CorrDiff

Downscales ERA5 reanalysis to kilometre-scale fields over Europe.

Field
Weather & climate
Method
Deep learning
Data it takes
gridded-fields
License
Apache-2.0

CorrDiff is a generative downscaling model trained against the COSMO-REA6 and REA2 reanalyses. It takes coarse ERA5 fields and produces fields at kilometre scale.

The small-scale structure is generated, so each run is one plausible realization consistent with the coarse input, and repeated samples differ. Treat an individual field as a sample from a distribution.

The training domain is European and the model does not transfer elsewhere without retraining. Weights are fetched at run time.

Upstream project → · Paper →

Catalog entry last checked 2026-09-08. All models →

Run CorrDiff on your data

The lab runs this on its own compute, in a container, with the inputs and parameters recorded alongside the result. Initial scoping conversations are free.

Contact the Lab