Dr. Joseph Pedlosky (Auth.) - Waves in The Ocean and Atmosphere - Introduction To Wave Dynamics-Springer-Verlag Berlin Heidelberg (2003) PDF

Waves in the Ocean and Atmosphere
Springer-Verlag Berlin Heidelberg GmbH

Joseph Pedlosky
Waves
intheOcean
and Atmosphere
Introduction to Wave Dynamics
With 95 Figures
, Springer
Author
Dr. Joseph Pedlosky

Woods Hole Oceanographic Institution
Department of Physical Oceanography
Clark 363 MS 21
Woods Hole, MA 02543
USA
e-mail: [email protected]
Library of Congress Cataloging-in-Publication Data Applied For
Bibliographic information published by Die Deutsche Bibliothek

Die Deutsche Bibliothek lists this publication in the Deutsche Nationalbibliografie;
detailed bibliographic data is available in the Internet at http://dnb.ddb.de
This work is subject to copyright. All rights are reserved, whether the whole or part of the material
is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation, broad-
casting, reproduction on microfilms or in any other way, and storage in data banks. Duplication of
this publication or parts thereof is permitted only under the provisions of the German Copyright
Law of September 9, 1965, in its current version, and permission for use must always be obtained from
Springer-Verlag. Violations are liable for prosecution under the German Copyright Law.
http://www.springer.de
ISBN 978-3-642-05564-5 ISBN 978-3-662-05131-3 (eBook)
DOI 10.1007/978-3-662-05131-3
© Springer-Verlag Berlin Heidelberg 2003
Originally published by Springer-Verlag Berlin Heidelberg New York in 2003.
Softcover reprint of the hardcover 1st edition 2003
The use of general descriptive names, registered names, trademarks, etc. in this publication does not
imply, even in the absence of a specific statement, that such names are exempt from the relevant pro-
tective laws and regulations and therefore free for general use.
Cover Design: Erich Kirchner, Heidelberg

Dataconversion: Buro Stasch, Bayreuth
Printed on acid-free paper - 32/3140 - 5432 10

Preface
For over twenty years, the Joint Program in Physical Oceanography of MIT and the
Woods Hole Oceanographic Institution has based its education program on a series of
core courses in Geophysical Fluid Dynamics and Physical Oceanography. One of the
central courses in the Core is one on wave theory, tailored to meet the needs of both
physical oceanography and meteorology students. I have had the pleasure of teaching
the course for a number of years, and I have particularly enjoyed the response of the
students to their exposure to the fascination of wave phenomena and theory.
This book is a reworking of course notes that I have prepared for the students, and I
was encouraged by their enthusiastic response to the notes to reach a larger audience
with this material. The emphasis, both in the course and in this text, is twofold: the de-
velopment of the basic ideas of wave theory and the description of specific types of waves
of special interest to oceanographers and meteorologists. Throughout the course, each
wave type is introduced both for its own intrinsic interest and importance and as a ve-
hicle for illustrating some general concept in the theory of waves. Topics covered range
from small-scale surface gravity waves to large-scale planetary vorticity waves. Con-
cepts such as energy transmission, reflection, potential vorticity, the equatorial wave
guide, and normal modes are introduced one step at a time in the context of specific
physical phenomena. Many topics associated with steady flows are also illustrated to
great benefit through a consideration of wave theory and topics such as geostrophic
adjustment, the transformation of scale under reflection, and wave-mean flow interac-
tion. These are natural links between the material of this course and theories of steady
currents in the atmosphere and oceans.
The subject of wave dynamics is an old one, and so much of the material in this book
can be found in texts, some of them classical, and well-known papers on certain aspects of
the subject. It would be hard to claim originality for the standard ideas and concepts, some
of which, like tidal theory, can be traced back to the nineteenth century. Other more recent
ideas, such as the asymptotic approach to slowly varying wave theory found in texts such
as Whitham's or Lighthill's, have been borrowed and employed to illuminate the subject.
In each case, references at the end of the text for each section indicate the sources that I
found particularly useful. What I have tried to do in the course and in this text is to weave
those ideas together in a way that I personally believe makes the subject as accessible as
possible to first-year graduate students. Indeed, I have tried to retain some of the infor-
mality in the text of the original notes. The text is composed of twenty one "lectures;' and
the reader will note from time to time certain questions posed didactically to the student
and certain challenges to the reader to obtain some results independently. A series of prob-
lem sets, which the students found helpful, are placed at the end of the text.
VI Preface
My teaching and research at the Woods Hole Oceanographic Institution has been
generously supported by the Henry L. and Grace Doherty chair in Physical Oceanog-
raphy for which I am delighted to express my appreciation. I also am happy to express
my gratitude for years of support from the National Science Foundation, which rec-
ognizes the inextricably linked character of research and teaching.
The waves course has been fun to teach. The fascination of the material seems to
naturally engage the curiosity of the students and it is to them, collectively, that this
book is dedicated.
Joseph Pedlosky
Woods Hole
May 05, 2003
Contents
1 Introduction ............................................................................. 1
Wave Kinematics .............. .. .. .................. 2
2 Kinematic Generalization ........................................ .................... 9
3 Equations of Motion; Surface Gravity Waves .................................... 19

First Wave Example: Surface Gravity Waves ......................................... 20
Boundary Conditions ................................................................... 23
Plane Wave Solutions for Surface Gravity Waves: Free Waves (Pa = 0) .......... 26
4 Fields of Motion in Gravity Waves and Energy .................................. 33

Energy and Energy Propagation ...................................................... 35
Addendum to Lecture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . ..... 39
5 The Initial Value Problem ............................................................ 41

Discussion ................................................................................. 49
6 Discussion of Initial Value Problem (Continued) ............................... 53
7 Internal Gravity Waves .............................................. . ................ 59

Group Velocity for Internal Waves .................................. 65
8 Internal Waves, Group Velocity and Reflection ................................. 67
9 WKB Theory for Internal Gravity Waves ......................................... 75

Normal Modes (Free Oscillations) ................................................... 79
10 Vertical Propagation of Waves:

Steady Flow and the Radiation Condition .. 91
11 Rotation and Potential Vorticity .................................................. 107
12 Large-Scale Hydrostatic Motions ...................................... 119

Potential Vorticity: Layer Model ........ .......................... . . . . . . . . . . . . . .. 120
Rossby Adjustment Problem................. ...................... . ............. 123
Energy ................................................................................... 129
VIII Contents
13 Shallow Water Waves in a Rotating Fluid;

Poincare and Kelvin Waves ........................................... . 133
Channel Modes and the Kelvin Wave ............................ . ................ . 136
The Kelvin Mode ....................................................................... 142
14 Rossby Waves ............ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 149
15 Rossby Waves (Continued), Quasi-Geostrophy ............................... 159

Quasi-Geostrophic Rossby Waves ................................................... 169
16 Energy and Energy Flux in Rossby Waves ...................................... 173

The Energy Propagation Diagram .................................................. 175
The Reflection of Rossby Waves ..................................................... 177
The Spin-Down of Rossby Waves ................ .................................. 180
17 Laplace Tidal Equations and the Vertical Structure Equation ............ 183
18 Equatorial Beta-Plane and Equatorial Waves ... .............................. 193

The Equatorial Beta-Plane ........................................................... 194
19 Stratified Quasi-Geostrophic Motion and Instability Waves .............. 205

Topographic Waves in a Stratified Fluid ........................................... 208
Waves in the Presence of a Mean Flow ............................................. 209
Boundary Waves in a Stratified Fluid .............................................. 211
Baroclinic Instability and the Eady Model ........................................ 213
20 Energy Equation and Necessary Conditions for Instability ............... 221

General Conditions for Instability .................................................. 227
21 Wave-Mean Flow Interaction ..................................................... 231
Problems ............................................................................... 239
References ............................................................................. 249
Index ..................................................................................... 253

Lecture 1
Introduction
A course on wave motions for oceanographers and meteorologists has (at least) two
purposes.
The first is to discuss the important types of waves that occur in the atmosphere
and oceans, in order to understand their properties, behavior, and how to include them
in our overall picture of the ocean and atmosphere. There are a large number of such
waves, each with different physics, and it will be impossible to discuss all of them ex-
haustively.
At the same time, a second purpose of the course is to develop the theory and con-
cepts of waves themselves. What are waves? What does it mean for a wave to move?
What does the wave do to the medium in which it propagates, and vice-versa? How do
waves (if they do) interact with one another? How do they arise? All of these are good
and fundamental questions.
In order to deal with both of these goals, the course will describe a series of differ-
ent waves and use each wave type to describe a different aspect of basic wave theory.
It will then be up to you to form the necessary connections and generalize the ideas to
all waves, at least on a heuristic basis. This will require you to sometimes retroactively
apply some new ideas developed in the discussion of wave type B, for example, back
to the application of wave type A discussed previously in the course.
In general, the physical ingredients will be stratification and rotation. But first, what
is a wave?
There is no definition of a wave that is simple and general enough to be useful, but
in a rough way we can think of a wave as:
A moving signal, typically moving at a rate distinct from the motion of the
medium.
A good example is the "wave" in a sports stadium. The pattern of the wave moves
rapidly around the park. The signal consists in the cooperative motion of individuals.
The signal moves a much greater distance than the motion of any individual. In fact,
while each person moves only up and down, the signal moves laterally (until it gets to
the costly box seats where it frequently dissipates).
Similarly in a fluid whose signal could be an acoustic pressure pulse, the surface
elevation of the ocean in a gravity wave, the rippling of the 500 mb surface in the tro-
posphere due to a cyclone wave, or the distortion of the deep isopycnals in the ther-
J. Pedlosky, Waves in the Ocean and Atmosphere

2 Lecture 1 0 Introduction
mocline due to internal gravity waves, the wave moves faster and further than the in-
dividual fluid elements. Thus, usually if
• u = the characteristic velocity of the fluid element in the wave, and

• c = the signal speed of the wave,
!!..«1 (1.1)
c
We shall see that this is also equivalent to the condition for the linearization of the
mathematical description of the wave physics.
Wave Kinematics
Before discussing wave physics, it is useful to establish some basic ideas and notational
definitions about the kinematics of waves. A more complete discussion can be found
in the excellent texts by Lighthill (1975) and Whitham (1974).
For simple systems and for small amplitude waves (i.e., when we linearize) we of-
ten can find solutions to the equations of motion in the form of a plane wave. This
usually requires the medium to be, at least locally on the scale of the wave, homoge-
neous. If l{J(xi,t) is a field variable such as pressure,
l{J(x,t) = l{J(Xi,t) = ReAei(Rox-rot) (1.2)
where
• A = the wave amplitude (complex so it includes a constant phase factor),
• K= the wave vector,
• (J) = the wave frequency, and
• Re implies that the real part of the following expression is taken.
We can define the variable phase of the wave eas

(}(x,t) = K ox-rot = kixi -rot (1.3)
where the summation convention is implied in the second form, that is,
max dim
kiXi ¢::} "LkjXj (1.4)
j=l
In the simplest case, A, (J) and kj are constants.
This begs the question of why we should ever observe a disturbance with a single
K = K and (J). To understand that we must do more work later on. But standing on a
beach and looking at the swell approaching it appears often to be the first order de-
scription of the wave field and a naturally simple case.
Of course, by Fourier's theorem (look it up now) we can represent any shape by a
superposition of such plane waves.
The function l{J we have considered above is constant on the surfaces (planes, hence
the name) on which e is constant, Le.,
kiXi - rot = constant ( 1.5)

Lecture 1 . Introduction 3
X2
A
Xl
() = constant
e= constant
Fig. 1.1. Schematic of wave crest Fig. 1.2. The plane wave showing the crests and
wave vector and wavelength
In two dimensions, for example, these will be the lines
R· x - wt = kjxj + kzxz - wt == kx + ly - wt (1.6)
(We will use the notation regularly, k j =k, k z =1, k3 =m, in Cartesian coordinates).
The directions of the lines of constant phase are given by the normal to those lines
of constant () (Fig. 1.1, Fig. 1.2), i.e.,
vo=vlR.x-wtJ=R (1.7)
or equivalently
~=~(k.x.-wt)=k. aXj =k-c5.. =k. (1.8)

aXi aXi }} } aXi } I} I
Define
K=IRI (1.9)
i.e., the magnitude of the wave vector. Then
R·x=Ks (1.10)
where s is the scalar distance perpendicular to the line of constant phase, for example
the crests where rp is a maximum.
The plane wave is a spatially periodic function so that rp(Ks) = rp(K[s + AD where
KA = 27t, since
ei(Ks) =ei (Ks+21t), ei (21t) == 1
Thus,
It = 27t (1.11)
K
4 Lecture 1 . Introduction
1.0
0.8
0.6
0.4
0.2
0.0
-0.2
-0.4
-0.6
-0.8
0
Fig. 1.3. 0 2 3 4 6 7
The wavelength of a plane wave f-------- A ~
Fig. 1.4.
e = 4n
The increase of phase in the direction of the
wave vector e = 2n
is the wavelength. It is the distance along the wave vector between two points of the
same phase (Fig. 1.3).
At any fixed position, the rate of change of the phase with time is given by
ao =-OJ (1.12)
at
OJ is therefore the rate of decrease of phase (note: as crests arrive, moving parallel to
the wave vector K, the phase will decrease at a fixed point (see Fig. 1.4).
How long do we have to wait until the same phase appears? The shortest wait occurs
when a time T has passed such that OJT = 2n. The time T is called the wave period,
and
T= 2n (1.13)
OJ
What is the speed of movement of the line of constant phase
O=kjXj -OJt=Ks-OJt (1.14)
Note that as t increases, s must increase to keep the phase constant (Fig. 1.5, 1.6).
as) ao tat OJ
(1.15)
at (J aotas K
y
R
Fig.l.S.
The movement with time of the e(t=O)
line of constant phase in the
direction of the wave vector x
0.5
- 0.5
-1
40
40
Fig. 1.6.
A plane wave in perspective view o '0 k =1, 1= 0.022
Be sure you understand the reason for the appearance of the minus sign:
{At constant e, de = 0 = Kds - mdt, so that ds / dt = w/ K}
We define the phase speed to be the speed of propagation of phase in the direction
of the wave vector.
phase speed: c = w/ K
Note that phase speed is not a vector. For example, in two dimensions the phase
speed in the x-direction would be defined such that at fixed y,
d8= 0 = kdx - wdt or (1.l6a)
W dB/dt
c =-=--- (1.l6b)
x k dB/dx
6 Lecture 1 . Introduction
Note that if the phase speed were a vector directed in the direction of K, its x-com-
ponent would be
-~ OJK~ OJ
C'I=--'I=- k #c (1.17)
K K K2 x
Therefore, it is clear that the phase speed does not act like a vector, and this is a
clue that this speed, by which the pattern of the wave propagates, may have less physi-
cal meaning that we would intuitively want to give to it.
Note that Cx is the speed with which the intersection of the moving phase line with
the x-axis moves along the x-axis (Fig. 1.7):
C
Cx =--
cosa
and as a goes to reI 2, Cx becomes infinitely large! This makes us suspicious that the
phase may not be the messenger of physical entities like momentum and energy.
In an interval length s perpendicular to the surface of constant phase, the increase
in phase divided by 21t gives us the number of crests in the interval. Thus, Fig. 1.8.
y
K
Fig. 1.7.
The small arrow shows the intersection point of
the line of constant phase and the x-axis x
1.0
1\ /\ 1\ f\ I
0.8
0.6
0.4
0.2
---.~ s
0.0
-0.2
KS / 2re = # of crests -0.4
-0.6
-0.8
-1.0
\/. \I. .\1 . \J . \1 .
o s 10 15 20 25 30 35
Fig. 1.S. A plane wave and the number of crests along the coordinate s
Thus also, the increase in phase along the wave vector is
af}
~f}= J-ds= JKds (1.18)
as
The more fundamental definitions have already been given; namely,
K=V8 (1.19a)
af}
OJ=-at (1.19b)
The former gives the spatial increase of phase, while the latter gives the temporal
(decrease) of phase.
In all physical wave problems, the dynamics will impose, as we shall see, a relation
between the wave vector and the frequency. This relation is called the dis per s ion
rela tion (for reasons that will be made more clear later), The form of the dispersion
relation can be written as:
OJ=il(kj ) (1.20)
Note that each wave vector has its own frequency. Often the frequency depends only
on the magnitude of the wave vector, K, rather than its orientation, but this is not al-
ways the case. Up to now, the wave vector, the frequency, the phase speed and the dis-
persion relation have all been considered constants, i.e., independent of space and time.
Lecture 2
Kinematic Generalization
Suppose the medium is not homogeneous. For example, gravity waves impinging on a
beach see of varying depth as the waves run up the beach, acoustic waves see fluid of
varying pressure and temperature as they propagate vertically, etc. Then a pure plane
wave in which all attributes of the wave are constant in space (and time) will not be a
proper description of the wave field. Nevertheless, if the changes in the background
occur on scales that are long and slow compared to the wavelength and period of the
wave, a plane wave representation may be locally appropriate (Fig. 2.1). Even in a ho-
mogeneous medium, the wave might change its length if the wave is a superposition
of plane waves (as we shall see later).
---+ I\, .---

1.0
0.8
0.6
0.4
0.2
0.0
-0.2
-0.4
-0.6
-0.8
-1.0
\I , \/ \/
0 10 20 30 40 50 60 70 80 90 100
1-------- LM ------ r.
Fig. 2.1. Schematic of a slowly varying wave

10 Lecture 2 . Kinematic Generalization
Thus, locally the wave can still look like a plane wave if AI LM « 1. In that case, we
might expect the wave to be described by the form:
rfJ(x,t) = A(x,t)e i8(x,t) (the real part of the expression is taken for granted), (2.1)
where A varies on the scale LM while the phase varies on the scale A. Thus,
1 aA _0(_1) (2.2a)
A aXi l LM
~=o(~)
aXi A
(2.2b)
so that
VrfJ =AeiOVe +ol L: J (2.3)
We define (guided by our experience with the plane wave):
K= ve local spatial increase of phase (2.4a)
-W= aB local increase of phase with time (2.4b)

at
Since the wave vector is defined as the gradient of the scalar phase, it follows auto-
matically that V x K = O.
Consider the increase of phase on the curve C1 from point A to point B in Fig. 2.2:
1B _ _ 1 __
nC I =-fK .dx =- fK.dx (2.5)
21t A 21t C1
--------, B
(2
Fig. 2.2.
A
Counting crests on two paths
AC1B and AC2 B () = constant
Lecture 2 . Kinematic Generalization 11
Now consider the same increase calculated on curve C2 :
IB _ _ 1 - _
nc z = - JK.dX=- JK.dX (2.6)
21t A 21t C z
The difference between them is
I z 21t
J J- -
nc -nc =1- - K·dX=- 1
21t
K·dX f- -
c1 C z Ctota!
(2.7)
=JJVxK.ndA
A
=0
Here we have used Stokes theorem relating the line integral of the tangent compo-
nent of K with the area integral of its curl over the area bounded by the closed con-
tour composed of the sum of the two curves C1 and C2 • Since the curl is zero, the two
calculations for the increase of phase must be independent of the curve used to do the
calculation.
Note that since
K=ve (2.8a)
ae
m=--at (2.8b)
it follows by definition that
aK +Vm=O (2.9)
at
in those cases where the wave vector and the wave frequency are slowly varying
functions of space and time (i.e., where it is sensible to define wavelength and frequency).
To better understand the consequences of the above equation, consider the fixed
line element AB in Fig. 2.3.
Fig. 2.3.
Conservation of crests along
the line AB A
Integrate the above conservation equation along the line element from A to B:
aB _ B
- JK.d.X+ JVm.d.X=o (2.10)
at A A
Using our previous definitions, in particular that Ks !21t is the number of crests in
the interval s, it follows from the above that
anAB = m(A) _ m(B)

(2.11 )
at 21t 21t
That is to say, the rate of change of the number of crests in the interval (A,B) is equal
to the rate of inflow of crests at point A minus the outflow of crests at point B, since
the frequency (divided by 21t) is equal to the number of crests crossing a point at each
moment. E.g.,
m(A) = rate of decrease of phase at point A (see Fig. 2.4)
Phase increasing in space
'T:(~
1.81- f \\
~
7\ \
\
\
\
/ \
\
l6[
\
\ \
\ \
1.4 \ I
I t+ 8t I
\
\
1.2 ~ I \
I I
1.0 ~ \. I
\
\ I I
0.81- I I
I I \
I I \
0.6
I
0.4 I
0.2
/
A /
0.0
a 2 3 4 5 6 7 8 9 10
Fig. 2.4. The movement of the phase through the interval AB

Lecture 2 . Kinematic Generalization l3
We may think of this as a statement of the conservation of wave crests. Namely,

the number of wave crests in a smoothly varying function ¢ as given above does not
change. The number in any local region increases or decreases solely due to the ar-
rival of preexisting crests, not to the creation or destruction of existing crests.
Now, let's suppose that we still have a local dispersion relation between frequency
and wave number but that the relationship slowly changes on scales that are long com-
pared with a wavelength or period due to changes, perhaps, in the nature of the me-
dium in which the wave is embedded.
In that case, the natural generalization of the dispersion relation is
0)= Q(kpXi,t) (2.12)
where the wave vector components and the frequency may themselves be functions
of space and time (slowly), and the dispersion relation is explicitly dependent on space
and time.
Thus,
dO) dn) dn dkj

-=- + -- (2.13)
dt dt K,x dk j dt
where the first term on the right-hand side is due to the explicit dependence of the
dispersion relation on time, as might happen if the temperature of a region through
which an acoustic wave were traveling were increasing with time.
We define the group velocity by the formula for each of its Cartesian components:
dn
Cgj (2.14)
dk j
for the component of the group velocity in the ph direction, or
Cg = 'VRQ (2.15)
It follows from a fundamental theorem in vector analysis that since the phase
is a scalar and the gradient operator is a vector, the group velocity is a true vector
(distinct from the phase speed). That is, it follows the law of vector decomposi-
tion.
Since, by our earlier definitions
dk j __ dO)
(2.16)
at"- dXj
we thus obtain
dO) dn dn dO)
-=---- (2.17)
dt dt dk j dXj
It therefore follows that
aw -" an
I' . denvatlvewlt
-+Cg ' vw;;-<=expIClI . . . h tune
. (2.18)
at at
Again, by similarly using
oki + ow =0
ot ox;
it follows that
oki + on + on ok; = 0 or (2.19a)

ot OXi ok j OXj
ok; on oki on
-+--=-- (2.19b)
ot ok j OXj oXi
Since the wave vector has no curl, it follows that
ok; = okj
OXj ox;
so the above equation can be rewritten:
i1K + (cg •V)K =-V n <= explicit dependence on space (2.20)

at
Note that the sum of derivatives on the left in the equations for the rate of change
of wave vector and frequency are the rate of change for an observer moving with the
group velocity.
So,
1. If the medium is independent of time, ~ ill propagates with the group velocity;
2. If the medium is independent of space, ~ K propagates with the group velocity.
If both (1) and (2) are true, both frequency and wave number propagate with the
group velocity:
on
cgi = oki
This is a vector, and we see here that real wave attributes propagate with this velocity. If
the dispersion relation is a function of space and/or time, the above equations tell us how
the frequency and wave number change as we move with the group velocity following a
wave. Further discussion can be found in Bretherton (1971) and Pedlosky (1987).
Example
We will soon see that free surface gravity waves (short enough so that rotation is
unimportant but long enough so that the wavelength is large) compared to the depth
have a dispersion relation:
w=kjiii
where H is the depth of the fluid and k is the wave number for this one-dimensional
example (Fig. 2.5).
The phase speed and group velocity are equal in this case:
Cg =C=(gH)1/2
If the depth is a function of x, then following a signal, since the dispersion rela-
tion is independent of time, the frequency will be constant for an observer moving
==
with the velocity cg C (gH) 1/2. For such an observer, with frequency constant,
k = const.! HII2, which implies that the wave will grow shorter (larger k) as the wave
enters shallow water. (It may become so short that it might break). Note that the
observer, following a particular frequency moving with the group speed will pro-
ceed at a rate:
dx
dt""(gH(X»1/2 (2.22)
For example. if H(x) is of the form H"" Ho( 1 - x! xo) where x is measured posi-
tive shoreward from some offshore position a distance Xo from the waterline (see
Fig. 2.5), the signal corresponding to a given frequency will proceed onshore such
that at a point x after an elapsed time t, the relationship between the elapsed time
and its onshore progress is
t = 2Xo(I-~1-X I Xo )!(gHo )1/2 (2.23)
Fig. 2.5.
Water wave running up a
sloped beach x=Q x=xo x
The above kinematic discussion doesn't tell us how the amplitude of the wave propa-
gates or, equivalently, how the energy in the wave moves. In some simple cases that
are general enough to be of interest, we can actually describe how the amplitude and
hence energy moves.
Consider the case of a homogeneous medium in which the governing equation for
the wave function ¢ is of the form
majat,ajaXi )I/J(xi,t) = 0 (2.24)
where IT is a polynomial in the partial derivatives with respect to space and time. A
simple example would be the Rossby wave equation:
--+--+
( a3 a3 /3-
ax 2 at
a)
axay 2 at
I/J=O (2.25)
so that in this case,
TI= :t[ :J:J+ ~( ~)]+/3 :X

i.e., the polynomial in the partial derivatives are in respect of x, y and t.
Suppose we look for an approximate solution of the form
~A(X,t)
h.I,
v
y
v/
/
1/
,.1//
80 100 120 140 160 180 200
r------------------------ LM ~
Fig. 2.6. A wave packet. The wave has wavelength A while its envelope has a scale LM
where A, k and 0) are slowly varying functions of time, i.e., where the solution has the
form of a one-dimensional wave packet (see Fig. 2.6), then
a¢
ax
=(i aeax A+ aA)e
ax
iB
(2.26)
=(I'kA + aA)
ax eiB , etc.
or
n¢ =0 => n(-im+~'ik +~)A = 0 (2.27)

at ax
Expanding the polynomial using the fact that the time and space derivatives of A
are small compared to m and k,
m-im,ik)+~ aA + an aA_ (2.28)

a(-lUJ) at a(ik) ax -0
The dispersion relation for plane waves comes from the disappearance of the first
term (which is the dominant one), namely
TI( -iOJ, ik) = 0 ~ Linear dispersion relation (2.29)
In the case above, this yields m = -f3 I k.

When this dispersion relation is satisfied, the remaining term yields the condition:
aA anIak aA =0 (2.30)
at anlam ax
where the derivatives of TI in the equation occur when TI is evaluated as a function of
frequency and wave number as in Eq. 2.29.
Since
anIak am) (2.31)

anlam =----ak il
it follows that
aA
-+c -=0
aA (2.32)
at g ax
Thus, the amplitude (and we can suppose) energy will propagate with the group
velocity and not the phase speed. Where the envelope (that is A) of the wave goes,
that is where the energy is. There is clearly no energy outside the wave envelope.
The reader should calculate the group velocity for this simple case of one-dimensi-
onal Rossby waves to see that the group and phase velocities are not the same. Similarly,
the argument presented here can be extended to any number of dimensions (try it).
It is also clear that one might be able to use similar ideas for inhomogeneous media.
Once again we see here the physical primacy of the group velocity over the phase
speed for the propagation of physical attributes of the wave.
Lecture 3
Equations of Motion; Surface Gravity Waves
For a rotating stratified fluid, the general equations of motion can be written as:
1. Momentum equation:
p[ !~ +2Qxu ] =-Vp+ .uV2U+ KV(V ·u) (if.u constant, K'is second viscosity) (3.1)
2. Mass conservation:
ap+V.(pU)=O ;and (3.2)

at
3. Thermodynamic energy equation:
ds =H
dt
where s is specific entropy and H is the nonreversible heat addition. This can be re-
written, assuming that s is a thermodynamic function of p and p,
cdT - dp =lP+!V 2T+Q+K'(V'U)2=H

- -aT (3.3)
P dt P dt p
Here, T is temperature, cp is the specific heat at constant pressure, a is the coeffi-

cient of thermal expansion, and lP is the dissipation function, i.e., the frictional trans-
formation of mechanical to thermal energy. If 'rij is the stress tensor and eij is the rate
of the strain tensor, lP = 'ri/ij (sums implied). Note that
a=-~(~~J p
(3.4)
For a perfect gas with a state equation p = pRT, the thermodynamic equation is
usually written in terms of the potential temperature:
RIC
8=T ( ~) p

20 Lecture 3 . Equations of Motion; Surface Gravity Waves
so that the thermodynamic equation becomes
1 dB H
(3.5)
B dt cpT
while for an incompressible liquid we can approximate the thermodynamic equation with
dp =_ap H (3.6)
dt cp
Here we have used the approximate state equation p =Po(1- a(T - To)) to relate
temperature in the thermodynamic equation to density. Be sure to note that when we
make the approximation of incompressibility in the mass equation (V . u"" 0), this does
not imply that dp / dt = 0 is the governing equation for density. Only if the dissipation H
can be neglected will that be true. That is a separate physical statement about the adia-
batic nature of the motion quite apart from the issue of compressibility. For a com-
pressible fluid, we would have, instead of dp / dt = 0, the statement ds / dt = O. For a
detailed discussion of the formulation of these equations, especially the thermody-
namics, see Batchelor (1967).
First Wave Example: Surface Gravity Waves
Perhaps the most familiar of waves in the ocean are the waves we see on the surface, either
from a ship or from the beach (or from the air). These are waves on the interface between
the water and the air (Fig. 3.1). The latter is so light compared with the former that we will
approximate the air as having zero density to eliminate any dynamical interaction with
the air to begin with. Theories of wave generation must include that coupling.
Consider a layer of liquid of uniform density and uniform depth. We suppose the
scale of the motion is small enough to be able to ignore the Earth's rotation and the
motion is small enough to be able to linearize this motion. In all such cases, we need
to ask ourselves whether these statements are sensible, and if so, for what range of
parameters? That is, if we ignore rotation, is there a limit, for example on the size of
the wave for which that is appropriate? We might already know, for example, that the
tides, which are a gravity wave response to the sun and the moon, do feel the effects of
the Earth's rotation, but, of course, they are of planetary scale.
1. Can we ignore rotation, friction and nonlinearity?

at
- To ignore rotation, compare a/ with Q ~ this implies that we need OJ» Q.
at
- To ignore friction, compare a/ with flk2,where k is a typical value of wave-
number ~ OJ» flk 2.
g~
~ ~ ~
~,o"'''0:L ~
Fig. 3.1.
The homogeneous layer of
fluid supporting surface
gravity waves
Lecture 3 . Equations of Motion; Surface Gravity Waves 21
- To ignore nonlinearity, compare a/at with respect to Ii . V ~ OJ»

uk or c » u ¢=:
this is the condition that the disturbance be wave-like, i.e., that the signal is
carried by the wave rather than the advective motion of the fluid.
2. Can we treat the fluid as incompressible?
Assume we can linearize. Suppose the motion is adiabatic. In general then, we
have
ds =0, s=s(p,p) (3.7)

dt
with the linearization
as
-=0=--+--
as ap as ap (3.8)
at ap at ap at
Thus,
~=_a~apap=(ap) ap (3.9)
at asjap at ap s at
From the theory of acoustics we know (or we can easily find out) that the speed of
sound in any medium is in fact given by the adiabatic compressibility of the medium.
That implies that if ca is the speed of sound in a fluid,
2
Ca -
_lap)
ap s
(One of the few scientific mistakes Newton made was to imagine that the speed of
sound was this derivative at constant temperature and not entropy).
So we have the estimate for the relation between a perturbation in the density and
the perturbation of the pressure:
bp=C~op (3.10)
We can, on the other hand, estimate the magnitude of the pressure fluctuation from
the horizontal momentum equation; if
Vp=o(p~~) then
bp=o(~OJ )
from which it follows from the relation between the pressure and density disturbances:
op =o( UOJ I (3.11)

p kc~)
Thus,
~ atSp
p at
=Jl ual)
c;k
We should compare this term, which is the estimate of the size of the local time
derivative in the mass conservation equation with a typical term in the remaining
combination of terms, namely, V . u= O(ku). Their ratio is thus
atSp
= 0(ual) al/k 2 2
-pdt - = - - =c- (3.12)
V·U kuk c; c;
Thus, as long as the phase speed of the wave is small compared to the speed of sound,
we can approximate the wave motion occurring as in an incompressible fluid for
which the equation for mass conservation reduces to the condition
V·u=o (3.13)
Note again that this does not by itself imply that dp / dt = O. A separate consider-
ation of the thermodynamics and the strength of the dissipation is required for that.
We now have a series of parameter tests we can make after the fact to check to see
whether the approximations of
l. linear motion
2. inviscid motion
3. incompressible motion
4. nonrotating dynamics
will be valid.
Assuming that these conditions will be met by the waves under consideration here,
the equations of motion reduce to the much simpler set:
ail
p-=-Vp-pgz
A
(3.14a)
at
V·u=O (3.14b)
z
where is a unit vector in the direction antiparallel to the direction of the local gravi-
tation.
We could have just waved our hands (perhaps appropriately for a course on waves)
and written down these traditional approximate equations. However, it is impor-
tant for each new investigation of a wave type to carefully consider a priori the condi-
tions required to achieve the approximate dynamics used for the physical description
of the wave to make sure that our physical system is no more complicated than
it need be, while at the same time, it should be consistent with the underlying physics
of the fluid.
The curl of our momentum equation (recall that we are considering a fluid of con-
stant density; the student is invited to use the thermodynamic equation to find the
condition for the validity of that approximation) yields
aVxil =0 (3.15)
at
So, if the vorticity is zero initially or at any instant (as it would be for an oscillatory
motion for which each field goes through zero periodically), it follows that it remains
zero for all time. If the curl of the velocity is zero, it follows from a fundamental fact
of vector calculus that the velocity can be represented by a velocity potential, l/J,
u=VI/> (3.16)
Note that only the spatial gradients of the velocity potential carry physical infor-
mation. Any arbitrary function of time can be added to I/> without changing the veloc-
ity field.
Since the motion is incompressible,
V· u=V· (Vl/J) =V21/> =0 (3.17)
The equation of motion within the fluid thus reduces to the elliptic problem gov-
erned by Laplace's equation:
(3.18)
This is an amazing simplification, and it should be a little disconcerting, because

we are looking to describe a wave motion. Laplace's equation, by itself, is certainly not
a wave equation. It describes among other things the electrical potential of static
charges as well as certain static gravitational fields but, alone, no dynamical wave
mechanism. The resolution of this seeming paradox is of course connected to the fact
that we have not yet considered the boundary conditions for our problem. There is no
more illuminating example of the importance of boundary conditions in the specifi-
cation of the problem than this case of surface gravity waves. All the dynamics are in
the boundary conditions. The internal equation, i.e., Laplace's equation merely relates
the horizontal and vertical structure of the motion field.
Boundary Conditions
The obvious boundary condition at the lower horizontal surface is that the normal
velocity vanishes there, i.e., w = 0 at z = -D, or
aI/> =0, z=-D (3.19)

az
The boundary conditions at the upper surface are significantly more interesting. Let's
call the departure of the free surface from its level "rest position" 1) (x,y, t) (Fig. 3.2),
z = 1J(X,y,t)
P = Pa(X,y,t)
Fig. 3.2.
A definition figure for variables
describing the motion in the Z = -0
o
surface gravity wave
which must be small (this will presently be made more explicit). Thus, we consider
the rippled free surface to only be slightly in departure from its rest state.
At the free surface, the physical boundary conditions are
1. the dynamic condition:
p(x,y,z,t) = Pa(x,y,t) Z = T/ (3.20)
and
2. the kinematic condition:
d¢ dT] dT]
w=-=-=>- (3.21a)
dZ dt dt
or
d¢ _ dT], Z=T]
a;- dt (3.21b)
We must now write these conditions completely in terms of the velocity potential, <jJ.
The linearized momentum equation is
dU = dV<jJ =_ Vp -gVz (3.22a)

at"- dt P
or
V{d¢ +..t+ gz}=o (3.22b)

dt p
The integral of the last equation implies that
d¢ +..t+gz=F(t) (3.23)
dt p
where F(t) is an arbitrary function only of time. We can always add a function that is
only of time to the velocity potential without changing the physical meaning of that
potential. Let's imagine that we have added such an additional term such that its de-
rivative with respect to time is equal to F(t). This allows us to write this linearized form
of Bernoulli's equation everywhere in the fluid in the form:
arjJ +..f+gz=o (3.24)

at p
Now let's apply this equation to the upper surface where z = 1](x,y,t) and P = Pa(x,y,t).
Thus,
arjJ + Pa + gTJ=o (3.25)

at p
A derivative of this equation with respect to time yields, using the kinematic con-
dition on the upper surface,
a2rjJ + arjJ =_~ aPa, at z=TJ (3.26)

at 2 g az p at
Note that each term in this boundary condition is linear.

The condition is applied at the unknown location z = 1]. Indeed, the position of the
free surface is, after all, one of the principal unknowns of the problem that we are try-
ing to predict. For the general nonlinear problem, this unknown location of the bound-
ary, at which the important boundary conditions are applied, is one of the most diffi-
cult aspects of the problem. However, we are considering only the linear small ampli-
tude problem, and it turns out that we can apply the boundary condition at the original
position of the interface, i.e., at z = 0. To see this take any term on the left-hand side of
the above boundary condition, generically call it G(x,y,1]) and expand it around 1] = 0.
Thus,
G(x,y,TJ)=G(x,y,O)+TJ aGI + higher order terms (3.27)

az z=o
The first term on the right-hand side of the above equation is of the order of the
amplitude of the motion, since G is one of the dynamic variables. Note that all the
dynamical variables are linear in the size of the amplitude of the motion.
The second term is of the order of G times the free surface height and is therefore
of the order of the amplitude squared. To be consistent with our linearization, such
quadratic terms must be neglected. That implies that each term in the boundary con-
dition stated above can be applied at z = 0; thus, we have
a2rjJ +arjJ =_~ aPa, at z=O (3.28)

at 2 g az p at
as the boundary condition on the upper surface, while again at the lower surface,
arjJ =0, z=-D (3.29)

az
Note that the upper boundary condition contains two time derivatives. This is the
mathematical source of the wave motion we will be describing. Its physical source is
the interplay between the gravitational force at the upper boundary providing a re-
storing force and the relation between the free surface elevation and the vertical ve-
locity at the upper surface.
We must also specify boundary conditions on the lateral boundaries. The simplest
problem we will consider will be that of a wave in an infinitely broad layer of fluid.
This is clearly an approximation, and we imagine that such a description will be valid
until the waves to be found propagate and interact with the inevitable lateral bound-
aries of the fluid. Until that time, we may provisionally just insist that the solutions
remain finite as x and y go to infinity. Useful references for formulation of the gravity
wave problem can be found in Kundu (1990), Lamb (1945) and Stoker (1957).
Plane Wave Solutions for Surface Gravity Waves: Free Waves (Pa = 0)
In Cartesian coordinates, Laplace's equation can be written as
a2¢ a2¢ a2¢

-+-+-=0 (3.30)
ax 2 ay2 az 2
We clearly can't have a three-dimensional plane wave because
1. the operator (the Laplacian) won't allow it, since an attempt to find such a plane
wave would lead to the condition
k; + k~ + k;=O (3.31)
which is impossible if all three components of the wave vector are real;
2. a plane wave won't satisfy the boundary condition al/JI az = 0 at z = -D;
3. the boundary conditions of finiteness as x and y get large imply that the horizontal
components of the wave number are real, and thus the vertical component, k3 or m,
must be purely imaginary.
We can find solutions, however, that are periodic in x, y, and t of the form
l/J = R(z)ei(kx+ly-rot) (3.32)
where again, the real part of the above equation is meant. Substitution into Laplace's
equation yields the ordinary differential equation for R(z),
d 2 R -K 2 R=0 (3.33a)
dz 2
K2 =k2 +12 (3.33b)
Note that all Laplace's equation will do is determine the structure with depth of
the plane wave solution in x and y.
The solution for R that satisfies the kinematic boundary condition at z = -D, i.e.,
that dR I dz = 0, is
R = A coshK(z + D) (3.34)
When this form is substituted into the boundary condition at z = 0, we obtain as a

condition for a nonzero solution for A
-arcosh(KD) + gKsinh(KD) =0 (3.35)
or
w=±.JgK tanhKD (3.36)
and
C= ~ =±(gD)1I2[ta~:Dr2 (3.37)
There are several important things to note about these results.
1. For each wave vector amplitude K, there are two waves propagating in opposite
directions, parallel and antiparallel to the wave vector. The frequency and phase
speed depend only on the wavelength, i.e., K and not on the orientation of the wave
vector.
2. The phase speed is different for different wavelengths in distinction to light waves
or sound waves. A pattern made out of a superposition of plane waves of different
wavelengths will have each component move at a different speed and hence the
pattern will disperse, which is why the relation between frequency and wave num-
ber is called the dispersion relation.
3. There are some important limiting cases to consider.
The maximum phase speed occurs when the wavelength (inverse to K) is long com-
pared to the depth, i.e., when KD « 1. Then the phase speed approaches (gD) 112 and is
independent of wavelength in that limit. In that case, when the phase speed is inde-
pendent of wavelength, the wave is called nondispersive. In that long wave limit or
for shallow water waves, OJ = K(gD)1I2.
On the other hand, when the wavelength is short compared with the depth, i.e., when
KD» 1, the dispersion relation becomes independent of depth and OJ= (gK)1I2, while
C= (g/ K)1I2. These deepwater waves are clearly dispersive. We will have to investi-
gate why the frequency and phase speed become independent of D in this limit.
Now that we have the phase speed, we can check our assumption of incompress-
ibility, that is, is C « ca' Since the maximum phase speed is given by the shallow water
limit for that condition, it will be satisfied if
.Jii5 «C a
or
D« c~/ g
For water, the sound speed is of the order of 1400 m S-I. That places a condition
on the depth such that for the incompressibility condition to be valid, we require
D « 200 km (pretty safe for oceanography, at least on Earth).
The nature of the dispersion relation is evident in Fig. 3.3 showing the frequency,
phase speed and group velocity as a function of wave number.
The dispersive nature of the waves can be contrasted to that of the standard "wave
equation" (which we will see in this course captures only a small fraction of wave phys-
ics in oceanography and meteorology). For light waves in a vacuum, sound waves and
waves on a thin string, the governing equation in one dimension is of the form
a2 rp _a2 a2rp =0 (3.38)

at 2 ax 2
whose general solution is known to be
q> = F(x + at) + G(x - at) (3.39)
consisting of two pulses traveling with the constant phase speeds ±a. The forms F
and G are determined by initial conditions after which the pulses travel without fur-
ill, C and cg versus kD

3.5ro----,-----,-----,-----,-----.----,-----,-----,----,-----,
3.0
2.5
2.0
1.5
1.0
c/(gD)1i2
........
0.5
------ ---------------
cg /(gD)·l ii ···········.· . ....... . ............. . . ......... .
0.00
2 3 4 5 6 7 8 9 10
kD
Fig. 3.3. Curves of frequency, phase speed and group velocity for surface gravity waves
ther change of shape. These are the classic nondispersive solutions for waves. In our
case, the waves are highly dispersive and the evolution of the wave shape with time
and unraveling the subsequent propagation of properties in the waves is a problem of
great subtlety and interest. It will eventually, as we might imagine from our earlier
discussion, come to depend on the character of the group velocity. For gravity waves,
with the dispersion relation quoted above
Cg ='Vj(m(K) = am K (3.40)
aK K
Thus, since the frequency is a function only of K, the group velocity is parallel to
the wave vector and hence parallel to the direction of phase propagation.
With or
= gK tanh KD,
am
2m-=g { tanhKD+ KD
2 } (3.41a)
aK cosh KD
or
2ccg =.K.{tanhKD+
K
K~
cosh KD
} (3.41b)
and
Cg 1{ 2KD} (3.41c)
~=2 1+ sin2KD
Thus the group velocity coincides with the phase speed for long waves (KD« 1),
while for short waves the group velocity is 112 the phase speed (see Fig. 3.4).
Wave packet
,/ ,
/ '\ • cg
/ \
/,
/,
I /I :\
, •C
Fig. 3.4.
"\ /
A wave packet propagating /
\
with the group velocity carries
a plane wave with crest moving , /
with the phase speed
0.9
0.8
-C 0.7
::.::
0
S 0.6
'';:::
v
c:
.2 0.5
n:I
\II
n:I
."
v 0.4
OJ
0.2
01~
'0 2 3 4 5 6 7 8 9 10
KD
1.0
0.9
0.8
\II
<1/
>
~ 0.7
~
.s;
~ 0.6
til
<1/
v
~ 0.5
~
\II
.Ev 0.4
.....
v'"
o 0.3
'';:::
~
0.2
0.1
OO~
. 0 2 3 4 5 6 7 8 9 10
KD
Fig. 3.5. a The group velocity as a function of wave number. b The ratio of the group velocity to
phase speed for surface gravity waves as a function of wave number scaled with fluid depth, i.e., KD
Suppose we have a wave packet carrying a short wave, KD « l.

The amplitude and K will move with the group velocity, while individual crests will
move with the phase speed. Since cg is half the phase speed for short waves, we will
see individual crests appearing at the rear of the packet and travelling through the
moving packet to disappear at the leading edge of the packet. Where do the crests go?
Well, they are only a feature of the pattern, and they appear and disappear like smiles.
It is the wave envelope moving with the group velocity that has physical content.
The ratio of the group velocity to the phase speed is shown in Fig. 3.5b as a func-
tion of wave number. They are equal for the longest waves, while for short waves the
group velocity is half the phase speed.
Lecture 4
Fields of Motion in Gravity Waves and Energy
Now that we have the dispersion relation, i.e., the dependence of frequency on wave
number (we define the magnitude, K, of the wave vector K to be the wave number),
we can ask what the fluid motion is in the wave field.
Our plane wave solution has been written in the form:
I/J = Aei(K.x--(j)t)coshK(z+D) (4.1)
Using the boundary condition at z = 0,
a¢
g17=--z=O (4.2)
at
since Pa has been taken to be zero for these free waves. We can therefore calculate the
free surface elevation from Eq. 4.1 and Eq. 4.2,
17= ( -iW]A ei(K·x-mt) cos hKD

g
and it is understood the real part of each expression is to be taken.

A is an arbitrary amplitude, and it will be useful to consider the amplitude of
the disturbance in terms of the amplitude of the free surface perturbation. So, let's
define
g
170 = I.(AW] coshKD
and take it to be real (this only defines the zero of the spatial phase, the point where
the free surface elevation is a maximum). This yields
1] =1]ocos(K·x-wt) ( 4.3a)
W coshK(z+ D) sin(K .x-wt) (4.3b)

¢=17o K sinhKD
From the velocity potential, we can calculate each velocity component, since u= VifJ.

