A new measurement approach for validating satellite-based above cloud aerosol optical depth

The retrieval of aerosol parameters from passive satellite instruments in cloudy scenes is challenging partly because clouds and cloud-related processes may significantly modify aerosol optical depth (AOD) and particle size, a problem that is further compounded by the 3D radiative processes. Recent advances in retrieval algorithms such as the “color ratio” method which utilizes the measurements at a shorter (470 nm) and a longer (860 nm) wavelength have demonstrated the simultaneous derivation of AOD and cloud optical depth (COD) for scenes where absorbing aerosols are found to overlay low-level cloud 15 decks. This study shows simultaneous retrievals of above-cloud aerosol optical depth (ACAOD) and aerosol-corrected cloud optical depth (COD) from airborne measurements of cloud-reflected and sky radiances using the color ratio method. These airborne measurements were taken over marine stratocumulus clouds with NASA’s Cloud Absorption Radiometer (CAR) during SAFARI 2000 field campaign offshore of Namibia. The ACAOD is partitioned between the AOD below aircraft (AOD_cloudtop) and above aircraft AOD (AOD_sky). The results show good agreement between AOD_sky and 20 sunphotometer measurements of the above aircraft AOD. The results also show that the use of aircraft-based sunphotometer measurements to validate satellite retrievals of the ACAOD is complicated by the lack of information on AOD below aircraft. Specifically, the CAR-retrieved AOD_cloudtop captures this “missing” aerosol layer caught between the aircraft and cloud top, which is required to quantify above cloud aerosol loading and effectively validate satellite retrievals. In addition, the study finds a strong anticorrelation between the AOD_cloudtop and COD for cases where COD <10 and a weaker 25 anticorrelation for COD >10, which may be associated with the uncertainties in the color ratio method at lower AODs and CODs. The influence of 3D radiative effects on the retrievals is examined and the results show that at cloud troughs, 3D effects increase retrieved ACAOD by about 3-10% and retrieved COD by about 25%. The results show that the color ratio method has little sensitivity to 3D effects at overcast stratocumulus cloud decks. These results demonstrate a novel airborne measurement approach for assessing satellite retrievals of aerosols above clouds, thereby filling a major gap that exists in the 30 global aerosol observations. https://doi.org/10.5194/amt-2020-246 Preprint. Discussion started: 21 July 2020 c © Author(s) 2020. CC BY 4.0 License.


Introduction
The uncertainties of aerosols measurements in the vicinity of clouds has implication for the direct shortwave radiative aerosol effect and forcing on the climate system. Also, aerosols are known to exert an indirect forcing on climate by altering cloud properties and precipitation. According to the last Assessment Report of the Intergovernmental Panel on Climate Change 35 (Boucher et al., 2013), the interactions between clouds and aerosols remain among the largest sources of uncertainty, pointing to a lack of good understanding of the aerosol-cloud system, and holding back progress in the enhancement of Earth system predictions/projections.
Space-based retrievals of aerosol optical properties in the vicinity of clouds is complex because of the difficulty in distinguishing the contributions from aerosols and clouds in top of the atmosphere (TOA) reflectance measurements. However, 40 in the last two decades, several studies have demonstrated new approaches for aerosol retrievals in the vicinity of clouds. The absorbing aerosols such as smoke plumes, desert dust, and volcanic ash have been monitored from satellite observations in the presence of clouds using the ultraviolet measurements of Total Ozone Mapping Spectrometer (TOMS)/ Nimbus 7 (Herman et al., 1997;Torres et al., 1998), Ozone Monitoring Instrument (OMI)/Aura (Torres et al., 2012), and the Scanning Imaging Absorption Spectrometer for Atmospheric Chartography (SCIAMACHY) (De Graaf et al. 2007). The near-UV retrieval 45 approach was extended to the visible and near-infrared spectral regions for the simultaneous derivation of aerosol optical depth and cloud optical depth based on Moderate Resolution Imaging Spectroradiometer (MODIS) measurements in regions where light-absorbing carbonaceous and dust aerosols overlay low-level clouds (c.f. Jethva et al. 2013, Sayer et al. 2016. Similarly, Waquet et al (2009) developed a method based on multiangle polarization measurements at visible and near-infrared wavelengths to retrieve aerosol properties over clouds and successfully applied it to measurements of the Polarization and 50 range of azimuthal directions as the aircraft flew in a circular flight track (c.f. Gatebe et al. 2003;Fig. 3). The quicklook RGB image in Figure 3 (R=1.04 µm, G=0.87 µm, and B=0.47 µm) illustrates measurements taken from 12:27 UTC to 12:54 UTC.
