The interest in air pollution over southern West Africa (SWA) has grown substantially in recent years (Knippertz et al., 2015). Population growth, urbanization, and industrialization have led to increasing local emissions from various sources in addition to natural ones. For instance, coastal city development in SWA is leading to more traffic and fuel consumption (Doumbia et al., 2018). Anthropogenic emissions are expected to continue increasing if no regulations are implemented (Liousse et al., 2014). Domestic fires, traffic, and waste burning are the most important sources of pollution in West Africa (Marais and Wiedinmyer, 2016; Bahino et al., 2018; Djossou et al., 2018). In the framework of the Dynamics-Aerosol-Chemistry-Cloud Interactions in West Africa (DACCIWA; Knippertz et al., 2015a) field campaign in SWA during June–July 2016 (Flamant et al., 2018b), measurements of the French Service des Avions Français Instrumentés pour la Recherche en Environnement (SAFIRE) ATR-42 aircraft showed fairly high background concentrations (i.e., outside of urban plumes) with PM1 mass concentrations dominated by secondary organic compounds that contribute 53 % to the total aerosol mass (Brito et al., 2018). For the urban pollution plumes of Abidjan, Accra, and Lomé, they found a doubling of PM1 mass concentrations. Air pollution is a main concern for human health leading to respiratory and other diseases (Lelieveld et al., 2015), but it may also affect local meteorology. For instance, using model sensitivity experiments, Deetz et al. (2018) showed that increasing aerosol loadings can lead to a reduced inland penetration of the Gulf of Guinea maritime inflow (Adler et al., 2019) and a weakening of the nocturnal low-level jet over SWA.

During the summer West African monsoon (WAM), the atmosphere over SWA is characterized by a mixture of pollutants from different sources as highlighted by Knippertz et al. (2017). In addition to the local pollution, long-range transport of dust from the Sahel and the Sahara as well as biomass burning aerosol from central and southern Africa affects the atmospheric composition. Mineral dust has been shown to affect radiation, precipitation, and many WAM features (e.g., Konare et al., 2008; Solmon et al., 2008; Stanelle et al., 2010; Raji et al., 2017; N'Datchoh et al., 2018). During this period, biomass burning is widespread in central and southern Africa, when plumes are carried westward by a jet between 2 and 4 km (Barbosa et al., 1999; Mari et al., 2008), while in West Africa activity peaks during the dry season from October to March (N'Datchoh et al., 2015). Biomass burning is an important source of aerosols and trace gases, with an estimated combined emission of several thousand teragrams per annum (Tg a-1) for tropical areas (Barbosa et al., 1999; van der Werf et al., 2003; Ito and Penner, 2004). For instance the estimated carbon emissions from both tropical fires and fuel wood use was 2.6 Pg C a-1 over the period 1998–2001 (van der Werf et al., 2003). Hao and Liu (1994) estimated that almost half of tropical biomass burning emissions come from Africa, with savanna fires contributing up to 30 % to the global total and 64 % to the African total. During the DACCIWA field campaign, a surprisingly high level of pollution was observed over the sea upstream of SWA. Haslett et al. (2019) found a significant mass of aged accumulation mode aerosol in the planetary boundary layer (PBL) over both continent and ocean. According to modelling work by Menut et al. (2018), biomass burning from central and southern Africa increases the level of air pollution in urban cities such as Lagos and Abidjan by approximately 150 µg m-3 for carbon monoxide (CO), 10–20 µg m-3 for ozone (O3), and 5 µg m-3 for particulate matter with diameters smaller than 2.5 µm (PM2.5).

