Wave breaking and bubble production generally begin when the wind speed rises above 5–10 m s-1 (Banner and Peregrine, 1993; Gemmrich and Farmer, 1999), although they can appear at wind speeds as low as 3 m s-1 (Asher and Wanninkhof, 1998). A universal description of wave breaking and bubble production in the open ocean is still lacking. The primary motivation for further research is the need to improve parameterisations of air–sea gas transfer, particularly for carbon dioxide (Wanninkhof, 2014), but also oxygen (Chiba and Baschek, 2010; Atamanchuk et al., 2020) and aerosol production (de Leeuw et al., 2011) for use in weather and climate models. The sensitivity of model outputs to the details of the subsurface bubble physics is still a matter of debate. However, many authors have identified a more complete knowledge of bubble presence and subsurface bubble physics as a necessary step in order to refine the current generation of models (Goddijn-Murphy et al., 2016; Blomquist et al., 2017; Gantt and Meskhidze, 2013).

There is a range of proven methods for detecting subsurface bubbles in the open ocean, although there is no universal technique that can cover the full range of observed bubble sizes, spatial patterns, and timescales. Bubbles are highly compressible and so the most commonly used methods are acoustical, using either sonar and acoustical backscatter (Thorpe, 1982; Trevorrow, 2003; Wang et al., 2011), acoustical resonators (Farmer et al., 1998a; Czerski et al., 2011; Farmer et al., 1998b), or passive acoustics (Deane, 2012). In general, active acoustical methods are limited to smaller bubble sizes (below 1 mm radius) and lower void fractions (below 10-4). For larger bubbles and higher void fractions, specialised bubble cameras have been developed (Stokes and Deane, 1999; Leifer et al., 2003; Al-Lashi et al., 2018a). Other systems have also been trialled, including holographic detection (Talapatra et al., 2012) and optical scattering techniques (Randolph et al., 2014).

The current evidence suggests that the highest void fractions (>10-3), which are associated with an actively breaking wave, are only found within a few tens of centimetres of the surface (Deane, 2016) or approximately match the scale of significant wave height Hs (Anguelova and Huq, 2012; Callaghan et al., 2016), although there is limited direct evidence for this at sea in very high wind conditions. Once wave breaking and the consequent turbulence have formed the initial bubble population, the bubble size distribution evolves rapidly under the influences of buoyancy and probably also dissolution (Deane and Stokes, 2002), but no new bubbles are formed. The largest bubbles will rise to the surface within a few seconds, forming visible whitecaps or bursting. Smaller bubbles remain in the water column; are advected by turbulent water flow, convection or Langmuir circulation (Trevorrow, 2003; Thorpe et al., 2003); and may either dissolve completely or eventually rise back to the surface. There is no consensus on how much of the initial population returns to the surface, and how the subsurface residence time probability distribution varies with bubble size. For practical reasons, studies either focus on processes very close to the surface (usually laboratory studies), or the deeper plumes observed at sea using upward-looking sonar, but rarely both at the same time.

Many studies have shown that small bubbles can form long-lasting plumes many metres in depth and with void fractions from 10-8–10-5. Goddijn-Murphy et al. (2016) noted that the bubble size contributing most to the void fraction during active breaking is bigger than a millimetre, and Graham et al. (2004) found that the bubble radius contributing most to void fraction in the deeper plumes is 100 µm.

Observations of these plumes have been made in a range of open-ocean conditions (Vagle et al., 2010; Graham et al., 2004; Wang et al., 2011; Trevorrow, 2003), mostly using sonar data to record backscatter intensity. This technique gives good spatial information but cannot provide void fractions without additional information on bubble size distributions. Comparisons of these datasets can be difficult because the definition of the bubble plume “edge” depends on the sonar frequency and the thresholding techniques used, which are not always clearly stated. However, the measured plume depths have been found to vary with wind speed.

