Comparing the simulated influence of biomass burning plumes on low-level clouds over the southeastern Atlantic under varying smoke conditions

Abstract

Biomass burning plumes are frequently transported over the southeast Atlantic (SEA) stratocumulus deck during the southern African fire season (June–October). The plumes bring large amounts of absorbing aerosols and enhanced moisture, which can trigger a rich set of aerosol–cloud–radiation interactions with climatic consequences that are still poorly understood. We use large-eddy simulation (LES) to explore and disentangle the individual impacts of aerosols and moisture on the underlying stratocumulus clouds, the marine boundary layer (MBL) evolution, and the stratocumulus-to-cumulus transition (SCT) for three different meteorological situations over the southeast Atlantic during August 2017. For all three cases, our LES shows that the SCT is driven by increased sea surface temperatures and cloud-top entrainment as the air is advected towards the Equator. In the LES model, aerosol indirect effects, including impacts on drizzle production, have a small influence on the modeled cloud evolution and SCT, even when aerosol concentrations are lowered to background concentrations. In contrast, local semi-direct effects, i.e., aerosol absorption of solar radiation in the MBL, cause a reduction in cloud cover that can lead to a speed-up of the SCT, in particular during the daytime and during broken cloud conditions, especially in highly polluted situations. The largest impact on the radiative budget comes from aerosol impacts on cloud albedo: the plume with absorbing aerosols produces a total average 3 d of simulations. We find that the moisture accompanying the aerosol plume produces an additional cooling effect that is about as large as the total aerosol radiative effect. Overall, there is still a large uncertainty associated with the radiative and cloud evolution effects of biomass burning aerosols. A comparison between different models in a common framework, combined with constraints from in situ observations, could help to reduce the uncertainty.

1 Introduction

Biomass burning aerosols (BBAs) influence the Earth's weather and climate , but their net impact is difficult to quantify because of the complex interactions between aerosols, radiation, and clouds. The two main chemical components of BBAs are pure black carbon (BC) and organic carbon (OC): the former predominantly absorbs and the latter mostly scatters solar radiation . The combined absorption and scattering properties of BBAs (direct aerosol effect) modify the radiative fluxes in the atmosphere, leading to temperature changes and, in general, to an energy redistribution that affects atmospheric stability and alters cloud evolution and precipitation (semi-direct aerosol effect). Moreover, BBAs can also serve as cloud condensation nuclei (CCN) and ice nuclei (INP) and alter the radiative properties of clouds and their lifetime (indirect effects; ).

Several tropical regions on the planet experience biomass fires every year that release considerable amounts of BBAs to the atmosphere, affecting not only local but also remote areas . The southeast Atlantic (SEA) stands out as one region receiving substantial quantities of BBAs from June to October . These aerosols originate from fires over the southwestern African savanna that are transported westwards by the predominant trade winds. In addition, the SEA hosts one of the largest semi-permanent stratocumulus (Sc) cloud decks on the planet, providing an ideal environment for frequent interactions between BBAs, Sc clouds, and radiation to occur. Another notable characteristic of the SEA is the typical occurrence of stratocumulus-to-cumulus transitions (SCTs) when air masses flow towards the Equator on the eastern side of the semi-permanent South Atlantic subtropical anticyclone . The so-called deepening–warming mechanism suggests that SCTs are primarily driven by increasing sea surface temperatures (SSTs) as the air moves from the subtropics to the tropics. Warmer SSTs lead to increased surface latent heat fluxes that increase the turbulent kinetic energy (TKE) and enhance turbulence in the boundary layer. The TKE increase strengthens the cloud-top entrainment which deepens the boundary layer. If the mixture between the warm entrained air and the cloudy air is positively buoyant, the mixing between the cloud and the sub-cloud layers is reduced. The latter eventually leads to a decoupled boundary layer, where the flux of surface moisture to the Sc cloud is diminished and the formation of cumulus clouds below the Sc is favored. According to this theory, the speed of the SCT is mostly modulated by the strength of the capping inversion of the marine boundary layer (MBL), whereas precipitation formation is less important . However, several recent studies have found that drizzle formation may indeed be highly relevant for SCTs, especially when aerosol concentrations are low and when feedbacks between aerosols, cloud droplet number, and precipitation are considered .

BBAs are frequently transported in the free troposphere (FT) over the SEA, above the Sc-topped MBL. The combined direct and semi-direct effects in the FT can impact the Sc and the MBL, mostly when the distance between the aerosol plume and the clouds is small . For instance, the absorption of SW radiation by the aerosols can warm the FT and strengthen the inversion at the top of the MBL, slowing down its deepening. found that the SW absorption in the FT can also reduce the subsidence rate, which can modulate the timing for potential contact between the aerosol plume and the underlying Sc clouds. Studies based on satellite data have also shown an increase in cloud cover and cloud thickness in the presence of BBAs in the FT . The BBA plumes over the SEA are typically accompanied by enhanced moisture originating from the continental boundary layer . The enhanced moisture transport is not directly caused by the biomass burning itself but happens to coincide with the transport of BBAs . Water vapor absorbs mainly at near-infrared and infrared wavelengths longer than 0.7 $\mathrm{\mu }$m . Thus, by increasing the downward longwave (LW) fluxes, moisture associated with BBAs in the free troposphere can suppress MBL deepening by reducing the net Sc top LW cooling, as has been shown in large-eddy simulations (e.g., ).

BBAs can also entrain and mix into the MBL, affecting clouds, precipitation, and the MBL evolution through aerosol indirect and semi-direct effects. The indirect effects resulting from a BBA plume can be manifested as a chain of processes. Following and , an increase in the cloud droplet number concentration ($N_{\mathrm{c}}$) can result in a reduction in cloud droplet size, leading to a higher cloud albedo and eventually precipitation suppression. This can potentially extend the cloud's lifetime and increase the cloud depth and liquid water path (LWP). However, precipitation suppression can also increase cloud-top radiative cooling, leading to an increase in the entrainment rate and a reduction in the LWP . The semi-direct effect of BBAs within the MBL can manifest itself as a temperature increase that leads to a reduction in the relative humidity (RH) and, consequently, a decrease in cloudiness . In addition to the above, the moisture associated with BBA plumes can result in that relatively humid air being entrained into the MBL, which can lead to an increase in cloud cover with increasing entrainment, instead of the opposite that typically would occur for a clean, dry free troposphere .

