Atmospheric Dynamics: Isaac M. Held

ATMOSPHERIC DYNAMICS
Isaac M. Held
Contents
1 Preliminaries 3
1.1 Brief Review of Vector Calculus . . . . . . . . . . . . . . . . 3
1.2 Eulerian and Lagrangian Perspectives . . . . . . . . . . . . . 5
1.3 Line segments, Volumes, and Conservation of Mass . . . . . 7
1.4 Diffusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.5 Thermodynamics of an ideal gas . . . . . . . . . . . . . . . . 9
1.6 Equation of Motion . . . . . . . . . . . . . . . . . . . . . . . 12
1.7 Energetics . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.8 Spherical Coordinates . . . . . . . . . . . . . . . . . . . . . . 16
1.9 Vorticity in Homogeneous Incompressible Flow . . . . . . . . 20
1.10 Planar vs. Spherical Two-Dimensional Flow . . . . . . . . . 23
2 Momentum fluxes in a barotropic model 29

2.1 Vorticity Mixing . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.2 Eddy Vorticity and Momentum Fluxes . . . . . . . . . . . . 32
2.3 Pseudomomentum . . . . . . . . . . . . . . . . . . . . . . . 34
2.4 Eddy Energetics . . . . . . . . . . . . . . . . . . . . . . . . . 41
2.5 Food for thought . . . . . . . . . . . . . . . . . . . . . . . . 43
1
1 Preliminaries
1.1 Brief Review of Vector Calculus

Consider two points, r and r + δ`, separated by an infinitesimal dis-
tance, where the components of the vector δ` in Cartesian coordinates
are (δx, δy, δz): δ` = x̂δx + ŷδy + ẑδz. Consider also a scalar field ξ, which
might for example, be the atmospheric temperature or density. Setting
δξ = ξ(r + δ`) − ξ(r), a linear approximation to this difference yields
∂ξ ∂ξ ∂ξ
δξ = δx + δy + δz . (1.1)
∂x ∂y ∂z
The gradient ∇ξ is defined so that
δξ = δ` · ∇ξ, (1.2)
that is,
∂ξ ∂ξ ∂ξ
∇ξ = x̂ + ŷ + ẑ . (1.3)
∂x ∂y ∂z
For any two points A and B and a path P connecting these points:
Z
ξ(B) − ξ(A) = ∇ξ · δ` . (1.4)
P
The line integral over any closed path of the gradient of a scalar vanishes,
as can be seen by setting B = A in this expression.
The divergence of the vector field F = (Fx , Fy , Fz ) in Cartesian coordi-
nates is
∂Fx ∂Fy ∂Fz
∇·F = + + . (1.5)
∂x ∂y ∂z
Gauss’s theorem relates the integral of the divergence of F over a volume
to the outward flux of F :
Z Z Z Z Z
∇ · F dV = F · n̂dA . (1.6)
V S
3
1. Preliminaries
Here n̂ is a unit vector perpendicular to the surface S bounding the volume

V and directed from the interior to the exterior of the volume.
The curl of a vector field A = (Ax , Ay , Az ), ω = ∇ × A, is defined in
Cartesian components, ω = (ωx , ωy , ωz ), as
∂Az ∂Ay
ωx = − ,
∂y ∂z
∂Ax ∂Az
ωy = − , (1.7)
∂z ∂x
∂Ay ∂Ax
ωz = − .
∂x ∂y
The curl of any gradient vanishes: ∇ × ∇ξ = 0, as does the divergence of

a curl: ∇ · (∇ × A). Stokes’ theorem relates the curl to the line integral
around a closed loop L:
Z Z Z
A · δ` = (∇ × A) · n̂dS . (1.8)
L S
Here S is any surface whose boundary is the loop L and n̂ is an outwardly

directed unit normal on that surface, with ”outward” defined consistently
with the orientation of the line integral around the loop, using the right-
hand rule.
Vector identities related to the divergence and curl can be generated
more easily by writing the curl in tensor notation.
X
(∇ × A)i = ijk ∂j Ak = ijk ∂j Ak (1.9)
j,k
Here ∂x ≡ ∂/∂x, each of the indices runs over (x, y, z), and the last ex-
pression uses the convention in which repeated indices are automatically
summed over, The tensor ijk is defined as equal to 0 if any of the two
indices are equal, 1 if (i, j, k) is an even permutation of (1, 2, 3) and −1 for
odd permutations (i.e, 123 = 231 = 312 = 1; 132 = 213 = 321 = −1). The
key identity, which can be confirmed by direct computation, is
ijk imn = δjm δkn − δjn δkm (1.10)
where δij equals 1 if the two indices are equal and 0 otherwise. One can
then show, for example, that
1
(∇ × A) × A = (A · ∇)A − ∇|A|2 (1.11)
2
4
1.2. Eulerian and Lagrangian Perspectives
by noting that
(∇ × A) × A)i = ijk j`m (∂` Am )Ak (1.12)
= −jik j`m (∂` Am )Ak
= −(δi` δkm − δim δk` )(∂` Am )Ak
= −Ak ∂i Ak + Ak ∂k Ai
Another useful identity that can be obtained with a similar manipulation
is
∇ × (A × B) = (A · ∇)B − (B · ∇)A + A∇ · B − B∇ · A . (1.13)
1.2 Eulerian and Lagrangian Perspectives

Let the vector field v(r, t) = (u, v, w) be the time-evolving velocity of a
fluid: in the time δt each infinitesimal parcel of fluid is displaced by vδt.
The time derivative of a scalar ξ(r, t) moving with the flow is
Dξ ξ(r + vδt, t + δt) − ξ(r, t)) ∂ξ
= lim = + v · ∇ξ (1.14)
Dt δt→0 δt ∂t
Expressing this material derivative in terms of the cartesian components of
the flow,
Dξ ∂ξ ∂ξ ∂ξ ∂ξ
= +u +v +w (1.15)
Dt ∂t ∂x ∂y ∂z
We can replace the scalar ξ in this expression by a vector, for example
the velocity itself, so that the acceleration, the rate of change of the velocity
following a particle of fluid, is
Dv ∂v ∂v ∂v ∂v ∂v
= + (v · ∇)v = +u +v +w (1.16)
Dt ∂t ∂t ∂x ∂y ∂z
In Cartesian coordinates, the unit vectors are independent of space and
time, so that
Dv Du Dv Dw
= x̂ + ŷ + ẑ . (1.17)
Dt Dt Dt Dt
The expression for, say, the x-component of the acceleration then takes the
same form as does the material derivative of a scalar, simply replacing ξ by
u in Eq. xx.
Rather than describe a fluid flow by thinking of the velocity field, and
other fields of interest, as a function of space and time (the Eulerian de-
scription), one can instead, at least formally, label individual fluid particles
5
1. Preliminaries
and follow how fields evolve following these particles (the Lagrangian de-
scription). Particles can be labelled by their location at some time t = 0.
The trajectory of a particle is obtained by solving
∂x(t, a)
= v(x, t); x(0, a) = a (1.18)
∂t
The acceleration of a fluid particle is then simply the partial derivative of
the velocity, holding the particle label fixed, but other interesting quantities,
such as the pressure gradient force, become harder to express. We will focus
entirely on Eulerian equations of motion, but a Lagrangian perspective,
in which one thinks about particle trajectories and how fields of interest
change along trajectories, will still be very important. Eq () has the form
of a general three-dimensional nonlinear dynamical system,
ẋ = u(x, y, z, t)
ẏ = v(x, y, z, t)
ż = w(x, y, z, t)
so it should not be surprising that trajectories can be exceedingly complex,

even chaotic, in what appear to be relatively simple velocity fields.
Particle trajectories can be difficult to visualize even in flows that are
so simple that the trajectories are not chaotic. Consider, for example, a
transverse wave in the presence of a uniform background flow (u, v, w) =
(U, v(x), 0) with v(x) = V sin(k(x − ct)), with U, V constants. As particles
in this flow move steadily towards positive x, their y-coordinate, η, varies
as
V
η= sin(k(x − ct)) (1.19)
k(U − c)
When the wave’s phase speed c approaches the speed of the background
flow U , the particle displacements due to the wave grow; the smaller U − c,
the more time particles spend in one phase of the wave before entering the
opposite phase.
This flow is non-divergent in the (x, y) plane, with ∂x u+∂y v = 0, and can
be described by a streamfunction ψ defined so that (u, v) = (−∂y ψ, ∂x ψ).
Here ψ = −U y + (V /k) sin(k(x − ct)). In the special case that the stream-
function is independent of time, particles will flow along streamlines, the
lines of constant ψ. A common trap that is difficult to avoid is to think that
streamlines in a time-dependent flow resemble trajectories. In the example
just described, the meanders in the streamlines due to the wave have am-
plitude V /kU , independent of the phase speed of the wave. A simple way
of obtaining the trajectories is to move to a reference frame that is moving
6
1.3. Line segments, Volumes, and Conservation of Mass
with the wave, so that the flow and its streamfunction are then independent
of time.
A more realistic picture of what happens to trajectories when the back-
ground flow speed equals a wave phase speed is obtained by adding shear
to the background flow, replacing the constant U with u = U (y) = Λy.
The resulting trajectories are illustrated in Fig. xx, as the reader may wish
to confirm by computing the streamfunction in the frame moving with the
wave. Near the location where U (y) = c there are regions where particles
are trapped by the wave and, on average, move with the wave phase speed.
These regions have a width proportional to (V /kΛ))1/2 . On either side of
these regions, particles are carried by the background flow. These critical
layer or ”cat’s eye” structures will play a role in our discussion of wave-mean
flow interaction in Ch. xx.
Another class of flows that results in counterintuitive trajectories is il-
lustrated by setting (u, v, w) = (A sin(ωt), 0, A cos(ωt)). If A is a constant,
the particle orbits are circles of radius A/ω in the x − z plane, traversed
counterclockwise. Now suppose that A is a function of z, but slowly varying
in the sense that A varies by a small fraction of itself over the distance A/ω
(that is, ∂A/∂z << ω). The result is that particles drift toward negative x,
if A increases with z, with speed (∂A/∂z)A/ω (Fig. xx). The time average
of the Eulerian flow is identically zero, but the Lagrangian mean flow of
the fluid particles is non-zero, an effect referred to as Stokes’ drift (owing
to Stokes’ classic analysis of this phenomenon in water waves). One can
also create the opposite situation, in which fluid particles do not move sys-
tematically, yet the Eulerian flow has non-zero time mean – one can simply
add a constant flow to balance the Stokes’ drift. This latter situation may
seem contrived, but it plays an important role in discussions of stratospheric
dynamics (Ch. xx)
1.3 Line segments, Volumes, and Conservation of Mass

