CAM6 with the finite-volume (FV) dynamical core (Lin and Rood, 1997) is used in this study. CAM6 treats important physical processes in the atmosphere, including radiative transfer, deep convection, cloud macrophysics, cloud microphysics, shallow convection, and planetary boundary layer turbulence. Cloud and aerosol interactions with longwave and shortwave radiation transfer are treated by the Rapid Radiative Transfer Model for GCMs (RRTM-G) scheme (Iacono et al., 2008; Mlawer et al., 1997). A double-moment scheme (Gettelman and Morrison, 2015) is used to describe the microphysical processes of cloud and precipitation hydrometeors in large-scale stratiform clouds, while the deep convection is represented by the Zhang and McFarlane (1995) scheme. CAM6 uses the Cloud Layers Unified By Binormals (CLUBB) scheme (Golaz et al., 2002; Larson et al., 2002) to unify the representations of cloud macrophysics, turbulence, and shallow convection.

The four-mode version of the Modal Aerosol Module (MAM4), which is an extension of the three-mode version of MAM (Liu et al., 2012), is used to describe the aerosol properties and processes in CAM6 (Liu et al., 2016). MAM4 uses the modal method to represent the size distributions of four aerosol modes: Aitken, accumulation, coarse, and primary carbon. The original MAM4 encompasses six aerosol species: black carbon, dust, primary organic aerosol, sea salt, secondary organic aerosol, and sulfate (Table 1). The primary organic aerosol here refers to non-marine sources of organic matter, usually from terrestrial biomass burning, fossil fuel, and biofuel burning. Aerosol species are internally mixed within a mode and externally mixed between modes. Then the lognormal size distribution can be determined for each mode based on a prescribed geometric standard deviation (Table 1). Different aerosol species are characterized by a variety of properties such as hygroscopicity, density, and optical properties (Table 2).

MAM in CAM6 adopts the modal approach, where aerosol species are assumed to be internally mixed within a mode and externally mixed between modes. MOA is emitted into the fine aerosol modes with different assumptions of mixing state with inorganic sea salt: (1) MOA is emitted into the Aitken and accumulation modes together with sea salt in the case of internally mixed with sea salt, or (2) MOA is emitted into the Aitken and primary carbon mode separately from sea salt in the case of externally mixed with sea salt. In addition, there is another assumption of whether the experimentally derived parameterizations of SSA mass emission flux represent the total emission of MOA and sea salt or only account for the emission of sea salt. In the former case, MOA will replace the mass and number emission fluxes of sea salt. In the latter case, MOA will add onto the sea salt mass and number emission fluxes. Burrows et al. (2018) tested different combinations of the two assumptions and found that the “internally mixed” and “added” MOA approach provides the most physically realistic configuration compared to the observations. Thus, in our study we use this configuration but acknowledge that current observations do not provide precise constraints on the mixing state.

While anthropogenic aerosol and precursor gas emissions are prescribed for model simulations, emissions of natural aerosols (e.g., SSA, dust) are calculated interactively in the model. SSA in MAM is emitted following the parameterization of Mårtensson et al. (2003) for dry particle diameters from 0.020 to 2.8 µm, and Monahan et al. (1986) from 2.8 to 10 µm. The Mårtensson et al. (2003) parameterization is derived from laboratory experiments in which particles were produced by bubble bursting using a sintered glass filter in synthetic seawater. The emission rate depends linearly on the sea surface temperature and is proportional to 10 m wind speed, raised to the power of 3.41 (Monahan et al., 1986; Gong et al., 1997).

In this study, several modifications are implemented in CAM6 in order to explicitly quantify the influence of marine organic matter on aerosols, clouds, and radiation. These modifications are comprised of (1) emission schemes of MOA, as introduced in Sect. 2.2.1, and (2) ice nucleation parameterizations for MOA, as introduced in Sect. 2.2.2.

