CMIP6 model ranking and machine learning–assisted ensemble evaluation using a multi-variable integrated MCDM–GDM framework

ABSTRACT This study proposes a Multi-Variable Integrated MCDM–GDM framework to evaluate and rank CMIP6 GCMs for precipitation (Pr), atmospheric maximum and minimum temperatures (Tasmax and Tasmin). The framework is demonstrated for the Upper Mahanadi River Basin, India, using 0.25° gridded IMD observations. In the proposed framework twenty-three (Pr), twenty-five (Tasmax), and twenty-four (Tasmin) bias-corrected CMIP6 models were assessed against several statistical performance indicators, such as mean error, bias, variability, and extremes, with AHP weights prioritising precipitation. Six complementary MCDM techniques were applied grid-wise and aggregated through a Group Decision-Making consensus to identify the best-performing CMIP6 models. On this basis, FGOALS-g3, MPI-ESM1-2-LR, MIROC-ES2L, KIOST-ESM, and NESM3 were identified as the top five CMIP6 models. Mean ensembles built from these five models achieved strong correlation (R2 = 0.78, 0.77, and 0.84 for Pr, Tasmax, and Tasmin), whereas machine-learning ensembles yielded only incremental gains (ΔR2 = 0.005 to 0.013). These findings confirm that the proposed framework establishes a more reliable selection of CMIP6 models than single-variable approaches, and that it is replicable and can be effectively applied to flood prediction and hydrological applications in the study region and other monsoon-dominated river basins.