Addition of resilience and recovery

[mcgregorian1]

Hi @teixeirak

The good news is maybe this won't be too bad? Don't want to jinx anything though...

First thing is that we need to decide how to deal with outliers. With resistance, we filtered out all values >2. As you can see below, the numbers vary per metric. To me it seems for recovery and resilience we'd want to filter out everything >3. What do you think? Do we have a biological reason for choosing a different threshold?

[teixeirak]

Sure. That should be fine.

You're right; I don't think it will be too bad.

[teixeirak]

@mcgregorian1 , I looked through the files you added, but don't see if with the top model coefficients (curious to see how it relates to height!). It also looks like you haven't yet implemented the decision in #133.

I'm quite curious to see what results we get. I think this will end up being a very nice addition / really enriching the story.

[mcgregorian1]

Sorry, you're right! The overall re-implementation of the code is not difficult, but I'm running into small things (making tables match, accommodating different top variables, etc) that are taking a little bit of time. I'll comment here when I've fully fixed things and added all the files - I'll also need to redo the resistance files as well since Reviewer 3 wanted marginal + conditional R2.

[mcgregorian1]

@teixeirak I've updated all files for each of the metrics, but I haven't transferred the values yet to Tables 4 and 5 for recovery and resistance. I can get to these tomorrow or Tuesday evening.

[teixeirak]

Many thanks, @mcgregorian1! I'm starting to digest these results.

A few key points: 1- both recovery and resilience have higher marginal R2s than resistance. That surprised me, but indicates that the results of these analyses are probably worth highlighting more than in an obscure appendix table. 2- height did not have consistent effects in the models for either recovery and resilience. For both metrics, there is no drought year (or combo) where height is in all top models with consistent coefficient. It therefore seems that it should be dropped from the models for these metrics. 3- The same is true of TWI (and TWI x height) 4- diffuse porous are much more resilient/ have higher recovery than ring porous. 5- higher (less negative) TLP --> lower resilience and recovery 6- higher PLA --> lower resilience

My current questions/ uncertainties: 1- what do we get when we remove TWI and height ? 2- It would help to see versions of Figs 1b, 3, and 4 for these metrics. I think its helpful to look at these and understand what's going on before we dive into incorporating them. 3- what is the context for this? I've never paid particular attention to resilience and recovery, and don't have the context for interpreting (have others looked at this? what have they found? what would we expect?) 4- how do we want to present this in manuscript? A minimum would be to give brief mention in the text and stick a couple tables in the appendix. A maximum would be to fully integrate these into abstract, hypotheses, etc. Given the key points summarized above, I'd lean towards the latter.

To fully integrate these, we'd need to: 1- add some hypotheses to table 1 (not bad) 2- add these metrics to table 3 (easy) 3- add density plots for resilience and recovery to Fig. 1 (not bad, but tricky space-wise) 4- add Rc and Rs to Fig. 3- not bad, but makes for a very full figure. 5- make Fig. 4 into a 3 row (Rt, Rc, Rs) x 5 col (ln[H], TWI, RP, PLA, TWI) grid. 6- add some tables, etc to the SI. 7- adjust the text throughout.

[teixeirak]

As you'll see, I just wrote a few of the coauthors who can be most helpful in interpreting these results.

[mcgregorian1]

My current questions/ uncertainties: 1- what do we get when we remove TWI and height ?

By considering height and TWI to be part of the "base" model, we were only looking at the effects of the leaf traits. Thus if we remove TWI and height, then the only thing we're left with is the top variables themselves (rp/tlp/pla). In other words, resistance ~ tlp + pla + rp + (1|sp) and same for recovery but minus pla.

2- It would help to see versions of Figs 1b, 3, and 4 for these metrics. I think its helpful to look at these and understand what's going on before we dive into incorporating them.

That can be done.

3- what is the context for this? I've never paid particular attention to resilience and recovery, and don't have the context for interpreting (have others looked at this? what have they found? what would we expect?)

I honestly haven't seen this mentioned much either, but I also haven't explicitly looked for this. I imagine this would take some searching through all papers that have cited Lloret et al.

4- how do we want to present this in manuscript? A minimum would be to give brief mention in the text and stick a couple tables in the appendix. A maximum would be to fully integrate these into abstract, hypotheses, etc. Given the key points summarized above, I'd lean towards the latter.

Based on the implications of this, I agree. I think we have to be cognizant of how much time everything will take. I agree I don't think the code should be that bad in terms of timing, but due to homework and such I won't be able to dedicate full evenings toward this like I did in July. I'll be interested to see what our coauthors say

[teixeirak]

My current questions/ uncertainties: 1- what do we get when we remove TWI and height ?

By considering height and TWI to be part of the "base" model, we were only looking at the effects of the leaf traits. Thus if we remove TWI and height, then the only thing we're left with is the top variables themselves (rp/tlp/pla). In other words, resistance ~ tlp + pla + rp + (1|sp) and same for recovery but minus pla.

Right. We do want to retain tables with the full results, as we still want to test hypotheses on height and TWI, but I don't think we want to allow them in the top models that we plot.

[teixeirak]

3- what is the context for this? I've never paid particular attention to resilience and recovery, and don't have the context for interpreting (have others looked at this? what have they found? what would we expect?)

I honestly haven't seen this mentioned much either, but I also haven't explicitly looked for this. I imagine this would take some searching through all papers that have cited Lloret et al.

I will work on this.

[teixeirak]

4- how do we want to present this in manuscript? A minimum would be to give brief mention in the text and stick a couple tables in the appendix. A maximum would be to fully integrate these into abstract, hypotheses, etc. Given the key points summarized above, I'd lean towards the latter.

