Identification of invariant states of agricultural landscapes based on hierarchical factor analysis of remote sensing information

The proposed work shows the possibilities of analyzing the time series of multispectral satellite images of medium resolution to identify invariant spatial structures under conditions of strong anthropogenic impact. The procedure for selecting spatial invariants is shown on the example of multi-year LANDSAT data series. The procedure is based on applying the method of hierarchical factor analysis of the multidimensional space of initial variables and segmentation of the resulting factor space into discrete states. Since each multispectral image is a cross-section of the state of the landscape cover and its ability to convert solar energy, each cross-section must have its own set of spatial invariants that describe most of the information about the work of the surface. From the invariant describing individual cross-sections, it is possible to construct new invariants that describe the stationary states of the time series as a whole. Using an iterative K-means procedure, the spatial differentiation of these states can be distinguished. The number of such states is determined by the principle of maximum entropy. The stability of the obtained discrete states is investigated using discriminant analysis, when the invariant states obtained from one time series serve as a training sample for another time series of images for the same territory. In this article, this approach is used to study the agricultural landscapes of the Samara region. We demonstrate high repeatability of discrete spatial invariants in the space of integral factors for time series separated in time by 10 years. Independently conducted field studies suggest that the selected stationary states can be correlated with soil types. This type of analysis makes it possible to establish the genetic properties of a territory, using remote methods, even under strong anthropogenic influence.