Ozone Pollution in China: Understanding Meteorological Drivers and Modelling Uncertainty

Weng, Xiang (2024) Ozone Pollution in China: Understanding Meteorological Drivers and Modelling Uncertainty. Doctoral thesis, University of East Anglia.

PDF Restricted to Repository staff only until 1 February 2027. Request a copy

Abstract

Surface ozone has emerged as a major air pollutant in China, marked by alarming increases from 2013 to 2019. Effectively mitigating ozone pollution is imperative, which requires a thorough understanding of its drivers and a critical assessment of emission reduction strategies.

To understand the drivers, this thesis focuses on elucidating the meteorological factors through the lens of a data-driven machine learning approach. Specifically, it uses two machine learning algorithms: the nonlinear-capable random forest regression (RFR) and the linear ridge regression (RR) to model the effect of meteorology on surface ozone across China from April to October, 2015–2019. Evaluation of the algorithms’ predictive skill reveals the importance of considering nonlinearity between local meteorological variables and ozone, as evidenced by RFR’s overall higher coefficients of determination (R2) with observations across China. Additionally, RR improves predictions by incorporating nonlocal meteorological predictors, which is particularly effective in southern China, suggesting the importance of large–scale meteorological phenomena on ozone pollution in this region. Both algorithms consistently identify temperature and humidity as the primary meteorological drivers in northern and southern China, respectively, with solar radiation being the most influential driver in the Yangtze River Delta and Sichuan Basin.

On the assessment of emission controls, this thesis highlights substantial uncertainty with WRF–Chem in conjunction with two different chemical mechanisms—CBMZ and MOZART. These mechanisms simulate discrepant summertime ozone changes by 2030 over China’s major city clusters: despite identical emission reductions, MOZART predicts worsening ozone pollution, whereas CBMZ indicates preliminary mitigation. This divergence primarily arises from discrepancies in their simulated ozone responses to nitrogen oxides (NOx) emission reductions. Subsequent process analyses, aided by a machine learning technique, reveal that these discrepancies mainly stem from differing dominant reaction pathways for HO2 and RO2—either primarily engaging in NO-to-NO2 conversion or ROOH formation—in the two mechanisms.

[This abstract contains characters which can't be rendered here, please see original abstract in .PDF]

Item Type:	Thesis (Doctoral)
Faculty \ School:	Faculty of Science > School of Environmental Sciences
Depositing User:	Kitty Laine
Date Deposited:	07 Apr 2025 14:19
Last Modified:	07 Apr 2025 14:46
URI:	https://ueaeprints.uea.ac.uk/id/eprint/98979
DOI:

Actions (login required)

View Item