Factors driving the seasonal and hourly variability of sea-spray aerosol number in the North Atlantic

N_SSA during late spring (NAAMES 2) was 3 times higher than during winter (NAAMES 1) and early spring (NAAMES 4) despite lower mean wind speeds (U₁₈) measured during NAAMES 2 (SI Appendix, Fig. S2 and Table S2). The higher N_SSA in late spring could be due to a contribution from SSA particles that may have a higher organic fraction (and less sea salt) than typical for SSA particles. This is supported by the higher c_p,660 (a measure of POC) as well as higher bacterial production rates measured during NAAMES 2 (Fig. 2), since the latter suggests enriched bioavailable dissolved organic carbon (DOC) concentrations in the surface microlayer. However, N_SSA did not correlate with DOC concentrations sampled in the upper 10 m (SI Appendix, Fig. S5), which may be because the small dynamic fraction of the bulk DOC is obfuscated by the large (refractory) DOC that constitutes the majority of bulk DOC (38). The higher N_SSA in late spring could also be due to other factors such as transport from productive coastal regions, precipitation effects on particle scavenging, whitecaps extent and bubble production, and differences in boundary layer mixing.

NAAMES measurements do show a strong and significant correlation between median d_m and median SST on seasonal timescales (R = 0.99, P < 0.05; SI Appendix, Fig. S6). This finding is supported by a strong negative correlation observed between the median scattering angstrom exponent (SAE, which is inversely proportional to particle size) and median SST on seasonal timescales (R = −0.98, P < 0.05; SI Appendix, Fig. S6). These findings represent observational evidence for a link between SST and d_m on seasonal timescales and support the hypothesis that SSA diameter is affected by the changes in film drainage time and bubble retention time prior to bursting due to alteration of physical seawater properties (i.e., viscosity and surface tension) (31, 39). Atmospheric chemical reactions could also increase SSA mode diameter, but photochemistry is expected to be more active in late spring than in early spring.

Correlations of median N_SSA and SSA mass concentrations with median local wind speeds were not significant for seasonal averages (SI Appendix, Fig. S6). The lack of seasonal dependence on wind speed may be due to the small dynamic range in seasonal winds, since the relative standard deviation (RSD) of 4 seasonally averaged local wind speeds was only 8.1%.

The variability in hourly local wind speed during each campaign was much larger than the 8.1% seasonal variability, with RSD values for wind speed on hourly timescales ranging from 32 to 48% (SI Appendix, Table S2). The larger variability of local wind speeds within each campaign provides data to assess the dependence of N_SSA on wind speed in the marine boundary layer.

N_SSA was moderately correlated with local wind speed (R = 0.5, P << 0.01) during all NAAMES (Fig. 3D), with N_SSA increasing from less than 5 cm⁻³ at low wind speeds (U₁₈ < 5 m s⁻¹) to more than 30 cm⁻³ at high wind speeds (U₁₈ > 20 m s⁻¹). These values for N_SSA fall within the range of published data from field observations (13, 14, 22, 35, 36).

The fact that N_SSA did not always track local coincident wind speed is not surprising since parameters such as upwind wind speed, precipitation, sea state, marine boundary layer mixing, atmospheric photochemistry, and SST also influence SSA number concentrations (1). However, the moderate correlation does show that wind speed explains 25% of the observed variability. An additional 6 to 25% of the observed variability between N_SSA and wind speed is explained by seasonal differences, with Pearson’s R improved from 0.50 (for all NAAMES) to 0.56 to 0.70 for individual campaigns (SI Appendix, Fig. S7). Separating the N_SSA versus wind speed measurements by campaign also reveals a lower increase in N_SSA with wind speeds for NAAMES 1 and 4 (with power law exponent below 1.5) compared to NAAMES 2 and 3 (with power law exponent above 1.9) (SI Appendix, Fig. S7). This observation provides further support that properties other than local wind speed significantly influence N_SSA.

The N_SSA parameterization from O’Dowd et al. (14) gives concentrations that are generally higher than the NAAMES N_SSA for wind speeds larger than 4 m s⁻¹ (Fig. 3D and SI Appendix, Fig. S8), in agreement with a previous study (12). There is no evidence that this discrepancy could be explained by a separate SSA mode in the accumulation size range (SI Appendix, Text S5), which may have important implications for climate models because several studies (9, 21, 40) have used the Gong (18) SSA production function optimized to the O’Dowd et al. (14) results. These models, therefore, use SSA number concentrations (hence CCN) that are likely higher than those measured in the western subarctic Atlantic.

SST and surfactants have been shown to influence SSA production both in the laboratory and in the field (16, 20, 22–26), but even the sign of the dependence of N_SSA on SST and surfactants remains uncertain (1, 11). At hourly timescales, NAAMES N_SSA are weakly correlated (R = 0.36, P << 0.01) to c_p,660 (a measure of POC) for 15-min measurements for NAAMES (Fig. 3E). The strength of this correlation between N_SSA and c_p,660 is increased (R = 0.51, P << 0.01) when measurements are limited to conditions of low wind speeds, although the data remain heteroskedastic (SI Appendix, Fig. S9). However, the link from POC concentrations to N_SSA is not expected since some POC components are too large to contribute to N_SSA. For POC to explain some N_SSA variability, we speculate that the POC correlation to N_SSA could mean that as POC is processed through foodweb processes that some fraction is transformed to a colloidal or dissolved form. Estimated heterotrophic bacterial production could provide a proxy for this production of submicron colloidal DOC, since bacteria can only take up low-molecular-weight organic compounds. The strong positive correlation between c_p,660 (a measure of POC) and bacterial production rates (R = 0.81, P << 0.01; SI Appendix, Fig. S5) provides initial support for this hypothesis, but the heteroskedastic distribution relating bacterial production rates to N_SSA indicates that other mechanisms should also be considered (R = 0.56, P < 0.05; SI Appendix, Text S10 and Fig. S5). Regardless of the processes that contribute organic carbon to bubbles and the microlayer (33, 41), both the additional labile DOC and possibly also smaller POC could form particles together with sea salt and also separately (42), potentially explaining the increased SSA number.

