Skip to content

Latest commit

 

History

History
324 lines (311 loc) · 13.4 KB

chapter_00_1_swe_decomp.md

File metadata and controls

324 lines (311 loc) · 13.4 KB

Shallow water equations

Governing Equations

The shallow water equations (SWE) have found utility as an academic tool to study and explain fundamental geophysical phenomena [see chapter 3, @vallis_atmospheric_2017] and also as the "dynamical core" of some numerical weather prediction models [for eg., a flux vector splitting scheme based formulation of the SWE [@lin_explicit_1997] was used in NOAA's GFS model until recently]1. The governing equations for a single layer shallow layer of fluid are: \begin{align} \label{eq:dtu0} \partial_t \mathbf{u} & = - (\mathbf{u}.\nabla) \mathbf{u} - c^2 \nabla h - f\mathbf{e_z} \times \mathbf{u}, \ \label{eq:dth} \partial_t h & = - \nabla \cdot (h \mathbf{u}), \end{align} where $\bf u$ is two-dimensional horizontal velocity vector, $c$ is the phase speed of gravity waves, $f$ is the system rotation, $h = (1 + \eta)$ is the non-dimensional scalar height field, and $\eta$ is surface displacement. In this inviscid formulation, the system of equations [@Eq:dtu0;@eq:dth] conserves the sum of kinetic and potential energy, defined as $E = E_K + E_P$, where $E_K= h|\bf u|^2$ and $E_P = c^2h^2 / 2$. Expanding the expression for $E_P$ in $\eta$ we get, $$ E_P = c^2(1/2 + \eta + \eta^2/2). $$ The first term is the constant background potential energy, while the two remaining terms are the potential energy associates with surface displacements. The second term, which is linear in $\eta$, is conserved by @eq:dtu0. The third term, $E_A = c^2n^2/2$, may be called available potential energy (APE), a concept introduce by @Lorenz:1955. The SWE also conserves the sum $E_K + E_A$. As a matter of fact, wave motions in SWE is characterized by equipartition between kinetic and available potential energy.

The inviscid SWE has another materially conserved invariant which is potential vorticity, $Q = (\zeta + f)/h$. Conservation of potential vorticity is crucial to explain several geophysical fluid motions [@vallis_atmospheric_2017]. Since $Q$ is a materially conserved quantity, so are higher powers of $Q$, including the quadratic potential enstrophy. From a turbulence perspective [@Warn1986;@LindborgMohanan2017], the non-quadratic nature of potential enstrophy restricts our interpretation of a potential enstrophy cascade. However, in the limit of strong rotation or QG limit, we can study its linearized form, $q = \zeta - f\eta$, which is approximately conserved.

In comparison with the QG equation which conserves quasi-geostrophic potential vorticity, the SWE permits both quasi-geostrophic (vortices) and ageostrophic (gravity waves) modes. On the one hand, the QG equation is similar to the incompressible 2D Navier-Stokes equations, and on the other hand, the SWE system has some similarities with the compressible 2D Navier-Stokes equations -- due to fact that [@eq:dth] has the same mathematical structure as the mass-conservation equation for advection of density $\rho$, and because shock waves are produced in the SWE [@Baines1998;@augier_shallow_2019;@vallis_atmospheric_2017].

