In remote, relatively pristine marine environments such as the Southern Ocean, the production of aerosols and clouds is predominantly governed by natural sources (McCoy et al., 2015). In order to represent these sources in Earth system models and project their response to climate change, the exchange of volatiles between the atmosphere and ocean requires rigorous constraint.

Dimethylsulfide (DMS) is derived from phytoplankton, and constitutes the largest natural source of non-sea-salt sulfate aerosol to the global troposphere of 10–20 nmol L-1 h-1 (Simó, 2001; Andreae and Crutzen, 1997), with an estimated annual input of 28.1 Tg S (Lana et al., 2011). Once in the atmosphere, DMS reacts to form sulfate aerosol, which acts as a source of cloud condensation nuclei (CCN). It has been hypothesized that DMS-derived aerosols may thus have a significant impact on the radiation budget (Charlson et al., 1987; Andreae and Crutzen, 1997; Ayers and Gillett, 2000), via direct scattering of sunlight and changes to cloud properties. However, more recent experiments highlight additional biogenic sources and pathways for the production of CCN, even in the absence of sulfate aerosol (Quinn and Bates, 2011; Bianchi et al., 2016; Kirkby et al., 2016). Current global flux estimates of DMS are poorly constrained, with estimates varying by as much as a factor of 2 (Lana et al., 2011).

Direct measurements of the air–sea exchange or flux (F) of a gas are challenging, and so F is often computed using an empirically determined gas transfer coefficient (k) and the air–sea concentration disequilibria (ΔC), according to the equation F=kΔC (Liss, 1983). The variability in flux estimates is widely considered to be driven by uncertainties in k (Zemmelink et al., 2004), which have been determined by a variety of methods including field observations using deliberately released tracers (Nightingale et al., 2000; Wanninkhof et al., 2004; Ho et al., 2011), wind and wave tank experiments (McGillis et al., 2000), global oceanic 14C uptake (Sweeney et al., 2007), and simultaneous measurements of waterside gas concentrations and air–sea flux (Huebert et al., 2004; Marandino et al., 2009; Bell et al., 2013). As gas exchange is primarily driven by shear-generated turbulence, k is often parameterized as a function of wind speed (Liss and Merlivat, 1986; Wanninkhof, 1992; Ho et al., 2006). However, gas fluxes are inadequately modelled by wind speed alone (Blomquist et al., 2006; Zemmelink et al., 2004), as other factors such as wave-breaking, sea state (e.g. Woolf, 2005; Asher et al., 1996), rain (e.g. Ho et al., 2000), and surface films (e.g. Schmidt and Schneider, 2011) also influence gas exchange at the sea surface. To enable prediction of gas fluxes for a range of compounds including DMS, the National Oceanographic and Atmospheric Administration (NOAA) Coupled-Ocean Atmospheric Response Experiment (COARE) model has been developed to incorporate many of the above factors. The model has been tuned to (Fairall et al., 2011) and validated against DMS eddy covariance field data (Blomquist et al., 2006; Yang et al., 2011).

The air–sea concentration disequilibria of DMS, and consequently air–sea exchange, are essentially controlled by the concentration in seawater ([DMS]) as atmospheric concentrations are typically at least 2 orders of magnitude lower. However, [DMS] is invariably measured at or below 2 m depth in both discrete and underway modes, and not at the sea-surface microlayer (SSM), the interface where gas exchange occurs. This assumes that there are no significant sources or sinks of DMS between the sample depth and the sea surface.

Dimethylsulfide concentration in the surface mixed layer is generally determined by the biomass, activity, and species composition of phytoplankton that produce dimethylsulfoniopropionate (DMSP), the precursor to DMS (Turner et al., 1988). Intracellular DMSP is regulated by factors such as nutrient availability and ultraviolet radiation dose (Archer et al., 2010; Toole and Siegel, 2004), whereas extracellular DMSP is influenced by grazing and bacterial processing (Yoch, 2002). To date, studies characterizing near-surface [DMS] have shown a decreasing gradient towards the interface, indicative of degassing to the atmosphere (Zemmelink et al., 2005). However, direct measurements of the air–sea flux of DMS by eddy covariance (EC) over coccolithophore-rich North Atlantic waters significantly exceeded those calculated from bulk seawater concentrations (Marandino et al., 2008). This discrepancy between predicted and observed fluxes was attributed to near-surface [DMS] gradients (above latitudes of 55∘ N; Marandino et al., 2008).

