The ORACLES field campaign is aimed at directly measuring the SEA ACAOD and its direct, indirect, and semi-direct radiative effects on climate via airborne sampling during three intensive operating periods (September 2016, August 2017, and October 2018) (Zuidema et al., 2016). The NASA P-3 flew as an airborne platform for in situ and remote sensing measurements of aerosols and clouds in all three campaigns, along with the NASA ER-2 high-altitude remote sensing platform in 2016 only. The 2016 deployment out of Walvis Bay, Namibia, included 15 successful flights for the P-3 from 27 August to 29 September (ORACLES Science Team, 2017). Nearly half of these research flights followed a routine flight path extending diagonally from 13∘ E, 23∘ S to 0∘ E, 10∘ S, and the other half focused on paths with increased chance of successful sampling with all instruments (see Fig. 1). All flights (P-3 and ER-2) were planned using the research flight planning software developed by LeBlanc (2018).

The 4STAR instrument determines in-flight aerosol optical depth (AOD) from airborne measurements of direct solar radiation. 4STAR incorporates a modular sun-tracking/sky-scanning optical head protruding above the aircraft fuselage, an instrument rack within the aircraft cabin, housing a computer, motion control, and two grating spectrometers, and an electrical umbilical and fiber optic cable connecting the optical head and the rack. This airborne sun tracker and sky radiometer has multiple operating modes (direct sun mode presented here; sky scan mode as in Pistone et al., 2019; and zenith under cloud mode as in LeBlanc et al., 2015), which are selected by an operator depending on the sky conditions. Using two spectrometers, 4STAR records hyperspectral radiation measurements spanning the continuous wavelength range from 350 to 1750 nm, with spectral resolution of 2–3 nm below 1000 nm and 3–7 nm at longer wavelengths. These hyperspectral radiation measurements yield AOD over the continuous wavelength range, broken only by prominent gas absorption lines. The full width of the field of view for the direct beam irradiance measurement is 2.4∘ with a radiometric deviations of less than 1 % across this span. The nominal calibration accuracy of AOD measurements from 4STAR is dependent on wavelength, time of day, tracking stability, stability of radiometric calibration, and various second-order corrections (such as removal of light absorption by trace gases). The accuracy is typically near 1 % in transmittance (at 500 nm), resulting in an AOD uncertainty of 0.01 at solar noon. The details of the calibration, corrections, and uncertainty assessment of 4STAR AODs are found in the Appendix.

A combination of in situ instruments is used to provide context for the AOD measurements. We use aerosol scattering from nephelometers from the Hawaii Group for Environmental Aerosol Research (HiGEAR), cloud droplet number concentration from the Artium Flight Probe Dual Range Phase Doppler Interferometer (PDI), and CO concentration from CO Measurements and Analysis (COMA), as described below.

We use the aerosol scattering coefficient at 550 nm, corrected for ambient outside relative humidity, which is calculated from nephelometer measurements operated as part of the HiGEAR extensive airborne measurement suite (similar to Howell et al., 2006). These nephelometers directly ingest aerosol from ambient air and together with other HiGEAR instrumentation provide size-resolved assessment of aerosol physical and chemical properties and their relationship to measured optical and microphysical behavior. The scattering coefficient of the aerosol is sampled with three-wavelength nephelometers (TSI 3563, at 450, 550, and 700 nm) while dependence on humidity is measured with paired single-wavelength nephelometers (Radiance Research M903 measuring at 540 nm; one had air humidified to 80 % relative humidity (RH), and the other did not control the RH). Comparisons between the dry Radiance Research and the TSI nephelometers are used to correct the truncation issues of the Radiance Research nephelometer, while the humidity dependence of the scattering coefficient is calculated from a gamma relationship obtained from the paired Radiance Research nephelometers (following Quinn et al., 2005). We also use an extinction coefficient at 550 nm, which is calculated from the corrected scattering and measured absorption coefficient. The absorption coefficient is measured in dry conditions using particle soot absorption photometers (PSAPs) from Radiance Research. The solid diffuser inlet efficiently samples particles <1 µm, with a 50 % cutoff at approximately 3 µm (McNaughton et al., 2007).

Cloud drop concentration was sampled from the PDI, mounted on a wing pylon of the NASA P-3. The PDI uses interferometry with a diagnostic technique for sampling cloud droplet size and velocity at the same time (e.g., Chuang et al., 2008; Small et al., 2009). The combined range of two lasers with differing wavelengths covers liquid cloud droplets sized 1 to 1000 µm or larger.