In this study, we carried out several numerical experiments to investigate the influence of MOA on aerosols and CCN and INP activities (Table 4). All simulations were performed for 10 years with prescribed climatological sea surface temperatures and sea ice. The first year of simulations was treated as model spin-up, and the last 9 years of simulations were used in analyses. The simulations were driven by the present-day (year 2000) aerosol and precursor gas emissions with given greenhouse gas concentrations. The model was run for 32 vertical levels from the surface up to 3 hPa with a horizontal resolution of 0.9∘ (latitudes) by 1.25∘ (longitude). We conducted two sets of experiments. The first set of experiments, as listed in Table 4, are used to test the model sensitivity to different MOA emission schemes. The baseline experiment (BASE) uses the default CAM6 model, which does not account for MOA emission and related physical processes. In addition to the BASE experiment, the B14 experiment addresses emission, advection, dry/wet deposition, and CCN effect of MOA using the Burrows et al. (2014) emission scheme. We also designed two additional experiments (G11 and NULL) to address the model sensitivity to emission methods. These simulations (B14 and G11) were conducted with the added and internally mixed MOA approaches, following Burrows et al. (2018). The INP effect of MOA is not considered in this set of experiments.

We conducted another set of experiments to investigate both CCN and INP effects of MOA, as listed in Table 4. The control experiment (CTL) is the same as BASE except that the D15 dust ice nucleation scheme was used to replace the CNT scheme in BASE because D15 gave a better model performance compared to observations in our previous study (Shi and Liu, 2019). The B14_D15, which is based on CTL, considers the MOA emission from B14 and the CCN effect of MOA. The B14_D15_M18 experiment, which is based on B14_D15, additionally considers the INP effect of MOA based on M18. The comparison between CTL and B14_D15 shows the CCN effect of MOA, while the comparison between B14_D15 and B14_D15_M18 shows its INP effect.

We further conducted three experiments to examine the model sensitivity to a different MOA ice nucleation parameterization (i.e., W15) in B14_D15_W15 and to two different dust ice nucleation parameterizations (i.e., N12 and CNT) in B14_N12_ M18 and B14_CNT_M18 by comparing them with the B14_D15_M18 experiment, respectively.

Given that a realistic representation of MOA emissions is a prerequisite for models to quantify its influence on ice nucleation, we evaluate three different MOA emission parameterizations in this section. We also analyze the processes contributing to MOA burden such as emission, transport, and removal because the burden pattern largely determines the INP distribution pattern. Comparisons with available observations are made to examine the performance of different MOA emission schemes.

Table 5 lists the annual global mean emissions and burdens of MOA and sea salt from different simulations. Overall, the G11 method generates the largest global MOA emission (27.1 Tg yr-1) followed by the B14 method (24.5 Tg yr-1). The magnitudes of MOA emissions are within the range of previous studies (Huang et al., 2018; Meskhidze et al., 2011; Langmann et al., 2008). The ratios of MOA emission to sea salt emission are 0.67 % and 0.74 % for the B14 and G11 experiments, respectively, which are also comparable to previous studies ranging from 0.3 % to 3.2 % (Huang et al., 2018; Meskhidze et al., 2011). The NULL approach only gives an annual global MOA emission of 4.6 Tg yr-1, with the ratio of MOA emission to sea salt emission of 0.13 %. These values are much lower than those of B14 and G11 approaches. We note that emissions and burdens of sea salt include the contribution from the coarse mode, which dominates the total sea salt emissions and burdens. We further evaluate aerosol mass mixing ratios and number concentrations in each aerosol mode in the B14 experiment, where MOA is added and internally mixed with sea salt. In B14, the ratio of MOA to sea salt mass burdens reaches up to 2.3 and 1.0 for the Aitken and accumulation modes, respectively. Number concentrations of accumulation mode aerosols near the surface are increased by up to 50 % over some regions of the Southern Ocean and Arctic.