The sun can be seen in the sky at about 33° view zenith angle, which also corresponds to the solar zenith angle, and a bright cloud system is seen on the image from view zenith angles 90-180°. The horizon coincides with the 90° view zenith angle, which is easily identified by the contrast between the sky and surface. In this image, the principal plane is defined by the 100 vertical plane containing the sun and the plane that is equidistant between two solar disks.
Note that the circular flight track during the BRDF measurements above the clouds (~650 m) is about 4 km in diameter, and with an aircraft bank angle of 20-30°, which is compensated by CAR to help maintain the full 180° view from zenith to nadir, the plane took ~3 minutes to complete an orbit. The marine stratiform clouds are generally characterized by a well-defined cloud top height corresponding to a strong boundary layer inversion. Given this viewing geometry of the cloud-105 aerosol system, the CAR measurements permit the retrieval of aerosol optical properties above clouds separated into above and below the aircraft, plus the cloud optical properties, using the color ratio method. These measurements provide the best data for validating above cloud aerosol retrieved from satellite measurements, analogous to the validation of cloud-free aerosol retrievals from satellites, which is typically done with observations from the AErosol RObotic NETwork (AERONET) groundbased sunphotometer network (Holben et al. 1998). 110

The color ratio method and its application to airborne observations
The color ratio (CR) method has been used to retrieve simultaneously the above-cloud aerosol optical depth (ACAOD) and aerosol-corrected cloud optical depth (COD) from OMI (Torres et al. 2012) and MODIS observations (Jethva et al. 2013;. The technique is physically based on the reduction of the ultraviolet (UV), visible (VIS), and near-infrared (NIR) radiation reaching the top of atmosphere, due to particle absorption above cloud. The effects of aerosol absorption have a 115 spectral signature, in which the absorption strength is found to be stronger at shorter wavelengths than at longer wavelengths.
This produces a strong color effect in spectral measurements, hence, the name color ratio method. The near-UV based color ratio algorithm has been applied to the long-term record of OMI to derive a global product of ACAOD (Jethva et al., 2018).
The ACAOD product has been validated against airborne measurements taken from HSRL-2 lidar operated during the ORACLES campaign conducted over the south eastern Atlantic Ocean. On the other hand, the ACAOD derived from the 120 visible/near-IR observations of MODIS was validated against the direct AOD measurements acquired from airborne AATS and 4STAR sunphotometers operated during different field campaigns (Jethva et al. 2016). In both OMI and MODIS validation studies, the satellite-retrieved ACAOD product was found to agree well with the airborne measurements within the expected uncertainty limits associated with the inversion technique, which mainly arises from the chosen aerosol model and its absorption properties. 125 Here, the CR method was applied to CAR observations, which include direct and diffuse solar radiance (or sky radiance), at nine spectral channels (see Fig. 4.). The direct solar component is given by the extra-terrestrial solar radiance attenuated by atmospheric absorption and scattering. On the other hand, sky radiance results from single and multiple scattering https://doi.org/10.5194/amt-2020-246 Preprint. Discussion started: 21 July 2020 c Author(s) 2020. CC BY 4.0 License. processes due to interaction of sunlight with aerosols and gas molecules. Atmospheric gas molecules (e.g. nitrogen, oxygen, carbon dioxide, ozone, water vapor, etc.) and aerosols are likely to strongly affect the solar radiance in the visible and near 130 infrared regions. The attenuation (scattering and/or absorption) by each atmospheric constituent is strongly dependent on wavelength and can be determined through the optical thickness using simple parametric models (e.g. Zibordi and Voss, 1989).