CO concentration from the in-situ-sampled air is reported using the COMA instrument, which includes the ABB/Los Gatos Research CO/CO2/H2O Analyzer modified for flight operations. It uses off-axis ICOS (integrated cavity output spectroscopy) technology to make stable cavity-enhanced absorption measurements of CO, CO2, and H2O in the infrared spectral region, technology that previously flew on other airborne research platforms with a precision of 0.5 ppbv over 10 s (Provencal et al., 2005; Liu et al., 2017).

New AERONET stations were set up to give context to ORACLES measurements in southwestern Africa along with two pre-existing stations in Namibia, neighboring the SEA. In addition to the new permanent sites, the highly spatially resolved DRAGON (Distributed Regional Aerosol Gridded Observation Networks; Holben et al., 2018) network of six AERONET stations were located near Henties Bay, about 100 km north of the NASA P-3 base station at Walvis Bay, Namibia, for the duration of ORACLES 2016. In addition to these stations, we use the data from the stations located at Walvis Bay Airport, Gobabeb, and Henties Bay in Namibia and Lubango and Namibe in Angola. The reported data from these AERONET sites and DRAGON represent the entire span of available sampled full-column AOD during the deployment time range, including potential local sources. To focus on the smaller aerosols of the lofted biomass burning aerosol (e.g., Pósfai et al., 2003) and reduce the influence of local sources such as large dust and sea salt aerosol particles, we report the fine-mode AOD, derived using the spectral deconvolution algorithm (O'Neill et al., 2003).

Recent advances in satellite imagery retrieval methodology enables the use of MODIS spectral cloud reflectances to obtain the overlying aerosol optical properties jointly with the cloud optical properties (Jethva et al., 2014; Meyer et al., 2015; and Sayer et al., 2016). The algorithm used here, MOD06ACAERO (Meyer et al., 2015), simultaneously retrieves the above-cloud AOD and cloud optical thickness and effective radius of the underlying marine boundary layer clouds while also providing pixel-level estimates of retrieval uncertainty that accounts for known and quantifiable error sources (e.g., radiometry, atmospheric profiles, and cloud and aerosol radiative models). MOD06ACAERO uses reflectance observations at six MODIS spectral channels from the visible to the shortwave infrared. Retrievals are run on both Terra (morning) and Aqua (afternoon) MODIS instruments with a constant aerosol–cloud vertical geometry and two different aerosol intrinsic property model assumptions. The aerosol models stem from either Haywood et al. (2003) or the standard MODIS Dark Target land aerosol product, which is the model used in this work (MOD04; Levy et al., 2009). The cloud forward model, ancillary data, and other retrieval assumptions are consistent with those of the operational MODIS cloud products (MOD06) (Platnick et al., 2017). Meyer et al. (2015) showed MOD06ACAERO retrieved cloud optical thicknesses and effective radius are consistent in range and values with the standard MODIS cloud products and larger than the standard above-cloud AOD product from the spaceborne CALIOP. Consistent with Meyer et al. (2015), we only report the AOD from MOD06ACAERO above clouds with an optical thickness of greater than 4 and AOD uncertainties lower than 100 %. Also note that for this work the retrievals are aggregated to a 0.1∘ equal-angle latitude/longitude grid.

For another comparison, we use the standard Dark Target aerosol retrieval from MODIS clear-sky pixels in the SEA that has been retrieving aerosol properties from reflectances measured since 2001 (Levy et al., 2013). We used 12 years of the high-resolution time series of the MODIS-retrieved fine-mode AOD sampled during August and September as a proxy for an ACAOD climatology similarly to Zuidema et al. (2016). Our use of the fine-mode total-column AOD to represent the smoke aerosol above clouds in this region is supported by the typically smaller size of the aerosol (Pósfai et al., 2003) and is used to exclude the coarse-mode aerosol which mostly consists of boundary layer sea salt and dust along the coast. The presence of biomass burning aerosol results in the fine-mode fraction vastly dominating the optical characteristics of above-cloud aerosol in the region (e.g., Yoon et al., 2012, and the fine-mode fraction by volume in Russell et al., 2014). When there is a significant amount of biomass burning aerosol in the boundary layer in addition to the aerosol above clouds, this fine-mode assumption is expected to be an overestimate.

