leaf biophysics refactor and functional unit tests

[rgknox]

Description:

Leaf biophysics (photosynthesis, stomatal conductance and leaf respiration) has been refactored. Major objectives in this refactor were to 1) extract relevant routines that occur at the leaf-scale and place them in their own module that does not use the fates scaling data structures (cohort, patch and site agnostic) and 2) to have functional unit tests that exercise each of these functions.

These methods do not yet use features in our other brand of unit tests (but could!), particularly the use of cmake and parameter reading.

Design Notes

Notes on Bisection and finding bounding (starter) values

Notes on re-deriving the Medlyn solve with a boundary layer

Colab Notebook that runs the unit test: link

Fixes: #1222

Collaborators:

@adrifoster @ckoven

Expectation of Answer Changes:

There may be subtle differences in our standard integration tests. I do not expect the same results within round-off, but the diffs should be looked through to see if anything is very large.

Checklist

If this is your first time contributing, please read the CONTRIBUTING document.

All checklist items must be checked to enable merging this pull request:

Contributor

• The in-code documentation has been updated with descriptive comments
• The documentation has been assessed to determine if updates are necessary

Integrator

• FATES PASS/FAIL regression tests were run
• Evaluation of test results for answer changes was performed and results provided

Documentation

• Technical Note update:
• User's Guide update:

Test Results:

TBD

CTSM (or) E3SM (specify which) test hash-tag:

CTSM (or) E3SM (specify which) baseline hash-tag:

FATES baseline hash-tag:

Test Output:

[rgknox]

Full model regression testing is now showing nigh-indistinguishable results with base:

Test description: BCI, inventory initialization, ST3, 5 years, 2-stream rad, 1 tropical evergreen PFT, parameter defaults otherwise:

https://drive.google.com/file/d/1HJadTpwo_MAdL5Sr8qWK4EFV98WPTlXy/view?usp=sharing

Removing draft status

Next steps:

1. evaluate gridded comparison while still retaining code that ensures near-b4b?
2. incrementally remove b4b preservation code and evaluate differences
3. clean up comments
4. evaluate timing differences (particularly related to ci precision)
5. introduce updates to btran (ie remove from stomatal intercept and add to Jmax)?

[rgknox]

I wanted to drop a note about expected changes with this refactor. There are a few places where results are changed with intent:

1. we previously would abort the solution for Ci (intracellular CO2) if we encountered a negative Anet. This presumably was because negative Anet generates trivial conductance (intercept) which is a trivial solution. However, if the solver is not given a chance to converge properly, the next updates to Ci could very well generate positive Anet. In the refactor, we remove the premature abort on solution
2. In the base model, we approach trivial solutions to cases with no leaves and no light kind of backwards. We see if the plant has light and then if the plant has leaves
3. the convergence tolerance on Ci in master (EDIT...corrected) is pretty small, but finishes its search after 5 iterations and accepts the answer regardless of convergence.

I think for many cases, the old solution that we have in main actually does get us a reasonable solution, but it breaks down and gives poor results in extreme conditions (particularly very dry conditions). This is probably due to a combination of points 1 and 3, and that we didn't offer a highly-stable alternative to the simple recursive numerical search of Ci (we now use bisection if a solution has not been achieved in 10 tries)

---

I show two plots. Each plot looks at conductance, assimilation and Ci over a range of temperatures at BCI, for 4 combinations of absorbed PAR (high and low) and humidity (high and low). They both assume a boundary layer conductance of 2.5 mol/m2/s, which is on the high side for is experienced at BCI. They both use Medlyn conductance, and default parameters.

*Note the turquoise line in the Ci plots is the compensation point.

Using original code and Ci tolerance:

Using new code and updated Ci tolerance:

---

Here is the same comparison using Ball-Berry:

Original code:

New Code: