Climate projections for Greece: Defining a regional sub-ensemble from the CMIP6 landscape

Abstract Most climate change impact studies, regardless of scope, traditionally rely on a predefined set of climate model simulations without thoroughly examining representativeness, model skill, and diversity. This approach risks overlooking regional nuances and limits the utility of projections for tailored adaptation strategies. In the Mediterranean—and particularly Greece, where climate risks are high—addressing these limitations is essential for reliable, actionable projections. The CMIP6 ensemble is extensive, but its size and internal variability pose challenges for regional use, leaving users to navigate an “ensemble of opportunity” with interdependent models and diverse historical and future behaviors. Here we evaluate 35 CMIP6 models over Greece against bias-adjusted GSWP3-W5E5 observations, assessing both annual and seasonal historical performance with multiple diagnostics (correlation, standard deviation, CRMSE, bias, RMSE) and summarizing skill via a composite Historical Performance Score (HPS): the harmonic mean of Taylor Skill Score (pattern fidelity) and a variability-aware bias score that penalizes systematic offsets relative to observed interannual variability. Future responses are analyzed for 2081–2100 (high-emission Shared Socioeconomic Pathway SSP5-8.5) using a quadrant framework based on temperature change ( $$\Delta$$ ) and late-century precipitation (); changes in maximum temperature ( $$\Delta$$ ) are also incorporated to characterize the amplification of hot conditions. By integrating model performance and ensemble spread, the methodology refines model selection to balance historical credibility with diversity of future outcomes, enabling compact regional sub-ensembles that capture the range from moderate to severe warming and from drier to wetter states. Results show that historical skill does not necessarily translate into capturing future extremes in warming or drying, but a carefully chosen sub-ensemble can maximize the range of $$\Delta$$ and outcomes while retaining high HPS. The EURO-CORDEX CMIP6 driving subset reflects central tendencies and underrepresents warmer-wetter boundary futures; at the same time, roughly half of its driving GCMs rank among the strongest historical performers for Greece by HPS, while the remainder are comparatively weak. These findings underscore the need for tailored, performance-and-spread-aware selection strategies to navigate complex multi-model ensembles in regional climate studies.