The Number of Classes in the Cluster Analysis of Space Images for Forest State Monitorin

Abstract

The possibility of combination of two cluster analysis tools for forests state monitoring was assessed. Multispectral space images of high resolution Landsat TM / ETM + and ultra-high resolution RGB-coating DigitalGlobe and their fragments were subjected to two-stage processing. At first unsupervised classification was performed using the ISODATA (Iterative Self-Organizing Data Analysis Technique) method. Then the thematic difference in the classification results was calculated. A relationship between the number of classes and the number of objects defined in the classification into two, four, six, ten classes is shown. The optimal number of classes for distinguishing different levels of structural organization of forest ecosystems are determined. When classifying into two classes in case of high-resolution images forest cover of selected fragments of the images is estimated. When classifying ultra-high resolution images into two classes, the ratio between the area of crowns and the area between the tree crowns is determined. When classifying high-resolution images into four classes, subclasses of more dense and sparse stand, subclasses with open spaces and areas covered with shrubs and woodlands are distinguished.