Currently, almost 60 % of the world population is living in urban areas, where they are exposed to emissions from the majority of anthropogenically produced air pollutants. Nitrogen dioxide (NO2) is a trace gas and key pollutant that can be considered a proxy for air quality and pollution in an urban environment, as it mainly originates from combustion processes such as burning of fossil fuels, which are mainly related to traffic and industry. NO2 plays an important role in atmospheric chemistry and can have a direct impact on human health. It is a short-lived species with a strong local character and concentrations that can vary strongly in both space and time. For the reasons stated, the monitoring and high-resolution mapping of the NO2 distribution is considered to be of great (social) relevance.

For about 2 decades, tropospheric trace gases, such as NO2, have been monitored and mapped at a global scale by spaceborne sensors like ESA's SCIAMACHY (Scanning Imaging Absorption Spectrometer for Atmospheric CHartographY), ESA's GOME (Global Ozone Monitoring Experiment), ESA–EUMETSAT's GOME-2, and NASA's OMI (Ozone Monitoring Instrument). See for example Richter and Burrows (2002), Beirle et al. (2010), Boersma et al. (2011), Hilboll et al. (2013), Valks et al. (2011), and Bucsela et al. (2013). However, the coarse spatial resolution of the order of a few tens of kilometres of these spaceborne air quality instruments makes them ineffective for studies of the NO2 field at the scale of cities and for resolving individual emission sources.

In the last decade, a number of studies have explored the potential of airborne imaging DOAS systems for higher-resolution mapping of the spatial distribution of tropospheric gases. The majority of these studies have focused on the retrieval of the NO2 field over urban areas and/or industrial sites, i.e. Heue et al. (2008), Kowalewski and Janz (2009), Popp et al. (2012), General et al. (2014), Lawrence et al. (2015), Schönhardt et al. (2015), Nowlan et al. (2016), Lamsal et al. (2017), Meier et al. (2017), Tack et al. (2017), Vlemmix et al. (2017), Broccardo et al. (2018), Merlaud et al. (2018), and Nowlan et al. (2018).

As the developed instruments vary in design, size, specifications, and data analysis applied, it is interesting to compare results from simultaneous observations. Here we present the first intercomparison study of NO2 VCDs, retrieved by the differential optical absorption spectroscopy (DOAS) analysis of visible spectra, observed by four different airborne imaging DOAS spectrometers. The instruments were operated simultaneously from two manned aircraft over Berlin during the ESA-funded AROMAPEX campaign that took place in April 2016.

The primary objective of the AROMAPEX project was to test and intercompare experimental airborne atmospheric imagers, dedicated to the geographical mapping of the spatial distribution of tropospheric NO2. AROMAPEX is also a preparatory step for forthcoming intercomparison and validation campaigns of satellite air quality sensors. In the coming years, a new generation of spaceborne instruments will be launched, providing information on atmospheric variables at much higher spatial resolution, of the order of a few kilometres. These measurements will be valuable for air quality, atmospheric composition, and climate monitoring studies and services. ESA launched Sentinel-5 Precursor (S-5P), a sun-synchronous low Earth orbit (LEO) mission (Ingmann et al., 2012) on 13 October 2017, and has planned the launch of the first Sentinel-5 (S-5) in 2021. Additionally, a range of geostationary (GEO) missions are planned: ESA's Sentinel-4 (S-4) (Ingmann et al., 2012), NASA's TEMPO (Tropospheric Emissions: Monitoring of Pollution; Chance et al., 2013; Zoogman et al., 2017), and KARI's GEMS (Geostationary Environmental Monitoring Spectrometer; Kim, 2012). The unprecedented characteristics of these instruments, such as higher spatial and temporal resolution, will create many new science opportunities, but also retrieval challenges. The AROMAPEX campaign and study are aimed at the preparation of the validation of trace gas products from future spaceborne systems and the study of satellite intra-pixel variability.

The paper is organised as follows: Sect. 2 presents the context of the AROMAPEX project and provides details about the set-up of the airborne campaign held in Berlin. Section 3 briefly introduces the four airborne imaging DOAS systems, operated during AROMAPEX. Section 4 describes the data analysis of the airborne observations for the retrieval and geographical mapping of the NO2 VCDs. In the following two sections, the resulting NO2 VCD distribution maps are discussed and compared with mobile car DOAS measurements. Section 7 discusses a quantitative assessment by intercomparing the co-located NO2 VCD products, retrieved from the four imagers.

