Evaluation of Optimized Multi-Model Ensembles for Extreme Precipitation Projection Considering Various Objective Functions

This study evaluated an optimized multi-model ensemble (MME) approach coupled with various objective functions to project the extreme precipitation. This study considered 60 stations in South Korea and used coupled model intercomparison project (CMIP) 6 GCMs under shared socioeconomic pathway (SSP) scenarios. The extreme precipitation characteristics were described by ten precipitation indices from expert team on climate change detection and indices (ETCCDI). The Generalized Reduced Gradient (GRG) algorithm was utilized to optimize the weights of all considered GCMs using four objective functions, namely Nash-Sutcliffe efficiency (NNSE), root mean square error (RMSE), Kling-Gupta efficiency (KGE), and percent bias (Pbias). The performance of optimized MME in extreme precipitation projection was compared to the equal-weighted MME. As a result, the optimized MME in this study demonstrated higher performances than the MME that simply averaged GCMs’ output in most regions, regardless of the objective functions. These results indicated that the optimized MME with KGE generally performed better than MMEs using other objective functions. This study emphasized the importance of efficient GCMs subset selection, unequal weights, and the use of appropriate evaluation metrics for the MME approach in extreme precipitation projection.