To reduce uncertainty in climate sensitivity and future global warming estimates, it is necessary to quantify the radiative forcing of aerosols. However, there is a large uncertainty in aerosol radiative forcing, and much of this uncertainty is related to the magnitude of indirect aerosol effects on clouds of -0.45 W m-2 with an uncertainty range of -1.2 to 0 W m-2 . Aerosols also impact clouds through “rapid adjustments” associated with aerosol–radiation interactions, otherwise known as semidirect effects (SDE). Available estimates suggest a relatively large SDE uncertainty of -0.3 to +0.1 W m-2 . The growing interest in the impact of aerosols on climate has stimulated the development of better physically based parameterizations of aerosols and aerosol–cloud interactions (ACI) in climate models. Nevertheless, the lack of understanding of external forcing on clouds remains one of the largest uncertainties in climate modeling and climate change projections.

One aspect of ACI is the possible influence of dust on marine stratocumulus clouds (MSc). North Africa is the world's largest dust source . Dust emissions from this region occur from both the hyperarid Sahara and the semiarid Sahel. Africa is responsible for approximately half of the global emissions with several hundred teragrams of dust being transported across the Atlantic towards the Americas throughout the year . This has consequences for air quality downwind as well as the radiative balance over the Atlantic, via scattering and absorption of solar radiation (and to a lesser extent absorption of terrestrial radiation), microphysical and thermodynamical effects on clouds , and tropical cyclone formation . The dominant mode of coupled ocean–atmosphere variability in the tropical Atlantic is called Atlantic Meridional Mode (AMM). show that this mode is linked to Saharan dust variability. The AMM is thermally damped, and thus direct ocean cooling from dust is required for the AMM to persist.

Along the western coast of Africa, extensive regions referred to as the semipermanent subtropical marine stratocumulus sheets exist, in which the stratocumulus cover exceeds 40 % and can be as high as 60 %. Therefore, they may be affected by the high concentrations of continental aerosols, in particular dust. Stratocumulus clouds strongly reflect incoming solar radiation and exert only a small effect on the outgoing longwave radiation. Overall they exert a strong negative net radiative effect that markedly affects Earth's radiative balance . Small changes in the coverage and thickness of stratocumuli are enough to produce a radiative effect comparable to that associated with increasing greenhouse gases .

A few observational studies show a relation between dust aerosols and cloud cover. show that there was a positive correlation between observed thin low cloud amount and mineral dust off the west coast of northern Africa. Observations during a dust storm suggest smaller cloud droplets and suppressed precipitation over the eastern Mediterranean . In another study, rainfall and dust load in the West African Sahel exhibit a negative correlation, which is explained by a larger number of cloud condensation nuclei (CCN) when the dust load is high, distributing available cloud water over a large number of droplets, thus suppressing droplet growth and precipitation . study the indirect effects of mineral dust on warm clouds during a Saharan dust-transport event. They show that clouds are affected strongly by dust and the effects segregate and vary systematically when classified by cloud precipitation regime, cloud-top temperature, and liquid water path (LWP). For non-precipitating clouds the estimated aerosol indirect effect (AIE) is -0.1 W m-2 over all temperature bands. Further classification by LWP (for all LWP > 150 g m-2) strengthens the AIE to approximately -0.2 W m-2. present an assessment of ACI from ground-based remote sensing under coastal stratiform clouds. They calculate ACI as the change in cloud droplet number concentration (Nd) with aerosol concentration for constant values of LWP. They show that the average ACI depends on the relative value of cloud LWP, methods for retrieving Nd, the aerosol size distribution, updraft velocity, and the scale and resolution of observations. show that over the tropical North Atlantic during summer, low cloud fraction increases by 3–10 % in response to high mineral dust loadings. compare dusty and pure cloud properties and radiative forcing over northwestern China (source region) and over the northwestern Pacific (downwind region). Dusty clouds are defined as clouds that extend into a dust plume environment (i.e., dust aerosols observed within 50 m of the cloud), while pure clouds are clouds having no dust aerosols within 500 m around them. They show that dust aerosols change the microphysical characteristics of clouds, reducing the cloud effective particle size and, possibly, cloud optical depth, LWP, and ice water path. They show that dust aerosols cause an instantaneous net cooling effect in the source and downwind regions, respectively. analyze the effect of dust storms on cloud properties and radiative forcing over northwestern China from April 2001 to June 2004. Due to changes in cloud microphysics, the instantaneous net radiative forcing is increased from -161.6 W m-2 for dust-free clouds to -118.6 W m-2 for dust-contaminated clouds. focus on tropical Atlantic Ocean (30∘ N to 30∘ S). They find that during a 10-year study period (July 2002–June 2012), dust intrusions from the Sahara into the tropical Atlantic in July cause a significant cloud cover up to 0.8–0.9 in the Saharan Air Layer. They suggest that the increase in cloud cover could be explained by the formation of shallow stratocumulus clouds below the temperature inversion with the assistance of settling Saharan dust particles.

