A novel framework for a multimodel ensemble of GCMs and its application in the analysis of projected extremes

Abstract

Methodologies for evaluating global climate models (GCMs) and generating multimodel ensembles have branched to meet the diverse needs of impact assessment studies. However, there has been limited focus on their intercomparison. In this context, a comprehensive framework comprising the major strands of GCM ranking and ensemble data generation is proposed in the current study. The framework incorporates error index-based rankings and Bayesian method-based weight distribution. Within the Bayesian analysis, a nonparametric (orthonormal) distribution is introduced in this study with the hypothesis that the efficiency of nonparametric distribution in modelling natural random phenomena will accentuate the efficiency of the ranking and ensemble process. Precipitation and temperature data of 21 GCMs from Coupled Model Intercomparison Project Phase 5 covering India is used to validate the proposed framework. Finally, the ensemble data generated from the framework is used for the analysis of projected extremes and their attribution. Results show that Bayesian framework-based rankings outperform other methods in 87.5% of instances. In the case of precipitation, the orthonormal distribution-based Bayesian ranking produces better results for 85.2% of India, while it produces the lowest error ensemble for 79% of the country. The data from the framework are compared to the widely used mathematical average of GCMs. It is found that for precipitation and maximum temperature, the weighted ensemble has closer proximity to the distributive properties of the observed data for the entire study area. Furthermore, uncertainty analysis shows that the ensemble data have minimum uncertainty for the entire area in the case of precipitation. Finally, the assessment of projected extremes shows low to medium confidence (~27%) in the attribution of precipitation extremes to anthropogenic causes under representative concentration pathway (RCP) 4.5, while under RCP 8.5, the confidence is around 80%. For temperature, it is greater than 90% in both scenarios.