[@Eq:dtu0] may also be written in the rotational form as: $$\label{eq:dtu} \partial_t \mathbf{u} = - \nabla |\mathbf{u}|^2/2 - c^2 \nabla h - \zeta \mathbf{\hat{e}}_z \times \mathbf{u}, $$ where, [$\zeta$]{acronym-label="zeta" acronym-form="singular+short"} represents absolute vorticity, i.e. the sum of vorticity and system rotation. Furthermore, some auxiliary equations are derived from the SWE. [@Eq:dtu0] and [@eq:dth] can be combined to form an equation for the total mass flux, [J]{acronym-label="J" acronym-form="singular+short"} $= h\mathbf{u}$: $$\label{eq:dtJ} \partial_t \mathbf{J} = -(\mathbf{u}.\nabla)\mathbf{J} - \nabla(c^2h^2)/2 - \zeta \mathbf{\hat{e}}_z \times \mathbf{J} - (\nabla. \mathbf{J})\mathbf{u}. $$ For a divergence free flow, a Poisson equation for h can be formulated. Taking the divergence of [@eq:dtu0], yields the Poisson equation: $$ \nabla^2 h = \frac{1}{c^2} \left[ \nabla.(\zeta \mathbf{\hat{e}}z \times \mathbf u ) - \nabla^2 \frac{|u|^2}{2} \right]. $${#eq:poisson} The spectral counterpart for [@eq:poisson] in tensor notation is: $$ -\kappa^2 \hat{h} = \frac{1}{c^2} \left[ ik_i (\widehat{\epsilon{ijk} \zeta_j u_k}) + \kappa^2 \frac{\widehat{u_i u_i}}{2} \right], $${#eq:poisson_fft} where the $\widehat{\text{ }}$ denotes the Fourier transform. To study interactions within SW turbulence it is useful to decompose the flow field. We consider two decompositions: the Helmholtz and the normal-mode decomposition.

Helmholtz Decomposition

The fundamental theorem of vector calculus [@baird_helmholtz_2012] states that any well-behaved vector field can be decomposed into the sum of an irrotational vector field and a rotational or non-divergent vector field. This allows us to express the velocity field as: \begin{align} \label{eq:helm_u} \mathbf{u} & = -\nabla \times (\mathbf{\hat{e}}_z \Psi) + \nabla \chi. \end{align}

For the sake of clarity, we shall denote the rotational and divergent parts of the velocity with the suffix r and d respectively, $$ \mathbf{u}^r = -\nabla \times \mathbf{\hat{e}}_z\Psi, \quad \mathbf{u}^d = \nabla \chi. $$

To find the projection operator for the divergent part, take the divergence of [@eq:helm_u], giving $\nabla \cdot \mathbf{u} = \nabla^2 {\bf \chi}$. This equation transforms into spectral space as $i{\bf k} \cdot \hat{\bf u} = -\kappa^2 \hat{\chi}$, implying: \begin{align*} {\hat{\bf u}}^d = & i {\bf k}\hat{\chi} = \frac{{\bf k}\cdot\hat{\bf u}}{\kappa^2} {\bf k}, \ {\hat{\bf u}}^r = & \hat{\bf u} - {\hat{\bf u}}^d, \end{align*} where, $\kappa = |\mathbf{k}|$, is the magnitude of the wavenumber vector. To obtain a similar decomposition, $h = h^r + h^d$, for the fluid depth one can use the Poisson equation [@eq:poisson_fft]. Since the Poisson equation requires a divergence free flow, the LHS of [@eq:poisson_fft] would correspond to the rotational part of the flow in the transformed plane, i.e. $\hat{h}^r$. Henceforth, the divergent part, $\hat{h}^d$ can be obtained by subtracting $\hat{h}^r$ from $\hat{h}$. While the Helmholtz decomposition is simple to compute and provide insightful results, it may be more revealing to apply a normal-mode decomposition -- especially in the case with system rotation, wherein potential vorticity is conserved and not vorticity.

