Introduce an interface to allow attaching a prognostic atmosphere #322
Conversation
simone-silvestri
changed the title
PrognosticAtmosphere…CliMA/ClimaOcean.jl into ss-glw-mk/branch-for-coupling-to-speedy
| @@ -28,6 +29,10 @@ function time_step!(coupled_model::OceanSeaIceModel, Δt; callbacks=[], compute_ | |||
| time_step!(sea_ice) | |||
| end | |||
|
|
|||
| # Time step the atmosphere | |||
| # TODO: allow different time-steps for atmosphere and ocean | |||
| time_step!(atmosphere) | |||
There was a problem hiding this comment.
Somehow you have to ensure the time-step is size Δt
There was a problem hiding this comment.
Numerically you can't change the time step in SpeedyWeather during the integration (well you can but it requires you to recompute the implicit solver). You can set it flexibly at initialization though. Whether ClimaOcean's dt is used in SpeedyWeather or SpeedyWeather's step is used in ClimaOcean I don't mind.
There was a problem hiding this comment.
I think that ClimaOcean should manage the time-step and that the component models time-step should be set based on what ClimaOcean has.
If component models have some restriction then we need to design a way to manage that. This also means that TimeInterval schedule cannot be used in the Simulation of OceanSeaIceModel.
Why don't we write a function like set_time_step!(atmosphere, dt) (also for ocean). Then SpeedyWeather will throw an error if dt is different from its internal unchangeable time-step.
There was a problem hiding this comment.
or set_Δt!(atmosphere, Δt).
(we have discussed elsewhere about the fact that we often use "time step" as a noun Δt as well as a verb for actually taking a time-step, ie time_step!... a little annoying)
There was a problem hiding this comment.
or set_time_step_size!(atmos, Δt)
There was a problem hiding this comment.
normally we choose our time step with Leapfrog(spectral_grid, Δt_at_T31=Minute(40)) with T31 being our default resolution and a linear scaling towards higher resolutions that's automatically applied. But with SpeedyWeather/SpeedyWeather.jl#650 I've just created a PR that also allows to set! the time step more explicitly
There was a problem hiding this comment.
Sorry for being dull --- do you mean that #650 makes it so that the user can choose the time-step when setting up a SpeedyWeather simulation?
This PR reorganizes a bit the code to allow for a prognostic atmosphere, which means adding some interfaces that can be extended in the atmosphere we want to couple with