Simple Slab Land #8

LenkaNovak · 2021-06-07T01:35:52Z

Dry heat advection-diffusion equation in an atmospheric single stack coupled to a prognostic surface temperature equation reflecting a slab land. Dry setup. This case is mainly for software design before implementing more complex cases.

Design: https://github.com/CliMA/CouplerMachine/blob/ln/slabland/experiments/SlabLand/README.md
Conservation tests: under the 16 June section

jb-mackay · 2021-06-07T22:34:35Z

Project.toml

@@ -13,6 +13,7 @@ KernelAbstractions = "63c18a36-062a-441e-b654-da1e3ab1ce7c"
 LinearAlgebra = "37e2e46d-f89d-539d-b4ee-838fcccc9c8e"
 MPI = "da04e1cc-30fd-572f-bb4f-1f8673147195"
 Oceananigans = "9e8cae18-63c1-5223-a75c-80ca9d6e9a09"
+Plots = "91a5bcdd-55d7-5caf-9e0b-520d859cae80"


We should keep our dependencies as trim as possible - so no Plots in the CouplerMachine dependencies (Pkg.rm("Plots") in the CouplerMachine environment). If you add Plots from the default Julia environment (no --project) then it will still be accessible when you want to use it in other projects without adding it to the dependency list:

julia -e 'using Pkg; Pkg.add("Plots")'

jb-mackay · 2021-06-19T00:11:32Z

experiments/SlabLand/README.md

+
+ (constant conductances and ConstantViscosity( FT(0), FT(0), FT(1e-5), u = 0 )
+
+## **Pipeline** (probably delete)


I think this pipeline could be worth holding onto for a little while - it's a clear walkthrough of what is happening during the coupling.

kmdeck · 2021-06-21T15:57:15Z

experiments/SlabLand/README.md

+- Upward-pointing net long-wave flux
+    $$R_{LW} = F_{sfc}^u - F_{sfc}^d = \epsilon \sigma \theta_{sfc}^4 - \epsilon F_{a}$$
+- Soil storage
+    $$G = \kappa_s (\theta_{sfc}-\theta_h)/h_s$$


\theta_{sfc} -> T_{sfc}

Good catch! :)

kmdeck · 2021-06-21T16:06:42Z

experiments/SlabLand/README.md

+
+3. Flux / state storage and communication
+    - the coupler stores 2 fields:
+        - `EnergyFluxAtmos` which collects `F_ρθ_accum / Δt` calculated and accumulated in Atmos (with $Δt$ denoting the period of the coupling cycle), and is called by land for its $T_{sfc}$ equation


maybe rename, e.g. BoundaryEnergyFlux?

…into ln/slabland

jb-mackay · 2021-06-21T22:46:08Z

docs/src/boundary_conditions_doc.md

@@ -0,0 +1,165 @@
+# Boundary Conditions (BCs)
+
+This section is an attempt to bridge mathematical considerations of BCs and their implementations in climate models.


I think this doc is great and very helpful, but may perhaps belong with the RQA work in ClimateMachine. What do you think @LenkaNovak ?

RQA said they will add it once they've revamped their BCs, so perhaps we can leave it here until then?

jb-mackay · 2021-06-21T23:54:30Z

experiments/SlabLand/coupler/put_get_calls.jl

@@ -0,0 +1,98 @@
+# Land put and get calls


I like the coupler directory set up and am glad we are separating out the pre- and post-step function definition from the balance laws; so this is great! There will probably be some other coupler-specific configuration/definitions that we'll need to do in the future so it's good to have a separate place for it.

kmdeck · 2021-06-22T16:40:45Z

experiments/SlabLand/README.md

+- Lower (coupled) boundary:
+    - ρ: `Impenetrable()`  
+    - u: `FreeSlip()`
+    - θ: `CoupledPrimary()`: $\frac{\partial θ}{\partial t} = - ∇ \cdot F_{tot}$ where $F_{tot}$ is the surface flux calculated by `calculate_land_sfc_fluxes(_...)` using Atmos and Land properties (see below) 


