2.1. The three dimensional system

We define a coordinate system $(\boldsymbol {x},t)=(x,y,z,t)$ , where $t$ is time. The $z$ direction is approximately bed normal with $b(x,y)$ the bed elevation (slowly varying) and the current exists in the region $z\gt b$ . The coordinate system is not necessarily so that $z$ is vertical, so that the first two components of gravity $\boldsymbol {g}$ need not be zero. We denote the angle of $z$ to the vertical by $\theta$ so that $g_3 = - g \cos \theta$ . In later sections (§§ 4 and 6) we restrict ourselves to a two-dimensional bottom current where $0\lt \theta \lt \pi /2$ , $g_1 = g \sin \theta$ and $g_2 = 0$ , as depicted in figure 1; we do not make these assumptions in the derivation. Velocity is denoted by $\boldsymbol {u} = (u,v,w)$ and volumetric concentration of particles by $\phi$ .

Our goal is to represent each property of the flow field as a depth-averaged quantity multiplied by a relatively steady shape function (i.e. slowly varying in $x,y,t$ ). Due to the turbulent nature of the flow each variable, say $f$ , has high wavenumber and high frequency fluctuations. For this reason we introduce Reynolds averaging, specifically ensemble averaging, with the average of a function $f$ denoted $ {\:\!\overline {\vphantom { \vphantom {\phi }\phi ^{\prime} }f}\:\!}$ and fluctuation $ \vphantom {\phi }f^{\prime} = f - {\:\!\overline {\vphantom { \vphantom {\phi }\phi ^{\prime} }f}\:\!}$ (e.g. Drew & Passman Reference Drew and Passman1999; Pope Reference Pope2000). The averaged variables, $ {\:\!\overline {\vphantom { \vphantom {\phi }\phi ^{\prime} }f}\:\!}$ , are sufficiently smooth to have their profiles represented by a slowly varying shape.

The (constant) density of water is $\rho _f$ and the density of the particles is $\rho _s$ ( $\rho _s\gt \rho _f$ for bottom currents); it is assumed that the only density variations come from the concentration so that the current density is $ {\:\!\overline {\vphantom { \vphantom {\phi }\phi ^{\prime} }\rho }\:\!} = \rho _f + {\:\!\overline {\vphantom { \vphantom {\phi }\phi ^{\prime} }\phi }\:\!} (\rho _s-\rho _f)$ . It is assumed that the particles are non-cohesive, dilute ( $ {\:\!\overline {\vphantom { \vphantom {\phi }\phi ^{\prime} }\phi }\:\!} \ll 1$ ) and Boussinesq ( $R {\:\!\overline {\vphantom { \vphantom {\phi }\phi ^{\prime} }\phi }\:\!} \ll 1$ where $R = (\rho _s-\rho _f)/\rho _f$ ) so that the bed elevation can be treated as constant in time and there are no high-concentration effects on the flow such as hindered settling or concentration dependent rheology.

To ensure that the the model only includes leading-order effects we introduce dimensional scales. We employ a time scale $\mathscr {T}$ , and length scales $\mathscr {L}_i$ corresponding to the length ( $i=1$ ), width ( $i=2$ ) and depth ( $i=3$ ) of the current. The current is assumed to be shallow, that is $\mathscr {L}_\alpha \gg \mathscr {L}_3$ for $\alpha \in \{1,2\}$ . (Throughout, the subscripts $\alpha,\beta,\gamma$ will be used for indices limited to $1,2$ .) This also aligns the coordinate system: $ {\partial b}/{\partial x_\alpha }$ scales as $\mathscr {L}_3/\mathscr {L}_\alpha \ll 1$ , the bed elevation is slowly varying. Considering the rotation of the coordinate system $\theta$ , the bed can be quite steep provided that $\mathscr {L}_\alpha /\mathscr {L}_3 \gg \lvert \tan \theta \rvert$ in the downslope direction. We assume the averaged velocities $ {\:\!\overline {\vphantom { \vphantom {\phi }\phi ^{\prime} }u_i}\:\!}$ scale as $\mathscr {U}_i = \mathscr {L}_i/\mathscr {T}$ , this can be viewed as a definition of the time scale. The scales of the turbulent transport terms are approximated by an eddy viscosity model. The details of the scaling analysis are given in Appendix A. Eliminating the small terms we deduce that at leading order

(2.1a) \begin{align} \frac {\partial \:\!\overline {\vphantom {\vphantom {\phi }\phi ^{\prime} }u_j}\:\!}{\partial x_j} &= 0, \end{align}

(2.1b) \begin{align} \frac {\partial {\:\!\overline {\vphantom { \vphantom {\phi }\phi ^{\prime} }\phi }\:\!}}{\partial t} + \frac {\partial }{\partial x_j} \left ({\overline{\vphantom{{{\phi }}}u_j}}\, {\overline{\phi }}\right) + \frac {\partial }{\partial z} \left ({\tilde {w} {\:\!\overline {\vphantom { \vphantom {\phi }\phi ^{\prime} }\phi }\:\!} + J_3}\right) &= 0, \end{align}

(2.1c) \begin{align} \frac {\partial {\:\!\overline {\vphantom { \vphantom {\phi }\phi ^{\prime} }u_\alpha }\:\!}}{\partial t} \!+\! \frac {\partial }{\partial x_{j}}\left ({{\:\!\overline {\vphantom { \vphantom {\phi }\phi ^{\prime} }u_j}}\, {\overline {\vphantom { \vphantom {\phi }\phi ^{\prime} }u_\alpha }\:\!} }\right) \!-\! \frac {\partial }{\partial z} \left ( {\! \tau _{3\alpha }^R + \nu \frac {\partial {\:\!\overline {\vphantom { \vphantom {\phi }\phi ^{\prime} }u_\alpha }\:\!}}{\partial z}}\right) \!+\! \frac {\partial p^T}{\partial x_\alpha } &= R g_\alpha {\:\!\overline {\vphantom { \vphantom {\phi }\phi ^{\prime} }\phi }\:\!} \quad \text { for } \alpha \in \lbrace {1,2} \rbrace, \end{align}

(2.1d) \begin{align} \frac {\partial p^T}{\partial z} &= R g_3 {\:\!\overline {\vphantom { \vphantom {\phi }\phi ^{\prime} }\phi }\:\!}, \end{align}

(2.1e) \begin{align} \frac {\partial k}{\partial t} + \frac {\partial }{\partial x_j}\left ({{\:\!\overline {\vphantom { \vphantom {\phi }\phi ^{\prime} }u_j}\:\!} k}\right) + \frac {\partial T_3}{\partial z} &= \tilde {\mathcal {P}} - \tilde {\epsilon }_K + R g_3 J_3, \end{align}

which represent conservation of volume, particles, $x$ and $y$ momentum, $z$ momentum and TKE, respectively. For three-dimensional variables (as used here) we use the convention that repeated indices should be summed over $\{1,2,3\}$ . We denote the constant settling velocity of the particles by $\tilde {\boldsymbol {u}}$ ( $\tilde {w}\lt 0$ for dense particles in a bottom current), the transport of particles by $\boldsymbol {J}$ , the Reynolds stress by $\boldsymbol {\tau }^R$ , the viscosity by $\nu$ , the combined effect of pressure and TKE relative to the hydrostatic ambient by $p^T$ and the TKE by $k$ which is transported by $\boldsymbol {T}$ , produced by $\tilde {\mathcal {P}}$ and the dissipated by $\tilde {\epsilon }_K$ . The hydrostatic approximation is common across the gravity current literature (e.g. Ellison & Turner Reference Ellison and Turner1959; Parker et al. Reference Parker, Fukushima and Pantin1986; Bonnecaze et al. Reference Bonnecaze, Huppert and Lister1993; Ungarish Reference Ungarish2020).

The form of the system (2.1) is generic, and can be derived by averaging a variety of models of dilute Boussinesq particle-laden flow. Following Parker et al. (Reference Parker, Fukushima and Pantin1986), a simple approach is applicable to situations where the particles are smaller than the Kolmogorov length scale. From the perspective of the turbulent micro-scales the particles act as a concentration field $\phi$ , similar to a concentration of salinity but moving at a speed $\boldsymbol {u} + \tilde {\boldsymbol {u}}$ , where by diluteness the velocity of the mixture is the same as the velocity of the fluid, and $k = ({1}/{2}) {\:\!\overline {\vphantom { \vphantom {\phi }\phi ^{\prime}} \vphantom {\phi }u_i^{\prime} \vphantom {\phi }u_i^{\prime}}\:\!}$ . Averaging we obtain (2.1) (strictly the unsimplified system (A1)) with

(2.2) \begin{align} \tau _{ij}^R &= - {\:\!\overline {\vphantom { \vphantom {\phi }\phi ^{\prime} } \vphantom {\phi }u_i^{\prime} \vphantom {\phi }u_j^{\prime} }\:\!}, & T_j &= {\textstyle \frac {1}{\rho }} {\:\!\overline {\vphantom { \vphantom {\phi }\phi ^{\prime} } \vphantom {\phi }u_j^{\prime} \vphantom {\phi }p^{\prime} }\:\!} + {\textstyle \frac {1}{2}} {\:\!\overline {\vphantom { \vphantom {\phi }\phi ^{\prime} } \vphantom {\phi }u_j^{\prime} \vphantom {\phi }u_i^{\prime} \vphantom {\phi }u_i^{\prime} }\:\!} - \nu \frac {\partial k}{\partial x_j}, & \tilde {\mathcal {P}} &= \tau _{ji}^R \frac {\partial {\:\!\overline {\vphantom { \vphantom {\phi }\phi ^{\prime} }u_i}\:\!} }{\partial x_j}, \end{align}

(2.3) \begin{align} J_j &= {\:\!\overline {\vphantom {\vphantom {\phi }\phi ^{\prime}}\vphantom {\phi }u^{\prime}_j \vphantom {\phi }\phi ^{\prime}}\:\!}, & \tilde {\epsilon }_K &= \nu {\:\!\overline {\vphantom {\vphantom {\phi }\phi ^{\prime}}\frac {\partial \vphantom {\phi }u_i^{\prime}}{\partial x_j}\frac {\partial \vphantom {\phi }u_i^{\prime}}{\partial x_j}}\:\!}. \end{align}

The particles are moved by the coherent fluctuations of the concentration and velocity, and energy in these velocity fluctuations dissipated by viscosity.

For larger particles, we must use a more rigorous phase-averaging approach. We use the results in Drew & Passman (Reference Drew and Passman1999), and the details are provided in the supplementary material. We will state the non-dilute version of the definitions of variables and later impose the dilute assumption. The velocity field of the mixture is the weighted average of the velocity of the fluid and solid phases

(2.4) \begin{align} {\:\!\overline {\vphantom { \vphantom {\phi }\phi ^{\prime} }\boldsymbol {u}}\:\!} &= \frac {1}{ {\:\!\overline {\vphantom { \vphantom {\phi }\phi ^{\prime} }\rho }\:\!} } \left ({ {\:\!\overline {\vphantom { \vphantom {\phi }\phi ^{\prime} }\phi }\:\!} \rho _f {\:\!\overline {\vphantom { \vphantom {\phi }\phi ^{\prime} }\boldsymbol {u}}\:\!} _f + (1- {\:\!\overline {\vphantom { \vphantom {\phi }\phi ^{\prime} }\phi }\:\!}) \rho _s {\:\!\overline {\vphantom { \vphantom {\phi }\phi ^{\prime} }\boldsymbol {u}}\:\!} _s }\right). \end{align}

In this formalism, the concentration is only defined during the ensemble average so that $ {\:\!\overline {\vphantom { \vphantom {\phi }\phi ^{\prime} }\phi }\:\!} =\phi$ . The particles move relative to the the mixture for two reasons, settling $\tilde {\boldsymbol {u}}$ and turbulent effects $\Delta {\:\!\overline {\vphantom { \vphantom {\phi }\phi ^{\prime} }\boldsymbol {u}}\:\!}$ , so that

(2.5) \begin{align} {\:\!\overline {\vphantom { \vphantom {\phi }\phi ^{\prime} }\boldsymbol {u}}\:\!} _s &= {\:\!\overline {\vphantom { \vphantom {\phi }\phi ^{\prime} }\boldsymbol {u}}\:\!} + \tilde {\boldsymbol {u}} + \Delta {\:\!\overline {\vphantom { \vphantom {\phi }\phi ^{\prime} }\boldsymbol {u}}\:\!}. \end{align}

The TKE is defined as

(2.6) \begin{align} k = \frac { {\:\!\overline {\vphantom { \vphantom {\phi }\phi ^{\prime} }\rho u_j u_j}\:\!} - {\:\!\overline {\vphantom { \vphantom {\phi }\phi ^{\prime} }\rho }}\, {\overline {\vphantom { \vphantom {\phi }\phi ^{\prime} }u_j}}\, {\overline {\vphantom { \vphantom {\phi }\phi ^{\prime} }u_j}\:\!} }{2 {\:\!\overline {\vphantom { \vphantom {\phi }\phi ^{\prime} }\rho }\:\!} }. \end{align}

In the dilute limit, $ {\:\!\overline {\vphantom {\vphantom {\phi }\phi ^{\prime}}\boldsymbol {u}}\:\!}$ tends to $ {\:\!\overline {\vphantom {\vphantom {\phi }\phi ^{\prime}}\boldsymbol {u}}\:\!}_f$ , $k$ tends to $({1}/{2}){\:\!\overline {\vphantom {\vphantom {\phi }\phi ^{\prime}} \vphantom {\phi }u_i^{\prime} \vphantom {\phi }u_i^{\prime}}\:\!}$ and the expressions in (2.2) apply. However, the transport of particles and the viscous dissipation appear different

(2.7) \begin{align} J_j & : = {\:\!\overline {\vphantom { \vphantom {\phi }\phi ^{\prime} }\phi }\:\!} \Delta {\:\!\overline {\vphantom { \vphantom {\phi }\phi ^{\prime} }u}\:\!} _j, & \tilde {\epsilon }_K & : = \nu {\:\!\overline {\vphantom { \vphantom {\phi }\phi ^{\prime} }\frac {\partial \vphantom {\phi }u_i^{\prime}}{\partial x_j}\frac {\partial \vphantom {\phi }u_i^{\prime}}{\partial x_j}}\:\!} - R {\:\!\overline {\vphantom { \vphantom {\phi }\phi ^{\prime}}\phi }\:\!} g_i \tilde {u}_i. \end{align}

The particles are transported by relative motion of the particle phase to the mixture, this relative motion a consequence of turbulence. The settling velocity is, by definition, that velocity at which the work done by gravity due to the settling motion is precisely balanced by the viscous dissipation in the fluid, and we take $\tilde {\epsilon }_K$ to represent all other viscous dissipation of TKE.

