Simulations were performed with HadGEM3-GA7.1 , which exhibits more realistic aerosol effective radiative forcing compared with preceding versions . Aerosol emission, evolution and deposition are simulated with the Global Model of Aerosol Processes (GLOMAP-mode), in which sulfate, sea salt, black carbon and particulate organic matter aerosol are represented in five log-normal size modes. These correspond to particle size ranges of ≤10 nm (nucleation mode), 10–100 nm (Aitken mode), 100–1000 nm (accumulation mode) and ≥1000 nm (coarse mode) . All modes are soluble, and an insoluble Aitken mode is also included. Mineral dust is represented in the model using a bin emission scheme .

Aerosol–cloud interactions are represented via the UKCA-Activate scheme , which simulates the number of aerosols activated into cloud droplets. CCN are defined as aerosols with a diameter ≥50 nm, which is the minimum size of aerosol that activates with a supersaturation of approximately 0.3 % . The number of activated aerosols is calculated via Köhler theory and depends on aerosol size, composition and number, along with the local temperature, pressure and vertical velocity . Because the grid cell sizes in global models are too large to resolve cloud base updraft velocity, a probability density function represents the likely distribution of vertical velocity within each grid box at each time step. The cloud droplet number concentration (Nd) is calculated from the number of activated aerosols at the cloud base, weighted by this probability density function . The number of cloud droplets subsequently influences the cloud albedo, as clouds with larger Nd (and smaller droplets) are optically brighter .

HadGEM3-GA7.1 scales marine DMS emissions by a factor of 1.7 to account for missing sources of marine organics, which yields a better representation of Nd compared with observations . Here, we use a modified configuration of the model, GA7.1-mod, which includes marine organics instead of DMS emission scaling. Furthermore, the GA7.1 standard configuration uses a simplified chemistry scheme, whereby chemical oxidants such as O3, OH, NO3 and HO2 are prescribed as “offline” monthly-mean climatologies in order to reduce computational time. In this study, the model used an online chemistry scheme, StratTrop (also known as CheST – Chemistry of the Stratosphere and Troposphere), which is a combination of the stratospheric and tropospheric chemistry schemes described by and , respectively.

The StratTrop scheme uses a Newton–Raphson solver and accounts for DMS oxidation via the gas-phase and aqueous-phase reactions shown in Table . The oxidation of DMS by OH proceeds by both an addition and abstraction pathway (the first two reactions listed in Table ), and can produce SO2 and MSA. The relative yields of these products are important, as SO2 leads to new particle formation, while other products such as MSA condense on existing particles, therefore increasing their size .

Gas-phase SO2 enters the liquid phase via an equilibrium approach described by Henry's law. Because SO2 dissociates in the aqueous-phase (Reactions  and ), it is more soluble than the equilibrium Henry's law constant (KH) implies. R1SO2⇌H++HSO3-R2HSO3-⇌H++SO32-

Therefore, the model uses an effective constant (KHeff) which for SO2 is related to KH by Eq. (). 1KHeff=KH1+kR1H++kR1kR2H+2

kR1 and kR2 are the equilibrium constants for the aqueous-phase dissociations shown in Reactions () and (). The hydrogen ion concentration (H+) is set as a global number in the model, equivalent to a constant pH of 5.

SSA is generated via a wind-speed-dependent parameterisation based on whitecap coverage . This function is based on the semi-empirical function by but improves the representation of small particles less than 0.1 µm in diameter. According to , the number of seawater droplets generated per square metre of sea surface, per increment of particle radius over 20 size bins, is calculated via Eq. (): 2dFdr=1.373u103.41r-A1+0.057r3.45×101.607e-B2.

The exponential terms A and B are defined by Eqs. () and (): 3A=4.71+Θr-0.017r-1.44,4B=0.433-log⁡(r)0.433, where r is the particle radius at a relative humidity of 80 %, Θ is an adjustable parameter that controls the shape of the size distributions, and u10 is the scalar horizontal wind speed at 10 m above the surface.

