Evaluating global climate models for the Pacific island region

Abstract

While the practice of reporting multi-model ensemble climate projections is well established, there is much debate regarding the most appropriate methods of evaluating model performance, for the purpose of eliminating and/or weighting models based on skill. The CMIP3 model evaluation undertaken by the Pacific Climate Change Science Program (PCCSP) is presented here. This includes a quantitative assessment of the ability of the models to simulate 3 climate variables: (1) surface air temperature, (2) precipitation and (3) surface wind); 3 climate features: (4) the South Pacific Convergence Zone, (5) the Intertropical Convergence Zone and (6) the West Pacific Monsoon; as well as (7) the El Niño Southern Oscillation, (8) spurious model drift and (9) the long term warming signal. For each of 1 to 9, it is difficult to identify a clearly superior subset of models, but it is generally possible to isolate particularly poor performing models. Based on this analysis, we recommend that the following models be eliminated from the multi-model ensemble, for the purposes of calculating PCCSP climate projections: INM-CM3.0, PCM and GISS-EH (consistently poor performance on 1 to 9); INGV-SXG (strong model drift); GISS-AOM and GISS-ER (poor ENSO simulation, which was considered a critical aspect of the tropical Pacific climate). Since there are relatively few studies in the peer reviewed literature that have attempted to combine metrics of model performance pertaining to such a wide variety of climate processes and phenomena, we propose that the approach of the PCCSP could be adapted to any region and set of climate model simulations.