Training recipe for Stormscope-goes-mrms
Hi,
After looking through phyicsnemo and earth2studio I'm struggling to piece together a training scope for the GEOS + MRMS Stormscope model. I can see the data ingests and regridding could be done by the data handlers in the earth2studio, but there's no details on handling of masks/missing data, data normalization, experts configs and the actual training pipeline. Is there any chance I've missed this code somewhere or it can be released? I am keen to adapt the training for Australian datasets.
Cheers,
Joshua S
The recommended entrypoint for training custom StormScope models is the general regional weather diffusion training recipe in PhysicsNeMo: https://github.com/NVIDIA/physicsnemo/tree/main/examples/weather/regional_weather_diffusion
I would watch that recipe as in the next few weeks we'll be adding some example configs for StormScope model architectures. The recipe is designed to be easy to adapt to custom datasets, so should be relatively straightforward to get started if you want to train a custom version over Australia.
Thanks for the pointer and prompt to keep an eye on the repo!