Comparisons Among Multi-model Ensemble Based on Different Ensemble Methods and Ensemble Member Sizes

Abstract

In this chapter, the results of three different multi-model ensemble mean methods are compared. In Sect. 3.1, the global mean surface air temperature and precipitation from multi-model weighted mean and equal-weighted mean are compared. The weight is given according to the correlation coefficients and relative bias between model and observation. The results from the multi-model weighted mean and the equal-weighted mean show no significant difference. In Sect. 3.2, the surface air temperature and precipitation in China from multi-model equal-weighted mean and Bayesian model averaging (BMA) method are compared. BMA is a multi-model ensemble method which considered each model’s posterior distributions and then weighted by posterior model probability. In our calculation, the training of BMA is based on observational data (derived from CRUTS 3.21) for the period of 1961–1990. The multi-model ensemble mean in Sects. 3.1 and 3.2 is calculated based on 22 CMIP5 models described in Chap. 1.