ROSSBY 

WAVES IN THE OCEAN (Part 1)

Stephen Riser, University of Washington
riser@ocean.washington.edu

Presented at National Institute of Oceanography, Goa 
April 5, 2010
 Rock

(1) H1
L

L L
( + waves ) c ≈ gλ → T = λ ≤
c 2

⎛ ρrock ⎞ ⎛ Vrock ⎞
H 2 = H1 + ⎜ ⎟ ⎜⎜ ⎟⎟
(2) ⎝ ρ water ⎠ ⎝ Apond ⎠
L
 Gaussian wind, 100 day oscillation

2000 km

2000 km

10 days 20 days 40 days 80 days

[Haidvogel and Rhines, 1983]

 Transverse and Longitudinal Waves

∇ i u = 0 , u ∼ e i ( kx + ly + m z − σ t ) ⇒ u i K = 0 K =( k , l , m )

particle motion
K transverse wave
wave crests (ocean waves)
wavenumber vector

wave crests

particle motion

longitudinal wave
K (sound waves)
wavenumber vector
 All linear wave problems will result in a governing PDE of the
form

ℜ (ϕ ) = 0 ( ℜ is a linear differential operator )

The approach to solving this PDE will generally be to assume

solutions of the form

ϕ ∼ e i ( kx + ly + m z − σ t )
K =( k , l , m )

This substitution will yield a relationship between the frequency

σ and wavenumber K in the form

σ = σ(k, l, m) = σ(K) K = wavenumber vector

→ the dispersion relation

 Phase speed (NOT phase velocity)….

∇ i u = 0 , u ∼ e i ( kx + ly + m z − σ t ) ⇒ u i K = 0 K =( k , l , m )

particle motion
K
wavenumber vector wave crests

Along the wavenumber vector, a line of constant phase moves at

a speed given by
σ
c=
K
Note that this applies only to a single, monochromatic wave, and
note that c is not a vector in the normal sense,

⎛σ σ σ ⎞
c ≠ ( cx , c y , cz ) ≠ ⎜ , , ⎟ ⎛ k l m⎞
K
K = ( k , l, m ) → =⎜ , , ⎟
⎝ k l m⎠ σ ⎝σ σ σ ⎠
“slowness”
A more useful description…..group velocity

Group velocity: the motion of a packet of waves

A(k)

ko k

The distribution of amplitudes of a wave packet in k-space

Suppose there is a wave packet moving in the +x direction, which can be
written as

ϕ ( x, t ) = ∫ A(k )e e ikx − iσ ( k ) t
dk
and further suppose that A(k) is peaked around a wavenumber ko.

In the vicinity of ko the dispersion relation is

A(k)
∂σ
σ ( k ) = σ ( ko ) + ( k − ko ) + ...
∂k
ko
k
so that ⎡ ∂σ ⎤
-i ⎢σ(ko )+ ko ( k -ko )+ ...⎥ t ko
ϕ ( x, t ) = ∫ A(k )e e ikx ⎣ ∂κ ⎦
dk
∂σ ⎡ ∂σ ⎤
ik(x- ko t)
i⎢ ko -σ ( ko ) ⎥ t
= ∫ A(k )e
ko
∂κ ⎣ ∂κ ⎦
dk e dk
⎡ ∂σ ⎡ ∂σ
∂σ
⎤ ⎤
⎡ ⎤ ι⎢
⎣ ∂κ
κο -σ ( ko ) ⎥ τ i⎢
⎣ ∂κ
κ
ko -σ ( ko ) ⎥ t

= ϕ ⎢x − = ϕ ( x − cgx t ,0)e
⎦ ⎦
κ
ο ο

t ,0 ⎥⎦ e
∂k
ko
⎣
 ⎡ ∂σ ⎤
i⎢ ko -σ ( ko ) ⎥ t
⎣ ∂κ
ϕ ( x, t ) ≈ ϕ ( x − cgx t ,0)e
⎦
k
o

modulation of the original packet

translation of the original packet at speed cgx

cgx = the x-group velocity; the speed at which energy of the packet moves
ϕ
ϕ ( x, to + Δt )
ϕ ( x, to )

t
to to+Δt

⎡ ∂σ ⎤
i⎢ ko -σ ( ko ) ⎥ t
⎣ ∂κ
ϕ ( x, t ) ≈ ϕ ( x − cgx t ,0)e
⎦
k
o
 4

Jet stream

5 3

Rossby waves in the atmosphere [500 mb pressure]

[ what about the ocean? ]
 A Short History of Rossby Waves

• Laplace (1799) formulated and discussed his tidal

equations and showed 2 types of solutions (“Motions of the
First Class”, and “Motions of the Second Class”).

