Abstract
Logarithmic Residual (LR), Least Upper Bound Residual (LUB) or Gray Scale Log Residual (GSLR) methods have been used for the purpose of converting digital numbers (DNs) of multiband remote sensing data to apparent band-to-band reflectance patterns. However, converted values by these methods show the specific distribution pattern in theN-dimension space defined by the N-band data, because these methods reduce the dimension toN-1. Thus, application of the spectral indices, which was previously developed for the original DN ofN-dimension, to the converted values may lead incorrect index values.
The authors propose Spectral Pattern Index (SPI), which allows us to measure a band-to-band pattern similarity (“spectral pattern” and “spectral contrast”) convertedN-1 dimension values by the LR, LUB and GSLR methods. The Landsat TM data of Teshionakagawa area in northern Hokkaido, Japan, was processed by this new index after conversion to apparent reflectance pattern using the GSLR method. The result show more detailed information on vegetation and lithology.