34 Lecture 4 . Fields of Motion in Gravity Waves and Energy
From the above formula for ljJ, we calculate the horizontal velocity vector and the
vertical component of velocity:
-
uH =170 ~KJ -
K cos(K·x-OJt) cos~K(z+D)
smhKD
(4.4a)
W =1]oaJsin(K .x-OJt) sinhK(z+ D) d1] sinhK(z+D)

(4.4b)
sinhKD dt sinhKD
From the Bernoulli equation
d¢
p=-pgz-P-at
we can calculate the pressure field in the wave. (Note that part of the pressure field has
nothing to do with the wave. That is the first term on the right-hand side; it is present even
in the absence of the disturbance). From the result from the velocity potential we obtain
p =-pgz+PgTJo cos(K .x-rot) coshK(z+ D)

coshKD
(4.5)
_ [COShK(Z+D) ]
-pg 1] coshKD z
There are some very important qualitative features to note before moving on.
1. The horizontal velocity, uH ' is in the direction of the wave vector and hence in the
direction of the propagation of the wave. This is not surprising for anyone who has
lolled in the surf and felt himself move back and forth in the direction of a wave as
it has passed by;
2. Each perturbation variable is proportional to the amplitude of the free surface el-
evation. That is, in this linear problem, the amplitude of every aspect of the motion
is proportional to the free surface elevation. This implies that products of any two
motion variables must be quadratic in the surface elevation (this is what we used to
linearize the surface boundary condition);
3. In the limit of deep water or equivalently short waves for which KD» 1, the as-
ymptotic forms of the hyperbolic functions imply that
eK(z+D)
coshK(z+D)
=e Kz z<O (4.6a)
sinhKD e KD
eK(z+D)
sinhK(z+D)
e Kz (4.6b)
sinhKD e KD
so that all the dynamical variables decrease exponentially from the free surface.
The scale of decrease, as imposed by Laplace's equation, is just the wavelength.
Hence, for waves whose wavelength is short compared to the depth, the motion
Lecture 4 . Fields of Motion in Gravity Waves and Energy 35
decays long before the bottom is reached. The wave field then does not sense the
presence of the bottom. This is why the frequency/wave number relation becomes
independent of D as KD gets large. It's a good rule of thumb to remember for grav-
ity waves; the depth of influence of the wave is its wavelength.
4. For very long waves, or equivalently, for shallow water, such that KD ~ 0, the
limiting form of the hyperbolic functions yields
KD« 1
_ =(ry)
UH D cK
K (4.7a)
w=ary(Z+D) (4.7b)
at
p=pg(ry-z) (4.7c)
In this limit, the horizontal velocity is independent of depth. Its magnitude is the ratio
of the free surface elevation to the depth multiplied by the phase speed. Thus, as long as
1] / D« 1, it will follow that uHf c« 1, which is the condition for linearization. The verti-
cal velocity is proportional to the rate of displacement of the free surface, linearly dimin-
ishing to zero at the bottom, and the pressure field is in hydrostatic balance in this limit.
It is left to the student to show that in the short wave limit, KD « 1, the condition
for linearization, u « c, leads directly to the condition 1]oK « 1. That is, the free sur-
face displacement divided by the wavelength must be small, i.e. the slope of the free
surface must be small.
Energy and Energy Propagation
The kinetic energy in a gravity wave per unit volume is simply

pl ul2
2
where the magnitude of the velocity is denoted by the vertical bars. Integrated over
depth we have the kinetic energy per unit horizontal area,
o 1_1 2
KE= f dz Pu2 (4.8)
-D
The potential energy per unit horizontal area is
Tf
PE = fpgzdz = pg (ry2 - D2) (4.9)
-D 2
Note that the term proportional to D2 in the PE is an irrelevant constant. Note, too,
that we have integrated to the free surface elevation 1] in the expression for PE but only
to z = 0 in the expression for KE. The reason for this is that we are calculating the en-
ergy to the second order in the wave amplitude, and to do this for the PE we must in-
clude the free surface displacement. If we were to extend the integral for KE to include 1)
in the upper limit, the correction to the expression for KE would be of O(u2 1), i.e., of
third order in the small wave amplitude and hence negligible. So the above integrals
as stated are each of order amplitude squared.
Now let's try to develop an equation for the propagation of wave energy. We start
from the governing equation, which is Laplace's equation for the velocity potential.
An excellent discussion of wave energy and its propagation can also be found in Kundu
(1990) and Stoker (1957).
We multiply that equation by the time derivative of the potential, viz.
0= ar/JV 2r/J=v.(ar/JVr/J)-Vr/J. aVr/J

at at at
(4.10)
=_~[(V r/J)2]+ v .(ar/J v r/J)
at 2 at
We recognize that the first term is (minus) the rate of change of kinetic energy per
unit volume.
Let's now integrate the above equation over the depth of the fluid.
a 0fdz(Vr/J)2/2+ 0fVH· ( vr/J-

-- ar/J)dz+ f -a (ar/J
0
- -
ar/J)dz=o (4.11)
at -D -D at _Daz az at
Here the symbol VH is the portion of the divergence in the horizontal plane, i.e.,
VH .Q= aQx + aQy

ax ay
where Q is any vector.
The last term in the equation above can be integrated, and the resulting terms evalu-
ated using the boundary conditions. Since aep/ az = 0 at z = -D, while
ar/J a" (4.12a)

az at
ar/J
-at =-g" , z=O (4.12b)
we obtain
a
- - 0fdz(Vr/J) 2 /2+ 0
fV w (ar/J) a"
Vr/J- dz-g,,-=O or (4.13a)
at -D -D at at
a
-[KE+PE]+VH·g=o - (4.13b)
at
0 ar/J
g = - f - V Hf/Jdz
_
(4.13c)
-D at
Lecture 4 . Fields of Motion in Gravity Waves and Energy 37
That is, the rate of change, locally, of the total energy per unit horizontal area is bal-
anced by the horizontal divergence of the fluX of wave energy, -G, a horizontal vector.
This horizontal flux can be easily interpreted physically, since
- a¢VH¢=(P+pgZ)UH (4.14)
at
and (p + pgz) = p', which is the part of the pressure field due to the wave activity. There-
fore, the energy flux vector is just the rate at which the pressure field in the wave is
doing work on the surrounding fluid. That rate of work yields the energy transfer from
one part of the fluid to another and hence the energy flux. We shall often be looking
for energy balance equations of the above type, Le.,
a = sources + d
-aE + nv . ..J ···
IsslpatlOn
at
that is, the rate of change of wave energy locally and its flux to other parts of the fluid
balanced by sources and sinks of energy. In the present case of a free, inviscid gravity
wave, both the sources and sinks are zero.
An interesting question arises here. If, as we believe, the important physical at-
tributes in the wave field propagate with the group velocity, can we relate the energy
flux vector to the group velocity?
First, let us calculate the kinetic and potential energy in the field of motion of the
plane wave we have been discussing. To make life easier for ourselves (always a good
idea) let us orient our x-axis to coincide with the direction of the wave vector. Then,
since the horizontal velocity is in the direction of the wave vector as shown above, there
will be only the x-component of the horizontal velocity to deal with along, of course,
with w. In this coordinate frame, K = k.
The potential energy is easy to calculate:
2
PE= pg1lo cos 2(kx-wt) (4.15)
2
This form oscillates between its maximum and zero during a wave period. The signifi-
cant quantity for our purposes is the average over a wave period, denoted by brackets, i.e.,
(PE) = pg1Jg
!
(4.17)
4
For KE we have
o 0
cos 2(kx-wt) cosh 2 k(z+D)
KE= Jdzp(u 2+w2)/2= Jp1Jgoi •
sinh2 kD
• 2 z/2 (4.18a)
-D -D +sm 2 (kx-wt) smh k(z+D)
sinh2 kD
and
( KE) = plJ2 oi ° -D°f cosh.2k(2z + D) dz-plJoO)

4smh kD
_ 2 2 sinh 2kD
----
8ksinh 2 kD
(4.18b)
= PlJ~gktanhkD sinh kDcosh kD = pglJ~
4sinh 2 kD 4
In deriving this result, we have first used the averaging of the cosine and sine terms
over a wave period, then the identity relating the square of the cosh and sinh terms to
cosh of twice the argument and then finally the dispersion relation itself to write 0l-
in terms of the wave number.
We note the important fact that averaged over a wave period (or a wavelength if we
were to average in x instead of t), the kinetic and potential energies are equal; that is,
there is equipartition of energy in the wave field between potential and kinetic en-
ergy exactly as in the oscillation of a pendulum.
The total energy averaged over a period is
(E)=(KE)+(PE)= pglJ~ (4.19)

2
Now let's calculate the energy flux vector in the x-direction and its average over a
period.
~x=-P
°J--dz=
afjJ afjJ
J 0 20)3
PlJo 2 cos 2(kx-0Jt)cosh 2 k(z+D)dz
-D at ax -D ksinh kD
(~
x
)= PlJ~0)3
2ksinh 2 kD -D 2
f[.!..+ cosh2k(z+ D)"L~
2 J
= P1]~ 0)3 [D + sinh 2kD ]
2ksinh 2 kD 2 4k
= PlJ~ ckD gtan h kD[-

1+sinh
-- -]
2kD
2sinh2 kD 2 4kD
= pglJ~ck(DI2) [.!..+ sinh2kD] (notethat2sinhkDcoshkD=sinh2kD)

sinhkDcoshkD 2 4kD
1 kD]
= pg1]~c [ 4+ 2sinh2kD
pglJ~
=-2- C
[1"2+ kD]
sinh2kD
2
= pglJo C
2 g
=(CgE) (4.20)
Lecture 4 . Fields of Motion in Gravity Waves and Energ·y 39
The important result obtained here is that for a plane gravity wave, the horizontal
flux of energy is equal to the energy itself multiplied by the group velocity. That
is equivalent to saying that the energy in the wave propagates with the group velocity.
That is, the energy equation may be written:
a(E) + VH .cg(E) =0 (4.21)

at
In a uniform medium where the frequency and wave number are essentially con-
stant, the group velocity will be independent of position. Thus for a wave packet, whose
averaged energy just depends on the distribution of its envelope of free surface height,
the above equation can be rewritten
-+c-g .V(E)=O
a(E) (4.22)
at
which states that for an observer moving laterally with the group velocity, the energy
averaged over one phase of the wave is constant. The energy in a slowly varying
packet travels with the group velocity in a homogeneous medium.
We will generalize this result to cases in which the energy is not simply contained
in a compact packet, and we will see that the generalization also allows us to think of
sequences of energy packets, each propagating with a group velocity appropriate for
the wave number of that particular packet, which together with its companions rep-
resents an arbitrary disturbance.
Addendum to Lecture
With the velocity field given by the velocity potential, we can calculate the trajecto-
ries of fluid elements in the plane wave. Let ~ and Sbe the x and z displacements of
the fluid elements around some original position (xo,zo). Then if the displacements
are small, we can linearize the Lagrangian trajectory equations:
dq =u(xo +q,zo + (,t) '" u(xo,zo,t) =w170 cos(kX-M) coshk(z + D)

dt (4.23a)
sinhkD
and similarly
d( w(xo,zo,t) =W170 sin(kx-M) sinhk(z + D)

dt= (4.23b)
sinhkD
Integration yields
q= -170 sin(kx-rut) coshk(z + D) (4.24a)

sinhkD
(= 170 cos(kx-rut) sinhk(z + D) (4.24b)

sinhkD
It follows that the trajectories are ellipses, i.e.,
~2
-+-=1
;;2 (4.25a)
L2 L2
x z
Lx =" coshk(z+D) (4.25b)

'/0 .
smhkD
L -11 sinhk(z+D)
z -'/0 (4.25c)
sinhkD
Thus, the orbits are flat at the bottom of the fluid layer where Lz = O. For deep wa-
ter, the two axes of the ellipse are equal (1] oekz ), so the orbits are circularly shrinking
in radius as z becomes more negative. For shallow water, the orbits reduce to essen-
tially horizontal lines parallel to the bottom. The student is asked to discuss the direc-
tion of motion along the ellipse as the wave passes overhead.
Lecture 5
The Initial Value Problem
It is not easy to see how a uniform or nearly uniform wave train can realistically emerge
from some general initial condition or from a realistic forcing unless the initial condition
or the forcing is periodic. That turns out not to be the case, and the ideas we have so far
developed about group velocity and energy propagation turn out to be invaluable in get-
ting to the heart of the general question of wave signal propagation. Indeed, it is the very
dispersive nature of the wave physics (i.e., the dependence of the phase speed on the wave
number) that is responsible for the emergence oflocally nearly periodic solutions. This
can be seen by examining the solution to the general initial value problem. This was first
done by Cauchy in 1816. It was also solved at the same time by Poisson. The problem was
considered so difficult at that time that the solution was in response to a prize offering of
the Paris Academie (French Academy of Sciences). Now it is a classroom exercise.
We will again consider a disturbance that is a function only of x and z (and t of course),
and we will consider the problem unforced by a surface pressure term, i.e.,Pa = 0.
The layer is again of depth D and it is initially at rest.
As initial conditions, we will take
1](x,O) = N(x) (S.la)
u(x,z,O) = ° ::} ljJ(x,z,O) = ° (S.lb)
The governing equation for the velocity potential is Laplace's equation, which for
two dimensions is
a2A1 a2rp_
'1'+--0 (5.2)
ax al
2
with boundary conditions:
arp =0, z=-D (S.3a)

az
W= arp - all
arp az -at
1 a2
z=o::::}---.t+ arp_ (S.3b,c)
at+gll=o at 2 g az -0, z=O

42 Lecture 5 . The Initial Value Problem
Since the region is infinitely long in the x-direction (in our approximation of a broad
swath of open water) and the coefficients of the differential equations and boundary
conditions are independent of x, it is appropriate and useful to represent the solution
as a Fourier Integral. You may want to brush up on the Fourier integral by looking at
anyone of number of standard mathematical texts, e.g. Morse and Feshbach (1953).
Thus, we write the velocity potential as
1 =
¢(x,z,t)= r;;- fiP(k,z,t)eikxdk (5.4a)
,,2n _=
with the dual return relation:
iP(k,z,t)= ~ J¢(x,z,t)e-ikXdx (5.4b)

,,2:rr_=
Note that the placement of the factors ..J2ii is somewhat arbitrary, and different con-
ventions are used. The only requirement is that the product of the constant before each
integral multiplies to 1/21t.
Similarly for the free surface elevation,
1 =
17(x,t) = r:::= fN(k,t)eikXdk (5.5a)
,,2:rr_=
1 = .
N(k,t)=- f17(k,t)e- 1kxdk (5.5b)
fiir_=
What we are doing is representing an arbitrary disturbance by an infinite sum of
plane waves in x, whose wave numbers are a continuous distribution over all k, which
is why an integral is required for the representation.
If the above representation for the potential is put into Laplace's equation, we ob-
tain as a condition for the solution that at each wave number k,
a2iP _k 2 iP = 0 (5.6)
az 2
while the boundary conditions become
a2iP aiP
-+g-=O, z=O (5.7a)
at 2 az
aiP =0, z=-D (5.7b)
az
and the similarity to the plane wave problem should be apparent. Indeed, the solution
for c[J can be written:
c[J(k,z,t) = A(k,t)coshk(z + D) I sinhkD (5.8)

Lecture 5 . The Initial Value Problem 43
This satisfies the boundary condition at z = -D. Satisfying the boundary condition
on z = 0 requires
d 2A+m(k)2 A =O (5.9)
dt 2
where
m(k)2 =gk tanh kD (5.10)
Thus, we can write
A(t) = a(k)eiW(k)t +b(k)e-iw(k)t (5.11)
so that
cP = (aeiOJt + be -iOJt) cosh k(z + D) (5.12a)

sinhkD
::::}
1
ifJ= .J2;; =~
fL(ae iOJt +be-iOJt )Je;kx coshk(z+D) (5.12b)
_= sinhkD dk
The solution for the velocity potential consists of a sum of waves. For each k, one is
moving to the left (the first term in square brackets) and the other is moving to the
right (the second term). Each one is moving with the frequency associated with the plane
wave at that k and with the vertical structure function of the plane wave at that k. The
total solution is the integral sum of all the plane waves excited by the initial conditions.
Since
aifJ)
T/--- (5.13a)
g - atz=o
T/(x,t)=- ~1[aeiOJt _be- iOJt ]imeikxcothkDdk (5.13b)

gv 21t -=
at t = 0 the velocity potential and its derivatives vanish. Thus for all k,
a(k) = -b(k) (5.14)
and using the dispersion relation oJ =gk tanh kD,

T/ =_1_ =f ib(k)kdk r .
.J2;; _= m Le'OJt + e-iOJt ] (5.15)
at t = 0
1 =
T/=T/o(x)= ~ fNo(k)eikxdk (5.16)
v 21t -=
which implies that
b(k}= No(k}m (5.17)

2ik
or
1 = N
J2; f -2-
17(x,t}=-- o(k}rLe iOJt +e-'OJt
. ]e,kxdk
. (5.18)
-=
which has a simple interpretation, namely, that half of the initial condition at each k
propagates to the left and the other half propagates to the right, each with the phase
speed, frequency, and wave number relation of the plane wave of that k. We might have
written down the above equation directly from our knowledge of the plane wave phys-
ics, but it is useful to go through the formal derivation at least once. Incidentally, now
that b(k} is known,
1
¢=i- -1 No kOJt _e-iOJt ]eikX coshk(z+D} (5.19)
J2; _= 2 sinh kD
This yields the formal solution to the problem, but it doesn't take much to realize
that a solution written as an infinite integral is not very revealing, and our real work
in understanding the physical nature of the initial value problem has just begun.
But first, to simplify things, let's assume that the initial condition on the free sur-
face height is an even function of x around the origin, namely, 1]o(x} = 17o( -x}. It fol-
lows from this that the Fourier transform of the initial condition, No(k} is an even func-
tion of k. To show this,
No(k}= ~ le- ikx 17(x)dx letx=-;, then (5.20a)

,,2n _=
No(k} = ~ leik~17(-;)d; = ~ le-i(-k)~17(;)d;

,,2n _= ,,2n _=
= No(-k} (5.20b)
where in the last step we have used the evenness of 1](x}. Since No(k} is an even func-
tion of k,
1 =
17= J2; _[NO (k}cos mteikxdk
1 =
= J2; _[No(k}cosmtcoskxdk
=
=·hln fNo(k}cosmtcoskxdk (5.21)
o
Thus, we have succeeded in reducing the interval to the range (0,00) in our k integra-
tion. Using a well-known identity for the product of cosine functions,
7]= ~- JNo(k)[cos(kx+OJt)+cos(kx-OJt)]dk
=
1t o
=Re IIINo(k)k(kx+IVt) +ei(kx-IVt} ]dk (5.22)

V--;o
where again we recall that (J) = (gktanhkD)ll2 (here we can take the positive root since
we have explicitly included both signs of the solution in the above formulae).
At this point, we are still in the position of having our solution given in terms of an
infinite integral. What can we say about the solution? Will some useful approximation
teach us anything?
For short times, i.e., for a very small t, we could expand the expression for TJ as a power
series in t, the first term of which is the known initial condition. That can allow us to ex-
amine the initial evolution of the disturbance. We might rightly object, saying that the ini-
tial evolution will depend very heavily on the arbitrary form of the initial condition. It will
be much more illuminating to ask about the solution after a long time has passed so that
the wave field can evolve to a state that reflects the general properties of the gravity wave
field. Can we say something more useful, then? It turns out we can, using a classical method
of approximating integrals of the type we have above: the method of stationary phase.
Our integrals for the free surface height are of the form
7]= IIINo(k)ktX(k)+eit\f/(k)]dk (5.23a)

V--; 0 2
X(k) =k(x It) + w(k)t (5.23b)
lfI(k) = k(x It) - w(k)t (5.23c)
and we would like to evaluate the integrals above for a large t and with the ratio x / t
fixed. This is equivalent to saying that for a large t, we are evaluating the integrals
moving away from the origin at the speed (arbitrary) U = x I t. So, for a large t, an
arbitrary x should be chosen, which is also large. That determines U = x I t, and we want
to find the value of the integral at that time and at that point.
The disturbance for x > 0 will be given by the second term in the above integral, so
consider the second integral in the equation for TJ. Suppose that the function lfI(k) does
not vanish on the semi-infinite k interval. Then we could change the dependent vari-
able of the integral from k to lfI, and obtain
[2= No eit\f/dlf/
7]=V--;! 2(dlf//dk) (5.24)
Integration by parts yields
No(k)
7_]1- - - - e it\f/ 1= - -1 =Jeil'/fd I fd/ - -
N
- (5.25)
itdlf//dk 0 ito dlf/dlf//dk
so that the disturbance would decay at least as fast as 1 / t (in fact it will decrease much
more rapidly, exponentially. See Lighthill, Waves in Fluids (1978) Jeffreys and Jeffreys
(1962) or Stoker (1957).
This rapid decay with time is due to the fact that while No(k) is a smooth function
of k, the sinusoidal behavior of the exponential produces a factor that oscillates very
rapidly when t is large as a function of k, so that contributions to the integral from
some interval in k are cancelled at k + L'1k by a factor of the opposite sign, as shown in
Fig. 5.1.
Thus, as long as ljI(k) increases smoothly with k, the factor eitlP(k) will oscillate very
rapidly as a function of k for a large t, unless in the neighborhood of some point k s' the
function ljf(k) does not increase with k, i.e., unless that point is a stationary point at
which
dvr (ks)=O (5.26)

dk
At such points, the phase function 1jf will not increase with k, and there is an op-
portunity for the integral to accumulate value in that neighborhood.
2.0 - -- - r - - - - - - - - r - - - - - r - - - - - - - . - - - - - . - - - - - - - - - ,
--- ------------------\----- ---- \

r-,
I- - -
1.5
No(k)
1.0
Re exp(itljf(k))
0.5
111111111
0.0 I ....·
- 0.5 ....
_,111.11 .. H V ,
-3 -2 -1
';=k' , V ~ Lilli!
o 2 3
Fig. 5.1. The behavior of the exponential factor for a large t showing the interval of stationary phase
To find such points of stationary phase:
1jI=kU-m(k), U=x/t (5.27a)
dV =O=U-dm/dk=x/t-cg(k) (5.27b)
dk
Thus at a given x and t, or for an observer moving away from the origin at a speed x / t,
the wave number of stationary phase, ks is given by that wave number whose group
velocity matches the velocity U = x / t (Fig. 5.2).
We note that for a given x / t, a stationary phase wave number can be found as
long as x / t is less than the maximum value of cg in the whole k interval. Since the
maximum value of the group velocity occurs for the longest wave and this maxi-
mum is --.JgfJ, we anticipate that for time t, the disturbance will be limited to a region
x:O:; t--.JgfJ. Thus, there will be a front moving out from the origin at the speed -1iD,
ahead of which the fluid will be essentially undisturbed and behind which the solu-
tion will be given by the asymptotic approximation to the integral we will now de-
velop (Fig. 5.3).
(g as a function of kD
1.0
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1 0
2 3 4 5 6 7 8 9 10
kD
Fig. 5.2. The curve of group velocity versus KD. The point of stationary phase corresponds to x / t = cg
Front
D;""""nc. Urn;,.d '0 ,h;, '09;on J

Fig. 5.3. x=Q
The interval for which the
disturbance can be found
for a large t x=.J9Dt
Consider the integral:
=
1= Jeitlf/ No(k)
- dk (5.28)
o 2
As we have argued, for a large t, the major contribution from this integral comes
from the interval in k near the stationary point ks' Near that point, we can write
dlfF
IIr(k)=lIr(k )+ - d 2 1fF
( k ) ·(k-k )+-(k )·(k-k)2 + ... (5.29)
'r 'r s dk s s dk2 s s
'-v-'
= Oby defn.
Thus the integral can be approximated as
I"" Jeit/(/(ks)eitlf/'(ks)(k-ksh2No(ks)dk/2
o (5.30)
IfF" == d 21fF/ de
Note that we have replaced No(k) by its value at the stationary point. This is valid
since only in this vicinity will the integral have an asymptotic value greater than 1 / t
and No is assumed to be a smooth function of k and hence much more slowly varying
than tlfl for a large t.
Thus,
I ""--e
No(ks) it/(/(ks) =Jeitlf/'(ks)(k-ks)2/2dk (5.31)
2 0
where the integral really extends over a region centered on ks'

Let
11fF"(ks)1 (k - ks )2 t = Ii (5.33a)
or
112
k-k _ { 2 )
(5.33b)
s- 11fF"(ks )It
This allows the integral to be written:
No(k s ) eiW(ks ) =J i/92sgnvr'(ks)dl?

(5.34)
I"" 112 e
2 {tl\VW;ks)l} _=
where the extension of the limits to plus and minus infinity follows from the relation
between k and () for a large t. The remaining integral is a standard one and can be found
in almost all integral tables:
le i /92 sgnvr'(ks)d l?= .,Jiiei (1t/4)sgn vr'(ks) (5.35)
Putting these results together leads us to our final formula for the asymptotic solu-
tion for the initial value problem for x > 0 and for a large t:
No(ks ) i(ljV(k, )+[1t/4)sgnY(k.»

,,=: 1vr"(ks )tjI/2 e (5.36a)
x
\V=-k-w(k) (5.36b)
t
Discussion
Now let's try to interpret the solution, valid for a large x and t, shown in the boxed
equation above.
We can think of the solution in the vicinity of the point (x,t) as a plane wave with
amplitude:
No(ks )
(5.37)
A= ~wW(ks)
and a phase
8(x,t)=W= ksx-w(ks)t (5.38)
Notice that since the wave number ks is a function of x and t through the station-
ary phase condition
dw (ks)=xlt
dk
the dependence of the phase of x and t can be rather complicated.

However, consider our generalized definition of wave number:
ae =k +x aks _ aw t aks
ax s ax ak ax
=ks + aks [x-cg(ks)d (5.39)
ax
Hence, the local variation of phase in x is equal to ks for x It = cg(ks), i.e., for an ob-
server moving away from the origin of the disturbance with the group velocity asso-
ciated with that wave number. Further, moving at that constant speed, the wave number
remains constant if the observer moves with that group velocity. Similarly,
ae
--=wCk aks- [x-c (k )t ]
)-
at s at g s
(5.40)
so that the frequency will be equal to wCks) for an observer moving at the group speed
at the stationary wave number.
Thus, at some point in the wave train, the disturbance will look like a plane wave
with the wave number and frequency (ws' ks) related by the dispersion relation, and these
local parameters will remain constant as the point moves away from the origin with
the group speed. In other words, the wave number and frequency propagate with the
group speed even though the original disturbance need not be anything close to a peri-
odic form. This is a result valid for a large t. What has happened is that the spectrum of
the disturbance sorts itself out wave number by wave number such that the part of the
disturbance with wave number ks propagates with its group velocity to the position
x = cgt. This happens for each k. The part of the spectrum with the fastest group veloc-
ity (the long waves in this case) will be found out in front and the slow waves will bring
up the rear. This explains why, although the initial disturbance may be quite different
from a plane wave (e.g., a gaussian in x), the disturbance with time can be locally ap-
proximated by a plane wave, justifying our earlier concentration on the properties of
plane and nearly-plane waves. It is the dependence of the phase speed and group speed
on k that disperses the original signal into a parade oflocal plane wave perturbations.
At any fixed x, the wave number will change with time as slower, shorter waves ar-
rive at that x.
Again, let's consider the phase
e = tlf'= ksx-wCks)t = kst[~- W(ks)] = kAx It -c(ks)] (5.41)

t ks
Thus, if we move in such a way as to keep the wave number constant, x It = cg(ks) *- c(ks)'
then the phase will change for the observer. Such an observer will see individual crests
and troughs moving past at a rate that depends on the difference between the group
and phase velocities. If, on the other hand, one wishes to follow an individual crest, so
that we set xl t = c(k), then from our above results,
k= ae =k + aks t[c-c ] (5.42)

ax s ax g
so that following an individual crest implies that the wavelength associated with that
crest will be changing with time. In this physics, you can always ride the same horse if
you want, but it may be repeatedly changing size if you insist on staying on the same
horse. In order to ride a horse that is always the same size, you will need to constantly
change horses (crests).
We can work this out analytically and explicitly in the limit when the water is so
deep that for all wave numbers possessing any reasonable amount of energy in the
spectrum, kD » 1. In this limit,
W= (gk)1I2 (5.43a)
cg =~(fr2 =~ (5.43b)
1.0
0.5
0.0
-0.5
lJ \1 \/ ..............
100 150 200 250 300

a Free surface as function of x, t = 10
1.0
0.5
0.0
-0.5
-1.0
0 2 4 6 8 10 12 14 16 18 20
b Free surface as function of t,x = 1OOg = 10m S-2
Fig. S.4. The upper panel (a) shows the free surface at a fixed time. Note the long waves out in front.
The bottom panel (b) shows the surface height field as a function of time at a fixed point. The low
frequency waves (small k) arrive first and the higher frequencies arrive later, since they have slower
group velocities
To find what the wave number of stationary phase is at the point x at time t,
1/2
c =xlt=~ ( ~ ) (S.44a)
g 2 ks
1 gt 2 _ 1
k =--~OJ(k )--gt/x (S.44b)
s 4 x2 s 2
(}(ks)=ksx-(gksY/2t=_~gt2 Ix (S.44c)
4
c(ks ) = 2x It (S.44d)
Note that at a fixed position, the wave number increases (waves get shorter) with
time, while the waves with the slower group velocities arrive later. At any given time,
the waves get longer (k gets smaller) as x increases. Note that the phase at any x and t
will change with time according to the ratio gf I x. To ride a particular crest, an ob-
server must then move so that x = t2, that is, the observer must accelerate with time to
keep up with a particular phase. To follow a particular wave number, the observer must
move at a constant speed equal to the group velocity for that wave number. Hence, for
dispersive waves, one can not simultaneously keep to the same phase and the same
wavelength, since the phase speeds and group velocities are not the same (Fig. 5.4).
Lecture 6
Discussion of Initial Value Problem (Continued)
We have seen that the initial spectrum of the waves, which is initially localized
in space, gets strung out with time so that at time t, each wave number appears at
x = clk)t. We might expect that the energy, if conserved, would also be distributed
by wave number, so that the amount of energy at wave number k in the original
spectrum at wave number k would also be found at the position x = cg(k)t for a large
enough time. This is as if the original disturbance is composed of an infinite number
of packets of constant wave number, each of which moves away from the origin of
the disturbance with its own group velocity. Each satchel of energy moves with
the group velocity (Fig. 6.1).
Let's try to make this more quantitative, and we will at the same time be able to
explain the inverse dependence of the amplitude on the square root of time found
in the last lecture. The energy in the gravity wave field is, as we have seen, propor-
tional to the square of the free surface displacement. By a fundamental theorem of
Fourier analysis,
E = pg f,72dx = pg ]N(k)1 2dk (6.1)

2_~ 2_~
which only states that we can count the energy in space or with the wave number spec-
trum.
Now, following an argument due originally to Rossby (1945) consider the energy
in a spatial interval between Xs and Xs + Ax s such that the center of this infinitesimal
interval is the place where the wave number ks is found at time t.
Again, x = cg(k.)t.
tcg (kl ) tcg (k2)
/'1 ' I
'-----~ L0-- k2
~ X
Fig. 6.1. A disturbance initiated by an initial source of energy, S, propagates away and is distributed
among "suitcases" of energy, each moving with its group velocity

54 Lecture 6 . Discussion of Initial Value Problem (Continued)
The x interval ~s will, for long times, be related to a wave number interval in the
original spectrum by the relation
~s
- taC-g
- J _
I1ks - ta
-co
-
2
2 I1ks (6.2)
ak k-k
- s
ak
The energy in that spatial interval from our asymptotic formula for the wave height
will be (assuming No is an even and real function)
2 N o(k s )2 2 k· "
1] ~= 11? cos (8( s)+lsgn(co (ks)7t/4)~s (6.3)
[tl~:~ (ks)l]
Averaging over a period and using the above expression for the interval length,
1]2~s = N o(ks )2 I1ks (half goes the other direction) (6.4)

2
The above expression is a function only of ks and so will remain constant for an
observer moving at the group velocity. Thus, the energy in the original spectrum in
the wave number interval 11k is conserved as it propagates outward with the group
velocity. The length of the interval that energy is contained in continuously and lin-
early extends with time, because the group velocity is slightly different at the leading
and trailing edges of the interval, since k is a continuous function of x at a given time.
In order to have the energy conserved, the product of the amplitude squared times
the interval length must be constant. Since the latter increase linearly with t, the am-
plitude must decrease like e 1l2 to conserve energy. This explains the square root fac-
tor in the result of the previous lecture. Note that the contribution to the wave ampli-
tude for those parts of the integral not near the stationary phase point will decline at
least as fast as 11 t. Then as time goes on, the stationary phase contribution will be-
come increasingly dominant.
In the sense described above, the energy propagates with the group velocity. That
is, energy present in the original spectrum at a given k finds itself at a position con-
sistent with the group velocity as the propagation speed for energy.
Looking carefully at the result for the amplitude, we note that there is a potential
difficulty with the expression for those values of k corresponding to the maximum (or
minimum, should one exist) of the group velocity. At such values of k,
a2 co _ aCg _
ak2 -ak- O (6.5)
and a singularity occurs. This, of course, coincides with a particularly interesting po-
sition in the wave train corresponding to (in the case of the maximum) the very front
of the wave train where the fastest group can be found. Since the second derivative of
frequency with respect to k vanishes at that k (this would be k = 0 for the gravity wave
Lecture 6 . Discussion of Initial Value Problem (Continued) 55
case), the expansion of mas a function of k around ks must be carried to a higher or-
der, i.e., to order (k - ks )3. A discussion of the asymptotics can be found in many texts
(e.g., Whitham 1974 or Stoker 1957).As one might imagine, since the group velocity is
changing much more slowly where the derivative of the group velocity is nearly zero,
the x-interval spreads more slowly, and the amplitude decreases more slowly in the
local area near the front, i.e., like t- 1/ 3• Indeed, it is easy to show that for the front of the
gravity wave train, the asymptotic formula previously derived must be replaced by
7] - 1 No(O) 1t (x-cot J
.fiii, D (C ot/2DJ/ 3Ai lD[cot/2D)1I3 ' co =.fii5 (6.6)
where Ai is the first Airy function that is a solution of the ordinary differential equation
d2~ -xAi =0 (6.7)

dx 2
so that it is oscillatory for negative values of its argument but exponentially decreas-
ing for positive values of its argument, as shown in Fig. 6.2.
Airy function representation of wave train near leading edge

0.6
0.4
0.2
0.0
-0.2
-0.4
-0.6
-0.8~!----~------~----~----~----~----~------L-----~----~-----J
- 10 -8 -6 -4 -2 0 2 4 6 8 10
X - cot
Fig. 6.2. The Airy function describing the behavior of the wave amplitude near the leading edge of
the advancing wave front
Wave packet
1.0
0.8
0.6
0.4
0.2
0.0
-0.2
-0.4
-0.6
-0.8
-1.0
-20 -15 -10 -5 0 5 10 15 20
Xo = 15, ko = 6.2832
Fig. 6.3. The wave packet described by Eq. 6.8
Another example that is also illuminating occurs when the initial spatial perturba-
tion is nearly a plane wave. Suppose that at time t =0, the spatial distribution of 1] is of
the form (Fig. 6.3)
1] = N oe-(X!Xo)2 cos k ox (6.8)
The wave packet is shown above. By using standard tables of integrals, it is easy to
show that the Fourier amplitude of the disturbance is
N(k)=_l_X e-(k-ko )2 x5 !4
J20 (6.9)
and is shown in Fig. 6.4.

Notice that the confinement length in x is x o, while the width of the spectrum is of
order 1 / xO. Thus, if the disturbance is broad in x, approximating a plane wave slowly
modulated by the long envelope, the spectrum is very narrow in k space. This, of course,
is the basic content of the quantum mechanical uncertainly principle, where k and x
are the momentum and position coordinates. We do not need to get very fancy here,
but it is important to note that with an N(k) so sharply peaked, the formula we previ-
ously derived for the evolution of the free surface,
Lecture 6 . Discussion of Initial Value Problem (Continued) 57
12.,------.------.------.------,------.------,-------,------.
1\
10
J
4.0 4.5 5.0 5.5 6.0
) \ 6.5 7.0 7.5 8.0
Fourier amplitude Xo = 15, ko = 6.2832
Fig. 6.4. The Fourier amplitude of the wave packet of Eq. 6.8
T](x,t)=--
..&
1
JNo(k)
=
-2- rle imt +e- lmt
. 1e'kxdk
. (6.10)
can be evaluated using the fact that for k distant from the spectral peak at k = ko (noth-
ing whatever here to do with stationary phase), the integrand is essentially zero.
_ No
'/71- ~ =Je-(k - k0 )2 Xo2 /4 i(kx-mt)
v21t _= e dk
'" ..&e e
J
No i(kox-w(ko)t) -(k-ko) 2x5/ 4 i([k-ko][x-Cgt]-i(k-ko)2liJ'(ko)/2)dk
e (6.11 )
The integral is a standard one and the result is
T] = xoNo ei(kox-liJ(ko)t)e-(x-cg(ko )t)2/ 4(illi (6.12)

2.ix
where
(ilX)2 = x~ / 4 + jro"(ko)t (6.13)

The origin of the Gaussian packet is now centered on the position cg(ko)t, and it
spreads (a little more algebra is needed to put this in real form, but the result is clear)
linearly in time, yielding an amplitude that decays like the inverse of the square root
of t. This is very similar to our stationary phase result for an arbitrary initial condi-
tion and emphasizes that the result we achieved there can be thought of as an infinite
collection of packets of the type described in this idealized example.
Lecture 7
Internal Gravity Waves
In both the atmosphere and the ocean, the fluid is density stratified, i.e., p = p(z) (it is
also a function of horizontal coordinates and time) so that usually dense fluid under-
lies lighter fluid. This stratification supports a new class of waves called internal waves.
Internal waves are designated as such, because the vertical structure of the waves is
oscillatory in z (contrast with the surface gravity wave) and most of the vertical dis-
placement occurs within the fluid as opposed to the upper boundary, as in the gravity
wave example we have just studied.
We will consider the problem in the simpler incompressible case appropriate for
the ocean. The generalization to the atmosphere is straightforward if a bit more com-
plicated (see, for example, Gill 1982 and also Lighthilll978, for the generalization).
For an incompressible, stratified, nonrotating fluid that experiences small pertur-
bations about a state of rest, the rest state is characterized by
Ii =0, p = Po(z) , P =Po(z) (7.1a-c)
apo =-Pog (7.1d)

az
If we examine small perturbations about a state of rest, the equations of motion,
assuming that the motion is frictionless and adiabatic!, are
au ap
Poa;=- ax (7.2a)
av ap
Poa;=- ay (7.2b)
aw ap
p, -=---pg (7.2c)
° at az
au +av +aw =0 (7.2d)
ax ay az
ap +w apo =0 (7.2e)
at az
1 The student is asked to remind him (her) self what is required to make these assumptions in a
consistent fashion.

60 Lecture 7 . Internal Gravity Waves
where we have written the total dynamical fields for density and pressure as
Ptotal=PO+P' P«PO (7.3a)
Ptotal = Po + P, P« Po (7.3b)
so that all the non-subscripted variables in the above equations are perturbation
quantities. Be sure to note again that the last equation is the equation for adiabatic
motion, i.e., the energy equation. The condition for incompressibility is expressed
by Eq. 7.2d.
A very simple special case of the above equations occurs when the horizontal ve-
locities are identically zero, and when the pressure perturbation is also zero and the
vertical velocity is independent of z. In that case, the first, second and fourth equations
are trivially satisfied, and the combination of the third and fifth equation leads di-
rectly to
Cl 2w +N2W=0 (7Aa)
Clt 2
N2 == -g Clpo (7Ab)
Po Clz
For consistency, the quantity N must in this special case be independent of z to al-
low w to remain independent of z. N is called the buoyancy frequency, or sometimes
the Brunt-Vaisala frequency or simply the Brunt frequency (depending on your na-
tional prejudice). Whatever it is called, the simple motion we have examined, columns
of vertical motion rising or falling with no variation in the vertical direction, oscillate
with the frequency N, which depends on the degree of vertical stratification. It is helpful
to compare this frequency with the frequency of surface gravity waves. For deep wa-
ter waves of wave number k for example (these are the relatively slow surface waves),
the ratio of the surface to internal wave frequencies is
fqnt. _....!!.-_ ( -Clpo )1/2 _ ( -Clpo A- )1/2

msurf. - .[ik - kpoClz - PoClz
(7.5)
"" ( Po
~ )1/2
Po
Here we have used the fact that the vertical scale of the surface gravity wave is its
wavelength A, and that scale times the vertical derivative of the density gives an esti-
mate of the overall change of density on that scale. Since, in the ocean, the density
changes by less than 0.001 over the total depth, the ratio of the frequencies is such that
the internal wave frequencies are always smaller than the surface wave frequencies.
This makes sense, since the gravitational restoring force for surface waves depends on
the difference between the density of air and water, while for the internal waves it de-
pends on the slight difference of density between adjacent strata of fluid.
Lecture 7 . Internal Gravity Waves 61
We can derive a more general equation for the internal wave field. Taking the hori-
zontal divergence of the horizontal momentum equations yields
a _ 2 a2W
Po at vh . Uh =-V hP =-PO azat (7.6)
with the aid of the continuity equation. In the above equation, the subscript h refers
to the two-dimensional operator in the horizontal plane. z is the vertical coordinate
antiparallel to gravity, and x and yare the horizontal coordinates.
Thus,
a2__
w '11 2
(7.7)
azat - hPI Po
The time derivative of the vertical equation of motion with the aid of the adiabatic
equation yields
a2w
-+Nw=----
2 1 a2p (7.8)
at 2 Po ataz
Note that for zero pressure fluctuation, the problem reduces to the case of the os-
cillation at frequency N.
Eliminating the pressure between the last two equations yields
a 2
2 1
- 2 [VhW+-- a (Po-
aw)] +NVhw=O
2 2 (7.9a)
at Po az az
2 _
Vh =-+-
a2 a2 (7.9b)
ax al
2
Before continuing to find solutions, let's examine the last term in the square bracket
on the left-hand side of the equation. This is
1 a ( aw) 1 apo aw a2w

Po az POaz = Po az az+ az 2
(7.1O)
The ratio of the two terms on the right-hand side is
aw/azapo/az ..!!:...-apo«1 (7.12)

Po a2 w/az 2 Po az
where d is the vertical scale of the vertical velocity w. Since that scale for internal waves
is less than the total depth of the ocean (it is usually of the order of the thickness of
the thermocline or less) the ratio is less than the total density change from top to bot-
tom in the ocean, a term, again, very much less than unity. Thus, in the governing equa-
tion for density, the derivative of the background density with respect to z may be ig-
nored, leading to the simpler governing equation:
a r. 2W]+N 2<72
-2lV
2
VhW=O (7.13)
at
where the Laplacian following the second time derivative is now the full, three-dimen-
sional Laplacian. Note that the structure is not spatially isotropic. The term multi-
plying N 2 involves only horizontal derivatives. In the presence of stratification, hori-
zontal and vertical directions have dynamically different significance. Note also that
if N is zero, we obtain, again, Laplace's equation for the vertical velocity, i.e., in the
absence of stratification, the flow would be irrotational.
In fact, it is left to the student to show that the three components of the vorticity
equation in this linearized example are
:t (v -uy)=O
x (7.14a)
:t(wy-vz)=-g[~l y
(7.14b)
:, (u,-w<)=g[:.l (7.14c)
for the z-,x- and y-components of the vorticity equation (subscripts in the above equa-
tions denote partial differentiation). If the perturbation density is zero, which will occur
if there were no density variation in the basic state, each component of vorticity would
be zero if initially so. Thus for internal gravity waves, we can anticipate that the rela-
tive vorticity will be different from zero.
Let us try to find plane wave solutions in three dimensions, i.e., we write
W = wocos(K ·x-OJt)
(7.15)
K .x =kx + [y + mz
Inserting this trial solution in the governing partial differential equation yields as
the condition for nontrivial solutions
m2K2 = N2K~ = N 2(k 2 + (2) (7.16)
or
m=±N Kh (7.17)
K
Since K2 = k2 + [2 + m 2 , the frequency can be written:
m=±NcostJ (7.18)
where tJis the angle between the wave vector K and the horizontal plane (Fig. 7.1).
Fig. 7.1. y
Schematic of wave vector orientation
t
dz
Fig. 7.2.
The force diagram along
a wave crest
The frequency depends only on the orientation of the wave vector and not its mag-
nitude. The frequency is therefore independent of the wavelength.
To get a better appreciation for the reason for this rather bizarre dispersion rela-
tion, consider a balance of forces along a line of constant phase, for example the crests
of the waves, as shown in Fig. 7.2.
Let Sbe the displacement of a fluid element along the line of constant phase. If the
wave vector is tilted to the horizontal at an angle ~, a displacement of an amount S
along the phase line yields a vertical displacement dz = S cos ~. This in turn yields a
buoyancy force in the vertical direction (positive upward) of
Fz =-,~.pg = dpo dz = dpo Scos 13

dz dz
The component of this force along the direction of the phase line is just
F = dpo Scos2 13
( dz
Since, by definition, there can be no variation of pressure along a phase line (noth-
ing in the wave field varies along a line of constant phase for a plane wave), there is no
pressure force along the phase line and the force balance reduces to
a2( _gap O (
Po at 2 - aZ or (7.19a)
a+
-
2
( N 2 cos 2 7J( = 0 (7.19b)
at 2
which recovers our dispersion relation for frequency of a harmonic oscillation. No-
tice that when if is 0, we recover the first simple case in which the frequency of oscil-
lation is exactly N. To understand the reason for that, note that for a plane wave, such
that all fields are of plane wave type
(u, v, w) = (uo, vo, wo)ei(kx+ly+mz-cot) (7.20)
the continuity equation imposes the condition
kuo+ lvo+ mwo= 0 or (7.21a)
j(. u=O (7.21b)
Thus, the fluid velocity in the three-dimensional plane wave is perpendicular to the
wave vector. The fluid velocity is along the crests of the waves, i.e., for internal waves,
the wave motion is transverse; that is, it is perpendicular to the direction of phase
propagation. Thus when the wave vector is horizontal, the motion of fluid elements in
the wave is purely vertical and with no variation of phase in z (m = 0) the vertical
motion will be independent of z. Those were the conditions of our introductory ex-
ample, and we see here that this is obtained when the wave vector is horizontal. It also
yields the maximum frequency for internal waves, i.e. (Omax = N.
Note too that the frequency is a constant on a cone in three-dimensional wave
number space (Fig. 7.3) where the elements of the cone make an angle if to the hori-
zontal. The frequency increases as the cone opens up, i.e., when the elements of the
cone are closer to the horizontal plane. This has important consequences for the di-
rection of the group velocity, since the frequency is a function only of if.
m
K
C9
k
Fig. 7.3.
The cone of constant frequency and the
direction of the group velocity
Group Velocity for Internal Waves
By definition, the three-dimensional group velocity is
_ ~aw ~aw k aWA
C =1-+)-+ - (7.22)
g ak al am
where 1, j, k are the three unit vectors along the x-, y- and z-axes, respectively. Let I/> be
the angle in the x-y-plane between the horizontal projection of the wave vector and
the x-axis (Fig. 7.4).
Then a simple calculation using
k 2 +12
2 N2 2 (7.23)
W = k2 +12 +m
yields
aw _ N m2 ~= N sintJ{sin1'Jcos¢} (7.24a)
a;;- K K2 Kh K
az-
aw _ N m2 _1_= N sin tJ{sin 1'Jsin¢}
K K2 Kh K
(7.24b)
aW --N-h-=--cosuslnu
K m N .0'_0
(7.24c)
am - K3 K
In particular, note that
waw
__ N2 2
m am - - K2 cos 1'J (7.25)
so that the vertical phase velocity is always opposite to the vertical group velocity. Waves
that appear to be propagating their phase upwards will be propagating their energy
downwards, and vice versa. This is evident from examining the dispersion cone in three
dimensions, keeping in mind that the frequency increases in a direction perpendicu-
lar to the elements of the cone as shown in Fig. 7.3.
Fig. 7.4.
The orientation of the wave vector k
Cg
Fig. 7.5.
The orientation of the group velocity with
respect to the wave vector K
Finally, note that
- _ N 2m 2 2 2 2
K,c g =kcgx+lcgy+mcgz =--4-(k +1 -Kh}=O (7.26)
mK
so that the group velocity is perpendicular to the wave vector and therefore in
the direction of the fluid velocity. Energy travels along the crests and troughs and
not perpendicular to them. For surface gravity waves, we had to get used to the fact that
the group velocity was not equal in magnitude to the phase speed. Now, for internal
gravity waves, we have to adjust to the remarkable fact that the group velocity is not
even in the same direction as the propagation of phase but at right angles to it (Fig. 7.5).
Lecture 8
Internal Waves, Group Velocity and Reflection
The rather unusual dispersion relation and the nonintuitive relation between group
velocity and the wave vector lead to some very unusual physical consequences.
Figure 8.1 is from Lighthill's book (1978) taken from a paper by Mowbray and Rar-
ity (1967). It shows the result of an experiment in which a small disk is oscillated in a
stratified fluid with a constant N at a constant frequency, w. We know that in such a case
the wave vectors will be aligned in a direction such that cosO = ±w/ N. There are four
such angles. The disturbance is limited to narrow bands leading away from the oscil-
lating disk. Since the energy must be moving away from the disk, it is not too hard to
see that starting with the band in the upper right hand quadrant, the direction of the
band must correspond to the direction of the outgoing group velocity moving upward
and to the right. Since this must be perpendicular to the wave number, and since the
vertical group velocity and phase speeds must be oppositely directed, it follows that
the dark bands in the figure are actually the crests of the internal waves, which form a
cross intersecting the little disk. A movie of the experiment would show those crests
moving rightward and downward in the upper right band corresponding to energy
moving upward and to the right. The situation is sketched schematically at the right
side. The student is invited to complete the picture for the other four quadrants. One
has to admit that the physics here seems very strange. But you'll get used to it.
Fig. 8.1. A photograph showing the lines of constant phase produced by a small disk oscillating with
constant frequency. Below a diagram is shown, indicating the lines of constant phase and the direc-
tion of the group velocity of the radiated waves (from Lighthill1978)

68 Lecture 8 . Internal Waves, Group Velocity and Reflection
Note that since the fluid element velocities are at right angles to the wave vector,
the pressure work term pu is perpendicular to the wave vector. Since we expect the
energy flux to be given by the pressure work term, this explains the perpendicular
orientation of K and cg (Can you derive the energy equation from the momentum and
thermodynamic equations we used at the beginning?).
Something even stranger appears to happen if we ask about the reflection of inter-
nal gravity waves from a solid boundary. Let us suppose we have a lower boundary
sloping upward to the right at an angle {3. We will suppose the incident wave and re-
flected wave are in the plane of the slope. It is easy to consider the generalization in
Fig. 8.2.
Suppose the incident wave has the representation
w = W,ei(k1x+m]z-W[t) (8.1)
where the I subscripts refer to the incident wave field.

