Impact of absorbing and non-absorbing aerosols on radiation and
low-level clouds over the Southeast Atlantic from co-located satellite
observations

Abstract. We use data derived from instruments onboard the Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observation (CALIPSO) and CloudSat satellites as well as meteorological parameters from reanalysis to explore situations when moist aerosol layers overlie stratocumulus clouds over the Southeast Atlantic during the biomass burning season (June to October). One main goal is to separate and quantify the impacts of aerosol loading, aerosol type, and humidity on the radiative fluxes, including cloud top cooling. To achieve our objectives we split the data into different levels of aerosol and moisture loadings. By using the aerosol classification available from the CALIPSO products, we also separate and compare situations with pristine air, with smoke, and with other (mixed) types of aerosols. We find a substantial number of cases with mixed aerosols above clouds that occur under similar meteorological conditions as the smoke cases. In contrast, the meteorology is substantially different for the pristine situations, making a direct comparison with the aerosol cases ambiguous. The moisture content is enhanced within the aerosol layers, but we do not find a monotonous increase of the relative humidity with increasing aerosol optical depth. Shortwave (SW) heating rates within the moist aerosol plumes increase with increasing aerosol loading and are higher in the smoke cases compared to the mixed cases. However, there is no clear correlation between moisture changes and SW absorption. Cloud top cooling rates tend to decrease with increasing moisture within the overlying aerosol layers, but the influence is relatively weak and confounded by the strong variability of the cooling rates caused by other meteorological factors (most notably cloud top temperature). No clear influence of aerosol type or loading on cloud top cooling rates is detected. We also do not find any correlation between aerosol loading and the thermodynamic structure of the atmosphere nor the cloud top height, i.e. no indication of a semi-direct aerosol effect. This result is consistent with previous studies that examined clearly separated aerosol and cloud layers (in our case at least 0.4 km).


to analyse the composition of the aerosol layers and to compare smoke versus non-smoke aerosol occurrences. One main goal of our study is to separate and quantify the impacts of aerosol loading, aerosol type, and humidity on the radiative fluxes within the aerosol layer as well as their potential influence on cloud top cooling. More specifically, we seek observational support for the model-based finding of reduced cloud top cooling from moist aerosol layers above the boundary layer. Furthermore, we examine if the loading and type of aerosol affect general cloud features such as cloud top height. In our study we use the satellite 70 data products to select cases where aerosols and clouds are separated from each other. This was not explicitly done by Deaconu et al. (2019), who in their analysis included all aerosols above clouds occurrences close to the coast of Angola. Another novel feature of our study is that we explore if the previously observed covariance between aerosol and moisture in the region implies a consistent and monotonous increase of humidity with aerosol loading. The observational data and methodology are described in Section 2. Our results are presented in Section 3 followed by a summary and conclusions in Section 4. 75 2 Datasets and methodology 2.1 CALIPSO, Cloudsat and ERA5 Table 1 displays a summary of the datasets, products and variables used in the study. The CALIOP (Cloud-Aerosol Lidar with Orthogonal Polarization) instrument on board CALIPSO provides information on aerosol and cloud optical properties with high vertical resolution. Furthermore, the CALIPSO V4 classification algorithm (Kim et al., 2018) discriminates between different 80 types of aerosols and clouds into ice or water phase (Winker et al., 2009). The aerosol type is determined using measurements of the integrated attenuated backscatter and the volume depolarization ratio as well as surface type and aerosol layer altitude (Omar et al., 2009). The ice-water phase is derived from the volume depolarization that allows to discriminate between the spherical cloud droplets and nonspherical ice crystals (Winker et al., 2009). An aerosol extinction-to-backscatter ratio (aerosol lidar ratio) is assumed in order to enable, in most cases, the calculation of extinction from lidar backscatter signals (Winker 85 et al., 2009).
Two datasets were used from the CALIPSO Version 4.20 (V4) Level 2 product: the Merged Aerosol and Cloud Layers Data and the Aerosol Profile Data. In the Merged Aerosol and Cloud Layers Data the information is reported by layers at a 5 km horizontal resolution. We used it in order to know the altitudes of the aerosol and cloud layers as well as the aerosol types. The full set of tropospheric aerosol types identified by the algorithm are: clean marine, dust, polluted continental/smoke, 90 The Aerosol Profile Data provides information as profiles with 60 m and 5 km of vertical and horizontal resolution, respectively and includes vertically resolved meteorological information derived from the Modern-Era Retrospective analysis for Research and Applications, Version 2 (MERRA-2). From this data set, we obtained: profiles of temperature and pressure, used in the computation of the potential temperature for exploring the stability of the atmosphere, profiles of relative humidity (RH), to examine the correlation between moisture above clouds and aerosol loadings as well as the aerosol extinction and col-100 umn optical depth of tropospheric aerosol, which both are related to the aerosol type and aerosol loading. The Cloud Profiling Radar (CPR) onboard CloudSat produces detailed images of cloud structures. The profiles of radiative fluxes and atmospheric heating rates usesd in our study were obtained from the 2B-FLXHR-LIDAR product (Henderson et al., 2013), which combines aerosol and cloud profiles from CALIPSO and CloudSat, weather data from the European Centre for Medium-Range Weather Forecasts (ECMWF), a dynamic land surface model (Lebsock et al., 2017) and radiative transfer model to compute the profiles 105 of radiative fluxes at a vertical resolution of 240 m (Lebsock et al., 2017).
To carry out the analysis, the products obtained from the Merged Aerosol and Cloud Layers Data and the Aerosol Profile Data from CALIPSO were combined with the radiative fluxes and atmospheric heating rates obtained from CloudSat. Since the spatial resolution between the satellite data sets differ, the CloudSat profiles were averaged to the 5 km horizontal resolution of CALIPSO. Finally, the ERA5 reanalysis (Hersbach et al., 2020) was used to characterize the governing meteorological 110 conditions during the period of analysis with special emphasis on winds.