Despite the fact that there are differences in the global annual mean value, B14 and G11 generate similar spatial patterns of MOA emission rates (Fig. 1), while G11 tends to give higher emission rates than B14. Large emission rates are located in the midlatitude storm tracks, equatorial upwelling, and coastal regions as shown in Fig. 1. These locations largely reflect the geographic distribution of primary ocean productivity as indicated by [Chl a] (in G11) or organic matter concentrations (in B14).

Here we illustrate the influence of surface wind speeds (Fig. S1 in the Supplement) on the emission of MOA. Although high MOA emissions are mostly co-located with vigorous oceanic biological activities, the oceanic area with smaller or larger wind speed tends to have a decreased or elevated emission rate relative to their biological activities. For instance, due to weak wind speeds (∼ 5 m s-1), a strong signal of oceanic organic matter concentration does not correspond to a large emission rate on the western coast of South America. On the contrary, because of strong wind speeds (∼ 10 m s-1), moderate emission rates are noticed over the subtropical northern Pacific Ocean and subtropical southern Indian Ocean despite relatively small [Chl a] or organic matter concentrations. This wind-speed-dependent pattern is more clearly shown in the B14 results than in the G11 results because in the B14 emission scheme FMOA/SSA is not related to the wind speed while SSA emission is proportional to the surface wind speed, as described in Sect. 2.2.1. Conversely, FMOA/SSA is inversely related to the wind speed in G11, which results in a more complicated relationship between wind speed and MOA emission rate in G11.

The global mean MOA burden is 0.097 Tg in B14, which is in close agreement with previous studies that suggested a range of 0.031 to 0.131 Tg (Huang et al., 2018; Burrows et al., 2018). The global distribution of MOA column burden shares a similar pattern between G11 and B14, with the peak burden around 1 mg m-2 over the midlatitude to high-latitude Southern Ocean (Fig. 1). Despite the fact that large burdens are usually related to locations of high emissions, they are also influenced by advection (dependent on 3-D wind), dry deposition (dependent on particle size), and wet deposition (dependent on precipitation). The oceanic regions with small annual precipitation rates (Fig. S1) lead to considerable accumulations of MOA in G11 and B14. For instance, the peak burdens with maximum values of 0.4 to 0.6 mg m-2 on either side of the Pacific tropical convection zone correspond to the subsidence induced dry zone (i.e., subsiding branch of Walker and Hadley circulations).

Zonally averaged vertical distributions of MOA mass mixing ratio illustrate the vertical transport of MOA (Fig. 1). Simulations from G11 and B14 exhibit a maximum value of 0.35 µg kg-1 within the boundary layer, located in 40–50∘ S of the Southern Ocean, while the maximum value is only 0.05 µg kg-1 in NULL. Globally, G11 shows slightly higher MOA mass mixing ratios over all latitudes compared with B14 and transports more MOA to high altitudes over the tropical regions. It is clear that MOA is accumulated in the lower troposphere, i.e. below 600 hPa in G11 and B14 and below 800 hPa in NULL. The reason is that MOA is generated over the oceans, especially over the storm track regions with high precipitation, limiting MOA mainly to the lower troposphere.

We further evaluate model simulated MOA concentrations with measurements at Mace Head (Ireland) and Amsterdam Island (Fig. 2). The B14 and G11 methods do well in capturing the observed seasonal variation of MOA concentrations at Amsterdam Island (Fig. 2a), although the model produces slightly higher MOA concentrations. At Mace Head, the two methods produce delayed concentration peaks by about 1 month compared with observations (Fig. 2b). The mass fraction of MOA in SSA (Fig. 2c) shows a better agreement between the model and observation. Both the simulated and observed organic mass fraction increase from March and reaches a peak in July, although the observed peak is broader. The sea ice extent prescribed in the model as a boundary condition has a strong seasonal variation over the Southern Ocean, as shown in Fig. S2. This can greatly impact the emission of MOA there (e.g., low emissions during the austral winter and early spring). The NULL approach does not reproduce observed seasonal cycles of MOA and significantly underestimates observed MOA concentrations due to the prescribed mass fraction (0.05) in the accumulation mode.

