Estimating Ground-Level NO₂ Concentrations Based On Multi-Source Satellite Data Products

Air pollution monitoring has great significance owing to atmospheric environmental pollution has a significant impact on human survival and development.With the continuous development of China's economy,people are suffering from serious nitrogen oxide pollution,especially in the eastern part of China.The trace gas nitrogen dioxide?NO₂?in the atmosphere plays an extremely important role.Therefore,this paper uses satellite remote sensing data and other related variables to estimate the ground-level NO₂ concentrations.The estimation of ground-level NO₂ concentrations are generally limited by the satellite bypass time.Therefore,this paper combines two satellite data to estimate the ground-level daily NO₂ concentrations.In addition,this paper also compares the spatial distribution difference of ground-level NO₂concentrations obtained based on two different satellite product data.Specifically,in this paper,we use the OMI-retrieved vertical column density?VCD?of NO₂ with a local bypass time at about 13:45 and GOME-2-retrieved VCD of NO₂ with a local bypass time at about 10:00 coordinating the meteorological data,emission data and other ancillary data to estimate ground-level NO₂ concentrations in2016,and compared the predictive performance of two machine learning models.The ten-fold cross-validation results of extra-trees model are R²=0.79 and RMSE=9.04?g/m³ and random forest model are R²=0.76 and RMSE=9.60?g/m³.Therefore,extra-trees model is used to estimate ground-level NO₂ concentrations,and the spatial distribution results of annual,seasonal and monthly NO₂ concentrations are analyzed,respectively.The results show that?1?North China Plain,Beijing-Tianjin-Hebei,Yangtze River Delta,Pearl River Delta Economic Region and some urban areas in the eastern part of China show a high average annual NO₂ concentrations.?2?The NO₂concentrations showed obvious seasonal characteristics.In summer,it was generally low-value distribution due to meteorological conditions,and the winter showed a general high-value distribution.?3?The NO₂ concentrations in different months also has a large difference.The average value of NO₂ concentrations in December is the largest,and the average value of NO₂ concentrations in August is the smallest.In addition,we estimated one year?2016.12-2017.11?of ground-level NO₂concentrations based on OMNO₂d data and POMINO data,respectively.Furthermore,we used extra-trees model to convey the non-linear relationship between NO₂concentrations and predictor variables,and compare the prediction accuracy with that of random forests model.The extra-trees model showed better prediction performance,with cross-validation R²=0.72?RMSE=9.20?g/m³?based on POMINO data and R²=0.70?RMSE=9.42?g/m³?based on OMNO₂d data,than the random forests model,with cross-validation R²=0.70?RMSE=9.45?g/m³?based on POMINO data and R²=0.69?RMSE=9.47?g/m³?based on OMNO₂d data.These demonstrate the ground-level NO₂ concentrations estimates from the extra-trees model are more reliable and robust,in addition,the models performed better when POMINO data as a variable input to the models.This study demonstrates that POMINO data could be a better source for mapping ground-level NO₂ concentrations across eastern China.