Area and time period
The Southeast Atlantic area selected for the study extends from 10 to 18°S and from 2 to 10°E. It is located over the Namibian stratus region identified by Klein and Hartmann (1993) and is close to the continent, where the biomass burning aerosol loadings are high and where the aerosol layer is on average centered above the low-level clouds (Deaconu et al., 2019). The 115 final extent of the area of study was determined based on a balance between having a sufficient number of cases while keeping the natural variability of meteorology and cloud properties relatively small. Our area of study is similar to the one used by Deaconu et al. (2019), but it is shifted 4°towards the west so that the entire domain is over the ocean. It is also 3°longer in the north-south direction.
The time period selected for the study is June to October for the years 2007 to 2010, i.e. covering the July-October period 120 when the dominant winds frequently transport biomass burning aerosols from continental sources towards the stratocumulus decks located over the Southeast Atlantic (Adebiyi et al., 2015). Following Deaconu et al. (2019), who studied June to August (JJA) of one year (2018), we also included the month of June. Here, we divide the full biomass burning season into two parts,

Selection and classification of cases
To study the effects of aerosols overlying clouds, we identify and contrast cases with and without aerosols above clouds. We also distinguish between cases with smoke aerosols and aerosols with other optical properties using the CALIPSO V4 Level 2 product on aerosol and cloud layers (cf. Section 2.1) as follows: 1. Smoke cases: Atmospheric columns in which aerosol layers(s) classified by the CALIPSO V4 algorithm as "elevated 130 smoke" are above and detached from clouds. The main characteristics of these cases are: -The presence of only one cloud layer in the atmospheric column with cloud top altitude between 0.75km and 2.5km.
Cases with cloud top altitudes lower than 0.75km are not considered to avoid the ground cluttered data in CloudSat retrievals, whereas the maximum altitude (2.5km) was chosen to only capture scenarios with shallow clouds.
-The presence of one or more aerosol layers above the cloud layer with a separation between the cloud layer and the 135 bottom aerosol layer in the range 0.4 to 6km. With the lower distance we expect to reduce the number of situations with possible contact between aerosols and clouds, whereas the higher altitude was selected to discard situations in 5 https://doi.org/10.5194/acp-2020-1089 Preprint. Discussion started: 22 October 2020 c Author(s) 2020. CC BY 4.0 License.
which aerosols are very far from the clouds. Situations with more than one aerosol layer above cloud are included only if the distance between the aerosol layers is smaller than 0.3 km.
2. Mixed cases: Cases with aerosol layer(s) that are not categorized as "elevated smoke" by the CALIPSO V4 algorithm.

