Global run of complex land model with ERA5 forcing

[kmdeck]

Purpose

Carry out global run of soil+canopy model on GPU and CPU

To-do

• Future PR: add in runoff. This requires altering our land model constructor, which will then involve changing a number of files more (to supply runoff instead of using the default). I will follow up this PR with a quick one adding that.

Content

1. Previously, we had only ever simulated Richards model globally. Because the soil properties vary spatially, certain functions that are only used by EnergyHydrology needed to be updated to be broadcastable. The primary change is because the parameter struct contains fields, but is not a field of a parameter structs. this means that functions that we broadcasted over a Ref of the parameter struct before now must be supplied each parameter field as an argument. This led to a LOT of changes (volumetric_heat_capacity, volumetric_internal_energy, volumetric_internal_energy_liq, phase_change_source, soil_resistance)
2. Parameters that were marked as FT in the EnergyHydrologyParameter struct now are a unionof FT and Field. Some parameters are only defined on the surface, rather than throughout the soil, so they need a different parametric type
3. added benchmark - on GPU
4. added experiment - GPU/CPU

Note: For the benchmark - currently about 1/4 of the time is in setup rather than solve, so maybe we need to change the setup of the simulation. however, making it longer leads to failures, so Im not sure yet.

Note: The experiment does not run on GPU in CI - on caltech - , but it does on GPU on clima (benchmark). For now I removed the run on caltech

---

• I have read and checked the items on the review checklist.

[juliasloan25]

This is awesome, thank you Kat! It would be nice to reduce some of the code duplication between the land.jl and global_soil_canopy.jl drivers (and maybe for the other benchmark runs too), but this is fine for now :)

[juliasloan25]

LGTM! Thank you for addressing the review so thoroughly :) I left a few more minor comments

[kmdeck]

LGTM! Thank you for addressing the review so thoroughly :) I left a few more minor comments

thank you for the quick and thorough review!

[Sbozzolo]

Suggested change

	K_sat_plant = 5e-9 # m/s # seems much too small?
	K_sat_plant = FT(5e-9) # m/s # seems much too small?

[Sbozzolo]

Are there FT missing here?

[kmdeck]

youre right! I will go through the file and update it in this PR #669

[Sbozzolo]

There's probably a FT missing around 0.0

[kmdeck]

we convert the BC into the FT type internally here:

ClimaLand.jl/src/standalone/Soil/Biogeochemistry/Biogeochemistry.jl

Line 456 in 0683b81

return FT.(bc.bc(p, t)) .+ FT.(ClimaCore.Fields.zeros(axes(Δz)))

(these are very allocation heavy and we need to update this code anyways!)

[Sbozzolo]

Typically, you don't need FT when you .= into a Field

[Sbozzolo]

Maybe using round is better?

[kmdeck]

Youre right, but I think LAI data is updated every 8 days, so floor vs round makes no real difference here. I am leaving this as is since we will remove it entirely when the bug fix is in.