search

ai2cm / fv3net

explore the FV3 data for parameterization
MIT License
16 stars 3 forks source link

Add vcm.DerivedMapping keys for lowest level T, q, u, v #2318

Closed AnnaKwa closed 1 year ago

AnnaKwa commented 1 year ago

The SST hybrid reservoir model uses several lowest level atmospheric inputs, which are saved in fortran diagnostics as 2D fields but only available at runtime via the full 3D fields. This adds entries to vcm.DerivedMapping so that just the lowest level 2D fields can be retrieved at runtime to use as inputs.

AnnaKwa commented 1 year ago

That's a good point. I checked out your concerns in this notebook. It looks like the difference in training fields vs lowest model level would be a source of systematic bias when testing out the hybrid SST reservoir online.

AnnaKwa commented 1 year ago

I looked into Andre's comment and the difference between the T2m, 10m u/v wind fields (used in training) and the lowest model level fields is pretty significant. https://github.com/ai2cm/explore/blob/master/annak/2023-07-17-example-reservoir-training/2023-08-25-check-fortran-near-sfc-vs-lowest-level.ipynb

I made a different PR to instead expose 10m winds (T2m is already available in the wrapper). Closing this PR.