The AROMAPEX campaign was held in Berlin from 11 to 22 April 2016. An overview of the area, flight plan, and main campaign sites is provided in Fig. 1. The four imaging DOAS systems were operated from two manned aircraft, performing time-synchronised flights at different altitudes: APEX (Airborne Prism EXperiment) was operated from the DO-228 D-CFFU aircraft of DLR (Deutsches Zentrum für Luft- und Raumfahrt) at 6.2 km a.g.l., while AirMAP (Airborne imaging DOAS instrument for Measurements of Atmospheric Pollution), SWING (Small Whiskbroom Imager for atmospheric compositioN monitorinG), and Spectrolite Breadboard Instrument (SBI) were operated from the Cessna 207T D-EAFU of FUB (Free University Berlin) at 3.1 km a.g.l. The cruise altitudes of both aircraft were well above the planetary boundary layer (PBL), containing the majority of tropospheric NO2. The aircraft operated from the Schönhagen airfield (see Fig. 1), 40 km southwest of Berlin, while the research teams were based at the Institute for Space Sciences of FUB, where measurements of additional atmospheric parameters were made.

The complex flight constellation was carefully planned in order to optimise the acquisition for trace gas retrieval purposes. Due to rainy and cloudy weather conditions at the beginning of the campaign, the two scheduled flights both took place on 21 April, the only clear-sky day during the campaign (see Table 1). The first flight took place in the morning from 09:34 to 12:01 LT and the second flight in the afternoon from 14:24 to 16:39 LT. The entire city of Berlin, as well as the semi-urban and rural area east and south of the city, was covered by both flights. An area of approximately 800 km2 was covered, consisting of 14 flight lines for the morning and afternoon flights. Note that due to a small delay of the Dornier aircraft, the second flight line of the morning flight was skipped in order to be better time synchronised with the Cessna. This explains the data gap in the retrieved APEX NO2 VCD distribution map (see Fig. 11). The absolute temporal offset between both aircraft above a certain position was 10 and 12 min on average for the morning and afternoon flights, respectively, with a maximum time difference of 24 min.

The flight plan consisted of adjacent straight flight lines, alternately flown from south to north and from north to south, with the first flight line in the west. Due to the large roll angles, spectra acquired during turns of the aircraft in between flight lines are not taken into account in the comparison. The flight plan approved by air traffic control (ATC) was initially larger than the area covered, in order to have some flexibility to adapt the actual flight pattern to the wind direction. Downwind of the sources, the maximum number of flight lines was retained in order to catch the urban plume. Upwind of the main known sources, the number of flight lines was reduced. In the case of the flights on 21 April, more flight lines were foreseen in the east as a result of the predicted west wind.

The AROMAPEX campaign is part of the AROMAT-I and AROMAT-II (Airborne Romanian Measurements of Aerosols and Trace gases) activities (Constantin et al., 2016), carried out in Romania in September 2014 and August 2015. The campaign was initially planned to take place in Bucharest, Romania, in summer 2015 but was eventually rescheduled to take place over Berlin in spring 2016 due to critical issues with the flight approvals over Romania for the DLR Dornier aircraft. AROMAPEX builds on the experience gained during the AROMAT campaigns (first flights with AirMAP and SWING together for NO2 and SO2 retrievals) and the BUMBA campaigns (Belgian Urban NO2 Monitoring Based on APEX remote sensing) held in April–June 2015 and July 2016 in Belgium (Tack et al., 2017).

The characteristics of the four airborne imaging DOAS instruments, which were operated during the AROMAPEX campaign, are only briefly discussed here with a focus on their differences, and the main specifications are summarised in Table 2. References are provided below, containing a more detailed and technical discussion of each instrument and data analysis. A mosaic of the four imaging instruments is shown in Fig. 2.

