HR: 09:45h
AN: H31L-08 [Abstracts]
TI: Comparison of Change Detection Techniques for Assessing Hurricane Katrina-Induced Damage to Forests
AU: Xu, Y
EM: yjxu@lsu.edu
AF: Louisiana State University Agricultural Center,School of Renewable Natural Resource, 106
RNR Bldg., Baton Rouge, LA 70803, United States
AU: * Wang, F
EM: fwang5@lsu.edu
AF: Louisiana State University Agricultural Center,School of Renewable Natural Resource, 106
RNR Bldg., Baton Rouge, LA 70803, United States
AB:
This study compared performance of four change detection algorithms with six vegetation indices derived from
Landsat Thematic Mapper (TM) imagery taken pre\- and post\-Hurricane Katrina. The overall goal of the study was
to select an optimal remote sensing approach for identifying disturbed forests by the hurricane in the Lower Pearl
River Valley, USA. The algorithms included univariate image differencing (UID), selective principal component
analysis (selective PCA), change vector analysis (CVA), and post-classification comparison (PCC). The indices
consisted of near-infrared to red ratios (RVI), normalized difference vegetation index (NDVI), Tasseled Cap index
of greenness (TCG), brightness (TCB) and wetness (TCW), and soil-adjusted vegetation index (SAVI). In addition
to the satellite imagery, "ground truth" data of forest damage were also collected through field investigation and
interpretation of post\-Katrina aerial photos. Disturbed forests were identified by classifying the composite and the
continuous change imagery with the supervised classification method. Results showed that the change
detection techniques largely affected the results with an overall detection accuracy varying between 59% and
86% and with a Kappa Statistics ranging from 0.15 to 0.72. Detected areas of disturbed forests were noticeable
in two groups: 186,625 \- 264,617 ha and 106,783 \- 124,205 ha. The PCC algorithm along with the composite
image contributed the highest accuracy and lowest errors (0.5%) in estimating disturbed forested land areas.
Both UID and CVA performed similarly, but caution should be taken when using selective PCA in detecting
hurricane disturbance to forests. Among the six indices, TCW outperformed the other indices owing to its
maximum sensitivity to the forest modification. This study suggests that compared with the detection algorithms,
proper selection of vegetation indices is more critical for obtaining a satisfactory result.
DE: 0468 Natural hazards
DE: 0480 Remote sensing
SC: Hydrology [H]
MN: 2007 Fall Meeting