研究报告

  • 赵楠,卢毅敏.基于XGBoost算法的近地面臭氧浓度遥感估算[J].环境科学学报,2022,42(5):95-108

  • 基于XGBoost算法的近地面臭氧浓度遥感估算
  • Remote-sensing estimation of near-surface ozone concentration based on XGBoost
  • 基金项目:国家重点研发科技专项 (No.2017YFB0503500);福建省科技计划项目(No.2020L3005)
  • 作者
  • 单位
  • 赵楠
  • 福州大学, 空间数据挖掘与信息共享教育部重点实验室, 福州 350108;福州大学, 福建省数字区域工程技术研究中心,福州 350108;数字中国研究院(福建),福州 350003
  • 卢毅敏
  • 福州大学, 空间数据挖掘与信息共享教育部重点实验室, 福州 350108;福州大学, 福建省数字区域工程技术研究中心,福州 350108;数字中国研究院(福建),福州 350003
  • 摘要:本文采用XGBoost机器学习算法,融合臭氧浓度地面监测数据、欧洲中期天气预报中心的ERA5数据集、中国多尺度排放清单模型构建的排放清单数据集、高分辨率遥感影像(TROPOMI_NO2、OMI_NO2)以及人口数据和DEM数据,构建训练估算数据集,开展近地面臭氧浓度估算研究.模型构建采用递归式特征消除法进行特征变量的选择,并对其进行十折交叉和自建模验证,R2分别为0.871和0.955,RMSE分别为12.8 μg·m-3和7.514 μg·m-3.同时进行了高分辨率遥感影像对估算结果的贡献分析,结果表明引入TROPOMI_NO2因子参与建模可校正近地面臭氧浓度普遍被低估现象.模型模拟结果显示臭氧浓度回归估算结果层次更加分明、条带现象消失、连续性和平滑性明显改善.
  • Abstract:In this paper, the Extreme Gradient Boosting(XGBoost) machine learning algorithm was used to estimate near-surface ozone concentration. By integrating the ground monitoring data of ozone concentration, ERA5 dataset of European Centre for Medium-Range Weather Forecasts (ECMWF), emission inventory dataset of Multi-resolution Emission Inventory for China(MEIC), high-resolution remote sensing images (TROPOMI_NO2, OMI_NO2), population data and DEM data, a training dataset was built to construct a remote sensing estimation model. A recursive feature elimination method was used to select features of the model, and the 10-fold cross-validation and self-modeling verification were performed. The R2 were respectively 0.871 and 0.955; the RMSE were 12.8 μg·m-3 and 7.514 μg·m-3 respectively. At the same time, the contribution of high-resolution remote sensing images to the estimation was analyzed. The results showed that the introduction of TROPOMI_NO2 factor to the modeling can improve the general underestimation of near-surface ozone concentration. The simulation results show that the ozone concentration regression estimation results are more distinct, the banding phenomenon disappears, and the continuity and smoothness are significantly improved.

  • 摘要点击次数: 210 全文下载次数: 408