摘要
连续的SO_(2)污染物时空分布情况对于监测和综合治理大气SO_(2)污染、提高环境空气质量具有重要意义,使用Sentinel-5P TROPOMI提供的近实时SO_(2)垂直柱浓度数据、ERA5气象再分析资料和DEM数据,基于梯度提升机器学习模型(CatBoost)对山东省2019年SO_(2)污染物时空变化进行分析。结果表明:①CatBoost模型拟合的SO_(2)浓度的各项评价指标均为最佳,拟合精度指标MAE(平均绝对误差)、RMSE(均方根误差)和R 2(决定系数)分别为2.72μg/m^(3)、4.23μg/m^(3)和0.809,十折交叉验证精度指标MAE、RMSE和R 2分别为2.66μg/m^(3)、4.29μg/m^(3)和0.814,同时CatBoost模型在不同时间尺度和年份上也具有良好的稳定性。②2019年1−12月山东省SO_(2)浓度呈现“高、低、高”的“U”型分布特点,1月SO_(2)浓度(25.5μg/m^(3))最高,8月(8.4μg/m^(3))最低。各季节SO_(2)浓度存在明显差异,表现为夏季(8.74μg/m^(3))<秋季(12.37μg/m^(3))<春季(14.12μg/m^(3))<冬季(20.18μg/m^(3)),呈冬春高、夏秋低的特征,且SO_(2)浓度年均值为13.36μg/m^(3)。③2019年山东省年均SO_(2)浓度总体呈现中部地区高、东部沿海城市低的空间分布特征,高值区多位于工企业集中地区,低值区主要分布在工业数量少、地势平缓的沿海一带。研究显示,CatBoost模型具备良好的稳定性和泛化能力,能够揭示山东省2019年SO_(2)浓度时空间分布特征与污染演化趋势之间的关系,可为山东省SO_(2)污染治理提供有效的方法和数据。
High spatiotemporal resolution sulfur dioxide(SO_(2))concentration data are of great significance for monitoring and comprehensive control of atmospheric SO_(2) pollution to improve air quality.This study estimated ground-level SO_(2) concentrations in Shandong Province during 2019 using the CatBoost gradient boosting machine learning model,integrating near-real-time SO_(2) column concentration data from Sentinel-5P TROPOMI,ERA5 meteorological reanalysis data,and DEM data.The results demonstrate the following:(1)The CatBoost model achieved the best effect in SO_(2) concentration estimation,with model fitting accuracy metrics showing a mean absolute error(MAE)of 2.72µg/m^(3),a root mean square error(RMSE)of 4.23µg/m^(3),and a coefficient of determination(R 2)of 0.809.The results of 10-fold cross-validation yielded MAE=2.66µg/m^(3),RMSE=4.29µg/m^(3),and R 2=0.814,demonstrating favorable stability across different temporal scales.(2)In 2019,the monthly SO_(2) concentrations in Shandong Province showed a U-shaped distribution,with the highest concentration in January(25.5µg/m^(3))and reaching the lowest concentration in August(8.4µg/m^(3)).The seasonal variation characteristics were obvious:summer(8.74μg/m^(3))<autumn(12.37μg/m^(3))<spring(14.12μg/m^(3))<winter(20.18μg/m^(3)),and the annual average was 13.36µg/m^(3).There was a significant seasonal variation,showing a distribution pattern of higher in winter and spring,and lower in summer and autumn.(3)From the perspective of spatial distribution,the annual average concentration of SO_(2) across Shandong Province in 2019 exhibited a clear geographical gradient.High concentrations were concentrated in the central regions,which was dominated by industrial areas,while lower concentrations were observed in the eastern coastal areas characterized by limited industrial activity and gentle topography.The results demonstrate that the CatBoost model exhibits robust stability and generalization capability,effectively elucidating the relationship between the spatiotemporal distribution characteristics of SO_(2) concentration estimation and its evolutionary trends in Shandong Province during 2019.This model serves as a scientifically validated tool to support targeted SO_(2) pollution control strategies in the region.
作者
李隆
杜宁
邓小东
张洪飞
龚德才
LI Long;DU Ning;DENG Xiaodong;ZHANG Hongfei;GONG Decai(Mining College of Guizhou University,Guiyang 550025,China)
出处
《环境科学研究》
北大核心
2025年第6期1241-1251,共11页
Research of Environmental Sciences
基金
贵州省科技计划项目(No.黔科合基础-ZK[2024]一般093)。
作者简介
李隆(1998-),男,贵州黔西人,1870466078@qq.com;杜宁(1969-),男,责任作者,贵州贵阳人,教授,硕士,主要从事大地测量与遥感研究,ndu1@gzu.edu.cn。