The boundary conditions for (2.1) (strictly the unsimplified system (A1)) are as follows. At the bed we impose no slip, i.e.

(2.8a) \begin{align} {\:\!\overline {\vphantom {\vphantom {\phi }\phi ^{\prime}}u_j}\:\!} = \vphantom {\phi }u^{\prime} = \vphantom {\phi }v^{\prime} &= 0 &&\text {for}& z &= b(x,y). \end{align}

Over a rigid bed we additionally have $w^{\prime}=J_3=0$ , and the treatment of the viscous boundary layer at the bed requires care (important in the analysis of §§ 4 and 6); for this reason, we include the full wall boundary layer in the model. For flows over an erosional bed we allow $w^{\prime} \neq 0$ , $J_3 \neq 0$ , so that the erosion is captured by $J_3$ . In the far field we require

(2.8b) \begin{align} \vphantom {\phi }u_j^{\prime} = J_j = {\:\!\overline {\vphantom { \vphantom {\phi }\phi ^{\prime} }\phi }\:\!} = p^T = \vphantom {\phi }p^{\prime} &= 0 &&\text {for}& z &=H(x,y,t). \end{align}

In Parker et al. (Reference Parker, Fukushima and Pantin1986, Reference Parker, Garcia, Fukushima and Yu1987), $z=H$ is taken to be $z \to \infty$ , and it is also assumed that $ {\:\!\overline {\vphantom { \vphantom {\phi }\phi ^{\prime}}u}\:\!} = {\:\!\overline {\vphantom {\vphantom {\phi }\phi ^{\prime} }v}\:\!} = 0$ in the far field, so that all entrainment comes from the $ {\:\!\overline {\vphantom { \vphantom {\phi }\phi ^{\prime}}w}\:\!}$ at $z \to \infty$ . A similar approximation is used for jets, mixing layers, wakes and boundary layers (e.g. Pope Reference Pope2000). The entrainment velocity is then defined, for a top-hat model, as

(2.8c) \begin{align} w_e = \frac {\partial h}{\partial t} - \overline {w}\,\Bigg |_{z \to \infty }. \end{align}

The function $h(x,y,t)$ is termed the depth of the current: for a top-hat model this is the location of the interface between the current and ambient. In gravity currents, the approximation $z\to \infty$ works well under a deep quiescent ambient fluid which is our primary focus, but interpretation becomes problematic with a weakly counterflowing ambient. We use a more careful formulation with $z=H$ some interface beyond the strong influence of the gravity current, but still with $H-b \sim \mathscr {L}_3$ . This change does not alter the depth-average model but aids interpretation. The equivalent condition in our formulation is

(2.8d) \begin{align} w_e = \frac {\partial }{\partial t} ({\varsigma _h h}) + {\left ({ \sum _{\alpha =1}^2 {\:\!\overline {\vphantom {\vphantom {\phi }\phi ^{\prime}} u_\alpha }\:\!} \frac {\partial H}{\partial x_\alpha } - {\:\!\overline {\vphantom { \vphantom {\phi }\phi ^{\prime}}w}\:\!} }\right)}\Bigg |_{z=H}. \end{align}

We require that, at the elevation $z=H$ , $ {\:\!\overline {\vphantom { \vphantom {\phi }\phi ^{\prime} }u_\alpha }\:\!}$ ( $\alpha \in \{1,2\}$ ) are small compared with their values within the current so that minimal momentum is transferred between the ambient and the current, ${ {\:\!\overline {\vphantom { \vphantom {\phi }\phi ^{\prime} }u_\alpha }\:\!} }_H \ll \mathscr {U}_\alpha$ . In turbulent flow $w_e\gt 0$ , the flow entrains ambient fluid. The depth function $h$ , the inclusion of the factor $\varsigma _h$ , and other consequences of this definition are explored in § 5. For now we simply note that typically $\varsigma _h \approx 1$ .

We close this subsection by performing some manipulations of the system (2.1). Firstly, the mean-flow kinetic energy (MKE) equation is obtained by multiplying (2.1c ) by $ {\:\!\overline {\vphantom { \vphantom {\phi }\phi ^{\prime} }u_{\alpha }}\:\!}$ and summing over $\alpha \in \{1,2\}$ and adding (2.1d ) times $ {\:\!\overline {\vphantom { \vphantom {\phi }\phi ^{\prime} }w}\:\!}$ , giving

(2.9) \begin{equation} \frac {\partial e}{\partial t} + \frac {\partial }{\partial x_j}\left ({\:\!\overline {\vphantom { \vphantom {\phi }\phi ^{\prime} }u_j}\:\!} e + {\:\!\overline {\vphantom {\vphantom {\phi }\phi ^{\prime} }u_j}\:\! p^T}\right) - \frac {\partial }{\partial z} \sum _{\alpha =1}^2\left ({\tau _{3\alpha }^R + \nu \frac {\partial {\:\!\overline {\vphantom { \vphantom {\phi }\phi ^{\prime} }u_\alpha }\:\!} }{\partial z}}\right) {\:\!\overline {\vphantom { \vphantom {\phi }\phi ^{\prime} }u_\alpha }\:\!} = - \tilde {\mathcal {P}} -\tilde {\epsilon }_M + R g_j {\:\!\overline {\vphantom { \vphantom {\phi }\phi ^{\prime} }u_j}}\, {\overline {\vphantom { \vphantom {\phi }\phi ^{\prime} }\phi }\:\!}, \end{equation}

where

(2.10) \begin{align} e & : = \frac {1}{2} \sum _{\alpha =1}^2 {\:\!\overline {\vphantom { \vphantom {\phi }\phi ^{\prime} }u_\alpha }\:\!} ^2, & \tilde {\epsilon }_M & : = \sum _{\alpha =1}^2 \nu \frac {\partial {\:\!\overline {\vphantom { \vphantom {\phi }\phi ^{\prime} }u_\alpha }\:\!} }{\partial z}\frac {\partial {\:\!\overline {\vphantom { \vphantom {\phi }\phi ^{\prime} }u_\alpha }\:\!} }{\partial z}. \end{align}

Technically, $e$ is only the portion of the full kinetic energy resulting from the $x,y$ components of the mean flow; this expression arises naturally under the shallow assumption $\mathscr {L}_1, \mathscr {L}_2 \gg \mathscr {L}_3$ and turns out to be the useful quantity in this context. Secondly, the GPE that would be released if the excess mass of the current $R {\:\!\overline {\vphantom { \vphantom {\phi }\phi ^{\prime} }\phi }\:\!}$ was moved from elevation $z$ to $b$ is $R g (z-b) {\:\!\overline {\vphantom { \vphantom {\phi }\phi ^{\prime} }\phi }\:\!} \cos \theta$ . The evolution equation for $(z-b) {\:\!\overline {\vphantom { \vphantom {\phi }\phi ^{\prime} }\phi }\:\!}$ is, by (2.1b ),

(2.11) \begin{align} \frac {\partial }{\partial t} (z-b) {\:\!\overline {\vphantom { \vphantom {\phi }\phi ^{\prime} }\phi }\:\!} & + \frac {\partial }{\partial x_j} (z-b) {\:\!\overline {\vphantom { \vphantom {\phi }\phi ^{\prime} }u_j}}\, {\overline {\vphantom { \vphantom {\phi }\phi ^{\prime} }\phi }\:\!} + \frac {\partial }{\partial z} (z-b) \left (\tilde {w} {\:\!\overline {\vphantom { \vphantom {\phi }\phi ^{\prime} }\phi }\:\!+ J_3}\right)\nonumber \\[-3pt] &\quad = \lbrack { {\:\!\overline {\vphantom { \vphantom {\phi }\phi ^{\prime} }w}\:\!} + \tilde {w}}\rbrack {\:\!\overline {\vphantom { \vphantom {\phi }\phi ^{\prime} }\phi }\:\!} + J_3 - \sum _{\alpha =1}^2 \frac {\partial b}{\partial x_\alpha } {\:\!\overline {\vphantom {\vphantom {\phi }\phi ^{\prime}}u_\alpha }}\, {\overline {\vphantom { \vphantom {\phi }\phi ^{\prime} }\phi }\:\!}. \end{align}

Again, technically this is not the full GPE, which is $R g_i x_i {\:\!\overline {\vphantom {\vphantom {\phi }\phi ^{\prime}}\phi }\:\!}$ , but the portion of the GPE that turns out to be the useful quantity for shallow flows.

2.2. Averaging over the depth

The equations in § 2.1 are integrated over the depth $b \leq z \leq H$ . The resulting system of equations is written in terms of the depth-averaged variables

(2.12) \begin{align} \Phi = \frac {1}{h} \int _b^{H} {\:\!\overline {\vphantom {\vphantom {\phi }\phi ^{\prime}}\phi }\:\!} \mathrm {d}{z}, \qquad U_1 \frac {1}{h} \int _b^{H} {\:\!\overline {\vphantom { \vphantom {\phi }\phi ^{\prime}}u_1}\:\!} \mathrm {d}{z}, \qquad U_2 = \frac {1}{h} \int _b^{H} {\:\!\overline {\vphantom { \vphantom {\phi }\phi ^{\prime}}u_2}\:\!} \mathrm {d}{z}, \qquad K = \frac {1}{h} \int _b^{H} k \mathrm {d}{z}. \end{align}

Recall that, in general, $h \neq H-b$ , and that under a deep quiescent ambient we may take $H \to \infty$ to simplify. For the concentration field $ {\:\!\overline {\vphantom { \vphantom {\phi }\phi ^{\prime}}\phi }\:\!}$ (as an example) at a single point in $(x,y,t)$ , the depth average $\Phi$ has eliminated all information about the variation of concentration with $z$ . We carry this information forwards using a shape function $\xi _\phi = {\:\!\overline {\vphantom { \vphantom {\phi }\phi ^{\prime}}\phi }\:\!} /\Phi$ which describes the variation of concentration relative to the depth average. It is informative to write the shape function not as a function of $z$ but of $\zeta : = (z-b)/h$ . For some gravity currents, a careful choice of $h$ will enable the shape functions to be invariant with respect to $(x,y,t)$ , all profiles of $ {\:\!\overline {\vphantom { \vphantom {\phi }\phi ^{\prime}}\phi }\:\!}$ being the same up to a rescaling in magnitude, $\Phi$ , and a vertical stretch, $h$ ; self-similarity. Generalising to all variables the decomposition takes the form

(2.13a) \begin{align} {\:\!\overline {\vphantom {\vphantom {\phi }\phi ^{\prime}}\phi }\:\!}(x,y,z,t) &= \xi _\phi (x,y,\zeta,t) \cdot \Phi (x,y,t), \end{align}

(2.13b) \begin{align} {\:\!\overline {\vphantom {\vphantom {\phi }\phi ^{\prime} }u_1}\:\!} (x,y,z,t) &= \xi _1(x,y,\zeta,t) \cdot U_1(x,y,t), \end{align}

(2.13c) \begin{align} {\:\!\overline {\vphantom { \vphantom {\phi }\phi ^{\prime} }u_2}\:\!} (x,y,z,t) &= \xi _2(x,y,\zeta,t) \cdot U_2(x,y,t), \end{align}

(2.13d) \begin{align} k(x,y,z,t) &= \xi _k(x,y,\zeta,t) \cdot K(x,y,t), \end{align}

where $\xi _\phi$ , $\xi _1$ , $\xi _2$ and $\xi _k$ are shape functions satisfying

(2.13e) \begin{align} \int _0^{\zeta _H} \xi _{\scriptstyle \bullet }(x,y,\zeta,t) \mathrm {d}{\zeta } &= 1, & \zeta _H & : = \frac {H-b}{h}. \end{align}

Consequently, $\Phi$ , $U_1$ , $U_2$ and $K$ are the depth-average values of density, velocity and TKE, while $\xi _\phi$ , $\xi _1$ , $\xi _2$ and $\xi _k$ capture the shape of the current. In particular, the expressions defined in table 1 are the features of the shape that influence the averaged properties of the current. The majority of the shape factors are of the form

(2.14) \begin{align} \sigma _{AB} &= \int _0^{\zeta _H} \xi _A \xi _B \mathrm {d}{\zeta }, &&\text {or}& \sigma _{ABC} &= \int _0^{\zeta _H} \xi _A \xi _B \xi _C \mathrm {d}{\zeta }, \end{align}

where we use $\xi _z = 2 \zeta$ . The exceptions are indicated by the $\tilde {\sigma }_{\scriptstyle \bullet }$ or $\varsigma _{\scriptstyle \bullet }$ notation. Note that our choice of normalisation (2.13e ) and shape factors (table 1 and figure 2) are different to those used by some authors, see Appendix B.

