Upper ocean observations from the north Pacific subtropical front during late winter demonstrate the generation of submesoscale intrusions by buoyancy loss. Prior to generation, a sharp thermohaline front was intensified by confluent flow of 1–2 × 10−5 s−1. Relative vertical vorticity, ζ, across a surface-intensified, along-front jet on the warm side of a frontal trough was 0.5 f. During the storm, buoyancy loss arose due to cooling of ∼650 W m−2 and down-front wind stress <0.5 N m−2 that generated a southward, cross-front Ekman transport of dense water over light. The resulting wind-driven buoyancy flux was concentrated at the front where it exceeded that due to convection by an order of magnitude. The intrusions appeared immediately following the storm both within the surface mixed layer and beneath the seasonal pycnocline. They were approximately 20 m thick and horizontally elongated in the cross-frontal direction. The near-surface intrusions had cool and fresh properties characteristic of the water underlying the seasonal pycnocline, whereas the subsurface intrusions were composed of warm and saline water from the surface. The apparent vertical exchange was constrained within a thin filament of 2 km zonal extent that was characterized by O(1) Rossby and Richardson numbers, pronounced cyclonic veering in the horizontal velocity throughout the surface mixed layer, and sloping isopycnals. The intrusion properties, background environmental context, and forcing history are consistent with prior numerical modeling results for the generation of ageostrophic vertical circulations by frontogenesis intensified by buoyancy loss, possibly resulting in symmetric instability.
 The dynamics of ocean fronts are currently receiving considerable attention due to the recent emergence, and perceived importance, of submesoscale dynamics in catalyzing both lateral and vertical upper ocean exchange [Thomas et al., 2008]. Defined as motions with Rossby numbers, (1), where is the relative vertical vorticity, and f is the Coriolis frequency; submesoscales are characterized by horizontal scales of 1–10 km and ageostrophic vertical velocities of O(100 m day−1), an order of magnitude larger than those predicted for geostrophic flows [Mahadevan and Tandon, 2006]. Numerical modeling typically indicates that submesoscales emerge from large-scale fronts where the presence of lateral density gradients and large vertical shears promote the development of thin filaments characterized by large strain and vorticity [Klein et al., 1998; Capet et al., 2008; Thomas et al., 2008; Wang and Jordi, 2011; Yoshikawa et al., 2012]. The large vertical velocities generated by submesoscales have important consequences for the subduction to the ocean interior of surface waters containing dynamical and biogeochemical tracers such as potential vorticity and dissolved gases [Thomas and Joyce, 2010].
 The mechanisms considered responsible for the generation of submesoscales within the surface mixed layer (SML) are frontogenesis, whereby confluent flow sharpens frontal gradients and locally increases the vorticity and lateral buoyancy gradient, , and forced and unforced instabilities [Mahadevan and Tandon, 2006; Thomas et al., 2008]. Among these mechanisms, particular interest has been focused on forced buoyancy loss at the ocean surface. Convective heat loss is known to intensify frontal ageostrophic circulations within submesoscale filaments [Yoshikawa et al., 2001], but more recently the buoyancy flux achieved by down-front winds through the cross-front Ekman transport of dense water over light has emerged as a potentially more prominent generation mechanism at strong fronts [Thomas and Lee, 2005; Lee et al., 2006; Thomas and Joyce, 2010; Nagai et al., 2012]. Recent efforts have attempted to diagnose the relative influence of cooling versus wind-driven buoyancy loss by quantifying the reduction in surface Ertel potential vorticity (EPV), , where is the absolute vorticity, due to each process [Yoshikawa et al., 2012]. In regions where EPV becomes negative, symmetric instability (SI) drives slantwise convection along isopycnals throughout a stratified but turbulent SML within narrow filaments of width O(1 km) [Taylor and Ferrari, 2010; D'Asaro et al., 2011].
 Submesoscales also restratify the SML by slumping lateral buoyancy gradients within mixed layer eddies (MLEs) [Boccaletti et al., 2007; Molemaker et al., 2005; Capet et al., 2008]. Previous observations have confirmed that the lateral density gradients associated with MLEs reach scales as small as a localized internal Rossby radius, , where N is the stratification within the SML and H is its depth [Hosegood et al., 2006]. Typical values of LR are O(1 km), which is an order of magnitude smaller than the previously suggested smallest scales of variability [Ferrari and Rudnick, 2000] and far below the horizontal resolution of regional and global scale numerical models.
 In this paper, we present observations of upper ocean, submesoscale intrusions made over two weeks during late winter at the north Pacific subtropical front (STF) (Figures 1a and 1b) to demonstrate the impact of wind-forced buoyancy loss on submesoscale upper ocean structures in addition to the restratifying influence of MLEs. Following a strong storm, during which wind stress and oceanic heat loss were especially large, intrusions appeared both within the surface mixed layer and beneath the seasonal pycnocline (Figures 1c and 1d). Similar frontal features have been previously observed and attributed to double diffusive interleaving [Gregg, 1980], submesoscale turbulence [Shcherbina et al., 2010], and advection by near-inertial waves [Beal, 2007]. To our knowledge, no previous observations have shown intrusions to occur both above and below the seasonal pycnocline. Furthermore, the cold, fresh intrusions observed by Shcherbina et al.  at the same location extended exclusively into the warm side of the front and were thus of the opposite sign to the warm and saline character of the intrusions presented here. We demonstrate here that the intrusions were most likely generated by ageostrophic secondary circulations at the frontogenetic front that were intensified by the substantial buoyancy loss experienced during the storm.
 Our analysis focuses on (i) the preconditioning of the front to the generation of submesoscale intrusions, (ii) the properties of the intrusions and the associated velocity field throughout their evolution, and (iii) their proposed generation by ageostrophic circulations resulting from buoyancy loss during the storm. Our results are consistent with numerical simulations of frontogenesis intensified by buoyancy loss [Yoshikawa et al., 2001; Thomas and Lee, 2005; Mahadevan and Tandon, 2006; Taylor and Ferrari, 2010; Yoshikawa et al., 2012] and supports the suggestions of Shcherbina et al.  that intrusions within the STF are a result of submesoscale turbulence. The details of our analysis follow in the main body of the paper but the key aspects of points (i)–(iii) above are as follows:
 1. Measurements were made over 2 weeks within the convergent flow field of a frontal trough of the STF where sea surface temperature gradients were intensified and modeling predicts the development of downdrafts that subduct surface waters to the underlying pycnocline [Mahadevan and Tandon, 2006; Yoshikawa et al., 2012]. A particularly sharp thermohaline front was identified in ship-based measurements during a large-scale survey at the beginning of the 2 week period and 6 days before the strong storm that generated the intrusions. The front, observed at 30.1°N where it separated warm and saline subtropical water to the south from cooler, fresher water to the north, was largely density compensated. Within a 10 km wide band located on the warm side of the front, confluent flow was estimated as − where V is the northward velocity, which is an order of magnitude larger than estimates based on the geostrophic velocities derived from altimetry with a 30 km horizontal resolution. Within the same band, density increased by >0.05 kg m−3 to the north, and an along-front jet of 0.4 m s−1 was surface intensified within the upper 20 m. The strong vorticity associated with the jet resulted in across the front. Near-surface stratification maintained positive EPV, but 2 weeks later, at the end of the observational period, EPV was nearly zero following two storms. The implication is that the cumulative impacts of the storms, of which the first was considerably more energetic, extracted EPV from the upper ocean. The front at the end of the observational period was considerably more diffuse than at the beginning, reflecting the same transient nature of frontal intensity as D'Asaro et al.  who observed negative PV for only 0.2 days at the Kurushio.
 2. The intrusions appeared following the first storm during which wind stress approached 1 N m−2 and heat loss ∼650 W m−2. Their temporal evolution was observed during repeated small-scale Lagrangian surveys of spatial scale 10 × 10 km centered on a drogued drifter. A prominent intrusion of high salinity water with properties of the surface layer was observed beneath the seasonal pycnocline while a fresh intrusion with properties of water from beneath the seasonal pycnocline was observed immediately above it at the surface (Figure 1). In both cases, the respective differences in of the intrusions relative to the surrounding water were ∼1°C and 0.4 psu. Both intrusions were ∼20 m thick and elongated in the cross-front direction. The subsurface intrusion had a horizontal extent of ∼2 × 8 km in orthogonal directions and was entirely detached from the surface. The intrusion lay horizontally and did not cross isopycnals, whereas the surface intrusion extended upward and sloped across isopycnals. The intrusions were embedded in a thin filament of 2 km zonal extent within which Ro and Richardson numbers, Ri, were O(1) due to vertical shear and cyclonic veering of the horizontal velocity. In addition to generating intrusions, mixed layer eddies reaching horizontal scales reaching LR appear to have rapidly restratified the surface mixed layer over its entire depth within 2 days of the storm.
 3. The absence of intrusions before the storm, and their appearance immediately following it, implicates buoyancy loss during the storm in their generation. Air temperature dropped by >4°C, triggering convective overturning at the surface and deepening of the mixed layer. In addition, wind stress had a significant down-front component in the direction of the along-front jet. In conjunction with the observed lateral density gradient and relative vertical vorticity at the front, a wind-driven buoyancy flux (WDBF) was generated by advecting dense water southward across the front. WDBF was intensified at the front, exceeding the convective air-sea buoyancy flux by an order of magnitude. In accordance with numerous modeling studies [Yoshikawa et al., 2001; Mahadevan and Tandon, 2006; Yoshikawa et al., 2012], we propose that buoyancy loss during the storm intensified the ageostrophic circulations resulting from the initial frontogenetic flow field within a narrow filament with O(1) Ro.
 The paper is structured as follows. We first describe the experimental details including location, sampling strategy, and instrumentation. The meteorological forcing and one-dimensional response of the SML is then presented before we focus on the frontal characteristics, specifically the extent to which susceptibility to frontogenesis can be identified. The properties of the intrusions are then discussed with reference to the salinity, density, and velocity fields before the wind-driven buoyancy flux is quantified. These results are then discussed from the perspective of, first, the relative roles of wind and cooling as forcing mechanisms for the intrusions. Second, the potential of symmetric instability in driving the observed mixed layer dynamics is explored; third, time dependence is considered due to the role of near-inertial oscillations. The role of submesoscale mixed layer eddies in restratifying the SML is addressed before a final thought on the implications to mode water formation is offered.
2. Experimental Background and Methods
2.1. Study Region and Sampling Strategy
 Data were acquired aboard the RV Wecoma between t = 62–76, where t is days elapsed in 2004 starting from 0 at 0000UTC, 1 January, within the region 29–31°N, 150–149°W (Figure 2). To monitor the temporal evolution of three-dimensional structures throughout the upper ocean, a Lagrangian reference frame was employed for the majority of the cruise during which an area of typically less than 10 × 10 km was repeatedly surveyed. These small-scale surveys, an example of which is depicted in Figures 1c and 1d, are referred to throughout the paper as a group. The origin of each group was defined by the position of a drifter drogued at a depth of 38 m and coordinates given as displacements from the drifter in kilometers to the east, and to the north, , respectively. An exception to this convention are two large-scale, exploratory surveys (legs of ∼100 km length), Alpha and Bravo, that were conducted at the beginning and end of the measurement period and for which positions are given in Cartesian coordinates.
 As submesoscales evolve over timescales of O(1 day), each group were required to be completed in less than 4 h so as to be treated as an instantaneous realization of the tracer and velocity fields. This approach is consistent with Thomas et al.  who consider the data collected within 0.07 inertial periods (1.3 h) from either side of the subpolar front as an instantaneous realization of the tracer and velocity fields. The key results in this paper relate to the intrusions that were monitored over 5 days throughout groups 27–46, corresponding to the period t = 69.4–75.0. The timing of all groups referred to throughout the paper are illustrated at the top of Figure 3.
2.2. Oceanographic Data
 The principal instrument used to measure the thermohaline and associated tracer fields was the Shallow Water Integrated Mapping System (SWIMS), a 300 kg towed body equipped with a Sea-bird CTD that depth cycles behind the ship in a saw tooth pattern. At a tow speed of ∼5 knots and sampling the upper 150 m of the ocean, each profile was completed in min and the horizontal distance between profiles was ∼350 m. In addition to standard CTD parameters, SWIMS also measured dissolved oxygen. To account for variations in saturation levels due to temperature and salinity changes, dissolved oxygen was converted to available oxygen uptake (AOU) as the observed value subtracted from the saturated value for each point given its TS properties. After doing so, it was apparent that all water above the permanent pycnocline was recently in contact with the atmosphere and had the same values of AOU, precluding the identification of subducted water by a higher oxygen content relative to its surroundings. Current velocities were obtained by the vessel-mounted acoustic Doppler current profiler (VM-ADCP) as 2 min ensembles in 4 m vertical bins between 7 and 200 m.
 The dissipation rate of turbulent kinetic energy, ε, is calculated from SWIMS data by computing Thorpe scales [Thorpe, 1977], LT, which estimates ε through the relationship Lo = 0.8 LT where Lo= is the Ozmidov scale. Several vertical profiles were also made using the loosely tethered Modular Microstructure Profiler (MMP) from which direct estimates of ε are obtained from measurements of microscale shear in the usual manner following Osborn .
2.3. Meteorological Data
 In addition to oceanographic measurements, meteorological data were acquired onboard the RV Wecoma, and version 2.6 of the COARE bulk algorithms [Bradley et al., 2000] were used to compute the air-sea fluxes. The air-sea buoyancy flux, Jb, was calculated from the meteorological and oceanographic measurements as:
where Q = Qshortwave + Qlongwave + Qlatent + Qsensible + Qrain (W m−2) is the total heat flux, = 4000 J kg−1 K−1 is the specific heat of water, g is the acceleration due to gravity, and α and β are the thermal and haline contraction coefficients computed from the daily mean SST and SSS measured by the intake at 3 m depth. E and P are evaporation and precipitation, respectively, and Ssurf is the surface salinity.
3. Meteorological Forcing and the 1-D SML Response
3.1. SML Response to Atmospheric Forcing
 The air-sea buoyancy forcing alternated between predominantly stabilizing and destabilizing due to the two storms experienced during the two weeks of measurements. Considered as four phases, labeled A–D in Figure 3, phases A and C were characterized by weak winds and negligible nighttime convection and were thus stabilizing. Phases B and D were largely destabilizing due to cold air temperatures and, during B in particular, strong winds. The difference in air and sea temperature (Figure 3c), ΔT = Tsea − Tair, reached 4°C during both storms. The significant aspects of the forcing and SML response are described below for each period:
 1. Prior to the first storm, wind stress, τ, remained N m−2 and was directed consistently to the south. ΔT−1°C and the resulting Jb remained largely negative (stabilizing), reaching a minimum value of −4.5 × 10−7 W kg−1. The base of the SML when defined by a density difference criteria of 0.1 kg m−3 was only 30 m due to the presence of the underlying seasonal pycnocline that was unusually shallow for the time of year. Throughout the upper 150 m, ε estimated from overturns barely exceeded noise levels (considered as −9.5 W kg−1).
