Configuration for stable centennial simulations at T21L8

[pierluigividale]

This is the configuration that ran stably for 100 years, then for a further 300 years. Plots available on my own GitHub repository.

Would be good for someone to revise and provide a critique.

Here is a copy of my notebook's cell:

using SpeedyWeather
# components
spectral_grid = SpectralGrid(trunc=21, nlayers=8)
output = NetCDFOutput(spectral_grid, PrimitiveWetModel, path="/Users/vidale/SpeedyWeather", id="Climate_Ocean_Land", output_dt=Minute(180))
add!(output, SpeedyWeather.TemperatureOutput(), SpeedyWeather.LargeScalePrecipitationOutput(), SpeedyWeather.ConvectivePrecipitationOutput())  # output also temperature and precip

# Some parameterisation choices
vertical_advection = WENOVerticalAdvection(spectral_grid)
large_scale_condensation = ImplicitCondensation(spectral_grid)
convection = SimplifiedBettsMiller(spectral_grid)
temperature=LandBucketTemperature(spectral_grid)
soil_moisture=LandBucketMoisture(spectral_grid)
land = LandModel(spectral_grid; temperature, soil_moisture)
ocean = SlabOcean(spectral_grid, mixed_layer_depth=50)
#longwave_radiation = NBandRadiation(spectral_grid, nbands=5)

#Try to fix ocean albedo
#albedo = Albedo(AlbedoClimatology(spectral_grid), AlbedoClimatology(spectral_grid))
albedo = Albedo(GlobalConstantAlbedo(spectral_grid, albedo=0.06), AlbedoClimatology(spectral_grid))
# create model, initialize, run

#time_stepping = Leapfrog(spectral_grid, Δt_at_T31=Minute(20))
time_stepping = Leapfrog(spectral_grid, Δt_at_T31=Minute(40)) #CPL this is the standar time step

### Entirely new run
model = PrimitiveWetModel(spectral_grid; vertical_advection, time_stepping, large_scale_condensation, convection, land, ocean, albedo, output=output)

### Continuation run
#initial_conditions = StartFromFile(path="/Users/vidale/SpeedyWeather", id="Climate_Ocean_Land",run_number=2)   # or specify `path` kwarg if not in this folder
#model = PrimitiveWetModel(spectral_grid; initial_conditions, vertical_advection, time_stepping, large_scale_condensation, convection, land, ocean, albedo, output=output)

add!(model, SpeedyWeather.AllOutputVariables()...)
simulation = initialize!(model)
# now offset via set! with add=true
set!(albedo.land, (λ, φ) -> 0., add=true)
run!(simulation, period=Year(100), output=true)

[milankl]

(if you put code into ```julia ... ``` blocks then they will be rendered as code blocks with Julia syntax highlighting)

[pierluigividale]

Ah! I was about to ask how to do that, because the format above is annoying.

[milankl]

Some comments, if you do add!(model, SpeedyWeather.AllOutputVariables()...) below then there's no need here
to provide PrimitiveWetModel to the NetCDFOutput as this just adds the default variables for that model
nor would you need to add temperature, precipitation etc

using SpeedyWeather

spectral_grid = SpectralGrid(trunc=21, nlayers=8)

output = NetCDFOutput(spectral_grid, path="/Users/vidale/SpeedyWeather", id="Climate_Ocean_Land", output_dt=Minute(180))
add!(model, SpeedyWeather.AllOutputVariables()...)

The large_scale_condensation = ImplicitCondensation(spectral_grid) and convection = SimplifiedBettsMiller(spectral_grid) are already default so you can shorten this part to

# Some parameterisation choices
vertical_advection = WENOVerticalAdvection(spectral_grid)
temperature=LandBucketTemperature(spectral_grid)
soil_moisture=LandBucketMoisture(spectral_grid)
land = LandModel(spectral_grid; temperature, soil_moisture)
ocean = SlabOcean(spectral_grid, mixed_layer_depth=50)

Albedo is fine but given you use the default time stepping now anyway just leave that out?

albedo = Albedo(GlobalConstantAlbedo(spectral_grid, albedo=0.06), AlbedoClimatology(spectral_grid))

removed those arguments then from the model constructor and whether you write ; output, or output=output is equivalent. The latter is Python syntax that Julian's simplified.

model = PrimitiveWetModel(spectral_grid; vertical_advection, land, ocean, albedo, output)
simulation = initialize!(model)

A manual albedo would need to be set now that's true but given you just +0 the line set!(albedo.land, (λ, φ) -> 0., add=true) doesn't do anything, just go straight to run!

run!(simulation, period=Year(100), output=true)

[pierluigividale]

Hi Milan,

thanks for pointing to redundancies, shall try to eliminate.

With that, there are some statements that I purposefully keep in there in order to remind me how to change things. I know that they are not necessary now, but, because I cannot remember how to change certain parameterisations and numerical schemes, if/when I want to, I inserted the statement that states the default, because then I can use that to search for my options.

An example is how to set the time step: every time I remove that statement, I forget, then I have to go back to the documentation and search, until I remember again. Probably not the most elegant way to do that, but every time I erased in the past, I ended up having to re-learn, which took me forever.

With that, YES, I can create one version with just the bare essentials.

