Automated Interpretation of Multi-Zone Space Images for Snow Depth Recognition: the Case of Western Yakutia

Abstract

The article presents a methodology for mapping the depth of snow cover in 5 areas of Western Yakutia using f ield data and automated interpretation of the depth of snow cover using the unsupervised classification method (classification without training) of a multi-spectral space image obtained in the spring in the area under consideration. Field snow surveys in the study area were carried out in March-April 2024 at 52 points. The depth of snow cover in March ranged from 28 to 70 cm, and its density from 0.12 to 0.21 g/cm3. Landsat-8 / OLI images closest to the dates of field snow surveys were used as initial images to identify differences in the distribution of snow depth in the areas under consideration. We created a map of the depth of snow cover for the areas under consideration Muna, Udachny, Aikhal, Nakyn and Mirny in two stages. The first stage included an analysis of the spatial differentiation of snow cover using a combination of 5–4–3 Landsat-8/OLI bands. Then, to interpret the depth of snow cover, this multispectral image was divided into classes using the unsupervised classification method in the ArcGIS 10.1 program, and the resulting classes were compared with field research materials. According to the results of the conducted study of snow depth mapping, it was revealed that the lowest snow depths are typical for the lower parts of the slopes, as well as for the slopes of windward western and northwestern exposures. The average thickness of the snow cover occurs in the middle and lower parts of the slopes of leeward and, less often, windward exposures. The greatest snow depths are formed on the watershed and upper parts of the slopes of leeward exposures, which is explained by the large amount of snow and increased turbulence of air masses in the upper parts of the watersheds. In addition, the greatest snow thickness is typical and for river valleys, in depressions, as well as on man-made landscapes and residential areas. Comparison of the results of automated decoding (uncontrolled classification) with field snow measurements confirmed the applicability of this method in differentiating the depth of snow cover.