Using Aurora model for prediction for a specific location

I am working on adapting Aurora for station-level prediction, where the target corresponds to a single fixed geographic location (one longitude/latitude pair).

Since Aurora applies spatial patching in the encoder, I was wondering:

• Is it supported to run the model using inputs corresponding to only a single grid cell (i.e., one lon/lat pair)?
• Or does the patching mechanism impose a minimum spatial resolution / minimum grid size that must be provided as input?

More specifically, is there a required minimum spatial extent (e.g., H*W grid) for the encoder to function correctly, even if the prediction target is a single station?
Any guidance on the recommended setup for station-level or point-based prediction would be greatly appreciated.

No files or tests are named. Start by tracing the encoder's spatial patching and the input-shape handling, then check whether single-cell inputs are accepted. Done means establishing the minimum H*W grid requirements and documenting or implementing the recommended setup for station-level prediction.