干旱气象 ›› 2025, Vol. 43 ›› Issue (4): 563-575.DOI: 10.11755/j.issn.1006-7639-2025-04-0563

• 论文 • 上一篇    下一篇

基于机器学习的环流分型与甘肃大气污染机制研究

刘宗瑞1(), 万紫悦1, 赵宇瀚1, 刘卫平2, 王若安3, 马玉霞1()   

  1. 1.兰州大学大气科学学院,甘肃 兰州 730000
    2.兰州区域气候中心,甘肃 兰州 730020
    3.甘肃省平凉市气象局,甘肃 平凉 744000
  • 收稿日期:2025-03-20 修回日期:2025-05-08 出版日期:2025-08-31 发布日期:2025-09-08
  • 通讯作者: 马玉霞(1974—),女,教授,研究方向为极端天气和气候及其健康影响。E-mail: mayuxia07@lzu.edu.cn
  • 作者简介:刘宗瑞(2000—),男,硕士研究生,研究方向为极端天气及其影响。E-mail: liuzr2023@lzu.edu.cn
  • 基金资助:
    国家自然科学基金面上项目(4235177);甘肃省自然科学基金项目(23JRRA1079)

Research on circulation classification and atmospheric pollution mechanisms in Gansu based on machine learning

LIU Zongrui1(), WAN Ziyue1, ZHAO Yuhan1, LIU Weiping2, WANG Ruoan3, MA Yuxia1()   

  1. 1. College of Atmospheric Sciences, Lanzhou University, Lanzhou 730000, China
    2. Lanzhou Regional Climate Center, Lanzhou 730020, China
    3. Pingliang Meteorological Bureau of Gansu Province, Pingliang 744000, Gansu, China
  • Received:2025-03-20 Revised:2025-05-08 Online:2025-08-31 Published:2025-09-08

摘要:

为揭示干旱半干旱地区天气形势与气象要素对大气污染的影响机制,基于2016—2022年欧洲中期天气预报中心(European Centre for Medium-Range Weather Forecasts,ECMWF)提供的第五代大气再分析数据集(ERA5),采用自组织映射神经网络(Self-Organizing Map,SOM)对700 hPa位势高度场和风场进行天气分型,并结合二次曲线拟合解析甘肃不同气候区典型城市气象要素与污染物的非线性关系。 结果表明:1)PM10和PM2.5质量浓度与气温整体呈负相关,而O3质量浓度随气温升高非线性增加;低风速(<1 m·s-1)和高风速(>4 m·s-1)下,PM10和PM2.5质量浓度较高,在静风及强风时分别因局地累积和沙尘输送导致颗粒物质量浓度升高,而1~4 m·s-1的风速有利于前体物积累,导致O3质量浓度升高;25%~75%的相对湿度条件下污染物质量浓度较高,但其影响存在区域差异,如干旱区的酒泉,在湿度小于25%条件下由于易发生沙尘天气PM10质量浓度较高,PM2.5由于吸湿增长作用,质量浓度随相对湿度升高而增加,O3在低湿条件下的消耗降低,其质量浓度随相对湿度升高递减。2)冬春季,以西南高压型和东部低槽型为主导,西南高压型下西部强西北风形成污染物输送通道,东部低压槽型下甘肃地区扩散条件差,其大气形势较为稳定,污染物易积累,导致PM10和PM2.5质量浓度显著升高。3)夏秋季,以高压型为主导,充沛的太阳辐射与高温条件促使边界层高度抬升,配合暖湿气流输送,为光化学反应提供有利环境,导致O3质量浓度较高。

关键词: 大气污染, 天气分型, 气象要素, 机器学习

Abstract:

To reveal the influence mechanisms of weather patterns and meteorological factors on air pollution in arid and semi-arid regions, this study utilized the fifth-generation atmospheric reanalysis dataset (ERA5) provided by the European Centre for Medium-Range Weather Forecasts (ECMWF) from 2016 to 2022. The Self-Organizing Map (SOM) neural network was employed to classify weather patterns based on the 700 hPa geopotential height field and wind field, combined with quadratic curve fitting to analyze the nonlinear relationships between meteorological factors and pollutants in typical cities across different climatic zones of Gansu Province. The results indicate that: 1) PM10 and PM2.5 mass concentrations generally exhibit a negative correlation with temperature, while for O3 its mass concentration increases nonlinearly with rising temperature. Under low wind speeds (<1 m·s-1) and high wind speeds (>4 m·s-1), PM10 and PM2.5 mass concentrations are elevated, which attributed to local accumulation under calm conditions and dust transport under strong winds, respectively. In contrast, wind speeds between 1 and 4 m·s-1 favor precursor accumulation, leading to higher mass concentration of O3. Pollutant concentrations are generally higher under relative humidity between 25% and 75%, though regional differences exist. For instance, in Jiuquan, the PM10 concentration is higher at humidity below 25% due to frequent dust events, the PM2.5 concentration increases with the rise of relative humidity due to the hygroscopic growth effect, while O3 concentration decreases with declining humidity owing to reduced consumption under dry conditions. 2) In winter and spring, the dominant weather patterns are the southwestern high-pressure type and the eastern trough type. Under the southwestern high-pressure pattern, strong northwesterly winds in western regions create pollutant transport pathways, whereas the eastern trough pattern results in poor diffusion conditions and stable atmospheric conditions in Gansu, facilitating pollutant accumulation and significantly increasing of PM10 and PM2.5 concentrations. 3) In summer and autumn, high-pressure systems dominate, with abundant solar radiation and high temperatures elevating the boundary layer height. Coupled with warm, moist air transport, these conditions provide a favorable environment for photochemical reactions, leading to higher O3 concentrations.

Key words: air pollution, synoptic weather patterns, meteorological elements, machine learning

中图分类号: