Diagnosing Mismatches between Simulations and Observations in Data-Model Comparisons using the CMIP5/PMIP3 Simulations

Abstract:

“Data-model comparisons” using paleo data and simulations can provide strong out-of-sample tests of the ability of climate models to simulate climates other than those of the present. Data-model comparisons have shown that such large-scale patterns of simulated future climates (like enhanced land/ocean and latitudinal contrasts, and changes in seasonality) are also present in paleoclimate simulations, and critically, in paleo observations, indicating that those patterns are robust features of climate change, and are not simply model artifacts. Data-model comparisons have also shown that there are mismatches between simulations and observations on regional scales that have persisted across generations of models. The source of the mismatches in individual models are often inferred through simple data analyses and visualizations, and the subjective application of our conceptual models of “how climate works.” The availability of multiple simulations, including ensembles, both within and among models, and across generations of models, allow more objective statistical approaches to be applied to test hypotheses about the specific mechanisms (e.g. atmospheric circulation, surface water-and-energy balances) that contribute to the mismatches. These approaches range from familiar methods for characterizing multivariate data sets to newer approaches that attempt to infer cause-and-effect relationships among such data.