Normal mode decomposition

@bartello_geostrophic_1995 demonstrated how the Boussinesq equations can be analysed using a normal-mode decomposition. Here, we follow the same procedure for the SWE. In order to isolate the geostrophic modes from the oscillating fast modes, we linearize the SWE followed by a normal mode or eigenmode decomposition. The linearized [@eq:dtu0] and [@eq:dth] can be written as \begin{align} \label{eq:dtu_l} \partial_t \mathbf u = & - c^2 \nabla \eta - f\mathbf{e_z} \times \mathbf u, \ \label{eq:dth_l} \partial_t \eta = & - \nabla \cdot \mathbf u. \end{align} Taking the curl and divergence of @eq:dtu_l gives the following evolution equations: \begin{align} \partial_t \zeta = & - f \delta \label{eq:dtcurl_l}, \ \partial_t \delta = & f \zeta - c^2 \nabla^2 \eta \label{eq:dtdiv_l}, \ \partial_t \eta = & - \delta \label{eq:dteta_l}, \end{align} where $\zeta$ is the relative vorticity and $\delta$ is the divergence. Representing the dependent flow quantities in terms of Fourier modes: \begin{align*} \begin{pmatrix} u \ v \ \eta \ \zeta \ \delta \end{pmatrix} (\mathbf{r},t) = \int \begin{pmatrix} \hat{u} \ \hat{v} \ \hat{\eta} \ \hat{\zeta} \ \hat{\delta} \end{pmatrix} e^{i(\mathbf k \cdot \mathbf{r} - \omega t)} \mathbf{dk} d\omega, \end{align*} allows us to rewrite the system of equations [@eq:dtcurl_l;@eq:dtdiv_l;@eq:dteta_l] as an eigenvalue problem: \begin{align*} i\omega \begin{Bmatrix} \hat{\zeta} \ \hat{\delta} \c\kappa\hat{\eta} \end{Bmatrix} = i \begin{bmatrix} 0 & if & 0 \ -if & 0 & -ic\kappa \ 0 & ic\kappa & 0 \end{bmatrix} \begin{Bmatrix} \hat{\zeta} \ \hat{\delta} \c\kappa\hat{\eta} \end{Bmatrix}. \end{align*} Let us define $A$ as the Hermitian matrix operating on the vector $\mathbf{W} = {\hat{\zeta}, \hat{\delta} ,c\kappa \hat{\eta} }^T$ which yields the familiar dispersion relation for the slow geostrophic mode and fast Poincaré wave modes: $$ \omega^{(0)} = 0,\quad \omega^{(\pm)}=\pm \sigma, $$ where, $\sigma = \sqrt{f^2 + c^2\kappa^2 }$. These are the eigenvalues of the matrix operator A. Since A is Hermitian, the corresponding eigenvectors are orthogonal and these are normalized as follows \begin{align*} \mathbf X^{(0)}_n = \frac{1}{\sigma} \begin{Bmatrix} -c\kappa \ 0 \ f \end{Bmatrix}, \quad \mathbf X^{(\pm)}_n = \frac{1}{\sqrt{2} \sigma} \begin{Bmatrix} f \ \mp i\sigma \ c\kappa \end{Bmatrix}. \end{align*} Let $X_n$ be the eigenvector matrix, which follows the property $X_n X_n^=I$, where $^$ represents the Hermitian transpose. It can be applied to diagonalize the system of equations as follows: \begin{align*} \partial_t \mathbf{W} = & [A] \mathbf{W}, \ \partial_t (X_n^* \mathbf{W}) = & X_n^[A]X_n X_n^\mathbf{W} = [\Lambda] (X_n^\mathbf{W}), \end{align} where $\Lambda$ is the diagonal eigenvalue matrix. Thus, the alternate diagonalized system of equations for the normal modes are given by: \begin{align*} \partial_t \mathbf{N} = & \begin{bmatrix} 0 & 0 & 0 \ 0 & -i\sigma & 0 \ 0 & 0 & i\sigma \end{bmatrix} \mathbf{N}, \end{align*} where \begin{align} \label{eq:nmode} \mathbf{N} = X_n^* \mathbf{W} = & \frac{1}{\sqrt{2}\sigma} \begin{Bmatrix} -\sqrt{2}c\kappa(\hat{\zeta} -f \hat{\eta}) \ f\hat{\zeta} + c^2\kappa^2\hat{\eta} - i\sigma\hat{\delta} \ f\hat{\zeta} + c^2\kappa^2\hat{\eta} + i\sigma\hat{\delta} \end{Bmatrix}. \end{align} The first mode represents linearized potential vorticity. The remaining two are the linearized ageostrophic or wave modes. In the absence of system rotation, i.e. $f=0$, the normal modes reduce to, $$ \mathbf{N} = \begin{Bmatrix} \hat \zeta \ \frac{1}{\sqrt{2}} (c\kappa \hat{\eta} - i\hat{\delta}) \ \frac{1}{\sqrt{2}} (c\kappa \hat{\eta} + i\hat{\delta}) \end{Bmatrix}. $${#eq:nmode_f0}

