COMPARATIVE CHARACTERISTICS OF FOREST MASSIVES USING SEGMENTATION AND CLUSTER ANALYSIS OF LANDSAT IMAGES

Abstract

For the purpose of monitoring the state of forest ecosystems, it is most effective to use the capabilities 2 of remote sensing methods. Fragments with an area of 990 km (scale 1:10,000) were identi ed on the LANDSAT multispectral satellite images of the territory of the Olyokminsky State Nature Reserve. Then polygons were saved at four levels of detail - 4, 16, 64, 256 with scales of 1:5,000, 1:2,500, 1:1,250, 1:675. When decrypting, an uncontrolled classi cation of polygons was carried out using the ISODATA method (Iterative Self-Organizing Data Analysis Technigue) into 2, 10 classes. Distribution curves of the forest cover index values for polygons of the 4th level of segmentation (detail) were built. The nature of the curves was close to the normal distribution, but the curve for the North polygon was atter, which indicates a more even distribution of the forest cover index over all ranges of values. According to the results of the classi cation into 10 classes, statistic processing was carried out with the calculation of indicators of the difference and similarity of polygons: the variance of the general population and the Fisher test (F-test). There were highlighted the most similar pairs of polygons at different levels of segmentation. The results of changes in the variance of the general population and the F-test at different levels of segmentation are considered. The rules for the transition between four levels of self-similarity (scaling) of multifractal structures are described. It was shown that the variance of the tetrad of each level changes by a constant value when moving between the levels of detail, regardless of the variance values within the tetrad. The value of the transition coef cients in the transition between the levels of segmentation (detail) is the difference between the average values of the variances of each of the levels. The value of the transition coef cients during the transition between the segmentation levels is determined by the difference between the variances of the current and the previous segmentation level.