مدل سازي خطر گردوغبار با استفاده از الگوريتم هاي يادگيري ماشين پشتيبان- مطالعه موردي استان اصفهان

Dust and sandstorms are common natural hazards in arid and semi-arid regions, posing an environmental challenge in recent years. The phenomenon of dust and sandstorms is influenced by various direct and indirect parameters, with climatic and edaphic factors being crucial among them. In this regard, this research focuses on modeling the risk of this phenomenon in Isfahan province due to its importance. Isfahan province, characterized by a dry and semi-dry climate and being adjacent to desert areas, is not exempt from this phenomenon and is under the influence of dust and sandstorm events throughout the year. Therefore, addressing dust and sandstorm events and modeling them to identify sensitive areas is of great importance in this province. For this purpose, machine learning models were employed as powerful tools for modeling this phenomenon and identifying sensitive areas. Hourly dust event codes from the statistical period of 1995 to 2021 were used for eva‎luating dust and sandstorm events. Additionally, maps of land use, soil science, geology, vegetation cover (NDVI index), precipitation, average temperature, relative humidity, maximum wind speed, and topographic humidity were prepared as influential factors on the dust and sandstorm phenomenon at the provincial level. Machine learning algorithms, including SVM, BRT, Glm, and Glmpoly, were used for modeling the risk of dust and sandstorms. The Receiver Operating Characteristic (ROC) curve was also used to compare the performance of the machine learning models. The results of the modeling showed that the area under the curve for BRT and Glm models had the highest and lowest values, respectively, indicating better performance of the BRT model compared to other models. Consequently, the results of this model were considered for creating a map of dust and sandstorm risk in Isfahan province, revealing a higher risk in the eastern half compared to the western part and other regions.

5

35647