HR: 0800h
AN: B21A-0038 [Abstracts]
TI: Classification of forested wetlands using ordination of multitemporal Landsat reflectance
AU: * Keim, R F
EM: rkeim@lsu.edu
AF: Louisiana State University, Baton Rouge, Baton Rouge, LA 70803, United States
AU: Zoller, J A
EM: jzoller@ene.com
AF: Ecology and Environment, Inc., 2 Riverway
Suite 625, Houston, TX 77056, United States
AU: Braud, D H
EM: dbraud@lsu.edu
AF: Louisiana State University, Baton Rouge, Baton Rouge, LA 70803, United States
AB:
There are important limitations in interpreting satellite imagery in dynamic environments. In forested wetlands,
where flooding and aquatic vegetation vary temporally, a multitemporal approach is needed for extracting stable
patterns. We used field measurements of forest composition and structure and seven cloud-free Landsat images
from a time span of six years to classify forest vigor in 95,000 ha of cypress-tupelo forested wetlands
surrounding Lake Verret in southern Louisiana. A principal component (PCA) ordination of the reflectance in
Landsat bands 3/4/5 for each image was the basis of the classification. No single Landsat band or image
dominated the first few PCs, so that the multitemporal and multispectral aspect of the data were fully expressed
in the ordination. Each Landsat pixel was classified as to forest vigor according to its scores in the first two PCs
by comparing pixel scores to those associated with field plots. The reflectance PCA, and thus the classification,
was directly interpretable in terms of ecosystem structure because the scored in the first two PCs in field-plot
pixels were correlated to field measurements of forest structure, such as leaf area index, stand density, and
understory composition, and because the field plots occupied interpretable regions of ordination space.
DE: 0439 Ecosystems, structure and dynamics (4815)
DE: 0480 Remote sensing
DE: 0497 Wetlands (1890)
SC: Biogeosciences [B]
MN: 2007 Fall Meeting