2.1. Governing equations and boundary conditions

We study the IKW flow via the numerical solution of the Boussinesq equations of motion for a rotating stratified fluid on an $f$ -plane with a $m$ -order viscosity/diffusion operator, written as

where $\text{D}/\text{D}t$ is the material derivative, $\boldsymbol{v}=(u,v,w)$ is the velocity vector, ${\it\rho}_{0}$ and ${\it\rho}$ are the ambient reference density and the Boussinesq density difference from ${\it\rho}_{0}$ . Further, $p$ is the pressure field, $\hat{\boldsymbol{k}}$ is the unit vertical vector (positive upward), whilst ${\mathcal{D}}_{m}$ represents an $m$ -order viscosity/diffusion operator acting on the momentum and the mass transport, defined as follows (Winters & D’Asaro Reference Winters and D’Asaro1997):

where ${\it\nu}_{m,j}$ and ${\it\kappa}_{m,j}$ are the $m$ -order viscosity/diffusivity coefficient, respectively, whose physical dimensions are $[L^{2m}/T]$ . The operator $\partial _{j}^{2m}$ corresponds to the $2m$ -order partial derivative with respect to the spatial coordinate $j$ (with $j=x,y,z$ ). $L_{j}$ and $n_{j}$ are the computational domain size and the number of equally spaced grid points in the $j$ th direction. $T_{{\mathcal{D}}}$ is a time scale chosen by trial and error to minimize the range of damped scales while maintaining numerical stability (Winters & D’Asaro Reference Winters and D’Asaro1997). The choice $m=1$ results in the standard Navier–Stokes equations, which should be solved down to the Kolmogorov length scale which is barely possible given the available computing power. Alternatively, for example, the viscosity could be chosen by trial and error to be just large enough to damp the downscale energy transfer for a given problem resolved over a fixed number of grid points. However, if the goal is to allow the inertial subrange to be as broad as possible for a given grid resolution, then larger values of viscosity are inefficient and smaller values require finer grids to resolve the dissipative scales and maintain stability. Similar arguments apply for the $m$ -order operators. We use $m=3$ for this study and note that the use of this type of closure implicitly assumes a downscale energy cascade with the rate of energy transfer controlled by the nearly inviscid processes in the inertial subrange. This rate can then be measured at small scales by directly evaluating the kinetic energy dissipation rate whose mathematical form depends on the value of $m$ chosen as shown in appendix A.

We impose no-flux and free-slip boundary conditions on the top, bottom and sidewalls of the cylindrical domain to avoid having to resolve small viscous boundary layers. The initial condition of the IKW is constructed from the linear normal mode solution derived by Csanady (Reference Csanady1967) for a discontinuous two-layer stratification in a cylindrical domain. This solution is written as

where ${\it\eta}_{i}(t,r,{\it\theta})$ is the modal structure of the interface displacement and ${\it\beta}_{1}=R_{i}^{-1}\sqrt{1-{\it\sigma}_{1}^{2}}$ . The radial shape of the solution is characterized by the first-kind modified Bessel function of the azimuthal gravest mode, $\text{I}_{1}$ (Abramowitz & Stegun Reference Abramowitz and Stegun1965), while the temporal and azimuthal components are given by the cosine periodic function. The wave frequency, ${\it\omega}_{1}$ , and dimensionless frequency, ${\it\sigma}_{1}={\it\omega}_{1}/f$ , of the IKW are obtained by solving the dispersion relation derived from the boundary condition for ${\it\eta}_{i}$ at $r=R$ (Csanady Reference Csanady1967; Stocker & Imberger Reference Stocker and Imberger2003):

From the linearized inviscid governing equations and solution (2.3), we obtain the density and velocity field for a discontinuous two-layer stratification (Csanady Reference Csanady1981; Antenucci & Imberger Reference Antenucci and Imberger2001). However, solution (2.3) cannot be directly used to specify the initial condition of an IKW in a smooth two-layer stratified fluid. Moreover, a derivation of continuous eigenfunctions after introducing a small transition scale ${\it\delta}_{i}$ into the density profile is also inconvenient because such solutions are both formally and practically restricted to have displacement amplitudes much smaller than ${\it\delta}_{i}$ . Here we wish to study the nonlinear transition regime, and waves in this regime have initial amplitudes substantially larger than ${\it\delta}_{i}$ .

For a given ${\it\delta}_{i}\ll h_{j},~j=1,2$ , the linear, discontinuous solution (2.3) can be specified with an amplitude ${\it\eta}_{0}$ significantly larger than ${\it\delta}_{i}$ . We prescribe such a solution, expecting that its subsequent evolution will be nonlinear. We then smooth out the discontinuities by allowing the density and momentum to diffuse over a finite time scale such that the interface thickness is of $\mathit{O}({\it\delta}_{i})$ . In contrast, solving the linear equations for a continuously varying but sharply transitioning stratification produces overturns in the density field when ${\it\eta}_{0}$ approaches or exceeds ${\it\delta}_{i}$ (see e.g. Kundu & Cohen (Reference Kundu and Cohen2004), Chapter 13, on normal modes in a continuously stratified layer).

