AI-based Downscaling of ERA5 with cGAN at 2.2 km over Italy

The daily high-resolution dataset for Italy, referred to as FAIR (Fast AI Reanalysis), is derived by AI-based downscaling of ERA5 reanalysis data, originally available at ≈31 km x 31 km horizontal resolution, to 2.2 km x 2.2 km. FAIR is released with a 6-day delay, corresponding to the release of ERA5 data on Copernicus. The downscaling is performed using a Conditional Generative Adversarial Neural Network (cGAN), developed by Manco I. et al. (2025). The dataset includes 2m-temperature fields and total accumulated precipitation. The high-resolution temperature data offer valuable spatial detail for investigating urban heat island effects and heatwaves, while the precipitation data are particularly useful for monitoring heavy rainfall and flood events. The cGAN model was trained on the 1990–2000 period and covers the entire Italian territory. The temporal coverage of the dataset is from 01/01/2025, with a daily temporal resolution. Whenever you publish research or applications based on this dataset, please include the following citation: Manco, I., Riviera, W., Zanetti, A., Briscolini, M., Mercogliano, P., & Navarra, A. (2025). A new conditional generative adversarial neural network approach for statistical downscaling of the ERA5 reanalysis over the Italian Peninsula. Environmental Modelling & Software, 188, 106427.