Mesocosm experiments were performed during two intensive campaigns: the first, during summer oligotrophic conditions (hereafter referred to as non-bloom conditions), occurred from 22 June to 10 July 2012 at the Station de Recherches Sous-marines et Océanographiques in the Bay of Calvi (BC), Corsica; the second, performed during winter pre-bloom conditions, took place from 21 February to 5 March 2013 in the Bay of Villefranche (BV), France. The two bays share many similarities in term of temperature, salinity, phosphate (PO43-), nitrate + nitrite (NOx), and silicate (Si) seasonal variations, and they both show typical bloom conditions in winter-spring and oligotrophic conditions in the summer, corresponding to a stratified water column. Pre-bloom conditions are also observed at both locations (Gazeau et al., 2015). The presence of pre-bloom and non-bloom conditions was confirmed by the order-of-magnitude difference in the average seawater chla concentrations (chlaBC, avg=0.069± 0.009 mgm-3, chlaBV, avg=1.005±0.125 mgm-3). Detailed site and experimental information for both campaigns that legitimize the comparison to test in these two locations at different seasons can be found in Gazeau et al. (2015).

Briefly, the mesocosms used in this study (volume of ∼ 52 m3) were fully described in Guieu et al. (2014). The mesocosms consisted of large bags made of two 500 µm thick films of polyethylene mixed with vinyl acetate (EVA, 19 %) with nylon meshing in between to allow maximum resistance and light penetration (Haikonene KY, Finland). Natural seawater was filtered through a mesh grid to remove large debris when deploying the mesocosms. In order to avoid actual atmospheric deposition, the mesocosms were covered with UV-transparent ethylene tetrafluoroethylene (ETFE) roofs, except during periods of sampling. In this way, transfer of rainwater/deposition was prevented, while preserving the sunlight irradiance of the mesocosms. The covers were elevated to ∼ 10 cm above the top of the mesocosms, allowing air to circulate to avoid a confinement effect in the trapped water. Among nine deployed mesocosms, three remained unmodified as controls and six were modified in terms of partial pressure of CO2, pCO2. The pCO2 levels used were slightly different between the two campaigns, as a consequence of different ambient pCO2 levels (i.e. ∼ 450 vs. 350 µatm at BC and BV, respectively). In the Bay of Calvi, the six targeted elevated pCO2 levels were 550, 650, 750, 850, 1000 and 1250 µatm. In the Bay of Villefranche, the levels were 450, 550, 750, 850, 1000 and 1250 µatm. These elevated pCO2 levels were reached by adding varying volumes of CO2 saturated seawater to the mesocosms. At both sites, seawater was pumped from near the mesocosms and sieved onto a 5 mm mesh sieve in order to remove large organisms. Pure CO2 was actively bubbled through the water for several minutes in order to achieve saturation; the water was then transferred to 25 L plastic containers for addition to the mesocosms. Depending on the targeted pCO2 level, 50L to more than 500 L were added. A diffusing system was used to ensure a perfect mixing of this CO2 saturated seawater inside the mesocosms. In order to minimize the stress induced by the addition of large quantities of acidified water, the acidification of the mesocosms was performed over 4 days, and the experiments started when the targeted pCO2 levels were reached. The CO2 levels were chosen in order to cover the range of atmospheric CO2 concentrations projected for the end of the century following various scenarios (RCP 2.6 to RCP 8.5; IPCC, 2013).

For the experiments described here, every morning approximately 5 L of surface water (taken within a 15 cm depth from the surface) was pumped from each mesocosm using a perfluoroalkoxy alkane (PFA) pump (Saint-Gobain Performance Plastics) activated by the pressurized air from a diving tank and connected to braided PVC tubing (Holzelock-Tricoflex, I.D. 9.5 mm). The water pump was flushed with seawater from the respective mesocosm prior to sampling. Samples were stored in large brown glass bottles outside (avoiding direct sunlight) until the experiments were performed that same day. During the BV campaign, the pump could not be used on 3 March 2013 due to unsafe sea conditions; water was instead manually sampled from the mesocosms with 2.5 L glass bottles while wearing long gloves. Additionally, due to dangerous wind and wave conditions, sampling was not performed on 5 March 2013 during the BV campaign. Instrumental failures occurred on 4 and 10 July 2012 at BC and on 28 February–1 March 2013 at BV.