 2. During the first storm, wind speed dropped to virtually zero before rapidly increasing as Tair dropped to a minimum value of 14.5°C, generating ΔT 5°C at t = 67.75. As τ increased to N m−2, the wind rapidly rotated in a clockwise sense and the cruise maximum, Jb = 2.6 × 10−7 W kg−1, was achieved. Qlatent attained cruise-maximum values of 445 W m−2, with Qnet reaching 650 W m−2. This phase may thus be categorized as being forced by both convective cooling and high wind stress; it is shown later that the component of τ directed down the front would further enhance convection through nonlinear Ekman transport across the front. As τ increased around t = 68.0, the SML rapidly deepened to 60 m and surface increased by 0.1 kg m−3 within a matter of hours. The maximum values of W kg−1 observed over the 2 weeks reached the seasonal pycnocline and vertically homogenized the SML. Stratification within the underlying seasonal pycnocline increased to s−2 within a thinner (∼20 m) and more clearly defined vertical layer.
 3. As the storm passed, Tair rose back to prestorm values (20°C) and eventually exceeded Tsea such that ΔT approached −2°C, i.e., the air-sea heat flux reversed sign and Jb remained largely stabilizing despite the nonzero wind stress. Wind stress remained ∼0.2 N m−2 until day 70 with a second peak of N m−2 at t = 68.5 at which time the wind was directed to the east. As the wind speed dropped at t = 70.0, the direction rotated to a southward heading and the SML restratified. A distinct layer within which s−2 appeared at 15 m at the beginning of day 71 and gradually deepened throughout the SML toward the seasonal pycnocline. The rapid restratification exceeded that due to insolation and is discussed further in Appendix A. Between , ε was negligible at the surface except for the uppermost 10 m as revealed by a series of MMP profiles, most likely due to direct mixing by surface waves generated by the preceding storm or contamination from the ship's wake. Turbulence within the intermediate layer between the seasonal and permanent pycnoclines was an order of magnitude larger than background levels. Bands in appear over time further indicating that the ship was repeatedly passing between areas of varying density, indicative of a lateral density gradient within the SML at a scale of less than 10 km spanned by the small-scale Lagrangian surveys.
 4. During the second, weaker storm Tair decreased to 14°C and ΔT increased to 5°C. Values of Qlatent, Qsensible, and Qnet were half those during phase B due to the weaker wind stress that increased from zero to N m−2, rotating rapidly from a prestorm southward heading to a northward and back again to a southward heading. The combined influence of increased τ and large ΔT generated a destabilizing 10−7 W kg−1 that eroded the near-surface stratification created during the preceding phase C. Values of ε approached 10−6 W kg−1 throughout the SML to the depth corresponding to the upper edge of the seasonal pycnocline. Note, however, that the turbulence did not appear to reach the SML base when considered as the depth at which increased above surface values by 0.1 kg m−3. In contrast to phase B, approximately one week earlier when τ reached 1 N m−2, the destabilizing buoyancy flux during this last phase was largely derived from the air–sea Jb as a result of elevated Qlatent. As will be discussed later, stronger τ directed down the front potentially elevates the buoyancy flux by adding a wind-driven component. This second storm was thus distinctly different to the first during phase B as wind stress was weaker, but convective heat loss was similar.
4. Oceanographic Context
 In this section, we present results from (i) remotely sensed sea surface temperature (SST) and altimetry measurements and (ii) high-resolution in situ SWIMS and VM-ADCP measurements from the exploratory large-scale frontal surveys, Alpha and Bravo, conducted at the beginning and end of the two week observational period during phases A and D above, respectively.
4.1. Regional Overview
 The cruise track largely followed the 18°C isotherm in a section of the STF that was characterized by horizontal SST gradients that varied in strength due to the meandering front and mesoscale eddy field (Figure 2a). Strongest SST gradients were observed in the frontal troughs that extended southward between cyclonic and anticyclonic eddies. Weaker SST gradients were found at the wave crests where the STF extended northward. The observations were made just upstream of a frontal wave trough centered on −149°E.
 Following deployment of the drifter at −150°E, the cruise track followed the frontal trough, traveling eastward with a cyclonic rotation. The rotation is only weakly apparent in geostrophic velocity vectors computed from altimetry (30 km horizontal resolution) (Figure 2b) but much more pronounced in the observed low-pass filtered (cutoff σ = 0.5 cpd) VM-ADCP velocities and drifter trajectory throughout the two weeks of observations (inset, Figure 2b).
4.2. Confluence and Frontogenesis
 Confluence concentrates the north-south frontal temperature gradients, rendering the front susceptible to frontogenesis. Altimetry-derived geostrophic velocities indicate that the observations were made within a region of varying, but weak, convergent and divergent flow (Figures 2c and 2d). Confluent flow was most pronounced to the west of survey Alpha, where , which is a factor of 5 smaller than values of ∼0.5 × 10−5 s−1 observed at the Kurushio [D'Asaro et al., 2011]. Altimetry-derived velocities have coarse spatial resolution and are based on a three-day mean during which time a strongly convergent flow field may have dissipated; D'Asaro et al.  observed a reversal in flow from strongly convergent to divergent within 1.5 days. Ship-based observations during Alpha identify a narrow zone of 20 km width that was centered on the front (Figure 4a), however, and across which s−1 when V is computed as the mean northward velocity within the upper 200 m.
 Horizontal strain rates
are greatest at the periphery of the strongest eddies with MSLA of 15 cm to the north and south of the study area. The peripheries of the eddies also correspond to the strongest SST gradients and St/f approached unity. Within the region in which the ship-based measurements were made, however, geostrophic strain was typically at least two orders of magnitude lower. In situ measurements cannot provide more accurate estimates of St due to our inability to quantify gradients in both velocity components simultaneously from the VM-ADCP data. Within the area presented in Figure 2, , with maximum values observed within the cores of the strongest eddies at 28 and 32°N. These values are unremarkable but ship-based observations suggest that Ro was likely to have been significantly enhanced in localized regions that were not resolved by the 30 km grid spacing of the altimetry data.
4.3. Frontal Thermohaline Context
 Detailed observations of the vertical structure of the front, and how it evolved during the 13 days between them, were obtained during Alpha and Bravo (Figures 1a and 1b). Alpha was conducted during day 62, 6 days before the storm on day 68 that we believe generated the intrusions, and was composed of one zonal and one meridional leg each of length 100 km. Bravo was completed on day 75 following groups 15–46 and was composed of two parallel, meridional legs of similar length to Alpha.
 The strong thermohaline front at 30.1°N was clearly defined and strongly localized during Alpha but had become weaker and more diffuse by the time Bravo was completed at 30.4°N. Horizontal gradients across the front during Alpha were /δx 0.2°C km−1 and δS/δx psu km−1 over a distance of <10 km. There was no evidence during Alpha of intrusions of the same magnitude (i.e., 0.4 psu over a spatial scale <1 km) seen after the storm.