140
Otherwise, the same criteria as for the smoke category are used for the selection of the altitudes and number of aerosol and cloud layers.
3. Pristine cases: Cases containing only a cloud layer with a cloud top altitude between 0.75km and 2.5km, i.e. the same characteristics as described for the smoke cases (above) but with no aerosol present above the cloud layer.

145
In this section, we will first examine the composition of the aerosol layer as determined by the CALIPSO V4 retrieval algorithm.
Thereafter, we will examine if and how the spatial and temporal distribution of the aerosol and cloud layers differ between the three groups of cases (as defined in Section 2.3) and to what extent differences in the prevailing meteorological conditions may prevent a fair comparison between them. Finally, we will analyze the influence of the aerosol layer and its composition on the radiative heating profiles and examine the main drivers of any influence (aerosol type, loading or RH).

Aerosol type occurrence
The frequency of occurrence for the different aerosol types found within the aerosol layers are shown in figure 1. The "elevated smoke", which corresponds to the smoke cases in our study, is the predominant type, representing 53% and 58% of the total aerosol layers found during JJA and SO, respectively. Among the remaining aerosol types (which correspond to the mixed cases in our study), the "Polluted continental/smoke" is predominant. Furthermore, the number of cases classified as "Elevated 155 smoke" and "Polluted continental/smoke" is greater during SO than JJA. This happens because during September, there is a maximum in the extent of the stratocumulus deck at the same time as there is a maximum in transport of continental aerosol over the Southeast Atlantic due to a strengthening of the anticyclone over southern Africa (Adebiyi et al., 2015). Figure 1 shows that during the biomass burning season there is a non-negligible number of "mixed" aerosol cases overlying the stratocumulus clouds over the Southeast Atlantic. The possibility exists of having some of these cases misclassified as not 160 being smoke for two reasons: first, some of the "mixed" cases can indeed contain a substantial amount of smoke, if the smoke is located below 2.5 km (cf. section 2.3); second, the CALIPSO algorithm itself can misclassify the aerosol type under certain circumstances (Kim et al., 2018), although the aerosol classification was improved from version V3 to V4 (used here) of the algorithm, resulting in an increase of the aerosol classified as smoke over the Southeast Atlantic . September-October (SO) during the period 2007-2010. "Other mixed" refers to situations with more than one aerosol layer, where one of the layers is not defined as "elevated smoke". Our smoke cases correspond to "elevated smoke" whereas our "mixed cases" contains the rest of the aerosol types.  (1107) 5 (120) 3.2 Temporal and spatial distribution of cases 165 Next, we examine the number of cases identified and their spatial (horizontal and vertical) distributions during the two periods (JJA and SO). If these characteristics differ substantially, then the cases may also be subjected to different meteorological conditions which may influence the outcome of any comparison. Table 2 shows the total number of days and the total number of profiles when "smoke", "mixed" and "pristine" cases were found. The number of aerosol profiles is greater during SO than during JJA for the reasons explained in section 3.1. In contrast,  at longitudes between 2 and 6°E, which means that more profiles matching the conditions described in Section 2.3 were found away from the coast rather than near the coast.

Clean Marine Dust
The altitudes of the tops of the cloud layers are shown in Figure 3 together with the top and base altitudes of the aerosol layers. The average altitude of the cloud tops is clearly higher in the pristine cases (between 1.3 and 1.4 km) compared to both aerosol cases (around 1 km). When an aerosol layer is present above a cloud layer, the maximum cloud top altitude is 1.7 km 180 (one smoke case in JJA), but in general there were very few cases with cloud tops higher than 1.5 km, a result consistent with Wilcox (2010). In contrast, the maximum cloud top altitude for the pristine cases is close to 1.9 km. Another notable feature is that the aerosol layer altitudes are on average higher during SO (4.2 km for smoke and 4.0 km for mixed cases) than during JJA note that all smoke cases have aerosol top altitudes higher than 2.5 km in accordance with the characteristics of the CALIPSO V4 aerosol type "elevated smoke".
A likely cause of the difference in aerosol altitudes between JJA and SO is the location of the Southern African Easterly Jet. This jet supports biomass burning aerosol transport from the continent to the ocean and is stronger and migrates to higher altitudes (between 650 and 600 hPa) during SO (Adebiyi and Zuidema, 2016). Another factor that could contribute to the 190 observed differences in the location of the aerosol layer is that land surface temperatures are higher in October (southern hemispheric spring) than in June (winter). Consequently, the top of the boundary layer, and the injection heights, may also be higher.