Normal mode inversion to primitive variables

To study the spectral energy budget, normal modes have to be transformed back to the primitive variable using another matrix operation, say $Q$. The normal modes can be represented by $\mathbf{N} = X_n^* \mathbf{W}$. Now, we form a new vector $\mathbf{B} = \mathbf{N}/\kappa$ which has the same dimension as velocity. Thus, \begin{align} \label{eq:bvec} \mathbf{B} = & \frac{1}{\sqrt{2}\sigma} \begin{Bmatrix} \sqrt{2} c\left(- \kappa^{2} \hat \psi + f\hat\eta \right) \ \kappa \left(c^{2} \eta + f \psi + i \phi \sigma\right) \ \kappa \left(c^{2} \eta + f \psi - i \phi \sigma\right) \end{Bmatrix}. \end{align} The vector can also be related to the primitive variable vector, $\mathbf{U} = {\hat{u},\hat{v},c\hat{\eta}}^T$ using transformation matrices as follows: \begin{align*} \mathbf{B} = & \frac{1}{\kappa}X_n^* \mathbf{W}, \ = & \frac{1}{\kappa}X_n^* [P] \mathbf{U}, \end{align*} where, $$ P= \begin{bmatrix}- i k_{y} & i k_{x} & 0 \ i k_{x} & i k_{y} & 0 \0 & 0 & \kappa \end{bmatrix}. $$ It is also straightforward to show that the magnitude of the normal modes equals the total energy as: $$ \mathbf{B}^\mathbf{B} = \frac{1}{\kappa^2} \mathbf{U}^ [P]^X_n X_n^ [P] \mathbf{U} = \mathbf{U}^\mathbf{U}, $$ implying, $$ (E_K+E_P)=\frac{1}{2}\sum_i B^{(i)}{B}^{(i)}= \frac{1}{2}\kappa^{2}(\psi {\psi}^* +\kappa^{2} \phi {\phi}^* + c^{2} \eta {\eta}^) = \frac{1}{2}(uu^ + vv^* + c^2\eta\eta^). $$ The primitive variables can be represented in terms of normal modes as, $$\mathbf{U} = [Q]\mathbf{B}$${#eq:uqmatb} where, the inversion matrix is \begin{align} \label{eq:qmat} Q = & \kappa [P]^{-1}{X_n}^{^{-1}} \ = & \frac{1}{\sqrt{2}\sigma\kappa} \begin{bmatrix} -i\sqrt{2}c\kappa k_{y} & \left(i f k_{y} + k_{x} \sigma\right) & \left(i f k_{y} - k_{x} \sigma\right) \ i \sqrt{2}c\kappa k_{x} & \left(- i f k_{x} + k_{y} \sigma\right) & - \left(i f k_{x} + k_{y} \sigma\right) \ \sqrt{2}\kappa f & c\kappa^{2} & c\kappa^{2} \end{bmatrix}. \end{align} In tensor notation, \begin{align} \label{eq:decomp_tensor_u} \hat{u}l = & \epsilon{lm3} ik_m\left[ -\frac{c}{\sigma} B^{(0)} + \frac{f}{\sqrt{2}\sigma\kappa} (B^{(+)} + B^{(-)}) \right] + k_l \frac{1}{\sqrt{2}\kappa} (B^{(+)} - B^{(-)}), \ \label{eq:decomp_tensor_eta} c\hat{\eta} = & \frac{f}{\sigma} B^{(0)} + \frac{c\kappa}{\sqrt{2}\sigma} (B^{(+)} + B^{(-)}). \end{align}

Footnotes

  1. See https://www.gfdl.noaa.gov/fv3/