HR: 16:30h
AN: GC34A-03    [Abstracts]
TI: Improved Semi-Arid Vegetation Type Differentiation at Community Level Using MISR Multi-Angular and Multi-Spectral Observations and SVM algorithms
AU: * Su, L
EM: sul@mail.montclair.edu
AF: Department of Earth & Environmental Studies, Montclair State University, 1 Normal Ave., Montclair, NJ 07043 United States
AU: Chopping, M J
EM: chopping@pegasus.montclair.edu
AF: Department of Earth & Environmental Studies, Montclair State University, 1 Normal Ave., Montclair, NJ 07043 United States
AB: Mapping accurately community types is one of main challenges for monitoring semi-arid grasslands with remote sensing. Multi-angle approach has been proved useful for mapping vegetation types in desert. Multi-angle Imaging SpectroRadiometer (MISR) Global Mode provides 275 meters spatial resolution 4 spectral bands (blue, green, red and near-infrared) observations at nadir, and red band 8 multi-angular observations (26.1, 45.6, 60.0 and 70.5 degrees from the vertical both forward and afterward of nadir). And MISR 233 orbit paths make up each 16-day global repeat cycle. In this paper, support vector machine (SVM)-based algorithms are used to classify vegetation types at community level in New Mexico desert. Besides the original MISR observations, the k parameter at red band, which is derived from RPV BRDF model, and structural scattering index, which is derived from volumetric weight at near-infrared band and geometric weight at red band of a linear semi-empirical kernel-driven BRDF model (RossThick-LiSparse-Reciprocal model), are introduced into the SVM algorithms as additional information dimensions. The k parameter is capable to expose surface cover heterogeneity at the sub-pixel scale resolution, and the structural scattering index represents the vegetation structural information in the pixel. So the two additional parameters can bring new information to the SVM classification. In order to get the structural scattering index, MISR 1100 meters spatial resolution near-infrared multi-angle observations are resampled to 275 meters pixel so that the volumetric weight at near-infrared band of the linear semi-empirical kernel-driven BRDF model can be obtained at same spatial resolution of geometric weight at red band. Preliminary results based on part data have showed that 72.9 precent or more accuracy can be expected.
DE: 0400 Biogeosciences
DE: 1630 Impact phenomena
DE: 1640 Remote sensing
DE: 1809 Desertification
DE: 9350 North America
SC: Global Climate Change [GC]
MN: 2005 Joint Assembly