The equations of motion (2.1), with the smoothed version of the initial condition (2.3), are numerically solved using the three-dimensional spectral model flow_solve (Winters & de la Fuente Reference Winters and de la Fuente2012). The model is based on expansions in terms of trigonometric functions of the dependent variables over regular meshes in a cubic domain ( $L_{x},L_{y},L_{z}$ ), inside which a cylindrical domain ( $R,H$ ) is immersed. The variables are expanded in cosine or sine series in each spatial component, depending on the boundary conditions to be satisfied. For example, in the vertical direction, the horizontal velocity components, $u$ and $v$ , and the density field, ${\it\rho}$ , are expanded in cosine series while the vertical velocity component, $w$ , is expanded in sine series to satisfy the free-slip wall and no-flux condition on the top and bottom boundaries. The boundary conditions on the curved sidewall are implemented using an immersed boundary approach introduced by Winters & de la Fuente (Reference Winters and de la Fuente2012).

2.2. Dimensionless numbers

The initial dynamics of the IKW are parametrized by four dimensionless parameters:

Here ${\mathcal{B}}_{i}$ is the Burger number (Antenucci & Imberger Reference Antenucci and Imberger2001) that characterizes the influence of rotation. If ${\mathcal{B}}_{i}\geqslant 1$ , rotation is weak, while if ${\mathcal{B}}_{i}<1$ , rotational effects are important. In the IKW case, as ${\mathcal{B}}_{i}\rightarrow 0$ , the wave energy becomes concentrated near the boundary with the fluid motions parallel to the shoreline. Notice that ${\mathcal{B}}_{i}$ corresponds to the positive root of the classic Burger number, $S$ , defined by Pedlosky (Reference Pedlosky1987, see § 1.3, equation 1.3.1).

${\mathcal{T}}_{sk}$ is the steepening parameter and it is interpreted as the rate at which the spatial differences of the internal wave celerity induced by ${\it\eta}_{0}$ lead to wavefronts and nonlinearities, $T_{s}\sim {\it\lambda}_{k}/{\rm\Delta}c_{i}$ , over an IKW period $T_{k}$ . Therefore, ${\mathcal{T}}_{sk}$ quantifies how fast an IKW with amplitude ${\it\eta}_{0}$ tends to produce a nonlinear wavefront (Boegman, Ivey & Imberger Reference Boegman, Ivey and Imberger2005; de la Fuente et al. Reference de la Fuente, Shimizu, Imberger and Niño2008) and thus to concentrate energy at smaller length scales in the near-shore region.

$T_{{\it\nu}k}\sim {\it\eta}_{0}^{2m}/{\it\nu}_{m}$ is the dissipation time scale for motions of vertical scale ${\it\eta}_{o}$ and ${\mathcal{T}}_{{\it\nu}k}$ gives the ratio of this time scale to the Kelvin wave period. To quantify the relative importance of viscous to nonlinear effects in the context of this flow, we have defined a dimensionless number that is the ratio of the (hyper-)viscous time scale to the nonlinear steepening time of the primary Kelvin wave:

In this sense, this ratio is similar to the standard Reynolds number that relates inertial to viscous forces. The definition of the $m$ -order hyper-viscous Reynolds number is (Lamorgese, Caughey & Pope Reference Lamorgese, Caughey and Pope2005; Spyksma et al. Reference Spyksma, Magcalas and Campbell2012)

which formally reduces to the classic viscous Reynolds number when $m=1$ :

High values of $\mathit{Re}_{m}$ imply energetic regimes, dominated by inertial forces, whereas low values of $\mathit{Re}_{m}$ imply flow regimes dominated by (hyper-)viscous forces. The parameter ${\mathcal{T}}_{{\it\nu}k}/{\mathcal{T}}_{sk}$ can be written in terms of the hyper-viscous Reynolds number as follows:

Note that ${\it\eta}_{0}/{\it\lambda}_{k}$ is the aspect ratio of the Kelvin wave and it is a measure of its hydrostaticity. Steeper, non-hydrostatic waves are more nonlinear and enhance the energy transfer to small scales.

${\mathcal{J}}$ is the gradient Richardson number, with ${\mathcal{N}}$ the buoyancy frequency and ${\mathcal{S}}$ the vertical shear in the horizontal velocity (Miles Reference Miles1961). In our problem, both ${\mathcal{N}}$ and ${\mathcal{S}}$ have maxima at the height of the density interface but the magnitude and spatial distribution of ${\mathcal{J}}$ will depend both on ${\it\eta}_{0}$ and ${\mathcal{B}}_{i}$ . The maximal shear ${\mathcal{S}}$ , for example, occurs in the shore region, near the maximum vertical displacement of the interface. For a fixed rotation regime, the initial minimum value of ${\mathcal{J}}$ decreases as ${\it\eta}_{0}$ is increased.

