CRA5 a high-fidelity compressed reanalysis atmospheric dataset for weather and climate research

The ERA5 reanalysis dataset, developed by the European Centre for Medium-Range Weather Forecasts (ECMWF), provides high-resolution, hourly global climate and weather data from 1950 to the present. However, its massive volume poses substantial storage and distribution challenges. To address this, we introduce CRA5, a highly compressed version of ERA5 generated by the neural network framework Aeolus. CRA5 reduces the 400 TB uncompressed float32 dataset to just 0.85 TB, achieving a 470× compression ratio. Notably, it offers over 100 times higher compression than the lossless GRIB files from the Copernicus Climate Data Store (CDS). Extensive experiments validate its numerical accuracy: CRA5 maintains consistent climatology and comparable power spectral density, yielding a mean absolute error of only 0.17 K for temperature across 37 vertical levels. Furthermore, it faithfully reconstructs extreme weather events and large-scale climatological patterns. By significantly lowering infrastructure barriers, CRA5 accelerates data access and facilitates broader collaboration in large-scale atmospheric research.