The passive DOAS technique uses measurements of scattered sunlight and the Lambert–Beer law to yield trace gas amounts and distributions in the atmosphere. While scattering causes smooth changes in the spectrum (e.g., λ-4 dependence for Rayleigh scattering), molecular absorption often has structured spectra. The total spectral attenuation is therefore split into a high-frequency part comprising the trace gas absorptions and a low-frequency part accounting for elastic scattering on molecules, aerosols, and clouds, as well as instrumental throughput. The latter part is described by a low-order polynomial. The effect of inelastic scattering known as the Ring effect , which is predominantly due to rotational Raman scattering leading to a filling in of Fraunhofer lines, is accounted for by a pseudo-cross-section σRing e.g.,. Similar spectral effects are caused by stray light inside the spectrometer when photons hit the detector at positions not corresponding to their wavelength. This is compensated for by applying another pseudo-cross-section σoff, for which often the inverse of the measured spectrum I is used. Further details about this so-called intensity offset correction and its similarity to spectral features produced by inelastic scattering can be found in . Lambert–Beer's law can then be expressed by the DOAS equation: τ=ln⁡I0I=∑iσi⋅SCi+σRing⋅SCRing+2+σoff⋅SCoff+∑papλp+r, where τ is the optical depth and the first sum is over all absorbers i having cross sections σi. The polynomial degree is p, and the residual term r contains the remaining (uncompensated for) optical depth, for example from measurement noise.

As measurements consist of spectra I and I0, Eq. () is defined at many wavelengths and solved in a linear least-squares fit returning the fit factors SCi and ap. While the polynomial coefficients ap are usually not used for further analysis, the so-called slant columns SCi=∫ρids are the integrated concentration ρi of absorber i along the light path s.

Recorded spectra contain almost no information about the altitude in which the absorption occurred. Thus, the sensitivity to different altitudes depends predominantly on measurement geometry. The measurement is more sensitive to tropospheric absorbers if the spectrum I is taken at small elevation angles above the horizon. This is due to the rather long light path through atmospheric layers close to the surface. On the other hand, the reference spectrum I0 is usually a zenith spectrum either measured at a small solar zenith angle (SZA) or taken close in time to the measured spectrum I (sequential), as for the zenith viewing geometry the light path through the atmosphere is short. The obtained SCi are therefore not absolute but the difference between measurement (I) and reference measurement (I0) and thus called differential slant column density (DSCD). As only DSCDs are used within this study, both terms are used synonymously in the following for simplicity. Furthermore, sequential reference fits are used throughout this study.

More details of the DOAS method can be found for example in and .

The IMPACT instrument, as deployed during the CINDI-2 field campaign (Sect. ), is shown in Fig. . It consists of a Czerny–Turner-type ANDOR Shamrock 303i imaging spectrometer equipped with a Newton DU940P-BU CCD camera with 2048pixels×512pixels covering a wavelength range from 394.5 to 536.4 nm. The CCD is cooled to -30 ∘C for reducing the dark signal (thermal electrons), while the spectrometer is actively temperature stabilized to +35 ∘C in order to avoid thermal (and therefore spectral) drifts. The spectrometer–CCD system is installed within a 19′′ rack that hosts at the same time all electronics and computers for instrumental control and operation. A 15 m long light fiber bundle consisting of 69 individual fibers (0.01 mm2 each) separates the indoor part (rack) from the telescope unit located outside. At both sides, the individual fibers are aligned vertically, i.e., stacked on top of each other (total height ∼9 mm), and sorted in a way that the uppermost fiber on the entrance side is also the uppermost fiber on the spectrometer side. However, as a result of the size of the CCD and the magnification characteristics of the spectrometer, light from the upper- and lowermost fibers does not hit the detector (these fibers are imaged outside the detector area), so that only 50 individual fibers fully mapped on the CCD are used. This is a nonoptimal setup as these fibers do not contribute to the used signal but enhance stray light inside the spectrometer. Although stray-light effects are compensated for by the intensity offset correction in the later DOAS fit (see Sect. ), light from these noncontributing fibers should be blocked in future applications to reduce potential problems with stray light.