The dissipation and mixing involved in the IKW evolution are characterized in terms of dimensionless turbulence activity parameters that quantify the dissipation rates of kinetic energy and buoyancy variance given by

where ${\mathcal{I}}_{m}$ denotes the turbulence intensity parameter expressed in terms of the hyper-viscosity approach of the kinetic energy dissipation rate, ${\it\epsilon}_{m}$ , and derived from the ratio of the Ozmidov and Kolmogorov scales (see appendix A). This parameter can be interpreted as the destabilizing effects of turbulent stirring compared to the stabilizing effects produced by the combined action of hyper-viscosity and buoyancy. Finally, ${\mathcal{K}}_{m}$ is the ratio of the diapycnal flux and a reference laminar diffusive flux, both in terms of the hyper-diffusion approach and the buoyancy, $b$ . This diffusivity parameter measures the increase in diffusive flux due to turbulent stirring and straining relative to the laminar, diffusive flux with the same background buoyancy that is not stirred or strained (see appendix B).

2.3. Set of numerical experiments

We consider a single rotation regime, ${\mathcal{B}}_{i}=0.25$ , and a range of dimensionless amplitudes ${\it\eta}_{0}/h_{1}\in [0.07,0.60]$ and aspect ratios ${\it\eta}_{0}/{\it\lambda}_{k}\in [8.8\times 10^{-4},7.5\times 10^{-3}]$ that allow ${\mathcal{T}}_{sk}\in [3.6,34.6]$ , ${\mathcal{T}}_{{\it\nu}k}\in [10^{-1},4\times 10^{4}]$ and initial minimum values of ${\mathcal{J}}$ between $\min \{{\mathcal{J}}_{0}\}=31.86$ , for ${\it\eta}_{0}/h_{1}=0.07$ , and $\min \{{\mathcal{J}}_{0}\}=0.13$ , for ${\it\eta}_{0}/h_{1}=0.60$ . With these parameters, we have performed six numerical experiments using spatial and temporal scales similar to those used in recent laboratory experiments carried out in a rotating table by Ulloa et al. (Reference Ulloa, de la Fuente and Niño2014). The dimensions of the domain are $L_{x}\times L_{y}\times L_{z}=1.95\times 1.95\times 0.2~\text{m}^{3}$ , and in its interior there is a circumscribed cylinder of radius $R=0.9~\text{m}$ and depth $H=h_{1}+h_{2}=0.07~\text{m}+0.13~\text{m}=0.2~\text{m}$ . The difference of densities is ${\rm\Delta}{\it\rho}={\it\rho}_{2}-{\it\rho}_{1}=15~\text{kg}~\text{m}^{-3}$ , the effective thickness of the interfacial transition is ${\it\delta}_{i}\approx 0.02~\text{m}$ , and the inertial frequency $f=0.361~\text{Hz}$ , with an IKW period of $T_{k}=60.45~\text{s}$ and a total simulation time $T_{ns}=3T_{k}=182~\text{s}$ . Figure 2 shows that the initial transition layer is resolved by about 12–14 grid points. The horizontal wavelength of the most unstable mode for a smooth, sheared density transition is about ${\rm\pi}$ times the interface thickness (see e.g. Kundu & Cohen Reference Kundu and Cohen2004, § 11.7, p. 506). This corresponds to about 10–11 times our horizontal grid spacing and so the most unstable mode is resolved on our spatial grid, though significantly smaller motions are not. Table 2 summarizes the parameters used in the numerical experiments.

Figure 2.

Density, ${\it\rho}(z)/{\it\rho}$ , and buoyancy frequency, ${\mathcal{N}}(z)/\text{max}({\mathcal{N}})$ , profiles in the initial condition of a discontinuous two-layer stratification ${\it\delta}_{i}=0$ (grey line) and a smooth two-layer stratification ${\it\delta}_{i}>0$ (red line) after a vertical diffusion over a time interval $t_{{\it\delta}_{i}}$ . The blue circles denote the equidistant grid points along the transition layer.

Table 1.

Glossary of symbols used in the text.