The retrieval and geographical mapping of NO2 VCDs, based on spectra acquired by the airborne imagers, consists of a three-step approach. First, the well-established DOAS technique (Platt and Stutz, 2008), based on the Beer–Lambert law, is applied on the observed backscattered solar radiation in the visible wavelength region (Sect. 4.1). For each analysed spectrum, this results in the retrieval of a slant column density (SCD), which is the concentration of NO2 integrated along the effective viewing path. SCDs depend on the optical path of the observation and are thus strongly dependent on the viewing geometry and the radiative transfer. In the next step, an AMF (Solomon et al., 1987) is computed for each observation by modelling an assumed state of the atmosphere and transfer of the solar radiation through the atmosphere, based on a radiative transfer model (RTM) (Sect. 4.2). AMFs are the factor between the slant and the vertical column, accounting for the effects of viewing and sun geometry, surface reflectance, aerosol scattering, and the NO2 vertical distribution. SCDs from the DOAS fit can then be converted to VCDs, which are the integrated amount of NO2 along a single vertical transect from the Earth's surface to the top of the atmosphere: VCDi=SCDiAMFi. VCDs are a more geophysically relevant quantity, independent of changes in the optical path length of the SCDs, e.g. due to high surface reflectance or a large solar zenith angle (SZA). VCD retrievals from different DOAS instruments can therefore be compared in a meaningful way. In a third and final step, the observations are combined with the recorded sensor position and orientation, allowing a proper geographical mapping of the NO2 VCDs (Sect. 4.3). The retrieval approaches and (the impact of) the parameter settings are only briefly discussed in the next sections. For full details on the APEX, AirMAP, SWING, and SBI retrieval approaches, we refer respectively to Tack et al. (2017), Meier et al. (2017), Merlaud et al. (2018), and Vlemmix et al. (2017).

The generated NO2 VCD distribution maps are shown in Figs. 11 and 12 for respectively the morning (09:34–12:01 LT) and afternoon (14:24–16:39 LT) flights on 21 April 2016. Note that all data sets are given the same NO2 VCD color-coding. Note as well that due to practical reasons and time restrictions during the project (time-inefficient retrieval code developed in the framework of a master student graduation project), the first and last two flight lines of the morning flight were not analysed in the processing of SBI level-2 data. The NO2 VCD maps were convolved by a Savitzky–Golay low-pass filter (Savitzky and Golay, 1964; Schafer, 2011). The filter was only applied for visualisation purposes and thus was not used for the quantitative comparison discussed in Sect. 7. Hourly averaged wind profiles were derived with an ADS–B receiver, collecting data from ascending and descending aircraft (Bütow, 2016). The Mode S transponder signals, sent out by most airliners, include all necessary information to calculate temperature and wind profiles. The accuracy of the derived profiles was improved by averaging a large number of data points, coming from different aircraft (see Fig. 13). Hourly averaged wind vectors, indicating the surface wind at flight time, are provided in Figs. 11 and 12. The NO2 horizontal distribution, observed by the different DOAS imagers, is consistent to a high degree. Note, however, the coarser spatial resolution of the SWING grid (see Sects. 3.3 and 4.3) and the non-continuous SBI grid due to the narrow FOV of the used telescope (see Sect. 3.4).

It is known from emission inventory data (Berlin Senate Department for Urban Development and the Environment, 2017) that an area with strong NOx emissions is located in the northwestern part of the city of Berlin. According to the emission inventory, potential strong NOx emitters are the power plant Reuter West (600 MW) and other industrial facilities close by, as well as the conference centre “Messe Berlin”. These sites were consequently covered by the flight plan. The wind was blowing from the west and patterns of enhanced NO2 can be clearly observed in the data, which are transported downwind from this area. The NO2 distribution is dominated by an exhaust plume with peak values of up to 2×1016 molec cm-2, crossing the city from west to east, and related to the large power plant Reuter West. The steam boilers are fired by hard coal and equipped with efficient flue gas scrubbers to generate electricity and heat simultaneously (Vattenfall AB, 2017). The large plume from the power plant is covered for more than 30 km downwind and is continuing towards the east, outside of the acquired region. According to a study of OMI tropospheric NO2 products over the Highveld region in South Africa, such plumes can be sufficiently stable to retain their structure for several hundreds of kilometres downwind (Broccardo et al., 2018). Enhanced levels of NO2 were indeed observed, approximately 65 km east of Berlin, where the Cessna 207T D-EAFU performed a sounding (not shown).

