The Effect of Seawater Salinity and Seawater Temperature on Sea Salt Aerosol Production

To improve our understanding of the impact of sea salt aerosols (SSA) on the Earth's climate, it is critical to understand the physical mechanisms which determine the size‐resolved SSA production flux. Of the factors affecting SSA emissions, seawater salinity has perhaps received the least attention in the literature and previous studies have produced conflicting results. Here, we present a series of laboratory experiments designed to investigate the role of salinity on aerosol production from artificial seawater using a continuous plunging jet. During these experiments, the aerosol and surface bubble size distributions were monitored while the salinity was decreased from 35 to 0 g kg−1. Three distinct salinity regimes were identified: (a) A high salinity regime, 10–35 g kg−1, where lower salinity resulted in only minor reductions in particle number flux but notable reductions in particle volume flux; (b) an intermediate salinity regime, 5–10 g kg−1, with a local maximum in particle number flux; (c) a low salinity regime, <5 g kg−1, characterized by a rapid decrease in particle number flux at lower salinities and dominated by small particles and larger bubbles. We discuss the implications of our results through comparison of the size‐resolved aerosol flux and the surface bubble population at different salinities. Finally, by varying the seawater temperature at three specific salinities we have also developed a simple parameterization of the particle production flux as a function of seawater temperature and salinity. The range of seawater salinity and temperature studied is representative of the global oceans and lower salinity water bodies such as the Baltic Sea.


The Sea Spray Simulation Chamber
SSA were generated in a temperature-controlled sea spray simulation chamber using a plunging jet. This system has been described in detail by Salter et al. (2014) and details are provided in the supplementary information (including a schematic of the experimental set-up, see Figure S1 in Supporting Information S1).

Particle Size Distribution Measurements
Following particle production by the plunging jet, the resulting aerosol particle-laden air was sampled through 0.8 m of tubing, a 1 m long Nafion dryer (MD-700-36F, Perma Pure LLC, Lakewood, New Jersey, USA), followed by another 1.9 m of tubing after which the flow was split. A condensation particle counter (CPC, type MCPC 1720, Brechtel Manufacturing Inc., Hayward, California, USA) with a flow rate of 0.36 L min −1 situated 0.4 m behind this split was used to enumerate the total number concentration for particles with D m > 0.01 μm. The size distribution of aerosol particles with 0.015 μm < D m < 0.906 μm, distributed over 37 size bins, was measured with a custom-built differential mobility particle sizer (DMPS) that was situated 0.5 m behind the split. The DMPS consisted of a 28 cm long Vienna type differential mobility analyzer and a condensation particle counter (CPC, type MCPC 1720, Brechtel Manufacturing Inc., Hayward, California, USA) with a flow rate of 0.36 L min −1 . A scan over all size bins was completed in 12 min and the measured size distributions were inverted and multiple-charge corrected using custom-built software.
An optical particle size spectrometer (OPSS) with a flow rate of 5 L min −1 (WELAS 2300 HP sensor and Promo LED 2000 H, Palas GmbH, Karlsruhe, Germany) was mounted 0.7 m above the sea spray simulation chamber and measured the particle size distribution in the optical diameter size range 0.150 μm < D < 10 μm, distributed over 59 bins. The OPSS operated with a white-light source in the wavelength range of 350-750 nm (Salter et al., 2015) and is therefore less affected by sizing ambiguities than OPSS instruments that use monochromatic light. The optical diameter measured by the OPSS depends on the refractive index of the sampled aerosol. Therefore, we have converted the measured optical diameters to volume equivalent diameters, D p , by assuming a refractive index of m = 1.54 − 0i for sea salt particles, which corresponds to the value for NaCl (Abo Riziq et al., 2007). The conversion was conducted using the software provided by the manufacturer (PDAnalyze Version No 2.024, Palas GmbH, Karlsruhe, Germany) and is based on instrument-specific Mie calculations. A 1.3 m long Nafion dryer (MD-700-48F, Perma Pure LLC, Lakewood, New Jersey, USA) with a sheath flow of 10 L min −1 was mounted horizontally in front of the OPSS in order to reduce the humidity of the sampled air. Temperature and relative humidity (RH) were monitored with two sensors (HYTELOG-USB, B + B Thermo-Technik GmbH, Donaueschingen, Germany) mounted in front of the sampling inlets of the OPSS and DMPS system. The RH measured in front of the DMPS and OPSS was 21.3% ± 0.4% and 17.1% ± 0.3%, respectively (mean ± standard deviation for the entire experiment). To account for changes in shape with RH, the diameters obtained from the DMPS and OPSS were shape-corrected using a dynamic shape factor calculated according to Zieger et al. (2017) to obtain volume equivalent diameters. Since Zieger et al. (2017) used the same sea spray simulation chamber and artificial seawater made in the same manner as the artificial seawater used in this study, their results should be applicable to our study. All sizing instruments were calibrated with polystyrene latex spheres.
