ANALYSIS OF METHODS FOR MAPPING THE VEGETATION COVER AT THE KAZAN-VESHENSKY SAND MASSIF

The article discusses two different ways to classify space images of Landsat-8 satellite by the example of the sand massif in the Kazan-Veshensky key area, the Rostov-on-Don region. The first method is a semi-automatic raster classification with training (SC) of selected reference areas, the second is a normalized vegetation index (NDVI). Taking into consideration K.N. Kulik typology of sands, the following classes are distinguished in the satellite image, i.e., open, slightly overgrown and overgrown sands. Also, shrub and herbaceous (plants in the vegetative state), pine forest plantations, and native forest with tree splits were marked as individual classes. The degree of sand overgrowth with native vegetation was estimated according to the projective cover. The resulting raster images were vectorized for further work. The estimation of cartographic image classification accuracy by the Cohen’s Kappa index was calculated. This work is necessary to identify the most reliable method for deciphering the selected area images. The resulting map can be used for initial assessment of phytoecological conditions of agrolandscapes in the sandy massif area. On the basis of cartographic images, by setting from 70 to 100 points in each class and checking their reliability by using archival satellite image for 07.07.2020, error matrices were complied, which permitted us to calculate the total interpretation accuracy. For semi-automatic classification with training, it constitutes 80.7%, and NDVI – 74.3%. According to the vegetation index with a smaller error, the classes of open sands and weakly overgrown sands were distinguished, in other cases the SC method turned out to be more accurate. Cohen’s Kappa coefficient in the semi-automatic classification with training is 77.4%, NDVI – 70.5%. The difference in classification accuracy is almost 7%. Thus, the optimal method for preliminary analysis of the Kazan-Veshensky sand massif key area using Landsat-8 satellite images is a semi-automatic classification with training.