The aerosol amount and humidity tend to co-vary within BBA-moist plumes, and their individual effects on MBL evolution and Sc clouds are therefore difficult to disentangle using observational data, e.g., from satellites . Their relative impacts might also vary depending on the meteorological situation and the magnitude of the perturbations . Therefore, a modeling perspective may be useful for examining the issue. used Lagrangian large-eddy simulations and an idealized SCT case to explore how SCTs can be influenced by a plume of enhanced moisture and smoke. They found that cloud-top entrainment and MBL deepening were reduced when the smoke layer was located above the Sc deck, due to smoke absorption and a strengthening of the inversion. When the plume entrained into the MBL, drizzle was suppressed, which, together with the enhanced moisture associated with the aerosol plume, contributed to Sc cloud sustenance and a delay in the SCT. With some modifications, also used the SCT case study. However, they obtained an acceleration of the SCT when the aerosol plume made contact with the Sc deck, due to an increased $N_{\mathrm{c}}$, smaller droplets, more evaporative cooling, and enhanced cloud-top entrainment. Note that the simulations by did not include prognostic aerosol concentrations, i.e., their model setup would not be able to produce any drizzle-driven acceleration in the SCT, which could explain some of the differences compared to .

One disadvantage of the studies by and is that they both used idealized meteorological conditions, representative of the northeastern Pacific. In contrast, used data from the joint ObseRvations of Aerosols above CLouds and their intEractionS (ORACLES) and CLouds–Aerosol–Radiation Interaction and Forcing: Year 2017 (CLARIFY-2017) campaigns to simulate an SCT case over the SEA. In their simulations, they also incorporated the effect of smoke on the large-scale circulation by forcing their large-eddy simulation (LES) model with output from a regional climate model. found that the large-scale thermodynamic and dynamic adjustments, which cannot be explicitly simulated by an LES model, had the largest impact on the SCT except when aerosol concentrations were very low. This delay was to some extent counteracted by local (within the MBL) semi-direct effects that decreased cloud cover. However, similarly to and , found that local semi-direct effects did not dominate the impact of BBAs on cloud evolution.

To summarize, BBA plumes and associated moisture (either overlying or mixed into the MBL) can influence the MBL, Sc clouds, and, consequently, SCTs over the SEA in multiple and sometimes counteracting ways. The complexity of the interactions has caused disagreement between previous modeling studies using large-eddy simulation. Most of these studies were based on idealized meteorology. Only one study has examined conditions representative of the SEA, and they focused on only one specific case . Since the location, timing, and levels of pollution of the BBA plumes influence the cloud and MBL evolution, the analysis of different situations can give a wider perspective of the possible ways in which the humid BBA plumes can affect low-level clouds over the SEA.

In this work, we use the MISU-MIT Cloud and Aerosol (MIMICA) LES model to simulate stratocumulus-topped boundary layers and SCTs for three different meteorological situations over the SEA, during August 2017, characterized by the presence of moist absorbing aerosol plumes interacting with the MBL. The three situations are chosen to study cases that are clearly different in the levels of pollution and moisture in the FT and also with respect to when the BBA plume appears in the FT and when it mixes into the MBL. We explore the individual influence of aerosols and moisture on the diel (24 h) cycle of the MBL, the Sc clouds, and on the SCT. Furthermore, we compare the impacts observed in the three situations and investigate the overall radiative effects of the absorbing aerosols and the enhanced moisture within the BBA plume. In Sect. , we describe the model setup and define some variables and parameters used in the analysis. Section  describes the results followed by a discussion and conclusions in Sect. .

2 Methods

2.1 Model setup and simulations

The MIMICA LES code solves a set of anelastic, non-hydrostatic governing equations. The model uses a two-moment bulk microphysics scheme where the mass mixing ratios and the number densities of five hydrometeor types (i.e., cloud droplets, raindrops, ice crystals, snow, and graupel) are defined as prognostic variables. In this work, we only use the microphysics associated with warm (no-ice) clouds and precipitation and use a radius of 25 $\mathrm{\mu }$m to separate cloud droplets from raindrops. Hydrometeor size distributions are defined by gamma functions. The model is coupled to a version of the four-stream Fu–Liou–Gu radiative transfer model which uses 6 bands for shortwave radiation (solar spectrum between 0.2 and 5 $\mathrm{\mu }$m) and 12 bands for longwave radiation (infrared spectrum between 2200 and 10 cm${}^{-\mathrm{1}}$). A two-moment aerosol module is used to characterize the aerosol population , with the possibility to define several log-normally distributed aerosol modes with different composition and size. Each aerosol mode is composed of a single aerosol type or a combination of four aerosol types (black carbon, organic carbon, sulfate, and sea salt). Aerosols can act as CCN following the ($\mathit{\kappa }$)–Köhler theory and are lost through cloud processing and precipitation. However, the model tracks the mass of activated aerosols and generates one aerosol particle from each evaporated droplet. When collision–coalescence processes occur, the model makes a simple assumption that the aerosols from the involved droplets merge into one upon evaporation.

To perform the simulations, we added explicit aerosol–radiation interactions to MIMICA. The implementation is an adaptation of the aerosol–radiation interactions used by in UCLALES–SALSA . MIMICA and UCLALES–SALSA share the same radiative transfer model , but in MIMICA the optical properties (optical thickness, single-scattering albedo, and phase function) are estimated for aerosol modes, whereas in UCLALES–SALSA they are estimated for aerosol bins. The real and imaginary refractive indices for each aerosol component are obtained from . The mean refractive index of the aerosol mode will be proportional to the volume fraction of each aerosol type.

For our simulations, we use a 9.6 km by 9.6 km horizontal domain with a resolution of 50 m. The vertical grid consists of 288 grid points from the surface to 6.5 km with a resolution of 10 m below 2.5 km and a vertical stretching above (vertical distance between grid points increases by 10 % each level). Aerosol properties described above are based on ORACLES-2017 measurements and follow . BBAs are represented by using a combination of black carbon (6.8 %) and organic carbon (93.2 %) in a single mode: the single-scattering albedo (SSA) of this internal mixture is approximately equal to 0.85. The initial geometrical mean diameter of the aerosol distribution is 185 nm with a fixed geometric standard deviation of 1.5. The hygroscopicity parameter ($\mathit{\kappa }$) is set to 0.2, consistent with , , and . We use a constant surface aerosol source of 70 cm${}^{-\mathrm{2}}$ s${}^{-\mathrm{1}}$ to maintain the background aerosol number concentration ($N_{\mathrm{a}}$) within reasonable values . For simplicity, the aerosol source is assumed to consist only of BBAs, as in and . The horizontal wind divergence rate is set to a constant value of $\mathrm{2.16}\times {\mathrm{10}}^{-\mathrm{6}}$ s${}^{-\mathrm{1}}$ in all cases.