Consider the evolution of an infinitesimal line segment δ` moving with the
flow v. The orientation and length of the vector δ` change due to the
difference in the velocity between the endpoints of the line segment:
Dδ`
= (δ` · ∇)v (1.20)
Dt
In Cartesian coordinates, the rate of change of the x-component of δ` is
Dδx ∂u ∂u ∂u
= δx + δy + δz (1.21)
Dt ∂x ∂y ∂z
7
1. Preliminaries
The first term stretches the segment in the x-direction, while the second
and third tilt the segment from the y- and z-directions into the x-direction.
Now follow an infinitesimal material parcel moving with the flow v and
examine its volume δV as a function of time. The divergence of v is the
fractional rate of change of this volume:
1 DδV
=∇·v (1.22)
δV Dt
Once can see this by choosing the small volume to be a cube at some initial
time. For the line segment oriented along an edge parallel to the x-axis, we
have, setting δy = δz = 0 in (??),
1 Dδx ∂u
= (1.23)
δx Dt ∂x
and
1 DδV 1 Dδx 1 Dδy 1 Dδz
= + + =∇·v (1.24)
δV Dt δx Dt δy Dt δz Dt
The mass of an air parcel δM is the density ρ multiplied by the volume
of the parcel δV . If the mass is conserved following the flow, then
DδM DρδV 1 Dρ 1 DδV
= =0→ =− = −∇ · v (1.25)
Dt Dt ρ Dt δV Dt
Conservation of mass can be expressed in this advective form
Dρ ∂ρ
= + v · ∇ρ = −ρ∇ · v (1.26)
Dt ∂t
or the flux form
∂ρ
= −∇ · (ρv) (1.27)
∂t
Given any scalar ξ that is conserved following the flow, except for the
source/sink per unit mass S, we have in advective form
Dξ
=S (1.28)
Dt
or flux form
∂ρξ
+ ∇ · (ρvξ) = ρS (1.29)
∂t
The advective form of these expressions are naturally most useful when
considering rates of change following fluid particles, while the flux forms
are most useful when considering the rate of change at a fixed point in
space.
8
1.4. Diffusion
For example, the diffusive flux of water vapor in the atmosphere is di-
rected down the gradient of the water vapor mixing ratio, q, the ratio of the
density of the water vapor to the density of the remaining (dry) air com-
ponent. It is this mixing ratio that is conserved following a parcel in the
absence of diffusion – the density of water vapor in the parcel will change
if the volume of the paercel changes in the absence of diffusion, but the
density of the dry air will change proportionally. If there are sources or
sinks of water vapor due to evaporation or condensation totalling S, then
the total mass of air is not conserved, but the mass of dry air is conserved.
Setting the density of dry air equal to ρd , and the water vapor mixing ratio
equal to q, we have Dq/Dt = S and
∂ρd q
+ ∇ · (ρd vq) = ρd S (1.30)
∂t
(The maximum value of the water vapor mixing ratio in the atmosphere is
abour 2%, so the difference between ρ and ρd is often ignorable.)
1.4 Diffusion
Up to this point we have ignored molecular diffusion, which is always
present, although it may be negligible for some contexts. Kinetic theory of
gases leads to Fick’s law, also empirically valid for liquids, that the diffusive
flux is proportional to, and directed down the gradient of the tracer mixing
ratio in question. Diffusion tends to equalize mixing ratios. The diffusive
flux F = −κ∇q must be added to the advective flux:
∂ρξ
+ ∇ · (ρvξ) = ρS − ∇ · F (1.31)
∂t
1.5 Thermodynamics of an ideal gas

The equation of state for an ideal gas can be written as
p = ρRT (1.32)
where T is the absolute temperature, and R is the gas constant. Alterna-

tively, p = nkT where n is the number of molecules per unit volume, and k
is Boltzmann’s constant. (Using this notation, R is inversely proportional
to the mean molecular weight.) For dry air, R ≈ 287(m2 /s2 )/K.
9
1. Preliminaries
On large scales the atmosphere is hydrostatic, with a balance in the

vertical equation of motion between gravity and the pressure gradient:
∂p p RT
= −ρg = − ; H ≡ , (1.33)
∂z H g
with pressure, and density, decreasing exponentially in the vertical with
scale height H for an isothermal atmosphere.
All motions that we are concerned with are adiabatic in the sense
that the evolution is sufficiently slow that molecular collisions maintain lo-
cal thermodynamic equilibrium within each (macroscopically) infinitesimal
fluid parcel. The first law of thermodynamics can be written as
De Dα
= −p + Q = −pα∇ · v + Q (1.34)
Dt Dt
where e is the internal energy per unit mass, α ≡ 1/ρ and Q is the external
heating. Alternatively, using the definition of the enthalpy per unit mass,
h = e + pα,
Dh Dp
=α +Q (1.35)
Dt Dt
In general, e and h are functions of pressure and temperature. The specific
heats at constant volume (or density) cv and constant pressure cp are defined
by
de dh
cv ≡ ; cp ≡ (1.36)
dT v dT p

so that cv DT /Dt = Q at constant volume and cp DT /Dt = Q at constant

pressure. For an ideal gas, h = e + RT and cp = cv + R.
Each degree of freedom excited at the temperatures of interest con-
tributes kT /2 per molecule to the heat capacity at constant volume, or
RT /2 per unit volume. The atmosphere consists mostly of diatomic molecules
with 5 relevant degrees of freedom (3 translational and 2 rotational, with
the vibrational degree of freedom unexcited by collisions at atmospheric
temperatures), so that cv ≈ 5R/2 and cp ≈ 7R/2. These expressions
turn out to be quite accurate for a dry atmosphere, resulting in a value
of cp = 103 m2 /(s2 K). Common ratios of the heat capacities are given their
own symbols:
cp R
γ≡ ≈ 7/5; κ ≡ ≈ 2/7 . (1.37)
cv cp
The mean surface pressure on Earth is roughly 105 P a. Dividing by the
gravitational acceleration g yields a mass per unit area for the atmosphere
of 104 Kg/m2 . The heat capacity (at constant pressure) of an atmospheric
10
1.5. Thermodynamics of an ideal gas
column is then 107 (J/m2 )/K, which is roughly the heat capacity of 2.4
meters of water.
The enthalpy equation is often written in the form
DT κT ω Q
= + (1.38)
Dt p cp
where ω is the traditional notation for Dp/Dt. (Ambiguities arising from

also using the symbol ω for vorticity, and frequency, are generally resolvable
from the context). Except on the smallest scales, pressure is a monotonically
decreasing function of z, and ω is referred to as the pressure-coordinate
vertical velocity.
The entropy per unit mass of an ideal gas is defined so that
Ds Q
= (1.39)
Dt T
From the enthalpy equation (), this requires
Ds 1 DT κ Dp D ln(T p−κ )
= cp ( − ) = cp (1.40)
Dt T Dt p Dt Dt
It is convenient to define potential temperature, Θ, by the expression
s = cp ln Θ (1.41)
in terms of which (xx) is satisfied by setting

p κ
∗
Θ=T (1.42)
p
where p∗ is an arbitrary reference pressure, conventionally chosen to be
105 P a, close to the mean sea level pressure. The potential temperature is
the temperature realized by adiabatic displacement to the pressure p∗ .
For an atmosphere in hydrostatic balance, then
1 ∂Θ 1 ∂T κ ∂p 1 ∂T g
= − = ( − ) (1.43)
Θ ∂z T ∂z p ∂z T ∂z cp
If potential temperature (or entropy) is independent of height, the lapse

rate, the rate of temperature decrease with height is the adiabatic lapse
rate, g/cp ≈ 9.8K/km.
A familiar parcel displacement argument relates the atmospheric lapse
rate to gravitational stability. Starting with an atmosphere in hydrostatic
balance, lift a small parcel adiabatically by a distance δz, assuming that in
11
1. Preliminaries
doing so one does not perturb the pressure field (an assumption that can be
justified), and compute the density of the lifted parcel. To the extent that
parcel’s density no longer equals that of the environment ρ0 , but differs by
an amount δρ, the pressure gradient no longer balances the gravitational
acceleration, and instead
Dw d2 1 ∂p δρ
= 2 δz = −g − = −g (1.44)
Dt dt ρ ∂z ρ0
But the density of the parcel, compared to that of the environment, given
an unperturbed pressure field, is
δρ δΘ ∂Θ
=− = −δz (1.45)
ρ Θ ∂z
with the second equality following from the fact that Θ is conserved follow-
ing the parcel. Therefore,
d2
δz = −N 2 δz (1.46)
dt2
where
g ∂Θ
N2 ≡ (1.47)
Θ ∂z
N is referred to as the buoyancy or Brunt-Vaisala frequency. An atmosphere
with a lapse rate smaller than the adiabatic lapse rate is gravitationally sta-
ble, N being a measure of this stability, while a super-adiabatic atmosphere
in which the lapse rate is greater than the adiabatic value is unstable.
1.6 Equation of Motion