Based on our analyses and comparisons with observations, we show that B14 implementation of MOA emission into CAM6 reasonably captures the concentrations and seasonal variations of MOA. Next we will study the MOA effects on clouds via acting as CCN (Sect. 3.2) and INPs (Sect. 3.3), based on model experiments with the B14 emission (Table 4).

After introducing MOA in the model, we notice an obvious increase in oceanic surface CCN concentrations at high latitudes. Figure 3 shows the spatial distribution of annual mean percentage changes in surface CCN concentrations at a supersaturation of 0.1 % due to MOA derived from the two experiments (CTL and B14_D15). From Fig. 3, the annual mean CCN concentration increases by 15 %–35 % over much of the ocean from 30 to 70∘ S, with a maximum increase of 45 % located over the Southern Ocean (60∘ S, 55∘ E). Other regions showing significant increases of CCN are over the pristine high latitudes, with increases of 25 %–35 % from 60∘ S to Antarctica in the SH and from 60 to 80∘ N in the NH. These results are comparable with previous results, with an average increase of 12 % and up to 20 % of CCN over the Southern Ocean (Meskhidze et al., 2011). Over low-latitude and midlatitude oceans, CCN changes due to MOA are smaller. Generally, the distribution of CCN change is consistent with the MOA emission pattern. The vertical profiles of CCN concentrations from the two model experiments and observations during the eight field campaigns are shown in Fig. 3. Clear increases of CCN concentrations in the boundary layer due to MOA are evident for campaigns over the ocean or coastal regions (SOCEX1, SOCEX2, ACE1, FIRE1, and ASTEX), with the maximum increase (26 %) in ACE1. Observed CCN from FIRE1 shows a strong inversion of CCN below 800 hPa, and this inversion is challenging for the model due to its coarse vertical resolution. An obvious underestimation of CCN in the model is noticed at FIRE3 over the Arctic Ocean in spring, which is attributed to the underestimated transport of air pollution caused by too strong wet scavenging in the model (Liu et al., 2012).

In order to examine the importance of MOA INPs, we compare modeled INPs from MOA versus dust and compare them with observations from several field campaigns at high latitudes (Fig. 4). Modeled INP concentrations from MOA are calculated online using M18 and W15 parameterizations (from B14_D15_M18 and B14_D15_W15 experiments, respectively), while dust INP concentrations are calculated online using D15, CNT, and N12 parameterizations (from B14_D15_M18, B14_CNT_M18, and B14_N12_M18 experiments, respectively). Modeled INP concentrations are computed based on aerosol concentrations at different temperatures and are selected at the same altitudes and locations as the observations. The measured INP data were obtained from Mace Head, the CAPRICORN campaign (Clouds, Aerosols, Precipitation, Radiation, and Atmospheric Composition over the Southern Ocean), Oliktok Point, and Zeppelin (McCluskey et al., 2018a, b; Creamean et al., 2018; Tobo et al., 2019).

As illustrated in Fig. 4, the M18 parameterization tends to underestimate observed INP concentrations except at temperatures colder than -25∘. On the other hand, the W15 parameterization overestimates observed INP concentrations except at temperatures warmer than -20∘. Under the same MOA scenario, the W15 parameterization is more efficient at producing INPs than M18. This is because the M18 parameterization was derived from MOA in the atmosphere, which accounts for the effect of physiochemical selective emission and aerosol chemistry in the air. In contrast, the W15 parameterization was derived based on the total organic carbon in sea surface microlayer samples, which contain higher organic mass concentrations compared with ambient MOA.

The dust INP concentration calculated with CNT shows an underestimation when temperature is warmer than -20∘ and an overestimation when temperature is between -30 and -20∘. This is consistent with previous work by Wang et al. (2014). The D15 parameterization indicates a clear underestimation. The N12 scheme has the better performance than D15 in Fig. 4. However, the field campaigns used in Fig. 4 are marine aerosol dominant or contained scenario campaigns. MOA is identified as an important INP source during these campaigns from measurements (McCluskey et al., 2018b, a). Thus, dust should not be expected to be the dominant INP, as indicated by the N12 scheme, which only considers dust INPs. This suggests that N12 may overestimate dust INPs, which is consistent with our earlier study (Shi and Liu, 2019). These results suggest that the N12 parameterization is more efficient in producing dust INPs than the D15 parameterization under the same dust loading. INP concentrations from N12 are calculated based on the coarse, accumulation, and Aitken mode dust aerosol, which account for fine dust particles, while INP concentrations from D15 are calculated based on the number concentration of dust particles with diameters larger than 0.5 µm (DeMott et al., 2015). Simulated total INPs, the sum of dust and MOA INPs from D15 and M18, gives a better agreement with observations than D15 and M18 alone, although underestimations still exist at warmer temperatures.

Figure 5 shows the comparison between simulated and measured INPs from five parameterization schemes as a function of temperature for the same field campaigns as in Fig. 4. Generally, an inverse linear relationship is revealed between log10(INPs) and temperature in the measurements. This relationship is also shown in simulated INP number concentrations from the empirical parameterizations (N12, D15, W15, M18). However, for CNT nearly constant INP number concentrations are presented at temperatures from -35 to -20∘, and then a rapid decrease is seen with increasing temperature when the temperature is warmer than -20∘. At temperatures higher than -15∘, nearly no INPs are produced by CNT, leading to the underestimation of INPs in the CNT method at these temperatures.

We notice higher INP number concentrations are produced from M18 compared with W15 at Zeppelin during March 2017. The most distinctive feature of this campaign is its very low aerosol loadings. For example, simulated SSA mass mixing ratio is around 0.6 µg kg-1 with the maximum value at 1.8 µg kg-1 below 850 hPa, and the dust mass mixing ratio is around 0.3 µg kg-1. We note that simulated dust INP number concentrations from N12 are always higher than those from D15 and that both N12 and D15 are more efficient in producing INPs than CNT when temperature is warmer than -20∘.

The global distribution pattern of annual mean MOA INP concentrations at 950 hPa at a temperature of -25∘ is similar to that of MOA column burden concentrations, as shown in Fig. 6a. The MOA INPs are spread over the oceans, with peaks (∼ 0.1 L-1) over 40 to 60∘ S in the Southern Ocean, the subtropical southern Indian Ocean, the subtropical Atlantic Ocean, and the subtropical eastern Pacific Ocean. Meanwhile, dust INP concentrations diagnosed at the same pressure and at the same temperature (Fig. 6b) are dominant over the NH and downwind of dust source regions in the SH (e.g., around Australia and extended to 50∘ S).

Figure 6c shows the horizontal distribution of the ratio of MOA INP concentration to dust INP concentration at 950 hPa. It is clear that MOA INPs are more important than dust INPs at 40∘ S south in the SH, where MOA INP concentrations can be up to 1000 times higher than those of dust INPs. The zonal mean vertical distribution of the ratio of MOA INP concentration to dust INP concentration is illustrated in Fig. 6d. The ratio peaks near 65∘ S, indicating that MOA INPs are more important than dust INPs over the Southern Ocean from the surface up to 400 hPa and that it extends poleward to 90∘ S. Above the 400 hPa altitude, dust particles are still more important INPs. Because dust particles are emitted over drier deserts (i.e., with lower precipitation and thus less wet scavenging), dust can be subject to long-range transport at high elevations. In contrast, most MOA particles are generated over the storm track regions with high occurrences of precipitation. Taking emission, transport, and wet scavenging of MOA and dust particles into account results in MOA INPs dominating below 400 hPa over the Southern Ocean, whereas dust INPs are generally more important elsewhere.

