Jacob Prince-Bieker

Hi, No, it should just be you have to change the input dimension size to 102, by default its 78 to match the input size in the paper, but I...

Yeah, but as that distance in the sin and cos is the great circle distance from the icosahedron node to the lat/lon point, I think it is quite similar. The...

Ah yeah, that makes sense, I guess one different way could be just decompose the distance into the north/south change and the west/east change, not in lat/long but in the...

Cool, thanks so much for pointing that out!

From the skim read I did, I am not sure it is either! The Euclidian distance, or great circle distance does work for our data, as it is on a...

> Sounds good! > > I'm really out of my depth here but I'm dimly aware that this might be quite an active area of research in academia. If you...

From the ECMWF Conference, there are apparently a lot more issues than generally just taking the observations and doing assimilation, one example here: https://github.com/DL-WG/LatentAssimilation but might need to be more...

Yeah, in the original paper, the author didn't mask pressure levels that go underground, so I'll probably leave it as is here, but could add an option for adding a...

Yes! That's the one. There have now been some other implementations of similar models that might be helpful, primarily Nvidia's GraphCast implementation here: https://github.com/NVIDIA/modulus/blob/main/modulus/models/graphcast/graph_cast_net.py

Hi, sorry for the delay, was on annual leave. For the repeating of the indicies and attributes, its repeating the latent graph to match the batch size of the input....