Aerosol sampling efficiencies were estimated with the Particle Loss Calculator Software (von der Weiden et al., 2009) and are shown in Figure S3 in Supporting Information S1. The losses of the larger particles measured by the OPSS were expected to be highest in the horizontally aligned Nafion dryer. The estimated losses also depend on the particle density, which is in turn dependent on how quickly the water within the particles evaporates during transit through the dryer (i.e., how much time the particles spend as droplets in the dryer vs. how much time they spend as salt crystals). Since the drying of the particles is a continuous process, we have used a density of 1500 kg m −3 for the loss calculations, representing a value for partially dried sea salt particles between the density of seawater, 1000 kg m −3 , and that of dry sea salt particles (including hydrates), 2017 kg m −3 (Zieger et al., 2017). Following all corrections, the data of the DMPS and OPSS were combined at measured particle sizes of 0.35 μm (i.e., the DMPS data was used for D p < 0.35 μm and the OPSS data was used for D p ≥ 0.35 μm). Figure S4 in Supporting Information S1 depicts the overlap between the size distribution measured by the two instruments.
MATLAB software version 2019a was used to conduct statistical tests on the size distributions. The Mathworks Curve Fitting Toolbox 3.5.9, which uses the method of least squares when fitting data, was used to establish empirical relationships. All fits presented in this study are non-weighted.

Surface Bubble Spectra
The bubble size distribution at the water surface was determined photographically in a similar fashion to that described in Salter et al. (2014). Details can be found in the supplementary information. The bubble film radius measured at the surface was converted to the actual bubble radius following Toba (1959) in order to calculate the bubble volume.

Experimental Set-Up
The experiments were conducted with artificial seawater consisting of Sigma sea salt (Sigma Aldrich, S9883; with ionic mass ratios comparable to those in seawater: 55% chloride (Cl − ), 31% sodium (Na + ), 8% sulfate (SO 2− 4 ), 4% magnesium (Mg 2+ ), 1% potassium (K + ), 1% calcium (Ca 2+ ), <1% other) that was dissolved in low organic carbon standard deionized water (MilliQ, > 18.2 MΩ), hereafter referred to as "DIW." Previous studies have shown that the amount of organic material in this salt is negligible (e.g., Salter et al., 2016). In order to estimate the effect of salinity on sea salt production (i.e., on SSA production fluxes, particle, and bubble size distributions), S was decreased stepwise (35,30,25,20,15,10,8,6,5,4, 3, 2, 1, 0 g kg −1 ) while the temperature T was kept constant at 20°C. This experiment was repeated at a higher salinity resolution in the range 10-5 g kg −1 (35, 17, 10, 9, 8, 7, 6, 5.5, 4.5 g kg −1 ) where significant impacts of changing S on the aerosol and bubbles were expected. Each experiment started when the measured S had settled to a constant value. We chose to decrease S rather than increase it since dilution of 100 L of artificial seawater with DIW proceeds much more quickly to steady state than the dissolution of additional salt.
In addition to the experiments conducted at a constant temperature, three temperature ramps were conducted at salinities 35, 17, and 6 g kg −1 to investigate the impact of changing the water temperature on particle production at different salinities. A salinity of 35 g kg −1 was chosen because it is typical of the global oceans, has provided a benchmark in past discussions of salinity and is a generally accepted oceanographic standard (e.g., Millero et al., 2008). Salinities of 17 and 6 g kg −1 were chosen to represent lower salinity waters such as the Black Sea and the Baltic Sea. During these temperature ramps, the water temperature was slowly decreased from 30°C to 0°C over a period of approximately 40 hr. Two temperature ramps were conducted at S = 35 g kg −1 to examine the reproducibility of the measurements. A summary of the experiments is provided in Table S2 in the Supporting Information S1.