 During Bravo, the front was characterized by isolated patches of elevated salinity (S > 34.8) to the north of the front that were vertically constrained to the 50 m deep SML. The higher salinity of these surface patches with respect to their surroundings indicates their origin to be from the equatorward, warm side of the front. Note that the fresher surface water with S < 34.5 psu observed to the north of the front during Alpha (apparent as blue shading in Figure 1a) was absent during Bravo. The differences in θ between Alpha and Bravo were broadly the same as in S except within the intermediate layer between the seasonal and permanent pycnoclines where θ lacked the spatial gradients that were apparent in S. The total change in θ, S across the front was approximately the same during Bravo compared to Alpha, but the change occurred in a more diffuse manner across a broader region and in number of smaller fronts. The change in frontal structure bears striking similarity to the changes over time observed by D'Asaro et al.  at the Kurushio where a sharp front evolved into a weaker, more diffuse front across which the net changes in thermohaline properties remained the same.
 In contrast to idealized scenarios in which an outcropping density front provides a clear pathway for the subduction of surface waters, the vertical structure of the upper ocean throughout our observations was complicated by the presence of both the seasonal and permanent pycnoclines at depths of 30–60 m and 120 m, respectively. During Alpha, the permanent pycnocline shoaled northward with a slope of 0.001, while above it the seasonal pycnocline was unusually shallow due purportedly to the unusually high rainfall during the winter of 2003/2004. Above the seasonal pycnocline there was, therefore, little evidence of an SML in the classical sense. The uppermost 30 m was weakly stratified and host to lateral density gradients, apparent by the 24.9 kg m−3 isopycnal intermittently reaching the minimum depth to which SWIMS sampled (10 m).
4.4. 2-D Frontal Sections
 The susceptibility of a large-scale front to intensification and the formation of ageostrophic motions can be diagnosed by, among other factors, the properties of an along-front jet and its associated relative vertical vorticity, potential vorticity (PV), and stratification. To assess the extent to which the front was preconditioned to the formation of such instabilities, we consider the meridional legs of Alpha and Bravo (eastern leg) as cross-front sections for which the eastward velocity component, U, constitutes the along-front flow in a 2-D approximation for PV, ,
 The results from the section during Alpha, which required 15 h to complete, should be considered as the preconditioning state prior to the generation of the intrusions, while the section during Bravo, which required 9 h to complete, reflects the frontal environment following vigorous forcing. During both surveys, there was negligible contamination of the velocity field by near-inertial oscillations. Note that there are no data available from depths less than 10 m, where modeling results have demonstrated the effects of frontogenesis are most pronounced [Klein et al., 1998; Capet et al., 2008; Wang and Jordi, 2011].
 During Alpha, near-surface at 15 m depth fluctuated by >0.05 kg m−3 over distances of ∼10 km between 24.88 and 24.98 kg m−3 (Figure 4a). At the front (identified by the black dashed line at 70 km in Figure 4a), increased rapidly at the northern edge of a surface-intensified frontal jet on the southern, warm side of the front (Figure 4b). Note that the meridional density distribution across the front was therefore such that an eastward wind stress (i.e., blowing in the direction of the along-front jet) would drive a southward transport of dense water over light. The frontal current penetrated to the permanent pycnocline at 120 m but was surface intensified within the upper 30 m where m s−1 across a ∼20 km zone. Within the along-front jet, PV approached zero and was typically lower, 5 × 10−10 s−3, to the south of the front compared to the north where the high PV seasonal pycnocline was nearly inseparable from the surface (Figure 4b). Between the high PV pycnoclines the weak stratification lowered PV to nearly zero except for a 20 km wide region coinciding with the front where N2 was an order of magnitude larger (Figure 4e), and due to the cyclonic vorticity associated with the northern edge of the jet (Figure 4d). The large at the front reduced PV to nearly zero at the surface. The break in the shallow PV barrier at the front is similar to that found in the Gulf Stream by Thomas and Joyce  where subduction followed extraction of PV by down-front winds. The data in Figure 4 are slightly smoothed; the break in the stratification and reduction in PV are more noticeable in the raw, unsmoothed data.
 Bravo was conducted 13 days after Alpha and followed the two storms and groups 15–47. Similar to Alpha, at the surface fluctuated over the same range of values (Figure 4f). There was no longer a distinct core to the surface current, which instead extended over a horizontal zone of 60 km (Figure 4g). Maximum m s−1 were half those observed during Alpha, but currents were still largely surface intensified on the warm side of the front. Due to the weak stratification across the upper 50 m, surface PV approached zero (Figure 4h). N2 between the two pycnoclines had increased following the deepening of the seasonal pycnocline following the first storm, thereby also increasing PV within the intermediate layer. The weakening of the frontal jet core caused a reduction of Ro, although the jet edge was still notable by at the front (Figure 4i). The PV distribution during Bravo thus reflects the increased importance of N2 due to the weakening in ζ.
 Vertical shear in the along-front velocity (approximated as U) peaked in the core of the frontal jet during Alpha at Sh2 = 5 × 10−3 s−2, where Sh computed over 4 m vertical intervals. If the lateral stratification was in thermal wind balance with the along-front flow such that , we would expect a lateral buoyancy gradient of 3.75 × 10−5 s−2. Calculating M2 along a depth of 120 m so that the sloping permanent pycnocline is accounted for, we find 8 × 10−6 s−2, somewhat smaller than required for thermal wind balance and implying an additional source of shear. We note, however, that the maximum shear is constrained to a very narrow region at the front and that values of shear throughout the majority of the section are commensurate with those required for thermal wind balance to be valid.
 During both surveys, the minimum horizontal length scales of density variability at a depth of 15 m were consistent with LR. Due to the deeper SML during Bravo compared to Alpha, LR increased from ∼2 km during Alpha to ∼5 km during Bravo. Small-scale lateral density gradients were particularly pronounced to the north of the front during Alpha and suggest the presence of mixed layer eddies that are implicated in the rapid restratification of the SML following the storm. A detailed analysis is presented in Appendix B.
5. Submesoscale Upper Ocean Structure and Evolution
 The previous section presented the large-scale context within which the intrusions evolved. We now present observations of the upper ocean salinity and density structure and the associated velocity field during the period t = 69–74.5 corresponding to groups 29–44.
5.1. Intrusion Structure
 The results from six groups conducted at approximately the same time each day have been chosen to depict the intrusion structure and the environment within which they evolved (Figure 5). As the temperature and salinity gradients were largely compensated, the properties of the intrusion are described in terms of salinity. At the scale of the groups, salinity within the SML varied in the zonal rather than in the meridional direction as would have been anticipated from the large-scale frontal distribution of water properties. Within the upper 60 m, salinity was vertically homogenous but decreased abruptly from S > 34.8 psu to the west of the drifter to <34.7 psu within a distance of <1 km of the drifter (Figure 5a).
 The decrease was not associated with a front but instead with a narrow filament of 2 km zonal extent within which fresher water was observed toward the surface approximately 2 km east of the drifter (Figure 5b). At 75 m depth, the saline intrusion was similarly horizontally constrained. The filament within which the intrusions were embedded was most apparent during group 36 (indicated by green ellipse in Figure 5d), which was somewhat larger in spatial coverage than the other small-scale surveys and resolved the increase in surface salinity 5 km to the east of the drifter. High salinity (S > 34.9 psu) was observed at the western and eastern edges of the sample region (Figure 5d) but was reduced to <34.7 psu within the filament just to the east of the drifter.