Bubble plumes are commonly described using two parameters: bubble penetration depth and an exponential decay constant that quantifies the decrease in bubble presence with distance from the surface (usually the only available measure is acoustical backscatter, so the decay constant actually refers to the decrease in scattering strength with depth). Reported plume depths vary from 5–25 m. Wang et al. (2011) made measurements in the most extreme conditions in the literature, with wind speeds up to 50 m s-1. They observed that the bubble plume depth increased linearly for wind speeds up to 35 m s-1 (where the depth had reached 20 m) but increased more slowly after that, reaching approximately 25 m at 50 m s-1. The exponential decay constant was found to be 0.8 m by Thorpe (1982), 0.7–1.5 by Crawford and Farmer (1987), and 0.5–3 m and strongly related to the plume depth by Trevorrow (2003). Graham et al. (2004) observed that e-folding depth for acoustical backscatter was related to wind speed, varying from 0.2 m for very low winds (6 m s-1) to 1 m for higher winds (12 m s-1) in 10 m deep water. To our knowledge, the studies by Vagle et al. (2012, 2010) are the only ones to use bubble size distribution measurements instead of backscatter as a basis for a parameterisation, producing bubble-size-distribution-dependent estimates of bubble-mediated fluxes of O2 and N2. There is a huge amount of detail in many of these papers, but the difficulties in adequately describing all relevant parameters and in making backscatter comparisons mean that no universally applicable description of these deep bubble plumes is agreed on. A full review of deep bubble plume measurements is beyond the scope of this paper, but an excellent review of the literature on this topic before 2004 can be found in Graham et al. (2004). Two important conclusions are that the penetration depth is most strongly related to wind speed, and that most bubbles come from shorter and steeper waves. A notable recent modelling study by Liang et al. (2012) suggests that the e-folding depth scales with friction velocity, and that wave age has a significant effect on plume behaviour.

Considerable attention has been paid to bubble source functions, but there are far fewer studies on bubble sinks. Although bubble dissolution is routinely and accurately modelled for bubbles coated with surfactant monolayers in the lab environment (Azmin et al., 2012; Lozano and Longo, 2009), bubble dissolution in the ocean is significantly more challenging. This is because bubbles are stabilised by a complex mixture of surfactants (Frew et al., 1990; Wurl et al., 2011; Zhou et al., 1998) and are likely to be stabilised still further by particulates and gel-like materials on their surface. In one of the very few explicit studies on this topic, Johnson and Wangersky (1987) showed that bubbles stabilised by a combination of surfactant and particulates would maintain their size for many hours, although a relatively small pressure increase could cause them to collapse very rapidly. It has generally been assumed that bubbles in the ocean dissolve gradually, but we are not aware of any direct evidence for this. Characterising the mechanism of bubble disappearance in the ocean is a necessary step towards understanding fluxes, and a considerable gap in current knowledge.

The surface ocean is a complex environment with many features that are expected to have a second-order influence on bubble production and their path through the water column. Some of these are the environmental conditions that affect breaking: the presence of swell, relative orientation of wind and waves, whether winds are rising or falling, and the surface heat flux. Relevant water parameters such as temperature and surfactant presence must also be considered.

It is widely accepted that the existence of either soluble or insoluble natural surfactants in ocean water is likely to influence bubble production, bubble residence time, and the contribution of bubbles to air–sea gas transfer and aerosol production. Laboratory studies have made some progress on these questions (Callaghan et al., 2013, 2017; Anguelova and Huq, 2012) but are hampered by the difficulty of conducting experiments with realistic natural surfactants. Characterising surfactants in the natural environment is a significant challenge.

Water temperature is expected to influence bubble processes at sea, but as yet there is no clear consensus on how. Salter et al. (2014) and Slauenwhite and Johnson (1999) observed a significant increase in bubble production at lower temperatures in laboratory experiments, and Salisbury et al. (2013) saw whitecap fraction decrease slightly as temperature increased in ocean satellite data. However, Callaghan et al. (2014) saw a 6 % increase in air entrainment in wave tank experiments as water temperature was raised from 5 to 30 ∘C.

