getting the 'year' analysis right

[teixeirak]

@ValentineHerr ,

I'm starting this as a new issue following up on #85.

We're still not getting things quite right with the analysis including year.

For POTR at CB, the GLS and decadal analysis not lining up (GLS says it's increasing, decadal analysis shows dramatic decrease):

The reason this is happening seems to be that the time range for POTR is limited to ~1970-1990 (following guidance in #85), and indeed there's little trend/ slight increase over that limited time range.

So, the issue is properly constraining the data included in the analysis. I'll review that in the next post.

[ValentineHerr]

Hmmm... given only the 3 last decades are analyzed in POTR, I wonder if it is picking up the pattern between ~23 and ~28 cm dbh as the difference between years is very big there, and increasing... Isn't dotted line meaning it is not significant trend?

[teixeirak]

Current analysis:

• We limited the analysis of each species to the range of years for which the DBH range represented by at least 3 individuals spanned at least half the total DBH range of that species, dropping species entirely if < 30 years met this criteria. (There are a lot of clarifications under issue #85, but I believe that description gets it.)
• Analysis has the problem described above, and also produces some unreasonable fits (most extreme below):

[teixeirak]

Hmmm... given only the 3 last decades are analyzed in POTR, I wonder if it is picking up the pattern between ~23 and ~28 cm dbh as the difference between years is very big there, and increasing... Isn't dotted line meaning it is not significant trend?

Yes, I think that's what's happening. And correct, not significant, but included in the best model. I don't think the model is wrong, but that we probably want to constrain it differently.

[teixeirak]

Hmmm, now I'm trying to unpack what we did and realizing that I mis-communicated what I had in mind at one point--specifically on my answer here, where I had originally envisioned "the total DBH range of the species". Sorry about that! But then I see that you ultimately caught that here, so I think we are doing what I originally had in mind. I've updated the description above to match that.

From looking at the POTR plot above, I'm surprised that we only get 3 decades by that criteria. I suppose it's because we require ≥3 cores. Perhaps the ≥3 cores criteria is unnecessary: maybe 1 or even 2 is enough. The more important criteria would be covering a good matrix of DBH x year.

[ValentineHerr]

If we keep the years that have dbh representing at least a third (instead of half) of the total dbh range we have this:

[ValentineHerr]

FYI, for POTR, all those years a very close to be included with the half dbh range criteria... 3rd column is the dbh range of the year, 4th column is the total dbh range for the species

[teixeirak]

Here are some thoughts:

• Do you think we can safely relax the criteria that a year can only be included when the DBH range in that year spans half the species' full DBH range? It is nice to have the better coverage in the plot above, but I wouldn't want to do that at the expense of reliability of the analysis.
• I think we could relax the n core required to 2
• 3 decades is too short, as evidenced by POTR. I think we should require 4-5 decades (probably 5) to include the species in the analysis.

My highest priority is to get a reliable method, as opposed to maximizing the number of species we can include in the analysis. The most valuable outcome here would be knowing what sort of sampling design we need to get a reliable answer as to how growth has been changing through time.

[teixeirak]

There are two criteria for reliability:

1. getting the direction of the trend right
2. getting the coefficients right, which is needed to reliably partition causation and project what might happen outside the scope of the existing data set.

Number 1 is obviously much easier, and I think it's sufficient for this paper. However, we should be careful in how we plot/ communicate the results. I think one thing that would help is to limit the y-axis (on all plots) to the maximum observed value. That will make it obvious when projections are running outside the reasonable scope (e.g., MEAZ at HKK).

[teixeirak]

Thinking more, perhaps it doesn't make sense to remove any data from the analysis, but rather to simply limit the analysis to species with a good DBH x year matrix (defined basically as we're already doing).

[ValentineHerr]

meaning at least 5 decades have years with dbh range representing half the total dbh range?

[teixeirak]

Yes, or maybe 1/3 the total DBH range? I'd really value your opinion here.