 Adjacent to the subsurface intrusion were plumes of cool, fresh water that suggest the upward transport of water from the permanent pycnocline. They were a persistent feature and are clearly resolved during groups 33 and 39, for which individual sections in the zonal (W-E) and meridional directions (SW-NE) are presented in Figure 6 (the corresponding legs are indicated by the black dashed lines at the surface in Figures 5c and 5e). The zonal leg bisected the subsurface intrusion, but the SW-NE legs surveyed ∼3 km to the west and missed the main structure of the filament. During group 39, the subsurface intrusion was split into two segments by one such plume of fresher water (indicated in Figure 6 by the white vertical arrows). Alternatively, this group captured the presence of two different intrusions, for which evidence exists during group 36 that shows multiple subsurface patches of elevated salinity throughout the wider area. The orientation of both sections fails to demonstrate any significant slope in the subsurface intrusion, but the upward sloping fresher intrusion had a slope of between 0.007 (group 39) and 0.013 (group 33). A further discussion of the importance of the slopes is provided on the subject of double diffusion in Appendix D.
 Both isohaline and isopycnal surfaces exhibit three-dimensional structure over the scale of the individual groups. To the west of the drifter, both isohalines and isopycnals were horizontal (Figure 7). Within 1 km of the drifter, the near-surface intrusion of fresher water caused the 34.72 psu isohaline surface to rise toward the surface while below it the subsurface intrusion can be identified at 75 m. As the deformation is associated with a filament rather than a front, the shoaling isohaline deepens back to its original depth <2 km to the east of the drifter. The subsurface saline intrusion did not translate horizontally or vertically during the observations, remaining centered at a depth of 75 m where vertical migration may have been inhibited by the seasonal and permanent pycnoclines above and below, respectively. Groups 33 and 34 resolved the intrusion's full extent as being <10 km and ∼2 km in the meridional and zonal directions, respectively. It was isolated from the surface and from any water with similar properties at the same depth.
 Isopyncals within the SML sloped sharply at the filament edge. During group 29, the SML was well mixed (Figure 7a); the 24.98 kg m−3 isopycnal was vertical near the position of the drifter and the SML base was delineated by the essentially horizontal 25.0 and 25.1 kg m−3 isopycnals. Two days later, during which time the air-sea buoyancy flux was stabilizing and τ was weak, isopycnals had begun to slope abruptly downward at the filament edge. The sloping isopycnals were persistent throughout the observational period and reflected the presence of lateral buoyancy gradients throughout the SML at scales commensurate with LR. The isopycnals sloped in the zonal direction, i.e., from west to east, in the same orientation as isohaline surfaces but with opposite slopes so that the two surfaces were required to cross. Thus, the vertical transport of surface water downward could have been achieved isopycnally if the subsurface saline intrusion originated at the surface and was subsequently subducted along the sloping isopycnals.
 The emergence of increasingly lighter isopycnals appearing throughout the surveys indicates both the restratification of the SML and the (re)establishment of the aforementioned lateral density gradients. The cumulative increase in N2 due to insolation would have contributed to restratification (Figure 12) but cannot explain the decrease of nearly 0.1 kg m−3 over four diurnal periods. Furthermore, Figure 3f provides clear evidence that following the complete destruction of SML stratification during day 69, 10−5 s−2 became established throughout the entire SML within two diurnal cycles. As the effect of insolation on restratification is surface intensified (upper 20 m), it is most likely that the observed growth in N2 throughout the SML was due to mixed layer eddies slumping the density surfaces acting over the entire depth of the SML.
5.2. Velocity Field: Vorticity, Strain, Richardson Number, and Rotation
 Submesoscale dynamics are associated with high shear, horizontal strain, and relative vorticity [Thomas et al., 2008]. In particular, positive vorticity has been shown to be enhanced within the narrow filaments in which submesoscale subduction occurs [Mahadevan and Tandon, 2006]. Elevated shear is further implicated in the regime within which symmetric instability drives slantwise convection at a lateral buoyancy gradient when [Taylor and Ferrari, 2010; Thomas and Taylor, 2010] and a clockwise rotation with depth. In this section, we demonstrate three specific aspects of the velocity field that implicate submesoscales in shaping the environment in which the intrusions evolved:
 1. The SML is highly sheared vertically and lowers Ri to O(1).
 2. This shear is generated by currents that rotate predominantly clockwise with depth throughout the SML and then weakly anticlockwise at the depth of the subsurface intrusion.
 3. The currents within the SML veer horizontally northward, i.e., positive ζ, at the position corresponding to the filament.
 We focus on Group 34 (t = 71.53–71.74) due to the velocity characteristics exhibiting their clearest signal and because the group was conducted at approximately midpoint through our observations.
5.2.1. Horizontal Gradients in Velocity: Vorticity and Strain
 Following removal of the near-inertial component (0.9 f − 1.1 f), velocity throughout the SML was predominantly eastward at all times due to the persistent frontal jet (see Figure 2b, inset). Surface currents rotated cyclonically from southward (negative V) to the west of the drifter to northward (positive V) at the position corresponding to the filament (Figure 8). The veering was constrained to the SML and did not extend beneath the seasonal pycnocline where the background frontal jet dominated and velocities remained primarily southward. Maximum zonal velocities ( m s−1) were consistently observed at the southern edge of the Lagrangian survey region. We interpret this as evidence that the small-scale surveys were located at the northern edge of frontal jet and therefore in a region of high positive ζ.
 The veering with horizontal position generated an increase in positive relative vertical vorticity, ζ, that extended to, and was maximum at, 70 m depth at the upper edge of where the saline intrusion was observed (Figure 9a). (1) where isopycnals descended 20 m. Maximum values ( ) were associated with the rapid cyclonic veering in V from a southward to a northward direction. An adjacent filament of negative Ro represents the periphery of the region that demarcates the intensification of the eastward current in the south western sector of the region. The adjacent regions of positive and negative vorticity over a horizontal scale of <10 km, and their magnitude, is consistent with observations of Thomas et al.  in the subpolar front of the Japan/East Sea. Strain was elevated only in patches, albeit within the central region broadly corresponding to the filament (Figure 9b).
5.2.2. Vertical Gradients in Velocity: Shear and the Richardson Number
 At the position corresponding to the filament, Ri within the SML attained values is <1, which is unstable to symmetric instability (Figure 9c). Critical values were constrained to the upper 50 m and within a narrow band of 2 km lateral extent. Beneath the SML, everywhere due to the increase in stratification that developed following the first storm throughout the intermediate layer.
 The rotation of the velocity with depth exhibited two distinct and persistent patterns. Throughout the SML, the velocity rotated clockwise at a rate of 0–5° m−1(Figure 9d). Highest rotation rates were at the surface where Sh and Ro were highest, and Ri was minimum, suggesting that the clockwise rotation provided the shear that lowered Ri to critical values. Beneath the seasonal pycnocline at ∼70 m, the velocity rotated anticlockwise with depth at a slower but consistent rate of O(1° m−1). The anticlockwise rotation was limited to the moderately stratified, intermediate layer between the pycnoclines within which the subsurface intrusion was observed (indicated in Figure 9d by the black 34.72 isohaline); within the underlying permanent pycnocline, the velocity reverted back to a weak clockwise rotation. As the velocities depicted in Figure 9 have had the near-inertial currents removed by band-stop filtering over the frequency range , the clockwise rotation throughout the SML should not be due to the downward propagation of the near-inertial wave that was generated on day 68. We note, however, that Alford et al.  observed a near-inertial wave radiating from the front nearby that had a frequency within 20% of f, and so we may be observing some residual energy from the wave. The change in the sense of rotation is at this stage unclear, but we propose that the transition from clockwise to anticlockwise is due to the dominance of the frontal jet beneath the seasonal pycnocline (where the anticlockwise rotation is very weak) compared to the SML where clockwise rotating currents attributable to either symmetric instability or the near-inertial wave are strongest. Details of the near-inertial component are discussed in Appendix C, and the potential role of symmetric instability and time-dependent currents are further discussed in sections 7.2 and 7.3.
