Source-apportionment and spatial distribution analysis of VOCs and their role in ozone formation using machine learning in central-west Taiwan

Manisha Mishra; Pin-Hsin Chen; Wilfredo Bisquera Jr; Guan-Yu Lin; Thi-Cuc Le; Racha Dejchanchaiwong; Perapong Tekasakul; Ciao-Wei Jhang; Ci-Jhen Wu; Chuen-Jinn Tsai

doi:10.1016/j.envres.2023.116329

Source-apportionment and spatial distribution analysis of VOCs and their role in ozone formation using machine learning in central-west Taiwan

Environ Res. 2023 Sep 1:232:116329. doi: 10.1016/j.envres.2023.116329. Epub 2023 Jun 3.

Authors

Affiliations

¹ Institute of Environmental Engineering, National Yang Ming Chiao Tung University, Hsinchu, 30010, Taiwan.
² Department of Environmental Science and Engineering, Tunghai University, Taichung, 407302, Taiwan. Electronic address: samlin@thu.edu.tw.
³ Air Pollution and Health Effect Research Center, And Department of Chemical Engineering, Prince of Songkla University, Songkhla, 90100, Thailand.
⁴ Air Pollution and Health Effect Research Center, And Department of Mechanical and Mechatronics Engineering, Prince of Songkla University, Songkhla, 90100, Thailand.
⁵ Environmental Protection Bureau, Yunlin County, Taiwan.
⁶ Institute of Environmental Engineering, National Yang Ming Chiao Tung University, Hsinchu, 30010, Taiwan. Electronic address: cjtsai@nycu.edu.tw.

PMID: 37276975
DOI: 10.1016/j.envres.2023.116329

Abstract

This study assessed the machine learning based sensitivity analysis coupled with source-apportionment of volatile organic carbons (VOCs) to look into new insights of O₃ pollution in Yunlin County located in central-west region of Taiwan. One-year (Jan 1 to Dec 31, 2021) hourly mass concentrations data of 54 VOCs, NO_X, and O₃ from 10 photochemical assessment monitoring stations (PAMs) in and around the Yunlin County were analyzed. The novelty of the study lies in the utilization of artificial neural network (ANN) to evaluate the contribution of VOCs sources in O₃ pollution in the region. Firstly, the station specific source-apportionment of VOCs were carried out using positive matrix factorization (PMF)-resolving six sources viz. AAM: aged air mass, CM: chemical manufacturing, IC: Industrial combustion, PP: petrochemical plants, SU: solvent use and VE: vehicular emissions. AAM, SU, and VE constituted cumulatively more than 65% of the total emission of VOCs across all 10 PAMs. Diurnal and spatial variability of source-segregated VOCs showed large variations across 10 PAMs, suggesting for distinctly different impact of contributing sources, photo-chemical reactivity, and/or dispersion due to land-sea breezes at the monitoring stations. Secondly, to understand the contribution of controllable factors governing the O₃ pollution, the output of VOCs source-contributions from PMF model along with mass concentrations of NO_X were standardized and first time used as input variables to ANN, a supervised machine learning algorithm. ANN analysis revealed following order of sensitivity in factors governing the O₃ pollution: VOCs from IC > AAM > VE ≈ CM ≈ SU > PP ≈ NO_X. The results indicated that VOCs associated with IC (VOCs-IC) being the most sensitive factor which need to be regulated more efficiently to quickly mitigate the O₃ pollution across the Yunlin County.

Keywords: Artificial neural network (ANN); Ozone; Positive matrix factorization (PMF); Source tracing; Volatile organic compounds (VOCs).

MeSH terms

Air Pollutants* / analysis
China
Environmental Monitoring / methods
Machine Learning
Ozone* / analysis
Taiwan
Vehicle Emissions / analysis
Volatile Organic Compounds* / analysis

Substances

Ozone
Air Pollutants
Volatile Organic Compounds
Vehicle Emissions