Climate Extremes in Europe: A Comparative Analysis of Climate Model Datasets

Ongoing climate change is increasing the need for reliable climate information to support adaptation, particularly for climate extremes whose frequency and intensity are projected to rise and cause substantial societal and environmental impacts. Adaptation planning often requires highly localized information, yet global climate models (GCMs) typically operate at coarse spatial resolutions (~100 × 100 km) and have limited skill in representing extremes. To address this, downscaling techniques are widely used to generate higher-resolution climate information. Statistical downscaling links large-scale model output to local observations, while dynamical downscaling employs high-resolution regional climate models driven by GCM boundary conditions. The strengths and limitations of each approach need to be evaluated.

In this study, temperature- and precipitation-related climate extreme indices were computed using multiple publicly available datasets, including global model outputs from CMIP5 and CMIP6, statistically downscaled products (NEX-GDDP-CMIP6 and CIL-GDPCIR), and dynamically downscaled regional simulations from EURO-CORDEX. All datasets were harmonized to a common spatial resolution to enable direct comparison. The analysis covers a historical period (1990-2019) and a future period (2071-2100) under a middle-of-the-road emissions scenario (RCP4.5/SSP2-4.5). Historical simulations were evaluated against a gridded observational dataset (E-OBS) and two reanalysis products (ERA5 and GMFD). Using daily temperature and precipitation data, 17 climate extreme indices were calculated, along with detailed analyses of mean conditions and two representative extremes: annual maximum temperature and maximum 5-day precipitation.

Downscaling generally improves the representation of the European climate compared to global models. Statistically downscaled and bias-corrected datasets perform better for mean and extreme temperature and for mean precipitation, while improvements for precipitation extremes are limited. Dynamically downscaled EURO-CORDEX simulations show systematic regional biases, particularly in Nordic regions, and generally produce higher precipitation extreme indices. No single dataset consistently outperforms others across all regions, with complex terrain and coastal areas remaining challenging. Despite performance differences, all datasets project similar overall trends in climate extremes under warming, although the magnitude and regional patterns vary. Uncertainties in observational and reanalysis datasets, especially for precipitation, further complicate model evaluation. Overall this analysis highlights the need for clearer guidance on dataset selection for adaptation applications.