A 20-year reference simulation (“REF”) was performed from 1989 to 2008 to evaluate the model. SSTs and greenhouse gas concentrations were based on observations. Emissions of NOx, CH4, CO, SO2, isoprene, monoterpenes, ethane, propane, formaldehyde, acetaldehyde, acetone, NH3, black carbon and organic carbon were prescribed based on the year 2000 . A further eight sensitivity simulations were performed, each of 10-year duration, from 1989 to 1998. These were designed to test the sensitivity of simulated aerosols to the choice of SSA source function and sulfate chemistry scheme, and are summarised in Table . All simulations used the DMS seawater climatology of and the DMS sea–air exchange parameterisation of . Simulations were run with N96 horizontal resolution (i.e. grid sizes 1.875∘×1.25∘ in size) and 85 levels between the surface and 85 km.

Analysis of aerosol measurements made on a 2018 Tangaroa research voyage on the Southern Ocean indicate that the dependency of SSA production on near-surface wind speed (u103.41) is overestimated by a factor of 2–4 via the source function (Eq. ). Recent research by indicates that Eq. () with SSA production dependent on u102.8 is a better fit to observed SSA concentrations in an environment dominated by high wind speeds such as the Southern Ocean. The “SSF” (SSA source function) simulation therefore aims to test this using HadGEM3-GA7.1-mod. CHEM1-SSF, CHEM2-SSF and CHEM3-SSF also use the SSA source function described by Eq. (), in combination with different sulfate chemistry schemes as described below. 5dFdr=2.6u102.8r-A1+0.057r3.45×101.607e-B2

The SSA source function is based on a series of in situ measurements of the total suspended sea spray concentration within the Southern Ocean boundary layer. The total concentration of sea spray was constrained from the number concentration size spectra measured with a PCASP-100X optical particle counter during a voyage from Wellington, New Zealand, to the Ross Sea in February–March 2018.

After the voyage, the Lagrangian particle trajectory model FLEXPART-WRF was used to develop source–receptor relations between the upwind environment and the in situ measurements. The source–receptor framework acted as a bridge through which several different formulas for the sea spray source function could be optimised. The newly optimised functions all found that the parameterisation produced too much sea spray at high wind speeds, as described by and previous studies including , and .

One of the newly optimised parameterisations developed by took a power-law form (i.e. Eq. ), similar to the parameterisation (Eq. ). We tested this parameterisation, as it was straightforward to implement in HadGEM3-GA7.1. show that the two power-law parameterisations differ primarily at high wind speeds, which are commonly observed over the Southern Ocean. For example, when u10=4 m s-1, both parameterisations predict the same SSA flux. However, when u10=11 m s-1, the SSA parameterisation predicts a SSA flux which is 40 % smaller than that predicted by .

validated their newly optimised parameterisations by comparing predicted SSA concentrations against airborne data collected on HIAPER (the NSF/NCAR High-performance Instrumented Airborne Platform for Environmental Research) as part of the SOCRATES (Southern Ocean Clouds, Radiation, Aerosol Transport Experimental Study) campaign. The goodness of fit between predictions and airborne measurements validated the use of the new parameterisations (including Eq. ) over the Southern Ocean.