In the telescope unit, light is collected and focused on the light fiber bundle with a commercial objective (1:1.4, focal length 8 mm). The instantaneous FOV of an individual fiber is determined by its dimension (active area) and the focal length of the objective and is about 0.8∘, both in the horizontal (azimuth) as well as in the vertical (elevation) direction. As the single fibers are stacked in the vertical dimension, the resulting hypothetical vertical FOV of the entire fiber bundle is ∼58∘, i.e., all 69 stacked single fibers. The part of the measurements used for the analysis yields a vertical FOV of ∼41∘ (50 individual fibers mapped on the CCD). The use of an objective instead of a single lens is necessary for overcoming spherical aberration and thus keeping the FOV constant for each individual fiber as the entrance slit has a considerable height (9 mm). This is different to the usual MAX-DOAS instruments where the light is focused on a very small spot-sized fiber entrance located on the optical axis, and therefore using a single lens is usually sufficient.

The vertical alignment of the sorted light fibers in combination with an imaging spectrometer – each fiber is mapped onto different CCD lines – allows taking measurements in multiple elevation angles simultaneously (see Sect.  for the calibration procedure of the elevation angle). Furthermore, the telescope hosts a visual camera taking snapshots for scene documentation with each measurement. The telescope unit is installed on an ENEO VPT-501 pan–tilt head, which allows pointing in any direction. However, as a result of the sufficiently large instantaneous vertical FOV, movements are performed in azimuthal direction only while the vertical tilt is kept constant (covering the elevation angles from -5 to +36∘) with the exception of zenith pointing for taking reference measurements.

Figure  shows an example image of the CCD for a typical off-axis measurement. The image quality (separation of single fibers) is best in the center of the CCD and blurred towards the edges. This is because the horizontal (spectrometric axis) and vertical (spatial axis) foci do not coincide everywhere in the focal plane (coincidence is optimized for the center of the CCD). The CCD can be placed in different positions, resulting either in good imaging or good spectrometric quality. Here, an intermediate flange was used placing the CCD in a position that is a compromise between imaging and spectroscopic performance. As a result, the slit function changes vertically across the detector from ≈1 nm FWHM (full width at half maximum) in the center of the CCD to ≈1.5 nm FWHM towards bottom and top rows. This was compensated for in the DOAS analysis by measuring and applying separate slit functions for different vertical binning ranges on the CCD associated with individual light fibers as defined in Sect. .

Ideally, an imaging instrument should be operated with a shutter or a frame transfer CCD in order to minimize the impact of illumination of the detector during readout. As the Newton DU940P-BU is not a frame transfer CCD and long-term operation of a shutter is limited by shutter lifetime, IMPACT measurements are taken without a shutter. As a result, the detector continues to be illuminated during the sequential CCD readout, leading to larger signals in those rows which are read out later. As the vertical position on the CCD corresponds to different elevation angles, this leads to a smearing of the CCD image and the corresponding viewing directions.

If illumination is assumed to be constant during measurements, a simple correction can be applied to the measured data. Starting from the very first line for which there is no smear effect, the original signal can be computed for each line successively by subtracting the additional illumination occurring during readout: 3Ij=Ijmeas-∑k=1j-1Ik⋅treadouttexposure, where Ij is the signal of row j without smear, Ijmeas is the intensity with smear, and treadout and texposure are the length of the duration of the readout of one line and the exposure time, respectively. While this correction works well in most cases, it can fail in situations where illumination changes rapidly, for example during measurements with broken clouds and high wind speeds.

Problems regarding the smear effect generally decrease with the ratio of exposure time to readout time because the relative contribution of illumination during readout then decreases. In other words, Ij approaches Ijmeas for treadout/texposure→0 (see Eq. ). To take advantage of this, an optical filter blocking parts of the sunlight was installed in the telescope unit. This allowed the increase of exposure times (typical IMPACT exposure times were then in the order of a few seconds) while avoiding saturation of the CCD. For every applied exposure time, dark images were recorded routinely and used to correct for dark current in the measurements prior to the DOAS analysis.