Ignoring viscous forces, the forces that we are typically concerned with in
the fluid dynamics of the atmosphere are the force of gravity and the force
generated by pressure gradients. Denoting the gravitational potential by
ΦG and the pressure field as p, we have for the inviscid equation of motion
in an inertial frame of reference
Dv ∂v
ρ = ρ( + (v · ∇)v) = −∇p − ρ∇ΦG (1.48)
Dt ∂t
The gravitational force exerted by the atmosphere on itself is negligible,
so we can think of ΦG as being specified independently of the state of the
atmosphere. Additionally, we will not be considering the gravitational tides,
so we can ignore the gravitational effects of the Sun and Moon and concern
12
1.6. Equation of Motion
ourselves only with the gravitational field of the Earth. Using the identity
(xx), one can rewrite this equation in the form
∂v 1 |v|2
= −ω × v − ∇p − ∇(ΦG + ) (1.49)
∂t ρ 2
For convenience, we typically move to a non-inertial coordinate system
rotating with the constant angular velocity of the Earth. (The Earth’s
angular velocity actually varies on a variety of time scales, and some of these
variations are of meteorological interest in that they are due to exchanges
of angular momentum between the atmosphere and the solid Earth, but
these variations are far too small to be of dynamical significance.) The
rotating frame of reference adds two additional terms to the equations of
motion: the centripetal acceleration, which can be written as the gradient of
a potential and added to the gravitational potential, and the all-important
Coriolis force.
The velocity v(r) of a fluid in solid body rotation with angular velocity
Ω is Ω × r, where r is vector pointing from the axis of rotation to the point
in question, Therefore, the velocity fields in rotating and inertial reference
frames are related by
vI (r, t) = vR (r, t) + Ω × r (1.50)
More generally, for any vector, the material derivatives in inertial and ro-
tating frames are related by
DI A DR A
= +Ω×A (1.51)
Dt Dt
We then have for the acceleration
DI vI DR vI
= + Ω × vI
Dt Dt
DR vR
= + Ω × vR + Ω × vI (1.52)
Dt
DR vR
= + 2Ω × vR + Ω × (Ω × r)
Dt
If R is defined to be the component of r perpendicular to Ω, we can write
1
Ω × (Ω × r) = −Ω2 R = − Ω2 ∇|R|2 (1.53)
2
The resulting equation of motion is
DR vR 1
+ 2Ω × vR = − ∇p − ∇Φ (1.54)
Dt ρ
13
1. Preliminaries
where
1
Φ ≡ ΦG + Ω2 |R|2 . (1.55)
2
If the flow is at rest in an inertial frame, then the flow is in hydrostatic
balance, with the pressure gradient in balance with the gravitational force.
Since the gradients are then parallel, surfaces of constant pressure are also
surfaces of constant gravitational potential, ΦG . If, instead, the flow is at
rest in the rotating coordinate system, then sufaces of constant pressure are
also surfaces of constant geopotential Φ.
If the Earth were a perfect sphere of radius a with a radially symmetric
gravitational potential, which we approximate near the surface as gz with z
the height above the surface, then the surface pressure at the pole would be
equal to the pressure at the equator at a height z = Ω2 a2 /(2g), or about 20
km. Assuming a scale height of roughly 6 km, the atmosphere would then
have a dramatic equatorial bulge, with the surface pressure at the equator
a factor of 10 or so larger than the surface pressure at the poles. Needless
to say, this bears no resemblance to the actual situation! The surface of the
Earth is, in fact, close to being a surface of constant geopotential, with an
equatorial bulge, and not a surface of constant gravitational potential. But
this bulge is small compared to the radius of the Earth, and, in addition, we
will always be working close enough to the surface that we can approximate
surfaces of constant geopotential as spheres. With this understanding, we
simply set the geopotential Φ equal to gz.
Using (xx), an alternative form of the equation of motion (xx) is
∂vR 1 |vR |2
= −(ωR + 2Ω) × vR − ∇p − ∇(Φ + ) (1.56)
∂t ρ 2
The vorticity of solid body rotation is simply 2Ω. This is most easily
confirmed by using Cartesian coordinates, with the z-axis along the axis of
rotation, so that (u, v) = (−Ωy, Ωx) and then computing the z-component
of the vorticity, ∂x v − ∂y u. The other components of the vorticity must
vanish by symmetry. The vorticity as viewed in the inertial frame and in
the rotating frame are related by
ωI = ωR + 2Ω (1.57)
We refer to ωI as the absolute vorticity and ωR (usually denoted simply as

ω), as the relative vorticity. Note how writing the RHS of (xx) in terms of
the absolute vorticity subsumes the Coriolis force. As we will consistently
use a rotating coordinate system in the following, we drop the subscript R
in the following unless there is some chance of ambiguity.
14
1.7. Energetics
1.7 Energetics
In the inviscid case, taking the dot product of the equation of motion with
the velocity yields an equation for the rate of change of the kinetic energy
density.
D 1 2
ρ |v| = −ρv · ∇Φ − v · ∇p (1.58)
Dt 2
or, for the rate of change of kinetic plus potential energy,
D 1 2
ρ ( |v| + Φ) = −∇ · (pv) + p∇ · v (1.59)
Dt 2
While we have ignored viscosity above, when we consider the energy
cycle in the atmosphere we need to keep in mind the dissipation of kinetic
energy by viscosity.
In the atmosphere there is a tremendous disparity between the scales of
motion that contain the bulk of the energy in the flow, on which most of our
interest will be focused, and scales at which molecular diffusion and viscosity
becomes significant. The time required to diffuse away a feature of scale L
in the absence of flow is L2 D, where D is the diffusivity , or viscosity in the
case of momentum. The ratio of this diffusive time scale to the advective
time L/U , is referred to as the Peclet number for tracers and the Reynolds
number for momentum. The order of magnitude of kinematic molecular
viscosities or diffusivities in the atmosphere near the surface is 10−5 m2 /s.
Large scale flows in the Earth’s atmosphere typically have velocity scales of
10 m/s and vertical scales larger than 1 km, leading to Reynolds number
of 109 . A hardly perceptible puff of wind near the surface with a speed of
1 m/sec and a modest vertical scale of 10 m still has advection dominating
viscosity by a factor of 106 .
The hope and expectation is that a theory for the circulation of the
atmosphere would not involve the value of the molecular viscosity in any
significant way. This expectation is based on the analogy with fully de-
veloped three-dimensional turbulence at very high Reynolds numbers, for
which it is found that the rate at which energy cascades to small scales
detemines the rate of dissipation of energy, while the value of the viscos-
ity simply determines the scale at which the dissipation takes place. In
fact, this is more than an analogy, since all significant dissipation in the
atmosphere (below heights of ≈ 100km at least), is presumed to occur in
patches of fully developed turbulence. Much of this dissipation occurs in
the planetary boundary layer, in which persistent turbulence is present near
the surface, but a significant fraction occurs in intermittent turbulence in
the free troposphere (”free” → ”above the planetary boundary layer”).
15
1. Preliminaries
(Actually, there is another process besides turbulent cascades that, a

bit surprisingly, dissipates a significant amount of kinetic energy in the
atmosphere. Raindrops fall at a rate determined by a balance between
viscous drag and gravity. Energy is dissipated by viscosity in this process.)
It is adequate for our purposes to use the expression for viscosity ap-
propriate for an incompressible flow (flow speedsa re small compared to the
speed fo sound, and the scales on which the viscosity acts are much smaller
than the scale height of the atmosphere). The additional term needed in
the equation of motion () is Dv/Dt = ... + ν∇2 v, where ν is the kinematic
viscosity.
Etc.
1.8 Spherical Coordinates

It should come as no surprise that spherical coordinates are useful in studies
of atmospheric dynamics. In addition to the radial distance from the origin
r, we use the symbol θ for latitude, which ranges from −π/2 at the south
pole to π/2 at the north pole, and λ for longitude, which ranges from 0 to
2π. The notation v = (u, v, w) is conventional for the (zonal, meridional,
vertical) components of the velocity field, with u postive for eastward flow,
v positive for northward flow, and w positive for a radially outward flow. In
the meteorological literature, and in common discourse, a wind with u > 0
is referred to as westerly (from the west), a wind with v > 0 as southerly,
and so forth. A note of caution: oceanographers speak of eastward, rather
than westerly, currents. Oceanographers want to know where currents are
taking their ship, while meteorologists as interested in where the air is
coming from.
It is helpful to be comfortable with the spherical coordinate expressions
for the gradient, divergence, and curl of a flow. We can write the expression
for an infinitesimal line segment in spherical coordinates as
δ` = λ̂(r cos(θ)δλ) + θ̂(rδθ) + r̂δr (1.60)
For a scalar field ξ, in order to retrieve δξ = δ` · ∇ξ the gradient must take

the form
1 ∂ξ 1 ∂ξ ∂ξ
∇ξ = λ̂ + θ̂ + r̂ (1.61)
r cos(θ) ∂λ r ∂θ ∂r
and for the material derivative
D ∂ u ∂ v ∂ ∂
= + + +w (1.62)
Dt ∂t r cos(θ) ∂λ r ∂θ ∂r
16
1.8. Spherical Coordinates
Care must be taken when considering the material derivative of a vector

in spherical coordinates, since the unit vectors are position dependent. For
example, the material derivative of the zonal component of the velocity,
Du/Dt, is not equal to the zonal component of the acceleration of a fluid
particle. Rather, with some effort one can show that the components of the
acceleration are
Dv Du uv tan(θ) uw
· λ̂ = − + (1.63)
Dt Dt r r
Dv Dv u2 tan(θ) vw
· θ̂ = + +
Dt Dt r r
Dv Dw u2 + v 2
· r̂ = −
Dt Dt r
where the additional terms on the right are referred to as metric terms.
One can remember the form of these terms by considering the angular
momentum and kinetic energy equations. The form of the metric terms in
the zonal (east-west) equation of motion can be obtained by considering the
material derivative of the angular momentum per unit mass, M = ur cos(θ).
(Here and in the following, the term angular momentum will always refer to
the component of the angular momentum vector along the axis of rotation.)
DM Du Dr Dθ
= r cos(θ) + u cos(θ) − ur sin(θ) (1.64)
Dt Dt Dt Dt
Du v
= r cos(θ) + u cos(θ)w − ur sin(θ)
Dt r
Du uv tan(θ) uw
= r cos(θ)( − + )
Dt r r
The fact that the metric terms must cancel when forming the rate of change
of kinetic energy can then be used as a mnemonic for the remaining terms
1 D|v|2 1 D(u2 + v 2 + w2 )
= (1.65)
2 Dt 2 Dt
Du uv tan(θ) uw
= u( − + )
Dt r r
Dv u2 tan(θ) vw
+ v( + + )
Dt r r
Dw u2 + v 2
+ w( − )
Dt r
Applying Gauss’s theorem to the infinitesimal volume element (with

volume r2 cos(θ)δrδθδλ) oriented along the coordinate axes, one finds that
17
1. Preliminaries
the divergence must be defined as

1 ∂u 1 ∂(v cos(θ)) 1 ∂(r2 w)
∇·v = + + 2 (1.66)
r cos(θ) ∂λ r cos(θ) ∂θ r ∂r
Applying Stokes’ theorem to an infinitesimal loop oriented along a pair of
coordinate axes, one can derive the expression for the component of the
vorticity normal to this loop. In particular, using a loop on the surface of
a sphere, we obtain for the radial component of the vorticity
1 ∂v 1 ∂(u cos(θ))
ωr = − (1.67)
r cos(θ) ∂λ r cos(θ) ∂θ
We have already seen that the vorticity of solid body rotation with
angular velocity Ω is 2Ω. The radial component of the vorticity of solid
body rotation is encountered so often that we denote it by the symbol f ,
also referred to as the Coriolis parameter :
f ≡ 2Ω · r̂ = 2Ω sin(θ) (1.68)
One can verify this expression from the spherical coordinate form of the
vorticity (xx) by setting (u, v) = (Ωr cos(θ), 0). The radial component of
the vorticity of solid body rotation is a monotonic function of latitude,
ranging from−π/2 at the south pole to π/2 at the north pole. We typically
denote the radial component of the relative vorticity as ζ, so that the radial
component of the absolute vorticity is f + ζ.
Computing the components of the Coriolis force in spherical coordinates,
we finally have for the equations of motion
Du uv tan(θ) uw 1 ∂p
− + = 2Ω sin(θ)v − 2Ω cos(θ)w −
Dt r r ρr cos(θ) ∂λ
2
Dv u tan(θ) vw 1 ∂p
+ + = −2Ω sin(θ)u −
Dt r r ρr ∂θ
2 2
Dw u + v 1 ∂p
− = 2Ω cos(θ)u − −g
Dt r ρ ∂r
with D/Dt defined by (xx).