During ORACLES, we sampled multiple types of scenes, some of which were described by Hobbs (2003), which had CAS (i.e., described in this paper as gaps) within aerosol layers and between aerosol and cloud layers. Some scenes show a gap between the aerosol layer and the clouds, some show no gap, and some show a gap between two aerosol layers. Examples of these cases have been collected via photography from the NASA P-3 and are shown in Fig. 2; they were similarly portrayed by Hobbs (2003). These photographs were selected for easier visual identification, although they are not always showing scenes with 100 % cloud cover. Aerosol appears visually darker than the background light blue sky when the observer is at or below the altitude of the aerosol layer (Fig. 2a, b). When the aerosol appears directly above clouds, it can be interpreted as a lighter colored haze extending from the cloud top, sometimes making it harder to distinguish between aerosol and cloud boundaries (Fig. 2c).

The AOD measurements that are used to quantify the aerosol above clouds in the presence of a gap, can extend thousands of meters vertically because the aerosol within a gap contributes minimally to the overall ACAOD. For conditions without gaps, where the lowest aerosol layer is touching the top of the cloud, the ACAOD is measured when the aircraft is immediately above clouds. To identify the measurements where 4STAR sampled ACAOD (including AOD measurements within a gap), we start with the periods of flights defined by the P-3 module flags as legs directly above clouds. These P-3 module flags were created using manual inspection of flight altitude time series and flight scientist mission notes from every flight (Diamond et al., 2018). We supplement these module flags with a manual inspection of the AOD time series from 4STAR and select each sample measured directly above a cloud layer and up to the bottom of the aerosol layer. The cloud layer was defined by a cloud drop concentration greater than 10 cm-3 as measured by the PDI. When the PDI was not operational, we used lack of sun tracking from 4STAR, high outside ambient relative humidity (>90 %), and/or visual inspection of in-flight video as the metric for being in clouds. The bottom edge of the aerosol layer is defined as the altitude that has a 10 % change in AOD and either a dry scattering coefficient at 550 nm of at least 50 Mm-1 or where the in-plume dry scattering coefficient drops by more than 75 %. Figure 3 shows profiles with color-coded vertical regions to demonstrate the selection of the ACAOD portion of the AOD measurements.

The relationship of the AOD at various wavelengths is used to determine the Ångström exponent (AE, or sometimes referred to as the extinction Ångström exponent) (Ångström, 1929), which is inversely related to the size of the aerosol particles. The AE for the sampled AOD is dependent not only on the size distribution of aerosol particles but also on the type of aerosol measured (e.g., Russell et al., 2010, 2014). As a first approximation, large aerosol particles will typically have small AE values, and small aerosol particles will have large AE values – e.g., an AE value between 0.1 and 1 for large marine aerosols (Sayer et al., 2012) or above 1.5 for small biomass burning or urban industrial aerosols (Russell et al., 2014, Fig. 6; LeBlanc et al., 2012). According to Dubovik et al. (2002), AERONET-derived AE values (computed between 440 and 870 nm) for biomass burning aerosols are between 1.2 and 2.1 in Bolivia or Brazil, whereas AE values from desert dust aerosol are between 0.1 and 0.9 in Saudi Arabia. The AE measured in the source regions of the biomass burning from SAFARI-2000 showed a range between 1.6 and 2.1 from Mongu, Zambia, during the biomass burning season (Eck et al., 2003). Here we evaluate AE using two methods: (1) by fitting a second-order polynomial to the logarithm of the AOD spectra from selected wavelengths from 355 to 1650 nm and finding its derivative at any one wavelength (here at 500 nm; AE500) – a similar method to O'Neill et al. (2001) and Shinozuka et al. (2011); (2) by using the negative of the slope of the AOD with wavelength on a logarithmic scale (two wavelengths are used here, 470 and 865 nm; AE470/865) – e.g., as in Dubovik et al. (2002).

When AOD spectra are not a straight line in a log–log plot but rather slightly curved, this indicates that the AE is wavelength dependent. The curvature of AE (spectral dependence of the AE) is related to the aerosol size distribution (e.g., Kaufman, 1993; Eck et al., 1999; O'Neill et al., 2001; and Yoon et al., 2012) and additionally to the aerosol absorption (Kaskaoutis and Kambezidis, 2008). The two methods to calculate AE (evaluated at different wavelengths) can be used to quantify the AE curvature and refine the aerosol size distribution or fine-mode fraction (e.g., Yoon et al., 2012).