An important and open question is the mechanism by which the biomass burning aerosols from central Africa get from the layer of midlevel easterlies into the PBL. Das et al. (2017) reported that global aerosol models tend to simulate a quick descent to lower levels just off the western coast of Africa, while Cloud-Aerosol Lidar with Orthogonal Polarization (CALIOP) observations suggest that smoke plumes continue their horizontal transport at elevated levels above the marine boundary layer. The strength and speed of subsidence vary amongst models and subregions. The hypothesis we investigate in this paper is that clouds play a considerable role in the downward mixing of biomass burning aerosol from the elevated plume. Most previous studies have focused on cloud-induced upward transport of aerosols and chemical species from close to their sources in the PBL to the free troposphere (e.g., Dickerson et al., 1987; Ching et al., 1988; Cotton et al., 1995). Using two-dimensional idealized simulations, Flossmann and Wobrock (1996) calculated the mass transport of trace gases across cloud boundaries and from the marine boundary layer into the free troposphere. The atmospheric condition was a well-mixed boundary layer up to about 500 m and above a moist layer of about 2.2 km capped by a very dry stable layer. They found that 60 % of the mass of an inert tracer is exported due to convective clouds. The same paper also examines the transport of SO2 including chemical reactions showing that only a small fraction is dissolved and reacts in the aqueous phase, while substantial amounts of SO2 are redistributed by clouds (see also Kreidenweis et al., 1997). Using a 1-D entraining–detraining plume model with ice microphysics, Mari et al. (2000) studied the transport of CO (inert tracer), CH3COOH, CH2O, H2O2, and HNO3, and they compared the results with observations from the Trace and Atmospheric Chemistry Near the Equator-Atlantic (TRACE-A) campaign. Convective enhancement factors at 7–12 km altitude, representing the ratios of post convective to preconvective mixing ratios, were calculated for both observed and simulated cases. Observed (simulated) values were 2.4 (1.9) for CO, 11 (9.5) for CH3COOH, 2.9 (3.1) for CH2O, 1.9 (1.2) for H2O2, and 0.8 (0.4) for HNO3. Pickering et al. (1996) showed an upward transport of CO , NOx, and hydrocarbon mixing ratios by convective clouds during the Brazilian phase of the TRACE-A experiment. Moreover, Yin et al. (2001) simulated trace-gas redistribution by precipitating continental convective clouds and found abundant highly soluble gases in their uppermost parts, while Halland et al. (2009) showed substantial vertical transport of tropospheric CO by deep mesoscale convective systems.

In contrast, rather few studies investigated the downward transport of elevated pollution through convective clouds. For the marine PBL, aerosol particles from the free troposphere have been identified to serve as cloud condensation nuclei in stratiform clouds with cloud entrainment contributing up to 20 % of the aerosol mass (Raes, 1995; Katoshevski et al., 1999). Over land, most studies concentrated on the Amazon rainforest. Based on campaign data during the wet season, Betts et al. (2002) showed that convective downdraughts rapidly transport air with high ozone down to the surface from around 800 hPa, suggesting a significant role of this process for the photochemistry of the PBL and surface ozone deposition. Gerken et al. (2016) even found evidence for transport of ozone-rich air from the midtroposphere to the surface, enhancing the volume mixing ratio in the boundary layer by as much as 25 ppbv on the regional scale, while Wang et al. (2016) demonstrated the injection of high concentrations of small aerosol particles into the PBL by strong convective downdraughts. In more general terms, Jonker et al. (2008) proposed a refined view of mass transport by cumulus convection relevant for the dispersion of aerosol. According to them, the descending motion near the cloud environment is significant and rather different than in a distant cloud environment, which is characterized by more uniform and quiescent dry descending motion.

This study uses simulations with COSMO (COnsortium for Small-scale MOdelling) (Baldauf et al., 2011) coupled online with Aerosol and Reactive Trace gases (ART; Vogel et al., 2009) to investigate cloud-induced transport of biomass burning aerosols from midlevel tropospheric layers into the PBL over the Gulf of Guinea and SWA. A 1-month simulation for July 2016 (i.e., during the DACCIWA field campaign) over a large domain will be evaluated with available observational datasets and combined with a detailed high-resolution case study, followed by idealized tracer experiments designed to quantify the vertical transport. The paper is organized as follows. Section 2 describes the satellite and reanalysis data as well as the model framework and simulation set-up used for this study. The model evaluation is presented in Sec. 3. In Sect. 4 the downward cloud venting process and its contribution to the vertical mixing of the biomass burning plume are assessed and discussed. Analysis of an artificial tracer to quantify the mass fraction of the biomass burning plume that mixes down into the PBL is given in Sect. 5. The last section presents a summary of the results and conclusions.