In the case of CAR measurements close to the sun (solar aureole), the signal from the direct solar radiance measurements saturate the detectors and therefore pixels that are especially close to the solar direction (scattering angles are ≤10°) should be excluded from any retrieval (Gatebe et al. 2010). The sky radiance distribution seen here is typical of clear skies (cloud free), 135 where the radiance of a point in the sky depends both on its position relative to the sun (i.e., azimuth angle) and on its airmass number (i.e., zenith angle). The sky radiance distribution is generally symmetrical about the principal plane, where the maximum value of the sky radiance for each wavelength is observed. This is illustrated in Fig. 4e: 45° » 315°; 90° » 270°; 135° » 225° for l >0.4 µm. The minimum values of sky radiance are found to be in the area directly opposite to the sun's position. 140 The CAR observations are indicative of the presence of absorbing aerosols above the clouds due to apparent brightening/darkening, which is evident when looking at the measured sky radiances /cloud bidirectional reflectance factor (BRF) (cf. Fig. 4). Aerosol loading has a strong influence, especially in the forward scattering directions (relative azimuth angle (φ) < 90° and φ>270°), with reflectances in the shorter wavelengths (e.g. 0.38 µm) larger by a factor of >3 relative to the longer wavelengths (e.g. 1.22 µm). This asymmetry is largely attributed to aerosol scattering and not to Rayleigh scattering, 145 as the latter is expected to exhibit symmetrical distribution in either scattering directions. More interestingly, there seems to be a strong aerosol absorption signal above clouds. It is well known that clouds reflect uniformly across the visible-near-IR spectrum, however, the presence of absorbing aerosols above clouds (in this case smoke transported from southwestern Africa) induces an overall absorption or darkening in the UV and shorter visible wavelengths, thus resulting in a strong reflectance gradient from UV to blue to near-IR spectrum, ~35% reduced reflectance at 0.34 µm compared to that at 1.04 µm, as seen in 150 Fig. 4f. Overall, the positive spectral gradient seen in Fig. 4f, is normally associated with cloud darkening at the shorter wavelengths (cf. Gautam et al. 2016).

The observations
Figures 5 and 6 show the full BRF of low stratiform clouds at selected wavelengths of 0.472 µm and 0.870 µm, respectively, 155 from each of the 16 different cases described in Section 2. The two wavelengths form the basis of the "color ratio" method for the simultaneous retrieval of above-cloud aerosol optical depth (ACAOD) and cloud optical depth (COD). The spectral BRF of stratiform clouds observed in the 16 cases is highly anisotropic due to a combination of factors ranging from cloud heterogeneity (including sub-pixel heterogeneity), solar illumination geometry, sensor viewing geometry, and cloud parameters such as optical thickness and effective radius (cf. Cornet et al. 2018). The 16 cases have a range of solar zenith angles (23°< SZA <34°). Measurements span an area of ~55 km (N-S) x ~12 km (E-W), with most cases (9 cases: cases h-p) concentrated over a much smaller area (~8 km x ~4 km) (cf. Fig. 1). The observations were taken at approximately the same altitude (Table 1:  rainbow), which is typical of water clouds (cf. Gatebe et al. 2003, where case h was analysed in detail). Figure 7 shows the derived spectral albedo (with atmosphere) for all the 16 cases at l=0.470 µm and l= 0.870 µm (see Table 2 for the spectral albedo (with atmosphere) for all the wavelengths). Clearly, Group 3 cases had higher spectral albedo and was optically thicker, while Group 2 cases from near the Namibian coastline had the lowest spectral albedo (with atmosphere). It is interesting to note that the spectral albedo remains almost constant in Group 2 cases, despite the change in measurement scale during the 180 spiral.
