Allow n-band radiative transfer

[Espeer5]

Purpose

This PR extends our radiative transfer functionality from 2 bands (PAR and NIR) to an n-band scheme. This will allow pairing with ClimaAtmos which provides 16 bands. Represents step one towards #876.

This PR must be a large number of lines due to updating of all of the experiments and tests.

To-do

• Update fundamental radiation structs
• Update radiative transfer computations in models
• Repair all experiments and tests to work with new scheme
• Add a test for using up to 16 bands of radiation
• Check benchmarks and long runs to ensure no significant regressions
• Verify GPU compatibility is not broken
• Restore windows-latest to tests when working

Content

• Adds a tuple-based type for the discretization of the λ spectrum
• All leaf & soil optics types are converted to tuple-valued fields
• Uses solar irradiance curve to allocate incoming shortwave to radiation bands
  • Also to compute PAR quantities from spectral absorption for photosynthesis
• Updates all tests and experiments to work with new structs / types

---

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

[Espeer5]

Can anyone help me out with the GPU compatibility issue here?

It's a bit odd to me because the same broadcast structure is used over ntuple in canopy_sw_rt_beer_lambert as in the canopy_sw_rt_two_stream, but only one of them seems to cause the issue. I'm a little stumped by it.

[Sbozzolo]

I can help with the GPU compat, but I have a couple of other things that I need to finish first.

[juliasloan25]

Can anyone help me out with the GPU compatibility issue here?

It's a bit odd to me because the same broadcast structure is used over ntuple in canopy_sw_rt_beer_lambert as in the canopy_sw_rt_two_stream, but only one of them seems to cause the issue. I'm a little stumped by it.

I expect that somewhere in canopy_sw_rt_two_stream you're broadcasting over floats, which will error if it's inside a GPU kernel (which this function will be since it's called with @.). E.g. see this thread. It's hard to tell what inside that function is a Tuple vs a float though, so I can't tell exactly where the issue is. Since this function is a lot more complex than canopy_sw_rt_beer_lambert there's more space to make an error in the broadcasting.

[juliasloan25]

Is it possible to apply canopy_sw_rt_two_stream for each element of \alpha_soil, \rho_leaf, \tau_leaf (just broadcasting normally) instead of wrapping them in tuples to get around the broadcasting?

[Espeer5]

Is it possible to apply canopy_sw_rt_two_stream for each element of \alpha_soil, \rho_leaf, \tau_leaf (just broadcasting normally) instead of wrapping them in tuples to get around the broadcasting?

Possibly. I was originally doing this but had some issues handling it correctly when α/ρ/τ were tuple-valued-fields vs. plain tuples. But you're right this must be the issue. Thanks