Clouds regulate climate by absorbing and scattering solar and thermal radiation. In present-day climate, clouds over the Southern Ocean account for 20% of the global net cloud radiative effect1. Despite their importance, Earth System Model simulations for the present-day consistently exhibit positive downwelling shortwave radiation biases over the Southern Ocean2,3,4. These biases have motivated efforts to collect more comprehensive Southern Ocean measurements of biogeochemical, aerosol, cloud, and precipitation properties3.

Sea spray and biogenic gases emitted by microbiota are the main sources of cloud condensation nuclei (CCN) and ice-nucleating particles (INP) in the Southern Ocean marine boundary layer. The region is mostly free of anthropogenic aerosol, offering a unique environment to study natural aerosol-cloud processes, critical for constraining radiative forcing of anthropogenic greenhouse gas and aerosol emissions5,6.

Here, we present measurements collected during the CAPRICORN-2 ship-based campaign2. We use in situ measurements of biogenic gases, aerosol composition and size, CCN, and INP, in conjunction with vertical profiles of cloud and precipitation properties and satellite-derived cloud droplet number concentrations7 (Nd) and effective radii (Re) in low-level clouds to explore aerosol-cloud interactions. With air mass back-trajectories (Supplementary Fig. 1), we identify periods influenced by the Antarctic continent (‘Antarctic’ air masses) and those influenced by the ocean (‘Oceanic’ air masses).

Periods of Antarctic air masses show significantly elevated surface CCN concentrations at 0.45% (CCN0.45 = 217 ± 73 cm⁻³) compared to Oceanic air masses (146 ± 101 cm⁻³, p < 0.001) (Fig. 1). Nd is also elevated during Antarctic air masses, and most of its variance during these periods is explained by surface CCN0.45 (R2 = 0.73, p < 0.05; Fig. 1, Supplementary Fig. 2). Similar enhancements in CCN concentrations across supersaturations from 0.25 to 1.05% (Fig. 2a), biogenic vapours (Fig. 2c, d; MSA and sulphuric acid) and submicron biogenic aerosol (Fig. 1, Fig. 2f, g; MSA and sulphate) are also observed during Antarctic air masses. These species are oxidation products of dimethyl sulphide (DMS), an important gas produced by ocean microbiota8. CCN0.45 during Antarctic air masses is very strongly correlated with these biogenic species (r > 0.8, p < 0.01) on daily timescales (Supplementary Fig. 2). CCN0.45 during Oceanic air masses is also strongly correlated with submicron MSA and sulphate (r > 0.8, p < 0.001) (but not their precursor gases), but also submicron sea salt (r = 0.72, p < 0.001). Unlike during Antarctic air masses, however, there is no significant relationship between surface CCN0.45 and Nd for during Oceanic air masses (r = 0.01; Fig. 1, Supplementary Fig. 2).

Fig. 1: The link between biogenic aerosol, CCN, and cloud droplet concentration over the Southern Ocean.
figure 1

a Daily location of the RV Investigator in January and February 2018 during CAPRICORN-2. b Cloud droplet number concentrations (Nd) of low-level non-precipitating clouds from MODIS overpasses as a function of daily averaged surface CCN concentrations at 0.45% supersaturation. A linear model was fitted separately for Oceanic and Antarctic air masses, with shaded areas representing the 95% confidence interval. Error bars represent the standard deviation of CCN and Nd retrievals across each day. In both panels, points are coloured by air mass origin and sized by PM1 Biogenic (MSA + sulphate) aerosol concentrations.

Full size image
Fig. 2: Aerosol, cloud, and precipitation differences across air mass-synoptic types.
figure 2

a Boxplots of CCN concentrations at supersaturations ranging from 0.25 to 1.05%. b Median aerosol number size distributions for different air mass-synoptic types, with shading representing the 25th and 75th percentiles. The Aitken mode for Oceanic and Antarctic air masses has central diameters of 33 ± 11 nm and 37 ± 14 nm, respectively, while the accumulation modes are centred around 91 ± 46 nm and 119 ± 33 nm, respectively. Inset shows distributions for diameters >0.7 µm. Shaded areas indicate the 25th and 75th percentiles. c–h Boxplots of gas-phase counts of MSA and sulphuric acid, PM1 MSA, PM1 sulphate, PM1 sea salt, and precipitation rates.

