Performance Evaluation of Cordex-sea Dataset Based on Multi-metrics and Weighted Ensemble Mean

Abstract

AbstractComprehensive performance evaluation of climate models on historical periods is crucial in achieving more accurate climate projections. This study aimed to evaluate the precipitation performance of nine regional climate models in the Coordinated Regional Climate Downscaling Experiment–Southeast Asia (CORDEX-SEA) dataset on spatiotemporal aspects and proposed schemes of implementing weighting factors to gain an ensemble mean with a better performance. Multi-metrics were used to measure the model skills. In the IPSL and GFDL regional climate models, extreme values in the category of errors were found. They uniquely occurred only at the lon_max boundary and only in certain months. The causes of such errors were further investigated. Hence, the maximum value screening must be carried out in the early model evaluation stage. Such errors may not be visible if the evaluation is only based on the mean value approach. Based on the Taylor diagram CNRM has the highest performance, followed by HadGEM2 and NorESM1, and the ensemble mean outperforms all those individual models. The implementation of the weighting factors shows that the weighted ensemble means produced better performances in terms of standard deviation ratio (0.98 to 1.19) than the unweighted ensemble mean (1.20). For the zonal mean, the weighted ensemble means (0.95) also outperforms other models (< 0.90). Even though the wavelet analysis indicates that all models and those ensemble means have deficient performances, especially in capturing interannual-to-decadal variability, the Fast Fourier Transform (FFT) analysis shows different results. In addition, the effect of bias correction is also confirmed.