First observation of tropospheric nitrogen dioxide from the Environmental Trace Gases Monitoring Instrument onboard the GaoFen-5 satellite

The Environmental Trace Gases Monitoring Instrument (EMI) is the first Chinese satellite-borne UV–Vis spectrometer aiming to measure the distribution of atmospheric trace gases on a global scale. The EMI instrument onboard the GaoFen-5 satellite was launched on 9 May 2018. In this paper, we present the tropospheric nitrogen dioxide (NO2) vertical column density (VCD) retrieval algorithm dedicated to EMI measurement. We report the first successful retrieval of tropospheric NO2 VCD from the EMI instrument. Our retrieval improved the original EMI NO2 prototype algorithm by modifying the settings of the spectral fit and air mass factor calculations to account for the on-orbit instrumental performance changes. The retrieved EMI NO2 VCDs generally show good spatiotemporal agreement with the satellite-borne Ozone Monitoring Instrument and TROPOspheric Monitoring Instrument (correlation coefficient R of ~0.9, bias < 50%). A comparison with ground-based MAX-DOAS (Multi-Axis Differential Optical Absorption Spectroscopy) observations also shows good correlation with an R of 0.82. The results indicate that the EMI NO2 retrieval algorithm derives reliable and precise results, and this algorithm can feasibly produce stable operational products that can contribute to global air pollution monitoring.


Introduction
The Environmental Trace Gases Monitoring Instrument (EMI) 1 is the first Chinese satellite-borne spectrometer with the aim to measure atmospheric pollutants from space. The EMI payload onboard the GaoFen-5 satellite was successfully launched on 9 May 2018. The GaoFen-5 satellite has a polar orbit at an altitude of 706 km. The Chinese EMI instrument is expected to contribute to the understanding of global air quality and atmospheric chemistry, similar to predecessor European and American satellite missions, e.g., the Ozone Monitoring Instrument (OMI) 2 and TROPOspheric Monitoring Instrument (TROPOMI) 3 . EMI has instrumental characteristics that are similar to OMI and TROPOMI, e.g., the local overpass time at~13:30, spectral coverage, push-broom imaging technique, and daily global coverage. Both EMI and TROPOMI (launched in 2017) are new-generation satellite-borne air pollutant sensors compared to the OMI that was launched in 2004. TROPOMI follows the heritage of OMI in both instrument design and trace gas retrievals, but with higher spatial resolution and signal-to-noise ratio. A prototype EMI nitrogen dioxide (NO 2 ) retrieval algorithm was developed before launch based on the OMI NO 2 retrieval. However, optimization of the NO 2 retrieval algorithm was necessary to adapt the unexpected issues of EMI after launch, especially spectral calibration.
Nitrogen oxides (NO x ), defined as the sum of nitrogen oxide and NO 2 , are the major pollutants contributing to ozone and secondary aerosol formation in the troposphere through photochemical reactions 4 . Sources of NO x include fossil fuel combustion, vehicle emissions, biomass burning, and lightning 5 . Due to rapid industrialization and urbanization in the past few decades, China has become one of the largest NO x emitters in the world 6 . As a result, China is experiencing a series of severe air pollution problems 7,8 . In addition to measuring NO 2 distribution directly from space, applications of satellite remote sensing may include estimations of pollutant emissions 9 , air quality trend detection 10 , model validation, and assimilation of satellite data 11 . Figure 1a illustrates the optical design of the EMI satellite instrument. The EMI instrument covers the ultraviolet (UV) and visible (Vis) spectral ranges from 240 to 710 nm with a spectral resolution of 0.3-0.5 nm. Light received by the telescope is depolarized by a scrambler and subsequently split into four spectral channels, the UV1 (240-315 nm), UV2 (311-403 nm), VIS1 (401-550 nm), and VIS2 (545-710 nm) channels. Each spectrometer is equipped with a two-dimensional chargecoupled device (CCD) detector, with one dimension used for spectral coverage and the other dimension used for spatial coverage. The EMI instrument scans in the nadir direction toward the earth's surface with an opening angle of 114°corresponding to a swath width of 2600 km, enabling daily global coverage with a nadir resolution of 12 × 13 km 2 and a local overpass time of 13:30 (Fig. 1b). The direct sun solar irradiance spectrum, typically used as a reference spectrum in the spectral analysis of the nadir radiance measurement, is introduced to the EMI telescope once a day using the quartz volume diffuser 12 . By using the unique absorption features of different trace gases in the UV-Vis range, the abundances of atmospheric pollutants can be retrieved from the difference between atmospheric and solar spectra.
In this paper, we present a new tropospheric NO 2 vertical column density (VCD, i.e., the vertical integral of NO 2 concentration from the earth's surface to the top of the atmosphere) retrieval algorithm dedicated to the EMI instrument. Details of the spectral retrieval, stratospheric-tropospheric separation of the NO 2 column and slant to vertical column conversion are presented. The first EMI retrieval of tropospheric NO 2 columns is compared to datasets from modern state-of-the-art European and American satellite sensors.

