On a global mean basis, BBA is estimated to exert a neutral direct radiative forcing of -0.1 to +0.1 W m-2 (Boucher et al., 2013). Even the sign of the global mean direct radiative forcing is in doubt because the single scattering albedo (SSA, the ratio of optical attenuation coefficients for scattering and extinction) of BBAs is close to the balance point between net reflection and net absorption of sunlight (e.g. Haywood and Shine, 1995). However, regionally, BBA plays a far more important role: nowhere is the uncertainty in the direct radiative effect and forcing more apparent than over the SE Atlantic than during the August–September dry season (Fig. 2).

Figure 2 shows the “direct” radiative effect derived from models participating in AEROCOM (Myhre et al., 2013b; Stier et al., 2013), indicating a regional hotspot for BBA forcing over the SE Atlantic but with significant uncertainty because BBA can exist either above the stratocumulus, decreasing the planetary albedo, or above open ocean, where it increases the planetary albedo. To accurately model the aerosol direct effect, models need to represent all of the following correctly: the magnitude and geographic distribution of the AOD, the wavelength-dependent SSA, the BBA vertical profile, the geographic distribution of the cloud, the cloud fraction, the cloud liquid water content, the cloud droplet effective radii, and the cloud vertical profile (Keil and Haywood, 2003; Abel et al., 2005; Samset et al., 2013; Stier et al., 2013). At a more detailed aerosol process level, we need to understand the optical properties of black carbon, organic carbon, and inorganic compounds as a function of mixing state and how these properties vary as a function of altitude and relative humidity and as a function of ageing from emission to deposition.

Another implication of BBA overlying cloud is that satellite retrievals of cloud that rely on visible wavelengths are generally biased low in cloud optical depth (COD) and effective radius (e.g. Hsu et al., 2003; Haywood et al., 2004; Wilcox and Platnick, 2009), with implications for remotely sensed correlative studies of aerosol–cloud interactions (Quaas et al., 2008). Recently, de Graaf et al. (2012) used high spectral resolution satellite data to show that the direct radiative effect of BBA over clouds in the SE Atlantic region could be stronger than +130 W m-2 instantaneously and +23 W m-2 in the monthly mean. These values are far stronger than those diagnosed in climate models which reach only +50 W m-2 instantaneously (e.g. de Graaf et al., 2014), suggesting that models misrepresent at least one of the key parameters noted above.

A further aerosol–radiation interaction occurs as a fast adjustment to the direct effect and is called the semi-direct effect (SDE), whereby the heating of the absorbing BBA layer and the reduction in surface temperature modify the atmospheric stability, surface fluxes, clouds, and hence radiation. Satellite observations over southern Atlantic stratocumulus have shown a thickening of cloud underlying BBA (Wilcox, 2012; Costantino and Breon, 2013), which could be a result of heating of the above-cloud column intensifying the cloud top inversion and reducing entrainment. Wilcox (2012) estimated that this produced a negative radiative effect that compensated for 60 % of the above-cloud positive direct effect. Large eddy model (LEM) simulations have been used to explore the detailed mechanisms of the semi-direct effect (e.g. Johnson et al., 2004; Hill and Dobbie, 2008) although they typically have relatively small domain sizes and therefore cannot account for the impact of aerosol in modifying synoptic-scale circulations. Global modelling studies are able to represent impacts on synoptic- and regional-scale dynamics and circulation patterns (e.g. Allen and Sherwood, 2010; Randles and Ramaswamy, 2010) but are unable to represent the detailed process-level mechanisms captured by LEMs. Studies in LEMs and global climate models have emphasized the importance of the vertical profile of aerosol and the degree of absorption (Johnson, 2004; Randles and Ramaswamy, 2010; Samset et al., 2013). Randles and Ramaswamy (2010) and Allen and Sherwood (2010) document the response to the semi-direct effect via atmospheric impacts on stabilization, reduced surface fluxes, and subsequent evolution of the modelled dynamical impacts. Johnson (2004) found the semi-direct effect to be 5 times smaller in global-scale models compared to LEMs, although these results are challenged by Allen and Sherwood (2010). In addition, internal variability masks local semi-direct effects, severely decreasing the statistical significance in previous studies of modelled semi-direct effects (e.g. Ghan et al., 2012) and our ability to assess their fidelity.

Despite considerable advances in modelling clouds using models of different resolutions, considerable uncertainties remain in modelling even the relatively simple cases of stratocumulus, owing to uncertainties in precipitation, decoupling, moisture budgets, and entrainment. Unsurprisingly, climate models show considerable inter-model biases in cloud fraction, liquid water path, effective radius, and COD when compared against satellite observations, leading to large discrepancies in the solar fluxes and hence the energy absorbed by the ocean in the region (e.g. Bodas-Salcedo et al., 2014).

Aerosol–cloud interactions, or “indirect effects”, remain one of the most elusive but key parameters in climate prediction (Stevens and Feingold, 2009; Boucher et al., 2013). For stratocumulus, the effect of increased CCN leading to cloud brightening can be modulated by changes in precipitation and subsequent changes to cloud water amounts through entrainment processes (e.g. Ackerman et al., 2004). Satellite-borne lidar studies of aerosol–cloud interactions in the region emphasize the critical role of the vertical profile of aerosol and cloud (Costantino and Breon, 2013) and the relative position of the two to each other (Chand et al., 2009). However, it is difficult to fully discern the level of interaction between clouds and aerosols because of the sensitivity of lidars in the free troposphere (Watson-Parris et al., 2018) and the attenuating effects of a thick layer of aerosols overlying clouds. Global bulk aerosol models and empirical representations of aerosol indirect effects are being replaced with microphysical aerosol models such as ECHAM5-HAM (Stier et al., 2005) and GLOMAP (also known as UKCA mode) (e.g. Mann et al., 2010; Bellouin et al., 2013), and more explicit representations of cloud and precipitation processes (Hill et al., 2015; Grosvenor et al., 2017) have also been developed. Such schemes require extensive evaluation, which is often achieved through multi-model intercomparison studies (e.g. Quaas et al., 2009) and comparison to observations. The spatial resolution of global numerical weather prediction (NWP) and climate models (typically 10–100 km) is widely recognized as inadequate for investigating essential aerosol–cloud interaction processes at the cloud scale (∼10 m; Lebo et al., 2017). Thus, relationships between sub-grid-scale variables such as cloud updraft velocity and entrainment from LEMs and their link to large-scale boundary layer variables are being sought but, while promising, are far from well established (e.g. Golaz et al., 2011; Malavelle et al., 2014). Simulations with HadGEM2 Coupled Large-scale Aerosol Scheme for Studies In Climate (CLASSIC) suggest that while BBA interaction with cloud may be limited by vertical stratification, it does enter the MBL and interact with cloud, producing a strong indirect effect in the region (Fig. 3a). However, the more sophisticated GLOMAP-MODE two-moment scheme leads to a much reduced aerosol indirect effect because an increase in aerosol mass does not necessarily lead to an increase in the aerosol number or CCN as the aerosol size distribution will tend to shift to larger sizes as more volatile organic precursors condense upon pre-existing aerosol particles (Fig. 3b). Therefore, when combined with a slight reduction in the cloud fraction associated with the aerosol-semi direct effect, some areas of the South Atlantic shown in Fig. 3b are subject to a positive radiative effect. The over-strong aerosol–cloud interaction in CLASSIC compared to GLOMAP-MODE has been noted in other studies that have used satellite retrievals to assess their validity (e.g. Malavelle et al., 2017).