Data of the IUP-Bremen MAX-DOAS instrument is used to validate corresponding IMPACT measurements (see Sect. ). Both instruments were set up side by side (∼2 m distance; see Fig. ). The MAX-DOAS instrument consists of a telescope unit (located outdoors) and two CCD–spectrometer systems (located indoors) measuring in the UV and visible. For validation of IMPACT observations (measuring in the visible), only data collected by the visible spectrometer are used, which is an ACTON-500 covering a spectral range from 406 to 579 nm at a resolution of ≈0.85 nm. The spectrometer was actively temperature stabilized to +35 ∘C. A Princeton NTE/CCD 1340/100-EMB with 1340pixels×100pixels was used for recording spectra leading to a spectral sampling of 7–8 pixels nm-1. The CCD was cooled to -30 ∘C to reduce dark signal.

Light was collected by a telescope unit mounted (similar to IMPACT) on a commercial ENEO VPT-501 pan–tilt head allowing pointing in any viewing direction. The instrument's FOV (≈1.1∘) was determined by a lens focusing incoming light on an optical fiber bundle (length ≈20 m), which was Y-shaped and connected the telescope with both spectrometers. It consists of 2×38=76 single fibers. An in-telescope shutter and HgCd line lamp allow dark and wavelength-calibration measurements, which were routinely performed. A very similar instrumental set up has been used in previous campaigns, e.g., CINDI and TransBrom .

The Second Cabauw Intercomparison of Nitrogen Dioxide measuring Instruments (CINDI-2) field campaign was carried out at the Cabauw Experimental Site for Atmospheric Research (CESAR), close to the villages of Cabauw and Lopik, the Netherlands, from 25 August to 7 October 2016. It was a successor of the first CINDI campaign in 2009 . CINDI-2 aimed at characterizing the differences between measurement approaches and systems and to progress towards harmonization of settings and methods . One key activity was a semiblind intercomparison of participating DOAS-type instruments from different international research groups. This intensive phase was scheduled for the time period 12–25 September 2016.

The measurement test site is located in a semirural environment, i.e., without strong local sources (except for a regional traffic road in the south potentially causing enhanced NOx levels during rush hour) but within the polluted region between Amsterdam, Rotterdam, and Utrecht.

In total, 23 groups and 31 DOAS-type instruments participated in CINDI-2. The instruments were mainly deployed at two container decks. At the lower level, 1-D MAX-DOAS instruments were pointing permanently in a common azimuth direction of 287∘ (clockwise from north) and performed vertical scanning sequences in this azimuth. Two-dimensional MAX-DOAS systems installed at the upper container deck (see Fig. ) providing a free view around the measurement site were following a rather complex measurement protocol prescribing the observation geometry on a 1 min time base. However, for comparison with 1-D instruments, a vertical scanning sequence was performed in the common azimuthal direction every hour.

The IMPACT instrument fulfilled two purposes during CINDI-2:

To participate in the semiblind intercomparison. For this reason, measurements were performed in the common azimuth direction of 287∘ every hour for 15 min, together with the 1-D and 2-D instruments.

In addition to the observation geometry, DOAS fit settings were also prescribed for the CINDI-2 semiblind intercomparison (Table ). These fit parameters have been used as well for the analysis of NO2 and O4 distributions within this study.

Figure  shows NO2 DSCDs from an example MAX-DOAS vertical scanning sequence on 23 September 2016 under good weather and viewing conditions in comparison to IMPACT results. Note that, due to instrumental restrictions, the elevation angles of IMPACT deviate slightly from the angles prescribed for the semiblind intercomparison, while the MAX-DOAS instrument follows exactly the prescribed angles. As a result, the column for the 1∘ MAX-DOAS elevation (blue triangle) should be slightly larger than the IMPACT slant column (blue circles) taken at the same time because the effective elevation of IMPACT is 1.4∘. Interestingly, this is not seen here (the NO2 slant columns of both instruments agree quite well). The reason might be small misalignments between both instruments, either in elevation or azimuth, or the NO2 profile shape (potentially in combination with differences in the FOV of both instruments).

Figure  demonstrates a striking advantage of imaging DOAS as measured NO2 slant columns reveal a short-term temporal variation, which is resolved by IMPACT but not by the MAX-DOAS instrument. As mentioned, the 1∘ MAX-DOAS observation matches the IMPACT observation taken at the same time, but then MAX-DOAS continues with the next elevation (2∘) while IMPACT repeats measurements of the complete elevation angle range. In the case of 1∘ (1.4∘) elevation, the NO2 slant columns change from ∼1.75×1017 to ∼1.40×1017 molec cm-2, which is about 20 %. This temporal variation is not captured by the MAX-DOAS instrument, with clear consequences for any profile retrieval on these data which assumes that measurements at different elevation angles probe the same atmosphere. This is further investigated in Sect. .