3.1. Damped linear regime (DLR)

The DLR is characterized by the evolution of an IKW that retains most of its initial linear features, with negligible nonlinear steepening and strongly controlled by viscosity. This regime describes the results of experiment 1 shown in figure 3(a,b); in this case, ${\it\eta}_{0}/h_{1}=0.07~({\it\eta}_{0}/{\it\delta}_{i}\approx 1.0)$ , ${\mathcal{T}}_{{\it\nu}k}/{\mathcal{T}}_{sk}\approx 3\times 10^{-3}<1$ and $\min \{{\mathcal{J}}_{0}\}\approx 31.86$ , which suggests an extremely viscous flow with a negligible steepening capacity and a strong hydrodynamic stability in the density interface. Therefore a linear and damped IKW dynamics along the $3T_{k}$ periods is expected. The PSD in figure 3(b) shows a single strong energy peak at ${\it\omega}/{\it\omega}_{k}=1$ , indicating that the initial IKW mode stored the energy of the flow.

3.2. Nonlinear regime (NLR)

The NLR is characterized by weak nonlinear degeneration, without dispersion of the IKW (in the KdV theory sense). This regime starts with the IKW steepening, as a consequence of a wave amplitude large enough to induce significant changes in the wave celerity, leading the wavefront and the formation of a solitary-type wave (Fedorov & Melville Reference Fedorov and Melville1995; de la Fuente et al. Reference de la Fuente, Shimizu, Imberger and Niño2008). This regime describes the results of experiment 2 shown in figure 3(c,d); in this case ${\it\eta}_{0}/h_{1}=0.18$ , ${\mathcal{T}}_{{\it\nu}k}/{\mathcal{T}}_{sk}\approx 1.9>1$ and $\min \{{\mathcal{J}}_{0}\}\approx 3.04$ , and therefore we expect a weakly nonlinear steepening controlled by viscosity and a stable flow around the density interface. In fact, figure 3(c) shows a slightly steep wavefront after the first wave period and a solitary-type wave structure around $t/T_{k}\approx 2.45$ . The PSD in figure 3(d) shows a wider energy cascade due to transfer of energy involved in the weakly nonlinear degeneration, with a main energy peak at ${\it\omega}/{\it\omega}_{k}=1$ , and two lower peaks of sub-inertial frequencies attributed to sub-azimuthal Kelvin waves (see the dot-dashed line).

3.3. Nonlinear and non-hydrostatic regime (NHR)

Nonlinear steepening and non-hydrostatic dispersion of the initial IKW (in the KdV theory sense) characterize the NHR. As the wave steepens, its azimuthal length scale decreases until the non-hydrostatic terms can be important enough to balance the nonlinear steepening, avoiding the wave breaking (Helfrich & Melville Reference Helfrich and Melville2006) and leading the degeneration of the IKW into a package of solitary-type waves (Grimshaw Reference Grimshaw1985). This process induces an important energy cascade from the basin scale to smaller scales. The NHR regime is identified in experiments 3–6 (see figures 3 e–l). In these experiments the Reynolds number increases from ${\mathcal{T}}_{{\it\nu}}/{\mathcal{T}}_{s}\approx 4\times 10^{2}$ , for ${\it\eta}_{0}/h_{1}=0.38$ , to ${\mathcal{T}}_{{\it\nu}k}/{\mathcal{T}}_{sk}\approx 10^{4}$ , for ${\it\eta}_{0}/h_{1}=0.60$ . Increasing ${\mathcal{T}}_{{\it\nu}}/{\mathcal{T}}_{s}$ corresponds to increasing nonlinearity and concentration of energy at smaller scales. Experiments 3 and 4 show hydrodynamic stable flows in the density interface region, with $\min \{{\mathcal{J}}_{0}\}\approx 0.58$ and $\min \{{\mathcal{J}}_{0}\}\approx 0.30$ , respectively, whereas experiments 5 and 6 show evidence of the emergence of hydrodynamic interfacial instabilities, with $\min \{{\mathcal{J}}_{0}\}\approx 0.20$ and $\min \{{\mathcal{J}}_{0}\}\approx 0.13$ , respectively.

3.4. Laminar–turbulent transition regime (TR)

The TR is characterized by the growth and occurrence of interfacial instabilities. Depending on how energetic the shear flow is, the instabilities can grow and generate turbulent patches in the density interface region or can be damped. The hydrodynamic stability of the IKW sheared flow can be studied via the gradient Richardson number, ${\mathcal{J}}$ , whose critical value ${\mathcal{J}}=0.25$ (Miles Reference Miles1961) is neither a necessary nor a sufficient condition to trigger interfacial instabilities in our sheared-type flow, but it is still a very useful criterion that can give us an order of magnitude. In our results only two experiments show the presence of interfacial instabilities, experiments 5 and 6, with $\min \{{\mathcal{J}}_{0}\}\approx 0.20$ and $\min \{{\mathcal{J}}_{0}\}\approx 0.13$ , respectively. Only they have initial ${\mathcal{J}}<0.25$ . Both experiments exhibit the emergence of Kelvin–Helmholtz-type billows on the wavefront of strongly nonlinear IKWs ( ${\mathcal{T}}_{{\it\nu}k}/{\mathcal{T}}_{sk}\sim 5\times 10^{3}{-}10^{4}$ ). However, the results of experiment 5 show fast damping of the interfacial instabilities, while the results of experiment 6 show a complex spatial and temporal dynamic, in which large scales, attributed to the IKW and internal solitary-type waves, coexist and interplay with turbulent small scales, associated with interfacial instabilities/breaking and turbulent patches.