An assessment of parametric uncertainty in the GLOMAP-MODE global model driven by ECMWF meteorology and observed low-level clouds (Lee et al., 2013) showed that BBA particles are one of the largest sources of uncertainty in CCN at cloud base. However, Lee et al. (2013) did not assess the effect of uncertainties in the physical model, which control the extent to which BBA and clouds mix, nor structural model uncertainties.

The stratocumulus decks of the SE Atlantic have been linked via global teleconnections to precipitation anomalies in Brazilian rainfall; SE Atlantic stratocumulus that is too bright can lead to precipitation deficits in the Nordeste and Amazonian regions (Milton and Earnshaw, 2007; Jones et al., 2009). Similarly, Atlantic sea-surface temperature gradients and the hemispherical asymmetry in the energy balance are strongly impacted by SE Atlantic stratocumulus (Jones and Haywood, 2012; Stephens et al., 2016), influencing the position of the intertropical convergence zone (ITCZ) and hence the African and Asian monsoon.

The last major international measurement campaign investigating biomass burning (BB) in southern Africa was the Southern AFricAn Regional science Initiative in 2000 (SAFARI-2000). The SAFARI-2000 dry-season intensive campaign focussed on the emissions, transport, and transformation of BBA plumes and the validation of satellite remote sensing retrievals of aerosol and cloud from the Terra satellite (Swap et al., 2002). The majority of investigations over the SE Atlantic were basic aerosol microphysics and cloud-free radiative impact studies (Haywood et al., 2003; Keil and Haywood, 2003; Osborne et al., 2004. Magi et al., 2008). Since SAFARI-2000, significant advances in airborne measurement of BC (e.g. Schwarz et al., 2008; McMeeking et al., 2011); organic and inorganic aerosol compounds (Morgan et al., 2010); and aerosol physical properties have occurred. In addition, improvements in the accuracy and sensitivity of measurements of aerosol optical properties, notably absorption (e.g. Sedlacek and Lee, 2007; Lack et al., 2008), have been made. Airborne lidar instrumentation and retrievals allow for concurrent mapping of vertical distributions of aerosols above clouds (e.g. Marenco et al., 2011). An extensive set of measurements of stratocumulus clouds has been performed during VOCALS off the Pacific coast of South America (Wood et al., 2011), with one of the foci being aerosol–cloud interactions (e.g. Yang et al., 2011; Painemal and Zuidema, 2013). However, the aerosol composition, sources, and interaction with the clouds in the VOCALS region are very different to those over the SE Atlantic, which is dominated by relatively strongly absorbing biomass burning aerosol (e.g. Haywood et al., 2003).

Model capabilities have also improved. At the time of SAFARI-2000, aerosol modelling was in its infancy, with only two global chemical transport models reporting the direct radiative forcing and cloud albedo indirect forcing of BBA in the IPCC report (Ramaswamy et al., 2003). Since 2000, the focus for aerosol–radiation interactions has shifted to areas where model results diverge (e.g. SE Atlantic; see Figs. 2 and 3). Global aerosol microphysics models have also been developed and are coupled to climate models and to cloud models at high resolution. Aerosol–cloud interactions are now studied at scales ranging from LEMs with resolutions of a few metres, through cloud resolving models, and limited area numerical weather prediction models to global models with resolutions of ∼100 km. New approaches to understand sources of aerosol uncertainty have also been developed (Lee et al., 2013). However, high-quality validation data in the SE Atlantic with which to challenge the global and cloud resolving models are almost entirely lacking.

With the rationale as described above, CLARIFY-2017 aimed to use the natural laboratory of the SE Atlantic to improve the representation of BBAs and clouds in models of a range of scales and increase the fidelity of aerosol–radiation and aerosol–cloud interaction processes and cloud representation and their impacts on local, regional, and global weather and climate. Experience suggested that these objectives were best achieved by conducting an intensive airborne field campaign with supporting surface and satellite measurements. The measurements were used to challenge and develop improved models at different spatial scales from the cloud scale to the global scale that couple aerosols, clouds, and radiation.

Specific key objectives of CLARIFY-2017 were as follows:

Key objective 1: measure and understand processes governing the physical, chemical, optical and radiative properties of BBAs in the SE Atlantic region.

An example of the flight patterns performed for determining the radiative effects of BBA in cloud-free and cloudy skies is shown in Fig. 6 based on preconceived ideas of what we would expect based on prior experience from SAFARI-2000 (Haywood et al., 2003).

The patterns shown in Fig. 6 were typically orientated so that the SLRs and profiles avoided running within 30∘ of the into-sun heading. This is to avoid making radiative transfer measurements where aerosol may have been impacted on the front face of the Eppley BBRs and SHIMS instruments, which could lead to a reduction in measured irradiance. Owing to the variability of cloud, the order of the runs was typically changed during CLARIFY-2017 so that the high-level SLR leg was followed by a reciprocal turn and profile descent followed by a reciprocal turn and SLR just above cloud top. This ensured the minimum length of time had elapsed between the two legs to minimize differences caused by changes in cloud fields below the aircraft (e.g. Peers et al., 2019, 2020). Because of the shape of this sequence, the pattern is known as a “Z pattern”.