We have separated all 4STAR measurements in the SEA into either ACAOD (11.5 h of measurements, from flags described in Sect. 3.1) or full-column AOD (0.9 h of measurements in level legs or profiles below 600 m in altitude). The full-column AOD is distinct from the ACAOD measurements as these necessarily require conditions without overlying cloud and thus will include the elevated biomass burning layer as well as any lower-level aerosol near the sea surface. We note that these two populations do not necessarily coincide directly in space and time but may be combined in a statistical sense. Figure 4 shows the distribution of those measurements, with roughly 1 sample s-1, at two wavelengths. The ACAOD at 501 nm (ACAOD501) from all samples (blue bars) has a mean, median, and standard deviation of 0.32, 0.33, and 0.15 respectively, with an absolute range of 0.02 to 1.04. The full-column AOD (pink bars) has a mean, median, and standard deviation of 0.36, 0.30, and 0.18, respectively, with an absolute range of 0.02 to 0.74. The larger mean AODs are likely representative of the combined aerosol burden from within the boundary layer as well as the typical plume observed aloft, although exhibiting larger variability as shown by the larger standard deviation. The smaller range of values for the full-column AOD as compared to ACAOD, is likely caused by the lower number of full-column AOD measurements and their differences in location and time compared to the ACAOD measurements. The small difference between the mean above-cloud and full-column AODs indicates that the majority of the AOD501 sampled in the region is due to the elevated layers of aerosol. In contrast, the AOD sampled at 1020 nm (AOD1020) is much larger for the full column than its above-cloud counterpart by nearly 70 %, with the full-column AOD1020 having a mean, median, and standard deviation of 0.15, 0.13, and 0.06 respectively, and the ACAOD1020 at 0.09, 0.09, and 0.05 respectively with a range of 0.01 to 0.75 (Fig. 4b).

Considered together, the ACAOD and full-column AOD (denoted by the total extent of the histogram bars in Fig. 4) represent what a satellite remote sensor would retrieve in the region if it were spatially and temporally colocated to the NASA P-3 aircraft and the retrievals would not discriminate between full-column and over-cloud measurements. The mean, median, and standard deviation of AOD501 for all combined measurements are 0.32, 0.33, and 0.15, respectively, though we note that this is dominated by the greater sampling of the ACAOD (N=39 229) vs. full-column AOD (N=3395). The uncertainty in the ACAOD sampled at 501 nm (1020 nm) by 4STAR due to instrumental artifacts and calibration (see appendix for more details) are 0.011, 0.01, and 0.008 (0.013, 0.012, and 0.012) for the average, median, and standard deviation, respectively.

The spatial distribution of the ACAOD501 is presented in Fig. 5. The ACAOD was averaged in nearly equidistant latitude and longitude bins (0.65∘ latitude by 0.6∘ longitude). We observe highest the ACAOD near the western coast of Africa at the northernmost parts of the sampled region, while the lowest ACAOD is in the south of the sampled region. The higher ACAOD extends to the west but with a reduced AOD compared to near the coast, consistent with the expected behavior of the climatological plume (see Fig. 1; Zuidema et al., 2016). The higher average ACAOD in the northernmost part of the sampled region is also observed in the fine-mode AOD from ground-based AERONET stations along the southern African coast (triangle symbols in Fig. 5).

The variability in standard deviation shows that, in the north, variability in the measured ACAOD is low (Fig. 5b). Note that the standard deviation here is calculated as a fraction of all samples, and we show the total number of flight days contributing to each bin to give context as to the temporal variability observed. The largest variability in the sampled ACAOD seems to be concentrated in the central portion of the measured region, around 18∘ S, 8∘ E, with the ACAOD standard deviation exceeding 0.15 over the 3–5 d sampled. This high variability is consistent with a day-to-day change in the location of the southern edge of the highest AOD in the aerosol plume climatology for September (Fig. 1 and Zuidema et al., 2016). Large variability is also observed near Walvis Bay, Namibia, outside the typical climatology for the biomass burning plume. This variability in the ACAOD is likely caused by local production of aerosol, observed to be mostly dust or large particles. This hypothesis is corroborated with ground-based measurements from an AERONET station located at the Walvis Bay Airport which show a large but variable coarse-mode fraction of AOD (average 58%±19% of coarse-mode fraction) and consistently larger aerosol effective radius from sky scan retrievals. The fine-mode fraction of the AOD sampled by AERONET near the Walvis Bay Airport also shows some variability (Fig. 5b), but this is dwarfed by the coarse-mode variability (not shown).

