A Comparison of the Performance of the Weighted Ensembles Means in CORDEX-SEA Precipitation Simulations

Abstract

Numerous studies stated that the performance of ensemble mean derived from multiple climate models generally surpassed the individual member model, and applying weighting factors potentially increase the ensemble mean of performance. This study aims to assess the performance of unweighted and weighted ensemble means of 9-modelled precipitation datasets in the CORDEX-SEA multi-model simulations for 1981-2005. The 9 datasets included: CNRM_a, ECE_b, GFDL_b, IPSL_b, HadGEM2_a, HadGEM2_c, HadGEM2_d, MPI_c, and NorESM1_d. The weighting factors were derived from the models' skill scores measured using five statistical-based metrics, namely Taylor, Pierce (SS), Tian skill score (Tian), Climate prediction index (CPI), and Performance and Independence (PI). The ERA5 and GPCP precipitation datasets were used as the references for comparison. Then, reliable metrics will be used to determine the weighting factor. The results found that three metrics namely Taylor, SS, and Tian were more reliable than the other two metrics (CPI and PI). Spatially, the weighted ensemble mean based on a random method was superior to other ensemble mean methods and individual models. We found that the CNRM_a and GFDL_b models were spatially performed best. In contrast, most the ensemble means was temporally less performed compared to the individual model. Our findings suggested that by removal of low performance models will significantly influence on the overall ensemble model performance. Further, the research may provide valuable considerations of climate models selection for climate projection assessments, especially in the Southeast Asia region.