In this paper, we will show the importance of Saharan dust contributions to ACI off the coast of northern Africa and, in particular, the importance of the SDE. Initial modeling studies found that the SDE causes a positive radiative forcing, thus warming the climate system . Furthermore, show that when absorbing aerosol coincides with shallow broken clouds, the radiative heating of absorbing aerosol reduces the cloud cover and increases the absorption solar radiation at the surface, resulting in a net positive radiative forcing. However, more recent modeling studies show that when absorbing aerosol resides above the cloud top, it can stabilize the underlying layer, enhancing stratocumulus clouds . , using large eddy simulation experiments, show aerosols may also yield increased cloud cover and surface cooling under certain scenarios. Although few observational studies exist to corroborate these model results, uses satellite data and shows that when smoke resides above stratocumulus clouds, the increased buoyancy of the air above the clouds inhibits the entrainment of dry air, which helps preserve humidity and cloud cover in the boundary layer. Similarly, show the overall effect of biomass burning was to enhance marine stratocumulus off the coast of California. , however, show that Amazonian biomass burning suppressed satellite-based cumulus cloud cover.

Here we quantify the radiative effects of Saharan dust on North Atlantic MSc. We use observations from multiple sensors in the NASA A-Train satellite constellation from 2004 to 2012 to evaluate the complex processes inherent in aerosol–cloud systems and to obtain estimates of aerosol–cloud radiative forcing for dust and marine stratocumulus clouds, including the uncertainties. The NASA data include CloudSat radar observations co-located with aerosol and cloud properties from Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observation (CALIPSO), Clouds and the Earth's Radiant Energy System (CERES), and ERA-Interim reanalysis data. We show that the SDE – relative to the first and second aerosol indirect effects – is the largest component of ACI and is also responsible for a seasonal reversal in the sign of ACI. A description of our datasets and methodology are provided in Sects. 2 and 3. Results are presented in Sect. 4, and a discussion/conclusion is presented in Sect. 5.

Cloud-Aerosol Lidar with Orthogonal Polarization (CALIOP) instrument on board the CALIPSO (Winker et al., 2009) has provided data since June 2006. This space lidar measures the backscatter signal at 532 and 1064 nm and the degree of linear polarization at 532 nm. CALIOP provides aerosol and cloud profiles with high vertical resolution of 30–60 m (up to 20 km) during its 16-day repeat cycle, and its beam diameter is 70 m at the surface . CALIOP has a very small swath width and the distance between two CALIPSO tracks is more than 2000 km in low and mid-latitudes. Thus, to produce statistically meaningful profiles, a significant averaging in time and space is needed .

CALIOP can discriminate between dust and other types of aerosols, which generally do not depolarize light. Due to CALIOP's sensitivity to polarization at 532 nm, the depolarization from scattering from nonspherical dust particles is a means to discriminate between dust and other aerosol species . CALIPSO categorizes aerosols into six subcategories: dust, marine, smoke, polluted dust, polluted continental, and clean continental . Compared to the Moderate-Resolution Imaging Spectrodiometer (MODIS) sensor, most studies show that CALIPSO underestimates dust aerosol optical depth (DAOD) of the order of 0.1 over the regions having strong mineral dust load (e.g., ). demonstrate improvements in CALIPSO dust extinction retrievals over northern Africa and Europe. The improvement is applied by corrections to the Saharan dust lidar ratio assumption for CALIPSO level 2 data, the separation of the dust portion in detected dust mixtures, and the averaging scheme introduced in the CALIPSO level 3 product For this study dust vertical profiles are obtained from . CALIPSO gives extinction coefficient of dust for 399 vertical levels. DAOD at each level is calculated as the vertical integral of dust extinction profile at 532 nm. By using CALIPSO it is possible to quantify how much dust is above clouds and how much is within or below clouds. CALIPSO data are available from 2007 to 2014 for this study.