The full-column AOD501 sampled by 4STAR and AERONET locations is presented in Fig. 5c, where its paucity of samples is apparent particularly in the central sampling region where the ACAOD shows higher than average values. The occasions where the P-3 sampled the full-column AOD occurred nearly always at the edges of the cloud layers. These full-column measurements were not inside pockets of open cells clouds (POC; Stevens et al., 2005; Wood et al., 2011). Full-column AOD measurements were more commonly measured past the southern edge of the stratocumulus cloud deck and where the marine boundary layer was both polluted by biomass burning or with a clean background (ORACLES Science Team, 2017). Where a direct comparison of the full-column AOD and the ACAOD is possible, the full-column AOD501 is on average 0.03 higher (a mean full-column AOD501 of 0.38 vs. a mean ACAOD501 of 0.35 at the same locations). This difference is nearly reproduced by AERONET, impacted by dust and sea salt in the boundary layer over land with overlying biomass burning aerosol, for the average fine-mode AOD501 (0.2) and total AOD501 (0.24).

An average ACAOD for this region can be calculated from these binned spatial statistics, representing a more even weighting of the ACAOD (equal spatial bins) as compared to averaging over the total number of samples which could be influenced by variability in sampling density. This averaging method attempts to reduce the spatial sampling bias by sampling the same area multiple times (like for the relatively low ACAOD near Walvis Bay) but at a cost of temporal resolution. The mean ACAOD501 and its mean uncertainty is 0.37±0.01, which is arguably more representative of the SEA region, as determined by the average of the mean within each spatial bin. The median ACAOD501 and median uncertainty of the region is 0.34±0.01, and the spatially averaged ACAOD501 standard deviation and the averaged standard deviation of its uncertainty is 0.05±0.004. The equivalent spatially averaged mean, median, and standard deviation of the ACAOD1020 and its uncertainty is 0.11±0.02, 0.09±0.01, and 0.02±0.004.

The spectral characteristics of the ACAOD are related to the aerosol intensive properties (shape, size distribution, absorption, and refractive index) (e.g., Kaskaoutis and Kambezidis, 2008; O'Neill et al., 2001). From all measurements of the ACAOD at wavelengths outside strong gas absorption, we created the ACAOD spectra representing the mean, median, and related standard deviation (Fig. 6), which is representative of the sampled ACAOD throughout this deployment. This ACAOD spectra are consistent with the mean, median, and standard deviation of the ACAOD at 501 and 1020 nm presented in Fig. 4. The ACAOD spectra for both the mean (0.38 at 452 nm; 0.13 at 865 nm) and median (0.38 at 452 nm; 0.12 at 865 nm) are easily within the mean uncertainty (0.013 at 452 nm; 0.008 at 865 nm) of the measured spectra. The standard deviation of the ACAOD (0.18 at 452 nm; 0.06 at 865 nm) is nearly equivalent to its mean at the longest wavelengths (longer than 1600 nm). This larger standard deviation at longer wavelengths can be caused by sporadic larger AODs at those longer wavelengths, agreeing with the notion of the intermittent presence of dust or marine aerosol, or alternatively, this may be linked to a lower signal-to-noise ratio of the 4STAR spectrometers. From the AE information, we can have a sense of the particle size, and we can have an educated insight into the aerosol type with the accompanying measurements and prior information for the region. To separate aerosol type (dust or sea salt), a more advanced aerosol classification method would be needed, such as the pre-specified clustering method described by Russell et al. (2014), which used the wavelength-dependent single-scattering albedo and refractive index.

There is a distinction between mean AE from the ACAOD vs. full-column AOD observed using both methods, AE500 and AE470/865, described in Sect. 3.2. The mean AE500 for the ACAOD and full column are 1.45 and 1.08, while the mean AE470/865 are 1.71 and 1.25, respectively (see blue and pink solid lines in Fig. 7). The distribution of AE in Fig. 7 seems to indicate that most of the ACAOD is influenced by fine-mode aerosol particles, which is consistent with aerosols that are aged biomass burning as reported by Eck et al. (1999) and with the in situ aerosol sizing measurements taken on board the NASA P-3 (albeit there are inlet passing inefficiencies for accurately sampling larger aerosol; Pistone et al., 2019). Even though the differences between full-column AOD and the ACAOD at 501 nm is small, the higher relative difference at 1020 nm significantly modulates the AE for above-cloud and full-column AE. This is consistent with the notion that even a relatively small population of larger aerosol particles (in this case likely sea salt), has a large impact on the AE, because of their larger AOD in the longer wavelengths (e.g., Yoon et al., 2012).

