Example with Real Atmospheric Data

[david5010]

Hi,

Thank you for sharing this open source project! I know that in one of the tests, there's an example using a randomly generated tensor as a test. However, is it possible to show an example using real atmospheric (GOES) data? In one of the issues, I did see that it's just a concatenation, but is the best way to convert the geospatial data into a tensor and pass it into the model and is there a specific ordering of the variables the model expects?

It would be an amazing addition to have a demo using actual GOES data.

[jacobbieker]

Hi,

We'd really love to have a contribution showing it with real GOES data! The issue is the GOES data itself is fairly large, and would involve either a lot of boilerplate in the README to open from AWS, or Planetary Computer. But an example script, etc. would be nice to have. Concatenation is the easiest way, and the way used in the MetNet papers, so that is what is done here, although there are probably other ways to do it better. We don't really have a trained model on GOES data, so no specific ordering is really necessary. But if you train it with GOES data, or other data, it just needs to follow the same ordering as in training.