6. Wind-Intensified Frontogenesis and Subduction
 The absence of intrusions prior to the storm and their immediate appearance following it implies their rapid, forced generation by buoyancy loss rather than by geostrophic forcing. Our observations were clearly made within a frontal trough where isotherms were compressed and the flow field was confluent, especially when estimated using ship-based observations. Numerical simulations show that strain-induced frontogenesis causes the subduction of density-compensated, warm, salty intrusions within frontal wave troughs over horizontal scales of <10 km [Wang and Jordi, 2011]. Frontogenesis is concentrated near the surface and accompanied by high Rossby numbers and strong cyclonic vorticity in precisely the same manner as we have observed during the Lagrangian groups. Other studies have also considered how frontogenetic circulations can be further intensified by buoyancy loss and have indicated an order of magnitude enhancement in vertical velocity relative to the unforced case [Yoshikawa et al., 2001; Mahadevan and Tandon, 2006; Wang and Jordi, 2011; Yoshikawa et al., 2012]. The vertical displacement required to subduct the surface water to beneath the seasonal pycnocline in our observations is ∼30 m and need not be dramatically enhanced relative to simulations of unforced subduction given that the transport could have taken longer than a day. The small horizontal scale of the filament is, however, consistent with simulations in which frontogenesis is intensified by buoyancy loss.
 Despite being unable to positively identify a region of negative EPV that would indicate conditions conducive to symmetric instability [Thomas et al., 2010; Nagai et al., 2012], EPV was low near the surface during Alpha which we consider to represent the preconditioning phase. Breaks in the underlying high EPV seasonal pycnocline were frequent and demonstrate that the pycnocline did not constitute an impenetrable barrier to subducted surface water. Despite lacking a frontal signature of the same magnitude as , surface density displayed a persistent gradient over 100 km and more pronounced gradients at scales of O(10 km).
 The buoyancy loss required to further reduce EPV and intensify ageostrophic circulations was driven by both cooling, which peaked at W m−2, and wind stress during the storm on day 68 (phase B). The latter generated a wind-driven buoyancy flux (WDBF) due to the interaction of the down-front wind, the lateral buoyancy gradients, and vertical vorticity associated with the along-front jet [Thomas and Lee, 2005; Mahadevan and Tandon, 2006]. To estimate the magnitude of the WDBF, the along- and cross-front components of the wind stress, and , respectively, were computed by rotating the wind stress vector to be aligned with the along-front current (estimating by low-pass filtering the velocity at 25 m depth, 0.5 cpd). Maximum τ was observed at when both components briefly reached a magnitude of 1 N m−2 (Figure 10). Thereafter, the along-front component of wind stress directed down the front, , reached a secondary peak of ∼0.5 N m−2 and remained at 0.1 N N m−2 until t = 70. The resulting Ekman transport
where is the along-front wind stress, is modified by the surface vertical vorticity of the along-front (geostrophic) flow, [Stern, 1965; Thomas and Lee, 2005]. It is assumed that the wind-induced frictional force is confined to an Ekman layer of thickness , where Az is the coefficient of turbulent viscosity that is furthermore much thinner than the vertical scale of the geostrophic flow, D, such that . Dense water from the northern (dense) side of the front is subsequently advected across the front to the south, generating convective instabilities and thus driving a WDBF across the base of the Ekman layer. The effective buoyancy flux then becomes
 The surface Ekman layer thickness, , was estimated to be 50 m and was thus a factor of 3 thinner than the depth to which the along-front jet penetrated, satisfying the criteria. The WDBF was computed for wind stress values of 0.2 and 0.5 N m−2, corresponding to the typical values observed during the storm and their peak values and by assuming two dimensionality across the front. Values from 15 m depth during the north-south leg of Alpha were used to approximate , ζ and (Figure 11).
 The maximum WDBF peaked at 1.8 × 10−6 m2 s−3 for = 0.5 N m−2 and maintained at 0.6 × 10−6 m2 s−3 for N m−2. The maximum air-sea buoyancy flux, Jb, was an order of magnitude smaller at t = 67.8 compared to WDBF for N m−2. At the front, WDBF remains larger than Jb, which peaked at 2.6 × 10−7 W kg−1 during the storm, for N m−2. Thus, WDBF significantly supplements, and during short periods completely overwhelms, the buoyancy flux achieved by heat loss alone and was likely sufficient to trigger and intensify ageostrophic secondary circulations within lateral bands with high Ro.
 We have presented observations of the north Pacific subtropical front throughout a period of alternating buoyancy flux. The emergence of submesoscale intrusions in both the surface-mixed layer and underlying seasonal pycnocline implicate a distinct period of high wind stress in the generation of the intrusions. Specifically, the orientation of the wind corresponds with that of the frontal jet and results in a wind-driven buoyancy flux that exceeded the air-sea buoyancy flux by up to an order of magnitude. The apparent dynamics bear resemblance to more than one single process, however. The relative influence of the likely candidates in shaping the upper ocean during our observations is discussed below.
7.1. Wind-Driven Versus Convective Buoyancy Flux
 Based on broadly similar values for the lateral density gradient and along front velocity, Mahadevan and Tandon  show how Ekman pumping intensifies ageostrophic circulations within filaments of 2 km width characterized by high cyclonic vorticity. The filament properties are consistent with those observed at the STF, which we propose resulted from the WDBF and cooling. Yoshikawa et al.  compared the effects of heat loss and a WDBF in simulations for the Kurushio and found that surface subduction driven by ASC was intensified to a larger degree by surface cooling than by wind stress, but that the vertical circulation is greatest when both mechanisms act together. A key difference between the two mechanisms lies in the depth to which subducted fluid reaches; down-front winds transport surface water to the middle of the pycnocline but not beneath it, whereas surface cooling uplifts isopycnals from greater depth, increasing the depth of penetration of surface waters. The double pycnocline in our observations precludes a direct comparison of the depth to which the subduction occurred and thus the degree to which wind or cooling effects dominated. This constitutes an outstanding issue that may be addressed by an appropriately configured numerical modeling study in the future that faithfully replicates the complex vertical structure of the upper ocean in our observations.
