The Ultraviolet Visible Hyperspectral Imaging Spectrometer (UVHIS), and high-resolution NO2 mapping from its first airborne observation

We present a novel airborne imaging differential optical absorption spectroscopy (DOAS) instrument: Ultraviolet 10 Visible Hyperspectral Imaging Spectrometer (UVHIS), which is developed for trace gas monitoring and pollution mapping. Within a broad spectral range from 200 to 500 nm, operated in three channels, the spectral resolution of UVHIS is better than 0.5 nm. The optical design of each channel comprises a fore-optics with a field of view (FOV) of 40°, an Offner imaging spectrometer, and a charge-coupled device (CCD) array detector of 1032 × 1072 pixels. A first demonstration flight using UVHIS was undertaken on 23 June 2018, above an approximate 600 km area in Feicheng, China, with a spatial 15 resolution of about 25 × 22 m. Measurements of nadir backscattered solar radiation of channel 3 are used to retrieve vertical column densities (VCDs) of NO2 with a mean fitting error of 2.6 × 10 molec cm. The UVHIS instrument clearly detected several emission plumes transporting from south to north, with a peak value of 3 × 10 molec cm in the dominant one. UVHIS NO2 vertical columns are well correlated with ground-based mobile DOAS observations, with a correlation coefficient of 0.65 for all co-located measurements, and a slight underestimation for polluted observations. This study 20 demonstrates the capability of UVHIS for NO2 local emission and transmission monitoring.


Introduction
Nitrogen oxides (NOx), the sum of nitrogen monoxide (NO) and nitrogen dioxide (NO2), plays a key role in the chemistry of the atmosphere, such as the ozone destruction in the stratosphere (Solomon, 1999), and the secondary aerosol formation in the troposphere (Seinfeld and Pandis, 2016). In the troposphere, despite lightning, soil emissions and other natural processes, 25 the main sources of NOx are anthropogenic activities like fossil fuel combustion by power plants, factories, and road transportation, especially in the urban and polluted regions. As an indicator of anthropogenic pollution, the amounts and spatial distributions of NOx attract large attention.
Compared to NO, Nitrogen dioxide (NO2) is more stable in the atmosphere. Based on the characteristic absorption structures of NO2 in the ultraviolet-visible spectral range, the differential optical absorption spectroscopy (DOAS) technique has been 30 https://doi.org/10.5194/amt-2020-225 Preprint. Discussion started: 24 July 2020 c Author(s) 2020. CC BY 4.0 License. well applied to retrieve the light path integrated densities from different platforms (Platt and Stutz, 2008). Combining the imaging spectroscopy technique, imaging DOAS instruments were developed in recent years to determine the trace gases temporal variation as well as the two dimensional distribution. Global horizontal distribution of tropospheric NO2 and other trace gases has been mapped and studied by several space-borne sensors, including SCIAMACHY (Scanning Imaging Absorption Spectrometer for Atmospheric CHartographY; Bovensmann et al., 1999), GOME (Global Ozone Monitoring 35 Experiment; Burrows et al., 1999) and GOME-2 (Munro et al., 2016), OMI (Ozone Monitoring Instrument; Levelt et al., 2006) and TROPOMI (TROpospheric Ozone Monitoring Instrument; Veefkind et al., 2012). The Environmental trace gases Monitoring Instrument (EMI; Cheng et al., 2019;Zhao et al., 2018), as the first designed space-borne sensor in China, was launched on 9 May 2018, on-board the Chinese GaoFen-5 (GF5) satellite. In terms of spatial resolution, the majority of these space-borne sensors are lower than 10 × 10 km 2 , except TROPOMI with a relative higher resolution of 3.5 × 5.5 km 2 . 40 In order to achieve spatial resolution higher than 100 × 100 m 2 for investigation of spatial distribution in urban areas and individual source emissions, several researches have applied the imaging DOAS instruments on airborne platforms. The airborne imaging DOAS measurement was firstly performed by Heue et al. (2008) over South African Highveld plateau. For the purpose of retrieval of urban NO2 horizontal distribution, Popp et al. (2012) and Lawrence et al. (2015) performed their measurements separately in Zürich, Switzerland, and Leicester, England. In 2013, an airborne measurement focusing on 45 source emissions was performed in China, over Tianjin, Tangshan and Bohai gulf (Liu et al., 2015). An inter-comparison study of four airborne imaging DOAS instruments over Berlin, Germany, suggests a good agreement between different sensors, and the effectiveness of imaging DOAS to reveal the fine-scale horizontal variability in tropospheric NO2 in an urban context (Tack et al., 2019).
Here we present a novel airborne imaging DOAS instrument: Ultraviolet Visible Hyperspectral Imaging Spectrometer 50 (UVHIS), designed and developed by Anhui Institute of Optics and Fine Mechanics, Chinese Academy of Sciences (AIOFM, CAS). As a hyperspectral imaging sensor with high spectral and spatial resolution, UVHIS is designed to be operated on aircraft platform for the purpose of atmospheric trace gas measurements and pollution monitoring over large area in a relative short time frame. By using the DOAS technique and geo-referencing, two dimensional spatial distribution of its first demonstration flight over Feicheng, China is also presented in this paper. 55 The organization of this paper is listed as follows: Sect. 2 presents a technical description of UVHIS system, as well as its preflight calibration results. Section 3 introduces the detailed information of its first research flight above Feicheng, China.
Section 4 describes the developed algorithm for retrieval and geographical mapping of NO2 vertical column densities from hyperspectral data. Section 5 present the retrieved NO2 column densities, and Section 6 compares airborne measurements with correlative ground-based data sets from a mobile DOAS system. 60 https://doi.org/10.5194/amt-2020-225 Preprint. Discussion started: 24 July 2020 c Author(s) 2020. CC BY 4.0 License.

