The Upper Indus Basin (UIB) plays a crucial role in water security and socio-economic development in Pakistan. Under changing climatic conditions, the sustainable management of the water resources of the UIB needs accurate and reliable projections of hydroclimatic variables. Given the limited assessments of hydroclimatic simulations from CMIP6 models in the region, this study assessed the uncertainties associated with the historical simulations of 16 CMIP6 GCMs in the UIB. The observations of 34 in situ weather stations were used as reference, while the performances of GCMs were assessed based on widely used evaluation indices, including correlation coefficient (CC), bias, relative bias (rBIAS), root mean square error (RMSE), Nash–Sutcliffe efficiency (NSE), Taylor diagram, and the performance diagram. Results of the evaluation indices indicated that most of the considered GCMs failed to represent the observed precipitation in the UIB. Correlations between the simulations of GCMs and the reference observations were generally low; CCs ranged from −0.24 to 0.16. All GCMs exhibited negative NSE values (ranging between −2.79 and −0.51). The values of RMSE (59.36 to 98.43 mm/month) and rBIAS (9 to 96%) were also very high. Among the considered GCMs, INM-CM4-8 and EC-Earth3-Veg-LR showed comparatively lower RMSE values, moderate rBIAS, and higher CC values. Three GCMs (MRI-ESM2-0, GFDL-ESM4, and CNRM-CM6-1) performed very poorly, with high negative NSE and significant overestimations. Among the 16 GCMs, EC-Earth3-Veg-LR had the highest composite score and better performance across all considered indices. The overall findings of this study indicated that none of the CMIP6-based GCMs (in their raw form) performed satisfactorily in the UIB of Pakistan; therefore, the application of bias-correction techniques is strongly recommended before direct application of these projections for climate impact and adaptation studies in this mountainous region. The results will be useful for the hydroclimatic data users and algorithm developers of global climate models.
Khan et al. (2026) studied this question.