Radiometric measurements above and below the BBA characterized broadband and spectral irradiances and radiances, provided aerosol vertical distribution from lidar, and enabled sea-surface reflectance characterization. Profiles through the BBA characterized the aerosol extinction and absorption coefficient from the EXtinction, Scattering and Absorption of Light for AirBorne Aerosol Research (EXSCALABAR) instrument and hence the aerosol optical depth and aerosol absorption optical depth. When BBA overlies cloud, SLRs above and below BBA provided remotely sensed estimates of cloud top droplet effective radius and liquid water path (LWP) from solar and microwave instrumentation.

Flight patterns for examining clouds typically resembled a series of stacked SLRs below cloud, within cloud, above cloud, and within aerosol. Typically, the patterns were used together with a series of sawtooths through the cloud to further characterize the variability of the cloud top and cloud base and to provide detailed characterization of cloud microphysical parameters within cloud and at cloud top from in situ measurements (effective radius, LWP, and liquid water content, LWC).

SLRs just below cloud base and just above cloud base were used to investigate CCN budgets, closure, and aerosol loss due to scavenging. SLRs below cloud, in cloud, and above cloud measured CCN, cloud droplet size distributions, and drizzle size distribution below cloud base to provide information on the entrainment process, the influence of entrainment on cloud microphysics, and constraints on BBA entrainment rates into cloud top. Vertical profile, sawtooth, and stepped profile measurements were made of the size distribution of cloud droplets and precipitation over the diameter size range 2 µm to 6 mm, capturing cloud droplets and precipitation. The onboard AMS, SP2, and OPCs were switched between the CVI inlet to measure droplet residuals and the total inlet to determine the size and composition of the nucleation scavenged and interstitial aerosol as a function of position and height in the cloud. Measurements higher in the cloud together with turbulence measurements examined the evolution of the cloud microphysics as condensational growth and coalescence occur.

All satellite overpasses, satellite observations from previous overpasses, model data, and observations from the AMF and from the KNMI lidar installed at the airfield were available to the planning teams (see Sects. 2 and 3). Owing to the high intensity of the flying programme, the flight planning teams were separated into two, an aircraft-based team flying the mission and a second ground-based team which prepared flight plans for the forthcoming flight. The team on the ground also was responsible for sending updates to the aircraft via SATCOM, providing updates of the cloud conditions from the geostationary satellites and measurements from the surface-based instrumentation. After a debrief of the flight, the ground-based team and aircraft-based team then swapped roles so that each team “owned” the flight from inception, through planning and execution. As per standard campaign operating procedures, a running tally of hours allocated to specific aerosol characterization, cloud characterization, aerosol–radiation and aerosol–cloud interactions, and other aspects such as pocket of open cell (POC) investigations was maintained during the campaign so that future flights could target any science gaps in the key objectives.

CLARIFY-2017 was able to show that the vertical structure is quite complex, with aerosols existing either predominantly in the MBL, predominantly in the residual CBL, or existing in both the MBL and residual CBL (Fig. 9).

Figure 9 shows that the CLUMP model is generally able to represent the distribution of aerosol in the MBL and the residual CBL. Figure 9 shows that the one notable exception is when a POC was observed over Ascension Island. During the POC event towards the end of the measurement campaign, the model does not accurately represent the close-to-pristine nature of the MBL (see Abel et al., 2020).

A hierarchical cluster analysis was performed based on the mean and maximum BBA concentrations and the altitude of the maximum concentrations in the free troposphere and the mean concentration within the boundary layer for each flight. The cluster analysis based on these criteria revealed two distinct groups, with the first group (G1) including flights C028–C032 (16–19 August 2017) corresponding to the period when the aerosol was solely in the MBL, with a mean concentration over the size range 0.1–3.0 µm (Table 2) measured by the PCASP instrument on the FAAM aircraft of 685 cm-3 in the MBL but just 35 cm-3 in the free troposphere. The second group contained two subgroups with flights C034, C035, C042, and C047–C050, i.e. those showing little aerosol in the MBL (∼78 cm-3) but much in the residual CBL (∼884 cm-3). These are denoted group G2. The mean synoptic geopotential height based on these two clusters and the associated wind speed vectors are shown in Fig. 10.

Figure 10 shows that, although south-east winds associated with a subtropical high dominated in the MBL at the location of Ascension Island (as indicated by the 925 hPa geopotential height contours and wind vectors) for both groups G1 and G2, the locations of the high-pressure centres were different. In G1, the centre of the high pressure was located around 40∘ S, 0–20∘ E, while under G2, the centre of the high pressure was around 30∘ S, 0–10∘ W. For both G1 and G2, the MBL around Ascension Island can be influenced by air of continental origin, but the MBL in G2 is also influenced by air recirculating around the subtropical high that does not pass over the African continent, owing to the non-geostrophic and divergent flow around high pressures. This recirculation characteristic of G2 appears to explain the relatively clean MBL during the periods 21–25 August and 31 August–4 September (Fig. 9). The geopotential height fields at 700 hPa show significant differences in airflows between the two clusters. In G1, where the cases were represented by a relatively clean free troposphere, there was no clear high pressure to the south-east of Ascension Island. In contrast, in G2, high pressure extended from Namibia and Angola to the island, with associated strong easterly winds that transport smoke from the African continent to the Ascension Island region in the residual CBL (Table 2).

This analysis is enhanced by an analysis of back-trajectories (HYSPLIT4; Draxler and Hess, 1997 using ERA5 reanalyses; Hersbach et al., 2020) presented in Fig. 11, which shows trajectories initiated every 1 h for August 2017 at 50 m (lower MBL), 330 m (mid-MBL), 1000 m (upper MBL), and 2000 m (lower residual CBL).

Back-trajectories originating at 50 and 330 m over Ascension Island indicate that the flow at these levels originates in oceanic regions. At 1000 m (upper MBL), the back-trajectories indicate some influence from land areas in northern Namibia and Angola; these areas experience seasonal burning (e.g. Abel et al., 2003), and thus some of the BBA detected in the MBL is likely to come from these regions. At 2000 m, in the residual CBL, the back-trajectories indicate an easterly flow and hence more northerly source of BBA (northern Angola, Gabon, Congo, Democratic Republic of Congo, Equatorial Guinea, and Cameroon) from an area where fires are most prevalent. BBA in the residual CBL only influences the microphysics of low-lying clouds after those aerosols are entrained into the MBL. Thus, any inferences of aerosol–cloud interactions that depend on relationships between the column AOD and CDNC (e.g. Quaas et al., 2008) may be erroneous in this region (Costantino and Bréon, 2013).

