Building upon the wind mixing and convection tutorial for submesoscale dynamics

[roxyboy]

Hi Oceananigans,

Thank you for developing and maintaining this extremely powerful model!

Amongst the examples in the tutorial, I'm particularly interested in the wind mixing and convection case (https://clima.github.io/OceananigansDocumentation/stable/literated/ocean_wind_mixing_and_convection/).
I want to build upon this in the context of submesoscale dynamics but would you think changing:

• the model to HydrostaticFreeSurfaceModel
• closure to CATKE
• advection schemes to Weno
• removing explicit diffusion
• turn off evaporative boundary condition so as to avoid a salinity run off
• and adjusting the domain parameters

would be a suitable test case?

I would like to study the spatiotemporal scales of the vertical buoyancy flux using data-driven methods.

cc @navidcy

[glwagner]

The physics sound interesting! Are you proposing to develop this as a new example? I think that is possible, but might have to run on GPU. That said we did recently open up the documentation build to GPU so this is a possibility (previously, all examples had to run on CPU and we were resource-limited so they needed to be cheap).

Computational resource limitations mean that the examples must be very carefully curated. That said we need more examples with the hydrostatic model for sure, and an example with CATKE would be nice.

My other comment is that the example is proposed to be developed from another one. However, to maximize the impact of each example, I would just emphasize that it should invoke as many differences as would be sensible from other examples. In other words lets think of this as simply writing a new case rather than modifying an existing one...

Nonlinear equation of state could also be appropriate, maybe Stokes drift too. For more novelty we can compute the surface fluxes using simple bulk formula instead of prescribing fluxes, perhaps.

[glwagner]

does picking up from the checkpoint succeed?

[roxyboy]

It seems that the model starts running from the checkpoint number but I'm not sure if this means that it's successful...

[glwagner]

you can compare the output from a model which is not checkpointed, and a model that is picked up from a checkpoint

[glwagner]

I think with an ImplicitFreeSurface you do not need to checkpoint the tendency of the free surface. For a SplitExplicitFreeSurface you probably do to be strictly correct on picking up from a checkpoint. However, I doubt that the dynamics of the simulation would be substantially affected (so it also depends on the accuracy you need here). We are planning a bit of an overhaul of the checkpointer in the near future so if you can raise an issue we can address it at that point.

[glwagner]

Sounds like the simulation described in this discussion is running which is pretty awesome! Everyone wants to see visualizations of the flow 😄