The form of the Coriolis force in the zonal equation can be appreciated
once again by considering the rate of change of angular momentum. In
solid body rotation, the angular momentum is the eastward wind Ωr cos(θ)
times the moment arm r cos(θ). The total angular momentum, expressed
in terms of the flow in the rotating frame, is then
M = r cos(θ)(u + Ωr cos(θ)) = r cos(θ)u + Ωr2 cos2 (θ) (1.69)
18
1.8. Spherical Coordinates
The material derivative of the first term has already been computed in (xx),
leading to the metric terms in the zonal equation of motion. Computing
the derivative of the solid body term
D 2 v ∂ ∂
Ωr cos2 (θ) = ( + w )Ωr2 cos2 (θ)
Dt r ∂θ ∂r
= r cos(θ)(−2Ω sin(θ)v + 2Ω cos(θ)w)
Combined with (xx), we see that the zonal equation of motion in the rotat-
ing frame can be written
DM 1 ∂p
=− (1.70)
Dt ρ ∂λ
The Coriolis force in the zonal equation emerges from the advection of
the angular momentum of solid body rotation. Since the Coriolis force is
perpendicular to the velocity, the components of the Coriolis force in the
meridional and vertical equations are such as to cancel the terms in the
zonal equation when computing the rate of change of kinetic energy, just
as for the metric terms.
Consider the special case of a steady (∂/∂t = 0) axisymmetric (∂/∂λ =
0) flow. The angular momentum is then conserved following the flow in
the meridional-vertical plane. In particular, suppose that in one branch of
this circulation there is meridional motion at fixed radial coordinate, r = a.
Then Ωa2 cos2 (θ) + ua cos(θ) is independent of latitude along this branch.
Suppose this flow starts at the equator with flow at rest with respect to the
surface of the Earth, u(θ = 0) = 0. Then u(θ) = uM (θ), where
sin2 (θ)
uM ≡ Ωa (1.71)
cos(θ)
With a equal to the mean radius of the Earth (a), Ωa ≈ 460m/s, so if
a circulation of this sort exits from the equator to 30 latitude, a rather
substantial flow of ≈ 130m/s is generated. We can talk interchangeably of
this flow as being created by the conservation of angular momentum as a
ring of air moves polewards and, therefore, closer to the axis of rotation, or
of the zonal flow as being accelerated as the poleward flow is turned to the
right (in the Northern Hemisphere) by the Coriolis force.
Note that a flow in solid body rotation has a latitudinal gradient in its
angular momentum, with the angular momentum decreasing polewards. In
addition, for an arbitrary zonal flow (u, v) = (u(θ), 0), the radial component
of vorticity is proportional to the latitudinal angular momentum gradient:
1 ∂M
= −r cos(θ)(f + ζ) (1.72)
r ∂θ
19
1. Preliminaries
In the angular momentum conserving zonal flow with u = uM , the radial

component of the vorticity vanishes.
The Rossby number is a measure of the magnitude of the radial compo-
nent of the relative vorticity as compared to the radial component of the
vorticity of solid body rotation. If U is a characteristic magnitude of the
horizontal flow and L a characteristic length scale for variations in this flow,
then
|ζ| U
Ro ∼ ∼ (1.73)
f fL
One can think of the Rossby number as one possible non-dimensional mea-
sure of the departure of the flow from solid body rotation, focusing on the
resulting distortion of the radial component of the vorticity.
1.9 Vorticity in Homogeneous Incompressible Flow

The equation of motion tells us how the flow will evolve, given the density
and pressure distributions. At a minimum, one also needs equations for
the evolution of the pressure and density in order to evolve the state of the
atmosphere forward in time. Conservation of mass provides an equation
for the evolution of the density. To evolve the pressure distribution in
the case of an ideal gas, one can, for example, first relate the pressure to
the density and temperature (or entropy) through the equation of state,
and then compute the evolution of temperature (or entropy) from a heat
equation. We will need to turn to this general situation, but we first consider
a much simpler case of an incompressible flow with uniform density. Our
interest in uniform density, incompressible flow is primarily in the special
case of two-dimensional flow on the surface of a sphere.
An incompressible flow is one in which the divergence is assumed to
vanish everywhere and at all times: ∇ · v = 0. Scaling arguments suggest
that incompressibility is a good approximation for an atmosphere when 1)
the Mach number, the ratio of the flow velocity to the speed of sound, is
small and 2) when the vertical scale of the flow is small compared to the
scale height of the atmosphere over which the density varies by O(1) in the
Earth’s gravitational field. We will discuss these scaling arguments in (xx).
A homogeneous flow is one in which the density is uniform in space
and time. (Inhomogeneous incompressible flows will occupy our attention
in due course.) In the homogeneous, incompressible, and inviscid case, by
taking the divergence of the equation of motion (xx) one obtains a Poisson
equation for the pressure of the form
∇2 p = −ρ∇ · ((v · ∇)v + 2Ω × v + ∇Φ) (1.74)
20
1.9. Vorticity in Homogeneous Incompressible Flow
where the right hand side at time t can be computed from knowledge of the
flow field at time t. With appropriate boundary conditions on the pressure,
one can then think of the incompressibility condition as determining the
pressure, and the evolution of the state of the system (v, p) is then fully
defined. In situations of this sort, we will refer to the flow evolution as
being determined by a prognostic equation, with an explicit time derivative,
with the pressure being determined by a diagnostic equation or a balance
condition, or as being slaved to the flow field.
For an incompressible flow, we have that
∇ × (v × ω) = (v · ∇)ω − (ω · ∇)v (1.75)
This follows from (xx), using the incompressibility condition and the fact
that ∇ · ω = 0. Taking the curl of the equation of motion in the form (xx)
and using this result, we have
∂ω
= −(v · ∇)(ω + 2Ω) + ((ω + 2Ω) · ∇)v (1.76)
∂t
or
D(ω + 2Ω)
= ((ω + 2Ω) · ∇)v (1.77)
Dt
Comparing to Eq. (xx), we see that the absolute vorticity in a homoge-
neous, incompressible flow satisfies the same equation as does an infinites-
imal material line segment. If a line segment is initially aligned along the
absolute vorticity vector, it will stay aligned as the flow evolves, with the
magnitude of the vorticity increasing or decreasing as the length of the line
segment increases or decreases. Because of this property, we often speak of
the vorticity as being tilted or stretched.
Suppose the existence of a scalar field that is conserved following the
flow, Dξ/Dt = 0. If a conserved scalar is not readily available, one can be
created by painting the flow with a continuously varying color. We then
have
D
[(δ` · ∇)ξ] = 0 (1.78)
Dt
since the quantity in brackets is simply the difference in the value of ξ
between the two endpoints of the material line segment, and these two
values are separately conserved as the segement moves with the flow. But
the vorticity and the material line segment satisfy the same equation, so we
have also proven, for homogeneous incompressible flow, that
D
[(ω + 2Ω) · ∇ξ] = 0. (1.79)
Dt
21
1. Preliminaries
Since this proof is a bit indirect, the reader may want to verify this equation
directly by working out the derivatives using the vorticity equation and the
equation for the conservation of ξ.
The quantity conserved here is referred to as the potential vorticity (P V
for short). Changes in the gradient of the conserved scalar cancel the effects
of the stretching and tilting of the vorticity, creating another conserved
scalar from the vorticity vector.
Another centrally important perspective on vorticity evolution is pro-
vided by Kelvin’s circulation theorem. This celebrated theorem states that
the circulation around a material loop C is conserved for (inviscid) homo-
geneous incompressible flow:
Z
D
v · δ` = 0 (1.80)
Dt C
We can prove this result by expanding the line integral into two terms
(replace the integral by a discrete sum if you find this notation confusing)
Z Z Z
D Dv Dδ`
v · δ` = · δ` + v · (1.81)
Dt C C Dt C Dt
The first term is zero because it is the integral of a gradient, from the
equation of motion, since the density is assumed constant. (The assumption
of uniform density is crucial here.) The second term vanishes because (xx)
implies that it is also the integral of a gradient:
Dδ`
v· = v · [(δ` · ∇)v] = δ` · ∇(|v|2 /2) (1.82)
Dt
Applied to an infinitesimal material loop, Kelvin’s theorem and Stokes the-
orem together imply that the component of the vorticity perpendicular to
the loop, multiplied by the area of the loop δA, is conserved:
D
[(ω + 2Ω) · n̂δA] = 0 (1.83)
Dt
We can rederive potential vorticity conservation from this result by assum-
ing that the loop lies in a surface of constant ξ, the conserved scalar. We
then have the normal to the loop proportional to the gradient of ξ:
D ∇ξ D hδA
0= [(ω + 2Ω) · δA] = [(ω + 2Ω) · ∇ξ ] (1.84)
Dt |∇ξ| Dt δξ
where h is the distance between two surfaces of constant ξ differing by δξ.
But hδA is the volume of the cylinder depicted in Fig xx, and is conserved
following the flow since we have assumed incompressibility. By construction
δξ is also conserved. We thereby recover xx.
22
1.10. Planar vs. Spherical Two-Dimensional Flow
1.10 Planar vs. Spherical Two-Dimensional Flow

Consider first the special case of inviscid two dimensional flow on a plane,
in an inertial reference frame, with identically zero vertical velocity. For the
three-dimensional flow to be incompressible, the flow must be non-divergent
on the plane, and can be represented by a streamfunction ψ(x, y):
∂ψ ∂ψ
v = (u, v, 0) = (− , , 0) (1.85)
∂y ∂x
The vorticity has only a vertical component in this special case:
∂v ∂u ∂ 2ψ ∂ 2ψ
ω = (0, 0, ζ); ζ = − = ∇2 ψ = + 2 (1.86)
∂x ∂y ∂x2 ∂y
It follows from the vorticity equation that the vertical component of the
vorticity is conserved following the flow in this 2D case, there being no
stretching or twisting:
∂ζ ∂ζ ∂ζ
= −u −v = −J(ψ, ζ) (1.87)
∂t ∂x ∂y
where we use the shorthand Jacobian notation
∂A ∂B ∂B ∂A
J(A, B) ≡ − (1.88)
∂x ∂y ∂x ∂y
We can rederive the result that the vertical component of vorticity is
conserved following the flow from Kelvin’s Circulation Theorem by simply
choosing an infintesimal loop lying in the plane and noting that the area
of the loop is conserved since the flow is non-divergence in the plane. Or
one can use potential vorticity conservation by choosing the scalar to be
the vertical coordinate z, which is conserved following the flow because the
vertical velocity is zero by assumption.
Conservation of the vertical component of vorticity (or simply ”vortic-
ity”) forms the basis of the fascinating theory of two-dimensional turbu-
lence. What happens to the dynamics of this flow if we add rotation about
the z-axis? (By adding rotation we mean that ψ is now the streamfunction
of a two-dimensional flow in a rotating reference frame.) Kelvin’s circu-
lation or PV conservation immediately yields the result that Eq (xx) is
completely unchanged, since the vertical component of the vorticity of solid
body rotation is a constant.
Now consider the case of 2-dimensional flow on the surface of a sphere
of radius a. (It may be useful at times to think of the fluid as having a finite
23
1. Preliminaries
but negligible thickness.) The flow is now assumed to be non-divergent on