Despite the challenge of maintaining a DMS source in a relatively thin (10–100 µm) layer at the air–water interface that is often subject to extreme physicochemical conditions (Zuev et al., 2001), a number of studies have examined and identified enrichment of DMS in the SSM, as summarized in Fig. 1 and references therein. Sea-surface microlayer thickness, as defined by near-surface biogeochemical gradients, is of the order of 100 µm (Zhang et al., 2003). Given the challenges of sampling this thin surface layer, the thickness has been operationally defined as 1 mm by Liss and Duce (1997). In the current paper we evaluate properties for both 100 µm and 1 mm SSM thicknesses. The physicochemical and biological properties of the SSM are often distinct from underlying waters, and may support enhanced biogeochemical activity (Liss and Duce, 1997). For example, the SSM is often enriched with surface-active organic material and bacteria, and is subject to elevated ultraviolet radiation and temperature (Cunliffe et al., 2013). DMS measurements in the SSM have identified both enrichment and depletion relative to sub-surface seawater (SSSW) concentrations; however, enrichment has tended to dominate (Fig. 1). The source and controls of this excess DMS have not been identified, and the assumption that the SSM may influence DMS emissions into the atmosphere remains untested.

A variety of devices have been successfully deployed for sampling biological assemblages and dissolved compounds in the SSM (Cunliffe and Wurl, 2014). Trace gas SSM analyses are more challenging given the difficulties of sampling a volatile gas in a thin film that is subject to airside and waterside turbulence. Indeed, laboratory experiments have shown that a proportion of DMS is inevitably lost during SSM sampling, regardless of the device used (Yang et al., 2001). The aim of the current work was to test the potential for near-surface processes to influence air–sea DMS exchange using a novel combination of direct sampling of the SSM and SSSW, and EC measurement of air–sea DMS flux. Measurements were made during the Surface Ocean Aerosol Production (SOAP) voyage (Bell et al., 2015; Law et al., 2016). The influence of biogeochemical variability on spatial and temporal variation in near-surface DMS enrichment and flux was assessed by measurements in three phytoplankton blooms of differing community composition in productive frontal waters east of New Zealand. This location is currently under-represented in the global DMS database and climatology (Kettle and Andreae, 2000; Lana et al., 2011). In addition, the meteorological and physical factors influencing near-surface [DMS] were also examined in this assessment of DMS enrichment in the SSM, and its potential contribution to air–sea flux.

The results of two independent techniques to assess the potential contribution of the SSM to the air–sea exchange of DMS provide intriguing evidence that this may be significant under certain physical and biological conditions. This study adds to a number of other reports of DMS enrichment in the SSM (Fig. 1), but raises challenging questions regarding the source and maintenance of elevated DMS in the SSM. Consequently it is instructive to consider the validity of these results and the physical and biological factors that may influence DMS in the SSM.

Near-surface gradients in dissolved gases have been reported previously for DMS and carbon dioxide (Zemmelink et al., 2005; Calleja et al., 2005), with potential implications for air–sea flux estimates. The vertical DMS profile in near-surface waters in B2 and B3 was uniform (see Fig. 4), indicating that DMS production and loss terms, such as ventilation, bacterial oxidation, and photolysis, were in balance (Galí et al., 2013). Furthermore, the profiles do not show significant near-surface depletion in [DMS], which has been previously reported and attributed to ventilation and photolysis (Kieber et al., 1996).

The presence of significant DMS enrichment in the SSM at the B1 stations (Table 1) is surprising, as vertical diffusion from the SSM would be expected to elevate [DMS] immediately below the SSM. As elevated [DMS] was not apparent at 1–2 cm (Fig. 4), this suggests that density stratification and/or preferential retention of DMS in the SSM suppressed vertical diffusive losses from the SSM. Elevated [DMSSSM] has been previously reported relative to concentrations at 25 cm depth, associated with near-surface density gradients arising from ice melt in the Weddell Sea (Zemmelink et al., 2005). The near-surface temperature data in the current study indicated episodic formation of a gradient in the upper 4 m at the B1 stations (see Fig. 7) and, assuming this gradient extended to the sea surface, the resulting stratification may have created optimal conditions for SSM enrichment, with concentration and retention of phytoplankton whilst suppressing diffusive loss to sub-surface water. Furthermore, if surfactants were present they may have suppressed ventilation across the air–sea interface (Salter et al., 2011) under these conditions, leading to an accumulation of DMS in the SSM.

The sea-surface microlayer sampling, storage, and analysis may have introduced potential artefacts, particularly for trace gases. The mesh screen sampling produced higher [DMSSSM] than the plate, potentially due to preferential retention of algal and suspended material on the mesh as previously reported (Turner and Liss, 1985). These authors also reported significant DMS enrichment coincident with elevated sub-surface productivity, and partially attributed the enrichment to “stressing of SSM organisms as a result of the sampling procedure”. This may have occurred in the current study in B1, as dinoflagellates are sensitive to shear stress (Wolfe et al., 2002), but this was not tested. However, in contrast to other applications (Cunliffe and Wurl, 2014), we avoided scraping the SSM off the glass plate to reduce transfer of particulate material and ventilation of DMS, and this may also have reduced shear stress and exposure time of the phytoplankton. Exposure to air during SSM sampling enhances DMS evasion, with ∼ 50 % loss at zero wind speed (Zemmelink et al., 2005), which suggests that the majority of previous DMS measurements in the SSM (see Fig. 1) are underestimates (Zemmelink et al., 2006).

