资源环境科技发展态势分析平台

This paper introduces and applies a new method to consistently estimate internal climate variability for all models within a multimodel ensemble. The method regresses each model's estimate of internal variability from the preindustrial control simulation on the variability derived from a model's ensemble simulations, thus providing practical evidence of the quasi-ergodic assumption. The method allows one to test in a multimodel consensus view how the internal variability of a variable changes for different forcing scenarios. Applying the method to the CMIP5 model ensemble shows that the internal variability of global-mean surface air temperature remains largely unchanged for historical simulations and might decrease for future simulations with a large CO2 forcing. Regionally, the projected changes reveal likely increases in temperature variability in the tropics, subtropics, and polar regions, and extremely likely decreases in midlatitudes. Applying the method to sea ice volume and area shows that their respective internal variability likely or extremely likely decreases proportionally to their mean state, except for Arctic sea ice area, which shows no consistent change across models. For the evaluation of CMIP5 simulations of Arctic and Antarctic sea ice, the method confirms that internal variability can explain most of the models' deviation from observed trends but often not the models' deviation from the observed mean states. The new method benefits from a large number of models and long preindustrial control simulations, but it requires only a small number of ensemble simulations. The method allows for consistent consideration of internal variability in multimodel studies and thus fosters understanding of the role of internal variability in a changing climate.