The UVHIS instrument
UVHIS is a hyperspectral instrument measuring nadir backscattered solar radiation in the ultraviolet and visible wavelength region from 200 to 500 nm. The instrument is operated in three channels at wavelength 200-276 nm (channel 1), 276-380 nm (channel 2) and 380-500 nm (channel 3) for minimal stray light effects and highest spectral performance. The main 65 characteristics of UVHIS are summarized in Table 1.
As shown in Fig. 1, the optical benches of three channels share a similar design, consisting of a telecentric fore-optics, an Offner imaging spectrometer, and a two dimensional charge-coupled device (CCD) array detector. The Offner imaging spectrometer consists of a concave mirror and a convex grating. Backscattered light below the aircraft is collected by a widefield telescope with a FOV of 40° in the across-track dimension. After passing through a bandpass filter and a 12.5 mm long 70 entrance slit in the focal plane, light is reflected and diffracted by a concave mirror and a convex grating. The dispersed light is imaged onto a frame transfer CCD detector, which consists of 1032 × 1072 individual pixels. For the purpose of alignment and slight adjustment of the spectrometer, only the central 1000 rows of pixel are well illuminated in the across-track dimension. In the wavelength dimension, the image covers central 1024 columns of pixels on the CCD detector while the left and right edge are used to monitor dark current. The spectral sampling and spectral resolution of all three channels can be 75 found in Table 1.
To reduce dark current and improve signal-to-noise ratio (SNR) of the instrument, CCD detector is thermally stabilized at −20°C with a temperature stability of ± 0.05°C (Zhang et al., 2017). However, the optical bench is not thermally controlled, because all the instrument is mounted inside the aircraft platform which is basically temperature consistent. UVHIS is mounted on a Leica PAV-80 gyro-stabilized platform that provides angular motion compensation. A high-grade Applanix 80 navigation system on-board is used to receive the position (i.e. latitude, longitude, and elevation) and orientation (i.e. pitch, roll, and heading) information, which is required for accurate geo-referencing. The UVHIS instrument telescope collect solar radiation backscattered from the surface and atmosphere through a fused silica window on the bottom of the aircraft. In the case of NO2 measurement, all observations in this study only use the channel 3.

