clarify how we select drought

[teixeirak]

I'm not sure its appropriate to include 1964 and 1965 at all. Looking back to issues #5 and #11, its not clear that 1965 or even 1964 should be included at all. Here's an old table (originally posted here) showing n sequences indicating pointer year (at 30% threshold): . 1965 does not show up at all, and 1964 shows up as a pointer year for only two chronologies. Here it looks like 1965 also had 2 chronologies with pointer years. I think that 1964 came in through the discussion in issue #11 (looking back to dry periods identified in Helcoski et al.). All this, combined with my efforts to describe how we defined drought as I was working on the methods description this AM, points to the need to be absolutely clear and systematic about how we define drought. Why do we include 1964 and not 1991? (From Fig. 1, it looks like 1991 should qualify....)

_Originally posted by @teixeirak in https://github.com/SCBI-ForestGEO/McGregor_climate-sensitivity-variation/issues/53#issuecomment-521733601_

[teixeirak]

@mcgregorian1, I'm starting a new issue on this because its a big enough topic to warrant its own discussion. As we're dealing with issue #53 (and changed results), it makes sense to go back and ensure that we're really happy with the drought definition.

Here's what I wrote in the methods: "We identified droughts within the time period 1950-2009, defining drought [@slette_how_2019] as events where peak growing season climatic conditions were anomalously dry relative to the long-term record and where tree growth was substantially reduced for at least some components of the community."

That is the basic concept that we want. I'd like to make sure our more detailed process for selecting the droughts is completely systematic, as our past effort was less clearly defined.

[teixeirak]

One side is to identify the years during which a significant portion of the ecological community suffered ≥30% reductions in growth. For the droughts that we've been considering, this threshold is met by 26-32% of individuals. I therefore think the 50% threshold that we've been using is too high to be very meaningful. We can work with the source tables (like this, referenced here: )

However, it would be much more straightforward (and scientifically justified) to just put all the species and canopy classes together. I think we discussed this once before, and it would require copying all of the cores over to a single .rwl file. If its faster to just do the math on existing output files like the one above, we could do that. (Where are those output files? I'm not sure if you've loaded them to GitHub.)

Bottom line here: let's identify the years in which >25% of tree suffered >30% growth reduction. If that's really a pain, we could continue to judge based on the number of tree groupings for which pointer years are identified, but that's somewhat arbitrary and hard to explain in the methods.

[mcgregorian1]

Ok sounds good.

Those rwl files are in this repo in the data folder under cores. I'm not sure if it would be faster to do the math on the tables we have already or reload the cores (probably reloading), but I'm going to have to take some time before re-checking all that due to the move.

---

From: Kristina Anderson-Teixeira notifications@github.com Sent: Thursday, August 15, 2019 2:17:51 PM To: SCBI-ForestGEO/McGregor_climate-sensitivity-variation McGregor_climate-sensitivity-variation@noreply.github.com Cc: McGregor, Ian McGregorI@si.edu; Mention mention@noreply.github.com Subject: Re: [SCBI-ForestGEO/McGregor_climate-sensitivity-variation] clarify how we select drought (#54)

@mcgregorian1https://nam02.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com%2Fmcgregorian1&data=02%7C01%7Cmcgregori%40si.edu%7Cd69eb9225e4748a9572b08d721ace30b%7C989b5e2a14e44efe93b78cdd5fc5d11c%7C0%7C0%7C637014898741933489&sdata=EvNDhmcziRWQ3uCyWe%2FZttVIXX3GKPdQhJa%2F8cYmyNk%3D&reserved=0, I'm starting a new issue on this because its a big enough topic to warrant its own discussion. As we're dealing with issue #53https://nam02.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com%2FSCBI-ForestGEO%2FMcGregor_climate-sensitivity-variation%2Fissues%2F53&data=02%7C01%7Cmcgregori%40si.edu%7Cd69eb9225e4748a9572b08d721ace30b%7C989b5e2a14e44efe93b78cdd5fc5d11c%7C0%7C0%7C637014898741933489&sdata=L63O0aaDMm2uStaQJSG4fPPNorwGQq%2F5jGjF2tCfGME%3D&reserved=0 (and changed results), it makes sense to go back and ensure that we're really happy with the drought definition.

Here's what I wrote in the methods: "We identified droughts within the time period 1950-2009, defining drought [@slette_how_2019] as events where peak growing season climatic conditions were anomalously dry relative to the long-term record and where tree growth was substantially reduced for at least some components of the community."

That is the basic concept that we want. I'd like to make sure our more detailed process for selecting the droughts is completely systematic, as our past effort was less clearly defined.

