Аннотация:The problem of the reality of borders has been the subject of discussion for two hundred yearssince the beginning of the formation of geography as a science. It is difficult to overestimate thepractical significance of this problem, based on the concept of boundaries, on this basis any formof landscape cover segmentation is performed, whether it be a geobotanical special map or forestinventory. At present, remote sensing data are the main tool for spatial assessment of the state ofthe landscape cover (and, therefore, its typification). However, the segmentation of individualimages does not provide an unambiguous division of landscapes, moreover, the very «intensity» ofdivision as a rule is taken out of parentheses. This problem can be considered from the mostgeneral positions - the representation of the ecological space as a set of invariant factors obtainedon the basis of the observed reflections in different spectral ranges (in this case, the system can beconsidered as thermodynamic) and the search for spatial variability to display these factors withdifferent spatial frequencies. The proposed approach solves several related problems that requirethe use of various methods. The first main task of extracting order parameters is carried out by theprincipal components method applied to the remote sensing dataset. With the proposed approach,the order parameters reflect the stationary state of the system unchanged in time and, therefore,are identical to the invariants for the considered time interval. The study of boundaries in a givenspace should theoretically reflect functional changes in ecosystems from the point of view ofthermodynamics. The boundary is defined as the maximum distance according to the Euclideanmetric (or other form of metric) relative to neighboring pixels. The boundaries are calculated fordifferent spatial frequencies, which, according to the Kotelnikov sampling theorem, should giveobjects of different hierarchies. Obviously, at a low frequency, the variety of boundaries issignificantly less than at a high frequency. Many boundaries are often very fragmented at lowfrequencies. At high frequencies, closed boundaries are often distinguished, highlighting smallobjects. Also there is a difference in boundary «thickness» representing different types ofboundaries. The analysis method used can be applied to time series of any variable ofmultispectral measurements and their various transformations