Weekend modeling, Nov 28   Robert Leverett
  Nov 28, 2005 06:43 PST 

ENTS

    Yesterday, John Eichholz and I modeled 3 more MTSF white pines for
volume, bringing the total number of modeled pines to 58. Per agreement
with Will Blozan, I've dropped limb volume estimates out of the totals
and settled for trunk volume. The latest regression equation is:

     Y = -53.38829012 - 2.25714004(X1) + 2.100125255(X2) + 5.904026908
(X3) + 11.82357878(X4) + 14.04687849(X5)

    where Y = trunk volume in cubic feet
             X1 = total height in feet
             X2 = diameter in inches at 2.5 ft
             X3 = diameter in inches at 4.5 ft
             X4 = diameter in inches at 50 ft
X5 = diameter in inches at 100 ft

   The multiple linear regression coefficient from the 58 sample points
is 0.963. As explained in a previous e-mail, this very high regression
coefficient is misleading in terms of using it as a reliable predictor
of trunk volume. For example, using the equation, two small trees in the
sample have a computed negative volume and the volumes of the 8 largest
trees are all understated. A nonlinear equation is needed, but one based
on multiple independent variables. The regression programs I have are
for multivariate linear or bivariate nonlinear.

   By excluding limbs, of the 58 pines modeled, only the Grandfather
pine in Monroe State Forest exceeds 1000 cubes. The question of whether
or not subsequent measurements of the tree will continue supporting a
thousand cubes begs to be answered. What must be done is to volume model
the trunk from two vantage points that are 90 degrees apart, i.e. test
for elliptical sections of the trunk.

   There is no question that the Grandfather tree is huge. It's CBH is
13.7 or 13.8 feet depending on where the midpoint of the slope is
located. Personally, I do believe the tree has a good crack at being
confirmed as a legitimate thousand cuber, but not by much. On the next
available weekend date with good weather, it will be back to Monroe I
will go.

    One of the 3 pines John and I modeled in MTSF was a very slender
tree. Its CBH is only 4.6 feet or 17.7 inches DBH. In this dimension,
the slender tree compares to one modeled in Mt Tom reservation. That
tree is 17.8 inches DBH. The volumes are virtually identical. A side by
side comparison follows:

Tree        Height    Diam at 2.5'     Diam at 4.5'    Trunk Volume

MTSF     133.3          18.5'              17.7"               76 cubic
ft

Mt Tom   106.6          18.8'              17.8"               75 cubic
ft

    The 26.7 ft height difference illustrates the dramatic height
advantage that Mohawk's pines enjoy over stands that are more typical of
what one commonly sees around New England.

    In an exercise to compare conical volume to computed volume I took
the diameter at 2.5 feet as opposed to on the ground and used that as
the base diameter. When the conical volume is compared to that computed
with the RD 1000, the results are all over the place, revealing the many
trunk departures from a regular cone shape. Interestingly, though, the
cone shape exceeds far more than trails the computed volume. So root
flare is still too much of a distorting factor at 2.5 feet. Using the
diameter at 4.5 feet, the cone shape understates the volume in 47 of the
58 sample trees. Using the diameter at 2.5 feet, the cone shape
overstates the volume in 40 of the 58 trees. A lot seems to be happening
between 2.5 and 4.5 feet above the base, though this is seldom obvious
from simple eye observation. Trees that have a big root flare start the
cone well outside the general space taken up by the trunk. That much is
obvious to the eye. For trees that remain cylindrical for a long trunk
distance and then take a nosedive to the top, the conical shape stays
within the space of the trunk for a long time.

     It would be extremely useful to find two diameters, taken at
standard heights, one of which fairly consistently overstates and the
other of which fairly consistently understates the modeled volume. The
diameters at 2.5 and 4.5 feet come fairly close to doing this. Perhaps
2 feet and 5.5 feet would accomplish this. Getting closer to the base
than 2 feet often gets one into so much root mass, that
circumference/perimeter determinations start to be challenging and when
a big tree is on a steep slope – forget it. However, to illustrate the
idea, the Jake Swamp tree cubes out to 561 cubic feet. The volume based
on the diameter at 2.5 feet is 588 cubic feet. The volume based on the
diameter at 4.5 feet is 480 cubic feet. The computed volume lies within
these boundaries. The Seneca and Grandmother trees have computed volumes
that fall within their 2.5 and 4.5 boundaries. However, the Grandfather
and Ice Glen pines do not. Oh well, onward with the mission.

Bob


Robert T. Leverett
Cofounder, Eastern Native Tree Society
RE: Weekend modeling   Don Bragg
  Nov 28, 2005 13:17 PST 

Bob--

Did you ever run your model without the intercept (the -53 value)?
Given how regression models are optimized, I would be surprised if it
would affect your regression coefficient much, and it would give you
another intuitive data point (volume = 0 cubic feet when total height =
0) that may improve the fit for smaller diameter trees.

I'm not sure how you fit your regressions, but I have seen a program on
the market (TableCurve, http://www.systat.com/products/TableCurve2D/)
that can fit a massive number of models quickly if you'd like to play
model form more. I believe they offer a preview version for people to
try for free, a good thing given its $500 price tag.

Just some thoughts...

Don Bragg
RE: Weekend modeling   dbhg-@comcast.net
  Nov 29, 2005 03:54 PST 
Don,
   Thanks for the ideas and the info on the regression software. I've just been using what is built into the Data Analysis package of Excel.
    I did put in an arbitrary 0 origin point inot my data, but that didn't cure the problem. So long as the model remains linear, the result seems yield negative intercepts. I'll drop the negative constant and recalculate the regression result.
Bob
RE: Weekend modeling and Excel problems   Don Bragg
  Nov 30, 2005 13:19 PST 

I wondered if you were using Excel...there has been considerable
consternation in some Ecology email lists (e.g., ECOLOG) about using
Excel for statistical analysis, including some definite errors due to
how things are calculated. Type 'regression problems in excel' into
Google, and you will find multiple webpages on this topic. I use other
programs to do regression in part to avoid the weaknesses of Excel--I
don't know if it will affect your outcomes, but it is worth
investigating.

If you'd like, I'd be happy to run some regressions using my statistical
software, and see what comes up (you'd have to email me your data)...

Don Bragg