Based on the implications of this, I agree. I think we have to be cognizant of how much time everything will take. I agree I don't think the code should be that bad in terms of timing, but due to homework and such I won't be able to dedicate full evenings toward this like I did in July. I'll be interested to see what our coauthors say

I don't think it will be too bad. Of course, we have to keep it to something we can finish by the end of the month, when this is due back.

[mcgregorian1]

3- add density plots for resilience and recovery to Fig. 1 (not bad, but tricky space-wise)

I've added standalone Fig 1b's for each metric along with a separate Fig 1 for each so you can see them. You're right, I'm not entirely sure...We could do 2 columns, with the timeseries in 1 and then have each density plot on top of each other in the second?

4- add Rc and Rs to Fig. 3- not bad, but makes for a very full figure.

I'm not sure this will be possible. I think the best I could do with the current format it's in is make a 3-row figure, but the text would be super small. Otherwise if I try to combine them all into one I feel that it will be too busy. Maybe this is one where we show one metric and have the other two be in SI?

5- make Fig. 4 into a 3 row (Rt, Rc, Rs) x 5 col (ln[H], TWI, RP, PLA, TWI) grid. Right. We do want to retain tables with the full results, as we still want to test hypotheses on height and TWI, but I don't think we want to allow them in the top models that we plot.

I've done this and also made each one separate. I'm not sure what to think about the 3x5 figure grid. What are your thoughts?

• Also, for Rc and Rs - let's say the top model for 1966 for Rc had height.ln.m in it. Did you mean I should ignore that and take the next best top model that doesn't have height.ln.m in it? Or did you mean to use that model but take out height?
  • In other words, the model was recovery ~ height.ln.m + rp + (1|sp). For the figures I uploaded tonight, I removed height to give me recovery ~ rp + (1|sp). As I was typing this I realized this is probably not the way to go, but the code is now written so changing it shouldn't take too long.

[teixeirak]

@mcgregorian1, thanks for all this! I broke out a few more specific new issues (#136, #137, #138) to make all of this easier to track. I'm still digesting everything, but starting with some bigger-picture thoughts.

First, the 3 metrics are related, such that if we know 2, we can calculate the third (e.g., Rs= Rt x Rc). If resistance is low for a certain set of trees, we know that either recovery is high or Rs is low. This has a couple important implications:

(1) we don't necessarily need to highlight all 3 to give the full picture. Of course we present Rt. I'd then see Rs as a higher priority than Rc, as Rs tells us the longer-term consequence of the drought. I think we should make available the results for all 3, but I'd probably focus on just Rt and Rs in Figs 1 (#136), 3 (#137), and maybe 4 (#138).

(2) We expect results to "add up". For example, Rt decreases with H, so its impossible that both Rs and Rc are truly independent of height. Indeed, when I went back and looked more carefully, removing models including the TWI x H interaction, Rs generally decreases with H (...nooooo! That's not good news!). Recovery is mixed, decreasing with H overall in all years and in 1977, more-or-less independent in 1966 and 1999.

[teixeirak]

Point 2 above implies that we probably want to include the same candidate terms for all 3 metrics (#133). I'll comment more under that issue.

[teixeirak]

Re-posting the list above of what's needed to incorporate Rs and Rc:

• [x] 1- add some hypotheses to table 1 (not bad)
• [x] 2- add these metrics to table 3 (easy)
• [x] 3- add density plot for resilience to Fig. 1 ( #136 )
• [x] 4- add Rs to Fig. 3 (#137),
• [x] 5- make Fig. 4 into a 3 row (Rt, Rc, Rs) x 4 col (ln[H], TWI, PLA, TWI) grid (#138)
• [x] 6- add some tables, etc to the SI.
  • [x] create density plot for Rc (fig. 1b parallel)
  • [x] add Rc density plot to appendix
  • [x] create Rc Fig.3
  • [x] add Rc Fig. 3 to appendix
  • [x] create parallel of Table S4 for Rs and Rc
  • [x] create parallel of Table S6 for Rs and Rc
  • [x] add S4/S6 parallels to SI
• [ ] 7- adjust the text throughout, including references (#135)

[mcgregorian1]

• [x] 6- add some tables, etc to the SI.

  • [x] create density plot for Rc (fig. 1b parallel)
  • [x] add Rc density plot to appendix
  • [x] create Rc Fig.3
  • [x] add Rc Fig. 3 to appendix
  • [x] create parallel of Table S4 for Rs and Rc
  • [x] create parallel of Table S6 for Rs and Rc
  • [x] add S4/S6 parallels to SI

@teixeirak did you mean SI when you said appendix here? I wasn't sure since we do technically have an appendix.

As it stands, though, everything is ready to be added.

[teixeirak]

Yes, I meant SI

[teixeirak]

@mcgregorian1 , I think all the analyses/ tables/ figures are now good, with a bit of minor cleanup as noted in other issues. The main thing that remains is changing the text. I've worked on the intro but have barely touched the methods, results, and discussion.

[mcgregorian1]

Ok, thank you! I'll start looking at more of the writing tomorrow morning.

[teixeirak]

I'll start looking at more of the writing tomorrow morning.

Great! I just did a bit of work on this in the draft. A few notes: 1- Our previous draft was 10 words under the limit, so we do not have space to add content without cutting. Please aim to keep the length of each section similar. 2- In the results section, I think we'll need to condense/cut some of the details about the individual year scenarios. Another target for cutting may be Rt_arima results. 3- In general, I've been using the term "drought tolerance" as an umbrella for Rt, Rc, and Rs, as defined in the first paragraph. When we're talking about a specific metric, we can use it's symbol.

[teixeirak]

I think we can call this done.