Other parameters reported to affect bubble formation, movement, and lifetime include phytoplankton presence (Kuhnhenn-Dauben et al., 2008), surface heat flux, upper ocean stratification (Vagle et al., 2012), whether winds are rising or falling (Hwang et al., 2016), and wind-wave Reynolds number (Salisbury et al., 2013; Toba et al., 2006; Scanlon and Ward, 2016; Brumer et al., 2017a).

This wide range of potential influences and lack of field data complicate the current efforts to improve parameterisations of ocean processes, a necessary step for improving weather and climate models, and to predict the biogeochemical consequences of our changing climate (Talley et al., 2016). For example, carbon and oxygen uptake in the North Atlantic have been found to be highly variable (Watson et al., 2009; Woosley et al., 2016), but the complexities are not yet understood. Improved parameterisations of carbon dioxide and oxygen flux across the ocean surface are needed to forecast the future uptake of these gases by the ocean. Bubble-mediated transfer is known to make a significant but poorly parameterised contribution in both cases (Goddijn-Murphy et al., 2016; Atamanchuk et al., 2020), and a better mechanistic understanding has been identified as critical for future improvements.

All the data presented here were collected during the HiWinGS expedition in 2013. This paper will give an overview of HiWinGS and the measurement methods used, describe the observed bubble void fractions during stormy conditions, and present evidence on the mechanisms of deep bubble plume formation. We will not address air–sea gas transfer directly but instead focus on bubble processes. The companion paper (Czerski et al., 2022) will present a separate analysis of the observed bubble size distributions and the implications for bubble advection and destruction processes, along with a summary of mechanistic understanding.

The overall aim of HiWinGS was to improve understanding of turbulent air–sea gas exchange processes in high wind conditions by making direct measurements of the fluxes of trace gases and physical parameters, combined with sea state, wave, and bubble physics (Brumer et al., 2017b; Kim et al., 2017; Blomquist et al., 2017; Yang et al., 2014).

The HiWinGS cruise took place between 9 October and 14 November 2013 on the R/V Knorr in the North Atlantic Ocean to the south of Greenland, a region known as a significant sink for CO2 and a period chosen to maximise the number and severity of the storms encountered. An overview of the expedition and the instruments deployed can be found in Blomquist et al. (2017) along with the major gas transfer results. Figure 1 shows the time series of the wind measurements over the whole expedition, with periods when the buoy carrying bubble sensors was in the water highlighted.

The bubble measurements used here were made during four buoy deployments: 18–21 October (station 3), 24–27 October (station 4), 1–4 November (station 6), and 7–10 November (station 7). The station numbers correspond to those in the Blomquist et al. (2017) paper. A summary of the conditions during each deployment is given in Table 1.

The bubble instruments were all mounted on an 11 m long free-floating spar buoy. This was deployed in advance of each storm, left to drift freely, and recovered after the storm. The buoy carried a bubble camera, acoustical resonators, an upward looking sonar, wave wires, a downward-looking foam camera, an inertial motion measurement unit, and an acoustic Doppler velocimeter (ADV). Figure 2 shows the position of the instruments on the buoy. A description of the buoy and its performance can be found in Pascal et al. (2011).

The two most significant characteristics of the buoy design relate to its buoyancy and its ability to orient into the wind. The buoy is split into two main sections: the main hull, a cylinder 25 cm in diameter and 6 m long which remained fully submerged at all times, and the top section, a cylinder 10 cm wide and 4 m long, which protruded through the sea surface. Above the top section, a watertight dome carried the inertial measurement unit, additional small cameras, and a specialised foam camera. The base is formed of two hexagonal damping plates which carried four 24V Deepsea Light & Power batteries with a combined weight of 80 kg, providing significant additional stability and reducing the natural vertical oscillation frequency of the buoy to a period of approximately 8 s. Small waves pass by without causing significant vertical motion, while the buoy rides over the larger swell. A lead ballast weight was suspended off-centre from the buoy base, forcing the buoy to sit in the water at an angle of approximately 8∘ to the vertical. Previous tests have shown that the hanging ballast has the effect of orienting the buoy into the wind (with the ballast on the downwind side), tracking the wind direction to within a few degrees (Pascal et al., 2011). The buoy was designed to keep the wave wires on the upwind side of the buoy, and all instruments except the ADV were also mounted on the upwind side. The ADV data show that the wind forcing on the dome pushed the buoy downwind at a speed comparable to the near-surface Stokes drift, producing a complex flow profile relative to the buoy. A full description is given in Appendix A. At the depths of the camera and resonator, the most likely situation is that both instruments were measuring in water which had flowed around the buoy from the downwind direction. However, given the constant vertical motion of the buoy and the turbulence in the surrounding water, we are confident that both instruments were making measurements that were representative of the bulk water around them.