In the following subsections, we will examine how the surface reflectance anisotropy impacts the retrievals of the optical depth (both clouds and aerosols) using the color ratio method. Figure 8 shows the retrieved AOD for aerosol layers located above the aircraft-level (AOD_sky)) derived from the 185 observed diffuse sky radiance by CAR. The retrievals were performed using a single-channel fit at 470 nm between the observed sky radiance aerosol look-up table accounting for the variations in AOD and geometry. Note that the aerosol model used for AOD_sky retrievals was the same for the inversion of AOD below aircraft (AOD_cloudtop). It is complicated to characterize and model the anisotropic effects of reflecting clouds with varying optical depths on the hemispherical diffuse sky radiances measured by CAR. Therefore, we adopted a simple approach to account for these effects, at least partially, by 190 retrieving AOD above the aircraft assuming an averaged underneath cloud optical depth field retrieved from the https://doi.org/10.5194/amt-2020-246 Preprint. Discussion started: 21 July 2020 c Author(s) 2020. CC BY 4.0 License.

The retrieved ACAOD and COD
AOD_cloudtop inversion for each CAR BRDF case. For the most part the hemispherical distribution of retrieved AOD_sky along the azimuth direction is found to be smooth and near-uniform suggesting that the sky retrievals of AOD aren't significantly affected by the cloud anisotropy and that the simple approach of assuming an averaged value of COD for the full azimuthal scan works reasonably well in capturing the cloud effects on the sky radiances. The angular pattern in cases a-d is 195 similar and in good agreement with the airborne direct sunphotometer measurements as discussed later ( Figure 12 and Table   1).
The retrieved AOD below the aircraft (AOD_cloudtop) for all the 16 CAR BRDF cases are shown in Figure  Another important observation in Fig. 9 is the increasing magnitudes of AOD above cloud for the cases e, f, and g. Table  215 sea level. It is expected that as the aircraft altitudes moves higher in the atmosphere, the CAR sensor would see an aerosol layer of greater geometrical thickness, thereby resulting in greater aerosol extinction and AOD. The retrieved AOD_cloudtop for these cases precisely demonstrates this effect by showing increasing magnitudes for higher aircraft altitudes.
The color ratio algorithm, along with the above-cloud AOD, also co-retrieves aerosol-corrected cloud optical depth, 220 which is shown in Figure 10. Unlike aerosol fields, both seen above and below the aircraft level show more homogeneous distributions, the cloud optical depth fields retrieved from most of the cases show a great deal of variability along the azimuthal plane. Except for the cases m, n, o, and p, all other cases (a through l) show overall higher cloud optical depth in the back scattering directions shown in the bottom hemisphere opposite to the Sun and between the azimuth angle 90° and 270°. Unlike polar orbiting satellite observations at a fixed geometry for a given overpass, the CAR measurements offer a complete picture 225 over all the viewing directions relative to the sun direction. This unique observational geometry provides increased information content that would allow quantification of the effects of angular reflectance distribution in remote sensing retrieval algorithms. Figure 11 shows the scatter plots of AOD_cloudtop vs COD for view zenith angles 0°-30° (blue color), 30°-60° (green color), and 60°-90° (red color), which shows very interesting patterns. The retrievals of AOD_cloudtop are found to exhibit a 230 systematic dependence on COD (similar to an exponential decay function), especially the blue color and green color dots,

The relationship between AOD_cloudtop and COD