For both campaigns, the mesocosms were located offshore in pelagic waters in order to measure primary marine aerosol properties and biogeochemical parameters of the water while minimizing contamination from anthropogenic sources. The mesocosms were reached via ocean kayak or boat. For both campaigns, sampling operations were performed from a mobile plastic platform that was moved via a rope network. The water temperature variances between BC and BV are quite drastic, given the time of year the experiments took place. From conductivity, temperature and depth (CTD) measurements, at BC the water temperature measured nearest to the surface varied from 21.8 to 25.2 ∘C; at BV, the temperatures ranged from 13.0 to 13.6 ∘C.

For the experiments described here, we focused on three different mesocosms: control mesocosm C3, and acidified mesocosms P3 and P6. Mesocosm P6 was the most acidified of all mesocosms (pCO2∼ 1250 µatm), and P3 was acidified to an intermediate level (pCO2∼ 750 µatm). This allowed a range of acidification effects to be analyzed.

Bubble-bursting experiments were performed using a square glass tank (20 l × 20 w × 25 hcm3), filled with 3.6 L of seawater (water depth of ∼ 10 cm), sealed with a stainless steel lid and continuously flushed with particle-free air (11 Lmin-1). The tank was constantly slightly over-pressured with particle-free air to ensure the absence of ambient room air. Aerosols were generated by splashing mesocosm seawater through plunging water jets, separated into eight jets via a flow distributor. The mesocosm seawater was re-circulated using a peristaltic pump; to minimize an increasing temperature of the seawater caused by constant re-circulation, a stainless steel heat exchanger was used on the seawater exiting the pump. The temperature of the water was recorded with a temperature sensor at the beginning and end of each experiment (for the BV campaign). Since no measurements of the bubble size distribution could be performed in such a small device, all water flow characteristics were performed according to the Fuentes et al. (2010a) settings, to reproduce the same bubble size distribution. The water flow rate was set to 1.8 Lmin-1, the height of the jets above the water surface was 9 cm and the penetration depth of the jets was ∼ 7.5 cm. Particle-free air was blown over the seawater (using a j-shaped tube ∼ 1.5 cm above the water surface) to mimic the wind blowing effect on the bubble-bursting process. Some of the water samples were also enriched by the addition of an organic rich microlayer from the same mesocosms (average 100 mL, range 50–170 mL). The enriched mesocosm samples were tested after the un-enriched sample to compare the effect of additional organic species.

Blank measurements were performed during the first 10 min of each experiment by verifying the aerosol concentration was zero in the particle-free air flushed tank. Between each water sample testing, the aquarium and tubing were rinsed with ultrapure water (> 18 MΩ cm), and clean water was re-circulated throughout the experimental setup for 10–15 min. Experiments were performed on the mesocosm water in different orders each day to make sure there were no experimental biases.

The aerosol flow was passed through a diffusion drier and was sent through a neutralizer into a differential mobility particle sizer (DMPS) and miniature continuous-flow streamwise thermal-gradient CCN chamber (CCNc) (Roberts and Nenes, 2005) to determine particle CCN activation properties. The neutralizer used was a variable-amplitude corona discharge which charges particles to the equilibrium charge distribution (Stommel and Riebel, 2004, 2005). For the BC experiments, the neutralizer voltage was ∼ 2.8 kV and for BV, ∼ 2.0 kV.