[Olivash]

running the same block of code, it crushes

julia> run!(simulation, period=Year(100), output=true) Weather is speedy: run Climate_Ocean_Land 1%|▏ | ETA: 2:48:29 (848.97 years/day)┌ Warning: NaN or Inf detected at time step 5843 └ @ SpeedyWeather ~/.julia/packages/SpeedyWeather/iLz8G/src/output/feedback.jl:144 Weather is speedy: run Climate_Ocean_Land 2%|▋ | ETA: 1:38:21 (143Weather is speedy: run Climate_Ocean_Land 2%|▌ | ETA: 1:38:18 (1432.Weather is speedy: run Climate_Ocean_Land 2%|▌ | ETA: 1Weather is speedy: run Climate_Ocean_Land 2% ETA: 1:38:14 (Weather is speedy: run Climate_Ocean_Land 2% ETA: 1:38:1Weather is speedy: run Climate_Ocean_LaWeather is speedy: run Climate_Ocean_Land 2%|█▍

[milankl]

Did you update to the current master version? See the installation docs how to do that!

[pierluigividale]

UPDATE: after we made the land surface parameters and ICs mutable, I have altered the script a little bit, to allow more soil moisture variability. There is still not enough, and many parts of the world end up with perennial SMA=1, which needs investigation, but the configuration is stable: I have run T21, T31 for 100 years and T63 for 10 years. All good.

No point continuing to tune, because we now have a sea ice model, and I intend to tune based on that new Atmosphere+Land+Ocean+SeaIce model, which I have run for 10 years at T63 and is currently running for 100 years. Future tuning will be based on that new model.

FOR THE PURPOSE OF CMSS2025 and the CTL simulation, this is the new setup for Atmosphere+Land+Ocean, but please note that you will have to update the model, as otherwise those land surface changes will be rejected.

using SpeedyWeather
# components
spectral_grid = SpectralGrid(trunc=31, nlayers=8)
output = NetCDFOutput(spectral_grid, PrimitiveWetModel, path="/Users/vidale/SpeedyWeather", id="Climate_Ocean_Land", output_dt=Minute(360))

# Some parameterisation choices
vertical_advection = WENOVerticalAdvection(spectral_grid)
temperature=LandBucketTemperature(spectral_grid)
soil_moisture=LandBucketMoisture(spectral_grid)
land = LandModel(spectral_grid; temperature, soil_moisture)
ocean = SlabOcean(spectral_grid, mixed_layer_depth=50)

#Try to fix ocean albedo
albedo = Albedo(GlobalConstantAlbedo(spectral_grid, albedo=0.06), AlbedoClimatology(spectral_grid))

#time_stepping = Leapfrog(spectral_grid, Δt_at_T31=Minute(20))
# change the time scheme to be more off-centred and enable other waves than Kelvin and WMMRGs.
#implicit = ImplicitPrimitiveEq(spectral_grid, \alpha=0.5...1)

# create model, initialize, run
### For an entirely new run
model = PrimitiveWetModel(spectral_grid; vertical_advection, time_stepping, land, ocean, albedo, output)

### For a Continuation run
#initial_conditions = StartFromFile(path="/Users/vidale/SpeedyWeather", id="Climate_Ocean_Land",run_number=15)   # or specify `path` kwarg if not in this folder
#model = PrimitiveWetModel(spectral_grid; initial_conditions, vertical_advection, time_stepping, land, ocean, albedo, output)

#Make sure to add all diagnostics to the model
add!(model, SpeedyWeather.AllOutputVariables()...)

simulation = initialize!(model)
# now offset via set! with add=true
set!(albedo.land, (λ, φ) -> 0.1, add=true)
model.land.geometry.layer_thickness[1]=0.2
model.land.geometry.layer_thickness[2]=2.0
model.land.soil_moisture.runoff_fraction=0.25
model.land.soil_moisture.mask=true
model.land.temperature.mask=true
set!(simulation, soil_moisture = 0.75)
initialize!(model.land, model)   # first arg: what you want to initialize, 2nd: always model (read only)
# Finally, let us run the model
run!(simulation, period=Year(100), output=true)

[pierluigividale]

New configuration after the changes implemented on Friday 8 August:

using SpeedyWeather
# components
spectral_grid = SpectralGrid(trunc=31, nlayers=8)
output = NetCDFOutput(spectral_grid, PrimitiveWetModel, path="/Users/vidale/SpeedyWeather", id="Climate_Ocean_Land_SeaIce", output_dt=Minute(360))

# Some parameterisation choices
vertical_advection = WENOVerticalAdvection(spectral_grid)

#Try to fix ocean albedo
albedo = Albedo(ocean=OceanSeaIceAlbedo(spectral_grid), land=AlbedoClimatology(spectral_grid))

### Entirely new run
model = PrimitiveWetModel(spectral_grid; vertical_advection, albedo, output)

simulation = initialize!(model)
# now offset via set! with add=true
set!(albedo.land, (λ, φ) -> 0.1, add=true)

### Continuation run
#initial_conditions = StartFromFile(path="/Users/vidale/SpeedyWeather", id="Climate_Ocean_Land_SeaIce",run_number=4)   # or specify `path` kwarg if not in this folder
#model = PrimitiveWetModel(spectral_grid; initial_conditions, vertical_advection, albedo, output)

# output both SST and sea ice concentration
add!(model, SpeedyWeather.OceanOutput()...)
add!(model, SpeedyWeather.LandOutput()...)
add!(model, SpeedyWeather.RadiationOutput()...)
add!(model, SpeedyWeather.PrecipitationOutput()...)
add!(model, SpeedyWeather.SurfaceFluxesOutput()...)
run!(simulation, period=Year(1), output=true)