Is the boundary condition $$\partial θ / \partial n = F_{tot}$$? (not the divergence of the flux)

Not quite, the way we set the BCs in our case is ultimately via the normal flux, i.e. fluxᵀn.ρθ += - F_tot / scaling_params (referred to as local_flux in different parts of the code), and then the model calculates the div of that.

ok! that is what I thought. I was just commenting on the way it is written. (the math expression does not reflect the BC, which is a Neumann BC on flux like you wrote - the math instead reflects the entire RHS at the boundary).

kmdeck · 2021-06-22T16:49:05Z

experiments/SlabLand/README.md

+5) Coupler converts Atmos field `state.F_ρθ_accum` to the coupler field `EnergyFluxAtmos` 
+6) Coupler sends its field `EnergyFluxAtmos` and transforms it into Land field `auxiliary.F_ρθ_prescribed`
+5) Land performs its own initialization and physics (`G`), which is then added to the  `auxiliary.F_ρθ_prescribed` (corresponding to $F_{tot}$ above) in the `source!` function to solve:
+$$T_s^{n+1} = T_s^n +\frac{F_{tot}^{n...}}{\rho_s c_s h_s}\Delta t$$


i also found this pipeline really helpful.

the use of F_tot is inconsistent here - in line 87 it refers to the non G fluxes, here it refers to G + non G fluxes.

edited, reread more carefully and understand the ^n notation now!

If one wanted to use a different explicit stepper on the land side (not forward Euler), is that possible? Would the "coupling cycle" dt be e.g. the different dts for the RK stages?

@kmdeck unless I'm mistaken, yes, the choice of the land timestepper will determine how $T_s^{n+1}$ is computed, and this does not have to be forward Euler, of course.

The "coupling cycle" refers to the synchronization timescale of the two solvers. At the beginning/end of the coupling cycle, coupled field information is exchanged. Within that coupling cycle, the individual components can take as many substeps as they need, but they don't exchange information.

thanks! @jb-mackay. I'll think about it more! May still follow up with ?s, but that sounds good.

codecov · 2021-06-22T18:13:08Z

Codecov Report

Merging #8 (28f775a) into main (eb2b6ac) will not change coverage.
The diff coverage is n/a.

❗ Current head 28f775a differs from pull request most recent head e82eab9. Consider uploading reports for the commit e82eab9 to get more accurate results

@@           Coverage Diff           @@
##             main       #8   +/-   ##
=======================================
  Coverage   40.98%   40.98%           
=======================================
  Files           4        4           
  Lines          61       61           
=======================================
  Hits           25       25           
  Misses         36       36

Impacted Files	Coverage Δ
src/CouplerMachine.jl	`100.00% <ø> (ø)`
src/CplModel.jl	`0.00% <ø> (ø)`
src/CplSolver.jl	`0.00% <ø> (ø)`
src/CplState.jl	`77.41% <ø> (ø)`

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update eb2b6ac...e82eab9. Read the comment docs.

kmdeck · 2021-06-22T20:23:39Z

experiments/SlabLand/README.md

+    - it is called by Atmos, which is assumed to be the model with the shortest timestep. This means that the fluxes are only calculated once and are consistent with Atmos's time stepping.
+
+2. Flux accumulation
+    - In order to integrate fluxes in time consistently with the Atmos time stepper, we define a new prognostic variable `F_ρθ_accum` whose source function accumulates the fluxes, so that $F\_ρθ\_accum = \int F_{tot} dt$


maybe this language was causing confusion? the source function doesnt accumulate the fluxes, it is the RHS term of the ODE d(F_rho theta_accum)/dt = F_tot

Umm, it basically calculates d (F_ρθ_accum) += \int F_{tot} dt. I will clarify the doc, but yes, accumulating via the source function is not ideal. In the CM that's the only easy option we have, but we will be giving feedback to the CMCore team on this to make it less confusing in the new version :).

hey! not being critical, just more clarifying for my own sake. I still dont understand the confusion re: using the language source - it seems as though this is exactly the same usage (a RHS term for a differential equation) as we always use it, except in this case the ODE is d(F_rho theta_accum)/dt = F_tot.

