consider more functional traits

[teixeirak]

@mcgregorian1, I think it would be worth adding some more functional traits to your analysis, but without going overboard.

We now have more functional traits for SCBI available in our hydraulic traits repo. It would be interesting to add some others: desiccation shrinkage (likely to be a meaningful trait); wood density, SLA (commonly analyzed traits that are probably less exciting/ less likely to be strong predictors).

Note that other traits may be available through TRY and/or NEON.

[mcgregorian1]

Ok. What traits were you thinking? I've heard that leaf area index is commonly used as a species trait, but I only see leaf area (fresh/dry) here. Would the wood density be from another dataset (I don't see it on the full traits table)?

[teixeirak]

Wood density should be there.

[teixeirak]

@mcgregorian1, this is the data sheet you'll want to look at for SCBI trait measurements. Variables that may be interesting include:

• [x] PLA_dry_percent (but see this issue) - check with Valentine that this is fixed
• [x] LMA_g_per_m2
• [x] mean_SPAD or Chl_m2_per_g (need @nkunert to advise on the difference)
• [x] WD (wood density)
• [x] hydraulic safety margin (based on hydraulic vulnerability curves, which we have for a subset of species)- I can provide details later.
• [x] traits from NEON

[mcgregorian1]

NEON traits that I think may be useful:

• canopy lignin/nitrogen/water content: good for seeing how canopy specifically is reacting in a particular year. Cons are that this data is only recent, and doesn't include suppressed / random intermediate trees
• ecosystem structure: apparently this has heights of the canopy. Cons is that it's not for specific trees.
• LAI: good for a snapshot, but again, only helpful for current year. Possibly this ratio could be mathematically sourced to our current DBH measurements, which means in that case we could predict LAI in the past drought years, but I'm not sure if that's sound.
• total biomass map: I'd say about the same reasonings as LAI
• heights: NEON has woody plant vegetation structure, which may work. I've noted this in #25

Overall, though, it seems iffy about using other data simply due to them being snapshots that I'm unsure I can extrapolate out to the past.

[teixeirak]

I can't look in detail right now, but a quick comment-- Probably none of the above (besides tree heights). I was thinking of Plant foliar physical and chemical properties and Plant foliar stable isotopes.

[mcgregorian1]

Other things to add in to model:

• [ ] HMS
• [x] SLA

HMS is the hydraulic safety margin, defined as "the minimum water potential observed in a given species minus either the P50 or P88 value" from Anderegg et al 2016. In other words, the water potential at min minus the water potential at 50 or 88. "P50 and P88 are defined as the water potentials at which a given species loses 50% and 88%, respectively, of hydraulic conductivity in a given tissue".

• HMS was determined to be the main predictor of drought-induced mortality across species
• Anderegg et al 2018 also found this using HMS50 (not using P88), plus LSA (leaf specific area). " the machine-learning algorithm consistently indicated that the standard deviation of HSM was one of the most important predictor variables...We found that higher diversity in the HSMs of the species in an ecosystem significantly buffers ecosystem latent heat flux response to drought."
  • P50 data can be found in the Xylem Functional Traits database based on this paper
  • it should be mentioned that Hartman et al 2018 say P50 is generally death for gymnosperms while P88 is death is angiosperms, but this is refuted by Anderegg's papers and Greenwood 2017.
  • SLA can be found here

SLA also used by Greenwood et al 2017, where lower SLA showed lower mortality responses

• specifically didn't use PDSI to determine drought because this other method (SPEI) is better.
• in the results, they sepecifically discuss the reason why height may not come out as significant in models looking at traits

For the record, Hartman et al 2018 (linked above) discuss research areas of need as opposed to they tested certain things and then determined certain results.

For Trugman et al 2018, I'm not sure we can use any existing data for our trees to confirm what the paper did. In theory we would need xylem biomass and leaf biomass if I'm understanding it correctly, but this would require another round of figuring out data.

