Shape Mismatch Between Input Batch and Predicted Output

I noticed that when running Aurora with ERA5 at 0.25° (but also with HRES T0 at 0.1°), the shape of the input batch tensors is 721×1440 (or 1801×3600), while the shape of the predicted outputs is 720×1440 (or 1800×3600). It seems that the South Pole is missing, as the metadata containing the latitude coordinates does not include the latitude of the North Pole (lat = -90°). Perhaps I’m missing something, but I wonder: in what kind of grid should we consider the model’s predictions to be represented if we want to validate the prediction? Is it a grid that is staggered relative to the one specified in the latitude and longitude metadata? Thanks

No file or test is named. Reproduce the ERA5 0.25° and HRES T0 0.1° cases, compare input and output tensor shapes with the latitude metadata, and establish which grid represents predictions for validation; done means the pole omission and alignment are explained.