Impact of Seawater Salinity on Particle Production
By varying the amount of sea salt dissolved in DIW and measuring the particle size distribution in the headspace of our chamber we have probed the impact of changing S on particle production. Figure 1 presents the aerosol particle number, surface, and volume size distributions for seawater salinities between 0 and 35 g kg −1 . For seawater salinities ≥15 g kg −1 the number size distribution had a local maximum near 150 nm and a second local maximum between 500 and 600 nm when presented in the form dN/dlogD p where N is the number of particles (see also Figure S5 in Supporting Information S1). It is worth noting that the size distribution was nearly independent of S in this range. If we assume the drop sizes at formation to be equal, then lower salinities should result in particles with less mass and smaller diameters. That the particle diameters in the high salinity range are approximately the same implies that the droplet formation process differs with S in this salinity range.
As S decreased below 15 g kg −1 the first local maximum shifted to approximately 70 nm and the second local maximum ceased to exist. This shift toward smaller particles is likely induced by changes in the surface bubble spectrum, which we will discuss in more detail in Section 3.2. Another prominent feature of the aerosol size distributions was a substantial increase in the concentration of particles <100 nm at S = 6 and 8 g kg −1 relative to both higher and lower S. It should be noted that the particle concentration at S = 0 g kg −1 was slightly greater than zero and should be considered as a baseline for the experiments. When presented in the form dA/dlogD p (where A is the particle surface), the surface size distribution exhibited a mode that was centered around 5 μm at salinities ≥15 g kg −1 and shifted to 2 μm at salinities ≤10 g kg −1 . The magnitude of dA/dlogD p decreased notably at seawater salinities <15 g kg −1 . Following our particle loss-corrections it is difficult to ascertain the changes to the particle volume size distribution dV/dlogD p . Indeed, it would appear that our measured size range missed a fraction of the largest particles.
A comparison to previous studies conducted by Mårtensson et al. (2003), Park et al. (2014), Tyree et al. (2007), and Zábori et al. (2012) is provided in Figure 2. To facilitate comparison, we have normalized the number size distributions of each study to the respective total number concentrations (N total ). Mårtensson et al. (2003) observed a shift in the mode diameter from 60 to 125 nm as S increased. However, in their study, the shift to larger sizes occurred at higher S between 9.2 and 33 g kg −1 . Tyree et al. (2007) also observed a shift to larger particle sizes (mode center from 62.5 to 98 nm) as S was increased from 1 to 10 g kg −1 . Similarly, Zábori et al. (2012) observed a shift in the center of the mode from about 142 to 225 nm as S was increased from 0 to 3 g kg −1 to 12-15 g kg −1 . Our study supports the findings of the two latter studies. With that said, it should be noted that Zábori et al. (2012) observed that the mean RH was 43% during their experiments in the seawater salinity range 0-18 g kg −1 . Since sea salt effloresces at an RH below 43% the particle diameters reported by Zábori et al. (2012) may not be dry diameters shifting their particles toward larger diameters. While Tyree et al. (2007) used a frit to generate SSA, Zábori et al. (2012) used a plunging jet. In the current study, although we observed a shift in the main particle number mode diameter with changing S (see Figure 1), we did not observe a monotonic increase in the main particle number mode at higher S which has been observed in previous studies (e.g., May et al., 2016;Park et al., 2014;Tyree et al., 2007). Since two of these previous studies used frits to produce bubbles (Park  Tyree et al., 2007), the bubble spectra present during their experiments were very likely different from those in experiments utilizing plunging jets (such as the current study). May et al. (2016) used multiple plunging jets and the authors make note of the shallow bubble plume which resulted. This likely impacted the bubble lifetime in this system and potentially explains the difference in their observed particle distributions compared to the current study.
The particle size distributions presented in Figure 1 highlight that changing S induced complex changes in the particle size distribution following bubble bursting. Therefore, in order to visualize these changes, Figure 3 presents particle number, surface, and volume concentrations that were obtained by integrating over the whole size range of the combined size distributions measured by the DMPS and OPSS as functions of S.
From Figure 3a, three distinct regimes can be identified, where each regime describes a salinity range with a different relationship between S and integrated particle number. The first regime occurred at seawater salinities >10 g kg −1 , where the integrated particle number decreased only gradually with decreasing S. The second regime occurred in the salinity range 5-10 g kg −1 , where a peak in integrated particle number production occurred in the first set of experiments. Finally, a third regime was observed at seawater salinities <5 g kg −1 , where the integrated particle number decreased rapidly as S decreased. The integrated concentrations were compared to the total number concentration measured by the CPC (see Figure S6 in Supporting Information S1). The measured total number concentrations exhibited the same three regimes, although shifted downward in magnitude by an average of 17% ± 2% for all salinities except 0 g kg −1 .