the sphere:
1 ∂u 1 ∂(v cos(θ))
+ =0 (1.89)
a cos(θ) ∂λ a cos(θ) ∂θ
which we can satisfy once again by defining a streamfunction
1 ∂ψ 1 ∂ψ
(u, v) = (− , ) (1.90)
a ∂θ a cos(θ) ∂λ
The radial component of the vorticity is
1 ∂v 1 ∂(u cos(θ))
ζ = − = ∇2 ψ (1.91)
a cos(θ) ∂λ a cos(θ) ∂θ
1 ∂ 2ψ 1 ∂ ∂ψ
= 2 2 2
+ 2 (cos(θ) ) (1.92)
a cos (θ) ∂λ a cos(θ) ∂θ ∂θ
Once again we use Kelvin’s Theorem, now using a loop lying on the
surface of our sphere. The result is that the radial component of the vorticity
is conserved following the flow. Equivalently, we can use potential vorticity
conservation, with the radial coordinate as the conserved scalar, to obtain
the same result. So the evolution equation for the radial component of
vorticity becomes simply
∂ζ u ∂ζ v ∂ζ
=− − (1.93)
∂t a cos(θ) ∂λ a ∂θ
All terms in this equation can be written in terms of the streamfunction.

The vorticity can be inverted uniquely to obtain the streamfunction by
solving Poissons equation (xx), which has a unique solution on the sphere.
For this 2D flow on a sphere, the vorticity field contains all information
about the dynamical state. To evolve the flow forward in time, given the
vorticity at some time t, we use conservation of vorticity to obtain the
vorticity at t + dt, after which we compute the flow consistent with this new
vorticity field (the flow induced by the vorticity distribution) by solving
Poissons equation for the streamfunction at t + dt.
Zonal averages, averages around a latitude circle, are of special interest
in theories of the atmospheric circulation. We denote the zonal average by
an overbar: Z 2π
1 C
u≡ udλ = (1.94)
2π 0 2πr cos(θ)
Here C is the circulation around the latitude circle. Applying Stokes Theo-
rem to this latitude circle, we find that the zonally averaged zonal wind at
24
any latitude is proportional to the radial component of vorticity integrated

over the polar cap bounded by this latitude circle. Rather than think about
the forces which modify the zonal mean zonal wind, we can, if we prefer,
think about the ways in which the radial component of vorticity integrated
over this polar cap can change.
A fundamental distinction between the spherical and planar cases results
when we add rotation. In this spherical case, the conserved quantity, the
radial component of the vorticity, is not uniform in solid body rotation, but
rather, is a monotonic function of latitude. In a rotating system we have
D(f + ζ) ∂ζ u ∂ζ v ∂(f + ζ)
= 0 =⇒ =− − (1.95)
Dt ∂t a cos(θ) ∂λ a ∂θ
or
Dζ
= −βv (1.96)
Dt
where the meridional gradient of the radial component of the vorticity of
solid body rotation is denoted by β:
1 ∂f 2Ω cos(θ)
β≡ = (1.97)
a ∂θ a
The importance of the background vorticity gradient is apparent when
one linearizes the dynamics about a state of rest in the rotating frame.
When we linearize equations, we typically start with a solution of the equa-
tions (in this case simply v = 0), and ask how small disturbances from this
state evolve. Being small, we assume that terms quadratic in the distur-
bance can be ignored. Denoting the perturbation with a prime, we find
that the the perturbations evolve according to
∂∇2 ψ β ∂ψ 2Ω ∂ψ
=− =− 2 (1.98)
∂t a cos(θ) ∂λ a ∂λ
By decomposing the initial condition into spherical harmonics (eignfunc-
tions of the Laplacian on the sphere), one can write the general solution to
this equation as a linear superposition of modal solutions (Rossby-Haurwitz
waves): X X
A`m P`m (cos(θ)) exp(i(mλ − ω`m t)) (1.99)
`=1,∞ m=−`,+`
with
2mΩ
ω`m = − (1.100)
`(` + 1)
Here P`m are associated Legendre polynomials. All of these modes prop-
agate westward. (The disturbance that propagates most rapidly westward
25
1. Preliminaries
(` = m = 1), is stationary in the inertial frame). This westward propagation

is clearly a consequence of the background vorticity gradient.
This background monotonic vorticity distribution provides large-scale
atmospheric and oceanic dynamics with much of its distinctive flavor, and
it can be important even when our interest is confined to a small enough
span of latitudes that the spherical geometry is otherwise irrelevant. For
this reason, we frequently make use of the tangent-plane, or β-plane, ap-
proximation to spherical two-dimensional flow, which simply amounts to
solving (xx) in Cartesian coordinates and assuming that β is a constant:
∂ζ ∂ψ ∂ζ ∂ψ ∂ζ ∂ψ
=− + −β = −J(ψ, ζ + βy) (1.101)
∂t ∂y ∂x ∂x ∂x ∂x
Linearizing this equation yields
∂ζ 0
= −βv 0 (1.102)
∂t
or
∂ ∂ 2ψ0 ∂ 2ψ0 ∂ψ 0
( 2 + ) = −β (1.103)
∂t ∂x ∂y 2 ∂x
Assuming a plane wave of the form
ψ 0 = <[Aei(kx+`y−ωt) ] (1.104)
results in the familiar Rossby wave dispersion relation

βk
ω=− (1.105)
k2 + `2
At fixed y, lines of constant phase propagate westward with the speed
c = ω/k = −β/(k 2 + `2 ). Longer waves, with smaller wavenumbers,
propagate faster. The background vorticity gradient is the source of the
anisotropic propagation; if the sign of the vorticity gradient were reversed,
the propagation would be eastward.
The local, tangent-plane approximation makes it easier to see the essence
of the dynamics resulting in Rossby waves, isolating this local dispersion
relation. Global modes, such as the Rossby-Haurwitz waves, or their coun-
terparts in more realistic models, do have a role to play in meteorology but
local Rossby wave-like dynamics is a dominant feature of the atmosphere.
One can add walls at northern and southern boundaries to create a finite
domain and mimic the spectrum of global modes obtained on the sphere,
and this reentrant channel geometry is quite common for this and other rea-
sons. But the walls create artificial features as well, unlike anything in the
26
atmosphere, and one also commonly considers a doubly-periodic (or toric)

geometry for the β-plane, in which the streamfunction in (xx) is assumed
to be periodic in both x and y, especially in studies of two-diomensional
turbulence.
27
2 Momentum fluxes in a barotropic
model
2.1 Vorticity Mixing

We continue to consider a homogeneous incompressible flow on the surface
of a sphere. Suppose that the fluid is initially in solid body rotation, with
angular velocity Ω. (From now on, in this context of two dimensional flow
on the sphere, the term vorticity will refer to the radial component of the
vorticity vector.)
Suppose that the fluid is now stirred by some external agent (see Figure
xx). We constrain the stirrer to act only within a well-defined range of
latitudes. Outside of this directly stirred region vorticity is conserved fol-
lowing the flow. This is not true within the stirred region, since the imposed
external stirring changes the vorticity on fluid particles.
Depending on the details of its motion, one can imagine that the stir-
rer might impart some horizontally integrated angular momentum to the
fluid, pushing the entire spherical shell either eastward or westward on av-
erage. In fact, the appropriate physical picture of this stirring of the upper
troposphere as due to vortex stretching associated with baroclinic eddy pro-
duction (getting ahead of ourselves) necessarily involves the vertical transfer
of angular momentum from the upper to the lower troposphere. From the
perspective of a model of the upper troposphere, stirring with this realistic
character would drain horizontally integrated angular momentum out of the
layer as a whole. We ignore this point for the time being as we temporarily
focus on the horizontal redistribution of angular momentum. Irrespective
of the dynamics of the vertical redistribution, it is the divergence of the
horizontal fluxes, when vertically integrated, that must balance the surface
stresses generated by the near-surface winds.
Once the stirring begins, the fluid will transmit the disturbance to other
latitudes (Fig. xx). Concentrating on a latitude circle outside of the directly
stirred region, we see that the air that has moved northwards through this
29
2. Momentum fluxes in a barotropic model
latitude has less vorticity than the air that has moved southwards, so there
is a net flux of vorticity southwards, down the mean vorticity gradient, as
the disturbance grows. The vorticity integrated over the polar cap bounded
by this latitude circle has decreased. By Stokes’ theorem, this implies that
the zonally averaged zonal flow, u, has decreased at this latitude.
The zonal mean flow is now easterly in a reference frame moving with
the initial solid body rotation. This argument can be made for any latitude
outside of the stirred region, on both the poleward and equatorward sides,
as long as the disturbance reaches this latitude. (It makes no difference if
one chooses to consider the latitude circle as the boundary of a polar cap
extending to the South Pole, rather than the North Pole.) Since they can
only redistribute angular momentum, the horizontal angular momentum
fluxes generated by the stirring must be directed into the stirred region!
This explanation for acceleration (∂u/∂t > 0) in the stirred region is a
bit indirect – one first argues for deceleration (∂u/∂t < 0) in the unstirred
regions that become populated by eddies spreading outwards from the stir-
ring; one then argues that the horizontal fluxes causing this deceleration
must be accelerating the stirred region. One way of making the argument
more direct is to assume that the stirring is concentrated in a short burst
that quickly creates eddies, which is then followed by a quiescent period
in which the stirring is turned off. During this latter period, vorticity is
assumed to be conserved following fluid particles throughout the flow. To
the extent that the disturbance spreads meridionally, eddy amplitudes can
be expected to decay in that part of the flow that was originally perturbed.
By a process opposite to that pictured in Figure 2, a poleward vorticity
flux will be generated in this region, to the extent that the flow reverts to
a more zonally symmetric form, and the zonal mean flow will thereby be
accelerated in the stirred region as the eddies decay. The compensating
deceleration occurs in those regions into which the disturbance spreads.
Suppose that the stirring is now turned off. If flow at some latitude
outside of the stirred region in Figure xx relaxes back to a zonal flow with
the vorticity on each particle being conserved, then the vorticity fluxes will
be reversed, and u will return to its initial value. If we want a stirring pulse
to generate a net deceleration at this latitude, some irreversible mixing must
take place. To the extent that eddy vorticity is mixed into the environment,
the vorticity transported during the decay phase will not fully compensate
the transport during the growth stage. The rectified effect of an event
in which some mixing occurs, in which pieces of vorticity break off and
are dispersed into the environmental vorticity distribution, will be a net
southward flux of vorticity and, therefore, deceleration of the flow in the
unstirred regions.
30
2.1. Vorticity Mixing
A bit of imagination is required to see the connection between this pic-

