Unsupervised Clustering for Identifying Dusty Days and Their Sources: A Long-Term Data Study in Sanandaj, Iran
Abstract
Dust storms are a persistent environmental hazard in the Middle East, yet long-term monitoring remains limited by sparse ground observations and the lack of objective, reproducible event definitions. This study develops a transferable, data-driven framework for objectively identifying dusty days in data-scarce regions and explicitly addresses two key issues: defining dust events using defensible multi-indicator criteria instead of arbitrary thresholds, and linking objectively detected events to dominant dust source regions. Long-term PM₁₀ records were combined with satellite and reanalysis aerosol indices, including Aerosol Optical Depth at 550 nm (AOD₅₅₀), Dust Optical Depth (DOD), and Dust Aerosol Optical Depth (DUAOD₅₅₀) from CAMS and MERRA-2. Quantitative thresholds derived using statistical analyses and K-Means clustering produced physically consistent criteria (PM₁₀ ≥ 75 μg m⁻³; AOD₅₅₀ ≥ 0.45; DOD ≥ 0.20; DUAOD₅₅₀ ≥ 0.20), capturing the episodic, right-skewed nature of dust variability. Among Boolean rules, OR logic provided the most comprehensive detection (616 dusty days) and remained robust under ±2% perturbations, while Meteorological Aerodrome Reports (METAR) dust and visibility codes confirmed higher recall than restrictive schemes. Using these objectively identified events, 72-h Hybrid Single-Particle Lagrangian Integrated Trajectory (HYSPLIT) backward trajectories with resampling, Principal Component Analysis (PCA), and K-Means clustering revealed two distinct regimes: regional transport from the Iraq-Iran borderlands and southern Mesopotamia, and longer-range intrusions from the Syrian Desert and northern-western Mesopotamia. Unlike previous single-indicator studies, this work provides a unified framework integrating objective thresholding, independent validation, and trajectory source attribution, supporting scalable monitoring, early warning, and risk management in arid and semi-arid regions.
Related
- “Late” effectors from Leptosphaeria maculans as tools for identifying novel sources of resistance in Brassica napus
- Unsupervised probing critical transitions in complex systems using generative diffusion models
- Seismic Precursors to the Blatten, Switzerland Landslide Revealed by Unsupervised Machine Learning
- Unsupervised ecosystem risk assessment in the Baltic Sea reveals bottom-oxygen depletion and fishing pressure as key threats in species-abundant hotspots
- Hydrogeochemical Vectoring to Concealed Porphyry Copper Systems Using Unsupervised Machine Learning: A Case Study from Central British Columbia, Canada
BibTeX
@misc{mirzaei2026unsupervised,
title = {{Unsupervised Clustering for Identifying Dusty Days and Their Sources: A Long-Term Data Study in Sanandaj, Iran}},
author = {Mohammad Momen Haji Mirzaei and Yashar Mohammadi},
howpublished = {EarthArXiv},
year = {2026},
month = sep,
doi = {10.31223/x5wr36},
url = {https://eartharxiv.org/repository/view/14739/},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Unsupervised Clustering for Identifying Dusty Days and Their Sources: A Long-Term Data Study in Sanandaj, Iran
AU - Mohammad Momen Haji Mirzaei
AU - Yashar Mohammadi
PY - 2026
DA - 2026/09/02
PB - EarthArXiv
DO - 10.31223/x5wr36
UR - https://eartharxiv.org/repository/view/14739/
AB - Dust storms are a persistent environmental hazard in the Middle East, yet long-term monitoring remains limited by sparse ground observations and the lack of objective, reproducible event definitions. This study develops a transferable, data-driven framework for objectively identifying dusty days in data-scarce regions and explicitly addresses two key issues: defining dust events using defensible multi-indicator criteria instead of arbitrary thresholds, and linking objectively detected events to dominant dust source regions. Long-term PM₁₀ records were combined with satellite and reanalysis aerosol indices, including Aerosol Optical Depth at 550 nm (AOD₅₅₀), Dust Optical Depth (DOD), and Dust Aerosol Optical Depth (DUAOD₅₅₀) from CAMS and MERRA-2. Quantitative thresholds derived using statistical analyses and K-Means clustering produced physically consistent criteria (PM₁₀ ≥ 75 μg m⁻³; AOD₅₅₀ ≥ 0.45; DOD ≥ 0.20; DUAOD₅₅₀ ≥ 0.20), capturing the episodic, right-skewed nature of dust variability. Among Boolean rules, OR logic provided the most comprehensive detection (616 dusty days) and remained robust under ±2% perturbations, while Meteorological Aerodrome Reports (METAR) dust and visibility codes confirmed higher recall than restrictive schemes. Using these objectively identified events, 72-h Hybrid Single-Particle Lagrangian Integrated Trajectory (HYSPLIT) backward trajectories with resampling, Principal Component Analysis (PCA), and K-Means clustering revealed two distinct regimes: regional transport from the Iraq-Iran borderlands and southern Mesopotamia, and longer-range intrusions from the Syrian Desert and northern-western Mesopotamia. Unlike previous single-indicator studies, this work provides a unified framework integrating objective thresholding, independent validation, and trajectory source attribution, supporting scalable monitoring, early warning, and risk management in arid and semi-arid regions.
ER -