As demonstrated in Fig. 9, comparison of the shape of the mean vertical distribution of aerosol extinction derived from the aircraft-based EXSCALABAR measurements and CLUMP NWP model shows a reasonable agreement, although generally the model does not extend the vertical distribution of aerosol high enough, and there is rather too much aerosol in the MBL. Simulations with the atmosphere only (i.e. not coupled to the ocean model) version of HadGEM3 (Hewitt et al., 2011) indicate that the discrepancy in the vertical distribution of aerosol is very likely due to the lack of account of aerosol radiative effects, in particular the model's neglect of aerosol absorption that “self-lofts” the air containing the BBA (Fig. 12).

While this “self-lofting” has been recognized for decades in smoke plumes (e.g. Westphal and Toon, 1991), the near-continental large-scale nature of the ascent rate and the counterbalancing descent elsewhere hints at a further impact of aerosol beyond aerosol direct, indirect, and semi-direct effects, that of teleconnections. Because atmospheric dynamics are constrained by physical laws of conservation of energy and momentum, any large-scale lifting of air masses must be balanced by the large-scale descent of air masses elsewhere. Figure 12b shows the spatial extent of the ascent (or the reduction in subsidence) over the region caused by absorption of the BBA. This suggests that the lack of inclusion of the radiative impacts of absorbing aerosols owing to computational constraints may have consequences on the performance of NWP models in accurately representing mean vertical velocities. In turn this may influence the strength of the Hadley and Walker circulations. However, it is acknowledged that the atmosphere-only simulations shown here neglect any dynamical changes that may be induced through changes in surface land temperatures or sea-surface temperatures (SSTs) which can induce changes in the thermally direct atmospheric circulation (e.g. Roekner et al., 2006; Sakaeda et al., 2011) and ocean heat transport, which has a large impact on the overall dynamical response (e.g. Hawcroft et al., 2018). The impacts of aerosol direct and semi-direct effects are also investigated in regional high-resolution regional climate models under the AEROCLO-sA measurement campaign (Mallet et al., 2020).

The performance of the HadGEM3 NWP model with CLUMP aerosol scheme compares favourably with the other modelling tools used in forecasting the aerosol spatial distribution. Note that, under the ORACLES project, Shinozuka et al. (2020) performed a multi-model analysis of vertical profiles of BBA against observations from ORACLES during the 2016 deployment when the NASA P3 aircraft was operating from Namibia over a wide area of the SE Atlantic closer to the African continent. Their results suggest that, for that region, each of the models analysed presents its own strengths, weaknesses, and biases, but one common feature is that all models tend to underestimate the height of the base of the smoke layer. This does not appear to be the case with the CLUMP simulations for the CLARIFY-2017 region, where the bottom of the residual CBL aerosol layer frequently corresponds to the top of the MBL (Fig. 9), forcing an accurate lower boundary for the residual CBL plume. The results comparing the CLUMP model to the observations suggest reasonable agreement in aerosol peak concentrations and in the total integrated extinction (Fig. 9), but, as noted earlier, the neglect of aerosol absorption in the NWP model appears to result in a peak aerosol concentration and upper bound of the plume that is approximately 1–2 km too low.

At the top of the MBL, a strong temperature inversion provides a strong energetic barrier to vertical mixing (e.g. Wood and Bretherton, 2006), providing an effective cap to cloud vertical extent. Figure 13 shows the observed boundary layer height as diagnosed from radiosondes launched from Ascension Island and the modelled cloud liquid water path. The Met Office NWP model represents the boundary layer height adequately over Ascension Island. The boundary layer height is important in retrievals of above-cloud aerosol properties from SEVIRI as errors impact the amount of water vapour above cloud that is assumed in the retrieval algorithm (Peers et al., 2019). Figure 13 also shows the cloud droplet number concentration as measured by the CDP instrument over the field campaign. There is clear evidence of the influence of aerosol–cloud interactions in the cloud-droplet number concentration. The cloud droplet number concentration is at its highest at the start of the measurement period, when the BBA is present in large quantities in the MBL, but abruptly transitions to its lowest value when the MBL is close to pristine. Low CDNC values are also found in the measurements when the POC is present towards the end of the deployment period (Fig. 13).

As a result of the strong vertical shear in wind speed, the time for an air parcel to leave the African continent to reach Ascension Island is shorter for aerosol higher up in the residual CBL than lower down in the CBL or in the MBL. Thus, in general, the aerosol at lower altitudes can be significantly older compared to that located at elevated altitudes (see analysis of ORACLES data by Dobracki et al., 2021). There is clear evidence from both Wu et al. (2021) and Dobracki et al. (2021) that aerosol higher up in the residual CBL exhibits a higher SSA (i.e. it is less absorbing on a per particle basis) than that lower down. Taylor et al. (2020) show that the mass absorption coefficient in the CLARIFY domain does not vary significantly with altitude. Wu et al. (2020) and Taylor et al. (2020) propose that the partitioning of a higher fraction of inorganic ammonium nitrate onto the existing particles at the colder temperatures associated with the higher altitudes explains the vertical structure in SSA in the region above Ascension Island. This explanation differs from that of Dobracki et al. (2021), who suggest loss of scattering organic material from the BBA as the aerosol ages based on data from the farther ranging ORACLES flights which encompassed a wider range of aerosol ages (see also Sect. 6.5).

Prior to the ORACLES, AEROCLO-sA, LASIC, and CLARIFY-2017 campaigns, airborne measurements of BBA size distributions in the region were sparse. Haywood et al. (2003) documented the size distribution of BBA during SAFARI-2000, both close to emission source, off the coast of Namibia, and in the vicinity of Ascension Island; both of these cases are of relevance for CLARIFY-2017. Although the data presented here are quality-assured, the specific analyses performed during CLARIFY-2017 inevitably differ due to different sampling locations, sampling periods, and case studies, etc. Therefore, we present a composite of the models that are used to fit data in these studies to allow for a quantification of the error introduced by the assumptions used in each analysis. Generally, these models use log-normal fits of the form: 1dni(r)dln⁡r=ni(2π)ln⁡σexp⁡-ln⁡ri-ln⁡rn22(ln⁡σ)2, in which ni(r) represents the number of aerosols of radius r for mode i, rn represents the geometric mean radius, and σ is the geometric standard deviation. Table 3 shows examples of the fits of this equation to the measured or retrieved size distributions during SAFARI-2000 and CLARIFY-2017, effective refractive indices, and the resultant single scattering albedo.

