update decadal plots

[teixeirak]

I'm splitting this off from issue #85, which is otherwise done. The goal is for @camerondow35 to make an updated version of the analysis of growth-DBH relationships by decade.

@camerondow35 this file tells you what years were kept for each species within each site and this file tells you what species were completely dropped from the Year (or dbh x clim interaction, or CO2) analysis at any site.

Originally posted by @ValentineHerr in https://github.com/EcoClimLab/ForestGEO-climate-sensitivity/issues/85#issuecomment-738306368

[camerondow35]

Working on this now - is there meant to be a second link here?

and this file tells you what species were completely dropped from the Year

[teixeirak]

I'm confused. What are you. expecting? Here's some more info from @ValentineHerr, originally posted in #85:

@camerondow35,

• are you focusing on SCBI only in your analysis?
• FYI, the core data pulled in that script is from here, which gets updated fairly often (but not so much for SCBI data). When it is updated, you may need to update the token in your script
• I am not sure how much of the GLS analysis you are using but if you ever need the latest version of the original script it is here.
• When I am done today (and after that script runs for a couple days), there should be a file here, called "all_Species_Year_to_keep_for_Year_or_clim_x_dbh.csv", that would let you know what species-year combo I kept for each site in the analysis including 'Year' or 'climate x dbh`.

[camerondow35]

Ok I see. There are two files, one with the dropped species and one with species year kept.

[camerondow35]

Slow progress here, sorry! I finally got the script working enough to produce sample plots. Were you thinking something like this? Plots for SCBI:

[teixeirak]

Yes; this is looking great! From a quick review:

(1) please adjust the colors for the decades to a continuous gradient so that it’s intuitive to pick out trends, and

(2) This uses Valentine’s parsed data set, right? I’m surprised that FRNI seems to have such an unstable pattern.

[camerondow35]

Are the colors here too similar, meaning are 1990 & 2000 too hard to tell apart?

This uses Valentine’s parsed data set, right? I’m surprised that FRNI seems to have such an unstable pattern.

Yes, I have subset the data to only include the species x year combos found in the two csv's Valentine made.

[teixeirak]

Looking great! But yeah, the colors are hard to tell apart-- especially in the red zone. The original color scheme worked well.

[teixeirak]

Regarding FRNI, let's see what the rest of the sites look like, and deal with that if it seems to be a systematic problem.

[camerondow35]

So the point of these plots is to show that some sites have lower BAI in recent decades, right? Would using a linear model work as well as the loess/gam?

[camerondow35]

I'm having an issue with running the same method for each site so instead of mixing methods I could just run them like this to keep it consistent. However, if you think mixing methods is fine I can go ahead and do that (some GAM some loess, maybe one linear - BCI is giving me trouble)

[teixeirak]

Hmmm... the linear fit seems to be okay based on SCBI, but it's inconsistent with what we're showing elsewhere in the manuscript. Would it work to use a polynomial fit, as Valentine did in the DBH responses figure (below)?

[camerondow35]

They all plotted at least! Here's the link so you can check them out: https://github.com/EcoClimLab/tree-rings_global-change/tree/master/decadal_plots

[teixeirak]

Thanks! I will review these carefully as soon as I get a chance.

[teixeirak]

Thanks, @camerondow35 . I have knit these into the SI file. I'll continue looking at these, but one first issue is that we're missing the plot for Niobrara, perhaps because it was added after Bianca wrote that code.

[teixeirak]

It would help to make the color palette more intuitive. Copying a suggestion from Ian McGregor on another project:

Can I suggest instead of rainbow to use something like one of the viridis palettes? In my opinion it makes it more intuitive plus those palettes are color-blind friendly. Your code would be something like

library(viridis)
colors <- magma(7) # or viridis, plasma, inferno

[teixeirak]

As far as the actual content of the plots, I'm going to be reviewing these more carefully in the context of the other results and will provide more feedback later.

[teixeirak]

@camerondow35 , I think overall these plots look fine.

The data are a bit out of date, as @ValentineHerr recently restored 2 decades of data that had been mistakenly removed from CB. So we'll need to re-run at some point with those data.

Plus, a few other formatting things:

• [x] it would also be helpful if you could set the minimum on the y axis to zero.
• [ ] please increase the font size so that it will be easily readable when reduced to fit a page. The screenshot below shows how it currently looks
• [x] add units to the axes: DBH (cm), BAI (cm^2)
• [x] It would be good to replace the species codes with the full name, as we've done so everywhere else. Codes and names are available in this table
• [x] It's only necessary to show the legend once, but, given that this is the SI, okay to repeat if dropping it adds too much coding time

[teixeirak]

@camerondow35 , is the following an accurate caption for these figures? "Plot visualizes the data included in the GLS model, separated by decade, using a probability density function. Transparent ribbons indicate 95% confidence intervals."

Also note that I'll be sending this around to coauthors later today and presenting in a virtual ForestGEO seminar on Wed. Then aiming to submit in a few weeks. (These plots don't need to be fixed up until submission, although I probably will show one in the presentation.)

[camerondow35]

That seems like an acceptable description but i'm not familiar enough with the term "probability density function" to make a 100% judgement call. I used the code geom_smooth(method = "lm", sd = TRUE, formula = y ~ poly(x,2,raw =TRUE)) for these plots.

[camerondow35]

Pushed updated plots