DMS oxidation chemistry is complex ; however, the set of reactions describing the conversion of gaseous DMS into sulfate aerosol in StratTrop (Table ) is simplified due to the computational cost of calculating chemical reaction rates. We tested three alternative reaction schemes with incremental increases in complexity, with the aim of identifying how sensitive Southern Ocean aerosols and clouds are to the choice of chemistry scheme. The three sulfate chemistry schemes investigated in our CHEM1, CHEM2 and CHEM3 simulations are described in Table . The CHEM1 and CHEM2 sensitivity simulations use the same aqueous-phase sulfate chemistry scheme as REF (i.e. the default StratTrop scheme included in HadGEM3-GA7.1-mod) but with increased complexity of the gas-phase chemistry. CHEM1 includes DMS oxidation by halogens as they have been shown to play an important role in the remote marine atmosphere . CHEM2 includes a gas-phase DMS oxidation scheme based on the scheme recently developed for the marine troposphere by . CHEM3 is identical to CHEM2 except that additional aqueous-phase sulfate reactions are included. This scheme is based on the aqueous-phase scheme by ; however, the oxidation reactions by OH are excluded as OH uptake into cloud droplets is subject to numerous uncertainties and is not currently implemented in HadGEM3-GA7.1-mod. The new scheme also includes aqueous-phase treatment of MSA, which is treated as a sink of DMS in HadGEM3-GA7.1-mod and does not oxidise to form aerosol. In the NODMS simulation, DMS emissions are switched off to help isolate its role in the annual AOD cycle over the Southern Ocean.

Figure  shows climatological monthly zonal-mean AOD between 40 and 60∘ S as simulated by HadGEM3-GA7.1-mod and observed by MODIS and MISR. The seasonality in AOD over the Southern Ocean is similar between MODIS and MISR, as shown previously by . The model generally agrees with the maximum, minimum and mean AOD observed by MODIS (Fig. a, b). However, the simulated seasonal cycle is out of phase. The model simulates too much aerosol in winter (JJA) and too little in summer (DJF) compared with satellite observations (Fig. c, d).

As discussed earlier, sulfate aerosol from biogenic sources and SSA predominantly contribute to AOD over the Southern Ocean. By performing a simulation with DMS emissions switched off (the NODMS simulation) and comparing it with the REF simulation (Fig. d), it is apparent that the model simulates primarily SSA during winter (July and August, 50–60∘ S). This result is consistent with the Aerocom (Aerosol Comparisons between Models and Observations) phase II models, which simulate a seasonal maximum in sea-salt AOD during winter at southern high latitudes, while sulfate AOD maximises in summertime.

In DJF, AOD is approximately 60 % lower in the NODMS simulation compared with REF, indicating that sulfate aerosol of marine biogenic origin is primarily produced during summertime when increased solar radiation and warmer waters make the ocean more biologically productive. Indeed, measurements at Baring Head (41∘ S, 179∘ E) indicate that sulfate in fine aerosol modes is mostly secondary sulfate from marine DMS emission, exhibiting an annual maximum in summertime, while coarse sulfate aerosol is mainly from sea salt and is relatively constant throughout the year .

Total AOD is calculated in HadGEM3-GA7.1 from adding together the individual contributions of dust AOD and the Aitken-mode (soluble plus insoluble), accumulation-mode and coarse-mode AODs. Aerosol particles in the soluble modes may activate to cloud condensation nuclei, and the contribution to total AOD from these modes is shown in Fig. a–c. Coarse-mode aerosol is the major contributor to total AOD due to its size and maximises in Southern Hemisphere autumn, winter and spring (Fig. c), implying that it is mostly SSA as discussed above. Accumulation-mode aerosol (Fig. b) shows a clear seasonal cycle which maximises during summer (as shown previously by, e.g. ), indicating that this is mostly aerosol of marine biogenic origin. Aitken- and accumulation-mode aerosol increases during springtime at 40∘ S, associated with long-range transport of particulate matter from South America, Australia and South Africa .

Given that HadGEM3-GA7.1-mod primarily simulates SSA during winter, we now examine SSA production in more detail. Zonal-mean near-surface wind speeds between 40 and 60∘ S are shown in Fig. a. The model agrees reasonably well with the ERA-Interim reanalysis , at least in the zonal mean. However, the actual position of the storm track tends to be zonally shifted in the model (not shown), which is associated with the model's shortwave radiation bias discussed in Sect. . While sparse observations over the Southern Ocean lead to some uncertainty regarding the comparative accuracy of near-surface wind data sets in reanalyses, indicate that ERA-Interim is the most reliable of contemporary reanalyses. Furthermore, ERA-Interim near-surface winds agree well with other reanalyses such as the National Centers for Environmental Prediction Climate Forecast System Reanalysis (NCEP/CFSR) and NASA Modern-Era Retrospective analysis for Research and Applications (MERRA) .

