Application of artificial intelligence and GIS technologies for zoning areas at risk of sand drift on highways

Abstract

Sand drift and sandstorms pose a complex hazard to highways in arid regions, causing reduced visibility, deterioration of pavement surface friction properties, and increased operating costs [1], [2]. According to recent estimates, drylands cover 40.6% of the Earth’s land surface, while sand and dust storms affect approximately 330 million people in more than 150 countries [1], [3]. This thesis proposes an approach based on the integration of artificial intelligence (AI), remote sensing, and GIS for identifying and zoning areas at risk of sand drift on highways into five categories. By combining Sentinel-1/2, Landsat, ERA5-Land, DEM, and road geometry data, it is shown that the probability of risk can be dynamically assessed for road sections. The literature review reports AUC values of 96.2% for the RF model, 0.94 for SVM, and IoU values of up to 89% for U-Net in desert road extraction [5], [6], [7]. As a result, a practical zoning scheme was developed to support monitoring, prevention, and investment prioritization for roads passing through arid regions of Uzbekistan and adjacent risk zones.