Write code outputting numbers to fill in forest C cycle diagrams

[teixeirak]

We will want to create a series of ensemble C cycle diagrams similar to this one (Anderson-Teixeira et al. 2016 GCB):

Forest types will be:

• tropical broadleaf ≥100 yrs
• tropical broadleaf <100 yrs
• temperate broadleaf ≥100 yrs
• temperate broadleaf <100 yrs
• temperate conifer ≥100 yrs
• temperate conifer <100 yrs
• boreal conifer ≥100 yrs
• boreal conifer <100 yrs

@ValentineHerr, please create code that outputs the statistics (calculated in issue #2) as text as in this figure for easy copy/pasting into the figure.

[teixeirak]

@ValentineHerr, see this file for mapping of variables onto C cycle diagram. I'm also recording conversion factors and data filters that should be applied. I will create an accompanying readme file to explain.

[teixeirak]

I've added a rough draft version of the C cycle diagram. There are some changes, including adding some additional variables and moving the stocks off to the side. @ValentineHerr, I hope this doesn't conflict with your progress on this issue. There's plenty of flexibility here, so please let me know what you think.

[ValentineHerr]

@teixeirak ,

I built the frame for the figure:

There will be minor adjustments but I believe it is looking good. Right? Are the stock supposed to be changing size like the arrows?

Next step will be to calculate the values and bring the right painting in the background (I created a folder here to store them. I probably will need some more guidance on how to use this file, especially variables like NEE_cum_C. Maybe we can talk in person when you come back. I need to continue on Iara and Craig's modelling this afternoon.

[teixeirak]

This is great; thank you! I'm so glad that you're able to automate this. It is generally looking good. There will be some minor cleanup/ adjustments, but the concept is solid.

A few adjustments for now:

• [ ] I made some more updates to the draft diagram (sorry!), so let's get those incorporated.
• [ ] I think I introduced confusion with the red lines; those were simply to indicate what had changed. All text and box outlines should be black.
• [ ] note that ANPP_repro and R_soil_het are in the wrong places.
• [ ] It's going to be hard to read text boxes on top of arrows or dark parts of the painting. Is it posisble to make semi-transparent text boxes behind those?
• [ ] I'm not sure if its Github or R, but the arrows look very pixelated. I'm assuming that will be improved in a higher resolution image?

[ValentineHerr]

I'll incorporate changes, no problem. I should be able to make the semi-transparent boxes. I think the resolution of the image can be improved. I just copy-pasted in github and probably lost a lot there.

[teixeirak]

Great! Yes, I think Github reduces the resolution. The figure at the top of this issue obviously became fuzzy when pasted here.

[teixeirak]

@bpbond, @nkunert, and @hmullerlandau, if you would like to review/ comment on the C cycle diagrams (other than trivial changes), now is the best time. Here is the rough draft version of the C cycle diagram that Valentine is automating in R. This file maps the ForC variables onto the C cycle diagram (e.g., indicates which are and are not mapped, groups similar variables).

[bpbond]

Wow @ValentineHerr nice work - cool to have this automated. I assume your code will be able remove the toucan for a temperate forest, and add a moose for boreal, etc.? :)

Seriously though, I like the figure, although the two Rsoil sub-arrows dominating the middle might be moved to the right, for visual symmetry and better clarity? (I.e. more like in @teixeirak 's original Powerpoint.)

[teixeirak]

I wouldn’t be surprised if @ValentineHerr could write R script to paint a moose, but we won’t go there because I don’t want to automate away my job!

Valentine nicely reproduced my earlier, ugly rough draft, so please base comments on this file under the assumption that she'll be able to replicate it.

[teixeirak]

@ValentineHerr, please check your code-- I'm looking at eddy flux (GPP, NEE, Reco) records from tropical forests, and something is wrong. I believe there should be only one record for young forests, and I'm counting at least 10 geographic areas for mature forests. I'm also noting that the young forests have a lot more data than they should based on the TropForC review paper.

[ValentineHerr]

Good catch, the script that created the variable averages per Biome was not handling age correctly. I think I fixed the problem. Let me know if you find anything else.