[ValentineHerr]

I think one third is probably enough, looking at POTR it looks like 1/2 is too restrictive...

[ValentineHerr]

I'll see how it looks for HKK.

[ValentineHerr]

Hmmm... MAEZ has 53 (just above 5 decades threshold) years with at dbh range at least 1/3 of total... so it keeps it in and if we don't actually remove the years that don't meet that criteria, the plot looks worse...

Doing 1/2 might be better, only 43 years meet that criteria there so MAEZ would be dropped.

[teixeirak]

Maybe that's what we need to do. We'll loose POTR, which is too bad because those results seem reasonable, but again I'd rather lose species and trust our analysis criteria.

[ValentineHerr]

Yeah... HKK would look then like this:

I'll see if we can do 40% of dbh range.

[ValentineHerr]

nope... would still be 53 years for MEAZ...

[ValentineHerr]

Other option is to keep 1/3 but say 6 decades...

[teixeirak]

Rather than n decades, perhaps say >1/2 the total time period?

[teixeirak]

By the way, why is CB ending in year 2000, when they were cored last year?

[teixeirak]

Rather than n decades, perhaps say >1/2 the total time period?

I believe this eliminates MEAZ.

[ValentineHerr]

1/2 total time period is 43 years for HKK, so MEAZ would be exactly at the limit if we take 1/2 dbh range.

Will check for CB.

[ValentineHerr]

POTR would be dropped too... only 31 years with 1/2 dbh range, and 1/2 time period is 48.5.

But let see what I find out about what happened to post 2000

[teixeirak]

By the way, why is CB ending in year 2000, when they were cored last year?

The raw data files have data up to 2020, but in Table S3 they stop at 2000.

[teixeirak]

1/2 total time period is 43 years for HKK, so MEAZ would be exactly at the limit if we take 1/2 dbh range.

meaning in or out?

[ValentineHerr]

meaning in or out? Right now I have it "inclusive" so it would be kept... but I can have it exclusive and it would be dropped.

[teixeirak]

How about 40% of DBH range over 2/3 of the time range? Given that we're testing for the effect of year, it seems to make sense to give it more coverage.

[teixeirak]

Those numbers could always be adjustable depending on the analysis (stricter for analyses that are concerned about getting the coefficients correct).

[ValentineHerr]

I like these numbers, (Yes, they are easily adjustable) POTR would be kept, MEAZ would be dropped, but I believe CHTA would also be dropped... (only 61 years meeting dbh criteria and 2/3 time range is 72).

[teixeirak]

Okay, let's go with that. It seems solid.

[ValentineHerr]

Wait, are we doing 2/3rd of the total period for the species of for the site? I think I did for the site but now am thinking it should be for the species, right?

[teixeirak]

Wait, are we doing 2/3rd of the total period for the species of for the site? I think I did for the site but now am thinking it should be for the species, right?

Right (time period for the species).

[ValentineHerr]

ok, let me see if that changes a lot

[ValentineHerr]

That is actually better, CHTA is kept.

Still investigating CB 2000-2020 issue (have to re-run code in dendro repo... I thought it is because all trees were cored dead and we remove the last 20 years of those but it looks like they were cored alive so I am still investigating but that code takes a while to run).

[teixeirak]

Oh, I think we didn't initially have confirmation that they were all live, so Bianca applied the conservative assumption that they were dead.

[ValentineHerr]

but now we are sure they were alive? that is what is in the code, but this file was not updated so it was overwriting the status to NA and 20 years were cut back.

[teixeirak]

Yes, CB trees were all cored live.

[teixeirak]

For writing up the methods, are we still saying we need at least 3 cores to meet the criteria? Or just 2?

[ValentineHerr]

I kept 3. I think that works. Looking back at this issue, I think 2 is not restrictive enough for some of portions of the figures we see.

[teixeirak]

Thanks!

[teixeirak]

I think this all looks good now!