ture and the Earths climate. In the context of this homogeneous non-
divergent model, the claim is that one should think of the upper troposphere
as being continually stirred in midlatitudes – by the vortex stretching cre-
ated by baroclinic eddy production, a process that we will examine in due
course The eddies generated by this stirring spread meridionally and mix
vorticity irreversibly, and thereby transport angular momentum so as to
accelerate the flow within the stirred region, and decelerate it to the north
and south. In a statistically steady state, torques are required to balance
these accelerations. The momentum flux convergence integrated through
the depth of the atmosphere must be balanced by torques at the surface.
Independent of the detailed manner in which angular momentum is trans-
ported vertically from the upper troposphere to the surface, we require mean
surface westerlies in midlatitudes to provide surface torques to balance the
acceleration due to the eddies in the stirred region, and surface easterlies
to the north and south.
It is difficult to develop an intuitive understanding of this acceleration

due to stirring if one focuses on the momentum or angular momentum
fluxes. Following a fluid particle, momentum is modified by pressure gra-
dients. In contrast, vorticity is conserved following the flow in our simple
2-dimensional homogeneous incompressible flow on a spherical surface. Be-
cause it is (we hope!) relatively easy to develop intuition about the fluxes of
such materially conserved quantities, we presume that vorticity fluxes are
inherently simpler to think about than are momentum fluxes.
The sign of the mean vorticity gradient is crucial to this argument, as

it determines whether mixing the fluid will result in deceleration or accel-
eration. It is not necessary for the model atmosphere to be initially in solid
body rotation for the argument to work as stated; it is only necessary that
the northward vorticity gradient be positive. But this gradient will even-
tually be destroyed in a region where the eddies are successful in mixing
the fluid, if there are no restoring forces acting. The deceleration in the
mixed region will then cease. Therefore, a restoring force for the vorticity
gradient, in the upper tropospheric regions where this mixing is occurring,
would be an essential part of this barotropic picture if we were to try to use
is as a basis for a consistent statistically steady model. But flow restoration
is a part of the problem that is difficult to talk about realistically without
generalizing to the inhomogenous baroclinic case, in which radiative forcing
can restore potential vorticity gradients.
31
2.2 Eddy Vorticity and Momentum Fluxes

We have seen how a simple vorticity mixing argument, combined with a pic-
ture in which the upper troposphere is preferentially stirred in mid-latitudes,
provides an intuitively appealing explanation for the zonal mean surface
wind field. Before generalizing to a baroclinic atmosphere, we pause here
to examine some of the concepts that emerge naturally as one attempts to
describe more quantitatively how and why eddies modify the zonal mean
flow in the atmosphere, including the key concepts related to pseudomo-
mentum conservation. These concepts have a general applicability, but it
can be difficult to develop an intuitive understanding of them if they are
first encountered in the most general settings. We continue to work in the
context of the simple two-dimensional, non-divergent model for the time
being.
A simple manipulation, using the definition of vorticity, makes the re-
lation between momentum fluxes and vorticity fluxes more explicit. Let
us agree to ignore spherical geometry for the moment and consider a two-
dimensional non-divergent flow on a plane, with x increasing eastwards and
y increasing polewards. From the definitions of vorticity, ζ and divergence
(which is assumed to vanish identically) we have
∂u ∂v
+ = 0
∂x ∂y
∂v ∂u
− = ζ
∂x ∂y
It follows directly that
1 ∂ 2 ∂
vζ = (v − u2 ) − (uv) (2.1)
2 ∂x ∂y
Let an overbar denote a zonal average, an average over x, where we assume
that the flow is periodic in the x-direction, just as on the sphere. A prime
denotes the deviation from this average. There can be no zonally averaged
meridional flow in this non-divergent model, so the total zonally averaged
northward flux of vorticity is given by the eddy flux, which, in turn, equals
the convergence of the northward eddy flux of eastward momentum:
∂ 0 0
vζ = v 0 ζ 0 = − uv (2.2)
∂y
A similar relationship exists on the sphere between the vorticity flux and
the convergence of the angular momentum flux:
1 ∂
v0ζ 0 = − 2
(cos2 (θ)u0 v 0 ) (2.3)
a cos (θ) ∂θ
32
2.2. Eddy Vorticity and Momentum Fluxes
In the cos2 (θ) term on the right hand side, one factor of cos(θ) takes into
account the moment arm needed to convert momentum to angular mo-
mentum, while the other factor is present due to the convergence of the
meridians. The latitudinal patterns of the zonal mean vorticity flux and
angular momentum flux contain the same information; we can generate one
from the other by differentiating or integrating. A southwards, negative,
vorticity flux, as would be produced by mixing of vorticity down the gra-
dient produced by planetary rotation, is equivalent to a divergence of the
eddy angular momentum flux.
One can stare at this derivation for quite a long time and still not
develop any intuition for why it exists. Yet we already have an intuitive
understanding of this relationship from the discussion in xx. A southward
flux of vorticity across some latitude circle removes vorticity from the polar
cap north of this latitude, which, by Stokes Theorem, implies a reduction
in the zonally averaged zonal flow. Assuming non-divergent flow on the
surface of the sphere, fluid motions can change the zonal mean flow only
through the convergence of eddy angular momentum fluxes, so the vorticity
flux must be proportional to this convergence:
∂u 1 ∂
=− (cos2 (θ)u0 v 0 ) = v 0 ζ 0 (2.4)
∂t a cos(θ) ∂θ
A seemingly innocuous but centrally important consequence of this relation

is that the northward vorticity flux vanishes when integrated over latitude:
Z
v 0 ζ 0 cos2 (θ)dθ = 0 (2.5)
The manner in which vorticity can be transported is evidently strongly

constrained. In particular, if it is not identically zero, the integrand in this
expression, the northward eddy vorticity flux, must be positive in some
regions and negative in others. This constraint holds whether the flow is
unforced and invisicid or forced and dissipative.
We evidently have to be careful when we try to think of vorticity as
being mixed downgradient. In a complex turbulent 2-dimensional flow one
might intuitively expect a preexisting gradient of vorticity to be mixed away
through downgradient fluxes. This intuition presumably would rely on the
fact that we expect the vorticity, like a passive tracer that is conserved
following the flow, to be cascaded to small scales, eventually destroying
large-scale gradients. But if the mean absolute vorticity distribution is
initially monotonic, as is the case if we are close to solid body rotation on
the surface of a sphere, one cannot mix vorticity downgradient everywhere,
33
for then the eddy vorticity flux would be everywhere of one sign. It is not
primarily our intuition about the downgradient turbulent mixing of vorticity
that we have to re-examine, but rather the possibility of the existence of
self-generated turbulence in the first place, when the flow is dominated by
a monotonic vorticity distribution. We can always stir the flow externally
to create eddies, but, as we have seen, our intuition that vorticity will be
mixed must then be altered to take into account the fact that the stirring
changes the vorticity on fluid particles.
A barotropic flow with a monotonic absolute vorticity distribution ev-
idently has a certain stability. In the context of linear theory this is a
well-known result, reviewed in the next section, associated with the names
of Rayleigh and Kuo, but one need not restrict oneself to linear theory when
thinking about the implications of this constraint, as first emphasized by
Arnold. In contrast, if the mean vorticity gradient does change sign, then
the fluid has the potential to spontaneously mix vorticity downgradient,
through a process we refer to as barotropic eddy production.
2.3 Pseudomomentum
To create a more explicit statistically steady model, it is helpful to explic-
itly linearize the equations about an arbitrary zonal flow, u(y). The mean
meridional, north-south, flow in in this non-divergent model is identically
zero, and an arbitarary zonal flow is an exact solution of the invisicid equa-
tion of motion. Our interest is in the evolution of small perturbatons away
from this state. We will assume a Cartesian model on a β-plane here for
simplicity, but the reader can verify that all of the results below can be
transferred to the sphere.
If we want to model a statistically steady state, we also need a way of
dissipating eddy vorticity perturbations. To generate the simplest possible
model, we simply damp the eddy vorticity uniformly. This damping will
be the source of irreversibility in this linear system, playing the role dom-
inated by irreversible wave breaking in more realistic models of the upper
troposphere.
Once linearized, the advection of vorticity reduces to the sum of two
terms, the zonal advection by the prescribed zonal flow of the perturbation
vorticity and the meridional advection by the perturbation meridional flow
of the basic state vorticity. The zonal flow u makes a contribution ζ =
−∂u/∂y to the vorticity. Therefore, we can write the basic state meridional
34
2.3. Pseudomomentum
vorticity gradient as
∂ζ ∂ 2u
γ≡β+ =β− 2 (2.6)
∂y ∂y
The dynamics of these linear disturbances is then governed by the equation
∂ζ 0 ∂ζ 0
= −u − γv 0 + s − κζ 0 (2.7)
∂t ∂x
here s is the stirring, κ is the inverse of the damping time. We assume for
the time being that γ is positive at all latitudes.
Multiplying both sides of this equation by the perturbation vorticity
and then averaging over x, we obtain an equation for the mean square eddy
vorticity, or eddy enstrophy.
1 ∂ζ 02
= −γv 0 ζ 0 + s0 ζ 0 − κζ 02 (2.8)
2 ∂t
Eddy enstrophy, in addition to being created and destroyed by stirring and
damping, is generated by a downgradient (southward in this case) eddy
vorticity flux. This is an interesting equation, but it becomes even more
interesting if we divide both sides by the mean vorticity gradient γ, and
replace the vorticity flux by the momentum flux convergence:
1 ∂ζ 02 ∂ 1 κ
= − (−u0 v 0 ) + s0 ζ 0 − ζ 02 (2.9)
2γ ∂t ∂y γ γ
We now have an equation that has the form of a conservation law for a
quantity P with flux F and source/sink Q:
∂P ∂F
=− +Q (2.10)
∂t ∂y
The conserved quantity is
1 02
P= ζ (2.11)
2γ
We refer to P as the density of pseudomomentum. The integral of P over
latitude is conserved in time, if there is no stirring nor damping, and if
there is no flux of P out of the domain.
On the sphere, one speaks of the pseudo-angular momentum (or is it the
angular pseudomomentum?), which has an extra factor of cos(θ) multiplying
ζ 2 /(2γ), so that, in the absence of soources and sinks, and taking into
account the convergence of meridians, the quantity
Z π/2 2
ζ
cos2 (θ)dθ (2.12)
−π/2 2γ
35
is conserved for linear disturbances.