Full size image

Importantly, the highest concentrations of CCN, biogenic precursors, and biogenic submicron aerosol are observed when Antarctic air masses flow directly from the East Antarctic ice sheet (‘Antarctic Outflow’; 259 ± 37 cm⁻³), rather than Antarctic air masses that spend more of their recent history near the coast (Fig. 2, Supplementary Fig. 1). Previous studies have highlighted the prevalence of high Nd over Southern Ocean high-latitudes7,9, speculating that the biologically productive waters near the Antarctic coastline play a crucial role. It is evident, however, that the enhanced CCN (and Nd) present in Antarctic air masses is a result not only of the availability of biogenic gases, but also of long-range transport and atmospheric processing10,11.

Oceanic air masses exhibit a pronounced Aitken mode (at 33 ± 11 nm) in their aerosol size distributions, likely from the entrainment of recently formed particles in the free troposphere12, with fewer accumulation-mode particles relative to Antarctic air masses (Fig. 2b)13. This contrasting pattern between air masses has also been observed near the Antarctic peninsula14, and explains the higher sensitivity of CCN in Oceanic air masses to supersaturation (Fig. 2a) and Aitken-mode number concentration (Supplementary Fig. 2). The relative lack of Aitken-mode particles and enhancement in the accumulation mode (at 119 ± 33 nm) for Antarctic air masses is likely a result of cloud processing15 although a higher availability of condensable gases, relative lack of coarse-mode particles (Fig. 2b inset) that act as a condensation sink, and less precipitation (see Fig. 2h) that can scavenge activated particles16 could also contribute.

Oceanic air masses contain higher concentrations of sea salt (Fig. 2e), also evident in the elevated concentrations of coarse-mode aerosols. The role of sea salt aerosol in cloud droplet formation over the Southern Ocean is complex, and depending upon updraft velocities, can suppress the activation of sulphate aerosols17. At sizes relevant for cloud droplet activation (around the accumulation mode), Oceanic air masses contain an external mixture of sulphates, sea salt and an internally mixed component containing both sulphates and sea salt18 (Supplementary Fig. 3a). The sea salt in Antarctic air masses is strongly correlated with local wind speed (r = 0.73, p < 0.001), however shows significant chloride depletion relative to oceanic air masses (Supplementary Fig 3b, 4). Sea salt likely plays only a minor role in cloud droplet activation for Antarctic air masses, with most accumulation mode particles composed of an internal mixture of sulphur-based compounds resulting from DMS oxidation. There is significant variability in DMS chemistry across the Southern Ocean, evident in the enhanced concentrations of MSA in Antarctic air masses relative to Oceanic air masses (Fig. 2c, f). The partitioning of MSA between the gas-phase and aerosol-phase is complex, with the potential for MSA to evaporate from aerosol19. Despite this complexity, it is evident that DMS exerts a significant influence on high-latitude Southern Ocean Nd.

While cloud Nd is modulated by CCN availability, INP availability and secondary ice formation processes determine the liquid-ice balance within clouds, influencing radiation and precipitation. INP concentrations across the Southern Ocean are extremely low20. Although INP concentrations from Oceanic air masses are, at times, higher than those from Antarctic air masses, this is only evident below approximately −25 °C (Fig. 3a). INP concentrations at −30 °C show a strong positive correlation with sub-micron sea salt concentration (r = 0.76; p < 0.001; Fig. 3b), suggesting biological material emitted from sea spray as the likely source18,21. It is therefore plausible that the lower concentrations of INP at these low temperatures for Antarctic air masses are explained by a relative lack of sea spray compared to Oceanic air masses, with possible degradation due to ageing22. The role that these INP play in modulating low-level cloud phase, however, is still unclear. There is a higher minimum occurrence of low-level supercooled liquid-contain (SLW) clouds in Antarctic air masses relative to Oceanic air masses (Fig. 3c), but this is only evident at temperatures above −15 °C, where INP measurements are below detection-limit for both Oceanic and Antarctic air masses. Previous studies have shown that rime splintering, a secondary ice formation process, is also important over the Southern Ocean23. Rime splintering, however, requires cloud droplets with radii larger than ~12 µm24. Antarctic air masses with elevated CCN result in higher Nd with smaller effective radii (Re = 10.9 ± 2.8 µm) compared to Oceanic air masses (Re = 13.3 ± 2.8 µm; Supplementary Fig. 5). The higher CCN for Antarctic air masses therefore decreases the likelihood of rime splintering and could contribute to the higher minimum occurrence of SLW-containing clouds in Antarctic air masses.

