Two separate experiments were conducted during July 2014 utilizing MARTs. Detailed information on the performance and operation of MARTs can be found in Stokes et al. (2013) and only a brief description will be provided here. SSA particles were generated in an enclosed 210 L acrylic tank via an intermittent plunging sheet of water operated on a computer-controlled 4 s on/4 s off cycle to allow for surface foam evolution and dissipation close to what would be observed for natural whitecaps in the ocean. The plunging cycle creates a constant, reproducible concentration of nascent SSA particles in the 90 L headspace, with SSA particle size distributions produced in the MARTs generally consistent with those observed for SSA particles from lab-generated breaking waves (Stokes et al., 2013). SSA particles sampled from the MARTs are primary, since the average residence time in the MARTs is much shorter than the timescale required for secondary processing of SSA particles (e.g., heterogeneous gas-phase reactions) (Lee et al., 2015). SSA particles from the headspace were sampled periodically each day by instruments that characterized SSA particle size distributions, composition, and optical and hygroscopic properties. SSA were sampled from the MART headspace and transported through an approximately 2 m long line of 9.5 mm conductive tubing into a laminar flow manifold from which the instruments sampled. The flow rates to each instrument (or group of instruments) from ports in the manifold differed (e.g., 1 L min-1 to the high-resolution time-of-flight aerosol mass spectrometer (HR-ToF-AMS, henceforth AMS) vs. 3 L min-1 to the cavity ring-down spectrometer (CRD) + scanning electrical mobility sizer (SEMS)). Although the airstreams were not split isokinetically, the aspiration efficiency was calculated to be near 100 % for the size range of particles sampled here (Kulkarni et al., 2011). Thus, the different instrument groups sampled similar particle populations from the manifold. Due to the limited headspace volume and flow restrictions, not all instruments in this study could sample simultaneously. During individual sampling periods, only a subset of the full instrument suite sampled from the MARTs. The flow rate of zero air going into the MART and the flow being pulled from the MART by the instruments, as well as sampling times for each group of instruments, are provided in Table 1. The flow into the MART was always 1 L min-1 greater than the combined instrument pull to ensure positive pressure in the headspace, which eliminated possibility of sampling of room air. The shape of the measured particle size distributions was relatively independent of flow rate for the range of flow rates considered while sampling from the MARTs, although the residence time of individual particles decreases as the flow rate increases (Stokes et al., 2013). The air pushed into the MART was produced by a zero-air generator (Sabio Instruments, Model 1001), with air flow controlled by a mass flow controller. The excess flow was released through a vent on the MART. The different flow rates through the MART and through the sample tubing for each sampling configuration led to some differences in the size distribution sampled by the downstream instrumentation.

The procedure for inducing phytoplankton blooms inside of the MARTs will be briefly described here; further general details can be found in Lee et al. (2015). The MARTs were filled with ∼ 120 L of seawater each collected from the Scripps Institution of Oceanography (SIO) pier in La Jolla, California, USA (32∘51′56.8′′ N, 117∘15′38.48′′ W). Debris and zooplankton were filtered out of the seawater with 50 µm mesh. Phytoplankton growth was induced by exposure to artificial or natural light and the addition of growth media, which is described further in Lee et al. (2015). The two independent MART experiments will be referred to, according to the MART location during the growth phase, as either “indoor” or “outdoor”. The indoor MART was illuminated using 5700 K full spectrum lights, while the outdoor MART was illuminated with sunlight. A key difference between these two experiments is the intensity of the photosynthetically active radiation (PAR) during growth. The PAR was much greater for the outdoor MART compared to the indoor MART (PAR ∼ 1000–1500 µE m-2 s-1 (Bouvet et al., 2002) vs. ∼ 70 µE m-2 s-1), which likely resulted in a much larger peak Chl a concentration for the outdoor MART, 51 µg L-1 (outdoor) vs. 10 µg L-1 (indoor). An additional difference between the two microcosms was that the seawater was collected on different days: 8 July for the indoor MART and 19 July for the outdoor MART. The conditions of the seawater at time of collection are detailed in Table 2.

