Regional models not supported

[sehoffmann]

Currently its not easily possible to evaluate a regional model without duplicating the ERA5 ground truth data set and slicing it accordingly.

The reason for this is the check happening at https://github.com/google-research/weatherbench2/blob/e6b1b9068ffdb9b07d6e315ceaef2de605aa20cc/weatherbench2/evaluation.py#L129C14-L129C34.

This check happens BEFORE(!) slicing the data according to the region config causing regional models to fail (if evaluated against a ground truth data set defined on the whole globe).

Please consider moving this sanity check after _impose_data_selection has been called, i.e. after

weatherbench2/weatherbench2/evaluation.py

Line 326 in e6b1b90

forecast_all_times = _impose_data_selection(

This trivial change would make it possible to support regional models without any additional changes if I am not mistaken.