Figure  shows a correlation plot between MAX-DOAS and IMPACT NO2 slant columns for several days within the semiblind intercomparison phase. For each MAX-DOAS elevation angle (color-coded) the closest IMPACT vertical scan (measured simultaneously) was selected. As a quality criterion, data were rejected if no IMPACT scan was found ±2 min around the MAX-DOAS measurement time (e.g., due to instrumental failures or saturated data). In addition, NO2 slant columns from IMPACT's simultaneous elevations were interpolated to the MAX-DOAS elevation angle.

Statistical values for the correlation plot are summarized in Table . In general, an excellent agreement is found with correlation coefficients of ≈98 % for the 30∘ elevation angle and even >99 % for elevation angles ≤15∘. The slope is close to 1 (within 8 %) and the offset is <1×1015 molec cm-2 with the exception of the 2∘ elevation, for which it is slightly larger.

In general, these intercomparison results agree well with the much more detailed (and official) intercomparison study from comprising all instruments participating at CINDI-2, although values are not identical. However, this is expected as the considered time periods are different. In addition, the comparison here is between IMPACT and a single MAX-DOAS instrument only, while in the official intercomparison exercise performed by a reference data set consisting of several instruments is used.

Figure  shows the campaign average of NO2 slant columns observed from IMPACT in all azimuths and elevation angles around the measurement site (note that due to instrument problems the entire semiblind intercomparison period is not captured here, but instead only 16–24 September 2016 is captured). For better visibility, the five lowermost CCD bins (corresponding to single fibers) pointing towards the ground have been removed as well as two CCD bins pointing effectively in almost the same direction as a result of the twisted fibers discussed in Sect. . Consequently, the panoramic view in Fig.  consists of 43 elevation angles on the vertical axis and 36 azimuth directions (-175 to +175∘ in 10∘ steps) on the horizontal axis. In addition, the fractional IMPACT elevation angles on the vertical axis have been rounded for better readability. We note that this has been done in subsequent figures and in the following discussion as well.

Obviously, the campaign mean NO2 distribution around the measurement site is rather homogeneous with a slight tendency to larger values in the southwest (between -165 and -75∘), which is most likely linked to a close-by local traffic road (in this azimuthal regime, the light path is almost along the road, which can be seen in Fig. ). Furthermore, the light path was obstructed by trees in ≈75 to 135∘ azimuth and elevation angles <5∘ which can be clearly seen by reduced NO2 slant columns in these directions – i.e., these small values are an effect of obstacles and the resulting short light path. In addition, obstruction by other instruments occurred in -25∘ and by a single tree in -115∘. In general, largest NO2 slant columns are found not in 0 or 1∘ but in ≈2∘ elevation, which is an effect of the instrument's FOV; i.e., surface effects are present in the 0∘ and (to a lesser extent) in 1∘ elevation angle as a result of the overlap of adjacent fibers mapped onto the CCD (see Sect.  and Fig. ).

The homogeneous long-term-averaged NO2 distribution around the measurement site is supporting the assumption of the absence of persisting strong local pollutants. However, much more variability is present on shorter timescales. This is demonstrated by Fig. a where the range of NO2 slant columns recorded on 20 September 2016 (maximum and minimum values) as well as the average of all applied azimuths in 4∘ elevation angle is shown (one data point for each panoramic image). Maximum values differ from the azimuthal mean by up to a factor of 2. This is quantitatively analyzed for the whole campaign in Fig. b showing the maximum relative difference, i.e., the ratio between maximum NO2 observed in any azimuth to the NO2 averaged over all azimuths. The maximum relative differences range from 10 % to 120 % for individual panoramic views and are ≈35 % on average. This is an unexpectedly high value indicating large spatial inhomogeneity on short timescales even for semirural measurement sites like Cabauw with no large local sources and very homogeneous long-term trace gas distributions. As a result, care has to be taken if ground-based (MAX-DOAS) measurements are used for satellite validation as a single viewing direction does not necessarily provide a good estimate of the NO2 columns within a satellite pixel. In this case, observations in many azimuths should be taken and averaged to reduce the variability present in satellite ground pixels. This is often done when validating satellite observations in urban areas where spatial gradients are expected; e.g., a validation of Ozone Monitoring Instrument (OMI) satellite pixels in an urban, polluted area taking into account not only the azimuthal inhomogeneity around the measurement site but also changes in the NO2 concentration along the light path (using 3-D DOAS) was presented by . The findings derived from IMPACT measurements suggest that similar efforts are necessary when validating satellite results even in semirural locations like Cabauw.