NO 2 retrieval overview
The retrieval of tropospheric NO 2 VCDs from satellite UV-Vis observations typically follows a state-of-the-art three-step approach. First, the total NO 2 slant column density (SCD) are retrieved from nadir radiance spectra normalized by the solar irradiance, using the differential optical absorption spectroscopy (DOAS) technique 13 . Subsequently, the stratospheric NO 2 columns are separated from the total NO 2 SCDs by assuming longitudinal homogeneity of stratospheric NO 2 , while neglecting the minor contribution of tropospheric NO 2 (usually on the order of 10 14 molecules cm −2 ) over remote clean regions 14,15 . Last, tropospheric NO 2 SCDs are converted to VCDs using air mass factors (AMFs) 16 . The AMF is defined as the ratio between SCD and VCD. It is a measure of the effective optical path length from the top of the atmosphere to the earth's surface and reflected to the satellite through the atmosphere: where M is the AMF, S denotes the SCD, and V represents the VCD. The AMF can be calculated with a radiative where V tropo and V strat denote tropospheric and stratospheric V, respectively. M tropo and M strat represent tropospheric and stratospheric M, respectively. Details of the stratospheric estimation and AMF calculation are provided in the "Materials and methods" section. Figure 2 shows an example of EMI NO 2 retrieval of S, V strat , and V tropo on 1 January 2019. Enhanced NO 2 levels are observed in Eastern China, India, and the Middle East.