Prevailing meteorological conditions
The atmospheric circulation governs the thermodynamic environment where clouds form. Even a small perturbation in the 195 prevailing wind pattern may affect the temperature and humidity profiles and thereby the characteristics of a stratocumulus cloud layer (Wood, 2012). It is therefore important to ensure that the different groups of cases are subjected to similar largescale circulation patterns and meteorology when investigating any influence of aerosol layers on the radiative fluxes and lowlevel cloud properties. Figure 4 shows the average horizontal wind direction for all three groups of cases at a level representative of the cloud 200 layer (900 hPa during both JJA and SO) and a level representative of the aerosol layer (700 and 625 hPa during JJA and SO, respectively, cf. Figure 3). The smoke and mixed cases have almost identical wind patterns, which is expected since they were often detected during the same days and since their temporal and spatial distributions were found to be similar (Section 3.2).
At 900 hPa, southeasterly winds dominate during both JJA and SO. At 700 (625) 2019)). Higher RH values are also observed in SO compared to JJA which can be linked to the strengthening of the easterlies during SO. During JJA, the RH within the aerosol layer is up to 8.5% higher for the smoke than for the mixed cases. The maximum difference in RH between the aerosol cases 220 reduces to only 3.4% just above the boundary layer during SO. Even though RH differences are small, extinction differences reach 0.05 which is close to the peak average extinction of the mixed cases (0.06) in SO. The potential temperature profiles show a shallower boundary layer with a stronger inversion in the presence of aerosols compared to the pristine situations which supports the cloud top heights differences observed in figure 3. It is likely that the difference in boundary layer height and cloud top altitudes is mainly caused by the northward shift of the anticyclonic circulation for the pristine cases ( Figure 4).

225
In summary, while the two aerosol cases have similar meteorological conditions, the pristine cases differ in terms of winds at 700 hPa, temperature and RH profiles. These differences constitute an obstacle to detect any aerosol influence on cloud properties and radiative fluxes when comparing aerosol versus pristine cases. This needs to be kept in mind in the following subsection.

Radiative heating profiles
230 Figure 5 (a-b) shows the average radiative heating profiles for the different cases. The main difference between the smoke and mixed cases in net radiation is found within the aerosol layer, where the smoke cases show a clear average heating during both JJA and SO while the mixed cases only show an average heating during SO. The differences in net heating is mainly caused by a difference in the SW fluxes as the differences in the LW fluxes are small. Within the cloud layer, the net radiative heating for the pristine cases is similar in magnitude compared to the aerosol cases. However, the similar magnitude is a result of 235 higher values of both SW heating and LW cooling. Above the boundary layer, the SW aerosol absorption is relatively small for the pristine cases. The LW cooling is also smaller than for the two aerosol cases, but still, the net radiative heating is always negative.
The average heating rates are sensitive to variations in the altitudes of the individual aerosol and cloud layers. We therefore identify the maximum LW cooling in the cloud layer (as a proxy for cloud top cooling) and the maximum SW heating in  are substantially lower for the pristine cases. A likely reason for the difference between the aerosol and pristine cases is the difference in cloud top heights (cf. Figure 3 and discussion in Section 3.5). The mean maximum SW heating within the aerosol layer is on average higher in the smoke cases than in the mixed cases. The spread in the heating rates is also larger which is  consistent with the wider aerosol optical depth (AOD) range observed in the smoke cases compared to the mixed cases (cf. Figure 6) 3.5 Influence of aerosol loading and relative humidity on heating profiles In this subsection we will focus on the smoke and mixed cases and examine to which extent variations in RH and aerosol loading affects the SW heating within the aerosol layer. Similarly, we will investigate if moisture and AOD variations have a 250 significant effect on the LW cooling rates at cloud top.
In this subsection we will focus on the smoke and mixed cases and examine to which extent variations in RH and aerosol loading affects the SW heating within the aerosol layer. Similarly, we will investigate if moisture and AOD variations have a significant effect on the LW cooling rates at cloud top.
To account for differences in aerosol loading, each group of aerosol cases was divided into three AOD intervals (low, middle 255 and high), with intervals chosen to give an even distribution of the number of profiles in each bin. The interval limits are different for each period (JJA and SO) and for each case since the AOD range varies in time and the smoke cases have a wider range of AOD values compared to the mixed cases (cf. Section 3.4). Figure 6 shows that the average extinction in the aerosol layer always increases with increasing AOD while there is no straight-forward relation between the AOD and average RH within the plume. During JJA, the average RH increases with increasing AOD for both groups of cases (smoke and mixed). via an influence on the CTA. However, we find no clear relation between the CTA and the RH (S-values below 0.09 and mostly non-significant). On the other hand, Figure 9 shows that CTA variations explain an important part of the variability of the cloud top LW cooling within a certain RH interval, illustrating the difficulty in isolating a signal of the RH impact on the cloud top radiative cooling. Note that in Figure 9, both periods and aerosol cases have been combined in order to obtain a sufficient 285 number of data points for each interval.