Fig. 3: Variation in INP concentrations and SLW occurrence for Antarctic and oceanic air masses.
figure 3

a Ice- nucleating particle concentrations between −30 °C and −15 °C. Grey shading indicates the typical range of INP concentrations from marine environments globally43. b Daily-averaged real-time INP concentrations at −30 °C as a function of PM1 sea salt concentration, with error bars representing the standard deviation of INP and PM1 sea salt measurements for each day. A linear model was fitted, with shaded areas representing the 95% confidence interval. c The minimum occurrence of supercooled liquid-containing clouds for different air mass origins as a function of temperature.

Full size image

Enhanced Nd can inhibit precipitation, and unglaciated clouds are also less likely to precipitate. Whether or not aerosol-cloud interactions have a large influence on precipitation at high Southern Ocean latitudes relative to meteorological processes, non-precipitating clouds also allow for further cloud-processing of aerosol that can shift the distribution from the Aitken mode up to the accumulation mode15, potentially sustaining higher concentrations of CCN. This is not the case for lower latitude Southern Ocean clouds, where coalescence scavenging during precipitation can be a significant sink of CCN and Nd16,25.

This study highlights the need to improve our understanding of biological, chemical, and physical processes that influence cloud properties across the Southern Ocean. These properties vary significantly between air masses that are influenced by purely marine air and the Antarctic continent. Fluxes of DMS and organic compounds that facilitate new particle formation26 must be constrained, and a thorough exploration of the microorganisms in the ocean and sea ice responsible for their production is needed. The chemistry that leads to particle formation and growth of aerosol to CCN sizes in the unique conditions around Antarctica also needs attention. The exchange and transformation of reactive gases and aerosols between the boundary layer and free troposphere, and how this is mediated by, and influences, cloud processing and phase, and precipitation, represents a substantial gap in our knowledge of this region.

The future is more uncertain. There is evidence that Antarctic sea ice has recently undergone a regime shift27. Sea ice extent is likely to decrease in the future, which would increase fluxes of reactive gases and sea spray aerosol, moisture and heat. On the other hand, ice-free coastal areas will likely increase in the future28, providing a potential new source of effective dust INP. The magnitude of these effects on cloud and precipitation properties is unknown. The uncertainty associated with these future changes in the Antarctic environment and ecosystem, combined with a warming climate, poses serious challenges to our ability to simulate future climate. We show that complex interactions between aerosols, clouds, and precipitation at higher latitudes are very different from those at lower latitudes of the Southern Ocean, and the role that biological emissions have in these interactions cannot be ignored in our efforts to model this environment9.

Methods

Voyage

CAPRICORN-2 (Clouds Aerosols Precipitation Radiation and atmospheric Composition over the Southern Ocean) took place on the Research Vessel (RV) Investigator from 11 January to 21 February 2018 between Hobart, Australia, and the ice edge of East Antarctica. Details of the deployment and an overview of the campaign can be found in McFarquhar et al.2.