Wind speeds over the Southern Ocean show a clear seasonality maximising between autumn and spring, and thus we expect more SSA to be produced during this time. As described in Sect. , the model uses the SSA source function of in which SSA generation scales according to a power law with wind speed. The correlation between simulated AOD and wind speed is shown for July 2003–2007 in Fig. . July was chosen as a representative wintertime month during which wind speeds are high over the Southern Ocean (Fig. a) and aerosol predominantly consists of sea salt (Fig. d). Comparing this with a similar regression model fit derived from MODIS and ERA-Interim data illustrates that the wind speed dependency of SSA production over the Southern Ocean is overestimated in the model as evidenced by the regression model fits obtained. This is supported by SSA measurements made on the 2018 Tangaroa research voyage, which indicate that SSA production requires a threshold wind speed below which no SSA is produced. The measurements also show that the SSA flux predicted by the parameterisation increases too quickly as a function of wind speed. These two effects result in overproduction of SSA at all wind speeds by a factor of 2–4 .

AOD over the Southern Ocean in the SSF simulation using the source function (see Eq.  and Table ) is shown in Fig. b. Compared with the REF simulation (Fig. c) the reduction in AOD is reasonably uniform throughout the year, with the reduction in coarse-mode AOD (shown for REF in Fig. c) between March and November particularly visible. Comparing to MODIS observations, the source function performs well during winter months when SSA is the dominant contributor to AOD (Fig. d). Changes in aerosol mode number concentrations and dry diameters in the SSF simulation are discussed later in Sect. .

Our finding that the SSA contribution to AOD is overestimated in the REF simulation is consistent with the Atmospheric Chemistry and Climate Model Intercomparison Project (ACCMIP) models, which overestimate annual-mean sea-salt AOD between 50 and 60∘ S compared to observations from AERONET Sun photometers . Our results are also consistent with previous work suggesting that the source function overestimates the SSA dependency on wind speed . found that using the GEOS-Chem chemical transport model, the SSA source function led to overestimation of coarse-mode SSA in the atmosphere by a factor of 2–3 at high wind speeds and suggested that the discrepancies are dependent on sea surface temperature. Similarly, showed that sea-salt surface concentrations are overestimated compared with observations at southern high latitudes in chemical transport model simulations using the source function. However, by implementing a weighting factor based on sea surface temperature as suggested by , their model simulated SSA concentrations that are in closer agreement with observations .

How SSA should ultimately be represented in global models remains the subject of ongoing research. Along with , other studies have found that SSA concentrations are correlated with sea surface temperature . More recently, demonstrated that variability in seawater composition may have just as large an impact on SSA production as temperature. Nonetheless, our results demonstrate that for the Southern Ocean during winter when SSA is the dominant contributor to AOD, reducing the wind speed dependency of SSA production results in good agreement between the model and observations. However, Fig. d points to the existence of partially compensating errors, namely that sulfate aerosol is underestimated in the model during summertime even more than suggested by the REF simulation. Given the importance of sulfur chemistry in the marine atmosphere, we now discuss the CHEM simulations and sensitivity of simulated aerosols and cloud microphysics to the choice of sulfate chemistry scheme.

Seawater DMS from the climatology used as input to HadGEM3-GA7.1-mod is shown in Fig. . Seawater DMS concentrations maximise in austral summer along the Antarctic continent following sea ice melt and the corresponding release of aerosol precursors by phytoplankton which grow on the underside of sea ice . In the climatology, the maximum summertime DMS concentration reached at southern high latitudes during DJF is up to 15 nM lower than in the older seawater DMS climatology. Using the ECHAM5-HAMMOZ model, showed that use of the climatology improved the simulation of DMS at Amsterdam Island, particularly during summertime when observed concentrations are large. However, the climatology includes large uncertainties as it was compiled from cruise observations interpolated to make a global climatology. These uncertainties translate to variations in Nd between 2 and 5 cm-3 in HadGEM3-GA7.1 .

