The sensitivity of Southern Ocean aerosols and cloud microphysics to sea spray and sulfate aerosol production in the HadGEM3-GA7.1 chemistry-climate model

With low concentrations of tropospheric aerosol, the Southern Ocean offers a ‘natural laboratory’ for studies of aerosol-cloud interactions. Aerosols over the Southern Ocean are produced from biogenic activity in the ocean, which generates sulfate aerosol via dimethylsulfide (DMS) oxidation, and from strong winds and waves that lead to bubble bursting and seaspray emission. Here we evaluate the representation of Southern Ocean aerosols in the HadGEM3-GA7.1 chemistry-climate model. Compared with aerosol optical depth (AOD) observations from two satellite instruments (the Moderate Resolution 5 Imaging Spectroradiometer, MODIS-Aqua c6.1 and the Multi-angle Imaging Spectroradiometer, MISR), the model simulates too-high AOD during winter and too-low AOD during summer. By switching off DMS emission in the model, we show that sea spray aerosol is the dominant contributor to AOD during winter. In turn, the simulated sea spray aerosol flux depends on nearsurface wind speed. By examining MODIS AOD as a function of wind speed from the ERA-Interim reanalysis and comparing it with the model, we show that the sea spray aerosol source function in HadGEM3-GA7.1 overestimates the wind speed 10 dependency. We test a recently-developed sea spray aerosol source function derived from measurements made on a Southern Ocean research voyage in 2018. In this source function the wind speed dependency of the sea spray aerosol flux is less than in the formulation currently implemented in HadGEM3-GA7.1. The new source function leads to good agreement between simulated and observed wintertime AOD over the Southern Ocean, however reveals partially compensating errors in DMSderived AOD. While previous work has tested assumptions regarding the seawater climatology or sea-air flux of DMS, we 15 test the sensitivity of simulated AOD, cloud condensation nuclei and cloud droplet number concentration to three atmospheric sulfate chemistry schemes. The first scheme adds DMS oxidation by halogens and the other two test a recently-developed sulfate chemistry scheme for the marine troposphere; one tests gas-phase chemistry only while the second adds extra aqueousphase sulfate reactions. We show how simulated sulfur dioxide and sulfuric acid profiles over the Southern Ocean change as a result, and how the number concentration and particle size of the soluble Aitken, accumulation and coarse aerosol modes 20 are affected. The new DMS chemistry scheme leads to a 20% increase in the number concentration of cloud condensation 1 https://doi.org/10.5194/acp-2019-629 Preprint. Discussion started: 15 August 2019 c © Author(s) 2019. CC BY 4.0 License.


Introduction
Clouds and aerosols play an important role in Earth's energy balance by absorbing and scattering solar and terrestrial radiation.
However, aerosol-radiation and aerosol-cloud interactions are leading sources of uncertainty in determining human influences 5 on climate (Myhre and Shindell, 2013). The Southern Ocean, one of the cloudiest regions on Earth, is remote from anthropogenic sources of aerosol thus making it an ideal environment in which to study aerosol-cloud interactions (Hamilton et al., 2014). Clouds forming in pristine regions such as over the Southern Ocean are highly sensitive to aerosol perturbations (Koren et al., 2014), however the specific roles that marine aerosols play in cloud formation are highly uncertain (Brooks and Thornton, 2018). 10 Marine aerosols are either primary or secondary in origin. Primary aerosols such as sea spray are directly injected into the atmosphere when breaking waves entrain air bubbles, which subsequently form whitecaps and burst. Secondary aerosols such as sulfate aerosol are formed from nucleation of sulfur-containing gases. Sea spray aerosol (SSA) is generated in significant quantities over the Southern Ocean by strong winds and waves (Murphy et al., 1998). SSA is an important contributor to the global-mean clear-sky AOD (Shindell et al., 2013), and its production is highly dependent on wind speed (Smirnov et al., 2003;15 Mulcahy et al., 2008;Glantz et al., 2009). A significant component of primary marine aerosol is sea salt with some fraction of organics (Fossum et al., 2018). Marine organic aerosols, along with sulfate aerosols, result from biogenic activity in the ocean (O'Dowd et al., 2004). Marine phytoplankton produce dimethylsulfoniopropionate (DMSP), which is broken down into several products including dimethylsulfide (DMS). Oceanic DMS emissions are the main source of atmospheric sulfur over the Southern Ocean, with an estimated 28.1 TgS transferred from the oceans globally into the atmosphere each year (Lana et al., 20 2011). Around coastal Antarctica, melting of sea ice elevates the seawater DMS concentration (Trevena and Jones, 2006), leading to a seasonal anti-correlation between sea ice extent and aerosol concentration (Gabric et al., 2018). When DMS is emitted into the atmosphere, it has a lifetime of 1-2 days and undergoes a series of chemical reactions to form sulfur dioxide (SO 2 ) which is further oxidised to form sulfate aerosol.
Aerosol particles emitted into the atmosphere can grow in size via condensation and coagulation. Depending on the aerosol 25 composition and meteorological conditions such as the cloud base updraft velocity (Rosenfeld et al., 2014), particles larger than 50 nm in diameter can be "activated" to cloud condensation nuclei (CCN) around which water vapour can condense and cloud droplets form. Generally speaking, liquid water clouds which have been perturbed by aerosols consist of more but smaller cloud droplets, and therefore scatter radiation more efficiently (Twomey, 1977;Boucher and Randall, 2013).
Previous work has confirmed that cloud droplet number concentrations (N d ) over the Southern Ocean are correlated with 30 marine biogenic activity (Thomas et al., 2010;Woodhouse et al., 2010). Meskhidze and Nenes (2006) identified that observed N d over a large phytoplankton bloom was twice as large compared to a region distant from the bloom. More recently, McCoy et al. (2015) found that N d is spatially correlated with regions of high chlorophyll-a, and that the spatiotemporal variability in N d is found to be driven mostly by high concentrations of sulfate aerosol at lower southern latitudes and by organic matter in sea-spray aerosol at higher latitudes.
The models participating in the fifth phase of the Coupled Model Intercomparison Project (CMIP5) simulated Southern Ocean sea surface temperature (SST) biases which are primarily linked to cloud-related errors in shortwave radiation (Hyder et al., 2018). SST biases affect the position of the storm track (Ceppi et al., 2014), which leads to cascading errors in global 5 climate models across the Southern Hemisphere and reduces confidence in projections of climate change and climate extremes in this region (Trenberth and Fasullo, 2010).
To understand potential connections between the representation of aerosols and clouds via the aerosol indirect effect, we investigate the representation of marine aerosols over the Southern Ocean in the Hadley Centre Global Environmental Model version 3, Global Atmosphere 7.1 (HadGEM3-GA7.1). An evaluation of cloud representation in the predecessor model HadGEM3-GA7.0 suggests that significant errors exist in the cloud scheme over the Southern Ocean, but they partially compensate one another (Schuddeboom et al., 2019). Furthermore, the aerosol forcing and climate feedback in this model is highly sensitive to the representation of DMS-derived sulfate aerosol (Bodas-Salcedo et al., 2019).
HadGEM3-GA7.1 is described in Section 2.1, and simulated AOD is evaluated relative to observations in Section 3.1. We then show how biases in simulated AOD during winter months can be addressed by implementing a new SSA source function 15 derived from measurements collected on the Southern Ocean (Section 3.2). Finally, while much prior work has focussed on testing the sensitivity of Southern Ocean clouds and aerosols to the choice of DMS seawater climatology and/or the DMS sea-air transfer function (Mahajan et al., 2015;Boucher et al., 2003;Fiddes et al., 2018;Korhonen et al., 2008;Woodhouse et al., 2010), we have investigated atmospheric DMS chemistry. We performed sensitivity tests in which different gas-phase and aqueous-phase sulfate chemistry schemes have been implemented. The resulting changes in simulated aerosols and cloud 20 microphysics are shown in Section 3.3.