The simulations are initialized (in the entire model domain) and later forced (only in the free troposphere), with meteorological fields and aerosol conditions given by trajectories calculated with the Goddard Earth Observing System model Version 5 (GEOS-5) . In the trajectories, the parcels' initial longitudes span from 0 to 12° E, while their initial latitude is set at 25° S. A fixed vertical level of 250 m for the parcels is assumed, disregarding any vertical movement. At a specific point, the horizontal wind velocities are calculated by linearly interpolating values from 16 neighboring spatiotemporal ($x$, $y$, $z$, and time) points. To calculate the incremental changes in the parcels' locations after each integration on the latitude and longitude coordinate, equatorial and polar radii of 6378.137 and 6356.752 km, respectively, are used. The model time step is 10 min in GEOS-5, and the simple forward Euler method is used for time integration. When the parcels approach land too closely to obtain the 250 m winds, the integration stops.

During MIMICA's simulations, the free troposphere is continuously nudged to the forcing (from GEOS-5 trajectories) temperature, $N_{\mathrm{a}}$, and mass fraction of water in air (water vapor mixing ratio) on a timescale of 30 min and on a timescale of 3 h to the horizontal winds. Note that this setup means that the meteorological fields from the GEOS-5 forcing data will always include FT semi-direct aerosol effects, as there are no GEOS-5 simulations where aerosol radiative effects are turned off . We define the nudging base as the maximum inversion height in the model domain plus 100 m and calculate the inversion height (or FT base) as the maximum vertical gradient of the potential temperature in each model column. The model is also forced with sea surface temperature (SST) values from GEOS-5 (Fig. a).

Figure 1

(a) Time series of sea surface temperature (SST) along the Lagrangian trajectories from GEOS-5. (b) Geographical location of the Lagrangian trajectories from GEOS-5.

[Figure omitted. See PDF]

We use three individual trajectories (cases) to force the LES, corresponding to periods starting on 3 different days (3, 16, and 31 August 2017; see Table ). The motivation to select these days is explained in Sects.  and . The trajectories pass over Ascension Island (8° S, 14.4° W) between 10:30 and 13:30 UTC on a specific day. In our simulations we only use the section of the trajectories within 64 h before and 3 h after noon (or closest available time) when they pass over Ascension Island (Fig. b).

Table 1

Trajectories used for the model simulations.

Trajectory ID	Start date	Start time	Latitude
AUG03	3 Aug 2017	21:00 UTC	11.11° S
AUG16	16 Aug 2017	21:00 UTC	11.60° S
AUG31	31 Aug 2017	21:00 UTC	11.10° S

For each case (or trajectory), four simulations are carried out: the control (CTRL) with the original values for each variable in the trajectory; an experiment called N100 with a fixed $N_{\mathrm{a}}=\mathrm{100}$ mg${}^{-\mathrm{1}}$ in the FT (this $N_{\mathrm{a}}$ value is among the lowest in the FT for AUG03, which is the cleanest among the three cases; see Sect. ); an experiment with no aerosol–radiation interactions (Aer-rad-off); and an experiment where the water vapor mixing ratio ($Q_{\mathrm{v}}$) in the FT was reduced to values between 0.1 and 0.4 g kg${}^{-\mathrm{1}}$ (DRY simulation). This range of $Q_{\mathrm{v}}$ values occurs in AUG03 CTRL at the beginning of the simulation when the FT is relatively dry. In total, we have 12 different simulations. In the experiments corresponding to the same day, the FT is always nudged (above the nudging base) towards the same forcing values of temperature and horizontal winds. Thus, changes in $N_{\mathrm{a}}$ (N100 experiment) or moisture (DRY experiment) or turning off the aerosol scattering and absorption (Aer-rad-off) are not going to affect the temperature and winds in the FT above the nudging base. This type of model setup also means that aerosol–radiation interactions and associated cloud adjustments will only be fully effective below the nudging base (the MBL and the lower 100 m of the FT).

The differences between the Aer-rad-off and the CTRL experiments are used to evaluate the direct and associated semi-direct aerosol effects in the MBL. Contrasts between the DRY and CTRL experiments are a consequence of the radiative impact of moisture in the FT on the MBL and the moisture entrainment into the MBL. By comparing Aer-rad-off and N100, we can obtain an estimate of the indirect aerosol effect. Finally, the differences between N100 and CTRL are due to all aerosol effects combined.

2.2 Parameters and variables used in the analysis

Here we briefly define some parameters and variables that we use in the analysis.

• Cloud cover is defined as the fraction of model columns with $\mathrm{LWP}>\mathrm{0.01}$ kg m${}^{-\mathrm{2}}$ .

• Vertically resolved cloud cover is the fraction of total columns at each model level with a cloud droplet mixing ratio of $Q_{\mathrm{c}}>\mathrm{0.001}$ g kg${}^{-\mathrm{1}}$ (following , who call this variable cloud fraction).

• Marine boundary layer turbulent kinetic energy (MBL TKE) is the domain-averaged TKE between the surface and the height of the inversion capping the MBL.

• The stratocumulus-to-cumulus transition (SCT) is defined as the time at which the cloud cover first decreases to half of its initial value .

• The decoupling parameter ($\mathit{\delta }{Q}_{\mathrm{t}}$) is calculated as the difference between the total water mixing ratio ($Q_{\mathrm{t}}$) from the bottom and the top 25 % of the MBL, according to and .

• Aerosol radiative effect at the top of the model domain is calculated as:

  • a.

    $F_{\mathrm{CTRL}}-F_{\mathrm{Aer}\text{-}\mathrm{rad}\text{-}\mathrm{off}}$ due to direct and semi-direct aerosol effects,

  • b.

    $F_{\mathrm{Aer}\text{-}\mathrm{rad}\text{-}\mathrm{off}}-F_{\mathrm{N}\mathrm{100}}$ due to indirect aerosol effects (we assume a negligible semi-direct aerosol effect because the aerosol absorption in N100 is small),

  • c.

    $F_{\mathrm{CTRL}}-F_{\mathrm{N}\mathrm{100}}$ due to all aerosol effects combined,

  where $F$ is the net incoming SW or LW radiation.

• Aerosol clear sky radiative effect at the top of the model domain (TOA) is derived from ${\mathrm{FCS}}_{\mathrm{Aer}\text{-}\mathrm{rad}\text{-}\mathrm{off}}-{\mathrm{FCS}}_{\mathrm{CTRL}}$, where FCS is the net outgoing SW or LW radiation with clear sky (no clouds).

• The radiative effect due to the enhanced moisture in the aerosol plume is calculated as the difference between the outgoing radiative fluxes in the DRY and the CTRL experiments: $F_{\mathrm{CTRL}}-F_{\mathrm{DRY}}$.

• The cloud radiative effect (CRE) at the top of the model domain is calculated as the difference between the upwelling clear-sky SW (or LW) fluxes and the upwelling all-sky SW (or LW) fluxes.

• The entrainment rate is calculated, following and , as the difference between the subsidence (large-scale divergence rate) at the top of the inversion and the change in the inversion height with time.

