Performance of CMIP6 GCMs in Representing Extreme Precipitation in Peru (1981–2014)

Extreme climate events, particularly precipitation extremes, pose significant risks to ecosystems, infrastructure, and socio-economic systems globally. In Peru, the diversity of its climate, driven by its complex topography, makes it highly vulnerable to such events, especially in the Andes and Amazon regions. This study evaluates the performance of 25 CMIP6 GCMs in simulating extreme precipitation events during both the wet and dry seasons at the national level. Gridded precipitation data from the PISCO product and CMIP6 model simulations for the period 1981–2014 were used to estimate extreme precipitation indices, including Rx1day, Rx5day, SDII, CDD, CWD, R10mm, and PRCPTOT. Performance was assessed using statistical metrics such as PBIAS, NRMSE, and the Pattern Correlation Coefficient (PCC), integrated through a TOPSIS ranking. Results indicate that NorESM2-MM, MPI-ESM1-2-LR, and CESM2 exhibit the best performance, achieving TOPSIS scores above 0.8. These models show high spatial correlation (PCC frequently >0.8) and relatively low biases. In contrast, models like FGOALS-g3 and CanESM5 show significant limitations, with PBIAS exceeding 80% in Rx1day and Rx5day and TOPSIS scores below 0.5. The ensemble reveals a persistent ‘drizzle bias,’ with wet day frequency (R1mm) generally overestimated by 20–40% in the wet season and by 40–80% during the dry season across most CMIP6 models. Furthermore, indices of temporal persistence (CWD and CDD) remain the most challenging, with CWD overestimations often exceeding 100–200%. These findings highlight the critical need for statistical or dynamical downscaling, together with bias correction, before using CMIP6 projections for local adaptation strategies in the Andes and Amazon regions.