Figure 2a shows a July 2016 average of the wind speed and streamlines at 925 hPa as simulated by COSMO-ART. Figure 2b shows the corresponding figure for the ERA-Interim reanalysis. The wind is southeasterly in the Southern Hemisphere and turns southwesterly along the Guinea Coast after crossing the Equator. This low-level monsoon flow advects relatively cool and moist air from the Gulf of Guinea onto the continent. In July the precipitation maximum is located around 10∘ N (e.g., Janicot et al., 2008), and westerlies penetrate far north into the continent and over the adjacent Atlantic Ocean. Apart from a slightly northward-shifted turning point and more fine-scale detail in the higher-resolved COSMO-ART data, the agreement with ERA-Interim in terms of the overall structure of the low-level flow field is good. However, there are some prominent differences in wind speed. ERA-Interim shows highest wind speeds in the Southern Hemisphere and a slow down towards SWA as well as a clear minimum over central Africa. COSMO-ART simulates a stronger monsoon flow and also significantly higher winds over central Africa. Maxima reach 15 m s-1 in both model and reanalysis. COSMO-ART shows a domain average of 6 m s-1, which is 1.4 m s-1 higher than ERA-Interim.

The wind field at 700 hPa is characterized by a broad easterly flow across most of the considered domain (Fig. 2c and d). A maximum is found over the Sahel known as the African Easterly Jet (AEJ), which typically peaks around 600 hPa (Parker et al., 2005) and is the result of the large meridional temperature gradient at low levels (Cook, 1999; Wu et al., 2009). The AEJ is well represented in COSMO-ART with a maximum wind speed of 10.4 m s-1 as compared to 9.73 m s-1 in ERA-Interim. Easterlies are also enhanced near the Equator to the south of an area with weaker flow over the Guinea coast. There are some subtle differences between COSMO-ART and ERA-Interim here, with the former showing a larger northward component over the ocean and slightly stronger winds. COSMO-ART also displays more fine structure in the Southern Hemisphere, where winds are overall weaker. Despite the moderate differences discussed above, we anticipate an overall realistic transport of biomass burning aerosol in the model, i.e., westward away from the hotspots in central Africa out to the Atlantic and then northward into SWA with the monsoon flow, if downward mixing occurs.

The simulated total cloud fraction averaged over July 2016 (Fig. 3a) is compared to observations from MODIS (Fig. 3b). SWA is very cloudy in summer with typical values ranging from 70 % to almost 100 % in agreement with a multi-year climatology presented in Hill et al. (2016). The cloud cover is overall adequately represented by COSMO-ART over land, particularly relative to the poor performance of many coarser-resolution climate models (Hannak et al., 2017). Cloud cover maxima stretch from southwestern Ghana to northeastern Côte d'Ivoire, along the Atakora chain at the border of Ghana and Togo, and over the Guinea Highlands of Liberia and Sierra Leone with overall satisfactory agreement between the two datasets. Towards the Sahel, to the north of 8∘ N, cloud fraction decreases in COSMO-ART but much less so in MODIS, which only shows a prominent minimum over central Côte d'Ivoire. Over the Gulf of Guinea, cloud cover is clearly overestimated by the model, suggesting a potential overestimation of cloud-induced mixing. The two local minima upwind of Abidjan and Accra may be related to coastal upwelling but are hard to verify with MODIS due to the coarser resolution.