For this two dimensional problem (no y wave number), the continuity equation is
simply
au + aw =0 (8.2)
ax az
so that a stream function can be used, where
alj/
u=-a;- (8.3a)
alj/
W=ih (8.3b)
and the incident wave, represented in terms of its stream function is
Ij/, = ~ ei(k]x+mIZ-W[t) ='Pjei(k]x+m]z-W]t) (8.4)

lk,
The solid boundary at which the reflection takes place satisfies z = xtan{3 so that
x z
the unit vector parallel to the boundary is iB = Xcos{3 + Zsin{3, where and are unit
vectors in the x- and z-direction, respectively.
Now the reflected wave will have x and z wave numbers and a frequency that are
not known a priori. How are they determined?
KI
Fig.S.2.
A wave packet with wave
number K] is incident on a
sloping surface
Lecture 8 . Internal Waves, Group Velocity and Reflection 69
We can write the reflected wave, generally, as
IJIr = Prei(krx+mrz-Wrt) (8.5)
The total wave field during the reflection process is the sum of the two waves,
IJI = IJI, + IJIr (8.6)
On the boundary where z = x tan {3, the total stream function must be a constant.
Without loss of generality, let that constant be zero. Thus, on z = x tan {3 we have
o= ~ei[(kI+mI tan P)X-Wjt 1+ ~ei[(kr +mr tan P)X-Wrt 1 (8.7)
This must be true for all t and for all x along the boundary. Clearly a single relation
between the amplitudes of the incoming and reflected waves will be unable to satisfy
Eq. 8.7, unless
cq = OJr (8.8a)
k, + m I tanf3 = kr + mr tanf3 or (8.8b)
K, . iB =Kr • iB (8.8c)
Thus, the frequency and the component of the wave vector parallel to the boundary
are both conserved under reflection. This is a general result for plane waves. What
is special for internal gravity waves is that the conservation of frequency implies that
since OJ = N cos if, the angle of the wave vector with the horizontal must be preserved
under reflection, regardless of the orientation with the boundary. For more familiar
problems where the reflection is specular, the wave vector component perpendicular
to the boundary is preserved. This is not the case here. Thus, under reflection both
the component of wave vector along the boundary and the horizontal component of
the wave vector must be preserved.
We can use a geometrical construction to see how this occurs (see Fig. 8.3).
In the construction, the reflected wave vector is determined by three considerations:
1. The component along the slope must be the same for both incident and reflected
wave;
2. The angle of the reflected wave vector to the horizontal must have the same magni-
tude as for the incident wave so that the cosine of the angle (frequency) is pre-
served under reflection;
3. The direction of the reflected wave vector must be such that the associated group
velocity is directed away from the slope.
We note that in this example, the magnitude of the reflected wave vector is much
greater than that of the incident wave. Therefore, the wavelength of the wave is not
preserved under reflection; indeed, the wavelength shortens as a consequence of the
reflection process.
Fig.S.3.
A sketch showing the wave
numbers of the incident and
reflected internal gravity waves
from a sloping surface K,
To determine the result analytically, let the magnitude of the incident wave vector
be K,. Then the component of the incident wave vector along the slope is K, cos( tJ - /3),
while that of the reflected wave along the slope is Krcos( tJ + 13). Note that we have used
the fact that tJ is preserved under reflection. Since these two terms must be equal to
satisfy the condition of no flow through the solid surface of the slope,
K,cos( tJ - 13) = Krcos( tJ + 13) (8.9)
Now define
a=n/2-tJ
Here a is the angle with respect to the horizontal of the group velocity of the inci-
dent packet and also the angle with respect to the horizontal of the reflected packet
(see Fig. 8.4). Note too that ill = N sin a.
If the definition of a is used in the equation for the equality of the along-slope com-
ponents of the wave vectors, we obtain
Kr = sin(a+ fJ)
(8.10)
Kj sin(a- fJ)
Note that as the angle of the incident group velocity approaches the angle of the
slope, the magnitude of the reflected wave number becomes infinite. Thus, as the fre-
quency is lowered, a gets smaller; when it coincides with 13, Kr becomes infinite. We
would anticipate that such short scales would be affected by friction and are likely to
be efficiently dissipated. So, 13 is a critical angle for the incident packet.
There is also a problem when a is less than 13 according to the above formula; since
neither of the wave number magnitudes can be negative, the left-hand side is always
positive, but the right-hand side becomes negative when a< 13. Clearly, the situation
KI
'9 1 '9 r
Kr
Fig. 8.4. A sketch showing the relation of the incident and reflected wave vectors
Kr~"'",
""""""
"K;'
Fig. 8.5. C9r

The reflection process when
the slope is steep (a < f3)
Fig. 8.6.
The reflection in a shallow
wedge
must be reconsidered in that case. Figure 8.5 shows the geometry of the reflection proc-
ess, then (see also Fig. 8.6). We see that when a> /3, as in the previous case, the reflec-
tion is forward along the slope. Now, when a < /3, the reflection must be backward (and
since a is preserved, forward reflection would put the wave packet inside the slope,
which is an impossibility). The back reflection leads to the relation (try it)
Kr sin(a+ 13) a result one might have guessed.

K] sin(f3 -a)
Fig. 8.7.
The reflection in a large, open
wedge
Since the reflection from a horizontal surface will be specular if the bottom slope
forms a wedge-shaped region with an upper horizontal surface, the reflection process
can lead to further surprises. If, for example, a wave packet enters the wedge with a
frequency such that a> [3, the reflection from both the bottom slope and the top sur-
face will be forward. The wave packet will bounce back and forth, advancing towards
the apex of the wedge, becoming shorter at every bottom reflection, and finally dissi-
pating in the apex of the wedge (Fig. 8.7).
If, on the other hand, the bottom slope is strong enough so that [3 > a (in the limit
it could be a vertical wall), the reflection from the bottom will be backward, and the
wave will leave the region of the wedge (Fig. 8.7).
Recall that all of these bizarre properties are due entirely to the fundamental phys-
ics of the wave that determines its frequency only in terms of the angle the wave vec-
tor makes to the horizontal, and since the frequency must be preserved under reflec-
tion, this places a terrific constraint on the reflection kinematics.
Up to now we have dealt with fluids in which the buoyancy frequency has been in-
dependent of z. In the ocean, N is certainly a function of z. It is large in the thermo-
cline and small in both the mixed layer and in the abyss. Before dealing with that vari-
ation, it is useful to discuss the equation for the energy in the wave field.
Multiplying each momentum equation by the velocity component in that direction,
we obtain
aKE ap
-=-u--v--w--wpg
ap ap
at ax ay az (8.11)
=V·pu-wpg
where
KE = Po ~2 + v2 + w 2 ] (8.12)
2
The step between the first and second equations uses the condition of incompress-
ibility, i.e., the divergence of the velocity vanishes. The last term on the right -hand side
of the equation for the kinetic energy is the transformation of potential to kinetic
energy. If heavy fluid sinks (p> 0, w < 0) or light fluid rises (p < 0, w> 0) (where we
recall that p is the density perturbation), then the kinetic energy will increase by the
conversion of gravitational potential energy. This last term can be written in conser-
vation form using the adiabatic equation, since from that equation it follows that
P 2 a 2 2
wpg=_~~=_g_ap (8.13)
Po N at 2PoN 2 at
Thus,
~[KE+ lp2 /2PoN 2J+-V ·pu=o (8.14)

at
It is tempting to consider the second term in the square bracket as potential en-
ergy. To see that in fact it is, it is useful to use the relationship between the Lagrangian
vertical displacement (; and the vertical velocity. For small displacements,
w=a( (8.15)
at
(Large displacements would require the total derivative in the above equation).
If this is used in the adiabatic equation, we obtain
N2
P=Po-( (8.16)
g
by a simple integration. We can therefore think of (; as the vertical displacement of

each isopycnal surface, such that the perturbed fluid element remains on its original
density surface. In turn, we can now write the energy equation as
a[ KE + Po
-
at 2
2r2] + V . pu = 0
N ~ (8.17)
so that the second term in the square bracket has exactly the same form as the poten-
tial energy of an extended spring in which the spring constant per unit mass measur-
ing the restoring force is the buoyancy frequency squared, i.e., N 2 •
It will be left for the student to show that for a plane wave, there is equipartition
between kinetic and potential energy and that the energy flux vector
pu=c~ (8.18)
where E is the sum of the kinetic and potential energy.

Note that for a plane wave in two dimensions, we can always align our coordinate
system for a single plane wave to align the wave vector in the x-z-plane.
Suppose the plane wave has the form
W= wocosE> (8.19a)
E>=kx+mz-wt (8.19t»
Then from the continuity equation,
m
u=wo-cos8 (8.20)
k
(note that this satisfies the condition that the fluid velocity be perpendicular to the
wave vector). From the relation between wand the vertical displacement r;, (or from
the adiabatic equation),
r =--smB
~
wo· (8.21)
OJ
Thus, the kinetic and potential energies averaged over a wave period are
(E)=~OW~{l+~: +::} (8.22a)
(E) = Po W2{k2 +m2} (8.22b)

2 0 k2
from which it follows that the horizontal and vertical components of the energy flux
are
cgx(E) = Pow~ (OJ) m 2

2 k e (8.23a)
Cgz(E)=-POW~ (OJ)m2 (8.23b)

2 m k2
Note again that the direction of the vertical energy flux is opposite to that of the
vertical phase speed OJ/ k. Indeed, the energy flux is perpendicular to the wave vector
as can be immediately verified.
Lecture 9
WKB Theory for Internal Gravity Waves
The buoyancy frequency is never really constant. Indeed, in the ocean there is a
significant variation of Nfrom top to bottom. Figure 9.1 (next page) from the Levitus
Atlas (1982) shows the distribution of N of the zonally averaged global ocean.
By assuming that N is constant in our calculations to this point, we have been say-
ing effectively that over the vertical distance Az = 21t / m, N2 changes only slightly. Al-
ternatively, we can state equivalently that N is a slowly varying function with respect
to the phase of the wave over which
&2 = aN 2 It «N 2 (9.1)
az z
The major change in N occurs in the vertical; that is, it is a stronger function of z
than of the horizontal coordinates. We already know from our earlier work on waves
in slowly varying media that
aK _ -
-+c ·VK=-vn (9.2)
at g
where Q is the local dispersion relation. If we consider N to be a function only of z,

this equation implies that the vertical component m will be the only component
of the wave vector that will alter as the wave traverses a region of variable N. Fur-
thermore, if we assume that N is independent of time, the dual equation for the
frequency
am _
-+c ·Vm=-
an (9.3)
at g at
shows that the frequency will be independent of time for an observer moving with
the group velocity. Since the wave packet itself will move with the group velocity, this
means that (k, 1, m) will be constant. It remains to be determined how m changes and
how the amplitude will vary in a region of varying N.
The governing equation is again
a2
-V2w+N2(Z)V2W=O (9.4)
at 2 h

76 Lecture 9 . WKB Theory for Internal Gravity Waves
Latitude
60' 30' EO 30' 60' 90'N
o
---...,../
g,~1\
~ -~'---- ',,",J~-----'~I~~\
l!1 \
I'
, II \ .' - _ _ _ 27.75 - "'\
\ ,',__
\
------27.80---1 / r \ j/- -I~ I",II,:\
:,'
I
',_,
"27.852
,/_, I~I "I
• J
r-- _//'0 I !,I~\",i'
... 'i 27.825 " \
, 27.85."" \ "
,_,-_,27.85 \'
JI
\.'
/ 27.85
I,
'. : .. I \
\
,' __ ~!/
, -- __ " -
_- .... ------- .. 5.0----------- ... -/'··- ... ---
2000
/------.J375----.,;'
. "\.,,--
\ (
,)
\\(j""'"
3000
.. ... . . ,/~."
," ~
\":
~.: "
Lecture 9 . WKB Theory for Internal Gravity Waves 77
Let's try to find a solution in the form
w =A(z)e i (kx+ly-mt+8(z)) (9.5)
and we will assume that the vertical variation of the phase is much larger that the ver-
tical variation of A; that is, we will assume that N is varying slowly enough in z so that
locally our solution will look like a plane wave.
We define
dO
m(z)=a; (9.6)
(Note that for a pure plane wave, 8 would be simply mz).

Inserting the hypothesized solution in the governing equation yields (z subscripts
denote differentiation)
-ai [_(k 2+Z2)A + A zz -0; A] - N2(e +Z2)A -iah20z A z +OzzA)=O or (9.7)
Azz + A[(N 2 -m2)K~ 02J - 2im201l2 j_)Ol!2 A)= 0 (9.8)

z z dZ ~ z
m2
We have assumed that 8z is order one while A zz / A « 1 and that the variation of 8z
and A are also small (the local plane wave approximation. This implies that the domi-
nant term in the equation is the curved bracket in the second term. This yields an ex-
pression for m or equivalently,
2 2= N 2_m2 Kh2
Oz =m ? or (9.9a)
m
m(z)= dO =[N 2(Z)_m 2 ]1I2 (9.9b)

dZ m2 Kh
or, equivalently, one may think of this relation as the necessary condition that the fre-
quency both satisfy the plane wave dispersion relation
m2 = N2(Z)K~
(9.10)
m 2 +K2h
while at the same time be independent of z. This yields for the vertical phase factor
0= z JKh,I N(z')-m
2
2
2
dz' (9.11)
Zo m
~ Fig. 9.1. Upper panel: Annual mean global potential density distribution in depth and latitude for
the world ocean. Lower panel: Annual mean of the buoyancy frequency as a function of latitude and
depth (reworked after Levitus 1982)
With the differential equation in the form

2 2
d 2W
- +N -OJ
2
K 2W=0
h (9.12)
dz 2 OJ
we might naively have expected the vertical structure for slowly varying N to look like
e{KhP7}
but instead it is the integral that enters the phase so that the vertical component of
the wave number vector is given by its local plane wave value.
The imaginary part of the equation for A (or equivalently, the next order term is
the slow variation with z) yields the constraint
~[Aml/2 ]=0 (9.13)

Jz
or
A(z)= A(zo) (9.14)

(mlmo)1/2
so that Zo and mo are evaluated at some arbitrary constant value of depth. As m gets
larger, i.e., in a region of larger N, the amplitude diminishes. This is easy to understand
physically. As the wave propagates vertically, the flux of energy must remain the same
at each z to avoid the pile up of energy and the local increase of amplitude with time.
We saw in the last lecture that the vertical energy flux could be written
Cgz (E)=_P OA2 mOJ

212
using A instead of wo for the amplitude. To keep the energy flux independent of z, and
since both frequency and horizontal wave number are independent of z, it follows
that A must go inversely with m1l2, which is the result we have already achieved. Thus
that behavior is simply a consequence of energy conservation.
If the wave propagates to an elevation where the frequency is greater than the local
value of N, the vertical wave number becomes purely imaginary and the disturbance
exponentially decays beyond that location. We can say then that the wave will be
trapped between regions in z where N matches the frequency of the wave.
To find the path of the wave packet in the vicinity of such a region, we can use the
ray equations, which for two dimensions are
dz =c =_N km (9.15a)
dt gz K3
dx m2
-=c =N- (9.15b)
dt gx K3
1.00 --,..---.....------r--....------r---.-------r--..,
r.
0.99
0.98
0.97
0.96
0.95
0.94
Fig. 9.2. 0.93

The ray path near a turning
point where the frequency OJ 0.92LI----~--~----~----~----~--~----~----~
equals N -0.02 -0.015 -0.01 -0.005 0.0 0.005 0.01 0.015 0.02
The path of the ray in the x-z-plane is given by
dz Cgz 0)
-=-=-klm=--- (9.16)
dx cgx ~N2 -ai
Consider the region near the point z* where N(z*) = 0). Expanding N 2around z* yields
N 2(z)=N 2(z*)-(z*-z)(N 2)' + ... (9.17a)
dz 0)
or (9.17b)
dx ~(-dN2 /dz)z*(z * -z)
z*-z= 0) (X*_X)2/3 (9.18)
~-(dN2 /dz)z*
The ray path has a cusp at the turning point (x*, z*) as shown in Fig. 9.2.
Normal Modes (Free Oscillations)
Consider a fluid bounded below by a flat bottom at z = -D and with a free surface whose
rest position is z = 0 (Fig. 9.3). Again, the fluid is incompressible and stratified. This
situation is a combination of the two problems previously studied. There should be
the possibility of surface gravity waves as well as internal waves due to the stratifica-
tion. The issue here is how they relate to each other and in addition, what the nature
of the internal waves in this bounded domain is.
Again, the governing equation is
02
-V2w+N2(Z)V~W=O (9.19)
ot 2
for unforced motions.

Z=TJ
~ n.~~
~ ~~ ~
N2
Po=Po(Z)
W=O---.......
Z=-o
Fig. 9.3. The definition figure for determining the normal modes of a stratified fluid with a free surface
The boundary conditions are
1. At the bottom,
w=O, z=-D (9.20)
2. At the free surface, we have both the kinematic condition
017 (9.21a)
w=at
where 1] is the free surface elevation, and
p(x,y,z = 17) =0 (9.21b)
Since 1] is supposed to be small (linear, small amplitude motions),
p(x,y,z=17)= p(x,y,O)+ op 17+ ... (9.22)

oz
We only want to keep terms that are linear in the amplitude of the motion on the
right-hand side of the above equation, since we are doing a consistent linearization of
the dynamics. Since the linearized form of the vertical momentum equation is
op =-(Po+p)g-P0 at
ow (9.23)
oz
each term in the above equation is of the order of the amplitude of the motion and so would
yield a quadratic term when multiplied by 1] in the expansion of the boundary condi-
tion, except the first term on the right-hand side of the equation for the vertical pressure
gradient, which yields op / oz in the absence of motion. Thus, considering only terms
that are linear in the perturbation amplitude, the pressure condition on the free surface is
p(x,y,z =17)=0 = p(x,y,O)- Pog17 (9.24)

It is important to realize that we are not assuming the motion is hydrostatic; all
we have used is the small amplitude approximation. We did not have to work very
hard in the surface gravity wave problem, because we had the Bernoulli equation
at our disposal to use at the surface. Since the fluid is stratified, the motion is no
longer irrotational and the velocity potential and resulting Bernoulli equation no
longer exist.
Thus at the upper boundary,
p(x,y,O) = Pogl] , z=O (9.25a)
al] =w (9.25b)
at
Eliminating the free surface elevation between the two conditions,
a (9.26)
at p(x,y,O) = Pogw , Z=O
We can operate on the above equation with the horizontal Laplacian
V~ ap = PogV~w (9.27)
at
and use the previously derived relation (from the divergence of the horizontal mo-
mentum equations and the continuity equation)
a2w
V~P=Po azat (9.28)
to obtain for the upper boundary condition in terms of w:
a2 a
__ ~
at 2 az -
2
°, z =°
gV h W = (9.29)
We will particularly be interested in the oceanographically relevant case where the

parameter DN2 / g« 1. This parameter can be interpreted in several ways. First of all,
it gives a measure, as we have seen before, of the total density difference over the depth
of the fluid divided by the mean density. This is very small for the ocean. Second, us-
ing our previous results, it can be seen as the ratio of the (square) of the maximum
internal gravity wave frequency to the surface wave frequency (squared) for a wave
whose wavelength is of the order of the depth of the fluid. We are interested, as noted,
in the case when this ratio is small, i.e., when the surface waves have higher frequency
and phase speeds than the internal waves. This helps separate the two wave types that
are described by the same set of equations given above. There is a hint then that in the
case when DN2 / g« 1, approximations to the governing equation will be in order if
we want to concentrate on one or the other of the waves.
Since the fluid is unbounded in the x and y-directions in this simple model, we can
look for plane wave solutions in the horizontal direction. Orienting the x-axis to be in
the direction of the horizontal wave number, this implies that we can look for solu-
tions of the form
w = W(z)ei(kx-mt) (9.30)
and insertion in the equation for wand its boundary conditions yields
2
d W +k2[N: -l]W=O (9.31a)
dz 2 OJ
W=O, z=-D (9.31b)
2 dW 2
OJ - - g k W=O, z=O (9.31c)
dz
Consider the case where N is constant and N2 > oJ. The solution of the Wequation
will be oscillatory in z and the solution that satisfies the boundary condition at z = -D
will be
W= A sinm(z + D) (9.32a)
m 2 =k2[ : : -1] (9.32b)
Note that the latter definition implies that were m known, the corresponding fre-
quency would be
OJ=+ Nk (9.33)
- Jk2+m2
which is a familiar result from our work on plane internal gravity waves. We can ex-
pect the above eigenvalue problem to yield quantized values of m so that the equa-
tion for the frequency in terms of k and m will be as in the plane wave case except that
m will no longer be a continuous variable but quantized.
The upper boundary condition yields the eigenvalue relation:
OJ2mcosmD =gk2 sinmD (9.34)
or using the relationship between frequency and wave number written above,
N 2m
gtanmD (9.35)
e+m2
It is useful to write the above condition in terms of non-dimensional wave num-

bers.
Let
m'=mD
(9.36)
k'=kD
Then the eigenvalue relation becomes
ND]
[-g-
2
_m' _ tanm' (9.37)
The roots of this can be found numerically. A graph of each side of the equation is
helpful in understanding the results (Fig. 9.4).
Figure 9.4 shows both the left- and right-hand sides of the dispersion relation for
the case where the parameter N2DIg is artificially large (0.1) and kD = 1. Still, the roots
of the relation corresponding to the intersections of the two curves are very close to
an integral multiple of 1t. In the above case, the first two are at mD = 3.1562 and 6.2947.
For smaller values of the parameter N2DIg, it is not possible to distinguish the curve
of the left-hand side of the equation from zero, in which case we can see either graphi-
cally or analytically from the equation itself that in the limit N 2D I g~ 0, the roots
approach
IGW dispersion relation kO = 1 N20lg = 0.1
0.1
-0.1
-0.2
-0.3
-0.4
_O.5'L________L -__- i__- i____ ~ __ ~ ______ ~L_ _ _ _ _~_L_ _ _ _ _ __L~_ _ _ _ _ _~
o 2 3 4 5 6 7
First two eigenvalues at mO = 3.1562 6.2947
Fig. 9.4. The dispersion relation 9.37

mD= jrr, j=1,2,3 ... (9.38a)
kN
(9.38b)
OJ = OJj =+ k 2 + /rrZ / DZ
In this limit, W(z) is very nearly
W(z) "" Asin jrr (z + D) = A( -1)j sin jrrz (9.39)

D D
so that w vanishes on both the lower and upper surfaces. For these internal gravity wave
modes for which OJ < N, the free surface dynamically acts as if it were rigid and the
eigensolutions are the same as if the upper surface were simply a rigid lid on which
w=O.
Now let's examine if there are solutions of the problem for OJ> N. If that is the case,
we can still use the same solutions and dispersion relation but we must realize that m
will now be purely imaginary. It might be clearer to just go back and rewrite the solu-
tion in terms of real variables. So for OJ> N, we write
qZ =kZ[1- ::] (9.40)
in terms of which the solution satisfying the lower boundary condition is
w = Asinhq(z + D) (9.41 )
which should look familiar from our surface wave studies. The upper boundary con-
dition now yields
l( NZDJ~=tanhq'
-g k,Z_q'Z
(9.42a)
q'=qD (9.42b)
OJZ = NZk z (9.42c)

-kz Z
-q
When the parameter N 2D / g is small, the left-hand side will be small except in the
vicinity of q' = k', which yields the only eigenvalue for which OJ> N (see Fig. 9.5).
This is the graph of the two sides of the eigenvalue relation when N < OJ. There is a
single root for qD which in the case when N2D / g = 0.01 is equal to 0.9884kD, q is very
nearly k.
This yields a frequency using the upper boundary condition:
OJ2 q cosh qD = gk2 sinh qD (9.43 )
or since q is very nearly k,

Dispersion relation for surface IGW mode N20lg = 0.Q1

2.0, "
1.5
1.0
0.5
0.01..........-= -
-0.5
-1.0LI__________ ________ __________ __________L __ _ _ _ _ _ _ _

~ ~ ~ ~L_ _ _ _ _ _ _ _ _~
0.0 0.2 0.4 0.6 0.8 1.0 1.2

Root at qO 0.98841 =
Fig. 9.5. The dispersion relation for the external mode
OJ '" ±,Jgk tanh kD (9.44)
and the eigenfunction is,
W(z) = sinhk(z + D) (9.45)
Both the eigenfunction and the eigenvalue in this limit are precisely the values ob-
tained for the surface gravity wave problem for a homogeneous fluid.
Thus, the full spectrum of oscillatory modes splits into two (unequal) parts. There is
first of all the free surface mode, which, when N2D I g« 1, does not even notice the strati-
fication. This is because the depth of penetration is of the order of the wavelength, and for
small values of N2DIg, the fluid motion in the wave, maximum at the surface and
exponentially decreasing into the fluid, does not encounter the density variation. The sec-
ond class of solutions whose frequencies are all less than N consist of internal gravity waves
whose vertical motion at the upper surface is negligible. Compared to w within the fluid,
the vertical velocity at the free surface is negligible and the frequencies of the modes and
their structures are approximately those for a fluid contained within two horizontal rigid
surfaces. It is left to the student to show that for eigenfunctions of the same amplitude, the
free surface displacement of the internal gravity wave compared to that of the free surface
wave is small, and this smallness is of the ratio of the respective frequencies.
For the internal gravity wave part of the spectrum using the rigid lid approxima-
tion,
w = W(z)ei(kx-wt) (9.46)
where W satisfies
dZW +kZ[NZ -l]W=O (9.47a)

dz z oi
W=O, z=-D,O (9.47b)
For N constant, the eigensolutions are the sine functions sin (j1tZ / D) and for the jlh
mode,
(lJj/N=kD/~kZDZ+ /1t z j=1,2,3 ... (9.48)
Note that for large k, the frequencies of all the modes approach N (the student should
think about the dispersion relation for plane internal gravity wave modes to under-
stand why this is so)(Fig. 9.6).
1.0 I I---,--,--,-----,---,-=~=::::::::==::::c::======j
0.9
0.8
0.7
0.6
~05
e .
0.4
0.3
0.2
2 4 6 8 10 12 14 16 18 20
wiN as a function of kD
Fig. 9.6. The dispersion relation showing the frequency as a function of wave number for the first
three internal gravity wave modes
The normal modes, of course, do not propagate energy vertically. Each mode in z
can be decomposed into two plane waves using each with vertical wave number of
opposite sign so that the eigenfunction can be thought of as the sum of an upward
and downward propagating mode whose energy fluxes vertically cancel.
There is energy propagation in the horizontal direction, and for each vertical mode:
aOJj _ jn ._
(9.49)
ak -ND(J.2n 2 + k2D2 )3/2J-1,2,3 ...
Note that for very large kD, the group velocity in the horizontal direction goes to
zero. The maximum group velocity for each mode is cgmax = ND / jn, which occurs for
the long waves, i.e., as kD ~ 0.
Since N is a relatively strong function of z in the natural ocean, it is important to
point out that qualitatively the eigenfunctions and eigenvalues of the more general
problem are similar to the case of constant N. There are, however, a few important fea-
tures of the solution to consider when N is variable.
Consider again the governing equation for the eigenfunction, W(z):
2 2
d W +k 2[N -l]W=O (9.50a)
dz 2 oi
W=O, z=-D,O (9.50b)
It should be clear that the solutions of the above problem satisfying the homoge-
neous boundary conditions on W must have frequencies less than the maximum value
of N in the interval -D < z < 0. This follows intuitively from the nature of the equa-
tion, for if the frequency is greater than Nmax, the coefficient in front of the last term
in the equation is always negative and the character of the solutions will be exponen-
tial rather than wavelike, and it will be impossible to satisfy the two homogeneous
boundary conditions. This also follows from multiplication of the Wequation by W,
which with integration by parts and use of the boundary conditions yields
2 JW 2]dz =°
_~o [(dWJ2 2(N
- dz + k oi -1 (9.51)
If OJ > N everywhere in the depth interval, both terms in the integral would be nega-
tively definite, and there would be no way to satisfy the condition that the integral
vanish. Similarly, it follows from the theory of standard Sturm-Liouville problems that
the eigenfunctions corresponding to different eigenvalues are orthogonal:
o
IWj Wk N 2 (z)dz=O, i*j (9.52)
-D
In addition, the eigenfunction corresponding to each higher eigenvalue has one
more zero of the function W(z) in the depth interval.
The character of the eigenfunctions are of interest. If
N~in ~oi ~N~ax (9.53 )

the character of W will be oscillatory in the depth interval in which N is greater than ill
and evanescent outside that interval. We can therefore expect some modes to have their
energy trapped in the region where the stratification is greatest, and these will be the
modes with the highest frequencies. There is a very good discussion of the general
problem in Gill's book, and Fig. 9.7 and 9.8 are taken from that book.
Figure 9.7 shows a characteristic profile of N in the subtropical North Atlantic. Note
the maximum of N in the region of the thermocline at about 750 meters. Figure 9.8
shows the first two eigenfunctions for that profile.
Potential temperature (OC/dbar) x 101

-0.10 -0.05 0.00 0.05 0.10 0.15 0.20 0.25 0.30
Brunt-Vaisala frequency (cph)

0 2345678 9 10
0
25
50
75
100
125
150
175
200
'0
x 225
0;
..0
:!!. 250
l!!
:::I
III
III
275
l!!
Q.
300
325
350
375
400
425
450
Fig. 9.7. 475

The distribution of N(z) in the
North Atlantic (from Gill 1982) 500
The figures on the left of the figure are essentially Wj. The second figure is essen-
tially the form of the solution in the long wave limit (more of this will be discussed
later), and the last figure is the shape of the pressure or horizontal velocity in each
mode, really the derivative of the function W. Note, as expected, the n = 1 mode has
no zeros for W (just like sinn z / D), while the second mode has a single zero (like
sin2n z / D). However, the location of the zero and maxima differ from the constant N
case. Note too that at great depth where N is small, the eigenfunctions are not oscilla-
tory in agreement with out qualitative discussion of the governing equation.
z w z hn z Pn
Fig. 9.S. Eigenfunctions for the buoyancy profile of Fig. 9.7 (from Gill 1982)
Lecture 10
Vertical Propagation of Waves:

Steady Flow and the Radiation Condition
There are numerous situations in which fluid flows over an obstacle, say a mountain
in the atmosphere, a sea mount, or a ridge in the ocean, and we would imagine that
internal gravity waves, if the fluid is stratified, would be generated. Such situations
are of interest in their own right, but additionally they force us to carefully examine
the radiative properties of the waves, which must be understood, sometimes, to actu-
ally solve the problem.
Consider the case of a stratified, incompressible fluid as studied in the preceding
lectures, except now we will imagine that the background state includes a mean ve-
locity in the x-direction, which is also a function of depth, z. If the dynamics are
inviscid, such unidirectional flows are themselves exact solutions of the equations of
motion. Indeed, consider the equation of motion in the zonal direction. In the absence
of rotation and friction, and for motion in the x-z-plane,
au au
-+u-+w-=---
au 1 ap (10.1)
at ax az pax
Before we start to examine the wave problem, first note that if the flow is periodic
in the x-direction, the x-average denoted by an overbar will yield
au
-=---
auw
(10.2)
at az
To obtain this equation, we have used the continuity equation and the assumption
that the x-average of all terms of the form a I ax[ finite 1will be zero. The equation above
describes how the mean current can change if there is a convergence of the flux of
x-momentum carried by vertical motion uw. Note that this term can be different from
zero even if u and w separately have zero x-average. It is therefore possible for the waves
that we will consider to alter the very flow that constitutes the character of the back-
ground in which the waves are embedded, but note that the change will be quadratic
in the amplitude of the wave motions (if the flux uw is due to the waves). Thus, in lin-
ear theory for the waves, the background flow can be approximated by its initial value,
but the above equation can be used to consistently calculate the small, second order
changes in the mean due to the waves, and sometimes we are very interested in that
change.

92 Lecture 10 . Vertical Propagation of Waves: Steady Flow and the Radiation Condition
To linearize the equations of motion, we write all variables as a sum of the order
one basic state plus a small perturbation. For our fields, that will be
Utotal =U(z)+u(x,z,t)+ ... (10.3a)
Wtotal = w(x,z,t) + ... (l0.3b)
Ptotal =Po(z)+ p(x,z,t)+ ... (l0.3c)
Ptotal = Po(z)+ p(x,z,t)+ ... (lO.3d)
where the lower case variables are the wave fields, and each is the order of the wave
amplitude and hence is small compared with the basic state. Again, the basic state
density and pressure fields, denoted by a subscript 0 satisfy the hydrostatic relation.
The linearized equations of motion, when the above decomposition is inserted into
the equations of motion and only terms that are linear in the wave amplitude are re-
tained, are
PO[~+U~]U+POWdU =_ ap (10.4a)
at ax dz ax
Po[~+U~]w+pg=-
at ax ap
az
(1Mb)
au + aw =0 (lO.4c)
ax az
[~+U~]p+wapo
at ax az
=0 (lO.4d)
Once again, we have assumed that the motion is incompressible and adiabatic
(Eq.lO.4d). In the continuity equation, we have assumed that the basic state density
changes by a very small amount over the vertical scale of the motion. Except for the
presence of terms proportional to U(z), this set of equations is identical to the equa-
tions we used to study internal gravity waves. Note that the last equation could be re-
written in the form
[ata axa]
-+U- b-wN 2 =0 (lO.5a)
b=.gp/po (10.5b)
It is convenient to introduce a compact notation for the linearized advective op-

erator, i.e.,
- [a a]
D=. -at+ U ax (l0.6)
Lecture 10 . Vertical Propagation of Waves: Steady Flow and the Radiation Condition 93
in terms of which our equations are
PoDu+ PowU z =-Px (10.7a)
PoDw=-pz -bpo (1O.7b)
Ux +wz =0 (10.7c)
Db=wN 2 (10.7d)
where, for ease, subscripts have been used to denote differentiation.

Note that using the linearized Lagrangian relation between vertical displacement S
and the vertical velocity
DS= w (10.8)
the adiabatic equation yields
b=(N 2 (10.9)
Multiplying each momentum equation by its velocity component and the adiabatic
equation by the buoyancy, b, with the aid of the continuity equation, we obtain the
energy equation
_{u
2 +W2 (2N 2 } _ dU
PoD - - - + - - +V·pu=-Pouw- (10.10)
2 2 dz
If the vertical shear of the basic state velocity U is zero, then we obtain the usual
conservation statement for the sum of the kinetic and potential energies. However, when
the shear is different from zero, there is a source term for the perturbation wave energy.
This source is proportional to the shear and its product with the product -Pouw, the
Reynolds stress. We are particularly interested in the sign of this term when averaged
over a wave period or wavelength, -Pouw. Consider the situation depicted in Fig. 10.1.
U(z) U(z)
Fig. 10.1. Illustrating the flux of momentum in the Reynolds stress. The velocity profile is smoothed
by the transfer of momentum by the waves
Think about a region in which dU / dz > O. If, on average, the perturbation zonal velo-
city U is positive whenever W is negative, the source term will be positive and the energy
of the wave field will tend to increase (of course the energy so produced could be locally
fluxed away). Is that likely? If a fluid element has w < 0 with dU / dz > 0, then it is coming
from a region where U is larger than where it arrives. If it retains to some degree its origi-
nal x-momentum, it will show up at its new location with a perturbation u, which is posi-
tive. Now the "if" of the last sentence is a big one,since there is no guarantee that other fac-
tors, such as the perturbation pressure, might not intervene to alter that simple prediction.
In fact, it often occurs that the Reynolds stress turns out to be zero even in the presence of
shear and perturbations u and w. When that correlation -Pouw is different from zero, it
provides a flux of u-momentum from one z-level to another. If that flux, as described above,
is downgradient, i.e., from larger U to smaller U, the perturbation energy will increase.
However, in fluxing mean momentum down the gradient it will tend to locally "flatten"
the profile of mean velocity as shown in the figure. Now from elementary considerations,
this internal mechanism cannot alter the overall mean momentum of the flow; that is,
JUdz = constant (10.11)

allz
However, the mean kinetic energy is
JU 2 Podz/2
allz
It is easy to see that if the integral of U is fixed and the profile of U is made more
flat, the variance of U, i.e., the kinetic energy of the mean flow will decrease.
This can also be seen directly by considering the momentum equation for the mean
flow in x as written above:
dU dUW
----- (10.12)
dt dZ
Multiplication by Ii yields
1 dU 2 _dUW dUUW -dU

---=-U---=-----+UW- (10.13a)
2 dt dZ dZ dZ
or
1 dU 2 dUUW -dU
--+----=UW- (10.13b)
2 dt dZ dZ
The first term in the above equation is the rate of change of the kinetic energy asso-
ciated with the mean flow (per unit mass). The second term on the left is a flux term,
which will integrate to zero if the flow is contained between horizontal plates where W
vanishes. The term on the right-hand side is a source or sink of kinetic energy of the
mean. Comparing it to the equation for the perturbation energy we see that if the
Reynolds stresses increase the energy of the perturbation wave field, they must at the
same time be decreasing the energy of the mean flow. (Note in our identification we
are using the fact that to order amplitude squared, Ii = U.) Thus, this term is an energy
transformation term representing an energy transfer between the mean (over a pe-
riod or wavelength) and the perturbations.
The vertical flux of wave energy is again pw. What can we say about it and its rela-
tion to the Reynolds stress? The earliest treatment of the problem can be found in
Eliassen and Palm (1960) and the student is referred to it for an fuller understanding
of the historical context.
We are going to be especially interested in steady flows over bumps and the result-
ing steady wave field generated by the interaction of the flow and the topography.
In the steady state in which there is no secular increase in the wave energy, the en-
ergy equation reduces, when averaged in x, to
d
dz {pw +PouwU}=PoUdUW (10.14)
az
after a simple integration by parts of the source term.
On the other hand, the steady momentum equation in the x-direction is
a dU
-[PoUu+ p]+ Pow-=o (10.15)
ax dz
Since the motion is two-dimensional and nondivergent, we can introduce a stream

function lJI such that
U=-If/z (1O.16a)
w = If/x (1O.16b)
which allows the momentum equation to be rewritten as
a[PoUu + p+ Polj/Uz] =0 (10.17)

dx
If the motion is periodic in x or if it vanishes as x~ infinity, then the quantity

inside the square bracket must itself vanish, so that
[PoUu+ p+ Polj/Uz] = 0 (10.18)
If this equation is multiplied by w,
POUWU+pW=_~dIf/2 dU (10.19)
2 dx POd;
where the relation between lJI and w has been used. An average over a wavelength in x
then yields the important result:
pw+ PoUuw=O (10.20)

From this it follows from the energy equation that
dUW =0 (10.21)
dZ
This implies that even if uw '" 0, its derivative with z must be zero if the wave field is
steady and there is no dissipation. Returning to the first equation of this lecture for the
mean flow, it also means that the wave field will not, under these circumstances, alter
the mean. This will occur only for that part of the wave field that has its quadratic prop-
erties varying with time, for example at the front of an otherwise steady wave field.
The relationship of the vertical energy flux and the Reynolds stress allows a simple
interpretive tool to characterize the sign of the energy flux. From pw + Pouw = 0, it
follows that
pW = -PoUuw = PoUlf/xlf/z
(10.22)
=PoUlf/x If/; =_Pou dZ ) u 2
If/z dX 'If
If the vertical energy flux is positive, this implies that the slope in the x-z-plane of
the lines of constant lJI (these are the phase lines of the wave) must be negative if U is
positive (Fig. 1O.2).
Now let's derive the governing equation for steady perturbations. By taking the
z-derivative of the x-momentum equation and subtracting from that the x-derivative
of the vertical momentum equation, we obtain
PoD[w x -u z ]- PoUzu x - PowzU z - PowU zz =-Pobx (10.23)
which is the equation for the y-component of vorticity. The term on the right-hand
side represents the baroclinic production of relative vorticity by horizontal density
gradients in the wave (this is the linearized part of Vp x Vp). With the aid of the con-
tinuity equation, this becomes
DV 2 lJ1 - lJIxUzz = -bx (10.24)
while the adiabatic equation is

- 2
Db = N lJIx (10.25)
>( de >c .. U
Fig. 10.2.
The orientation of wave crests
to yield upward wave radiation
Eliminating b between the two equations yields the final equation for the stream
function:
D V If/+N 2 If/xx= D- If/x Uzz

-2 2
(10.26)
We note that if U = 0, the equation reduces to the equation for internal gravity waves
a at
we obtained in the last lecture, since now i5 = I if U is zero.
However, we are interested in the case where U is not zero but where the flow and
the wave field are steady. In the steady state, i5 = Ua I ax,
and so the governing equa-
tion for 1fI can be written in the compact form
U2 _
axa2 [
2
(NU U
2 U
J1
V21f/+ _2_ -E.. If/ =0 (10.27)
For motion that is periodic in x (or which vanishes at infinity) we can integrate the
above equation twice to obtain
N2 -E..
V21f/+ ( __ U J
If/=O (10.28)
U2 U
For solutions that are periodic in x with wave number k, we can look for solutions
of the form
If/ = ¢(z )e ikx (10.29)
where it is understood that we take the real part of the solution.

Thus, ¢ satisfies
d2 ¢ +U 2(z)-k2)¢=0 (10.30a)
dz 2
£2 ={ ~: _U;z} (10.30b)
A particularly illuminating example occurs when we consider the flow over a bumpy
lower boundary, whose elevation is given by the periodic form
h =ho coskx
(which is why we chose k for the wave number of our solution).

The lower boundary condition is
w= dh =U ah (10.31)
dt ax
where the last equality depends on the solution being steady and linear.
We will imagine that the upper boundary is very far away and idealize that by con-
sidering that z runs between 0 at the lower boundary and 00 for large positive z.
Since w = alf// ax, an integration of the lower boundary condition in x yields
If/= Uh, z=0 (l0.32a)
or
Ij/=Uf!oRee ikx => ¢=Uf!o, z=O (l0.32b)
A glance at the differential equation for cp shows that the character of the solution
depends on whether
£2 >k2
or
£2 <k2
In the former case, the solutions will be oscillatory in z, while in the latter case they
will be exponential in z. For the simple case of constant U, this change in character has
a simple physical interpretation. In this case, £2 = N 2 / U2• In the frame in which the lower
boundary is fixed and the flow is moving to the right with speed U, the motion is steady.
Let's put ourselves in a frame moving to the right with the basic flow. Then the unper-
turbed fluid appears to be at rest, but it is being disturbed by a lower boundary with a
ripple of wave number k moving to the right at speed U. This will force a response with
the forcing frequency of the boundary disturbance, which is Uk. If that frequency is
greater than the maximum internal gravity wave frequency, N, we clearly can't have a
wavelike response in z. The condition that Uk < N is simply £2 = N2 / U2 > k2 •
Let us first examine the other case where £2 < k2•
In this case, the solution for cp is
¢=Ae- mz +Be mz
(10.33)
r
m=lk2-£2 ]112 >0
Now, for large positive z, we would like the solution to remain bounded, i.e., as
z ~ 00, we want cp to remain finite. This clearly requires that we choose B = O. The
remaining constant is determined by the condition at z = 0, namely,
A=Uf!o (10.34)
Thus,
Ij/=ReUf!oe-mZeikx =Uf!oe-mz coskx= Ue-mZh(x) (10.35)
from which we obtain

w = ¥'x = -Uhoke -mz sin kx (10.36a)
U= -¥'z = -Uhome -mz cos kx (10.36b)
¥'x¥'z = U2JJgmke -2mzsinkxcoskx = 0 (10.36c)
Note that the streamlines are in phase with the topography and simply diminish
exponentially with height above the bottom. Furthermore, the last equation tells
us that the Reynolds stress and hence the vertical energy flux are identically zero
when averaged over a wave period. That seems reasonable. In this parameter regime,
no internal gravity wave can be excited (the frequency is too large), and without a wave
response there is no upward radiation of energy. Note also that the vertical displace-
ment
( = ¥' = hoe -mz cos kx = h(x)e-mz (10.37)

U
is exactly in phase with the topography. Where the bottom goes up, the streamline
follows it.
Let's calculate the drag on the mountain by the pressure in the wave field. For a
bottom with a relief h = h(x),
ah
f p-dx
21tlk
Drag= (10.38)
o ax
That is, the drag is the pressure times the projection of the topography that presents
a face perpendicular to the x-axis. There will be a drag if there is higher pressure
on the face of the slope upstream compared to the pressure on the face downstream.
We can easily calculate the pressure from the steady zonal momentum equation when
dU I dz = O. In that case,
Px=-PoUU x => P=-PoUu=PoU¥,z (10.39)
from which it follows in the present case,
p =-poU2mh(x)e- mZ (10040)
Note that for this case, the pressure is in phase with the topography. The lowest
pressure (the largest negative anomaly) occurs over the ridge crest, and the pressure
is symmetric about the crest; indeed,
21tlk 21tlk
Drag= fp(x,0)h xdx=-poU2m fhhxdx=O (10041)
o 0
Thus, in this parameter regime there is no wave radiation and no drag on the to-
pography. The absence of drag is not too surprising. We know that for a homogene-
ous, incompressible, irrotational flow in the absence of friction there is no drag. Clearly,
if Uk is much greater than N, the fluid will respond as if the stratification were zero
and the zero drag result of potential flow is anticipated. It is perhaps a little strange
that this result holds up to the equality (at least) where Uk = N, but as we shall see, the
drag is directly related to the ability of the flow to support a wave, and the threshold
Uk = N is precisely that boundary between wave and no wave.
Now let's consider the more interesting case when Uk < N. This puts the fluid in the
parameter regime in which internal gravity waves can be generated. In this case, we define
2
m 2 =£2 _k2 =~_k2 >0 (1Q.42)
U2
so that the solution for the wave is
¢(z) =Ae imz + Be- imz (1Q.43)
We still have the boundary condition at z = 0 that
¢=Uho, z=O (10.44)
However, the condition that the solution be finite at infinity is no help at all in re-
jecting either the A or B solution for cp, so that with the boundary condition at z = 0,
we will have one condition (equation) for the two unknowns, A and B. How did we get
into this pickle?
The essence of the difficulty is related to the two infinities we have introduced into
our problem by our simplifications. First, we have assumed that since the upper bound-
ary is so far away, we may idealize the region as infinite in z. Of course, for waves radi-
ating upward, that will hold only for a finite time. Second, we have decided to exam-
ine the steady problem after all transients have radiated away and that requires that
in principle, an infinite amount of time has passed so there is clearly a conflict be-
tween the two assumptions. Had we solved the initial value problem, i.e., if we had
considered the problem for finite t while the z-domain was infinite, it would be clear
that since the group velocity is finite, for all finite t the disturbance should go to zero
as z goes to infinity. However, in the problem what we have done is to invert this limit.
By examining the steady problem, we let time go to infinity first and then we must ask
how the solution looks for large z. Stated this way, it is clear that we are asking for the
partial solution that is valid within the advancing front that was set up long ago by
the initial disturbance. Inside that front the solution may be steady, but without con-
sidering the initial value problem, there seems no way, with the boundary conditions
so far applied, to specify the steady solution (find A and B). Because we have chosen
to solve a simple, physically incomplete problem, we must add some physics to take
the place of the initial value problem we have chosen not to solve (because it would be
so complicated). Indeed, it would be a pity to have to go through the whole initial value
problem just to determine which combination of A and B is correct in the steady state.
The physics that we must add is called the radiation condition. Simply put, it states
that we must decide on the direction of the wave radiation in the steady state that is
physically pertinent for our problem. If we consider the problem as one in which a
disturbance is formed at the topography by its interaction with the current and then
radiates upward, the condition should be that the radiation be outgoing radiation;
that is, the energy flux in the z-direction should be positive.
We have seen before that the vertical energy flux is
pw =PoUlf/xlf/z (10.45)
where again an overbar denotes an average in x over a wavelength.