Model description
Simulations were performed with the Hadley Centre Global Environmental Model version 3, Global Atmosphere 7.1 (HadGEM3-GA7.1) (Walters et al., 2019;Mulcahy et al., 2018), which exhibits more realistic aerosol effective radiative forcing compared 25 with preceding versions (Mulcahy et al., 2018). 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 30 (Woodward, 2001).
Aerosol-cloud interactions are represented via the UKCA-Activate scheme (West et al., 2014), 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 (Abdul-Razzak and Ghan, 2000). 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 (N d ) is calculated from the number of activated aerosols at the 5 cloud base, weighted by this probability density function (Mulcahy et al., 2018). The number of cloud droplets subsequently influences the cloud albedo, as clouds with larger N d (and smaller droplets) are optically brighter (Twomey, 1977).
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 N d compared with observations (Mulcahy et al., 2018). 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 10 configuration uses a simplified chemistry scheme, whereby chemical oxidants such as O 3 , OH, NO 3 and HO 2 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 Morgenstern et al. (2009) andO'Connor et al. (2014), respectively. 15 The StratTrop scheme uses a Newton-Raphson solver, and accounts for DMS oxidation via the gas-phase and aqueous-phase reactions shown in Table 1. The oxidation of DMS by OH proceeds by both an addition and abstraction pathway (the first two reactions listed in Table 1), and can produce SO 2 and methane sulfonic acid (MSA). The relative yields of these products are important as SO 2 leads to new particle formation, while other products such as MSA condense on existing particles therefore increasing their size (von Glasow and Crutzen, 2004;Hoffmann et al., 2016). 20 Gas-phase SO 2 enters the liquid phase via an equilibrium approach (Warneck, 2000) described by Henry's Law. Because SO 2 dissociates in the aqueous-phase (Reactions R1 and R2), it is more soluble than the equilibrium Henry's law constant (K H ) implies.
25 Therefore, the model uses an effective constant (K H ef f ) which for SO 2 is related to K H by Eq. (1). SSA is generated via a wind speed-dependent parametrisation based on whitecap coverage (Gong, 2003). This function is based on the semi-empirical function by Monahan et al. (1986), but improves the representation of small particles less than 0.1 µm in diameter. According to Gong (2003), the number of seawater droplets generated per square-meter of sea surface, per increment of particle radius over 20 size bins is calculated via Eq. (2): The exponential terms A and B are defined by Eq. (3) and (4): Where r is the particle radius at a relative humidity of 80%, Θ is an adjustable parameter that controls the shape of the size 10 distributions and u 10 is the scalar horizontal wind-speed at 10 m above the surface. 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 2. All simulations used the DMS seawater climatology of Lana et al. (2011) and the DMS sea-air exchange parametrisation of Liss and Merlivat (1986). 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.

Simulations performed
Analysis of aerosol measurements made on a 2018 Tangaroa research voyage on the Southern Ocean indicate that the 20 dependency of SSA production on near-surface wind speed (u 3.41 10 ) is overestimated by a factor of 2-4 via the Gong (2003) source function (Eq. (2)). Ongoing research by Hartery et al. indicates that Eq. (5) with SSA production dependent on u 2.8 10 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. (5), in combination with different sulfate chemistry 25 schemes as described below.
dF dr = 2.6u 2.8 10 r −A (1 + 0.057r 3.45 ) × 10 1.607e −B 2 DMS oxidation chemistry is complex (von Glasow and Crutzen, 2004), however the set of reactions describing the conversion of gaseous DMS into sulfate aerosol in StratTrop (Table 1)  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 3. 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 5 DMS oxidation by halogens as they have been shown to play an important role in the remote marine atmosphere (Boucher et al., 2003;von Glasow and Crutzen, 2004;Chen et al., 2018

Satellite-based observations
Model output is compared to daily-mean aerosol optical depth (AOD) data derived from Moderate Resolution Imaging Spectroradiometer (MODIS)-Aqua measurements, collection 6.1 (Platnick et al., 2003;Sayer et al., 2014) and monthly-mean AOD derived from the Multi-angle Imaging Spectroradiometer (MISR). MODIS is a passive imaging radiometer that measures reflected solar and emitted thermal radiation across a 2330 km swath, providing near-daily global coverage over land and ocean 20 at the Equator and overlap between orbits at higher latitudes. MODIS was deployed on the Aqua satellite in May 2002. Here the MODIS Level 3 data product with a spatial resolution of 1 • × 1 • (latitude/longitude grid) is used for AOD at 550 nm. A number of inconsistencies and potential retrieval problems, which have been identified in past MODIS products, have been remedied in MODIS collection 6.1. The data used in this study were obtained using the combined Deep Blue (land retrieval only) and Dark Target (ocean and land retrieval) approaches (Sayer et al., 2014). In this study we use MODIS measurements 25 from 2003 to 2007; a period characterised by a notable absence of volcanic eruptions reaching the lower stratosphere as discussed below. Since MODIS data are limited at high latitudes in the visible band, we spatially and temporally co-locate MODIS and model data before calculating climatological monthly means (Schutgens et al., 2016).
A previous study by Remer et al. (2008) showed that over oceans, MODIS-retrieved AOD agrees well (within the expected uncertainties) with observations obtained from the ground-based Aerosol Robotic Network of sun photometers (AERONET, the equator, coverage time varies from 2 to 9 days, respectively. The MISR AOD product has been validated with respect to AERONET (Kahn et al., 2010;Garay et al., 2017), and also shows good agreement with AOD Level 3 data from MODIS (Mehta et al., 2016).
Here the MISR Level 3 data product with a spatial resolution of 0.5 • × 0.5 • is used for total AOD at 555 nm, with the AOD retrieval algorithm dependent on surface types such as vegetated areas, dark water bodies and high contrast terrain  observed by MODIS (Fig. 1a,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. 1c,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. 2d), it is apparent that the model simulates primarily SSA during winter (July and August, 50-65 • S).

