OpenMeteo has started a public dataset (see https://github.com/open-meteo/open-data) that is archiving multiple different providers of weather data forecasts. The archive only goes back to December 2023, so currently probably isn't super helpful for training, but would be interesting to have support as the archive gets larger, and we might want to finetune models or try different initializations for ensembles, like in #85.

Context

Being able to use multiple different NWPs as comparisons/finetuning/initializations could help a lot in seeing how these models compare to other, more physics-based simulations. They also include models that have much higher resolutions than ERA5, or for limited areas, even HRES/ENS. This then relates to being able to use models with adaptive meshes or with a nested high resolution area inside the global model (#78, #3).

Possible Implementation

Similar to #86 except the data is not in Zarr, but a format more suited to site-level forecasts. Would want to get global or local grids of data from it, so would require some reshaping of the data.