In order to investigate the local maxima in particle production in the salinity range ∼5-10 g kg −1 more deeply, a second experiment was carried out with increased salinity resolution in this range. This experiment also showed a local maximum in particle production in this range. However, the exact salinity at which this local maximum in particle production occurred differed between the two experiments (6 g kg −1 for the first experiment, 4.5 g kg −1 for the second experiment). The relative error in number concentration between the two experiments was 8% ± 5% (maximum 13%). Although we suspect that an offset in the OPSS calibration and changes in the OPSS set-up may have caused these deviations (in the first experiment the OPSS was operated on an external pump while in the second experiment it was operated on the internal pump -although both pumps were set to the same flow of 5 L min −1 ), accounting for these differences did not reduce the differences in the integrated concentrations between the two experiments completely. This suggests that some properties of the seawater used in the two experiments differed. Since seawater temperature was controlled we can exclude this as a factor. Instead, it may have been that slight changes in the amount of organic matter present in the seawater impacted the bubble bursting process. Modini et al. (2013) reported a significant decrease in particle production efficiency when surfactants were added to NaCl solution. They further reported that the addition of surfactants stabilized the bubbles and thereby lead to an increased bubble persistence, allowing the bubbles to drain and thin, which in turn resulted in a significant reduction in film drop numbers. Although great care was taken to clean the experimental set-up as thoroughly as possible with ethanol, it is not possible to eliminate all sources of organic contamination that might be present in the experimental set-up. Zábori et al. (2012) observed a local maximum in particle number production in the salinity range 3-9 g kg −1 that is consistent with the increased particle production at salinities 6 and 8 g kg −1 observed in the current study. Note that the salinity bins in Zábori et al. (2012) are wider than the steps used in the current study. We suspect that the observed changes in particle number with S result from changes in the bubble spectra that are discussed in Section 3.2. From Figure 4 it becomes apparent that the changes in modal diameter and particle concentration with salinity observed from this study compare reasonably well to the findings for NaCl from Russell and Singh (2006), who fitted both the modal diameter and number concentration with a one-sixth power law (see their Table 2). It should be noted that the modal particle concentrations measured by Russell and Singh (2006) are several orders of magnitude higher. This difference can be attributed to the different experimental set-ups (i.e., they used a bubbler based on the design by Blanchard & Syzdek (1988) to produce bubbles, while we used a plunging jet).
Both particle surface and volume decreased monotonically as S decreased (Figures 3b and 3c). The strong dependence of both particle surface and volume concentration on S likely results from the lower ion concentrations and subsequent lower solute concentrations of the dried particles (assuming the initial droplet was the same size) as S decreased as previously hypothesized by Mårtensson et al. (2003) and Slade et al. (2010). Although Mårtensson et al. (2003) only measured the impact of S on particle production at three different salinities (0, 9, and 33 g kg −1 ), they also observed monotonically decreasing particle volume with decreasing S. Closer inspection of Figure 3c reveals two distinct salinity regimes. At salinities ≥10 g kg −1 the particle volume concentration decreased more rapidly as S decreased compared to salinities lower than 10 g kg −1 . While the particle volume concentrations from the two experiments agree very well in the low salinity range, they deviate substantially for S > 10 g kg −1 . We suspect that these differences are caused by the aforementioned differences in the operation of the OPSS (internal vs. external pump) that could possibly have resulted in losses of super-micron particles during the second experiment. As we could see from our loss estimations, the sampling efficiency of super-micron particles depends strongly on the density of the SSA which is in turn dependent on how quickly the particles are moved through the dryer. Even small changes in the number of large particles will have a substantial impact on the integrated particle surface and volume concentration and thus also on the effective radius. The relative error of the repeated salinity experiments were 10% ± 8% for surface concentration (maximum 22%) and 17% ± 15% for volume concentration (maximum 41%).  Russell and Singh (2006). Values from the current study are presented as mean and standard deviation.