Finally, the modelled mixing ratio of surface CO (Fig. 4a) is evaluated with satellite data from MOPITT for July 2016 (Fig. 4b). A gridded monthly mean of CO from MOPITT is computed using the daily-mean CO retrieved for the 1000–900 hPa layer. Some areas have too frequent cloud contamination and therefore do not allow for the computation of a representative monthly mean (white shading in Fig. 4b). Overall the spatial patterns of CO concentration are captured by the model with some discrepancies. Over central Africa widespread burning is evident with a larger magnitude and spatial extent in the model as compared to MOPITT. From there, a plume of enhanced concentrations stretches northwestward in both datasets, but again values in COSMO-ART are somewhat larger and therefore reach more remote parts of the Atlantic Ocean. This also supports a potential overestimation of the import of pollution from central Africa into SWA in the model. In addition, COSMO-ART simulates marked pollution plumes over Nigeria associated with Lagos, the oil fields in the Niger Delta (flaring activities), and the Sahelian city of Kano, which are hard to verify with MOPITT due to cloud contamination. However, emissions from Kano, where clouds are less frequent than in the south, are likely overestimated. Emissions from other large cities (e.g., Accra, Kumasi, Abidjan) in contrast appear relatively weak in COSMO-ART. This may be at least partly due to uncertainties in standard emission inventories (Liousse et al., 2014). Despite some overall discrepancies, we argue that the two fields are similar enough to draw conclusions on the importance of cloud-mixing process in the model, particularly because the fields are relatively similar over the ocean.

In Sect. 3 we present simulated monthly-mean conditions. We will now focus on a case study for 2 July 2016 to illustrate the impact of meteorology on the spatial and temporal distribution of CO. We will especially focus on the role of convective clouds on the vertical distribution of CO.

The discussion in Sect. 4 suggests that long-range-transported biomass burning aerosol from central Africa can be mixed into the PBL over the Gulf of Guinea in association with convective clouds. We will refer to this process as downward cloud venting in the following in contrast to the more classical upward cloud venting (e.g., Cotton et al., 1995).

In general, processes that can support the transport of biomass burning aerosols from free-tropospheric layers into the PBL include (i) large-scale subsidence and thus vertical advection (Katoshevski et al., 1999), (ii) turbulent mixing through the PBL top, and (iii) vertical transport associated with convective clouds. With respect to point (i) we can state that the cross sections in Fig. 8 do not show clear indications of a systematic sinking of the biomass burning plume, suggesting that for the situation presented in Sect. 4 synoptic-scale subsidence is not a leading factor.

To investigate the relative importance of processes (ii) and (iii), we designed an idealized tracer experiment. For the simulations starting at 2 July 2016 at 00:00 UTC, initial profiles of a tracer were prescribed within the D1 and D2 domains. The idealized tracer has a concentration of 1 ppmv between 2 and 4 km and is zero elsewhere. Chemical reactions as well as dry deposition are neglected in order to isolate effects of transport. At the lateral boundaries the tracer concentrations were held constant at the initial profile such that only mixing within the domain can change tracer concentrations. Two types of simulations were done: one with and one without turbulent diffusion of the tracer. The idea behind this is to separate this effect from that of downward cloud venting. The simulations were carried out for a period of 2 d (2–3 July 2016).

For the larger domain D1, Fig. 9 shows the percentage of tracer mass located between 2 and 4 km, between 1 and km, and below 1 km; the latter two with and without turbulent tracer diffusion. As the PBL is usually quite shallow over the ocean (as indicated, for example, by the low cloud base in Fig. 8d), the lowest layer should in most cases comprise the entire PBL and possibly also the lower part of the free troposphere with some variations in space and time. All values are averaged between 9∘ W and 1∘ E, and the different panels show time evolutions along different latitude circles.

Over the open ocean at 5 and 0∘ S (Fig. 9a, b), where mostly shallow cumuli are present, the concentration in the layer between 2 and 4 km stays fairly high over the 2 d period with well over 80 % still present at the end of the simulation. The increase in the tracer mass in the intermediate layer from 1 to 2 km starts almost immediately after the beginning of the simulation and reaches a plateau well above 20 % during day 2. This indicates that this increase cannot solely be the result of vertical mixing near the shown latitude circle but must also be related to horizontal transport of tracer that was mixed downwards upstream. The final tracer amount is almost independent of whether turbulent diffusion is considered or not, indicating the importance of downward cloud venting (compare dashed and solid red lines in Fig. 9a and b). The tracer reaches the layer below 1 km at the end of the second day parallel to a marked increase in the layer above. The tracer mass then slowly increases to reach final values ranging around 15 %. Here, the vertical mixing clearly is a combination of turbulent diffusion and cloud-induced mixing with the former contributing on the order of one-quarter to one-third. It is interesting to note that the southernmost section has a higher percentage of cloud-induced mixing than that over the equatorial cold tongue. Below we will show evidence that the higher SST in the former region likely supports cloud formation and associated mixing.