Simulated surface atmospheric DMS concentrations in the REF simulation agree reasonably well with measurements from the SOAP and SOIREE voyages (Fig. a), although the spread in measurements varies by almost 1000 ppt. The model does not capture such a large spread in variability, likely because it provides output averaged over coarse horizontal grid cells, and SOAP sought out the highest chlorophyll-/DMS-containing waters at the time of the voyage. To examine the seasonal cycle in atmospheric DMS, we compare model results with measurements obtained from Amsterdam Island (Fig. b) and Cape Grim (Fig. c). For both observations and the model, the summertime maximum coincides with the peak of phytoplankton productivity. At Amsterdam Island, the REF simulation underestimates DMS in January by 55 % and overestimates it in July by a factor of 3. However, at Cape Grim, DMS is overestimated throughout the year in the REF simulation and simulates approximately 5 times too much DMS in January. The large DMS concentrations simulated at Cape Grim relate to the seawater DMS climatology, which shows a region of high DMS productivity nearby (Fig. b).

In all three CHEM simulations, the change in the simulated surface atmospheric DMS concentration is negligible relative to the magnitude of the seasonal cycle in DMS (Fig. b, c). Figure a shows DMS in the lowest 2 km of the atmosphere over the Southern Ocean. In the CHEM simulations, the DMS concentration is 7 %–13 % larger than in REF. This likely relates to the rate constant for the first DMS + OH reaction listed in Tables  and , which is an order of magnitude smaller in the new scheme tested compared with the existing StratTrop scheme, implying that it will proceed more slowly, and therefore less DMS will be oxidised.

SO2 concentrations decrease with height above 0.5 km altitude (Fig. b), which is the approximate cloud base (Fig. ). Surface SO2 concentrations are almost 30 ppt lower in the CHEM1 simulation compared with REF. This is likely due to the implementation of reactions between DMS and halogens (DMS + BrO and DMS + Cl), which may convert the sulfur in DMS to DMSO and SO2 (rather than only SO2; see Table ). In particular, the DMS + BrO reaction has been shown to be particularly important in the remote marine troposphere . Measurements at Baring Head (41∘ S, 179∘ E) during February and March 2000 indicate that the SO2/DMS ratio of clean boundary layer air originating from over the Southern Ocean is approximately 0.06–0.26 . Our simulated ratios of surface SO2/DMS over the Southern Ocean (40–60∘ S) agree with this measured range for the REF, CHEM1, CHEM2 and CHEM3 experiments (0.19, 0.11, 0.15 and 0.13, respectively).

Examining the global H2SO4 distribution in the REF simulation reveals that H2SO4 mixing ratios over the Southern Ocean in DJF are larger than any other region (not shown), consistent with the ECHAM5-HAMMOZ model . H2SO4 concentrations are increased by ∼0.015 to 0.025 ppt relative to REF in the CHEM2 and CHEM3 simulations (Fig. c) due to the extra DMS oxidation reactions added.

Figure  shows vertical profiles of aerosol mode number concentration and particle dry diameter over the Southern Ocean in the REF and sensitivity simulations. For reference, the mean mass fraction of cloud liquid water in DJF over the Southern Ocean is shown in Fig.  to illustrate that the aerosol profiles we examine are situated within the cloud layer. In the SSF simulation, decreasing the dependency of SSA generation on wind speed means that the number concentration of accumulation- and coarse-mode particles decreases by 30 %–50 % (Fig. b and c). The particle dry diameters in these modes are largely unchanged (Fig. e and f). However, the soluble Aitken mode changes; the number concentration increases by ∼40 % in the SSF simulation relative to REF and the average particle dry diameter decreases by 10 nm (Fig. a and d). Initially this was unexpected, as SSA is emitted only into the accumulation and coarse modes in the model, and not the Aitken mode . The change in the Aitken mode likely comes from smaller non-SSA particles (e.g. sulfate aerosol) being unable to coagulate on larger SSA particles, as these are reduced in number.

