Direct Observations of Coastally Generated Near‐Inertial Waves During a Wind Event Authors: Samuel M. Kelly, Erica L. Green, Ian A. Stokes, Jay A. Austin, Andrew J. Lucas, Jonathan D. Nash Published: 2024 ABSTRACT AbstractWind over the ocean generates near‐inertial velocities. In the open ocean, horizontal variability in the inertial frequency and mesoscale vorticity generate internal waves that transport energy laterally and drive diapcynal mixing in remote locations. In the coastal ocean, horizontal variability is produced by the coastline. This study analyzes observations along a straight coastline in Lake Superior, which acts as a “natural laboratory” for the coastal ocean. Depth‐profiles of velocity, temperature, and turbulent miscrostructure were collected during a 96 hr repeat survey from 3 to 20 km offshore in Aug 2018. Wind work was 2 mW m−2 and generated 0.2 m s−1 near‐inertial velocities that were inhibited within two internal Rossby radii (6 km) of the coast. The velocities are interpreted as a superposition of a “forced flow”, which is horizontally uniform, and a “wave flow”, associated with offshore propagating near‐inertial waves. A 1D momentum equation skillfully predicts r2=0.82 the horizontally averaged near‐inertial velocities and the TKE shear production, which matches the 1 mW m−2 observed TKE dissipation rate. The offshore propagating wave has an energy flux of 10 W (m‐coastline)−1 and a downward energy flux of 1 mW m−2. These results suggest that most near‐inertial wind work is lost directly to TKE shear production, but some energy is transferred to offshore propagating waves that may help catalyze shear instability away from the coast. Introduction Steady and slowly varying winds over the ocean produce Ekman transport, while abrupt changes in the wind generate near‐inertial motions. Near‐inertial motions have large amounts of kinetic energy and produce most of the velocity shear in the upper ocean (Alford et al., 2016; Thomas & Zhai, 2022). They contribute to upper ocean mixing, which sets the mixed‐layer depth, transition‐layer depth, and upper‐ocean stratification (Large & Crawford, 1995). Upper‐ocean mixing also impacts surface temperature and velocities, which control the air‐sea exchanges of heat and momentum (Csanady, 2001). A detailed understanding of near‐inertial motions is thus important to accurately predict seasonal stratification and future climate. These challenges are amplified in large lakes and the coastal ocean where the ocean surface boundary layer (OSBL; note that we use standard oceanographic terms to describe Lake Superior in this manuscript to emphasize the relevance of the study to oceanography) composes a large fraction of the water volume, so that surface fluxes have large and immediate impact on the vertical profiles of heat and momentum (e.g., Austin & Allen, 2011; Bai et al., 2013). The frequency and vertical wavenumber spectra of near‐inertial motions are set by the wind stress [τ(z=0,t)], inertial frequency (f), and upper ocean buoyancy frequency [N(z)]. The first‐order frequency response is that of a forced harmonic oscillator (Pollard & Millard, 1970). Near‐inertial motions are excited by both impulsive wind events at mid latitudes (D’Asaro, 1995) and oscillatory winds at latitudes where the diurnal sea‐breeze is near‐inertial (Lucas et al., 2014). The depth to which the wind stress penetrates and accelerates horizontal velocities in the OSBL sets the vertical scales of motion (D’Asaro et al., 1995). Small vertical scales are excited when the mixed layer is shallow and there is little velocity below the mixed layer. Larger vertical scales are excited when the mixed layer is deep and/or wind energy penetrates into the stratified transition layer beneath the mixed layer. Near‐inertial motions propagate as internal waves if their frequency, vertical wavenumber, and horizontal wavenumber lie along the internal‐wave dispersion curve. Strong winds are often associated with large‐scale (e.g., 500‐km diameter) storms, so that only internal waves that are very near the inertial frequency are generated (Gill, 1984). Such waves have small horizontal and vertical group speeds, causing them to propagate slowly and dissipate locally. Other mechanisms exist that produce smaller‐scale horizontal variability, generating superinertial internal waves that propagate with faster group velocities. For example, near‐inertial motions may dephase due to variations in f due to latitude (the β effect; D’Asaro et al., 1995) and mesoscale vorticity (the ζ effect; Thomas et al., 2020). A third, more abrupt, mechanism for internal‐wave generation is the presence of a coastline (Kelly, 2019; Kundu et al., 1983; Millot & Crépon, 1981; Pettigrew, 1980). In Lake Superior, the