The size distributions for BBA from Peers et al. (2019) and Wu et al. (2021) are consistent with those determined from SAFARI-2000, although Haywood et al. (2003) chose to describe the accumulation mode with two log-normal distributions rather than a single log-normal distribution. The corresponding refractive indices retrieved over the CLARIFY period (16 August–7 September 2017) derived from AERONET Version 2 algorithms for the Ascension Island site are 1.47–0.020i at a wavelength of 550 nm. The smaller value of the real and imaginary part of the refractive indices from AERONET compared to the in situ retrievals documented in Table 3 likely reflects that they represent column-averaged properties, and hence there is a contribution from aerosol components such as sea salt and sulfate of DMS origin within the MBL (Wu et al., 2021; Taylor et al., 2020). Wu et al. (2021) state a mean SSA at 550 nm of approximately 0.81 at 2 km altitude in the residual CBL rising to 0.86 at 5 km altitude and assign the difference in SSA to the thermodynamic impact of temperature on the partitioning of inorganic nitrate into the aerosol phase.

The composition of the aerosol in different layers was measured with an Aerodyne Compact Time-of-Flight airborne AMS (C-ToF AMS; Table 1), which provided organic mass, nitrate, sulfate, and ammonium mass concentrations and an SP2 to determine the BC mass concentration. Vertical profiles of these different chemical components averaged across each of the regimes, together with campaign-average compositions, are shown in Fig. 14. These data have enabled a detailed characterization of the composition of aerosol in the region (Wu et al., 2021). During periods when the residual CBL is filled with BBA, large quantities of predominately organic aerosol are present. The composition fractions (average ± SD – standard deviation) were organic aerosol (OA; 61±5) %, BC (13±3) %, SO4 (11±4) %, NO3 (8±3) %, and NH4 (7±2) % (Wu et al., 2021; Taylor et al., 2020), suggesting a BC / OM (organic matter) ratio of around 0.21. Notably, the inorganic components are present in significant mass concentrations, and the fraction of ammonium and nitrate present increases with altitude. The MBL displays a greater proportion of sulfate than the residual CBL, while the mass fractions of nitrate, BC, and OA are lower in the MBL. This increased sulfate mass fraction in the MBL is likely due to the formation of sulfate from DMS oxidation in the MBL since similar concentrations are present during clean (MBL) periods. The main chemical properties and processes governing the particulate chemistry are discussed in detail by Wu et al. (2021).

A key objective of CLARIFY-2017 was to assess, to the highest degree of accuracy possible, aerosol optical properties, with a particular focus on the aerosol SSA, owing to its strong influence on the aerosol direct radiative effect. Such an objective precludes the use of filter-based observations which are subject to a wide range of empirical corrections (e.g. Bond et al., 1999; Davies et al., 2019) and have been shown under some conditions to yield uncertainties in absorption of over 200 % (Lack et al., 2008; Cappa et al., 2008). Recognizing the crucial importance of aerosol absorption for understanding aerosol–climate interactions, the UK Met Office has developed and tested a new state-of-the-art spectroscopic instrument for accurate measurement of aerosol optical properties. The instrument (EXSCALABAR; Table 1) employs multiple cavity ring-down extinction and photo-acoustic absorption spectrometers to determine multi-wavelength measurements of optical attenuation coefficients for dry, humidified, and thermally denuded aerosols to high precision and accuracy (Davies et al., 2018, 2019; Cotterell et al., 2019, 2020). Davies et al. (2019) examined the biases in filter-based retrievals of the aerosol absorption. While these biases were far more modest than those derived by Lack et al. (2008) and Cappa et al. (2008), they remained of a significant level (∼20 %) for aged BBA and depended on the correction scheme; the biases were reduced to levels of <11 % using advanced two-stream radiative transfer correction schemes (Müller et al., 2014) but took values up to 21 % when using the more common correction scheme of Bond et al. (1999). For other aerosol sources such as urban aerosols, Davies et al. (2019) report an overestimation of absorption from filter-based measurements using the correction scheme of Bond et al. (1999) of up to 45 %.

Davies et al. (2019) performed an analysis of the SSA of aerosol dominated by BBA in both the MBL and the residual CBL derived from EXSCALABAR and presented detailed probability distributions of the derived SSA, finding mean values of 0.84, 0.83, and 0.81 at 467, 528, and 652 nm respectively. Wu et al. (2021) extended this analysis, reporting column-weighted dry SSAs derived from EXSCALABAR, and found a mean and standard deviation of 0.85±0.02, 0.83±0.03, and 0.82±0.03 at 405, 550, and 658 nm respectively in the residual CBL, with evidence that the SSA increased with altitude in the residual CBL. Interestingly, these mean values are somewhat lower than those obtained during ORACLES-2016 measurements (Pistone et al., 2019; SSA at 500 nm of 0.85–0.88) but are in agreement with those from radiometric measurements, which do not rely on filter-based absorption instrumentation, derived from nine above-cloud flights of the NASA P3 aircraft during the 2016 and 2017 ORACLES campaign (Cochrane et al., 2020).

In the MBL, Wu et al. (2021) report SSAs and standard deviations derived from EXSCALABAR of 0.86±0.02, 0.85±0.03, and 0.84±0.03 at 405, 550, and 658 nm respectively. Zuidema et al. (2018b) report SSAs from the ARM site (at ∼330 m a.s.l., i.e. residing in the MBL) of 0.78±0.02 for August 2016–2017 and 0.81±0.03 for September 2016–2017 (interquartile range) at 529 nm, which suggest stronger absorption than the study of Wu et al. (2020). Zuidema et al. (2018b) acknowledge that the filter-based systems are dependent on the artefact-correction algorithm and use the mean of the correction from Virkkula (2010) and Ogren et al. (2010) algorithms. However, their filter-based measurements agreed with measurements made with those from an AERODYNE CAPS-SSA instrument deployed in July–September 2017; both yielded values of 0.77 at 529 nm. Without these additional measurements, the apparent discrepancy between the ARM and CLARIFY measurements could have been attributed to the remaining biases associated with filter-based correction algorithms. Davies et al. (2019) showed that these correction algorithms typically overestimate aerosol absorption, and without moving to more advanced two-stream correction algorithms (Müller et al., 2014), these correction algorithms underestimate the SSA by around 0.03 to 0.04 at 550 nm for measurements made during CLARIFY-2017. In addition, the MBL mass absorption coefficients are consistent between CLARIFY (Taylor et al., 2020) and LASIC (Zuidema et al., 2018b), indicating that it is the scattering measurements that differ between the two campaigns, rather than the more challenging absorption measurements. Work is currently underway to fully investigate these discrepancies. One possibility is that the impactor used in sampling the aerosol for CLARIFY may not correspond exactly to that for the AMF inlet (1.3 and 1.0 µm aerodynamic diameter respectively). Thus, a fraction of supermicron sea salt aerosols may increase the SSA for the CLARIFY-2017 measurements.

