I’m trying to use the existing data_loader_hrrr_era5 (as referenced in examples/generative/stormcast/datasets/data_loader_hrrr_era5.py) to train the regression model (e.g. via regression.train). The documentation describes the folder layout under <location>—with separate era5/ and <hrrr_dataset_name>/ directories, each containing per-year .zarr files inside their respective train/, valid/, and test/ subdirectories. However, it’s not clear:
Specifically, I’d like to know:
For ERA5 per-year .zarr files, how the variables should be structured (dimensions, coordinate names, etc.).
For HRRR per-year .zarr , what the internal layout must look like and how the variables should be structured (dimensions, coordinate names, etc.).
Without knowing the exact schema, it’s difficult to build a custom Zarr export that the loader can consume. Any examples, minimal specs, or references to how the original datasets were organized (paths + internal variable names) would be extremely helpful. Thanks in advance for any pointers!