• Entrainment fluxes of aerosol and moisture at the top of the MBL are calculated by multiplying $N_{\mathrm{a}}$ and the mixing ratio of specific moisture ($Q_{\mathrm{v}}$) with the entrainment rate, respectively.

2.3 Observations

We evaluated the simulated time evolution and diel cycle of cloud cover and liquid water path using geostationary satellite sensor data. More specifically, we used cloud cover and LWP from SEVIRI retrieved by NASA's Langley Research Center using the Satellite Cloud and Radiation Property Retrieval System algorithms . Both variables are averaged in boxes of 0.185° diameter around each data point in the GEOS-5 trajectories. LWP retrievals are screened to only include data with solar zenith angles below 70° and cloud fractions above 90 %. We complement this information with retrievals from the latest (third) edition of the CM SAF CLoud property dAtAset using SEVIRI (CLAAS-3; ). CLAAS-3 provides high-resolution (0.05° in space and 15 min in time) cloud property retrievals, which is useful when evaluating the model performance in terms of simulating the time evolution of clouds.

3 Results

3.1 Control simulations

As highlighted in Sect. , the three cases analyzed (AUG03, AUG16, and AUG31) differ in the levels of pollution and moisture in the FT and in terms of when the BBA plume appears in the FT and when it mixes into the MBL (see Figs. –). Nevertheless, there are some characteristics that are common for all simulated CTRL cases. Firstly, there is a visible covariance between $N_{\mathrm{a}}$ (Fig. ) and specific humidity (Fig. ) within the FT BBA plumes, as typically happens over the SEA (see Sect. ). Secondly, some well-known features of the MBL diel cycle can be observed in all three cases: during the nighttime, there is an increase in the cloud cover (Fig. ); in the vertically resolved cloud cover (Fig. ); in the LW cooling at the top of the MBL (Fig. a–c); and in the MBL TKE (Fig. d–f). There is also a fast increase in the height of the MBL inversion with time (Fig. j–l). During the daytime, the cloud cover and liquid water path (Fig. ) decrease at the same time as there is a reduction in the MBL TKE and a slower deepening of the MBL compared to nighttime conditions. Cumulus clouds start to develop under the Sc after some time in the simulations: around sunrise of the second day in AUG03, near the beginning of the simulation in AUG16, and after sunset of the first day in AUG31 (Fig. ). There is a general increase in the MBL decoupling during or after the second night (Fig. m–o). The second day is characterized by a sharp decrease in the cloud cover and the vertically resolved cloud cover in all CTRL cases. The SCT, according to the criteria defined in Sect. , happens on the second day in all CTRL simulations: around midday in AUG03 and AUG16 and in the afternoon in AUG31. Precipitation is small in all CTRL cases (Fig. g–i), which suggests that, for these cases, drizzle formation does not have a substantial impact on the SCT in MIMICA. The aforementioned description of the MBL evolution in the three situations is consistent with a deepening–warming type of SCT, primarily driven by the SST increase (Fig. a). The diurnal CRE is dominated by the SW fluxes (Fig. ). The domain-averaged CRE (Fig. ) becomes less negative with each day of simulation as a result of the reduction in cloudiness.

Figure 2

Temporal evolution of simulated profiles of the horizontally averaged aerosol number concentration ($N_{\mathrm{a}}$) from MIMICA. Results are shown for all experiments (CTRL, Aer-rad-off, N100, and DRY) and all cases (AUG03, AUG16, and AUG31). Dashed lines separate nighttime ($N$ in the upper subplots) from daytime (D). $Z$ is the altitude.

[Figure omitted. See PDF]

Figure 3

As in Fig.  but for the water vapor mixing ratio ($Q_{\mathrm{v}}$). Simulations are shown for AUG03, AUG16, and AUG31.

[Figure omitted. See PDF]

Figure 4

As in Fig.  but for the vertically resolved cloud cover (VRCC). Simulations are shown for AUG03, AUG16, and AUG31.

[Figure omitted. See PDF]

Figure 5

(a–c) Cloud cover and (d–f) liquid water path along the three trajectories as a function of time from SEVIRI (NASA) and the MIMICA simulations (liquid water path is in-cloud in MIMICA). The cloud cover and liquid water path output from MIMICA have a time resolution of 15 min, and values are smoothed using a 1 h moving average.

[Figure omitted. See PDF]

Figure 6

Temporal evolution of the simulated (MIMICA) domain-averaged (a–c) LW cooling rate (DT LW) at the MBL top (corresponds to the maximum horizontal domain-averaged LW cooling rate), (d–f) MBL TKE, (g–i) cloud-top entrainment rate ($W_{\mathrm{e}}$), (j–l) inversion height ($z_{\mathrm{i}}$), and (m–o) decoupling parameter ($\mathit{\delta }{Q}_{\mathrm{t}}$). All output variables have 15 min temporal resolution. DT LW and TKE values are smoothed using a 1 h moving average. The entrainment rate is smoothed with a 2 h moving average.

[Figure omitted. See PDF]

Figure 7

Temporal evolution of the simulated (MIMICA) domain-averaged (a–c) cloud droplet number concentration ($N_{\mathrm{c}}$), (d–f) cloud droplet mean size ($D_{\mathrm{c}}$), and (g–i) accumulated precipitation ($R_{\mathrm{accum}}$). The output of $N_{\mathrm{c}}$ and $D_{\mathrm{c}}$ have a 4 h time resolution, while the temporal resolution for the $R_{\mathrm{accum}}$ output is 15 min.

[Figure omitted. See PDF]

Figure 8

Simulated (MIMICA) domain-averaged diurnal SW and LW cloud radiative effects (in W m${}^{-\mathrm{2}}$) in AUG03 (a, d, g, j), AUG16 (b, e, h, k), and AUG31 (c, f, i, l). The total radiative effect (SW $+$ LW) is the number above (if positive) or below (if negative) the stacked bars. Note that this number (total radiative effect) does not match the values on the $y$ axis since it is the sum or difference of the SW and LW bars.

[Figure omitted. See PDF]

Despite the similarities, there are also clear differences between the three cases. AUG03 is initially characterized by a relatively clean FT (Fig. ). Around midday of the second day and until the end of the simulation, the BBA plume becomes apparent in the FT, approximately between an altitude of 4000 m and the top of the MBL. The maximum values of the domain-averaged $N_{\mathrm{a}}$ in the plume are around 600 mg${}^{-\mathrm{1}}$ and are reached towards the end of the simulation near the MBL top. The entrainment of aerosol in the MBL remains relatively low during the whole simulation compared to in AUG16 and AUG31 (Fig. a–c).

Figure 9

Temporal evolution of the simulated (MIMICA) domain-averaged aerosol ($N_{\mathrm{a}}$) and water vapor mixing ratio ($Q_{\mathrm{v}}$) fluxes at the top of the MBL for AUG03, AUG16, and AUG31.