Conservation of momentum can be understood as a consequence of
translational invariance of a physical system; pseudomomentum conserva-
tion equations of this type can be shown to be consequences of the transla-
tional invariance of the basic state on which the waves propagate. The flux
of pseudomomentum, F, is often referred to as the Eliassen-Palm flux. In
this case, the flux of pseudomomentum is the negative of the momentum
flux,
F = −u0 v 0 (2.13)
The source/sink for pseudomomentum consists of two terms Q = S − D,
associated with the stirring and damping respectively. The reader might
benefit from deriving the analogous results on the sphere.
We now have for the mean flow modification due to the waves,
∂u ∂P
= v0ζ 0 = − +S−D (2.14)
∂t ∂t
Mean flow modification due to waves, in problems analogous to this two-
dimensional example, is a consequence of wave transients (the rate of change
of pseudomomentum), wave sources (S), and wave dissipation (D). This
mean flow modification, being quadratic in the amplitude of the wave mo-
tions, can be ignored initially in the computation of the RHS. If this mean
flow modification is sustained and no other forcing of the mean flow comes
into play that constrains the evolution, then the mean flow change will
eventually affect the eddy dynamics.
Consider the special case in which there are no wave sources or sinks,
so that S = D = 0. Once again suppose that initially the disturbance is
tightly localized in latitude, but as time evolves it spreads meridionally. As
we have already seen, we expect that the mean zonal flow will increase in
the region initially occupied by the eddy, and will decrease in those regions
into which the eddy spreads. Expressing the eddy amplitudes in terms of
the pseudomomentum allows us to determine by how much the mean flow
is modified: the change in u is precisely equal in amplitude, and opposite
in sign, to the change in P.
For the case of an arbitrary linear disturbance, with no sources or sinks,
the total pseduomomentum, the integral of the pseudomomentum density
over the domain, is conserved. If the vorticity gradient γ is everywhere
positive, this density is positive definite. As a result, if a disturbance is ini-
tially infinitesimal, it must always remain infinitesimal in order to conserve
pseudomomentum; therefore the flow is stable.
There are various formal definitions of stability; the type we are refer-
rring to here is often called stability in the sense of Liapunov, in which one
36
2.3. Pseudomomentum
bounds the growth of a quadratic, positive defiinite measure of the size of

the disturbance. It captures the essense of our intuitive notion of stability,
that an O() disturbance cannot grow to be O(1). Stability in this sense
potentially depends on the way in which we measure size. Rather than
pseudomomentum, suppose that we measure the size of a disturbance by
integrating its enstrophy over the domain. Eddy enstrophy is not conserved,
but it is easily checked that we have stability in the sense of enstrophy also.
Eddy enstrophy can grow somewhat by moving from a latitude of small γ
to one of larger γ, but the amount of growth possible is bounded as long as
γ is bounded.
One can understand why the pseudomomentum takes the the form that
it does by the following argument. Suppose the flow is unperturbed at time
t = 0 in the neighborhood of some latitude y = 0. As some later time t, a
perturbation has propagated into this region. Locate the initial positions
of the particles that now find themselves at y = 0 at time t, and define
η(x : t) to be the y-coordinate of the curve defined by these initial points.
The difference between the initial and final zonal mean flows is determined
by the amount of vorticity that has passed through y = 0 in the intervening
time. Since vorticity is conserved, this integrated transport is
Z η(x;t)
−1 1
u(t) − u(0) = L ζI (y, 0)dxdy ≈ γη 2 (2.15)
0 2
where L is the length of the channel, and the final expression on the RHS is
obtaining by assuming, for small displacements, that we can approximate
ζI ≈ C + γy and also set η = 0.
1
P = γη 02 (2.16)
2
. We can interpret η 0 as the meridional particle displacement required to
create the vorticity perturbation, etc.
If, instead, the mean vorticity gradient changes sign with latitude, then
there is the potential for growth since the contributions to the pseudomo-
mentum from different latitudes can cancel. This is the Rayleigh-Kuo crite-
rion for barotropic instability: the mean vorticity gradient must change sign.
Rayleigh considered the non-rotating plane-parallel case, for which this cri-
terion reduces to the statement that the flow curvature, ∂yy u must change
sign. Kuo extended this result to the rotating sphere, or equivalently, the
β-plane considered here. While the mathematics is the same, the result is
profoundly different. In the non-rotating case, a flow stable by the Rayleigh
criterion is exceptional: the simplest jet profile, say u = sech2 (y/L), will
have not one but two inflection points at which ∂yy u = 0. But on a rotating
37
sphere, due to the background monotonic vorticity distribution represented

here by the positive constant β, stability is the rule rather than the ex-
ception if the jet is sufficently weak. Rotation on the sphere is stablizing
to these barotropic flows because of the associated background monotonic
vortiicty distribution.
Returning to the general case with stirrring and damping, but now as-
suming that the eddy field is in a statistically steady state, we have instead
∂u ∂F
= =S−D (2.17)
∂t ∂y
The mean flow will be accelerated by the eddies where the source of pseudo-
momentum S is greater than the sink D. In a true statistically steady state,
this acceleration in the upper troposphere would be balanced by momentum
transport from the upper to the lower troposphere, where it would, in turn,
be balanced by surface stresses. The stress exerted by the atmosphere on
the surface is then simply proportional to the upper tropospheric S − D.
As long as the eddies spread in latitude, we expect S − D > 0, and surface
westerlies, in the source region.
The simple case of a steady unforced, inviscid wave is of special interest,
for then the vorticity flux is zero and the momentum flux is independent
of latitude, independent of the structure of the mean flow u through which
the wave propagates.
An alternative perspective on the horizontal redistribution of angular
momentum, based on the properties of Rossby waves, is often encountered
and can be very useful. It requires that the flow remain linear and that the
meridional variation of the zonal mean flow be slow enough so as to allow
one to define a local dispersion relation and a group velocity. One can then
show that the Rossby waves propagating to the north from the source region
generate a southward flux of eastward angular momentum, and that the
waves propagating to the south generate a northward flux. The convergence
of eastward momentum in the source region accelerates the flow, while the
divergence in the regions into which the waves are propagating produces
deceleration.
The meridional group velocity of a Rossby wave is
∂ω 2βk`
= 2 (2.18)
∂` (k + `2 )2
The wave propagating northward must have positive meridional group ve-
locity, or k` > 0. But the eddy zonal and meridional velocities in the wave
are
u = A`sin(kx + `y − ωt) (2.19)
38
2.3. Pseudomomentum
v = −A ksin(kx + `y − ωt) (2.20)

so the momentum flux averaged over x is simply
k`A2
uv = − (2.21)
2
which is negative in this case. Equatorward of the source, on the other hand,
we must choose k` < 0 to insure a southward group velocity, resulting in a
positive momentum flux. Figure xx shows the pattern of tilted streamlines
consistent with this choice of signs for the product k`. Superposition of
this pattern onto the mean westerlies produces the familiar tilted trough
structure of midlatitude eddies.
This linear wave perspective can be very useful, but it disguises the ex-
planation for the sign of the mean flow acceleration, making it appear to
depend on the details of the Rossby wave dispersion relation. We have seen
that the pattern of acceleration and deceleration follows directly from vor-
ticity conservation, without any restriction as to the linearity or nonlinearity
of the flow. Indeed, the connection between these two explanations seems
mysterious at first glance (even at second glance) and is worth pursuing.
Why does F = −uv turn out to be the meridional flux of P ? We have
already encountered one fact that helps us understand this: the meridional
group velocity of a Rossby wave packet is opposite in sign to the eddy
momentum flux, so it is of the same sign as F . This makes sense, because
P is simply a particular measure of eddy amplitude. If we are looking at
a well-defined wave packet moving meridionally, P must move with it, and
the sign of the flux of P must reflect this fact. Exploring this point further,
assume that there are no sources or sinks for simplicity and consider the case
of a wave packet moving with a well-defined group velocity G. Avoiding
the formal derivation, we expect to be able to write down that
∂P ∂ GP
=− (2.22)
∂t ∂y
If G is independent of y, this equation just expresses the fact that the
pseudomomentum moves with the wave packet. We place G inside of the
y-derivative to cover the case in which G varies with y, to insure that the
total amount of pseudomomentum is conserved. For a wave packet, we
therefore expect that F = GP .
We can confirm this expression for the special case of uniform flow, U =
const with γ = β. Consider once again a wave packet consisting of a carrier
wave with zonal and meridional wavenumbers k and `, and an envelope
with streamfunction amplitude A. As described above, the momentum
39
flux is −k`A2 /2 and the meridional group velocity of the wavepacket is

G = 2βk`/(k 2 + `2 )2 . Additionally, ignoring derivatives of the amplitude
modulation as compared to derivatives of the phase of the wave, the vorticity
is ≈ (−(k 2 + `2 )A/2, so the pseudomomentum density is
(k 2 + `2 )2
P ≈ A2 (2.23)
4γ
and, indeed, F ≈ GP ! Despite the fact that I have worked through this
derivation numerous times, I never fail to be impressed by the away in
which the various wavenumber factors cancel, as they must, to produce this
simple result.
It takes some effort to get accustomed to the idea of using pseudomomen-
tum as a measure of eddy amplitude, rather than a more familiar measure
such as eddy kinetic energy. In some respects, eddy kinetic is actually a
more complicated quantity than is the pseudomomentum. For example,
eddy kinetic energy, unlike pseudomomentum, is not conserved for linear
disturbances on a sheared flow, even in the absence of wave sources and
sinks. If we multiply Eq () by U and integrate over latitude, we obtain an
equation for the rate of change of the zonal kinetic energy. The negative of
this expression is the rate of change of eddy kinetic energy, since the total
energy of the flow is conserved as long as the flow is inviscid and unforced.
One can write the result in a couple of different ways; continuing to ignore
spherical geometry the result is:
∂ u02 + v 02
Z Z
∂
=− u v 0 ζ 0 dy = (u u0 v 0 )dy (2.24)
∂t 2 ∂y
Integration by parts has been used to obtain the last expression. The fa-
miliar consequence is that eddy kinetic energy is generated by an eddy
momentum flux that is directed down the mean velocity gradient. On the
sphere, it is the gradient of angular velocity, U/cos(θ) that is important in
this regard, as the reader may wish to verify.
We therefore have the result that a wave packet propagating from strong
westerlies into weaker westerlies will lose energy, for its momentum flux
will be directed upgradient; conversely, a packet propagating from weak to
strong westerlies will gain energy. When we see an eddy gaining energy from
the mean flow we are often tempted to look for an instability, a temptation
that should be resisted in the case of a disturbance propagating through a
shear flow that is stable by the Rayleigh-Kuo criterion.
40
2.4. Eddy Energetics
2.4 Eddy Energetics