This raises the question as to how DMS enrichment is maintained in the SSM whilst ventilation is occurring across the air–sea interface. Zemmelink et al. (2006) calculated a DMS residence time in the SSM of the order of 40–60 s, and consequently a very high production rate would be required to maintain enrichment.

To maintain the observed and calculated enrichment in the SSM, DMS production must dominate over loss terms such as photolysis and bacterial oxidation and occur at a significantly greater rate than previously reported for the open ocean. Indeed, the PRSSM estimates in Table 1, which are determined indirectly from FEC and are independent of the SSM concentration measurements, significantly exceed reported DMS production rates for sub-surface waters (Simó, 2004). For example, in a compilation of 65 studies the maximum gross DMS production rates of 10–20 nmol L-1 h-1 (Simó, 2004) were up to 2 orders of magnitude lower than the calculated PRSSM based upon a 1000 µm SSM thickness.

Microorganisms in the SSM are exposed to extreme physicochemical conditions, including high irradiance (Zuev et al., 2001), whereas the DMS production rate estimates reported in Simó (2004) were from dark incubations that exclude the influence of light on DMS production. The conversion of intracellular DMSP to DMS is considered to be sensitive to both the quantity and spectra of light (Sunda et al., 2002; Archer et al., 2010), and so exposure to high irradiance in the SSM will have a significant influence on DMS production. This is supported by the “DMS summer paradox” where higher DMSP and DMS levels have been observed in shallow mixed layers that are exposed to high light levels (Simó and Pedrós-Alió, 1999). Laboratory and field experiments have also demonstrated that DMS has a positive, dose-dependent response to solar radiation (Galí et al., 2013; Sunda et al., 2002; Vallina et al., 2007). In particular, gross DMS production is stimulated by ultra-violet radiation (UVR), which causes a reduction in algal cell integrity and enhanced release of DMSP, DMS, and cleavage enzymes, and also up-regulation of intracellular DMSP cleavage (Galí et al., 2013). No relationship was observed between either [DMSSSM] or [DMS6.0m] with incident solar radiation in the current study, although this was confounded by differences in other factors such as phytoplankton biomass and community composition. The SSM was often sampled in the morning (08:00–09:30), which may suggest that the high DMS EFs in B1 may be a response to a night–day change in irradiance. Rapid changes in light can stimulate intracellular and dissolved DMSP production in coccolithophores (Darroch et al., 2015), with low-light cultures exposed to irradiance (including UVR) exhibiting an increase of 24–62 nmol L-1 h-1 DMS (Archer et al., 2010). These production rates are still 1–2 orders of magnitude lower than many of the calculated PRSSM for B1 (Table 1), but nevertheless confirm the potential for rapid DMS accumulation in response to increased light stress. Deck board incubations of SSM and SSSW from B2 and B3 stations showed that DMS production in the light was approximately double that in the dark (Cliff Law, personal communication, 2016), consistent with other reports (Galí et al., 2013). The highest net production rate of 3.7 nmol L-1 h-1 in the light (Cliff Law, personal communication, 2016) was again substantially lower than the calculated PRSSM in Table 1. Bacterial inhibition by high summertime UVR in the SSM (Zemmelink et al., 2006; Slezak et al., 2007) can decouple DMS production and consumption, with increased DMS observed in sub-surface waters (Vila-Costa et al., 2008). The coincidence of elevated phytoplankton DMS production, inhibition of bacterial DMSP consumption, and tolerance of bacterial DMSP degradation under elevated UV-A reported for BATS (Levine et al., 2012), provides further support for potential DMS accumulation in the SSM. However, the absence of a significant difference in DMSP cycling between light and dark incubations of SSSW during SOAP (Lizotte et al., 2016) suggests bacterial oxidation was not inhibited by light, although this was not measured in the SSM.

The different phytoplankton community composition of the three blooms may have influenced DMS enrichment in the SSM, particularly as all the blooms contained phytoplankton that are significant DMSP producers. B2 and B3 contained a higher proportion of coccolithophores but, despite evidence of their increased production of DMS and DMSP under high light stress (Archer et al., 2010), DMS levels were low in these two blooms. Conversely, B1 was dominated by dinoflagellates (> 50 % of the phytoplankton biomass) and [DMSSSSW] levels and SSM enrichment were significantly higher. Dinoflagellates are significant DMSP producers, with intracellular DMSP content and DMSP lyase activity that generally exceeds that reported for coccolithophores (Caruana and Malin, 2014). Enzymatic cleavage of DMSP is currently viewed as the primary process for DMS production, and dinoflagellates have been identified as capable of converting DMSP to DMS (Steinke et al., 2002, and references within). An association between elevated DMS and dinoflagellate biomass has been observed by a number of empirical studies (e.g. Zhang et al., 2014; Zindler et al., 2012).

