| 引用本文: | 马德栗,湛甜,鞠英芹,王凯,李斌,杜良敏.2026.湖北省13个地级行政区O₃浓度演变及预报研究[J].地球环境学报,17(3):810-827 |
| MA Deli,ZHAN Tian,JU Yingqin,WANG Kai,LI Bin,DU Lianming.2026.Research on the evolution and prediction of O₃ concentration in 13 prefecture-level administrative regions of Hubei Province[J].Journal of Earth Environment,17(3):810-827 |
|
| |
|
|
| 本文已被：浏览 95次 下载 4次 |
码上扫一扫！ |
| 湖北省13个地级行政区O₃浓度演变及预报研究 |
|
马德栗1，2，湛甜1，鞠英芹3，王凯1，李斌4，杜良敏1
|
|
1. 湖北省气候中心，武汉 430074 ；2. 潜江市气象局，潜江 433199 ；3. 湖北省气象工程技术中心，武汉 430074 ；4. 山西省忻州市气象局，忻州 034000
|
|
| 摘要: |
| 文章采用2015—2023年湖北省13个地级行政区O3-8h逐日质量浓度和气温、降水、相对湿度及风速等气象因子数据集，分析O₃时空演变特征，开展基于动态时间弯曲（DTW）法的O₃质量浓度日变化聚类，并探索O₃日质量浓度多种机器学习方法预报及效果评估。结果表明：2015—2023年湖北省O₃年平均质量浓度在79.30—99.79 μg/m³，呈微弱上升趋势，春、夏季浓度较高，秋、冬季较低；月平均浓度在48.7—120.0 μg/m³，呈明显的“M”双峰型特征，中东部地区极值分别出现在6月和9月；2020年之前中东部O₃出现“周末效应”，2020年之后则相反。基于动态时间弯曲法和手肘法将湖北省O₃日质量浓度分为4类，其中，武汉至鄂东北大别山南麓的Ⅲ类地区O₃质量浓度最高，超过国家一级标准占比集中在42%—46%，超过二级标准占比集中在7%—9%。利用机器学习模型沙普利加性解释（SHAP）方法分析典型城市影响O₃浓度的主要气象因子，确定4类地区的O₃预报特征量，对比4种机器学习预报方法在宜昌、荆州、武汉和咸宁4个城市的逐日O₃浓度预报效果，长短期记忆网络（LSTM）表现最优。运行LSTM预报2023年5—10月13个地级行政区逐日O₃质量浓度大于国控一级，成功率均在70%以上，漏报率5%—12%，具有较好的预报能力。 |
| 关键词: 动态时间弯曲（DTW） O₃浓度预报 沙普利加性解释（SHAP） |
| DOI：10.7515/JEE2024116 |
| CSTR：32259.14.JEE2024116 |
| 分类号: |
| 基金项目:2023年度武汉市气象科技联合项目（2023020201010581）；2014年度中国地质大学（武汉）科学技术研究课题 |
| 英文基金项目: |
|
| Research on the evolution and prediction of O₃ concentration in 13 prefecture-level administrative regions of Hubei Province |
|
MA Deli1，2，ZHAN Tian1，JU Yingqin3，WANG Kai1，LI Bin4，DU Lianming1
|
|
1. Hubei Climate Center, Wuhan 430074 , China ；2. Qianjiang Meteorological Bureau, Qianjiang 433199 , China ；3. Hubei Meteorological Engineering and Technology Center, Wuhan 430074 , China ；4. Xinzhou Meteorological Bureau of Shanxi Province, Xinzhou 034000 , China
|
| Abstract: |
| Background, aim, and scope With the accelerated development of urbanization, O₃ pollution in China is becoming increasingly serious, not only affecting economic and social development, but also posing a serious threat to human physical and mental health. Meteorological factors such as temperature, precipitation and relative humidity are important conditions for O₃ pollution. Hubei Province is located in the middle reaches of the Yangtze River Economic Belt, and in recent years, O₃ pollution has become increasingly serious, especially in the Wuhan urban area, where O₃ concentrations exceed the standard in summer and has become the primary air pollutant. The spatial and temporal variations and influencing factors of air quality in major urban areas of Hubei Province are still unclear. This study focuses on the spatial and temporal evolution of O₃ and concentration forecasting to provide data support and references for understanding urban air quality conditions and developing environmental management. Materials and methods This article analyses the spatial and temporal characteristics of O₃ concentration using a dataset of the maximum daily 8-h average O₃ mass concentration, daily temperature, precipitation, relative humidity, and wind speed from 13 prefecture-level administrative regions in Hubei Province during 2015—2023: Shiyan, Xiangyang, Yichang, Enshi, Suizhou, Jingzhou, Jingmen, Wuhan, Ezhou, Huanggang, Xiaogan, Huangshi, and Xianning. Based on the dynamic time warping (DTW) method, clustering of daily O₃ mass concentration changes was carried out, and various machine learning methods for predicting and evaluating daily O₃ mass concentration are explored. Results From 2015 to 2023, the annual average concentration of O₃ in Hubei Province ranged from 79.30 μg/m³ to 99.79 μg/m³, showing a slight upward