Table 1.

Definitions and properties of the shape factors. In the definitions we use the subscripts $\alpha,\beta,\gamma$ to indicate numerical values of $1$ or $2$ . We state the equations in this section where the shape factor appears, or else the first equation where it appears. The values for top-hat flow are calculated using (2.15).

Figure 2.

Violin plots of the distribution of values that each shape factor can take, computed from the dataset compiled by Fukuda et al. (Reference Fukuda, de Vet, Skevington, Bastianon, Fernández, Wu, McCaffrey, Naruse, Parsons and Dorrell2023) (Simmons et al. (Reference Simmons, Azpiroz-Zabala, Cartigny, Clare, Cooper, Parsons, Pope, Sumner and Talling2020) are field data, all others are experimental). For each shape factor (marked to the left) we plot the probability density for the distribution (computed using a kernel method) as a black line. In the computation, the data are weighted to account for the large number of samples from some sources. The data points for each shape factor are plotted at the horizontal location of their value, and given a random vertical displacement within the kernel. Here, $h$ is calculated by setting $\sigma _{11}=1$ , and $\varsigma _h$ by (5.1) with $\delta =10^{-2}$ .

There are two special cases of flows we will consider, along with the general case. Firstly, when the shape of the current is in self-similar form, the shape functions do not depend on $x$ , $y$ or $t$ , meaning the shape factors in table 1 are constants. Secondly, for top-hat flow, the shape functions take the form

(2.15) \begin{align} \xi _{\phi } &= \begin{cases} 0, & 1 \lt \zeta, \\ 1, & 0 \lt \zeta \lt 1, \\ \varsigma _\phi & \zeta = 0, \end{cases} & \xi _1 = \xi _2 = \xi _k &= \begin{cases} 0, & 1 \lt \zeta, \\ 1, & 0 \lt \zeta \lt 1, \end{cases} \end{align}

which yields the special values in table 1.

We next integrate the equations in § 2.1 over the depth $b \leq z \leq H$ using the results in Appendix C and applying the boundary conditions (2.8) (Ellison & Turner Reference Ellison and Turner1959; Parker et al. Reference Parker, Fukushima and Pantin1986). The indices $\alpha,\beta,\gamma$ take a value of $1$ or $2$ and sums range over these values. Conservation of fluid volume is, by (2.1a ),

(2.16) \begin{gather} \frac {\partial }{\partial t} (\varsigma _h h) + \sum _\beta \frac {\partial }{\partial x_\beta } \Big (\underbrace { h U_\beta }_{{\text {volume flux}}} \Big) = S_h = \underbrace { w_e \vphantom {\big)}}_{{\qquad \text {entrainment velocity}}}. \end{gather}

Conservation of particle volume is, by (2.1b ),

(2.17) \begin{gather} \frac {\partial }{\partial t}\left ({h \Phi }\right) + \sum _\beta \frac {\partial }{\partial x_\beta } \Big ( \underbrace { \sigma _{\beta \phi } h U_\beta \Phi \vphantom {\big)}}_{{\text {particle flux}}} \Big) = S_{\Phi } = - \underbrace { \varsigma _{\phi } w_s \Phi \cos \theta \vphantom {\big)} }_{{\text {deposition}}} + \underbrace { E_s. \vphantom {\big)} }_{{\text {erosion}}} \end{gather}

Conservation of momentum is, by (2.1c ), using (2.1d ) to calculate the hydrostatic pressure,

(2.18a) \begin{gather} \frac {\partial }{\partial t} \left ({h U_\alpha }\right) + \sum _\beta \frac {\partial }{\partial x_\beta } \Big (\underbrace { \sigma _{\beta \alpha } h U_\beta U_\alpha \vphantom {\big)}}_{{\text {momentum flux}}} \Big) + \frac {\partial }{\partial x_\alpha } \Big (\underbrace { {\frac {1}{2}} \sigma _{z\phi } R g h^2 \Phi \cos \theta \vphantom {\big)}}_{{\text {pressure}}} \Big) = S_{\alpha }, \end{gather}

(2.18b) \begin{gather} S_{\alpha } = - \underbrace { R g h \Phi \cos \theta \frac {\partial b}{\partial x_\alpha } \vphantom {\bigg)} }_{{\text {pressure on bed slope}}} \qquad - \underbrace { u_{\alpha \ast }^2 \vphantom {\bigg)} }_{{\text {basal drag}}} \qquad + \underbrace { R g_\alpha h \Phi. \vphantom {\bigg)} }_{{\text {downslope gravity}}} \end{gather}

Conservation of MKE is, by (2.9),

(2.19a) \begin{align} &\sum _\beta \frac {\partial }{\partial t} \left ({ {\textstyle \frac {1}{2}} \sigma _{\beta \beta } h U_\beta ^2 }\right) + \sum _{\beta \gamma } \frac {\partial }{\partial x_\beta } \Big ( \underbrace {{\textstyle \frac {1}{2}} \sigma _{\beta \gamma \gamma } h U_\beta U_\gamma ^2 \vphantom {\big)}}_{{\text {MKE flux}}}\Big)\nonumber \\ &\qquad \qquad + \sum _\beta \frac {\partial }{\partial x_\beta } \Big ( \underbrace {{\textstyle \frac {1}{2}} \tilde {\sigma }_{\beta z\phi } U_\beta R g h^2 \Phi \cos \theta \vphantom {\big)}}_{{\text {pressure work}}} \Big) = S_M = \tilde {S}_M - \underbrace {h \mathcal {P} - h \mathcal {B}_M \vphantom {\big)}}_{{\text {energy transfer}}}, \end{align}

(2.19b) \begin{align} &\tilde {S}_M = \sum _\beta \underbrace { \sigma _{\beta \phi } U_\beta R g_\beta h \Phi \vphantom {\Big)}}_{{\text {work by downslope gravity}}} \qquad \qquad - \underbrace {h \epsilon _M \vphantom {\Big)}}_{{\text {MKE dissipation}}}.\end{align}

Conservation of GPE is, by the product of (2.11) and $R g \cos \theta$ ,

(2.20a) \begin{align} \frac {\partial }{\partial t}\!\left ({ {\textstyle \frac {1}{2}} \sigma _{z\phi } R g h^2 \Phi \cos \theta }\right) & + \sum _\beta \frac {\partial }{\partial x_\beta } \Big (\! \underbrace {{\textstyle \frac {1}{2}} \sigma _{\beta z\phi } U_\beta R g h^2 \Phi \cos \theta \vphantom {\big)}}_{{\text {GPE flux}}}\! \Big) = S_G = \tilde {S}_G + \underbrace {h \mathcal {B}_M + h \mathcal {B}_K \vphantom {\big)}}_{{\text {energy transfer}}},\nonumber\\[4pt] \end{align}

(2.20b) \begin{align} \tilde {S}_G = - \sum _\beta \underbrace { \sigma _{\beta \phi } U_\beta R g h \Phi \cos \theta \frac {\partial b}{\partial x_\beta }}_{{\text {variation in datum}}} - \underbrace { w_s R g h \Phi (\cos \theta)^2}_{{\text {energy loss to settling}}}. \end{align}

(The variation in datum arises because we measure GPE relative to the local bed elevation in (2.11)). Conservation of TKE is, by (2.1e ),

(2.21a) \begin{gather} \frac {\partial }{\partial t} ({ h K }) + \sum _\beta \frac {\partial }{\partial x_\beta } \Big ( \underbrace {\sigma _{\beta k} h U_\beta K \vphantom {\big)}}_{{\text {TKE flux}}} \Big) = S_K = \tilde {S}_K + \underbrace {h \mathcal {P} - h \mathcal {B}_K \vphantom {\big)}}_{{\text {energy transfer}}}, \end{gather}

(2.21b) \begin{gather} \tilde {S}_K = - \underbrace {h \epsilon _K \vphantom {\big)}}_{{\text {TKE dissipation}}}. \end{gather}

Summing (2.19)–(2.21) yields the equation for conservation of total energy, $E$ ,

(2.22a) \begin{align} &\frac {\partial }{\partial t} ({h E}) + \sum _{\beta \gamma } \frac {\partial }{\partial x_\beta } \Big (\underbrace {{\textstyle \frac {1}{2}} \sigma _{\beta \gamma \gamma } h U_\beta U_\gamma ^2 \vphantom {\big)}}_{{\text {MKE flux}}} \Big)\nonumber \\ &\qquad \qquad + \sum _\beta \frac {\partial }{\partial x_\beta } \Big ( \underbrace {\sigma _{\beta k} h U_\beta K \vphantom {\big)}}_{{\text {TKE flux}}} + \underbrace {\varsigma _{\beta z\phi } U_\beta R g h^2 \Phi \cos \theta \vphantom {\big)}}_{{\qquad \text {GPE flux and pressure work}}} \Big) = S_E, \end{align}

(2.22b) \begin{align} S_E &= - \sum _\beta \underbrace { \sigma _{\beta \phi } U_\beta R g h \Phi \cos \theta \frac {\partial b}{\partial x_\beta } \vphantom {\bigg)} }_{{\text {variation in GPE datum}}} + \sum _\beta \underbrace { \sigma _{\beta \phi } U_\beta R g_\beta h \Phi \vphantom {\bigg)} }_{{\text {work by downslope gravity}}} \qquad \qquad \nonumber \\ &\quad - \underbrace { h \epsilon _T \vphantom {\big)} }_{{\text {total dissipation}}} \qquad - \underbrace { w_s R g h \Phi (\cos \theta)^2, \vphantom {\big)} }_{{\text {GPE loss to settling}}} \end{align}

where the total energy (per unit mass) is defined as

(2.23) \begin{equation} E = \sum _\beta \underbrace {{\textstyle \frac {1}{2}} \sigma _{\beta \beta } U_\beta ^2 \vphantom {\big)}}_{{\text {MKE}}} + \underbrace {K \vphantom {\big)}}_{{\text {TKE}}} + \underbrace {{\textstyle \frac {1}{2}} \sigma _{z\phi } R g h \Phi \cos \theta \vphantom {\big)}}_{{\text {GPE}}}. \end{equation}

While we will call $E$ the total energy going forward, it is technically only the portion of the total energy that is useful for describing shallow flows; see the discussion below (2.10) and (2.11). Throughout our stating of the depth-averaged system, we have used the definitions of settling velocity, erosion, basal shear velocity and depth-average total, mean flow and turbulent dissipation as

(2.24) \begin{equation} \begin{gathered} \begin{aligned} w_s \cos \theta &= - \tilde {w}, & E_s &= { J_3 }\Bigg |_b, & u_{\alpha \ast }^2 &= { \nu \frac {\partial {\:\!\overline {\vphantom { \vphantom {\phi }\phi ^{\prime} }u_\alpha }\:\!} }{\partial z} }\Bigg |_b, \end{aligned} \\ \begin{aligned} \epsilon _T &= \epsilon _M + \epsilon _K, & h \epsilon _M &= \int _b^{H} \tilde {\epsilon }_M \mathrm {d}{z}, & h \epsilon _K &= \int _b^{H} \tilde {\epsilon }_K \mathrm {d}{z} - {T_3}\Big |_b. \end{aligned} \end{gathered} \end{equation}

The terms marked as ‘energy transfer’ are the depth-averaged TKE production, turbulent buoyancy flux and mean-flow buoyancy flux, defined as

(2.25a) \begin{align} \mathcal {P} & : = \frac {1}{h} \int _b^{H} \tilde {\mathcal {P}} \mathrm {d}{z}, \end{align}

(2.25b) \begin{align} \mathcal {B}_K & : = \frac {R g \cos \theta }{h} \int _b^{H} J_3 \mathrm {d}{z}. \end{align}

(2.25c) \begin{align} \mathcal {B}_M & : = \frac {R g \cos \theta }{h} \int _b^{H} {\:\!\overline {\vphantom { \vphantom {\phi }\phi ^{\prime} }w}}\, {\overline {\vphantom { \vphantom {\phi }\phi ^{\prime} }\phi }\:\!} \mathrm {d}{z}, \end{align}

respectively. We understand $\mathcal {P}$ as the rate of conversion of MKE to TKE, $\mathcal {B}_K$ as the rate of conversion of TKE to GPE and $\mathcal {B}_M$ as the rate of conversion of MKE to GPE.

We briefly note that in figure 2 and other figures going forward, the data considered are for unidirectional currents ( $U_2=\partial /\partial y=0$ ) and the depth is defined following Ellison & Turner (Reference Ellison and Turner1959) and Parker et al. (Reference Parker, Fukushima and Pantin1986) by setting $\sigma _{11} = 1$ ; that is take

(2.26) \begin{equation} h = \left ({ \int _b^{H} {\:\!\overline {\vphantom { \vphantom {\phi }\phi ^{\prime} }u}\:\!} \mathrm {d}{z} } \right)^2 \Bigg / { \int _b^{H} {\:\!\overline {\vphantom { \vphantom {\phi }\phi ^{\prime} }u}\:\!} ^2 \mathrm {d}{z} }. \end{equation}

In our analysis $h$ is arbitrary, and will be discussed thoroughly in § 5.