An equation for the rate of change of the kinetic energy for this incompress-
ible, two dimensional flow can be obtained by taking the dot product of the
equation of motion with v,
∂ 1 2 1 p
( |v| ) = −∇ · [v( |v|2 + )] (2.25)
∂t 2 2 ρ
Kinetic energy changes locally becuse it is advected in from elsewhere and
because for the work performed by the pressure force. Energy transfer due
to pressure forces complicates atmospheric energetics. Averaging around a
latitude circle, and considering the Cartesian β-plane for simplicity,
∂ 1 2 ∂ 1 p
(u + v 2 ) = − v( (u2 + v 2 ) + ) (2.26)
∂t 2 ∂y 2 ρ
The total kinetic energy at any latitude can be decomposed into two
parts: the energy in the zonal mean flow, Z = u2 /2, and the energy in the
deviations from the zonal mean flow, E = (u02 + v 02 )/2, the latter referred
to as the eddy kinetic energy. (Recall that v = 0 in this incompressible 2D
flow0. From xx we have
∂Z ∂
= uv 0 ζ 0 = −u u0 v 0 (2.27)
∂t ∂y
and subtracting from (xx) results in the eddy kinetic energy equation
∂E ∂ 0 0 ∂ 0 1 02 p0
= u uv − v ( (u + v 02 ) + uu0 + ) (2.28)
∂t ∂y ∂y 2 ρ
∂u ∂ 0 1 02 p 0
= −u0 v 0 − v ( (u + v 02 ) + ) (2.29)
∂y ∂y 2 ρ
For small eddies, we retain only the quadratic terms on the RHS,
∂E ∂ 0 0 ∂ 1
= u uv − (uu0 v 0 + v 0 p0 ) (2.30)
∂t ∂y ∂y ρ
∂u 1 ∂ 0 0
= −u0 v 0 − vp (2.31)
∂y ρ ∂y
so that, integrating over the domain (assuming no fluxes at the boundaries
of the domain)
Z Z Z
0 0
∂u
Edy = − u v dy = − Zdy (2.32)
∂y
41
Eddy energy grows if the eddy momentum fluxes are downgradient on av-
erage. Unlike pseudomomentum, eddy kinetic energy is not conserved for
linear waves on a shear flow. Rather, energy moves from the zonal mean
compartment to and from the eddy compartment.
In the version (xx) of the eddy energy balance, it is traditional to refer
to the first term on the RHS as the conversion from zonal to kinetic energy,
associated with a local downgradient eddy momentum flux, and the second
term as the divergence of an energy flux v 0 p0 . But this decomposition into
conversion and transport is somewhat arbitrary, and we could equally well
refer to the first term in (xx) as the conversion and the second term as
convergence of an energy flux v 0 p0 + uu0 v 0 . (Motivation for referring to this
term as an energy flux follows from the fully nonlinear form (), from which
one can see that the term uu0 v 0 emerges from the flux of kinetic energy).
One can argue that this latter form is more natural, from the form of the
equation for Z, or by considering the case of a steady, unforced, inviscid
wave. Recall that for this special case the eddy vorticity flux vanishes but
the momentum flux does not, in general. So in such a wave, for which
there is no change in time of the eddy kinetic energy, there is a cancellation
between non-zero conversion and non-zero energy flux convergence using (),
while the two terms on the RHS side of () vanish individually.
We can combine the energy equation for small disturbances with pseu-
domomentum conservation to obtain
Z
∂
(E − uP)dy (2.33)
∂t
The integral of P is conserved, so if some P moves to a region of smaller

u, the integral of E must decrease. The intergrandin () is referred to as
the density of pseudoenergy, since a conservation equation of this form can
be obtained for analogous problems whenever the basic state on which the
waves propagate is independent of time and satisfies the equations of mo-
tion.
The reader can check that this discussion of energetics carries over to
the sphere. In particular, the spherical version of (xx) is
Z π/2 Z π/2
1 ∂ u
E cos(θ)dθ = − u0 v 0 cos(θ) ( ) cos(θ)dθ (2.34)
−π/2 −π/2 a ∂θ cos(θ)
On the sphere, eddy energy grows if the angular momentum flux is down
the gradient of the angular velocity.
42
2.5. Food for thought
2.5 Food for thought

The fact that the eddy flux of momentum in parts of the troposphere is
directed up the angular velocity gradient, so that kinetic energy is trans-
fered from eddies to the mean flow, has resulted in the teminology negative
viscosity being applied to these eddies. This term should be avoided in
discussions of the atmospheric climate. We have seen that energy transfer
from eddy to mean flow occurs whenever a disturbance propagates from
stronger westerlies to weaker westerlies, ie., whenever the eddy generation
region is localized within the zone of strong westerlies. If the dominant
eddy source were localized in the tropics, the eddies would instead extract
energy from the mean flow as they propagated into midlatitudes from the
tropics.
The term negative viscosity is sometimes used as a synonym for the
inverse energy cascade in two-dimensional flow. This inverse cascade will
be discussed in lecture ??. But one has to keep in mind that transferring
energy to larger scales is not equivalent to transfering energy to the zonal
mean flow. Think of the instability of a zonal jet in our two-dimensional
homoegenous incompressible flow. One can prove quite generally that as
an unforced flow evolves in two dimensions, the mean scale at which the
energy resides (as defined by the horizontal wavenumber weighted by the
kinetic energy spectrum) must increase (see, for example, Salmon, p. xx).
Yet the unstable waves must still extract kinetic energy from the mean flow,
this being the only source of energy in this problem. It is useful to try to
picture how the energy spectrum of the eddy and zonal mean flow evolve
to accomplish this.
It is also interesting to generalize the derivation leading to the relation-
ship between the zonal mean flux of vorticity and the meridional flux of
pseudomomentum . Before taking the zonal mean, the poleward vorticity
flux is related to the latitudinal derivative of uv and the longitudinal deriva-
tive of (v 2 u2 ). We have seen that uv is related to the meridional flux of
pseudomomentum. v 2 u2 is related to the zonal flux of psedomomentum in
precisely the same sense. To see this, note that the zonal group velocity of
our Rossby wave is
Gx = U + β(k 2 − `2 )/(k 2 + `2 )2 (2.35)
The rest of the derivation proceeds as before leading to the generalization:
vζ ≈ ∇ · (GP ) (2.36)
A stirred barotropic model with linear damping can be used to simulate
the observed eddy momentum fluxes in the upper troposphere rather well,
43
as recently discussed by DelSole (2001). Given that linear damping is pre-

sumably not a good model for the actual eddy dissipation mechanisms in
the upper troposphere, it is surprising how accurate a fit to the observations
can be generated in this simple way.
This mixing argument makes the claim that the zonally averaged dis-
tribution of surface winds (and the associated surface pressures and mean
meridional circulation) is determined by the location of the region in which
the system is ”stirred” by baroclinic instability. Suppose that the baroclinic
excitation happened to be concentrated in two distinct latitude spans per
hemisphere, rather than in one, as on Earth. Would there be two separate
regions of surface westerlies and a five-cell circulation? The answer is def-
initely yes. Just such a pattern can emerge in atmospheric models when
the eddy size is sufficiently small compared to the radius of the planet, as
occurs when the rotation rate is increased. Numerical models clearly show
that, given enough room, baroclinic excitation prefers to be organized into
distinct, latitudinally confined storm tracks, even when (actually, especially
when) the forcing creating the instability varies gradually in latitude. These
numerical simulations confirm the correspondence between the preferred re-
gions of excitation, the surface wind distribution, and the mean meridional
circulation. The complication is that the spatial pattern of the stirring can
be the result of self-organization that does not follow in a self-evident way
from the structure of the external forcing.
44

Atmospheric Dynamics: Isaac M. Held

Uploaded by

Copyright:

Available Formats

Atmospheric Dynamics: Isaac M. Held

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Atmospheric Dynamics: Isaac M. Held

Uploaded by

Copyright:

Available Formats

ATMOSPHERIC DYNAMICS

2 Momentum fluxes in a barotropic model 29

1.1 Brief Review of Vector Calculus

Here n̂ is a unit vector perpendicular to the surface S bounding the volume

The curl of any gradient vanishes: ∇ × ∇ξ = 0, as does the divergence of

Here S is any surface whose boundary is the loop L and n̂ is an outwardly

ijk imn = δjm δkn − δjn δkm (1.10)

1.2 Eulerian and Lagrangian Perspectives

so it should not be surprising that trajectories can be exceedingly complex,

1.3 Line segments, Volumes, and Conservation of Mass

1.5 Thermodynamics of an ideal gas

where T is the absolute temperature, and R is the gas constant. Alterna-

On large scales the atmosphere is hydrostatic, with a balance in the

so that cv DT /Dt = Q at constant volume and cp DT /Dt = Q at constant

where ω is the traditional notation for Dp/Dt. (Ambiguities arising from

in terms of which (xx) is satisfied by setting

If potential temperature (or entropy) is independent of height, the lapse

1.6 Equation of Motion

We refer to ωI as the absolute vorticity and ωR (usually denoted simply as

(Actually, there is another process besides turbulent cascades that, a

1.8 Spherical Coordinates

δ` = λ̂(r cos(θ)δλ) + θ̂(rδθ) + r̂δr (1.60)

For a scalar field ξ, in order to retrieve δξ = δ` · ∇ξ the gradient must take

Care must be taken when considering the material derivative of a vector

Applying Gauss’s theorem to the infinitesimal volume element (with

the divergence must be defined as

with D/Dt defined by (xx).

In the angular momentum conserving zonal flow with u = uM , the radial

1.9 Vorticity in Homogeneous Incompressible Flow

∇ × (v × ω) = (v · ∇)ω − (ω · ∇)v (1.75)

1.10 Planar vs. Spherical Two-Dimensional Flow

but negligible thickness.) The flow is now assumed to be non-divergent on

The radial component of the vorticity is

All terms in this equation can be written in terms of the streamfunction.

any latitude is proportional to the radial component of vorticity integrated

(` = m = 1), is stationary in the inertial frame). This westward propagation

results in the familiar Rossby wave dispersion relation

atmosphere, and one also commonly considers a doubly-periodic (or toric)

2.1 Vorticity Mixing

A bit of imagination is required to see the connection between this pic-

It is difficult to develop an intuitive understanding of this acceleration

The sign of the mean vorticity gradient is crucial to this argument, as

2.2 Eddy Vorticity and Momentum Fluxes

A seemingly innocuous but centrally important consequence of this relation

The manner in which vorticity can be transported is evidently strongly

is conserved for linear disturbances.

bounds the growth of a quadratic, positive defiinite measure of the size of

sphere, due to the background monotonic vorticity distribution represented

v = −A ksin(kx + `y − ωt) (2.20)

flux is −k`A2 /2 and the meridional group velocity of the wavepacket is

2.4 Eddy Energetics

The integral of P is conserved, so if some P moves to a region of smaller

2.5 Food for thought

as recently discussed by DelSole (2001). Given that linear damping is pre-

You might also like

ijk imn = δjm δkn − δjn δkm (1.10)