With the above solution for rp, we have
If/= Aei(kx+rnz) + Bei(kx-rnz) (10.46)
The part of the solution with coefficient A has its phase lines tilting upstream (for
U> 0), while the B solution has its phase lines tilting downstream (Fig. 10.3).
We saw earlier that the energy flux would be upward for phase lines tilting upstream
so we would expect the A solution to be the one that gives us the physically acceptable
solution of outgoing radiation. Let's calculate the energy flux explicitly.
With pw = Po UlI'xll'z , we first calculate the energy flux in the A solution. Recall that
we must use the real part of the solution. Thus, using" to denote complex conjugation,
If/= A ei(kx+rnz) +£e-i(kx+rnz) (10.47)

2 2
so that with ()= kx + mz
pw =PoUlf/xlf/z = ikpoU [A iO + A *e- iO ] im [Ae i8 _ A *e- i8 ]

2 2
- mkpoU [ _ 21AI2 + A 2e 2i8 + A *2 e-2i8 ] (10.48)
mk 1 12 >0
=-PoUA
2
It is important to note that in calculating the quadratic product of pw, we had to
use the real part of lI', which involves both the linear solution and its complex conju-
gate. Had we erroneously used only the term proportional to eie, we would have ob-
tained only terms like eM from the product, and these have zero average over a wave-
length. The correct answer given above, whose average is different from zero, depends
on using the full real parts to calculate the flux terms.
Fig. 10.3.
The tilt of the wave crests in
the two solutions. The
A solution has energy
propagating upward; the
B solution has energy
propagating downward "A" solution "8" solution
If we were to make the same calculation with the B solution, the result would be
- mk 2
pw = --PoUIBI < 0 (10.49)
2
Since by convention both m and k are positive, it is the A solution that represents
outgoing radiation, and the B solution represents incoming radiation from infinity.
Consistent with our physical description, it is the A solution which is appropriate to
our problem, since that is the solution that satisfies the radiation condition of out-
going energy flux. It is important to realize that the B solution is a perfectly fine so-
lution physically. It represents, in the context of the steady problem, wave energy mov-
ing downward from some source far away and above the topography. There may in-
deed be problems in which that radiation would be appropriate, but our specification
of outward radiation is the additional condition that we must add to render our solu-
tion both unique and appropriate to the physical problem we have in mind.
Using the condition that z = 0 yields A = Uho so that
Vf=ReAei(kx+mz) =Uhocos(kx+mz) (10.50)
this yields, as expected, phase lines tilting upstream (Fig. 10.4).

From the earlier result (Eq. 10.39),
p =-PoUu= PoUVfz =-poUhomsin(kx+mz) (10.51)
Thus on z = O,p = -poUhomsinkx, which is now 90° out of phase with the topogra-
phy. High pressure now occurs on the upstream face of each ridge (where x is -reI 2)
while the downstream face has low pressure. This leads to a net force on the topogra-
phy so that (Fig. 10.5)
kx + mz = constant
/
Fig. 10.4. A schematic of the solution satisfying the radiation condition. The phase lines tilt against
the current
p high plow
/~~
"'-,. ~~~""=7 "Ix)
PIx)/
Fig. 10.5. The pressure distribution with respect to the topography for the case where waves are radiated
dh
Drag = fo p-dx
2rrlk
dx
km
= PoU hQ-
2
2 2
(10.52)
Note that for km > 0, the correct choice of the phase orientation, the drag is posi-
tive. This drag has nothing to do with friction; it is simply the response to the wave
energy that is radiated away to infinity by the topography. Indeed, in the case where
there was no wave radiation, there was no drag. Note the relation between the wave
drag on the topography and the flux of energy; using our result for the amplitude A,
- 32mk
pW=PoU hQ-=DragxU (10.53)
2
so that the rate at which energy is radiated away from the topography is precisely equal
to the rate at which the drag is doing work on the topography and thus is equal to the
rate at which the topography is doing work on the fluid.
There have been other techniques introduced to deal with the apparent indetermi-
nacy of the solution for which we have used the radiation condition. An alternative is
to introduce a small amount of friction, for ease, proportional to the velocity, and re-
calculate the constant m. With the presence of friction, m will be complex and one
solution will exponentially increase with z, while the other decreases. Choosing the
solution with the exponential decrease and then letting the size of the friction go to
zero reproduces the solution found here by the radiation condition. The student should
think through the physical reason for why this is true, and those with a background
that includes the Laplace transform should also see why it is equivalent to the steady
state chosen by the initial value problem.
We noted above that the choice of solution by the radiation condition is equivalent
to choosing the solution that has the energy flux upwards. Let's spend a moment re-
viewing how that would enter explicitly in the steady problem.
Let prime variables denote velocities and positions seen by an observer moving to
the right at the speed U. For such an observer, there will be no mean flow and as al-
ready mentioned, that observer will see a rippled lower boundary moving to the left
with speed -U forcing the fluid with a frequency Uk.
lO4 Lecture 10 . Vertical Propagation of Waves: Steady Flow and the Radiation Condition
To move into the primed frame, we introduce the relations
x = x' + Ut'
(l0.54)
t= t'
so that the phase in the resting frame ei(kx-wt) becomes ei(kx'+[kU-m]t') in the moving
frame. We define the intrinsic frequency as the frequency seen in the frame moving
with the mean flow so that it is the frequency with respect to a locally stationary fluid:
% =m-kU or (l0.55a)
m=Uk+% (l0.55b)
Thus, in the original frame in which the lower boundary is at rest, the frequency is
the sum of the intrinsic frequency plus the Doppler shift Uk. We know that the in-
trinsic frequency for internal gravity waves, i.e., in the absence of a mean flow, is
Nk
r.l_-+ ~ (l0.56)
~u - - vk2+m~
In the case under consideration, we are taking into account a disturbance which is
steady in the rest frame so that the frequency ill must be zero. This means we must
choose the negative sign in the above equation for the intrinsic frequency so that the
Doppler shift downstream cancels the phase propagation upstream.
The condition of zero frequency chooses the wave number component m such that
N 2_ k 2
m2 = _
U2
a result already obtained from our differential equation for 1/>. Although the frequency
is zero, its derivative with respect to wave number is different from zero. Therefore
the group velocity in the resting frame is
Nm 2
cgx
-- am
ak =U - (k 2 +m
2)3/2 (1O.57a)
Nkm
c
gz (2
k +m
2)3/2
(1O.57b)
Using the above results for m yields
Uk 2
cgx=-2- (l0.58a)
K
Ukm
cgz =-2- (l0.58b)
K
so that the direction of the group velocity is downstream and upwards even though
the phase lines are tilting upstream.
Thus,
Cgz =m (10.59)
cgx k
so that cg is parallel to the wave vector K. In the frame in which there is no mean flow,
our earlier results (or a recalculation using the intrinsic frequency) yield
Nm 2
C
gxo
=---
K3/2
(1O.60a)
Nkm
C
gzo
=--
K3/2
(10.60b)
so that in the frame of no mean flow, as previously noted, the group velocity is at right
angles to the wave vector
Cgzo k
(10.61)
cgxo m
The relation between the two is illustrated in Fig. 10.6. It is a simple matter to show
that the two group velocities are orthogonal to one another; that is, cg ' Cgo = O.
The difference between the two group velocities is precisely equal to the mean flow
in the x-direction U, which carries energy of the wave field downstream. The two wave
vectors and their relation to the mean flow are shown in Fig. 10.7. A very clear discus-
sion of the relationship is to be found in Lighthill (1978).
A final remark is in order about the relation between the drag and the effect on the
mean flow. Since the drag represents a force on the fluid by the topography, one ex-
Fig. 10.6.
The wave vector and the
direction of the group velocity
for the steady, radiated internal
gravity wave
o
Cgo Cg
Fig. 10.7.
The relationship between the
group velocity vector for the
steady wave and the wave in a
resting medium Um 2 //f Uk 2 //f
pects that the mean flow will decelerate as a consequence. However, we earlier noted
that for steady waves in the absence of dissipation,
auw =0 and since

az
au auw
-=---
at az
where the right-hand side is calculated from the wave field, there is apparently no
deceleration of the mean flow. How is this apparent paradox resolved? (Note: usually
a paradox is a sign of incomplete thought, not an out and out error).
As mentioned above in our discussion of the radiation condition, there will be an out-
going front, ahead of which there is no wave signal, behind which we will have established
the steady wave field of our calculation. At a time t, the front will have moved to a
distance cgzt above the topography into the fluid. At the position of the front, the above
momentum balance holds. So ahead of the front uw = 0 while behind the front (Fig.1O.8)
uw=u 2 hii mk
2
Thus,
auw <0
az
and is large at the front. It is at the front that the deceleration occurs, and only there.
Integrating across the front yields a local change in the mean
flu= U 2 hJmk -uhJK2/2

2Cgz
Note that the change is of the order of the small parameter (h oK)2, which is the
square of the steepness of the topography.
uw = 0 . - - - Front
...................................................................................................................................................................................................
uw=llh~mk
2
r
cgzt
1
Fig. 10.8. The waves radiated from the topography and the Reynolds stress developed behind the
advancing front
Lecture 11
Rotation and Potential Vorticity
For motions whose time scales are of the order of a day or greater, or more precisely
when the frequency of the wave motion is of the order of the Corio lis parameter or
less, the effects of the Earth's rotation can no longer be ignored. Such waves are evi-
dent in both oceanic and atmospheric observational spectra. Figure 11.1 taken from
the article of Garrett and Munk (1979) shows a power spectrum of vertical displace-
ment of an isotherm. We see a great deal of variance at frequencies less than N (as we
might expect) with a peak near the Coriolis frequency f = 2.Q sin O.
Some waves, such as gravity waves, are affected by rotation while others are prima-
rily due to rotation, and of these there are different types with different characteristic
time scales.
Consider first an unbounded fluid. To simplify the analysis, we will start by assum-
ing that it is incompressible, inviscid and uniformly rotating. We will also assume that
the vertical scale of the motion is much smaller than the scale over which the density
1031~~r----"----~
102
10'
.r::.
c.
u
~
E 10°
10-'
10-2
Fig. 11.1.
A power spectrum of internal
wave energy (after Cairns and 0.Q1 f 0.1 1.0 N 10
Williams 1976) cph

108 Lecture 11 . Rotation and Potential Vorticity
would change by an 0(1) amount, i.e., we assume that if m is the vertical wave num-
ber of the motion,
1 dPo «m (1Ll)
Po dZ
For small perturbations we write, as before,
Ptatal = Po + P , P « Po (11.2a)
Ptatal = Po + P , P « Po (11.2b)
dPO =-Pog (11.2c)

dZ
Then the linearized equations of motion for a fluid whose rotation axis is anti-par-
allel to the direction of gravity are
p [dU _ tv ] = _ dp (11.3a)
o dt dX
Po [dV ] dp
at+ fu =- dy (11.3b)
dW dp
Po-=---pg (11.3c)
dt dZ
dU dV dW
-+-+-=0 (11.3d)
dX dy dZ
db 2
--wN =0 (11.3e)
dt
b =gp / Po' N 2 = -g- dPo

-- (11.3f)
Po dZ
Our goal is to derive a single equation for either the pressure or vertical velocity to
serve as our wave equation. To start, we take the x-derivative of the y-momentum equa-
tion and subtract from that the y-derivative of the x-momentum equation; that is, we
are taking the curl of the horizontal momentum equations to obtain
dW
dS =-f(ux+vy)=fa; (11.4)
dt
Note that the Coriolis parameter in this study is assumed independent of position,
an assumption that will be relaxed later in the course. The component of vorticity
parallel to the z-axis is
Lecture 11 . Rotation and Potential Vorticity 109
(=v x -U y (11.5)
It is important to note that in the presence of the background rotation, z-de-

pendent vertical motion will give rise to relative vorticity by stretching the vortex
lines associated with the overall rotation of the fluid. If instead of taking the curl
of the horizontal momentum equations we take the divergence, Le., the x-derivative
of the x-momentum equation and the y-derivative of the y-momentum equation,
we obtain
:t (u + y)- f (
x v =- V~p
Po
( 11.6a)
or
a2w + f(= V~p (11.6b)

ataz Po
If we eliminate w between the vorticity and the divergence equation,
a2 ( + l( =f V~p (11.7)
at 2 Po
while eliminating the vorticity between the same two equations yields
~ aw + f2 aw = f V~p (11.8)
at az2 az Po
If the perturbation is independent of horizontal position, the right-hand side of

the above equation would be identically zero, and that would give rise to a harmonic
oscillation at the Coriolis frequency f This is analogous to the situation we saw for a
nonrotating fluid in which disturbances independent of z gave rise to oscillations with
frequency N. In the latter case, we discovered N as a limiting frequency of oscillation.
We shall discover the same thing in the presence of rotation for oscillations with
frequency f
If we take the time derivative of the vertical momentum equation and use the adia-
batic equation to eliminate the buoyancy b,
a2 w 2 1 a2 p
-+Nw=---- (11.9)
at 2 Po ataz
so that again, disturbances independent of z will oscillate with frequency N. We now
have two limiting cases to keep an eye on. If we eliminate the pressure between Eq. 11.1
and Eq. 11.2, we obtain a single equation for w (take the horizontal Laplacian of Eq. 11.2
and the vertical derivative of Eq. 11.1):
a rn2 w 1+ f2 --+
2
-Lv a w N 2n2 2
-0 (11.10)
at az
VhW-
2 2
When f is zero, this recovers the wave equation for internal gravity waves studied
earlier. To derive this equation, we have used repeatedly the approximation that the
vertical derivative of the background density is small with respect to the vertical de-
rivative of w, i.e.,
1
---«--
apo 1 aw (I 1.12)
Po az waz
Before proceeding to the solution of the wave equation for w, it is illuminating to
examine the equations of motion a bit more carefully.
First of all, note that if we knew the pressure, we could easily find the horizontal
velocity uh from the easily derived equation
a2Uh 2- 1 a f
-2-+ f Uh =---Vhp+-zxVp
A
(I 1.13)
at Po at Po
z
where is the unit vector in the z-direction parallel to the rotation (and anti-parallel
to gravity).
Second, we can find an equation for the pressure by eliminating w between Eq. 11.8
and Eq. 11.9 instead of the other way around. We obtain
-a [_V2p+
a2 a2
f2~+N2V~p ] =0 (I 1.14)
at at 2 az 2
This is almost the same equation we obtained for w. There is an extra time deriva-
tive operating on the whole equation. For motions of nonzero frequency that would
make no difference, but we must be careful. A first integral of the equation yields
a2 a2
_V2p+ f2~+N2V~p=.Q(x,y,z) (I 1.15)
at 2 az 2
The question is what is Q? Its existence is connected with the conservation equa-
tion, whose first integral is the above equation. It is also clear that it ought to be de-
termined by initial data since it is independent of time. Clearly, then, it should be re-
lated to some quantity that is conserved during the motion.
If we return to the vorticity equation
as- =f aw (I 1.16)
at az
and use the adiabatic equation to eliminate w, we obtain
at
-
~(S- f a(b/ N 2 ))=0
az
(I 1.17)
This is, for the simple model we are considering, the form of the conservation of
potential vorticity q, where
q=( _ fdblN 2 (11.18)

dz
this form can be easily interpreted. Let's call (to avoid confusion with the symbol for
vertical relative vorticity) the vertical displacement of a fluid parcel, Z. From the adia-
batic equation, Z = b I N 2 so that
q=( _ f dZ (11.19)
dz
If the potential vorticity (pv) is conserved, the spreading apart of density surface,
i.e., dZ I dz > 0 in the presence of background planetary vorticity, f, will give rise to a
corresponding increase in relative vorticity to keep q constant. This should be famil-
iar from simple layer model treatments of pv.
It seems likely that our conserved quantity Q is related to q. Can we show what the
relation is? To simplify the derivation, we will take N constant.
From
f(= V~p (11.20a)

Po -wzt
Wzt =bzzt I N 2 (11.20b)
V~p _l_~~( _ pz -Wt)

=> f( =-;;;- N dz dt l Po
2 2
(11.20c)
= V~p
--+
l_pzztt +_l-w zttt
2 N2 (11.20d)
Po N Po
and
-f 2bz IN 2 = -f22 (-pzz-w zt ) (11.21)

N
it follows that
bz } 2 I I l f2
f { ( - f -2 =VhPlpo +-2-Pzztt +-2 Wzttt+-2PzzIPo+-2 Wzt or (11.22)
N Npo N N N
2
fq=VhPI l pzzl Po +-2
Po +-2 I 2)
- Wttz + f Wz
d(
N N dt
2 l I d2 (2 ) (11.23)
=VhPI Po +-2 pzzl Po +-2 -2 VhPI Po
N N dt
_ -I - [d 2V 2p f2 d 2p N2n2 ]
- ---+ - + vhP
PON 2 dt 2 dz 2
so that comparing with the earlier equation, we have, finally,
Q=(PofN 2 )q (11.24)
Thus, the conserved quantity in the wave equation for the pressure is a simple
multiple of the potential vorticity. This has very important consequences:
1. Since aq I at = 0, the initial data that gives the pv is sufficient to determine Q, and it
remains unaltered throughout the motion;
2. The oscillating part of the wave field has no potential vorticity. This also follows
from the conservation of potential vorticity, since if the motion is periodic, the
conservation equation aq I at = 0 becomes simply mq = O. If the frequency is not
zero, the potential vorticity must be zero.
Therefore, the pressure and velocity field may be divided into two parts. There is a
wave part that carries no potential vorticity and a steady part, which is a steady par-
ticular solution of the P Eq. 11.3.
Let's write the total solution for p, b and the velocity:
p= Pg (x,y, z) + Pw(x,y,z,t)
b= bg(x,y, z) +bw (x,y,z, t) (11.25)
U= ug (x, y, z) + Uw (x, y, z, t)
where the g subscripted variables are independent of time and the w subscripts refer
to the wave-like part of the motion.
For the steady, linear part of the motion, the balances are
1 apg
fV g = Po ax (11.26a)
1 apg
fUg =- Po ay (11.26b)
Wg=O (11.26c)
apg
bg =- az (11.26d)
that is, for the steady part of the solution, the horizontal velocity is in geostrophic and
hydrostatic balance, the vertical velocity is zero and most importantly,
abg 1N 2 =_V2p
q=Sg-f-
1 f a2pg
+ ___ (11.27)
az
Pof h g N 2 2 az
(Note, it is because the geostrophic W is zero that the wave equation for w does not
contain the extra time derivative that the pressure equation does, because there is no
nontrivial steady solution for w).
Thus, if the potential vorticity is determined by the initial conditions, the above
elliptic equation for the geostrophic pressure completely determines the steady part
of the solution, since once the geostrophic pressure is determined, both the steady
density perturbation and the geostrophic horizontal velocities can be calculated from
the geostrophic pressure. Note that the vertical velocity in the steady geostrophic so-
lution is zero as a consequence of the steady form of the adiabatic equation. Thus, the
steady geostrophic part of the solution can be determined independently of the wave
part, in terms of the initial value of the potential vorticity.
Since the wave field carries no pv, the wave part of the pressure field is determined
from the homogeneous part of Eq. 11.3; thus,
a 2
_V2p a
+f2~+N2V2p =0
2
(11.28)
at2 w az 2 h w
The initial conditions on the wave pressure must satisfy that part of the initial pressure
field, which contains no pv. Thus, if the total initial perturbation pressure is P;otal (x, y, z, 0),
Pw(X,y,z,O)= P;otal(X,y,Z,O)- Pg(x,y,z) (11.29)
We will see later how to exploit the potential vorticity conservation for the initial
value problem, but now let's return to the wave problem. Let
w =Woei(kx+ly+mz-mt) (11.30)
substitution in the wave equation yields the dispersion relation,
2 k 2 12
oi=f2 m +N2 + (11.31)
k2 +12 +m2 e +12 +m2
Note that if m were zero, the frequency would be N, while if k and I were zero, the
frequency would be f. Let Kh be the magnitude of the horizontal component of the wave
vector, i.e., Kh =..Jk2+12 ,such that (see Fig. 11.2)
sine= ml K (11.32a)
cosO=Khl K (11.32b)
r
K=le+12+m2 ]1/2 (11.32c)
so that
oi = f2 sin 20+ N 2cos 20 (11.33)
Since
N 2 -oi =N 2 -lsin2 O-N 2cos 2 0
=(N 2 -l)sin 2 0 (11.34)
oi -l =(N 2 - f2)cOS 20
~ y
Fig. 11.2. The three-dimen-

sional wave vector, whose
orientation determines the
frequency of the inertial-
gravity wave x Kh
and since almost everywhere in the ocean N» f, it follows that the frequency of the
plane wave will satisfy f::; co::; N, which explains the rapid fall off of the spectrum of
observed internal waves at N. The observed peak at fis related to the geometrical prop-
erties of the forcing. Usually, the horizontal scale is much greater than the vertical scale.
In this case, a convenient way to write the frequency relation is
2 2
co2 = f2 +(N 2 _ f2) Kh '" f2 + N 2 Kh
K~ +m2 m2
so if Kh « m, the excited frequencies will be close to f

With the dispersion relation for the frequency, the student should check that the pv
in the wave is zero by calculating both the relative vorticity and b, as well as forming q.
Now let us orient the coordinate system so that the wave vector lies in the x-z-plane.
This makes the y-wave number zero. Note that velocity in the y-direction will be dif-
ferent from zero, since
dV w =-fuw
dt
if 1is zero.
Then
2
co2 = f2_m__ + N 2 _ __
e (11.3Sa)
m 2 +k2 m 2 +k2
co dCO = N 2km 2 / K4 _ f 2km 2 / K4 (l1.3Sb)

dk
so that
am = [N2 _ f2 ]m2k (11.36a)

ak K4
and similarly,
am =_[N2 _ f2] k 2m (11.36b)

am K4
Thus the components of the group velocity are
(N 2 - f2)
(cgx,C gz )- " mk(m,-k) (11.37)
mK
Once again, the vertical component of the group velocity is opposite to the vertical
phase speed
m
m Cgz <0
and the group velocity is perpendicular to the wave vector (because, again, with an incom-
pressible fluid, a three-dimensional plane wave has its fluid velocity orthogonal to the wave
vector, and the group velocity will be in the direction of the energy flux vector pu).
In the limit where k « m, the dispersion relation is
m 2 =f2+N 2 _k
2
(11.38)
m2
and waves with such frequencies are called Poincare waves.
Now that w or the wave pressure is determined, it is easy to calculate each of the other
velocity components and the density perturbation so that the energy in the plane wave
may be found. The student should check whether equipartition between kinetic and
potential energy obtains for the internal gravity waves in the presence of rotation.
If
w = Wocos(kx+mz-cot)
it follows that
p _ (N 2 _ f2\V;0 cos(kx+mz-cot)
-- (11.39a)
Po mm
m m
u =--w =--Wo cos(kx+mz-cot) (11.39b)
k k
av = - fu ~ v =f -m- Wo sm(kx+mz
-
.
-cot) (11.39c)
at mk
b=-(N 2 /m)Wosin(kx+mz-0Jt) (11.39d)
Hodograph of horizontal velocity
0.6
0.4
0.2
0.0
-0.2
-0.4
-0.6
-1.0 -0.8 -0.6 -0.4 -0.2 0.0 0.2 0.4 0.6 0.8 1.0
f! ro = 0.5
Fig. 11.3. The ellipse traced out by the position of the horizontal velocity vector during one wave period
Note that the relations for u and v imply that the horizontal velocity vector will rotate
in the horizontal plane clockwise with time for 0) > 0 and m > 0, i.e., for downward
energy propagation. Indeed, the horizontal velocity vector traces out an ellipse
(Fig. 11.3)
2
u 2 + _v_ m2
f 2/
_ 2
(11.40)
2 - Wo-~2
0) k
whose major axis is along the x-axis and whose minor axis, along the y-axis, is smaller
by a factor flO).
Finally, note that if the fluid is contained between two lateral boundaries a distance D
apart, the equation for the normal modes in that region will be (with the rigid lid ap-
proximation)
w = W(z)ei(kx+ly-ax) (l1.41a)
d2 w 2 (N 2 _0)2)
(l1.41b)
-2-+Kh 2 2 W=O
dz (0) -f)
W=O, z=O,-D (11.41c)

p(z) ----. N(z) ----. Z(z) ----.

o e- 1NoO.75NoNo m=0.75No
r -j=l
-4
I I
-2
I I
2I
Po f
Fig. 11.4. Normal modes for three frequencies. The density distribution and the distribution of N(z)
are shown in the first two panels on the left (redrawn from Garrett and Munk 1976)
The mode shapes will depend on the frequency, and the discussion of the domains
of oscillation and evanescence in z are similar to the nonrotating case. Figure 11.4, taken
from the previously referenced article by Garrett and Munk (1976), shows some ex-
amples.
Lecture 12
Large-Scale Hydrostatic Motions
For many motions in both the ocean and the atmosphere, the horizontal scale far ex-
ceeds the vertical scale of the motion. For example, motions in the ocean occurring in
the thermocline will have a vertical scale of a kilometer or less, while the horizontal
scales might be of the order of hundreds of kilometers. Motions in the ocean induced
by traveling meteorological systems will have such large scales. If the motion has such
disparate scales in the vertical and horizontal, we can expect important influences on
the dynamics. First of all, we would expect that the vertical velocity will be small com-
pared with the horizontal velocity, since the motion consists of nearly flat trajecto-
ries. That in turn could mean that the vertical acceleration is small. Such dynamical
consequences often allow simplifications to our treatment of the physics, and we are
always looking for such simplifications so that we can make progress with more diffi-
cult problems; not just make life easier for ourselves.
We need to define clearly what "small" means in a dynamical context. As an example,
let's review the results of the plane internal gravity wave with rotation. As we saw in
the last lecture, if
W =Wo cos(kx + mz - 0Jt) (12.1)
then the pressure and horizontal velocity is
2 2)
-P (N -OJ W;0 cos(kx+mz-OJt) (12.2a)
Po mOJ
m
u = - - Wo cos(kx + mz - 0Jt) (12.2b)
k
Therefore, the ratio of the vertical acceleration to the vertical pressure gradient is
POw t
pz
=o( 2
OJ2 2 )
N -OJ
h'l e OJ2
W 1
=N2~2 + f2 m 22
K K
or (12.3)
POW t =o(t +(N 2 - t)e /K2)

pz 2 2
(N - f2)m / K2
(12.4)
e f2
=_+_K 2 /m 2
m 2 N2

120 Lecture 12 . large-Scale Hydrostatic Motions
Now consider the case where the horizontal scale is much larger than the vertical
scale of the motion. This implies that k « m, and so K2 I m 2 "" 1. Since kIm is small and
in oceanographic settings fl N is small, we see that the vertical acceleration is small
compared to the vertical pressure gradient. Later we will see more directly how this
comes about by scaling the equations of motion, but here we can see it from the solu-
tion of our problem. This implies that to
o[(~r.~l.
the vertical acceleration in the vertical momentum equation can be ignored compared
with the vertical pressure gradient so that to this order, that equation is replaced by
the hydrostatic approximation
ap
az =-pg (12.5)
for the perturbation as well as the mean resting state. That is, the motion has such a
weak vertical acceleration that although the fluid is in motion, the pressure can be
calculated from the hydrostatic equation as if the fluid were at rest. This is character-
istic of motions whose horizontal length scales for the motion are large compared to
the vertical scales of the motion.
However, note that although w is small with respect to u,
aw =O(mWo) (12.6a)
az
au =O(Wok(mlk»=o(aw) (12.6b)
ax az
so that w cannot be neglected in the continuity equation. The vertical velocity is small
with respect to the horizontal velocity, but the fast derivative in z in comparison with
x compensates.
We will examine such hydrostatic motions and waves starting with a simple ho-
mogeneous model.
Potential Vorticity: Layer Model
Consider a layer of inviscid fluid with, to begin with, a flat bottom and a uniform
density p. The fluid is rotating with a constant angular velocity Q - fl 2, whose axis is
opposite to the gravitational force (see Fig. 12.1).
This is meant to be a model for a small segment of the ocean, whose lateral scale,
while being much greater than the depth, is small enough so that the dynamical ef-
fects of the Earth's sphericity can be ignored. We therefore use Cartesian coordinates.
Further, it is convenient to define P =pIp in terms of which the linearized equations
of motion are, when the hydrostatic approximation is used,
Lecture 12 . Large-Scale Hydrostatic Motions 121
Q
p=Pa
1- p o
~ z=-o
Fig. 12.1. A shallow layer of fluid rotating and of constant density
du _ jv=_ dP (12.7a)
dt dx
dv dP
-+fu=-- (12.7b)
dt dy
dP
O=-g-- (12.7c)
dz
du + dv + dw =0 (12.7d)
dx dy dz
From the hydrostatic equation, it follows that
P =g(7J-z)+ Pa(x,y,t) (12.8)
where ry is the free surface height and Pa is the atmospheric pressure field at the free
surface. Using the result of the calculation of the pressure, the horizontal momentum
equations become (subscripts for derivatives)
ut - fv = -g7Jx - Pax (12.9a)
vt + fu = -g7Jy - Pay (12.9b)
Since the forcing terms on the right-hand side of the momentum equations are in-
dependent of z, it is consistent to look for solutions for u and v that are also indepen-
dent of z. This allows us to integrate the continuity equation immediately to obtain
D(u x + vv) + w(h) - w(-D) = 0
Since w is zero at the bottom (z = -D) and is equal to dry I dt at the free surface, the
equation for mass conservation is simply
d7J +D(ux+vy)=O (12.10)

dt
122 Lecture 12 . Large-Scale Hydrostatic Motions
The horizontal convergence of velocity times D yields the horizontal conver-

gence of volume. This must be compensated for by an increase in the free surface
elevation TJ.
The elimination of the pressure from the horizontal momentum equations leads
to the vorticity equation
(=vx-u y (12.l1a)
f a"
a( =-f(u x +v y )= D at (12.l1b)
at
or
~[(
at
-L,,]=o
D
(12.l1c)
which is the statement of potential vorticity conservation for the linear, single layer
model when f is constant. We define the pv as
q=(_L"
D
(12.12)
To obtain a wave equation for disturbance, we start by taking the divergence of the
horizontal momentum equations:
:t [u +Vy]-
x j( =-gV 2 ,,-V 2 Pa (12.13)
which with the continuity equation yields
D a2
1 at" -j(=-gV 2 ,,-V 2 Pa
2 (12.14)
The vorticity can be eliminated from this equation with the aid of the equation re-
lating the vorticity to the potential vorticity so that
a2 __
v2,, _ _1 --.!1. f2 f
,,=V 2 Pa/g--q (12.15)
c~ at 2 c~ g
where Co = -vgI5 is the gravity wave speed for long waves in a nonrotating fluid.
Note that the potential vorticity is, by the conservation statement, independent of
time. Thus, once again we can separate the solution for TJ into a steady part (which will
be in geostrophic balance) and an unsteady part associated with the waves and which
will carry no potential vorticity. The geostrophic part of the field will absorb the con-
sequences of the initial distribution of potential vorticity, while the remainder of the
initial conditions, the part containing no pv, will radiate away as gravity waves. Again,
the steady part can be calculated independently of the wave part so that the final steady
state after the wave has radiated away can be calculated independently of the time-de-
pendent wave problem. l Historically, this has given rise to a set of interesting adjust-
ment problems, starting with the classical paper of Rossby (1938) (see also Gill 1982).
In these investigations, the question is asked: suppose we start with an initial distribu-
tion of velocity and free surface elevation not in geostrophic balance. How does the fluid
adjust to eventual geostrophic balance, and what is the final geostrophically balanced
state? The first part of the question requires the solution of the wave radiation problem
(not easy), while the second part, the ultimate geostrophic state, is very easy because of
the conservation of pv. We will give a classical example here to see how this works. It
should be clear that the process is of general application when pv is conserved.
Rossby Adjustment Problem
Consider a layer of fluid in which at time t = 0, a slab of fluid occupying the range
between a and -a is set into motion with a uniform velocity U along the x-axis, and at
the same initial instant the free surface elevation is zero (Fig. 12.2). Suppose the at-
mospheric pressure forcing is zero. We would expect, somehow, that eventually the free
surface will deform, producing a pressure gradient in the y-direction to balance at least
part of the initial x-velocity. The question is of the original motion: how much ends
up in steady geostrophic balance and how much of the original energy is radiated away
in the form of gravity waves? Thus,
v=1]=O, t=O (I2.16a)
IYI<a
u={~ IYI>a
(I2.16b)
/
/
//
./
./
.
./
u=O u=U //// u=O
./
Y /./
./ ",
k";
o -0
Fig. 12.2. The initial condition in a homogeneous layer of fluid before geostrophic adjustment
1 We are assuming that the atmospheric pressure forcing has no steady part. Otherwise it is easy to
show that the response to such forcing is just an inverted barometer response in which the velocity
is zero and 1/ = -Pal g, a rather dull solution from the point of view of wave dynamics.
The potential vorticity, q, at t = 0 can be easily calculated. The initial free surface
height is zero, so the only contribution to the potential vorticity comes from the rela-
tive vorticity:
au [J(y+a)-J(y-a)]
s=--=-U (12.17)
ay
where 8(x) is the Dirac delta function. It is zero except where its argument is zero,
where its value is infinite and has the property that its integral over the origin of its
argument is one. It is the derivative of the step function H(x). Since the original zonal
velocity can be written as the sum of two step functions
u =U[H(y+a)-H(y-a)]
(see Fig. 12.3) the result for the vorticity follows directly.
The potential vorticity is thus
q =(s _Lry]
D t=O
=-U(J(y+a)-J(y-a)) (12.18)
0.7
0.6
0.5
0.4 ~
uy = 8(y+ a) uy = -8(y- a)
0.3
~
0.2
0.1
0.0
-2 -1.5 -1.0 -0.5 0.0 0.5 1.0 1.5 2.0
U(y) a = 1 U = 0.66
Fig. 12.3. The initial velocity as a function of y showing the two delta functions in the vorticity
In the steady state in which the final adjusted state is to be calculated,
V217-f: 17=- uf [J(y+a)-J(y-a)] (12.19)

Co g
Note that the problem is forced entirely by the distribution of pv. Since we expect u
and TJ to be functions only of y in the final state, the equation can be simplified to
f2 Uf
17yy - - 2 17=--[J(y+a)-J(y-a)] (12.20)
Co g
which must be solved on the infinite y-interval between plus and minus infinity. Since
the delta functions are zero except at the zeros of their arguments, the right-hand side
of the above equation is zero except at the two points ±a. Note that what we are really
doing is finding the Greens function response for TJ to two point sources of pv at the
two points ±a.
The velocity field u is initially even about the origin y = 0, and there is nothing in
the linear problem above that will break that symmetry for u. If u is an even function
of y, TJ, whose derivative with respect to y yields the geostrophic u, must be an odd
function of y.
Thus the solution can be written
ASinh(YIA) IYlsa
{
17= Ae-(y-a)/Asinh(aIA) y~a (12.21)
-Ae(y+a)/Asinh(al A) ys-a
where A is the deformation radius defined by
A=~ (12.22)
f
The deformation radius is an intrinsic length scale and measures the tendency for
gravity to smooth disturbances out horizontally against the tendency for rotation to
link the fluid together vertically along the rotation axis. If the fluid were stratified,
instead of co' the appropriate speed for defining the deformation radius would be the
internal gravity wave speed for a particular vertical mode. Hence, for a stratified fluid
there will in general be an infinite number of deformation radii. In the present case,
we have only one for the homogeneous layer. Like many other fundamental quantities
in GFD, this one is named after Rossby and is often called the Rossby deformation
radius. We will shortly see why the word deformation is used.
We have used the anticipation of antisymmetry for TJ to write the solution in terms
of a single unknown constant A. We have also chosen the solution so that the free sur-
face elevation is continuous at the point ±a. Otherwise, since u is proportional to the
y-derivative of TJ, we would generate infinite velocities at those points.
To determine the constant A, we return to the differential equation for 11 and

integrate it over a small neighborhood around the point y = a, i.e., from y = a - E to
Y = a + E. We will then let E~ o. Carrying out the integration and remembering that
the integral of the delta function is unity when the interval includes the zero of its
argument, we obtain
17 y (a +E)-17 y (a -E) = Uf I g (12.23)
In the limit where E~ 0, this yields, using the limits of the solutions on each side
of the point y = a,
- ~Sinh(aIA)- ~COSh(aIA)=uflg
(12.24)
~A=- AUf e- alA
g
This completes the solution. Collecting our results, we have
Co
j
_e- alA sinh(y I A)
17ID=~ -e-yIAsinh(aIA)
e ylA sinh(al A)
IYI::;a
y~a
y::;-a
(12.25)
from which the geostrophic zonal velocity of the final state can be calculated from
u=_.K. 17
f y
j
e-aIACOSh(YIA)
u/U= -e-yIAsinh(aIA)
-eyIAsinh(al A)
IYI::;a
y~a
y::;-a
(12.26)
Note that u is not continuous as the point +a and -a; the jumps in the velocity of
the initial conditions persist to the final steady state that is forced the by delta func-
tion sources of potential vorticity at ±a, which give rise to "kinks" in the free surface
elevation at those points where the slope of 11 is discontinuous. Note that v is zero in
the steady state, although it is certainly not zero in the waves whose radiation is es-
sential to reach the steady state.
The solution for the adjusted steady state has some curious and nonintuitive prop-
erties.
Figure 12.4 shows the solution for 11 and u for the case where a = 1 and A is 10, i.e.,
when the deformation radius is large compared to the geometrical scale of the flow.
Note that in the figures the zonal velocity, whose profile is shown coming out of
the paper, is positive.
1.2r'-----,,-----,------,------,------,------,------,------,
1.0
------------------1
0.8
ulU
0.6
0.4
0.2
I
5*17 10
0.0
~----- _ _ _ J
-0.2 'L-____- ' -______.L-_ _ _ _--'-_ _ _ _ _ _- ' -_ _ _ _ _ _L -____--'-______- ' -____- - - '
-2.0 -1.5 -1.0 -0.5 0.0 0.5 1.0 1.5 2.0
Ulco = 0.2, },,=colf=10, 0=1 ---+Y
Fig. 12.4. The free surface height (solid) and the x-velocity (dashed) for the case where the deforma-
tion radius is ten times the current half width a
The velocity profile has not changed much; it is still nearly the sum of step func-
tions of the initial data, but the free surface elevation that was initially zero has changed.
The fluid, under the influence of the Coriolis force, has slid to the right of the direc-
tion of flow to set up a pressure gradient with high pressure to the right of the current
and low pressure to the left of the current (looking downstream). This final adjusted
state seems intuitively attractive, and indeed is often the example used for illustrative
purposes and close to the one Rossby originally used (note: the free surface height
has been multiplied by 5 for clarity).
The results become a good deal stranger when the deformation radius is as small or
smaller than the geometrical scale, i.e., when It < a. For example, when It is equal to a,
we get the situation shown in Fig. 12.5.
Note that now the reduction of the zonal velocity in the center of the region is much
more evident. The free surface is tilting to support the flow geostrophically, but note the
reverse flow in the region beyond ±a. Also note that the characteristic decay scale for the
deformed free surface is just the deformation radius, hence the name. The really fun-
damental role of the potential vorticity is particularly evident when the deformation
radius is small with respect to the scale a. Figure 12.6 shows the case when It is O.la.
In this limit, the final flow consists of two vortex sheets limited to regions of the
order of the deformation radius around each edge where the delta functions of the pv
are maintained. The free surface elevation is symmetric about each delta function, and
the two are nearly nonoverlapping. The structure is very distant from the original pic-
,,
0.6
I t
," '
,,, " "
/
/
0.4 f- ", ~ u/U
,,
0.2 f-
,,
o.ol ---
~
~----
,,
"
",,
",'"
,, ,,
/
/
-0.2 f- /
/
,
I
.~ y
-04 [ "
-0.6
-5 -4 -3 -2 -1 0 1 2 3 4 5
Uleo = 0.2, A=colf= 1, a=l
Fig. 12.5. As in Fig. 12.4, except that the deformation radius equals the current half width
0.5 , ,
,,,"-
0.4f-
"" u/U
'I
,,,
0.3 f- 1\
'I
,' ,
,,
0.2 f-
I
0.1f-
,, ,,
\
, I
,, ,,
....---
/
",
0.01
,, ,
",
/
-0.1 f- \
,, , I
I
-0.2 f- ', , ,, ''

',
"I,
-0.3 f-
"
,
"
-0.4 f- "I, "
.I
"
I
-0.5'
-2.0 -1.5 -1.0 -0.5 0.0 0.5 1.0 1.5 2.0
U!co = 0.2, A = co/f= 0.1. a=l ----.y
Fig. 12.6. As in Fig. 12.4 except that the deformation radius is small with respect to the current half
width, i.e., 0.1
ture of the first figure and emphasizes that the resulting adjusted geostrophic flow is
determined by the distribution of pv, not by the original distribution of momentum.
Given the remarkable difference between the final states and the initial state, it is
important to realize that it is the waves, which are not described, that radiate away that
part of the initial condition that will not move to geostrophic balance. One measure
of the amount of radiation is the difference between the initial energy and the final
energy in the adjusted state.
Energy
By forming the product of the horizontal momentum equations with each velocity
component and the product of the continuity equation with the free surface height, it
is easy to show that the equation for the energy per unit horizontal area is (in the ab-
sence of atmospheric pressure forcing)
:t 2
[D( u :v2 J+ gt]+ V '(gTJu)D=O (12.27)
Note that the kinetic energy involves only the horizontal velocity. Consistent with
the hydrostatic approximation, the vertical velocity is too small to contribute. The
energy flux term is the horizontal velocity times the pressure, which in this case is given
hydrostatically by the free surface elevation.
In the adjustment problem just discussed, the initial energy is all kinetic energy,
since the initial free surface elevation is zero. That initial total energy is
DU 2
Einitial =--x2a=DU 2a (12.28)
2
We could compute the final energy by using our results for u and T] and integrating
over the whole y-interval. There is, though, an easier indirect way to do the calculation.
Starting from the equation relating T] and the potential vorticity,
KV 2TJ- L TJ=q (12.29)

f D
Multiplication by gDT] / f yields
2 2 D
D~ V· (TJV TJ)- D~(V TJ)2 - gTJ2 = qTJ gf (12.30)
f f
If the disturbance vanishes at infinity, the divergence term, the first term on the
left-hand side of the equation, will have zero integral over the whole domain. Recog-
nizing that the kinetic energy of the geostrophic velocities is given by the second term
(divided by 2), we finally obtain for the geostrophically balanced state
fEgeos.dA =- ;7 fTJqdA
where the integral is over the whole domain. In the case just considered, q is the sum
of two delta functions, so
= 2 =
fEgeos.dy=- ~o f1JU [8(y-a)-8(y+a)]dy
_= f_=
=- ;f
c2 u
(1J(a)-1J(-a») (12.31)
= U2D ..1(I_e- 2aIA )

2
The ratio of the final energy to the initial energy will give us a measure of how much
is retained in the geostrophic state and how much is radiated away by the gravity waves.
That ratio is
Egeos. l_e- 2aIA

(12.32)
Einitial 2a I ..1
so that the ratio is a function only of the parameter a I A =fa I co' i.e., the ratio of the
width of the current to the deformation radius. The energy ratio as a function of that
parameter is shown in Fig. 12.7.
1.0 i ,
EgeostrOphlC
,
/ Elnltlal i
0.9
0.8
0.7
0.6
0.5
0.3
0.2
0.1
O.O ~
i ----~----~----~----~----~~----~----~----~----~--~
o 2 3 456 7 8 9 10
a /;. = fa / Co
Fig. 12.7. The ratio of the final geostrophic energy to the initial energy
The energy ratio is unity when the deformation radius is large, i.e., when the rota-
tion is negligible. That limit is easy to understand, since if there were no rotation, the
initial state of no free surface elevation and a uniform flow in the region between y = +a
and -a would be an exact steady solution. Nothing would happen, and no energy would
be radiated. At the other extreme, when the deformation radius is small with respect
to a, the motion is limited to a narrow region around the end points at ±a, and the
energy, almost all of which is kinetic, is of the order of U2D* A" i.e., it is of the order
of It. As the deformation radius becomes small, the energy retained in the geostrophic
motion becomes small, and in this limit, most of the energy of the initial state is radi-
ated away as gravity waves.
The complete problem was studied by Cahn (1945), an associate of Rossby's. Cahn
presents the full-time dependent solution showing the evolution to the time indepen-
dent state. The analysis is more complicated than we have space for, but I encourage
you to at least examine the graphical results of his calculation.
Lecture 13
Shallow Water Waves in a Rotating Fluid;

Poincare and Kelvin Waves
We now examine the nature of the waves which serve, among other things, to sculpt
the geostrophic final state from an arbitrary initial state. These waves, as we noted
earlier, have no potential vorticity, because in the simple models we are examining,
the conservation of pv is simply the statement:
aq =0 (13.1a)
at
q=s -17 i (13.1b)
D
Thus, for periodic motion for which the time derivative can be replaced by multi-
plication by the frequency, q must vanish. The wave part then satisfies the wave equa-
tion derived in the last lecture:
1 a217 l
'11 217- 2 - 2 - - 2 17=0, Co =..[ii5 (l3.2)
Co at Co
If there were no rotation, we would get the classical, nondispersive wave equation.
In one dimension that equation would be
17xx -17tt/ c~ =0 (13.3)
Its well-known solution is
17 = F(x-cot)+G(x+cot) (l3.4)
where F and G are arbitrary functions of their arguments. The functions can be deter-
mined by initial data. The important thing to note here is that the shape of the distur-
bance remains fixed with time, and each function translates F to the right and G to
the left with the speed co. The unchanging shape is a reflection of the fact that for
nonrotating shallow water, the phase speed is independent of wave number; the wave
is nondispersive, and so no change of shape occurs.
Returning to the case where the rotation is different from zero, we can find plane
wave solutions in the x-y-plane of the form