N_SSA was not correlated with SST (R = −0.2; Fig. 3F) although SSA mass concentrations were weakly correlated with SST (R = 0.30, P < <0.01; SI Appendix, Fig. S10), consistent with laboratory and field observations (11, 21, 28, 29). In addition, d_m increased linearly with SST (Fig. 3C) from 0.38 ± 0.17 µm for 0 °C < SST < 5 °C to 0.77 ± 0.19 µm for 20 °C < SST < 25 °C (R = 0.60, P << 0.01). The correlation of the 4-campaign dataset is equal to or higher than that of the 4 separate campaigns (Pearson’s R were 0.4 for NA1, 0.2 for NA2 and NA3, and 0.6 for NA4), consistent with the expectation that the larger dynamic range of SST in the 4-campaign dataset would provide a stronger correlation. Based on these results, we developed a new expression for the impacts of SST on d_m (Fig. 3C):which is a parameterization that could be introduced into future climate models. The measured increase in d_m with increased SST is consistent with laboratory experiments (16, 20, 29, 31). For example, the mean diameter of the SSA distributions increased monotonically with SST, from 0.08 µm at SST = 0 °C to 0.18 µm at SST = 25 °C for Mårtensson et al. (16) and from 0.8 µm at SST = 2 °C to 2.0 µm at SST = 30 °C for Salter et al. (29) (Fig. 3C).

SST, local wind speed, and biomass measured during NAAMES explain some of the variability in SSA number concentrations and diameter. The retrieved SSA size distributions are broadly consistent with SSA parameterizations from O’Dowd et al. (14) and Lewis and Schwartz (1) (Fig. 4). However, the O’Dowd et al. (14) parameterization (based on measurements in the northeastern Atlantic) is clearly lower than the NAAMES N_SSA for particles with diameter larger than 0.5 µm, at all wind regimes and for all seasons. These larger SSA particles constitute more than 90% of retrieved SSA mass, which contributes the most to the direct aerosol radiative effects under open ocean conditions. For example, at low wind speeds, N_SSA was a factor of 2 larger for NAAMES 1, 3, and 4 than O’Dowd et al. (14) predictions and a factor of 5.0 larger for NAAMES 2 (climax biomass) for particles larger than 0.5 µm. Similarly, at high wind speeds, N_SSA was a factor of 1.5 larger for NAAMES 1, 3, and 4 than O’Dowd et al. (14) predictions and a factor of 9.2 larger for NAAMES 2. These results imply that global climate models with SSA production functions optimized to O’Dowd et al. (14) data may underestimate SSA concentrations for diameters larger than 0.5 µm (hence direct radiative effects), but may overestimate overall SSA concentrations (hence CCN) by a factor of 4 to 7 (SI Appendix, Fig. S8) for conditions where the single (broad) modes used here are sufficient to represent SSA. The limited evidence found for additional small particles for NAAMES 1 (SI Appendix, Text S5) could be relevant for SSA particle number concentrations (but likely not CCN), although additional investigation would be needed to attribute them quantitatively to SSA.

At low local wind speeds, NAAMES N_SSA were within 5 to 20% of N_SSA predicted by the Lewis and Schwartz (1) SSA parameterization for NAAMES 1, 3, and 4 (although the size and width of the SSA particle modes differ, Fig. 4), but a factor of 3 larger than predicted values during NAAMES 2 (late spring biomass climax). At high wind speeds, N_SSA predicted by Lewis and Schwartz (1) is higher than NAAMES N_SSA by a factor of 1.7 to 1.8 for NAAMES 1, 3, and 4, but smaller than NAAMES N_SSA by a factor of 1.7 for NAAMES 2.

If the O’Dowd et al. (14) SSA parameterization is used to calculate the scattering from SSA particles [using Mie theory (43); SI Appendix, Text S11], then the resultant scattering coefficient (b_sca-450) is lower than measured scattering during NAAMES by a factor of 2 to 5 on average (SI Appendix, Fig. S11). In addition, Mie-calculated b_sca-450 using the O’Dowd et al. (14) parameterization explains only 11 to 40% of the variability in measured b_sca-450 (SI Appendix, Fig. S12). These comparisons illustrate that the representation of SSA size distributions by the bimodal O’Dowd et al. (14) parameterization biases the scattering and consequently the direct radiative forcing of SSA (44).

Effects of SST and bacterial production on SSA and their scattering have begun to be incorporated in global climate model studies (29, 45–48). Our work shows that the single-mode SSA particle distributions retrieved from ambient measurements that maximize correlations to local wind speeds provide a factor of 4 to 7 fewer N_SSA (hence CCN) than parameterizations based on the O’Dowd et al. (14) bimodal SSA size distribution. Furthermore increases in SST lead to larger SSA mode diameters and not accounting for these increased particle sizes makes the direct radiative effects of SSA too low by a factor of 2 to 5 for the subarctic western Atlantic region measured here. The weak dependence on the c_p,660 biomass indicator could also explain some additional variability in sea spray number concentration, but the complexity of the ocean ecosystem drivers means that a simple linear dependence does not represent this relationship. Longer duration measurements of open-ocean seawater and aerosol properties would be needed to evaluate ecosystem-based models of the surface ocean.