where larger values of AOD_cloudtop correspond to lower values of COD and crawling along the x-axis on the right as COD increases. An exception to this rule are the retrievals made at the higher view zenith angles, 60°-90° (red color), where the retrieved ACAOD remains low (<0.2) despite an increase in the COD, which seems unrealistic and confirms some of the limitations of the color ratio method. Another exception is seen in cases e, f and g, where AOD_cloudtop vs COD show no 235 clear dependence on viewing zenith angle and COD was around 5, indicating that these observations belonged to either clearsky or partially cloudy sky or thin heterogeneous scenes, for which the assumption of fully overcast thick homogeneous pixels made in the CR algorithm breaks down. The relationship betweent the two retrieved quantities appears to be confined for COD<10, after which both retrievals are found to be not correlated with each other. Such observed dependence was expected as noticed in the color ratio algorithm introduced in Jethva et al. (2013). The uncertainties in satellite ACAOD inversion is 240 known to be larger at lower CODs. This is because the retrieval domain space, i.e., color ratio versus reflectance at a longer wavelength, at lower CODs becomes narrower with steep changes in the color ratio, especially at COD<10. Therefore, any uncertainty in the assumptions made in the retrieval algorithm, i.e., single-scattering albedo, an assumption of fully overcast pixels, and linear interpolation between the nodes where reflectances and its ratio of a joint aerosol-cloud scene behaves nonlinearly would result in the amplification of the error in the retrieved ACAOD. These artifacts are more pronounced at lower 245 values of both ACAOD and COD, where uncertainties in the retrieved ACAOD could reach 40% to 80% at COD<10 and ACAOD<0.5 typically observed in the present CAR AOD retrievals (Jethva et al., 2013, Table II). Figure 12 shows the two main aerosol-above-cloud retrieved parameters, namely AOD_sky, when CAR views upward flying above the cloud field, and the AOD below aircraft (AOD_cloudtop), when CAR views downward measuring the cloud field averaged over all the viewing directions (see also, Table 1, columns 6-9). The summation of AOD_sky and 250 AOD_cloudtop provides the column AOD above the stratocumulus cloud fields (ACAOD), as retrieved from CAR measurements over marine stratus clouds during SAFARI 2000 in the southeast Atlantic region. In addition to the two aerosolabove-cloud parameters retrieved from CAR, Fig. 12 also shows simultaneous COD retrievals using CAR measurements as well as AOD retrievals from the AATS sunphotometer that made coincident measurements of AOD on the UW CV-580 flights.
The AOD retrievals from AATS are based on direct Sunphotometer measurements and therefore represent aerosol loading 255 above the aircraft-level.
In the case of flight transects shown in Fig. 1, the AATS AOD retrievals were largely obtained above the marine stratocumulus clouds. However, when the cloud top is well separated from the aircraft, i.e., the altitude of aircraft is higher than that of the cloud tops; the AATS measurements do not capture the aerosol layer below the aircraft as the instrument is always pointing upwards, toward the Sun. Therefore, the reported AOD data from AATS is not representative of the total 260 column AOD above clouds, unless the aircraft is flying at the same altitude where cloud top is located. Often, the altitude difference is not negligible, for example, during the SAFARI flights shown in Fig. 3, there was a clear separation of ~600 m between the aircraft and cloud top. Specifically, the CAR-retrieved AOD_cloudtop captures this "missing" aerosol layer caught between the aircraft and cloud top, which is in addition to the AOD_sky retrieved above the aircraft level. The latter quantity is equivalent to that retrieved by AATS, whereas AOD_cloudtop is the remainder of the column AOD that we retrieve from 265 CAR in this study. For these reasons, Jethva et al. (2016) in validating MODIS-retrieved ACAOD for the same September 13, 2000 AATS flight extrapolated the airborne measurements from the respective altitudes to cloud-top using a detailed profile https://doi.org/10.5194/amt-2020-246 Preprint. Discussion started: 21 July 2020 c Author(s) 2020. CC BY 4.0 License. measurements and associated altitude-AOD polynomial in order to make the comparisons between satellite and airborne measurements consistent.
