Scholarlink

<p>Hazardous air pollution increasingly threatens environmental quality and public health. However, limited ground-based monitoring constrains understanding of long-term PM<sub>2.5</sub> variability and its meteorological drivers. We aim to investigate long-term PM<sub>2.5</sub> trends, meteorological influences, and predictive performance using statistical and machine-learning models based on the Modern-Era Retrospective Analysis for Research and Applications (MERRA-2) data from January 2000 to August 2025 across the East–Central–Southern Africa domain (20°S–0°N, 20°–40°E). PM<sub>2.5</sub> peaks at 15–25&#xa0;µg/m<sup>3</sup> (June–September) and drops below 7&#xa0;µg/m<sup>3</sup> (April). A statistically significant long-term increase of 0.0500&#xa0;µg/m<sup>3</sup>/yr (95% CI: 0.0216 to 0.0784&#xa0;µg/m<sup>3</sup>/yr; p = 0.0006) keeps PM<sub>2.5</sub> annual averages (10–13 µg/m<sup>3</sup>) above the WHO-2021 guideline (5&#xa0;µg/m<sup>3</sup>). Standardized regression coefficients from the multiple linear regression model show that thermal variables drive PM<sub>2.5</sub> changes most: Temperature (air (29.65%), skin (26.64%), and dew point (21.47%)), and precipitation (13.46%), while weaker effects (2.44–3.32%) are observed for pressure, wind speed, and boundary layer. Ensemble models (Gradient Boosting and Random Forest) performed best (slope = 0.97 ± 0.01, R2 = 0.85), followed by Ridge (0.96 ± 0.01, R2 = 0.75) and Lasso (0.94 ± 0.01, R2 = 0.64). Long-term drift of small negative residual trends (–0.034 to –0.076&#xa0;µg/m<sup>3</sup>/yr) was observed, while in recent years (2023–2025), variability increased (± 6&#xa0;µg/m<sup>3</sup>). Overall, the findings and methodology provide a scalable framework for assessing PM<sub>2.5</sub> risks in data-scarce regions worldwide and offer important insights for air quality management and understanding interactions between climate variability and air pollution.</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

Statistical and Machine Learning Modeling of Long-Term PM2.5 Variability Across East, Central, and Southern Africa

  • 期刊论文

  • 发表日期：

    2026年4月29日

查看全文
  • Irankunda Elisephane,
  • Moriasi Kevin Ming’ate

摘要

Hazardous air pollution increasingly threatens environmental quality and public health. However, limited ground-based monitoring constrains understanding of long-term PM2.5 variability and its meteorological drivers. We aim to investigate long-term PM2.5 trends, meteorological influences, and predictive performance using statistical and machine-learning models based on the Modern-Era Retrospective Analysis for Research and Applications (MERRA-2) data from January 2000 to August 2025 across the East–Central–Southern Africa domain (20°S–0°N, 20°–40°E). PM2.5 peaks at 15–25 µg/m3 (June–September) and drops below 7 µg/m3 (April). A statistically significant long-term increase of 0.0500 µg/m3/yr (95% CI: 0.0216 to 0.0784 µg/m3/yr; p = 0.0006) keeps PM2.5 annual averages (10–13 µg/m3) above the WHO-2021 guideline (5 µg/m3). Standardized regression coefficients from the multiple linear regression model show that thermal variables drive PM2.5 changes most: Temperature (air (29.65%), skin (26.64%), and dew point (21.47%)), and precipitation (13.46%), while weaker effects (2.44–3.32%) are observed for pressure, wind speed, and boundary layer. Ensemble models (Gradient Boosting and Random Forest) performed best (slope = 0.97 ± 0.01, R2 = 0.85), followed by Ridge (0.96 ± 0.01, R2 = 0.75) and Lasso (0.94 ± 0.01, R2 = 0.64). Long-term drift of small negative residual trends (–0.034 to –0.076 µg/m3/yr) was observed, while in recent years (2023–2025), variability increased (± 6 µg/m3). Overall, the findings and methodology provide a scalable framework for assessing PM2.5 risks in data-scarce regions worldwide and offer important insights for air quality management and understanding interactions between climate variability and air pollution.