| 引用本文: | DENG Shudan,HU Litiao,LIU Riyang,GAO Ming,SHAO Yanchuan,MA Zongwei.Revealing Drivers of PM2.5 Pollution Events by Explainable Machine Learning[J].Environmental Monitoring and Forewarning,2025,17(1):35~42 |
| |
|
| 摘要: |
| 细颗粒物（PM2.5）污染对人体健康和社会经济均有负面影响，为了解PM2.5 重污染事件形成的关键驱动因素，利用气象参数和大气污染物排放清单数据构建随机森林模型，模拟长三角地区2017年冬季4次重污染事件中的PM2.5质量浓度，并借助沙普利加和解释（SHAP）机器学习方法识别重污染事件的驱动因素。研究结果表明，气象要素对重污染事件中ρ（PM2.5）有着复杂的影响，其中降水量、地表净太阳辐射和露点温度都是影响4次重污染事件中ρ（PM2.5）的重要气象驱动因素；一次排放污染物中，交通源排放的二氧化硫（SO2）、农业源排放的氨气（NH3）和溶剂使用排放的挥发性有机物（VOCs）对ρ（PM2.5）也有较为重要的影响。 |
| 关键词: 细颗粒物污染 驱动因素 可解释性 机器学习 沙普利加和解释 |
| DOI：DOI：10.3969/j.issn.1674-6732.2025.01.006 |
| 分类号:X823 |
| 基金项目:国家自然科学基金项目（72234003，71921003） |
|
| Revealing Drivers of PM2.5 Pollution Events by Explainable Machine Learning |
|
DENG Shudan1， HU Litiao1， LIU Riyang1， GAO Ming2， SHAO Yanchuan1， MA Zongwei1*
|
|
1. State Key Lab of Pollution Control and Resources Reuse， School of the Environment， Nanjing University， Nanjing, Jiangsu 210023， China； 2. Jiangsu Academy of Environmental Industry and Technology Corp.， Nanjing, Jiangsu 210036， China
|
| Abstract: |
| PM2.5 pollution has serious negative impacts on human health， society， and economy. In order to develop more effective control plans for PM2.5 pollution， ensure the health of the people， and promote sustainable development of the environment， it is crucial to understand the key driving factors behind the formation of PM2.5 pollution events. In this study， we constructed a random forest model using meteorological parameters and atmospheric pollutant emission inventory data to simulate the PM2.5 concentration during the 2017 winter pollution events in the Yangtze River Delta region， and identified the driving factors of heavy pollution events using SHAP（SHapley Additive exPlanations） explainable machine learning. The research results indicate that meteorological factors have a complex impact on PM2.5 concentration in heavy pollution events， among which precipitation， surface net solar radiation， and dew point temperature are important meteorological driving factors； SO2 emissions from transportation sources， NH3 emissions from agricultural sources， and VOCs emissions from solvent use also have a significant impact on PM2.5 concentration. |
| Key words: PM2.5 pollution Driving factors Explainability Machine learning SHAP |