Algorithm improvements
A prototype EMI NO 2 retrieval is developed before launch. The prototype algorithm is very similar to the ig. 2 An example of EMI NO 2 retrieval on 1 January 2019. The total SCDs (S), stratospheric VCDs (V strat ), and tropospheric VCDs (V tropo ) retrieval of NO 2 are shown in a, b, and c, respectively. Note that satellite ground pixels affected by clouds are indicated in white operational OMI NO 2 retrieval 17 . However, due to unexpected issues, i.e., low signal-to-noise ratio at the edges of the spectral channels, bad irradiance measurement due to a diffuser calibration issue, and spectral saturation issue, the NO 2 retrieval setting must be further optimized to address these issues. A series of sensitivity tests, including cloud correction, fitting wavelength range, reference selection, and spectral precalibration, have been performed to optimize the settings for tropospheric NO 2 VCD retrieval. Table 1 lists the updated retrieval settings of the EMI NO 2 retrieval. Parameters used in the OMI QA4ECV NO 2 retrieval 17 are also listed for reference.
The EMI NO 2 fitting range is shifted slightly from 405-465 nm (OMI operational NO 2 setting 17 ) to 420-470 nm to avoid the lower signal-to-noise ratio region at the edges of the VIS1 channel 12 . Figure 3 illustrates an example of the retrieval of NO 2 SCD, i.e., the NO 2 amount integrated along the optical path in the atmosphere, by applying the DOAS fit to the EMI-measured spectrum.
The spectral saturation issue (i.e., the analogue photon signal reaches the maximum digital value of the CCD detector) is critical for EMI observations over bright clouds due to its high surface reflectance. Supplementary Figure 1 shows the global spatial pattern of the root mean square (RMS) of the spectral fitting residual, cloud radiance fraction from TROPOMI observations, and the true color image from the MODIS-Aqua instrument on 1 January 2019. The spatial pattern of the fitting residual RMS is correlated to the cloud pattern. Therefore, we filtered pixels with relatively large spectral fitting residuals, i.e., the RMS values >0.004.
The key calibration data measured during the on-ground calibration 12 seem unsuitable for EMI on-orbit measurements due to the degradation and stability of the instrument in the complex space environment (e.g., cosmic radiation exposure 18 and possible instrument changes since launch 19 ). Therefore, we recalibrated the EMI earth radiance measurements by comparing the EMI radiance to TROPOMI measurements and RTM simulations. An advantage of DOAS is that it does not rely on precisely calibrated radiance and is less sensitive to the variability in radiometric calibration than other methods based on discrete radiance (e.g., SBUV and TOMS ozone retrieval algorithms 20 ). Figure 4 shows the comparisons of NO 2 SCDs for one orbit on 4 January 2019 retrieved using these spectral fitting scenarios: (a) current settings of the EMI NO 2 retrieval listed in Table 1; (b) using the measured irradiance spectrum as a reference; and (c) same as in (a) but without spectral precalibration. Irradiance spectra measured by EMI are currently accounting for some calibration issues, and these issues are probably related to the interference of the space environment on the hemispheric reflectance of solar diffusers 18 . NO 2 SCDs retrieved with irradiance as a reference show large biases and errors, particularly the central part of the measurement swath (Fig. 4b). Therefore, it is not optimal to use the direct sun irradiance spectra as a reference. To avoid the influence of abnormal irradiance spectra, we use cloud-free earth radiance measurements over the Pacific Ocean as a reference 21 . Compared to using solar irradiance as a reference, using earth radiance as a reference greatly reduced the spectral noise in the fit residual (Fig. 4a, b), which is likely related to the differences between spectra measured with the solar and earth-viewing modes 21 . The mean RMS of fitting residual by using earth radiance as a reference over cloud-free regions is 30% smaller than that with irradiance as a reference, as shown in Fig. 4a, b. Although using radiance as a reference improved the spectral retrieval, the radiance reference also contains a NO 2 absorption signal. Therefore, we must calculate the SCD offset to compensate for the residual NO 2 signal in the reference spectrum. The SCD offset is calculated using the NO 2 AMFs multiplied by the a priori NO 2 profile taken from the GEOS-Chem model simulations ( Supplementary Fig. 2). The NO 2 simulation over clean remote regions is generally consistent with independent satellite observations, with a monthly mean bias of <0.26 × 10 15 molecules cm −2 (Supplementary Fig. 3). Then, the SCD offset is added back to the NO 2 SCDs. Note that the reference spectra are selected for each cross-track row to minimize the cross-track bias due to instrument artifacts. The systematic cross-track bias in EMI NO 2 SCDs (the so-called "stripes", see Supplementary Fig. 4) is also observed for the OMI and TROPOMI products, and this bias can also be mitigated by using earth radiance as the reference spectrum 21 .
To account for the small variation in the spectral alignment due to the thermal variation in space 12 , we calibrated the additional spectral shift or squeeze and instrument slit function through cross-correlation with a high-resolution solar spectrum atlas 22 prior to the NO 2 DOAS fitting. The precalibrated measurement spectra lead to an~30% smaller SCD fitting uncertainty than using initial calibration parameters (Fig. 4c), as well as a fit residual, and the SCD is nearly unchanged (within~3.3%).