One reason for the observed spatial inhomogeneity of NO2 is the transport and passing of polluted air masses. Figure  shows the temporal evolution of NO2 slant columns in all applied azimuth directions (vertical axis) and 4∘ elevation angle on 20 September 2016. The data gap around 14:00 UTC is due to an instrumental failure. Besides moderately enhanced NO2 towards the evening, a clear transport event occurred around 10:00 UTC. Between 09:00 and 10:00 UTC, increased NO2 slant columns appear in all azimuth directions between 25 and ∼175∘ (south). Between 10:00 and 11:00 UTC, the maximum of NO2 is then traveling from an azimuth angle of β1≈30∘ to β2≈-70∘ (see geometrical considerations in Fig. ).

The wind direction on 20 September 2016 was quite variable with low absolute wind speeds. However, the mean wind direction was ≈75∘ (see Table  for meteorological conditions). If the plume is transported by the wind, the direction of smallest distance r to the measurement site is α≈-15∘ (see Fig. ). The assumption here is a straight trajectory s (blue arrow) of the plume and thus the smallest distance r (dashed line) to the measurement site is perpendicular to it. As can be seen in Fig. , this coincides roughly with the direction of the largest NO2, although slant columns are not necessarily largest at the smallest distance r as the magnitude depends also on the (unknown) plume's shape and relative contribution of the light path through it.

The spatial distance traveled in Δt=1 h (10:00 to 11:00 UTC) can be estimated from wind speed: 5s=vwind⋅Δt. The angles between r and the trajectory's start/end points (i.e., plumes's positions at 10:00 and 11:00 UTC) are γ1=|β1|+|α| and γ2=|β2|-|α| (Fig. ). The distances s1, s2 and s are then given by (omitting the sign of γ1 and γ2) 6s1=r⋅tan⁡(γ1),7s2=r⋅tan⁡(γ2),8s=s1+s2=r(tan⁡(γ1)+tan⁡(γ2)). As a result, the smallest distance r to the measurement site is 9r=vwind⋅Δttan⁡(γ1)+tan⁡(γ2). Note that this calculation is in principle true for the 0∘ elevation angle only, whereas measurements in 4∘ were used here. However, this was neglected for simplicity as the effect is small and below the uncertainty introduced by the variety of assumptions made. For a mean wind speed of 1.2 m s-1 measured at the Cabauw meteorological tower, a smallest distance of r≈1.8 km is obtained (s≈4.3, s1≈1.8, s2≈2.5 km).

Figure  shows the measurement site's surrounding with the smallest distance r and plume's trajectory between 10:00 and 11:00 UTC indicated as blue arrow. Obviously, the origin of the transport event cannot be precisely identified, but it could be linked to a regional industrial park that is close to the starting point of the plume's trajectory. This speculation is supported by the fact that increased values of NO2 are already found slightly earlier (≈ 09:30 UTC) in northeastern directions (see Fig. ). In addition, increased NO2 slant columns are seen in the zenith direction as well (not shown). This indicates that parts of the plume were overpassing the measurement site and thus a large spatial extent of the plume perpendicular to the direction of propagation, most likely as a result of the unstable wind direction. Finally, the fact that the 4∘ elevation angle is clearly enhanced although the plume was overpassing the instrument as well means that the plume is close to the ground, which is usually an indication for a close-by origin. This is supported by vertical NO2 profiles retrieved in Sect. .

As already mentioned, one of IMPACT's objectives is to enable aerosol and trace gas profile retrievals rapidly in every direction around the measurement site.

In addition to NO2, IMPACT measurements enable the oxygen dimer O4 to be retrieved from the same DOAS fit (Table ). As O4 is a collision complex of O2 molecules, it depends on pressure only and is therefore a measure of the light path e.g.,and references therein.