The simplest measure of bubble presence at a single point is void fraction, the ratio of air to water volume. Figure 3 shows the void fraction measured simultaneously by bubble camera and resonator, and compared with sonar backscatter for a 45 min period on 2 November, during wind speeds of 18 m s-1. The patterns in time match very well and provide confidence that all systems were working as intended. Measured void fractions from the camera and resonator are shown on both linear and logarithmic scales to emphasise the point that a measurable bubble void fraction was present at 2 m depth at all times, even though the linear plots show distinct peaks. At 4 m, the void fraction is below the detection limit (10-8) of the resonator almost all the time. This suggests that the sonar used may also have had a detection limit of approximately 10-8, or possibly that bubbles were entirely absent at 4 m between the distinct plumes. We note that the bubble size ranges sampled by the resonator and camera are very different (radii of 6–173 µm and 20 µm–10 mm respectively). Extrapolating the observed bubble size distribution to 1 µm in radius suggests that the maximum possible contribution to the total volume from bubbles below 20 µm is extremely small (significantly less than 1 %), while the small number of bubbles larger than ∼ 100 µm reaching the depth of the resonator make a negligible contribution to the total void fraction; a direct comparison between the two instruments is thus appropriate.

We observe that although the major features on all three timelines match, plumes are visible on the sonar at times when there are very low void fractions at 2 and 4 m depth. This is a consistent feature and shows the limitations of sonar data in isolation. Significant backscatter can be caused by small resonant bubbles, but this does not necessarily indicate an equally significant void fraction in those plumes. We note that the linear scale on the y axis exacerbates this effect on this plot, but it highlights the care needed not to over-interpret bubble plumes observed by sonar alone. The observed profile of sonar backscatter is comparable with that seen by Vagle et al. (2010) when the different backscatter thresholds of their data are taken into account. It is also difficult to make a direct quantitative comparison between the sonar backscatter at 2 m and the measured camera void fraction. The plume patterns agree well, but the regions of highest void fraction (the centres of big plumes) may be acoustically shielded (Deane, 2016) by the smaller bubbles around them, producing a low sonar backscatter signal for some high void fraction regions.

Figure 4 shows 1 min averages of void fraction at 2 and 4 m for station 6 (1–4 November) for periods when the value at 4 m was above the noise level. For most intervals there is no strong correlation between bubble presence at 2 and at 4 m. Some of this may be due to plume shearing (see Sect. 3.3), but this does not explain all of the observed variation. We note that this seems to be a general feature and that although large identifiable plume structures exist that are connected vertically, they are overlaid on a significant general background level of bubbles which is highly heterogeneous and shows no strong vertical connection. Where the void fractions are correlated at both depths, the ratio implies an e-folding distance of 0.5 m (see Sect. 3.3.3 for further analysis of e-folding depths).

The buoy dome carried a foam camera at 2 m height which recorded the upwind ocean surface and was designed to observe the details of waves breaking directly in front of the buoy (Al-Lashi et al., 2018b) for comparison with subsurface measurements. Although it had a fisheye lens, the effective field of view (3–4 m upwind of the buoy) was insufficient to collect breaking wave statistics. It was positioned to look for direct correlation between surface-breaking waves and the bubbles immediately beneath them. Very few (<10) active breaking waves were seen in its field of view during the measurement periods, partly because data were only captured during daylight hours, but also because the short duration of active breaking limits the probability of an event falling within the field of view. None of the observed breaking waves were directly correlated with bubble presence at 4 m depth, which is consistent with deep plume formation being independent of wave breaking. Figure 5 shows one example where an active whitecap was observed by the foam camera and appeared to precede the formation of a deeper bubble plume.