7.2. Symmetric Instability
 If we further consider that the buoyancy loss due to cooling and the WDBF during the storm was sufficient to generate regions of negative EPV, downwelling due to frontogenesis would be further intensified by forced symmetric instability [Taylor and Ferrari, 2010; D'Asaro et al., 2011; Yoshikawa et al., 2012]. The SML evolves into two distinct dynamical regimes when a lateral density gradient is forced by either heat loss or destabilizing wind stress [Taylor and Ferrari, 2010]. The upper, near surface layer is essentially the traditional mixed layer within which both vertical and horizontal gradients are weak following homogenization by convection, as was the case in our observations immediately following the storm (Figure 7a). Below this, a second layer develops within which symmetric instability dominates and N2 is nonzero, lateral density gradients persist, and turbulence is generated by shear instabilities. This second regime is characterized by a bulk Richardson number, . The persistent Ri <1 regime observed within the high-vorticity filament to the east of the drifter is further consistent with observations from the Kurushio, where enhanced dissipation rates were attributed to unbalanced ageostrophic frontal flows arising from symmetric instability [Nagai et al., 2012]. While to an extent circumstantial, the sloping isopycnals throughout the SML in our observations are reminiscent of the slanting isopycnals in simulations of symmetric instability [Taylor and Ferrari, 2010; Thomas and Taylor, 2010], as is the clockwise rotation with depth in the uppermost 30–50 m. A definitive identification of the role played by symmetric instability would require similar large-scale surveys to Alpha to have been completed during and immediately following the storm to identify regions of negative PV.
7.3. Time Dependence: Near-Inertial Oscillations
 In addition to generating three-dimensional mixed layer instabilities, unsteady winds also generate time-dependent, near-inertial waves such as that generated by the storm during day 68 and which had surface currents of equal magnitude to the frontal jet. While our data do not enable us to directly estimate the contribution of time-dependent motions to an estimate of vertical motions in a manner similar to Thomas et al. , we note the clockwise rotation of the horizontal velocity with depth following the storm when the near-inertial currents were at their most intense. As suggested by D'Asaro et al. , it is therefore possible that the near-inertial motions play a role in the rapid confluence and divergence at the front and thus why the front was only weakly frontogenetic during Alpha became strongly frontogenetic 6 days later once the near-inertial wave had been generated. Alford et al.  recently attributed the generation of a near-inertial wave with a frequency of within 20% of f to the adjustment of the STF in the same region where the Rossby number was 0.2–0.3, similar to our observations. The downward propagating wave intensified ageostrophic shear in a manner that would also contribute toward the time-dependent motions implicated in the intrusion generation.
7.4. Mixed Layer Eddies and Restratification
 The rapid restratification of the SML following the storm would not be expected to result from forced ageostrophic circulations. Instead, our observations support the modeling results of Mahadevan and Tandon  that more than one process acts simultaneously to shape the upper ocean structure and evolution at submesoscales. In addition to the characteristics of the intrusions, the persistent lateral density gradients within the shallow SML and the associated cyclonic vorticity are consistent with baroclinic MLEs trapped in the upper 20 m arising from surface frontogenesis [Wang and Jordi, 2011]. Yoshikawa et al.  demonstrate that MLEs quickly restratify the SML following cooling and wind-forcing and are thus the likely mechanism facilitating restratification throughout the SML during phase C following the storm. The horizontal wavelength of the most unstable mode of MLEs is , where Uo is the along-front velocity [Fox-Kemper and Ferrari, 2008]. Values of Uo = 0.2 m s−1 and Ri = 1, as was repeatedly observed throughout the region where the subduction was observed, give k = 3 km and is therefore in good agreement with the scale of the observed lateral density gradients throughout the SML (Appendix B).
 The eventual fate of the intrusions in our observations is uncertain due to the cessation of sampling at the time when salinity gradients were observed to become weakened. The intrusion lost its coherent signature after approximately 5 inertial periods, however, which corresponds to the time at which a tracer subducted along an the isopycnal surface due to MLEs is predicted to undergo exponential growth [Badin et al., 2011]. The upper ocean was also clearly characterized by moderately high horizontal strain but large vertical shear, each of which would have contributed to the enhanced erosion of the intrusion. A definitive identification of the mechanisms leading to their diffusion is beyond the scope of the current study but warrants future study if similar features are able to be targeted in the ocean.
7.5. Impact on Water Mass Transformation
 A final note relates to the possible role played by submesoscales in water mass transformation. The current study site includes the region within which Eastern Subtropical Mode Water (ESTMW) is known to be created, yet the details of the processes primarily responsible for its transformation are as yet unknown. Suga et al.  suggest that ESTMW is formed by lateral induction and small cross-isopycnal flow, but later estimates of the annual subduction rate of ESTMW based on high-resolution climatology reveal that the low PV of ESTMW is due to either a small-density advection rate (cross-isopycnal flow) or a large subduction rate characterized by a thick winter mixed layer and mixed layer front [Suga et al., 2008]. Our observations suggest that the formation of ESTMW may be linked to submesoscale frontal processes along the STF in the same manner as Yoshikawa et al.  propose the cooling of a baroclinic current to impact on intermediate water formation through the generation of frontal downdrafts. Wind stress is ubiquitous, if somewhat irregular over a timescale of days, as is the front itself and the relative vorticity associated with frontal jet [Shcherbina et al., 2010]. The submesoscale vertical exchange that we have observed could thus be a quasi-permanent process throughout the STF and influence the formation of a regional water mass.
Appendix A:: Diurnal Restratification
 Following Hosegood et al. , the relative influence of insolation on SML restratification was quantified for a single daytime during phase C when the stabilizing air-sea heat flux, lack of nighttime convection, and weak wind stress suggests that the increase in near-surface N2 may have resulted from insolation. To diagnose the relative roles of slumping lateral buoyancy gradients throughout the SML (apparent as banding in during C) to restratification, the predicted increase in N2 due to the depth-dependant shortwave heat flux, Qshortwave, was estimated for day 71.
 The predicted decrease in at a depth of 15 m due to the integrated Qshortwave throughout the daytime of day 71 was only 0.01 kg m−3. Due to the near exponential decay of insolation with depth, the difference in between 15 and 30 m, (15–30 m), due to insolation remained significantly 0.01 kg m−3 (dashed line in Figure 12a) which is an order of magnitude smaller than the observed changes. The temporal fluctuations in Figure 12a are a result of the horizontal variability in the vertical density gradients as SWIMS was towed around the drifter. Over a distance of less than 10 km, 10−5 s−2 between 15 and 30 m due to strongly inclined isopycnals within the sample region. Therefore, we are unable to quantify the proportion of restratification due to insolation because the signal is so weak in comparison to existing density variability over horizontal scales of <10 km and vertical gradients between 15 and 30 m depth.
Appendix B:: Wavelet Analysis and Surface Lateral Density Gradients
 Following the same procedure as Hosegood et al. , lateral gradients within the SML of the temperature and salinity contributions in the linear equation of state, spice, and density were quantified at a depth of 15 m as a function of scale l and position xo using wavelet analysis. Wavelet coefficients [Torrence and Compo, 1998] of the temperature and salinity contributions to density are:
where is the horizontal gradient in potential temperature or salinity, is the complex conjugate of the Morlet wavelet, eiQx, and Q is the quality , which was chosen for consistency with Ferrari and Rudnick  and Hosegood et al. .
 The front at 30.1°N in Alpha was well defined, localized, and characterized by pronounced thermohaline gradients (Figure 13a). The magnitude of individual contributions of θ and S to the density change across the front at a distance of 70 km exceeded 0.5 kg m−3 over a horizontal distance of 10 km. They almost cancelled each other in their net effect on density; ρ increased from south to north by ∼0.1 kg m−3 over 20 km. Marked peaks in density were observed to the north of the front, specifically at 75, 95, and 120 km.
 In contrast, the front during Bravo was composed of two weaker thermohaline fronts at 18 and 37 km, across each of which the θ, S contributions to ρ were ∼0.2 kg m−3 and thus a total change in density of 0.4 kg m−3.