[Figure omitted. See PDF]

AUG16 starts with a strong and humid aerosol plume in the FT with average $N_{\mathrm{a}}>\mathrm{3000}$ mg${}^{-\mathrm{1}}$ near the MBL top. The BBA plume is already in contact with the MBL top at the beginning of the simulation and eventually entrains into the MBL, resulting in a relatively clean FT and a very polluted MBL (with $N_{\mathrm{a}}>\mathrm{1000}$ mg${}^{-\mathrm{1}}$) during the second half of the simulation. This situation differs from the cases analyzed by , , and in which the absorbing aerosol layer in the FT is initially clearly separated from the cloud layer and only after a certain time makes contact with the Sc and entrains into the MBL.

In AUG31, the FT just above the MBL top remains relatively polluted and humid all the time, with $N_{\mathrm{a}}$ values above 1000 mg${}^{-\mathrm{1}}$. There is a gradual increase in $N_{\mathrm{a}}$ with time in the MBL as a consequence of the entrainment of the aerosol plume (Fig. c). This entrainment is, however, not as strong as the one in AUG16 (Fig. b), and at the end of the simulation most of the pollution remains above the MBL.

3.2 Comparison between the control simulations and the observations

Figure  compares cloud cover and liquid water path along the trajectories between SEVIRI (NASA) and the MIMICA simulations. In the case of SEVIRI (NASA), for each of the days (AUG03, AUG16, and AUG31), we averaged three contiguous trajectories (four in AUG16) that are expected to have identical thermodynamic profiles (e.g., the trajectory matching the simulation AUG03 and two other trajectories from SEVIRI (NASA) located at a distance less than 285 km from the trajectory simulated in AUG03). We did this in order to reduce the lack of satellite observations due to the filtering applied to the dataset.

The SEVIRI (NASA) retrievals of cloud cover suggest a predominance of overcast conditions at the beginning of the three trajectories that later alternate with more broken cloud conditions (Fig. a–c). MIMICA also simulates cloud conditions along the trajectories that are initially overcast and later evolve into more broken cloud scenes. The sparse LWP data retrieved from SEVIRI (NASA) have the best agreement with the LWP simulated by MIMICA in AUG03, while the biggest differences are observed during the first day of AUG31. The comparison may be affected by the fact that our retrieval samples (SEVIRI) have three (or four) values per time, which corresponds to the number of contiguous trajectories selected to calculate the average in each situation (AUG03, AUG16, and AUG31).

In order to obtain a more robust statistical evaluation, we compare diel cycles averaged during August over 10 years of cloud cover data and the LWP from CLAAS-3 with the equivalent variables for each of the trajectories in the MIMICA simulations (Fig. ). There is a clear diel cycle of cloudiness and LWP in both the CLAAS-3 dataset and the MIMICA simulations. As expected, the cloudiness and the LWP are at a minimum in the early afternoon. The cloud cover peaks early in the morning and in the late evening in both the observations and the simulations. However, the cloud cover reaches its minimum a few hours earlier in the simulations compared to the observations, and the amplitude of the diel cycle is stronger. The values of the LWP are higher in the CLAAS-3 dataset than in MIMICA. Given the inherent differences in resolution between the CLAAS-3 retrievals and the MIMICA simulations, we find the model produces reasonable results.

Figure 10

Average diel cycle of (a) cloud cover and (b) liquid water path from SEVIRI (CLAAS-3) and MIMICA CTRL simulations. Values from SEVIRI (CLAAS-3) correspond to 10 years of data during August over the geographical area covered by the trajectories. The retrievals of the LWP are based on the shortwave channels; hence they are available only during the sunlit hours. The shading shows the standard deviation over an area about 500 km wide containing the three trajectories. For MIMICA, time-averaged values along each trajectory are shown.

[Figure omitted. See PDF]

3.3 Influence of direct and semi-direct aerosol effects (Aer-rad-off vs. CTRL)

In this subsection, we compare the Aer-rad-off simulations with the corresponding CTRL cases to study the influence of aerosol direct and semi-direct effects on the simulated cloud properties. Above-cloud semi-direct effects are excluded from the comparison because the forcing conditions in the FT are the same in both cases. For AUG03, the pollution levels are relatively low and mainly present during the second half of the simulation. In addition, most of the BBA plume remains above the Sc deck and the MBL (Fig. ). Thus, any impacts associated with the semi-direct effect are relatively weak and mainly noticeable during or after the second day. The main feature when comparing CTRL (including aerosol–radiation interactions) with Aer-rad-off (without aerosol–radiation interactions) is a reduction in cloud cover during the second day when aerosol–radiation interactions are turned on (Fig. d–f). The daytime average difference in SW heating between CTRL and Aer-rad-off is generally small during the second day (Fig. d). However, a closer inspection (not shown) suggests that the aerosol SW heating in CTRL causes a temperature increase and a relative humidity decrease that most likely explains the reduction in cloudiness (initially the temperature increase only occurs above the MBL top but later increases within the MBL as the clouds break up). The diurnal average CRE (driven mostly by the SW fluxes) is reduced in absolute values in CTRL compared to Aer-rad-off during all 3 d (Fig. ). This reduction can be explained by a combination of two factors: the smaller cloud fraction in CTRL during the daytime compared to in Aer-rad-off (especially during the second day in the three cases), which reduces the albedo, and the reduction in contrast in albedo between clear and cloudy skies (which makes the clear-sky scenes brighter in CTRL compared to in Aer-rad-off because the aerosols have a higher albedo than the sea surface). Differences in precipitation are small between CTRL and Aer-rad-off (Fig. ).

Figure 11

Simulated (MIMICA) mean daytime (between 6 and 18 h) profiles of SW heating for the 3 d for each case and each sensitivity experiment.

[Figure omitted. See PDF]