Figure 4.

Results of experiment 6: evolution in time $t/T_{k}\in [0,3]$ of the vertical velocity $w$ at the density interface region ${\it\rho}(t,\boldsymbol{x})={\rm\Delta}{\it\rho}/2$ . (a)  $t/T_{k}=0$ , (b)  $t/T_{k}=0.128$ , (c)  $t/T_{k}=0.25$ , (d)  $t/T_{k}=0.5$ , (e)  $t/T_{k}=0.75$ , (f)  $t/T_{k}=1.0$ , (g)  $t/T_{k}=1.5$ , (h)  $t/T_{k}=1.75$ , (i)  $t/T_{k}=2.0$ , (j) $t/T_{k}=2.25$ , (k) $t/T_{k}=2.75$ , (l)  $t/T_{k}=3.0$ .

The first three regimes have been analysed in previous numerical (de la Fuente et al. Reference de la Fuente, Shimizu, Imberger and Niño2008; Sakai & Redekopp Reference Sakai and Redekopp2010) and experimental studies (Wake, Ivey & Imberger Reference Wake, Ivey and Imberger2005; Ulloa et al. Reference Ulloa, de la Fuente and Niño2014), whereas the TR on IKWs has been observed only in field data (Lorke Reference Lorke2007; Preusse et al. Reference Preusse, Peeters and Lorke2010) and to date in numerical experiments.

4.1. Spatiotemporal distribution of turbulence

Figure 5 shows the vertical average of the turbulence intensity parameter, $\langle {\mathcal{I}}_{m}\rangle _{H}$ ; the dashed line defines the radial position of the internal Rossby radius of deformation, $R_{i}$ , from the boundary to the interior. Figure 5(a) shows the initial distribution of $\langle {\mathcal{I}}_{m}\rangle _{H}$ induced by the IKW itself. Kinetic energy dissipation is concentrated within $R_{i}$ and decays to low values towards the centre of the domain. The nonlinear dynamics of the IKW induces distinct zones of elevated values in the near-shore region, associated with turbulent patches produced by the breakdown of interfacial instabilities (see figure 5 b–d). As the initial IKW degenerates, the turbulence intensity increases at the front of each solitary-type wave along the shore region (see figure 5 d–f) achieving values of $\langle {\mathcal{I}}_{m}\rangle _{H}\sim \mathit{O}(10^{2})$ , while most of the offshore region shows a low activity with values of $\langle {\mathcal{I}}_{m}\rangle _{H}\sim \mathit{O}(10^{-1})$ . However, after the first period, and particularly after the strong reactivation of the pre-existing turbulent patch, there is a steady increase of $\langle {\mathcal{I}}_{m}\rangle _{H}$ to the interior of $R_{i}$ , with values in the range $\mathit{O}(10^{0})\leqslant \langle {\mathcal{I}}_{m}\rangle _{H}\leqslant \mathit{O}(10^{1})$ (see figure 5 f–h). This process is again observed after the second period (see figure 5 j). After three periods elevated values of $\langle {\mathcal{I}}_{m}\rangle _{H}$ are distributed throughout the domain (see figure 5 k,l). Nevertheless, there are clear differences in the intensity between the near-shore and interior regions, where the internal Rossby radius of deformation (dashed line) seems to define a transition length scale of the turbulent activity along the radial component.

Figure 5.

Results of experiment 6: spatiotemporal distribution of the turbulence intensity, $\langle {\mathcal{I}}_{m}\rangle _{H}$ . (a)  $t/T_{k}=0$ , (b)  $t/T_{k}=0.128$ , (c)  $t/T_{k}=0.25$ , (d)  $t/T_{k}=0.5$ , (e)  $t/T_{k}=0.75$ , (f)  $t/T_{k}=1.0$ , (g)  $t/T_{k}=1.5$ , (h)  $t/T_{k}=1.75$ , (i)  $t/T_{k}=2.0$ , (j)  $t/T_{k}=2.25$ , (k)  $t/T_{k}=2.75$ , (l)  $t/T_{k}=3.0$ .

