Regional and seasonal mean inter-comparison across CMIP3 and CMIP5 climate model ensembles

We examine the hypothesis that the new generation of climate models, the Climate Modeling Intercomparison Project version 5 (CMIP5) suites of models, improves over the previous generation, CMIP3 suites. The bases of the hypothesis are that higher grid resolutions, more sophisticated physics, or comprehensive earth system model components, may enhance finer scale projections. The ensemble of models is evaluated in terms of their ability to simulate precipitation, near-surface air temperature, and wind speed at regional and seasonal scales. The metrics considered for the evaluation include past performance skills, quantified here through bias maps, as well as model agreement, assessed through inter-model comparisons. The bias maps use the Global Precipitation Climatology Project (GPCP) for precipitation and NCEP-II for temperature and wind speed as reference data. Other metrics considered in the study include meridional profiles, multi-model bounds, and mean square errors or differences. Historical performance and future model agreements are based on multi-model ensemble statistics as well as pairwise comparisons of model versions in CMIP3 and CMIP5. To compare model projections in the future, greenhouse gas emissions scenarios and concentration pathways in CMIP3 and CMIP5 are selected to enable effective comparisons of projections across the model versions. While the models and scenarios across CMIP3 and CMIP5 may not always be directly comparable, our premise is that projections still need to be evaluated across versions to understand the science, improve predictions, and inform adaptations. Here we employ 11 carefully selected model-pairs, multi-model ensembles, and emissions scenarios across the two model generations. For future projections, of particular interest are regions and seasons where the sign of the change differs within or across model generations, as well as any changes in multi-model ensembles. Preliminary results suggest that newer generation global circulation models do not appear to improve significantly either in terms of skills (past performance) or consensus (model agreement). However, in certain cases, the possibility of solidifying important insights may be suggested. In addition, regional and seasonal differences appear to persist between model pairs across generations, for example, over regions such as South America and parts of Africa and Asia.