To illustrate the various retrievals, we consider flight measurements from cases h-p. The COD associated with the marine 270 stratocumulus clouds appear relatively uniform, varying between 15 -20 for measurements from cases h-m (Fig. 12). These retrievals (for cases h-m) are based on some of the most homogeneous clouds observed during the three separate circular measurements obtained from transects a-d, e-g and h-p. These relatively homogeneous and similar sets of circular transects are also noted in the BRF polar plots shown in Fig. 6h-m. The simultaneous retrievals of Sky_AOD show moderately high aerosol loading, AOD = 0.5 across circles h-m, which is in very close agreement with the AATS_AOD retrievals. The 275 consistency in AOD retrievals (above the aircraft level) between the two disparate measurement approaches, i.e. AATS and CAR, is generally found throughout data obtained from the 16 cases (a-p), as indicated by the high correlation (R 2 = 0.92) between the two retrievals shown in Table 1. However, the central distinction here is that the CAR approach also allows us to directly retrieve aerosols above clouds that are present below the aircraft level (AOD_cloudtop). For instance, in cases h-m, the AOD_cloudtop is 0.2, implying the total above-cloud column AOD is 0.7 or 40% higher relative to the above aircraft-level 280 AOD. Overall, we find AOD_cloudtop ranging between 0.17 and 0.40 from the 16 cases shown in Fig. 12, indicating a notable enhancement of the overall presence of aerosols above clouds. These observations show that a significant aerosol layer is not captured in previous measurements, indicating the strength and effectiveness of near-simultaneous multiangular measurements scanning the sky and surface, as demonstrated in this study using CAR measurements.

The influence of 3D effects on the retrieved ACAOD and COD 285
Numerous earlier studies indicate that passive remote sensing of both cloud and aerosol properties can be significantly impacted by three-dimensional (3D) radiative processes (e.g., Marshak and Davis, 2005;Wen et al., 2006; http://i3rc.gsfc.nasa.gov/Publications.html). Since the impact of 3D effects is different for different observations and retrieval algorithms (e.g., Cornet et al., 2018), we next examine the impact of 3D effects on the CAR aerosol and cloud retrievals discussed above. Our goal does not lie in providing quantitative estimates of 3D effects; instead we examine whether 3D 290 effects are likely to play a substantial role in shaping the behavior of CAR-retrieved cloud and aerosol optical depths. Figures 5k, 6k, 9k and 10k as a representative of heterogeneous areas with potentially significant 3D effects. The figures show that around 60° azimuth angle, CAR observed a roughly 300 m wide and very long trough in which the retrieved COD drops by roughly 50% while the retrieved ACAOD increases by roughly 50% . Figures 9,   10, and 11 show that this behaviour is not unique, and that in many cases with COD values below 10 or sometimes even 20, 295 the retrieved AOD values increase sharply as COD decreases. In principle, this behaviour appears consistent with earlier findings that showed 3D effects to increase retrieved AOD values for pixels that were surrounded by brighter (thick-cloudcovered) areas (e.g., Wen et al., 2013).

Our tests consider the scene shown in
To examine 3D influences in CAR retrievals, we performed 1D and 3D radiative transfer simulations using the Monte Carlo model that powers the online simulator of 3D radiative processes that was created as part of the I3RC (Intercomparison 300 of 3D Radiation Codes) project and is publicly available at http://i3rcsimulator.umbc.edu/. The key simulation parameters are listed in Table 3 and the simulation results are shown in Table 4.
