Reliable assessment of precipitation extremes is essential for managing hydro-climatic risks in any region. Yet achieving such reliability remains challenging as considerable uncertainty persists among CMIP6 Global Climate Models (GCMs) due to structural differences and biases in simulating regional precipitation behaviour. This study evaluates the performance of thirteen statistically downscaled and bias-corrected CMIP6 models in reproducing eight ETCCDI precipitation indices over the Kosi River Basin. Model skill was assessed using eight statistical indicators with objective criterion weights derived through the Criteria Importance Through Inter-criteria Correlation (CRITIC) method to minimise subjective bias. Four multi criteria decision making (MCDM) approaches TOPSIS, VIKOR, EDAS, and PROMETHEE-II were then applied to generate integrated model rankings and identify optimal ensemble subsets. Findings indicate that MPI-ESM1-2-HR, INM-CM5-0, and BCC-CSM2-MR consistently outperformed others while ACCESS-CM2 and NorESM2 variants showed weaker agreement. Among the ensemble configurations, the eight-member ensemble (AMME8) provided the best balance of accuracy and uncertainty reduction closely replicating the observed inter-relationships among precipitation extremes and achieving the optimal symmetric uncertainty. Future projections using the AMME8 ensemble indicate a marked intensification of precipitation extremes under both SSP245 and SSP585. The far future (2061–2100), particularly under SSP585 shows the strongest amplification with increases of up to 47% in annual precipitation, 60% in heavy rainfall days, and nearly 79% in extremely wet days, suggesting heightened flood risk across the basin. These findings underscore the importance of index specific model evaluation and optimal ensemble formulation for generating reliable regional climate projections to support water resource planning and climate adaptation strategies.
Singh et al. (Sun,) studied this question.