资源环境科技发展态势分析平台

We propose a framework that enables the evaluation of a large number of climate models by numerous performance metrics, which can be customized toward a specific impact assessment perspective under climate change (e.g., agriculture, flood control, or else). The customization is performed by weighting the performance metrics. Three criteria are applied to combine a set of diagnostics for creating a single performance index, namely, summation of rank (SR), Euclidean distance of the cluster analysis (CA), and that of Empirical Orthogonal Function analysis (EOF). These indices are then used to objectively select optimal ensemble subsets by applying a culling method. The model evaluation and multimodel ensemble selection in the Indochina Region as a study area are performed on precipitation for two cases: a nonweighted case applying equal weights for all 36 metrics, and a weighted case focusing on the evaluation for agricultural drought monitoring, as an example, with and without model independence and skill weights. We demonstrate that the optimal ensemble subsets of this framework improve significantly the distribution of monthly precipitation data compared to those of the best single model or the full model ensemble during the historical period. The optimal ensemble subsets of CA and EOF criteria are improved more than those of the SR criterion. The performance of the optimal ensemble subsets is also confirmed in the future projection for the RCP8.5 scenario by implementing model-as-truth experiments. A simple and user-friendly decision graph of all model members for the ensemble selection is developed, and its usefulness is demonstrated.