In Table 4, each row, the left column indicates whether or not below-CAR aerosols (BCA) were considered, what the cloud optical depth was at the trough center, and whether the simulations considered 1D or 3D radiative processes. The indicated uncertainties come from Monte Carlo simulation noise. 305 Since COD retrievals are shaped mainly by the 0.87 µm reflectance values, 3D BRFs exceeding 1D BRFs by about 25% for COD=7 indicates that 3D radiative processes significantly enhance CAR BRFs and thus the COD values retrieved in the center of the trough-which means that 3D effects make the COD drop in the trough appear less deep than it really is. This behavior is consistent with earlier studies showing that radiative smoothing (caused by the diffusion of photons scattered from thick to thin areas) make horizontal cloud variability appear less strong than it really is. Several studies proposed counteracting 310 this effect by artificially roughening the retrieved COD fields (e.g., Marshak et al., 1998;Zinner et al., 2006), but these methods are yet to gain wide usage. By performing additional simulations, we found that if we decreased COD at the center of the trough from 7 to 4.7, 3D simulations would yield 0.87 µm BRF values around 0.32-thus resulting in hypothetical retrievals yielding COD=7 (similar to the actual CAR retrievals). We note, however, that the value of 4.7 depends on our assumption of cloud base altitude (hence cloud geometrical thickness), and so it is somewhat uncertain. 315 Regarding aerosol retrievals, the results in Table 4 imply that 3D effects enhance by about 3-10% the impact of below- Since CR is the key signal in our ACAOD retrievals, this implies that 3D effects are likely to increase retrieved ACAOD 320 values by 3-10%.
To understand this result, we need to consider both the radiative smoothing discussed above for COD retrievals, and the 3D process often called "bluing" (e.g., Marshak et al., 2008). Bluing occurs when nearby thick clouds reflect more sunlight than the clouds in the field-of-view do, and some of the extra reflection is then scattered into the instrument field-of-view by air molecules and aerosol particles that reside between the cloud and the sensor. As expected, Table 4  However, the table also reveals that given a certain 0.87 µm BRF value, 3D and 1D processes yield fairly similar 0.47 µm BRFs and thus color ratios: BRF0.47 µm, COD=4.7,3D ≈ BRF0.47 µm, COD=7,1D and CR3D, COD=4.7 ≈ CR1D, COD=7.
The weak impact of 3D effects on CR is likely due to two factors. First, while the bluing process implies a larger molecular and aerosol scattering enhancement at 0.47 µm than at 0.87 µm (i.e., a higher CR), this is partially compensated by the aerosol 330 absorption cross section being larger at 0.47 µm than at 0.87 µm. Second, much of the 3D effects that cause the enhancements apparent in Table 4 are likely caused by the in-cloud radiative smoothing process discussed above, which causes similar relative enhancements in trough BRFs at 0.47 µm and 0.87 µm: Cloud droplets, which cause radiative smoothing through multiple scattering, have similar scattering properties at 0.47 µm and 0.87 µm.
We note that simulations (not shown) indicate that 3D effects would have similar or even weaker influence on ACAOD 335 retrievals over the linear trough if the measurements were taken not by CAR flying only 600 m above the clouds, but by a satellite passing overhead. This is because the compensating effect of aerosol scattering and absorption and the spectrally neutral in-cloud radiative smoothing cause 3D relative enhancements that are spectrally quite neutral.