Over the Gulf of Guinea between 4 and 4.7∘ N (i.e., the latitude range used for Fig. 10) there is a more pronounced decrease in the tracer mass in the layer between 2 and 4 km down to about 65 % after 2 d (Fig. 9c). Consistently, the 1–2 km layer gains more tracer mass and exceeds 30 % on the second day with a more continuous rise than farther south (Fig. 9a and b). Turbulent diffusion appears to play a more important role here, but the overall contribution is still fairly small. The increase in mass below 1 km, however, does not match these differences, reaching values similar to that over the open tropical Atlantic but with a similar contribution from cloud-induced mixing of about 60 %. This result illustrates the complicated configuration of differential advection at different height levels combined with spatially differing vertical transport.

To investigate the aspect of SST influence further, Fig. 10 shows the final mixing state in the layer below 1 km after 2 d of integration, i.e., the right-hand-side intercept of the black curves in Fig. 9 but for steps of 1∘ latitude. Plotted against SSTs for the same longitudinal range, a close correspondence is evident in the Southern Hemisphere. Both mixing and SSTs have a marked maximum around 6–7∘ S followed by a common minimum around 2∘ S. To the north, SSTs increase to levels even higher than in the Southern Hemisphere but the tracer mass increases only little. One likely reason for this is the smaller tracer concentrations aloft as evident from Fig. 9c. Another factor may be the flow of near-surface air over the cooler equatorial water, leading to a decrease in buoyancy. It is also possible that enhanced shallow subsidence closer to the coast (see Fig. 8f) helps suppress vertical mixing into the PBL.

Finally over land, i.e., between 5 and 10∘ N, the vertical exchange maximizes leading to a reduction in the 2–4 km layer down to almost 50 % (Fig. 9d). Consistently, tracer mass in the intermediate layer increases more strongly up to well over 44 %, while tracer mass below 1 km reaches 23 %. A clear diurnal cycle is evident, particularly in the lower layer, with vertical mixing mostly occurring in the early afternoon when the PBL is deepest. The suppressing of turbulent diffusion reduces the tracer mass by 20 % with some evidence of a diurnal cycle in the differences evident from Fig. 9d. As expected, dry mixing is more important in the early afternoon, while cloud-induced mixing peaks later. The important role of clouds in vertical mixing over land is consistent with the large cloud cover shown in Fig. 3. The diurnal cycle is also evident at 1–2 km, where switching off turbulent diffusion leads to a net increase in this layer during the afternoon.

Figure 11 shows results analogous to Fig. 9 but for the model domain D2 and for the areas A–D indicated in Fig. 5. Their latitudinal extent corresponds to Fig. 9c, but the longitudinal extent varies between the different panels. Taken together, the four panels of Fig. 11 stretch from 7∘ W to 1∘ E, while Fig. 9c stretches from 9∘ W to 1∘ E. Zooming down into such small areas illustrates the impact of short-lived intense local mixing events. While all four subregions show a marked decline in tracer mass in the 2–4 km layer (blue lines), the evolution is sometimes bumpy and final values range from 41 % in Fig. 11a to about 74 % in Fig. 11c and d. Mixing into the layer immediately below the biomass burning plume (i.e., 1–2 km, red lines) begins a few hours after the start of the simulations with even a transient reduction in some areas, mostly during morning hours. It is possible that the diurnal cycle in monsoon flow and cloudiness over the near landmass contributes to such fluctuations. Some correspondence is seen between “loss” events in the upper-layer and “gain” events in the middle layer, but the connection is not always clear cut, indicating effects of horizontal transport and possibly also mixing upwards from the main aerosol layer. As the 1–2 km layer is above the PBL in most cases, it is no surprise that switching off turbulent diffusion has very little affect, apart from the easternmost domain where clouds appear to be rather inactive (compare red dashed and solid lines in Fig. 11).

