Towards an Objective Characterization of Climate Model Performance

This study is based on previous work where we measured the performance of models in terms of their ability to simulate the observed climate mean state for a wide range of quantities. We have shown that the mean of a multi-model ensemble consistently outperforms any individual simulation. Now an important question is how to construct an optimally weighted multi-model mean which maximizes the strengths while minimizing the weaknesses discovered in the model conglomerate. Consequently, we explore ways to reduce our rather comprehensive choice of climate quantities into a much smaller subset. Our goal is to derive an unbiased description of model performance retaining a significant proportion of information while neglecting a considerable amount of data redundancy. Statistical methods as diverse as cluster analysis and principal component analysis are shown to be successful in producing a minimal collection of climate quantities which are distinctly useful in model evaluation. To the first order, this subset consists of two variables: one primarily representing model physics, while the second mainly represents model dynamics. We apply these results to the IPCC-AR4 ensemble and demonstrate how it can be used to construct an optimally weighted average of many models.