 The increase in the minimum horizontal scale of density variance during Bravo reflects the increase of LR (indicated by the white line in Figure 13) following the storm that depended on the surface mixed layer (and therefore H). During Alpha, density gradients equivalent to 0.01 kg m−3 at a wavelength of 10 km reached length scales of 1 km (the minimum resolved wavelength) immediately to the north of the front at 75 km (Figure 13b) and correspond with LR, further implicating MLEs in the SML dynamics.
 Direct observations indicate lateral density gradients near the front that are larger than those indicated by the wavelet analysis. Figure 14a illustrates the horizontal density gradients at a depth of 15 m during Group 21, which was conducted during the period of minimum τ as Tair dropped prior to the storm on day 68. Within individual legs, horizontal density gradients, , of 0.01 kg m−3 km−1 were observed over distances <10 km. Equivalent gradients were observed throughout the shallow surface layer above the seasonal pycnocline throughout the period between Alpha and Bravo, providing direct quantitative measurements of the magnitude and scale of lateral density gradients at the surface.
Appendix C:: Near-Inertial Velocities
 Time-dependent motions preclude application of the quasi- and semigeostrophic omega equations. Near-inertial oscillations are a common feature of the upper ocean and can furthermore be trapped within the negative vorticity troughs of frontal jets [Kunze and Sanford, 1984], leading to an intensification of their associated vertical shear. The impulsive, clockwise rotating wind stress on day 68 generated a near inertial oscillation whose associated currents persisted until day 73, corresponding to a duration of five inertial periods (Figure 15). The magnitude of the near-inertial currents, , reached a maximum of 0.1 m s−1 just below the depth of the drifter's drogue (38 m), thereby causing the drifter to stall every 24 h as was directed to the west, opposing the eastward flowing frontal jet. Maximum were observed at the base of the SML but energy propagated beneath the seasonal pycnocline to the underlying, weakly stratified layer that separated the pycnoclines.
 Despite maximum at 50 m, the largest shear associated with the near inertial velocities, , occurred at 120 m as the deeper energy was rapidly attenuated within the permanent pycnocline (Figure 15b). Maximum was similar at both 50 and 120 m depth, however, reaching 3.5 × 10−4 s−2 throughout the period t = 68.5–70.5 when the intrusions were generated but over a broader depth and longer time within the deeper layer.
 Velocities at 100 m depth were approximately half those at 50 m and exhibited a 180° phase shift relative to the surface. The corresponding vertical wavelength of ∼200 m is double that observed at the front at 28°N during the same cruise [Hosegood et al., 2008] and by Kunze and Sanford , also at the STF, but very close to that observed by Alford et al. . In the two former cases, the near-inertial waves were generated in the negative vorticity trough on the warm side of the front and were required to shorten their vertical wavelength to satisfy the dispersion relation
where the effective Coriolis frequency
is the planetary vorticity plus half the geostrophic vorticity, ζ. As the wave propagates downward, feff is reduced due to weakening of the geostrophic vorticity, causing a reduction in vertical wavelength and an increase in wave shear. The wave in our observations appears to have been generated in the region of positive vorticity associated with the front (Figure 4), assuming that the drifter remained at the northern edge of the frontal jet. Thus, while near-inertial currents clearly play a significant role in modulating the total velocity field, they do not conform to previous observations of trapped waves at fronts but certainly introduce the complication of time-dependent motions to the interpretation of the submesoscale fields.
Appendix D:: Double Diffusive Interleaving
 Intrusions arising from double diffusion (DD) have been shown to be hydrostatically stable despite potentially sloping across isopycnals [Roden, 1964; Fedorov, 1978; Gregg and McKenzie, 1979; Gregg, 1980], as was observed here in the case of the surface intrusion. The vertical salinity structure of the STF in our observations was, as is common throughout the subtropics, conducive to the salt-fingering regime of DD, while the spatial dimensions of the subsurface saline intrusion are consistent to within an order of magnitude with previous observations. The vertical thickness, H, of isopycnal intrusions formed through DD are predicted to be
where n = 0.56 is the density flux ratio and is the lateral density change due to salinity [Ruddick and Turner, 1979]. From an observational perspective, the background density gradient in equation (D(1)) is difficult to quantify accurately yet introduces significant uncertainty to the predicted thickness of the resulting intrusion. Taking values of the lateral salinity and vertical density gradients from Alpha as the initial state in which the intrusion was formed gives H of O(100 m). This is an order of magnitude larger than typical estimates due to the large salinity gradient in our case; for example, Gregg  found H = 7.9 m at Ocean Weather Station P during late summer at 50°N for the same vertical density gradient. The smaller thickness was due to a lateral change of salinity of only 0.04, which is clearly much smaller than the change across the front of 0.4 in the present case.
 The slope of the fresher intrusion within the SML was impossible to ascertain with any certainty due to the complexity of its orientation. Based on groups 33 and 39, we estimated the slope as between 0.013 and 0.007. The subsurface saline intrusion exhibited no slope and was horizontal as far as our observations were able to identify. Intrusive motions may be driven by baroclinicity if the intrusion slope lies between the horizontal and the slope of the isopycnal surfaces referred to as the “wedge” of baroclinic instability [May and Kelley, 2002]. The horizontal orientation of the subsurface intrusion would appear to preclude this mechanism, however. A further possibility is the role of near-inertial velocity perturbations in driving the intrusions, a mechanism suggested for generating intrusions within the Agulhas current [Beal, 2007]. While displaying some consistencies with our observations, the occurrence here of the subduction and upwelling within a thin filament characterized by elevated ζ suggests that an alternative mechanism is more appropriate.
 The fundamental inconsistency of the observed intrusion with DD interleaving lies in the implied downward displacement of the subsurface intrusion based on its θ, S properties and its appearance as an isolated feature rather than an intrusive tongue connected to its source. Assuming that the intrusion did not originate from the same depth at the warm side of the front, the θ, S properties of the intrusion indicate that it must have originated at the surface and been displaced across the strongly stratified seasonal pycnocline within a timescale of 1–2 days given the lack of any intrusions prior to the storm. It is well understood, however, that for warm-saline intrusions, the buoyancy flux associated with the salt fingering (SF) on the underside of the intrusion generally exceeds that due to diffusive convection (DC) at the upper interface, and that as a result the intrusion crosses isopycnals as it slopes upward [Turner, 1978]. Such a scenario would require the observed intrusion here to have originated at a depth of >70 m on the warm side of the front, a requirement for which there is no evidence. DC can cause warm saline anomalies to sink if the vertical density ratio, , falls within the range [Kelley et al., 2003]. A small region at the surface exists within the criteria for DC are satisfied but corresponds to the cooler, fresher water to the north of the front from where the water contained in the subsurface intrusion is not derived (Figure 16). The front itself is characterized by conducive for SF, and we thus consider the DD was not responsible for the intrusions.
 This work was funded by the U.S National Science Foundation under grant OCEO326280. We are indebted to Frank Bradley for setting up the rain gauges and providing their outputs and to Carter Ohlmann for computing the solar transmission profiles from the SPMR data. We thank Jack Miller, Eric Boget, Paul Aguilar, John Mickett, Avery Snyder, Glenn Carter, Dave Winkel, Steve Bayer, and Andrew Cookson for their help in collecting the data and the captain and crew of the RV Wecoma for their assistance during difficult conditions.