It is clear from the results of Wu et al. (2021) and Taylor et al. (2020) that many of the aged BBA particles in the vicinity of Ascension Island consist of a core of black carbon with a thick coating of organic and inorganic material (shell / core diameter ratio ranging from around 2.3 at the surface to approximately 2.6 at 5 km a.s.l.). Over the wavelength range 405–660 nm there is little evidence of absorption by organic “brown” carbon, but there is clear evidence of absorption enhancement via a lensing effect, whereby incident radiation is focussed onto the absorbing core of black carbon; this effect was also suggested by Zuidema et al. (2018b). Taylor et al. (2020) show that aerosol optical properties are not well represented when using Mie scattering theory and volume weighting of refractive indices that is currently used in many GCMs (general circulation models) such simple mixing rules are unable to simultaneously represent both the mass absorption coefficient and the SSA of the BBA. While the models documented in Table 3 utilize volume weighting of refractive indices, the resultant mass absorption coefficient (i.e. the mass-normalized aerosol absorption cross section) using a straightforward Mie theory model with these volume-weighted effective refractive indices does not agree with measurements derived from the EXSCALABAR, AMS, and SP2 instruments (Taylor et al., 2020). Internally consistent optical closure of both the optical parameters and the mass absorption coefficient can be improved using core / shell Mie scattering treatment of a black carbon core and an organic and inorganic coating but can be most accurately reconciled using more complex semi-empirical parameterizations of mixing state (Taylor et al., 2020).

BBA measured in the vicinity of Ascension Island was always very aged (>7 d from emission; Wu et al., 2021) and consisted of a thick coating of organics and inorganics surrounding an insoluble black carbon core (Taylor et al., 2020). While these measurements alone do not allow us to estimate the impacts of ageing on aerosol physical and optical properties, the same instrumentation has been flown during other campaigns, e.g. Methane Observations and Yearly Assessments (MOYA; Allen et al., 2017; Davies et al., 2019; Wu et al., 2020), that made measurements much closer to the source regions of the biomass burning over continental Africa (Davies et al., 2019; Wu et al., 2020). Wu et al. (2020) use identical measurement systems to those used during CLARIFY-2017 and report mean shell / core diameter ratios for BBA of as little as 1.07 (SD 0.10) for BBA less than 30 min subsequent to emission, increasing to 1.39 (SD 0.06) for BBA 3–6 h subsequent to emission, and 1.66 (SD 0.07) for BBA 9–12 h subsequent to emission. The shell / core ratios of 2.3–2.6 determined during CLARIFY-2017 suggest that the coating has continued to thicken as the BBA ages and the constituent components become increasingly internally mixed. Davies et al. (2019) also use nominally identical EXSCALABAR instrumentation during MOYA to determine a BBA SSA of around 0.91 at wavelengths close to 550 nm for BBA that has aged by 9–12 h since emission. Scanning electron microscope measurements made as long ago as SAFARI-2000 suggested that, on emission, black carbon consisted of individual spherules in chain-like structures (Posfai et al., 2003). Owing to surface tension effects, these chain structures collapse to more compact cores when coated by organic aerosol that was either formed at source or through the condensation of semi-volatile organic species within a few seconds from emission (e.g. Posfai et al., 2003; Abel et al., 2003). The black-carbon chain-like structures have a higher fractal dimension and a higher absorption efficiency compared to the more compact cores (Chakrabarty and Heinson, 2018). Together with the condensation of organic or volatile inorganics, which are predominantly scattering in nature, one might expect the SSA to increase with time (e.g. Abel et al., 2003). However, this condensation of scattering species can have the opposite effect, acting effectively as a lens focussing radiation on the absorbing core. Additionally, oxidation and nitration of the organic components could lead to an increase in absorption by “brown” carbon (e.g. Saleh et al., 2015), but conversely photochemical bleaching of BBA particles has been noted in laboratory studies (e.g. Zhong and Jang, 2014).

Trajectory simulations prove that aerosol high up in the atmosphere is generally younger (Dobracki et al., 2021; Sect. 6.1). From ORACLES measurements, Dobracki et al. (2021) suggest that the changes in the aerosol SSA as the aerosol ages are due to a reduction of organic material through evaporation. However, the vertical profile of the chemical composition of the BBA may be complicated by differences in the thermodynamic structure of the residual CBL and the condensation of inorganic nitrate into the aerosol phase (Wu et al., 2021; Taylor et al., 2020). Taken together, at the process level, the competing effects of fractal-chain collapse, evolving lensing effect from increasing coating thicknesses, changing absorption of the BrC coating, and the details of condensation and evaporation of volatile aerosol components make the ageing process particularly complex and not attributable to a single change in aerosol microphysics. What is clear is that the BBA measured during CLARIFY appears to be more strongly absorbing than that emitted at source, indicating that as BBA ages, the mechanisms that increase absorption outweigh those that may decrease absorption. However, at the time of writing, the contribution of the processes documented above remains an open issue and will undoubtedly be the subject of further work.

Aerosol–radiation interactions have been investigated at several scales. Peers et al. (2019, 2020) have developed a novel above-cloud aerosol detection algorithm from the geostationary SEVIRI instrument. Importantly, this retrieval accounts for the impacts of water vapour in the relatively wide SEVIRI spectral bands by assimilating humidity profiles from the Met Office NWP global model, leading to improvements in the accuracy of the retrievals (Chang and Christopher, 2016). A comparison against above-cloud retrieval algorithms developed from MODIS (Meyer et al., 2015) has been performed revealing some systematic differences, but overall the agreement in cloud and aerosol properties is satisfactory (Peers et al., 2020). The geostationary nature of the SEVIRI satellite instrument means that, unlike polar orbiting satellite retrievals which require precise colocation, coherent comparisons between aircraft and SEVIRI retrievals are possible. A number of cases have been investigated, with encouraging agreement between aircraft and SEVIRI retrievals (Peers et al., 2020). Additional work has confirmed the strong magnitude of the above-cloud direct radiative effect using OMI and MODIS (De Graaf et al., 2019), and a comparison of the above-cloud direct radiative effect from various space-borne sensors has been performed (De Graaf et al., 2020).

