ASSESSMENT OF FACTORS AFFECTING MUDFLOW ACTIVITY USING MACHINE LEARNING METHODS

Abstract

The mountainous northern slope of the Greater Caucasus has been actively developed in recent years, with new resource-intensive industries emerging in the region. Recreational and tourism activities have also seen significant growth. In these circumstances, an important aspect of ensuring security in this region is the study of hazardous natural processes. The most destructive ones are avalanches, landslides, and mudflows, the effects of which often have catastrophic consequences. The paper presents an analysis of data on the characteristics of mudflows conducted with the use of machine learning methods. The study aims to identify the main factors affecting the formation of mudflows in the mountainous areas of the North Caucasus, for which there is no field observation data on the type of hazardous phenomena. The materials of the Mudflow Hazard Inventory for the south of the European Part of Russia were used as input data for the construction of the mudflow type classification model. Different machine learning models, including neural networks, SVM, and logistic regression, were compared in terms of the classification of mudflow types. The results of the study demonstrate a significant superiority of the neural network-based model over the other algorithms. Based on the results obtained for the characteristics of the mudflow basins missing in the Inventory, a schematic map of mudflow basins in the Republic of North Ossetia–Alania, categorized by type of mudflow, was developed. It was compiled with the use of QGIS 3.22.2, which is a software package designed for the creation, visualization, and analysis of spatial data. Vectorization technology was used in the creation of the interactive map. The results obtained can be applied to improve systems for monitoring and forecasting mudflow processes, as well as to develop more effective measures to prevent and mitigate their consequences.