Preflight calibration 85
In order to reduce errors in spectral analysis, spectral and radiometric calibration in the laboratory before actual airborne observation is very necessary.
For radiometric calibration, we use an integrating sphere with a tungsten halogen lamp for channel 2 and channel 3. While for channel 1, a diffuser plate with a Newport xenon lamp is used for a sufficient ultraviolet output. With the help of a well calibrated spectral radiometer to monitor the radiance of calibration system, digital numbers (DNs) from CCD detectors of 90 three channels can be converted to radiance correctly. Uncertainty of absolute radiance calibration of UVHIS is 4.89 % for channel 1, 4.67 % for channel 2 and 4.42 % for channel 3.
The preflight wavelength calibration was also performed in the laboratory, using a mercury-argon lamp and a tunable laser as sources. We model the slit function of UVHIS using a symmetric Gaussian function. Spectral registration and slit function calibration are achieved by least square fitting of characteristic lines in collected spectra. Table 2 lists the retrieved full-width 95 at half maximums (FWHMs) for channel 3.

Research flight
The first demonstration flight above Feicheng city, Shiheng town and neighbouring rural areas was performed on 23 June 2018, aiming at producing NO2 field maps of large area in a relative short time frame. UVHIS was operated from the Y-5 aircraft at altitude of 3 km above sea level, higher than the height of planetary boundary layer (PBL), with an average aircraft 100 ground speed of 50 m/s. An overview of the observation area and flight lines is provided in Fig. 2. The aircraft took off at local noon from the airfield in Pingyin county, about 19 km northwest of the centre of the field. An area of approximately 600 km2 was covered in 3 h, under clean sunny and cloudless conditions with low-speed southerly winds.
Research flight consists of 13 parallel lines in the east-west direction, starting from the lower left corner in Fig. 2. The distance between adjacent lines is 1.5 km, while the swath width of each individual line is about 2.2 km. Gapless coverage 105 between adjacent lines can be guaranteed in this pattern because of the adequate overlap. To validate the NO2 column densities retrieved from UVHIS by comparison to ground measurements, mobile DOAS measurements were performed inside the research area on the same day. As shown in Fig. 5, the measurements of the mobile DOAS system basically circle around the steel factory and the power plant, which are the presumed major emission sources inside the observation area.
At a typical flight altitude of 3 km in the Feicheng demonstration flight, the 40° FOV results in a swath width of 2.2 km on 110 the ground. In the condition of spatial binning by 10 pixels across-track, the spatial resolution of the ground pixel in this direction is about 22 m. Meanwhile along-track spatial resolution is determined by the product of the detector integration time and the aircraft ground speed, with some variation due to aircraft pitch angle. At typical aircraft ground speed of 50 m/s and integration time of 0.5 s, the along-track spatial resolution of the ground pixel is about 25 m.

Data processing chain 115
NO2 vertical column density (VCD) retrieval algorithm of UVHIS consists of four major steps. First, some necessary preprocessing procedures are required before any spectral analysis of UVHIS data. Next, UVHIS spectral data after preprocessing are analyzed in a suitable wavelength region by application of the well-established DOAS technique. After that, air mass factors (AMFs) are calculated for every observation based on SCIATRAN radiative transfer model to convert the slant column densities (SCDs) to vertical column densities. In the final step, combining with sensor position and 120 orientation information, NO2 VCDs are geo-referenced and overlaid onto Google satellite map layers.

Preprocessing
The preprocessing procedure before spectral analysis includes data selection, dark current correction, spatial binning, and inflight calibration. First, the spectral data acquired during U-turns of aircraft are removed in the processing because of the large and changing orientation angles. Also a threshold of radiance values is set to neglect some over-illuminated ground 125 pixels inside the flight area, which are usually caused by presence of cloud or water mirror reflection.
The dark current correction is performed based on the measurement at the start of the entire flight by blocking the foreoptics, which is necessary to improve instrument performance and reduce analysis error in DOAS fit.
In order to increase the SNR of the instrument and the sensitivity to NO2, the raw pixels of imaging DOAS are usually aggregated in across-track direction and across-track direction. According to photon statistics when only shot noise is 130 considered, the SNR should rise with square root of number of binned spectra. However, this improved SNR of instrument results a reduction of spatial resolution. In data analysis of Feicheng flight, we use binning of 10 pixels in across-track direction, resulting in a ground pixel size of about 25 × 22 m 2 .
Since the wavelength-to-pixel registration and the slit function shape of UVHIS could change compared to laboratory calibration results, in-flight wavelength calibration is essential for the next DOAS analysis. This in-flight wavelength 135 calibration is achieved by fitting the measured spectra to a high-resolution solar reference (Chance and Kurucz, 2010) with slit function convolution and wavelength shift. The nominal wavelength-to-pixel registration determined in laboratory calibration, is used as initial values in the iteratively fitting procedure for convergence to the optimal solution. The effective shifts and FWHMs of different across-track position are plotted in Fig. 3. Results at three wavelengths are shown: blue for 430 nm (the start of the analysis wavelength region), green for 450 nm (the middle of the analysis wavelength region) and 140 red for 470 nm (the end of the analysis wavelength region).

