A new method of multi model ensemble to improve the simulation of the geographic distribution of Köppen-Geiger climate classification

Multi-model ensembles (MMEs) have been demonstrated as a useful method for improving the results of models, and previous research found the unweighted ensemble average to be similar to a simple weighted ensemble average. Our goal was to improve the results of MMEs by assigning suitable weights for each model. In this study, the simulation of the geographic distribution of the Köppen-Geiger climate classification was found to be poor using nine general circulation models (GCMs), with the percentage of the inconsistent areas between the GCMs and the observations (CRU) covering approximately 40-52 % of the total land area. Two-thirds of the nine GCMs failed the agreement tests between the GCMs and CRU. We found that the geographic distribution of Köppen-Geiger climate classification was simulated better using a new weighted ensemble average method than by any single GCM or by the unweighted ensemble average. The percentage of inconsistent area between the MMEs and CRU was reduced to about 35 % of the total land area. The agreement between the MMEs and the observation is 0.6340, at a "good" level. The geographic distribution of inconsistent regions between the MMEs and CRU have a banding distribution along the boundaries of two adjacent climate types, which confirms that the boundary regions of two different climate types are simulated poorly by models. Experiments that replaced temperature or precipitation in the MMEs using CRU relevant data suggest that temperature was simulated better than precipitation. Therefore, a new method is proposed for improving the simulation performance of the geographic distribution of climate classification using GCMs.