Overall, the results discussed above imply that 3D radiative processes had a significant impact on retrieved cloud optical depths, but also that the 3D impacts on retrieved ACAOD values is fairly small and is not the main reason for the retrieved 340

Conclusion
In conclusion, the study accomplished the simultaneous retrieval of above cloud total aerosol optical depth (ACAOD) and aerosol-corrected Cloud Optical Depth (COD) from airborne CAR measurements of cloud-reflected and sky radiances using the color ratio method. The ACAOD is partitioned between the AOD below the aircraft (AOD_cloudtop) and the AOD above 345 the aircraft (AOD_sky) with full angular coverage provided by the CAR measurements. The study demonstrates a novel measurement approach for retrieving and quantifying aerosols above clouds, in particular recovering the aerosol layer between cloud tops and aircraft level that is missed in typical airborne sunphotometer measurements of above-cloud-aerosols. Overall, this work provides a path forward for filling a critical gap in aircraft-based sunphotometer measurement strategies that are currently used to validate satellite retrievals of the ACAOD. 350 The results show a strong anticorrelation between the AOD_cloudtop and COD for cases where COD <10, and a weaker anticorrelation for COD >10. The impact of 3D radiative effects on the retrievals is examined and the results show that at cloud troughs, 3D effects increase retrieved ACAOD by about 3-10% and retrieved COD by about 25%. This indicates that the color ratio method has little sensitivity to 3D effects at overcast stratocumulus cloud decks. The results also display good agreement between CAR and sunphotometer measurements of the above aircraft AOD. However, the results also show 355 that the use of aircraft-based sunphotometer measurements to validate satellite retrievals of the ACAOD is complicated by the lack of information on AOD below the aircraft, indicating the strength and effectiveness of near-simultaneous multiangular measurements scanning the sky and surface, as demonstrated in this study using CAR measurements.     Table 1). The circular flight tracks were performed primarily for the airborne measurements of bidirectional reflectance distribution function (BRDF) (cases a-d and h-p), and in a few instances (cases e-g) represent vertical profiles for physical and chemical measurements. The marine stratus clouds were extensive as seen by the MODIS/Terra instrument on the same day around 09:25 UTC (see the map inset). The CV-580 flight 515 began just prior to 10:00 UTC and ended at 13:00 UTC. The enlarged map is derived from GWELD product generated browse image (Roy and Zhang 2019).    have lower BRF values as shown by the blue colors. A prominent feature of the marine stratocumulus clouds is the presence of the cloud bow ring associated with scattering by water droplets and with a peak at ~75° zenith angle in the antisolar direction. Figure 6: BRF at 0.874 µm obtained at different solar zenith angles (23°< SZA <34°) and locations over the marine stratocumulus off the Skeleton coastline in Namibia for the 16 cases described in Table 1. A prominent feature of the marine stratocumulus clouds is the presence of the cloud bow ring associated with scattering by water droplets and with a peak at ~75° zenith angle in the antisolar direction.      Table 1). The circular flight tracks were 555 performed primarily for the airborne measurements of bidirectional reflectance distribution function (BRDF) (cases a-d and h-p), and in a few instances (cases e-g) represent vertical profiles for physical and chemical measurements.
The marine stratus clouds were extensive as seen by the MODIS/Terra instrument on the same day around 09:25 UTC (see the map inset). The CV-580 flight began just prior to 10:00 UTC and ended at 13:00 UTC. The enlarged map is derived from GWELD product generated browse image (Roy and Zhang 2019).  viewing zenith and azimuthal angles, and covering an area defined by a diameter of about 4 km on the surface (assuming the aircraft is flying 600 m above the surface). The unique feature of these measurements is the solar disks, 575 which define the start and end point for each circle. A prominent feature of the marine stratocumulus clouds is the presence of the cloud bow ring associated with scattering by water droplets and with a peak at ~75° zenith angle in the antisolar direction. 590 Figure 5: BRF at l=0.472 µm for different solar zenith angles (23°< SZA <34°) and cloud optical thickness. The marine stratocumulus are often extensive and flat, but contain areas that have thinner clouds or even open cells that allows radiation to penetrate through and therefore have lower BRF values as shown by the blue colors. A prominent feature of the marine stratocumulus clouds is the presence of the cloud bow ring associated with scattering by water droplets and with a peak at ~75° zenith angle in the antisolar direction. 595 https://doi.org/10.5194/amt-2020-246 Preprint. Discussion started: 21 July 2020 c Author(s) 2020. CC BY 4.0 License. Figure 6: BRF at 0.874 µm obtained at different solar zenith angles (23°< SZA <34°) and locations over the marine stratocumulus off the Skeleton coastline in Namibia for the 16 cases described in Table 1. A prominent feature of the marine stratocumulus clouds is the presence of the cloud bow ring associated with scattering by water droplets and 600 with a peak at ~75° zenith angle in the antisolar direction. Figure 9: Retrieved aerosol optical depth above clouds and below the aircraft (ACAOD). Pixels without valid retrievals are shaded white.