The plume is clearly confined until it reaches the central part of the city. Then, the plume broadens towards the east and appears to be more inhomogeneous. This is mostly due to the contribution of emissions from traffic and local sources in the city, but part of the apparent inhomogeneity may be caused by time differences among subsequent flight lines in combination with a dynamically changing NO2 field, as well as the synoptic view of different NO2 layers, which are subject to slightly different wind regimes. As the dominant plume crosses the city centre and ring road, city traffic-related NO2 cannot easily be differentiated from it. Examples of differentiating between industrial and traffic emissions have been discussed in earlier studies such as Popp et al. (2012), Meier et al. (2017), and Tack et al. (2017).

Parallel to the Reuter West exhaust plume and just south of it, a second major west–east-oriented plume is detected by all DOAS imagers in the morning data. The plume seems to originate from a power and ventilation station at the Messe Berlin conference centre. A third clear line source pattern of enhanced NO2 is observed further southeast and seems to be transported from the highways A100 and A113 and industrial buildings surrounding the highways. The NO2 levels are, however, lower than in the two main plumes.

In the southern part of the acquired region, upwind of the city, the pollution levels are much lower due to the lack of major sources in this predominantly suburban, rural area. NO2 VCD map statistics are summarised in Table 6: for the morning flight, NO2 levels range between 1×1015 molec cm-2 in the south and 20×1015 molec cm-2 in the dominant plume, with a mean of 7.3±1.8×1015 molec cm-2. The mean NO2 VCD is relatively low because of the acquisition of a large background area.

The afternoon data set (see Fig. 12) largely exhibits the same NO2 distribution. Although slightly higher peak values of up to 23×1015 molec cm-2 are observed, the mean VCD of 6.0±1.4×1015 molec cm-2 is lower than for the morning flight. The main exhaust plume, related to the Reuter West power plant, can be observed again. However, the afternoon plume appears to be broken close to the source, which may originate from interruptions in the emissions or plume displacements between overpasses. We checked if two similar looking NO2 hotspots detected in two adjacent flight lines, and indicated by a white asterisk in Fig. 12, could be the same plume feature, transported over the acquisition time of both locations. The measured distance between the two points is approximately 2.3 km. Based on the average wind direction and wind speed of 3.6 m s-1 and the interval in acquisition time, we determined empirically that the plume feature should have moved over 2.8 km. Differences are expected by variations from the average wind speed and different wind speeds at plume height than the assumed surface wind.

The plume is less confined than in the morning and more expanded in north–south direction, which could be related to the weaker wind from the west (around 3.6 m s-1 at the surface). The wind direction is also more unstable during the afternoon flight, with the surface wind changing from 301∘ at 15:00 LT to 273∘ at 16:00 LT, and 287∘ at 17:00 LT. The slightly different structures observed in the plume, by APEX and AirMAP, could be explained by a combination of (1) the strong spatio-temporal variability in the NO2 field, (2) the delay of up to 20 min in acquisition of the NO2 field from the Dornier and the Cessna, and (3) the fact that the maps are built from adjacent flight lines within the time frame of a few hours. Based on the average wind speed of 3.6 m s-1 and taking into account the delay of up to 20 min in acquisition time of the NO2 field, we estimate that the plume features have been transported over a distance of 4.3 km to the east-southeast within this time interval.

The plume related to the Messe Berlin power station is not detected in the afternoon observations, while the plumes transported from the highways A100 and A113, running south of the city, can be observed again. In the southern part, the background levels seem to increase smoothly to the east. A large artefact is identified in the south (see white dot in Fig. 12), resulting in enhanced APEX NO2 VCDs and decreased AirMAP VCDs. The difference is approximately 1×1016 molec cm-2. The artefact seems to be strongly correlated with a crop field and was identified as winter rapeseed. A possible explanation is that the spectral signature of this crop is spectrally correlated with the NO2 cross section, affecting the retrievals in a different way depending on the chosen fitting interval. The effect could be similar to the sand–soil signature, discussed in Richter et al. (2011) and Merlaud et al. (2012). Note that a number of smaller similar artefacts are observed in the south, related to the same type of crop.

