摘要
提出了一种基于梯度提升树优化的近地面O_(3)浓度时空分布估算模型.基于O_(3)地面观测站点数据、高分辨率大气成分卫星数据(TROPOMI、AIRS)、ERA5气象再分析资料、以及地表覆盖和地形数据,本文研究了京津冀地区近地面O_(3)与对流层O_(3)及其前体物、气象因素、以及下垫面资料之间的相关关系,对比分析了不同梯度提升树算法模型(GBDT、XGBoost、LightGBM)的估算精度.结果表明,3种模型整体上均可对近地面O_(3)进行精确估算,GBDT、XGBoost、LightGBM的决定系数R^(2)分别为0.9489、0.9547、0.9495,均方根误差RMSE分别为13.85,13.26,13.76μg/m^(3),XGBoost模型的精度相对最高;通过采用过滤法、相关性分析法以及递归特征消除法筛选特征,对XGBoost估算模型进行了优化,在保证模型精度前提下,降低了特征复杂度,优化后模型估算精度可达到R^(2)=0.9549,估算速率提升了约17%,为区域尺度O_(3)浓度时空分布建模与估算提供了一个精细而高效的方法模型.
This paper proposes an improved model for estimating the temporal and spatial distribution of near-surface ozone concentrations based on the gradient boosting tree optimization algorithm.First,in-situ near-surface O_(3) concentrations data,high-resolution satellite observations of atmospheric compositions(TROPOMI,AIRS),ERA5meteorological reanalysis data,and land cover and terrain data were used to investigate thecorrelations among O_(3),tropospheric O_(3),O_(3) precursors,meteorological factors and underlying surfaces in the Beijing-Tianjin-Hebei region.Then,estimation models for near-surface O_(3) concentrations based on three different gradient boosting tree algorithm models(GBDT,XGBoost,LightGBM)were developed,and the estimation accuracies of different models were compared and analyzed.Results showed that all three models could accurately estimate near-surface ozone concentrations,with the XGBoost model demonstrating the highest accuracy.The coefficients of determination R^(2) of GBDT,XGBoost,and LightGBM were 0.9489,0.9547,and 0.9495 respectively,and the root mean square errors RMSE were 13.85μg/m3,13.26μg/m^(3) and 13.76μg/m^(3) respectively.Using filtering methods,correlation analysis,and recursive feature elimination,the XGBoost estimation model was optimized,feature complexity was reduced while maintaining model accuracy.After optimization,the estimation accuracy of the optimized model reached R^(2)=0.9549,increasing the estimation efficiency by about 17%.This paper provided a sophisticated and efficient method for modeling and estimating the temporal and spatial distribution of ozone concentration on a regional scale.
作者
梁晓霞
谢东海
韩宗甫
宋世鹏
张欣欣
顾坚斌
余超
LIANG Xiao-xia;XIE Dong-hai;HAN Zong-fu;SONG Shi-peng;ZHANG Xin-xin;GU Jian-bin;YU Chao(College of Resource Environment and Tourism,Capital Normal University,Beijing 100089,China;State Key Laboratory of Remote Sensing Science,Aerospace Information Research Institute,Chinese Academy of Sciences,Beijing 100101,China;School of Computing,Beijing University of Posts and Telecommunications,Beijing 100876,China;School of Electrical Engineering,Nantong University,Nantong 226019,China)
出处
《中国环境科学》
EI
CAS
CSCD
北大核心
2023年第8期3886-3899,共14页
China Environmental Science
基金
国家重点研发计划重点专项课题(2022YFC3700102)
国家自然科学基金面上项目(42171393)。
关键词
近地面臭氧
对流层臭氧
梯度提升回归模型
卫星遥感
特征选择
时空分布
near-surface ozone
tropospheric ozone
gradient boosting regression model
satellite remote sensing
feature selection
spatial and temporal distribution