In order to parameterize the effect of S on SSA production, we have attempted to describe the integrated particle number concentration as a function of S by defining the following empirical equation: where N denotes the particle number concentration integrated over the combined size range (0.015 < D p < 10 μm), k = 80 cm −3 and m = 438 cm −3 (the fits presented here are also included in Figure 3). As is clear from Figure 3a this empirical relationship does not account for the local maxima in particle production we observed at salinities between 5 and 10 g kg −1 despite being a reasonable fit to the data at other salinities (r 2 = 0.80, which describes the goodness of fit).
For salinities ≥10 g kg −1 the relationship between S and particle concentration in the headspace of the chamber can be described with the empirical linear equation: where = 2 cm −3 kgwater g −1 salt and m = 578 cm −3 for the first experiment (r 2 = 0.88). The second experiment has only three data points in this range, but indicates a linear relationship similar to that in the first experiment.
The particle surface area concentration showed a linear relationship with S: where = 47 μm 2 cm −3 kg water g −1 salt and m = −37 μm 2 cm −3 for the first experiment (r 2 = 0.98) and = 36 μm 2 cm −3 kg water g −1 salt and m = 22 μm 2 cm −3 for the second experiment (r 2 = 0.96). For the two distinct regimes observed in the particle volume, we have defined the following empirical relationships to describe the dependence of particle volume on S: The coefficients k and m are given in Table 1. It is worth noting that the slopes for the two experiments differ by a factor of around 2 at S > 10 g kg −1 . The difference in dV/dS above and below 10 g kg −1 suggests that two different particle formation processes may have dominated in these salinity ranges. The fit to V p is consistent with particle formation where the particle mass M p is proportional to the salinity of the artificial seawater (Mårtensson et al., 2003;Nilsson et al., 2021;Stavn et al., 2009):

Dependence of the Surface Bubble Spectra on Seawater Salinity
Figure 5a presents the bubble density in the field of view as a function of the bubble film radius for the measured range of salinities. At S ≥ 10 g kg −1 the bubble density peaked at a bubble radius of about 0.3-0.4 mm. As S decreased there was a shift toward larger bubble sizes and a decrease in bubble number density, particularly for bubbles <1 mm. The change in the bubble spectra as a function of S is more clearly demonstrated in Figure 5b, where we can see that the dependence of the integrated surface bubble volume on S exhibits similar behavior to the dependence of aerosol particle number on S (e.g., Figure 3a). That is, as S decreased from 35 to 10 g kg −1 the surface bubble volume decreased moderately. Then, as S decreased from 10 to 5 g kg −1 , a local maximum

and m With 95% Confidence Intervals (in Brackets) for the Empirical Relationship of the Aerosol Volume and Salinity S for Both Salinity Experiments (see Equation 4), as Well as the Coefficient of Determination, r 2
in bubble volume was observed. Finally, as S decreased below 5 g kg −1 a rapid decrease in the surface bubble volume was observed.
A similar decrease in the bubble density with decreasing S has been reported in several previous studies. A summary of these studies is given in Lewis and Schwartz (2004). Many of these investigators (e.g., Asher et al., 1997;Carey et al., 1993;Cartmill & Su, 1993;Marrucci & Nicodemo, 1967;Monahan & Zietlow, 1969;Slauenwhite & Johnson, 1999) reported a higher number of smaller bubbles with radii <1 mm in seawater than in freshwater which is consistent with our bubble size distribution. The higher abundance of smaller bubbles in seawater than in freshwater has been attributed to coalescence inhibition and an increased break-up of subsurface bubbles in seawater (e.g., Lessard & Zieminski, 1971;Lewis & Schwartz, 2004;Slauenwhite & Johnson, 1999). The inhibition of coalescence in seawater relative to freshwater has been attributed to the different electrolytic properties of fresh-vs. seawater. The transition in bubble coalescence has been observed to happen in the salinity range 0-10 g kg −1 (Craig et al., 1993;Lessard & Zieminski, 1971) which is in agreement with the observations in the current study. Notably, previous studies (Carey et al., 1993;Craig et al., 1993;Drogaris & Weiland, 1983;Lessard & Zieminski, 1971;Marrucci & Nicodemo, 1967) have observed that bubble coalescence is significantly reduced at ionic strengths in the range 0.1-0.2 mol kg −1 seawater which corresponds to salinities between 5 and 10 g kg −1 . This agrees well with the peak in particle concentrations and bubble volume observed in the current study. Inline with our observations, this suggests that in regions with salinities >10 g kg −1 SSA production is likely to differ only marginally from the major oceans.