two‐layer structure means that internal waves have a vertical mode‐1 structure (Austin, 2013). Observations of the ratio of kinetic‐to‐potential energy indicate a dominant frequency of about 1.05f−1.10f and horizontal wavelength of about 60 km (Austin, 2013). As near‐inertial motions impinge on the coast, the no‐normal‐flow condition leads to a reduction in near‐inertial velocities, which we call “coastal inhibition”. The divergence/convergence of cross‐shore velocity leads to the upwelling/downwelling that pumps the base of the mixed layer and radiates coastally generated near‐inertial internal waves (CNIWs). Pettigrew (1980) and Millot and Crépon (1981) developed a two‐layer theory describing the response to an abrupt change in long‐shore wind stress and compared it with mooring data (see also Baines, 1986). The theory was later revisited by Kundu et al. (1983), who extended it to continuous stratification, and Shearman (2005), who compared the theory with additional observations collected by moorings on the New England Shelf. CNIWs also occur in the (non‐tidal) Baltic Sea, and have been associated with both open water (van der Lee & Umlauf, 2011) and boundary mixing (Lappe & Umlauf, 2016). These previous analyses have identified coastal inhibition of near‐inertial velocities, time lags between wind and density perturbations due to offshore wave propagation, and polarization relations between density perturbations and velocities that were consistent with theory. However, previous observational analyses have not been able to separate the directly forced motions from the propagating internal waves, or quantify the energy fluxes associated with offshore (and downward) propagating near‐inertial internal waves. Kelly (2019) revisted the theory of CNIWs by extending it to continuous wind forcing and formally separating the evolution equations for the forced and internal wave motions. He also quantified CNIW energy fluxes due to continuous winds using both analytical solutions and numerical simulations over arbitrary coastal topography. The results indicated that 0.03 TW may go into CNIWs globally, although energy fluxes are typically less than 100 W (m coastline)−1. Kelly (2019)'s attempt to isolate CNIWs in mooring observations over the New Jersey continental shelf was largely inconclusive, possibly because of “contamination” by non‐stationary internal tides. CNIWs are predicted to be ubiquitous in the coastal ocean, but are difficult to isolate and quantify in complex settings. Here we examine summer observations of near‐inertial motions along Minnesota's North Shore of Lake Superior. This unique setting provides an ideal “natural laboratory” for observing CNIWs because (a) Lake Superior has energetic near‐inertial velocities and negligible internal tides (Austin, 2013), (b) the coast is straight and has little long‐shore variability (Figure 1), and (c) the coast is essentially a vertical wall, reaching 125 m depth less than 3 km from shore. A shipboard survey captured the time evolution of the flow along the cross‐shore and vertical axes, leaving only long‐shore variability unsampled. Here we use this novel data set to address several questions.What are the relative strengths and spatial structures of the subinertial and near‐inertial flows near the coast?What are the energy pathways from wind work to turbulent kinetic energy (TKE) dissipation?Is the observed flow consistent with the theory of CNIW generation? And, is the observed CNIW energy flux consistent with theory?The answers to these questions are critical for understanding how near‐inertial energy propagates downward and offshore along coastlines, redistributing the kinetic energy that is available to drive diapycnal mixing in large lakes and throughout the global coastal ocean. What are the relative strengths and spatial structures of the subinertial and near‐inertial flows near the coast? What are the relative strengths and spatial structures of the subinertial and near‐inertial flows near the coast? What are the energy pathways from wind work to turbulent kinetic energy (TKE) dissipation? What are the energy pathways from wind work to turbulent kinetic energy (TKE) dissipation? Is the observed flow consistent with the theory of CNIW generation? And, is the observed CNIW energy flux consistent with theory? Is the observed flow consistent with the theory of CNIW generation? And, is the observed CNIW energy flux consistent with theory? Western Lake Superior is 100–200 m deep and has steep coastal bathymetry. The survey was conducted along a 18 km cross‐shore line just west of Grand Marais, MN. NOAA buoy 45006 is located at the Western Mooring (WM) site. Review of CNIW Theory Equations of Motion Near‐inertial currents in Lake Superior are primarily