Finally in the layer below 1 km tracer concentrations are fairly low throughout most of the period in all four subregions. At the end of 2 July moderate increases are seen in the two westernmost domains. These appear to be related to the showers discussed in the context of Figs. 5–8 in the previous section. In the afternoon and evening hours of 3 July a more marked mixing event occurs in the two western domains with concentrations peaking around 18:00 UTC. Reflections of these events can be seen in the upper layers as well. Turbulent diffusion practically plays no role in the mass increase in the lower layer. The event on 3 July contributes most to the large-scale increase seen in Fig. 9c. This discussion illustrates the influence of localized intense mixing events on tracer concentrations in the PBL, which will largely be missed out by models that parameterize moist convection.

Recent observational and modelling work has revealed significant concentrations of biomass burning aerosol reaching SWA in the PBL and contributing to a deterioration of air quality there (Brito et al., 2018; Menut et al., 2018; Haslett et al., 2019). It has been suggested that this plume stems from the extensive fires in central Africa during the WAM season. Here we investigated potential transport pathways of the aerosol. While previous studies discussed subsidence to the west of the African continent to be an important mechanism, here we identify – to the best of our knowledge – for the first time that downward cloud venting is one of the processes by which biomass burning aerosol from middle tropospheric layers is mixed into the PBL over the Gulf of Guinea.

This study heavily relied on high-resolution simulations using the COSMO-ART model for July 2016. COSMO-ART enables us to simulate both meteorological fields and CO distributions over a domain including SWA, central Africa, and the adjacent tropical Atlantic. The simulated wind speed and direction are broadly in agreement with the ERA-Interim reanalysis, although COSMO revealed a somewhat stronger midlevel export from central Africa and faster monsoon flow. Regarding cloud cover, COSMO-ART reproduces areas of maximum and minimum clouds over SWA but overestimates it over the Gulf of Guinea. The spatial distribution of CO is used as a tracer to detect the biomass burning plume. Compared to observations, the simulated CO concentration captures the main spatial patterns, but the central African biomass burning plumes appear to be overestimated.

For a particular case study (2 July 2016), we conducted simulations for realistic conditions and for idealized experiments, with a passive tracer initially restricted entirely to the 2–4 km layer. The main results are schematically illustrated in Fig. 12, showing an ensemble of clouds with different vertical extents. The biomass burning aerosol is first transported out to the tropical Atlantic with a strong easterly flow at midlevels (yellow arrow in Fig. 12). Its base is often well above the usually shallow oceanic PBL (sketched to reach 800 m in Fig. 12). Both turbulent diffusion and cloud-induced mixing cause a vertical transport to below 1 km with the latter contributing more than two-thirds over most areas. Individual cloud-induced mixing events can be detected that are associated with deeper clouds, precipitation, and downdraughts, leading to surface cold pools. Concentrations of the biomass burning plume aerosol below 1 km reach about 15 % of the initial mass in the 2–4 km layer after 2 d in our tracer experiments. Details of the mixing depend crucially on cloud depth and precipitation intensity (as indicated by solid and stippled arrows in Fig. 12). Once in the lower layers biomass burning can be carried northward with the southerly or southwesterly monsoon winds (indicated by yellow arrows in Fig. 12). It is conceivable (but not shown here) that the strong shear between near-surface southerlies and midlevel easterlies helps tilting convective clouds when they form and thereby increases evaporation at cloud edges, downdraught formation, and mixing. In addition, we found a meridional gradient in the effectiveness of downward transport irrespective of the actual sources of biomass burning aerosol. The largest PBL input occurs over the warm waters of the Southern Hemisphere (around 7∘ S) with a marked decrease towards the equatorial cold tongue. Towards the north, SST increase again but mixing efficiency does not reach the same levels as in the Southern Hemisphere, possibly due to differences in vertical stability.

This study is largely based on a case study to illustrate the potential importance of downward cloud venting over the Gulf of Guinea. Further investigations are needed based on longer simulations and other models to get more robust statistics. Moreover, the tracer experiment we present here was performed for an inert tracer such as CO with no sedimentation and no deposition. The central African biomass burning plume contains large amounts of aerosols, which do sediment and can be washed out by rainfall into the ocean. The magnitude of this, however, remains an open question that needs to be addressed in future studies.