Height measurements

[gregjohnsonbiometrics]

I plotted the height data (startHT) over stand age (STDAGE) restricting it to 5 year age points (e.g., 10, 15, 20) and limiting it to ages up to 100 for clarity. There appears to be a consistent decrease in height to age after STDAGE 50. It makes me suspicious there's something going on there. It could be that the older population of sample trees come from different (lower) productivity areas (reserves for instance).

OWNCD, and STATECD don't seem to show any thing interesting.

Any ideas?

[gregjohnsonbiometrics]

Here's an interesting graph for Douglas-fir heights. The x-axis is HT40 at the time of the height measurement. The y-axis is the measured height (startACTUALHT) / STDAGE which I'm calling ht_mai. As the stand's dominant height increases, the average height growth per year declines in a pretty uniform way (the line in the graph is a GAM smoother, I dropped the individual points as they are super variable). The curve gets pretty wonky after about HT40 of 50 feet, but it's more or less hovering around a constant out to STDAGE 150.

I'm trying to explore ways to measure/assess productivity given the data we have at hand. I was hoping an approach like this would shed some light.

[holoros]

Nothing to add from my perspective, but interesting findings that are a bit concerning. No idea on how to explain the observed trends, except for sample size issues? I do think a quantile approach might help, but this is one of the reasons we used total biomass increment rather than height for our site productivity modeling in Maine.

My understanding is that STDAGE is based on a single tree core taken from a dominant at the time of plot establishment and then updated. John Shaw gave a presentation on this several years ago at a FIA Science Symposium, https://research.fs.usda.gov/treesearch/50180.

I am still trying to fit a North American climate site index model, but it is proving more difficult than expected. I'll keep at it. I am trying a spatially informed random forest to better capture potential trends. Just a lot more computationally demanding than a regular non-spatial random forest, which I might need to end up using.

[DMtree-DougFir]

First graph. Assuming the y-axis is height and not height growth as you describe. If so, I wonder if this is showing that the older stands are dominated but lower site index stands. This is the old problem with B201 and why those curves are biased. Related to this, there is probably an imbalance of data from coastal plots and Rocky MT plots. I wonder about plotting the H40 values by age (and possibly splitting it by site class or site equation or something like that. Am not sure if box widths would tell anything interesting about numbers of trees (plots)?

The second graph is a head scratcher, but again, is it possible that the old stands are dominated by high elevation trees that may have a more prolonged height growth trend (Curtis and others 1974). Again, plotting by the assigned site curve type might give some clues.

The "FOREST INVENTORY AND ANALYSIS NATIONAL CORE FIELD GUIDE FOR THE NATIONWIDE FOREST INVENTORY" (Version 9.4. September 2024) says core 3 overstory dominant/codominant trees at plot establishment (page 47) although it suggests some possible "local" field procedures are possible.

There is one other complication that may have some (unknown) impact. For explorations we (I) extracted and computed per acre values for the whole plot and "ignored" the fact that some plots have multiple conditions on them. This would not necessarily impact tree values (ages) as I merged the condition (age, site) to the individual tree (and tree growth) records. For plots that have multiple conditions (or actually are partially non-forest conditions) the use of the TPA_UNADJ to compute things like H40 could be impacted (under-estimated?). I think this is somethings we need to think about and discuss how to handle better.

[gregjohnsonbiometrics]

Here are graphs of HT40 over STDAGE with evenly spaced (by number of observations) site index classes. Note that site index is from over 11 site equations for multiple species and at least 2 base ages (so a pig's breakfast). Even so, as we'd expect, the HT40 trends are consistent with site height curve trends.

The second graph is the same as the first with the individual points removed.

The third graph is limited to STDAGE <= 70 to get a better look at the HT40 around base age 50.

[gregjohnsonbiometrics]

Trying to judge height growth productivity without site index, I fit a curve to the Douglas-fir startHT40 data:

$startHT40 = b_0 * (1 - exp( -b_1 * STDAGE ) )^{b_2}$

the parameter estimates are:

$b_0$ = 3.837e+01, $b_1$ = 1.256e+02, $b_2$ = 1.017e-03, $b_3$ = 4.715e-01

I then computed the ratio $HT40_{ratio}$ of predicted HT40 to actual. -- values over 1.0 indicate plots performing above the mean HT40 performance and under 1.0 performance below the mean.

The first graph below shows the individual tree annualized height growth over initial height colored by 4 ratio classes. The second graph is the same with the individual points removed for clarity.

The trends tend to make sense where we have the most data.

The last graph is a map of $HT40_{ratio}$ across all the states contributing data.

Sample sizes by $HT40_{ratio}$ classes:

HT40_ratio	n
[0.13,0.50)	5869
[0.50,1.00)	79702
[1.00,1.50)	74673
[1.50,4.86]	25396