In general, the NO2 VCD results of the two flights show very similar spatial patterns. All four DOAS imagers allow us (1) to retrieve the NO2 horizontal variability at city scale and (2) to resolve local emission sources. Despite the coarser spatial resolution of SWING, the instrument is able to detect all the relevant patterns of enhanced NO2. The distribution maps show that the NO2 tropospheric columns (1) have an inhomogeneous distribution, (2) can be highly variable, and (3) can exhibit strong gradients in an urban context. Due to the relatively coarse spatial resolution of current spaceborne air quality sensors and the local representativeness of ground-based observations, airborne data sets currently provide a unique way to measure and visualise the horizontal distribution of pollutants at the scale of cities.

As mentioned in Sect. 4.4, 3-D effects of the radiative transport are not taken into account in this study. It is expected that the effective spatial resolution assigned to the VCDs will be reduced by up to 2 orders of magnitude. Nevertheless, the different data sets will be affected in nearly the same way (same NO2 field, same SZA, but slightly different viewing geometry), reducing the impact of 3-D effects of the radiative transport on the intercomparison results of this study.

The APEX NO2 VCD retrievals of the morning and afternoon flights have been compared with an independent correlative data set acquired by a mobile car DOAS system, in a way similar to that of APEX acquisitions over Belgium (Tack et al., 2017) and AirMAP (Meier et al., 2017) and SWING (Merlaud et al., 2018) acquisitions over Romania. During AROMAPEX, mobile car DOAS measurements were performed by the University of Galati (UGAL), the Max Planck Institute for Chemistry in Mainz (MPIC), and the Royal Belgian Institute for Space Aeronomy (BIRA). In this study, we only validate the APEX NO2 VCDs based on the UGAL car DOAS observations, as this data set contains most of the NO2 variation, covering background areas as well as large parts of the key NO2 plumes. Note that harmonisation and intercomparison of different car DOAS observations, performed during several campaigns, including AROMAPEX, are currently ongoing and a full comparison with airborne retrievals will be the focus of a future study.

Details on the instrumental set-up of the UGAL zenith-sky car DOAS system and the NO2 retrieval approach can be found in Constantin et al. (2013). In both the morning and afternoon, the car followed a route departing from the FUB Institute for Space Sciences building towards the city centre and back. The route covered a large part of the major east–west-oriented plume. For the comparison, a VCD is extracted from the generated APEX NO2 maps for each co-located mobile measurement. Mobile observations are averaged in case of sampling of the same APEX pixel. The time series of the car DOAS VCDs along with the APEX VCDs are plotted at the respective car positions in Fig. 14a and b for the morning and afternoon flights, respectively.

The time series are in good agreement for both the morning and afternoon and exhibit largely the same NO2 distribution with low values close to FUB, located in the southwest of the city, and increased levels of NO2 closer to the city centre and downwind of the major plumes. Note that gaps in the APEX time series are related to parts of the route outside of the airborne acquisition area. The NO2 VCDs measured along the route during the morning by APEX and car DOAS are respectively 7.0 and 8.0×1015 molec cm-2 on average, and for the afternoon flight 7.3 and 8.5×1015 molec cm-2, respectively. The mobile measurements seem to be representative for the whole data set as the averages are close to the mean values for the full NO2 VCD distribution maps, being 7.4×1015 and 6.3×1015 molec cm-2 for morning and afternoon, respectively (see Table 6). In general, an overestimation of car DOAS VCDs or underestimation of APEX VCDs can be observed. This can also be observed in the scatter plots and linear regression analysis, provided in Fig. 15a and b for the morning and afternoon flights, respectively. The correlation coefficients are 0.86 and 0.96, respectively. For the afternoon flight, the slope and intercept are strongly affected by underestimation of the APEX VCDs between 13:30 and 14:30 LT.

The NO2 column at a certain geolocation is not sampled by both instruments at the same time and variability in local emissions and meteorology can lead to differences. The absolute time offset between car DOAS and airborne observations can be up to 2 h and is provided in the scatter plots as well. There is however not a clear difference in the spread for measurements with a small or large time offset, which can lead to the assumption that the NO2 field was relatively stable during the time of measurements.

Efforts were made to ensure the comparability of the correlative data sets, but nevertheless the scatter can be largely explained by sampling of different air masses due to the viewing geometry, differences in the sensitivity to NO2, observation time differences in combination with NO2 variability, and instrumental and algorithmic conceptual differences and related errors and uncertainties. Ongoing work is focusing on harmonisation of (1) retrieval settings for the car DOAS observations and (2) a priori input, e.g. the NO2 profile, aerosols, and other properties related to the radiative transfer.