baroclinic because the 100–200 km wide western‐basin is smaller that the barotropic deformation radius (λ0=gH/f≈350 km, where g is gravity and H is depth), meaning barotropic near‐inertial waves generated at opposing coastlines cancel each other within a fraction of inertial period (e.g., Shearman, 2005; van der Lee & Umlauf, 2011). Furthermore, our observations (Section 3) indicate that baroclinic currents account for 86% of near‐inertial kinetic energy. Baroclinic motions can be further decomposed into vertical modes based on the background stratification (e.g., Gill, 1982; Kelly, 2016). The result (not shown) is that 85% of baroclinic energy is associated with the first vertical mode (E. Green, personal communication), because the background stratification is approximately two layer (Austin, 2013). The linear, Boussinesq, hydrostatic, baroclinic equations of motion are. (1a) ∂tu+fkˆ×u=−∂xp+∂zτ (1b) 0=−∂zp+b (1c) ∂tb+N2w=0 (1d) ∂xu+∂zw=0 where long‐shore variability is ignored ∂y≈0, u=(u,v) is horizontal velocity, w is vertical velocity, p is pressure, b is buoyancy, and τ is the sum of turbulent (Reynolds) and viscous stress. Pressure, buoyancy, and stress are normalized by a reference density ρ0=1000 kg m−3. The surface and bottom are located at z=0 and z=−H(x), respectively. Baroclinic u, p, and ∂zτ are obtained by removing depth averages, while b and w are entirely baroclinic (Kelly, 2016). Barotropic‐baroclinic topographic coupling is neglected because the bottom buoyancy frequency and topographic slope are weak (N<10−4 s−1 and ∂xH<1%, respectively) offshore of 3 km, also barotropic currents are weak. The energy equation is formulated by multiplying Equation 1a by u, Equation 1b by w, summing, and depth integrating (2) τ⋅u0⏟W=∂t∫−H012|u|2+12b2N2dz⏟E+∂x∫−H0updz⏟F+∫−H0τ⋅∂zudz⏟PS+τ⋅u−H⏟D, where the stress term has been integrated by parts. The equation has been rearranged so that the total rate of wind work (W) is equal to the sum of energy sinks. The sinks from left to right, roughly in order of importance, are the energy tendency, cross‐shore energy flux divergence, TKE shear production, and energy dissipation by bottom‐drag. Microstructure profiles of TKE dissipation do not extend to the bottom boundary layer, so we implement a free‐slip boundary condition and parameterize the unresolved bottom boundary layer using a quadratic drag law. Energy dissipation in the bottom boundary layer is (3) D=ρ0CD|u|−H3, where CD=0.0025 is a typical drag coefficient and bottom velocity is the average velocity over the weakly stratified bottom layer (100–150 m depth in Lake Superior). Energy loss to total bottom drag is small during the experiment (D=0.17±0.09 mW m−2) because currents in the bottom layer are weak. Since bottom drag is small and non‐linear (i.e., it does not naturally separate into barotropic and baroclinic components), we simply omit bottom drag from our subsequent energy balances and treat it as a source of uncertainty. Forced and Wave Flows Approximating the stress as spatially constant, the equations of motion can be averaged in space to form a forced flow, driven by wind stress, (4) ∂tuF+fkˆ×uF=∂zτ and a residual flow, representing inviscid internal waves, (5) ∂tuW+fkˆ×uW=−∂xp, where u=uF+uW, so summing Equations 4 and 5 recovers Equation 1a. The utility of the separation is that Equation 4 is a 1D harmonic oscillator (at each depth) and uW can be combined with Equations (1b), (1c), (1d) to form the wave equation (6) ∂tt+f2∂z1N2∂zuW+∂xxuW=0 which is mathematically analogous to the wave equation that governs internal tides (Llewellyn Smith & Young, 2002). Note that p, b, and w are components of the wavefield because τ and uF are spatially uniform and produce no vertical velocity. The wave equation is only forced through the kinematic boundary condition at the coast u=0 at x=0, which requires uW=−uF at x=0. Kelly (2019) solved Equations 4 and 6 analytically for continuous winds and arbitrary stratification using Laplace transforms and normal modes, respectively. The individual flows have simple energy balances for horizontally uniform stress. The depth‐integrated energy equation for the forced flow is (7) τ⋅uF0⏟WF=∂t∫−H012|uF|2dz⏟EF+∫−H0τ⋅∂zuFdz⏟PSF (omitting bottom drag). The equation indicates that wind work WF either increases kinetic energy ∂tEF or drives TKE shear production PSF. The energy equation for the wave equation is (8) −∂x∫−H0uWpdz⏟FW=∂t∫−H012|uW|2+12b2N2dz⏟EW, where wave energy‐flux convergence −∂xFW is an energy source that must correspond to a local increase in wave energy ∂tEW. The separate energy equations, Equations 7 and 8, are simpler to interpret than the total energy equation (Equation 2) because they ignore cross terms that contribute to total wind work and energy