Files in this item

FilesDescriptionFormat

application/pdf

application/pdf9210819.pdf (5MB)Restricted to U of Illinois
(no description provided)PDF

Description

Title:Modeling red pine tree mortality: An artificial neural network approach
Author(s):Guan, Biing Tzuang
Doctoral Committee Chair(s):Dawson, Jeffrey O.
Department / Program:Crop Sciences
Discipline:Crop Sciences
Degree Granting Institution:University of Illinois at Urbana-Champaign
Degree:Ph.D.
Genre:Dissertation
Subject(s):Biology, Biostatistics
Agriculture, Forestry and Wildlife
Artificial Intelligence
Abstract:The main objective of this study is to seek new modeling techniques to improve the ability of projecting tree mortality in forest growth and yield simulation. Multi-layer feed-forward artificial neural networks (ANN) are adopted to achieve the goal. The premise of ANN modeling approach is the ability of such networks to approximate any measurable or continuous function to any desired degree of accuracy, given enough complexity and training.
In this study, two types of tree mortality models are developed based on red pine (Pinus resinosa Ait.) data collected from the Great Lakes region. Diameter at breast height (DBH) and annual diameter growth (ADG) are the explanatory variables in both types of models, and annual survival rate is the response variable. For the first type of models, training set consists of data obtained based on a cross-classified scheme. For the second type of models, individual tree records are used to construct training set. Training method for the first type of models is the back-propagation method, and networks are trained on serial computers. A method based on the fast simulated annealing is used to train models of the second, and the trainings are performed on a massively parallel computer. In addition to several goodness-of-fit and performance statistics, a model-based comparison approach is also developed to assess the performance of ANN mortality models against a benchmark statistical model.
Results from this dissertation suggest that ANN mortality models not only fit the training data better than the benchmark model, but also expect to perform better in the future, provided that the training set are representative. Model-based comparisons show that ANN mortality model in general have lower prediction biases, but with larger prediction variances, than the benchmark model. Mean squared error criterion suggests that ANN mortality models are expected to perform better in the future, provided the training data are representative.
A brief review of modeling tree mortality in forestry growth and yield projection, as well as an overview of neural computing approach, is also presented in this study. Other issues related to the use of artificial neural networks in forestry related modeling are also discussed.
Issue Date:1991
Type:Text
Language:English
URI:http://hdl.handle.net/2142/20780
Rights Information:Copyright 1991 Guan, Biing Tzuang
Date Available in IDEALS:2011-05-07
Identifier in Online Catalog:AAI9210819
OCLC Identifier:(UMI)AAI9210819


This item appears in the following Collection(s)

Item Statistics