2.3. Interpretation as a classical volumetric model

As discussed in the introduction, the purpose of this derivation is to produce a system of equations for the same unknown functions as Parker et al. (Reference Parker, Fukushima and Pantin1986), but allowing for the imposition of shape functions selected by the modeller to improve accuracy. We have also incorporated arbitrary slowly varying topography $b$ , again this can be specified by the modeller, along with generalising to three-dimensional flows (that is, we include variation in $y$ ). The equations governing volume (2.16), particles (2.17) and momentum (2.18) are precisely these generalisations, and the case $\varsigma _h=1$ has been used previously (Dorrell et al. Reference Dorrell, Darby, Peakall, Sumner, Parsons and Wynn2014; Sher & Woods Reference Sher and Woods2015; Negretti et al. Reference Negretti, Flòr and Hopfinger2017). For the energetics, we have derived a set of three equations for the separate contributions from MKE (2.19), GPE (2.20) and TKE (2.21), along with their sum, which describes the evolution of the total energy (2.22). Parker et al. (Reference Parker, Fukushima and Pantin1986) present their model with an equation for the evolution of TKE, and so we may naively think that (2.21) is the appropriate equation. However, the key consideration is not the component of energy modelled, but the closures required to complete the model. As was found by Parker et al. (Reference Parker, Fukushima and Pantin1986) in their top-hat model, the energy transfer terms are intimately related to other model closures, and we must therefore eliminate them from the model. That is not to say these terms are not important (Odier et al. Reference Odier, Chen and Ecke2014), but rather that they cannot be specified independently. The only equation which does not include energy transfer terms is the equation for total energy (2.22). This forms the model: the functions $h$ , $\Phi$ , $U_\alpha$ , $K$ are solved for using the system of equations (2.16)–(2.18) and (2.22). To close the system the modeller must specify the topography, $b$ , and shape of the concentration, velocity and turbulence fields through the shape functions in (2.13) which give rise to the shape factors in the model through the expressions in table 1. Additionally, the modeller must specify the parameters and closures present in the model from Parker et al. (Reference Parker, Fukushima and Pantin1986), including entrainment of ambient fluid at speed $w_e$ , erosion of the bed at rate $E_s$ , drag from the bed with shear velocity $u_{\alpha \ast }$ and viscous dissipation of energy at rate $\epsilon _T$ .

3.1. Derivation of energetic consistency requirements

We begin by rearranging the equations for MKE (2.19a ) and GPE (2.20a ) so that the time derivatives can be easily substituted from the governing equations for volume, particles and momentum (2.16)–(2.18)

(3.1a) \begin{align} S_M = - \left (\!{\sum _\beta \frac {1}{2} \sigma _{\beta \beta } U_\beta ^2}\right) \frac {\partial h}{\partial t} & + \sum _\beta \sigma _{\beta \beta } U_\beta \frac {\partial }{\partial t}(h U_\beta) + \sum _\beta \frac {1}{2} h U_\beta ^2 \frac {\partial \sigma _{\beta \beta }}{\partial t}\nonumber \\ & + \sum _\beta \frac {\partial }{\partial x_\beta } \left (\!{ \sum _\gamma \frac {1}{2} \sigma _{\beta \gamma \gamma } h U_\beta U_\gamma ^2 + \frac {1}{2} \tilde {\sigma }_{\beta z\phi } U_\beta R g h^2 \Phi \cos \theta }\right)\!, \end{align}

(3.1b) \begin{align} S_G = \frac {1}{2} \sigma _{z\phi } R g h \Phi \cos \theta \frac {\partial h}{\partial t} & + \frac {1}{2} \sigma _{z\phi } R g h \cos \theta \frac {\partial }{\partial t} \left ({h\Phi }\right) + \frac {1}{2} R g h^2 \Phi \cos \theta \frac {\partial \sigma _{z\phi }}{\partial t}\qquad\qquad\nonumber \\ & \qquad\qquad\qquad\quad\qquad + \sum _\beta \frac {\partial }{\partial x_\beta } \left ({ \frac {1}{2} \sigma _{\beta z \phi } U_\beta R g h^2 \Phi \cos \theta }\right). \end{align}

Due to the structure of the source terms $S_M$ , $S_G$ and $S_K$ , it is not possible to rearrange them into expressions for $\mathcal {P}$ , $\mathcal {B}_K$ and $\mathcal {B}_M$ . Instead, two of these can be expressed in terms of the remaining one. We choose to express $\mathcal {P}$ and $\mathcal {B}_K$ in terms of $\mathcal {B}_M$ , which yields

(3.2) \begin{align} h \mathcal {P} &= - h \mathcal {B}_M - \left ({ S_M - \tilde {S}_M }\right),& h \mathcal {B}_K = - h \mathcal {B}_M + \left ({ S_G - \tilde {S}_G }\right), \end{align}

with the second equality in the TKE equation (2.21a ) then stating simply $S_{E} + S_K + S_G = \tilde {S}_{E} + \tilde {S}_K + \tilde {S}_G$ . The reason for this choice is that $\mathcal {B}_M$ is a property of the mean flow, and can therefore be deduced from the equations describing the mean flow, whereas the others are properties of the turbulence requiring closure. A consequence of this choice is that $h \mathcal {P}$ will be derived as the implied loss of MKE, while $h \mathcal {B}_K$ is derived as the implied gain of GPE, and this will be seen in the expressions to follow. To deduce $\mathcal {B}_M$ , we employ the continuity equation

(3.3) \begin{align} \frac {\partial {\:\!\overline {\vphantom { \vphantom {\phi }\phi ^{\prime} }u}\:\!} }{\partial x} + \frac {\partial {\:\!\overline {\vphantom { \vphantom {\phi }\phi ^{\prime} }v}\:\!} }{\partial y} + \frac {\partial {\:\!\overline {\vphantom { \vphantom {\phi }\phi ^{\prime} }w}\:\!} }{\partial z} &= 0, &&\text {thus}& {\:\!\overline {\vphantom { \vphantom {\phi }\phi ^{\prime} }w}\:\!} &= - \int _b^{z} { \frac {\partial {\:\!\overline {\vphantom { \vphantom {\phi }\phi ^{\prime} }u}\:\!} }{\partial x} + \frac {\partial {\:\!\overline {\vphantom { \vphantom {\phi }\phi ^{\prime} }v}\:\!} }{\partial y} }\Big |_{z^{\prime}} \mathrm {d}{z^{\prime}}. \end{align}

Substituting into the definition of $\mathcal {B}_M$ (2.25c ) and decomposing into shape functions (2.13) to rewrite in terms of shape factors, we obtain the expression for $h \mathcal {B}_M$ given below. To obtain expressions for the other turbulent energy transfers, we begin with the expressions in (3.2), into which we substitute our expression for $h \mathcal {B}_M$ , the rearranged energy equations (3.1) and eliminate time derivatives using the governing equations (2.16)–(2.18). These manipulations are algebraically challenging, and verification of the results using the computer algebra package Maple are provided as supplemental information. Firstly, the depth-integrated rate of conversion of MKE to TKE is

(3.4) \begin{align} h \mathcal {P} = & \frac {1}{2} \left ({\sum _\beta \frac {\sigma _{\beta \beta }}{\varsigma _h} U_\beta ^2}\right) w_e + \sum _\beta \sigma _{\beta \beta } U_\beta u_{\beta \ast }^2 - h \epsilon _M\nonumber \\ & + \sum _\beta (\sigma _{\beta \phi }-\sigma _{\beta \beta }) U_\beta R h \Phi \left ({ g_\beta - g \cos \theta \frac {\partial b}{\partial x_\beta } }\right)\nonumber \\ & + \frac {1}{2} \sum _{\beta \gamma } \left ({ - \frac {\sigma _{\gamma \gamma }}{\varsigma _h} + 2 \sigma _{\gamma \gamma } \sigma _{\beta \gamma } - \sigma _{\beta \gamma \gamma } }\right) U_\gamma ^2 U_\beta \frac {\partial h}{\partial x_\beta }\nonumber \\ & + \frac {1}{2} \sum _{\beta } \left ({ - \frac {\sigma _{\beta \beta }}{\varsigma _h} + 4\sigma _{\beta \beta }^2 - 3\sigma _{\beta \beta \beta } }\right) h U_\beta ^2 \frac {\partial U_\beta }{\partial x_\beta }\nonumber \\ & + \frac {1}{2} \sum _{\beta \neq \gamma } \Bigg\lgroup \left ({ - \frac {\sigma _{\gamma \gamma }}{\varsigma _h} + 2 \sigma _{\gamma \gamma } \sigma _{\beta \gamma } - \sigma _{\beta \gamma \gamma } }\right) h U_\gamma ^2 \frac {\partial U_\beta }{\partial x_\beta }\nonumber\\ & \qquad\qquad\ \ + 2 \left ({ \sigma _{\gamma \gamma } \sigma _{\beta \gamma } - \sigma _{\beta \gamma \gamma } }\right) h U_\beta U_\gamma \frac {\partial U_\gamma }{\partial x_\beta } \Bigg\rgroup \nonumber \\[3pt] & + \sum _{\beta } \left ({ \sigma _{\beta \beta } \sigma _{z\phi } - \varsigma _{\beta z\phi } }\right) U_\beta R g h \Phi \cos \theta \frac {\partial h}{\partial x_\beta }\nonumber \\[3pt] & + \frac {1}{2} \sum _{\beta } \left ({ \sigma _{\beta \beta } \sigma _{z\phi } - \tilde {\sigma }_{\beta z\phi } }\right) U_\beta R g h^2 \cos \theta \frac {\partial \Phi }{\partial x_\beta }\nonumber \\[3pt] & + \frac {1}{2} \sum _\beta \Bigg\lgroup { - \frac {1}{\varsigma _h} \frac {\partial }{\partial t} \left ({\varsigma _h \sigma _{\beta \beta }}\right) + \sum _\gamma \left ({ 2 \sigma _{\beta \beta } \frac {\partial \sigma _{\beta \gamma }}{\partial x_\gamma } - \frac {\partial \sigma _{\beta \beta \gamma }}{\partial x_\gamma } }\right) U_\gamma }\Bigg\rgroup h U_\beta ^2\nonumber \\[3pt] & + \frac {1}{2} \sum _\beta \left ({\sigma _{\beta \beta } \frac {\partial \sigma _{z\phi }}{\partial x_\beta } - \frac {\partial \tilde {\sigma }_{\beta z \phi }}{\partial x_\beta } + \tilde {\sigma }_{D\beta z \phi }}\right) U_\beta R g h^2 \Phi \cos \theta. \end{align}

The terms in the first line are the energy required to accelerate entrained fluid, the work done by drag and the MKE lost to viscous effects. The second line results from an imbalance between work done by downslope gravity on the mean velocity and that from velocity/density profiles, along with the imbalance of the implied work by varying GPE datum and the vertical transport of material. The third line is the result of imbalances in the energy changes associated with varying depth from a changing volume of fluid, work done by depth-average momentum and the depth-resolved transport of MKE; the fourth, fifth, and sixth lines similarly with the effect of varying velocity. The seventh and eighth lines result from an imbalance of work done by the depth-average pressure/GPE and depth-resolved transport of these quantities. The ninth line includes the effective variation in MKE from the shape functions varying temporally and spatially, and similarly the variation in MKE flux. The tenth line includes the imbalance between depth-average and depth-resolved pressure work from varying shape functions, along with the vertical transport generated by varying $\xi _\beta$ . Next, the depth-integrated rate of conversion of TKE to GPE is

(3.5) \begin{align} h \mathcal {B}_K = & \frac {1}{2} R g h \cos \theta \Bigg \lgroup \frac {\sigma _{z\phi }}{\varsigma _h} \Phi w_e + \lbrack { 2 - \sigma _{z\phi } \varsigma _\phi }\rbrack w_s \Phi \cos \theta + \sigma _{z\phi } E_s\nonumber \\ & + \sum _\beta\! \left (\!{ - \frac {\sigma _{z\phi }}{\varsigma _h} - \sigma _{\beta \phi }\sigma _{z\phi } + 2\varsigma _{\beta z \phi } }\right)\! \Phi \frac {\partial}{\partial x_\beta} {(h U_\beta)} + \sum _\beta \left ({ - \sigma _{\beta \phi }\sigma _{z\phi } + \sigma _{\beta z \phi } }\right) U_\beta h \frac {\partial \Phi }{\partial x_\beta }\nonumber \\ & + \Bigg\lbrack {\varsigma _h \frac {\partial }{\partial t} \left ({ \frac {\sigma _{z\phi }}{\varsigma _h}}\right) + \sum _{\beta } \left ({ - \sigma _{z\phi } \frac {\partial \sigma _{\beta \phi }}{\partial x_\beta } + \frac {\partial \sigma _{\beta z \phi }}{\partial x_\beta } + \tilde {\sigma }_{D\beta z \phi } }\right) U_\beta }\Bigg\rbrack h \Phi \Bigg \rgroup. \end{align}

The terms in the first line are the GPE generated by entrainment, the energy expended holding particles in suspension less the amount deposited, and the energy required to erode. The first term in the second line is the imbalance in GPE from varying volume flux as a consequence of changing depth, depth-average transport of GPE and depth-resolved transport of both GPE and pressure; the second term similar with varying concentration. The third line includes the effect of varying shape factors, including the direct change of GPE, the imbalance between depth-average and depth-resolved GPE flux, along with the vertical transport generated by varying $\xi _\beta$ . Finally, the depth-average rate of conversion of MKE to GPE is

(3.6) \begin{align} h \mathcal {B}_M = & \frac {1}{2} R g h \Phi \cos \theta \sum _\beta \Bigg ( 2 \sigma _{\beta \phi } U_\beta \frac {\partial b}{\partial x_\beta } - \tilde {\sigma }_{\beta z\phi } h \frac {\partial U_\beta }{\partial x_\beta }\qquad\quad\qquad\nonumber\\ & \qquad\qquad\qquad\qquad\qquad\qquad\quad + \left ({\sigma _{\beta z\phi }-\tilde {\sigma }_{\beta z\phi }}\right) U_\beta \frac {\partial h}{\partial x_\beta } - \tilde {\sigma }_{D\beta z\phi } h U \Bigg), \end{align}

where the terms are the increase in GPE from flow generated by varying bed elevation, the work done by pressure due to velocity gradients, the imbalance in the change of GPE and work done by pressure due to changing depth and the vertical transport generated by varying $\xi _\beta$ .