As a case study, Fig.  shows the measured intensity (a) and O4 slant columns (b) from one IMPACT scan (acquisition time ∼15 min) on 24 September 2016 under excellent viewing conditions. The position of the sun is clearly visible at ∼125∘ azimuth (solar azimuth angle, SAA) and ∼25∘ elevation. O4 slant columns close to the sun are reduced as a result of shorter average light paths due to strong forward scattering of aerosols. This is validated by simulated O4 slant columns for the same measurement geometry without aerosols, i.e., pure Rayleigh scattering (c) and with aerosols (d). The simulations have been performed using the radiative transfer model SCIATRAN in its version 3.4.4. As input for the simulation, an exponential decrease (0.1 km-1 surface value, AOD=0.2) was used as the aerosol extinction profile, and a Henyey–Greenstein (HG) parametrization of the aerosol phase function with an asymmetry factor of g=0.75 and a single-scattering albedo SSA=0.95 was applied. These values were obtained from a close-by Cabauw AERONET station ().

Simulated O4 slant columns for pure Rayleigh scattering differ strongly from measured O4 columns, both in absolute values and in the azimuthal distribution. In particular, the largely reduced columns around the sun are not reproduced by the simulation showing slightly reduced columns at the SAA and SAA + 180∘ only as a result of the Rayleigh phase function. In contrast, simulated columns including aerosols agree much better with measured columns and cover the azimuthal distribution (Fig. d). Thus, the comparison between the simulated and measured azimuthal distribution of O4 columns can be used to retrieve information about the aerosol properties and in particular its phase function.

Retrievals of aerosol properties, e.g., from AERONET stations, are usually based on intensity measurements in the solar almucantar, which in this case is the azimuthal distribution in ≈25∘ elevation. With MAX-DOAS it is also possible to incorporate O4 measurements in the retrieval of aerosol microphysical properties and phase function as suggested by . demonstrated a corresponding retrieval based on intensity and O4 measurements in different azimuths and found that the largest sensitivity is gained from measurements in the aureole region of the sun, therefore requiring a small FOV, protection against direct sunlight and the capability to perform automated measurements in the azimuth. While measurements very close to the sun are challenging for IMPACT due to its large FOV, two important aspects can be investigated as a result of IMPACT's capability to record full 2-D maps very rapidly around the measurement site:

Is there a potential for O4 measurements in almucantars different than the solar almucantar to contribute to/support aerosol retrievals?

For research question (1), it is a limitation that sky radiometers (e.g., within the AERONET network) and current state-of-the-art MAX-DOAS instruments are measuring in only one viewing geometry at a time. A scan along the solar almucantar then provides observations at different scattering angles. In contrast to these instruments, IMPACT measures many almucantars at the same time, in the case study shown in Fig. , both above and below the solar almucantar. The geometrical scattering angle (single-scattering case) has been calculated for every viewing geometry and is plotted in Fig. c. Obviously, almucantars above and below the solar almucantar provide slightly different scattering angles and might therefore complement the classical retrieval.

However, not all almucantars should be used, and, even if exploiting the solar almucantar only, a threshold for the lowest usable elevation angle should be regarded (research question 2). The reason is that a retrieval of, for example, the aerosol phase function requires the azimuthal distribution of measured O4 to be caused by the aerosol phase function only. In contrast, in the observations it is caused by the combined effect of (1) phase function and (2) varying aerosol load and extinction profile in different azimuth directions as well as along the light path, i.e., in different distances from the instrument. For measurements taken at large elevations, the aerosol load and profile can be assumed to be homogeneous as the horizontal distance around the measurement site from which information is obtained (in a single-scattering case this is the distance to the scattering point projected to the ground) is short. For small elevations, this horizontal extent around the measurement site is much larger – in a first approximation it is scaling with 1/tan(elevation), if only averaging in the boundary layer is considered and the last scattering point is above the boundary layer height. Thus, for small elevations the aerosol load and profile can change substantially along the light path.

