资源环境科技发展态势分析平台

This study examines potential benefits of performance-based multi-model ensembles (MMEs) in projecting the impacts of climate change on extreme precipitation indices over East Asia (EA) using the data from 19 GCMs in the coupled model intercomparison project 5 (CMIP5). The Taylor skill score is adopted as the measure of the model skills in simulating the spatial and interannual variability of the selected extreme precipitation indices over four EA regions. The overall rank based on the total skill score (TSC) is used to construct two skill-based MMEs, MME of high-skill, MMH (MME of low-skill, MML) that include the top (bottom) seven models, in addition to the simple ensemble of all 19 GCMs (ENS). Inter-GCM consistency is measured using the signal-to-noise ratio (SNR). In the present-day period, MMH yields higher skill scores than MML and ENS for almost all extreme precipitation indices as well as regions. Regional variations in biases, inter-model consistency, and TSC are large. The inter-model consistency is highest for Northern China and Manchuria and is lowest for Southern China. The most notable differences in the key properties of climate change signals from the three MMEs among the three ensembles are that the climate change signals from MMH and ENS exceed the 90% significance level in much larger areas than those from MML. However, the differences in the climate change signals between MMH and MML are generally below the 90% significance level. The SNR of the projected climate change signals shows that MMH yields more consistent climate change signals than ENS/MML. Both the SNR differences and the area in which statistical significance exceed the 90% level suggest that constructing climate change signals from a group of higher-skill models may yield more reliable projections than constructing MMEs from the entire models or a group of lower-skilled models.