Figure 3.

The difference between the coefficients of the energy transfer terms and the top-hat approximation to these terms (3.7), plotted using the same format as figure 2. Note that the majority of coefficients vanish in a top-hat model, in which case the coefficients are plotted without modification. There are sections dedicated to: $h\mathcal {P}$ (3.4); $h \mathcal {B}_K$ (3.5); $h \mathcal {B}_M$ (3.6); the pseudo-equilibrium simplification (4.4); and the volume-free production (5.14), as indicated to the right.

Using the values of the shape factors in table 1, top-hat flows simplify to

(3.7a) \begin{align} h \mathcal {P} &= U_1 u_{1\ast }^2 + U_2 u_{2\ast }^2 + {\textstyle \frac {1}{2}} (U_1^2 + U_2^2) w_e - h \epsilon _M, \end{align}

(3.7b) \begin{align} h \mathcal {B}_K &= {\textstyle \frac {1}{2}} R g h \cos \theta \left ({ \Phi w_e + \lbrack { 2 - \varsigma _\phi }\rbrack w_s \Phi \cos \theta + E_s }\right), \end{align}

(3.7c) \begin{align} h \mathcal {B}_M &= {\textstyle \frac {1}{2}} R g h \Phi \cos \theta \left ({ 2 U \frac {\partial b}{\partial x} + 2 V \frac {\partial b}{\partial y} - h \frac {\partial U}{\partial x} - h \frac {\partial V}{\partial y} }\right). \end{align}

For two-dimensional flow ( $U_2={\partial b}/{\partial y}=0$ ) with negligible mean-flow dissipation in the boundary layer ( $\epsilon _M=0$ ), the first two expressions match those from Parker et al. (Reference Parker, Fukushima and Pantin1986), and the third is consistent though never explicitly stated.

However, using the values from figure 2 we find that the additional terms from a complete analysis (3.4)–(3.6) are not negligible. The difference between the coefficients and the top-hat approximation is plotted in figure 3. This can be understood as representing each of $h \mathcal {P}$ , $h \mathcal {B}_K$ and $h \mathcal {B}_M$ as the sum of the top-hat model and a correction based on the shape of the current, and a coefficient of $\pm 1$ means the correction from including shape factors is as large as the contribution from the top-hat model. Of course, we are only comparing the size of the coefficients, and the size of the term also depends on what this coefficient multiplies, but this analysis gives an indication of which terms are important. The region where the terms are of magnitude less than $10\,\%$ of the top-hat terms is indicated by a grey band, and while a few of the coefficients do lie in this region where they can (arguably) be neglected, many of them lie substantially outside of this region. Thus, we cannot neglect the effect of shape on the energetics.

3.2. Using energetic consistency requirements

The governing equations we have derived includes balances for volume (2.16), particles (2.17), momentum (2.18) and total energy (2.22). This system is similar to that proposed by Parker et al. (Reference Parker, Fukushima and Pantin1986), but crucially allows for the specification of flow shape in a model that captures energetics. Shape factors are known to have a substantial influence on prediction in models which do not capture energetics (Dorrell et al. Reference Dorrell, Darby, Peakall, Sumner, Parsons and Wynn2014). This modelling framework has the potential to be much more accurate because it allows for the inclusion of additional physics. However, these additional physics appear in the model as closures specified by the modeller, and there is a possibility that the selected closures are unsuitable in a subtle way. Suppose that we take a selection of developed closures for shape functions (Islam & Imran Reference Islam and Imran2010; Abad et al. Reference Abad, Sequeiros, Spinewine, Pirmez, Garcia and Parker2011), entrainment (Ellison & Turner Reference Ellison and Turner1959; Parker et al. Reference Parker, Fukushima and Pantin1986; Cenedese & Adduce Reference Cenedese and Adduce2010) and erosion (Parker et al. Reference Parker, Fukushima and Pantin1986; Dorrell et al. Reference Dorrell, Amy, Peakall and McCaffrey2018; Guo Reference Guo2020), in addition to making some reasonable choice of drag coefficient, and constructing some expression for the total dissipation (while method for this construction is presented in Parker et al. (Reference Parker, Fukushima and Pantin1986), it would be better to have an empirically verified closure). Then we can simulate a current using the governing equations, and post-process the results to obtain the energy transfer that implicitly occurred using (3.4)–(3.6). It is extremely unlikely that independently developed closures will somehow yield the energy transfer seen in real currents.

Consequently, the purpose of (3.4)–(3.6) is as consistency requirements to be used during the development of closures. When a set of closures for the governing equations have been developed using data from experiment or simulation, a final verification can be performed by using the developed closures to predict the energy transfer, which can be compared with the measured energy transfer. This final verification, if successful, demonstrates that the set of closures are energetically consistent and will produce the correct evolution of TKE. While the TKE may not always be the subject of interest, having a model which is tightly constrained to real physics gives confidence in the model in general.

For these consistency requirements to be useful in closure development, two things are required. Firstly, the balances must be verified. Provided we are dealing with a shallow Boussinesq gravity current under a deep unstratified ambient, this is principally verification that there have been no algebraic mistakes in the derivation. Secondly, some simplification of the consistency requirements must be made in order to make them usable for model development. The supplied Maple script can be used to apply simplifications to specific circumstances. The next section provides an example of simplifying the system to a specific simulation configuration, and verifying the accuracy of both the full and simplified form of the energetic transfer expressions.

4.1. Two-dimensional pseudo-equilibrium flow

In this subsection we consider two-dimensional steady flow down a smooth slope ( ${\partial }/{\partial t} = {\partial }/{\partial y} = U_2 = b = 0$ , $\theta \neq 0$ , use notation $U \equiv U_1$ ). Under such conditions we may expect the fluid to reach a pseudo-equilibrium configuration after flowing a sufficient distance, where the shape enters self-similar form (shape factors constant) and the following dimensionless parameters become independent of $x$ :

(4.1) \begin{align} \frac {U}{\sqrt {R g h \cos \theta }}, && \frac {w_s\cos \theta }{\sqrt {K}}, && \frac {U^2}{2 K}. \end{align}

Up to constant coefficients, these are the Froude number (ratio of flow speed to velocity scale), the Rouse number (ratio of particle settling to upwards turbulent diffusion) and the ratio of MKE to TKE, respectively. These conditions imply

(4.2) \begin{align} \frac {\partial }{\partial x} \left ({h \Phi }\right) = \frac {\partial U}{\partial x} = \frac {\partial K}{\partial x} = 0,\end{align}

as used by Parker et al. (Reference Parker, Fukushima and Pantin1986). Simplifying the governing system (2.16)–(2.18) and (2.22) using the pseudo-equilibrium conditions (4.2)

(4.3a) \begin{gather} \frac {\partial h}{\partial x} = \frac {w_e}{U}, \end{gather}

(4.3b) \begin{gather} 0 = \varsigma _{\phi } w_s \Phi \cos \theta - E_s, \end{gather}

(4.3c) \begin{gather} \left ({ \sigma _{11} U^2 + {\textstyle \frac {1}{2}} \sigma _{z\phi } R g h \Phi \cos \theta }\right) \frac {w_e}{U} = R g h \Phi \sin \theta - u_{1\ast }^2, \end{gather}

(4.3d) \begin{align} & \left ({ {\textstyle \frac {1}{2}} \sigma _{111} U^2 + \sigma _{1k} K + \varsigma _{1z\phi } R g h \Phi \cos \theta }\right) w_e\qquad\qquad\qquad\qquad\qquad\qquad\qquad\qquad\qquad \nonumber \\& \qquad\qquad\qquad\qquad\qquad\qquad = \sigma _{1\phi } U R g h \Phi \sin \theta - h \epsilon _T - w_s R g h \Phi (\cos \theta)^2. \end{align}

Figure 4.

The depth-integrated energy transfer calculated using the pseudo-equilibrium balance (4.4), plotted using the same format as figure 2. The value of $h \mathcal {B}_M$ can be positive or negative, so it is split over two lines.

To obtain consistency requirements for the energetics in pseudo-equilibrium we substitute (4.2) and (4.3) into the full expressions for $h \mathcal {P}$ , $h \mathcal {B}_K$ and $h \mathcal {B}_M$ (3.4)–(3.6) to obtain

(4.4a) \begin{align} h \mathcal {P} &= \sigma _{1\phi } U u_{1\ast }^2 + \left ({ 2 \sigma _{11} \sigma _{1\phi } - \sigma _{111} }\right) {\textstyle \frac {1}{2}} U^2 w_e \notag \\ & \quad + \left ({ \sigma _{z\phi } \sigma _{1\phi } - \sigma _{1z\phi } }\right) {\textstyle \frac {1}{2}} R g h \Phi \cos \theta w_e - h \epsilon _M, \end{align}

(4.4b) \begin{align} h \mathcal {B}_K &= {\textstyle \frac {1}{2}} R g h \Phi \cos \theta \left ( { \tilde {\sigma }_{1z\phi } w_e + 2 w_s \cos \theta }\right), \end{align}

(4.4c) \begin{align} h \mathcal {B}_M &= {\textstyle \frac {1}{2}} R g h \Phi \cos \theta \left ({ \sigma _{1z\phi } - \tilde {\sigma }_{1z\phi } }\right) w_e. \end{align}

This simplification puts $h \mathcal {P}$ and $h \mathcal {B}_K$ in a format where they are easily compared with top-hat expressions in (3.7), revealing an alteration in the magnitude of the terms and the introduction of some additional terms, see figure 3. Compared with (3.7) the turbulent production $h \mathcal {P}$ typically has a smaller contribution from the energy needed to accelerate entrained fluid up to speed, but a new contribution from the energy required to uplift the mass, while the turbulent production induced by entrainment is reduced, and similarly the production by the mean flow $h \mathcal {B}_M$ has a negative contribution from the energy consumed by the entrainment uplift. The magnitude of the pseudo-equilibrium transfer in real currents is plotted in figure 4, showing that $h \mathcal {B}_M$ is small for this regime (but the non-equilibrium terms may be large). It is difficult to compare the sizes of $h \mathcal {B}_K$ and $h \mathcal {P}$ without knowing the mean-flow dissipation $h \epsilon _M$ . Note that the currents used to plot figure 4 are not necessarily in pseudo-equilibrium balance, see § 6.

Figure 5.

Plots computed from the DNS data of Zúñiga et al. (Reference Zúñiga, Balachandar, Yang, Zhang, Smith, Loppi, Cantero and Kerkemeier2024). (a-c) The time-averaged fields of concentration, velocity and TKE, the depths $h$ (2.26) and $\tilde {h}$ (5.1) shown in dashed and dash-dot white respectively. (d) The spatial variation of the depth-average quantities. (e) The spatial variation of the shape factors. (f) The residual in the full depth-averaged equations (2.16)–(2.18) and (2.22) or the pseudo-equilibrium equations (4.3), computed from the simulation data using (4.5). In the legend the equation is indicated using the corresponding conserved quantity. (g) The residual in the energy equations (2.22) and (4.3d ) split over contributions from MKE, GPE and TKE in (2.19)–(2.21), dividing by full energy flux rather than the flux for the specific equation.

To investigate the pseudo-equilibrium dynamics we employ data from a Direct Numerical Simulation (DNS) originally published in Zúñiga et al. (Reference Zúñiga, Balachandar, Yang, Zhang, Smith, Loppi, Cantero and Kerkemeier2024). (For similar simulations see Salinas et al. Reference Salinas, Cantero, Shringarpure and Balachandar2019a ,Reference Salinas, Cantero, Shringarpure and Balachandar b , Reference Salinas, Zúñiga, Cantero, Shringarpure, Fedele, Hoyal and Balachandar2020, Reference Salinas, Balachandar, Shringarpure, Fedele, Hoyal and Cantero2021, Reference Salinas, Balachandar and Cantero2022, Reference Salinas, Balachandar, Zúñiga, Shringarpure, Fedele and Cantero2023, and for an experimental configuration see Odier et al. Reference Odier, Chen and Ecke2014.) Here, a conservative ( $w_s=E_s=0$ ) gravity current flowed down a slope of $\theta = 3^{\circ }$ over $0\lt x\lt 730$ ( $b=0$ ), having been fed into the domain at $Re = 5650$ at $x=0$ , $0\lt z\lt 2$ , the units of their simulation such that $Rg = 1/\sin \theta$ . The flow was simulated until it became statistically steady in time, and then time-averaged statistics were computed. The resulting flow is shown in figures 5(a)–5(c). We compute the depth-averaged variables and shape factors directly through integration over $0 = b \leq z \leq H = 60$ , the full height of the simulation domain, see figure 5(d,e). On $x\gt 300$ the depth averages satisfy well the pseudo-equilibrium conditions (4.2), $K$ being the last to become steady. However, the concentration shape $\xi _\phi$ continues to slowly evolve over the entire length of the simulation as can be seen through the shape factors $\sigma _{z\phi }$ , $\sigma _{1z\phi }$ and $\tilde {\sigma }_{1z\phi }$ .

