Influence of weather and air pollution on concentration change of PM2.5 using a generalized additive model and gradient boosting machine

• Published in: Atmospheric environment (1994) 2021-06, Vol.255, p.118437, Article 118437
• Main Authors: Cheng, Bowen, Ma, Yuxia, Feng, Fengliu, Zhang, Yifan, Shen, Jiahui, Wang, Hang, Guo, Yongtao, Cheng, Yifan
• Format: Article
• Language: English
• Subjects: Air pollution; Generalized additive model; Gradient boosting machine; Meteorological factors; PM2.5
• ISSN: 1352-2310; 1873-2844
• DOI: 10.1016/j.atmosenv.2021.118437

Particulate matter with a diameter of ≤2.5 μm (PM2.5) is a critical air pollutant that adversely affects human health and the ecological environment. Using the data of daily air pollutant concentration and meteorological elements from 2013 to 2020 in Lanzhou, China, we employed the generalized additive model (GAM) and gradient boosting machine (GBM) approaches to analyze the relationship between PM2.5 concentration and environmental factors. The results revealed that the annual average PM2.5 concentration in Lanzhou, China, decreased by 3.98 μg/m3 per year during the study period and the variation of PM2.5 concentration was influenced by many factors. Mean temperature, air pressure, relative humidity, and O3 concentration had significantly negative effects on PM2.5 concentration, whereas CO and NO2 concentrations had significant positive effects. PM2.5 concentration was also high during static winds or high-speed winds. Among these influencing factors, the interactions of meteorological factors and air pollution also had strong correlations with PM2.5 concentration. We found that the multi-factor model could better explain the influence of environmental factors on PM2.5 concentration than the single-factor model. The fitting degree of the GBM was better than that of the GAM model; their mean absolute errors were 11.30 and 11.85, respectively. We also found that the GBM model was more precise for estimating daily PM2.5 concentration, whereas the GAM was more suitable for long-term trend analysis. •The variation of PM2.5 concentration is influenced by multiple factors and their interactions.•The multi-factor model could better explain the influence of environmental factors on PM2.5 concentration.•The fitting degree of the GBM was better than that of the GAM model.•GBM was more suitable for short-term prediction, whereas GAM was more suitable for long-term trend analysis.