Evaluating the enhanced benefits of multi-model ensemble mean from NEX-GDDP-CMIP6 versus native CMIP6 Models in accurately representing historical temperature patterns in South America

We assessed the additional value of the NEX-GDDP-CMIP6 (NEX ensemble) temperature models compared with that of their original CMIP6 models across seven climatic sub-regions in South America from 1950 to 2014, aiming to identify the best-performing models with the smallest biases. Using ERA5 data, the 22 NEX ensemble temperature models and an equivalent number of native CMIP6 models were evaluated. All models and reanalysis data were remapped to a standard latitude-longitude grid with a spatial resolution of 0.25° × 0.25° using bilinear interpolation. Model performance was evaluated at seasonal and annual timescales, with metrics computed separately for each aggregation period and then averaged regionally. We applied six complementary performance metrics—mean bias, correlation coefficient, mean absolute error, root mean square error, modified Kling–Gupta efficiency, and Taylor skill score—to quantify accuracy and consistency. The comparison between NEX and CMIP6 models was conducted at the ensemble mean level to identify systematic differences in skill. The NEX ensemble model consistently demonstrates a tendency to overestimate temperatures, particularly within the NWS, NSA, SAM, and SES sub-regions, with observed biases reaching up to 1.84 ± 2.3°C. Contrastingly, the 22 native CMIP6 model showed mixed biases, exhibiting significant underestimations in the NES and SSA. Statistical comparisons using a two-tailed t-test (p < 0.05) demonstrate that improvements in the NEX ensemble are significant in various high-altitude regions. The NEX model exhibits enhanced performance in estimating annual temperature in high-altitude areas with average MAE reductions of 4.50 ± 3.1°C, whereas the native CMIP6 ensemble model displays greater accuracy in select low-altitude regions with MAE reductions of 3.70 ± 2.8°C. Our findings emphasize the need to improve future climate projections derived from both the NEX ensemble and the native CMIP6 datasets. It will help identify temperature biases in the models and evaluate their performance across various regions. Such insights are essential for conducting climate change research, including drought forecasts, forest fire evaluations, and other studies where temperature plays a critical role.