A Method of Mapping Seasonal Dynamics of Open Sand Areas and Deflated Territories Based on High-Resolution Satellite Data and Machine Learning

The increase in the areas of open sands and deflated territories in recent years has become a significant issue in the southeastern part of the European territory of Russia. Degradation of vegetation cover leads to intensification of deflation processes, culminating in catastrophic dust storms, which result in an even greater expansion of open sand areas. In this context, satellite monitoring of seasonal and long-term changes in deflated areas devoid of vegetation cover is highly relevant. Automated classification methods for satellite images necessitate a labor-intensive procedure for creating reference samples for each date, and expert interpretation is characterized by even greater labor costs. A potential solution may involve the use of universal expert threshold values for the maximum and minimum values of vegetation indices and spectral brightness coefficients. This study proposes an approach for forming reference samples based on expert thresholds, which has been substantiated using aerial imagery obtained from unmanned aerial vehicles. It was established that the spectral brightness coefficient in the red range and the Normalized Difference Vegetation Index exhibit the closest correlation with the proportion of open sand area within a pixel. The proposed approach enables generation of maps of open sands and deflated territories at least three times a year on the basis of three-month composite images from Sentinel-2 with a spatial resolution of 10 m. The verification of the results demonstrated the applicability of the method for mapping seasonal dynamics of open sand areas and deflated territories in the Northwestern Caspian region.