Modeling wind erosion susceptibility of Eastern Iran using machine learning.

Wind erosion and dust storms pose significant environmental and socio-economic threats to Iran's drylands, driven by both natural processes and human activities. This study addresses a critical gap by employing advanced machine learning (ML) techniques to generate high-resolution spatial susceptibility maps for wind erosion in the highly vulnerable eastern and northeastern regions of Iran. We evaluated ten distinct ML models using a comprehensive set of environmental predictors including climatic, soil, topographic, and geological variables to identify areas most prone to erosion. Our results demonstrated that the Random Forest (RF) and Ensemble (ESMs) models outperformed others, achieving high predictive accuracy based on ROC, TSS, and Kappa metrics. These models revealed that approximately 27-30% of the study area is susceptible to moderate or high-intensity wind erosion. Key drivers influencing erosion distribution were identified as soil texture (particularly sand content), specific bio-climatic factors (temperature seasonality and precipitation patterns), and topographic features such as elevation and wetness index. The resulting susceptibility maps provide a vital decision-support tool for policymakers and land managers, enabling targeted mitigation strategies in priority zones. Without effective intervention, these areas face accelerated land degradation, threatening agricultural sustainability and potentially exacerbating socio-economic challenges such as rural-to-urban migration. This study underscores the value of machine learning in environmental hazard assessment and offers a scalable framework for wind erosion management in arid regions worldwide.