134 Lecture 13 . Shallow Water Waves in a Rotating Fluid; Poincare and Kelvin Waves
17 = Aei(kx+Zy-ax) (13.Sa)
K= xk+ yl (l3.Sb)
K =~k2 +12 (13.5c)
which yields for the frequency the dispersion relation:
m=±{f2 +c5[e +121Y'2 (13.6)
It is important to note that the horizontal velocity has a nonzero divergence, since
_ 1
V'h ·u=--17t
D
so the group velocity does not have to be perpendicular to the wave vector. Indeed, we
can tell immediately from the dispersion relation that the group velocity will be in the
direction of the wave vector.
Note that the frequency has a minimum value of ±f. That is, these waves all have
frequencies greater than the Coriolis parameter. If Il is the wavelength, the increase of
the frequency of the wave above f will depend on the ratio of the wavelength to the
deformation radius, coIf. If the wavelength is large compared to the deformation ra-
dius, the frequency will be close to f.
We may easily calculate the two components of the group velocity:
dm 2 k 2 k
-=Co-=Co - - - (13.7a)
dk 0.1 ~ f2 + C5 K2
dm 2 I 2 I
-=co-=co--- (13.7b)
dl 0.1 ~f2 +c5 K2
Thus, the group velocity is in the direction of the wave vector and in the same direc-
tion as the phase speed. Note that while the group velocity goes to zero as the wave
number goes to zero (large wavelengths), the phase speed becomes infinite in that limit;
this is another indication of the physical irrelevance of the phase speed as a messenger
of real information.
By eliminating v between the two horizontal momentum equations, one obtains a
simple relation between u and 1], i.e.,
d 2u 2 d 217 d17
- + f u=-g--fg- (13.8)
dt 2 dXdt dy
similarly for v,
d 2v 2 d217 d 17
- + f v=-g-+gf- (13.9)
dt 2 dtdY dX
This allows us to solve for u and v in terms of 1] unless the operator of the left-hand
side is null, which will happen for oscillations exactly at the inertial frequency, i.e.,
when 0.1= ±f.
Lecture 13 . Shallow Water Waves in a Rotating Fluid; Poincare and Kelvin Waves 135
For all other frequencies, we have the relations (after aligning the x-axis with the
wave vector):
17 =1Jo cos(kx - 0Jt) (l3.11a)
U= 170 OJ cos(kx-OJt) (l3.11b)

D k
V= 1Jo Lsin(kx-OJt) (13.11c)

D k
so that again the velocity vector traces out an ellipse in the x-y-plane whose major axis
is in the direction of the wave vector, and its minor axis shorter by an amount OJ! fis at
right angles. For positive frequency and wave number, the velocity vector in the wave
moves clockwise as the wave progresses through a period. Note that the fluid velocity
is smaller than the phase speed by the (small) parameter Tlo! D.
We also note that the maximum group velocity is co' and this occurs for the short-
est waves. The longest waves have the slowest group velocity. Therefore, were we to do
the adjustment problem, we would expect that we would see the short waves speed
away from the adjusting current first, and after a long time, a long swell of waves would
finally move away from the vicinity of the current. This is exactly what Cahn found as
seen in Fig. 13.1 redrawn from his paper.
1.0
0.5
(Ylo)
• n
a
a
E
.g
0.0
-0.5
-- ..,., ~ - ........
~
"-'
V
I
'0 +0.5
~
-
~
I
:J
~
t::
'"a. 0.0
I "'-
~
I-"-
-0.5
lJ III
(y> 0)
-1.0 I I I
Horizontal unit = deformation radius Co / f
Fig. 1 3.1. The free surface height as a function of time showing at first the passage of the fast short
waves and then the longer waves with slower group velocities (after Cahn 1945)
Channel Modes and the Kelvin Wave
Up to now, we have not considered waves in a domain bounded horizontally. Normally,

what that does is introduce certain conditions that quantize the horizontal wave num-
ber. However, in the case of a rotating fluid, there are some surprises. Consider the
wave motion in a channel of width L (Fig. 13.2).
Again, the equation of motion for the wave is
1 '(PTJ f2
V 2TJ-----TJ=0, co =.[ii5 (13.12)
c~ dt 2 c~
On the boundaries of the channel, which have been oriented along the x-axis, the
y-component of the velocity, v, must vanish. Since
d2 v 2 d2 TJ d TJ (13.13)
- + f v=-g-+gf-
dt 2 dtdY dX
the boundary condition becomes
d2TJ dTJ (13.14a)

O=-g-+gf-
dtdY dX
y=O,L (13.14b)
The domain is infinite in the x-direction, and so we can look for wave modes of the form
TJ = rr(y )e(ikx-M) (13.15)
so that 11(y) satisfies the ordinary differential equation:
d2rr
dy2
+{oi- c~
f2
k2 }rr =0 (13.16)
this is subject to the boundary conditions;
drr +kLrr =0 (13.17a)

dy OJ
y=O,L (13.17b)
y=L
Oil
y=O
Fig. 13.2. The channel of width L in plan view in which gravity modes occur
Note that if there were no rotation, f = 0, a possible solution would be 1] indepen-

dent of y with w = ±kco. This would be the lowest cross stream mode and the solution
with the lowest frequency. Higher modes of the form cos jny / L would also be possible
with frequencies
w=±co~e + in 2 / L2
It will be of interest to examine how the rotation alters this simple structure of the
problem (see also Gill 1982 and Pedlosky 1987). The cause of the change will be found
in the mixed boundary condition at y = 0 and L, which mixes the function and its de-
rivative, and which explicitly involves the frequency.
It is useful to define the constant
2 f2
R2=~_k2 (13.18)
c~
so that Ris something like a y-wave number. The solution can then be written
1](y} = AsinRy + BcosRy (l3.19)
The constants A and B are not both free but must be chosen to satisfy the bound-
ary conditions. Applying the boundary condition on y = 0 first yields
kf 0Jii
AR+B-=O ::::;.B=--A (13.20)
w kf
The same boundary condition applied at y = L yields
Mcosn-BRsinn+ kf {Asinn+Bcosn}=O (13.20)

w
which when combined with the first equation relating A and B yields
sinn[wf!2 + kf] = 0 (13.21)

kf w
When the definition of Ris used to evaluate the square bracket in the condition above,
we obtain the final eigenvalue relation for the modes in the rotating channel, i.e.,
sinn[w2g2 +k2 f2 ]=sinn[ cv2 lcv c~t )-ecv

2 2 +k2 f2l
(13.22)
=sinn[ :; -e }w2 - f2 )=0

There are apparently three possible ways in which this eigenvalue relation, or dis-
persion relation linking m and k, can be found.
1. m= ±f
2. m= ±kco
3. sinJiL =0
The first of these is immediately suspicious, since as we have noted, when the fre-
quency is exactly equal to f, we can no longer use Eqs. 13.8 and 13.9 that relate the
velocities to the free surface elevation. We will have to examine this case very care-
fully, and in fact, it will turn out that this is a spurious root.
The second possibility does not look much more promising, since it appears to yield
the frequency relation for a y-independent mode for a nonrotating fluid. Note that the
boundary conditions
d'if + kL'if =0, y =O,L (13.23)

dy m
do not allow a nontrivial y-independent solution iff is not zero.
With all this doubt in mind, let's start with the term that looks like the most ordi-
nary of eigen conditions, namely
sinJiL = 0 (13.24)
The solution of this condition is
£L=nn, n=1,2,3 ... (13.25)
where we note that we have started with n = 1. The solution corresponding to n = 0

would yield, from the boundary condition at y = 0, B = O. But if I' =0, the remaining
term proportional to A would be the sine of a zero argument. Hence, the whole solu-
tion becomes trivially zero.
The physical reason why this occurs is related to the relation between 1] and v. If
the free surface height were independent of y, we would have
v= gf a'7(x,t) (13.26)
f2-m 2 ax
which would be nonzero at the boundaries y = O,L, unless 1] were identically zero ev-
erywhere. Hence, the first nontrivial term must be n = 1. Using the definition
2 f2
1'2 m - k2 (13.27)
c~
this yields the dispersion relation for m for each n,
w~ =l +c~[e +n 2n 2/ L2] (13.28)
This is exactly the dispersion relation for the plane Poincare wave we deduced ear-
lier, except that the y-wave number is quantized in multiples of rc/ L with the major
exception that the n = amode is not allowed. Now, in the unbounded case, there is such
a y-independent mode. In addition, whenfis zero there is such a mode allowed. What
has happened to that lowest mode? Something is missing, since it makes no physical
sense that the addition of the smallest rotation of the system can eliminate the lowest
mode previously allowed. We have a problem here we must be sure to clear up. For
now though, let's go ahead as if we have not noticed this vexing apparent paradox and
examine what the modes that are allowed are like.
Using the relation between A and B and choosing A to measure the elevation of the
free surface,
lJ=lJo[cos(nrcy/L)- ~ Sin(nrcy/L)}oS(kX-OJt) (13.29)
Note that the y-structure depends on the phase speed of the mode; that is, in the
square bracket the relative importance of the sine term with respect to the cosine term
depends on w/ k. To keep the equations uncluttered, the subscript n on the frequency
has been suppressed, but the student should recall that for each n, the frequency is
given from the dispersion relation above. Since for each k there are two roots for w
differing in sign, it follows that the cross channel structure will differ for waves going
to the right and waves going to the left. Using the relations between the velocities and
the free surface height
gf alJ _ g _a2
lJ
V(j2 - W )
2
= ax atay (13.30)
one easily finds that
_ ~2 + c2n2rc 2 / L2 ]
v= lJo 0 sin(mty/L)sin(kx-OJt) (13.31)
D OJnrc/ L
The y-component of velocity contains only the sine term, since of course it has to
vanish on y = 0 and 1. Similarly, the velocity in the x-direction can be found and is
u = lJo [--.:Lcos(nrcy / L)- fL sin(n1ty / L)]COS(kX-OJt) (13.32)

D (w/ k) nrc
Looking back at the formula for the free surface elevation (Eq. 13.29), we note that TJ
will vanish for those values of y for which
tan(nrcy / L) = wnR. n = nrc {I + c~(e +n 21t2 /L2 )}1I2 (13.33)

kf L f2
1.0 I ,,'"",
I I 1/,11 11111111 I
1,'1/"'1 IIIIIIII I
0.9 111111 \ \ \
J , ' I , ' II
I I ' ",,
111"'11 \
1111111\ ,
111111 \ \ \ _ /
"
II J,' 1/:: I I III \ \ \ \
0.8 111'\\\ \
I " l \ \ \ \ ..... __ / I ' / , III
',","1
III \ \ \ \ I I I I 1"
11\1\ \ , _ ...... 1'/1 1 ,
0.7 1\\\\\ /1'/1 1 ,
~\\\' ......""'- /1/111
" - - / I I J
\ \ \..... "'" I I I
/
1\' --//
\ '\........ .... .... ./
O.5f-- ~
l
/
..------ /
-----
1 / - - - - - ..... \
I/,"-- .....
1 1 / - - - ...... , '
I I I ..- -.... ,,\ \
1 / / - - - ......
11/ _-_ ,I'
,'
0.3
I I I;'
I f I , / /,...
..... \ \ \
...... " '\. \ \ I :/,'/:/ ..... :,\\~
"1/1/--,,'\1\ 111/1/--," II'
'J I I / / - \ \ \ \ I I Ifllll- \,'\\1
0.2 ""I' // ',\1\'"
",,,', /_, \\',111
""1'1/
lilli', / - , \\'\"1
'\\1\111
11111 1 , I \ ' \ \ 1 '11 III J I I I I \ \ \ \ \ III

I l ' " J I I/'..... \ \ \ ' ,'II " ' " II J /'..... \ \ \ I ,'"
0.1 II'" J I I I \ \ \ I I ,'II 11111" I I \ \ \\\,111
111111 , I I \ I I I I ,"I 11'''''1 I \ , \,',111
I I ' " II , , \ , " , III
,'III 11111/ II 1 \ \ I, ~ ':::
-1.5 -1.0 -0.5 0.0 0.5 1.0 1.5 2.0
n = 1, kL = 3.1416, ca I f2L 2 = 10, (0 If = 14.0852
Fig. 13.3. The elevation of the free surface for a low rotation mode, wIf = 14.0852. The gravest mode
corresponding to n = 1 is shown
1.0
,\\\lId'
, \ 111111' ,,',,/I J'
11'11/11 ,. ...... \\'''''1
\ \ \ ' 11"1
'I,
I"
/11 / I /
/I /I I I
11'11 " , I \ 1 I \1"1
0.91- """,,,
, I I ",,,, 1,' 11 " , ,
\
~ I 1 ' 1,"1
111111// I
1,,11111 J
I
I I ""Il' I,: 1111 I , , ' ,',1111 I " 1111 I I
0.8
I- ,"11',,'
""",,,
1',"1 I , I I ,,',,'II ',,11111
I 111",,' l,'li' , I \
1\ " " I I
f ,,',,'II
, ' , I 11", I" 1111' ,
' , , " 111 I
' " " II' 1,' \\I \ \
,,' \I , ,\ \ I I I ' " 'II '\,"" \ \
071:-'
,
I
11''',,'
I I f 1/,,' I,'", \ \ ,
~
I I I J ,'II
/ 1 I '1/'1, '11" 1 I \
',\1 II \ \
,
/ I lf'll' I,' 11 \ \ \ , I I I , "I '11\1\\ \
I" \ \ \ ,
l i l l i ' \ ' _ ...... I l f l l l l
I,' \ \ \, /'
/ I II/,,' I, \ II \ \ '
0,6 ~. . . / /1/ / f J :
I 1\ \ , \ ..........
I I '"
1 I I I I I I \ \ \ \ , ....
..... / 1 I I J I 1 , ' \ \ ' _ ..... / " 1 / / I \ ' "-
D,S ~ . . . / /' / / ' I \ \ ' ...... 1 ' I /
1\\ ' '
...... /,' 1\ \ , ...... _ ..... ,. ' I 1
\ ,' \ ......
...... / , I , \ ' , ...... - - ...... / / 1 1
, '
---- ............ / \
/
0.4 /
0.3 -----....
......, -----
. . . . . .....,
0.2
,/ /" - ...... -
",/ ..... -- - ...."" .... ,
.... ' , \ \
" I' // ........ -
",/ ... -_,',\\
- ............ ,
,','/I// . . . - . . . ,'\\\~ ,','/,/,. . . . - . . ,'\\\~

1// 1 / / ' - - ' \ \ \ \ 1 111 1 / / - - , ' \ \ \ ,
0.1 I J / " / ..... _ \\\\1\ JIII" ...... _ '\\\\1
111//1/ '\\\\11 JII/", '\\\\11
II J I I I I / - , " ' \ \ \1, "I/I' I / - , " \ \ \ II,
O.O~-U~~~~LL-L--~~~LULLL-~~~~~~LL~--~~~~~~
-2.0 -1.5 -1.0 -0.5 0.0 0.5 1.0 1.5 2.0

n=1, kL=3.1416, d/f2L2=0.1, (O/f=1.7245
Fig. 13.4. The gravest mode for the case oflarge rotation. wlf= 1.7245 and the deformation radius is
about 1/3 of the channel width
When f = 0, these coincide with the infinities of the tangent function, i.e., at y at
odd half multiples of L. So, in the limit as f-----7 0, all these modes coincide with the
cosine modes of cross stream wave number nn / L for n > O. So, once more we are per-
plexed to find that even though we have an infinite number of modes, we are missing
the lowest mode corresponding to n = O.
Let's first look at the structure of the modes we have found. First let's look at the mode
for small f or when the deformation radius is much larger than the channel width. For
col fL = 10, the contours of the free surface height are shown in Fig. 13.3 for an x-wave
number n/ L.
Note that for this case, at low rotation the lowest mode has a node at about the half-
way point in y in the channel. On the other hand, when the rotation rate is large so
that the deformation length is about a third (actually 0.316) of the channel width, the
form of the free surface is as shown in Fig. 13.4, and we note that now the zero level is
much closer to the lower boundary.
Both of the cases above are for wave patterns propagating to the right. If the fre-
quency is negative so that the pattern propagates to the left with the same speed as
the above example, the free surface height instead looks like in Fig. 13.5.
The pattern is essentially the same except that the nodal line in y has shifted to-
wards the boundary at y =L as the pattern propagates to the left.
The student is left to discuss the group velocity in these modes in the x-direction.
Note that there is no energy flux in the y-direction, and it is left to the student to ex-
plain why that is in terms of individual plane Poincare waves.
1.01 ""'IIIIil\\\\ Jjiiilli::~'I\\\ " / //,'/::: 111111\1 i }Illllilltllll'
:~\\\\\',-.,/,' '/,'::
0.9 11 \ \ \ \..... .,",' I I, I
1\\ \', - / I I I J
\',:---: ... '/ /

I \ \ \. -- _ .... / I J I
0.8 \
\
, '- ... I
0.71-1_ _-+-_____-+-_____-+-_____-+-__--1
0.6 ,
/
I
/
",..---
/ ,
~
I ,/ .... - - - ,/,/ ..........

I I , / , , - - .....
:://I/~--",\\"
I / I ' / "" - - .......... , \ \
0.5 , ' , ' , ' , " / .... _ _ , .... , \ \ I
::: / / / - - ',\ '- ""~

I I I I I \. \ \ \ I
I , I I I I" .... \ \ \ \ I
..... - , ' , ' \ 1 11 /
,I' \. \ \ \ \ "
II/'ll
I r I I I 1'/ ..... \ \ \ ,I I
:::::,'/,~-, '- \",,~~:

1/ J I I I
II/"I, / - , \. , \ \ 1 11
11,'11, / \. \ ,,1,11
I II II I I I \
\ I I I I I I
II r J" ,
I" Ilf I 1 1'''' , \ I ' I II'
I \ . \ \ I \1 ,,1
III"" I ,-, \ ' ,11\\1
I" I " , I I \ \ \ I I " \1
I I I " II I I \ I I I ' I I ,I 1 11
11/11I,f I ' \ 1,1 1
111111,1 I
'1IIIll/ I
\ \1,1 1 1 11
111\11111 1',,"111
1,"1', I' \\ '
II,til
ll
1
',1 ,, '
'11 11111 11 1 ,1 1 1 11
',,111'1 I ,III,II,I
',,11111 I 111111111
',,'1111 I I,I,,II11
'I'"
'1111111 I 1/1,11111
0.1 1.11111 I J, 1,, 1 ,1 1 '1\ \\

1,"111 ,, ,1,"/11
'" " " ,
1i,11I1 I \
11,"11 \ \ I
I , ,,'llll
/ / , I IIII I I , " 1\ \ \ I' / ,11 , 1 1
I' '11\ \ \. / I I I "ll' " ." I \ \ \ , 1",
1 ,1"l"
1 ,,1
11.1 1 ,11 11.1 1 ,,1
-1.5 - 1.0 0.0 -0.5

0.5 1.0 1.5 2.0
n=l, kL=3.1416, c~/f2L2= O.1,wlf=-1.7245
Fig. 13.5. As in the previous figure but for a wave propagating to the left
The Kelvin Mode
Let's now examine the second possibility as a solution of the eigenvalue condition,
namely that
OJ=±keo (13.34)
As we noted, this is a rather unexpected possibility, since it is the dispersion rela-

tion for y-independent, nonrotating, long surface gravity waves. Our fluid is rotating,
and as we noted above, no solution independent of y is a possible solution in the chan-
nel. So, it is of interest to examine the possibility with some care. Using the definition
2 f2
OJ-
fi2 k2 (13.35)
2
eo
we find that for this case,
fi=±iflco (13.36)
so that the cross channel wave number is purely imaginary. Let's look at the solution
corresponding to the positive imaginary root (it is left to the student to repeat the analy-
sis for the negative root to demonstrate that nothing new is discovered; the negative
root only serves to interchange the identity of the two solutions we will shortly find).
Using the relation between A and B from the boundary condition at y = 0,
kf 0Jfi
Afi+B-=O =>B=--A (13.37)
OJ' kf
and writing the sine and cosine in their exponential form, we obtain
If(y) = 170 {e- fy I Co [1 + OJ/ keo]- efy /co [1- OJ/ keo]} (13.38)
Here we have redefined A = iT/o'

The solution consists of two parts. The first term exponentially decreases from the
lower boundary at y = O. The second term exponentially decreases in the -y-direction
from boundary happens at y = L. The scale for the exponential decrease from either
boundary is the deformation radius. Note that this solution can occur only in the pres-
ence of lateral boundaries in order to keep the solution finite for all y.
For the solution propagating to the right for which OJ = keo' the seeond term vanishes,
and the total solution restoring the x- and t-dependence is the right moving Kelvin
wave:
17 =17oe-fylc o cos(kx-OJt) (13.39)
Note that such a solution would be valid for the region y > 0 if only a single wall
were present and the fluid were semi-infinite in the +y-direction. We obtained the
solution using only the boundary condition at y = 0, and we must check that it also
satisfies the v = 0 condition at y = L. In fact, let's calculate v for all y in the channel
using the relation
v(f2 -ai) =gfT/x - gT/yt
=-gfksin(kx-kcot)+ gL kco sin(kx-kcot) (13.40)

Co
=o!
The cross channel velocity is identically zero for all values of y in the channel, and
so of course this satisfies the boundary conditions trivially at y = 0 and L. Moreover,
calculating u,
u(l-ai)=-gfT/y - gT/xt
f2
= g-T/- gk 2coT/ (13.41)
Co
=.K.(l- k2C~)T/ =.K.(f2 -ai)T/

Co Co
or
u=_K aT/ (l3.42)

f ay
so that the long channel velocity is in geostrophic balance with the pressure field, al-
though the motion is unsteady and the frequency is not small with respect to f
If we choose the other root (J) = -kco so that the wave is traveling to the left, the so-
lution consists of the same wave, now a maximum at the boundary at y = L exponen-
tially decreasing in the direction towards the lower boundary at y = o. Again, as you
can check, the cross channel velocity is exactly zero, and the long channel velocity is
in geostrophic balance. Note that in regions where the free surface elevation and the
u-velocity are in phase and if one is positive, so is the other.
Note that as f----7 0, the mode becomes independent of y and
T/ ~ T/o cos(kx-mt)
which is the "missing" lowest mode of the nonrotating case. That mode in the pres-
ence of rotation maintains its character of having no cross channel velocity and does
so by introducing a sloping free surface elevation that exactly balances geostrophi-
cally the Coriolis acceleration of u. Indeed, it is illuminating to examine the original
equations using the a priori condition that v is identically zero, i.e.,
fu=-gT/y (13.43a)
ut =-gT/x (13.43b)
T/t =-Du x (13.43c)

Combining the last two of these equations yields

2
17tt = Co17xx (13.44)
one solution of which is 1] = F(x - cot,y) where F is an arbitrary function. The first
equation determines the y-structure of F. Since from that equation fU t = -g1]yt' we find
that from the second equation -fg1]x = -g1]yt = gCo1]yx' the second equality follows from
the x - cot structure of the function F. This yields the differential equation
a f
av 17x +-17x =0 (l3.45)
/ Co
from which the exponential y-structure of the solution follows immediately. An im-
portant consequence of this approach is that we see that the form of the Kelvin wave
in x is arbitrary. Any function of the argument x - cot is legitimate for the x-t-struc-
ture, and as we could see from its dispersion relation, the form in x is unchanging with
time as the wave propagates, because the frequency relation is nondispersive, i.e., the
frequency is a linear function of k.
While the Poincare waves have a minimum frequency
Wmin = ~2 + C~1t2 / L2 p'2 (13.46)
the Kelvin wave has no minimum. As k ~ 0, the frequency will go to zero. So if we

had a rather narrow channel for which the minimum frequency was quite a bit higher
than f, a forcing at or below the inertial frequency would not be possible in the chan-
nel for Poincare waves. The dynamical equations (linear, rotating shallow water) we
have been studying are often called the Laplace Tidal Equations, because they are
exactly those used to discuss the tidal response to solar and lunar forcing. Naturally,
one has to include the effects of sphericity, which we have not done, but qualitatively
we can see there would be difficulty of the tidal forcing at semi-diurnal or diurnal
periods to effective produce a Poincare wave response in a narrow sea. Instead, the
response is more likely to be a Kelvin wave signal propagating around the boundary
of the sea.
To get a feeling what that might look like, consider the superposition of two Kelvin
waves of equal amplitudes propagating on both boundaries of the channel represent-
ing an incoming wave on one boundary balanced by an outgoing wave on the other
boundary. We are neglecting the (difficult) problem of the reflection of the Kelvin wave
at one of the ends of the channel if it is close, but we will assume it is far enough away
to ignore.
The sum of the two Kelvin waves would be
17 = 170 [cos(k[x-cotJ)e -JYleo + cos(k[x + cotJ)e -(L-y)f leo 1 (13.47)
where we have introduced a constant term in the second wave so that each wave has
the same maximum amplitude. Note that the second wave is decreasing as y diminishes.
Note that at y = L 12, the channel mid-point, the free surface height is
17 = 170 [cos(k[x - cot 1) + cos(k[x + cot])]e - fL I2c o
(13.48)
= 2170 cos kx cos kcote - fL 12co
so that for all t the free surface elevation vanishes and is therefore fixed in time at the
points
Y =Ll2, kx= j1t/2, j = 1,2,3 ... (13.49)
These fixed points for the elevation are called amphidromic points in tidal theory.
Figures 13.6-13.8 show the sum of the two Kelvin waves at several times over a wave
period. The asterisks mark the amphidromic points.
Figure 13.6 shows the case where kL is 1t and the figure is drawn for the time t = T 14
where T is the wave period 21t1 kco. Figure 13.7 shows the free surface elevation some
time before when t = 0.245 T.
Note that the amphidromic points on the zero contour of free surface height have
remained stationary as the phase of the disturbance rotates around it. Figure 13.8 shows
the situation at the later time t = 2.55T.
Again, note that although the phase lines have altered their tilt considerably, the
amphidromic points remain stationary.
1.0~~~~~~~~~~~:/'J7n7ITIT~~~'-~~~~rrn\\~'-~~~
__ I 1,/,"" I
I~'l\\\\\"
1111\\\' I I" "" 111111\\ ' - / I I,,""
0.9 1111\\' ~/ I f / " ' , 111111\' Ilf/""
11111\ '_I II"" 1111\\\ If/"', ~/
11111\'-1 /1 "'1
\\\\, . . . -,'/1,
111\\' /I'iI,
0.8 1\\\,' /",1,
111,,~-/I",
11\\'--//'"
11\'
\\,,
/
____ /1,
I', 11\\,-_//1',
II" /' II,
0.7 I' ~ __ / / I I\,,~-/III
I \ \. ....... - --- I I
0.6 " \
, ...... /
/
0.51 >I< >I< >I< >I<
0.4 ,/
1/
..... --
- ......
.......
'-
',
" ,',----
1/ - ........
....... "
',
0.3 /// . . . -',\1 11/..-_',\\
11// ',,\1 1 111/ ",,\1 1
1111 ....... - , \\' 11/1.,......-, \\'
0.2 J i l l I ............ '\\\1 I I I / I ........... ' \ \ \ 1
11//'/ \., "I 11//'/ ,\ 1\1
1111'/ , - , "\1\1 I l f l ' , / - , \\\1\1
0.1 '/1111/ ............. \ \ ' \ \ 1 1 '"111/,..... . . . .
\\\\\1 1
1111111/ , I I I III 1111111/ ,\11 11 1
I I I I I I I ! / --, \\\\"1 1 IIIIffl1 /' - , \,\,'11 1
0.0 I I I I
-2.0 -1.5 -1.0 -0.5 0.0 0.5 1.0 1.5 2.0
fL/ Co =0.5, t/ period =0.25, kL =3.1416
Fig. 13.6. The superposition of a left and right traveling Kelvin wave of equal amplitude in the chan-
nel. Note the amphidromic points marked by the asterisks. The pattern is shown at a time where the
phase lines in yare vertical
1.0 II I II I
1111111" _ 11III1111 1111111" _ 11III1111
11\\\\\' ....... //1/111'1 1 \ \ \ \ \ \ ' ....... / 1 ' , " " \
0.9 1111\1' 11I1111 1111\1' /111111
JII\ " _ / / 1 1 1 ' \ \\\\,"_//11/11
1111\' 11I111 1111,' 11I111
0.8
I,' I,',
\ \ \ \ , ....... - 111'1 \ \ \ \ , ....... - II,',
I,'
I I I ' ....... _ / 1 1 \ \ ....... _ /
II", /,' 1 \ \ ' .......
1'''-//'
\, ...... /1
I'"
\'''''' /
/1
\. " ,/
" \. ' ...... , /
,
,, ,,
, ,
'- '-
, ,
, "
/
/ " I
/
_......" \ \
- ,I 1/'" , I
I I / '....... _........' " \ \ I\
1/' ........ " \ \ \
I I / "\ \ \ \ I I / _....... \ \ \ \
I I / "\ \ \ \ I J I / "\ \ \ \
I I I I / ........ -...." \ \ \ \ \ 1111/ / - '111\
I I I I I I _ '\1 I I I 11111 I _ ',11\11
111111/ "\1\11/ 1111I I / ',1\1 I I I
0.1 111111/ / , \11"1\ I J II I I I ............ \ \ \ \ 1\
111111 I I
11111111/
I... . . . . . ,'.1
' \ I \ "1\
II
11111I1 /
11111111
I... . . . . ./
_ ' \ 1\\1111
, .1 II
- 1.5 -1.0 -0.5 0.0 0.5 1.0 1.5 2.0

fL / Co = 0.5, t / period =0.245, kL = 3.1416
Fig. 13.7. As in the previous figure but at a later time. The phase lines have moved but the amphidromic
points are fixed on the phase line
1.0 I 1/ I, II
1111\\\ ,_1/11',,11 1"111\\ ,_111,'1/11
0.9
111\\\' .......... /','/1" ,11\\\\\ ...... /'1,'11/1
11\1' //11111 1"\\1' /111111
1 \ \ \ \ ' _ 1 / '1/11 ,1\\\\ ' _ I I "III
111 1 ' / /1111 11\\1' / 11111
0.8 i l l ' .............. ,/1111 , 1 \ \ \ ............. / " 1 "
1 \ \ ' ............. /,'1 \ \ \ ' ............ /111
1\' //11 I" //11
0.7 , ...... /// I ' ....... 1/1
'- ,/ / / '- ...... / / /
/ I
0.6 / / I
/
./ ./
0.5 /
0.4 ,/" / ,
I IF \ / // \
1// '\ 1// 'I
1///_,' 1///_,'
0.3
,'I ,'I
I I
11// 11//
1/// ....... - , \ \ 1 111/ "......-, \ ' 1
1111'/-- \\'1' 1111//-.. \\\1
1111/1 '1\111 1111/1 '" 11
1111 1 1 / - ' \ \\\11 1 1 1 1 / 1 / - ' \ \\\1
0.1 1111/1/ / . . . . \\\\111 111111/ / . . . . \ \ \ \ 1
11111111 ,111\111 111/1111 ,\11\11
1111111/ / - , \111\111 1111/11/ / 1111\/1
0.0 II I \1 I
-2.0 - 1.5 -1.0 -0.5 0.0 0.5 1.0 1.5 2.0
fL / Co = 0.5, t / period =0.255, kL = 3.1416
Fig. 13.8. The same Kelvin wave superposition at a later time
We now turn our attention to the third possibility for an eigen solution, namely
(t) = ±f, and we noticed that in this case it was not possible to solve directly for the ve-
locity field from the free surface height, since for example
a
2v 2 2TJ aTJ
- + f v=-g-+gf-
a (13.50)
at 2 atay ax
the operator on the left-hand side is trivially zero and would give an infinite ampli-
tude for v. We must return to the original equations, i.e.,
v t + fu=-gTJ y (13.5Ia)
Ut - fv =-gTJx (13.5Ib)
Let's examine a solution oscillating like e- ijt, i.e., with (t) =f, and see if it is possible.
Then
-ifv + fu =-gfiy (13.52a)
-ifu - fv = -giki7 (l3.52b)
Note that the determinant of the coefficients of u and v is zero, but if the second
equation is multiplied by i and subtracted from the first, we obtain
dfi +kfi=O (l3.53a)

dy
so that
fi = T/oe- ky (l3.53b)
At the same time using one of the two momentum equations,
. gk-
U=-IV+- TJ (l3.54)
f
If this is placed in the equation for mass conservation,
1ft +D(u x +vy)=O (13.55)
we obtain
-iff[ + ikD(iku + v y) = 0 (13.56a)
~ -ifi7 + ikD{ij =g~fi}+DVy=O (l3.56b)
~ _kV=iffi(I_C~k2J=ifTJo(I_C~eJe-kY
v
y Dl f2 Dl f2
(13.56c)
which yields the solution for v:
v=Ae ky _ iflJo
2Dk
(1- C~k2)e-kY
f2
(13.57)
However, v must vanish on y = 0 and L. We can make v vanish on y = 0 by the proper

choice of A so that
v= iflJo 2k 2 ] sinhky
. [ 1-~ (13.58)
Dk f
The only way v can vanish on y = L is if the coefficient in front of the sinh term van-
ishes, in which case v is identically zero and the exponential decay rate for T] is e-fy1cO ,
while the frequency, f, is also keo. Thus, there is no possible wave solution with fre-
quency f except at a single wave number at which point the solution is indistinguish-
able from a Kelvin wave. We conclude that the full solution of the problem consists of
an infinite number of Poincare waves plus the Kelvin wave. The dispersion diagram
for the complete problem is shown in Fig. 13.9.
Note that for very large k, all the modes approach the dispersion curve for the Kelvin
wave. Note that there are two Kelvin modes, one for each boundary.
w/fvs.kL for rotating channel Poincare and Kelvin modes

8 r• --------,---------,---------,--------,---------,--------,
6
n=3
4
n=2
2f- .n.=..L-.
-21=~.,..,·~~-
-4
-6
-8LI________ ~ __________L __ _ _ _ _ _ _ __ L_ _ _ _ _ _ _ __ L_ _ _ _ _ _ _ _ ~ ________ ~
o 2 4 6 8 10 12
fL leo = 2, kL
Fig. 13.9. The full dispersion relation of the Poincare and Kelvin waves
Lecture 14
Rossby Waves
When we consider waves oflarge enough scale, the sphericity of the Earth can no longer
be ignored. Rossby was the first to point out that the most significant effect of the
Earth's sphericity is that it rendered the Coriolis parameter f = 2Q sine, a function of
latitude. Since the large scale motions in the ocean are nearly horizontal, the only com-
ponent of the Coriolis acceleration that really matters is the one involving the hori-
zontal velocities, and therefore only the local vertical component of the Coriolis pa-
rameter is dynamically significant. Otherwise, for scales that are large but still sub-
planetary, a Cartesian coordinate system can be used to obtain at least a qualitatively
correct view of the dynamics. Such an approximation in which the variation of the
Coriolis parameter with latitude is treated but in which the geometry is otherwise
Cartesian is called the beta-plane approximation, and we shall use it without a de-
tailed justification. The student is referred to Pedlosky (1987) for a careful derivation.
In this course, we will use the heuristic approach outlined above.
In this way, we take as the governing linear equations of motion
Ut - tv =-g1Jx (14.1a)
Vt + f u =-g1Jy (14.1b)
where now
f =2.QsinO "" 2.Q sin 00 + 2.QcosOo R(O-Oo)+ .. ·

R
(14.2)
. 2.Q cos 00
= fo+ fly , fa = 2.Q sm 00 ' j3=--~
R
where eo is a mid-latitude point about which we have expanded the Coriolis param-
eter, and R is the Earth's radius. The relation between latitude and the y-variable fol-
lows from (see Fig. 14.1)
y = R(O-Oo), and note that j3 =af lay (14.3)
If L is a characteristic magnitude of the north-south scale of the motion, then the

change of f compared to its characteristic value is
j3!:J.y / fo = O(LI R)« 1 (14.4)
as the principle parameter restriction for the validity of the beta-plane approximation.

150 Lecture 14 . Rossby Waves
Q
z
Fig. 14.1.
The tangent ,B-plane at the
central latitude eo
--*--/17
o h8
/ Do
Fig. 14.2. The homogeneous layer of fluid on the ,B-plane
At the same time, we will let the depth of the fluid in the absence of motion, which
we have called D, be a function of position as well (Fig. 14.2). Thus,
D=Do -hb(x,y) (14.5)
Returning to the continuity equation and integrating it over the depth of the fluid,
assuming again that since the pressure gradient is independent of depth, we may take
the horizontal velocities independent of depth:
D(u x +vy)+w(top)-w(bottom) =0 (14.6a)
w(top)=l]t (14.6b)
w(bottom) = u ·Vhb (14.6c)
The last condition follows from the kinematic condition that at the bottom, the
velocity must be parallel to the bottom so that a horizontal velocity flowing across the
Lecture 14 . Rossby Waves 151
gradient of the bottom depth produces a vertical velocity in order that the total veloc-
ity is parallel to the bottom. Putting the equations together yields the equation for mass
conservation:
7Jt + D(u x +vy)+u· VD =0 (14.7a)
or
7Jt +V·(Du)=O (14.7b)
(note that VD = -Vhb ).

Now let's form the vorticity equation by cross-differentiating the momentum equa-
tions to eliminate the pressure term. We obtain, remembering that f is a function of y,
(t+ f(u x +v y )+j3v=O (14.8)
and then with the mass conservation equation we can eliminate the horizontal diver-
gence of velocity:
St + f ( -Ii- Ii ·VD) +f3v=O

11t D (14.9)
or
~[(
at - f!l]+U.Vf
D
-uf·VDID=O (14.lOa)
or equivalently,
~[( - f!l]+DU.vL=o (14.10b)

at D D
The first term in this equation is the rate of change of the potential vorticity. The
second term is the inner product of the mass flux with the gradient of the ambient
potential vorticity fl D. Up to now, with constantfand constant D that term has been
zero, and the potential vorticity has been constant at each point. However, in the pres-
ence of a gradient of the potential vorticity preexisting in the absence of motion, the
potential vorticity will not be constant at each point, even though it will be conserved
following a fluid element. When the background potential vorticity is not constant,
waves may now possess nonzero potential vorticity.
Let's try to estimate the order of magnitude of the frequency of such a wave in which
the rate of change of pv is produced by motion in the field of varying ambient poten-
tial vorticity. The magnitude of the rate of change of pv can be estimated as:
• wU / L, where U is the characteristic fluid velocity in the wave and L is the charac-
teristic horizontal scale of the wave (so that derivatives in x and y go like 1/ L).The
last term we can estimate as f3u, and this yields an estimate of the frequency;
• w = O(f3L), that is, of the order of the gradient of ftimes the north-south excursion
of the fluid element. The ratio of w to f will then be:
• wI f = O(f3L110) « 1, if the beta plane approximation holds. That is, these waves, in
distinction to the Poincare waves, will have frequencies less than the Coriolis param-
eter; they will have time scales long compared to a day and be parametrically separated
from the spectrum of gravity waves. Note, too, that this wave, again in distinction to the
Poincare and Kelvin waves, owes its very existence to the presence of rotation. We need
to discover the relationship between the Rossbywave, as this f3 dependent wave is called,
and the earlier gravity waves we have discussed. We must formulate an equation that
governs both and then see how each wave type emerges from the governing equation.
To do so, it is helpful to introduce the transport variables
U=uD (I4.11a)
V=vD (I4.11b)
where D is the undisturbed depth. In terms of these variables,
Ut - fV = -gD1]x (I4.12a)
Vt +jU=-gD1]y (I4.12b)
1]t +Ux + Vy =0 (14.12c)
Cross-differentiating the momentum equations yields
(VX -Uy)t + f(U x + Vy )+ j3V =-g(Dx1]y -Dy1]x) (14.13)
The divergence of the momentum equations yields
(Vy +Ux)t - f(Vx -U y )+ ,8U=-g[(D1]y)y + (D1]JJ (14.14)
We can eliminate the vorticity-like term between the two equations by taking the
time derivative of the divergence equation and adding to it the vorticity equation
multiplied by f to obtain
a
9t(Vy +Ux )+ !3(Ut + fV)=-g-V ·(DV1])+ gf](1],D) (14.15)
at
In the above equation, we have introduced two new operators,
9t(a)=( :t22 + f2]a (14.16a)
](a,b)=axby -aybx (14.16b)
where a and b are arbitrary functions of x, y and t.

The equation for mass conservation allows us to eliminate the divergence of the
transport to obtain
-'J'-17t + p[U t + jV]=-gY' ·(DY'17t)+ gf](17,D) (14.17)
Note that were f3 = 0, we would have a single equation in 1], and it would in fact be
the equation previously derived for gravity waves in the presence of rotation with the
important exception of the derivatives of D on the right-hand side. To obtain a single
equation in 1], we must work a little harder.
As before, we can derive equations relating U and V to the free surface elevation.
From manipulating the momentum equations we obtain, as before,
IJiU = -gD1Jxt - gDf1Jy
(14.18)
IJi V = - gD1J yt + gDf1Jx
Using these relations, we can eliminate U and V from the previous equation for 1]
to finally obtain
a
-'J'-[gY' ·(DY' 17)-'J'-17]- gf'J'-J(17,D)+ p(- gD17xtt )-2/3 gDf17yt + PgDf 217x =0 (14.19)
at
, i '2
'---v--------' '------,,---'
3a
'-r-'
3b
'-----v--'
3c
This is a single equation for the free surface height. It is valid (or should be) for
both Poincare and Rossby waves, but given that the former have frequencies greater
than f and the latter have frequencies less than f, some terms in the equation may be
important for one wave and not for the other.
Let's estimate the various bracketed terms in the above equation for the case of the
waves that have frequencies greater than f We will estimate each term separately and
then their ratios:
(1) = 0(0)3 c~ 1L2)17 (2) = 0(gf0)2hb / L2)17
(3a) =0(fJc~0)21L)17 (3b) =O(fJc~fO)IL)17 (3c)=O(fJc~f21L)17
Here we have estimated the operator 'J'- = O(ar) and have used L to estimate hori-
zontal derivatives.
The ratio (2) / (1) is
(2) fg0)2hbL2 f hb
(14.21)
(i) L20)3C~ = 0) Ii
For waves that have frequencies 2': f, it follows that for such waves the second term
will be small with respect to the first term, since we have assumed hb « D. Similarly,
(3a) _ fJL < fJL «1 (14.22)

(1)- 0) - f
as a consequence of the beta plane approximation.

