基于TROPOMI NO_(2)、CO及HCHO重构数据的近地面O_(3)浓度估算研究

Estimation of near-surface O_(3) concentration based on TROPOMI NO_(2),CO and HCHO reconstruction data

以TROPOMI大气成分产品为代表的卫星遥感数据在近地面臭氧(O_(3))浓度估算中表现出很好的潜力。由于云及反演算法局限性等原因,TROPOMI的大气成分产品存在一定的数据缺失,使得基于此的近地面O_(3)浓度估算数据覆盖度较低。为此,本文采用经验正交函数分解插值法(DINEOF)对TROPOMI NO_(2)、CO、HCHO原始数据产品进行缺失像元重构,并基于重构数据产品采用极限梯度提升算法(XGBoost)估算了中国大陆地区2019年—2021年高覆盖度的日最大8 h平均O_(3)浓度(MDA8 O_(3))。对比研究了DINEOF方法对TROPOMI NO_(2)、CO、HCHO原始数据产品的缺失像元进行重构后再建模估算O_(3)的方法(方案1),使用TROPOMI NO_(2)、CO、HCHO原始数据产品并对其缺失像元赋空值并输入其他特征变量来建模估算O_(3)的方法(方案2),有TROPOMI NO_(2)、CO、HCHO原始数据产品的建模结果和无TROPOMI NO_(2)、CO、HCHO原始数据产品,只有其他特征变量的建模结果相结合的方法(方案3),均可提升O_(3)模型估算结果的覆盖度。实验表明:方案1的结果最好,其十折交叉验证R^(2)=0.86,RMSE=15.86μg/m^(3),模型精度和方案2基本一致且高于方案3,在重构区域的模型精度最高(训练集R^(2)=0.82,RMSE=15.57μg/m^(3)),且当重构区域出现O_(3)重污染时(浓度大于160μg/m^(3)),能明显改善模型高值低估现象,结果的空间分布更合理。方案1估算的2019年—2021年近地面MDA8 O_(3)的平均覆盖度从33.6%提升到97.2%,使用TROPOMI NO_(2)、CO、HCHO重构数据产品建模估算O_(3)可提升模型性能和模型结果的覆盖度。

Satellite remote sensing data represented by TROPOMI atmospheric composition products show good potential in the estimation of near-surface O_(3)concentrations.Given the limitations of cloud and inversion algorithms,data for TROPOMI atmospheric composition products are lacking,resulting in low coverage of estimation results.Therefore,the DINEOF method was used to reconstruct the missing cells of TROPOMI NO_(2),CO,and HCHO original data products and estimate the maximum daily 8 h average O_(3)concentration(MDA8 O_(3))of Chinese mainland high coverage from 2019 to 2021 based on XGBoost.In this study,three schemes to improve the coverage of O_(3)model estimation results are compared.Scheme 1 reconstructs the missing cells of TROPOMI NO_(2),CO,and HCHO original data products based on the DINEOF method and uses the reconstructed data to model and estimate O_(3).Scheme 2 is based on TROPOMI NO_(2),CO,and HCHO original data products,null values are assigned to their missing cells,and only other characteristic variables are entered to model and estimate O_(3).Scheme 3 uses a combination of modeling results containing TROPOMI NO_(2),CO,and HCHO original data products and modeling results that do not contain TROPOMI NO_(2),CO,and HCHO original data products but with other characteristic variables.Experiments show that the results of scheme 1 are the best;its 10-fold cross-validation results are R^(2)=0.86 and RMSE=15.86μg/m^(3).The model accuracy is basically the same as scheme 2 and higher than that of scheme 3,and the model accuracy in the reconstruction region is the highest(training set R^(2)=0.82,RMSE=15.57μg/m~3).When O_(3)heavy pollution occurs in the reconstruction region(concentration greater than 160μg/m^(3),the underestimation of the high value of the model can be remarkably improved,and the spatial distribution of the results is more reasonable.The average coverage of the near-surface MDA8 O_(3)estimated in scheme 1 increased from 33.6%to 97.2%from 2019 to 2021.Using TROPOMI NO_(2),CO,and HCHO refactor data products to model and estimate O_(3) can improve model performance and coverage of model results.