• Margules (1893) examined the free oscillations of a

rotating, planetary atmosphere, confirming the existence of
Laplace’s second class solutions.

• Hough (1898) examined the free oscillations of a global

ocean of uniform depth, again confirming the second class
solutions, and found the eigenfunctions of the solution on a
sphere.

• Rossby (1939) introduced the β-plane approximation and

was able to study the transformed Laplace tidal equations
in detail for the atmosphere, resulting in important
advances in the study of Laplace’s motions of the second
class.
 Laplace’s Tidal Equations (LTE)
Ω ∂u g ∂η
− 2Ω v cosθ = −
θ
∂t a sinθ ∂φ
∂v g ∂η
+ 2Ω u sinθ =
∂t a ∂θ
∂η h ⎡ ∂ ∂u ⎤
h + ⎢ ( −v sinθ ) + ⎥ = 0
∂ t a sinθ ⎣ ∂θ ∂φ ⎦

h = mean ocean depth η = perturbed sea level

u = ( u,v,w ) a = Earth radius
π
θ= − latitude; φ = longitude; Ω = rotation rate
2
The LTE can be combined to form a single PDE in
terms of the variable η (sea level):

Put η ∼ ei( sφ -σ t ) ( where s is any integer ) to find

4Ω 2 σ
ℜ (η ) = εη ( where ℜ is a linear operator ) ε= γ =
gh 2Ω

1 ⎡ ∂ ⎧ 1 ⎛ ∂ ⎞⎫ s ⎛ γs ∂ ⎞⎤
ℜ= ⎢ ⎨ 2 ⎜ s cosθ − γ sinθ ⎟⎬ + ⎜ − cosθ ⎟⎥
γ sinθ ⎣ ∂θ ⎩ γ − cos θ
2
⎝ ∂θ ⎠ ⎭ γ − cos θ
2 2
⎝ sinθ ∂θ ⎠⎦

This is a formidable PDE to solve; it is not generally

separable in θ and λ. Approximate solutions have been
found by Margules, Hough, Longuet-Higgins, and others.
Hough’s approximate solution: eigenfunctions on
of the LTE on a sphere in terms of Legendre
polynomials (later called Hough functions).

−4

−2
Amplitude A(θ )

+3

+5
+1

η (φ ,θ , t ) ∼ A (θ ) e i ( sφ −σ t )

Latitude (deg)
 Rossby’s Contribution: The β - plane

π
f = 2Ω sin λ ; λ = latitude ; θ = −λ
2

f − f o = 2Ω ( sin λ − sin λo ) ≈ 2Ω ( sin λo + [ λ − λo ] cos λo + ...... − sin λo )

Ω
⎛ δY ⎞
f − f o = 2Ω cosλo [ λ − λo ] = 2Ω cosλo ⎜
λ
⎟ δy
⎝ R ⎠ λo

2Ωcosλo
f = fo + β y ; β= ; y = δY
R

the β− plane approximation; replaces spherical coordinates with Cartesian coordinates

[see Veronis (1963) for a complete treatment]
 The β − plane

2Ωcosλo
f = fo + β y ; β= ; y = δY
R
Note that only the first term of the Taylor series in λ has
been kept; this implies that β y << fo for consistency.
[unless fo = 0]

Thus,
βy ⎛ 2Ω cos λo ⎞ y
<< 1 ⇒ ⎜ ⎟ << 1 ⇒ y << R tan λo
fo ⎝ R ⎠ ( 2Ω sin λo )