For the CCNc-DMPS system, aerosol flow passed first through a TSI-type DMA (length 44 cm) selecting particle sizes in 26 channels ranging from 10 to 400 nm by stepping the voltage over an integration time of about 8.5 min. Immediately after the DMA, the aerosol flow was split between the CCNc and a TSI CPC model 3010. The DMA sheath flow rate was 9 Lmin-1, and the sample flow rate was 1 Lmin-1 in BC; 7.5 and 1.35 Lmin-1 were used in BV. For the BV campaign, the aerosol flow was split 1 Lmin-1 for the CPC and 0.35 Lmin-1 to the CCNc. In the CCNc, a total aerosol flow rate of 100 sccm with a sheath-to-aerosol flow ratio of 5 was used. The CCNc operated at specific temperature gradient (dT) settings, testing two different supersaturations (SSs). For the BC campaign, a temperature gradient of 6 ∘C (dT6) was used in the column and the top temperature of the column varied as the ambient temperature changed (Ttop-Tamb= 2 ∘C); in BV, dT6 and a temperature gradient of 3 ∘C (dT3) were tested and the top temperature of the column was always set at 30 ∘C. The data are plotted as activated fraction vs. particle diameter and fit with a sigmoid curve, from which we obtain the activation diameter at each dT (see Asmi et al., 2012 for more details). The CCNc system was calibrated with atomized (NH4)2SO4 and NaCl solutions at the beginning, end and throughout each campaign. The activation diameter of the calibration was then used to calculate the corresponding supersaturation; this supersaturation was then used for all mesocosm experiments. The activation diameters and corresponding supersaturations for each dT for both campaigns are shown in Table 1 (dT6 = 0.39 % SS, dT3 = 0.08 %). The range of SS values used in this work is typical of those reported in natural clouds. Anttila et al. (2009) found cloud SS values from 0.18 to 0.26 % for low-level clouds in northern Finland, Hegg et al. (2009) obtained a SS range from 0.2 to 0.3 % for clouds over the California coast, and Asmi et al. (2012) found SS values from 0.1 to 0.3 % at the puy-de-Dôme station in central France.

Every day at 8:30 (LT), depth-integrated sampling (0–10 m) was performed in each mesocosm using 5 L Hydro-Bios integrated water samplers. Samples for pigment analyses were filtered (2 L) onto GF/F. Filters were directly frozen with liquid nitrogen and stored at -80 ∘C. Measurements were performed on an HPLC from filters extracted in 100 % methanol, disrupted by sonication and clarified by filtration (GF/F Whatman). Samples for microbial diversity (2 mL) were fixed with 0.5 % final concentration glutaraldehyde, frozen in liquid nitrogen, and then transferred to a -80 ∘C freezer. Virus-like particles, heterotrophic and autotrophic prokaryotes abundances were measured with the use of Flow Cytometry (Beckton Dickinson FACS Calibur model). Total organic carbon (TOC) was measured instead of dissolved organic carbon (DOC) in order to avoid contamination during filtration. However, the TOC measurement is referred to hereafter as DOC, due to the low concentration of particulate organic carbon in both sites (averaged over all mesocosms and all sampling times, BC: 4.32± 0.91 µMol, BV: 11.49± 5.50 µMol, which was typically less than 10 % of TOC). DOC concentrations were determined on 20 mL samples by high temperature oxidation with a Shimadzu 5000A TOC Analyzer. Transparent exopolymeric particle (TEP) concentrations were measured spectrophotometrically according to a dye-binding assay (Engel, 2009). Samples (250 mL) were filtered onto 0.4 mm pore size polycarbonate filters under low vacuum (< 100 mmHg), stained with 1 mL of Alcian blue solution (0.02 g Alcian blue in 100 mL of acetic acid solution of pH2.5) and rinsed with 1 mL of distillate water. Filters were then soaked for 3 h in 6 mL of 80 % sulfuric acid (H2SO4) to dissolve the dye, and the absorbance of the solution was measured at 787 nm, using acidic polysaccharide xanthan gum as a standard.