DOAS analysis
After preprocessing, the observed UVHIS spectra are analysed by the application of QDOAS (Danckaert et al., 2020) software in order to retrieve NO2 slant column densities. The basic idea of the DOAS approach is to separate broadband signals like surface reflectance and Rayleigh scattering, and narrow-band signals like trace gas molecular absorption. The 145 fitting window is within 430 and 470 nm wavelength region, considered to contain strongly structured NO2 absorption features, and with low interference of other trace gases such as O3, O4. Absorption cross-sections of NO2 and other trace gases and a synthetic Ring spectrum are simultaneously fitted to the logarithm of the ratio of the observed spectrum and a reference spectrum. These cross sections are made by convolving the high-resolution cross sections with in-flight wavelength calibration results for all across-track positions. Further details of the DOAS analysis setting can be found in 150 Table 3.
For each analysed spectrum, the direct result of the DOAS fit is differential slant column density (dSCD), which is the difference of NO2 integrated concentration along the effective light path between the studied spectrum and the selected https://doi.org/10.5194/amt-2020-225 Preprint. Discussion started: 24 July 2020 c Author(s) 2020. CC BY 4.0 License. reference spectrum (SCDref). Reference spectra is acquired over a clean rural area upwind of the urban and factory areas, in the lower left corner of Fig. 2. In the quite homogeneous background area, several spectra were averaged to increase the 155 SNR of the reference spectrum. For the purpose of avoiding across-track biases, a reference spectrum is required for each across-track detector because of its intrinsic spectral response. Under the assumption that the spatial and temporal variability of the stratospheric NO2 field is negligible during the test flight, this approach eliminates the stratospheric NO2 contribution, making the results only sensitive to the tropospheric portion. According to Popp et al. (2012), the residual NO2 amount in the background spectrum is estimated to be 1 × 10 15 molec cm −2 . 160 A sample NO2 DOAS fit result and corresponding residual of UVHIS spectra is illustrated in Fig. 4, with a differential slant column density (dSCD) of 5.05 ± 0.38 × 10 16 molec cm −2 and a RMS on the residuals of 4.56 × 10 −3 . The direct output of the applied DOAS fitting algorithm are dSCDs.

Air mass factor calculations
SCD is the integrated concentration along the effective light path of observation, which is strongly dependent on the viewing 165 geometry and the radiative transfer. VCD is the integrated concentration along a single vertical transect from the Earth's surface to the top of the atmosphere, which is independent of changes in the light path length of SCD.
As shown in Eq. (1), dSCDi from the DOAS fit can be converted to VCDi by dividing the AMFi which accounts for enhancements in the light path (Solomon et al., 1987). In this study, NO2 AMFs have been computed using the SCIATRAN 170 (Rozanov et al., 2014) radiative transfer model (RTM). SCIATRAN model numerically calculates AMFs based on a priori information on the parameters that change effective light path, such as sun and viewing geometry, trace gas and aerosol vertical profiles, surface reflectance etc. Coastal aerosol band (433 to 450 nm) is selected because its bandwidth is relatively narrow, and this band is basically inside 180 the DOAS fitting window (Vermote et al., 2016). (4) Since no accurate trace gas vertical profile is available during the flight, a well-mixed vertical distribution (box profile) of NO2 in the PBL is assumed. However, accurate PBL height is also unavailable, a typical height of 2 km is a reasonable guess in the case of sunny summer day in mid-latitude area in China. (5) Aerosol optical Depth (AOD) information used in AMF calculation is MODIS AOD product at 470 nm (Remer et al., 2005), https://doi.org/10.5194/amt-2020-225 Preprint. Discussion started: 24 July 2020 c Author(s) 2020. CC BY 4.0 License.

