A New Framework for Estimating and Decomposing the Uncertainty of Climate Projections

Climate projections obtained by running global climate models (GCMs) are subject to multisource uncer-tainties. The existing framework based on analysis of variance (ANOVA) for decomposing such uncertainties is unable to include the interaction effect between GCM and internal climate variability, which ranks only second to the main effect of GCM in significance. In this study, a three-way ANOVA framework is presented, and all main effects and interaction ef-fects are investigated. The results show that, although the overall uncertainty (O) is mainly contributed by main effects, in-teraction effects are considerable. Specifically, in the twenty-first century, the global mean (calculated at the grid-cell level and then averaged, and likewise below) relative contributions of all main effects are 54% for precipitation and 82% for temperature; those of all interaction effects are, respectively, 46% and 18%. As the three-way ANOVA cannot investi-gate the uncertainty components resulting from uncertainty sources, it is improved by deducing the relationship between uncertainty components resulting from uncertainty sources and those resulting from the main effects and interaction effects. By the improved three-way ANOVA, Ois decomposed into uncertainty components resulting from the emission scenario (S), GCM (M), and internal climate variability (V). The results reveal that Ois mainly contributed by M in the twenty-first century for precipitation, and by M before the 2060s whereas by S thereafter for temperature. The robustness of the V characterization is explored by investigating the variation of Von the number of included ensemble members. The extent of the underestima-tion of the V contribution is roughly an average of 4% for precipitation and 1% for temperature.