Of the four dominant dinoflagellate species observed in B1, Gyrodinium has been reported in association with high DMSP concentrations in the field (see Table 1, Caruana and Malin, 2014). Some dinoflagellate species migrate to the surface during the day, which influences the vertical distribution of associated DMSP and DMS. For example, a 10-fold increase in [DMS] was recorded due to diel vertical migration of a dinoflagellate bloom in the St. Lawrence River (Merzouk et al., 2004). Analysis of phytoplankton community composition at the B1 stations showed only one dinoflagellate genus, Ceratium, which was more abundant at 1–2 cm relative to 2 m (data not shown), although this family does not generally exhibit high intracellular DMSP.

The EC data provide further evidence of a contribution of near-surface DMS production to air–sea flux, notably the close coincidence of significant EC- and SSM-derived residual [DMS] during B1 (Fig. 5c). The validity of this evidence is in part dependent upon generation of robust k values from the COARE model. Comparison with observational DMS data sets has confirmed that the COARE gas transfer model is a good predictor of k for DMS in most conditions (Blomquist et al., 2006; Yang et al., 2011), including the SOAP voyage (Fig. 5b). A discrepancy with COARE has been reported under high winds (> 11 m s-1) in the North Atlantic, with lower measured k values attributed to the suppression of turbulence due to wind–wave interaction, by Bell et al. (2013). In the current data analysis this suppression would result in a lower [DMSinf], in contrast to the elevated values observed. In addition, the largest deviations between [DMSinf] and [DMS6.0m] during B1 occurred at mid-range wind speeds (6–8 m s-1, Fig. 6a), where Bell et al. (2013) found good agreement with COARE. Consequently previous analysis does not indicate any significant bias in the COARE parameterization that could account for the high [DMSinf] during B1.

Spatial decoupling of airside and waterside measurements inevitably introduces error into the estimate of residual [DMS]. For example, Bell et al. (2015) identify a spatial offset between measurements of DMS flux and seawater DMS of up to 2 km during SOAP. However, this is unlikely to have generated the significant differences between [DMSinf] and [DMS6.0m] observed in B1, as these anomalies were observed when the ship was stationary or travelling slowly (< 2 knots), when wind speeds were < 10 m s-1 (see Fig. 5a). During these conditions, the flux footprint (Bell et al., 2015) would be much smaller. In addition, [DMSinf] exceeded 20 nmol L-1 on a number of occasions during B1, whereas [DMS6.0m] rarely exceeded 20 nmol L-1 throughout the entire voyage, suggesting that horizontal transport of DMS in the marine boundary layer from another bloom was not the source of the anomalously high [DMSinf] during B1.

Dimethylsulfide fluxes are traditionally computed using [DMS] at depths below the air–sea interface; consequently significant near-surface DMS has important implications for flux estimates. Sub-surface [DMS] between 1 and 160 cm depth was relatively uniform at all stations on the Chatham Rise, in contrast to suggestions that DMS concentration should decrease near the air–sea interface as a result of surface sinks (Kieber et al., 1996). Although near-surface DMS gradients were generally absent, a significant exception was recorded in a dinoflagellate bloom during light to mid-range wind speeds (i.e. < 10 m s-1) and near-surface temperature stratification. On several occasions in this bloom, significant enrichment of DMS in the SSM coincided with measured DMS fluxes that exceeded predicted fluxes calculated using sub-surface [DMS] and the COARE algorithm. Although SSM enrichment of DMS (see Table 1) and anomalously high air–sea DMS fluxes have previously been reported (e.g. Marandino et al., 2008, 2007), this study's results are the first to link these two phenomena.

There are some aspects of this data set that are surprising and require further investigation to establish the significance of the SSM to air–sea DMS flux. For example, the study raises questions as to how significant DMS enrichment is maintained in the SSM without influencing the [DMS] in the underlying water. In addition, the elevated SSM [DMS], both measured and inferred from flux measurements in the dinoflagellate bloom B1, necessitates a substantial in situ DMS production in the SSM. To maintain this enrichment, DMS production is required at a rate that significantly exceeds previous estimates for the open ocean (Simó, 2004). Nevertheless, the two independent approaches used in this study indicate that the SSM may influence DMS air–sea flux under certain biogeochemical and meteorological conditions, and so production at the air–sea interface may contribute to anomalously high DMS fluxes recorded in other regions of high productivity (Marandino et al., 2009, 2008).

The [DMS6.0m] dataset can be downloaded from http://saga.pmel.noaa.gov/dms/select.php.