Abstract:

Near-surface ozone (O₃) is a secondary atmospheric pollutant, primarily generated through photochemical reactions by nitrogen oxides (NO_x) and volatile organic compounds (VOCs). Coastal urban agglomerations often exhibit a higher degree of industrialization and larger emissions of precursors, leading to more severe O₃ pollution. Meteorological conditions and local circulation systems play critical roles in modulating the formation, accumulation, and transport of O₃. Situated in the East Asian monsoon climate region, Guangdong Province is bordered by the South China Sea to the south and the continent to the north. This unique geographical setting makes it frequently subject to the significant influence of mesoscale circulation systems, such as sea-land breezes. Understanding the interaction between sea-land breezes and O₃ pollution is therefore of great scientific and practical importance for regional air quality management and sustainable environmental development. In this study, Zhuhai and Shanwei were selected as representative cities. Based on monitoring data of air quality and meteorological elements, combined with statistical analysis and machine learning methods, the variation characteristics of O₃ concentration and the driving effect of sea-land breezes were systematically investigated. A random forest model was used in this study, with meteorological factors (e.g., temperature, relative humidity, wind speed, wind direction, air pressure) and pollutant concentrations (e.g., PM_2.5, PM₁₀, NO₂, SO₂, and CO) as input variables to construct an O₃ concentration evaluation model for distinguishing sea-land breeze scenarios. The research results show that: 1) there are significant differences in the long-term variation trends and seasonal characteristics of O₃ concentrations at the two stations. At the Jida Station in Zhuhai, O₃ concentration decreased slowly from 2016 to 2020 and fluctuated and increased from 2020 to 2024, with a seasonal distribution characterized by high values in autumn and low values in summer. At the Xincheng Middle School Station in Shanwei, the O₃ concentration increased continuously from 2016 to 2019 and fluctuated and decreased from 2020 to 2024, with relatively high concentrations in spring and autumn and relatively low concentrations in summer and winter. 2) The daily distribution of sea-land breezes and their intensifying effect on O₃ pollution are regional. From 2016 to 2024, the number of sea-land breeze days at Zhuhai Jida Station and Shanwei Xincheng Middle School Station was 553 days and 863 days respectively. The former was concentrated in January-February, August-September, and November-December, while the latter was concentrated in March-April and August-September. The O₃ exceeding standard rate and concentration on sea-land breeze days at both stations were higher than those on non-sea-land breeze days, but the difference was more significant in Zhuhai. In autumn at Zhuhai, the O₃ exceeding standard rate (96.8%) and the concentration difference (49 μg·m⁻³) were both the largest, and the difference in the exceeding standard rate was prominent in summer and winter. In Shanwei, the difference was the largest in winter (concentration difference of 22 μg·m⁻³) and the smallest in spring and summer. The wind rose diagram verifies that the O₃ concentration in the sea-breeze direction at both stations is significantly higher than that in the land-breeze direction. 3) Long-distance transport combined with local sea-land breeze circulation significantly exacerbates coastal ozone pollution. During the coastal ozone pollution episode in Guangdong from November 17 to 27, 2023, ozone concentrations in Zhuhai and Shanwei on days with sea-land breeze were significantly higher than on days without such circulation. The transport of pollutants from the Yangtze River Delta and the southeastern coast to the Guangdong coastal area, driven by northerly to northeasterly winds, coupled with local sea-land breeze circulation, intensified the pollution through a “transport-formation-recirculation” mechanism under favorable meteorological conditions. A comparison between the two locations indicates that, whether in the core polluted areas or in low ozone areas, sea-land breeze circulation is a key meteorological driver exacerbating coastal ozone pollution. 4) The impacts of pollutants and meteorological factors on O₃ depend on the sea-land breeze scenarios. The relative importance and correlation patterns of various influencing factors varied significantly between sea-land breeze and non-sea-land breeze days. ① Among the pollutants, PM₁₀ and PM_2.5 are the core correlation factors for the O₃ concentrations at the two stations, with the highest correlation coefficients and importance rankings. At the Zhuhai Jida Station, O₃ is negatively correlated with NO₂ and positively correlated with SO₂. In Shanwei, the correlation and importance of particulate matter are significantly weakened on days without sea-land breezes. ② In terms of meteorological factors, on sea-land breeze days, the Zhuhai Jida Station is more affected by temperature, wind direction, and relative humidity, while in Shanwei, relative humidity and temperature are the primary factors. In the scenario without sea-land breezes, the correlations of all meteorological factors are significantly reduced. 5) Distinguishing sea-land breeze scenarios can improve the simulation accuracy of O₃ concentration. The random forest models developed in this study demonstrated that explicitly distinguishing sea-land breeze days significantly improved the predictive performance for O₃ concentrations. For Zhuhai Jida Station, the model achieved an R² of 0.83 for both sea-land breeze and non-sea-land breeze days, compared to 0.82 when no scenario distinction was made. For Shanwei Station, the R² reached 0.83 on sea-land breeze days, versus 0.74 on non-sea-land breeze days and 0.72 in the non-differentiated model. Across all models, R² values exceeded 0.71, with root-mean-square error ε_r≤21 μg·m⁻³. Notably, the normalized root-mean-square error ε_r for sea-land breeze day models was the lowest among all scenarios—0.13 for Zhuhai and 0.12 for Shanwei, indicating superior model stability and accuracy under these well-defined circulation regimes. In summary, this study provides comprehensive evidence that sea-land breeze circulations exert a substantial driving effect on coastal O₃ pollution in Guangdong Province. By integrating long term monitoring data, statistical analysis, and machine learning modeling, this study elucidates the differential effects of pollutants and meteorological factors under various sea-land breeze scenarios and demonstrates the value of scenario-based modeling in enhancing model accuracy. The findings contribute to a clearer understanding of the meteorological driving mechanisms behind ozone pollution and offer significant scientific guidance for formulating differentiated ozone control policies in the coastal regions of Guangdong in future practice.