Figure 5 shows that the observed breaking event generated a shallow plume of bubbles within the top 2 m. After about 40 s, during which time a few waves have passed, this shallow plume appears to join or form a deeper plume. The buoy was drifting downwind throughout this period. Our interpretation is that the buoy and the shallow plume drifted into a convergence region between two Langmuir cells, where bubbles have already accumulated, and so intercepted an existing deep plume. Figure 6 shows the corresponding water velocity data from the ADV at 3.8 m. The dashed box represents the time period covered by Fig. 5, with a few minutes either side shown for context. There is a pronounced sideways flow feature as the deeper plume appears, followed by a period of flow in the opposite direction, which would be consistent with Langmuir cells at an angle to the wind. The vertical flow is almost all downward throughout this period, indicating a convergence zone. It seems likely that the breaking wave was independent of the Langmuir cell. The foam patch at the surface was observed to move slowly downwind relative to the buoy. Small long-lasting bubbles from this breaking wave are likely to have remained in the convergence zone, contributing to the persistent plume there. The combination of the ADV and sonar data provide convincing evidence that the deep plumes are due to the convergence of Langmuir cells and are not directly connected to the breaking process. We also note that Fig. 6 also shows a 2–3 min period of sustained downward flow around 18:30 UTC, which can be compared to the features in Fig. 3 at the same time. There is an increased void fraction at both 2 and 4 m at 18:30 UTC, although it is an order of magnitude lower at 2 m than the period between 18:10 UTC and 18:15 UTC. It seems likely that the relative lack of bubbles during the downward flow at 18:30 UTC is due to a lack of bubbles in the surface layer, demonstrating that the bubble distribution in the surface layer is highly heterogeneous.

Void fraction data lack the detail of bubble size distributions but provide a straightforward measure to follow plume evolution over time, and to analyse overall patterns of bubble presence and absence. At 2 m depth, bubble presence fell below measurable levels (a void fraction of 10-8) for only 4 % of the total data acquisition time. As shown in Fig. 3, structure is visible on the log scale at all times at the camera depth, and consequently categorising individual plumes is a difficult task and labels implying plume presence or absence should be treated with care.

All the bubble camera data discussed below are an amalgamation from all four deployments, just over 23 h of data. The resonator data cover a longer period (52 h over three deployments), but the void fraction at 4 m only rose above the resonator noise level of 10-8 in 9 % of all resonator measurements.

There are a total of 8 h when both the camera and resonator data were of high quality and measuring simultaneously, during wind speeds between 14 and 21 m s-1. In that time, there were 30 identifiable plume-like features measured at 4 m (several were close to the noise level, so the exact number is very sensitive to the threshold chosen), lasting between 30 s and 6 min. A total of 20 of these matched up directly with plumes at 2 m; i.e. they were closely related in time and had the same shape. It is striking that the peaks at 4 m consistently lag those at 2 m by between 0 and 66 s (equivalent to a 0–9 m horizontal offset, given the buoy drift speed). Figure 10 shows an example of a typical event and the measured offset. The void fraction pattern is not symmetric in time – there is a sharp incoming rise and a slow tail-off at both depths.

All of the measured offsets bar one were positive (so bubbles were observed at 2 m before 4 m as the buoy drifted downwind), which suggests either that the plume shapes were consistently tilted towards the wind during the period of these measurements, or that plumes always become narrower with depth.

In general, the plumes seen in the sonar data are not symmetrical and are often highly irregular in shape. Apart from the offsets, there is no consistent discernible skew in plume shapes. Symmetrical triangular plumes are suggested by Zedel and Farmer (1991), but they process their data using the assumption that the plume is symmetric.