Summary and conclusions
We have used CALIPSO and CloudSat retrievals for the years 2007-2010 to study situations when moist aerosol layers overlie low-level clouds over the Southeast Atlantic during the biomass burning season (June -October). We divided our data into two periods, June-July-August (JJA) and September-October (SO) to reduce the effect of seasonal meteorology changes on the studied aerosol-cloud interactions. Furthermore, we used the CALIPSO V4 aerosol classification algorithm to separate cases with pristine air above clouds, smoke aerosols above clouds and other types of (mixed) aerosols above clouds.
The pristine cases displayed a clear difference in the large-scale wind pattern compared to the other two types of cases with aerosols above clouds. Easterly winds predominated in the smoke and mixed aerosol cases, which is also a prerequisite for  The two aerosol cases (mixed and smoke) displayed similar large-scale winds. They were both also associated with enhanced levels of moisture in the free troposphere, which is typical for biomass burning plumes that are advected from the continent K Day 1 Figure 9. Histograms of the mean and the standard deviation (STD) of the maximum SW heating at the aerosol layer and the minimum LW cooling at the cloud layer as functions of CTA and RH. Both periods (JJA and SO) and both aerosol cases (smoke and mixed) were used. (Haywood et al., 2003;Adebiyi et al., 2015;Deaconu et al., 2019). However, we did not find a monotonous increase of the RH of the aerosol layer with increasing AOD. In fact, we found a significant and negative correlation between AOD and RH during SO.
According to the CALIPSO V4 aerosol classification algorithm, and in agreement with our expectations, smoke was the dominant aerosol type overlying the stratocumulus clouds during the biomass burning season. Nevertheless, a substantial 305 amount of other kinds of aerosols were also detected within the pollution plumes. One explanation for the obtained result cloud be that the CALIPSO algorithm miss-classifies some of the smoke aerosols as other aerosols. Another explanation could be that other aerosol types than smoke indeed occasionally dominate the pollution plumes. Chazette et al. (2019) observed a mixture of different aerosol types, mostly polluted dust and smoke, in the free troposphere over the coastal regions of Namibia (near the area of our study) during the biomass burning season. Their results are consistent with our findings and merits a broader 310 definition of the pollution plumes overlying the stratocumulus clouds.
Our analysis clearly showed that the SW heating of the aerosol layer increased with higher aerosol loading and that the heating rates were higher in the smoke cases compared to the mixed aerosol cases. Changes in the RH of the aerosol layer always had a negligible impact on the SW heating rates, in agreement with the findings by Yamaguchi et al. (2015) and Deaconu et al. (2019). We found no clear impact of changes in AOD or aerosol type on the thermodynamic structure of the 315 atmosphere nor the cloud top height, i.e. no indication of a semi-direct aerosol effect. Previous studies have suggested that there is a weak overall semi-direct effect of elevated smoke layers over the Southeast Atlantic and that the gap between the absorbing aerosol layer and the underlying cloud must be small (less than 0.5 km) to detect a significant influence (Herbert et al., 2020;Costantino and Bréon, 2013;Adebiyi and Zuidema, 2018). Our results are consistent with these studies as we selected cases with a minimum distance of 0.4 km to avoid any potential contact between the aerosol layer and the cloud.