Modeling wind erosion susceptibility of Eastern Iran using machine learning

Area of Science:

• Environmental Science
• Geosciences
• Data Science

Background:

• Wind erosion and dust storms present significant environmental and socio-economic challenges in Iran's drylands.
• These hazards are exacerbated by natural processes and human activities, necessitating effective management strategies.

Purpose of the Study:

• To develop high-resolution spatial susceptibility maps for wind erosion in vulnerable eastern and northeastern Iran.
• To identify and rank the most effective machine learning models for predicting wind erosion risk.
• To pinpoint critical environmental factors driving wind erosion in the region.

Main Methods:

• Evaluated ten machine learning (ML) models, including Random Forest (RF) and Ensemble models (ESMs).
• Utilized a comprehensive dataset of climatic, soil, topographic, and geological variables as predictors.
• Assessed model performance using metrics such as ROC, TSS, and Kappa.

Main Results:

• Random Forest and Ensemble models demonstrated superior predictive accuracy for wind erosion susceptibility.
• Approximately 27-30% of the study area exhibits moderate to high susceptibility to wind erosion.
• Key contributing factors identified include soil texture (sand content), bio-climatic variables (temperature seasonality, precipitation), elevation, and wetness index.

Conclusions:

• High-resolution susceptibility maps generated by ML models serve as crucial decision-support tools for land managers and policymakers.
• Targeted mitigation strategies in identified priority zones are essential to prevent land degradation and associated socio-economic impacts.
• The study highlights the utility of machine learning for environmental hazard assessment and provides a scalable framework for arid region wind erosion management.