Air mass classification

Air mass back trajectories were calculated using HYSPLIT29 every hour at a height of 100 m above sea level from the RV Investigator’s position. The Global Data Assimilation System 0.5° meteorological data was used for the HYSPLIT trajectory calculations. Air mass backwards trajectories were categorised into three categories. Trajectories with a mean terrain height over the past 120 h greater than 300 m and a mean latitude south of 65°S were categorised as “Antarctic”. Those that had a mean latitude north of 45°S and a radon concentration above 100 mBq m−3 were categorised as “Australia”. All other trajectories were considered to be “Oceanic”, with mostly marine influence within the last five days. After removing hours that were influenced by local exhaust contamination from the ship, the “Antarctic”, “Oceanic”, and “Australian” air masses accounted for 228, 537, and 33 h, respectively. In our case, the relative occurrence of “Antarctic” and “Oceanic” air masses is very insensitive to the choice of mean terrain height threshold used in the classification. The brief influence of “Australian” air masses was excluded from the analysis presented in this Brief Communication.

Antarctic air masses were further split according to the synoptic types defined in Truong et al.30, that used k-means clustering on thermodynamic properties measured by the radiosondes launched from the RV Investigator. A synoptic type was assigned to each hour of the dataset according to the closest radiosonde launch in time. Three synoptic types were identified for Antarctic air masses; “Antarctic Cyclone”, representing Antarctic air masses close to cyclone centres (130 h); “Antarctic Outflow”: Antarctic air masses that had outflow from the Antarctic continent (45 h); and “Antarctic Post-front”: Antarctic air masses post-cold-front and northwest of cyclone centres (35 h). This Brief Communication focuses on the distinction between the Antarctic air masses and synoptic types and the Oceanic air masses (see Supplementary Fig. 1), and so differences between synoptic types for the Oceanic air masses were not considered in this analysis.

In situ data

The in situ data measured on the RV Investigator underwent data filtering to remove periods of contamination from the ship’s exhaust stack. First, an automated exhaust identification algorithm was applied31 and then rapid increases in CN, black carbon, carbon monoxide and carbon dioxide concentrations were then used to manually identify additional periods of exhaust contamination. These steps, fully described in Humphries et al.32 removed 14% of the data over the CAPRICORN-2018 measurement period.

CCN concentrations at 0.25, 0.35, 0.45, 0.55, 0.65, and 1.05% were measured with a DMT CCN-100. Measurements for each supersaturation were collected for 10 min before ramping to the next supersaturation, giving 10 min per hour at each supersaturation33. Significance testing for comparing CCN concentrations for different air masses was done using a Mann-Whitney U test.

INPs were collected on aerosol filters and measured using an Ice Spectrometer for temperatures between −30 and −10 °C. Real time INP concentrations at −30 °C presented here were also measured using a Continuous Flow Diffusion Chamber. More details are described in Moore et al.20.

The non-refractory sub-micron aerosol composition was measured with an Aerodyne Time-of-Flight Aerosol Chemical Speciation Monitor (ToF-ACSM). Separately, PM1 aerosol was also collected on 47 mm quartz filters, sampling for between 20 and 48 h. The anion and cation concentrations from these PM1 filters were measured with a Dionex ICS-3000 reagent-free ion chromatograph. Further details can be found in Humphries et al.32,33. Although a revised fragmentation table and dataset are available for the ToF-ACSM, we do not use it here because it did not offer any significant improvement in the correlations with the PM1 filter measurements described below.

While the ToF-ACSM offers a high temporal resolution for aerosol composition, the total concentration for different species depends on a fragmentation table, which might not be reliable for some species, such as MSA. Averaging the ToF-ACSM data over each filter collection period shows that the calculated concentration of ToF-ACSM MSA is underestimated by a factor of 8 ± 4. There was a strong correlation (R2 = 0.63) between the filtered and ToF-ACSM concentrations of MSA, however, so we applied a correction to the ToF-ACSM MSA concentration to take advantage of its high temporal resolution. We fitted a linear regression model to the ToF-ACSM and filtered MSA data. Taking into account the standard deviation of MSA concentration from the ToF-ACSM over the filter collection periods and an assumed uncertainty in the filtered concentration of MSA ( ± 5%), we ran a Monte Carlo simulation to estimate the uncertainty in the linear regression coefficients. Based on these coefficients, we then calculated a corrected ToF-ACSM MSA concentration with a mean uncertainty of ±27%. Other organic species are not presented here, but were generally below 0.05 µg m−3.

