| 基于可解释性机器学习的广州市城区与对照站点O3和PM2.5污染特征及差异分析 |
| 摘要点击 328 全文点击 23 投稿时间：2025-04-17 修订日期：2025-07-31 |
| 查看HTML全文
查看全文 查看/发表评论 下载PDF阅读器 |
| 中文关键词 广州 城区 对照站点 O3 PM2.5 差异 极限梯度提升（XGBoost） Shapley加性解释方法（SHAP） |
| 英文关键词 Guangzhou urban station control site O3 PM2.5 discrepancy extreme gradient boosting （XGBoost） Shapley additive explanation （SHAP） |
| DOI 10.13227/j.hjkx.202504227 |
| 作者 | 单位 | E-mail | | 梁宝玲 | 广东省广州生态环境监测中心站, 广州 510006 | liangbl3@alumni.sysu.edu.cn | | 孙启斌 | 广东省东莞市气象局, 东莞 523086
粤港澳大湾区气象研究院, 东莞市城市生态环境气象工程技术研究中心, 东莞 523086 | | | 张金谱 | 广东省广州生态环境监测中心站, 广州 510006
中国科学院广州地球化学研究所有机地球化学国家重点实验室, 广东省环境资源利用与保护重点实验室, 广州 510640 | | | 梁永健 | 广东省广州生态环境监测中心站, 广州 510006 | | | 蔡明甫 | 生态环境部华南环境科学研究所, 广州 510655 | | | 裴成磊 | 广东省广州生态环境监测中心站, 广州 510006 | peichenglei@163.com |
|
| 中文摘要 |
| 基于2019~2023年广州市O3和PM2.5数据，开展城区与对照站点O3和PM2.5质量浓度的分布和差异特征分析. 结果表明，城区和对照站点O3质量浓度呈现高值波动趋势，PM2.5质量浓度下降显著，O3和PM2.5质量浓度在城区和对照站点差异逐渐缩小，ΔO3与ΔPM2.5的变化趋势从2019~2020年间的协同性逐渐转为2020~2023年的差异性. 城区和对照站点O3月变化分别呈现双峰和三峰分布的特征，PM2.5则一致呈现“U型”分布. 城区和对照站点O3质量浓度呈现秋季>夏季>春季>冬季的特征，O3质量浓度上升趋势显著且O3质量浓度峰值出现时间延迟，由夏季转为秋季，PM2.5质量浓度呈现冬季>秋季>春季>夏季的特征. 基于极限梯度提升（XGBoost）模型和Shapley加性解释方法（SHAP）分析结果表明，地表净太阳辐射和风速是影响城区和对照站点O3质量浓度的最大的因素，超标天时二者贡献程度增加，对照站点对O3前体物（NO2）的光化学生成敏感性更为显著，NOx减排有利于城市背景O3质量浓度的降低. NO2、相对湿度和CO是影响城区站点PM2.5质量浓度的主要因素，对照站点的前3个贡献要素则是相对湿度、NO2和风速. 敏感性分析结果进一步揭示：地表净太阳辐射≥135 W·m-2、风速≤1.8 m·s-1和高于ρ（NO2）下限值（城区/对照站点：34/22 μg·m-3）O3质量浓度显著增加；而在静稳条件（风速≤1.6 m·s-1）、湿度≤78%、高于ρ（NO2）下限值（城区/对照站点：40/22 μg·m-3）和ρ（CO）下限值（城区/对照站点：0.8/0.7 mg·m-3）PM2.5质量浓度增加显著. |
| 英文摘要 |
| Based on the O3 and PM2.5 mass concentration from 2019 to 2023 in Guangzhou， this study analyzed the spatial distribution and difference in O3 and PM2.5 mass concentrations between the urban and control sites. The results revealed that O3 mass concentrations at both the urban and control sites exhibited a high-value fluctuation trend， while PM2.5 mass concentrations declined significantly. The difference in O3 and PM2.5 mass concentration between the urban and control site gradually narrowed. The variation trends of ΔO3 and ΔPM2.5 shifted from a synchronous pattern during 2019 to 2020 to a divergent pattern during 2020 to 2023. Monthly variations in O3 mass concentrations displayed a bimodal pattern at the urban station and a trimodal pattern at the control site， whereas PM2.5 mass concentrations consistently exhibited a “U-shaped” distribution. Seasonal variations showed that O3 mass concentration followed the characteristics of autumn>summer>spring>winter， with a notable upward trend in O3 levels and a shift in peak occurrence from summer to autumn. In contrast， PM2.5 mass concentrations were highest in winter， followed by those in autumn， spring， and summer. Implementing the extreme gradient boosting （XGBoost） model and the Shapley additive explanation （SHAP）， the analysis identified surface net solar radiation and wind speed as the most significant factors affecting O3 mass concentrations at both the urban and control sites. Their contributions increased on polluted days， and control sites were more sensitive to the photochemical formation of ozone precursors （NO2）， indicating that NOx emission reductions were more effective in lowering background O3 levels. For PM2.5 mass concentrations， NO2， relative humidity， and CO were the primary influencing factors at the urban station， while relative humidity， NO2， and wind speed were the top contributors at the control site. Sensitivity analysis further revealed that O3 mass concentrations significantly increased under conditions of surface net solar radiation≥135 W·m-2， wind speed≤1.8 m·s-1， and NO2 concentrations exceeding the lower threshold （urban/control sites： 34/22 μg·m-3）. PM2.5 mass concentrations significantly increased under stagnant conditions （wind speed≤1.6 m·s-1）， humidity≤78%， and concentrations exceeding the lower thresholds of NO2 （urban/control sites： 40/22 μg·m-3） and CO （urban/control sites： 0.8/0.7 mg·m-3）. This study provides insights into the long-term evolution of O3 and PM2.5 mass concentrations at urban and control sites and offers quantitative evidence to support the development of differentiated strategies for controlling regional complex pollution. |
