Statistical Methodology for Evaluating Process-Based Climate Models

In climatology, there are mainly two types of models used, that is, global circulation/climate models (GCMs) and regional climate models (RCMs). GCMs can be run for the whole globe, while RCMs can be run only for a part of the globe. In this chapter, we provided a general statistical methodology for evaluating process-based (GCM or RCM) climate models. To bridge observed and simulated data sets, statistical bias correction was implemented. A meta-analysis technique is used for selecting a model or scenarios, which have good performance compared to others. For model selection and ensemble projection, Bayesian model averaging (BMA) is used. Posterior inclusion probability (PIP) is used as model selection criterion. Our analysis concluded with a list of best models for maximum, minimum temperature, and precipitation where the rank of the selected models is not the same for the listed three variables. The outputs of BMA closely followed the pattern of observed data; however, it underestimated the variability. To overcome this issue, 90% prediction interval was calculated, and it showed that almost all the observed data are within these intervals. The results of Taylor diagram show that the BMA projected data are better than the individual GCMs' outputs.