The marine aerosol size distributions remained fairly stable during a given experiment, which lasted around 1 h for each water sample. The aerosol size distribution also remained stable throughout the course of each campaign, with a similar distribution shape. Four lognormal modes were fit to the average size distributions of each campaign, with results summarized in Fig. 1 and Table 2. The average size distributions were taken from 214 size distributions for BC and 182 size distributions for BV, and size envelopes are also included within Fig. 1. In order to investigate the size of the aerosol independently of the concentration, the size distributions were normalized using the total aerosol number concentration. We found that the primary marine aerosol size distributions were best described using the three expected modes: nucleation mode (around 20 nm), Aitken mode (around 40 nm), and accumulation mode (around 100 nm); and an additional fourth mode around 260 nm. Using only three modes for the fitting procedure could not satisfactorily represent the primary marine aerosol size distribution. A mode at around 300 nm was also found by Sellegri et al. (2006) during bubbling experiments for which the effect of wind on the surface breaking bubbles was simulated in a similar experiment set-up. The 300 nm mode was interpreted by the authors as the result of a thicker bubble film where the bubbles are forced to break by the wind instead of reaching a natural breaking thickness. The four average lognormal modal diameters determined (18.5± 0.6, 37.5± 1.4, 91.5± 2.0, 260± 3.2 nm) were present in both BC and BV. The lognormal mode fitting was also used to determine the particle number fraction at each lognormal modal diameter. Looking at the number fractions on a daily temporal scale for SS = 0.39 %, the lognormal mode number fractions remained relatively constant throughout both campaigns, though differences were noted between the campaigns (Fig. 2, Table 2). Throughout the campaign in BC, the fractions of Modes 1–3 were approximately equal in magnitude (0.297), whereas in BV, the magnitude of the Mode 2 (the Aitken mode) fraction relative to the other modes was dominant (0.48). These same trends were observed for all experiments using SS = 0.08% and all the enriched mesocosm samples (Figs. 3, 4). When augmenting the bacterial abundance in seawater, Collins et al. (2013) observed an increased particle fraction of the smallest lognormal mode diameter with no change to the shape or magnitude of the size distribution; this was attributed to the replacement of internally mixed salt/organic particle types by insoluble organic type particles. Previous studies have also indicated changing size distributions or mode number fractions with increasing organic material (Fuentes et al., 2010b; Sellegri et al., 2006). In the present study, we will examine which chemical component is linked to the increase of the Aitken mode particles in Sect. 3.3.

Many other studies have found different lognormal mode distributions of both artificial and natural seawater samples, though many have similar modal sizes. Differences in the size distribution of laboratory generated primary marine aerosol found in the literature seem to depend on the method used to generate them. Fuentes et al. (2010a) observed four modes (modal sizes 14, 48, 124, 334 nm) generated from plunging-water jet experiments with artificial seawater. Mode 4 was believed to be linked to splashing water from the jet mechanism. Plunging-water jet experiments were found to most closely mimic the size distribution of ambient primary marine aerosol (Fuentes et al., 2010a) while generating sufficient aerosol for characterization measurements. In a separate study (Fuentes et al., 2010b), four modes represented both artificial and natural seawater (modal sizes 15, 45, 125, 340 nm) well. Similar to our findings, increasing organic content was found to increase the number fraction of Mode 2 while decreasing the relative fractions of the other modes. In similar experiments with Baltic seawater collected between May and September, Hultin et al. (2011) observed either two lognormal modes (site: Askö, 86, 180 nm) or three lognormal modes (site: Garpen, 93, 193, 577 nm). Sellegri et al. (2006) tested synthetic sea salt with a weir and observed three lognormal modes (4 ∘C: 30, 85, 200 nm; 23 ∘C: 45, 110, 300 nm). After adding SDS, they noticed an increase in the fraction of particles at the smallest lognormal diameter. In synthetic seawater experiments with sintered glass filters, Mårtensson et al. (2003) observed one submicron lognormal mode (100 nm); Tyree et al. (2007) observed a lognormal mode at the same diameter using artificial and natural seawater with pore diffusers. The Mårtensson et al. (2003) and Tyree et al. (2007) studies using sintered glass filters or pore diffusers report relatively different size distributions compared to those obtained by plunging jet experiments, likely due to the bubble formation processes. Collins et al. (2013) observed three lognormal modes in seawater wave channel experiments (∼ 90, 220, 1000 nm), with changing number fractions as described in the above paragraph.

Temperature has also been shown to affect size distributions and aerosol number concentrations (Hultin et al., 2011; Mårtensson et al., 2003; Zábori et al., 2012a). The initial and final water temperatures were measured over the course of an experiment (∼ 1 h, BV only); the water temperature was found to increase by 4.6± 1.2 ∘Ch-1 on average due to the constant water recirculation through the peristaltic pump and an insufficient heat exchange system (encompassing all experiments, water temperatures ranged from ∼ 11.7–26.8 ∘C from the initial to final measurement). This increase is 2–5 times higher than the 1–2 ∘Ch-1 temperature increase measured by Zábori et al. (2012a). Even with this large temperature range, we did not observe that the increase in the water temperature affected the shape of the size distribution. We were not able to make qualitative statements about the relationship between the number concentration and temperature, with the level of uncertainty in the wind flow stability within these experiments. Acidification had no effect on the aerosol size distribution, as no clear differences were found between mesocosms C3, P3, and P6 within a given field campaign.