I can chat with Valeria more though since she was the one who brought it up!

kmdeck · 2021-06-22T20:26:28Z

experiments/SlabLand/balance_law_interface_atmos.jl

+)
+    init_state_auxiliary!(m, m.physics.orientation, state_auxiliary, geom)
+
+    state_auxiliary.T_sfc = 0


is this consistent with the IC of the land model?
https://github.com/CliMA/CouplerMachine/pull/8/files#diff-7385dec13548164d3ca2e93987be013394bcfa736c47de38fa9d0b36e5860e01R119

Yes, at the coupled boundary, aux.T_sfc (of Atmos) is set to be equal to state.T_sfc (of Land) at the beginning of each Atmos coupling cycle. This line just initializes the elements of the MPIStateArray that are not at the boundary (so they should never be used). Technically, this line should not be needed but I was occasionally getting Nans if the whole field was not initialized. 🤷

kmdeck · 2021-06-22T20:31:08Z

experiments/SlabLand/balance_law_interface_slabland.jl

+
+"""
+    Boundary conditions
+"""


Are these BC methods needed yet? Right now the PDE for land is just an ODE...

Eeep! Well spotted, I've just removed them! :)

kmdeck · 2021-06-22T20:38:02Z

it seems like a lot of the files in this PR are not specific to the slab land experiment, and looks like some are already in the repo (e.g. utilities/sphere_utils, grid stuff, numerics (filters...), even physics). maybe can be reorganized to have a single copy of these files?

kmdeck · 2021-06-22T20:43:36Z

experiments/SlabLand/parameters_initialconditions.jl

+        xmax = 10000,   # horizontal extent of the SingleStack in the x direction [m]
+        ymax = 500,     # horizontal extent of the SingleStack in the y direction [m]
+        zmax = 10000,   # vertical extent of the SingleStack [m]
+        g    = 9.81,    # gravitation acceleration [m /s^2]


worth using Clima parameters at this stage?

I think for this illustrative example it's ok to leave the params list as is, but it's definitely worth doing for the more formal test cases that we will be publishing as part of CliMA. :)

LenkaNovak · 2021-06-24T20:56:47Z

it seems like a lot of the files in this PR are not specific to the slab land experiment, and looks like some are already in the repo (e.g. utilities/sphere_utils, grid stuff, numerics (filters...), even physics). maybe can be reorganized to have a single copy of these files?

Good point. These are copied from the RQA interface. We are keeping these folders under each example separate, because the RQA interface was evolving as we were writing these different test cases, so we wanted to have a stable backend for each case. Once the Aleph.jl repo/module (where the RQA code will live) is ready though, we will import these files from there.

jb-mackay

Looks good @LenkaNovak. Once you have addressed @kmdeck's comments feel free to merge. As mentioned, we'll have to re-tidy this and some other experiments when the RQA interface if formalized.

kmdeck

looks good to me @LenkaNovak @jb-mackay. nice job!!

Lenka Novakova added 2 commits May 31, 2021 01:59

init SlabLand SingleStack in RQA interface

3fe7917

diffusive atmos slab land conservation passes

8976f6e

jb-mackay reviewed Jun 7, 2021

View reviewed changes

Lenka Novakova and others added 6 commits June 10, 2021 13:32

cleanup

83994d9

clean up + docs

7dd2f14

Delete SlabLand.md

5b8df7f

interim save

8a4c47d

Fix docs/Manifest.toml

d2fcfa3

Fix typos in bc doc.

