Hi,

Is there data comparing the performance of GraphCast_small vs that of the regular, high-resolution version of GraphCast? I would assume that the GraphCast_small wouldn't perform quite as well, but I am curious as to by how much.

Also, for the GraphCast_operational model, what does it mean for the model to be fine-tuned on HRES data from 2016 to 2021 after already being trained on earlier ERA5 data?

Thanks in advance for any insight!