Year of Award

2009

Document Type

Dissertation

Degree Type

Doctor of Philosophy (PhD)

Degree Name

Forestry

Department or School/College

College of Forestry and Conservation

Committee Chair

Carl Seielstad

Commitee Members

Lloyd Queen, Woodam Chung, David Affleck, Anna Klene

Publisher

University of Montana

Abstract

Airborne laser scanning data is used to discriminate between Douglas-fir (.Pseudotsuga menziesii), ponderosa pine (Pirns ponderosa), lodgepole pine (.Pinus contorta), and western larch (Larix occidentalis) in a mixed coniferous forest of western Montana, USA. Laser-derived structural and intensity variables are investigated to classify tree species at individual tree and plot-level dominant species. Linear Discriminant Analysis is applied to discriminate between species, and in combination with Maximum Likelihood Classification, used to map species across the landscape. Oneway ANOVA tests indicate that proportions of first and single returns and mean intensities are significantly different between species (p-value < 0.001) at both individual tree and plot-dominant species levels. A single variable in the Linear Discriminant Analysis (LDA), e.g., mean or standard deviation intensity, can produce classification accuracy ranging from 49-61% at the dominant species level and 37-52% for individual trees. The accuracy can be improved to 95% and 68% respectively by using multiple variables, including proportions of return type, intensities, and canopy heights. Adding proportion of return-type improves classification accuracy at the dominant species level, but not for individual trees. The inclusion of both mean and standard deviation of canopy heights produce higher accuracy at both levels. Validation of the landscape classifications is performed using a stand database consisting of predominant and secondary species and with gridded, fixed-area plots. Assuming that stand homogeneity is at least within dominant species criteria (>70%), the application of intensity and canopy height variables generates a classification accuracy of 45% that is increased to 53% by including mixed species (stands without a clear dominant species) in the error analysis. A second method based on Maximum Likelihood Classification (MLC) using two layers, (1) the modified LDA-species based layer, and (2) percent canopy cover (PCC) layer, improves accuracy up to 75%. Unlike the modified LDA, in which the accuracy increases with incorporation of mixed species, the MLC produces lower accuracy (38%) when these stands are included. Both methods, the modified LDA and MLC produce best results for Douglas-fir followed by lodgepole pine and ponderosa pine, while western larch is difficult to identify. Almost without exception, the classification identifies the correct mix of species within each mixed polygon, but field data do not currently support validation of individual pixels within stands.

Download

Share

COinS
 

© Copyright 2009 Agus Suratno