Before going any further, we validate the depth-average model (2.16)–(2.18) and (2.22). In steady state each equation is of the form ${\mathcal {F}}/{x} = \mathcal {S}$ for some flux $\mathcal {F}$ and source $\mathcal {S}$ , and we calculate the residual in each as

(4.5) \begin{align} \mathcal {R} = \frac {1}{\mathcal {F}} \left \lvert { \frac {\partial \mathcal {F}}{\partial x} - \mathcal {S} } \right \rvert,\end{align}

were the flux and source are computed directly from the simulation data. Consequently, $\mathcal {R}^{-1}$ is the distance over which cumulative error would produce an $\mathcal {O}{(1)}$ change to the flux if we were to integrate (4.5) with the given source. (In fact, since the residuals alternate sign, the change to the flux may be seen as a random walk and thus residuals accumulate over a distance of $\mathcal {R}^{-2}$ .) By figure 5( f), once the current has established ( $x\gt 100$ ) the length scale of residual accumulation is around $10^5$ , vastly longer than the simulation domain. Splitting the residual in the energy equation over its components (figure 5 g) we see a similarly good agreement. The larger residual on $x\lt 100$ is a consequence of the assumption that horizontal scales are much larger than vertical by which we omitted terms in (2.1). We also plot the residual in the pseudo-equilibrium balance, ensuring the computed residual is simply the direct simplification of (4.5) employing the assumptions. Once the flow has reached the balance ( $x\gt 300$ ) the residual in the system (figure 5 f) is no larger than for the full equation, showing that this is an accurate simplification, however, the residual in the split energy equations (figure 5 g) is slightly larger with an accumulation length scale of $10^4$ , which we will find is a consequence of the varying shape factors.

Figure 6.

Further plots computed from the DNS data of Zúñiga et al. (Reference Zúñiga, Balachandar, Yang, Zhang, Smith, Loppi, Cantero and Kerkemeier2024). (a) The properties of turbulence, entrainment and drag computed directly from DNS data. (b–d) Comparing the values of $h\mathcal {P}$ , $h\mathcal {B}_K$ and $h\mathcal {B}_M$ direct from simulation to the expressions in the full model (3.4)–(3.6), along with the simplifications of self-similar flow ((3.4)–(3.6) neglecting derivatives of shape factors), the pseudo-equilibrium balance (4.4) and top-hat flow (3.7). (e) The coefficients of the additional terms in (4.4) with respect to top-hat flow. ( f) The dimensionless mean-flow dissipation (red) and the approximation using Reichardt (Reference Reichardt1951) (black dashed).

The properties of the turbulence and other source terms used in the residual analysis are computed directly from the simulation data using (2.8d ), (2.24) and (2.25), and are plotted in figure 6(a). Crucially, we find that the viscous dissipation of the mean flow $h \epsilon _M$ is almost equal to the turbulent production $h \mathcal {P}$ . Whether $h \mathcal {P} \simeq h \epsilon _M$ in general is discussed later.

The values of $h\mathcal {P}$ , $h\mathcal {B}_K$ and $h\mathcal {B}_M$ are plotted in figures 6(b)–6(d). We see that the values of the full model expressions from (3.4) to (3.6) agree very well with the exact values from the simulations across all $x$ (including small $x$ where the curves go out of the figure, the small amount of noise at large $x$ arises from the $x$ derivatives). Reducing to self-similar form by neglecting the $x$ derivatives of shape functions in the consistency requirements (3.4)–(3.6) introduces a large error in the proximal region $x\lt 300$ (see figure 6 b–d), but in the distal region $x\gt 300$ where the flow is equilibrated the error is small and provides a reasonable approximation. Simplifying further to the pseudo-equilibrium expressions (4.4) does not increase the error in the distal regions, the assumptions on the depth-averaged variables (4.2) being satisfied to a much greater degree of accuracy than the assumption that shape functions are independent of $x$ . The top-hat approximation increases the error more substantially. Plotting the coefficients of the difference of the top-hat approximation and the pseudo-equilibrium approximation in figure 6(e) (the same coefficients as in figure 3) we find that they are moderately large, which is the cause of the error.

4.2. Mean-flow dissipation

Given the comparison in figures 6(b)–6(d), the greatest error in approximating the turbulent processes arises from neglecting the mean-flow dissipation $h \epsilon _M$ because $h \mathcal {P} \simeq h \epsilon _M$ . We ask, is this an unusual property of this flow or something we should expect to see in general? The mean-flow dissipation is a property of near bed flow. Working in dimensionless wall variables $u^+ = {\:\!\overline {\vphantom { \vphantom {\phi }\phi ^{\prime} }u}\:\!} /u_{1\ast }$ and $z^+ = z u_{1\ast }/\nu$ we see that

(4.6) \begin{align} h \epsilon _M = \int _0^{H} \nu \left ({\frac {\partial {\:\!\overline {\vphantom { \vphantom {\phi }\phi ^{\prime} }u}\:\!} }{\partial z}}\right)^2 \mathrm {d}{z} = u_{1\ast }^3 \int _0^{H u_{1\ast }/\nu } \left ({\frac {\partial u^+}{\partial z^+}}\right)^2 \mathrm {d}{z^+}. \end{align}

While internal shear layers also technically contribute to this dissipation, these produce a change in velocity of order $U$ and thus to have a leading-order contribution the the mean-flow dissipation they must have a thickness of order $\nu U^2/u_{1\ast }^3$ so that the Reynolds number of the shear layer is of order $U^3/u_{1\ast }^3$ . In environmental currents the Reynolds number of shear layers is much larger, even in the presence of jet sharpening (Dorrell et al. Reference Dorrell, Peakall, Darby, Parsons, Johnson, Sumner, Wynn, Özsoy and Tezcan2019), and so the contribution to the mean-flow dissipation is only from the viscous sub-layer between the bed and the logarithmic layer. This motivates a definition of the dimensionless mean-flow dissipation as

(4.7) \begin{align} \epsilon _M^+ : = \int _0^{H u_{1\ast }/\nu } \left ({\frac {\partial u^+}{\partial z^+}}\right)^2 \mathrm {d}{z^+} = \frac {h \epsilon _M}{u_{1\ast }^3}. \end{align}

As a rough estimate, we may expect that there is a viscous sub-layer with $u^+ = z^+$ over $0\lt z^+\lt 10$ , and a turbulent log layer where $u^+ = \kappa ^{-1} \ln z^+ + \text {const}$ on $z^+\gt 10$ , where $\kappa = 0.41$ is the von Kármán constant, resulting in the approximation $\epsilon _M^+ \approx 10 + {1}/{10 \kappa ^2} = 10.6$ showing that even the log layer does not contribute significantly to the value. The estimate can be improved using the matched asymptotic of Reichardt (Reference Reichardt1951) (see Kadivar, Tormey & McGranaghan Reference Kadivar, Tormey and McGranaghan2021) which yields a value of $\epsilon _M^+ = 9.15$ . We plot $\epsilon _M^+$ in figure 6( f) and find that it lies very close to this estimate, but there is some slow downstream evolution which is similar in kind to the evolution of the shape factors in figure 5(e).

Figure 7.

Extrapolation of the dissipation to high Reynolds numbers, using three datasets of channel flow simulations: $+$ from Lee & Moser (Reference Lee and Moser2015), $\times$ from Kaneda & Yamamoto (Reference Kaneda and Yamamoto2021) and $\circ$ from Orlandi (Reference Orlandi2019). Here, $\epsilon _M^+$ is red and $\epsilon _K^+$ is blue. (a) The channel flow data and the gravity current data from Zúñiga et al. (Reference Zúñiga, Balachandar, Yang, Zhang, Smith, Loppi, Cantero and Kerkemeier2024) (solid). (b) Curves of best fit for the channel flow dissipation. Plotted faintly are samples from a probability distribution over curves, showing the uncertainty of the best-fit curve. (c) Extrapolation of the dissipation best fit, and ratio of the extrapolations (purple). We also show a power-law extrapolation of the best-fit ratio through the final data point. Two abscissae are given for (c) showing $Re_\tau$ and $Re$ . (d) The Reynolds number for particulate gravity currents, plotted using the same format as figure 2.

While it is true that $\epsilon _M^+$ is finite at high Reynolds numbers, the dimensionless turbulent dissipation, $\epsilon _K^+ : = h \epsilon _K / u_{\ast 1}^3$ , grows at high Reynolds number, so that $\epsilon _M^+/\epsilon _K^+ \to 0$ as $Re \to \infty$ . But how large does $Re$ need to be for $\epsilon _K^+$ to dominate? To answer, we use data from channel flow between two boundaries (sometimes called plane Poiseuille flow) from Lee & Moser (Reference Lee and Moser2015) and Kaneda & Yamamoto (Reference Kaneda and Yamamoto2021). For this analysis we will employ a Reynolds number $Re_\tau$ where the length scale is the half-height of the channel and the velocity scale is $u_{\ast 1}$ . Figure 7(a) shows that, sufficiently far downstream of the release, the gravity current of Zúñiga et al. (Reference Zúñiga, Balachandar, Yang, Zhang, Smith, Loppi, Cantero and Kerkemeier2024) has the same dissipations as channel flow (length scale in $Re_\tau$ is flow depth (2.26)). From the data, the dissipations have curves of best fit given by

(4.8) \begin{align} \epsilon _M^+ &= 9.17 \cdot (\ln Re_\tau - 5.01)^{-0.0148}, & \epsilon _K^+ &= 3.62 \cdot (\ln Re_\tau - 3.28)^{0.856}, \end{align}

with coefficients of determination ( $R^2$ ) of $0.8867$ and $0.9997$ , respectively (figure 7 b). To account for the uncertainty in parameters we use a Bayesian approach. The uncertainty of the data is estimated as the root-mean-square difference between the points and the respective best-fit line. We use a uniform (improper) prior for the parameters, so the Bayesian approach is equivalent to a likelihood approach. Sample curves from the resulting posterior distribution are plotted faintly in figure 7(b,c) showing almost no scatter. Extrapolations of these curves, and their ratio, are plotted in figure 7(c), which has axes for both $Re_\tau$ and $Re$ . The latter is calculated with the channel-average velocity as the velocity scale (equivalent to depth-average velocity for gravity currents), the two Reynolds numbers related by $Re_\tau \simeq 0.09 Re^{0.88}$ (Pope Reference Pope2000). To account for error in the functional form in the extrapolation we also show a power-law (straight line in log scale) extrapolation of the best fit curve from the final data-point (dashed). We say that the mean-flow dissipation is small when it is below $10\,\%$ of the turbulent dissipation, and negligible when below $1\,\%$ . The small threshold is reached somewhere in the interval $10^{18} \lesssim Re_\tau \lesssim 10^{21}$ ( $10^9$ for power-law extrapolation). The threshold is substantially above the largest values ever measured in gravity currents, $Re_\tau \lesssim 10^7$ (figure 7 d, green points from Simmons et al. Reference Simmons, Azpiroz-Zabala, Cartigny, Clare, Cooper, Parsons, Pope, Sumner and Talling2020). Thus, mean-flow dissipation is relevant at the Reynolds numbers of all geophysical currents.

Figure 8.

(a) Extrapolation of the dissipation to high roughness, $\circ$ are data from Orlandi (Reference Orlandi2019). Here, $\epsilon _M^+$ is red, $\epsilon _K^+$ is blue and their ratio is purple, with $Re_\tau$ given above. Curves of best fit are shown, and plotted faintly are samples from a probability distribution over curves, showing the uncertainty of the best-fit curves. (b) The mean particle diameter for particulate gravity currents, plotted using the same format as figure 2.

The data from channel flow considered so far are all from smooth channels, and particulate gravity currents always flow over a rough bed, the roughness appearing at the particle scale and (for environmental currents) at the sales of benthic fauna and bedforms (Olu et al. Reference Olu, Decker, Pastor, Caprais, Khripounoff, Morineaux, Baziz, Menot and Rabouille2017; Sen et al. Reference Sen, Dennielou, Tourolle, Arnaubec, Rabouille and Olu2017; Azpiroz-Zabala et al. Reference Azpiroz-Zabala2024). A large body of research exists into flow over rough beds, see the recent review of Kadivar et al. (Reference Kadivar, Tormey and McGranaghan2021). There are a great many parameters which can be used to characterise the bed roughness as documented by Thakkar, Busseb & Sandhama (Reference Thakkar, Busseb and Sandhama2017). For research on how roughness effects the log-law region see Shringarpure et al. (Reference Shringarpure, Cantero and Balachandar2012)and Bilgin & Cantwell (Reference Bilgin and Cantwell2023), and for the additional layers of the flow which are present in flow over roughness see Nikora et al. (Reference Nikora, Koll, McEwan, McLean and Dittrich2004), Mazzuoli & Uhlmann (Reference Mazzuoli and Uhlmann2017)and Forooghi et al. (Reference Forooghi, Stroh, Schlatter and Frohnapfel2018). We use data from Orlandi (Reference Orlandi2019), representing the roughness using the equivalent sand roughness $k_s^+$ which is calculated from the log layer by fitting $u^+ = (1/\kappa) \log (y^+/k_s^+) + 8$ (e.g. Schlichting & Gersten Reference Schlichting and Gersten2016). The best fit curves in figure 8(a) are

(4.9) \begin{align} \epsilon _M^+ &= 38.9 \cdot (k_s^+ + 5.62)^{-0.719}, & \epsilon _K^+ &= 15.5 \cdot (k_s^+ + 70.3)^{-0.201}, \end{align}

with coefficients of determination $0.9978$ and $0.7126$ . For the mean-flow dissipation to be small we require $k_s^+ \gtrsim 500$ and to be negligible $k_s^+ \gtrsim 10^4$ . We expect these thresholds to be only weakly dependent on $Re_\tau$ , because $\epsilon _K^+$ is largely independent of $k_s^+$ and only varying by a factor of $\sim 10$ over the $Re_\tau$ of gravity currents, while $\epsilon _M^+$ is independent of $Re_\tau$ . To reach $k_s^+ \gtrsim 10^4$ requires bedforms $\gtrsim 10^4 d_{50}^+$ high (figure 8 b), green points from Simmons et al. Reference Simmons, Azpiroz-Zabala, Cartigny, Clare, Cooper, Parsons, Pope, Sumner and Talling2020), which may occur in natural settings but not in experiments. Mean-flow dissipation is therefore important for experimental flows and moderately sized turbidity currents, but not for large currents over bedforms.

