Enhancing wind erosion hazard assessment: a novel framework combining integrated land susceptibility to wind erosion (ILSWE) index and machine learning algorithms (case study: Saravan area, Southeast Iran)

Abstract

Wind erosion poses a significant threat to arid and semi-arid ecosystems globally, particularly in the Middle East, a major source of dust emissions. Iran, with extensive arid and semi-arid landscapes, experiences substantial economic and ecological damage from wind erosion, exceeding US$18 billion annually. This study developed a novel methodology for mapping wind erosion hazard in Saravan County, southeast Iran, addressing the critical need for accurate hazard assessment and targeted mitigation strategies. An initial wind erosion inventory map was created using the Integrated Land Susceptibility to Wind Erosion (ILSWE) model, incorporating various factors like climatic erosivity, soil erodibility, soil crust, vegetation cover, and surface roughness. This inventory was then used to train and validate three machine learning (ML) models (Bagged CART, Random Forest, and XGBoost). Model performance was evaluated using Receiver Operating Characteristic (ROC) curve analysis, with the Random Forest model achieving the highest accuracy (AUC = 0.95). Results indicated that 42.7 % of the study area is classified at high or very high hazard for wind erosion, primarily located in western Saravan, characterized by degraded rangelands with sparse vegetation. Key factors influencing wind erosion hazard included elevation, clay content, and calcium carbonate content. This research demonstrates the efficacy of integrating the ILSWE model with ML techniques for accurate mapping of wind erosion hazard, providing valuable information for prioritizing mitigation efforts and promoting sustainable land management practices in arid and semi-arid environments. The developed methodology offers a transferable framework for wind erosion assessment in other vulnerable regions worldwide.