By testing two temperature gradients, it is possible to look at the hygroscopic properties of different size particles based on the different supersaturations at which they are activated. The activation diameter time series measured at SS = 0.39 % (dT6) is shown in Fig. 5. In BC, there is little variation temporally between the control and acidified mesocosms tested (Dp, 50, avg=46.47± 0.88 nm), indicating that acidification does not have a large direct effect on CCN activity. Additionally, the enriched samples showed similar behavior to the non-enriched waters, indicating that the addition of the organic-rich microlayer had little effect on the water uptake for the aerosols. For the experiments incorporating the enriched microlayer, the entire organic-rich volume was added a few minutes before starting the water jet system, rather than being continuously introduced to the tank throughout the entire experiment. This could have led to microlayer depletion over the course of an experiment, explaining why no visible difference was seen between the enriched and un-enriched samples. However, no clear difference was seen between the first (when the microlayer was present) and subsequent size distributions during a given microlayer enriched experiment. In past experiments, the addition of organics to bubbling experiments have shown changes in the size distribution (Sellegri et al., 2006), particle number concentration (Fuentes et al., 2010b; King et al., 2012; Tyree et al., 2007), and CCN activity (Collins et al., 2013; Fuentes et al., 2011). Other experiments have shown no visible change (Moore et al., 2011) from the addition of organics, similar to the experiments performed in this study. In some studies, the concentrations and/or nature of some of the organic surfactants were unrealistic (King et al., 2012; Moore et al., 2011; Sellegri et al., 2006). At BV, the average activation diameter for SS = 0.39 % over the course of the experiments was Dp, 50, avg=59.48± 1.1 nm (Fig. 5), while at SS = 0.08% the average activation diameter was Dp, 50, avg=141.91± 10.8 nm (Fig. 6). Activation diameters larger than the corresponding salt standards (here, shown as NaCl), indicating higher organic presence, are observed more at BV than BC for SS = 0.39 % due to the organic pre-bloom conditions, whereas the non-bloom water conditions were very stable at BC. For SS = 0.08 %, the activation diameters are very similar to the NaCl standard, which signifies a lower organic fraction for larger particles. Activation diameters for individual mesocosms in both campaigns are shown in Table 3.

There is an anti-correlation of the activation diameter at SS = 0.39 % with the ambient average air temperature in BV (Fig. 5), though no correlation exists with SS = 0.08 % (Fig. 6), or in BC (Fig. 5). However, more daily temperature variance was observed in BV than BC, based on the time of year of the campaigns. This anti-correlation could indicate an additional temperature impact on the emission of small particles (∼ 50 nm) and their chemical composition, though this effect is unclear and undocumented in the literature.

We investigated if the observed differences in the activation diameters from BC and BV could be linked to the different operating techniques used for the campaigns. As indicated previously, Ttop in BC was variable, changing as the ambient temperature changed. As the ambient temperature changed throughout the day, the temperature in the column would also change, leading to possible temperature instabilities throughout the course of an experiment. On the contrary, in BV, Ttop was fixed at 30 ∘C, a temperature higher than the daily temperature variability. In this way, temperature fluctuations in the column were avoided. However, in observing the measured temperatures throughout the column for both campaigns and all the experiments, the temperatures of the column remained quite stable for both methods of operation, so we believe that these effects are very minor. Additionally, it has been shown that organics can volatilize in the CCN column due to the temperature gradient (Asa-Awuku et al., 2009), biasing observed CCN activity. It is possible that this occurred for both campaigns based on the relatively high measured operating temperatures observed in the column; if organic material was volatilized, the activation diameters would increase from those shown here.

