摘要
本研究以东北地区平原型城市——吉林省白城市为例,采用2015—2022年多源数据(空气质量、气象、卫星、遥感),在系统分析重度污染事件形成成因的基础上,筛选了重度污染事件PM_(2.5)质量浓度预测的最佳机器学习算法。结果显示,在2017年以前,白城市重度污染发生频率较高,主要发生在秋末冬初和深冬;2017年以后重度污染天数显著减少。重度污染主要分为4种类型,即本地排放型、传输主导型、气象诱导型、复合污染型,其中复合污染型比例最高。本文构建了重度污染发生期间的空气质量、气象、遥感数据为基础的机器学习预测算法,对白城市PM_(2.5)质量浓度进行了预测,XGBoost算法表现最优,R^(2)为0.92,均方根误差(RMSE)为24.6μg/m^(3),显著优于随机森林(R^(2)=0.87)和支持向量机(R^(2)=0.67)等算法。本研究为东北地区平原型城市重度污染的预测提供了一种简洁、易掌握、精度高的流程和算法,有利于大气环境日常管理。
Severe air pollution disrupts traffic,poses significant health risks,and adversely affects economic development.Therefore,systematically analyzing the causes of severe pollution in typical urban areas and accurately predicting the occurrence of severe pollution events is of considerable scientific and practical importance.The northeastern region of China,a major hub for heavy industry and agriculture,is characterized by its northernmost latitude and longest heating period,making its emission sources and meteorological conditions highly representative.This study focuses on Baicheng City,a plain-type city in Jilin Province,and utilizes multisource data from 2015 to 2022,including air quality,meteorological,satellite,and remote sensing data.Through a systematic analysis of the underlying causes of severe pollution events,we identified the most effective machine learning algorithm for predicting PM_(2.5) concentrations during such events.The results indicate that,prior to 2017,Baicheng City experienced a high frequency of severe pollution events,primarily occurring in late autumn,early winter,and deep winter.However,after 2017,the number of severe pollution days significantly declined.Severe pollution events were classified into four primary types:local emission-driven,transmissiondominated,meteorologically-induced,and composite pollution,with composite pollution being the most prevalent.A machine learning-based prediction algorithm was developed using air quality,meteorological,and remote sensing data during severe pollution episodes to forecast PM_(2.5) concentrations in Baicheng.Among the tested models,the XGBoost algorithm demonstrated the best performance,with an R^(2) of 0.92 and a root mean square error(RMSE)of 24.6μg/m^(3),significantly outperforming other algorithms such as Random Forest(R^(2)=0.87)and Support Vector Machine(R^(2)=0.67).This study provides a straightforward,accessible,and highly accurate process and algorithm for predicting severe pollution events in plain-type cities of northeastern China,offering valuable insights for the effective management of atmospheric environmental conditions.
作者
秦杨
翟帅
石博文
张梅
陈卫卫
Qin Yang;Zhai Shuai;Shi Bowen;Zhang mei;Chen Weiwei(Jilin Province Ecological Environment Monitoring Center,Changchun 130032,Jilin,China;Key Laboratory of Wetland Ecology and Environment,Northeast Institute of Geography and Agroecology,Chinese Academy of Sciences,Changchun 130102,Jilin,China;College of Geography,Changchun Normal University,Changchun 130032,Jilin,China)
出处
《地理科学》
北大核心
2025年第8期1720-1732,共13页
Geographical Science
基金
吉林省生态环境厅环境保护科研项目(吉环科字第2024-05号)
国家重点研发计划项目(2023YFF0807202)资助。