On 9 July, particles from the indoor MART were sampled immediately following nutrient addition. Further sampling was delayed until chlorophyll a (Chl a) concentrations exceeded approximately 12 µg L-1, which occurred 7 days after nutrient addition. (The same seawater as used in the indoor MART was added to a separate MART and sampled immediately after collection and before nutrient addition. However, the resulting particle size distribution from this MART differed substantially from those measured from the indoor MART, with a much greater contribution of large particles. Thus, the measurements from this separate MART are not directly comparable to the measurements from the indoor MART and are not considered further.) The outdoor MART was only sampled after Chl a concentrations exceeded approximately 12 µg L-1, which occurred 3 days after nutrient addition. This delay in sampling from when the water and nutrients were first added to the MARTs is necessary because the plunging process can lead to lysis of the phytoplankton cells during this vulnerable growth period, which will inhibit phytoplankton growth (Lee et al., 2015). Ultimately, SSA from the initially collected water from the indoor MART was sampled on 9 July, and subsequent sampling commenced periodically from 19 July through 31 July, i.e., beginning 11 days after the water was collected. Sampling from the outdoor MART did not commence until 22 July, continuing through 1 August. During the growth period and the off-sampling periods, air was gently bubbled through the tank to provide aeration. Sampling from the MARTs was performed daily once the threshold Chl a concentrations were reached.

A variety of online and offline measurements were made to characterize water composition and particles generated within the MARTs. A general sampling schematic is shown in Fig. 1 and a list of the instrumentation used is given in Table 1. As only a limited number of instruments were able to sample concurrently from the MART due to flow limitations, the individual sampling configurations (i.e., groupings of instruments sampling at the same time) are indicated; three specific instrument groupings are considered. The sampling times of each group relative to Group 1 are listed in Table 1 and a more detailed schematic of Group 1 optical and sizing measurements is provided in Fig. S1 in the Supplement. (The specific timing was dictated by the broader goals of IMPACTS.) A general description of the key instrumentation used as part of this study is provided below. Group 1 sampled for 1.5 h, Group 2 sampled for 2 h, and Group 3 sampled for 1 h each day that sampling was conducted.

The daily and study average merged size distributions for each MART are shown in Fig. 2a (indoor MART) and b (outdoor MART). The day-to-day variations in the size distributions were generally small. The average SSA particle number-weighted size distributions from both MARTs peaked around dp,m=100 nm and were relatively broad. The observed concentration of supermicron particles (dp,m > 1000 nm) was somewhat lower than that previously reported from a MART (Stokes et al., 2013) and likely reflects greater gravitational losses of supermicron particles in the long sampling line used here (Fig. S4). Since the hygroscopicity measurements discussed in this study are based on measurements made using polydisperse distributions, it is useful to determine the effective scattering-weighted particle diameters that characterize the MART size distributions. The study-average integrated scattering for each MART was calculated from Mie theory using the observed dry particle size distributions (Fig. 2c). The dp,m at which 50 % of the total scattering occurs were 570 nm for the outdoor MART and 530 nm for the indoor MART and particles with dp,m > 1000 nm contributed < 10 % of the total scattering in both MARTs, indicating that the derived GF(85 %) values for these two experiments are most sensitive to submicron particles with dp,m values between about 400 and 800 nm.

The extent of agreement between the observed bsca for dry particles and the values calculated from Mie theory using measured size distributions (Eq. 3) has been assessed (Fig. S5). The calculated bsca are ∼ 15 % lower than the observed bsca for both PMall and PM2.5, which is outside the combined uncertainty for the CRD and size distribution measurements (which is ∼ 11 % from error propagation). Some of the difference may result from differential losses between or within the sizing instruments and the CRD, although this seems generally unlikely to explain the entire difference, as losses of particles in the submicron range should be small. There is greater scatter in the PMall light scattering comparison than there is from the PM2.5 comparison, which likely results both from the APS measurements being made at a different time than the CRD and SEMS measurements and the need for dilution correction. Some of the difference between the observed and calculated bsca may be attributable to the assumption of spherical particles in the calculations, although similar closure was obtained (within 16 %) between observed and calculated bsca for atomized NaCl, suggesting that this is unlikely to explain the difference. It is possible that the diameters measured by the SEMS may have been too small. Wiedensohler et al. (2012) reported that sizing errors between instruments can be up to 10 %. If the measured diameters are increased by 8 %, then a 1:1 agreement between the measured and calculated extinction values is obtained. However, tests conducted during the study in which a second DMA was used to size-select monodisperse particles in the range 100–300 nm indicated agreement between the instruments to within 1 %. Additional tests after the study using 220 nm monodisperse polystyrene latex spheres demonstrated the SEMS sizing was good to better than 1 %, suggesting that sizing inaccuracies cannot explain the difference absent some fundamental problem with the data inversion procedure for size distributions (Lopez-Yglesias et al., 2014), which seems unlikely. Uncertainty in the assumed refractive index value for the dry particles may explain a small fraction (< 5 %) of the difference. Additionally, if the dry particles had retained some water in the CRD but not the SEMS, then the observed bsca would be larger than the calculated value. However, the RH in the CRD dry channel is much lower than the efflorescence RH for NaCl (∼ 45 %, Biskos et al., 2006), and thus it seems unlikely that residual water would have contributed substantially to the difference. Regardless of the explicit reason for the difference in calculated and observed absolute values of bsca, since the calculation of f(RH) depends on the ratio between the bsca for wet and dry particles, such absolute differences do not strongly affect the retrieval of GF(85 %) values. We have tested the sensitivity of the retrieval method to an 8 % increase in the particle diameters. The retrieved GF values are increased by a marginal amount (0.015–0.03) when the diameters are increased, and thus such potential sizing uncertainty does not affect the main conclusions presented here.