The hygroscopicity of the aerosol was determined using kappa-Köhler theory (Petters and Kreidenweis, 2007) following Asmi et al. (2012). Using the activation diameter and numerical iteration, the kappa value was determined when the maximum of the saturation curve was equal to the supersaturation in the CCNc, following, SDp=Dp3-Dp, 503Dp3-Dp, 5031-κexp⁡4σwMwRTρwDp, where S is the supersaturation, Dp is the diameter of the droplet, Dp, 50 is the dry diameter, R is the gas constant, T is temperature, and σw, Mw, and ρw are the surface tension, molecular weight and density of water, respectively. Lower kappa values correspond to more hydrophobic, or organic-like, particles. In BC, the average mesocosm kappa value at SS = 0.39 % was κavg, BC=0.95± 0.17, In BV, the average mesocosm kappa values for SS = 0.39 and 0.08 % are κavg, BV=0.45± 0.13 and 0.78± 0.14, respectively. This indicates that the smaller particles (measured at the higher SS) were higher in organic material. The kappa values obtained at SS = 0.08 % are more representative of the kappa that would be obtained for a bulk chemical composition, as most of the aerosol mass is comprised in the accumulation mode for submicron particles. Our value then falls well into the suggested range of the kappa average of marine aerosol, κmarine=0.72± 0.24 (Pringle et al., 2010). Little variance was seen between the control and acidified mesocosms.

Using the calculated kappa values, we determined the organic fraction using a kappa closure equation, κtotal=εorgκorg+1-εorgκinorg, where εorg is the bulk volume fraction of organic material and κ is the kappa of the organic or inorganic material. Following Collins et al. (2013), we used κinorg= 1.25, a good proxy for seawater, and κorg= 0.006. In BC, the organic fraction ranged from -0.21 to 0.46 (average, 0.24± 0.14) for SS = 0.39 %. For BV, the organic fraction ranged from 0.43 to 0.80 (average, 0.64± 0.11) for SS = 0.39 % and from 0.19 to 0.55 (average, 0.38± 0.11) for SS = 0.08 %. Previous studies have also found mass organic fractions ranging from 30 to 80 % in sea spray aerosol studies of water from the North Atlantic, Sargasso Sea near Bermuda, and Pacific water near La Jolla, California (Collins et al., 2013; Facchini et al., 2008; Keene et al., 2007). Negative organic fractions were calculated during the BC campaign due to the sensitivity of Eq. (2) to the value of κinorg over κorg. We show results here using a κinorg value supported by literature rather than determine non-realistic κinorg values to provide positive organic fractions. Therefore, the values shown in this work can be considered low estimates of the organic fraction. Table 3 shows average kappa and organic fraction values of each mesocosm for both campaigns.

The organic fraction of the Aitken mode particles (obtained from measurements performed at SS = 0.39 %) is significantly increased during the BV experiment compared to the BC experiment. This indicates that the Mode 2 fraction increase observed in the size distribution is due to the presence of organic matter, in agreement with the observations of Collins et al. (2013).

The ratio of cloud condensation nuclei (CCN) to condensation nuclei (CN) decreased slightly during the pre-bloom period for SS = 0.39 %: at BC, CCN / CNaverage=0.55± 0.03, whereas at BV, CCN / CNaverage=0.45± 0.07. The change between CCN / CN during the oligotrophic and pre-bloom conditions was likely due to the combined effects of a higher organic fraction and higher Mode 2 to Mode 1 ratio during pre-bloom conditions, likely caused by the increasing organic content of the water due to the pre-bloom. For SS = 0.08 % measured at BV, CCN / CNaverage=0.15± 0.02.