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHubhttps://nam02.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com%2FSCBI-ForestGEO%2FMcGregor_climate-sensitivity-variation%2Fissues%2F54%3Femail_source%3Dnotifications%26email_token%3DAJNRBEMPYKKHZF6RWLZPPDLQEWMU7A5CNFSM4IMAR2KKYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOD4MSTIA%23issuecomment-521742752&data=02%7C01%7Cmcgregori%40si.edu%7Cd69eb9225e4748a9572b08d721ace30b%7C989b5e2a14e44efe93b78cdd5fc5d11c%7C0%7C0%7C637014898741943485&sdata=Qu4dsnamPOrMOQsCBMh%2Fgc4pvNlRaFuPq1Wija6Ag1g%3D&reserved=0, or mute the threadhttps://nam02.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com%2Fnotifications%2Funsubscribe-auth%2FAJNRBEOT2EUHC6XKYZIPFQ3QEWMU7ANCNFSM4IMAR2KA&data=02%7C01%7Cmcgregori%40si.edu%7Cd69eb9225e4748a9572b08d721ace30b%7C989b5e2a14e44efe93b78cdd5fc5d11c%7C0%7C0%7C637014898741943485&sdata=bQSWIEw2Nx81qcJ67DFfB77gP%2F8D0wkQpOgy7prVefw%3D&reserved=0.

[teixeirak]

I have combined them for you (this was easy). The file is here.

[teixeirak]

The second aspect of selecting droughts is verifying that the climate was anomalously dry. I think this is tricky because every drought is different. The two metrics we have on hand are May-August PET-PRE and PDSI. They don't always seem to tell the same story, which makes sense considering that droughts can be very different in nature.

I think the best approach is what we did previously-- identifying years when lots of trees suffered reduced growth and then verifying that those were anomalously dry. As a reminder, we have a plot of PDSI that you made here:

[teixeirak]

I have combined them for you (this was easy). The file is here.

@mcgregorian1, please prioritize running this when you have a chance, and save the output listing % trees with >30% growth reduction. I'll then think about how best to select droughts.

[teixeirak]

I just spent a bit more time looking at the NOAA data and thinking about options.

• In terms of PDSI, 1966, 1977, and 1999 were the driest (1991 got drier in the fall, but that would be after tree growth stopped). Specifically, they had the lowest PDSI’s of the study period (1950-2009) in May, June, July, and August (each ranked separately). 1991 ranks 10th for May-July, and 4th for August. They were also 3 of the 4 driest in terms of mean MJJA [PET-PRE]. Based on this, I think they make most sense from a climatological standpoint.

• 1991 was coming out as a marker year, and a drought, but objectively doesn’t seem to live up to these others. It may have reduced growth because the previous 2 years were wet, and high moisture during the previous growing season reduces growth (not sure about 2-yr lag). There may also be some lagged effect of gypsy moth. Bill McShea remembers the worst year at SCBI being in 1988 or 1989, but they were definitely in the area in 1991 (based on maps from SNP), which may have contributed to this being a marker year. In fact, I think that 1991 should be thrown out regardless because of the gypsy moths.

Based on this, I’d recommend that we go with following criteria : (1) the year had to be identified as a pointer year for at least 4 species/canopy position groupings (or preferably use all-trees pointer years, assuming you can do that fairly quickly), (2) among the driest years in terms of May-Aug PDSI (see above), (3) mean May-Aug [PET-PRE] ≥ 80 mm mo-1, placing it among the top 10 in a century (2001 also qualifies here, but not in the other categories). This would give us 1966, 77, and 99. This is essentially what we did before, except that we then added 1964-65 to the 1966 drought.

[mcgregorian1]

@teixeirak sorry it took so long for me to get back. We've gotten our first R assignment (using data.table as a syntax) and it's taking me much longer than I'd hoped.

First off, thank you for looking more into the drought metrics. I agree with using just 66,77,99. Below are some results from me running this and some notes:

combined cores in rwl file

Thanks again for putting all this together.

pointer years

With all the years as one unit, I get this signal for pointer years (using the >25% trees saw >30% growth reduction). As you can see, 1991 is clearly on top

• note: this table is the full set of calculated pointer years

running the models

The top candidates are now roughly the same as they were before, except for 2 major things:

• "year" is no longer considered significant (with dAIC>2). I'm confused as to why
• "TLP" is also no longer significant
• I've saved these as separate files than their predecessors with "0209" as a suffix on the file names (files are here).

best overall models

• I've also saved these files with the 0209 suffix. These of course exclude "year" and "TLP" as top variables.

I'm confused why year wouldn't make it as a top variable now, going so far as to have a dAIC of <-2 for the "all_years" model. There's been nothing major done to the data other than a reorganization of it, unless the (wrongly-created) average of 1964-1966 I was using prior to the rehaul we did at the beginning of August was faulty from the start and I just never realized that. Actually, maybe that is the reason - looking back at the overhaul where we did the correct math for a 64-66 average (midway through #53), "year" still comes out as a top variable.

[teixeirak]

Great, thanks for running these analyses!

We're all set in terms of defining drought. Please go ahead and count the selection of 1966, 77, and 99 as final. I've just updated the manuscript methods & results related to identifying years accordingly. We can close this issue, but first I'll finish commenting on your comments above.

[teixeirak]

I'm confused why year wouldn't make it as a top variable now, going so far as to have a dAIC of <-2 for the "all_years" model.

I'm not surprised. 1964-66 came out as the weakest drought year, presumably because 1966 was diluted by 1964 and 1965. Now they are more even.

[teixeirak]

* I've saved these as separate files than their predecessors with "0209" as a suffix on the file names (files are [here](https://github.com/SCBI-ForestGEO/McGregor_climate-sensitivity-variation/tree/master/manuscript/tables_figures)).

Thanks! Let's go ahead and treat these as the master versions (overwrite previous results).