Figure 10c compares the horizontal offset with the ratio between the void fractions at 2 and 4 m. It is notable that there are no cases with both a very high void fraction ratio and a high horizontal offset. There are two possible interpretations of the void fraction ratio between the two depths. The first is that when bubbles are advected downwards to form a plume, the initial ratio is 1 : 1, and then bubbles are destroyed more quickly at depth. In this case, older plumes would have a higher ratio, and the ratio is driven by mechanisms acting at 4 m. The second is that only a small proportion of bubbles are ever advected downwards, and that therefore high ratios are driven by intense bubble plumes near the surface which dissipate over time. In this case, the ratio will decrease as the plume ages. Our observations are that the void fractions at 4 m only ever cover a narrow range (as shown in Fig. 8), which points towards the second case. With this interpretation, Fig. 10c shows that the plumes with the largest horizontal offsets seem to be older.

Figure 10d shows the wind speed at the time each horizontal offset was measured. The largest offsets were seen during periods of the highest winds. This leads to one possible explanation: that the size of the Langmuir cells increases with wind speed, and the offset is due to advection patterns that scale with the cell.

The offsets could also result from the shear associated with Stokes drift or other near-surface currents. The orbital motion associated with a passing wave is almost circular but also includes a small forward drift which is highest at the surface and decreases with depth. However, to explain the direction of the consistent offset, this drift would have to be dominated by the swell moving in the upwind direction rather than wind sea and wind-driven currents in the downwind direction. It is not possible to be sure about the surface flows in this case; however, it is thought that the Stokes drift from wind sea will dominate that from the opposing swell at the surface but not at depth (Webb and Fox-Kemper, 2015). Figure 10d is hard to reconcile with Stokes drift in the upwind direction due to the opposing swell. There is no discernible pattern between the swell significant wave height and the horizontal offset (not shown), although there were two major swells present at that time, as discussed in Appendix A. We lack the data to provide a clear explanation for the consistent plume offset, given the uncertainty about the current profile in the top few metres in these conditions. The offset matches what might be expected if the Stokes drift from swell at this depth dominated the wind sea Stokes drift during this period, but the best available flow profile estimates from the literature suggest that this is unlikely. However, the offsets were consistent and require further investigation.

A better understanding of the horizontal flow profiles could lead to a method for estimating the time since a plume formed. An assumption about the shape of the plume at its moment of “formation” would be needed, the simplest being that the plume is vertically aligned. If an estimate of the shear caused by near-surface currents was available, it would be possible to calculate the time needed for the observed horizontal separation to be generated, and therefore the “age” of the deep plume. As discussed further below, this is not likely to be the time since the formation of the individual bubbles, but the time since the collection of bubbles was advected downward from the shallow bubble layer to form a deep plume. There is considerable uncertainty associated with applying this method here, but estimates using our data imply plume ages of the order of 10 min. We note that bubble plumes were observed to shear by Crawford and Farmer (Crawford and Farmer, 1987), who also suggested that the shape of a sheared plume might give an indication of age but did not attempt age estimates.

Sideways shear over long time periods may also explain the existence of plumes at 4 m that are not obviously connected to 2 m plumes. As the wind direction and speed change, the spatial distribution of plumes will constantly change as the local horizontal flow profile moves bubble plumes around. Changes to Langmuir circulation patterns have been observed to occur over tens of minutes (Smith, 1992; Farmer and Li, 1994).

The model of plume development which is consistent with our data is as follows:

Breaking waves cause the formation of shallow bubble plumes, confined to the upper metre or so of the water column. The continual injection of bubbles into this layer produces a near-permanent bubble population with a probability distribution that depends on wind conditions and a structure that depends on both Langmuir circulation and near-surface shear currents. The maximum void fraction of approximately 10-4.5 observed at 2 m depth can be explained if there are mechanisms acting in the shallow layer that limit the bubble population which can survive beyond the first minute or so after a breaking wave. However, this continuous layer appears not to extend to 4 m depth.

The most likely explanation for the advection generating deep plumes is Langmuir circulation. This is consistent with Farmer and Li (1994) who estimated that surface bubbles could be of the order of 100 s old before they reached 2 m depth in the downward flow of Langmuir circulations (at wind speeds of 10–15 m s-1). Thus, bubble populations have significant time to evolve before the deep plumes are formed.