5b075e1

jb-mackay reviewed Jun 19, 2021

View reviewed changes

Lenka Novakova added 2 commits June 19, 2021 12:55

advection added + conserving BCs

3b5f3e1

resol conf

f378219

LenkaNovak marked this pull request as ready for review June 19, 2021 20:02

LenkaNovak requested a review from kmdeck June 19, 2021 20:13

kmdeck reviewed Jun 21, 2021

View reviewed changes

jb-mackay added 2 commits June 21, 2021 09:36

Clean up dependencies

bb26473

Merge branch 'ln/slabland' of https://github.com/CliMA/CouplerMachine …

a88b8b2

…into ln/slabland

jb-mackay reviewed Jun 21, 2021

View reviewed changes

Fix small typos in bc documentation

83d0634

jb-mackay reviewed Jun 21, 2021

View reviewed changes

kmdeck reviewed Jun 22, 2021

View reviewed changes

jb-mackay added 5 commits June 22, 2021 09:50

Clean up docstrings and package imports

0082c21

Fix manifest

b2c81fb

Merge branch 'main' into ln/slabland

3469150

Remove duplicate entries in manifest

be274e4

Rename register_cpl_field to coupler_register.

28f775a

kmdeck reviewed Jun 22, 2021

View reviewed changes

jb-mackay approved these changes Jun 24, 2021

View reviewed changes

Lenka Novakova and others added 2 commits June 24, 2021 16:53

review clean up

ecf8b97

slab numerical flux bug fix

e82eab9

kmdeck approved these changes Jun 25, 2021

View reviewed changes

LenkaNovak merged commit e743c9f into main Jun 25, 2021

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Simple Slab Land #8

Simple Slab Land #8

LenkaNovak commented Jun 7, 2021 •

edited

Loading

jb-mackay Jun 7, 2021

jb-mackay Jun 19, 2021

kmdeck Jun 21, 2021

LenkaNovak Jun 24, 2021

kmdeck Jun 21, 2021

jb-mackay Jun 21, 2021 •

edited

Loading

LenkaNovak Jun 24, 2021

jb-mackay Jun 21, 2021

kmdeck Jun 22, 2021

LenkaNovak Jun 24, 2021

kmdeck Jun 24, 2021

kmdeck Jun 22, 2021

kmdeck Jun 22, 2021 •

edited

Loading

jb-mackay Jun 24, 2021

kmdeck Jun 25, 2021

codecov bot commented Jun 22, 2021 •

edited

Loading

kmdeck Jun 22, 2021

LenkaNovak Jun 24, 2021

kmdeck Jun 24, 2021

kmdeck Jun 22, 2021 •

edited

Loading

LenkaNovak Jun 24, 2021

kmdeck Jun 22, 2021

LenkaNovak Jun 24, 2021

kmdeck commented Jun 22, 2021 •

edited

Loading

kmdeck Jun 22, 2021

LenkaNovak Jun 24, 2021

LenkaNovak commented Jun 24, 2021

jb-mackay left a comment

kmdeck left a comment


		(constant conductances and ConstantViscosity( FT(0), FT(0), FT(1e-5), u = 0 )

		## Pipeline (probably delete)

		@@ -0,0 +1,165 @@
		# Boundary Conditions (BCs)

		This section is an attempt to bridge mathematical considerations of BCs and their implementations in climate models.

Simple Slab Land #8

Simple Slab Land #8

Conversation

LenkaNovak commented Jun 7, 2021 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

jb-mackay Jun 21, 2021 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

kmdeck Jun 22, 2021 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

codecov bot commented Jun 22, 2021 • edited Loading

Codecov Report

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

kmdeck Jun 22, 2021 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

kmdeck commented Jun 22, 2021 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

LenkaNovak commented Jun 24, 2021

jb-mackay left a comment

Choose a reason for hiding this comment

kmdeck left a comment

Choose a reason for hiding this comment

LenkaNovak commented Jun 7, 2021 •

edited

Loading

jb-mackay Jun 21, 2021 •

edited

Loading

kmdeck Jun 22, 2021 •

edited

Loading

codecov bot commented Jun 22, 2021 •

edited

Loading

kmdeck Jun 22, 2021 •

edited

Loading

kmdeck commented Jun 22, 2021 •

edited

Loading