flux (e.g., τ⋅uW0 and uFp). Cross terms in the energy flux point in non‐intuitive directions and oscillate at the horizontal scale of the waves, but they have little meaning because they average to zero over a wavelength and are perfectly canceled by the cross‐terms in the wind work (Kelly, 2019). The difficulty of interpreting cross‐terms in the energy flux has been previously examined in detail with respect to standing waves (e.g., Nash et al., 2004). By approximating stress as spatially uniform, the above decomposition ignores the non‐linear nature of turbulent stress. In reality, shear production is driven by instabilities that depend on the full velocity and cannot be decomposed into forced and wave components. Therefore, we simply use the preceding decomposition as a straw man, and expect that PSF may explain some, but not all, of the observed TKE dissipation. Observations The R/V Blue Heron repeated an 18 km cross shore out‐and‐back transect 20 times during a 96 hr period starting 20 Aug 2018 (Figure 1). The average ship speed was 4 kts (2 m s−1). Wind and atmospheric pressure were measured on the ship mast (10 m height) and about 50 km southeast at NDBC buoy 45006. Wind stress was computed from wind speed following Large and Pond (1981). During the survey, the wind shifted from downwelling favorable (east) to upwelling favorable (west) and the wind stress increased from 0 to 0.1 N m−2 (Figure 2). Buoyancy flux was computed from the meteorological buoy observations using clear sky radiative fluxes (Dickey et al., 1994) and bulk estimates of sensible and latent heat flux (Fairall et al., 1996). Winds stress was weak prior to the 96 hr survey that started on 20 AUG 2018 (gray shaded region, with triangles indicating when this ship was at the offshore end of the line; a). Wind direction shifted from easterly (downwelling favorable) to westerly (upwelling favorable) on 21 AUG (b). The surface buoyancy flux was 10−8−10−7 W kg−1 during nighttime cooling (positive values indicate nighttime convection in the mixed layer, negative/stabilizing daytime values are not shown). A Rockland VMP‐250 profiled microstructure velocity shear, temperature, and fluorescence to approximately 40 m depth every 2–5 min, corresponding to 250–500 m horizontal resolution. The instrument was loosely tethered to the ship using fishing line so that it freefell at approximately 0.8 m s−1. It was recovered using a 12 V electric fishing reel. The VMP was periodically recovered to charge its battery and download data. While the VMP charged, an RBRduet fast (16 Hz) T.D sensor profiled temperature and pressure to 40 m depth every 2 min. In total, we collected 1377 VMP and 522 duet profiles during the survey. The VMP data were processed following Cannon et al. (2021) to obtain the TKE dissipation rate. We relate the TKE dissipation rate to shear production, by assuming a steady state TKE balance of PS+PB=∫−H0ϵdz and a constant flux Richardson number, Rif=−PB/PS≈0.15, where PB is TKE buoyancy production and Rif is sometimes called the “mixing efficiency” (Moum, 1996). We note that while Rf≈0.15 is typical, observed values have ranged from 0 to 0.5 (Ivey et al., 2018). The observational estimate of shear production is (9) PS,obs=11−Rif∫−H0ϵdz. The R/V Blue Heron has a hull‐mounted RDI Broadband 150 kHz (BB150) acoustic Doppler current profiler (ADCP). Currents were obtained in 2 m bins and averaged every 2 min using UHDAS processing software. Reliable data were collected from 8 m depth to 8 m above the bottom (150 m). Data closer to the boundaries were contaminated by sidelobe interference. Since these regions are generally well mixed (see Section 4.1), we extrapolated velocities to the surface and bottom using nearest neighbors. The east‐north velocities were rotated into cross‐ and long‐shore components by minimizing the barotropic (depth averaged) flow variance in the cross‐shore direction. The cross‐shore direction was found to be 77° clockwise from east, similar to the transect angle of 63°, so long‐ and cross‐transect quantities only required a 14° rotation into the long‐ and cross‐shore directions. In order to construct a composite picture of horizontal, vertical, and temporal variability, the velocity and meteorological data were slightly smoothed at the timescale of VMP/duet casts and linearly interpolated onto the time of each cast. Vertical variability was interpolated into uniform 0.1 m bins. Next, casts were binned into half‐overlapping 0.5 km cross‐shore bins. Observations from individual transects (typically 3–6 casts in each 0.5 km bin) were averaged and then data from separate transects were linearly interpolated onto an hourly grid. Background temperature was computed as the average depth of each isotherm to prevent