Assessing the reliability of climate models, CMIP5

In spite of the yet incomplete subsample of the 5th phase of the Coupled Model Intercomparison Project (CMIP5) model ensemble to date, evaluation of these models is underway. Novel diagnostics and analysis methods are being utilized in order to explore the skill of particular processes, the degree to which models have improved since CMIP3, and particular features of the hindcasts, decadal and centennial projections. These assessments strongly benefit from the increasing availability of state-of-the-art data sets and model output processing techniques. Also paleo-climate analysis proves to be useful for demonstrating the ability of models to simulate climate conditions that are different from present day. The existence of an increasingly wide ensemble of model simulations re-emphasizes the need to carefully consider the implications of model spread. Disparity between projected results does imply that model uncertainty exists, but not necessarily reflects a true estimate of this uncertainty. Projections generated by models with a similar origin or utilizing parameter perturbation techniques generally show more mutual agreement than models with different development histories. Weighting results from different models is a potentially useful technique to improve projections, if the purpose of the weighting is clearly identified. However, there is yet no consensus in the community on how to best achieve this.
These findings, discussed at the session “Assessing the reliability of climate models: CMIP5” of the World Climate Research Program (WCRP) Open Science Conference (OSC), illustrate the need for comprehensive and coordinated model evaluation and data collection. The role that WCRP can play in this coordination is summarized at the end of this paper.