4.3. Summary of comparison with simulations

There are two distinct alterations to the energetics of Parker et al. (Reference Parker, Fukushima and Pantin1986) that have been discussed in this section: flow shape and mean-flow dissipation. Regarding flow shape, there is a clear hierarchy of approaches shown in figure 6: the top-hat model is the least accurate, then the approaches which include flow shape but not its variation, and most accurate is the model which includes the variations of flow shape. When fully including flow shape and providing closures from the simulation, the model approaches the accuracy of the full simulation, showing that the energetics can, in principle, be accurately captured by a depth-average model. This accuracy improvement is relevant for predictive models (§ 2.3) because the flow shape is incorporated into the models. Conversely, the mean-flow dissipation is only relevant for assessing the consistency requirements (§ 3), which are important for model development and interpretation. It is not included in the predictive model outside of the total dissipation, which can be approximated separately. Nonetheless, it can make a leading-order contribution to the implied energetics which is important for the use of consistency requirements in closure development (§ 3.2), and the understanding of particle suspension (§ 6).

5.1. Gravity current depth

To this point we have not discussed the depth $h(x,y,t)$ ; the model presented is independent of the definition. A wide variety of measures of depth can be used (e.g. Salinas et al. Reference Salinas, Balachandar, Zúñiga, Shringarpure, Fedele and Cantero2023), and the symmetry group in Appendix D can be used to transform between different measures. For two-dimensional flow ( $U_2= {\partial }/{\partial y}=0$ ) Ellison & Turner (Reference Ellison and Turner1959) measure the depth by enforcing $\sigma _{11} = 1$ , equivalent to (2.26). To be consistent with previous work we have used (2.26) in all figures, and state explicitly whenever it is used with most expressions being for general depth. The advantage of (2.26) is that it enables self-similar flow to be captured whenever it occurs, if instead depth is taken to be some arbitrary function then even if the flow profiles are self-similar the shape functions in (2.13) will still depend on $(x,y,t)$ . As a general point, whatever definition of depth is used in a depth-averaged model, it must enable self-similar flow to be captured.

There is another measure of depth in the system: $\tilde {h} : = \varsigma _h h$ appears in the definition of entrainment (2.8d ) which gives rise to the volume equation (2.16). By (2.8a ) the entrainment across $z=b+\tilde {h}$ should be the same as across $z=H$ , so $b+\tilde {h}$ can be taken to be any elevation above the region in which the velocity field is influenced by the presence of the current. However, the surface $z=b+\tilde {h}$ moves with the velocity field, and we wish for the shape factor $\varsigma _h=\tilde {h}/h$ to be a constant for self-similar flow. This means that it must be on the edge of where the velocity field is influenced by the current, that is

(5.1) \begin{align} \int _{b}^{b+\tilde {h}} \sqrt { {\:\!\overline {\vphantom { \vphantom {\phi }\phi ^{\prime} }u}\:\!} ^2+ {\:\!\overline {\vphantom { \vphantom {\phi }\phi ^{\prime} }v}\:\!} ^2} \mathrm {d}{z} &= (1-\delta) \int _b^{H} \sqrt { {\:\!\overline {\vphantom { \vphantom {\phi }\phi ^{\prime} }u}\:\!} ^2+ {\:\!\overline {\vphantom { \vphantom {\phi }\phi ^{\prime} }v}\:\!} ^2} \mathrm {d}{z} &&\text {for}& 0 \lt \delta &\ll 1. \end{align}

Our definition of $\tilde {h}$ is comparable to that used in some experimental investigations of entrainment, in that it seeks a level at which the bed-parallel velocity has become small (Odier et al. Reference Odier, Chen and Ecke2014; Maggi et al. Reference Maggi, Negretti, Hopfinger and Adduce2023). In these, the entrainment rate is defined for steady flow in terms of the velocity of fluid flow across this surface, $w_e = - w(z=\tilde {h})$ . However, in a depth-average model, the conditions which arise most naturally are

(5.2) \begin{equation} \begin{aligned} w_e &= \frac {\partial }{\partial x} (Uh) &&&& \text {for spatially evolving steady flow, and} \\ w_e &= \frac {\partial \tilde {h}}{\partial t} &&&&\text {for temporally evolving uniform flow.} \end{aligned} \end{equation}

The condition for steady flow is well known (Ellison & Turner Reference Ellison and Turner1959; Negretti et al. Reference Negretti, Flòr and Hopfinger2017; Martin, Negretti & Hopfinger Reference Martin, Negretti and Hopfinger2019; Maggi et al. Reference Maggi, Negretti, Hopfinger and Adduce2023), the uniform case less so. The definition of $\tilde {h}$ is important because it controls the time evolution of depth through conservation of volume (2.16): steady currents should have the same entrainment rate as uniform currents when they have the same flow properties. Further investigation is required to establish whether (5.1) gives the same entrainment in these two scenarios. Throughout, we use (5.1) with $\delta = 0.01$ .

5.2. Relating entrainment to turbulence

The discussion above reveals the ad hoc nature of the volume equation (2.16). The interface $z = \tilde {h}$ is not a physical property, but rather is constructed through heuristic arguments, with different researchers coming up with different constructions. The entrainment rate $w_e$ then measures the flow rate across our constructed interface. This formulation is a historical artefact: the original depth-averaged model was presented by Saint-Venant (Reference Saint-Venant1871) for open-channel flow, and in that context there is a clear interface between air and water determining $z = \tilde {h}$ . The source term is no longer entrainment, but could be interpreted as heavy rain inflating the volume of the river or canal. There is no such clarity for gravity currents: sometimes the upper interface is moderately sharp, but often it is highly diffuse and the attempt to represent it as a mathematical surface is artificial.

Likewise, the physical interpretation of entrainment $w_e$ is not clear: it is some measure of the progressive mixing and dilution of the current, but one dependent on the imposed interface. The physical origin of entrainment is known to be turbulence, in particular it is the buoyancy flux which leads to the fluid becoming more dilute and disperse (Strang & Fernando Reference Strang and Fernando2001; Arneborg et al. Reference Arneborg, Fiekas, Umlauf and Burchard2007; Wells et al. Reference Wells, Cenedese and Caulfield2010; Odier et al. Reference Odier, Chen and Ecke2014). There is a natural velocity which arises from these turbulent fluctuations, which we will term the turbulent buoyancy velocity,

(5.3) \begin{equation} w_{\mathcal {B}} : = {2 \int _b^H J_3 \mathrm {d}{z}}\Bigg / {\int _b^H {\:\!\overline {\vphantom { \vphantom {\phi }\phi ^{\prime} }\phi }\:\!} \mathrm {d}{z}} = \frac {2 \mathcal {B}_K}{R g \Phi \cos \theta }.\end{equation}

For a top-hat model (table 1) of a compositional current ( $w_s=E_s=0$ ) the consistency requirement for $\mathcal {B}_K$ (3.7b ) gives $w_e = w_{\mathcal {B}}$ , consistent with Wells et al. (Reference Wells, Cenedese and Caulfield2010). For a particle-driven current, (3.7b ) becomes

(5.4) \begin{equation} w_{\mathcal {B}} = w_e + \lbrack { 2 - \varsigma _\phi }\rbrack w_s \cos \theta + \frac {E_s}{\Phi }. \end{equation}

This can be used to eliminate $w_e$ from the governing equations, after which the top-hat volume equation becomes

(5.5) \begin{gather} \frac {\partial h}{\partial t} + \sum _\beta \frac {\partial }{\partial x_\beta } (h U_\beta) = w_e = w_{\mathcal {B}} - \lbrack { 2 - \varsigma _\phi }\rbrack w_s \cos \theta - \frac {E_s}{\Phi }.\end{gather}

Different approaches have been taken to close the entrainment in model of particle-driven currents. Suppose that $\hat {w}_e$ is a closure of entrainment in compositional currents. Historical models either neglect entrainment outright (Bonnecaze et al. Reference Bonnecaze, Huppert and Lister1993), or assume that closures can be transferred directly from compositional to particle-driven currents as $w_e = \hat {w}_e$ (Parker et al. Reference Parker, Fukushima and Pantin1986). However, there is a range of physical processes in particle-driven currents that are not present in compositional currents, and so it is not obvious that the aggregate effect of these processes (entrainment) is transferred so easily. Instead, it is reasonable to suppose that the effects of turbulence are most similar, the large-scale vortices performing the mixing will view the particles as a concentration field. Thus $w_{\mathcal {B}}$ can be approximated as the same in the two classes of current. That is, if $\hat {w}_{\mathcal {B}}$ is a closure for buoyancy velocity in compositional currents (top-hat gives $\hat {w}_{\mathcal {B}} = \hat {w}_e$ ) then $w_{\mathcal {B}} \simeq \hat {w}_{\mathcal {B}}$ . Thus

(5.6) \begin{equation} w_e \simeq \hat {w}_e - \lbrack { 2 - \varsigma _\phi }\rbrack w_s \cos \theta - \frac {E_s}{\Phi }, \end{equation}

we expect particle-driven currents to be modified relative to compositional currents by some detrainment, driven by settling and a reduction of available energy due to erosion. In the limit of no turbulence and a horizontal bed ( $w_{\mathcal {B}} = E_s = \theta = 0$ ) with vertical continuity near the bed ( $\varsigma _\phi =1$ ) this becomes the standard expression for dilute laminar detrainment $w_e \simeq - w_s$ (in this case $w_e = - w_s$ at early times, the approximation is exact despite the top-hat criterion not always being satisfied as time advances, Dorrell & Hogg Reference Dorrell and Hogg2010). For bypass flows where erosion balances deposition ( $E_s = \varsigma _\phi \Phi w_s \cos \theta$ ) we find that $w_e \simeq \hat {w}_e - 2 w_s \cos \theta$ .

The approach (5.6) is comparable to the approach of Toniolo et al. (Reference Toniolo, Parker, Voller and Beaubouef2006) who linearly interpolated between the dynamics of compositional currents and laminar deposition to model their experiments. Bolla Pittaluga et al. (Reference Bolla Pittaluga, Frascati and Falivene2018) effectively use $w_e=\hat {w}_e - w_s \cos \theta$ , which is subsequently adopted by Ma et al. (Reference Ma, Parker, Cartigny, Viparelli, Balachander, Fu and Luchi2024). The energetic interpretation of detrainment presented here gives a clear guide of how it is best implemented.

To understand the impact of detrainment on the turbulence, we first reconstruct the TKE equation from Parker et al. (Reference Parker, Fukushima and Pantin1986) by substituting the top-hat consistency requirements from (4.4) into our TKE equation (2.21), yielding

(5.7) \begin{align} \frac {\partial }{\partial t} ({ h K }) + \sum _\beta \frac {\partial }{\partial x_\beta }(h U_\beta K) & = \sum _\alpha \left ({ U_\alpha u_{\alpha \ast }^2 + {\textstyle \frac {1}{2}} U_\alpha ^2 w_e }\right)\nonumber \\ & \quad - {\textstyle \frac {1}{2}} R g h \cos \theta \left ({ \Phi w_e + \underbrace {\lbrack { 2 - \varsigma _\phi }\rbrack w_s \Phi \cos \theta } + E_s }\right) - h \epsilon _T. \end{align}

The term with an underbrace has a rather strange interpretation. If $\varsigma _\phi \lt 2$ it is reasonable; the TKE is expended holding particles against gravity. However, for $\varsigma _\phi =2$ particle suspension costs no energy, and for $\varsigma _\phi \gt 2$ TKE is generated by particle settling ( $1 \lesssim \varsigma _\phi \lesssim 5$ by figure 2). This term has long been recognised by Parker as greatly troubling (private correspondence). Rewriting using $w_{\mathcal {B}}$ we arrive at

(5.8) \begin{align} \frac {\partial }{\partial t} \left ({ h K }\right) + \sum _\beta \frac {\partial }{\partial x_\beta }(h U_\beta K) & = \sum _\alpha \left ({ U_\alpha u_{\alpha \ast }^2 + \overbrace {{\textstyle \frac {1}{2}} U_\alpha ^2 w_e }}\right)\nonumber \\ &\quad - {\textstyle \frac {1}{2}} R g h \Phi \cos \theta w_{\mathcal {B}} - h \epsilon _T, \end{align}

with $w_e$ computed using (5.4). This reformulation, and reinterpretation as $w_{\mathcal {B}}$ being the specified closure and not $w_e$ , removes the troubling term. Energy is now simply expended generating buoyancy velocity according to some closure for $w_{\mathcal {B}}$ , as would be the case in a compositional current. The expression with the overbrace is the MKE converted to TKE during the process of accelerating entrained fluid. Provided that the entire term is positive (i.e. the net entrainment is positive, $w_e\gt 0$ ) there is no problem with interpretation. However, in a situation where the particle dynamics does not just reduce the entrainment ( $w_e \lt w_{\mathcal {B}}$ ) but actually creates detrainment ( $w_e \lt 0$ ) there is a loss of TKE in the model. This is because the top-hat model does not account for the momentum loss to the ambient fluid in this case. This can be resolved by adding another source term to the momentum equation (2.18) which is $0$ for $w_e \geq 0$ and $w_e U_\alpha$ when $w_e\lt 0$ , and similarly a term to the MKE and total energy equations (2.19) and (2.22) which is $w_e (U_1^2+U_2^2)/2$ when $w_e\lt 0$ . Tracking through the influence of these extra terms, the overbraced term in (5.8) is zero when $w_e\lt 0$ .