10
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 15 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 (Li et al., 2018).
Total AOD is calculated in HadGEM3-GA7.1 from adding together the individual contributions of dust AOD and the Aitken mode (soluble + insoluble), accumulation mode and coarse mode AODs. Aerosol particles in the soluble modes may activate to 20 cloud condensation nuclei, and the contribution to total AOD from these modes is shown in Fig. 2a-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. 2c), implying that it is mostly SSA as discussed above. Accumulation mode aerosol (Fig. 2b)

Simulated sea salt aerosol
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 40-60 • S are shown in Fig. 3a. 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 Section 2.1, the model uses the SSA source function of Gong  Table 2) is shown in Fig. 3b. Compared with the REF simulation (Fig. 3c) the reduction in AOD is reasonably uniform throughout the year, with the reduction in coarse mode AOD (shown for REF in Fig. 2c) between March and November particularly 15 visible. Comparing to MODIS observations, the Hartery et al. source function performs well during winter months when SSA is the dominant contributor to AOD (Fig. 3d). Changes in aerosol mode number concentrations and dry diameters in the SSF simulation are discussed later in Section 3.3.
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 20 between 50-60 • S compared to observations from AERONET sun photometers (Shindell et al., 2013). Our results are also consistent with previous work suggesting that the Gong (2003) 2011) seawater DMS climatology, which shows a region of high DMS productivity close by (Fig. 5b).
In all three CHEM simulations, the change in the simulated surface atmospheric DMS concentration is negligible relative 25 to the magnitude of the seasonal cycle in DMS (Fig. 6b,c).  Tables 1 and 3, 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.
Surface SO 2 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 SO 2 (rather than only SO 2 ; see Table 3). In particular, the DMS+BrO reaction has been shown to be particularly important in the remote marine troposphere (Chen et al., 2018;Boucher et al., 2003;von Glasow and Crutzen, 2004 (Fig. 7c) due to the extra DMS oxidation reactions added.  Fig. 9 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. 8b and c). The particle dry diameters in these modes are largely unchanged ( Fig. 8e and f). However, the soluble Aitken mode changes; the number concentration increases by ∼40% in the SSF simulation 15 relative to REF and the average particle dry diameter decreases by 10 nm (Fig. 8a 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. 8c,f). 20 In CHEM2 and CHEM3 simulations there are more smaller particles in the accumulation mode which are smaller on average ( Fig. 8b,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 At southern high latitudes, the number fraction of SSA CCN is larger than any other region on the globe (Quinn et al., 2017).
Therefore owing to the reduced aerosol abundance in the SSF simulation, CCN concentrations also decrease by up to ∼13% relative to REF (Fig. 10a). In the CHEM simulations, CCN concentrations decrease by -18% (CHEM1) to +25% (CHEM2 30 and CHEM3), which is likely linked to the changes in accumulation mode aerosol shown in Fig. 8. The changes in CCN in the CHEM simulations translate to changes in N d over the Southern Ocean of -13% in CHEM1 to +20% in CHEM2 and CHEM3 (Fig. 10b). Bodas-Salcedo et al. (2019) show that in HadGEM3-GA7.1, the simulated seasonal cycle in N d over the Southern Ocean is primarily driven by seawater DMS emissions. While the model captures the observed seasonality in N d , the magnitude is too low, which was also reported by Mulcahy et al. (2018). However, the CHEM2 and CHEM3 simulations bring the model into better agreement with N d observations.
Of all the CHEM and CHEM-SSF sensitivity simulations, AOD simulated in the CHEM1 simulation agrees most favorably with MODIS (Fig. 11a), and the root-mean square error of 0.029 is the same as it is when comparing REF and MODIS (Fig. 1c).
However, the seasonal bias remains. The CHEM1-SSF simulation shows good agreement with MODIS during austral winter 5 but underestimates summertime AOD and N d (Fig. 10 and 11d). 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.
However, given that the chemistry schemes used in the CHEM2 and CHEM3 simulations also show the best agreement with . 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. Ocean. Simulated wintertime AOD agrees favourably with observations as a result, but points to partially compensating errors in the formulation of sulfate aerosol production, which maximises over the Southern Ocean in summer as a result of marine biogenic activity. We performed simulations to test the sensitivity of Southern Ocean clouds and aerosols to alternative gasphase and aqueous-phase chemistry schemes associated with sulfate aerosol. The schemes tested here lead to changes in 30 simulated DMS, SO 2 , H 2 SO 4 and aerosol particle sizes and number concentrations. In particular, the CHEM2 and CHEM3 schemes tested lead to increases in CCN and N d of up to 20%, leading to better agreement between simulated and observed N d .