At a latitude of 20°, for example,

y << R tan λo = ( 6.3 × 103 ) ( 0.36 ) = 2.3 × 103 km

 Use of the β − plane

∂u ∂η
− fv = − g
∂t ∂x
u and v momentum equations

f = fo + β y
∂v ∂η
+ fu = − g
∂t ∂y

∂η ∂ ∂
+ ( hu ) + ( hv ) = 0 integrated continuity equation
∂t ∂x ∂y

Put u, v,η ∼ e −iσ t

 These can be combined into a single PDE for v :

⎛ iβ ∂ σ −
2 2
⎞ v f
f = fo + β y
∇ v+⎜
2
⎟ + v=0
⎝σ ⎠ ∂x gh

Case (1): near the Equator fo = 0, so that f = β y .

⎛ β ⎞
−i⎜ + kx ⎟
put v=e ⎝ 2σ ⎠
F ( y) (separation of variables)

d 2 F ⎧⎪ ⎡ σ 2 β 2 ⎤ β y ⎫⎪
2 2
+ ⎨⎢ − 2 − k ⎥ −
2
⎬F = 0
⎪⎩ ⎣ gh 4σ
2
dy ⎦ gh ⎪⎭
[ a parabolic cylinder ODE; solution are Hermite functions ]
 Fo F1

y/Δ y/Δ

F2 F3

y/Δ y/Δ

The first few Hermite functions, showing the meridional structure of

equatorially-trapped waves (note: Δ ∼ 300 km)
The separated ODE yields the dispersion relation

gh ⎛ σ 2 βk ⎞
⎜ −k −
2
⎟ = 2n + 1
β ⎝ gh σ ⎠
n = 0, 1, 2, ……

which is cubic in σ and thus has 3 roots for σ for

any choice of n .
 στ

σ σ
<0 >0
κ κ

Δ ∼ 300 km
∂σ ∂σ τ ∼ 2 days
>0 <0
∂k ∂k

kΔ

The dispersion relation for near-Equatorial waves, determined from

the eigenvalue ODE; note that there are several distinct types of
waves.
Case (2): away from the Equator, so that f = fo + β y .

∂u 1 ∂p
− fv = −
∂t ρ ∂x The Rigid Lid Approximation

∂p
= − g ρ ⇒ p ( z ) = − g ρ ( z− ζ )
∂v 1 ∂p ∂z
+ fu = −
∂t ρ ∂y 1∂p ∂ζ
= g ρζ at z = 0 ⇒ =g
ρ ∂z ∂z
∂u ∂v
+ =0
∂x ∂y

The same basic equations, with a rigid lid approximation.

Put u, v, p ∼ e − iσ t
 ∂u 1 ∂p
− fv = − geostrophic balance:

∂t ρ ∂x generally accurate to
within 1-2% error at
mid-latitudes on time
∂v 1 ∂p scales > a few days and
+ fu = − length scales > 10s of
∂t ρ ∂y kilometers

∂u ∂v
+ =0
∂x ∂y
note: 3 equations, 3 unknowns
[resulting PDE admits a variety of solutions]

However, using the geostrophic approximation,

we can make an initial estimate of the solution:

1 ∂p 1 ∂p
u≈− v≈
ρ f o ∂y ρ f o ∂x
[geostrophic balance: a filter]
 1 ∂p 1 ∂p
u≈− v≈ [like a streamfunction]
ρ f o ∂y ρ f o ∂x
∂ψ ∂ψ p
u=− v= ψ =
∂y ∂x ρ fo

Cross-differentiate the u and v equations, substitute for the streamfunction,

and put the results in the continuity equation to yield single PDE for ψ :

∂ 2
∇ ψ + βψ x = 0 the Rossby wave equation
∂t [ a vorticity equation ]
 ∂ 2
∇ ψ + βψ x = 0
∂t
∂ ⎛ ∂v ∂u ⎞ ∂ ∂ 2 ∂ ∂y ∂f
⎜ − ⎟ =
∂t ⎝ ∂x ∂y ⎠ ∂t
(ψ xx + ψ yy ) =
∂t
∇ψ βψ x = β v = ( )
β y = β =
∂t ∂t ∂t
∂ζ rate of change of rate of change of
=
∂t relative vorticity planetary vorticity