Parameters in RTM
because neither ground-based aerosol measurement is performed, nor any AERONET station data near the flight area is 185 available. Like the NO2 profile, the aerosol extinction box profile is constructed from the PBL height and AOD. Single scattering albedo (SSA) is assumed to be 0.93, and asymmetry factor is assumed to be 0.68 for aerosol extinction profile.
The radiative transfer equation in SCIATRAN is solved in a pseudo-spherical multiple scattering atmosphere, using the scalar discrete ordinate technique. Simulations are performed for sensor altitude of 3 km above sea level, and wavelength of the middle of the NO2 fitting windows, i.e. 450nm. A NO2 AMF look-up table (LUT) was computed, with different RTM 190 parameter settings provided in Table 4. For each retrieved dSCD, an AMF was linear interpolated from the LUT based on the sun geometry, viewing geometry, and surface reflectance.

RTM dependence study
1. AMF dependence on the surface reflectance  Fig. 6, indicate that the relation between surface reflectance and AMF is non-linear. Especially when surface reflectance is blow 0.1, AMF increase with surface reflectance rapidly.
Generally speaking, the AMF should be higher in the case of a bright surface reflectance, because more sunlight is reflected 205 from the ground back to atmosphere and then recorded by the airborne sensor. Compared to rural areas, urban and industry areas usually exhibit an enhancement in value of surface reflectance and subsequently an increment of AMF. As shown in Fig. 7, the dependency of the AMF on the surface reflectance is very strong. Also a strong variability of surface reflectance and AMF can be observed in these areas.
In this paper we use Landsat 8 surface reflectance product (Vermote et al., 2016) of coastal aerosol band (band 1) as input 210 parameter in AMF calculation. The surface reflectance is retrieved through atmospheric correction, using the Second Simulation of the Satellite Signal in the Solar Spectrum Vectorial (6SV) model (Vermote et al., 1997). Since there is no overpass on the same day inside UVHIS research flight area, we choose surface reflectance product on 3 May 2018 considering the sunny weather condition and no cloud presence. The spatial resolution of Landsat is about 30 m, which is https://doi.org/10.5194/amt-2020-225 Preprint. Discussion started: 24 July 2020 c Author(s) 2020. CC BY 4.0 License. slightly larger than UVHIS. A resampling of Landsat 8 surface reflectance product based on nearest neighbour interpolation 215 is performed for every UVHIS ground pixel.

AMF dependence on profiles
Based on airborne UVHIS retrieval product, the horizontal distribution of NO2 can be detected, but the vertical distribution of NO2 in the atmosphere is not available. The assumptions we made for profile shape of trace gas and aerosol extinction do not consider the effective variability during research flight, which can be expected in an urban area. Focusing on the impact 220 of different profile shapes on the AMF computation, sensitivity tests of two different NO2 profiles which are closer to ground surface are performed: well-mixed NO2 box profile of 0.5 and 1 km height. Compared to the box profile of 2 km which is close to the height of PBL, AMFs decrease by an average of 13 % in the case of a box profile of 1.0 km, while AMFs decrease by an average of 22 % in the case of a box profile of 0.5 km.
Depending on the relative position of aerosol and trace gas layer, the optical thickness and scattering properties, aerosols can 225 enhance or reduce the AMF in different ways [28]. If an aerosol layer is located above the majority of the trace gas, the aerosols with high SSA have a shielding effect as less solar scatter light pass through the trace gas layer, leading to a shorter light path. On the other hand, if aerosols and the trace gas are present in the same layer, aerosols can lead to multiple scattering effects which extend the light path and result a larger AMF. According to the simulations of a well-mixed aerosol box profile of 2 km and a pure Rayleigh atmosphere, AMFs are slightly higher (about 1 %) compared to pure Rayleigh 230 scenario.

