Hi all,

Currently, the normalisation stats for variables are stored in a single torch file.
As by design, they are limited to the variables used by ArchesWeather. However, it would be nice to support all variables (on all pressure levels).

I would suggest to provide the stats in a netcdf file. In this way, the statistics can be load similar to the era5 data, with the same indexing strategy and the potential user does not have to create new torch files.