AUG16 shows the clearest impacts among all cases of aerosol–radiation interactions on simulated cloud properties and MBL evolution due to the high levels of pollution (Fig. ). During the first day, the BBA plume is in contact with the Sc deck, and there is also a substantial amount of pollution present within the MBL. However, the average daytime SW heating within the MBL does not differ substantially between CTRL and Aer-rad-off (Fig. b) due to the overcast conditions. Consequently, the differences in cloud parameters between CTRL and Aer-rad-off (e.g., cloud cover and LWP; Fig. ) are small during the first day of the simulation. The differences between CTRL and Aer-rad-off become more pronounced after the second night, when the aerosol plume is mainly located within the MBL (Fig. ). The general reduction in cloud cover during the second day (see Sect. 2.1) allows more solar radiation to penetrate the MBL, causing a strengthening of the direct and semi-direct effects of the absorbing aerosol that amplifies the reduction in cloud cover in CTRL (including aerosol–radiation interactions) compared to in Aer-rad-off. Figure b shows that the LW cooling at the top of the MBL is lower in CTRL compared to in Aer-rad-off during the second day, which contributes to a decrease in the MBL TKE (Fig. e). Furthermore, the absorption of solar radiation produces stronger SW heating just below the cloud (around 2 K d${}^{-\mathrm{1}}$ difference between CTRL and Aer-rad-off; Fig. e) compared to close to the surface (around 1 K d${}^{-\mathrm{1}}$ difference between CTRL and Aer-rad-off). This heating profile stabilizes the MBL in CTRL, which also favors reduced buoyancy production. The net result is a clear reduction in cloud cover in CTRL compared to in Aer-rad-off (Fig. e). The entrainment rate is also substantially lower in CTRL compared to in Aer-rad-off (Fig. h) and actually reaches negative values during the second day of the simulation, in agreement with a reduction in the inversion height with time (Fig. k). The above processes cause the MBL to be around 200 m shallower in CTRL than in Aer-rad-off during the last day of the simulation. There is a negligible difference in drizzle production between Aer-rad-off and CTRL, since the large amount of aerosols in the MBL effectively suppresses precipitation (Fig. h). As in AUG03, the CRE in AUG16 is less negative in CTRL than in Aer-rad-off (Table ), but the difference between the simulations is larger due to the larger amounts of BBAs in the AUG16 case.

For AUG31, the impact of the direct and semi-direct aerosol effects within the MBL are weaker than in AUG16, which is expected since the MBL is less polluted and the BBA plume remains mainly above the cloud (Fig. ). The mean SW heating difference between CTRL and Aer-rad-off is largest during the third day, reaching a maximum value of around 1 K d${}^{-\mathrm{1}}$ just below the cloud (Fig. i). Therefore, the differences between CTRL and Aer-rad-off regarding the cloud variables are smaller than in AUG16 (Fig. ). Similarly to AUG03 and AUG16, the CRE is less negative in CTRL than in Aer-rad-off (Fig. ).

To summarize, all three cases show clear differences in cloud cover due to direct and semi-direct effects. These differences occur regardless of the level of pollution, although they are more pronounced when pollution is high and during broken cloud conditions (during the second and third day of the simulation) compared to overcast conditions (first day of the simulation). In other words, the efficiency of the direct and semi-direct aerosol effects appears to increase with broken cloud conditions. For the AUG16 case, the SW heating during the daytime is clearly larger within and below cloud in CTRL compared to in Aer-rad-off (Fig. ). For the other two cases, the differences in mean daytime SW heating within and below clouds are less pronounced.

3.4 Influence of aerosol indirect effects (N100 vs. Aer-rad-off)

In order to obtain an individual estimate of the aerosol indirect effect, we compare the N100 and Aer-rad-off simulations. Note that we cannot compare N100 with CTRL as this would show the combined direct, semi-direct, and indirect aerosol effects. The low $N_{\mathrm{a}}$ in N100 produces mean daytime SW heating profiles in the FT similar to those in Aer-rad-off (Fig. ), showing that the impact of aerosols on radiation is small in N100 and that it is reasonable to compare Aer-rad-off and N100 to derive an estimate of the aerosol indirect effect.

In general, the indirect aerosol effect should primarily be visible in the cloud microphysical parameters. In all Aer-rad-off simulations, the $N_{\mathrm{c}}$ is also higher than in N100 (Fig. a–c). This increase leads to a reduction in the average cloud droplet radius (Fig. d–f) and an inhibition of drizzle (Fig. g–i). The largest impact on the drizzle production occurs for AUG16, most likely as a consequence of the relatively high humidity content of the BBA plume for this case. The precipitation sensitivity to $N_{\mathrm{a}}$ is most pronounced during the nighttime, when most of the drizzle is produced.

Somewhat surprisingly, the AUG03 DRY simulation displays an increase in accumulated precipitation during the last night with slightly higher values than in N100. In DRY, the convection is slightly more organized than in N100; thus it is more efficient in generating precipitation. Since there is a moisture perturbation in DRY, differences between DRY and N100 are also not solely attributable to the indirect effect. Furthermore, AUG03 has the smallest differences (compared to AUG16 and AUG31) in the forcing conditions between the experiments (CTRL, Aer-rad-off, N100, and DRY) which also leads to the smallest differences in precipitation between N100 and the rest of the simulations.

The difference in cloud cover between N100 and Aer-rad-off (Fig. ) varies with time and is not consistent for all cases, in particular during the daytime. During the nighttime, and especially towards the end of the simulation, there is a tendency for N100 to show lower cloud cover values than Aer-rad-off does, i.e., that higher aerosol concentrations lead to more clouds, which is consistent with the lower precipitation values and the aerosol indirect effect. However, the impact of the indirect effect on cloud cover evolution is generally only seen if there is a substantial drizzle, and this is not the case for our simulations, where the drizzle is small even in the N100 case (thus, the ability of MIMICA to simulate drizzle should be evaluated further). Consequently, there are small and inconsistent differences between N100 and Aer-rad-off that prevent us from making a robust assessment of the indirect effect of BBA plumes on cloud evolution. Furthermore, for AUG16, the cloud cover differences between Aer-rad-off and N100 during and after the second day of the simulation are smaller than between each of those experiments and CTRL. This suggests that, for this situation, the aerosol heating effect (semi-direct effect) is more important than the indirect aerosol effect in terms of affecting the Sc cloud properties and the transition into cumulus.

The highest $N_{\mathrm{a}}$ values in Aer-rad-off compared to N100 produce the highest (in absolute terms) SW CRE in all Aer-rad-off cases (Fig. ). There are two reasons: firstly, the higher $N_{\mathrm{c}}$ and smaller droplet radii increase the cloud albedo; secondly, aerosols are transparent to radiation in Aer-rad-off, producing darker clear skies and brighter cloudy skies. In contrast, the SW CRE is generally smaller in CTRL than in N100, i.e., when both direct and indirect effects are considered. The main reasons for this difference are the clear-sky albedo which is higher in CTRL than in N100 due to more aerosols and that the cloud cover tends to be smaller in CTRL than in N100 during the daytime.

3.5 Influence of moisture in the BBA plume (DRY vs. CTRL)

The impact of moisture within the BBA plume on cloud evolution is examined by comparing CTRL with DRY. Figure  shows that there is a clear difference in cloud cover between all CTRL and DRY cases, but the impact differs between AUG16 and the other two cases.

