Estimating democratic averages across GCMs, RCMs, and scenarios in the CMIP5-based EURO-CORDEX simulation ensemble

The EURO-CORDEX CMIP5-based model simulation ensemble is very large, with 8 GCMs being downscaled by 11 RCMs for three different RCP scenarios, resulting in a total of 136 simulated GCM-RCM-Scenario combinations at the time of this work. Still, the GCM-RCM-Scenario 3-dimensional combination matrix is not at all complete.

With a simple linear ANOVA-based technique some missing simulations can be emulated simply by taking appropriate combinations of linear terms calculated from the matrix of existing simulations. This means that approximate ensemble averages can be calculated, where all GCMs as well as all RCMs have the same weight in the averaging.

We aim at emulating values of various fields for combinations, which have not actually been performed, in order to get a more “democratic” picture
of ensemble means. We will show results for a number of standard variables: Seasonal means of temperature, precipitation, and wind strength as well as some extremes. Also, a complete sub-matrix has been systematically analysed with successive numbers of simulations being left out, in a bootstrap procedure, to estimate variability and errors of ensemble members as well as averages using the ANOVA technique as compared to the complete sub-matrix.

We will demonstrate use and necessary adaptations of this technique for filling the entire EURO-CORDEX matrix based on large sub-sets of all existing simulations.