The temporal variation in Chl a concentrations, the derived GF(85 %), and various particle composition metrics are shown in Fig. 3 for the indoor MART. As has been previously observed in microcosm experiments, the measured Chl a time series exhibits a distinct peak (Lee et al., 2015), which in this case occurred on 16 July at a value of 10 µg L-1. This Chl a concentration is around the upper end of values observed for large phytoplankton blooms observed in the oceans, in particular near coastal regions (O'Reilly et al., 1998). After the peak the Chl a concentration dropped relatively quickly to around 1.5 µg L-1 (15 % of the peak) and then eventually to ∼ 1.4 µg L-1 (14 % of the peak). The DOC concentrations varied from 240 to 350 µM C, increasing rapidly when the Chl a concentration peaked and then staying relatively constant around 320 µM (Fig. S6A). The peak DOC range is somewhat larger than values typically observed for blooms in the ocean, which are only ∼ 130–250 µM C (Kirchman et al., 1991; Norrman et al., 1995). The temporal variation in heterotrophic bacteria concentration was similar to that for DOC, and heterotrophic bacteria concentrations ranged from ∼ 1 × 106 to 1.2 × 107 mL-1 (Fig. S6A), which is comparable to concentrations observed in the ocean (Li, 1998).

The GF(85 %) values determined for the indoor MART ranged from 1.79 to 1.9 and exhibited distinct temporal variations, decreasing from 1.88 ± 0.04 on 16 July, just as the Chl a peaked, to a minimum range of 1.79 ± 0.03 to 1.80 ± 0.01 from 17 July to 18 July when the Chl a concentration dropped to 3.41 ± 1.89 µg L-1 and then recovering back to 1.90 ± 0.03 on 7/20 (Fig. 3a). The range of these values is 10–15 % lower than the value of ∼ 2.1 for pure (inorganic) sea salt (Ming and Russell, 2001). There is a 1-day lag between the peak in Chl a and the (temporary) depression in GF(85 %).

The NR-OM volume fractions of SSA varied from 0.29 to 0.50 throughout the course of the indoor MART microcosm experiment (Fig. 3). The observation of such large εorg values is consistent with the substantial depressions in the GF(85 %) values relative to pure inorganic sea salt (2.1). The temporal variation in the εorg was generally similar to that of the GF(85 %) values, with smaller GF(85 %) values corresponding to larger εorg values, although the peak in εorg is somewhat sharper than the dip in the GF(85 %). The inverse relationship between the GF(85 %) and εorg is consistent with organic compounds being less hygroscopic than sea salt. The O : C ratio of the NR-OM had an average value of 0.25 ± 0.05 (1σ), which is similar to the value of 0.20 ± 0.08 reported by Frossard et al. (2014) for primary NR-OM that was generated from the open ocean using the “sea sweep” (Bates et al., 2012). The O : C ratio of NR-OM in the indoor MART generally increased with time, from 0.17 to 0.30, but also exhibited a temporary decrease on 17 July, the day when the GF(85 %) and εorg both first dropped. Since O : C often correlates with hygroscopicity for organics (at least for multi-component mixtures), this behavior may indicate a general increase in the hygroscopicity of the NR-OM with time (Cappa et al., 2011; Massoli et al., 2010). However, since the hygroscopicity of organic aerosol with O : C values in this range has generally been found to be small, the observed variations in O : C may not have a noticeable impact on the overall behavior of the GF(85 %) values.