∂
⎡⎣∇ 2ψ + β y ⎤⎦ = 0
∂t
⇒ ∇ 2ψ + β y = ζ + f = constant = PV y

x
particle motion
Assume wavelike solutions of the form

i ( kx + ly −σ t )
ψ =e wavenumber vector wave crests

transverse waves
This substitution yields

−iσ ( k + l 2 2
) + iβ k = 0
which becomes the dispersion relation for Rossby waves,

βk
σ =−
k +l2 2

σ β
A single Rossby wave
always propagates to
Note the east-west
=− <0 the west (eastward
k +l2 2
phase speed:
k propagation not
allowed) !

Take k ∼ 2π /1000 km, l = 0, then σ ∼ 2π /(44 days) → σ /k ∼ 40 cm/sec .

For the North Pacific, about 104 km wide, this wave would require
about 229 days to cross the basin.
 βk
σ =−
k +l
2 2
Rossby wave dispersion relation

Notes:

(1) The longest waves ( k, l → 0 ) have the highest temporal frequencies, σ → ∞ .

(2) Thus, to observe high-frequency waves, observations must be made over

large spatial scales; or,

(3) To observe short waves, observations must be collected over a long time.

Taken together, (1)−(3) suggest it will be

difficult to observe these waves in the ocean.
Group velocity….

⎛ ∂σ ∂σ ⎞
cg = ⎜ , ⎟
⎝ ∂k ∂l ⎠

⎛ β (k 2 − l2 ) ⎞
⎜ 2 β kl ⎟
cg = ,
⎜ ( k 2 + l 2 ) ( k 2 + l 2 )2 ⎟
2

⎝ ⎠

Result: the east-west group velocity of a packet of Rossby waves

can be either in the east or the west direction, but the east-west
phase speed for a single wave can be only in the westward
direction !
 βk l
σ =− short waves long waves

k +l
2 2
(k2, l2) (k1, l1)
• •
cg1
cph1
cg2
cph2

• k

⎛ β ⎞
⎜ − , 0 ⎟
⎝ 2σ ⎠

2 2
⎛ β ⎞ ⎛ β ⎞
⎜ k + ⎟ + l 2
= ⎜ ⎟
⎝ 2σ ⎠ ⎝ 2σ ⎠

Rossby Wave Dispersion Diagram

 ROSSBY WAVES IN THE OCEAN (Part 2)

Stephen Riser, University of Washington
riser@ocean.washington.edu

Presented at National Institute of Oceanography, Goa, April 5, 2010
Effects of stratification…. ∂w
∼ f
∂z

∂ ⎡ 2 1 ∂ ⎛ 1 ∂ψ ⎞ ⎤
⎢∇ ψ + 2 ⎜ 2 ⎟ ⎥ + βψ x = 0
∂t ⎣ f o ∂z ⎝ N ∂z ⎠ ⎦
vortex stretching term

ρoo f o
ρ = ρoo + ρo ( z ) + ρ ′ ( x, y , z , t ) ⇒ ρ ′ = − ψz
Boussinesq approximation
g
g ∂ρo
N2 = −
ρoo ∂ z

Look for solutions of the form

ψ = φ ( z)e i ( kx + ly −σ t )
 ψ = φ ( z)e i ( kx + ly −σ t )

φ ( z ) = cos mz , m = nπ / H
solution for N = constant, with boundary conditions that
w = 0 at the sea surface and bottom

For a more realistic solution, take N = Noeaz (exponential stratification),

⎡ ⎛ cn ⎞ ⎤
⎢ ⎛c Yo ⎜ ⎟ ⎥
n az ⎞ ⎝ a ⎠ ⎛ n ⎞ az
c
φ ( z ) = e ⎢Y1 ⎜ e ⎟ −
az
J1 ⎜ ⎟ e ⎥
⎢ ⎝a ⎠ J ⎛ cn ⎞ ⎝ a ⎠ ⎥
o⎜ ⎟
⎢⎣ ⎝a⎠ ⎥⎦
→ Y, J = Bessel functions, cn = eigenvalues
Vertical modal structure….
u, v, p w, ρ
[ exponential N(z) ]