Key words: coastal area, sea-land breeze, ozone pollution, machine learning, random forest model, Guangdong

摘要：

为研究海陆风对广东沿海大气O₃浓度的影响，该研究以珠海、汕尾为典型城市，基于空气质量与气象要素监测数据，结合统计分析与机器学习方法，系统探究O₃浓度变化特征及海陆风的驱动作用。研究以气象因子和污染物因子为输入变量，构建区分海陆风场景的O₃浓度随机森林评估模型。研究结果表明，1）2016－2024年珠海和汕尾研究站点海陆风日O₃浓度及超标率均高于无海陆风日，且海陆风日和无海陆风日O₃质量浓度分别在秋季和冬季差异最大，差异值分别是49.0 μg·m⁻³和22.0 μg·m⁻³。2）与无海陆风日相比，海陆风日海风方向上的O₃浓度明显偏高。3）2023年11月17－27日，在静稳天气及区域输送背景下，海陆风回流加剧O₃污染，珠海、汕尾海陆风日O₃值分别为167.2 μg·m⁻³和128.2 μg·m⁻³。4）海陆风日PM_2.5、PM₁₀是影响O₃浓度的主要污染物；就气象因子而言，相对湿度与O₃浓度的相关性最大，风向和温度是影响珠海站点O₃浓度重要程度最高的气象变量，在汕尾站点则是相对湿度和温度。5）选取气象因子和污染物因子建立O₃浓度随机森林模型，两个研究站点海陆风日的随机森林模型的R²分别达0.83和0.82，标准化均方根误差ε_r分别低至0.13和0.12。该研究结果明晰了海陆风对沿海O₃污染的驱动效应，可应用于广东沿海地区O₃污染差异化防控及策略制定。

关键词: 沿海地区, 海陆风, 臭氧, 机器学习, 随机森林, 广东