The NO2 VCD maps, retrieved based on data from the different imagers, are quantitatively compared in this section. For the pixel-wise comparison, all NO2 maps were harmonised to ensure comparability and gridded to the same regular grid size of 0.0045∘, roughly corresponding to the spatial resolution of the whisk-broom SWING instrument (see Sect. 4.3). Regridding to a coarser spatial resolution also reduces the impact of fine-scale NO2 differences that can occur due to (1) different sensitivities to NO2, related to the instrumental characteristics, platform altitude, and retrieval algorithm; (2) the slightly different viewing geometries; (3) time differences in the observation of a dynamic NO2 field; and (4) imperfect georeferencing. In order to avoid averaging of measurements from adjacent flight lines, only the central half of the swath has been compared, i.e. for the APEX instrument with a FOV of 28∘, only the observations within ±7∘ off-nadir have been compared. For APEX, AirMAP, and SWING, this corresponds to a swath of roughly 1500 m. These data sets are compared with the full swath of SBI, which is 450 m due to its very narrow field of view of 8.3∘. The harmonised and intercompared NO2 VCD maps are shown in Fig. 16 for the morning flight.

In Fig. 17, the distribution of the slant errors from APEX and AirMAP retrievals is provided for the morning flight (upper panel). The slant error, σDSCDi, can be estimated from the fit residuals in the DOAS analysis, as indicated in Sect. 4.4. Structures that are correlated with the surface reflectance or the NO2 field cannot be observed. Some flight lines exhibit slightly larger slant errors, which is probably related to small instabilities in the spectral performance. For the APEX retrievals, slant errors are generally larger (mean slant error of 3.1×1015 molec cm-2) when compared to AirMAP (mean slant error of 2.1×1015 molec cm-2). The larger slant errors for APEX retrievals, as well as the larger variability, can be attributed to limitations related to the spectral performance of the APEX instrument, i.e. spectral resolution, sampling rate, and robustness of the slit function in operational conditions, as discussed extensively in Kuhlmann et al. (2016) and Tack et al. (2017). As most of the fit errors are absolute errors and do not scale with the NO2 signal, the distribution of the relative slant errors (relative to the retrieved slant columns) is provided in Fig. 17 (lower panel) as well. The relative slant error is on average 37 % and 24 % for APEX and AirMAP retrievals during the morning flight, respectively. For smaller NO2 abundances, e.g. upwind and south of the city centre, the relative error is largest. In the background area, the relative slant error is often very high in the case of the APEX observations and retrievals being close to the detection limit. The high retrieval uncertainty in these areas can result in the presence of slightly different structures in the retrieved NO2 VCD maps.

Time series of the pixel-wise VCD comparison are provided in Fig. 18, with APEX data in green, AirMAP in red, SWING in blue, and SBI in purple for (a) the morning flight and (b) the afternoon flight, respectively. All data sets have been compared to the AirMAP NO2 VCDs. Note that the time on the x axis corresponds with the UTC time recorded by the Cessna 207T D-EAFU, and thus it is the valid recording time for the AirMAP, SWING, and SBI instruments. The absolute time difference between overpasses from the two aircraft was 10 to 12 min on average for the morning and afternoon flights, with a maximum difference of 24 min. The corresponding scatter plots and orthogonal linear regression analyses are provided in Fig. 19. The color-coding of the lower plots indicates the absolute time offset between the observations from the two aircraft.

The time series exhibit strong NO2 peaks, which correspond to the crossings of the west–east-oriented main plume related to the Reuter West power plant. As mentioned in Sect. 2, the flight lines were flown perpendicular to it. The data gaps correspond to the roll movements of the aircraft in order to prepare the acquisition of the next flight line. An overall good agreement can be observed among all observations, for both low and high retrievals. Pearson correlation coefficients are close to or higher than 0.9 for the morning and afternoon flights, while the linear regression analyses show slopes close to unity and generally small intercepts. As expected, the best agreement is observed among the data sets collected from the same aircraft, as similar air masses were sampled. A very good fit is observed between AirMAP and SBI. This can be partly explained by the fact that the SWING and APEX retrievals contain more noise, mainly due to the instrument characteristics, resulting in a slightly larger spread.

