Currently, the fallback on ERA5 climatological data relies on the files stored in suppor material, which have been produced to handle the missing data for EERIE when inferring with NeuralGCM. These data are a single representative day for 2010. This may have unexpected consequences during the inference.
To overcome this issue, different approach are possible:
- A dynamic retrieve from ERA5 zarr store in google for the year analyzed (if amip historical simulation)
- The creation of a more comprehensive year with ERA5 zarr store in google of monthly data from Copernicus CDS.
- ...
Currently, the fallback on ERA5 climatological data relies on the files stored in suppor material, which have been produced to handle the missing data for EERIE when inferring with NeuralGCM. These data are a single representative day for 2010. This may have unexpected consequences during the inference.
To overcome this issue, different approach are possible: