We use the UCLALES–SALSA model (Tonttila et al., 2017) that combines the large eddy simulator UCLALES (Stevens et al., 1999, 2005; Stevens and Seifert, 2008) with a detailed sectional model of aerosol microphysics SALSA (Kokkola et al., 2008, 2018) extended with cloud and precipitation processes. SALSA simulates microphysical processes such as coagulation, sedimentation, partitioning of water between vapour and liquid phases, activation of aerosol particles to cloud droplets, and growth of droplets to form precipitation.

Aerosol is described by two externally mixed modes, each consisting of a set of size bins with unique chemical composition. Tracking the two externally mixed modes allows us to separately follow the fate of the background aerosol and sea spray emitted during the simulation. The aerosol particle bins are based on the dry diameter of the particles. In the current study, we used 15 bins ranging from 3 nm to 10 µm (sizes here and further on are given as dry diameter unless stated differently). This description is extended to also cover the cloud droplets so that particles that activate as cloud droplets are moved to a parallel bin structure identical to the one for aerosols. Condensation/evaporation and cloud droplet activation are computed according to Köhler theory (Köhler, 1936), taking into account the size and composition of particles in every size bin. Drizzle formation is parameterized using the autoconversion scheme developed by Seifert and Beheng (2001). In UCLALES–SALSA, the autoconversion schemes essentially move the cloud droplets that grow to drizzle-size from the dry-size-based cloud bins to the wet-size-based rain bins, allowing accurate modelling of precipitation formation via condensation and coalescence processes (Ahola et al., 2022). The parameterization is a computationally efficient alternative to explicitly simulating the initial rain formation step based on collisions between cloud droplets (Tonttila et al., 2021). We use seven rain bins covering wet diameters from 50 µm to 2 mm. For every bin (aerosol, cloud, and rain), the number concentration, the amount of all chemical species, and water amount are tracked. All microphysical processes (condensation, evaporation, coagulation/coalescence, sedimentation, etc.) are simulated identically for aerosol, rain, and cloud droplets. The model structure and applied algorithms are explained in more detail by Tonttila et al. (2017).

Several different sea spray emission parameterizations were included in the model. In various simulations in this study (Table 1), we make use of the emissions parameterizations by Mårtensson et al. (2003), Gong (2003), Fuentes et al. (2010), and Monahan et al. (1986). In majority of the presented simulations (case names starting with F10 in Table 1), the emission fluxes for submicron (dry) particles are modelled following Fuentes et al. (2010) who parameterized the impact of sea water DOC content on emitted aerosol number size distribution. As Fuentes et al. (2010) do not provide scaling from their experimental setup to emission from open ocean surface, we have scaled their flux to reproduce the number flux of Mårtensson et al. (2003) in the size range covered by both parameterizations (20–450 nm) at our base case sea surface temperature of 10 ∘C by multiplying it with 5500. The temperature dependence of submicron sea spray emission for this parameterization is implemented following Forestieri et al. (2018), the smaller modes having no temperature dependence and largest one increasing linearly. The flux of supermicron particles is computed following Monahan et al. (1986). To study the impact of the uncertainties in the sea spray emission parameterizations, extra simulations are made using the temperature-dependent scheme of Mårtensson et al. (2003) extended with the Monahan et al. (1986) scheme for sizes coarser than 2.8 µm (M03 in Table 1) and Gong (2003) scheme (G03 in Table 1). Monahan et al. (1986) and Gong (2003) schemes are coupled with temperature dependence from Jaeglé et al. (2011). For sea salt, the dry diameter is assumed to be half of that at 80 % relative humidity, as the error of this relation was reported to be below 5 % by Sofiev et al. (2011).

The selected sea spray flux size distributions are shown in Fig. 1. As the Fuentes et al. (2010) scheme (black) was scaled to reproduce the flux of Mårtensson et al. (2003) at 10 ∘C (MO3, green line), the emissions by these schemes are relatively similar. The parameterization by Gong (2003) (G03, blue line) was included in this study to cover the wide range of fluxes predicted by various sea spray emission parameterizations, as it differs from the other used parameterizations by much lower emission of small particles. As seen from comparing the black (F10) and brown (F10-DOC = 512 µM) lines on Fig. 1, the impact of the largest concentration of dissolved organic carbon considered by Fuentes et al. (2010) (512 µM) on the emission flux of the Aitken mode sea spray is comparable in size with the temperature effect shown for the Mårtensson et al. (2003) parameterization (light to dark green), while there is almost no effect on the rest of the size range.

The organic fraction of the emitted particles is computed following the wind speed, particle size, and chlorophyll a (Chl a) dependent parameterization of Gantt et al. (2011). Experiments with non-zero Chl a start with G11 in Table 1. As the relation between Chl a and DOC in sea surface is complex (Fuentes et al., 2010; Van Pinxteren et al., 2017), this scheme is applied independently of the DOC-dependent sea spray flux parameterization. When using it, no changes are made to the total sea spray flux or the emission size distribution, but for each size bin, part of the sea spray is emitted as primary organic matter instead of sea salt. Internal mixing of sea salt and organic matter is assumed. Hygroscopicity parameter of pure sea salt is set to 1.0 (Christiansen et al., 2020), and 0.1 is used for organic matter which is in the range reported by Kuang et al. (2020) for SOA, by Fuentes et al. (2011) for algal exudate, and only slightly lower than what was reported by Mayer et al. (2020) for marine organic species. The effect of organic content on particle hygroscopicity is demonstrated by the yellow lines on Fig. 1. With high chlorophyll a concentration (2 µg L-1), the small particles up to 0.3 µm in diameter consist almost entirely of organic matter while the large ones stay almost free of organics, leading to large difference in their hygroscopicities. Primary marine organics are treated as nonvolatile.

Partitioning of semivolatile organic species between aerosol and gas phase has been implemented using the volatility basis set approach (VBS, Donahue et al., 2006; Robinson et al., 2007). VBS is coupled to SALSA identically to what was used by Mielonen et al. (2018). We use three volatility bins with saturation concentrations 0, 1, and 10 µg m-3. Our approach deviates slightly from the classical implementations of volatility basis set (VBS) (e.g. Farina et al., 2010) that assume instant equilibrium between the gas and aerosol phases. In particular, we first use the VBS framework to compute the equilibrium vapour pressure of each semivolatile species for every aerosol, cloud, and precipitation size bin according to their organic content. These values are then used for computing diffusion-limited condensation and evaporation using the analytical predictor of condensation scheme (Jacobson, 1997).

As the study concentrates on clean marine areas, we focus on biogenic isoprene and monoterpenes as SOA precursors. Their fluxes from marine surface can either be prescribed or computed from their concentrations in sea water following Wanninkhof (2014).

Isoprene and monoterpenes are oxidized by the OH and NO3 radicals and ozone using reaction rates from Atkinson et al. (2006). The concentrations of the gas phase oxidants are prescribed at levels representative of longer-term averages and kept constant over the simulation period of 12 h, and their emission, production, and consumption are not explicitly modelled. The stoichiometric coefficients give the yields of the different semivolatile organic species (described by the volatility bins) resulting from precursor oxidation. The coefficients for monoterpenes originate from Kokkola et al. (2014) and for isoprene these are derived from the values used by Henze and Seinfeld (2006). The oxidation reactions together with the stoichiometric coefficients are listed in Table S1. Further oxidation of the semivolatile VBS species towards lower volatility bins is left out of these relatively short simulations. Due to low water solubility of the fresh SOA, the impact of liquid water on partitioning is ignored.

The model also includes an additional aqueous phase SOA formation route. We use a simplified model for the aqueous formation of SOA from isoprene epoxydiols (IEPOX) and glyoxal, similar to what was used by Mielonen et al. (2018). For partitioning the IEPOX and glyoxal to aqueous phase we account for the liquid water amount, and to take into account the “salting-in” effect (Kampf et al., 2013) we use higher effective Henry's law constants for aerosol than for cloud and rain droplets. For glyoxal, we apply the effective Henry's law constants from Kampf et al. (2013). The ionic strength has been reported to have an impact also on IEPOX uptake (Gaston et al., 2014) but its magnitude has a large uncertainty. Thus, for aerosol phase we use the measurements of Nguyen et al. (2014), and for cloud droplets and rain we have lowered the value to 105 M atm-1, same as used by e.g. Jo et al. (2019). As the background aerosol defined by Ackerman et al. (2009) for this case study is a simplified proxy that consists of pure ammonium bisulfate and does not reflect the complexity of real marine aerosol, we do not account for particle composition in more detail. The reactive partitioning model does not include any aqueous phase chemical reactions or irreversible processes. Gas phase oxidation reactions of IEPOX and glyoxal are included and are listed in Table S1 in the Supplement. To reduce the computational demands, the aqueous processes were turned on only in a dedicated sensitivity simulation.

The parameters for the LES simulations are based on the second research flight of the DYCOMS-II nocturnal aircraft campaign near the Californian coast in July 2001 (Stevens et al., 2003), characterized by lightly drizzling deck of closed-cell stratocumulus transitioning to heavily drizzling open cells. Additional CCN can substantially weaken the precipitation and retain cloud water in weakly precipitating marine boundary layer making it more sensitive to changes in aerosol concentration than strongly precipitating or non-precipitating regimes (Wang et al., 2011). Ackerman et al. (2009) adapted the DYCOMS-II case for LES intercomparison and demonstrated its aerosol sensitivity. The case has been previously used by Tonttila et al. (2017) for evaluating the UCLALES–SALSA model, demonstrating good skill to capture the cloud properties. Apart from including the marine emissions, the model setup and inputs used for the current study are almost identical to those used by Tonttila et al. (2017), with only a couple of small differences. Firstly, the model domain has been enlarged from 6 to 10 km to better represent the scale of the open cell structures. While by the end of the simulations the size of the open cells exceeds 10 km (Fig. S1), preliminary tests showed that the 10 km domain allowed to simulate the transition process in this case with sufficient accuracy (Fig. S2). Secondly, to reduce computational demands, the radiative transfer parameterization of Ackerman et al. (2009) is used instead of the four-stream scheme. Thirdly, the original autoconversion parametrization used by Tonttila et al. (2017) is replaced by the double-moment scheme by Seifert and Beheng (2001). Although the onset of precipitation is seen earlier with the time-independent autoconversion approach than with the collision-based rain formation, the schemes produce consistent precipitation rates. Sensitivity of the results to the autoconversion scheme is shown in the Supplement, Sect. S1.3.

The model domain covers 10×10 km area with 60 m horizontal resolution. Vertically, the domain consists of 97 layers reaching up to 1.5 km, layer thickness increasing from 5 m at the surface to 80 m at the model top. The resolution, optimized for this case by Ackerman et al. (2009), is less than 25 m for all in-cloud and below-cloud layers, and 5 m in the regions with largest gradients (near surface and cloud top). The 12 h long simulations were run with 1 s time step. If needed, the time step is automatically reduced during the simulation to maintain the numerical stability of the transport processes. Within a single model time step, the condensation equation and thus water partitioning between gas and liquid phases is solved in sub-time steps of 0.05 s to avoid problems with small aerosol particles that quickly respond to changes in conditions.

For each grid cell, the background aerosol was initialized with a bimodal distribution of ammonium bisulfate aerosol defined by Ackerman et al. (2009). The first hour of the simulations is used as a spin-up period to allow the turbulence to build up and boundary layer to settle. In addition to drizzle formation, the marine emissions are also inactive during spin-up to avoid build-up of emitted components in the lowest model layers while the turbulence is still developing. Horizontally averaged vertical profiles and domain average parameters are saved every 60 s; 3D fields are saved once per hour.

Affecting the cloud processes by changing CCN number and properties could lead to feedbacks to the sea spray emission through changing wind fields. These kinds of feedback processes are out of the scope of the current study but could mask or amplify the direct effects of the studied parameters. Thus, to reduce the impact of model feedbacks, the sea spray emission module is forced by constant wind speed of 10 m s-1, selected to match the midpoint of the maximum annual sea spray production range (7–16 m s-1) reported by Grythe et al. (2014). Tables 1 and 2 list the 26 model simulations discussed in this paper for the sea spray and VOC emissions, respectively. The simulations are set up to evaluate one effect at a time in order to be able to directly compare the magnitude of the different effects. The impact of specific changes to the emission fluxes is evaluated by comparing with the no-emission control, and the F10 experiment with SST 10 ∘C and no DOC that we refer to below as the base case. The no-emission control was run without any sea spray or VOC emission (NoEms in Tables 1 and 2). This simulation is almost identical to the one described by Tonttila et al. (2017), with the above mentioned differences in selected parameterizations.

The results regarding the primary sea spray flux parameterization and its temperature dependence are discussed in Sect. 3.2, while Sect. 3.3 concentrates on the impacts of sea water organic content either by flux enhancement or sea spray composition. Unless specified otherwise in Table 1, the setup and parameters identical to the base case were used for all other sea spray simulations discussed in these sections.

The model simulations with isoprene and monoterpene emissions are listed in Table 2. In order to evaluate the impact of these VOCs, we utilized the measurements of Kim et al. (2017), who during the HiWinGS measurement campaign measured the air concentrations and marine fluxes of both isoprene and monoterpenes over the Northern Atlantic.

In majority of cases, the VOC concentrations in the model were initialized with steady-state mass mixing ratios reached by the model at cloud-level (1 ng kg-1 of monoterpenes and 2 ng kg-1 of isoprene) by the end of a 5 d long preliminary simulation on reduced horizontal grid. VOC fluxes in this preliminary simulation were kept constant at the mean levels observed by Kim et al. (2017) (0.83 pmol m-2 s-1 of isoprene and 0.44 pmol m-2 s-1 of monoterpenes), and oxidant concentrations were prescribed at levels representative of clean marine areas (20 ppb of ozone, 0.1 ppt of OH, and 0.01 ppt of NO3 radicals, marked as “default” in Table 2). Boundary layer ozone concentrations close to this were measured during DYCOMS-II campaign (UCAR/NCAR, 2006). Similar ozone and OH concentrations were also used by Kim et al. (2017) in their calculations.

Section 3.4 discusses the changes in the cloud layer due to the SOA formed from the measured mean and maximum VOC fluxes and concentrations. The sensitivity of the SOA cloud effects to the background aerosol is discussed in Sect. 3.5.

Figure 2 illustrates the development of the cloud layer in the UCLALES–SALSA simulations. At the beginning of the simulation (left column), the clouds consist of closed cells with 1–2 km diameter and very little drizzle is observed. As the simulations progress (middle column), the size of these structures grows and light drizzle is formed. In the simulations without sea spray emissions (upper row), the clouds start to break after about 6 h while the amount of drizzle increases. By the end of the run (right panel), the closed cells have turned into large open cells with heavy drizzle. The horizontal size of the structures grows from couple of kilometres to more than 10, exceeding the size of the 10 km model domain (Yamaguchi and Feingold, 2015). In the F10 case simulations that include sea spray (lower row), the changes are slower, and by the end of the run, the clouds have not yet fully reorganized.

Comparing the no-sea-spray control run (grey) in Fig. 3 with the base case (F10, black) shows that including the sea spray emission to the simulations can substantially increase the number of cloud droplets (dashed lines in panel a), and thus lead to smaller cloud droplets (panel b) delaying the drizzle (panel e) and slowing down the removal of the background aerosol (panel f). This introduces a positive feedback to the system as the slower washout of the aerosol results in higher CCN concentration. As seen from the cloud water path plotted in panel c, this can delay the transition to open cells by several hours.

In order to evaluate the impact of uncertainties in sea spray emission, simulations were made with two alternative sea spray parameterizations: (i) G03 – Gong (2003) and (ii) M03 – Mårtensson et al. (2003) extended with Monahan et al. (1986) scheme for coarse sizes. While the emission flux for dry diameters below ∼0.5 µm in the base case was scaled to match that of the M03 scheme (green) and supermicron flux in both cases is computed according to the same Monahan et al. (1986) scheme, the different shapes of the size distributions (see Fig. 1) still lead to somewhat different outcomes. The M03 scheme produces more particles with diameters below 0.1 µm and fewer larger ones, which leads to activation of large number of smaller particles resulting in smaller average cloud droplet size.