The explicit co-variation of εorg and the GF(85 %) values is shown in Fig. 4. Assuming that volume mixing rules apply (i.e., the Zdanovskii–Stokes–Robinson (ZSR) mixing rules; Stokes and Robinson, 1966), the overall effective GF (GFobs) can be estimated as GFobs=(1-εorg)⋅GFseasalt3+εorg⋅GForg31/3, where GForg is the GF value for pure OM and GFseasalt is the GF value expected for pure sea salt. The ZSR line connecting GFseasalt(85 %) = 2.1 and GForg(85 %) = 1.0 provides the minimum value expected for any fraction of OM relative to sea salt. Low GForg(85 %) values (∼ 1.0) have been observed for fatty acids (Vesna et al., 2008), which have been found in SSA particles in the atmosphere (Mochida et al., 2002) and were observed in SSA produced in a related mesocosm experiment (Wang et al., 2015; Cochran et al., 2016). Values above this line indicate that the GF of the OM in the sampled SSA particles is, on average, greater than 1. Equation (5) was fit to the data shown in Fig. 4 to determine an average value for GForg(85 %) for the indoor MART. The best-fit GForg(85 %) was 1.16 ± 0.09, which falls between GForg=1.0 expected for non-hygroscopic OM, such as fatty acids, and GForg expected for more soluble OM, such as sugars (e.g., GForg(85 %) = 1.24 ± 0.04 has been observed for glucose; Mochida and Kawamura, 2004). (The uncertainty on GForg(85 %) is the fit uncertainty and does not account for uncertainties in either the CE for SSA particles or ρOM.) This value for GForg can be interpreted as an optically weighted average for the OM component of the SSA particles sampled here. Although the derived GForg values are reasonable, as the CE is relatively uncertain for sea spray particles (Ovadnevaite et al., 2012; Frossard et al., 2014) and the ρOM is also uncertain, the derived value for GForg should be viewed as suggestive only.

One important issue to consider in assessing the quantitative nature of the derived GForg(85 %) value is that the mass-weighted organic fractions used to determine εorg do not necessarily have the same weighting with respect to particle size as the GF(85 %) values do. The scattering-weighted median diameter, relevant to the GF(85 %) measurements, was dp,m=530 nm. The mass-weighted size distribution of OM can be estimated from the size distribution of a particular tracer ion in the AMS mass spectrum, specifically the ion at m/z = 43, which is indicative of total organic mass (Fig. S7). (Interference from the large signal contribution from inorganic ions prohibited explicit determination of the total NR-OM size distribution, and thus only the m/z = 43 ion is used.) The peak in the NR-OM mode occurs at dva ∼ 1 µm, which corresponds to dp,m=560 nm. This is comparable to the scattering-weighted median diameter, suggesting that the temporal variations in the GF(85 %) and εorg can be compared. However, the AMS aerodynamic lens imposes a 50 % cut diameter of ∼ 1 µm (Canagaratna et al., 2007) and thus the observed peak in m/z 43 likely occurs at a diameter that is somewhat too small compared to the actual distribution.

There was also a strong temporal correlation observed between the GF(85 %) values and the number fractions of ATOFMS mass spectra categories. Specifically, the decrease in GF(85 %) values after the bloom peak corresponded to a shift from particles that generated mass spectra dominated by sea salt ion markers (SS type) to particles with strong spectral signatures indicative of organic or biological influence (SSOC and Mg type). (The measurements made just after nutrient addition on 9 July are a notable exception, discussed below.) The inverse relationship between the GF(85 %) and the SS-type particle spectra is consistent with the inverse relationship between the GF(85 %) and εorg, providing additional confidence that the temporal variations in GF(85 %) are driven by variations in particle composition. The SSOC mass spectral type is identified in large part by the presence of carbon-containing peaks (CN- and CNO-) and K+ and Ca+ in the mass spectrum (see Fig. S2). The Mg-type spectra are characterized by their large Mg+ peak, which has been previously attributed to the presence of biological material (e.g., bacteria) (Guasco et al., 2013; Prather et al., 2013), as well as by smaller Ca+ and K+ peaks (Fig. S2). Both the SSOC and Mg spectral types are depleted in peaks corresponding to Na or NaCl. It is important to note, however, that dried SSA particles sampled by the ATOFMS can be spatially chemically heterogeneous, with shells depleted in Na and rich in Mg, K, and Ca (Ault et al., 2013). Thus, some fraction of the particles identified as having Mg- or SSOC-type spectra may be partially explained by the incomplete ionization of sea salt particles (Sultana et al., 2016). However, variations in the thickness of this Na-depleted shell likely reflect variations in the total particle organic content. Therefore, increases in the fraction of SSOC- or Mg-type mass spectra generated suggest a net increase in SSA particle organic content. On 9 July a substantial fraction of particles containing large Fe+ peaks were observed; on all other days the Fe spectral-type particle fraction was negligible. The Fe spectral-type particles may have had an influence on the observed GF(85 %) values when present with high relative abundance. However, their large fraction on 9 July is a result of nutrient addition and not biological changes in the seawater.