The azimuthal distribution of the turbulence intensity parameter has features of baroclinic instabilities. This is also observed in figure 6, which shows the relative vertical vorticity, ${\bf\omega}\boldsymbol{\cdot }\hat{\boldsymbol{k}}$ , normalized by $f$ at $z/H=0.8$ (horizontal plane) and the density interface structure ${\it\rho}(t,\boldsymbol{x})={\rm\Delta}{\it\rho}/2$ in purple. Initially we observe the formation of one turbulent patch in the shore region around $t/T_{k}=0.25$ (see figure 6 c), which grows and spreads along the shore until approximately the first period (see figure 6 d–f). After the first period, the initial turbulent patch is separated into two (see figure 6 d,h,i) that also remain trapped on the shore, and a third vortex structure grows on the shore region during the second period (see figure 6 k,l). It is observed that turbulent patches tend to aggregate in large horizontal motions that scale with $R_{i}$ . Figure 6(m) shows a close-up of the vertical vorticity at $t/T_{k}=3$ , where three large-scale vortex-type motions are identified in the near-shore region (see dashed ellipses). We can study the existence of baroclinic instability via a simplified model. Pedlosky (Reference Pedlosky1970) considered a quasi-geostrophic two-layer model in an inviscid flow rotating on an $f$ -plane, with uniform velocities $U_{1}$ and $U_{2}$ in each layer $(U_{1}\neq U_{2})$ . The flow is unstable to disturbances with horizontal wavelengths, ${\it\lambda}$ , larger than ${\rm\pi}R_{i}$ (see e.g. Pedlosky Reference Pedlosky1987, Chapter 7, p. 556). Then, considering the azimuthal wavelength of the IKW, ${\it\lambda}_{k}=2{\rm\pi}R$ , and the internal Rossby radius here adopted, it is obtained that the flow induced by the IKW admits the growth of baroclinic instabilities because ${\rm\pi}R_{i}/{\it\lambda}_{k}={\mathcal{B}}_{i}/2=0.125$ . In addition, the interaction between the vortex-type motions and the vertical shear flow driven by the IGW could be supporting the emergence of baroclinic instabilities (Sakai Reference Sakai1989; Gula, Plougonven & Zeitlin Reference Gula, Plougonven and Zeitlin2009; Flór, Scolan & Gula Reference Flór, Scolan and Gula2011). Note that these large-scale vortex motions are not observed in the interior of the basin.

Figure 6.

Results of experiment 6: spatiotemporal distribution of the vertical vorticity, $({\bf\omega}/f)\boldsymbol{\cdot }\hat{\boldsymbol{k}}$ , at $z/H=0.80$ . (a)  $t/T_{k}=0$ , (b)  $t/T_{k}=0.128$ , (c)  $t/T_{k}=0.25$ , (d)  $t/T_{k}=0.5$ , (e)  $t/T_{k}=0.75$ , (f)  $t/T_{k}=1.0$ , (g)  $t/T_{k}=1.5$ , (h)  $t/T_{k}=1.75$ , (i)  $t/T_{k}=2.0$ , (j)  $t/T_{k}=2.25$ , (k)  $t/T_{k}=2.75$ , (l)  $t/T_{k}=3.0$ , (m)  $t/T_{k}=3.0$ .

Figure 7 shows the time series of extreme and bulk values of the turbulence intensity parameter, within and outside $R_{i}$ . The red vertical lines identify events ( $e_{i}$ ) with high turbulent activity (local maxima) in the time series of experiment 6 (– $\bullet$ –), whose horizontal positions can be identified in figure 5(b,d,g,i,j). In figure 7(a) the values of $\min \{{\mathcal{I}}_{m}\}$ are comparable for all experiments except experiments 1 and 2 (blue lines), with $\min \{{\mathcal{I}}_{m}\}\sim \mathit{O}(10^{3})$ . Values of $\max \{{\mathcal{I}}_{m}\}$ , however, increase with increasing nonlinearity, causing the separation of the time series between $\mathit{O}(10^{-1})\leqslant \max \{{\mathcal{I}}_{m}\}\leqslant \mathit{O}(10^{2})$ (red lines). In the case of experiment 6, peak turbulence intensity values are $\mathit{O}(10^{5})$ times the background laminar value. These results show a significant heterogeneity of ${\mathcal{I}}_{m}$ in both time and space. We have divided the domain into two volumes: $V_{in}$ and $V_{out}$ denote the volumes within and outside $R_{i}$ , respectively, and over these volumes bulk quantities of ${\mathcal{I}}_{m}$ have been calculated. Within $R_{i}$ (see figure 7 b), time series of $\langle {\mathcal{I}}_{m}\rangle _{V_{in}}$ show distinct differences between each experiments. The highest values and the most interesting temporal fluctuations are observed in experiment 6 (– $\bullet$ –). Outside $R_{i}$ (see figure 7 c), time series show that $\langle {\mathcal{I}}_{m}\rangle _{V_{out}}$ tends to grow as a function of time, and shows only one peak at $e_{2}$ attributed to the formation of the first turbulent patch; figure 5(d) shows that this event ( $e_{2}$ ) affects a small region outside $R_{i}$ . Comparing both volumes, early in time, the distinction between the turbulent activity within and outside $R_{i}$ is quite important, but by the later times in the experiments, ${\mathcal{I}}_{m}$ is only about a factor of 3 higher (see figure 7 b,c for the most energetic case) and the tendency is to reduce the difference between both regions.

