Using 1° resolution instead of 0.25°

[KieranMalandain]

Hi Aurora team,

My lab has access to 1° ERA5 data, but Aurora requires 0.25° in its primary model. Would it be possible to pass this 1° data straight to Aurora, or must we use torch.nn.functional.interpolate(...)?
We are fine-tuning for MJO prediction, so 1° resolution is sufficient for us to calculate RMM indices.

Many thanks for your work!
Kieran

[wesselb]

Hey @KieranMalandain!

Thanks for opening an issue. Your best bet will likely to give the one-degree data directly to Aurora and fine-tune the model. Note that fine-tuning is necessary, as the model hasn't been trained on one-degree directly so it won't work out of the box, but fine-tuning should work well and be effective.