Journal of Atmospheric and Environmental Optics, Volume. 20, Issue 2, 176(2025)

Estimation method of atmospheric particulate matter mass concentration based on machine learning and multi-source data

YANG Huirong1,2, HE Nanteng1,2, BU Lingbing1,2、*, MO Zusi1,2, FAN Zengchang1,2, ZHOU Xiaomeng1,2, and SU Xin1,2
Author Affiliations
  • 1Key Laboratory for Aerosol-Cloud-Precipitation of China Meteorological Administration, Nanjing University of Information Science and Technology, Nanjing 210044, China
  • 2School of Atmospheric Physics, Nanjing University of Information Science and Technology, Nanjing 210044, China
  • show less
    Figures & Tables(8)
    Flow chart of model building
    Rank of importance of input factors. (a) PM2.5 under clean background; (b) PM2.5 under polluted background;(c) PM10 under clean background; (d) PM10 under polluted background
    [in Chinese]
    Optimal factor verification results of ML models. (a) Clean environment (RF model: 8 and 9 input factors for PM2.5 and PM10 respectively); (b) polluted environment (GRNN model: 6 and 8 input factors for PM2.5 and PM10 respectively)
    Horizontal distribution of PM2.5 (a) and PM10 (b) mass concentration estimated by optimal ML model (19:00, December 21, 2021, Beijing time)
    Comparison between the data of the state Control site and the model inversion data for PM2.5 (a) and PM10 (b)
    • Table 1. Optimal model and number of input factors under different conditions

      View table
      View in Article

      Table 1. Optimal model and number of input factors under different conditions

      Particulate matter in different environmentsOptimization modelNumber of optimal input factors
      PM10 (clean)RF9
      PM2.5 (clean)RF8
      PM10 (polluted)GRNN8
      PM2.5 (polluted)GRNN6
    • Table 2. Models used in different backgrounds and their verification results

      View table
      View in Article

      Table 2. Models used in different backgrounds and their verification results

      Particulate matter in different environmentsOptimization modelNumber of optimal input factorsR2EMA/(μg·m-3)ERMS/(μg·m-3)
      PM10 (clean)RF90.7310.0213.79
      PM2.5 (clean)RF80.847.4310.98
      PM10 (polluted)GRNN80.7613.1817.76
      PM2.5 (polluted)GRNN60.888.8713.80
    Tools

    Get Citation

    Copy Citation Text

    Huirong YANG, Nanteng HE, Lingbing BU, Zusi MO, Zengchang FAN, Xiaomeng ZHOU, Xin SU. Estimation method of atmospheric particulate matter mass concentration based on machine learning and multi-source data[J]. Journal of Atmospheric and Environmental Optics, 2025, 20(2): 176

    Download Citation

    EndNote(RIS)BibTexPlain Text
    Save article for my favorites
    Paper Information

    Category:

    Received: Jan. 31, 2023

    Accepted: --

    Published Online: May. 30, 2025

    The Author Email: Lingbing BU (lingbingbu@nuist.edu.cn)

    DOI:10.3969/j.issn.1673-6141.2025.02.006

    Topics