CALIPSO gives only two or three extinction coefficient values per month per grid box, and thus it is not possible to use daily CALIPSO to infer statistical relationships. Daily DAOD is obtained from Monitoring Atmospheric Composition and Climate (MACC) reanalysis. The MACC global reanalysis consists of a long-term reanalysis (2003–2012) with the coupled MACC system with data assimilation of aerosol optical depth (AOD) from MODIS satellite data. Different aerosol species (sea salt, dust, organic matter, black carbon, and sulfate) are included in MACC . AOD is also obtained from MACC at 0.55 and 0.865 µm wavelengths.

The CERES products include both solar-reflected and terrestrial radiation from the top of the atmosphere to the Earth's surface. Daily data of cloud properties such as effective cloud-particle radius (re), cloud optical thickness, cloud cover, and LWP are from the CERES Aqua Single Scanner Footprint (SSF) Edition 3A dataset. Daily values of clear-sky albedo from 2004 to 2012 are also derived from CERES for this study. All satellite data are obtained on a 1∘ × 1∘ resolution.

The MSc regime is defined by lower-tropospheric stability (LTS) and vertical velocity. To calculate potential temperature, daily temperature is obtained from the ERA-Interim reanalysis at 1000 and 700 hPa levels at 1∘ × 1∘ resolution. Daily mean vertical velocity at 500 hPa is also obtained from ERA-Interim.

Here we present the annual and seasonal radiative effect of dust on MSc, as estimated by both Method 1 and Method 2. The annual mean aerosol–cloud radiative effect estimated by Method 1 is -1.5 ± 1.4 W m-2 (Table 1). The negative radiative effect indicates that dust modifies MSc in a way that results in a cooling effect over the study area. Method 1 separates the aerosol–cloud radiative effect into two terms (Eqs. 4 and 5). Figure  shows the first aerosol indirect effect for different seasons. In all figures white areas indicate missing values, where no data for dust or clouds exist or insufficient data exist to calculate the partial derivatives.

The first indirect effect is stronger where the dust load is larger and the stratocumulus regime exists for a longer time (see Figs.  and ). The annual mean first indirect effect is -0.3 ± 0.3 W m-2, and it varies from -0.7 ± 0.6 W m-2 in summer to 0.1 ± 0.5 W m-2 in winter (Table 1). The larger negative radiative effect during summer, compared to spring and fall, is consistent with a greater abundance of both MSc and dust during summer.

Figure shows the combination of the CLE and SDE (i.e., the second term in Method 1). Similar to the cloud albedo effect, the CLE + SDE is negative during summer, fall, and spring and positive during winter. Moreover, CLE + SDE also exhibits a summertime maximum (negative), which is again consistent with the greater abundance of MSc and dust during summer. For all seasons the second term is much larger than the first term. The second term varies from -3.2 ± 2.5 W m-2 in summer to 0.9 ± 2.9 W m-2 in winter, with an annual mean of -1.2 ± 1.4 W m-2. This shows the importance of CLE and SDE in the study area.

Method 2 yields similar conclusions on the magnitude of the total aerosol–cloud radiative effect, as well as the seasonal variation. The annual mean aerosol–cloud radiative effect for Method 2 is -1.5 ± 1.6 (Table 1), and it varies from -4.3 ± 4.1 in summer to 0.6 ± 1 in winter. Method 2 separates the radiative effect into intrinsic and extrinsic parts, which are shown in Figs.  and , respectively. The intrinsic effect dominates the radiative effect in this method. Like Method 1, the radiative effect is more negative over areas with larger dust load and a higher percentage of days with MSc.

