Comparison of Three Bayesian Approaches to Project Surface Air Temperature Changes over Japan Due to Global Warming

Abstract

We applied three Bayesian approaches with two metrics (bias and trend) to the sensitivity study of the projection of surface air temperature (SAT) changes over Japan for winter and summer due to global warming. In addition, we investigated the differences in projected probability density functions (PDFs) produced by the Bayesian approaches with the two metrics and the arithmetic ensemble mean (AEM) using available simulations from the Coupled Model Inter-comparison Project phase 3. The magnitude of model weights differs among the three Bayesian approaches with the same metrics, although the models with large weights are identical. However, models with large weights differ between the two metrics. Similar projected SAT changes of the models with large weights result in narrower PDFs of SAT changes than those of AEM. The peaks and means of the Bayesian-weighted PDFs are almost identical to those of the AEM's PDF. However, in some cases, 5, 75, and 95 percentiles of the PDFs differ distinctly among the three Bayesian approaches. Therefore, impact researchers should investigate the differences in the PDFs of SAT changes due to Bayesian approaches and metrics.