• the maximum survivable xylem damage (MSXD) a tree could undergo during drought increased under conditions with more available water and more available CO2. MSXD decreases as tree size increases, and larger trees with survivable xylem damage took longer to recover than smaller trees, because returning to an optimal ratio of xylem to leaf biomass takes longer (ratio increases with tree size).
• based on current available data and modeling, this paper suggests post-drought carbon allocation = proxy for tree fitness (very good explanation in the discussion). Rebuilding damaged xylem increases whole-plant C gain, so if this takes a long time, trees can die from C starvation
• in general, mortality should have peak at drought event and then smaller peak later due to lag

[mcgregorian1]

I've checked the HSM data from the TRY database. They only have water potential data for ("qual" "quru" "cagl" "fram" "qupr" "litu"), which are the same ones we have the curves for from Nobby's calculations. However, the TRY database only has enough data to give a HSM estimate for "qual" and "quru."

I also checked SLA, which has data for 11 species. Taking those averages, I did a rough run of the models and it does not appear in the top 5.

[teixeirak]

Thanks for checking TRY. For those traits (HSM and SLA), its preferable to use our own data.

[mcgregorian1]

Krista and I just spoke, and these were the notes:

• [x] Run the model with no individual traits, assuming these are covered by including the "sp" effect. See what comes out of it.
• [x] SLA is the inverse of LMA. We already have LMA specific to SCBI, so take out SLA.
• [x] don't worry about getting NEON chemical data. We need to keep moving progress forward on this.
• [ ] Get HSM. To do this, need P50.
• [x] Make issue in hydraulic traits repo, assign to krista, ask for guidance on which functions to use for calculating P50. Do we use tlp as the min water potential? Not mentioned in Anderegg.
• [x] Make issue, ask nobby about difference for Chl/SPAD
• [ ] the main traits effects we want to keep in the model are tlp, rp, HSM, LMA, PLA, wood density, and mean SPAD/Chl (we have decided to not include sap_ratio)
• [ ] the main biophysical effects we want in the model are position_all, height.ln.m, distance.ln (to water), and elev.m. Otherwise, also have year and species

[mcgregorian1]

For the record, running the model with everything included gives the following for the top 10.

[mcgregorian1]

Running the model with no hydraulic traits (keeping only position, height, elevation, distance to water, year, and species) yields the following for top 10. This follows what we were finding in the earlier model runs.

[mcgregorian1]

These are the coefficients for the best model (# 8 above)

And these are for the full model (# 16 above). The main difference is that elev and distance have a positive coefficient.

[mcgregorian1]

I've run the model with all leaf traits (with HSM (using TLP as a proxy)), and this is what I'm seeing.

[mcgregorian1]

Running the traits-only model with p50 and p80 (calculated as of 11 june), gives this, in both instances Chlorophyll dominates the influence.

Edit as of 12 June: adding HSM (from the equation given for Pmin by Zhu in an email) does not come out in the top model. In fact, the top two models do not change.

[mcgregorian1]

Hi @teixeirak. I've checked the NEON data looking at 2018 (for an example), and according to this, there's really not much variation between the different heights. The first graph shows the temperatures recorded at each of the heights (10,20,30,40,50m), and the second graph shows them all overlaid. I'm not sure there's anything significant here that we could pull out.

[teixeirak]

Let's drop Chl. Reasons are (1) I don't see a clear biological mechanism as to why it would affect drought response, (2) @nkunert mentioned that its a fairly uncertain measurements, and (3) the coefficeint goes opposite directions in the two top models listed above.

[mcgregorian1]

Is it possible that biologically, Chl is a kind of proxy for total leaf area? Thus, it kind of subsumes PLA and LMA, since the measure of Chl in a certain species is related to how much light the species can take in, which affects growth patterns? Or am I missing something there?

[teixeirak]

Let's treat this as the final traits list:

Ring Porosity Percent Leaf Area Leaf Mass Area Wood density TLP P50 P80

This is what's in your current list in the manuscript, but with chlorophyll removed.

[mcgregorian1]

Sounds good, of course noting P50 and P80 will most likely only make an appearance in the supplementary info

[teixeirak]

I think we should have a table testing each trait individually (issue #36 ), and they can show up there. They just won't be in the final best model.

[teixeirak]

Closing (obsolete).