Figure 7.

Evolution in time (scaled by $T_{K}$ ) of (a) the maximal and minimal values of the turbulence intensity parameter, (b) the bulk turbulence intensity parameter within $R_{i}$ , $\langle {\mathcal{I}}_{m}\rangle _{V_{in}}$ , and (c) the bulk turbulence intensity parameter outside $R_{i}$ , $\langle {\mathcal{I}}_{m}\rangle _{V_{out}}$ . Legend: – $\bullet$ –, exp. 6; – $+$ –, exp. 5; –▼–, exp. 4; – $\times$ –, exp. 3; –▪–, exp. 2; –♦–, exp. 1.

4.2. Turbulence intensity and effective diffusivity

Figures 8(a) and 8(b) show time series of the bulk averages of the turbulence intensity parameter, $\langle {\mathcal{I}}_{m}\rangle _{V}$ , and the diffusivity parameter, $\langle {\mathcal{K}}_{m}\rangle _{V}$ , respectively.

Figure 8.

Evolution in time (scaled by $T_{K}$ ) of (a) the bulk turbulence intensity parameter, $\langle {\mathcal{I}}_{m}\rangle _{V}$ , and (b) the bulk diffusivity parameter, $\langle {\mathcal{K}}_{m}\rangle _{V}$ . Panel (c) exhibits $\langle {\mathcal{K}}_{m}\rangle _{V,T}$ versus $\langle {\mathcal{I}}_{m}\rangle _{V,T}$ . Legend: – $\bullet$ –, exp. 6; – $+$ –, exp. 5; –▼–, exp. 4; – $\times$ –, exp. 3; –▪–, exp. 2; –♦–, exp. 1.

The temporal structure of $\langle {\mathcal{I}}_{m}\rangle _{V}$ (see figure 8 a) is forced by the turbulent activity in the near-shore region (see peaks denoted by vertical dashed lines) but with lower magnitudes because of the low turbulent activity on the offshore region. Meanwhile, the time series of $\langle {\mathcal{K}}_{m}\rangle _{V}$ can be clustered into three groups (see figure 8 b): a first group with experiments 1–3, a second group with experiments 4–5 and a third group with experiment 6. The first group contains experiments with the lowest and almost constant value of $\langle {\mathcal{K}}_{m}\rangle _{V}$ , the second includes experiments with weak fluctuations of $\langle {\mathcal{K}}_{m}\rangle _{V}$ , and the third group has the highest values and the strongest fluctuations of $\langle {\mathcal{K}}_{m}\rangle _{V}$ . Experiment 6 is unique in that it yields substantially increased values of $\langle {\mathcal{K}}_{m}\rangle _{V}$ that are clearly correlated with individual breaking events (– $\bullet$ – and vertical dashed line in figure 8 b). The temporal structure of $\langle {\mathcal{K}}_{m}\rangle _{V}$ indicates that episodic increases of the bulk diffusivity parameter are not directly correlated with the bulk-averaged turbulence intensity.

Figure 8(c) shows the volumetric and temporal averages, $\langle \cdot \rangle _{V,T}$ , of ${\mathcal{I}}_{m}$ and ${\mathcal{K}}_{m}$ . Red symbols correspond to $\langle {\mathcal{K}}_{m}\rangle _{V,T}$ against $\langle {\mathcal{I}}_{m}\rangle _{V,T}$ while the clusters of symbols around each red symbol correspond to $\langle {\mathcal{K}}_{m}\rangle _{V}(t)$ against $\langle {\mathcal{I}}_{m}\rangle _{V}(t)$ (data shown in figure 8 a,b). Further, the standard deviations (s.d.) of the bulk values $\langle \cdot \rangle _{V}$ over time are shown (bars in the red symbols). We only see significant temporal variability in experiment 6, shown by – $\bullet$ – in figure 8(b) and the cluster of circles in figure 8(c). The large deviations seen in experiment 6 are a consequence of having approximately steady values of turbulence intensity (see figure 8 a) but with very episodic elevated mixing (see figure 8 b), suggesting that an instantaneous and local sampling of dissipation can be very misleading in terms of predicting the instantaneous mixing rate. On the other hand, averaging over all the events we identify a simple relation between $\langle {\mathcal{K}}_{m}\rangle _{V,T}$ and $\langle {\mathcal{I}}_{m}\rangle _{V,T}$ . For values of $\mathit{O}(10^{-2})\leqslant \langle {\mathcal{I}}_{m}\rangle _{V,T}\leqslant \mathit{O}(10^{-1})$ we find that $\langle {\mathcal{K}}_{m}\rangle _{V,T}\sim \mathit{O}(10^{0})$ , i.e. that the total diffusivity is not substantially elevated over that expected by laminar diffusion. For $\mathit{O}(10^{-1})\leqslant \langle {\mathcal{I}}_{m}\rangle _{V,T}\leqslant \mathit{O}(10^{1})$ , however, the observed diffusivity is greater than that expected in a laminar flow and increases approximately linearly with turbulence intensity. The data fit a power law $\langle {\mathcal{K}}_{m}\rangle _{V,T}\sim a\{\langle {\mathcal{I}}_{m}\rangle _{V,T}\}^{b}$ , with parameters $a=12.2$ and $b=1.1$ . In the next subsection we describe the Kelvin wave evolution in the near-shore region and analyse the vertical distribution of ${\mathcal{I}}_{m}$ during the events $e_{2}$ and $e_{3}$ .