It also follows that
(3b)_flLf < flL «1 (14.23a)

(1)- oi - f
(3c) _ flL f2 ~ flL «1 (14.23b)
(1)- 0)3 f
Thus, for waves whose frequencies exceed f, the governing equation within an er-
ror of the order of (f3L / fo' hb / Do) remains the same equation as before; namely,
a [ C2V21J_--.!J...-
9\-
at 0
a2 1'21J ] =0
at 2 )0
(14.24)
so that we will obtain the same Poincare and Kelvin waves as before and the new terms
in the governing equation will give rise, at most, to small corrections to the frequency
and structure (one might be interested in finding those corrections but our main con-
ceptual point here is that they are just corrections to the basic rotational-gravity waves
we have already found).
On the other hand, for 0)« f, the balance of terms will be quite different. For ex-
ample, the operator 9\ = o(f) and each term can be estimated to have the order
(1)=(0Jj2C~ /L2)1J (2) = (j3 ghb /L2)1J
(3a) = (fJc~0)2 /L)1J (3b)=(fJc~0Jj /L)1J (3c)=(fJc~t /L)1J
Therefore, for frequencies in the range of the estimated Rossby wave frequency,
(2) -L hb =0(1)
(1)- 0) D
(3a) _ flL 0) «1
(1)- f f
(14.25)
(3b) _ flL «1
(1)- f
(3c) = flL =0(1)
(1) 0)
so that for low frequency motions, the approximation to the governing equation is
f2 :t [C~V21J - f21J]- f3 g](1J, D) + /lgDf 21Jx =0 (14.26a)
or
~[V21J-f:
at Co
1J]+P1Jx+L!(1J,hb)=O
Do
(14.26b)
Recalling that f =fo + [3y and that the second term is much smaller than the first,
we have a uniform approximation to the above equation as
ot
[2 f~]
-o V 17-- /
2 17 + J(17,f + fohb Do)=O
Co
(14.27)
where we have also used the smallness of hb with respect to Do. Note that in this equa-
tion, supposedly valid for low frequency waves, f and D are considered constants ex-
cept in places where they are spatially differentiated. We will have to work a little harder
to justify this heuristic derivation, but the outlines of the scaling justification should
be clear at this point.
Let's look for plane wave solutions, and to make the notation simple to begin
with let's examine the simple case where hb is a function only of y and such that its
derivative with respect to y is constant, i.e., a constant bottom slope. Plane waves of
the form
17 =Aei(kx+ly-mt) (14.28)
will be a solution of the above equation, if
fJk (Rossby wave dispersion relation). (14.29)

llc5
0)=
k2 +12 +
Here we have defined
/J = fJ + fo Ohb (14.30)
oy
as an "effective [3."
There are a number of astonishing properties of the dispersion relation. The dis-
persion relation itself is shown in Fig. 14.3.
In Fig. 14.3, the frequency is scaled by [3col fo and the wave numbers by the defor-
mation radius Col fo' and the y wave number has been chosen to be fol co'
First of all and most striking is that for each positive k there is only one value of m and
it is always < O. So, the phase speed of Rossby waves is always towards negative x (in this
t3
case where the bottom slope is in the y-direction and where we assume that > 0). The
topographic slope could have the opposite sign, and the wave could have its phase propa-
gate to positive x, but the important thing is that there is only one value for the phase
speed. Previously, for all the gravity waves we have studied, for each wave propagating
to the right, there was one propagating to the left with a frequency of the same mag-
nitude. This is not true for the Rossby wave. Space is no longer dynamically isotropic.
The dynamics recognizes, for example, which way north is by the direction of the in-
crease of f. Once there is a special direction in space picked out for the wave, all its
properties will manifest that non-isotropy.
The phase of the wave propagates in such a way that an observer, riding on the wave
crests and looking in the direction of propagation, would see higher ambient poten-
tial vorticity on his right.
0.00 " - - . - - - - - - - , - - - , - - - . - - - - - - - , - - - , - - - , - - - - - - , - - , - - - - - ,
-0.05
-0.10
-0.15
-0.20
-0.25
-0.30
-0.35
-0.40~1----~----~----~----~----~----~----~----~----~-----"
o 2 4 6 8 10 12 14 16 18 20
Rossby dispersion relation /co / f = 1, O)min = -0.35355 kco / f ----+
Fig. 14.3. The dispersion relation for Rossby waves
The maximum frequency (numerically) as a function of k will occur when

(
k=1!2+fi/c~
)112 (l4.3la)
-OJmax /J (14.3lb)
2~12 + f02 / c~
Over both k and I, the maximum frequency will occur when the y wave number is
zero (i.e., when the y wavelength is very much less than the deformation radius) so
that the overall maximum of the Rossby wave frequency and thus the minimum of
the Rossby wave period is
-OJ -
max(allk,/) -
/Jeo
2fo (l4.32)
Second, it is dynamically impossible at this level of approximation to distinguish

between the effect of the Earth's sphericity and the effect of a uniform bottom slope
on a flat Earth in providing the necessary ambient potential vorticity gradient to sup-
port the Rossby wave. This fact has often been used with profit to construct labora-
tory models of oceanic waves or circulations that depend on the effect of {3 by instead
introducing a uniformly sloping bottom.
The fact that the phase speed is always directly to the west (if we think about the
planetary beta factor) seems puzzling. What would happen to wave energy in a basin
if it always moved westward and never had a chance to return eastward? Of course,
we are already alert to the fact that the energy moves with the group velocity and not
the phase speed, so it is important to calculate the group velocity.
From the dispersion relation (and for now we will simply write f3 for the effective
ambient pv gradient),
k2 _(12 + f02/ C~)

(I4.33a)
C
gx
=j3 [e+[2+ fo2 /Co2]2
2k[
Cgy = j3 [k 2 + [2 + fi /c~ y (l4.33b)
The group velocity in the x-direction is of two signs, although the phase speed is
always negative. For k2> [2 + N / C6, i.e., for waves that are short in the x-direction, the
group velocity component in the x-direction is positive, while for long waves in x, that
is, k2< [2 + fa2/ C6, the group velocity is negative, i.e., westward. Long waves have their
energy propagating westward, and short waves have their energy propagating east-
ward.
Figure 14.4 shows the group velocity in the x-direction as a function of k (scaled
with the deformation radius).
0.2·,----,,----,-----.-----,-----.-----.-----.----,-----,-----.
0.1
0.0
- 0.1
-0.2
cgx!(f3L~)
- 0.3
-0.4
2 4 6 8 10 12 14 16 18 20
leo / fo = 1 kLo --+
Fig. 14.4. The group velocity in the x-direction for the Rossby wave
The group velocity in the x-direction is, of course, zero when k2 = 12 + f02 I C6, and it
has its positive maximum at
k=~~2+ fi IC5r/2
That maximum positive group velocity in the x-direction is
fJ (14.34)
Cgxmax 8~2 + fi I c5]
while its minimum group velocity, or equivalently, its maximum negative group ve-
locity occurs at k = 0, the longest waves in x, and is equal to
C5 2
Cgxmin =- fJ f02 =- flLD (14.35)
where we have introduced the notation LD == Col fo for the deformation radius. Note that
the maximum speed to the west is eight times greater than the speed to the east. The
westward moving long waves have a much swifter speed of energy propagation than
the shorter waves, whose energy moves eastward.
The group velocity in the y-direction can have either sign, depending on the sign
of the product of k and 1. Note that since
OJ
C =- (14.36a)
y 1
2z2 <0 (14.36b)

cgylc y (k 2 +12 +llLt)
the group velocity in the y-direction is oppositely directed to the phase speed in the
y-direction. This is reminiscent of the oppositely opposed phase and group speeds in
the vertical direction for internal gravity waves, and it is left to the student to develop
and complete the analogy.
Lecture 15
Rossby Waves (Continued), Quasi-Geostrophy
For the Poincare wave, w?:: f, and so the wave motion is not in geostrophic balance,
while for the Rossby wave,
w~ jJLd = flea / f (15.1)
so that
wif = jJLd < jJL

f -j
(15.2)
where L is the scale of the motion. Thus, for Rossby waves, the frequency is less than f
so that in the x-momentum equation, for example,
1 2
,-'-, ,-'-,
Ut - tv = -g1Jx (15.3)
term (1) will be less than term (2) by the order of w / f The velocity will be in approxi-
mate geostrophic balance to that order. This is similar to the hydrostatic approxima-
tion in which the vertical pressure gradient can be calculated as if the fluid were at
rest, even though it is motion, because the vertical accelerations are very small when
the aspect ratio D / L of the motion is small. Here the horizontal pressure gradient is
given by the Corio lis acceleration as if there were no acceleration of the relative ve-
locity, i.e., as if the flow were uniform in space and time even though it is not because
that acceleration is very small compared to the Coriolis acceleration.
These simple intuitive ideas form the basis of a formal theory, quasi-geostrophy, that
systematizes that idea (Pedlosky 1987). The reason why we have to be formal is that other-
wise (and brutal historical experience shows the foolishness of taking the careless path) it
is not clear how to proceed in the approximation beyond its initial step to arrive at an equa-
tion of motion that is dynamically consistent and conserves in appropriate approximate
form all the conservation principles present in the original, more complex set of equa-
tions. We want the simplified set so that we can penetrate more deeply into the low fre-
quency limit of the dynamics, which is of special interest in oceanography and meteorol-
ogy, but we clearly want to do it right. It is always easy to do it wrong.
For example, if the Coriolis parameter varies, where can that variation be ignored
and where must it be maintained? The same question will hold with regard to the varia-
tion of the depth, which we saw in the last lecture acted dynamically similarly to the
{3-effect. If the motion is in geostrophic balance at the lowest order, how can we con-

160 Lecture IS . Rossby Waves (Continued), Quasi-Geostrophy
sistently calculate its evolution in time or its structure in space? Geostrophy only tells
us that if we know the velocity, we can calculate the pressure, or vice-versa, but it does
not tell us how to calculate either of them from initial or boundary data. Our task now
is to take up this question, and our goal is to derive a set of equations for the low fre-
quency motion of the fluid, in this example a homogeneous layer of fluid, that is sim-
pler than the initial set but rich enough to allow us to go beyond the investigation of
simple plane wave theory.
To do so, we must bring to the analysis certain physical ideas. Nothing here is, to
begin with, automatic. Based on our experience, we describe a set of consistent pre-
sumptions and find the dynamics consistent with those presumptions. If those a priori
ideas are valid and physically interesting, the resulting equations will give us interest-
ing results; otherwise, they will not.
We presume, a priori that the time scales of the motions of interest are long com-
pared to 11f. Or, more formally, if T (think of a wave period) is the time scale of the
motion such that
ata = O(I/T) (15.4)
then we presume
fT »1, i.e. (OJ/ f)« 1 (15.5)
We also assume that there is a length scale, L, which characterizes the horizontal
scale of the motion such that horizontal derivatives can be estimated by 11 L. Further,
we assume that there is a scale for the fluid velocity, U, which characterizes the mo-
tion of the fluid. This means that the nonlinear part of the total derivative, i.e., terms
like uvx' will be of the order of U2 1L, and this introduces an advective time scale,
Tadvective = L I U. The condition that the advective time scale be long compared to the
rotation period is fTadvective» 1 or equivalently that
U
£=-«1 (15.6)
fL
where £ is the Rossby number. Actually, we will define the Rossby number in what
follows in terms of the constant value off at the reference latitude so that
U
£=- (15.7)
foL
will be a constant.
We need to carefully estimate all the terms in the equations of motion and obtain an
easy way to keep track of their relative sizes. That is most efficiently done by introduc-
ing non-dimensional variables. These non-dimensional variables will be 0(1) if we have
chosen the scale for time, length and velocity correctly for the motion of interest.
Lecture 15 . Rossby Waves (Continued), Quasi-Geostrophy 161
We introduce non-dimensional variables as follows; they will temporarily be denoted

by primes.
(x,y) = L(x',y') (15.8a)
(u, v) = U(u', v') (15.8b)
t= Tt' (15.8c)
We must also scale the free surface height TJ. How should we do that in a way that is
consistent with our scales for velocity, length, and time and our presumption that the
motion is of low frequency? We anticipate that the motion to the lowest order will be
both hydrostatic and geostrophic (almost; that is where the quasi comes in) so that to
the lowest order we expect that
gVTJ = O(foU) (15.9)
but if our estimates of spatial scale are correct,
gVTJ = O(gTJ I L) (15.10)
therefore,
~=O(fo~L J (15.11)
and so
!L=O(foULJ=tJ02L2 =t:(~J2 (15.12)

Do gDo gDo LD
where the deformation radius is, as it was defined earlier,
I
LD = col fo = (gDo)1I2 fo (15.13)
For motions whose lateral scale is of the order of the deformation radius, we can
expect that the proportional change in layer thickness due to the motion, i.e., TJ I D will
be of the order of the Rossby number and hence small. If L is much larger than the
deformation radius, we may still be able to consider the proportional change in layer
thickness due to the motion as small, if the product is
( foUL =
gDo
J t:(~J2
LD
«1
i.e., for a small enough E.

162 Lecture 15 . Rossby Waves (Continued), Quasi-Geostrophy
We therefore formally introduce the scaling for the free surface height,
I] = Do£FI]' (15.14a)
F= f02L2 (L)2 (15.14b)

gDo - LD = 0(1)
This allows us to write for the total thickness of the fluid
D=Do(l+£FI]'- ~:) (15.15)
and we will assume throughout that
~«1
Do
Similarly, with the beta-plane approximation in the representation of f, we take
f=fo(l+~/fo)=fo((l+~ Y') (15.16)
and we assume that the parameter {3L / fo« 1.

Now we insert each of these relations into the equations of motion; for example, in
the x-momentum equation we have
U ou' U 2 r "u x'+Y,,]

--+-lu Ul' - I' (
JO jJL ')'
1+-y Y =-JOUI]
I' ' x' (15.17)
Tot' L fo
Dividing by the factor foU yields
£T ~~: +£[u'u'x.+Y'U,y.]-(I+ ~ Y}'=-I]'x' (15.18)
where
1
£T=-«1 (15.19)
foT
Similarly for the y-momentum equation we obtain
B
T
OY' + £ru' y' . +y' y' .]+
ot' ~ x y
(1 + jJLfo Y')u' = -I]' .
y
(15.20)
The conservation of mass equation in dimensional units is
I]t +(uD)x + (YD)y =0

and by inserting the scaling variables above and the form for D, i.e.,
D=Do(l+EF1]'-~ )
we obtain
Do EF1]'t.+Do U [{u'(1 +EF1]'-hb I Do )1x. + {v'(1 +EF1]'-hb I Do)}

T L Y
.]= 0 (15.21a)
or
ETF1]'t·+u'·"il'(EF1]'-hb I Do)+(1 +EF1]'-hb I Do)"il· u'=O (15.21b)
At this point, our equations look as if they have come down with a bad case of acne;
the primes make the equations look very ugly. The traditional thing to do at this point
is to improve the aesthetic quality of the development by dropping primes. Henceforth,
unless otherwise noted, unprimed variables will be non-dimensional, and we will use
asterisks to denote dimensional variables, e.g., x* = Lx.
Our dynamical system of equations can now be neatly written as
au - nv u+
ET-+CU· flL)'zXU=-v
- ( 1+-y - n 1] (15.22a)
at fo
ET
at +u·"il(EF1]-hb IDo)+(I+EF1]-hb /Do)"il·u =0
a1] (15.22b)
These equations contain several small parameters. There are the two Rossby num-
bers ET and c, as well as a measure of the sphericity factor {3L 110, and of course, the de-
viation of the rest thickness of the layer in absence of motion from the constant mea-
sured by hbl Do. We will assume that the parameter F is order one, i.e., that the hori-
zontallength scale is of the order of the deformation radius. We will expand the equa-
tions of motion in an asymptotic series in c and assume that each of the Rossby num-
bers is of the same order, that is, that
E _ UT =Ulc=O(I) (15.23)
ET - L
and if we want to subsequently linearize the resulting equations, we can assume at the
end of our labors that this parameter is small with respect to one. We will also assume
that {3L I fo is of order c, which implies that
_E U
(15.24)
flL/ fo = flL2 =0(1)
The Rossby number, c, is a ratio of the relative vorticity, of the order U I L to the plan-
etary vorticity, f It is assumed small. The ratio above U I {3L 2 is the ratio of the relative
vorticity gradient to the planetary vorticity gradient and that can be order one. That is
because the relative vorticity varies relatively fast on the scale L, while the planetary
vorticity varies more slowly on the scale of the Earth's radius, R. The fact is that L I R « 1
is a requirement of the beta-plane approximation.
So we expand each variable in the series:
u(x,y,t,t:) =uo(x,y,t)+t:U1(x,y,t)+ ... (15.25a)
1J(x, y,t,t:) = 1Jo(x, y,t) + C1Jl (x, y, t) + ... (15.25b)
Note that each subscripted variable is independent of £. Thus, when this series is
inserted in the equations of motion, like orders in £ must balance for the equations to
be valid for £ small, but arbitrary. This leads to the following set of equations.
Collecting the 0(1) terms in the momentum equation,
zx Uo = -"\1110 (15.26)
or in component form,
Uo = -11oy (15.27a)
Vo = 110x (15.27b)
which is simply the geostrophic balance at the lowest order (note that the variation of
the Coriolis parameter does not enter at this order; it as iff were constant in the low-
est order geostrophic balance). Note that as a consequence of geostrophy,
dUO + dVo =0 (15.28)

dX dy
The geostrophic velocity, with constant f, is horizontally nondivergent.

When we look for order one terms in the mass conservation equation, the develop-
ment depends on whether hbl Do is order one or less. We will assume it is 0(£) so that,
as the beta effect, that variation does not enter at the lowest order. It is left for the stu-
dent to discover what the dynamics will look like if hbl Do is 0(1). If hbl Do is 0(£),
then all terms in the equation for 11 are of order £, noting that the horizontal diver-
gence of the O( 1) velocity vanishes.
We then are left at order one, with only the diagnostic relation between the pres-
sure gradient and the geostrophic velocity with no way, at this order, to calculate the
evolution of the fields, e.g., to discuss Rossby waves. We must go to a higher order in
our expansion to do so. It is precisely for this reason that higher order small terms
must be considered and that we must be exquisitely careful to consider all small terms
that are of the same order. It is for this reason that we have gone through the scaling
and the non-dimensionalization so that we can be sure we are not leaving a small term
out while considering others. We need to keep the dynamics consistent if the final re-
sult is to be physically sensible.
At O(e), i.e., keeping terms of O(e), the conservation of mass and momentum equa-
tions yield
ET aT/o
-F--+Fu
E at
n - n(
- o 'v1]o-uo 'v - hb +v'UI=O
EDo
J n- (15.29a)
ET auG fJL2 _
--+uouo x +VOUo y -v1- y-vo --1]lx (I5.29b)
E at U
ET avo fJL2
--+uovox +VOVOy +uj + y--uo =-1]ly (15.29c)
E at U
Here we have used the fact that
fJL _ fJL2 = 0(1)

Efo - U
and that
~=0(1)
EDo
in identifying terms of order e. Note that these O(e) equations describe the rate of
change with time of the 0(1) velocities and free surface elevation. However, the equa-
tions contain the O( e) variables as well, and so the system does not seem closed at this
order. This is a little bit worrisome. Let us press on, though, by eliminating the O(e)
free surface elevation from the momentum equations by cross differentiating. Using
the fact that the 0(1) velocities have zero divergence, we obtain an equation for the
evolution of the relative vorticity;
e fJL2
..L SOt +UOSO X +voSoy +-vo =-(v1y +uIJ (15.30)
£ U
where So = vox - UOy is the relative vorticity.
The interpretation of the above equation is rather interesting. The left-hand side
of the equation yields the total rate of change of the sum of the relative plus planetary
vorticity following a fluid element; in dimensional variables this would just be
ds+f
dt
The right-hand side of the equation is minus the product of the Coriolis parameter
at the reference latitude,fo and the divergence, i.e., -fo(ux + vv )' In our scaling universe, the
Coriolis parameter is order one, and the horizontal divergence is O(e), so the product is of
the same order as the rate of change O(e) of the order one relative vorticity. Note that this
source of vorticity normally would contain the convergence not only in the presence of
the reference Coriolis parameter but of the full vorticity f + S. However, those corrections
are of a higher order in Rossby number, it would not be consistent to keep them, and
indeed, they do not appear in the non-dimensional vorticity equation we have derived.
This is one of the advantages of the careful bookkeeping that the method does for us.
We still are in some difficulty, apparently, because the rate of change of the relative
vorticity is given by the higher order divergence, which we don't know. We can elimi-
nate the divergence, though, through the use of the equation for mass conservation.
Thus,
~«(o - F1]o) + Uo . v{h + hb/£l)o}= 0 (15.32)

dt
We have defined
do = eT ~+(uo. V) (15.33a)
dt- e dt
, jJL2
/3=- (15.33b)
U
Indeed, the equation can be written in conservation form more simply as
do r '
-l(o-F1]o+j3y+hb/eDO =0
] (15.34)
dt
All variables in the above equation are 0(1), and the equation is a conservation
equation for an 0(1) variable. What is that quantity? By now you should have the feel-
ing from its form that is the potential vorticity or some suitable approximation to it
valid for a small Rossby number. We shall check that shortly, but first we need to make
a very important point. The equation is a single equation in several variables, So' 110
and the two velocity components. However, the 0(1) geostrophic relation allows us to
write all the variables in terms of the free surface elevation, since
Uo =-1]Oy (15.35a)
Vo =1]ox (1S.35b)
(0 = vox -UOy = 1]oxx +1]Oyy = V21]0 (15.35c)
Noting that the lowest order free surface elevation plays the role of a stream func-
tion for the 0(1) geostrophic velocities, we define
If/= 1]0 (15.36)
in terms of which the above equation can be rewritten:

d [ 2
.-JLLV Ij/- FIj/ + fly + hb/ cDo =0 (15.37a)
A ]
dt
or
ST ~[V2'1f -F'If] +'lfx ~[V2'1f-F'If]-'lfy[V2'1f -F'If]

S at cy (15.37b)
+ 'lfx ~y +hb/SDoL - 'lfy[.By+hb/SDoL = 0
Thus, we have attained a governing equation in the single variable Iff. We introduce
the notation for the Jacobian,J(a,b), of any two functions a and b:
J(a,b) =axby -aybx

The equation of motion is thus
c a[ ]
...I..- lv 21j/-FIj/ + J(Ij/,V 21j/-FIj/+ fly+hb/cDO) =0 (15.38)
A
c at
This equation forms the heart of our analysis of quasi-geostrophic motion, but
before we proceed to its analysis and in particular its role in wave theory, it is useful
to understand the origin of the equation in a more heuristic manner than our careful
asymptotic derivation.
For a single layer of fluid, in dimensional units, the equation for conservation of
potential vorticity, assuming only that the motion is hydrostatic and the horizontal
velocities are independent of z, is
~[; + f]=o (15.39a)

dt D
D=Do+TJ-hb (15.39b)
D is the total depth, and it departs from a constant value by a small amount; in-
deed, we can approximate the potential vorticity
_; + f =; + fo+fly ",_1 «( + fo+fly)(1-{TJ-hb}IDo +... ) (15.40)

q- D Do +TJ-hb Do
using the expansion for 1 I (1 + £) = 1 - £ + 0(£)2.

Keeping only those terms in the above product that are of either order one or of
the same order as the relative vorticity, we get
fo
q"'-+=-~~~~~-=
[; + fly - foTJ I Do + fohb I Do ] (15.41)
Do Do
The first term in this approximation for q is an irrelevant constant. The conservation
equation then applies to the second term, which aside from a multiplicative constant yields
~(( + fly -
dt
fo" + fohb = 0
Do Do
J (15.42)
At the same level of approximation, the geostrophic relation yields, in dimensional units,
u=-.K.."y (15.43a)
fo
v=.K.."x (15.43b)
fo
Thus, the pv equation becomes
.K..~[V2,,_ fi" + fJfoY + f02hb] =0 (15.44)

fo dt gDo g gDo
Defining (remember these are in dimensional units)
If/=.K.." (15.45)
fo
the potential vorticity equation becomes
dq "'~[V21f/_+If/+fohb +fly]=O (15.46)

dt dt LD Do
If we were to scale x and y as we did earlier in the lecture, the above equation would
become the dimensionless dynamical equation we obtained earlier in our more care-
ful scaling and asymptotic expansion method. That care allowed us to be sure that in
our estimate of potential vorticity we included all the correct terms (and no more)
and that we could replace f with its reference value in the geostrophic relation and
definition of stream function.
The end result is that for low frequency motions with a small Rossby number,
the governing equation of motion is the potential vorticity equation in which
all terms are evaluated using their hydrostatic and geostrophic approximations
in terms of the pressure field, in this case, the free surface height.
Note that the total derivative in dimensional units is
d a alf/ a alf/ a
-=-+----- (15.47)
dt at ax ay ay ax
As we remarked earlier, for quasi-geostrophic motions the role of the bottom slope
mimics that of the beta effect, and one can use a constant bottom slope in the labora-
tory to model the dynamical effect of the Earth's sphericity. We can see, perhaps more
easily from the non-dimensional form of the equation, that the relative importance of
the beta effect and the bottom slope depend on the ratios
jJL2 fohb
U U/L
The first is the ratio of the planetary vorticity gradient to the relative vorticity gra-
dient, and that ratio is typically unity, although as the horizontal scale grows larger the
beta effect tends to dominate, because for a given scale for U, the relative vorticity and
its gradient decrease with L. The topographic term can also be thought of as a ratio of
the contribution by the topography to the potential vorticity gradient lfohbl Do) I L to
the relative vorticity gradient U I L 2 , and again as L increases, the topographic term
tends to dominate (Note that the equivalent topographic beta is of the order
fohb
fJtopog = DoL
so the ratio given above can be written
fJtopogL2
U
to complete the analogy). These terms are the contributions to the ambient poten-
tial vorticity, that is, the potential vorticity preexisting in the absence of any motion.
When the ambient potential vorticity dominates, i.e., when the above ratios become
very large, the first approximation to the potential vorticity equation is just
{V/,~+f~b J=o (15.48)
The stream function must then be constant along lines of constant ambient pv in
the x-y-plane, which is an extraordinarily strong constraint. Breaking that constraint
requires either a source of potential vorticity that will nudge fluid elements off the
lines of constant ambient pv or regions in which dissipation (hitherto ignored) or
nonlinearity become important. These considerations are of vital importance in the
theory of the general circulation of the ocean, but pursuing them further here would
divert us from our goal of understanding the physics of waves. Therefore, we return
to the quasi-geostrophic potential vorticity equation (qgpve) given above. We will
work in dimensional units, although our systematic derivation leaned heavily on our
scaling and asymptotic approximations for a small Rossby number.
Quasi-Geostrophic Rossby Waves
We return to the qgpve and examine the nature of Rossby waves in the presence of an
ambient potential vorticity gradient. For simplicity, we will take the ambient gradient
to be a constant. We define the ambient pv as
Q=~+ fohb/DO (15.49)

and assume its gradient is constant. The linearized form of the qgpve is then
ata LVr, 2If/-a 2]

If/ +If/xQ y -If/yQx =0 (15.50)
The quantity a2 is 1/ Lband has the dimensions of a wave number (squared). With
the above assumptions on Q, the equation has constant coefficients, and if we consider
an infinite region, we can find plane wave solutions in the form
If/ =Aei(kx+ly-M) (15.60)
which requires that
-iw(-k2 _12 -a 2)+ikQy -ilQx =0
_ [Qyk - Qxd __ ' . [ f( x VQ ]

~ OJ- - - - Z 2 2 (15.61)
K +a
=-f(.[ ZXVQ]
K2+a 2
where f(is the two-dimensional wavevector,K2= k2 + 12,while i is the vertical unitvec-

tor. Finally, the frequency can be written
f(.(zxVQ)
OJ (15.62)
K2+a 2
So the frequency depends on the projection of the wave vector on the direction
perpendicular to VQ, i.e., it depends on the projection of the wave vector along the
lines of constant ambient pv. Since the geostrophic velocity is perpendicular to the
wave vector (why is this so?), the frequency depends on the degree to which fluid ele-
ments cross ambient pv contours. If the fluid flows along pv contours, i.e., if it . VQ = 0,
the time derivative in the linearized qgpve would be zero: no wave (Fig. 15.1).
y
VQ R
~v X
Fig. 15.1.
The relation between the wave
vector, the ambient pv gradient
and the direction of the phase ixVQ
speed
Note that the phase speed in the direction of the wave vector is
_ K
OJ
C=--
KK
z
and the inner product of this pseudo vector with the vector x V Q yields
c.(zxVQ)= [i.{zxVQ}y (15.63)

K 2 +a- ;::0
Thus, the phase always moves in the direction such that an observer riding on a
crest will see larger values of Q on her right. As discussed earlier, the phase propaga-
tion is in a single direction, essentially such as to make an acute angle with the isop-
leths of ambient pv and to be guided by its gradient.
The case of a flat bottom arose when VQ = f3 j was discussed in the last lecture. What
we need to do now is to develop a clear picture of the direction and magnitude of en-
ergy propagation in the Rossby wave. This is rendered a bit tricky because of the dis-
concerting fact that the obvious candidate for the energy flux at the lowest order in
Rossby number pu is horizontally nondivergent. That is, if the velocity is geostrophic,
both its divergence and its inner product with grad p vanish identically. That is not a
useful tool for calculating the transfer of energy. The difficulty is only resolved by
noting that as in the case of the dynamics, the energy flux will involve the 0(£) contri-
butions of the pressure to calculate its gradient and 0(£) contributions to the velocity
to calculate the velocity's horizontal divergence. That awkwardness can be avoided by
dealing directly with the qgpve, as we shall show in the next lecture.
Lecture 16
Energy and Energy Flux in Rossby Waves
In discussing the energy and its flux for Rossby waves, we encounter the problem that
the natural definition of the energy flux at the lowest order pii is horizontally non-di-
vergent and therefore has no effect on the wave energy. To discuss the real energy flux,
one has to include the divergent, non-geostrophic 0(£) part of the velocity field as well
as the pressure contribution at this order. This would be a messy business, and what is
worse is that the solution of the quasi-geostrophic potential vorticity equation doesn't
give us these quantities as part of the solution. Is there a way we can describe the en-
ergy flux entirely within the quasi-geostrophic framework? The answer is yes, and it
follows from a direct consideration of the linear quasi-geostrophic equation. First,
though, let us orient the y-axis in the direction of the gradient of the ambient potential
vorticity, VQ, and call the magnitude of the gradient f3 for obvious reasons. As long as
the gradient is a constant, there is no loss of generality. It will be up to the student to try
to generalize these results when the gradient is not constant. The linear qgpve is
~[V2V'-a2V'l+paV' =0 (16.1)
at ax
If we multiply the equation by the stream function, we obtain
V· (1jIV V't)- V V" V V't - a 2V'V't + PV'V'x =0 (16.2)
which can be rewritten:
~[(V V')2 + a2V'2]+ V

.{_1jIV V't _~XpV'2} =0 (16.3)
at 2 2
x
where is a unit vector in the x-direction. From the definition of the stream function,
it is clear that the first term in the square brackets is the kinetic energy
1
_(U 2+V2)
2
which is associated with the 0(1) geostrophic motion. The second term is potential
energy, since
t02 g2lJ2 glJ2

a2V'2 _
(16.4)
-2-- 2c~ t02 = 2Do

174 Lecture 16 . Energy and Energy Flux in Rossby Waves
(multiplication of the whole equation by Do is necessary to give the total energy in the
water column, but this obvious step is trivial).
Therefore, the first term in the above equation will be the sum of the kinetic and
potential energies in the wave field. The term in the curly bracket is a vector, S, whose
divergence alters the local wave energy. Note that S is given entirely in terms of the
geostrophic stream function, IJI.
Thus, we have the usual energy flux equation:
dE -
-+V·S=O (16.5a)
dt
S= -IJIV IJIt - xf31J12 /2 (16.5b)
To get a better feeling for the flux vector S, consider a Rossby wave packet
IJI= Acos(kx + Zy - rot) (16.6)
where the amplitude A is a slowly varying function of space and time. Let's calculate
the energy. The kinetic energy is
H ] A2
KE =-Lv/; + v/~ =_(k 2 +Z2)sin2(kx+Zy-rot) (16.7)
2 2
and averaged over a period,
(KE) =£(k 2 +Z2) (16.8)

4
while the potential energy, which similarly averaged, is
(PE) = A2 a2 (16.9)
4
Thus, the total energy in the Rossby wave averaged over a period (or wavelength) is
2
(E)=~(K2+a2), K2=k2+Z2 (16.10)
4
Now we need to calculate the flux vector S. Using the notation
i} = kx + Zy - rot
- V/2
S =-r;N V/t - f3-x
A
2 (16.11)
A2
= -Acos(iJ){roKAcos(iJ)} - f3-cos 2(iJ)
2
Lecture 16 . Energy and Energy Flux in Rossby Waves 175
this becomes, when averaged over a wave period,
- A2 Kk A2 A
(16.12)
(5)=fJ 2 (K2+a2) -fJ x 4
To arrive at this, we have used the dispersion relation
fJk
W= (16.13)
(K2 +a 2)
Part of the energy flux vector is in the direction of the wave number, and a part lies
along the x-axis. It is useful to decompose the flux vector into its x and y components.
y
If is the unit vector along the y-axis,
[i
(5) = fJ A24 l 2k2 -e _/2 _a 2 ]+ y(~ I]
K2+a 2 K2+a 2 )
=
fJA2(K2
4 +a
2)[A(k 2 -/ 2 -a 2 ]
xl(K 2 +a 2)2 +
A(
Yl
2k/
(K 2 +a 2 )2
J] (16.14)
=cg(E)
where we have used the formula derived earlier for the group velocity of Rossby waves;
namely,
_ fJ[Ak 2 -/ 2 -a 2 2k/A ] (16.15)

Cg = X (K 2 +a 2)2 + y (K 2 +a 2)2
This allows us to write the energy equation:
a(E) +V.Cg(E)=O (16.16)

at
The Energy Propagation Diagram
As noted earlier, to obtain the full energy flux written in terms of the pressure work
term, one would have to include the effects of the order Rossby number (t:) velocity.
That velocity is not horizontally nondivergent. Therefore, the total velocity required
for the calculation of the energy flux does not satisfy the condition that it would be
perpendicular to the wave vector. That in turn implies that the group velocity will not
be perpendicular to the wave vector (nor parallel to it). To discuss the relation between
the wave vector's direction and the direction of the group velocity (which is, from above,
the direction of the energy flux), we will employ a graphical development due origi-
nally to Longuet-Higgins (1964). Consider waves of frequency
fJk
W=- -- 2 (16.17)
k2+/2+a
We will use the convention that k > 0, so that m is < O. For a given OJ, the possible
locus of wave numbers in the k-l-plane satisfies
k
e +12 + /3-+a 2 =0
m
=:>
(16.18)
(k_{L})2 +12 =L_ 2

-2m 40i
a
The wave vector must therefore lie on a circle (see Fig. 16.1) in the k-l-plane cen-
tered at the point
(Lo)-2m'
with radius
~ 4m
/3 2
2
_a 2
Note that for the circle's radius to exist, the frequency has to less than the maxi-
mum Rossby wave frequency [3/ 2a.
The point of the circle closest to the origin lies on the k-axis at a distance
k
m
=L_~
-2m
/3
4m
2
2
_a 2 (l6.19)
--- \( -ffJ2/4
\/, OJ 2 -a2
Fig. 16.1. The energy propagation diagram

When the deformation radius Co / fo = 1 / a is very large so that a ----7 0, the point
k m ----7 O.
Now let's calculate the energy flux vector )S(. From our earlier results,
IS)= A2 k~- A 2p X
\ 2 K2+a 2 4
= A2 (-W)[k -~]
2 2( -w)
(16.20)
P- l ]
A2- [ k - -
--
- 2(-w) (-2w) ,
A2 -
=--OW
2(-w)
where the vector oW shown in Fig. 16.1 is directed from the origin of the wave num-
ber circle to the point on the circle corresponding to wave number K. Note that the
length of the vector is constant for all wave numbers on the circle, and of course, so is
the frequency, so that for all waves with waves at that frequency, the magnitude of the
energy flux is constant as long as the amplitude is the same for the waves. The dia-
gram is very helpful in visualizing the relation between wave number vector and group
velocity, and it is immediately apparent that that relation is not a simple one. The group
velocity is neither perpendicular nor parallel to k, and indeed in some cases, it is nearly
in the opposite direction. This is particularly helpful in visualizing the process of re-
flection.
The Reflection of Rossby Waves
Consider a straight western boundary of a basin sloping at an angle ewith respect to

the x-axis as shown in Fig. 16.2. Suppose a Rossby wave in the form of a packet of a
beam of energy impinges on the wall from the right (the east).
~
Fig. 16.2. The reflection of Rossby waves from a western wall oriented at an angle e with respect to
the x-axis. The energy propagation diagram is shown adjacent
In order for the wave energy to be moving eastward and northward as in the fig-
ure, the incident wave vector Ki must lie on the segment of the circle nearest the ori-
gin. That is, the wave must be a relatively long wave. For each frequency and y wave
number, l, there are two choices of k, determined by the dispersion relation:
k=~+~ /3
2(-m) - 4m
2
2
_a 2 (16.22)
The root with the plus sign corresponds to shorter waves and a larger k and hence
with group velocities to the east, while the root with the minus sign corresponds to a
group velocity directed westward and is the root that must be chosen to represent the
incident wave. We represent the incident wave as
If/i = Aiei(kiX+liY-l11jt) (16.33a)
and the reflected wave has the form
If/r = Arei(krx+lrY-llV) (16.33b)
and during the time of interaction with the wall, the total stream function for our linear
problem is the simple sum of the two waves:
If/ = If/i + If/r (16.34)
y "t .......~~~
./ OW,
R;
:><
:7'
K,
x
Fig. 16.3. The incident and reflected wave numbers and their position on the wave propagation circle
Let i be the tangent vector to the boundary (see Fig. 16.3):
i =xcosO+ ysinO
On the boundary, x and yare related by y = x tan e. Thus, on the wall where the to-
tal stream function must be a constant (and we may choose the constant to be zero),
\1'=0 =A;ei[(ki+li tan8)x-tqtl + ~ei[(kr+lrtan8)x-aVl (16.35)
For this to be true for all x along the wall and for all time, it is necessary that
mr = tq (16.36a)
kr + Ir tanO = ki + Ii tanO (16.36b)
=>Ar=-Ai (16.36c)
The first condition that the frequency be conserved under reflection (which we have
seen before in our study of internal waves) means that both the incident and reflected
wave must lie on the same wave number circle. The second condition requires that the
component tangent to the wall of both the incident and reflected waves must be equal.
These, plus the radiation condition that the reflected wave has its energy directed away
from the wall is sufficient to determine the position of the reflected wave on the wave
number circle. Note that the magnitude of the wave number is not conserved under
reflection. Indeed, the wave number vector is lengthened in the reflection process, i.e.,
the reflected wave has a shorter wave length than the incoming wave. Note that since
the amplitude is conserved, the energy of the reflected wave per unit horizontal area
is larger than the energy of the incoming wave. Yet energy must be conserved. Since
both the incoming and outgoing waves are on the same wave number circle and the
amplitude of the wave is preserved, the magnitude of the vector OW is preserved un-
der reflection. Thus, the energy flux of the reflected and incident waves must be the
same. High-energy, slow-moving wave packets leaving the wall are balanced by rela-
tively low energy, rapidly moving packets impinging on the wall. It is left for the stu-
dent to verify that the group speeds meet that condition. It is also left as an exercise
for the reader to show that the angle of incidence of the group velocity is equal to the
angle of reflection of the reflected wave packet with respect to the boundary. That is,
the reflection process is specular.
In the special case when e = 1t 12, i.e., a wall along a longitude, the y wave number
is conserved under reflection and the incident and reflected x-wave numbers satisfy
ki = (!:m) -~ :::2 _a 2
(16.37)
k =~+~ j32 _a2

r (-2m) 4m2
For very low frequency waves, the discrepancy for the x wave number will be very
great, and the western boundary of an ocean acts then as a source of very short (in x)
scale of energy. The reflected wave will have the same zonal velocity as the incident
wave, but its meridional velocity will be much larger.
The group velocity in the x-direction in the limit of very short x wave number will
be of the order
Cgx =o(~)
and is directed eastward. In the presence of a large-scale (Sverdrup dynamics) zonal
current drift, U, the net group velocity will be
Cgxnet =U +L
e
If U were negative, all scales with k larger than {fJ7fJ would not escape from the
generation region. This gives, as a characteristic length scale for a zone of high wave
number energy near the western boundary, 8 1 =~ U / f3, which is the characteristic scale
of the western boundary current in the inertial theory of the Gulf Stream. When the
large-scale flow is directed eastward, U is positive and the energy is not trapped. This
corresponds to the fact that purely inertial models for the Gulf Stream fail in regions
of eastward Sverdrup flow. It is left to the student to calculate the characteristic dis-
tance over which the eastward propagating energy, when U < 0, would decay in the
presence of lateral friction (by calculating the diffusion time for wave number k and
using the group velocity) to deduce the scale of frictional models of the Gulf Stream.
This is a good example of how an understanding of fundamental wave dynamics can
give us insight into even the problems of steady circulation theory.
The Spin-Down of Rossby Waves
In regions far from lateral boundaries, the principal dissipative agent is bottom friction. It
is beyond the scope of this course to review in a complete fashion the nature of the viscous
boundary layer, the Ekman layer, and the student can refer to several texts (e.g.,
Greenspan 1968 or Pedlosky 1987) for a full discussion. Physically, and for simplicity
let's think of a flat bottom there will exist a region of the order of <\ = (Ay! 10)112 near
the bottom boundary where the vertical shear of the velocity will be strong so as to
allow the fluid to satisfy the no-slip condition at the bottom. The thickness of this re-
gion, <\, depends on the Coriolis parameter and the coefficient of vertical mixing of
momentum Ay ., and it is generally very thin. In that layer, the fluid dynamics is no longer
geostrophic, and the presence of friction allows fluid to flow across lines of constant
pressure from high pressure to low pressure. Under a region of cyclonic vorticity where
there will be a low pressure center, the flow in the boundary layer will converge to-
wards the cyclone's center. Since the flow is incompressible, that lateral convergence
must lead to a vertical flux of fluid out of the boundary layer into the geostrophic re-
gion above. That vertical velocity is given by the relation:
1/2
w(x,y,O) = wE =( ~ ) ( (16.38)
2fo
where Sis the vorticity of the geostrophic flow. Note that since Sis of the order U I L,
the vertical velocity satisfies our general scaling expectation between wand the hori-
zontal velocity, i.e., that w = O( Ud I L) where d is the vertical scale of the motion. In
the boundary layer, d is ~ = (AJ fo)1I2, and so the result is certainly plausible. The stu-
dent is encouraged to examine the cited reference for details.
For the purposes of the wave problem in quasi-geostrophic flow, the effect is to al-
ter the equation for mass conservation for the layer ofgeostrophic flow. That comprises
most of the layer (see Fig. 16.4) except for the boundary layer.
Redoing the vertical integral of the continuity equation now yields
D(u x + v y) = -w( top) + w(bottom)

dT/
=--+WE
dt (16.39)
1/2
=_dT/+ ( ~) (
dt 2fo
Redoing the steps leading to the potential vorticity equation then yields an extra
term on the right-hand side of the vorticity equation and the potential vorticity equa-
tion such that now we have
dq "'~[V2¥F-~¥F+ fohb + j3y]=_ fo ( ~)1/2 V 2¥F

dt dt LD Do Do l2 fo (16.40)
1 2
=--'11 ¥F
Ts
/TJ
c:== ~s Do
~. / "• J, tDE
Fig. 16.4. The genesis of vertical motion pumped out of the bottom Ekman layer by cyclonic geo-
strophic motion above the bottom
where Ts is a characteristic decay time for the system due to bottom friction:
Ts=~ (16.41)
(Av/o / 2Y'2
Note that this time, called the spin-down time (or spin-up time for optimists),
increases with the depth and decreases as the mixing coefficient and the rotation get
larger.
For linear Rossby waves, the wave equation becomes
a 2
-[\7 Ij/-a
2
/J1f/ 1 2
=--\7 Ij/ (16.42)
at Ij/J+
x Ts
Solutions for plane waves can be found in the form:
Ij/ = Ae -at ei(kx+ly-M) (16.43)
where (J is the frictional decay rate. Inserting the above form in the wave equation
yields, after equating real and imaginary parts of the dispersion relation,
(J
K2 (1
K2 +a 2 Ts
J (16.44a)
k
w=-/J K2 +a2 (16.44b)
The frequency wave number relation is unchanged, and the decay rate is in fact one
over the spin-down time slightly modified by the scale. Note that when the horizontal
scale is very large compared to the deformation radius, K « a, the decay rate is small,
while for short length scales for which the above equality is reversed, i.e., when the
scale is small compared to a deformation radius, the decay rate becomes independent
of scale (This would exactly occur if there were an upper rigid lid instead of a free
surface. Why?).
For our previous work on waves to have relevance it is necessary that we can ob-
serve at least several oscillations before the wave decays. That is the basis of our ap-
proximation that inviscid theory is pertinent to the wave problem. So we have been
implicitly assuming all along that
T
wT =21t s » 1 (16.45)
s Tperiod
Lecture 17
Laplace Tidal Equations

and the Vertical Structure Equation
Let's return to the linearized wave equations before the gravity waves are filtered out by
the quasi-geostrophic approximation. What we will see now is that the analysis of the ho-
mogeneous model can be carried over, in important cases, to the motion of a stratified
fluid. A vertical modal decomposition can be done for these cases, and we will be able to
show that the equations for each vertical mode are analogous to the equations for the single
layer. Exactly what that relationship is will be the subject of our development that follows.
To keep the discussion simple, we will consider hydrostatic motion but not necessar-
ily geostrophic motion. We will also relax the fJ-plane approximation and consider lin-
earized motion on the sphere. Our coordinates will be () for latitude, cfJ for longitude,
and z for the elevation above the Earth's spherical surface as seen in Fig. 17.1. The ve-
locities in the zonal, meridional and vertical directions will be u, v and w. As before, we
will separate the pressure and density into the values those variables have in the rest
state plus a small perturbation due to the motion
Ptatal =Po(z)+ p(fj),(J,z,t) (l7.1a)
Ptatal = Po(z)+ p(fj),(J,z,t) (l7.1b)
and we will assume that the density field of the basic state Po changes only slightly
over the vertical extent of the fluid (for a compressible fluid like the atmosphere, see
Andrews et. al. 1987), i.e.,
Dapo «1
poaz
()
Fig. 17.1.
The coordinate system for the hydrostatic
equations of motion on the sphere

184 Lecture 17 . laplace Tidal Equations and the Vertical Structure Equation
For linearized, inviscid motion, the equations for the perturbations become
,--"-,,-A-----------.
a b
Po [ Ut -2.QsinOv+2.QcosOw 1
=--_¢-
P
RcosO
(17.2a)
Po[V t + 2.QsinOu ]=_Po (17.2b)

R
c d
-~-
Po 2.Q cosOu = - pz - gp (17.2c)
u¢ (vcosO)o
--+ +w =0 (17.2d)
RcosO RcosO z
Pt+WdPO -0 (17.2e)
dZ -
We have used subscripts for differentiation. R is the (constant) Earth's radius. We have
also assumed in the last equation that the motion is adiabatic. In the momentum equa-
tions, we have included each of the components of the Coriolis acceleration, 2,Q x ii, and
please note that while the contribution in Eq. 17.2a of the component of the Earth's ro-
tation that is tangent to the Earth's surface, 2.Q cose, involves the weak vertical velocity
(we are assuming the vertical scale of the motion is much less than its horizontal scale),
its contribution in the vertical equation of motion depends on the much stronger zonal
velocity. The issue here is if we ignore this contribution in the zonal momentum equa-
tion, can we also consistently ignore it in the vertical equation of motion? It is often
said that if one approximation is made, the other must be made; otherwise if term (b)
is absent but term (c) retained, the dot product of the velocity with the momentum equa-
tions would have the Coriolis force doing work on the fluid, an obvious absurdity since
it is always perpendicular to the velocity. But saving us from absurdity is not a justifica-
tion for an approximation. We must show term (c) is small if term (b) is. Moreover, the
smallness of term (c) in its equation must be measured by the same parameter of small-
ness as term (b) is measured in its equation. If there were different parameters that
measured the relative smallness of those terms in each of their equations, we might find
a situation where one parameter was small and the other 0(1). So, we must see whether
a sensibility scaling argument will let us always ignore both terms simultaneously.
The ratio of term (b) to term (a) is obviously of the order of the vertical to horizon-
tal velocity. If the scale of the former is Wand that of the latter is U, we know from the
continuity equation that
~ =0(~)«1
where D and L are the vertical and horizontal scales of motion. Thus if 8 = D I L « 1,
we can ignore term (b) compared to term (a) in the zonal momentum equation. If the
Coriolis acceleration enters at the lowest order into the dynamics, then this tells us
that the scale of the pressure field, P, must be P = Po2.QUL.
Lecture 17 . Laplace Tidal Equations and the Vertical Structure Equation 185
This does not imply that there is a balance between the horizontal pressure gradient
and the Coriolis acceleration, only that they are both O( 1) terms in the momentum equa-
tion. With that scaling for the pressure, the vertical pressure gradient will be of the order
pz =O(PoZ.QUL/ D) (17.3)
so that the ratio of term (c) to term (d) will be
PoZ.Qucos{} O(DlL) (17.4)

pz
Thus, if D I L « 1, the horizontal component of the Earth's rotation can be ignored
in both the zonal and vertical momentum equations. Our final set of approximate equa-
tions is therefore
Po[U t -Z.Qsin{}v]=-~ (17.5a)

Rcos{}
Po[Vt +Z.Q sin {}u ]= - Po (17.5b)

R
O=-Pz -gp (17.5c)
u¢ (V cos {})o
--+ +wz=O (17.5d)
Rcos{} Rcos{}
Pt +w oPo =0 (17.5e)
oz
As before, we can write the last equation as
Pt-WPON2/g=O, N2=_-.L dpo (17.6)

Po dz
and the use of the hydrostatic equation allows us to eliminate P completely from the
equations by writing the adiabatic equation as
Pzt + PowN 2 =0 (17.7)
We will consider only those situations in which the bottom of the fluid is flat at z = -D
and the top consists of a free surface. When the bottom is flat, we can separate the variables
in the problem into a function of z and a function of horizontal and time variables. Fol-
lowing the treatment in Moore and Philander (1977) and Pedlosky (1987), we write,
(vUJ = [U(r/J,{},t)]F(Z)
V(r/J,{},t)
(17.8a)
w= W(c{>, 8, t)G(z) (17.8b)
-.L =g1J(r/J,{},t )F(z) (17.8c)

Po
186 Lecture 17 . Laplace Tidal Equations and the Vertical Structure Equation
Note that U should not be confused with our earlier use of the same symbol for the
horizontal velocity scale. It is now a variable and a function of horizontal coordinates
and time. The functions F and G are functions only of z, and they must be determined
by the physics. Our first goal is to derive the governing structure equations for those
functions. Finally, the variable 1] is a function that represents the horizontal structure
of the pressure field. At this stage, it has nothing at all to do with the actual elevation
of the fluid's free surface.
If we insert these forms in the equations of motion, we find first from the horizon-
tal momentum equations, since F(z) is a common factor,
gTJ¢
Ut - fV =- Rcose (17.9a)
Vt + fU =- gTJe (17.9b)
R
Aside from the use of spherical coordinates to represent the horizontal pressure
gradient, these are exactly the linearized momentum equations for a single layer of
fluid whose velocity components are U and V and whose free surface elevation is 1].
The same process for the continuity equation is not so simple: w depends on the
function G while u and v are proportional to F. This leads to
U¢ (vcose)e Gz
--+ +W-=O (17.10)
Rcose Rcose F
All terms except the ratio of Gz / F are independent of z, while each term in this ratio
is a function only of z. The only way this can be consistent (this is familiar from the
standard separation of variables) is if the ratio is a constant. We define the constant as
G 1
~= constant =- (17.11)
F h
The constant h is called the equivalent depth (we shall see why shortly) but at this
stage of our analysis, it is only a separation constant. This allows us to write the conti-
nuity equation in the suggestive form:
~+ (vcose)e + w =0 (17.12)
Rcose Rcose h
Inserting the forms into the adiabatic equation in its form involving the vertical
pressure gradient Eq. 17.7 yields
G N2
TJt+W--=O
Fz g
(17.13)
G N2
~TJt+W--=O
Gzz gh
Lecture 17 . laplace Tidal Equations and the Vertical Structure Equation 187
and it is clear, as in our discussion of the continuity equation, that the coefficient of
the term W in the last equation must be a constant. We choose that constant to be -1.
There is no loss of generality in doing this. Choosing any other constant would only
alter the definition of h. The properly skeptical student should try it.
With this choice, the adiabatic equation becomes
7lt =W (17.14)
which is not a boundary condition but is rather the separated form of the adiabatic
equation, although the form is delightfully suggestive of the boundary condition for a
single homogeneous layer. With the above choices for the separation constant, the
function G now satisfies (and it is here that it would be clear that any other choice than -1
would only alter the definition of h)
N2
Gzz+-G=O (17.15)
gh
This is a homogeneous differential equation with, generally, non constant coeffi-

cients, since N is a function of z and with a free parameter h. The problem is not com-
plete until the boundary conditions are established.
In order to have w vanish on z = -D, we must take
G(z)=O, z=-D (17.16)
At the free surface, the conditions are that the free surface displacement, which here
we will call zT> satisfies
w =WG(ZT) aZT (17.17)

at
while the total pressure is atmospheric pressure, which we will take to be a constant
(zero), thus
Ptotal = PO(ZT)+ g'lF(ZT)

(17.18)
dpO
'" PO(O)+-ZT + ... + g'lF(O)
dz
keeping only linear terms.