If we had imaged the entire surface of the water generating particles during our experiments we could have combined our measurements of the surface bubble number with our particle measurements to provide an estimate of the impact of S on the number of particles produced per bubble. However, since our photographs cover only a fraction of the water surface, that is located close to the plunging jet, and the bubble density declines toward the edges of the sea spray tank to avoid wall effects on the bubble spectra (see Figure S2 in Supporting Information S1), it was not possible to estimate the total amount of bubbles at the water surface at any one time. As such, we were not able to provide such an estimate. Instead, we have normalized the concentration of particles with the number of bubbles in the field of view (Figure 5c) to provide an estimate of the rate of change of the concentration of particles produced per bubble with varying S. These estimates indicate that the normalized concentration of particles was approximately constant for S > 10 g kg −1 . In contrast, as S decreased below 10 g kg −1 the normalized concentration of particles increased rapidly with decreasing S reaching a maximum at S = 2 g kg −1 . The transition regime at salinities 5-10 g kg −1 that was so evident in both particle number (Figure 3a) and bubble volume (Figure 5b) also exhibits notably different behavior for this metric. One possible explanation for the occurrence of this transition regime is a rapid change in the surface bubble size distribution from the predominance of many small bubbles at higher salinities that tend to expel only very few droplets, to the predominance of fewer larger bubbles at lower salinities that break up into more numerous droplets.
The number of jet drops and film drops produced when a bubble burst depends on the bubble size (Lewis & Schwartz, 2004). Bubbles with radii >1 mm tend to produce more film drops while bubbles with radii <1 mm produce more jet drops (Lewis & Schwartz, 2004). As such, our observation that more sub-millimeter bubbles were present at salinities >10 g kg −1 likely coincided with an increased production of jet drops. Since we observed no notable changes in super-millimeter bubble density we conclude that film drop emissions were less impacted by changes in S than jet drops, which is in line with the observations of Harb and Foroutan (2019). Since jet drops tend to expel super-micron particles (Lewis & Schwartz, 2004), a higher proportion of jet drops will shift the particle size distribution mode to larger particles. This agrees well with the shift to larger particle sizes for salinities >10 g kg −1 that we observed. However, it is important to note that other studies (e.g., X. Wang et al., 2017) have observed that a substantial fraction of sub-micrometer particles can also be produced from jet drops. Figure 6 presents our bubble size spectra at S = 0 and 35 g kg −1 alongside those of Harb and Foroutan (2019), who used a plunging sheet of water to entrain air, and Salter et al. (2014), who used the same experimental set-up as that used in the current study. All of these experiments were conducted at temperatures of around 20°C. The bubble size distribution obtained in the current study at S = 35 g kg −1 agrees fairly well with the bubble size distribution measured by Harb and Foroutan (2019), whereas we observe considerably fewer bubbles at all sizes for S = 0 g kg −1 than Harb and Foroutan (2019). In the observations made by Salter et al. (2014) at S = 35 g kg −1 , fewer bubbles of all sizes were present than in the current study (with differences of up to two orders of magnitude for bubbles with radii <1 mm). This is surprising given that Salter et al. (2014) used the same experimental set-up used in the current study and even more surprising in that they observed slightly higher particle concentrations than those obtained in the current study suggesting that, if anything, more bubbles may have been present in their study. Their pictures were re-analyzed in the current study, leading to the same results. We can only speculate on the possible explanations for this. For example, the bubbles imaged by Salter et al. (2014) may have been slightly smaller than those of the current study and therefore outside the range of detection.

Temperature Dependence of Particle Production at Three Distinct Salinities
Temperature ramps were conducted at salinities of 35, 17, and 6 g kg −1 ( Figure S8 in Supporting Information S1). Since we have conducted two temperature ramps at S = 35 g kg −1 we were able to compare these two experiments for consistency. To do so, we have used the MATLAB two-sample Kolmogorov-Smirnov test (Massey, 1951) to compare the integrated particle number, surface and volume concentrations at each temperature of these two temperature ramps. It is worth noting that the integrated number concentrations for both experiments were almost identical for T ≥ 14°C (at a probability value of p = 0.99 and at a significance level of 5%), while they differed slightly at lower temperatures with a relative error of 10% ± 5%. The differences in surface and volume between the two experiments exceeded the standard deviations of the individual experiments with 12% ± 12% deviation for surface concentration (maximum 43%) and 19% ± 14% for volume concentration (maximum 48%), respectively. As such it is not surprising that the integrated concentrations were found to be different at probability values of p = 1.9 × 10 −6 and p = 8.9 × 10 −8 , respectively. We suspect that these differences result from very small changes in the experimental set-up that are beyond our control.