Immersion freezing on MOA in mixed-phase clouds requires that there are cloud droplets at temperatures colder than -4∘. Ice nucleation consumes cloud liquid water and thus will compete with other processes for cloud liquid water (e.g., autoconversion of cloud water to rain, accretion of cloud water by rain and snow). This competition is expected to result in a reduction in the ice nucleation rate of MOA compared with the offline calculation of ice nucleation rate, as in McCluskey et al. (2019). Figure 7 shows the annual zonal mean ice production rates from the immersion freezing of MOA and dust, which are calculated online for the cloud ice production tendency in the B14_D15_M18 experiment. Over the NH, the immersion freezing of dust dominates the primary ice production, giving an averaged ice production rate of 5 kg-1 s-1, which can increase to up to 20 kg-1 s-1 over 40∘ N at 400 hPa (Fig. 7b), while the MOA ice production rate is around 1 kg-1 s-1 (Fig. 7a). However, in the Arctic boundary layer, the MOA fraction of total ice production rate is around 0.6–0.7 (Fig. 7c), indicating that MOA INPs are more important for generating ice crystals than dust INPs in that region. Over the SH, the immersion freezing rate of MOA dominates the primary ice production below 400 hPa with the MOA fraction close to 1. The zonal average ice nucleation rate of MOA is around 1 kg-1 s-1, and increases to up to 5 kg-1 s-1 at 65∘ S in the Southern Ocean at 400–600 hPa. The immersion freezing rate of dust is around 1 kg-1 s-1 above 500 hPa, and smaller than 0.1 kg-1 s-1 below the 600 hPa altitude in the SH. Analysis of the seasonal variation of ice nucleation rate of MOA indicates that a maximum rate of about 16 kg-1 s-1 occurs at 400–600 hPa over 60∘ S in July (austral winter). In summary, the annual mean immersion freezing of MOA dominates the primary ice production over the SH below 400 hPa altitude and in the Arctic boundary layer.

Table 6 displays the differences of cloud and precipitation variables between the CTL and B14_D15_M18 experiments. With added MOA aerosol, the global annual mean surface concentration of CCN at 0.1 % supersaturation changes from 103.3 cm-3 in CTL to 106.6 cm-3 in B14_D15_M18. This increase of 3.28 cm-3 is comparable to other model estimates of 3.66 cm-3 (Burrows et al., 2018) and 2.6–3.0 cm-3 (Meskhidze et al., 2011). The vertically integrated cloud droplet number concentration (CDNUMC) increases by 7.5 × 104 cm-2 (5.25 % in percent change) on the global annual mean and by 1.1 × 104 cm-2 (0.94 %) and 3.2 × 105 cm-2 (16.89 %) over 20–90∘ S during the austral winter (June–July–August) and summer (December–January–February), respectively, by comparing B14_D15_M18 with CTL. This reflects a strong seasonal variation of MOA emissions due to changes in the sea ice extent and biological activity. The global annual mean liquid water path (LWP), ice water path (IWP), longwave cloud forcing (LWCF), and total cloud fraction (CLDTOT) do not show obvious changes between CTL and B14_D15_M18. The global annual mean shortwave cloud forcing is stronger by -0.41 W m-2 due to MOA. During the austral summer over 20–90∘ S, we notice an increase of 4.57 g m-2 (5.10 %) in LWP and a 1.35 % (2.52 %) increase in low-cloud fraction. As a consequence, shortwave cloud forcing (SWCF) is enhanced by -2.87 W m-2 (Table 6), which is comparable to -3.5 W m-2 estimated in Burrows et al. (2018). Ice number concentration on -15 ∘C isotherm level increases by 9.34 % during the austral winter. There does not appear to be a significant change in LWCF, which is consistent with the result of Huang et al. (2018).