finescale smoothing (Desaubies & Gregg, 1981). Temperature was converted to density using the freshwater equation of state (Chen & Millero, 1986) and the buoyancy frequency was estimated following McDougall and Barker (2014) and using linear least squares fits over 1‐m to estimate the stable density gradient in the presence of noise and small overturns. Baroclinic pressure was computed by depth integrating the density perturbation (from the hydrostatic background density) multiplied by gravity and removing the depth average. Barotropic pressure is unknown because it depends on the sea surface height (SSH), which was not observed. Buoyancy frequency was extended to the bottom using nearest‐neighbor interpolation so that vertical modes could be calculated following Kelly (2016). Results The full time series of temperature, velocity, and TKE dissipation rate are difficult to interpret because periodicity occurs due to both the ship path and inertial oscillations (Figure 3). Short‐period (4–5 hr) oscillations in the time series are prominent due to the ship passing through cross‐shore variability as it steamed on and offshore. After the wind stress increase at 60 hr (Figure 2), near‐inertial motions drive the largest isotherm displacements, velocities, and TKE dissipation in the record (Figure 3). To better interpret the data, we sort the observations by transect and examine their simultaneous cross‐shore and temporal variability. Timeseries of temperature (a), velocity (b and c), and TKE dissipation rate (d) during the 96 hr survey. Triangles indicate when this ship was at the offshore end of the line. The arrow on the dissipation color bar indicates the instrument noise level (7×10−10 W kg−1). Mean Water Column Structure Mean temperature is about 17° C at the surface and 4° C at the bottom, with a thermocline between 8 and 25 m depth (Figure 4a). Stratification is strong in the thermocline (N=0.03 s−1, indicating a 3 min buoyancy period) and weak below 50 m (N≈10−4 s−1 or 17 hr period; Figure 4b). Because the deep buoyancy frequency is lower than the inertial frequency, freely propagating internal waves do not reach the bottom. The thermocline is slightly wider near the coast, implying either enhanced mixing or isopycnal straining by coastal currents. Vertical modes have large curvature from 8 to 25 m depth (Figure 5). The mode‐1 eigenspeed is c1=0.33 m s−1 and higher modes are much smaller (i.e., less than 0.1 m s−1) confirming the quasi‐two‐layer stratification and justifying the extrapolation of velocity profiles to the surface and bottom boundary. The mode‐1 deformation radius is λ1=c1/f=3.1 km. Mean temperature during the survey ranges from 4–16°C (a). VMP and duet profiles were collected to 50 m and the local depth is 150 m. Stratification (Buoyancy frequency squared; N2) is maximum at 10 m and somewhat more diffuse near the coast (c). Shear squared S2 is strongest in the surface mixed layer and upper thermocline. S2/4 is plotted so that values larger than N2 indicate stratified shear instability, Ri<1/4. The TKE dissipation rate is largest in the surface mixed layer (10−7 W kg−1), decreases through the thermocline, and is weakest in the deep water (10−9 W kg−1; d). Isotherms are plotted every 2°C from 6–16°C. The arrow on the dissipation color bar indicates the instrument noise level (7×10−10 W kg−1). The mean temperature profile 10 km offshore displays a two layer structure (a) with strong stratification at 12 m (b). Vertical modes have curvature between 5 and 25 m depth and eigenspeeds cn that decrease rapidly with mode number (c). Velocity shear squared S2 is highest in the mixed layer and upper thermocline (Figure 4c), consistent with a turbulent mixed layer over a strong thermocline. Shear extends offshore, suggesting it is driven by open lake near‐inertial velocities rather than mean coastal currents. The water column is typically stable to stratified shear instability (i.e., Ri>1/4 or N2>S2/4). Noise in the velocity shear is about 5×10−4 s−2, preventing an accurate estimate of stability in the weakly stratified deep water. The mean TKE dissipation rate is ϵ=10−7 W kg−1 near the base of the mixed layer and is approximately uniformly distributed across the transect, implying that lake‐wide processes (as opposed to coastal processes) drive mixed‐layer turbulence (turbulence data is removed above 5 m depth due to ship‐wake contamination). Below the mixed‐layer, turbulence monotonically decreases to ϵ=7×10−10 W kg−1 at the base of the thermocline, which we interpret as the VMP noise floor. Depth integrals of this noise are about 0.1 mW m−2, slightly smaller than uncertainty due to unobserved bottom drag (Section 2.1). Subinertial Flow The transects occurred at 47.5° N, where the inertial period