The difference in average AE evaluated at different wavelengths, (AE500–AE470/865) is -0.26 for the ACAOD, which is very similar to the combination of AE470/865 and AE difference (centered at an AE difference of -0.2 and AE470/865 of 1.85) sampled by the Mongu AERONET station within the biomass burning source region of southern Africa (Yoon et al., 2012). The full-column average AE difference of -0.17 with an AE470/865 of 1.25 is typical of coarse-mode dominant, with Mie theory predicting 30 %–40 % of fine-mode fraction for this combination of AE difference and AE values (Yoon et al., 2012). This large coarse-mode fraction is corroborated by the in situ measurements of large marine aerosol particles during the boundary layer flight segments during ORACLES and reports of local dust in the boundary layer sampled at the AERONET Mongu station.

The spatial patterns (Fig. 8) of the above-cloud AE470/865, calculated from each AOD measurement, help indicate the potential changes in aerosol intensive properties measured during ORACLES 2016. For the sampled region, the spatial mean AE470/865 (AE500), obtained by averaging the mean of each bin over the entire region, is 1.65 (1.44), with a spatial average of the medians of 1.66 (1.48) and a spatial average of the standard deviation within each bin of 0.10 (0.06). This same spatial averaging method was also used in Sect. 4.1. The spatial statistics of AE470/865 and AE500 for the full-column AOD is lower than its ACAOD counterpart by 0.4 for the mean and by 0.3 for the median, with similar standard deviations. The smallest AE470/865 (similarly for AE500, not shown) is observed in locations near the coast in the southern part of the sampling region and south of the routine flight paths. A distinctively smaller than average AE470/865 value is also observed near Walvis Bay, Namibia. This low AE470/865 may be coincident with dust or marine aerosol within the sampled column of the ACAOD at altitudes of 300 to 3700 m. Farther from the coast, there is a small tendency towards decreasing AE values, present in multiple flights, from about 1.8 to 1.6 at 5 to 3∘ E, as compared to similar latitudes near the coast. At those same locations (not shown), the AE500 of the above-cloud aerosol does not show a similar trend, possibly indicating a change in aerosol composition and size. There is, however, a trend of higher AE500 near the center of the region (7 to 11∘ E, 20 to 15∘ S), by more than 0.2 as compared to the furthest west points. Similar to the map of the standard deviation of the ACAOD (Fig. 5), a larger standard deviation in AE is observed near 18∘ S, 8∘ E (Fig. 8), at the variable southern edge of the climatological mean aerosol plume in an area with multiple sampling days. The high standard deviation in AE in this region is associated with an ACAOD between 0.2 and 0.45, with AE from 0.2 to 1.2. These aerosols, sampled over more than 1 d, may not be uniquely biomass burning, but the low AE may indicate that there is water vapor condensation on aerosol by neighboring mid-level clouds, observed in a few flights in that region. Farther northwest, a nearly equivalent number of days were sampled, but the standard deviation of the AE470/865 is lower, indicating lower day-to-day variability. In the northern near-coast region, there are multiple bins that were sampled during only 1 d; here the standard deviation should not be taken to represent the actual variability in the aerosol but rather of the sampling accuracy within a day.

To contextualize the ACAOD sampled during the ORACLES 2016 measurements, we compared the ACAOD measured directly below the aerosol layers from the NASA P-3 to those retrieved from MODIS satellite measurements (both standard aerosol Dark Target and above-cloud retrievals). We focus on the diagonal routine flight paths (southeast to northwest), where the P-3 sampled the same locations numerous times over the course of the month-long deployment and the MODIS pixels within 15 km of the P-3 tracks. The sampled ACAOD for each of the routine flights (identified by their day in Fig. 9a) is compared to its equivalent above-cloud aerosol retrieved from the combination of MODIS sensors from Aqua and Terra using the MOD06ACAERO methodology described by Meyer et al. (2015) (Sect. 2.5). When comparing the ACAOD from 4STAR and MOD06ACAERO for each sampling day, a general agreement for most days is observed with some high deviations at certain longitudes for MOD06ACAERO, albeit with day-to-day variability as to the direction of the agreement. For example, MOD06ACAERO was high compared to 4STAR measurements on 12 September near 7∘ E, and higher than average ACAOD was measured by both 4STAR and MOD06ACAERO near 3∘ E on 31 August and 4 September.

We compile the daily 4STAR ACAOD and MOD06ACAERO values to a mean and median (spanning the August–September 2016 ORACLES deployment period), which we then compare to a proxy of ACAOD climatology based on the standard MODIS Dark Target fine-mode aerosol retrieval (Fig. 9b, c). The ACAOD proxy is the monthly averaged MODIS fine-mode AOD for clear-sky pixels that have been aggregated from its original high resolution to 1∘ in latitude and longitude following the diagonal routine flight track of the P-3. The above-cloud aerosol is fine-mode dominant (Sect. 4.2), while the boundary layer aerosol is coarse mode dominant. The general longitudinal dependence and magnitudes of the mean ACAOD as measured by 4STAR are consistent with the MODIS fine-mode climatology, with larger ACAODs in the western region (Fig. 9b).