In a recent study, Rinaldi et al. (2013) showed that chla was the best biological surrogate for predicting organic enrichment in sea spray. Chla parameterizations are currently being used in models to account for the organic content of seawater. We find a strong linear correlation with same-day measurements of organic fraction (from SS = 0.39 %) and total chla concentrations (R2= 0.781, p< 0.00001) shown in log–log scale in Fig. 7, following Organic fraction[%]=42.28⋅(chla)[mgm-3]+22.98. Similar correlations were also found with a number of pigments: chlorophyll c1 + c2 (R2= 0.783, p< 0.00001), 19′-butanoyloxyfucoxanthin (R2= 0.711, p< 0.00001), alloxanthin (R2= 0.699, p< 0.00001), sum carotenes (R2= 0.773, p< 0.00001) and 19′-hexanoyloxyfucoxanthin (R2= 0.736, p< 0.00001) (Fig. 8). Various studies have found linear correlations between the organic fraction of aerosols measured at a receptor site and chla concentrations observed by satellite along the back trajectory (Langmann et al., 2008; O'Dowd et al., 2008; Rinaldi et al., 2013; Vignati et al., 2010); others have found exponential fittings (Gantt et al., 2011) with the same methodology or a Langmuir functional relationship (Long et al., 2011) using a model with experimental data from Facchini et al. (2008) and Keene et al. (2007). Figure 7 shows many of the existing chla-organic fraction parameterizations in the literature, including this work. It is clear that many of the parameterizations from the North Atlantic Ocean also describe the correlation in the Mediterranean Sea fairly well, even though the methodologies for most of them are very different from the one used in this study. The parameterization derived in this work lays at higher organic content when compared to the other parameterizations, even though it does not include secondary organic contributions as do many of the satellite-receptor site studies. This is likely due to the size dependence of the organic fraction that we observed in the BV data. Our parameterization is applicable for Aitken mode type aerosols and would probably shift towards lower organic content for accumulation mode particles. It is also possible that organic components in Mediterranean primary marine aerosol are of multiple origins and not solely linked linearly to chla-rich species. Bacteria have also been observed to affect the organic material in seawater (Gruber et al., 2006; Jiao et al., 2010; Ogawa et al., 2001). We find a correlation with heterotrophic prokaryotes (R2= 0.476, p=1.3×10-5), virus-like particles (R2= 0.161, p= 0.025), autotrophic prokaryotes (R2= 0.499, p< 0.00001) and Synechococcus abundance (R2= 0.143, p= 0.033), shown in Fig. 9a–d. In a wave channel experiment on natural seawater doped with Zobell growth medium, bacteria and phytoplankton (Dunaliella tertiolecta) cultures, Prather et al. (2013) also observed a link between heterotrophic bacteria and organic fraction while no correlation with chla was found, highlighting the necessity to study complex systems of all biological material (phytoplankton, prokaryotes, organic matter) for marine aerosol. Most likely, the observed differences between Prather et al. (2013) and this work have to do with the localized biogeochemical nature of the different experiments, causing variance in the chemical composition and organic fraction of the marine aerosol. An additional correlation (R2= 0.477, p=1.2×10-5) exists with TEPs (Fig. 9e), a surface-active complex, variable mixture of organics (Filella, 2014; Passow, 2012). During BC, there is also a sigmoidal correlation between organic fraction and DOC concentrations (χ2= 0.411, p< 0.00003); data are unavailable for BV (Fig. 9f). Sigmoid fits are also shown in Figs. 8 and 9 for all biological parameters where they could be determined. Sigmoid fits might be more appropriate to use in many cases, to conceptually constrain the organic fraction of the primary marine aerosol to one regardless of the chla concentration. We have included both linear and sigmoid fits, with their respective R2 and χ2 values for completeness.

We also wanted to see if correlations existed between different biogeochemical parameters and the temporal relative mode fractions shown in Fig. 2. Relative fractions of Modes 3 and 4 (91.5 and 260 nm, respectively) showed no clear correlations to any parameter. However, strong anti-correlations were observed between the Mode 1 (18.5 nm) relative fraction and the abundances and concentrations of virus-like particles, heterotrophic prokaryotes and all pigments previously discussed, except for alloxanthin, which had a positive correlation (correlations not shown). The relative fraction of Mode 2 (37.5 nm) showed strong positive correlations with the abundances and concentrations of virus-like particles, heterotrophic prokaryotes, TEPs and all the pigments discussed, except for alloxanthin, where no correlation was observed (correlations not shown). This further supports the idea of an increase in the Mode 2 (Aitken mode) relative fraction during periods of high biological activity due to the higher concentrations of organic material, at the expense of Mode 1.

The control and acidified mesocosms showed no significant differences in terms of correlations between organic fraction and different biogeochemical parameters. For studies of marine aerosol, this indicates that any acidification effects on these biological parameters impacts the physical and chemical parameters of the aerosol much less than the natural variances caused by organic pre-bloom and bloom periods. It is not yet clear whether this observation can extend beyond the western Mediterranean Sea. However, it is important to note that due to the oligotrophic nature of the Mediterranean, even during the pre-bloom conditions at BV, the chla concentrations and abundances of other parameters are still much lower than could occur in places like the North Atlantic Ocean.