trend. Concentrations were higher in spring and summer and lower in autumn and winter. The monthly average concentration ranges from 48.7 μg/m³ to 120.0 μg/m³, with a clear double-peak pattern characterized by extreme values in most areas occurring in June and September, respectively. Before 2020, the average quality concentration of Oon weekends in the central and eastern regions was higher than that on weekdays, resulting in a weekend effect. Based on the dynamic time warping method and elbow method, the daily O₃ mass concentration in Hubei Province was classified into four categories, in which the highest O₃ mass concentration was found in Category Ⅲ, from Wuhan to the southern foot of the Dabie Mountains in northeastern Hubei, where the exceedance rates of the national first- and second-level standards were the highest. The shapley additive explanations (SHAP) values of the main meteorological factors affecting O₃ mass concentration in four different regions were analyzed, and the characteristic variables of O₃ forecasting in typical cities were determined. Comparing the forecasting effects of four machine learning forecasting methods in different areas, the long short-term memory (LSTM) method performed best. Discussion The O₃ mass concentrations in 13 prefecture-level administrative regions in Hubei Province are strongly influenced by season and region. Due to multiple factors such as reduced human activities, the level of O₃ pollution in 2020 decreased compared to that before the pandemic. The average O₃ mass concentrations of 2020 in Hubei Province decreased in spring and summer, with the largest decrease in the summer months, which was attributed to reduced emissions from industrial and mobile sources during the pandemic, including pollutants such as NO₂, SO₂, CO, and VOCs. It is worth noting that in summer, a significantly high O₃ concentration area appeared in the Shiyan and Xiangyang areas of northwestern Hubei, which was mainly affected by mountainous terrain and sinking trans-mountain airflow, resulting in summer temperatures prone to extremes, promoting the generation of local O₃, and further exacerbating O₃ pollution. Similar to areas such as Shanxi and Yunnan, a weekend effect is observed in central-eastern Hubei Province, but an anti-weekend effect is observed after 2020, which may be mainly related to reduced emissions from pollutant sources and changes in human activity pattern. In the summer of 2023, a high O₃ pollution period occurred in several cities in Hubei Province, with serious O₃ pollution in May and August, and the analysis of the effectiveness of four machine learning models for O₃ concentration prediction found that the long short-term memory network was the most effective. This indicates a certain level of forecasting ability, but the parameters still need to be further optimized. Conclusions The concentration of Ois influenced by various meteorological factors such as temperature, precipitation, and wind speed, and the LSTM method has a certain predictive ability for the daily O₃ concentration prediction. Recommendations and perspectives This study reveals the changing pattern of O₃ concentration and its relationship with meteorological factors, and provides a predictive method, which provides a basis for the prevention and control of O₃ in Hubei Province. Meanwhile, these insights provide a scientific foundation for future strategies for reducing air pollution in central Chinese cities. |
| Key words: dynamic time warping (DTW) O₃ concentration prediction shapley additive explanations (SHAP) |
|
|
|
|
|
