Scholarlink

<p>Airborne particulate matter (PM₁₀) poses major environmental and health challenges in arid regions, yet its variability remains poorly understood due to complex climate-land interactions. This study aims to identify key climatic and environmental processes influencing PM₁₀ variability in the Lake Urmia Basin (LUB) from 2000 to 2023 using a comprehensive set of 19 drivers. We applied Singular Spectrum Analysis and the Mann-Kendall test to detect long-term climatic trends and used Spearman’s correlation to explore pairwise associations. To model nonlinear and interactive effects, we developed Generalized Additive Models (GAMs) and interpreted relationships using Partial Dependence Plots (PDPs). Our results reveal intensified aridity, characterized by rising air and soil temperatures, increased evaporation, and reduced precipitation and soil moisture, which collectively weakened natural dust-suppression mechanisms and elevated PM₁₀ levels. GAMs achieved high predictive performance (R² = 0.88, RMSE = 3.31, MAE = 2.41&#xa0;µg/m³) and highlighted key nonlinear and interaction effects, including unimodal PM₁₀ responses to temperature and synergistic amplification between barren land and aerosol optical depth. Conversely, moisture-related variables mitigated PM₁₀ through wet deposition and soil stabilization. These findings demonstrate the dominant influence of dryness and land degradation on dust formation while underscoring the value of interpretable machine learning for disentangling complex environmental processes and guiding air-quality management and climate adaptation strategies in dust-prone regions such as the LUB.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Determining Nonlinear and Interactive Processes Driving PM₁₀ Variability via Interpretable Machine Learning

  • 期刊论文

  • 发表日期：

    2026年2月18日

查看全文
  • Razi Sheikholeslami,
  • Hadiseh Khaleghi,
  • Ali Karimaee

摘要

Airborne particulate matter (PM₁₀) poses major environmental and health challenges in arid regions, yet its variability remains poorly understood due to complex climate-land interactions. This study aims to identify key climatic and environmental processes influencing PM₁₀ variability in the Lake Urmia Basin (LUB) from 2000 to 2023 using a comprehensive set of 19 drivers. We applied Singular Spectrum Analysis and the Mann-Kendall test to detect long-term climatic trends and used Spearman’s correlation to explore pairwise associations. To model nonlinear and interactive effects, we developed Generalized Additive Models (GAMs) and interpreted relationships using Partial Dependence Plots (PDPs). Our results reveal intensified aridity, characterized by rising air and soil temperatures, increased evaporation, and reduced precipitation and soil moisture, which collectively weakened natural dust-suppression mechanisms and elevated PM₁₀ levels. GAMs achieved high predictive performance (R² = 0.88, RMSE = 3.31, MAE = 2.41 µg/m³) and highlighted key nonlinear and interaction effects, including unimodal PM₁₀ responses to temperature and synergistic amplification between barren land and aerosol optical depth. Conversely, moisture-related variables mitigated PM₁₀ through wet deposition and soil stabilization. These findings demonstrate the dominant influence of dryness and land degradation on dust formation while underscoring the value of interpretable machine learning for disentangling complex environmental processes and guiding air-quality management and climate adaptation strategies in dust-prone regions such as the LUB.