AMF dependence on sun and viewing geometries
As can be seen in Fig. 5, the effect of sun and viewing geometries on AMFs is very small. Based on previous study of Tack et al. (2017), the changing SZA have the greatest effect on AMFs, among other sun and viewing geometries. The SZA varies from 12.8° to 37.4° during the 3 h research flight, and AMF at the end of the flight is 10 % larger than at the beginning of the 235 flight when other parameters are assigned to mean value (Fig. 8). Generally, a larger SZA could result a longer light path through the atmosphere and thus a larger AMF.

AMF dependence on the analysis wavelength
The dependence of AMF on analysis wavelength is shown in Fig. 9. It is obvious that the AMF increase with the analysis wavelength. This could be explained by Rayleigh scattering characteristics: photons at shorter wavelengths are more 240 scattered before NO2 layer than photons at longer wavelengths, leading to the reduced sensitivity to AMF at shorter wavelengths. In the DOAS analysis wavelength window of 430-470 nm, the increase in AMF is about 2 %. https://doi.org/10.5194/amt-2020-225 Preprint. Discussion started: 24 July 2020 c Author(s) 2020. CC BY 4.0 License.

Geo-referencing and mapping
Accurate geo referencing is essential for emission source locating and data comparison, and can be achieved with sensor position and orientation information recorded by navigation system and IMU on-board. After geo-referencing, the NO2 245 VCDs are gridded to combine overlapped adjacent measurements, with a spatial resolution of 0.0003° × 0.0002°.
Corresponding to 27 × 22 m 2 , the grid size used is slightly larger than the effective spatial resolution of the UVHIS for the purpose of reducing the number of empty grid cells. All VCDs are assigned to a grid cell based on its centre coordinates, and several VCDs in one grid cell are unweighted averaged. The final NO2 VCD distribution map is plotted over satellite Maps layers in QGIS 3.8 software (QGIS development team, 2020). 250

Results
The NO2 VCD two-dimensional distribution map is shown in Fig. 10 for the research flight on 23 June 2018. In general, with a high performance of UVHIS in spectral and spatial resolution, Figure 10 shows fine-scale NO2 spatial variability to resolve individual emission sources. The NO2 distribution is dominated by several exhaust plumes with enhanced NO2 concentration compared to other areas, which share a largely consistent transportation pattern from south to north. The largest plume with 255 peak values of up to 3 × 10 16 molec cm −2 , is transporting for at least 9 km downwind inside the observation area. The map is built from 13 overlapped adjacent flight lines, the acquisition time of every track is about 10 minutes. Due to temporal discontinuity of flight lines and dynamic characteristics of NO2 field, biases can be observed between adjacent flight lines, especially inside the polluted area when wind direction is changed.
Inside flight area, the predominant NO2 sources are mainly related to anthropogenic emitters in the northwest part. These 260 sources include a power plant, a steel factory, two cement factories, several carbon factories etc. A strong gradient from south to north is shown in all plumes, consistent with the southerly wind direction. The largest plume originates from an emitter inside a steel factory, which is represented by number 3 in Fig. 10. This dominant plume is transporting at least 9 km from south to north and seems to be continuing outside the flight region. As the plume spreads further north, it becomes much broader. However, it is discontinuous and inhomogeneous in the distribution map, because of time differences among 265 adjacent flight lines, as well as a dynamically changing NO2 field which is related to wind direction and speed. As shown in Fig. 10, this plume reaches its peak value outside the steel factory, at a small valley about 1 km to the north. This enhanced level of NO2 may be caused by terrain factor which contributes to the accumulation of pollution gases.
Number 4 in Fig. 10 represents an isolated exhaust plume originated from another emitter inside the steel factory, with a peak value of 1.4 × 10 16 molec cm −2 . A weaker plume with peak values of 1.0 × 10 16 molec cm −2 is also detected by UVHIS, 270 which seems to originate from the power plant. Indicated by number 2 in Fig. 10, this power plant is less than 2 km south of the steel factory. Number 1 in Fig. 10 indicates several carbon factories, which are located on the left side of the flight area.
Several plumes with peak values of 1.5 × 10 16 molec cm −2 , gradually merge together during the transportation downwind.