Figure 9.

Results of experiment 6: evolution in time $t/T_{k}\in [0,3]$ of the stratification, ${\mathcal{N}}^{2}/{\mathcal{N}}_{0}^{2}$ , in the area ${\it\theta}\in [0,2{\rm\pi}]$ , $z/H\in [0,1]$ at $r/R=0.98$ . (a)  $t/T_{k}=0$ , (b)  $t/T_{k}=0.128$ , (c)  $t/T_{k}=0.25$ , (d)  $t/T_{k}=0.5$ , (e)  $t/T_{k}=0.75$ , (f)  $t/T_{k}=1.0$ , (g)  $t/T_{k}=1.128$ , (h)  $t/T_{k}=1.5$ , (i)  $t/T_{k}=2.0$ , (j)  $t/T_{k}=2.25$ , (k)  $t/T_{k}=2.75$ , (l)  $t/T_{k}=3.0$ .

4.3. Flow evolution in the near-shore region

Figure 9 shows a summary of the vertical stratification in experiment 6, ${\mathcal{N}}^{2}/{\mathcal{N}}_{0}^{2}$ (where ${\mathcal{N}}_{0}^{2}\equiv \max \{{\mathcal{N}}^{2}(t=0)\}$ ), in the area ${\mathcal{A}}_{{\it\theta}}$ defined by $z/H\in [0,1]$ , ${\it\theta}\in [0,2{\rm\pi}]$ and $r/R=0.98$ , for 12 different times $t/T_{k}\in [0,3]$ . The region ${\mathcal{A}}_{{\it\theta}}$ allows us to observe the IKW evolution near the boundary where the wave displacements are maximal and where the maximal values of the turbulent activity are observed (see figure 5). Figure 9(b) shows KH-type instabilities in the early stages of the experiment ( $t/T_{k}\approx 0.128$ ) because of the strong initial shear near the crest of the IKW. Figure 9(c) shows the collapse or breakdown of the shear instabilities at $t/T_{k}\approx 0.25$ . This process induces the formation of a turbulent patch and a local thickening of the density interface region that propagates along the shore region (see figures 9 d and 5 c). Simultaneously, the IKW has begun to steepen and form a solitary wave train (see figure 9 d–f). After the first period, the leading wave of the solitary wave train interacts with the pre-existing stirred region (see figure 9 g). The interaction generates interfacial breaking near the trough of the leading wave and a turbulent patch that subsequently spreads along the shore, over the solitary wave train (see figure 9 g,h). This sequence of events is also observed after the second period (see figure 9 j), where interfacial instabilities on the trough and tail of the leading wave can be seen. After $t/T_{k}\approx 3$ , the initial IKW has evolved into a solitary wave train that covers the area ${\mathcal{A}}_{{\it\theta}}$ and a significant thickening and weakening of the density interface as a consequence of vertical mixing is observed (see figure 9 k,l).

Figure 10.

Results of experiment 6: evolution in time $t/T_{k}\in [0.25,0.75]$ of the turbulence intensity parameter (in log scale) and radial vorticity in the region ${\it\theta}\in [0,{\rm\pi}]$ , $z/H\in [0,1]$ at $r/R=0.98$ , during the peak of $\langle {\mathcal{I}}_{m}\rangle _{V}$ , denoted by $e_{2}$ in figures 5(d), 7 and 8. (a)  $t/T_{k}=0.25$ , (b,c)  $t/T_{k}=0.31$ , (d,e)  $t/T_{k}=0.38$ , (f,g)  $t/T_{k}=0.44$ , (h,i)  $t/T_{k}=0.50$ , (j,k)  $t/T_{k}=0.56$ , (l,m)  $t/T_{k}=0.63$ , (n,o)  $t/T_{k}=0.69$ , (p,q)  $t/T_{k}=0.75$ . VE: vortex ejection.