We see a less favourable slope for the VCD comparison between AirMAP and SWING for the afternoon flight. In the case of low VCDs, a positive bias can be observed for the first flight lines and an opposite effect for the last flight lines. This is also visible in the NO2 VCD maps (Fig. 12): the west–east-oriented smooth increase in the background levels is less present in the SWING retrievals. All SWING retrieval parameters and results were carefully checked and we observed a possible polarisation dependency, which could impact the retrievals. SWING was initially designed to be operated from an UAV, which has repercussions on the size of the instrument. Although a quartz fiber was used, the straight fiber was only 5 cm long, which limited its efficiency at depolarising the incident light. Future manned aircraft missions with SWING are planned to be performed with a slightly adapted design, including a longer quartz fiber. As discussed earlier, SWING is a compact instrument without temperature stabilisation or tracking. A temperature dependence could be another possible cause, affecting the retrievals.

This study presents the first intercomparison of NO2 VCDs, retrieved from four different airborne imaging DOAS instruments. APEX performed flights for the retrieval and high-resolution mapping of NO2 VCDs for the first time over Switzerland (Popp et al., 2012) and Belgium (Tack et al., 2017), AirMAP over Germany (Schönhardt et al., 2015) and Romania (Meier et al., 2017), and SWING over Romania (Merlaud et al., 2018). After being tested individually during dedicated campaigns (except for SBI, which was deployed here for the first time), the experimental airborne imagers were operated simultaneously over the city of Berlin, in a unique but complex constellation, during the AROMAPEX 2016 campaign. In contrast to APEX and AirMAP, SWING and SBI are compact instruments initially designed to be operated from an UAV and 12-Unit CubeSat, respectively. APEX, AirMAP, and SWING have a comparable swath width of 3 km, while SBI has a swath of 450 m. The spatial resolution is better than 100 m for APEX, AirMAP, and SBI (push-broom scanning) and approximately 325 m for SWING (whisk-broom scanning).

The study demonstrates that the NO2 distribution over a large city region can be mapped accurately with high spatial resolution and in a relatively short time frame (typically a few hours). The observations allow us to differentiate local emission sources and reveal the fine-scale horizontal variability in tropospheric NO2 in an urban context, eventually contributing to an increased understanding of trace gas distributions and related chemical and dynamical processes in urban areas. For the morning flight (09:34–12:01 LT) on 21 April 2016, NO2 levels range between 1×1015 molec cm-2 upwind of the city and 20×1015 molec cm-2 within the dominant plume, with a mean of 7.3±1.8×1015 molec cm-2. The afternoon data set (14:24–16:39 LT) largely exhibits the same horizontal NO2 distribution. Although slightly higher peak values of up to 23×1015 molec cm-2 are observed, the mean VCD of 6.0±1.4×1015 molec cm-2 is lower when compared to the morning flight.

The NO2 VCD products of the four airborne imagers have been qualitatively and quantitatively compared. The data sets are consistent to a high degree after harmonisation of the parameter settings, AMF LUT, and gridding algorithm. Pearson correlation coefficients are higher than 0.9, while the linear regression analyses show slopes close to unity and generally small intercepts. This demonstrates the robustness of both the instruments and the applied retrieval approaches. Small discrepancies remain, however, due to a combination of (1) instrumental differences, e.g. SNR, spatial and spectral resolution, and temperature stabilisation; (2) observation differences, e.g. platform altitude, overpass time over a dynamic NO2 field, and viewing geometry; and (3) algorithmic differences, e.g. retrieval of surface reflectance product and DOAS fitting parameters.

The AROMAPEX study is seen as a preparatory step for forthcoming calibration and validation campaigns for the new generation of spaceborne air quality sensors, such as S-5P, S-4, and S-5. In less than 2.5 h, a (sub)urban area of approximately 23 km by 32 km was covered by the imagers, which is the equivalent of about 30 TROPOMI pixels (see Fig. 20). The AROMAPEX study assures a suite of reliable instruments that can be deployed separately from each other for future satellite validation. The high-resolution NO2 maps, generated from the airborne data, are unique data sets to study the satellite NO2 intra-pixel variability and to link between global and regional monitoring from space, local air quality models, and ground-based observations.