Mapping of salty aeolian dust-source potential areas: Ensemble model or benchmark models?

Area of Science:

• Environmental Science
• Remote Sensing
• Machine Learning

Background:

• Dust emissions significantly impact human health, environment, agriculture, and transportation.
• Accurate assessment of land susceptibility to dust is crucial for mitigation strategies.

Purpose of the Study:

• To evaluate the capability of different machine learning models in analyzing land susceptibility to dust emissions.
• To identify dust-source areas and predict susceptibility using advanced modeling techniques.

Main Methods:

• Identified dust-source areas using Aerosol Optical Depth (AOD) from MODIS (2000-2020) and field surveys.
• Employed Weighted Subspace Random Forest (WSRF) and compared it with General Linear Model (GLM), Boosted Regression Tree (BRT), and Support Vector Machine (SVM).
• Determined the importance of various dust-driving factors.

Main Results:

• The WSRF model demonstrated superior performance over benchmark models, achieving over 97% accuracy, Kappa, and probability of detection, with a false alarm rate below 1%.
• Spatial analysis revealed a higher frequency of dust events around the outskirts of Urmia Lake.
• High to very high dust emission susceptibility was identified in salt lands (4.5%), rangelands (2.8%), agricultural lands (1.8%), dry-farming lands (0.8%), and barren lands (0.2%).

Conclusions:

• The Weighted Subspace Random Forest (WSRF) model is highly effective for precise mapping of land susceptibility to dust emissions.
• The study provides critical insights for dust management and environmental planning, particularly in arid and semi-arid regions.