Herbert et al. (2020) examined semi-direct effects (sometimes referred to as rapid adjustments to aerosol–radiation interactions) from absorbing aerosol layers overlying the southern Atlantic stratocumulus deck using large eddy simulation (LES) modelling. Herbert et al. (2020) diagnose SDEs (W m-2) from changes in the cloud resulting in modelled fast feedbacks. SDEs diagnosed in this way in the region appear to have a strong diurnal cycle, peaking in the morning, so daily averaged SDE is much weaker than instantaneous values would suggest. Aerosol layers located immediately above the cloud exert strong SDE by affecting the temperature inversion at the top of the MBL and reducing the entrainment of air into the stratocumulus. The LES simulations suggest that this SDE weakens considerably with increasing distance between aerosols and clouds, with SDE exerted by aerosol layers 250 m away from the cloud top roughly half of that of layers located just above the cloud top and almost no SDE when the aerosol layer is 500 m above the cloud top. An analysis of lidar profiles from the NASA Cloud–Aerosol Transport System (CATS) lidar (5 km resolution, V3-00, mode 7.2, level 2 Daytime Operational Layer Data Product, 1064 nm wavelength), which, owing to the longer wavelength, has been shown to be able to penetrate through the thick BBA layers over the SE Atlantic (Deconu et al., 2019), suggests that in some 27 % of cases the aerosol base is within 500 m of the cloud top and therefore close enough to exert a SDE (Fig. 15). Additionally, in 22 % of cases the whole BBA layer is within 2 km of the cloud top; yet only 3 % of cases are within 1 km. Of course, this analysis is over a far greater area than that sampled in the vicinity of Ascension Island (Fig. 7) and includes areas off the coast of Angola and Namibia where “clear slots” (Hobbs, 2003; Haywood et al., 2003; Redemann et al., 2020) are more evident.

Note that these results contrast with others. The global-model-focussed work of Che et al. (2020) suggests that aerosol SDEs are stronger than those from direct and indirect effects (see section 6.9). In common with earlier studies (e.g. Johnson et al., 2004), the magnitude and the sign of the SDE are dependent on the relative location of BBA and clouds, as BBA can either increase the underlying cloud LWP or decrease the surrounding droplet numbers depending on whether the BBA are above or inside the cloud. Zhang and Zuidema (2019) also found that the cloud top inversion was often weaker when aerosol was likely present in the free troposphere, during August 2016–2017, rather than stronger, and attributed this to a meteorological effect; the residual of the CBL containing the aerosol is also often cooler than the air it is replacing. Such disagreements may stem from the relative ability of different models to resolve changes in boundary layer moisture and in the temperature inversion above stratocumulus, the lack of weakening of subsidence from BBA heating in LES studies (Myers and Norris, 2013), and confounding effects from meteorology.

The magnitude of aerosol–cloud interactions has been the subject of intense debate over the past 2 decades. While there is clear evidence of the Twomey effect (Twomey, 1977) from recent comprehensive satellite assessments of ship-tracks and degassing volcanoes (e.g. Malavelle et al., 2017; Christensen et al., 2014; Toll et al., 2017, 2019), there is generally a lack of clear evidence of the cloud lifetime or Albrecht effect (Albrecht, 1989) as the impact appears to be strongly dependent on the atmospheric state (e.g. Chen et al., 2015; Toll et al., 2019). On a cloud-by-cloud basis, there is considerable evidence from observational studies of subtropical stratocumulus (e.g. Comstock et al., 2004; Wood, 2005; Possner et al., 2020) that if there is an increase in the CDNC at a fixed cloud liquid water content, then the auto-conversion process by which cloud droplets grow to a size big enough to initiate precipitation should be inhibited.

On a statistical basis throughout the CLARIFY-2017 campaign, a strong cloud response to aerosol loading is observed (Fig. 16; see Barrett et al., 2021b, for more details), where increased aerosol concentrations (Na) under polluted MBL conditions (median Na=437 cm-3, interquartile range 305–569 cm-3) resulted in greater cloud droplet concentrations (median CDNC = 111 cm-3, interquartile range 62–170 cm-3) than were observed for clean conditions (median Na=53 cm-3, interquartile range 25–93 cm-3, median CDNC = 7 cm-3, interquartile range 1–34 cm-3), where clean and polluted conditions are defined using a CO threshold of 83 ppb. There was also a corresponding influence on the cloud droplet effective diameter, consistent with the Twomey effect (Twomey, 1977), whereby the cloud droplets were observed to be significantly smaller under polluted conditions than compared to clean conditions (cloud effective radii of 9 and 18 µm respective median values) despite the observation from the LASIC AMF that the LWP generally increases with pollution levels for the CLARIFY time period (Barrett et al., 2021b). Generally the cloud top altitudes (Zct) were found to be higher under polluted conditions than clean conditions. The influence of pollution on precipitation was also examined; approximately 42 % of in-cloud data points (defined as total water content from the Nevzerov probe > 0.01 g m-3) contained drizzle (defined where drizzle water content > 0.01 g m-3) in clean conditions compared to approximately 24 % in polluted conditions. Care must be taken when interpreting these initial results as they may be influenced by the overall sampling strategy and the potential influence of the BBA impacts on reduced subsidence (Myers and Norris, 2013); further work is ongoing examining the co-variability of cloud microphysical and meteorological influences. However, these conclusions do appear consistent with data derived from independent carbon monoxide (CO) and precipitation data obtained from the ARM mobile site and precipitation measurements from the Met Office located on Ascension Island. We extend the analysis period to August–September for 2016 and 2017 and precipitation data from the Met Office at Ascension Island which are less likely to be impacted by orographic effects than precipitation at the LASIC AMF site. Figure 17 shows the resulting distributions of CO concentration for precipitating and non-precipitating days.

Figure 17 suggests that precipitation does appear to be inhibited in polluted days when compared to non-polluted days as evident from the means of the distributions (77.8±30.4 (2 SD) for the 51/122 precipitating days and 95.7±60.8 (2 SD) for the 71/122 non-precipitating days). The mode CO is identical for the precipitating and non-precipitating days, while the CO distribution for the non-precipitating days shows evidence of a far longer tail to high values of CO. Note also that the patchy nature of precipitation over the Ascension Island region means that non-precipitating days will likely contain days when there is precipitation in the vicinity, but there is no precipitation detected at the Met Office site. Of course, we acknowledge that this is a purely statistical analysis that uses CO as a proxy for CCN and does not account for cause and effect, nor does it take account of covariability that might influence precipitation such as the boundary layer height, cloud top height and LWP, but these results appear consistent with the results from Zhang and Zuidema (2019), who analysed disdrometer data and polluted and non-polluted conditions determined from the AMF LASIC deployment. The ability of BBA to act as CCN over the SE Atlantic is further elucidated under the ORACLES project (Kacarab et al., 2020).