In the CHEM simulations, the coarse mode remains largely unchanged regardless of the chemistry scheme used (Fig. c, f). In CHEM2 and CHEM3 simulations, there are more smaller particles in the accumulation mode which are smaller on average (Fig. b, e), which has implications for cloud microphysics. As discussed earlier, soluble aerosols such as sea salt and sulfate with a diameter ≥50 nm can become activated to CCN, corresponding to a supersaturation of ∼0.3 %. Simulated CCN and Nd over the Southern Ocean are shown in Fig. . In the REF simulation, summertime-mean CCN concentrations at 800 m above the surface average 120 cm-3, which is the same as measurements reported at Cape Grim (41∘ S, 145∘ E, 0.23 % supersaturation; ) and Princess Elisabeth Antarctic Research Station (72∘ S, 23∘ E, 0.3 % supersaturation; ).

At southern high latitudes, the number fraction of SSA CCN is larger than any other region on the globe . Therefore, due to the reduced aerosol abundance in the SSF simulation, CCN concentrations also decrease by up to ∼13 % relative to REF (Fig. a). In the CHEM simulations, CCN concentrations decrease by -18 % (CHEM1) to +25 % (CHEM2 and CHEM3), which is likely linked to the changes in accumulation-mode aerosol shown in Fig. . The changes in CCN in the CHEM simulations translate to changes in Nd over the Southern Ocean of -13 % in CHEM1 to +20 % in CHEM2 and CHEM3 (Fig. b). show that in HadGEM3-GA7.1, the simulated seasonal cycle in Nd over the Southern Ocean is primarily driven by seawater DMS emissions. While the model captures the observed seasonality in Nd, the magnitude is too low, which was also reported by . However, the CHEM2 and CHEM3 simulations bring the model into better agreement with Nd observations.

Of all the CHEM and CHEM-SSF sensitivity simulations, AOD in the CHEM1 simulation agrees most favourably with MODIS (Fig. a), and the root mean square error between 40 and 60∘ S has decreased slightly (from 0.032 to 0.028) following the original REF and MODIS comparison (Fig. c). However, the seasonal bias remains. The CHEM1-SSF simulation shows good agreement with MODIS AOD during austral winter but underestimates summertime AOD and Nd (Figs. b and d). CHEM2-SSF and CHEM3-SSF show the reverse; simulated summertime AOD agrees well with MODIS but wintertime AOD is too high, even with the new SSA source function included. In terms of simulating Southern Ocean AOD accurately, we recommend CHEM1 for future studies. However, given the improvements in Nd in the CHEM2, CHEM3, CHEM2-SSF and CHEM3-SSF simulations relative to observations, the CHEM2 and CHEM3 DMS chemistry schemes allow for a more accurate representation of cloud microphysical properties over the Southern Ocean. Furthermore, the CHEM2 and CHEM3 schemes represent a fundamentally improved representation of DMS chemistry over the default scheme and improve process-based understanding of sulfate aerosol formation over the Southern Ocean.

A large source of uncertainty in our investigation into aqueous-phase chemistry lies with the constant cloud water pH in the model (assumed to be 5 everywhere). Changes in cloud water pH have substantial impacts on aerosol particle size distributions and CCN concentrations, particularly in parts of the Northern Hemisphere where reductions in anthropogenic SO2 emissions since the mid-1980s have increased cloud water pH . show that the effect of pH on particles larger than 50 nm in diameter (which may activate to CCN) over the Southern Ocean is not negligible. Aqueous-phase chemistry may also be affected in the model due to the lack of persistent low-lying cloud over the Southern Ocean . Aqueous-phase chemistry is more efficient at processing sulfur-containing gases than gas-phase chemistry, but cloud droplets are needed to allow in-cloud droplet chemistry to occur. Future work will focus on these issues and on evaluating changes to clouds and aerosols outside the Southern Ocean region when these changes are implemented.