This effect is clearly present in Fig. b: measured O4 slant columns have a distinct maximum in small elevations centered around ≈-25∘ azimuth (ranging from ≈-60∘ to 25∘ azimuth), which is not reproduced by simulated O4 columns. As illustrated in Fig. c, this is not the location of largest scattering angles (occurring at ≈-55∘ azimuth only) and therefore not related to the O4 maximum expected in backscattering direction (due to preferred forward scattering and consequently larger light paths in backscattering direction). Furthermore, if the O4 maximum was an effect of the phase function, a second maximum would appear close to the ground at ≈-85∘ azimuth (given that the aerosol profile would not change with the azimuth), because scattering angles in -25 and -85∘ azimuth are identical (see Fig. c). Obviously, no second O4 maximum is present at -85∘ azimuth, indicating that the aerosol load seen in small elevation angles changes with the viewing azimuth. In particular, the observed maximum in O4 slant columns at -25∘ azimuth indicates smaller aerosol loads close to the ground (longer light paths) in this direction. As a result, almucantar scans in small elevation angles should not be used to retrieve aerosol information.

In order to quantify this finding, Fig. a shows two specific azimuthal distributions of measured (solid) and simulated (dashed) O4, i.e., two horizontal cross sections of Fig. b and c, for elevation angles of 4 and 25∘ (solar almucantar), respectively. While the agreement between measurement and simulation is very good in 25∘ elevation, differences in 4∘ are much larger, both in absolute values and in shape. Figure b shows the same data but plotted as a function of scattering angle. The solid line represents scattering angles counterclockwise from the position of the sun (SAA = 125∘) and the dashed line clockwise. For the solar almucantar, both lines agree quite well with each other as well as with the simulation (green line), indicating that the aerosol seen in 25∘ elevation is rather homogeneous around the measurement site and aerosol parameters used in the simulation are realistic. In contrast, the 4∘ almucantar does not match the simulation and – more importantly – O4 columns observed clockwise from the incoming direction show severe differences and another shape than those recorded counterclockwise. This cannot be explained with the aerosol phase function, which is symmetrical. This supports the conclusion that inhomogeneous aerosol content around the measurement site is seen in 4∘ elevation, i.e., close to the ground. This is furthermore supported by aerosol extinction profiles retrieved with BOREAS (Fig. ) showing smaller values close to the ground between -50 and 25∘ azimuth. However, the BOREAS aerosol retrieval for this day is challenging due to the relatively small absolute aerosol load (AOD ≈ 0.2), and consequently Fig.  should not be overinterpreted (the general patterns appear to be reliable, but individual values should be regarded with care).

To elaborate a threshold of usable almucantars and to test their potential for aerosol retrievals, various SCIATRAN simulations have been performed based on different aerosol parameters. For each set of parameters, resulting correlation coefficients between measured and simulated O4 azimuthal distributions are shown in Fig. d as a function of elevation angle. Aerosol parameters leading to largest correlations are then compared to independently measured quantities from the AERONET station.

The blue curve in Fig. d corresponds to the original simulation shown in the previous plots using g=0.75 and SSA=0.95. For small elevations, correlation coefficients increase rapidly. This is due to a combination of the observed obstruction by trees discussed above and true inhomogeneities of the O4 azimuthal variation. The steep increase is followed by a much shallower increase until a plateau is reached at ≈10∘. For very large elevations >30∘, correlation coefficients decrease slightly, most likely as an effect of smaller O4 columns and thus poorer statistics.

It is found that changes in the SSA (red line) lead to almost the same results; i.e., the pure analysis of the shape of O4 columns at a specific almucantar is (not surprisingly) insensitive to the SSA.

The green and the magenta line were performed with the same SSA as the original simulation but larger asymmetry factors g. Resulting correlation coefficients are clearly smaller.

To conclude, the variation of O4 columns along almucantars contains information about the asymmetry factor g. As can be seen from Fig. d, the value of g=0.75 measured by the close-by AERONET station leads to the largest correlation coefficients. However, it should be mentioned that simulations using smaller asymmetry factors (not plotted) show a similar performance unless g reaches very small values (g<0.5). Consequently, the simple approach of using correlation coefficients as performed here is not a sufficient way to determine g with good precision. However, the potential of using O4 (ideally together with intensity) in more sophisticated retrievals appears to be promising.

For the two initial research questions the following can be concluded:

In general, different almucantars recorded simultaneously by IMPACT have slightly different scattering angles, meaning that the information content they provide is not redundant. Consequently, these almucantars have a potential to be used in future retrievals of the aerosol phase function. In particular, use of almucantar O4 columns turned out to contain information about the asymmetry factor g but to be insensitive to the SSA.