Discussion
The tropospheric NO 2 VCDs retrieved from EMI spectra are first validated against the OMI QA4ECV NO 2 products and the operational TROPOMI NO 2 products 23 . EMI measurements are compared to the OMI and TROPOMI products due to their similar instrument characteristics, i.e., the push-broom design, spectral bands, and near-noon overpass time at~13:30. Note that the TROPOMI NO 2 product generally followed the OMI QA4ECV NO 2 retrieval algorithm, but TROPOMI has a higher signal-to-noise ratio and spatial resolution 23 . amplitudes of NO 2 VCDs compared to OMI and TRO-POMI, while finer-scale details of NO 2 are captured by the satellite instrument with a higher spatial resolution. The EMI dataset overestimates NO 2 VCDs by up to 50% over polluted regions, such as the North China Plain (NCP) and India (Fig. 5d) compared to the TROPOMI observations. The spatiotemporal correlations between EMI NO 2 and TROPOMI NO 2 were also evaluated. For data taken from January to August 2019, the correlation coefficient (R) of daily mean NO 2 VCD time series over NCP between EMI and TROPOMI is 0.90, while the spatial correlation coefficient (R) of mean NO 2 VCDs over the NCP is 0.92 (Fig. 6). The remaining discrepancies between EMI and TROPOMI are mainly due to the NO 2 vertical profile used in the tropospheric AMF calculation, while the spectral fitting method (<3%) and stratospheric estimation method (<10%) only show a minor contribution ( Supplementary Fig. 5). The EMI tropospheric NO 2 VCDs are also compared to the ground-based NO 2 measurements from the Multi-AXis Differential Optical Absorption Spectroscopy (MAX-DOAS) instruments over northern China. A good agreement with a Pearson correlation coefficient (R) of 0.82 is found between the two datasets during January-August 2019 (Fig. 7, Supplementary Fig. 6). However, EMI generally underestimates tropospheric NO 2 VCDs by 30% compared to MAX-DOAS. The biases can be explained by the difference in spatial coverage between the ground-based and satellite observations 24 Table 1. b same as in a but using solar reference. c same as in a, but without spectral precalibration. The resulting NO 2 SCD, relative uncertainty, and RMS of the fitting residual are shown in the upper, middle, and bottom panels, respectively. The fitted NO 2 SCD and its uncertainty are masked in the white color when RMS > 0.004 new EMI tropospheric NO 2 retrieval provides reliable results for the investigation of air pollution distribution.

Materials and methods
The stratosphere-troposphere separation The stratospheric contribution of NO 2 must be subtracted from the total NO 2 column to derive the tropospheric NO 2 column. In the EMI NO 2 retrieval, we used the STRatospheric Estimation Algorithm from Mainz 14 to estimate the stratospheric contribution, which is based on the assumption that there is negligible contribution of tropospheric NO 2 columns over the remote Pacific and cloudy pixels in the middle latitudes. The weighting factors based on cloud and polluted regions,  which determines their impacts on the stratospheric estimate, are assigned to each satellite pixel. Subsequently, spatial smoothing based on weighted convolution is used to estimate the global stratospheric column.

NO 2 AMF calculations
The EMI NO 2 AMFs of each atmospheric layer (i.e., Box-AMFs) are calculated at 445 nm by the linearized pseudospherical vector model VLIDORT 26 version 2.7. In addition to the solar and satellite-viewing geometries provided in the level 1 data, additional atmospheric and surface information are needed in the AMF calculations. Surface albedo at 442 nm is taken from the OMI minimum earth's surface Lambertian equivalent reflectance 27 and interpolated to the EMI footprints. Considering the same local overpass time between EMI and TROPOMI, cloud top pressure and cloud fraction from TROPOMI 28 are used for the calculations of EMI NO 2 AMFs. A priori NO 2 profiles are taken from the high-resolution (~20 km) WRF-Chem simulations for the China domain and from GEOS-Chem simulations at the resolution of 2 × 2.5°for the global domain ( Supplementary Fig. 7). The spatial resolution of the NO 2 a priori profile is reportedly one of the dominant uncertainty sources during the NO 2 AMF calculations 29 . To expedite the calculation, these box-AMFs are precalculated and stored in the six-dimensional lookup table. Then, the box-AMF for each EMI observation can be derived by interpolating within the lookup table.