A time derivative of the last equation combined with the kinematic condition then
yields
gWG(O) =g'ltF(O) (17.19)
but from the continuity equation, this implies that Z = 0:

G(O)=F(O)=hGz(O)=>Gz-Glh=O, z=o (17.20)
which is the final condition for G. We then have an eigenvalue problem for the func-
tion G(z), whose eigenvalues are h. Note that the problem could also be written in terms
of F(z).
Using the above relations between F and G, we obtain as an equally valid alterna-
tive problem:
(~) +_1 F=O

2
N z gh
(17.21a)
Fz =0, z=-D (17.21b)
N2
Fz+-F=O, z=O (17.21c)
g
The advantage of the second formulation is that the eigenvalue h is not in the bound-
ary condition.
The equations for either G or F can be solved numerically, and the eigenvalue is found
along with the structure of the solution in z. Insight into the nature of the problem can
be gained by examining the case for the constant N.
In that case, the solution for G(z), which satisfies the boundary condition at
z= -D,is
2 N2
G=Asinm(z+D), m =-- (17.22)
gh
m is the vertical wave number of the solution (it will be quantized since the region is
finite), and
Az_-21t
- (17.23)
m
will be the vertical scale of the motion in the mode that has that value of m as the ver-
tical wave number. Note that the vertical scale of the motion is not h. Indeed, if we
define the vertical scale height for the density
Po (17.24)
hp =- dPo/dz
as the scale over which the density changes by its own magnitude (this is much greater
than the depth of the ocean for realistic oceanic density gradients), the vertical scale
of the motion is given by
Az =21t~hhp (17.25)
so that the vertical scale of the motion is essentially the geometric mean of the equiva-
lent depth and the density scale height.
The eigenvalue relation for h is obtained from the final boundary condition at z =0
and yields
1 .
mcosm D --smmD=O (17.26a)
h
N2
tan(mD}=mh=- (17.26b)
gm
or
2
tan(mD}=(N D]_I_ (17.27)
g mD
We note as we have before that
N 2 Dlg= _Ddp
__ D ~p
0 = _ " , _ 0 «1 (17.28)
Po dz hp Po
Thus the roots of the dispersion relation split into two classes. The first class has roots
for which mD is D{l}. In that case, the right-hand side of the dispersion relation is es-
sentially zero, and the solutions correspond to the zeros of the tangent function, or
mD= jrr, j=I,2,3, ... (17.29)
There are an infinite number of such roots corresponding to
m= jrr (17.30)
D
and since m2 = N2 1gh, the associated equivalent depth for mode j is
N 2 D2
ghj=~ {l7.31}
J rr
Note that for this mode, the horizontal equations will contain an equivalent long
wave gravity wave speed:
c.] =Vrgi;
o"j
= ~D rgD
Jrr « ,,}Sv {l7.32}
These equivalent speeds are the long wave speeds for internal gravity modes of ver-
tical mode number j and are much slower than the homogeneous phase speed for long
waves -Jgi5.
The modal structures for each j are simply
Gj =sin(jrczl D}, j=I,2,3 ... (17.33a)
Fj = cos(jrrz I D}, j = 1,2,3 (17.33b)

and although these form a complete set for the representation of w (the sine series is
complete), it is clear that the set of functions Fj , while containing an infinite number
of functions, which are all orthogonal, is not complete, since the cosine series lacks
the constant term. In other words, we have not found the barotropic mode that con-
tains zero vertical velocity.
We must reexamine the dispersion relation. We previously assumed that mD was
0(1). That may not always be the case. Indeed, as mD~ 0, the dispersion relation
becomes
tanmD"'mD= N 2 D
gmD
(17.34)
=>m 2D2 _ N 2D
g
but by definition,
N2
m2 (17.35)
gh
or for this mode subscripted zero,
ho=D (17.36)
or
(N2
mo=Vgv
and so
moD= ~N2D2
--«1 ( 17.37)
gD
so that the function
Fo =cosmo(z+D) (17.38)
hardly varies at all in z, i.e., the function is very nearly z-independent. This is the
barotropic mode.
To sum up, for linear, inviscid motion of a stratified fluid on the sphere, when the
fluid has a flat bottom, we can separate the motion into an infinite number of vertical
modes. Each mode satisfies a set of equations for its horizontal structure, which is
identical to that of a homogeneous layer of fluid possessing a long gravity wave speed
Cj= (gh/ 12 • That is, it behaves as a homogeneous layer with the equivalent depth hj'
which itself is one of the eigenvalues of the vertical structure equation. This is the only
way stratification enters the problem, i.e., by determining the equivalent depths and
phase speeds and by determining the vertical structure of the modes. Note that for
each mode in z, the vertical structure is maintained as that mode propagates, reflects
or dissipates as time progresses. Thus, all our previous work on the dynamics of
Poincare, Kelvin and Rossby waves for a homogeneous layer can be carried over, mode
by mode, to a stratified layer as long as the motion is hydrostatic and the bottom is flat.
If the bottom is not flat, it is not possible to separate the motion into individual modes.
The topography will mix the modes together, and the modal description is no longer
useful. The equations for the horizontal structure are called Laplace's tidal equa-
tions, because in their original application to a homogeneous fluid, they are with ap-
propriate added forcing terms the equations for the ocean tides. To recapitulate those
equations we have
g7J¢
Ut - jV =- Rcose (17.39a)
Yr + fU = - g7Je (l7.39b)
R
~+ (vcose)e + W =0 and (17.39c)

Rcose Rcose h
7Jt =W (l7.39d)
Note that if the last two equations are combined,
U¢ (vcose)e]
7Jt+ h[- - + =0 (17.40)
Rcose Rcose
and the correspondence to the dynamics of a shallow layer of fluid of depth h is

complete.
Lecture 18
Equatorial Beta-Plane and Equatorial Waves
The equator is a special region dynamically, most obviously because there the vertical
component of the Earth's rotation vanishes. It turns out to be, in consequence, a re-
gion in which certain linear waves have unusually strong signals and are involved in
some important atmospheric and oceanic phenomena such as the Quasi-Biennial
Oscillation in the atmosphere and the El Nino (ENSO) phenomenon in the ocean (and
atmosphere). Good, useful references that describe in detail those phenomena are
Andrews et al. (1987) for the former and Philander (1990) for the latter.
To see intuitively why the equator might be such as special zone, consider heuristi-
cally a Poincare wave packet near the equator with frequency
OJ= (r2 +c~(k2 +[2)) 112 (18.1)
and we note that near the equator where! vanishes, the y-dependence of the Corio lis
parameter cannot be neglected. The dispersion relation is of the class of relations
discussed in our first lecture where the relation between frequency and wave num-
ber also explicitly includes a dependence on a spatial variable, in this case latitude or
locally, y, i.e.,
OJ= Q(K,f(y)) (18.2)
As we noted in the first lecture, the wave vector for a slowly varying packet satisfies
dK =-\1Q =-/3 L; (18.3)

dt OJ
y
where is a unit vector in the meridional direction. Remember that the gradient on
the right-hand side of the equation for the rate of change of the wave vector is the
gradient with respect to the explicit dependence of the dispersion relation on spatial
variables, in this case only y.
That means that as the packet propagates, the frequency and x-wave number, k, will
be constant in the packet; only [will change. The dispersion relation for Poincare waves
implies that as the packet moves to higher latitudes where f is larger, the y-wave num-
ber must decrease to keep the frequency constant. Finally at the latitude Yc such that
!(Yc)= [OJ2 -c~e r 2 (18.4)

194 Lecture 18 . Equatorial Beta-Plane and Equatorial Waves
the y wave number vanishes. Beyond that point, I becomes imaginary, leading to spa-
tial decay with y for y > Ye. This produces a trapping zone around the equator in which
the wave energy, which would normally disperse in two dimensions, is trapped within
a wave guide as a consequence of the increase with latitude of the square of the Co-
rio lis parameter, an effect that is clearly symmetric about the equator. Similarly, for
Rossby waves where
k
=- fJ ---=-----:--.,,-----c- (18.5)
e +12 + f2 /c~
(f)
the same trapping effect must occur. Note that for both Poincare and Rossby waves, the
meridional component of the group velocity vanishes when the Y wave number van-
ishes so that the wave energy will not cross the critical latitude and will be reflected
back into the equatorial wave guide. Also note that as the Coriolis parameter van-
ishes, the minimum frequency of Poincare waves approaches the maximum frequency
of Rossby waves, and so the two wave types cannot be expected to be as well-separated
in the frequency domain as they are in mid-latitudes.
Thus, overall, the equatorial band will act as a wave guide for both gravity and
Rossby waves. We expect the wave modes to be trapped meridionally and the prop-
agation to be basically along the equator. This means the waves will generally not
disperse their energy over more than the zonal direction, and consequently the
amplitude of the waves and their influence can be anticipated to be more impor-
tant for equatorial dynamics in general than in mid-latitudes. It remains for us to
move beyond this heuristic discussion to find the nature of the waves in the equa-
torial zone.
The Equatorial Beta-Plane
We will assume that the wave motions have a large enough horizontal scale to ensure
that the wave motion is hydrostatic. We will also only consider cases in which the ocean
bottom is considered flat, and in fact, we will not consider any interaction with the
bottom. In that case, as we saw in the last lecture, we can resolve the wave motion on a
set of vertical normal modes, each mode yielding an equivalent depth hn and a corre-
sponding long wave speed cn' both of which come from the eigenvalue problem de-
scribed in the previous lecture.
For the linear inviscid problem, the equations of motion are
oUn _ jV, =__g_ o 17n (18.6a)

ot n RcosB o¢
oVn _ fU =_1... o17n (18.6b)

ot n R oB
o17n +h [_1_0U n +_l_~(V COSB)] =0 (18.6c)

ot n RcosB o¢ RcosB oB n
If the motion is limited to a narrow region,L, around the equator such that L / R « 1,
we can expand the trigonometric functions in the above equations, i.e.,
Lecture 18 . Equatorial Beta-Plane and Equatorial Waves 195
f = f(O)+ fly+ ... = fly+O(LI R)3 (18.7a)

'-v--'
=0
cosO =1 +O(LI R)2 (18.7b)
This allows us to define the local Cartesian coordinate system:
x = RcosO[¢-¢o]= R(¢-¢o)+O(LI R)2 (18.8a)
y=RO+O(LlR)2 (18.8b)
In these terms, the equations become the simpler set, valid on the equatorial beta-plane:
Unt -fJyVn =-glJnx (18.9a)
Vnt + fJyU n =-glJn y (18.9b)
lJnt + hn[Un x + Vn y 1=0 (18.9c)
Of course, we must check after the fact that our solution does satisfy the condition of
being localized in the vicinity of the equator. It is also easy to add forcing terms to each of
the momentum equations to represent the action of a wind stress, and the exercise is left
to the student to trace the development of the equations with such forcing terms present.
Now the heuristic discussion at the start of the lecture leads us to anticipate that the
wave modes will be contained in a wave guide, a sort of naturally produced equatorial
channel. In that case, we might anticipate that the modes will be analogous to the modes
we found in the channel problem for mid-latitudes. In that case, we found Poincare, Kelvin
and Rossby modes. It was a straightforward business in the mid-latitude case to write the
problem in terms of the free surface height. With the strong variation of fin the equatorial
case, it turns out be far simpler to pose the problem in terms of the meridional velocity
(we noted in the mid-latitude case that the eigenstructure for the meridional velocity was
far simpler than for either the free surface perturbation or the zonal velocity). How-
ever, based on our experience in the mid-latitude channel, we also might anticipate that
we should be alert to a wave mode for which the meridional velocity is identically zero,
for that is one of the chief characteristics of the Kelvin wave. Hence, before we formu-
late the wave problem in terms of the meridional velocity, we should check to see whether
a mode exists in which Vn is identically zero. If that were so, we would have
Ut =-glJx (18. lOa)
flyU=-glJ y (18.l0b)
lJt +hUx =0 (18.lOc)
We have suppressed the explicit subscript notation, and the student is expected to
realize that the following development is pertinent to each mode n, each with its own
equivalent depth h.
Eliminating the free surface elevation between the first and third equations yields
Utt - ghU xx =0 (18.11)
which is again the classical one-dimensional wave equation whose general solution is
U± =U±(x±ct,y) (18.12)
One solution, U+ = U+(x + ct,y), propagates to the west with no change of shape,
while the other solution, U- = U-(x - ct,y), propagates eastward with no change of
shape. These are reminiscent of the Kelvin waves in a channel. In that case, the right-
moving wave "leaned" against the lower wall, and the left-moving wave "leaned" against
the upper wall. In the present case, there are no walls, only the equator itself, and we
have to check whether either of these solutions has a y-structure consistent with equa-
torial trapping of the disturbance. The y-derivative of the x-momentum equation and
the x-derivative of y-momentum equation to eliminate the free surface term yields
Uyt=/JyU x (18.13)
Now for each possible solution we have
r+
U =±cU r+ (18.14)
which when inserted in Uyt = [3yUx yields
aa'Y (U~)+ /Jyc (U~ )= 0 and
=> U~ = e±(f3y2 I c) p± (x ± ct) (18.15)
where P is an arbitrary function.

Thus, if the region is unbounded in y only, the (-) solution is acceptable, since the
(+) solution diverges at IYI = 00. There is only an eastward moving Kelvin wave mode:
U . =e-(fJi/c)p(x-ct) (18.16)
KelvIn
with a corresponding free surface height,

_ c _(f3y2 I c)
17Kelvin --e P(X-ct) (18.17)
g
The mode moves eastward with the long wave speed. For each vertical mode n, that
wave speed is c = cn = ~. The decay scale is
Leq =(~ f2
which is the equatorial deformation radius.
Table 18.1.
Quantities for the first five
baroclinic modes (Moore and
Philander 1977)
5 2 44 138
To get a feeling for the quantities involved, Table lS.l (from the article by Moore
and Philander 1977) gives the pertinent quantities for the first five vertical baroclinic
modes. The quantities refer to the Equatorial Atlantic but are typical.
The appropriate time scale is determined by the relation
= cn ' => T = ~ fJcn

1
Leq
T
Note that the equatorial deformation radius depends on the square root of the long
wave speed, not on the speed itself as in mid-latitudes. In mid-latitudes, the deforma-
tion radius is
Cn
LD=f (1S.1S)
but at the equator f = f3y. If we set y to be of the order of the equatorial scale, this yields
Cn
Leq = fJL (1S.19)
eq
whose solution yields our previous definition. Note that the time scale above satisfies
T = fJLeq (1S.20)
as in Rossby waves.
For the barotropic mode, the equivalent depth h is of the order of the fluid depth, D.
This yields
Co = fii5 = 0(200) m S-1 (lS.21a)
Leq =3000km (1S.21b)
so that the "trapping" scale is of the order of the planetary scale. In that case, the equa-
torial wave guide has little sense, since it is global and the barotropic mode must be
considered separately. Fortunately, most of the equatorial response that seems to be
relevant is in the baroclinic modes, and indeed the equatorial Kelvin wave has been
clearly identified in the equatorial regions (Eriksen et al. 19S3).
Note that the y-structure of the Kelvin mode is a Gaussian and that U is in geo-
strophic balance with the pressure field.
We now return to the task of deriving a wave equation for the rest of the wave modes.
Taking a time derivative of the x-momentum equation and using the mass conserva-
tion equation yields
Utt -fJyVt =-g17xt =gh(Uxx + Vxy )
=> [U tt - ghUxx = fJyVt + ghVyJ (18.22)
while operating in the same way on the y-momentum equation yields
Vtt + fJyU t =-g17yt =gh(Uxy +Vyy)
=> [Vtt -ghVyy =-fJyUt + ghUxy ] (18.23)
We now operate on the above equation for V (Eq. 18.23) with the operator
( a2 a2 )
l at 2 -c
2
ax 2
and use the equation for U (Eq. 18.22) to obtain
l(~_C2~)(
at 2
~-C2~)V=_fJy( ~-C2~)U
ax 2 lat 2 ay2 lat 2 ax 2 (18.24)
a (a a
+ c axay lafT- C (Jx2
2 2 2
2 2 )
U
or
( a2 a2 )( a2 a2 ) a [ (JV 2 (J2V ]
lafT- C
2
ax 2 lat 2 -C
2
al V=-fJyai fJyat+ C axay
(18.25)
2 a2 [fJy (JV 2 (J2V]
+C axay at+ C (Jxay
Carrying out the algebra implied by the above products yields the final equation
for V:
~[~{V2V -~Vtt _ jJ2( V} + fJVx to

at at c 2 J C
(18.26)
Note that the final equation is a local conservation statement for the quantity in
the square bracket. The student at this point is allowed to guess what that quantity
really is and to verify the presumption. Also note the similarity of the equation to the
wave equation at mid-latitude where the factor f32y2 is replaced by f. Also be sure to
recall that this equation holds for each vertical mode, i.e., for each n with a correspond-
ing vertical structure function and a corresponding equivalent depth.
Let's try to find plane wave solutions in x and t of the form
v = Aei(kx-M)IfF(Y) (18.27)
where 1jf satisfies the ordinary differential equation:
d 2 1fF ~2 _f32_Y__
2 k] 1fF=0
k2-f3- (18.28)
dl + [ c 2 c2 aJ
Note that beyond a certain critical value of y, the form of the equation implies eva-
nescent (or exponentially growing) behavior. Equatorward of that latitude, the func-
tion 1jfwill be oscillatory in y. We can put the equation in standard form by introduc-
ing a meridional coordinate scaled on the equatorial deformation radius:
( JII2
Y=l; ~ (18.29)
In terms of which,
d2 1fF +[aJ2 _k2~_ ck _~2]1fF=0 (18.30)

d~2 f3c f3 aJ
This, interestingly enough, is exactly the Schroedinger equation for the quan-
tum mechanical oscillator, and the solutions have been extensively studied. In the
mathematics of special functions, this is the Hermite equation. It is well-known
(see for example Schiff 1955) that the only solution that is bounded at infinity is of
the form
e-qZ 12
1fF=1fF'(~) H.(~) (18.31)
} ~(2j j!1t1l2) }
This, aside from the complicated constant in the denominator to make the func-
tions orthonormal, is a Gaussian in the meridional coordinate multiplied by one of
an infinite set of polynomials H/';) called the Hermite polynomials. The orthogonal-
ity condition is
~
flfFjlfFid~ =oij (18.32)
The Hermite polynomials are generated according to the rule:
H/~) =(-oj e~2 /2 d J e -qZ 12 (18.33)

d~J
The first few polynomials are
Ho= 1
Hl=2~
H2 = 4~2_ 2 (18.34)
H3 = 8~3-12~
H4 = 16~4 - 48~2 + 12
Note that the solutions divide into odd and even functions because of the symmetry
in y of the governing equation for 1fI. Each of these functions, finite at infinity, satisfy
d 2 V/ + [(2j + 1)_q-2 ]V/=O (18.35)

dq-2
The first four of the eigenfunctions are shown In Fig. 18.1.
'Po 'PI
1.0 • 71< 1.5 ., - - . . . . , - - - - - , - - - , - - - - - ,
1.0
0.8
0.5
0.6
0.0
0.4
- 0.5
0.2
- 1.0
0.0-4 - 1.5 -----'-------'-----'---~

o o
'L
-2 2 4 -4 -2 2 4
'P2 'P3
3 ir---,---~---r_--~ 6 .r ---,---~--_r---_,
4
2
o
o
-2
-1
-4
-2 I "I _6 L'----L----'-----'---~
-4 -2 o 2 4 -4 -2 o 2 4
Fig. 18.1. The first four eigenfunctions for the equatorial wave amplitude with latitude
For our solution to correspond to one of these eigenfunctions and hence satisfy a
condition of finiteness at infinity, we must have
oi k 2c k c . .
-----=2)+1, )=1,2,3, ... or ( 18.36)
f3c fJ m
m2 = c2(k 2+(2j + 1)1 L~q)+ f3k c2 (18.37)

m
This should be compared to the dispersion relation at mid-latitude for Poincare
waves for a value of f~ 0, while the last term on the right-hand side is similar to the
effect produced by the beta term at mid-latitudes.
This is a cubic for frequency in terms of x-wave number, and it is easier to solve the
quadratic for k in terms of m. This yields the dispersion relation in the form
]112
k=-L±~ [ (fJ _ 2m)
2
_ 8j fJ (18.38)
2m 2 m c c
Before discussing the full form of this relation, it is useful to discuss limiting cases.
If m = O(kc), the last term in Eq. 18.37 would then be of O({3c), which compared to the
first term is
O(~J
Thus, if the gravity wave speed is much greater than the Rossby wave speed, c > {31 k2,
then the last term can be neglected and we obtain the approximate dispersion rela-
tion for the Poincare waves:
m=±c(k2 +(2j+1)j'2 (18.39)
On the other hand, if m is small, we would obtain a balance between the last term
in Eq. 18.37 or the approximate equation for the equatorial Rossby mode, i.e.,
m= fJ (18.40)
e +(2j+1)1 L~q
It is also easy to solve Eq. 18.38 when j = O. In that case,
k=_L±~(fJ _ 2m) (18.41)

2m 2 m c
The two roots are then
k=- m (18.42a)
c
k=_fJ+ m (18.42b)
m c
The first root yields a wave moving with the gravity wave speed to the west, and
this yields as we have seen before an unbounded solution for the zonal velocity in y
and must be rejected. The other root does yield a bounded solution. At low frequen-
cies it looks like a Rossby wave; that is
OJ",,_fJ
k
while for large frequencies, it looks like a pure gravity wave, OJ "" kc.
Thus, there are two classes of solutions for each wave number. There is a set of higher
frequency modes similar to the mid-latitude Poincare modes and a set oflow frequency
modes corresponding to the Rossby modes. In addition, there is the Kelvin mode that
exists only in its eastward traveling form. A single wave, which is discussed above, is
often called the mixed Rossby-gravity wave or the Yanai wave, which straddles the
two wave types. Each mode corresponding to a different j index goes along with the
eigenfunction, lfIj(y / Leq) for its V-field except the Kelvin mode that has only a zonal
velocity whose shape is given by lJIo' the Gaussian.
The full dispersion relation is shown in Fig. 18.2. It is standard practice for the equa-
torial problem to consider only positive frequencies and to let the x-wave number run
over positive and negative values.
4.0,.~__~~~--------r--------.---------r--------.--------'
3.5
3.0
2.5
2.0
1.5
1.0
0.5
wave
0.0'4 3 2j 3 ~ o,/
-1 2
kLequatori.,
Fig. 18.2. The dispersion diagram for equatorial waves. The line bending upwards connects the ex-
trema in the curve of frequency vs. wave number
In Fig. 18.2, the wave number is scaled with the equatorial deformation radius and
the frequency is scaled with the characteristic time scale so that the frequency is given
in units {fJC. Note that both the Yanai wave and the Kelvin wave have only positive
group velocity in the x-direction. Indeed from Eq. 18.37 it is easy to show that
am[2m +~]=c[2k +~] (18.43)

ak fJc m2 fJ m
or
C 2km +1]
[p
am (18.44)
ak
[2;2 +~]
The group velocity vanishes on the line
m=_L (18.45)
2k
This line in the (m,k)-plane separates westward from eastward group speeds and
also marks the locus of the extrema in the frequency wave number plane.
If we insert the above condition in the dispersion relation, we obtain the value of
the minimum Poincare frequency for each j:
112
mmin =( ~) [u + 1)1/2 + j1l2]' Poincare (18.46a)
and
112
mmax =( ~) [(j + 1)112 - /12 ], Rossby (18.46b)
For each j, these give the points of reversal of the sign of the group velocity. Note
that the minimum of the Poincare frequency is not at k = 0 but slightly displaced to
negative k as a consequence of the beta effect. Note that the difference between the
minimum Poincare frequency for j = 1 and the maximum Rossby wave frequency for
j = 1 satisfy
mmm
. p'
omcare r::: + 1
,,2
m =--r-=5.828 ... (18.47)
max Rossby " 2- 1
so that both groups of waves are in the same range of parameter space. Any attempt
to plot the Rossby waves and Poincare waves of the same diagram in mid-latitude would
be nearly impossible, since the frequencies are so disparate in size (this is, after all,
the basis of quasi-geostrophy in mid-latitudes).
For each j, there is a corresponding eigenfunction for the V-field from which the
zonal velocity field can be calculated, i.e.,
Vj = All'} (;)ei(kx-OJ/) (l8.48a)
j
U =A ji(f3c)l/Z[(ljl/Zlf/j-l(;) +(j+ljl/ZIf/}+l(;)] (l8.48b)
2 [m+kc] 2 [m-kc]
and this holds for the Poincare, Rossby and Yanai waves. It is left to the student to work
out the corresponding free surface elevations. The Kelvin wave of course has only a
zonal component, and its amplitude is proportional to lfIo( ~). The fact that each j mode
consists of two eigenfunctions, lfIfor j's one greater and one less than the j for V, ren-
ders the reflection problem rather complex. That, plus the physical fact that the Kelvin
mode only exists in its eastward form makes the reflection problem from the eastern
and western boundaries quite different, and the student is referred to the references
given above for a detailed description of that problem.
Finally, we recall that each of the solutions above represents the contribution of a
particular vertical mode with mode number n. Hence, each frequency and eigenfunction
really should carry two indices, one for its horizontal structure (j) and one for its verti-
cal structure (n).
Lecture 19
Stratified Quasi-Geostrophic Motion

and Instability Waves
We return from our brief visit to the equator and investigate the low frequency mo-
tions in mid-latitudes that occur in a stratified fluid. The motion we consider will be in
near {quasi-)geostrophic balance, but we will develop the equations in an informal,
heuristic way, leaning heavily on the formal analysis of Lecture 15. We will also employ
the beta-plane approximation so that we are assuming that two parameters, £ = U/ foL,
b = {3L / fa, are both small. That being the case, the lowest order balances in the horizon-
tal momentum equation imply that
U=- Py (19.1a)
Pofo =-Ij/y
v= Px (19.1b)
Pofo =lj/x
Note the beta-plane use of the constant reference value of the Coriolis parameter.
As a consequence of that balance, it follows that at O( 1), the horizontal velocity is non-
divergent, so that for an incompressible fluid,
aw =O(£,b) (19.2)
az
If wvanishes at any z at the lower or upper boundary or approximately vanishes there,
it follows that w itself is small. Indeed, w is smaller by a factor of £ or b compared to its
geometrical scaling UD / L. In that case, the vorticity equation that arises at order £ can
be written
a( a(
-+u-+v-+ fJv= fo-
a( aw (19.3)
at ax ay az
In the vorticity equation, the contribution to the advection of vorticity due to w is neg-
ligible, since w is of a higher order in Rossby number than u and v, but its influence is felt
by the stretching term on the right-hand side. Small as w is, the weak stretching is ampli-
fied by the large Coriolis parameter, the planetary vorticity, which is O{ c- 1) larger than the
relative vorticity, which makes up for the smallness of w in the vorticity budget. In the above,
(=V X -u y =v 21j/ (19.4)

206 Lecture 19 . Stratified Quasi-Geostrophic Motion and Instability Waves
At the same time, the motion, which is assumed as usual to be adiabatic, satisfies
dp +u dp +v dp +w dpo =0 (19.5)
dt dx dy dz
Again, the vertical velocity is negligible in providing a contribution to the advec-

tion of the perturbation density, but it does enter the advection term by its contribu-
tion to the advection of the large background density gradient, dpol dz (large, that is,
with respect to the density gradients associated with the motion; we shall still assume
that the background density varies slowly compared to the vertical scale of the motion).
Using the hydrostatic equation and the standard definition of the buoyancy fre-
quency allows us to write the density equation as
~(dP)+WN2pO =0 (19.6a)
dt dz
d d d d
-=-+u-+v- (19.6b)
dt dt dx dy
N2=_L dpo (19.6c)

Po dz
Eliminating w between the adiabatic equation and the vorticity equation and tak-
ing care to use the properties of the geostrophic velocity, we obtain as the governing
equation for the geostrophic stream function
d d
[ -+If/x--If/ - d][ V 2 If/+-
d (f02
-- dlf/)] +fJlf/x=O (19.7)
dt dy y dx dz N 2 dz
This is the quasi-geostrophic potential vorticity equation (qgpve). It is impor-

tant to note that although the motion is three-dimensional, i.e., a function of x, y and z,
the advective term in the equation only reflects the effects of horizontal advection due
to the rotation-induced smallness of w. A more systematic derivation is given in
Pedlosky (1987).
At the lower boundary, the kinematic condition is
1 d dp
w=uhbx +Vhby =- N 2po dt dz (19.8)
Using the geostrophic relations for the horizontal velocities and the relation
between p and If/, we obtain for the boundary condition at z =-D
d d N2 _
-~+ J(If/,lf/z)+-J(If/,hb)-O (19.9a)
dt dz fo
J(a,b)=axb y -aybx (19.9b)

Lecture 19 . Stratified Quasi-Geostrophic Motion and Instability Waves 207
Ignoring the deviation of the free surface with respect to that of the internal iso-
pycnals (the same discussion as for internal waves), the boundary condition at z = 0 is
just w = 0, which in terms of the geostrophic stream function is
ddlj/
ataz+ !(Ij/,lj/z) =0, z=O (19.20)
Let's examine that approximation a bit more carefully. From the adiabatic equation,
the characteristic size of w generated within the fluid is of the order
Wint = o( UfUL )
LN 2 D
(19.20)
where we have used the geostrophic scaling for p and the scaling UI L for the advective
time derivative. On the other hand, the vertical velocity at the free upper surface will
be of the order
wz=O=dll/dt =O(
l ~J=o(UfULJ
(19.21)
LPog l LPog
The ratio is of the order
wz=o
--=--«1
N 2D (19.22)
Wint g
and so to the lowest order, W is zero at the free surface, which is the condition used above.
Let's look first for baroclinic Rossby waves. Let the bottom be flat and assume the motion
is small amplitude so that we can linearize the dynamics. The problem then becomes
~[V21j/+~(
dt
f~ dlj/Jl+ p(}1j/
(}z N dz dx
2
=0 (19.23a)
d21j/ =0 (19.23b)
dzdt
z=O,-D (19.23c)
We can find plane wave solutions in the form

Ij/ = Aei(kx+ly-OJt)<p(z) (19.24)
where <p satisfies the ordinary differential equation
~(f~ d<PJ+{A_
dz N dz (-m)
2
k2-l2}<P=0 (19.25a)
d<P =0 (19.25b)
dz
z=O,-D (19.25c)
This is a standard Sturm-Liouville eigenvalue problem, and indeed it is the

same problem we discussed for the vertical structure equation for Laplace's tidal equa-
tions. For the simple case where N is a constant, the solutions can be found imme-
diatelyas
cP(z) =cos(nrrz I D), n = 0,1,2, ... (19.26)
(note that n = 0 is a nontrivial case) from which it follows that for each n,
OJn = fJk LD=ND (19.27)

k2 +12 + n 2rr2 ' fo
L2
D
The n = 0 mode is the barotropic mode. The horizontal motion is independent of z.

The higher modes, each with a lower frequency for the same horizontal wave number,
have zero vertically averaged horizontal velocity. For each n, the dispersion relation is
exactly what we found for a homogeneous layer of fluid, except that now the term in-
volving the deformation radius is the baroclinic deformation radius, and the barotropic
mode is approximated by the limit where the deformation radius of the free surface is
considered infinitely large compared to the LD given above. This is the same approxi-
mation that allowed us to ignore wat z = o. Again, mode by vertical mode, we can apply
all the results of our investigations of the Rossby wave in a homogeneous layer to each
vertical baroclinic mode.
Topographic Waves in a Stratified Fluid
Consider now the case where N is again constant, but a bottom slope exists and we
ignore the beta effect. This last condition implies that the horizontal length scale is
small enough that in the linear vorticity balance, mK2 » 13k, which we must check af-
ter the fact. In this case, the potential vorticity equation (linearized) is merely, for the
same periodic plane wave in x and y,
-I.~d2cP2_N 2 2
. 2 K cP] =0, K2=k2+12 ( 19.28)
dz JO
We assume for simplicity that the bottom is sloping uniformly in the y-direc-
tion and that the upper surface at z = 0 is very far away (we have to quantify this idea
shortly) so that the region can be considered infinite in z. Then the lower boundary
condition is
2
-imcPz +ik~hb cP=O, z=-D (19.29)
fo Y
The solution that decays away from the lower boundary and so remains finite with
distance from the lower boundary is
cP = Ae -KN(z+D)! fo (19.30)
(the student should now show that the condition that the upper surface appears to be
infinitely far away from the lower boundary is simply KL D » 1, i.e., that the wavelength
be short compared to the deformation radius).
Using this solution in the lower boundary condition yields the dispersion relation
khb N
O)=--_Y- (19.31)
K
This is actually a very remarkable result. It has some similarities to the dispersion rela-
tion for the Kelvin wave. Here we have a single boundary at z = - D and a wave with a single
direction of propagation. The frequency, as in the case of the Kelvin wave, is indepen-
dent of the rotation. On the other hand, again like the Kelvin wave, the trapping scale
depends on f, only now, the trapping increases as f decreases. This bottom trapped wave
has a vertical trapping scale 8 such that if A, is the wavelength,fol N8= 1. Another way
to look at the wave is to note that the bottom slope introduces a topographic beta effect:
fohb y
Ptopog =----r;-
in terms of which
kLD
0) =- Ptopog K
which has something of the character of a Rossby wave.
Waves in the Presence of a Mean Flow
Instead of linearizing about a state of rest, let's return to the full, quasi-geostrophic
potential vorticity equation
ata aya Yax-a][V If/+-aza(f02

[-+If/x--lf/ -- alf/Jl + Plf/x =0
N az
2
2
(19.32)
and imagine that the wave is embedded in a mean zonal flow. That is, we will write the
stream function as P(y, z), which represents a mean zonal flow that is an exact solu-
tion of the qgpve, and add to it a wave perturbation so that
If/= P(y, z) + cp(x,y, z, t) (19.33)
Note that in the basic wave-free state, the zonal flow and the accompanying density
anomaly are
U(y,z)=-If'.Y (19.34a)
p(y,z)=_foPo If'. (19.34b)

g Z
From these definitions or equivalently the thermal wind equation,
U =_-.L rip (19.34c)

Z foPo dy
d
[ -+U
dt
-d][ v 2 rp+-
dX
d (f02
-- drpJ] +
dZ N 2 dZ
drp- d- [ 'P +-
d (f02
- d- +fJrpx =0
dX dy YY dZ N 2 dZ
'PJ] (19.35a)
or
[~+U~][V2rp+~(f02
dt dX
drpJ]+ drp[fJ- U _~(f02 dUJ]=o
dZ N dZ dX 2 dZ N dZ YY 2
(19.35b)
This equation is the perturbation form of the qgpve. The presence of the mean flow
has produced two very important changes. First, the local time derivative has been
changed to a linearized form of the advective derivative in which the additional term
U~
dX
represents the advection by the mean flow. Equally important (if not more so) is the
fact that the planetary gradient of vorticity, f3, is now supplemented by the contribu-
tion of the mean flow to the potential vorticity gradient of the basic state. That is, the
meridional potential vorticity gradient is now
dq =fJ-U _~(fi U
dy YY dZ N2 Z
J (19.36)
This is analogous to the way, for a homogeneous fluid, the bottom topography
supplements the beta effect to provide an altered potential vorticity gradient in which
the wave propagates. However, as we shall see, the effect of the mean flow can do more
than simply alter the frequency.
The boundary condition at Z = -D in this linearized problem becomes
d d) rpZ +rpx~y +-rpxhb

(-+U- N 2
=0, z=-D or (19.37a)
dt dX fo Y
(~+u~)rpz
dt dX
+rpx[-Uz + N hb
fo
2
Y
]=0, z=-D (19.37b)
The last term in the square brackets can be written in a rather suggestive form:
N
-hb
fo
2~ -UzfolN
Y
2
] Nfo2[hb
=-
Y
-
g Py -]
fo (-gpo)
fo
(19.38)
N [ hb y + p~
=To
2
N [ hb y - dy
15 ] =To dZJ p
2
1
so that the boundary condition contains terms involving the difference between the
slope of the boundary and the slope of the basic state's isopycnal surfaces as they in-
tersect the boundary. At the upper boundary, which we assume is flat,
(~+U~)cpz
at ax +CJd-uz]=o, z=O (19.39)
Boundary Waves in a Stratified Fluid
Consider the situation in which for simplicity we ignore beta completely and suppose
that the bottom boundary is flat, i.e., hb = O. Instead we will consider the case where
the mean flow in the x-direction is sheared in the vertical so that
U =Uo + Uz{z+ D} (19.40)
where Uz is a constant, i.e., a flow with constant vertical shear. This is supported by a
horizontal density gradient and hence a sloping density surface in the y-z-plane. In-
stead of the bottom sloping and the basic state density surfaces being flat, as in the
case of the bottom trapped topographic wave we studied earlier, we now have the bot-
tom flat and the density surfaces sloping. From the form of the boundary condition,
however, these might have some equivalence. Let's see. For the case where N is con-
stant and where the lower and upper boundaries are well-separated (in the sense of
the topographic boundary wave discussed above), the qgpve is again, for QJ,
d 2QJ N2 ]
(-iOJ+Uik) [ - 2 - - 2 K2(/J =0, K 2 =e+1 2 (19.41)
dz fo
leading again to the interior solution:
II' = Aei(kx+ly-aJt)e -KN(z+D)! fa (19.42)
The boundary condition at z = -D now yields the relation
-(Uoik-iOJ)KN / fo -ikUz =0 or (19.43a)
OJ=U k+~fo U (19.43b)

o K N z
OJ +U
c=-=Uo+~ (19.43c)
k KN
The propagation consists of two parts. The first is a simple advection by Uo' which
is the basic state velocity at z = -D. The more interesting contribution is from the ver-
tical shear, or equivalently, the slope of the isopycnals relative to the lower surface.
Indeed, the result for the frequency can be written as
k(az/ay)p N
OJ-Uok= K (19.44)
which should be compared to the relation for the bottom trapped topographic wave.
In this simple case, this shows the equivalence between the sloping isopycnals and the
sloping surface (the change in sign is prefigured by the differing signs in the bound-
ary term)
N
fo
2
Y
dZJ
[hb - dy
p
1 (19.45)
Now let's instead consider a wave localized near the upper boundary. The potential
vorticity equation is the same, but the solution decaying away from the boundary is
rp= Aei(kx+ly-ox)ezKN / fa (19.46)
The boundary condition on z = 0 now yields
[(U o +UzD~k-im]KN -Uzik =0 (19.47a)

fo
kUz fo
:::>m= (Uo+U D) k- - - (19.47b)
z K N
Comparing this result to the case where the wave is trapped near the lower bound-
ary, we see two differences. First, the advective velocity is different because the shear
makes the advecting velocity larger at z = 0 (assuming positive shear). Second, the
intrinsic frequency, i.e., the frequency seen by an observer moving with the local ba-
sic flow, had the opposite sign compared to the former case. It is the slope of the
isopycnals relative to the boundary, and this has changed from the previous situation.
It is interesting to ask whether there is any wave number for which the two frequen-
cies or the two phase speeds of these apparently independent waves could be equal. If
that were the case, it could be possible for the two waves to effectively interact. Equat-
ing the two phase speeds in the two cases leads to
(Uo+UzD)- Uz fo =U + U z fo
K N 0 K N
(19.48)
:::>K=2 fo =2..
ND LD
When the wave number is twice the inverse of the deformation radius, both bound-
ary waves, one moving towards positive x relative to the local flow at its boundary and
one moving towards negative x relative to the local flow at its boundary, are moving at
the same speed relative to a fixed frame. In that case, we might wonder whether a par-
ticular mode can be produced from the interaction of these two waves. Note that when
KL D = 0(1), the assumption that the two boundaries are well-separated fails, and we
have to consider the solution from first principles. A surprise results when we do.
Baroclinic Instability and the Eady Model
We return to our stratified layer with the shear flow previously described. The layer
has an overall thickness D, and N is constant as well as the shear Uz • Again, we ignore
the beta effect. The algebra is a bit more standard if we move the position of the ori-
gin in Z to the lower boundary so that 0 ::; Z ::; D. The basic flow is thus
U=Uo +Uzz (19.49)
The boundary conditions are
(-im+ik(U o+Uz z)}Pz -~ ikl/J=O (19.50a)
z=O,D (19.50b)
while the equation for l/J is

2
d-2l/J !!......K
2
2 l/J=0 (19.51 )
dz 2 fo
The general solution for l/J is
l/J = Acosh(flz) + Bsinh(flz) (19.52a)
NK
(19.52b)
11'= fo
Applying the boundary conditions at Z = 0 and Z = D yields two equations for A

andB:
Z =0, -cf.1B- jiA =0 and (19.53a)
Z = D, (UzD- c) [uAsinhCuD)+ ,llBcosh(,uD)]

(19.53b)
-Uz [Acosh(,llD) + B sinh(,uD)]= 0
where
_ m
c =~-Uo (19.53c)
k
The equations above are two homogeneous, linear, algebraic equations for the
constants A and B. The condition for nontrivial solutions is that the determinant of
the coefficients vanishes. This yields a quadratic equation for c:
c2-UzDc +U/(~ coth(,llD)- ;2 J=o (19.54)

This yields two solutions:

1/2
C = UzD ± UzD [ 1-4 coth(uD) +4_1_ ] (19.55a)
2 2 (fJD) (JiD)2
or
1/2
_ UzD +
-C- 4 _ JiDcoth(JiD) + 1
Uz [(fJD)2 ]
(19.55b)
2 - Ji
The useful identity
1
cothx =-{tanh(xI2)+coth(xl 2)}
2
finally allows us to write the equation for the phase speed as

1/2
OJ UzD Uz JiD JiD JiD fJD
c =- =Uo + - ± -[ ( - - coth(-»)(-- tanh(-») ] (19.56)
k 2 Ji 2 2 2 2
7,.--.-,-----~----._----._---,-----.-----.----_.----,_--__,
o I ..-
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0
p*D
Fig. 19.1. A graphical solution of Eq. 19.57. Each term is plotted versus its argument
Since x ~ tanh x for all x, the second factor in the square bracket is always positive.
The first factor will change sign where
f.1D 12 = coth( f.1: J (19.57)
The graphical construction of the intersection (Fig. 19.1) shows that the critical
value of liD '" 2.3994.
Note that this corresponds to a value of K = 2.39971 LD rather close to the heuristi-
cally motivated value from the previous discussion. For wave numbers less than this
critical K, the frequency will be complex.
When c is complex, i.e., when c = cr + ic~ the behavior in time consists of an oscillation
and an exponential growth for positive ci' i.e., the time factor becomes e- ikct = e-ikcrt e kCit
with a growth rate
kU
cq = _ z [(coth(f.1D 12)- f.1D1 2)(f.1D 12-tanh(j.iD12)]1/2 (19.58)
f.1
Figure 19.2 below shows the real part of the phase speed measured with respect to Uo
and scaled by Up. The dotted line shows the imaginary part of c also scaled with Up,
0.8 ,
. - - - - - - r - - - - , - - - - . , - - - - - - r - - - - - , - - ---r------.-------,
0.7
0.6
0.5 1 c,
0.4
0.1
0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0

/10---+
Fig. 19.2. The Eady dispersion relation K scaled on deformation radius. The thin curves show the
real and imaginary parts of the phase speed. The heavy line made of "o's" shows the growth rate
while the imaginary part of the frequency, the growth rate scaled by Uz or the shear, is shown
to be the line formed by the circles. The figure is drawn for the case where I =O.
Note that for each wave number for which a positive imaginary part of c exists, there
is another solution with the same real part of c but whose imaginary part is negative.
This follows from the fact that the equation of the perturbation field is real, so that if
cP is a solution with eigenvalue c, its complex conjugate cP* will be a solution with
eigenvalue c*. Since the dynamics is inviscid and thus reversible in time, the physics
must include the possibility for a cunningly chosen initial condition to return a dis-
turbance to zero amplitude (exponentially slowly).
The model described above was initially described by E. T. Eady (1949). This ex-
planation and the paper by Jule Charney (1947) were the first to correctly describe
the instability process now known as baroclinic instability of which the Eady model
is perhaps the simplest example. The accomplishments of both these independent
analyses are staggering. Not only did Eady and Charney, who were working indepen-
dently, correctly identify the physical process responsible for synoptic scale waves in
the atmosphere (and ocean), but they had to derive a version of quasi-geostrophy at
the same time. For those of you starting graduate school, it will give you a standard to
strive for to know that this represented Charney's Ph.D. thesis.
For K greater than the critical value Kc = 2.3999/ LD , both roots for the phase speed
are real. As K gets very large, each root approaches the value of the zonal velocity at
Eady eigenfunctions 110 = KNO / fo

1.0,~ , , , ,
z
0.9
0.8
0.7
0.6
o.s
0.4
0.3
0.2
0.1
0.0 I II
-1.0 -0.9 -0.8 -0.7 -0.6 -O.S -0.4 -0.3 -0.2 -0.1 0.0
110 = 6, Crl = 0.83333, Cr2 = 0.16667, Cil = 0
Fig. 19.3. The stable Eady eigenfunction for KLD = 6
one of the boundaries at z = 0 or z = 1, and the eigenfunction resembles the trapped

boundary wave of the semi-infinite interval model. Figure 19.3 shows an example when
KLD = 6. The eigenfunction with c / (UzD) about 0.8 is shown. Note its intensification
near the upper boundary.
In this limit, there is no interaction between the upper and lower boundaries and
the wave is stable. We shall see later a theorem that will explain why, in the Eady model,
such an interaction is necessary for instability.
As K approaches its critical value from above, the two roots for C coalesce. For K
less than the critical value, there are two roots which are complex conjugates.
The growth rate is
Wj =kCi(K)=kC{ Jk 2 +l2 ) (19.59)
That is, the complex phase speed, as we can see from the original eigenvalue prob-
lem is a function only of the total wave number. The growth rate is the imaginary part
of that phase speed multiplied by the x-wave number, i.e., by the component of the
wave vector in the direction of the basic velocity. The largest growth rate will there-
fore occur for a given K when k is largest, i.e., when the y-wave number, l, is zero. In
the figure showing the growth rate, I have chosen the case where 1= O. The maximum
growth rate occurs for k on the order of 1.6/ LD , which gives a quarter wavelength of
just under LD itself. This is the basic explanation for the presence in both the atmo-
sphere and the ocean of synoptic scale eddies with the scale, preferentially, of the de-
formation radius.
The fact that the instability is maximized for 1= 0 is related to the energy source
for the waves. Since the motion is horizontally divergent to the lowest order (geo-
strophic), the perturbation velocity is perpendicular to the wave vector. With the wave
vector oriented in the x-direction, the perturbation velocity will be directed across the
current in the y-direction.
Now from the thermal wind relation,
P y = foU z
Po g
and using the relation
Po
_z= N2
__
Po g
it follows that the slope of the isopycnals in the basic state is
ozJ
oy
= foU z
N2
(19.60)
p
Motion in the y-direction will therefore move fluid down the density gradient, and
the fluid motion has a chance to release the potential energy that is stored in the slop-
ing density surfaces, a slope required to balance the vertical shear of the current in
which the wave is embedded.
The important point to keep in mind is that these waves are self-excited; since they
are unstable, the slightest perturbation of the basic flow will produce a spectrum of
growing waves, and we anticipate that at least until the amplitude becomes large enough
for nonlinearity to be important, we will see the most unstable wave dominate the spec-
trum. That is, we don't need an external forcing mechanism to produce the wave, in
distinction to all the wave types we have discussed before. We will have to discuss more
completely the energy source for the waves that is in the basic current, but it should be
intuitively clear that the sloping density surfaces are a potential source of energy if the
perturbations on average can level those surfaces releasing potential energy to pertur-
bation kinetic energy.
For the unstable wave, the function cPwill be complex, since c is complex and A and B
will thus be complex. It is useful to recognize this and write the stream function:
'If = cP(z )ei(kx-mt)

