An evaluation of the CMIP3 and CMIP5 simulations in their skill of simulating the spatial structure of SST variability
- 588 Downloads
The natural sea surface temperature (SST) variability in the global oceans is evaluated in simulations of the Climate Model Intercomparison Project Phase 3 (CMIP3) and CMIP5 models. In this evaluation, we examine how well the spatial structure of the SST variability matches between the observations and simulations on the basis of their leading empirical orthogonal functions-modes. Here we focus on the high-pass filter monthly mean time scales and the longer 5 years running mean time scales. We will compare the models and observations against simple null hypotheses, such as isotropic diffusion (red noise) or a slab ocean model, to illustrate the models skill in simulating realistic patterns of variability. Some models show good skill in simulating the observed spatial structure of the SST variability in the tropical domains and less so in the extra-tropical domains. However, most models show substantial deviations from the observations and from each other in most domains and particularly in the North Atlantic and Southern Ocean on the longer (5 years running mean) time scale. In many cases the simple spatial red noise null hypothesis is closer to the observed structure than most models, despite the fact that the observed SST variability shows significant deviations from this simple spatial red noise null hypothesis. The CMIP models tend to largely overestimate the effective spatial number degrees of freedom and simulate too strongly localized patterns of SST variability at the wrong locations with structures that are different from the observed. However, the CMIP5 ensemble shows some improvement over the CMIP3 ensemble, mostly in the tropical domains. Further, the spatial structure of the SST modes of the CMIP3 and CMIP5 super ensemble is more realistic than any single model, if the relative explained variances of these modes are scaled by the observed eigenvalues.
KeywordsCMIP Climate variability Model evaluation Eigenvalue projection
We like to thank Tobias Bayr, Johanna Baehr, Katja Lorbacher and Timofej Woyzichowzki for fruitful discussions and comments. The comments of two anonymous referees have helped to improve the presentation of this study substantially. The ARC Centre of Excellence in Climate System Science (CE110001028) and the Deutsche Forschung Gemeinschaft (DO1038/5-1) supported this study. The slab ocean model simulations were computed on the National Computational Infrastructure in Canberra.
- Jolliffe I (2002) Principal component analysis, 2nd edn. Springer, New YorkGoogle Scholar
- Santer BD, Taylor KE, Gleckler PJ, Bonfils C, Barnett TP, Pierce DW, Wigley TML, Mears C, Wentz FJ, Brüggemann W, Gillett NP, Klein SA, Solomon S, Stott PA, Wehner MF (2009) Incorporating model quality information in climate change detection and attribution studies. Proc Natl Acad Sci USA 106:14778–14783CrossRefGoogle Scholar