As the parameterization of G03 (blue) was originally developed to reduce the overestimation that was evident when extending the Monahan et al. (1986) formula towards smaller sizes (Gong, 2003), its fine particle flux is significantly lower than the fluxes other parameterizations give (about 5 times lower for 0.1 µm particles and more than an order of magnitude for smaller), while for the supermicron range it gives slightly higher flux (Fig. 1). Compared with fine particles that extend the cloud lifetime, the effect of the coarse fraction is different – acting as giant CCN (GCCN), they speed up the collision–coalescence process that can lead to earlier drizzle formation (Houghton, 1938). For this reason, the cloud water path (panel c in Fig. 3) reduces even faster with G03 emission (blue) than without any sea spray (grey).

The importance of temperature effects on sea spray emissions was tested by varying the SST only in the sea spray emission without affecting any other fluxes in the LES. Both the temperature dependency parameterization of Forestieri et al. (2018) coupled with the Fuentes et al. (2010) sea spray flux and Mårtensson et al. (2003) temperature-dependent parameterizations (F10 and M03 on Fig. 4) were used. In addition to the 10 ∘C base case, two alternative sea surface temperatures were simulated, covering the range from 0 to 25 ∘C.

As seen from the dark red and blue lines in Fig. 4, for the case of the temperature dependency parameterization of Forestieri et al. (2018) used in the base case setup with the Fuentes et al. (2010) emission scheme that keeps the emission of fine modes constant regardless of SST, the effects are not large. The substantial differences between the light blue, green, and red lines in Fig. 4 demonstrate that in the Mårtensson et al. (2003) scheme, the temperature effects are much larger, as smaller number of fine particles and larger amount of coarse fraction is emitted with warmer SST (Fig. 1), both working in the same direction of speeding up the transition. In the cold conditions of 0 ∘C SST (light blue in Fig. 4), the clouds only start to reorganize at the very end of the simulation.

As demonstrated by Terai et al. (2012), the ability of aerosols to suppress precipitation is relatively weak for clouds thicker than ∼200 m. In our case, the clouds start out about 400 m thick, and indeed the drizzle increases and the cloud liquid water path starts reducing by the end of the simulation, even with the case with the highest sea spray emission (M03 at 0 ∘C SST, light blue in Fig. 4). In the base case, about 1.5 million particles are emitted from a square metre of sea surface per second, around 2/3 of those in the size range to activate as CCN. The loss rate of particle number due to coalescence processes in the cloud is about 2.5 times higher than the emission of potential CCN, leading to the continuous decrease in cloud droplet number. In the highest emission case (M03 at 0 ∘C), the emission of potential CCN is only ∼10 % higher than in the base case, indicating that the noticeably longer cloud lifetime than in the base case could be partly also due to lower number of GCCN emitted in colder conditions.

Organic content of surface sea water can impact the aerosol by either changing the emission size distribution or changing the composition and thus the hygroscopicity of the emitted particles. To be able to directly compare the impacts these different pathways have on the cloud layer, we applied them separately in two different simulations. The impact of changing the size distribution was tested via a simulation (F10-DOC = 512 µM) otherwise identical to the base case but the dissolved organic carbon content in sea water was changed from zero to 512 µM, the maximum value for the parameterization. This results in increased emission of sub 0.1 µm particles (Fig. 1). In this simulation, sea spray composition and resulting hygroscopicity was not changed. In order to evaluate the impact of changing hygroscopicity, another simulation (F10-G11-Chl a= 2 µg L-1) was run with sea water chlorophyll a concentration set to 2 µg L-1, and the organic fraction of the particles was computed by the Gantt et al. (2011) aerosol size-dependent scheme. In this simulation, DOC was set to zero. The results of these simulations are shown in Fig. 5 (flux enhancement from dissolved organic carbon in sea water with brown line and hygroscopicity reduction due to high organic fraction of small particles with yellow line). Both effects are relatively small compared to the temperature impacts of the M03 parameterization (Fig. 4) and in the expected direction. The larger fine mode emission flux due to DOC leads to more cloud drops activating and a slight delay in sub-cloud precipitation (dashed lines in panel e) and thus slightly longer lifetime for the closed cell structures. The reduction of fine particle hygroscopicity (yellow line) has somewhat more noticeable effect and expectedly in the opposite direction – reduced hygroscopicity of the fine aerosol leads to smaller number of particles activating, earlier drizzle, and loss of clouds.