The aerosol–cloud radiative effect is weakly positive during boreal winter. The presence of non-dusty aerosols could also be a reason of the large uncertainty. show that, in winter, Saharan dust is not the predominant aerosol species over our study area. In winter, non-dusty aerosols, such as carbonates (organic and black carbon), sea salt, and sulfates, also significantly contribute to the total AOD over the tropical North Atlantic. Absorbing aerosols, such as organic and black carbon, produce mainly a positive semidirect radiative effect, which is similar to the dust effect. Sulfates and sea salt, non-absorbing aerosols, produce a negative indirect radiative effect, acting as effective CCN. Thus, non-dusty aerosols, producing either positive or negative radiative effects, significantly contribute to the large uncertainty of the aerosol–cloud radiative effect in winter.

In Method 1, CLE and SDE dominate the total aerosol–cloud radiative effect. Since the sign of the dust–cloud radiative effect is affected by the height of dust column , to investigate the role of the SDE over the region we look at the vertical profile of Saharan dust from CALIPSO. Figure  shows that during winter, most of the dust burden resides between 0 and 1 km. In contrast, during spring there are two peaks of Saharan dust: the large peak resides within the marine boundary layer (between 0 and 1 km), and a smaller peak resides above the boundary layer. During summer, similar to spring, there are two peaks, but most of dust resides above the boundary layer. During fall the amount of dust is less than in other seasons and most of dust burden resides between 0 and 1 km, with some dust between 1 and 4 km. The horizontal solid and dashed red lines in Fig.  are average CERES MSc cloud-top heights ±1σ, respectively, for each season. The average cloud-top heights in summer and spring are lowest with 1.9 ± 0.43 and 1.98 ± 0.41 km, respectively, and highest in winter and fall with 2.2 ± 0.3 km as shown in Fig. . CALIPSO shows that 88.3 ± 8.5 % of dust resides below 2.2 km in winter. During summer, however, there are two peaks, with 35.6 ± 13 % below 1.5 km and 44.4 ± 9.2 % between 2 and 4 km.

To be more clear we plot cloud-top height for different seasons. Figure  shows the MSc cloud-top height over the study area for all seasons. In summer for most of our study area the cloud-top height is less than 2 km, while in winter it is more than 2 km. The cloud-top height in spring and fall is between summer and winter. Since MSc form within the boundary layer, a considerable amount of dust resides above the clouds during summer. We use cloud-top height for those days where the vertical profile of dust extinction coefficient from CALIPSO is available and calculate how much dust is above the top of MSc. The extinction coefficient of dust for each level is obtained from CALIPSO and vertically integrated to calculate DAOD for each grid box, and then extinction coefficients above the CERES cloud-top heights are vertically integrated and divided by DAOD to give the percent of dust above the clouds. The computation is done on a 1∘ × 1∘ grid. Figure  shows that 61.8 ± 12.6 % of the dust resides above MSc during the summer; only 11.9 ± 10.8 % resides above MSc during the winter. In spring (fall) 35 ± 19.8 % (31.2 ± 15.9 %) of the dust resides above MSc. show that during the summer, the Saharan air layer is found to be thicker and higher near Africa at 1–5 km. During winter, it occurs in the altitude range 0–3 km off the western Africa. This is consistent with the vertical profile of Saharan dust in our study. This vertical profile analysis helps to explain the relatively weak first term of Method 1, relative to the second term. It also implies the second term is dominated by the SDE.

To investigate this more, we plot the two partial derivatives that constitute the second term of Method 1. Figures  and show the sensitivity of the cloud fraction and LWP to a relative change in AI for all seasons. Using Eq. (8) to calculate the sensitivity of cloud fraction to a relative change in AI leads to a nonlinear distribution, and thus the statistical significance of the Eq. (8) is evaluated using a bootstrap test. Note that by using Eq. (8) (i.e., only cloud fraction changes mediated by Nd) the effect of absorbing aerosol on meteorology and subsequently cloud cover is suppressed (i.e., a part of the SDE). Figure shows that the sensitivity of cloud fraction to AI is relatively weak. It also shows that this sensitivity is positive (negative) during summer (winter) for most of the study area, which shows that cloud fraction increases (decreases) when AI increases. Figure  shows that the sensitivity of LWP to AI dominates the second term of Method 1. During winter, most of the study area features a reduction in LWP with respect to the AI. During summer, however, this sensitivity is generally positive. Considering the seasonal contrast in the amount of dust above MSc during summer versus winter, the seasonal reversal of these sensitivities – which drive the reversal in the total aerosol–cloud radiative effect – is consistent with the importance of the SDE.