In AUG03 and AUG31, there are enhanced levels of moisture above and in contact with the MBL, in particular during the second half of the simulation (AUG03) or after the first day of the simulation (AUG31; Fig. ). This additional moisture could favor the high values of cloud cover observed in CTRL compared to in DRY (Fig. ). However, differences in moisture entrainment between CTRL and DRY are generally small, in particular for AUG03 (Fig. d). The most likely reason for the higher cloud cover in CTRL compared to in DRY is instead that the MBL is shallower in the former (Fig. j–l). Enhanced levels of moisture above the MBL can reduce the net LW cooling at cloud top (see Sect. ). Figure a and c also show that before midday of the second day, the LW cooling at the MBL top is generally lower in CTRL compared to in DRY in AUG03 and AGU31. In CTRL, the reduced LW cooling leads to a reduction in the MBL TKE and cloud-top entrainment, which reduces the MBL growth compared to DRY (Fig. ). The presence of more clouds in CTRL than in DRY results in a more negative daytime CRE, in particular during the last 2 d of the simulation.

The effect of moisture on cloud evolution is more complex for the AUG16 case. During the first 1.5 d of the simulation, moisture is clearly enhanced in the FT in CTRL compared to in DRY (Fig. ), meaning that moister air is also progressively entrained into the MBL (Fig. d–f). Similarly to the AUG03 and AUG31 cases, the additional moisture contributes to higher cloud cover and LWP values in CTRL compared to in DRY for AUG16 (Fig. ). Since the cloud cover and LWP in DRY are substantially lower than in CTRL during the first night and day, the domain-averaged LW cooling at the top of the MBL is reduced (Fig. b). Therefore, unlike in AUG03 and AUG31, the domain-averaged LW cooling at the MBL top is not consistently higher in DRY than in CTRL for the AUG16 case during the first 1.5 d of the simulation (Fig. b). Between the second midnight and the following morning of AUG16, most of the humid BBA plume has entrained into the MBL in the CTRL simulation. In CTRL, moist entrainment maintains the Sc cloud deck during this period, while the progressive reduction in humidity above the MBL facilitates MBL deepening. In contrast, DRY experiences an early (before sunrise) cloud break-up with an associated reduction in the domain-averaged MBL top LW cooling, MBL TKE, and entrainment rates with respect to CTRL, which leads to a slowing of the MBL growth (Fig. k). During and after day 2 of AUG16, the MBL in CTRL remains moister than in DRY, particularly in the lower half of the MBL (Fig. ). During this period, the differences in cloud cover between CTRL and DRY are not consistent, especially between midday of day 2 and the third midnight. The stronger MBL decoupling in CTRL compared to in DRY prevents moisture fluxes from the lower MBL to reach the cloud base in CTRL, which to some extent limits cloud growth in CTRL with respect to DRY.

3.6 Quantification of the radiative effects of aerosols and moisture

We compare the overall radiative impact of aerosols due to the direct and semi-direct effects, due to the indirect effect, and due to all aerosol effects. Figure  shows that the time-mean domain-averaged radiative effect at the top of the model domain is dominated by SW radiative effects. In general, positive values in the SW (i.e., a warming) can be caused by the absorption of aerosols above clouds (which reduces the upwelling SW radiation at the top of the model domain) or by a reduction in cloudiness or cloud albedo. In contrast, the SW effect is negative if the cloud albedo or cloudiness increases or if aerosols are present under clear-sky conditions. In the LW, a positive effect (warming) can be caused by an increase in cloudiness or moisture.

Figure 12

Domain-averaged and time-averaged radiative effect (in W m${}^{-\mathrm{2}}$) in AUG03 (a), AUG16 (b), and AUG31 (c) due to aerosol direct and semi-direct effects (CTRL $-$ Aer-rad-off), indirect aerosol effects (Aer-rad-off $-$ N100), all aerosol effects combined (CTRL $-$ N100), and enhanced moisture (CTRL $-$ DRY). The mean values are calculated over 64 h of MIMICA simulations. The total radiative effect (SW $+$ LW) is the number above (if positive) or below (if negative) the stacked bars. Note that this number (total radiative effect) does not match the values on the $y$ axis since it is the sum or difference of the SW and LW bars.

[Figure omitted. See PDF]

The mean direct and semi-direct aerosol effect is positive for AUG16 and AUG31 but negative for AUG03. Figure  shows that the SW radiative effect is positive during the first 2 d (day 1 and day 2) in the three cases, with higher values in AUG16 and AUG31 as a result of the presence of absorbing aerosols over the Sc cloud deck and the reduction in cloudiness in CTRL compared to in Aer-rad-off. However, during the third day, the radiative effect becomes negative as the increase in clear-sky albedo (in CTRL compared to in Aer-rad-off) becomes larger than the decrease in albedo due to fewer clouds. The mean indirect aerosol effect is negative for all three cases (AUG03, AUG16, and AUG31). This result is expected as there was no substantial impact on cloud cover and because the smaller cloud droplets in CTRL compared to in N100 increased the cloud albedo. The total radiative effect from the combination of direct, semi-direct, and indirect aerosol effects is also negative for all cases. This shows that for these three cases, and averaged over the whole simulation time, the net effect of the biomass burning aerosols is to cool the system, with the indirect effect dominating over the direct and semi-direct aerosol effects.

Figure 13

Mean daytime radiative effects (in W m${}^{-\mathrm{2}}$) due to aerosol direct and semi-direct effects (a–c), indirect aerosol effects (d–f), all aerosol effects combined (g–i), and enhanced moisture (j–l) in AUG03 (a, d, g, j), AUG16 (b, e, h, k), and AUG31 (c, f, i, l). The total radiative effect (SW $+$ LW) is the number above (if positive) or below (if negative) the stacked bars. Note that this number (total radiative effect) does not match the values on the $y$ axis since it is the sum or difference of the SW and LW bars.

[Figure omitted. See PDF]

The enhanced moisture associated with the BBA plume also leads to a negative mean radiative effect (cooling) for all three cases (Fig. ). The main reason is that the enhanced moisture helps to sustain the Sc cloud deck which increases the albedo of the system. Note that for all three cases, the cooling caused by the enhanced moisture is about as large as the sum of all the aerosol effects (direct $+$ semi-direct $+$ indirect).

4 Discussion and conclusions

In this study, we have used large-eddy simulation in a Lagrangian setup to explore how the stratocumulus cloud cover, boundary layer evolution, and stratocumulus-to-cumulus transitions over the southeast Atlantic are affected by the individual impacts of absorbing aerosols and moisture within biomass burning plumes. We initialized and forced our model with meteorological conditions corresponding to three different periods during August 2017 (AUG03, AUG16, and AUG31). These situations were clearly different with respect to the levels of pollution and water vapor in the plume and also regarding when the plume appeared in the free troposphere and when it mixed into the cloud layer and the MBL. The selection of three cases can be useful in generalizing some of the impacts of the BBA layers on the Sc clouds and SCTs. The analysis and comparison of multiple situations also helps to give a wider perspective on the topic, since there is only one previous study that investigates the effects of biomass burning aerosols on SCTs using large-eddy simulation with meteorological conditions specific to the southeast Atlantic .