1 3

z/H 2 1
3 2

u, v, p modes

[ constant N(z) ]

⎡ ⎛c ⎞ ⎤
⎢ ⎛c Yo ⎜ n ⎟
n az ⎞ ⎝ a ⎠ ⎛ cn ⎞ az ⎥
φ ( z ) = e ⎢Y1 ⎜ e ⎟ −
az
J1 ⎜ ⎟ e ⎥
⎢ ⎝a ⎠ J ⎛ cn ⎞ ⎝ a ⎠ ⎥
o⎜ ⎟
⎢⎣ ⎝a⎠ ⎥⎦

w, ρ modes
N = N o e az , a = 0.693 km −1 , N o = 6.04 × 10 2 day −1
cn = (2.014, 4.279, 6.532)
H = 5.4 km
Dispersion relation….

βk
σn = − 2
For a given k and l , the
baroclinic Rossby waves
f
k + l + 2 cn
2 2 o 2 have a longer period than
the barotropic waves
No
Take k ∼ 2π /500 km, l = 0, n =1, then σ ∼ 2π /(370 days) ; σ /k ∼ 1.3 cm/sec

For the North Pacific, approximately 104 km wide, this wave would take about
20 years to cross the basin.

Thus, baroclinic Rossby waves might travel very slowly (compared

to barotropic waves) and would seem to be inherently nonlinear;
this calls into question their possible existence.
The linear wave assumption….

u i∇u (U / L )
2
U
∼ << 1 ⇒ << 1
∂u σU σL
neglected nonlinear terms ∂t

σ
⇒ U << σ L ∼
K

particle speed phase speed

For a linear wave, the speed of the

fluid particles in the wave has to be
much less than the wave phase speed;
otherwise the wave is nonlinear and
likely unstable.
Other important issues related to simple Rossby waves:

• Effects of topography
In a 5000 m ocean at
βy βy δh 30°N, a change in depth
<< 1 ; suppose ∼ = 0.01 ⇒ of 50 meters over a N/S
fo fo h distance of ∼ 85 km is the
same size as the β-effect
in the vorticity equation.

• Wave-mean flow interactions

• Effects of coastlines (reflection properties)

• Nonlinearities
Pacific Island tide gauge stations

[Wunsch and Gill, 1976]

 Observations
[4-5 day band]

S N

model

[Wunsch and Gill, 1976]

S N
 The Pacific TAO Array

[Rossby wave model]

20°C isotherm depth during the
1992-1993 El Niño event

[Observations]

[Kessler and McPhaden, 1995]

Measured quantities from
the TAO array during the
1997-1998 El Niño event

Fits of a few equatorial

waves to the observations

[McPhaden and Xu, 1999]

Observations of Rossby Waves from Altimetry 39°N

32°N

21°N

[Chelton and Schlax, 1996]

 13 April 1993 31 July 1993

Rossby wave propagation is clearly evident near the Equator; at

other latitudes is is more difficult to see, and the sea level field is
less spatially coherent.

[Chelton and Schlax, 1996]

 The approximate ratio of particle
speed to phase speed as a function
of latitude for the Pacific

Result: nonlinearity is
generally very large except
near the Equator!
[Chelton and Schlax, 1996]
The trajectories of 18
acoustically-tracked floats
at a depth of 1300 m in the f
western N. Atlantic during h

the period 16 May through

11 July, 1979. Note the
wavelike motion with
particles oscillating on
NE/SW trajectories.

[Price and Rossby, 1982]

 •

The inferred dispersion curve from the Price-Rossby study

[Price and Rossby, 1982]

 relative

error

topographic
planetary

Estimates of individual terms in the potential vorticity equation from the

1300 m float array, showing approximate balance.

[Price and Rossby, 1982]

 Summary

• Rossby waves are the “motions of the second class”

originally found by Laplace and studied by many others.

• Rossby’s contribution was to formulate the theoretical

problem in Cartesian coordinates, using his β-plane
approximation, allowing useful solutions to be found.

• The properties of Rossby waves (their dispersion

relation) make them inherently difficult to observe.

• It seems likely that nonlinear effects in Rossby waves

are quite strong outside of the near-Equatorial ocean.