GEOSTATS: The Spatial Forest
- Dear Statistician,
This letter informs you of the forestery reference in Rev Ciencia Forestal en México 19 (76) 1994: 141-152 by Bertil Matérn of the Dept of Forest Biometry, University of Agricultural Sciences, Umea, Sweden. The Poisson forest (Keuls et al., 1963), the angle count method (Bitterlich, 1984) and the central satellite model (Warren, 1991) are emphasized. Basal area at breast height going to square meters per hectare is the main target.
Using Goodlands (1971, J Ecol) work on the Brazilian Cerrado quite apart from Matérns brilliant review, the writer calculated these relations for 4 niches1) grassland, 2) matorral (short open woods), 3) light forest & 4) forest. The relation of the 4 sites (X) to basal area (Y) is logistic, site by number of tree species fits: Species = a + b(Site)3 + c exp-Site, and Site by tree height fits: Height = a + b(Site) + c(Site)2. The coefficients of determination are at unity, and these relations are ideal. As the X and Y axes have only 4 points, perfect curves must result. These relations may help in the statistical management of new data from different niches.
When the foliar morphometric averages in 4 mezquitales (mesquite woods) are so compared, logistic equations explain site by leaflet length and site by rachis length, because the curve that runs through any 4 points is likely to be logistic, sine, exponential, extreme value, rational, etc. Also, using (or overusing!) averages is just not the same as running a fit through the many original data points. Bias must arrive, because the experiential judgement of the researcher is called upon for expression of reasonable caution in the use of such models since many work as well as another. When several thousand raw data points are run, the results could be DF Adj r2 = 0.25 or less, leaving one to wonder what happened.
More, of course. See Means, JE et al, Software for computing plant biomassBIOPAK users guide, USDA, Forest Serv, Gen Tech Rep PNW-GTR-340, which refers to a DOS program on Internet that has a library of equations. Go to http://www.fsl.orst.edu/lter/data/software/biopak.htm
Best Regards, Paul R. Earl
PS See Software at http://www.fs.fed.us/links/software.shtml