ERA5-PrecipSR

This dataset is a regional NetCDF dataset designed for precipitation super-resolution and spatial downscaling tasks. It is constructed based on the ERA5 hourly data on single levels from 1940 to present provided by the Climate Data Store.
The dataset is organized in a one-sample-per-file format and has been split into training, validation, and test sets, containing 17,892, 2,237, and 2,236 samples respectively. Each sample corresponds to a multivariate meteorological field at a specific time and over a specific region, including high-resolution total precipitation (tp), surface pressure (sp), 2-meter air temperature (t2m), relative humidity (rh), wind speed (wind), land-sea mask (land_mask), as well as the corresponding low- and high-resolution latitude and longitude coordinates.
Notably, low-resolution precipitation is not stored as an independent variable, but is dynamically downsampled from high-resolution precipitation. This design reduces data redundancy while preserving the core variables required for conducting physics-informed precipitation reconstruction research.