A strong CCN effect of MOA on clouds (in terms of significant changes in CCN and CDNUMC) tends to occur only in the SH over 40–60∘ S, while a strong INP effect (in terms of significant changes in cloud ice mass and number concentrations) is notable over 50–70∘ in both hemispheres (Fig. 8). Over 40–60∘ S, a significant increase from 70 to 90 cm-3 in the annual zonal mean surface CCN concentration is observed. The CCN concentration there is nearly 30 % higher in B14_D15 and B14_D15_M18 than in CTL. As a result, CDNUMC increases from 2.6 × 1010 m-2 in CTL to 3.0 × 1010 m-2 in B14_D15 and B14_D15_M18 over 40–60∘ S, leading to an increase in LWP due to the aerosol indirect effect (Fig. 8). Furthermore, we notice a stronger SWCF at 40–60∘ S by 3 W m-2 in B14_D15 compared with CTL. After considering the INP effect of MOA in the model, we notice that cloud ice number concentration and cloud ice mass mixing ratio increase in mixed-phase clouds, which led to a slight decrease in CDNUMC. As indicated in Fig. 8b and d, cloud ice number concentration increases from 4500 kg-1 in B14_D15 to 5500 kg-1 in B14_D15_M18 at ∼ 60∘ S, with cloud ice mass mixing ratio increased by 0.25 mg kg-1. Over 60∘N, cloud ice number concentration increases from 4200 kg-1 in B14_D15 to 5200 kg-1 in B14_D15_M18, with cloud ice mass mixing ratio increased by 0.1 mg kg-1.

Figure 9 shows the seasonal variations of cloud properties and cloud radiative forcing averaged over the 20–90∘ S in the SH in response to the introduction of MOA as CCN and INPs. The seasonal variation of surface CCN concentration at 0.1 % supersaturation shows a maximum value of 72 cm-3 in the austral summer and a minimum value of ∼ 50 cm-3 in the austral winter in CTL. Similar seasonal variation patterns are also noted for CDNUMC and LWP. With the inclusion of MOA in the model, B14_D15 and B14_D15_M18 produce more surface CCN, with an increase of up to 14 cm-3 (∼ 20 %) in January compared with CTL. Accordingly, CDNUMC increases from 2.1 × 1010 m-2 in CTL to 2.5 × 1010 m-2 in B14_D15 in January, and LWP increases from 93 g m-2 in CTL to 97 g m-2 in B14_D15 in January. As a consequence, SWCF is stronger by -3.5 W m-2 in B14_D15 compared to CTL during the austral summer. We also notice that CCN, CDNUMC, and SWCF show smaller changes during the austral winter due to weaker oceanic biological activity and larger sea ice extent.

Different from the warm cloud features above, seasonal variations of ice properties in mixed-phase clouds (i.e., cloud ice mass mixing ratio and number concentration on -15 ∘C isotherm level, IWP) clearly show winter maxima. After introducing the INP effect of MOA in the model, ice number concentration on -15 ∘C isotherm level increases by comparing B14_D15 with B14_D15_M18, with obvious increases of up to 27 % in June. Ice mass mixing ratio on -15 ∘C isotherm level increases by 0.19 mg kg-1 (13 %) in June. Increases in both cloud ice number and mass contribute to the increase of IWP by 0.5 g m-2 in austral winter. The seasonal change of LWCF is not well correlated with changes in ice number concentration and mass mixing ratio in mixed-phase clouds because LWCF is controlled more by high clouds. Our introduction of MOA INPs mainly occurs in mixed-phase clouds and thus has only a small influence on LWCF.

As shown in Table 7, the CCN effect of MOA on SWCF is strongest in the austral summer, with the value of -2.78 W m-2 over 20–90∘ S in the SH. In contrast, the INP effect of MOA on LWCF is strongest in the austral winter, with a value of 0.35 W m-2 (Table 8). For the net cloud forcing (SWCF + LWCF), the CCN effect of MOA is 2.65 W m-2 in austral summer, and the INP effect is 0.65 W m-2 in austral spring over the 20–90∘ S. The annual global mean CCN and INP effects of MOA on the net cloud forcing are -0.35 and 0.016 W m-2, respectively. From an annual mean perspective, the CCN effect of MOA on SWCF is -0.84 W m-2 over 20–90∘ S and is about twice as much as the global mean value (-0.41 W m-2), which indicates that the global annual mean SWCF change due to MOA is dominated by SH contributions.