Number 5 and Number 6 in
The urban areas of Feicheng city and Pingyin county are located on the right side, and on the upper left corner of the flight area respectively. Compared to the industrial areas mentioned above, the pollution levels of the rural and urban areas are much lower due to the lack of contributing sources, ranging from 1 to 4 × 10 15 molec cm −2 .

Uncertainty analysis 280
The total uncertainty on the retrieved NO2 VCDs is composed of three parts: (1) uncertainties in the retrieved dSCDs; (2) uncertainties in reference column SCDref; (3) uncertainties in computed AMFs. Assuming that these uncertainties originating from independent steps are sufficiently uncorrelated, the total uncertainty of NO2 VCD can be quantified as follows: The first uncertainty source, σdSCDi, originates from DOAS fit residuals and is a direct output in QDOAS software. This 285 dSCD uncertainty is dominated by shot noise from radiance, electronic noise from the instrument, uncertainties from the cross sections and errors from wavelength calibration. In this study, spatial binning of 10 pixels is performed to reduce this DOAS fit residuals, with a mean slant error of 4.8 × 10 15 molec cm −2 . Compared to mobile DOAS system, this DOAS fit error is approximately two times larger. Unlikely situation of mobile DOAS system, it becomes a major contributor to the total uncertainty in the case of an airborne sensor, especially in a clean area. 290 The second uncertainty source, σSCDref, is caused by the NO2 residual amount in the reference spectra. Since no direct measurement of the NO2 column is performed in the reference area, an uncertainty of 100 % is assumed on the estimated NO2 background amount. This results in an uncertainty of 1.0 × 10 15 molec cm −2 in the background column.
The third uncertainty source, σAMFi, derives from uncertainties in the parameter assumptions of radiative transfer model inputs. According to previous studies (Boersma et al., 2004;Pope et al., 2015), it is treated as systematic and depends on the 295 surface albedo, NO2 profile, aerosol parameters, and cloud fraction. (1) Since the research flight took place under cloudless conditions, cloud fraction is neglected in this case.
(2) Results of the dependence tests in Sect. 4.3.2 suggest that the surface albedo has the most significant effect on the AMF. According to Vermote et al. (2016), the uncertainty of LANDSAT 8 surface reflectance product of band 1 is 0.011. (3) Based on the sensitivity study performed in Sect. 4.3.2, the uncertainty related to the a priori NO2 profile shape is lower than 22 %. (4) According to the performed simulations of a pure Rayleigh 300 atmosphere, the uncertainty related to aerosol state is estimated to be less than 1 %. (5) Because of the high accuracy of the https://doi.org/10.5194/amt-2020-225 Preprint. Discussion started: 24 July 2020 c Author(s) 2020. CC BY 4.0 License.
viewing and sun geometry, and its low impact on the AMF computation revealed in the previous section, the uncertainty related to viewing and sun geometry is expected to be ignorable. Therefore, combining all uncertainty sources in quadrature, a mean relative uncertainty of 24 % on the σAMFi is obtained.
Based on above discussion, the total uncertainties on the retrieved NO2 VCDs of all observations of the research flight are 305 calculated. They typically range between 1 and 5 × 10 15 molec cm −2 , with a mean value of 2.6 × 10 15 molec cm −2 .