is 16.2 hr (the inertial frequency is f=1.08×10−4 s−1). Filtering the total flow with a 32 hr running mean isolates the sub‐inertial flow. The subinertial cross‐shore flow is negligible (about 0.01 m s−1; not shown) and the long‐shore flow is primarily baroclinic, reaching 0.1 m s−1 (Figure 6). The baroclinic long‐shore jet shifts from negative (westward toward Duluth, MN) to positive (eastward toward Thunder Bay, ON) as the winds shift from downwelling to upwelling favorable. The isotherms upwell along the coast, eventually ventilating the 16°C isotherm. Upwelling is contained within about 6 km of the coast, which is about two mode‐1 Rossby radii. Geostrophic currents determined from cross‐shore baroclinic pressure (not shown) closely match the low‐passed current observations. Long‐shore subinertial currents (low‐passed with a 32 hr running mean) are separated into barotropic (left) and baroclinic (right) components. Baroclinic components shift from westward (downwelling favorable) to eastward (upwelling favorable) as the winds shift. Baroclinic currents are consistent with geostrophic currents (not shown) determined from baroclinic pressure gradients. Isotherms are plotted every 2°C from 6–16°C. Subinertial flow is most significant for long‐shore transport. Barotropic and baroclinic currents of 0.02 and 0.1 m s−1 correspond to daily horizontal displacements of 2 and 9 km, respectively. However, subinertial flow is quite weak compared to the near‐inertial flow. Thus, subinertial flow produces little velocity shear and is unlikely to drive TKE dissipation on its own. The sub‐inertial energy balance is constructed by low‐passing the variables with a 32‐hr running average, computing the energy terms, and smoothing over a 32‐hr window. Sub‐inertial wind work is primarily balanced by the energy tendency (Figures 7a and 7b). That is, wind work lifts the isotherms and increase the energy in the geostrophic coastal jet. The increase in the strength of the coastal jet is also associated with a small offshore transport of energy, which results in a flux divergence at 2–3 km and a flux convergence at 4–5 km (Figure 7c). This is not due to wave‐radiation, but rather the offshore expansion of the coastal jet. PS is assumed negligible because it is driven by shear which is dominated by near‐inertial motions. Time‐averaging the terms over the survey indicates about 1 mW m−2 of sub‐inertial wind work is balanced by a similar increase in energy near the coast. Mismatch between sources and sinks indicate that uncertainty in the balance is also about 1 mW m−2, somewhat higher than the noise floor of 0.2 mW m−2 due to VMP noise and neglecting bottom drag. The subinertial energy balance is between wind work (ρ0τ⋅u) (a) and energy “sinks” due to energy tendency (∂E/∂t), (b) and flux divergence (∂xF), (c). Most wind work and energy gain occur within two Rossby Radii (6 km) of the coast (d). The dashed line at 0.2 mW m−2 indicates the uncertainty due to VMP noise and neglecting bottom drag. Near‐Inertial Flow Near‐inertial (high passed with a 32 hr running mean) velocity amplitudes are 0.05 m s−1 prior to the wind event around 60 hr. From 60 to 88 hr near‐inertial velocities reach 20 cm s−1 (Figure 8). Onshore near‐inertial flow (60 and 76 hr; Figure 8) depresses the isotherms leading to positive buoyancy in the thermocline (64 and 80 hr; Figure 9). Offshore flow (68 and 84 hr; Figure 8) raises the isotherms leading to negative buoyancy in the thermocline (72 and 88 hr; Figure 9). This back and forth motion pumps the thermocline up and down by about 5 m at the coast. Inertial velocities decay to zero within 6 km of the coast consistent with coastal inhibition (Shearman, 2005). Cross‐shore near‐inertial velocities (high‐passed with a 32 hr running mean) have 16 hr period. Onshore flow depresses isotherms and offshore flow elevates isotherms. During slack periods (64, 73, 80, and 88 hr), residual velocities have shorter wavelength and slope downward, consistent with coastally generated internal waves. Isotherms are plotted every 2°C from 6–16°C. As in Figure 8, except colors indicate near‐inertial buoyancy perturbations. During periods of slack cross‐shore flow (64, 72, 80, and 88 hr), residual cross‐shore velocities are consistent with coastally generated internal waves because they have shorter lateral scales and move downward with distance offshore. Successive periods of slack flow reveal features that have moved offshore by about 5 km per inertial period, indicating a group speed of about 0.1 m s−1 or a frequency of about 1.05f for a mode‐1 wave (cg1=1−f2/ω2c1, where ω is the wave frequency). These waves perturb should perturb the thermocline by a few meters as they propagate, but offshore and downward propagation of