The peak in this climatology occurs near 1∘ E along the diagonal, whereas the 4STAR ACAOD broadly peaks closer to 3∘ E, and MOD06ACAERO subsampled to routine flights is closer to 2∘ E. The larger mean MOD06ACAERO at 7∘ E as compared with 4STAR and the climatology is likely due to anomalously high days skewing the mean (such as 12 September). On the eastern end, between 10 and 12∘ E, 4STAR measured a much lower ACAOD (below 0.1) than the climatology and MOD06ACAERO but measured a higher ACAOD (0.27) at the easternmost edge of the routine flight path, near 14∘ E. The easternmost 4STAR measurements are within 0.05 of the averages from AERONET ground-based measurements over the same routine flight days, which are higher by ∼0.15 than monthly averages from AERONET measured during August or September 2016. For the entire longitude span investigated here, the 4STAR ACAOD averaged 12.2 % lower than the climatology (difference of 0.04 AOD) and 16.0 % lower than MOD06ACAERO for September (12.1 % of the August mean) along the routine flight track.

The longitudes with the smallest difference between the subsampled MOD06ACAERO and the monthly averages shows where the flight sampling is adequate to represent monthly mean, whereas for regions with large differences, the sampled ACAOD is not representative of its monthly mean. The peak mean ACAOD for all August and September MOD06ACAERO at the most western edge of the region, near 0∘ E, is shifted to the east in the mean MOD06ACAERO subsampled for routine flights.

The largest differences between the monthly mean MOD06ACAERO for September 2016 and the subsampled MOD06ACAERO (around 2, 6–7, and 10∘ E) suggest that sampling in that region is not representative of the monthly mean. There is good agreement between the MOD06ACAERO subsampled and the monthly mean in other longitudes (within 0.05) suggesting that the 4STAR ACAOD can be compared to monthly statistics at those locations. In these locations, the 4STAR ACAOD had a bias of about 0.05–0.08 for most of the flight tracks (4STAR being lower than the subsampled and monthly mean MOD06ACAERO). There is a divergence near the coast (12∘ E) between the 4STAR ACAOD and MOD06ACAERO, showing a longitudinal trend in this bias by greater than 0.1.

Similar longitudinal dependence of the ACAOD is observed in the medians as with the means but with greater differences at most longitudes between the 4STAR ACAOD and MOD06ACAERO. Differences between the mean and the medians are shown here to reduce impact of outliers in our sparsely sampled data. The MODIS fine-mode climatology medians peak twice in the western edge, near 1 and 4∘ E, whereas the measured 4STAR ACAOD peaks at 1∘ E; MOD06ACAERO also peaks at 1∘ E and again at 7∘ E, like its means. Median and mean differences for both MOD06ACAERO and 4STAR seem to move their respective maximum farther west and increase matching farther east (notably at 9∘ E), indicating a changing ACAOD distribution with longitude.

Overall, the ACAOD sampled by 4STAR is slightly lower than the MOD06ACAERO counterpart for averages and medians over the same days; additionally, it is lower than the MODIS AOD fine-mode climatology. The peak for September 2016 was more eastward than what the MODIS AOD fine-mode climatology indicates, with 4STAR measurements peaking even more to the east than those from MOD06ACAERO. This shift in peak ACAOD is likely related to differences in meteorology and associated wind patterns or a shifting of the biomass burning source locations for September 2016 as compared to the 12-year climatology. The assumption that all fine-mode AOD in clear sky retrieved by MODIS over 12 years is representative of the above-cloud AOD should be revisited, as this assumes that (1) no aerosol in the marine boundary layer contributes to the fine-mode AOD and (2) aerosol in clear sky is representative of the above-cloud aerosol. As far as the first assumption is concerned, a polluted marine boundary layer with non-negligible black carbon concentrations was observed at times during ORACLES 2016 (ORACLES Science Team, 2017), which would indicate that the proxy ACAOD from MODIS 12-year climatology may be an upper bound of the ACAOD. The synoptic scale of the near-constant ACAOD values (see Fig. 1) spans both the marine stratocumulus clouds and neighboring clear sky pixels for given days, leading credence to the second assumption.

