Skip to Content

Deep learning for Chinese NOx emission inversion and the integration of in situ observations: a case study on the COVID-19 pandemic

NO2 column density measurements from satellites have been widely used in constraining emissions of NOx. However, satellite constraints are affected by the cloud coverage and the low sensitivity to the surface. As a consequence, changes in surface NOx emissions might not be well captured by satellite NO2 column measurements. Ground-based stations provide in situ measurements on surface NO2 concentrations, which is more directly representative of NOx emissions. The in situ observations could be therefore used as the complementary information to further constrain satellite-derived emissions. We build a deep learning (DL) model to estimate Chinese NOx emissions using surface NO2 concentrations. A multi-stage training strategy is applied to integrate the Ministry of Ecology and Environment of China (MEE) observation network into the satellite-derived Tropospheric Chemistry Reanalysis (TCR-2). We train the DL model from 2005 to 2018, and evaluate the DL model performance for 2019 and 2020 using both dependent and independent data sets. The evaluation of the DL-based system shows a high correlation coefficient (R=0.96) after the first training stage. We demonstrate that the DL-based analysis shows an improved variability in NOx emissions during the Chinese New Year (CNY) holidays, which is highly consistent with the Baidu "Qianxi" mobile data. The DL-based analysis shows that an 18\% anomalous drop in the Chinese NOx emissions is caused by the COVID-19 pandemic 20--30 days after the 2020 CNY, relative to the same period in 2019. This finding is consistent with both OMI-derived and TROPOMI-derived TCR-2 data products.

Event series  Brewer-Wilson Seminar Series