Bin Yu

4. SPATIAL STATISTICS AND REMOTE SENSING

4.1 Identification of Conifer Tree Species

One potential use of successful classification rules based on remote sensing data is tree species identification. It is important to know what is possible from the multispectral data because not everything about a tree is reflected in the selected spectral bands. Spectrometer on the other hand provides intensity readings of hundreds of bands (hyperspectral) so the readings are virtually curves. Such hyperspectral data can come from ground spectrometer or satellites such as TRW, to be launched this year, which has 384 bands from 400 nanometers to 2,500 nanometers. A spectrometer has been used to collect curves on the ground on 600 some trees of 6 species by Professor Gong's group. It is of interest to analyze these curves to understand the within and between species variations and therefore set limits on what is possible with Landsat multispectral data. Spectral reflectance curves have been used for classification purpose in many other areas rather than remote sensing. Jointly with Professor Peng Gong at UC Berkeley and Dr. Mark Hansen at Bell Laboratories, we have been investigating using hyperspectral curves to classify the 6 conifer tree species.