Comparison to mobile DOAS measurements
In order to compare UVHIS NO2 VCDs to ground-based measurements, mobile DOAS observations were performed on 23 June 2018 as well. This mobile DOAS system is composed of a spectrum acquisition unit and a GPS module. The spectrum collection unit consists of a spectrometer, a telescope, an optical fiber, and a workbench. The FOV of this telescope is 0.3°, 310 and its focal length is 69 mm. The spectrometer used is a Maya 2000 Pro spectrometer, with a wavelength range of 260-420 nm and a spectral resolution of 0.55 nm. Zenith-sky observations of mobile DOAS are adopted for minimal blocking of buildings and trees in this research. The important properties of the mobile DOAS system and its NO2 retrieval approach are shown in Table 5.
Basically, the route of the mobile DOAS is designed to encircle the power plant and steel factory which are supposed to be 315 predominant sources. For the comparison, mobile DOAS observations are first gridded to the same sampling of UVHIS pixels, then the VCD of UVHIS NO2 results is extracted for each co-located mobile measurement. An overview of the mobile DOAS measurements over UVHIS NO2 layer is shown in Fig. 11. It is obvious that NO2 distributions of mobile DOAS system and UVHIS exhibit similar spatial characteristics, which low values are located in the south of the steel factory and power plant, and high values are near the downwind of several plumes. 320 Figure 12 shows scatter plots with VCDs retrieved by UVHIS on the x-axis and mobile DOAS VCDs on the y-axis. The corresponding results of linear regression analysis are also provided in Fig. 12, with a correlation coefficient of 0.65, a slope of 1.04, and an intercept of 3.27 × 10 13 . In general, an overestimation of mobile DOAS VCDs of increased value can be observed. The absolute time offset between mobile DOAS and airborne observations can be up to 1 h, which means that both instruments cannot sample the NO2 column at certain geolocation simultaneously. Considering the variability in local 325 emissions and meteorology, it is reasonable that the differences between these two instruments exist. Besides, the averaging effect of the area inside an UVHIS pixel can also lead to the underestimation of UVHIS compared to mobile DOAS system.

Conclusions
In this paper, we present the newly developed Ultraviolet Visible Hyperspectral Imaging Spectrometer (UVHIS), with a broad spectral region from 200 to 500 nm, and a high spectral resolution better than 0.5 nm. The instrument is operated in 330 three channels at wavelength 200 to 276 nm (channel 1), 276 to 380 nm (channel 2), and 380 to 500 nm (channel 3) for https://doi.org/10.5194/amt-2020-225 Preprint. Discussion started: 24 July 2020 c Author(s) 2020. CC BY 4.0 License. minimal stray light effects and highest spectral performance. The optical design of each channel consists of a fore-optics with a FOV of 40°, an Offner imaging spectrometer, and a CCD array detector of 1032 × 1072 pixels.
Also we present the first NO2 retrieval results from the UVHIS airborne observation in June 2018. The research flight above Feicheng, China, covered an area of about 20 × 30 km 2 within 3 h, with a high spatial resolution about 25 × 22 m 2 . We first 335 retrieve the differential NO2 slant column densities from nadir observed spectra by application of DOAS technique, to a mean reference spectra over a clean area. Then we convert those NO2 slant columns to vertical columns using air mass factors derived from SCIATRAN model with Landsat 8 surface reflectance product. Total uncertainties of NO2 vertical columns are in range of 1 to 5 × 10 15 molec cm −2 , with a mean value of 2.6 × 10 15 molec cm −2 .
The two-dimensional distribution map of NO2 VCD demonstrates that UVHIS is adequate for trace gas pollution monitoring 340 over a large area in a relative short time frame. With the high spatial resolution of UVHIS, different local emission sources can be distinguished, fine-scale horizontal variability can be revealed, and trace gas emission and transmission can be understood. For the flight on 23 June 2018, NO2 distribution is dominated by several exhaust plumes which exhibit same south to north direction of transmission, with a peak value of 3 × 10 16 molec cm −2 in the dominant plume. Comparisons of UVHIS NO2 vertical columns to mobile DOAS observations show good agreement overall, with a correlation coefficient of 345 0.65 for all co-located measurements. However, an underestimation of the high NO2 columns of UVHIS is observed relative to the mobile DOAS measurements.
High-resolution information about the NO2 horizontal distribution, generated from UVHIS airborne data, is unique and valuable compared to ground-based instruments and space-borne sensors. In future study, UVHIS could be applied in validation of satellite trace gas instruments, and in connection between local point observations, air quality models, and 350 global monitoring from space.    (2013) Ring effect Chance and Spurr (1997) Polynomial term Order 5 Intensity offset Order 1 490    (2013) Ring effect Chance and Spurr (1997) Polynomial term Order 5 Intensity offset Order 1 505 510 515 520 525 https://doi.org/10.5194/amt-2020-225 Preprint. Discussion started: 24 July 2020 c Author(s) 2020. CC BY 4.0 License.