Diameter Distributions of Longleaf Pine Plantations-A Neural Network Approach
Abstract
The distribution of trees into diameter classes in longleaf pine (Pinus palustris Mill.) plantations does not tend to produce the smooth distributions common to other southern pines. While these distributions are sometimes unimodal, they are frequently bi- or even tri-modal and for this reason may not be easily modeled with traditional diameter distribution models like the Weibull whose form is unimodal and monotonic. Neural networks, a development of artificial intelligence research, can take on any form that is found in the data, allowing the prediction of many phenomena without the assumption of a given model form. To see if this new technique has merit in forest modeling, the diameter distributions of several longleaf pine stands were modeled using a neural network process. Weibull distribution models were also fit to the data using regression and parameter recovery techniques. These approaches were then compared. The neural network procedure appeared to provide slightly better predictions than either of the traditional methods. Results of this comparison are presented in this poster. This early result indicates that this new technique is promising and deserves further investigation.