The bubbles present at 2 m depth show a clear dependence on the surface forcing, with void fraction increasing with wind-wave Reynolds number above a threshold of RHW=2×106 or a wind speed of 16 m s-1. It is interesting that the Reynolds number threshold corresponds to that at which wind speed parameterisations of the CO2 transfer velocity diverge (Brumer et al., 2017a), implying a change to bubble-mediated gas exchange. The void fractions in 2 m plumes are also higher during falling winds than rising, with some evidence for a similar pattern at 4 m.

The most likely explanation for the wind speed hysteresis appears to be changes to bubble destruction mechanisms during rising and falling winds. It is likely that periods of increased wave breaking at higher wind speeds create surface waters with a higher concentration of oxygen and nitrogen. If the wind falls more quickly than the saturation state can adjust, bubbles produced during falling winds may last longer than those during rising winds because they are in water with higher gas saturation. A period of falling winds at high wind speeds implies that the recent winds were even higher, and the data support the idea that plumes of small bubbles last longer during these periods of (presumably) higher gas saturation.

Dahl (2003) saw increased acoustical scattering (assumed to be due to bubbles) when winds were falling compared to periods when they were rising, although these data were taken at low wind speeds (2–10 m s-1). Liang et al. (2017) suggested that the transfer of oxygen and nitrogen into the ocean was greater when winds were rising rather than falling, and at 19 m s-1 the flux of those two gases was almost doubled in the rising case compared with the falling case. They observe that breaking waves are fewer in number but larger in size in developing seas, suggesting that this may cause bubbles to be entrained to a greater depth, favouring gas transfer. This may leave a longer-lasting population once winds start to fall. It is also possible that bubbles dissolve faster in rising seas, leaving fewer to be detected. This underlines the point that bubble presence may not necessarily be directly related to gas flux in a simple way, since the presence of bubbles merely indicates that they have formed, been advected to the measurement point, and have not yet been destroyed.

It is hard to separate out the effect of potential changes in bubble production and changes due to subsurface processes. Whitecap observations may provide a first-order proxy for bubble production. Whitecap fractions during HiWinGS were observed to decrease as wave age increased (Brumer et al., 2017b), but Fig. 9 here shows no difference in observed void fraction at 2 m at high and low wave ages. Callaghan (Callaghan et al., 2008) observed significantly higher whitecap coverage during periods of falling winds compared with rising winds, for wind speeds between 10 and 24 m s-1. However, the same separation was not seen in satellite data during a later study (Goddijn-Murphy et al., 2011). These studies and our data do not provide a conclusive steer on the likelihood of bubble production being directly affected by rising or falling winds, although the bubble presence at 2 m depth clearly is. However, the reversal of the pattern of void fraction with wind speed at the highest and lowest winds (when there is a bias due to the most recent likely conditions) suggests that bubble lifetime is likely to be a more significant influence than bubble formation on the shift in bubble presence during rising and falling winds.

In order to build a more complete picture of plume evolution, the two bubble layers (the shallow plumes in the top metre and the deeper plumes which have been advected downwards) need to be studied simultaneously. Although the processes in both cases are ultimately driven by similar phenomena – high winds causing breaking waves and contributing momentum to the subsurface flow – the critical mechanisms for the two layers are likely to differ. This may explain why no simple links have been found between deep bubble plumes and surface forcing conditions: very few measurements have been made in the top metre to track the upper bubble field, and the deeper measurements have rarely had the auxiliary data needed to explain the processes forming them. It is clear that data from different depths within both the shallow plumes and the deeper plumes, correlated with flow and gas saturation data, are needed to follow the mechanisms driving our observations.

We have no direct measurements of the surfactant activity during HiWinGS. Chlorophyll was measured; levels were highest during the first buoy deployment (approximately 1.5 µg L-1) and generally decreased as the winter approached, reaching a low of 0.3 µg L-1 in the last deployment (Table 1). However, recent studies (Sabbaghzadeh et al., 2017) suggest that there is no universal relationship between chlorophyll levels and surfactant activity.