This remedy for $w_e\lt 0$ in the TKE equation is specific to the case of the top-hat model, and is not required in the model with shape functions. Instead, the detrainment of fluid adjusts the velocity shape function to be raised relative to the concentration profile, capturing the real physics at play. The principal goal of this manuscript is to include flow shape in the modelling framework, which is important to the flow generally and not just for detrainment. The pseudo-equilibrium case in § 4 is quite straightforward, the consistency requirement (4.4b ) yields

(5.9) \begin{equation} w_e = \frac {w_{\mathcal {B}} - 2 w_s \cos \theta }{\tilde {\sigma }_{1z\phi }}. \end{equation}

Note that, typically, $\tilde {\sigma }_{1z\phi }\lt 1$ (figure 2) and thus for compositional currents $w_e\gt w_{\mathcal {B}}$ . The expression for $w_e$ can be substituted into the pseudo-equilibrium governing equations (4.3) to obtain the balances written in terms of the buoyancy velocity $w_{\mathcal {B}}$ . Perhaps more importantly, we can obtain the consistency requirements for turbulent production and mean-flow buoyancy flux by substitution into (4.4a ) and (4.4c ), which are

(5.10a) \begin{align} h \mathcal {P} &= \sigma _{1\phi } U u_{1\ast }^2 + \frac { 2 \sigma _{11} \sigma _{1\phi } - \sigma _{111} }{\tilde {\sigma }_{1z\phi }} {\textstyle \frac {1}{2}} U^2 \left ({ w_{\mathcal {B}} - 2 w_s \cos \theta }\right) \notag \\ & \hspace {2cm} + \frac { \sigma _{z\phi } \sigma _{1\phi } - \sigma _{1z\phi } }{\tilde {\sigma }_{1z\phi }} {\textstyle \frac {1}{2}} R g h \Phi \cos \theta \left ({ w_{\mathcal {B}} - 2 w_s \cos \theta }\right) - h \epsilon _M, \end{align}

(5.10b) \begin{align} h \mathcal {B}_M &= {\textstyle \frac {1}{2}} R g h \Phi \cos \theta \left ({ \frac {\sigma _{1z\phi }}{\tilde {\sigma }_{1z\phi }} - 1 }\right) \left ({ w_{\mathcal {B}} - 2 w_s \cos \theta }\right). \end{align}

We can also use (5.9) to get an approximation for entrainment in pseudo-equilibrium particulate currents

(5.11) \begin{align} \tilde {\sigma }_{1z\phi } w_e + 2 w_s \cos \theta = w_{\mathcal {B}} &\simeq \hat {w}_{\mathcal {B}} = \tilde {\sigma }_{1z\phi } \hat {w}_e, &&\text {thus}& w_e &\simeq \hat {w}_e - \frac {2 w_s \cos \theta }{\tilde {\sigma }_{1z\phi }}. \end{align}

5.3. A volume-free energetic model of gravity currents

The approach used up to now is to rearrange a consistency requirement for buoyancy flux $h \mathcal {B}_K$ to express entrainment $w_e$ in terms of the buoyancy velocity $w_{\mathcal {B}}$ , which we expect to have the same closures in particle-driven and compositional currents. We then substitute for $w_e$ wherever we see it. To apply this to the general case we would need to rearrange (3.5) and substitute the result into the volume equation (2.16). This is algebraically complex; here we present an equivalent but simpler approach (the equivalence is shown in the supplemental Maple document). The consistency requirement (3.5) is derived from the GPE equation (2.20) (see (3.2)), as

(5.12) \begin{align} \frac {\partial }{\partial t}\left ({ {\textstyle \frac {1}{2}} \sigma _{z\phi } R g h^2 \Phi \cos \theta }\right) & + \sum _\beta \frac {\partial }{\partial x_\beta } \left ({ {\textstyle \frac {1}{2}} \sigma _{\beta z\phi } U_\beta R g h^2 \Phi \cos \theta }\right)\qquad\qquad\qquad\qquad\qquad \nonumber \\ & \qquad\qquad\qquad\qquad = \tilde {S}_G + \frac {1}{2} R g h \Phi \cos \theta w_{\mathcal {B}} + h \mathcal {B}_M. \end{align}

Here, $\tilde {S}_G$ is defined in (2.20b ) and contains the effects of bed variation and particle settling; $h \mathcal {B}_M$ can be found in (3.6) derived as the uplift of concentration by the mean flow, and is a consequence of the incompressibility condition (2.1a ).

We can now offer an alternative to the classical volumetric framework for modelling gravity currents, which was outlined in § 2.3. The governing equations are those for particles (2.17), momentum (2.18), GPE (5.12) and total energy (2.22). This model is closed by imposition of shape functions along with buoyancy velocity $w_{\mathcal {B}}$ ; erosion $E_s$ ; drag $u_{\alpha \ast }$ ; and dissipation $\epsilon _T$ . Consistency requirements for this system of equations can be derived using the volume equation (2.16) to obtain an expression for entrainment $w_e$ , and either the MKE equation (2.19) or the TKE equation (2.21) to obtain a closure for turbulent production $h \mathcal {P}$ . This is equivalent to rearranging the expressions in § 3, see there for interpretation of terms. Substituting the definition of $w_{\mathcal {B}}$ from (5.3) into the consistency requirement for $h \mathcal {B}_K$ (3.5) and rearranging for $w_e$ we obtain

(5.13) \begin{align} w_e = & \frac {\varsigma _h}{\sigma _{z\phi }} \Bigg \lgroup w_{\mathcal {B}} - \lbrack { 2 - \sigma _{z\phi } \varsigma _\phi }\rbrack w_s \cos \theta - \sigma _{z\phi } \frac {E_s}{\Phi }\nonumber \\ & + \sum _\beta \left ({ \frac {\sigma _{z\phi }}{\varsigma _h} + \sigma _{\beta \phi }\sigma _{z\phi } - 2\varsigma _{\beta z \phi } }\right) \frac {\partial }{\partial x_\beta } (h U_\beta) + \sum _\beta \left ({ \sigma _{\beta \phi }\sigma _{z\phi } - \sigma _{\beta z \phi } }\right) \frac {U_\beta h}{\Phi } \frac {\partial \Phi }{\partial x_\beta }\nonumber \\ & + \Bigg\lbrack { - \varsigma _h \frac {\partial }{\partial t} \left ({ \frac {\sigma _{z\phi }}{\varsigma _h}}\right) + \sum _{\beta } \left ({ \sigma _{z\phi } \frac {\partial \sigma _{\beta \phi }}{\partial x_\beta } - \frac {\partial \sigma _{\beta z \phi }}{\partial x_\beta } - \tilde {\sigma }_{D\beta z \phi } }\right) U_\beta }\Bigg\rbrack h \Bigg \rgroup. \end{align}

The two formulations of entrainment, $w_e$ from velocity of ambient fluid fluid into the current and $w_{\mathcal {B}}$ from the turbulent mixing, are thus related as is expected but has not previously been shown. (Note that the appropriate way to approximate a current by a top-hat model when using a closure for buoyancy velocity is to define depth by $\sigma _{z\phi }=1$ , so $h$ is twice the elevation of the centre of excess mass, Arneborg et al. Reference Arneborg, Fiekas, Umlauf and Burchard2007; Anjum, Mcelwaine & Caulfield Reference Anjum, Mcelwaine and Caulfield2013.) The expression for $w_e$ , (5.13), can be substituted into the consistency requirement for $h \mathcal {P}$ (3.4) yielding

(5.14) \begin{align} h \mathcal {P} &= \frac {1}{2} \left({\sum _\beta \frac {\sigma _{\beta \beta }}{\sigma _{z\phi }}U_\beta ^2}\right) \left ({ w_{\mathcal {B}} - \lbrack { 2 - \sigma_{z\phi } \varsigma _\phi }\rbrack w_s \cos \theta - \sigma _{z\phi }\frac {E_s}{\Phi } }\right) + \sum _\beta \sigma _{\beta \beta }U_\beta u_{\beta \ast }^2 - h \epsilon _M\nonumber \\ &\qquad + \sum_\beta (\sigma _{\beta \phi }-\sigma _{\beta \beta }) U_\beta R h\Phi \left ( { g_\beta - g \cos \theta \frac {\partial b}{x_\beta } }\right)\nonumber \\ &\qquad + \frac {1}{2} \sum _{\beta\gamma } \left ({ \sigma _{\gamma \gamma } \left\lbrack { 2 \sigma _{\beta\gamma } + \sigma _{\beta \phi } - 2 \frac {\varsigma _{\beta z \phi}}{\sigma _{z\phi }} }\right\rbrack - \sigma _{\beta \gamma \gamma }}\right) U_\gamma ^2 U_\beta \frac {\partial h}{\partial x_\beta}\nonumber \\ & \qquad + \frac {1}{2} \sum _{\beta } \left ({ \sigma_{\beta \beta } \left\lbrack { 4\sigma _{\beta \beta } + \sigma _{\beta\phi } - 2 \frac {\varsigma _{\beta z \phi }}{\sigma _{z\phi }}}\right\rbrack - 3\sigma _{\beta \beta \beta } }\right) h U_\beta ^2 \frac{\partial U_\beta }{\partial x_\beta }\nonumber \\ &\qquad + \frac{1}{2} \sum _{\beta \neq \gamma } \Bigg\lgroup \left ({ \sigma_{\gamma \gamma } \left\lbrack { 2\sigma _{\beta \gamma } + \sigma _{\beta\phi } - 2 \frac {\varsigma _{\beta z \phi }}{\sigma _{z\phi }}}\right\rbrack - \sigma _{\beta \gamma \gamma } }\right) h U_\gamma ^2\frac {\partial U_\beta }{\partial x_\beta } \nonumber\\ &\qquad\quad\qquad\qquad + 2 \left ({ \sigma _{\gamma \gamma } \sigma _{\beta \gamma } - \sigma _{\beta \gamma \gamma } }\right) h U_\beta U_\gamma \frac {\partial U_\gamma }{\partial x_\beta } \Bigg\rgroup \nonumber \\ &\qquad + \frac {1}{2} \sum _{\beta \gamma } \sigma _{\gamma \gamma } \left ({ \sigma _{\beta \phi } - \frac {\sigma _{\beta z\phi }}{\sigma _{z\phi }} }\right) \frac {h U_\gamma ^2 U_\beta }{\Phi } \frac {\partial \Phi }{\partial x_\beta }\nonumber \\ & \qquad + \sum _{\beta } \left ({ \sigma _{\beta \beta } \sigma _{z\phi } - \varsigma _{\beta z\phi } }\right) U_\beta R g h \Phi \cos \theta \frac {\partial h}{\partial x_\beta }\nonumber \\ &\qquad + \frac {1}{2} \sum _{\beta } \left ({ \sigma _{\beta \beta } \sigma _{z\phi } - \tilde {\sigma }_{\beta z\phi } }\right) U_\beta R g h^2 \cos \theta \frac {\partial \Phi }{\partial x_\beta }\nonumber \\ &\qquad + \frac {1}{2} \sum _\beta \Bigg\lgroup - \frac {1}{\sigma _{z\phi }} \frac {\partial }{\partial t} \left ({ \sigma _{z\phi } \sigma _{\beta \beta } }\right) \nonumber\\ & \qquad+ \sum _\gamma \left ({ \sigma _{\beta \beta } \frac {\partial }{\partial x_\gamma } \lbrack { 2\sigma _{\beta \gamma } + \sigma _{\gamma \phi } }\rbrack - \frac {\partial \sigma _{\beta \beta \gamma }}{\partial x_\gamma } - \frac {\sigma _{\beta \beta }}{{\sigma _{z\phi }}} \left\lbrack { \frac {\partial \sigma _{\gamma z \phi }}{\partial x_\gamma } + \tilde {\sigma }_{D\gamma z \phi } }\right\rbrack }\right) U_\gamma \Bigg\rgroup h U_\beta ^2\nonumber \\ &\qquad\qquad\quad\qquad + \frac {1}{2} \sum _\beta \left ({ \sigma _{\beta \beta } \frac {\partial \sigma _{z\phi }}{\partial x_\beta } - \frac {\partial \tilde {\sigma }_{\beta z \phi }}{\partial x_\beta } + \tilde {\sigma }_{D\beta z \phi } }\right) U_\beta R g h^2 \Phi \cos \theta. \end{align}

The consistency requirement for mean-flow uplift of particles (3.6) is not affected, and is used to close (5.12). The accuracy of these expressions is verified in figure 9 in the same way as the consistency expressions for the classical volumetric model were verified in figure 6, § 4. The coefficients in (5.13) and (5.14) calculated for real currents are included in figure 3.

Figure 9.

Equivalent plots to those in figure 6 but for the case of the volume-free energetic model (§ 5.3).

This new formulation does not contain an equation for volume, and eliminates the need for an arbitrary imposition of a interfacial surface. This model is closed directly by a measure of the turbulence, the buoyancy velocity (5.3). The only time when the interfacial surface $z=\tilde {h}$ needs to be specified is when using the consistency relationship for entrainment (5.13), which depends on $\varsigma _h = \tilde {h}/h$ . This model is volume free in that it does not explicitly include a volume equation, and instead the evolution is governed by the energetics. The way that incompressibility (2.1a ) enters the formulation is through the derivation of mean-flow uplift of concentration (3.3), and in deducing the implied entrainment rate (5.13).

A key observation is that, even for compositional currents, the entrainment $w_e$ is no longer equal to, or even proportional to, the buoyancy velocity $w_{\mathcal {B}}$ (5.13). Consequently, in the general case, we require new closures for $w_{\mathcal {B}}$ . Early results on this have been documented by Wells et al. (Reference Wells, Cenedese and Caulfield2010), but further work is required to parametrise the stirring and mixing of the density field. See Caulfield (Reference Caulfield2021) for a review of this research area.