Held Suarez temperature forcing

[milankl]

In addition to #269 also implements the temperature relaxation

[milankl]

The last commits reduce drastically the time spend in the parameterizations, which was (boundary layer drag only) about 2x slower than for physics=false

[milankl]

@maximilian-gelbrecht the last commits remove the idea to use abstract types in default_parameters.jl to dispatch initial conditions, orography, etc and uses concrete structs instead so that they directly can hold associated parameters. E.g.

run_speedy(initial_conditions=ZonalWind(u₀=25))
run_speedy(orography=ZonalRidge())
run_speedy(boundary_layer=LinearDrag(drag_time=12))

meaning that orography=ZonalRidge will now throw an error because it passes on the type and not an instance of it. Adding the () will, however, initialize with default values.

[milankl]

T63, 8 levels, surface temperature with Held Suarez forcing and full orography from the last commit

out3.mp4

[maximilian-gelbrecht]

@maximilian-gelbrecht the last commits remove the idea to use abstract types in default_parameters.jl to dispatch initial conditions, orography, etc and uses concrete structs instead so that they directly can hold associated parameters. E.g.

run_speedy(initial_conditions=ZonalWind(u₀=25))
run_speedy(orography=ZonalRidge())
run_speedy(boundary_layer=LinearDrag(drag_time=12))

meaning that orography=ZonalRidge will now throw an error because it passes on the type and not an instance of it. Adding the () will, however, initialize with default values.

Nice! I was already thinking about suggesting that change.

[milankl]

@white-alistair I had to adapt the thermodynamics tests as we now centrally calculate the pressure (and log pressure) on every level in get_column! so that it doesn't have to be recalculated for every parameterization. Meaning column.pres is now a vector of nlev+1 with the last element being the surface

Will also eventually overhaul the vertical interpolation as recalculating the interpolation weights for every grid cell can be avoided. That actually sounds like it should be part of Geometry (in contrast to a lot of other things in there, that don't seem like "Geometry"...)

[milankl]

@white-alistair there seems to be some inconsistencies with the units of pressure across dynamics and various places in the physics. At the moment we have

• log(Pa) only in the dynamics
• convert to Pa at every level for the physics through $p_k = \sigma_k \exp(\log(p_s))$ (i.e. 1 exp per grid point)
• pres_thresh_cnv that is normalised (by 1000hPa, I assume)
• parameterizations that do pres > pres_thresh_cnv so clearly assume the pressure of the column to be normalised too

Just a note to self (and you) that we may want to unify this somehow and make consistent. I believe we should calculate general quantities like pressure that clearly belog to the column centrally in get_column! and not in individual parameterizations.

[milankl]

And just to add: I believe at some point it makes sense to implement hybrid $\sigma$-pressure coordinates, because it seems to be widely used standard. It's not suuuper difficult to implement for the dynamics, but it obviously interferes with the physics so want to put this off for now, but it's on my todo wishlist.