Certain flight days provided a particular wealth of information on cloud properties in the region. On 19 August 2017, for example, a large cloud feature south of Ascension Island about 1.5∘ in size was sampled at five different altitudes during straight and level runs. The aircraft followed a line of strong echoes on its flight instrument radar. Droplet size distributions and out-of-cloud samples of the aerosol size distributions were obtained at each level up to and above the cloud top height, which was around 2.5 km in altitude. This case study as the cloud changed from an overcast stratocumulus system to organized convective clouds is documented by Cui et al. (2021) and a model–observation comparison by Gordon et al. (2020).

The role that the free tropospheric BBA plumes observed over the SE Atlantic play in modulating the evolution of the underlying clouds via microphysical perturbations is dependent on where and when the BBA plumes mix down into the boundary layer. CLARIFY airborne and LASIC ground-based measurements from a case study of both a pocket of open cells (POC) and the surrounding stratiform cloud topped boundary layer highlighted that the efficiency of this entrainment of aerosol can depend on the form of the underlying cloud structure (closed vs. open cellular convection), with a marked reduction in entrainment of BBA in the region of open cells (Abel et al., 2020). An analysis of satellite imagery in Abel et al. (2020) demonstrates that these open cellular cloud regions occur regularly in the offshore environment surrounding Ascension Island during September. If the findings from this case apply more broadly, then this low susceptibility of open cells to intrusions of overlaying BBA could have important implications for aerosol indirect effects in the region, especially given that global climate models are generally not capable of simulating mesoscale features such as POCs, due to their coarse resolution and often relatively simplistic representation of aerosol–cloud–precipitation interactions.

The data collected during the CLARIFY campaign are already proving a valuable resource for modelling studies. In addition to the modelling work in support of flight planning (Sect. 1), preparatory work funded by the project involved testing the GLOMAP-mode aerosol microphysics scheme in the Unified Model at convection-permitting resolution over the south-east Atlantic (Gordon et al., 2018). The scheme was evaluated against satellites and data from the ARM site at Ascension Island and shown to perform well. The Unified Model global configuration was subsequently also shown to predict properties of smoke aerosol realistically in an evaluation against CLARIFY measurements of extinction (Che et al., 2020) and ORACLES 2016 measurements (Shinozuka et al., 2020).

One aspect that is interesting is that the direct radiative effect of above-cloud aerosol appears to be fairly independent of model resolution. This might be thought of as somewhat of a surprise because finer resolution models can include higher grid box mean AODs and CODs, and they might therefore be expected to give a wider range of direct radiative effects when compared to coarser resolution versions of the model (Fig. 18).

The simulations shown in Fig. 18 are nudged simulations with the Unified Model performed for August–September 2006 to coincide with POLDER observations. The coarser resolution models show a more spatially homogeneous spatial distribution of the direct radiative effect than the higher resolution models, but when averaged over the domain, the probability distribution of the direct radiative effect is a similar shape. That the model direct effects do not diverge as a function of model resolution is a testament to the validity of utilizing a model with identical underlying physics in the Unified Model framework.

A complementary global modelling paper (Che et al., 2020) examines a wider area of the south-east Atlantic in an atmospheric UK Earth System Model (UKESM1) configuration of the Unified Model, at N96 resolution, representative of typical climate model simulations to assess regional and global climate impacts over a longer time period. This work highlights the complex interaction of radiation, microphysics, and dynamical feedbacks. Decomposition of the radiative effects shows that the regional direct radiative effect is generally positive when the biomass burning plume is above the stratocumulus deck (with July–August average +7.5 W m-2), as the surface albedo of the underlying clouds is fairly high (e.g. Keil and Haywood, 2003). However, in UKESM1, rapid adjustments (semi-direct effect) enhance cloud albedo and more than compensate the direct effects, resulting in a net negative radiative effect over the region (July–August average -0.9 W m-2). Microphysical effects of aerosol–cloud further increase cloud albedo and the associated negative radiative effect. In the global mean, rapid adjustments due to biomass burning (semi-direct effects) appear negligible. In a separate work using the same model configuration, Che et al. (2021) perform a source attribution of CCN and resulting cloud droplet numbers for the CLARIFY domain. Che et al. (2021) estimate that during the biomass burning season, upper tropospheric binary nucleation between sulfuric acid and water (Vehkamäki et al., 2002) and BBA contributes a similar amount of CDNC,and these are the most important two sources of CCN in this region. This highlights the importance of upper tropospheric nucleation and subsequent subsidence in subtropical areas for the local cloud regime (e.g. Clarke, 1993) and highlights a route for significant microphysical aerosol effects on clouds, at considerable distance from anthropogenic source regions.

In a separate work based on using the same model configuration, Che et al. (2021) perform a source attribution of CCN and resulting cloud droplet numbers for the CLARIFY domain. Che et al. (2021) note that CCN from upper tropospheric binary nucleation contributes approximately 50 % to BBA impact on droplet numbers in the region, with negligible contributions from sea salt and boundary layer nucleation. This highlights the importance of upper tropospheric nucleation and subsequent subsidence in subtropical areas for the local cloud regime (e.g. Clarke, 1993) and highlights a route for significant microphysical aerosol effects on clouds, at considerable distance from anthropogenic source regions.

Further work on regional modelling includes the coupling of the GLOMAP-MODE two-moment aerosol microphysics to the CASIM two-moment cloud microphysics scheme, which allows for more refined studies of the indirect effect (Gordon et al., 2020). The current intensity of modelling activity suggests that the CLARIFY data set will be a valuable resource for model evaluation for many years to come.

The measurements obtained during the CLARIFY measurement campaign will also, together with data from ORACLES, LASIC, and AEROCLO-sA, contribute a key constraint on the representation of biomass burning aerosols in current climate models as part of the ongoing AeroCom Aircraft intercomparison study (https://wiki.met.no/aerocom/phase3-experiments#baseline_aircraft_experiment, last access: December 2020).