Analysis of Semivariogram Features to Infer Land Cover Textures in Satellite Imagery

Given the importance of identifying Land Covers (LC) for various fields of study, such as the environment and natural disasters, work has been going on to develop tools and sources that provide information on them. In this sense, it is possible to obtain spatial information about LCs with satellite images. In order to obtain said spatial information, it is necessary to extract and determine a set of characteristics that can best describe and infer differences between types of coverage. In this article, an analysis and selection of semivariogram characteristics is carried out to describe the spatial variation in the textures of CTs. The selection of characteristics improves the identification of TC types and the performance of semantic segmentation methods in satellite images. The images used for the analysis experiments corresponded to multispectral satellite images from the Sentinel 2 satellite of COPERNICUS of the European Space Agency (ESA) and Landsat 8 of the United States Geological Survey (USGS). Semivariograms and their characteristics for patches of different TCs (agricultural vegetation, vegetation, soil, water, and urban) were obtained. In addition, coefficients of variation of the characteristics in the semivariograms were analyzed between different and the same hedges. Finally, relevant characteristics were selected considering the behavior of the coefficient of variation and those obtained with the Relieff algorithm.