Since the bulk of the dust resides above MSc during summer, aerosol–cloud microphysical interactions (including AIE and CLE) would be muted. Thus, AIE and CLE would be smaller than SDE. Moreover, the SDE would be negative, as observed by the CLE + SDE term of Method 1. also shows that absorbing aerosols overlying MSc largely do not interact with the clouds. However, the aerosols still result in cloud thickening by a dynamical feedback related to the enhanced stability of the atmosphere, which yields an increase in the cloud albedo. This is consistent with , who show that absorbing aerosol above MSc result in increased stability, which strengthens the inversion and reduces cloud-top entrainment of the overlaying dry air, thereby enhancing the underlying clouds. show that the dust found in the atmosphere is substantially coarser than represented in current global climate models. As coarse dust warms the climate, the temperature inversion is stronger and yields thickening of the underlying clouds. show that in response to increased dust load over the tropical North Atlantic in summer, MSc also increase, and this is linked to increases in atmospheric stability, reductions in boundary layer height, and moistening of the lower atmosphere.

During winter, when the total aerosol–cloud radiative effect reverses sign and becomes positive, most of the dust burden resides within or below the clouds. When absorbing aerosol coincides with the cloud, the heating favors cloud clearing and thinning, thus reducing the cloud albedo and yielding a positive radiative effect . In contrast, aerosol indirect effects do not drive cloud clearing/thinning and thus do not contribute a positive radiative effect. Therefore, over our study area, we conclude that the SDE is the most important aerosol–cloud effect resulting in an overall negative radiative effect. The SDE is also strong enough to change the sign of total aerosol–cloud radiative effect from negative to positive during the winter.

To estimate the aerosol–cloud radiative effect of Saharan dust on North Atlantic MSc, we use observational data from several different satellites from 2004 to 2012. The aerosol–cloud radiative effect is estimated using two different methods, following (Method 1) and (Method 2). The annual mean aerosol–cloud radiative effect estimated by Method 1 is -1.5 ± 1.4 W m-2. Estimating the radiative effect using Method 2 yields similar results, with an annual mean of -1.5 ± 1.6 W m-2. Thus, both methods show that Saharan dust modifies MSc in a way that has a cooling effect over the North Atlantic Ocean. Both methods also yield a seasonal maximum negative radiative effect during summer, which is consistent with more Saharan dust and MSc during summer. Furthermore, both methods yield a reversal in the sign of the aerosol–cloud radiative effect, which switches from negative to positive during the winter season. In Method 1, the radiative effect varies from -3.8 ± 2.5 W m-2 during summer to 1 ± 2.9 W m-2 during winter; similarly, Method 2 varies from -4.3 ± 4.1 during summer to 0.6 ± 1 during winter.

Method 1 allows us to separate the cloud albedo effect (first term of Method 1) from the CLE and the SDE (second term of Method 1). The cloud albedo effect, which varies from -0.7 ± 0.6 W m-2 in summer to 0.1 ± 0.5 W m-2 in winter, is relatively small compared to the CLE + SDE, which varies from -3.2 ± 2.5 W m-2 in summer to 0.9 ± 2.9 W m-2 during winter. This shows the importance of the second term, the combination of the CLE and the SDE.

To gain insight as to whether CLE or SDE dominates the second term of Method 1, we use CALIPSO data to quantify the amount of Saharan dust that resides above MSc. The analysis shows that 61.8 ± 12.6 % of Saharan dust resides above MSc during summer, but only 11.9 ± 10.9 % resides above MSc during winter. This seasonal dependence in the location of the dust, relative to MSc, shows the importance of the SDE.

When most dust resides above the clouds during summer, aerosol–cloud microphysical effects that involve the co-location of aerosol and cloud, such as the second aerosol indirect effect (CLE), would likely be muted relative to the SDE. Moreover, the positive value of the aerosol–cloud radiative effect during winter, when most dust resides within MSc, indicates that the SDE is dominant – which is the only mechanism by which a negative aerosol–cloud radiative effect can be obtained. We conclude that aerosol–cloud radiative effects associated with Saharan dust and North Atlantic MSc are dominated by the semidirect effect.