The change in sea salt particle number concentration with T (decreasing with increasing T) agrees qualitatively with many earlier studies that used a similar experimental set-up (e.g., Bowyer et al., 1990;Hultin et al., 2011;Mårtensson et al., 2003;Salter et al., 2014Salter et al., , 2015Zábori et al., 2012). We note that the difference between dN/dT for the 17 and 35 g kg −1 experiments is small, while the 6 g kg −1 experiment has a significantly smaller amplitude in dN/dT. Salter et al. (2014) concluded that it was changes to the bubble size distribution that were driving changes to particle production at T ∼ 10°C. In a similar fashion, the data generated during the current study suggest that it is changes to the bubble spectra, especially the ratio between bubbles with radii larger than 1 mm and bubbles with radii smaller than 1 mm, that are driving changes to the particle size and number concentration as salinity changes (see Figures 3 and 5c). However, the fact that the experiment at S = 6 g kg −1 exhibits the weakest trend in number concentration with T suggests that the temperature effect might be proportionally smaller for low salinity waters than it is for high salinity oceans.

Estimation of In Situ Emissions From Our Laboratory Sea Salt Simulation
In order to compare our laboratory sea salt simulation to pre-existing sea salt emission flux parameterizations and pre-existing in situ emission estimates based on eddy covariance aerosol fluxes, we have converted the number concentrations to particle production fluxes following Salter et al. (2015). The particle number per logarithmic size increment and second (that was obtained from the multiplication of the dN/dlogD with the sweep air flow Q sweep ) was divided by the volume of entrained air Q air (T) (m 3 s −1 ) that was determined as a function of temperature by Salter et al. (2014).
The particle production rate f (per m 3 ) was then multiplied by a wind speed dependent parameterization of the oceanic air entrainment flux following Long et al. (2011) in order to derive the size-resolved interfacial flux d dlog p (m −2 s −1 ): where ent = (2 ± 1) × 10 −8 ⋅ 3.41 10 is in units of m 3 m −2 s −1 and U 10 is the wind speed at 10 m above sea level in m s −1 . Figure 7 shows a comparison of our data at S = 35 g kg −1 and T = 20°C to the parameterizations of Mårtensson et al. (2003) and Salter et al. (2015) at U 10 = 9 m s −1 . Given that these two source parameterizations strictly apply to 33 and 35 g kg −1 salinity, respectively, they can only be compared to the high salinity experiments we conducted. While the production flux obtained from this study is only slightly higher than the parameterization of Salter et al. (2015), the parameterization of Mårtensson et al. (2003) yields fluxes that are roughly one order of magnitude higher than those obtained from our high salinity experiments. These large differences between experiments with similar laboratory set-ups underline the limited extent to which parameterizations derived from laboratory experiments can accurately represent a real word phenomena. For comparison to the low salinity experiments we conducted, we include SSA flux estimates based on ambient eddy covariance aerosol measurements over low salinity waters ∼8 g kg −1 (Nilsson et al., 2021). To achieve this comparison we assume that the volume of entrained air in our experiments depends on the water temperature according to Salter et al. (2014) and that it does not change across the salinities investigated in our experiments. As such, we assume that changing the salinity only impacts the subsequent break up of air cavities and bubble coalescence with impacts on the sub-surface bubble spectra. A large number of studies (Anguelova & Huq, 2018;Blenkinsopp & Chaplin, 2011;Chanson et al., 2006;Harb & Foroutan, 2019;Loewen et al., 1996;Q. Wang & Monahan, 1995;Wu, 2000) have conducted air entrainment experiments at different salinities and then registered the bubble cloud with photographic or acoustic methods from which the total air volume of the bubbles and the void fraction have been estimated. Of these studies, the majority support our assumption that the volume of entrained is independent of seawater salinity (Anguelova & Huq, 2018;Blenkinsopp & Chaplin, 2011;Loewen et al., 1996;Q. Wang & Monahan, 1995). However, several of the studies observed that the volume of entrained air varied across experiments with water of differing salinity (e.g., Chanson et al., 2006;Harb & Foroutan, 2019;Wu, 2000). Given the inconclusive nature of these results, our emission estimates should be regarded as preliminary until better data on the dependence of entrainment on salinity is available.

