Weighting of model results for improving best estimates of climate change

摘要

Climate projections from multi-model ensembles are commonly represented by the multi-model mean (MMM) climate change. As an alternative, various subjectively formulated schemes for performance-based weighting of models have been proposed. Here, a more objective framework for model weighting is developed. A key ingredient of this scheme is a calibration step quantifying the relationship between intermodel similarity in observable climate and intermodel similarity in simulated climate change. Models that simulate the observable climate better are only given higher weight where and when such an intermodel relationship is found, and the difference in weight between better and worse performing models increases with the strength of this relationship. The method is applied to projections of temperature change from the Third Coupled Model Intercomparison Project. First, cross-validation is used to estimate the potential of the method to improve the accuracy of climate change estimates and to search for suitable predictor variables. The decrease in cross-validation error allowed by the weighting is relatively modest but not negligible, and it could potentially be increased if better predictor variables were found. Second, observations are used to weight the models, to study the differences between the weighted mean and MMM estimates of twenty-first century temperature change and the sensitivity of these differences to the predictor variables and observational data sets used.