Assessment of CMIP6 models’ performance for South America: multilevel perspective

Abstract

The selection of global climate models (GCMs) that represent the main features of the present climate is crucial for reducing uncertainties when assessing climate trends or driving regional climate models. Based on this premise, the main goal of this study is to assess the skill of 17 GCMs from the Coupled Model Intercomparison Project - Phase 6 (CMIP6) in representing basic features of the present climate (1979–2014). As the analyses are performed for different atmospheric variables and vertical levels, the study is considered multilevel. Different statistical metrics are used to compare the simulations with ERA5 reanalysis focusing on 5 subdomains of South America (SA): northern Amazon Basin (NAMZ), southern Amazon Basin (SAMZ), Northeast region of Brazil (NEB), South American Monsoon (SAM) and La Plata Basin (LPB). Spatial and temporal performance is synthesized through a comprehensive rating index, which integrates multiple metrics into a single performance score, enabling a consistent comparison and overall ranking of model skill across regions and variables. Among the results, model performance varies across regions and variables. Models such as CNRM-CM6-1, GFDL-ESM4, MPI-ESM1-2-HR, and NorESM2-MM stand out for their superior performance across multiple variables and regions, whereas INM-CM4-8, KIOST, and NESM3 show limitations, particularly in simulating upper-level circulation and precipitation processes. Therefore, the main conclusion is that models should be chosen based on their performance in the region of interest. Seventeen CMIP6 GCMs were evaluated over SA using multiple variables and atmospheric levels. Model performance varies strongly by region, with greater dispersion in tropical sub-domains and higher consistency in subtropical regions. CNRM-CM6-1, GFDL-ESM4, MPI-ESM1-2-HR, and NorESM2-MM, consistently outperform other models across variables and regions. The CRI framework integrates multiple metrics, enabling reliable regional model ranking. Best-ranked models better reproduce the observed monthly precipitation climatology and variability. The proposed ranking provides a consistent basis for selecting boundary conditions for dynamical downscaling over SA.