An evaluation of the model results against satellite retrievals showed that the model reproduced the diel cycle of cloud cover and liquid water path reasonably well from a climatological perspective. For all three periods examined, the simulations showed SCTs that were broadly consistent with the theory of a deepening–warming transition , which is in agreement with results from previous studies under relatively high aerosol concentration conditions (e.g., ). The drizzle increased slightly when aerosol concentrations were reduced, and there was a clear decrease in the cloud droplet number concentration. However, none of the experiments showed signs of a relevant impact of drizzle on the SCT, not even when the aerosol concentrations were lowered to background levels. The reason is likely that there was little drizzle formation in all cases and that experiments with an even lower aerosol concentration are needed to explore a potential “drizzle-driven” transition in MIMICA (e.g., ). These findings further amplify the existing interest in exploring the SCT in different LES models within a common framework. In this context, the Southeastern Atlantic Stratocumulus Transitions with Aerosol–Rain–Radiation interactions (SEA STARR) large-eddy simulation intercomparison project is currently being conducted in order to investigate the SCT in several LES models (including MIMICA) under the same meteorological conditions and aerosol forcings and to improve our comprehension of the factors responsible for the differences in the SCT between models.

The semi-direct effect of absorbing aerosols that were in contact with, or mixed within, the MBL was found to be substantial, especially in highly polluted situations, and in particular during the daytime and during broken cloud conditions. Our simulation results imply that biomass burning aerosols have the potential to speed up SCTs through local semi-direct effects. However, the influence will be dependent on the state of the SCT and on the time of the day when the absorbing aerosol makes contact with the cloud deck. In our simulations, the forcing conditions were the same in all experiments, with impacts of the biomass plume on temperature and winds always included in the free troposphere. Therefore, we were not able to explore any free tropospheric semi-direct aerosol effects on cloud cover (as in, e.g., ) or the impacts of large-scale subsidence changes caused by aerosol heating (as in ). Thus, we cannot say if semi-direct effects of aerosols in the free troposphere can reduce the relative importance of local semi-direct effects in the MBL as shown in , , and .

The moisture associated with the aerosol plume had a clear impact on the MBL and SCT for the three periods analyzed. When located mostly above the Sc deck, the LW radiative effect of the humidity slowed down the MBL deepening, in particular during the night. These results are consistent with those obtained by and . However, in contrast with and , we did not find that the LW radiative effect of the water vapor above the MBL caused cloud breakup. The reason for this difference is that the enhanced moisture in the FT was never clearly separated from the cloud in our simulations. Thus, the moisture from the plume could always entrain the cloud and the MBL, which led to an increase in cloud cover and a delay in the SCT, in agreement with .

We note that the conclusions drawn regarding the plume impact on the SCT may depend on the definition of the SCT and that there are different ways to define the SCT in the literature. Here we have used the definition employed by and (see Sect. ) based on a cloud cover threshold of 50 %. Recently, used a similar approach. However, they only regarded the SCT as happening when the cloud cover remained below 50 %, either during the following 24 h of the simulation or until the end of the simulation. If we use the same approach as , the SCT happens 1 d later in several of our simulations (including two of the control simulations: AUG03 and AUG31). investigated the decrease in the domain mean albedo at the beginning and end of their simulations to obtain a measure of the amplitude of the SCT, and thereby also an estimate of the pace of the SCT, following . This criterion is not directly comparable with the ones based on cloud cover. A common metric among different studies would be useful to make comparisons more robust. Since all the mentioned criteria regarding the SCT have their flaws and do not fully capture the complexity of the phenomena, we have avoided rigorously using the 50 % threshold in cloud cover to conclude that the semi-direct effect favors while the moisture delays the SCT. Instead, we have looked at the differences in the evolution of the cloud cover between the experiments and used the 50 % threshold as a “soft” reference for the SCT.

In our simulations, the indirect aerosol radiative effect always dominated over the direct and semi-direct radiative effects, and the absorbing aerosol plume produced an average net radiative cooling effect over the 3 d of the simulation of about $-\mathrm{4}$ to $-\mathrm{9}$ W m${}^{-\mathrm{2}}$. These results are consistent with , who used a regional model to investigate the effects of biomass burning aerosols during 2 months over the SEA. However, our estimate can be influenced by the fact that our experiments did not include any aerosol effects in the FT and that our BBA plumes were not clearly separated from the cloud deck. For instance, the semi-direct effect of absorbing aerosols located above and separated from the cloud deck might contribute to additional cooling as in . Our simulations also showed that the moisture accompanying the absorbing aerosol in the biomass burning plume produced an additional cooling effect that was as about as large as the total aerosol radiative effect itself.

Code and data availability

The modeling datasets used in this study are available at 10.17043/baro-perez-2023-biomass-burning-1 . MIMICA model source code is available at https://bitbucket.org/matthiasbrakebusch/mimicav5/src/master/ .

Author contributions

ABP and MSD designed the experiments. MSD provided the methodology to force MIMICA. ABP implemented the aerosol–radiation interactions in the model with inputs from JT and HK and under the supervision of AMLE, JS, and MS. ABP performed the simulations, provided all visualizations, and wrote the article with inputs and revisions from all coauthors. AMLE, FAMB, and AD revised the article and participated in the analysis of results. AD provided the SEVIRI (CLAAS-3) dataset. DP and HL provided, respectively, the SEVIRI (NASA) dataset and the Lagrangian trajectories used to force the model.

Competing interests

The contact author has declared that none of the authors has any competing interests.

Disclaimer

Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims made in the text, published maps, institutional affiliations, or any other geographical representation in this paper. While Copernicus Publications makes every effort to include appropriate place names, the final responsibility lies with the authors.

Special issue statement

This article is part of the special issue “New observations and related modelling studies of the aerosol–cloud–climate system in the Southeast Atlantic and southern Africa regions (ACP/AMT inter-journal SI)”. It is not associated with a conference.

Acknowledgements

We are grateful for the technical assistance provided by Matthias Brakebush (SU) and Hamish Struthers (NSC) in running the MIMICA LES model using the NSC resources. We also thank Jens Redemann for his valuable comments on the article.

Financial support

This study has been funded by the Swedish National Space Agency (grant no. 16317). Additional funding has been provided by the European Union's Horizon 2020 research and innovation programme (FORCeS, grant no. 821205) and the Swedish Research Council (grant no. 2020-04158). The computations and data handling have been enabled by resources provided by the National Academic Infrastructure for Supercomputing in Sweden (NAISS), partially funded by the Swedish Research Council (grant no. 2022-06725). Hyunho Lee has been supported by the National Research Foundation of Korea (grant no. NRF-2021R1C1C1012804). The article processing charges for this open-access publication were covered by Stockholm University.

Review statement

This paper was edited by Hailong Wang and reviewed by two anonymous referees.