Use global MODIS LAI for all simulations; remove modis_lai_fluxnet_sites artifact

[juliasloan25]

Purpose

We've been using 2 different MODIS LAI artifacts: one global map for global runs, and one dataset for 4 fluxnet sites to use for site-level simulations. After #1279, we have the ability to read in spatial maps to Point/Column spaces, so we can use the global map for global and site-level runs.

This PR switches all site-level (fluxnet) runs to use the global map, and removes the site-specific LAI artifact.

Closes #1224

Content

• rename find_modis_year_paths -> modis_lai_multiyear_paths
• use modis_lai_multiyear_paths instead of modis_lai_single_year_path (if we have multiple years of data we may as well use it since this artifact isn't too large). Single year artifact still exists and is tested
• modis_lai_multiyear_paths only includes an extra year of data if the end date is in December, instead of always including an extra year of data
• in modis_lai_multiyear_paths, remove PeriodicCalendar extrapolation from LinearInterpolation. Since the data spans multiple years, we get uneven time intervals (e.g. between last date of one year and first date of the next year) so we can't use PeriodicCalendar with it
• rename final_date -> end_date in src/Artifacts.toml for consistency
• always pass longlat to domain for fluxnet simulations (this required some restructuring of drivers)

LAI values at the start of the simulation

site ID	site-level LAI artifact (main)	global LAI artifact (this PR)
US-MOz	0.320855	0.343325
US-Ha1	0.594469	0.532508
US-Var	.0962558	1.0596
US-NR1	0.488428	0.40729

Note that the values from the global LAI artifact are identical whether we have use_lowres_clm true or false.

Changes in results

For most variables output from the 4 fluxnet sites, the R-squared value compared to data remained the same (variables are LHF, SHF, LWU, SWU, GPP, ET). RMSD values mostly decreased (suggesting better agreement with data) but RMSD changes range from 2.3% increase (Ha1 GPP) to 20% decrease (Ozark GPP). Overall it seems like this change slightly improves results, but does not have a strong effect.

---

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

[juliasloan25]

I moved this into the driver scripts so that we have access to lat and long when we construct it. This file will be removed soon in a different PR, so it doesn't matter that this is duplicating some code.

[ph-kev]

I am not sure if I am the best person for reviewing this, but everything looks reasonable to me.

[ph-kev]

I find it strange that start_date isn't actually the start date.

[juliasloan25]

Yeah I agree. I think we used to call it ref_date and it may be clearer if we go back to using that. But this is the convention we use throughout ClimaLand right now

[kmdeck]

could we maybe discuss this more? I think we should get rid of t0 and tf and only work with start and stop_date

[kmdeck]

(at least in fluxnet runs; this is also what we do in global runs)

[imreddyTeja]

Some of the results are slightly changed, which is probably expected.

Have you checked how much/if the LAI values differ for the fluxnet sites when using modis_lai_fluxnet_sites vs the global modes artifact?

The software changes look good to me

[juliasloan25]

Some of the results are slightly changed, which is probably expected.

Have you checked how much/if the LAI values differ for the fluxnet sites when using modis_lai_fluxnet_sites vs the global modes artifact?

I added a table to the PR description showing the LAI values at the start of the simulation for each site, using the site-level artifact vs the global artifact. They look pretty similar to me. I didn't go through and check the values over time, but I think this is a good enough check to see that the values are reasonable.

[juliasloan25]

I realized that the way I'm computing maxLAI doesn't recover the previous behavior. Since we evaluate the TVI at a point at the start date to get LAI_init, taking maximum(LAI_init) just returns the initial value at that point. I can't find a straightforward way to take a maximum of the TVI values over time, and since we use the global artifact now it also isn't very straightforward to take a maximum of the dataset values only at this point in space. We would have to find the index of the lat/lon in the dataset that are closest to our space, and take a maximum over all values at that point over time. This would defeat the purpose of using ClimaUtilities to handle the interpolation and file reading.

I would prefer to stop using maxLAI to set RAI and capacity - maybe we can just use the initial LAI instead? But I'd like to discuss this with Kat before making the change, so I'll wait to merge until next week