Progress towards primitive equations

[milankl]

Quick overview where we are in terms of a primitive equation dynamical core and parameterizations

Dynamical core

• Geopotential calculation Geopotential calculation #188
• Virtual temperature Virtual temperature calculation #197
• Gradients of (logarithm of) surface pressure Surface pressure tendency #189
• Layer thickness-weighted vertical averages of u,v,div Surface pressure tendency #189
• Vertical velocity / mass fluxes Vertical velocity dsigma/dt #191
• vertical advection Vertical advection of u,v + tendencies (vor, pres grads) #198 Overhauled vertical advection #211
• pressure gradient terms Vertical advection of u,v + tendencies (vor, pres grads) #198
• log surface pressure tendency -Dbar in spectral; a,b work arrays; humidity advection; vorticity_flux_divcurl! with flags #204
• temperature equation Temperature equation and primitive equation time stepping #205 Reformulated temperature equation #210
• add kinetic energy to geopotential Bernoulli potential for PrimitiveEquation #196
• realistic initial conditions Jablonowski and Williamson initial conditions #249
• semi-implicit corrections Semi-implicit corrections for primitive equations #251
• bug hunting Primitive equation dry core bughunt #256 Jablonowski test case #265

Parameterizations

• Surface drag Boundary layer scheme from Held and Suarez #269
• Held Suarez temperature relaxation Held Suarez temperature forcing #270
• Shortwave radiation Add shortwave radiation parametrization #169
• Longwave radiation Add longwave radiation parametrization #177
• Large scale condensation Parametrization of Large-Scale Condensation #82
• Convection Convection #128 Convection #414
• Surface fluxes Surface fluxes of heat, moisture and momentum: Land and ocean seasonal climatology #399
• Soil temperature Surface fluxes of heat, moisture and momentum: Land and ocean seasonal climatology #399
• Vertical diffusion

@maximilian-gelbrecht @white-alistair @dmey

[milankl]

I hope I can tick off many of the dynamical core parts in the next two weeks, but some things seem to be trickier than I originally thought: For example our structure of the DiagnosticVariables split into layers makes absolute sense for many parts of the dynamical core, but not for the vertical advection (used for u,v,T,q), which is a bit annoying. I need to see what's the best to implement that (naive Fortran-SPEEDY copy vs a restructured version with some parallelisation in mind). I'm currently tending towards a naive copy.

[white-alistair]

Useful overview! How would you estimate the relative difficulty and/or time required for the remaining issues?

[milankl]

• Virtual temperature: Easy $T_v = T(1+\mu q)$, but it's currently unclear to me whether this is needed in spectral or grid-point space or both or when to compute it to avoid a spectral transform of the fields.
• Vertical advection: Medium, it's not that tricky, but the vertical advection operator acts across layers, and I don't know yet how to structure this given that we currently work mostly on layers with diagn.layers[k].grid_variables.var_name
• pressure gradient terms: Medium, the gradients are already computed for surface pressure tendency and the geopotential too, but it still may require some thinking.
• temperature equation: Medium, the implementation should be straight forward, but the maths will take me a bit to get my head around
• add kinetic energy to geopotential: Easy, that's very similar to how it's done in the shallow water model
• semi-implicit corrections: Harder, as the maths will be adapted to the scaling of our equations and if it's not implemented correctly then things will keep blowing up as the gravity waves aren't dampened.

about the physics: You or @dmey probably know better than me.

[maximilian-gelbrecht]

• Vertical advection: Medium, it's not that tricky, but the vertical advection operator acts across layers, and I don't know yet how to structure this given that we currently work mostly on layers with diagn.layers[k].grid_variables.var_name

Don't know how the vertical operators exactly look like, but I wrote those get_var functions for the PrognosticVariables with output / setting initial conditions in mind. For working with the layers like that, it might be worth it to write non-allocating versions that return views where needed.

[dmey]

Should be able to do surface fluxes, skin temperature and soil temperature around the end of the year. Note that these will depend on extra work like land data. I am not sure about sea/ocean stuff--sea ice I thought we can just use the climatology at least for starters, ocean mixed-layer I would need more info. Vertical diffusion after that but will depend on workload and priorities. @milankl do you have a date in mind?

[milankl]

No date, it's done when it's done. I'm just happy that everyone contributes, whenver that is. Happy for you to start on those you suggested. I don't mind being more involved with vertical diffusion and/or ocean mixed-layer, but I won't focus on them before the dynamical core is up and running reliably.

[milankl]

With #251 merged (that was painful!) technically the drycore is now complete 🥳 However, there are still some bugs/errors in the explicit terms: T63 8levels after ~12 days, vorticity in lowest layer, top: with semi-implicit time stepping and 3x the time step, bottom: without.

So the semi-implicit time stepping does its job, but the error occurs somewhere else.

[milankl]

Moving this to v0.6 milestone as what I had intended for v0.5 is basically completed. I believe there's still something wrong as outlined in #309 but that's not for now.

[milankl]

Although there are still bits and bobs missing, overall the primitive equation dynamical core is completed and so I'm closing this issue in favour of other more specific issues.