near‐inertial buoyancy is less clear (Figure 9). The near‐inertial energy balance is constructed by high‐passing the variables with a 32‐hr running average, computing the energy terms, and smoothing over a 32‐hr window. Near‐inertial wind work is primarily balanced by the energy tendency (Figures 7a and 7b). In particular, the large wind event around 60 hr increases wind work and near‐inertial energy. About one inertial period later, wind work is negative, so the wind decelerates the flow. Energy‐flux divergence is about five times smaller than wind work and is both positive and negative. Observed shear production PS,obs is also smaller than wind work, but 1 mW m−2 during the wind event. All of the energy terms are typically larger than the uncertainty due to bottom drag and VMP noise (0.2 mW m−2). Averaging the near‐inertial energy balance over the entire survey produces a null result because positive wind work at the start of the wind event is largely canceled by negative wind work at the end of the event. Therefore, the energy balance is just time‐averaged from 48 to 80 hr, when wind work is positive. During this time, wind work is about 2 mW m−2 and decays to zero within 6 km of shore (Figure 10e). Wind work is balanced by a combination of increasing near‐inertial energy, offshore energy flux convergence, and PS,obs=0.5 mW m−2, which is nearly constant in the cross‐shore direction. The near‐inertial energy balance is between wind work (ρ0τ⋅u); (a) and energy “sinks” due to energy tendency (∂E/∂t), (b), flux divergence (∂xF), (c), and shear production (PS) (d). Most wind work and energy sinks occur two Rossby Radii (6 km) offshore of the coast (e). The dashed line at 0.2 mW m−2 indicates the uncertainty due to VMP noise and neglecting bottom drag. The Forced Flow We isolate the forced velocity by horizontally averaging the observed velocity from 6 to 20 km offshore. The near‐coast region is excluded so that the coastal inhibition zone and geostrophic jet do not bias the offshore average. Averaging is a low‐pass filter with a filter width equal to the domain size. Thus, it does not remove waves with wavelengths much larger than the domain. To estimate and remove the average associated with long waves, we use the horizontally averaged velocities to force Equation 6 and numerically solve for the wave velocities following Kelly (2019). We then compute the horizontal average of the modeled wave velocity and subtract it from the observed forced velocity (This correction could be refined over several iterations, but we found that iterative solutions diverged unacceptably from the initial estimate). In practice, the horizontally averaged speeds are about 0.2 m s−1 and the correction due to averaging long waves is about 0.08 m s−1, although the largest wave corrections are in quadrature with the horizontal averages, so they amount to a slight phase shift rather than a 40% amplitude modulation. The forced velocity below the surface is almost entirely in phase with the surface velocity (allowing for a 180° phase shift across the thermocline due to the absence of barotropic flow). Ignoring the out‐of‐phase velocity signal, which is an order of magnitude smaller than the in‐phase signal, allows us to separate the time and depth dependence of forced velocity uF(z,t)=uˆF(t)∂zΣ(z). Substituting this expression into Equation 4 reveals that stress can be written τ(z,t)=τ(0,t)Σ(z), where Σ(z) is the baroclinic stress profile and τ(0,t) is the wind stress. Velocity data can be used to estimate ∂zΣ(z), but an integration constant is needed to determine Σ(z). We determine this constant by noting that ∂zΣ(z) has zero depth average, so Σ(0)=Σ(−H), and ignoring bottom stress requires Σ(0)=0. Note that because the total stress at the surface equals the wind stress, this normalization implies that the barotropic stress profile is normalized to one at the surface (see Stokes et al., 2024, for details on of projecting stress profiles on to vertical modes). The observed stress and stress‐divergence profiles were obtained by normalizing every profile of uF(z,t) so that the surface value was 1 m s−1, and then time‐averaging the depth profile to obtain ∂zΣ(z) and then Σ(z). The observed forced velocity is less than 0.1 m s−1 at the surface until the storm at 64 hr when it increases to 0.2 m s−2 (Figure 11). The stress and stress‐divergence profiles (Figures 11c and 11d) decrease monotonically with depth, consistent with a mixed layer to 8 m depth (also imposed by the velocity extrapolation) and a transition layer from 8 to 15 m. The stress divergence profile is almost exactly the same as the mode‐1 structure function (Figure 11d), indicating the wind preferentially excites mode‐1 velocities and little else. The forced surface velocity (blue) is well predicted by the generalized slab model (red) in both the cross‐shore (solid) and long‐shore (dashed) directions (a). Most wind work in the slab model goes toward increasing the forced kinetic energy. A smaller amount is lost to TKE production due to shear in the forced velocity (b). Modeled TKE production is similar to observed TKE dissipation. The implied stress profile decreases linearly through the mixed layer and quadratically in the transition layer (upper thermocline; c). The stress divergence closely resembles the mode‐1 structure function (with a barotropic offset; d). The estimated shear production from the observed stress profile is in close agreement with observed TKE dissipation (e). To verify the dynamics of the forced velocity, we used the observed wind stress and stress profile to numerically integrate Equation 4 using the fourth order Runge‐Kutta algorithm. This integration is equivalent to solving a so‐called “slab model” (Pollard & Millard, 1970; Stokes et al., 2024). These models often include drag to provide damping and regularize the model (Plueddemann & Farrar, 2006), but our integration is short enough that drag has a negligible impact on the flow. However, the short integration is highly sensitive to the (uncertain) initial conditions. Therefore, we adjusted the model initial conditions until the numerical solution was in optimal agreement with the observations (r2=0.82; Figure 11a). The high r2 indicates that the observed forced velocity is largely predictable from Equation 4. The energy equation for the forced velocities (Equation 7) is evaluated using the observed wind stress and the numerical solution to Equation 4. Wind work increases from 0 to 4 mW m−2 during the storm. EF increases at a rate of 3 mW m−2 and about 1 mW m−2 is lost to shear production. There are two ways to estimate the shear production. Shear production can be calculated from τ(z,t)=τ(0,t)Σ(z) using the explicit expression for PSF in Equation 7 or estimated from the observed TKE dissipation rate using Equation 9. The two estimates have similar time variability (Figure 11b) and agree well in the transition layer between 8 and 15 m depth (Figure 11e). However, observed TKE dissipation is larger than predicted near the surface, potentially due to other sources of OSBL turbulence such as boundary layer shear, breaking waves, and convective mixing. For example, OSBL TKE dissipation during convection is typically equal 45%–100% of the surface buoyancy flux (Ivey & Imberger, 1991; Moum & Smyth, 2001), which was near 10−7 W kg−1 during nighttime surface cooling (Figure 2; the buoyancy flux was negative/stabilizing during the day). Observed TKE dissipation below the thermocline is dominated by instrument noise. The Wave Flow Wave velocity is modeled by solving Equation 6 following Kelly (2019) with the observed uF as the coastal boundary condition. The observed wave velocity is the total velocity minus the observed forced velocity (defined above). Total surface velocities are inhibited near the coast and nearly in phase along the cross‐shore line (Figure 12a). Conversely, both the observed and modeled wave velocities are maximum at the coast and propagate offshore (Figures 12b and 12c). Baroclinic surface pressure due to CNIWs is just total pressure because the forced flow does not have any vertical velocities. Observed pressure is similar to modeled pressure at the coast but its offshore phase propagation is faster (i.e., it has flatter phase lines; Figures 12d and 12e), suggesting it is closer to the inertial frequency. Total depth integrated energy flux (F) is weak and oscillates between on and offshore because it is “contaminated” by the cross term uFp. Observed and modeled wave energy flux FW is always offshore and 10 W (m coastline)−1 after the wind event at 60 hr. Total surface velocity displays coastal inhibition (a) while observed and modeled wave velocity decay away from the coast (c and d). Baroclinic wave pressure at the surface (d) resembles modeled pressure (e), but shows less offshore propagation. Total depth‐integrated energy flux (f) is both on and offshore, while observed and modeled energy flux display strong offshore propagation after the storm (g and h). Vertical velocity is estimated from observations of ∂xu using the continuity Equation 1d and assuming long‐shore uniformity. Horizontal uWp and vertical energy fluxes (wp) are estimated and averaged during the wind event (48–90 hr). Offshore fluxes are 1 W m−2 and enhanced near the surface. Vertical energy fluxes are about 1 mW m−2 and downward at the coast. Vertical energy flux oscillates between up and down farther from the coast, suggesting internal‐wave reflections between the well mixed bottom and surface layers, where N