[teixeirak]

Looks good. I'll let you know if I catch anything else.

Another minor thing-- How are you scaling the arrows for young forests where the flux is a function of age? Let's scale according to the predicted value for forests at age 50.

[teixeirak]

Also, please switch order of equations at the bottom of the diagram so that they're in sequential order.

[teixeirak]

@ValentineHerr, some more small modifications on this (also listing the two above):

• [x] How are you scaling the arrows for young forests where the flux is a function of age? Let's scale according to the predicted value for forests at age 50.
• [x] please switch order of equations at the bottom of the diagram so that they're in sequential order
• [x] There are a couple instances (see tropical broadleaf young) where fluxes with no records are showing up as NA, 0/0/0
• [x] It would be nice to have a quick visual mechanism for differentiating results based on a good sample size versus those based on just a few records. My initial idea is to put the means/equations in bold (say those with >= 10 geographic.areas), but I'm open to other ideas if you can think of something better.

[ValentineHerr]

I believe I covered every thing above. I decided to use dashed lines for the arrows where there is less than 10 geographic.areas. Let me know if you don't like it.

Should I separate BNPP variables? I am thinking that we could have BNPP stop half way and split into 2 arrows for fine and coarse. That way those arrows could also have attributes relative to the amount of data we have and to the mean value (or predicted value at age 50).

[teixeirak]

Yes, good idea to separate BNPP. Please put BNPP_coarse on the right side (closer to the tree).

I think the dashed lines work well, but we might think about the sample size cutoff. 10 geographic areas is a fairly high bar.

[ValentineHerr]

could be 10 plots ?

I pushed my changes on BNPP.

[teixeirak]

I'd prefer to stick with geographic areas, as I'm concerned about pseudoreplication. Let's lower the threshhold to 7.

Please do the same for the stocks-- outline boxes with dashed lines when <7 geographic areas.

[ValentineHerr]

Ok I've changed the threshold to 7 and made the dash line for stocks too.

Next and hopefully last step for this issue will be to put the right background behind each diagram.

[teixeirak]

This is looking great.

Could you please change to fitting functions of ln[age], as in issue #11? Note that we may eventually put in more specialized functions for certain variables, but ln[age] seems to be fairly good as a generic function.

[ValentineHerr]

Just as a note, the fitting functions won't be exactly the same as in issue #11 since we are working one biome at a time here. In other words, Biome won't be a fixed effect in these equations.

[teixeirak]

That's fine, and actually preferable in that we expect differences in the slope with age across biomes that can't be detected through those analyses. An alternative to consider for issue #11 would be plotting these relationships instead of the model with biome as a fixed effect, but that would exclude biome-variable combinations with lower sample sizes, so let's leave it as is for now.

[ValentineHerr]

I made the following changes but none of the results turned out significant:

• Add plot.name as nested random effect (I had removed it because we had too little data per Biome, I believe)
• run the mixed model only when we have at least 10 records and 3 unique combination of geograpgic.area and plot.name (I had only at least 3 geograpgic.area)
• us log10(age) instead of age. I am using log10 (here and in issue #11) because it is easier to interpret the slope that way. We can say "for any increase of an order of magnitude in age, response increases by X amount"...
• don't do linear model if mixed model doesn't turn out significant.

All the equations disappeared in the diagrams. If we remove plot name from the random effects, we get some of them back (and actually we get extra more for Temperate Conifer YOUNG, probably due to the log of age...)

Let me know what you think.

[teixeirak]

Plot.name should be left out (for young forests only) because plot names often indicate different sites in a chronosequence. For mature forests, plot. name should either be included as a random effect nested within geographic.area or we should compute an average for each plot (weighted by sampling time) before conducting the analysis.

[teixeirak]

The rest sounds good.

[ValentineHerr]

I thought the modeling was only for young forest. I was following issue #2

[teixeirak]

You're right, sorry for the confusion.

[teixeirak]

This is basically done, so I'm closing. Minor cleanup will be part of issue #18.