Satellite UV-Vis spectroscopy: implications for air quality trends and their driving forces in China during 2005–2017

Abundances of a range of air pollutants can be inferred from satellite UV-Vis spectroscopy measurements by using the unique absorption signatures of gas species. Here, we implemented several spectral fitting methods to retrieve tropospheric NO2, SO2, and HCHO from the ozone monitoring instrument (OMI), with radiative simulations providing necessary information on the interactions of scattered solar light within the atmosphere. We analyzed the spatial distribution and temporal trends of satellite-observed air pollutants over eastern China during 2005–2017, especially in heavily polluted regions. We found significant decreasing trends in NO2 and SO2 since 2011 over most regions, despite varying temporal features and turning points. In contrast, an overall increasing trend was identified for tropospheric HCHO over these regions in recent years. Furthermore, generalized additive models were implemented to understand the driving forces of air quality trends in China and assess the effectiveness of emission controls. Our results indicated that although meteorological parameters, such as wind, water vapor, solar radiation and temperature, mainly dominated the day-to-day and seasonal fluctuations in air pollutants, anthropogenic emissions played a unique role in the long-term variation in the ambient concentrations of NO2, SO2, and HCHO in the past 13 years. Generally, recent declines in NO2 and SO2 could be attributed to emission reductions due to effective air quality policies, and the opposite trends in HCHO may urge the need to control anthropogenic volatile organic compound (VOC) emissions.


Supplementary Texts On the OMI instrument
As seen from measured earthshine radiance as well as solar irradiance, the OMI instrument shows low optical degradation and high spectral stability over the mission time. OMI irradiances have degraded by 3-8% while radiances have changed by 1-2% 1 . The wavelength shift of the OMI instrument remains to be within 0.02 nm 1 . The Signal-to-noise ratio of OMI has been gradually decreasing over the years due to the expected CCD degradation 2 . The phenomenon called "row anomaly (RA)" occurred since 2007 and changed over time 1 , which affects the level 1b data at all wavelengths for particular viewing directions or cross-track positions. To overcome these addressed issues such as optical degradation and row anomaly, we have implemented specified methods during the retrieval of each trace gas. For example, a systematic cross-track biases were existed in the NO2 SCDs retrieval, possibly caused by the imperfect calibration in the solar irradiance. Therefore, a de-striping procedure was implemented to correct such stripe-like patterns 3 . And the optical degradation in solar irradiance and signal-to-noise ratio (SNR) significantly could affect the retrieval of weak absorbers such as HCHO and SO2. For HCHO SCD fittings, nadir reference spectrum from radiance data was used instead due to the large uncertainty in the solar irradiance. For SO2 retrieval, we applied "soft calibration" to OMI radiance, and thus the SO2 fitting residual was largely reduced 4 . Note that the retrieval uncertainties of these gases are reported to be increasing slightly over the mission time, as described in Zara et al., 2019 5 , due to the optical degradation. For the RA effect, we excluded all satellite RA-affected pixels during the regridding of level 2 data.

OMI NO2 retrieval
Finally, the tropospheric NO2 VCDs per orbit were re-gridded to the Level 3 product with a resolution of 0.1× 0.1° over eastern China by using a novel P-Spline method 11 . Note that satellite pixels were filtered out first if satisfying any of the following rejection criteria: cloud radiance fraction larger than 0.3; root mean square (RMS) of the spectral fitting larger than 0.002; pixels affected by the row anomaly; solar zenith angle larger than 70°; and other quality flags.

OMI HCHO retrieval
For HCHO SCDs retrieval, we generally followed the nonlinear least-square fitting methods as described in González Abad et al., 2016 12 , based on the so-called BOAS (basic optical absorption spectroscopy) approach. During the AMF calculations, monthly a priori profile of HCHO from the WRF-Chem model was used over China. See retrieval algorithm details in our previous study 13 . A constant value method 11 was also implemented during the HCHO re-gridding, with data screening criteria applied (similar to that of NO2).

OMI SO2 retrieval
A constant value method 11 was also implemented during the SO2 re-gridding, with data screening criteria applied (similar to that of NO2).
For SO2 retrieval, we implemented an OEM approach with full radiative simulations to directly retrieve SO2 VCDs simultaneously with ozone profile 4, 14,15 . Compared to the previous algorithm for the GOME-2 instrument 15 , the following updates are included for OMI SO2 retrieval: 1) fitting wavelength range is optimized at 312-326 nm; 2) monthly a priori atmospheric profile generated by the GEOS-Chem model was used 8 .
During NO2 SCDs retrieval, we generally follow the common DOAS configurations as suggested in QA4ECV NO2 project 5 . The NO2 spectral fit was selected in the wavelength range of 405-465nm and performed with the QDOAS software package 6 . For NO2 AMF simulations, stratospheric and tropospheric NO2 AMF were calculated pixel-by-pixel by the VLIDORT model at version 2.7 7 . During the RTM calculations, a priori NO2 profile was obtained from the monthly GEOS-Chem simulations 8 at a resolution of 2 × 2.5°; and additional information such as cloud fraction, cloud top pressure, and surface albedo, was taken from the operational OMI O2-O2 cloud dataset 9 . During stratosphere-troposphere separation on the total NO2 column, we estimated the stratospheric contribution from the total NO2 column based on a modified reference sector method, i.e., the STREAM algorithm 10 .

Ground-based validations
Our improved retrieval of trace gases including NO2, HCHO and SO2 have been widely used in ground-based MAX-DOAS measurements in eastern China, and shown overall better agreements compared to the operational OMI products [16][17][18][19] . Please refer to these literatures for detailed validation performances on the presented satellite data. Figure S1 showed the spatial distribution of linear regression slopes for annual NO2, SO2, and HCHO.