In order to evaluate the importance of the particle size dependence of the organic fraction, an extra sensitivity simulation (F10-G11-Chl a= 2 µg L-1-noSizeDep) was performed where the maximum organic fraction was attributed to emitted sea spray of all sizes (orange line in Fig. 5). Lower water uptake of the coarse sea spray due to its large organic fraction substantially reduces the GCCN effect, keeping the cloud field stable for noticeably longer.

In this section, we investigate the impact of secondary organic aerosol formed from marine isoprene and monoterpenes emissions on the cloud layer. The simulations are listed in Table 2.

The mean fluxes observed by Kim et al. (2017) during the HiWinGS measurement campaign over the Northern Atlantic (0.83 pmol m-2 s-1 of isoprene and 0.44 pmol m-2 s-1 of monoterpenes) were emitted in the K17-meanFlx case. The model was initialized with the steady-state VOC in-cloud mixing ratios reached by UCLALES–SALSA with the mean fluxes – 1 ng kg-1 of monoterpenes and 2 ng kg-1 of isoprene. The mean VOC air mixing ratios measured during the HiWinGS cruise were substantially higher – 85 and 23.5 ng kg-1 of monoterpenes and isoprene respectively – than the ones reached by the model with the mean fluxes. Especially for monoterpenes, this imbalance between the marine fluxes and air concentrations was also noticed by Kim et al. (2017). For monoterpenes, the occasionally observed high peaks also caused the mean to be much higher than the median (15 ng kg-1). In order to investigate the impact of such background concentrations, we run additional simulations where the model was initialized with both of those mixing ratios (K17-meanFlx-meanInic and K17-meanFlx-medInic).

As the VOC fluxes also exhibited high spatial and temporal variability and values much higher than the mean were regularly observed during the campaign, an extra simulation (K17-maxFlx) was run with the reported maximum fluxes, which were more than an order of magnitude larger than the mean (18 pmol m-2 s-1 of isoprene and 27 pmol m-2 s-1 of monoterpenes). As these high emissions are not expected to be long-lasting, the model was initialized with the steady-state conditions of 1 and 2 ng kg-1 to see the impact of short-term emission peaks.

To study the impact of the selected oxidant concentrations, the simulations were repeated with doubled ozone concentration and OH and NO3 increased by an order of magnitude, representative of areas with higher anthropogenic impact, such as Northern Atlantic (indicated with -hiOx in Table 2 and case names).

As no sea spray was emitted in these simulations, we evaluate the effects of the emissions by comparing with the NoEms case. As seen from Figs. 6 and 7, SOA formed from marine monoterpenes and isoprene can have as strong impact on clouds as the direct sea spray emissions at 10 m s-1 wind speed. However, the K17-meanFlx case (green line) with the mean fluxes and steady-state background VOC concentrations does not differ from the case without VOC emission (grey) noticeably even with the higher oxidant levels. As the mean monoterpene air concentration observed by Kim et al. (2017) was influenced by the rare but high peak values, we ran simulations initializing the model also with their median observed mixing ratio, which we considered more representative of an average day. The simulation with low oxidant concentrations representing clean marine atmosphere (light blue line in Fig. 6) also does not differ from the NoEms case in a noticeable manner, while some impact can be observed in more polluted atmosphere (blue line in Fig. 7). The mean initial concentration (yellow) has the largest effect, but the maximum fluxes (light and dark red in Figs. 6 and 7) can also slow down the transition to open cell structures by hours even with low initial concentration. The impact is dependent on the rate of oxidation that is defined in the model by prescribed oxidant concentrations.

In order to elucidate the individual contributions from isoprene and monoterpenes, two simulations were made based on the K17-maxFlx-hiOx simulation by zeroing either isoprene or monoterpene emission and emitting the maximum observed flux of the other VOC (K17-onlyIsop with only isoprene and K17-onlyMtrp with only monoterpenes emissions). As seen from the sensitivity studies where either isoprene or monoterpene emissions were zeroed (green and pink in Fig. 8), the whole SOA production in the model and its effect on clouds was solely due to monoterpenes.

