Integrating veros with an atmosphere in `ClimaOcean.jl`

[simone-silvestri]

Hello Veros team.

Thanks for the great code! It is very nice to see the development of a performant python-native ocean model.

On the julia side, we are developing a coupled framework, called ClimaOcean, and would like to have the option to include arbitrary ocean models, as this would enable us to generalize the coupled model and utilize another model that provides ensemble simulations for comparison with our own ocean model.
Veros would be the perfect starting point given the compatibility between Julia and Python.

Therefore, with @nutrik we were working towards coupling Veros with an atmosphere in our julia coupled framework, ClimaOcean. The relevant PR is CliMA/ClimaOcean.jl#602. We believe that this exercise would benefit both, as we would be able to couple another model in ClimaOcean, and Veros would obtain an out-of-the-box coupler that can be tested with prescribed and prognostic atmospheres (we are planning/have started to couple with ClimaAtmos.jl and SpeedyWeather.jl)

If there is interest on your side to leverage this coupled infrastructure, the last bit we need to ensure a smooth and performant coupling is the possibility of prescribing boundary conditions for temperature, salinity, and velocities.

At the moment, I believe there are a couple of assumptions in Veros that would need relaxing and generalizing:

• BC for temperature and salinity are a climatological relaxation (we would need a prescribed flux since fluxes are computed at each timestep in ClimaOcean)
• Interpolation from monthly climatology for stresses should be generalized to a container for prescribed fluxes that can be filled in. At the moment, we need to fill all the climatological maps to make sure interpolation works to give us the correct stress, which is a bit of a waste.

This would basically allow us to remove these lines, which are not so elegant a hack to work around these assumptions:
https://github.com/CliMA/ClimaOcean.jl/blob/8aa6b13c96472b88de76fa90f60da10471b1ac13/ext/ClimaOceanPythonCallExt/veros_state_exchanger.jl#L86-L114

Thanks again, and excited to make this happen if you are interested!

[simone-silvestri]

This is a short snippet of the video produced by the script
which forces the 4-degree Veros simulation using the 3-hourly JRA55 atmosphere

veros_4deg_forced_by_JRA55.online-video-cutter.com.1.mp4

[dionhaefner]

Thanks @simone-silvestri for sharing your results, and sorry for not reacting until now (I was on paternal leave).

I'm trying to understand the requirements a bit better in order to suggest a good path forward. Specifically:

Interpolation from monthly climatology for stresses should be generalized to a container for prescribed fluxes that can be filled in. At the moment, we need to fill all the climatological maps to make sure interpolation works to give us the correct stress, which is a bit of a waste

What is being interpolated here? Which climatological maps specifically? The code you link looks reasonably clean to me --- what specifically are you trying to eliminate?

[simone-silvestri]

Thanks for the answer. And also sorry for my slow follow-up. I now understand I was trying to use to use a prebuilt setup (the global 4 degree) that has a set_forcing defined in the class which rewrites whatever forcing I define.
I am trying to avoid needing to rebuild a new setup in python, as I would like to completely hide the python part.
The objective here is to get an example running in CI, so I think the easier option is to just substitute the set_forcing from the class with a no-op:

# Disable the hardcoded forcing function from the specific Veros setup so that our forcings are not overwritten
ocean.setup.set_forcing = @pyeval("lambda state: None")

Which should do the trick.

[dionhaefner]

Making sure I understand this correctly – are you using a hacked version of Veros that takes care of the atmosphere interaction? And you want to disable setup-specific set_forcing functions to make sure your changes are not being overwritten?

In that case it sounds much easier to me to not modify Veros source code, and implement atmosphere interactions as part of set_forcing?