Forest list archive: msg00028

[Prev][Next][Index][Thread]

Predicting diameter variability -Reply



>     Therefore, it may be useful to relate standard deviation of
>diameter and available stand characteristics such as average
>diameter, stand density, and age.

>     If you know anything about this issue (including
>references), I, and probably other netters, would appreciate
>hearing from you.

The diameter distribution methods used in the New Zealand STANDPAK,
mentioned by Euan Mason, are described in:

1) Goulding, C.J. and Shirley, J.W. 1979.  A method to predict the
yield of log assortments for long term planning.  Pp. 301-315 in
Elliott, D.A. (ed) "Mensuration for Management Planning of Exotic
Forest Plantations", New Zealand Forest Service, Forest Research
Institute Symposium No. 20.

2) Goulding, C.J. 1986.  ?  In: Levak, H. (ed), "1986 Forestry
Handbook".  New Zealand Institute of Foresters (Inc.), Wellington.
1986.

3) Garcia, O. 1984.   New class of growth models for even-aged
stands:
Pinus radiata in Golden Downs Forest.  New Zealand Journal of
Forestry Science 14:65-88.

These do exactly what Boris was talking about, estimating
distributions from stand-level data.  1) and 2) (not entirely sure
about the contents of 2, I do not have it at hand) follow earlier
unpublished work by Clutter and Allison, and by Levak, using
regressions on the stand variables to estimate the basal area of the
minimum dbh tree and the variance of the individual tree basal
areas.  Together with the mean basal area, these values are then used
to calculate the parameters of a 3-parameter Weibull distribution of
basal areas, which is subsequently converted to a distribution of
diameters.  In 3) a regression for the coefficient of variation of
the basal areas (equation 22), together with the mean basal area, are
used to obtain a two-parameter Weibull (if you plot Weibulls with the
same mean and variance but different location parameters on top of
each other you will se why!)  Working with basal area distributions
instead of diameters has been found to give more stable
relationships, and also makes the method-of-moment estimators to be
almost fully efficient (the shape parameter is larger.)  Of course,
if the tree basal areas follow a 2-parameter Weibull so do the
diameters (this is not true with three parameters).

Now, despite what my New Zealand friends said, I believe that this
kind of approach will give you distribution estimates as good as you
have the right to expect.  The real problem is that talking about
tree diameter distributions in general does not make much sense.  To
repeat it one more time, usually TREE SIZES ARE NOT RANDOMLY
DISTRIBUTED ON THE GROUND.  Competition induces short-range negative
spatial correlations (the neighbor of a large tree is likely to be
small). It seems to me that this is pretty obvious to any forester,
and it is explicitly built-in in individual-tree (distance-dependent)
growth models.  Often more noticeable, microsite variation induces a
somewhat longer-range positive correlation (trees close by are likely
to grow in more similar conditions than trees further apart).  This
one has been totally ignored in all the individual-tree growth models
I know of.  The consequence of all this is that the distribution of
tree sizes in a stand (what we usually want) can be, an usually is,
totally different from the distribution in sample plots (what we
usually get).  More generally, a "diameter distribution" will vary
unpredictably with the size of the piece of land considered:

Garcia, O. 1992.  What is a diameter distribution? In: Minowa, M.
and Tsuyuki, S.  (eds.) "Proceedings of the Symposium on Integrated
Forest Management Information Systems".  Japan Society of Forest
Planning Press.
(see also reference 3, above, and page 1896 in Can.J.For.Res. 24,
1894-1903, 1994.)

And these are not just theoretical niceties, there is increasing
empirical evidence that the effects can be quite substantial.
There is no denying that distributions of output products are
important for forest management, and we would very much like to have
them.  But realistically, we must content ourselves with rather rough
estimates, and no amount of mathematical or statistical
sophistication will change that.  Besides the spatial structure
issues, it is "well-known" that decent estimates of anything beyond
second moments requires samples of astronomical size.  It seems
possible, however, to estimate stand-level variances from inventory
plot data when there are several plots in the stand, and that is a
subject worth pursuing.

It never ceases to amaze me that nobody seems to see facts that are
so obvious.  But I suppose that playing with distributions has
already become a minor industry, and it is such a fun!  As Kendall
and Stuart noticed in The Advanced Theory of Statistics, "the fitting
of distributions to observational data has a certain intrinsic
interest which is apt to outrun its statistical usefulness".

My apologies for this long diatribe.  Now that I got it out of my
system, on with your next diameter distribution publication!

-------
Oscar Garcia  -  Visiting Professor  -  og@kvl.dk
The Royal Veterinary and Agricultural University  (KVL)
Department of Economics and Natural Resources, Unit of Forestry
Thorvaldsensvej 57,  DK-1871 Frederiksberg C,  Denmark
Tel. +45 35 28 22 47.  Fax +45 31 35 78 33



Follow-Ups:

[Metla] [Main Index] [Thread Index]

Mail converted by MHonArc 1.1.0