To save the computational resources, the aqueous phase SOA formation from isoprene had been turned off in all the previously discussed simulations. The impact of these processes was investigated with the K17-onlyIsop-aqSOA setup which was otherwise identical to K17-onlyIsop but accounted also for aqueous phase SOA formation through isoprene epoxydiols and glyoxal. Isoprene products indeed became visible when the aqueous phase production was turned on (blue line in Fig. 8). However, their impact still stayed significantly lower than that of monoterpenes.

In the high VOC emission case (K17-maxFlx-hiOx, dark red in Fig. 6), the transitioning timescale to open cell structure is extended by ∼3 h. As the mass fraction of SOA stays very small (< 2.5 % within 1 km height from surface), its unexpectedly large impact requires further investigation. The background aerosol is defined by Ackerman et al. (2009) as bimodal distribution of ammonium bisulfate with high number concentration of Aitken mode aerosol below the sizes that readily act as CCN (Fig. 9). Comparing the simulations with and without marine VOC emissions (Fig. 10, left panel) shows that the condensing organic vapours help these particles to grow to the next size bin, though even then they mostly stay too small to contribute much to the activated CCN (Fig. 10, right panel). However, both the aerosol and the cloud droplet concentration of the coarser bins stay higher in the simulation, with the SOA formation indicating that the large impact stems from positive feedback of the extra particles delaying the drizzle that would otherwise scavenge larger fraction of the particles from the whole size range.

Considering this, we performed two sensitivity studies – one with removing the fine size mode entirely (K17-maxFlx-aerInic = cr) and another initializing the model with sea salt aerosol with size distribution defined by Fuentes et al. (2010) emission scheme (K17-maxFlx-aerInic = sslt) as background (see Fig. 9). In the latter case, the concentration was normalized to reproduce the same number of cloud droplets as the noEms case immediately after the model spin-up. The setups were otherwise identical to K17-maxFlx-hiOx simulation. Corresponding no-emission control simulations without any VOCs were run for both of the alternative initial conditions (NoEms-aerInic = crs and NoEms-aerInic = sslt). The results of these simulations are shown in Fig. 11.

Comparing the three emission-free control simulations (grey, pink, and cyan lines in Fig. 11) shows that the reduced initial aerosol number concentration itself leads to slightly shorter lifetime of the closed cell regime in both of the alternative background cases although both of the alternative initial conditions reproduce the same cloud droplet number in the beginning of the simulations (dashed lines in panel a). Comparing the pink line with the red one shows that the effect of VOC emission disappears completely when the Aitken mode is removed from the background aerosol. Thus, we conclude that the large VOC effect critically depends on the underlying background aerosol distribution. Marine aerosol regularly exhibits a bimodal distribution (Hoppel and Frick, 1990; Hudson et al., 2015), where the dip between the modes is caused by activation of large enough particles and their subsequent growth due to cloud processing or removal through wet deposition. Both in-cloud coalescence and wet deposition of activated particles lead to reduction of particles available to act as CCN. The small mode consists of primary and secondary particles that are too small for cloud activation in their current size but capable of growing to CCN size by condensation of organic vapours. Larger number of small particles are likely to exist downwind marine areas with high biological activity as a result of DMS emission, oxidation, and following aerosol nucleation (Sanchez et al., 2018). However, as seen from Fig. 1, some of the sea spray parameterizations also produce noticeable number of particles in this size range, although they are not separated from the larger particles by a gap. Thus, it is interesting to consider the case where the background aerosol consists of freshly emitted sea spray. Comparing the simulations without and with VOC emission (cyan and teal lines in Fig. 11) shows that condensation of the organic vapours on the wide fresh sea salt distribution from Fuentes et al. (2010) also delays the transition to open cell regime, although in lesser extent. Thus, significant impact of marine VOCs on stability of the cloud layer is more likely in areas with high biological activity, as they can provide both larger amount of Aitken mode particles and the organic vapours to condense on them to grow them to CCN size.