forced RCMs produced better K-G simulations but the RegCM simulated a wetter climate in the Iberian Peninsula and the Mediterranean, whilst RACMO2 and RCA produced drier climate zones in the Mediterranean and Eastern Europe. Each of them overestimated the ET zone in the Alps.

**Figure 2.** Simulated K-G climate classification according to E-OBS (**A**) and in ALADIN (**B**), HIRHAM (**C**), RegCM (**D**), RAMCMO2 (**E**) and RCA (**F**) without bias correction.

We applied the following bias correction methods: i) the empirical quantile mapping (eQM) of precipitation and temperature [15], ii) quantile mapping of precipitation and temperature based on a gamma + Generalized Pareto Distribution (gpQM) [39], iii) the power transformation of precipitation [40,41] the variance scaling of temperature [23], and iv) the local intensity scaling (LOCI) [42].

The daily mean precipitation and temperature values were used for the bias corrections.