Similarly, because the ToF-ACSM vaporiser is set to 600 °C, it only vaporises a small fraction of sea salt-related ions. Based on the sea salt concentration of the PM1 filters, we calculated the total PM1 sea salt concentration from the concentration of chloride-related fragments in the ToF-ACSM using the same process as above for MSA. The uncertainty in this estimate was high (±57%) due to the relatively weak relationship between the ToF-ACSM chloride and PM1 filtered sea salt concentrations (R2 = 0.24). This relatively weak correlation is likely due to a combination of the ToF-ACSM detection limits for chloride (particularly during Antarctic air mass periods with little sea salt), variability in internal parameters of the ToF-ACSM, as well as variability in the relative contributions of different sea salt species.

The aerosol size distribution was measured with three separate instruments. This included a GRIMM Nano-SMPS (3.8–409 nm), a TSI long-column SMPS (15.1–661 nm), and a TSI APS (523 nm–19.8 µm). A merged size distribution was calculated for each hour of the campaign, taking the GRIMM Nano-SMPS size bins up to 35.4 nm, the TSI long-column SMPS size bins up to 615 nm, and the APS size bins above 615 nm. The APS diameters were converted from aerodynamic diameter to electrical mobility diameter assuming dry sea salt properties with a density of 2.2 g cm−3 and a shape factor of 1.0834. A bimodal lognormal fit was applied to the hourly-averaged merged size distribution data following Modini et al.35 to estimate number concentrations of the Aitken and accumulation modes.

Reactive gases were measured using an Aerodyne Time-of-Flight Chemical Ionisation Mass Spectrometer (ToF-CIMS) with (HNO3)nNO3- (n = 0,1) as reagent ions. We extracted ionised counts of MSA (CH3SO3- and (HNO3)CH3SO3-) and sulphuric acid (HSO4- and (HNO3)HSO4-) and normalised them to the sum of reagent ions. An ionisation efficiency calibration was not done for these species and therefore we only report normalised counts.

Aerosol hygroscopic properties were measured using a Hygroscopicity Tandem Differential Mobility Analyser (H-TDMA). In this study, the initial differential mobility analyser (DMA) alternated size selection between 40, 100, and 150 nm every 6 min. The sample was then passed through a nafion set up to expose the aerosol to a relative humidity of 90%, before being passed to an SMPS to measure the hygroscopic growth factor. The operation of the H-TDMA is described in detail in Cravigan et al.36.

Cloud and precipitation data

Nd were determined from MODIS Level 2 retrievals for non-precipitating low-level clouds following Mace et al.7. All available derived counts within 2.5° latitude and longitude of the RV Investigator’s position for each day were taken and averaged to compare with the surface measurements. The uncertainty associated with MODIS-derived Nd is described in Grosvenor et al.37 and estimated as 54% for 1 by 1° spatial retrievals. Further details on MODIS-derived Nd specific to the Southern Ocean are discussed in detail in Mace et al.7, with potential biases in the assumed adiabaticity and the cloud droplet size distribution variance unlikely to significantly influence the outcome of our study.

Cloud reflectivity was measured by a BASTA 95 GHz Frequency Modulation-Continuous Wave Doppler radar on the RV Investigator38. Cloud backscatter and depolarisation were measured with a Leosphere Raman UV-polarisation Lidar on the RV Investigator. Cloud phase as a function of height was determined by merging data from cloud lidar and radar along with meteorological data from ERA5, following Alexander and Protat39. Here, pixels that were identified as mixed-phase or supercooled liquid were classified as “SLW-containing”. The altitude-dependent occurrence of SLW-containing clouds is treated as the minimum occurrence due to the possibility of undetected liquid cloud layers above ice clouds that fully attenuate the lidar signal. Precipitation properties at the surface were measured by an ODM470 optical disdrometer following Protat et al.40.