More accurate quantification of model-to-model agreement in externally forced climatic responses over the coming century

Abstract:

Separating model-to-model differences in the forced response (U_MD ) from internal variability (U_IV ) in causing uncertainty in future projections of key climatic variables is important, but challenging. Quantifying U_MD is necessary for assessing confidence in future projections, while estimating the magnitude of U_IV is crucial to understand the range of possible futures that might occur. Separating these uncertainties is limited in the traditional use of multi-model intercomparisons because many models only have a small number of realizations. Part of the problem arises because under the same forcing scenario projections from two different model runs can differ because of internal variability as well as differences in the models forced response to external forcing. The other part of the problem arises because climate models are not all independent, with many models sharing components and code. This reduces the effective number of degrees of freedom of an ensemble containing such models, which can in turn lead to both an overestimate of the statistical significance of projected changes and overconfidence.

By utilizing a new archive of six largely independent single model initial-condition large ensembles (SMILEs) we can simply separate the contribution of U_MD and U_IV in the response to strong forcing (RCP8.5) for projections of temperature (ΔT), precipitation (ΔP) and for the first time their temporal variability (ΔT_var & ΔP_var). We present a new method for analyzing multi-model ensembles, which takes into account their lack of independence, and show that this gives similar results to the SMILEs. We confirm that U_MD is larger than U_IV for ΔT and ΔP across the globe. For ΔT_var and ΔP_var, U_IV dominates between 20^o and 80^o latitude in both hemispheres. We present global maps indicating large regions of agreement in the sign of the change, but often disagreement in the magnitude. For example Europe warms in all SMILEs, but the magnitude of the warming can vary by 4^oC. In the tropical Pacific ΔT_var and ΔP_var, we find large model-to-model differences in the sign of the forced change itself, a result contrasting with previous assessments.