Particle Production Flux as a Function of Temperature and Salinity
We have derived a parameterization of the particle production flux as a function of T and S. A parameterization of the effective radius as a function of T and S can be found in Section S3 in the Supporting Information S1 (see also Figures S7 and S9 and Tables S3 and S4 in Supporting Information S1). The dependence of total particle production flux on T could be described with an equation building on the inverse tangent function (arctan) with correlation coefficients of r 2 ≥ 0.99 for all three salinities (see also Figure 8a): where F p is the particle production flux (in m −2 s −1 ) integrated over the whole size range (D min = 0.015 µm and D max = 10 μm in this study) and the coefficients a to d are given in Table 2.

Summary and Conclusions
In this study, we have performed a series of laboratory experiments to investigate the role of salinity on particle production and bubble spectra using a continuous plunging jet. We were able to describe the changes in particle number concentration and volume concentration with salinity and identified three distinct salinity regimes in the particle number production. For salinities between 10 and 35 g kg −1 , the particle number concentration was nearly constant. Between salinities 5 and 10 g kg −1 , we observed a local maximum in particle number concentration that coincided with a transition in the surface bubble spectra toward larger bubble sizes and decreased bubble density at lower salinities. Below 5 g kg −1 we observed a rapid decrease in particle number concentration, but only a gradual decrease in particle volume concentration. Above salinity 5 g kg −1 , this decrease in volume concentration  with decreasing salinity was considerably steeper. Furthermore, we have observed a shift of the particle mode centered at 150 to 70 nm and a decrease in particles >200 nm as the salinity was decreased below 15 g kg −1 . The observed shift to smaller particles with decreasing salinity is attributed to the linear relationship between salinity and particle volume. Changes in the particle production were further associated with changes in the surface bubble spectrum that exhibited higher numbers of bubbles with radii <1 mm at higher salinities, while the aerosol concentration normalized to the number of bubbles in the field of view peaked at salinities below 5 g kg −1 .
We see in the current study that a carefully repeated experiment with identical salinity (identical salt mixture), identical sea spray simulator set-up and cleaning and preparation procedures and identical sampling system and aerosol instrumentation revealed the two main features: A low and a high salinity regime with a near linear relationship between the salinity and particle production, and between a more unpredictable region with a peak in particle production. However, despite this, in the cases where we repeated the experiment at the same temperature, we observed that while the main features were reproduced, there were significant differences in magnitude between them (up to 40%-50%, particularly in the volume concentration). Although we cannot confirm the source of the variability between our experiments this suggests that the cycle of air entrainment, bubble formation, bubble bursting, and droplet formation is highly sensitive to very low amounts of surface-active contaminants and that this may account for the differences we observed.
Additionally, temperature ramps, ranging from 30°C, which is representative of tropical waters, to 0°C, representative of polar waters, were conducted at three distinct salinities (35 g kg −1 to represent the global average oceanic salinity, 17 g kg −1 , representative of lower salinity waters such as the Black Sea, and 6 g kg −1 , representative of water bodies such as the Baltic Sea). In terms of aerosol production, the experiments conducted at salinity 35 and 17 g kg −1 did not differ markedly and exhibited a similar dN/dT trend to Bowyer et al. (1990), Hultin et al. (2011), Mårtensson et al. (2003, Salter et al. (2014Salter et al. ( , 2015, and Zábori et al. (2012). In line with the findings of Salter et al. (2014), we attribute this temperature trend to a shift in the surface water bubble population with changing water temperature. The temperature ramp at S = 6 g kg −1 exhibited aerosol production that was qualitatively similar to the 17 and 35 g kg −1 ramps, but with a weaker amplitude (smaller dN/dT values). It may be that the two different processes, both mediated through changes in bubble population by either changes in temperature or salinity, interfere with each other for low salinity waters. Finally, we have derived a simple parameterization of the particle production flux as a function of salinity and water temperature based on the observed relationships that can be used to model the impact of salinity on sea salt production.

Data Availability Statement
The particle and surface bubble size distribution data from the salinity and temperature experiments conducted in this study are available at the Bolin Center for Climate Research Database via https://doi.org/10.17043/ zinke-2021-laboratory-3 (Zinke et al., 2022).