The temporal variation in Chl a concentrations, the derived GF(85 %) values and various particle composition metrics are shown in Fig. 5 for the outdoor MART. Like the indoor MART, the Chl a concentrations exhibited a characteristic rise and fall for the microcosm experiment. However, the maximum Chl a concentration was 51 µg L-1, which is 5 times higher than the indoor MART and likely due to greater PAR in the outdoor MART. Such high Chl a concentrations are well above those typically observed in the ocean. However, the Chl a concentration rapidly declined to 6 µg L-1 2 days after the peak and then continued to decrease over the next week to < 1.5 µg L-1. Both DOC and heterotrophic bacteria concentrations increased as the bloom progressed until they stabilized around the point when Chl a concentrations had returned approximately to their pre-bloom levels, with DOC concentrations ranging from 200 to 300 µM C and heterotrophic bacteria concentrations from 1 × 106 to a peak of 1.7 × 107 mL-1 (Fig. S6B).

The GF(85 %) values ranged from a maximum of 1.99 ± 0.03 to a minimum of 1.78 ± 0.04, again lower than what would be expected for pure sea salt (by 5–15 %). Unfortunately, no pre-bloom measurements were possible for this experiment, with the first particle measurements made for all instruments when the Chl a concentration was peaking. The smallest GF(85 %) values were observed towards the end of the microcosm, when the Chl a concentrations were at their lowest point (< 1.5 µg L-1). The GF(85 %) values exhibited two sequential decreases after the Chl a peak, the first after 3 days and the second after 6 days. The range of GF(85 %) values for the SSA particles from the outdoor MART was similar to that for the indoor MART.

The observed εorg values for the SSA particles were similarly comparable to the indoor MART (0.26 to 0.46). In general, decreases in GF(85 %) values corresponded to increases in εorg values, with the exception of the measurements on 7/30 (Fig. 5c). Following the peak in Chl a, the εorg values increase gradually over a few days, then stabilize, and finally exhibit an additional increase 6 days after the bloom peak (Fig. 5a). The temporal variations in εorg tracks neither the Chl a nor DOC concentrations. This suggests that, perhaps, biological processing and its impact on the composition of organics in the sea water, more so than absolute organic concentrations, is important for determining the abundance of organic matter transferred into SSA particles (Rinaldi et al., 2013; Lee et al., 2015; Quinn et al., 2014). However, further experiments will be needed to confirm this hypothesis.

Values of GForg(85 %) have again been estimated using the ZSR relationship (Eq. 5) and the observed GF(85 %) and εorg. Assuming GFseasalt (85 %) = 2.1, the derived GForg(85 %) = 1.23 ± 0.10 (Fig. 4). This GForg(85 %) for the outdoor MART is comparable to that obtained from the indoor MART, suggesting that the OM generated from the outdoor MART has similar hygroscopicity as the OM sampled from the indoor MART.

As with the indoor MART, there was a reasonable temporal correlation between the GF(85 %) values and the number concentration of SS spectral-type particles from the ATOFMS. The two most abundant non-SS particle mass spectra types were SSOC type and Mg type, with all other types contributing negligibly. This is again an indication that the temporal variation in the GF(85 %) values corresponds to an evolution of the composition of the SSA particle population and is consistent with the variation in εorg.

The Supplement consists of six figures that provide additional support of the conclusions presented in the paper.