Predictor Screening, Calibration, and Observational Constraints in Climate Model Ensembles: An Illustration Using Climate Sensitivity

Abstract

limate projections have been remarkably difficult to constrain by comparing the simulated climatological state from different models with observations, in particular for small ensembles with structurally different models. In this study, the relationship between climate sensitivity and different measures of the present-day climatology is investigated. First, it is shown that 1) a variable proposed earlier that is based on interannual variation of seasonal temperature and 2) the seasonal cycle amplitude are unable to constrain the range of climate sensitivity beyond what was initially covered by the ensemble. Second, it is illustrated how model calibration helps to reveal potentially useful relationships but might also complicate the interpretation of multimodel results. As a consequence, when ensembles are small, when models are neither independently developed nor structurally identical, when observations are likely to have been used in the model development and evaluation process, and when the interpretation of the relationships across models in terms of well-understood physical processes is not obvious, care should be taken when using relationships across models to constrain model projections. This study demonstrates the pitfalls that might occur if emergent statistical relationships are prematurely interpreted as an effective constraint on projected global or regional climate change.