Additionally, the filtering of MOD06ACAERO to only apply to retrievals over opaque water clouds (with optical thicknesses greater than 4) may lead to systemic biases in the ACAOD. Aerosols embedded within clouds have been shown from spaceborne polarimeter measurements to skew ACAOD retrievals (Deaconu et al., 2017). Although based on different retrieval principles, having aerosol embedded within clouds would likely produce a reflectance spectrum in MODIS measurements similar to the aerosol above clouds, leading to biased high retrievals of the ACAOD that include the optical impact of cloud-embedded aerosols.

The vertical profiles of AOD showcase the large variability in the gap size and location along the atmospheric column (Fig. 10). The ACAOD flag, described in Sect. 3.1, allows assessment of the frequency of cases where there is and is not a gap between the aerosol layer and clouds, (Fig. 2b and c), though it is not able to identify more complex scenes with a gap within aerosol layers. During any one profile, the vertical extent of the continuous measurements flagged as ACAOD quantifies the gap between the cloud top and aerosol layer bottom. For cases in which this vertical extent is near 0 m (within an uncertainty of 60 m), it is said that the profile has no gap between aerosol and cloud. Unlike previous studies from spaceborne lidars (Devasthale and Thomas, 2011; Rajapakshe et al., 2017), we found that within the entire region sampled by the NASA P-3 the gap does not linearly decrease towards the west in a near-monotonic fashion (Fig. 14). Figure 14a shows the meridionally averaged gap extent for all the samples, convolving the temporal and latitudinal variations. The smallest gap extent is observed at longitudes west of 2.0∘ E, similar to what is observed in CALIOP measurements (not shown; Deaconu et al., 2019), but this may be biased due to the low number of days sampled (only a maximum of 3 d, with six different profiles), resulting in a relatively large impact of the meteorological state compared to the driving impact of the climatology. The largest average gap is not nearest to coast but rather midway into this sampling region at about 7.5∘ E and is observed over 5 nonconsecutive days spanning 31 August to 20 September, with gaps larger than 1 km observed on 6 September at 18.2∘ S, on 10 September at 17.8∘ S, and on 14 September at 16.1 to 17.7∘ S. Similarly, a local maximum in gap extent near 7.5∘ E was described by Rajapakshe et al. (2017) using CATS and CALIOP measurements observed in nighttime. Nearer to coast, between 8.5 and 11.5∘ E, there is a region of close to near-zero gap extent, with median extents below 500 m. Combined in larger longitude spans with a higher number of samples (Fig. 14b, c, and d), omitting the profiles taken over land during take-off and landing at 14.5∘ E, the mean of the gap extent distribution peaks between 5 and 10∘ E. Another way to view this dependence with longitude of the distribution is the proportion of the total profiles or cases that have a gap of less than 60 m (near zero for this analysis) or through the larger distance defined by McGill et al. (2015) as clouds embedded within an aerosol layer (CEAL; 360 m), denoted by the gold and brown colors in Fig. 14. We see a region where 0 % of the 3 d (4 profiles) measured a near-zero gap extent at 5.5∘ E, and 0 % of the 3 d (16 profiles) are considered CEAL cases at 5.5 to 7.5∘ E. The peak of the cases that have no gap or CEAL occurs at the westernmost edge, with a secondary peak between 8.5 and 11.5∘ E. For all measurements, the proportion of CEAL cases is observed here at 48 %, a statistically significant lower value (p value of 0.027) than reported for a larger region sampled with CATS (60 %) by Rajapakshe et al. (2017).

The direct radiative effect of aerosol above clouds is not likely to be significantly modified by whether the aerosol is touching the top of the cloud or not but rather by the modulation of inherent aerosol and cloud properties. The direct aerosol radiative effect varies by only 1 %–3 % when considering changes in height above clouds of the back carbon aerosol layer (Zarzycki and Bond, 2010). Alternatively, for the indirect aerosol–cloud interactions, we have observed aerosol layers touching the top of the clouds. We have observed more direct contact between clouds and aerosol by up to 12 % for CATS as reported by McGill et al. (2015) and potentially by more than 40 % for CALIPSO as compared to Devasthale and Thomas (2011), this increases the potential of a larger indirect effect. However, touching of the aerosol and cloud is not always the best indicator of potential aerosol–cloud interactions for indirect effects, especially when considering that there may have been past interactions between a specific cloud and aerosol layer (e.g., Diamond et al., 2018). The exact representativeness of these results, including the aerosol layer vertical distribution, from airborne sampling to the natural world, are investigated in future studies (e.g., Shinozuka et al., 2019). There is likely a large interannual variability and geographical sampling variations in the SEA, which could skew the comparison between airborne and satellite sampling.