The CorrDiff-Mini training dataset of ERA5 (low-resolution) reanalysis and the corresponding HRRR (high-resolution) analysis fields at a resolution of approximately 3 km per pixel. It is intended to be used with the CorrDiff-Mini code in Modulus to enable training a lightweight CorrDiff version for educational and testing purposes and as a baseline for implementing custom versions of CorrDiff.
The data are collected from the period 2018-2021 from the HRRR CONUS domain covering the continental United States and some surrounding regions. The samples are 64x64 pixels in size, with a spatial resolution of approximately 3 km per pixel.
The ERA5 variables, intended to be used as the CorrDiff input, include temperatures (t), geopotential height (z), west-east (u) and south-north (v) winds and specific humidities (q), each as 1000, 850, 500 and 250 hPa pressure levels, as well as the 10-meter winds, 2-meter temperature, total column water vapor, surface pressure and mean sea level pressure. The HRRR variables, intended as the CorrDiff output, include 10-meter winds, 2-meter temperature and total precipitation.
NOAA High-Resolution Rapid Refresh (HRRR) Model was accessed on Aug 2024 from https://registry.opendata.aws/noaa-hrrr-pds. Data is produced by the US Government and licensed as such: https://registry.opendata.aws/noaa-hrrr-pds/
Disclaimer: For each dataset an user elects to use, the user is responsible for checking if the dataset license is fit for the intended purpose.