(19.61)
=IcP(z )Iei(kx+a(z)-mt)
In the above, we have separated the amplitude function rpinto its modulus and its phase
a(z) = arg( cP) (19.62)
Eady eigenfunctions p =KNO / fo

,,,
1.0. • • I
,,
0.9
.... a(z) of decaying wave
0.8 I
I
I
0.7 I
I
I
0.6 I
I
/
1(/)1
/
0.5
I
/
I
0.4 I
I
I
0.3 I a(z) of unstable wave
I
I
0.2 I
I
I
,,
0.1 I
O.OLI-4--------L---------~----------~--------~----------~--------~
-3 -2 -1 0 1 2 3
pO = 1.61, Cr l = 0.5, Cr2 = 0.5, Cil = 0.19243
Fig. 19.4. The modulus (solid) and phase (dashed) for the stable (ci < 0) and unstable (ci> 0) waves.
The phase of the unstable mode increases with height
A line of constant phase in the x-z-plane will be a line on which
ph = kx + a(z) = constant (19.63)
The slope of a line of constant phase is therefore
(~:)Ph
k
(19.64)
da/dz
Figure 19.4 shows the modulus of the eigenfunction and its phase for a wave num-
ber very near the wave number of maximum growth rate. Both the unstable wave and
its complex conjugate are shown. Of course, the modulus of ([J is the same for both.
Note that the function a increases with z for the unstable wave and decreases with z
for the stable wave.
For the unstable wave, the fact that a increases with z means that a line of constant
phase of the unstable wave has a negative slope in the x-z-plane; that is, it leans against
the current shear. Intuitively, for a passive tracer we would expect isolines of the tracer
to be pitched over in the direction of the shear. The unstable wave has an active structure,
and to extract energy from the basic flow it must lean against the shear in the current.
Lecture 20
Energy Equation
and Necessary Conditions for Instability
To get a better feeling for where the source of the instability is, it is useful to develop
an equation for the perturbation energy for waves in the presence of a mean flow that
contains both horizontal and vertical shear. This entire subject is enormous, and we
will only scratch the surface in our discussion. The text by Gill (1982) and Pedlosky
(1987) contain ample discussion for further reading.
We start with the governing equation for the linear perturbations derived in the
last lecture:
[~+U~][VllP+~(fol
at ax az N dlPj]+
dz
alP[aq]=o
ax ayl
(20.I)
where the term in the last bracket is the potential vorticity gradient in the y-direction
associated with the basic flow. It contains contributions from {3, the relative vorticity
gradient and the gradient of the thickness between isopycnal surface in the basic state
aq_p_u _~(folauj
ay - YY az Nl az
(20.2)
~P-u,,-f" ~W~Ll
where z(y) is the position of an isopycnal in the basic state.
To derive the energy equation, we follow the steps we took earlier in finding the
energy flux vector for quasi-geostrophic Rossby waves. We multiply the qgpve by the
stream function and manipulate the derivatives to work the form into a budget for
the energy. The details are a good deal more tedious here, because U is a function of y
and z and there are many "extra" terms. It is these terms that are the most illuminat-
ing. The details of the derivation will be left for the student. The result with no fur-
ther approximation can be written
a
-E+V·S+-
at
fil lPzt+UlPzx
- a (lP-
az N
{}j fi Uz
=-lPxlPyUy-lPxlPz-l
N
(20.3a)
E = (V lP)l + follP; (20.3b)

2 2N l

222 Lecture 20 . Energy Equation and Necessary Conditions for Instability
- [(-+u-
S=-rp a a )Vrp ]+x{a-
--.!L~+UE+rp(Vrp.vU)
2 av
} +rpu~ (20.3c)
at ax ay 2 ax
x
where is a unit vector in the x-direction. The energy E is the sum of the kinetic en-
ergy and the potential energy. The second term in E is the potential energy, since
firp; _l(p/ PO)2

(20.4)
2N2 - 2N2
which we recognize from our discussion of internal waves as the representation of the
potential energy in the wave field. The horizontal flux vector is similar to that which
we found for Rossby waves. The local time derivative in the first term is replaced by the
linearized advective derivative, and the beta term is replaced by the full potential vor-
ticity gradient. This is supplemented by the advection of energy in the x-direction by
the mean flow plus two other terms. These terms are more difficult to interpret easily,
but they are related to corrections to the higher order work terms done by the geo-
strophic pressure correction. The horizontal divergence of this flux vector has its com-
panion in the z-direction.
If the fluid is contained within solid walls in z and y so that the boundary conditions
at z = 0 and z = -D are as described in the previous lecture, and if the perturbation is
either periodic in x or vanishes for large positive and negative x, then the volume integral
of the flux terms will contribute no net term to the energy balance for the perturbation
energy. This is really just a consequence of the definition of energy flux. The flux vector
moves the energy from one place to another without creating or destroying energy.
However, there are two terms: these terms are on the right-hand side of the
energy equation that in general, do not integrate to zero when the volume integral
is carried out. The first of these is already familiar from our discussion of the en-
ergy flux in internal gravity waves in a mean current. Using a bracket to denote a
volume integral, this term is
(rpxrpyU y) = -(UVU y)
and thus is the integral of the horizontal Reynolds stress times the horizontal shear of
the basic current. If the perturbation carries larger values of zonal momentum to re-
gions of lower momentum tending to smooth out the mean lateral shear, i.e., if when
Uy > 0 and v < 0 we also have u > 0 so that the perturbations "remember" that they have
come from a region of large zonal momentum compared to their destination, the mean
shear will be flattened with a consequent increase in wave energy as the energy of the
basic current is reduced. Such an energy transfer requiring only horizontal motions
occurs in ordinary shear flow instability of a homogeneous fluid with lateral shear and
is termed barotropic instability.
In the Eady model we discussed in the last lecture, the basic current has no hori-
zontal shear so that this energy transformation process is absent. The remaining term
on the right-hand side is the pertinent one for that process. Using the relation between
geostrophic stream function and density perturbation,
Lecture 20 . Energy Equation and Necessary Conditions for Instability 223
.(2
~2 qVpPz =-V(pg) PO~2 =-Vpg ily p
gp (ilzJ (20.5)
This baroclinic energy transformation term is proportional to the transport in

the y-direction, i.e., in the direction of the density gradient of perturbation density.
If, for example Py > 0, ¢::> Uz > 0, on average (when integrated over the volume of the
fluid) parcels moving to positive y carry a negative density anomaly and fluid elements
moving from positive y carry a positive density anomaly. The product - vp > 0 and so
the energy of the wave field will increase. That is, if the wave field produces a flux of
density from regions of high to low density of the basic state, tending to smooth out
the basic horizontal density gradient, this will flatten the slope of the mean isopycnal
surfaces and release energy for the perturbations.
It is really a form of convection. In ordinary convection in which fluid is heated
from below, energy is released by having warm, light fluid rise and cold, heavy, dense
fluid sink. Here the situation is a bit more complex, but fluid coming from the region
of larger mean density moving to smaller mean density will tend to sink as it moves
laterally and vice-versa for the fluid moving in the opposite direction.
Notice that to have this transformation term positive, we need
0< Uz {jJx{jJz =U z{jJz2-(jJx =-Uz {jJz2(ilz)

- (20.6)
{jJz ilx rp
so that the product of the basic vertical shear multiplied by the slope of the isolines
of constant cfJ in the x-z-plane must be negative. That is, the phase lines of constant
perturbation of the geostrophic stream function must lean against the shear, as
we already noted from the Eady model. Here we see that it is a necessity to release
the potential energy locked up in the sloping isopycnal surfaces of the basic state.
Figure 20.1 shows a cross-section in the x-z-plane of a growing Eady mode. The solid
lines are the geostrophic stream function, and the dashed curves yield contours of
perturbation density. Note that the former lean against the shear, and the latter lean
with the shear. This phasing assures that on average the density flux is down the mean
density gradient.
In both the ocean and the atmosphere, horizontal density gradients are sources
of baroclinic eddy energy, and the eddies springing from the self-excited baroclinic
unstable waves typically have scales of the order of the appropriate deformation
radius.
Eady himself introduced a very simple argument to make plausible the con-
vective nature of the instability. He suggested considering the virtual displacement
of a fluid element in the y-z-plane. The isopycnals are sloping with an angle r
such that
tanr=(ilZJ = foUz
ily _ N 2
P
so that the slope is due to the existence of the vertical shear (Fig. 20.2).
Eady mode for pO =KNO / fo = 1.66

1.0
zlC
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0.0
0.0 0.5 1.0 1.5 2~ 2.5 3.0 3.5
Wi = 0.30924, c, = 0.5 x/Ld----.
Fig. 20.1. A cross-section in the x-z-plane of the growing Eady mode. The solid lines are isolines of
perturbation pressure, while the dashed lines show isolines of perturbation density
p= constant
j'o,z
,
~ ,~A
I~
Fig. 20.2.
The Eady wedge of instability
Consider a displacement of a fluid parcel from point A to point B as indicated in

the figure. Assuming the fluid parcel at A preserves its density when it arrives at B, it
will arrive there with a density anomaly:
t5p=PA -PB (20.7)

But if the displacement is small,
ap ap
PB = PA +-Az+-.1y (20.8a)
az ay
l J
l
ap ap
op=- -Az+-.1y
az ay
=_ ap Az 1+ ay
ap/ap
az
j (20.8b)
az Az
.1y
so that the anomaly of buoyancy will be
gOP =N2Az[1- zy)p ] (20.9)

Po .1z/.1y
If .1y is zero or if the slope of the basic isopycnals is zero, this reduces to the result
we obtained in reasoning out the restoring force for internal gravity waves. In that case,
we had a restoring force (a positive density anomaly for a positive Az) giving rise to a
force proportional to Az and with N2 as the spring constant (per unit mass). Now, how-
ever, if
0::::; Az::::; azJ

.1y ay p
the buoyancy anomaly will be negative and the arriving fluid parcel will have lower den-
sity than its surroundings. The resulting buoyancy force will then encourage a continued
displacement and the release of energy. That is, if the motion occurs so that on average
the fluid elements slope upwards within a wedge determined by the slope of the den-
sity surfaces with respect to the horizontal, the gravitational energy available will power
continued displacement rather than restoration to its initial position ~ instability.
From this point of view, the instability is a type of slanted convection requiring vertical
shear to yield the slope of the isopycnals and allowing the existence of the wedge of
instability.
This simple explanation has been criticized (Heifetz et al. 1998), since the wave func-
tion is not a plane wave in the x-z-plane, so it is not possible to avoid considering the pres-
sure perturbation in the force balance on the parcel. However, the basic argument on
the basis of the buoyancy force is compelling and, I feel, illuminating. In Fig. 20.3 we
show a snapshot of the v and w velocities in the y-z-plane at a particular value of x (quarter
wavelength). The solid lines are the isopycnals, and the arrows show the trajectory in
the y-z-plane.
Eady v and w for pO = KND I fa = 1.66, p*x = 0.7854

1.0
zlC
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0.0
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5
Wi = 0.30924, Ci = 0.18629, C, = 0.5 ylLd---.
Fig. 20.3. The instantaneous perturbation velocities in the y-z-plane are shown as arrows. The solid
lines are the basic state isopycnals
It is of interest to try, on the basis of our work up to this point and general di-
mensional analysis ideas, to estimate the characteristic growth rate of baroclinic,
unstable waves. In the presence of a vertical shear, a layer of depth D might be ex-
pected to have the imaginary part of its phase speed to scale with Up. The growth
rate would then be
CO; = kCj =O(kDUz ) (20.10)
and if the wavelength is of the order of the deformation radius,
k=O(fo/N D) (20.11)
so that
CO; =o(f~z )=N( ~;)p (20.12)
(note in the last form its apparent independence of j).

For mid-ocean flows, we might estimate
U _1 cms- 1
z- --
lOs cm
N=5X10- 3 S-1
fo =10-4 S-1
10- 6
=:>0>. =__ S-1
1 5
which yields an e-folding time of about 60 days.
General Conditions for Instability
The Eady model is a very simple one and hardly realistic. The investigation of more
realistic velocity structures usually requires considerable numerical work, and it is hard
to make general statements. It is useful to have some a priori ideas of when self-ex-
cited waves can be expected within geostrophic flows. There are a series of theorems
giving necessary conditions for instability. The student is referred to Chapter 7 of
GFD for a detailed discussion. Here we present only the most well-known theorems.
This class of theorem dates back to the original work of Lord Rayleigh.
Let us assume that our basic current is again directed in the zonal direction, but
imagine that it is now a function of both y and z and that the beta effect is not negli-
gible. It is important to note that once the current is not zonal, there are very few theo-
rems that are directly applicable.
However, the case is of interest, and it may be provide some general picture of what
is required even in the nonzonal case.
Again, if we look for plane waves in x (not y, since now the linearized potential vor-
ticity equation has nonconstant coefficients in y) the governing equation
[~+U~][V29'+~(
at ax
f~ a9')] + a9'[aq]=o
az l N az ax ay 2
(20.13)
admits solutions of the normal mode form
9' =P(y,z)eik(x-Cf)
where IJ'satisfies
(U-J~( f02 ap)+ a2p -ep]+ aq P=o

t az l N 2 az al ay
(20.14)
subject to boundary conditions on the bottom (which we take to be z = 0) and the top
(z = D):
(U-C)~+If"( -Uz + ~02 hb J=O, y Z=O (20.1Sa)
(U -C)~ +If"(-Uz ) =0, z=D (20.1Sb)
and we insist that there exist two lateral boundaries at Y = ±L where the disturbance
vanishes. Of course, L could be infinite.
If we multiply the equation for tpby its complex conjugate after having divided by
(U - c), and if we then integrate over the region of the problem in the y-z-plane, we
obtain, with the aid of the above boundary conditions
[f/
- Dfdz LfdY ----%-~12 +llf"yl 2 +k211f"12 ]
o -L N
+ DfdZ Lfdyjllf"12 dq/dY }+ LfdY f021tpl2 {U}_ (20.16)

U -c N2(U -c) z z-D
o -L -L
f0211f"12 {U 2 }
-t
L
d
Y N2(U -c) z
N
--h
fo by z;O
=0
The first term in this integral condition is always real and negative definite. If c is
complex, the remaining terms will have an imaginary part. Indeed, if we just write down
the imaginary part of the above equation using
1 (U -cr +icj)
(20.17)
(U -c)
IU- cI2
we obtain
fo dZ fdyjllf"12 dQ/2dY } + fdY

-L IU-cl -L N
~0211f"12 2{Uzt;D
IU-cl
c·'
I
=0 (20,18)
1/11f"12 {U 2
L
fdY --hb }
N
_c1
0
-L N 2 1U 2 z fo y z;O
For instability to occur, i.e" for the imaginary part of c to be different from zero,
the collection of integrals in the square bracket must add to zero.
For example in the Eady problem, the potential vorticity gradient in the interior of
the fluid is exactly zero, i.e.,
dq = fJ -U _~( f02 dUJ

dY yy dZ N 2 a-;
=P-Uyy -fo :z[1 :U (20.19)
=0
For instability to occur, the two boundary terms must be able to cancel each other.
For the Eady problem, there is no topography and Uz is positive at both boundaries so
that the cancellation is possible. However, as we saw, it is necessary that the wave number
be small enough so that the wave extends to both boundaries. If the eigenfunction were
zero at one of the boundaries, only one of the boundary terms in the above constraint
would survive, and it would be impossible to satisfy the condition for instability.
We noted earlier the term
z
2
U - N hb
fo Y
ex: [z )
Y P
-hb ]
Y
(20.20)
Thus, if the topography of the lower boundary were sloping upward more steeply
than the isopycnals (which have constant slope in the Eady model), the contribution
from the lower boundary term would add to that of the upper boundary, and instabil-
ity would be impossible. Topography can thus eliminate the instability and stabilize
the flow.
If both boundaries are boundaries of constant density, the boundary terms in the
integral condition vanishes. In that case, for instability, the gradient of the potential
vorticity must be both positive and negative in the y-z-plane. Potential vorticity of a
single sign would be (in the absence of the boundary contributions) a stable distribu-
tion. The simple exemplar is, naturally, pure Rossby waves.
In Charney's model, there is no contribution from the upper boundary (it is infi-
nitely far away) and the potential vorticity gradient is positive. Instability is possible
because the positive contribution from the pv integral is cancelled by the contribu-
tion from the lower boundary. Note that these conditions are necessary conditions for
instability, not sufficient conditions. It sometimes occurs that the necessary condition
is met and the flow is still stable. There are very few useful sufficient conditions that
can be found.
Problems
Problem Set 1
1. As discussed in the text we can consider the generalization of a plane wave to have
the form for waves of slowly varying properties;
Ij/ = A(x)eifl(x,y,z,t)
and we have defined the wave number vector as the gradient of the phase e.
a Show that the condition, for example, that the x-wave number is slowly varying
(i.e., that the local definition of a wave number makes sense) is that:
exx «1
ex 2
and carefully interpret this result, i.e., what does the condition mean and why
should the condition be imposed? Do the same for the frequency.
b Consider a circular water wave, perhaps formed by a stone thrown in a pond,
At
whose free surface elevation is given by:
T7 = ei(Kr-mt)
where ro is a constant and r is the circular radius:

r=(x 2 +l)l!2
Assuming the wave is slowly varying, find the x- and y-wave numbers of the
wave field at each point in the x-y-plane.
c Using your results in (a), under what circumstances will the assumption in (b)
be sensible? This should depend on 1\ and r.
2. The dispersion relation for Rossby waves which we will derive later, might be ap-
proximated as
(j)=_ j3k
(k 2+12) , (j)~O
where k and I are the x- andy-wave number components. f3is a parameter. For the plan-
etary problem, it can be shown that f3 is a measure of the Earth's rotation and sphericity
or it also could be related to the slight slope of the bottom of the fluid (as we shall see).
240 Problems
Let's suppose it is the latter, and in our lab we have let
fJ = fJ(y) = fJo(1-[y / Lf)
where L is large compared to a wavelength of the wave.

a Find the x- and y-components of the group velocity.
b Derive the ray equations for the variation for k, 1and w.
c Show that along the ray path k and ware constant so that
1=[fJo(1-!~/L2)k k2 J'2
d Find the position Yo where the group velocity in the y-direction vanishes. Note
from the ray equation for 1 that 1 continues to decrease at that point (i.e., be-
comes negative). Discuss the implications of that for the trajectory of a wave
packet which initially starts near y = o. Sketch the path in the x-y-plane.
3. A particular wave has the form
f/J = Ae i8 (x,t)
{}=-gt 2 /4x
a What is the x-wave number?

b What is the frequency?
c Under what conditions is it sensible to talk about a slowly varying frequency?
d At what speed need you move to see a constant frequency and wave number?
e Moving at that speed, what is the relation between frequency and wave number?
f At what speed do you have to move at to see a constant phase, e, (i.e., stay on a
particular crest)? Is that speed constant with time?
Problem Set 2
1. Acoustic waves in their pure form are small, adiabatic perturbations of a medium
of otherwise uniform density and pressure. Assuming that the specific entropy can
be written as 5 = s(p, p), show that the governing equations of inviscid motion for
disturbances propagating in the x-direction are
a
au
Po-=--
ap
at ax
ap au
-=-Po- and
at ax
~J ap --~J ap
ap 0 at ap 0 at
where 0 subscripts denote variables in the uniform, unperturbed state.
Problems 241
b Then show that p satisfies
rip
-=c-
2 rip
Jt2 a Jx 2
Identify the sound speed ca and discuss the nature of the solutions of the
equation. Do signals disperse? What significance does this have for communica-
tion by speech?
For a perfect gas like air, p = pRT, and under adiabatic transformation it fol-
lows from the standard thermodynamic relations that (Jp I Jp}s = yRT, Y= cpl CV '
What is the sound speed at room temperature?
2. Consider the atmospheric pressure field
Pa =Pocos(kx-at)
moving over an infinite body of water of depth D. Find the resulting periodic solu-
tion of the water after all initial transients have decayed.
3. Consider a small circular pond of depth D. Suppose the radius of the pond is R. Find
the free modes of oscillation for the free surface under gravity. Be sure to carefully
state the boundary conditions at the lateral boundary of the basin. Which mode has
the lowest frequency? If D = 3 meters and R is 10 meters, find that frequency. What
is the corresponding frequency in a small water glass (give an estimate)?
(Hint: Find solutions in the form I/J = F(r) cosh K(z + D} e ime e- iwt, and you may
be surprised to discover which m yields the lowest frequency.}
Problem Set 3
1. For a plane gravity wave of the form
TJ =TJo cos(kx- OJ/:}
we assumed in the text that we could neglect (a) nonlinearity, (b) friction, (c) com-
pressibility, and (d) planetary rotation.
Check these assumptions and discuss, in each case, what non-dimensional pa-
rameter measures the goodness of the approximation. Make sure you write the
condition in terms of quantities given in terms of 110' k, D and properties of the
fluid such as g and v. Be careful to distinguish the conditions when k D is both large
and small. You may use sensible values of the wavelength, depth, etc. to get an idea
of what limits these parameters set.
2. Consider a rectangular tank of sides Lx' Ly and depth D filled with homogeneous,
incompressible fluid. Suppose the fluid in this small basin is forced by a surface
pressure of the form
Pa =Po cos(kx- at}

242 Problems
Find the linear, forced solution (note: Since the problem is linear, the response
must be oscillating at the forcing frequency, but the spatial structure will be modi-
fied by the geometry of the basin). Be sure to carefully pose the boundary condi-
tions on the side walls. What Fourier series in x and y is appropriate for the bound-
ary conditions?
When will resonance occur?
What is the solution for small d?
What do we mean by small d?
3. In class we derived an energy equation for a layer of fluid supporting gravity waves
in the case when the applied atmospheric pressure was zero. Redo the calculation
when Pa-:f. o.
Problem Set 4
1. Consider the motion of a homogeneous layer of fluid of constant density and of

depth D. At t = 0, the surface of the fluid is flat but there is a vertical velocity such
that
w(x,z =O)=Wo(x)
Formulate the initial value problem and find the solution for 1J(x, t) in terms of
a Fourier integral and discuss the solution without reproducing the details of the
derivation of the stationary phase argument.
2. Energy in the internal gravity wave frequency range is generated at z = Zo with an

x-wave number k and a z-wave number m.
a Find the path of such a packet of energy in the x-z-plane (i.e., find dz / dx for the
group velocity ray). Estimate the time it would take the packet to reach a depth D
if it starts near the surface and if you assume N is independent of z.
b Discuss how you would do the problem in part (a) if N2 = N5exP(Z/ d), where dis
the thermocline scale (about 1000 meters) and the vertical wavelength of the
gravity wave is much less than d.
3. Consider a stratified fluid with constant N in an infinitely long channel of width L

with a rigid lid. Suppose that at t = 0,
w =wo(x)sin 1tZ / D
wt=O
Find the solution of the initial value problem if Wo is an even function of x (hint:
Note that with the initial condition as given, a solution for all t > 0 can be found in
the form w = W(x, t) sin(1t z / D)). Qualitatively discuss the solution after you have
obtained it.
Problems 243
Problem Set 5
1. Consider a plane, internal gravity wave in a fluid with a constant buoyancy fre-
quency. Calculate both the kinetic and potential energies and discuss whether there
is equipartition of the energy (note, it is convenient to choose a coordinate system
so that the wave vector lies in the x-z-plane).
2. The vorticity components along the three coordinate axes are
ow _ov
O1x = oy oz
ou _ow
O1y = oz ox
ov _OU
O1z = ox oy
a Derive, from the linear equations of motion, equations for the rate of change of
these vorticity components. In particular, show how the horizontal gradients of
density produce vorticity and physically interpret your result.
b Calculate the vorticity in a plane internal gravity wave when N is constant.
3. Consider the reflection of an internal gravity wave from a sloping surface. Show
that the energy flux normal to the surface of the incident wave is equal to the en-
ergy flux of the reflected wave. We showed in class that the energy densities of the
incident and reflected waves were not equal. Is energy conserved?
Problem Set 6
1. Reconsider the normal mode problem for internal waves in the case where N2 < 0, i.e.,
when heavy fluid is initially on top of lighter fluid so that oPo I OZ > O. Let the fluid be
contained in a layer of depth D between two rigid surfaces and let N2 be constant.
a What are the frequencies of the normal modes? Are they real? Interpret your
result in terms of growth of the disturbance.
b For what wavelengths will the perturbations grow the fastest?
c Given the length scale for maximum growth rate, what effect do you think fric-
tion or heat conduction would have in determining the wavelength of maximum
growth?
2. Calculate the normal modes of internal gravity waves for a stratified fluid with N2 > 0
when the fluid is contained in a box with sides oflength Lx and Ly and with depth D.
You may assume the upper boundary is a rigid lid. Find the free modes of oscillation
and their frequencies (hint: The boundary condition on x = 0, say, is u = o. That
implies that op I ox = 0 there for all z and so that (0 lox) (op I oz) = 0 on x = o. You
can use that to write the condition in terms of w).
244 Problems
3. Find an expression for the frequency leading to critical angle reflection of internal
gravity waves in terms of the local value of OJ, N and the slope of the topography.
Calculate the condition on frequency for realistic oceanic values. Do you believe
the Coriolis effect can be ignored for such frequencies? How would you decide?
Problem Set 7
1. Consider a layer of fluid with a buoyancy frequency N (constant). The fluid is flowing in
the positive x-direction with constant velocity U. The base of the fluid is rippled such that
h = hmcoskx
where h is the (small) departure of the bottom of the fluid from a flat surface. The upper
surface of the fluid is level and rigid at a distance D from the bottom.
a Find the steady solution for the flow (it is nonrotating).
b Discuss whether resonance can occur and interpret your result.
c Calculate the drag on the rippled boundary. Are you surprised (hint: Consider an
explanation in terms of the net radiation of energy and the relation between
work done and drag)?
2. Achannel,semi-infinite in the x-direction (O:S; x:S; 00) of depth Dand width L (O:S; y:S; L),
contains a stratified fluid of constant buoyancy frequency N. The fluid is contained
between two level, rigid horizontal boundaries. At x = 0, a wave maker continuously
imparts to the fluid a velocity in the x-direction,
u = ReUOcos(rcz/ D)e irot
Find the periodic response of the fluid to the periodic forcing. If (0 is less than N,
carefully describe how you determine the proper condition on the solution for
large x (hint: You may have to apply a radiation condition).
3. Consider a plane wave in an unbounded, stratified fluid with constant buoyancy

frequency N and constant Coriolis parameter f For simplicity, align the x-axis of
the system so that the wave vector has no y-component.
a Is the energy equipartitioned between kinetic and potential energy? Is there a
particular wavelength for which equipartition occurs?
b In terms of the stream function amplitude, determine the vorticity and the po-
tential vorticity.
Problem Set 8
1. Consider the motion of a rotating, homogeneous layer of water of depth D. Let the
layer be infinite in horizontal extent. Suppose that at t = 0, the elevation of the free
surface above its resting value is given by
h = ho' -a:S; y :s; a
and is zero elsewhere. The velocity at t = 0 is also zero.

Problems 245
a Find the equation governing the free surface displacement in the steady geo-
strophic portion of the solution.
b Show that at y = ±a, both TJ and its first derivative are continuous in the geo-
strophic solution.
c Find the steady solution for the x-velocity.
d Using the relation between the free surface elevation and the potential vorticity,
find the energy in the steady geostrophic state. Discuss, as a function of the
ratio of the deformation radius to the length interval 2a the percentage of the
initial energy radiated away by gravity waves.
2. Show that for the linearized motion of a layer of homogeneous, rotating fluid, that
the relation between the free surface height and the velocities can be written:
a2 ii 2- a
- + f u=-g-Y'TJ + gfkxY'TJ
A
at 2 at
(you may find it useful to write the above in component form).
What happens for an oscillation for which the frequency exactly matches the
Coriolis parameter?
3. Consider a rotating fluid of depth D contained in the region x ~ 0, -00 ~y ~ 00. Sup-
pose that along a wall at x = 0, the velocity in the x-direction is given by
u =Uoei(ly-OX)
where it is understood that the real part of the above term is relevant.
Find the solution for the free surface height in x > O. Distinguish the case when
the frequency is greater or less than f
Problem Set 9
1. Consider the Kelvin wave in a channel of width L. If the free surface elevation has
the form
77 = 770 cos(kx - OJt)e - f y/co
a Find the relative vorticity in the wave and its potential vorticity.
b Calculate the kinetic and potential energy in the wave and check for equipartition.
c Discuss the trajectory of fluid elements as the wave passes.
2. A Poincare wave with x-wave number kl «0) and y wave number £1 approaches a
wall at x = 0 from the right.
a What angle does the group velocity make with the x-axis?
b What is the frequency of the wave?
c If the amplitude of the free surface height in the incident wave is AI' find the com-
plex amplitude of the free surface height of the reflected wave and its x- and y-
wave numbers (hint: Be sure to carefully write out the condition u = 0 at the wall).
246 Problems
3. a Derive the governing equation for the velocity component, v, in for a layer of
rotating fluid of constant depth in the channel as discussed in class and discuss
its solutions of the form v = v (y) ei(kx-OJt).
b From v, how would you find u and T/? For what frequencies does this relation fail?
(Hint: To find u in terms of v, take the time derivative of the x-equation of mo-
tion and use the continuity equation to eliminate T/t' then use that relation to
find T/ in terms of v)
Problem Set 10
1. Consider a plane Rossby wave with a free surface
17 = 170 cos(kx + £y -0Jt)
a Calculate the kinetic and potential energy in the wave. Check for equipartition.
Is there a particular wavelength for which equipartition obtains if it is not true
generally?
b Calculate cgx as a function of k. Where does it have its largest positive and nega-
tive values? Where is it zero? For a wave with a wavelength A = SO km, estimate
the period of the wave for a fluid of depth 4 km (the precise value will depend
on the orientation of the wave vector).
2. a Show that a single plane Rossby wave is an exact solution of the nonlinear quasi-
geostrophic potential vorticity equation (qgpve) (hint: First calculate the relative
vorticity in the wave and show it is a constant multiple of the stream function).
b Show that an arbitrary sum of plane Rossby waves will be a solution of the non-
linear qgpve if the magnitude of the wave vector of each wave is identical. Note
that the frequencies of the waves will differ. Suppose, instead, you have a set of
waves of varying wavelengths but whose wave vectors are co-linear?
3. Consider a channel of width L on the beta plane, i.e., 0 ::; y ::; L. The bottom is flat. At
x = 0, a wave maker produces a zonal velocity of the form
u = Uo cos(ny lL)e- iOJt
a Find the resulting Rossby wave for the region x > o.

b Do the same for the region x < O.
Problem Set 11
1. Reconsider the development of the quasi-geostrophic equations when
jJL =0(1), hb =0(1) and E'« 1

f D
Derive the governing potential vorticity equation in this limit. Under what cir-
cumstances could the Rossby wave frequency satisfy OJ «f?
Problems 247
Fig. p.Ol.
A definition figure for problem 2. The arrow
shows the direction of the incident wave energy
2. Consider the reflection of a westward propagating Rossby wave. Its group velocity
is directed west-northwestward in a direction that slopes 45° to the northeast from
a latitude circle (see Fig. p.Ol).
a If the frequency is given, how would you determine the wave number vector of
the incident wave?
b Discuss the reflection of the wave. In what direction is the reflected group veloc-
ity? What is the wave vector of the reflected wave?
3. Suppose we model the southern boundary Ys of the Gulf Stream as a rippling sur-
face propagating eastward. We prescribe that boundary as
Y =Ys + Yo sink(x-ct), Yo« Ys
Now consider the oceanic region south of that boundary (i.e., y ~ Ys).
Describe the resulting possible wave radiation in the region Y < o. Consider both
positive and negative values of c.
Problem Set 12
1. Consider the dynamics of a Rossby wave triad as discussed in class. From the prop-
erties of the function P(Kn, Km)' show that the ens trophy in the triad
3
V= 'L,E/Kj+a 2 )
j=!
is conserved where Ej is the energy in each wave component (note: This implies
that
3
'L,EjKj
j=!
is also conserved).
248 Problems
2. From the quasi-geostrophic equations, show that for a fluid in an infinite region
whose motion is limited to the finite part of the x-y-plane, the total enstrophy
V= snS--a2\1/)dxdY
is conserved.
3. Consider the reflection of a linear Rossby wave from a western boundary oriented
in the north/south direction (parallel to the y-axis). Calculate the ens trophy of the
incident and reflected waves. Is the emerging enstrophy flux equal to the incident
flux? If not, what is the mechanism for the non conservation? Discuss your result
and its implications. What is the situation if the reflection occurs at a northern
boundary that lies along a latitude circle?
References
Lecture 1
Lighthill J (1978) Waves in fluids. Cambridge University Press, Cambridge, pp 504

Whitham GB (1974) Linear and nonlinear waves. John Wiley & Sons, New York, pp 636
Lecture 2
Bretherton FP (1971} The general linearized theory of wave propagation. In: Reid WH (ed) Math-
ematical problems in the geophysical sciences, vol 1. American Mathematical Society, pp 61-102
Pedlosky J (1987) Geophysical fluid dynamics. Springer-Verlag, New York, pp 710
Lecture 3
Batchelor GK (1967) An introduction to fluid dynamics. Cambridge University Press, London, pp 615
(especially Chapter l}
Kundu PK (1990) Fluid mechanics. Academic Press, San Diego, pp 638 (especially Chapter 5)
Lamb H (1945) Hydrodynamics, 6th edition. Dover Publications, New York, pp 738
Stoker JJ (1957) Water waves. Interscience, New York, pp 567
Lecture 4
Kundu PK (1990) Fluid mechanics. Academic Press, San Diego, pp 638 (especially Chapter 5)
Lecture 5
Jeffreys J, Jeffreys BS (1962) Methods of mathematical physics. Cambridge University Press, Cam-
bridge, pp 716 (especially Chapters 14 and 17)
Lighthill J (1978) Waves in fluids. Cambridge University Press, Cambridge, pp 504 (especially Chap-
ter 3, Section 3.7)
Morse PM, Feshbach H (1953) Methods of theoretical physics, vol 1. McGraw-Hill, New York, pp 997
(especially Section 4.8)
Stoker JJ (1957) Water waves. Interscience, New York. pp 567
Lecture 6
Rossby CG (1945) On the propagation of frequencies and energy in certain types of oceanic and
atmospheric waves. J Meteorol 2:187-204
Whitham GB (1974) Linear and nonlinear waves. John Wiley & Sons, New York, pp 636 (especially
Section 11.4)
Lecture 7
Gill AE (1982) Atmosphere-ocean dynamics. Academic Press, Harcourt Brace & Co., San Diego, pp 662
(especially Chapter 6)
Lighthill J (1978) Waves in fluids. Cambridge University Press, Cambridge, pp 504 (especially
Chapter 4)
Munk W (1981} Internal waves and small scale processes. In: Wunsch C, Warren BA (eds) Evolution
of physical oceanography. MIT Press, pp 264-291
250 References
Lecture 8
Lighthill J (1978) Waves in fluids. Cambridge University Press, Cambridge, pp 504 (especially Chapter 4)
Mowbray DE, Rarity BSH (1967) A theoretical and experimental investigation of the phase configu-
ration of internal waves of small amplitude in a density stratified liquid. J Fluid Mech 281-16
Lecture 9
Levitus S (1982) Climatological atlas of the World Ocean. U.S. Dept. of Commerce, US. Government
Printing Office, PP 173
Lecture 10
Eliassen A, Palm E (1960) On the transfer of energy in stationary mountain waves. Geofys PubI22:1-23
Lighthill J (1978) Waves in fluids. Cambridge University Press, Cambridge, pp 504
Lecture 11
Cairns JL, Williams GO (1976) Internal wave observations from a midwater float, 2. J Geophys Res
81:1943-1950
Garrett C, Munk W (1979) Internal waves in the ocean. Annu Rev Fluid Mech 11:339-369
Lecture 12
Cahn A (1945) An investigation of the free oscillations of a simple current system. J Meteorol 2: 113-119
Rossby CG (1938) On the mutual adjustment of pressure and velocity distributions in certain simple
current systems, II. J Mar Res 1:239-263
Lecture 13
Cahn A (1945) An investigation of the free oscillations of a simple current system. J MeteoroI2:113-119
Pedlosky J (1987) Geophysical fluid dynamics. Springer-Verlag, New York, pp 710
Lectures 14, 15
Pedlosky J (1987) Geophysical fluid dynamics. Springer-Verlag, New York, pp 710 (especially Chapter 6)
Lecture 16
Greenspan HP (1968) The theory of rotating fluids. Cambridge University Press, London, pp 327
Longuet-Higgins MS (1964) On group velocity and energy flux in planetary wave motions. Deep-Sea
Res 11 :35-42
Pedlosky J (1987) Geophysical fluid dynamics. Springer-Verlag, New York, pp 710 (especially Chap-
ters 3 and 4)
Lecture 17
The derivation of the Laplace tidal equations in oceanic and atmospheric contexts can
found in many references, among which:
Andrews DG, Holton JR, Leovy CB (1987) Middle atmosphere dynamics. Academic Press, New York,
pp 489 (especially Chapter 4)
Moore DS, Philander SGH (1977) Modeling the tropical oceanic circulation. In: Goldberg ED, McCave
IN, O'Brien JJ, Steele JH (eds) The Sea, vol 6. Wiley Interscience, pp 319-361
References 251
Lecture 18
Andrews DG, Holton JR, Leovy CB (1987) Middle atmosphere dynamics. Academic Press, New York,
pp 489 (especially Chapter 8)
Eriksen CC, Blumenthal MB, Hayes SP, Ripa P (1983) Wind generated Kelvin waves observed across
the Pacific. J Phys Oceanogr 13:1622-1640
Moore DS, Philander SGH (1977) Modeling the tropical oceanic circulation. In: Goldberg ED, McCave
IN, O'Brien JJ, Steele JH (eds) The Sea, vol 6. Wiley Interscience, pp 319-361
Philander SG (1990) El Nino, La Nina, and the southern oscillation. Academic Press, San Diego, pp 289
Schiff LI (1955) Quantum mechanics. McGraw-Hill, New York, pp 417 (especially Chapter 4)
Lecture 19
Charney JG (1947) The dynamics of long waves in the westerlies. J MeteoroI4:135-162

Eady ET (1949) Long waves and cyclone waves. Tellus 1:33-52
Lecture 20
Gill AE (1982) Atmosphere-ocean dynamics. Academic Press. Harcourt Brace & Co., San Diego, pp 662
Heifetz E, Alpert P, Da Silva A (1998) On the parcel method and the baroclinic wedge of instability.
J Atmos Sci 55:788-795
Lecture 21
Charney JG, Drazin PG (1961) Propagation of planetary scale disturbances from the lower to the
upper atmosphere. J Geophys Res 66:83-109
Edmond HJ, Hoskins BJ, Mcintyre ME (1980) Eliassen-Palm cross sections for the troposphere. J Atmos
Sci 37:2600-2616
Index
A - frequency 60,72,73,75,243,244
-, distribution 75
acceleration, vertical 119, 120 -, standard definition 206
adjustment - profile, eigenfunction 89
-, geostrophic 123
-, problem 123 c
advection 205,206,211
-, horizontal 206 Cartesian coordinate system 3, 26, 149
-, of energy 222 channel
Airy function 55 - mode 136
amphidromic point 145 - problem, mid-latitudes 195
approximation rotating 137
- beta-plane 183,205 -, velocity 143
- hydrostatic 120 Charney model 229
- geostrophic 123, 168 circulation
- Eulerian 236
B - residual 234, 236
-, mean 236
background -, meridional 234
-, density 206 -, velocity, residual 235
-, flow 91 compressibility 20,241
balance -, adiabatic 21
- geostrophic 123, 143 condition
- hydrostatic 35 - arbitrary initial 58
-, quasi-geostrophic 205 - boundary 23,25,26
basic - dynamic 24
- current, energy 222 -, kinematic 24
- flow 221 cone of constant frequency 64
-, state 221 conservation equation 11 0
Bernoulli equation 24, 34, 81 continuity equation 61,68,91-93,96,120,150,
beta effect 227 186, 187
-, topographic 209 convection 223
beta factor, planetary 157 coordinates, Cartesian 3, 26, 149
beta plane 205,246 Coriolis
- approximation 149,150,152,153,162, - acceleration 143,149,159,184,185
164,183,205 - effect 244
-, equatorial 193-195,204 - force 127, 184
bottom friction 180, 182 - frequency 107, 109
boundary - parameter 107,108,134,149,152,159,
- condition 23,25,26 165,180,193,194,205,244,245
-, mixed 137 crest 6,31
- lateral 26, 142 current
-, wave 211 -, basic, energy 222
Brunt-Viiisiilii frequency 60 -, equatorial 231
buoyancy 109,225 cyclone
-, anomaly 225 - center 180
-, force 63, 225 -, wave 1
254 Index
o - eigenfunction 216
- mode 223,224
data, initial 133 - model 213,216,217,222,223,227,229
decay -, problem 228,229,233
-, rate 182 Earth
-, with time 46 - rotation 20,107,185,239
deep water wave 60 -, equatorial 193
deformation -, sphericity 120,149, 156,239
- length 141 eddy
- radius 125-128,131,134,140-142,158, - energy, baroclinic 223
161,177,208,215,217 -, synoptic scale 217
-, baroclinic 208 eigenfunction 85, 88
-, equatorial 196, 197, 199, 203 eigensolution 84
-, Rossby 125 eigenvalue 84, 85
density 21 -, condition 138,142
- anomaly 225 -, problem 82,217
- background 206 -, relation 82,83, l37, l38
- basic state 92 Ekman layer 180, 181
- distribution 117,233 El Nino (ENSO) 193
-, annual mean 77 elevation, free surface l39,140
- disturbance 21 Eliassen and Palm flux vector 233
- flux 223 energy 17,53,129,131,173,222
-, horizontal meridional 233 - advection 222
- gradient, horizontal 211, 223 - baroclinic transformation 223
- mean 236 - budget 221
- perturbation 222 - conservation 78, 243
- surface 73 - equation 73,93,96,129,221,222
-, sloping 218 -, thermodynamic 19
- vertical scale height 188 - equipartition 38,73
-, wave flux 237 - fields of motion 33
depth, equivalent 186 - flux 68,96,101,103,173,175
Dirac delta function 124 - horizontal 39
dispersion - magnitude 177
- diagram 202 - radiation condition 102
- Eady 215 -, vector 37,38,177,221
- relation 7,13,17,27,33,43,50,63,64,67, -, vertical 74,78,96,99,101
83,113,138,190,203,209 - gravitational 225
- Kelvin wave 209 - gravity wave field 53
- linear 17 - kinetic 37,72-74,94,129,173,222,245,
- local 75 246
- Poincare wave 193 -, change 94
-, Rossby wave 156 -, equation 72
displacement, vertical 73, III - pertubation 93
-, power spectrum 107 - potential 35,37,72-74,173,222,245,246
dissipation function 19 - propagation 35,41,54
disturbance 48,50,53,67,78 -, diagram 175,176
- Fourier amplitude 56 -, downward 116
- initial 50 - ratio l31
- shape 133 - transfer 171,222
-, wave equation 122 - transformation 19,72
domain 136 -, term 95
Doppler shift 104 enstrophy flux 248
drag 99, 103, 105 entropy 21
-, on mountain 99 -, specific 19
-, on topography 99, 103 equation
-, positive 103 - adiabatic 61, 74
dynamics, nonrotating 22 -, motion 60
- Bernoulli 24, 34, 81
E - continuity 61,91-93,120,186
- energy 19,73,93,96,129,221,222
Eady - free surface elevation l39
- dispersion relation 215 - Hermite 199
Index 255
- horizontal momentum 21,205 - stratified 205,208,211

- hydrostatic 185,206 -, motion 183
- internal wave field 61 - unbounded 107
- Laplace 23, 26, 36, 41, 42 - velocity 135
-, tidal 183 flux
- mass conservation 151,153 - enstrophy 248
- mode, vertical 183 - incident 248
- momentum 19,23,93,95 - momentum 93
-, horizontal 121 - vector 174,175,222
-, linearized 186 -, horizontal 222
- motion 19,91,92 force
-, hydrostatic 183 - buoyancy 63
-, linearized 92 - gravitational 26
-, vertical 61 -, restoring 60
- perturbation energy 221 forcing 41
- potential vorticity 181,231,246 Fourier
- ray 240 - amplitude 56, 57
- Schroedinger 199 - analysis 53
- structure, vertical 183 - integral 42, 242
- thermodynamic 20 - theorem 2
- vorticity 62,122,151,205 -, transform 44
- wave energy propagation 36 free surface 51,127
- wind, thermal 210 - deformation radius 208
equator 193, 196, 197 - displacement 36,187,245
-, trapping zone 194 -, rate 35
equatorial - elevation 33-35,81,127,129,135,139,
- beta-plane 193,195,204 140,161,162,245
- current 231 -, amplitude 34
- deformation radius 196 -, equation 139
- wave -, height 45,127,129,245
-, dispersion diagram 202 free wave 33
-, eigenfunctions 200 French Academy of Sciences 41
-, guide 194 frequency 7,209,240
equipartition 245,246 - conservation 69
-, energetical 38, 73 - dependence on wave number 33
equivalent depth 186 - intrinsic 104
Eulerian circulation 236 -, definition 104
evanescence 117 - propagation 50
expansion, thermal 19 - rate of change 14
- relation to wave number 35
F - slowly varying 240
-, wave orientation 63
fields of motion 33 friction 20,91, 103, 180,241,243
flow function, dissipation 19
- basic 98, 221
- geometrical scale 126 G
- geostrophic 129,227,237
-, vorticity 181 gas, perfect 19,241
- mean 105,221,235 geostrophic
-, momentum 94 -, approximation 123
- quasi-geostrophic 181 -, state, ultimate 123
- unidirectional 91 -, stream function 206,207,222
fluctuation, zero pressure 61 gradient, horizontal density 96
fluid gravitational restoring force 60
- compressible 20 gravity wave 1,9, 15
- density stratified 59 - field 33
- element, trajectories 39 -, energy 53
- incompressible 21,22 -, equation 61
- motion 33,217 -, properties 45
-, low frequency 160 - internal 59,66,67,69,82,91,92,243
- nonrotating 109,122 - crest 67
- rotating 120,121,136,142,245 -, energy 242
256 Index
- generation 100 L
- group velocity 65
- incident 70 Lagrangian
- maximum frequency 81 -, pathway 234
- mode 86, 243 -, trajectory equations 39
- reflection 68, 70, 243, 244 Laplace
- rotation 119 - equation 23, 26, 36, 42, 81
-, theory 75 -, two-dimensional 41
-, speed 201 -, vertical velocity 62
Greens function 125 - operator 26
group velocity 13,14,17,30,39,41,47,50,52, - tidal equations 144, 183, 191,208
54,66,67,105,115,157 -, transformation 103
- as function of wave number 30 layer
- definition 13 - stratified 213
- direction 64, 175 -, geostrophic flow 181
- internal wave 65 line of constant phase 63
- maximum 158 linearization 2, 21, 25, 35, 80
- minimum 158 -, restriction 231
- Poincare wave 134 linearized momentum equation 186
- orientation 66 Lord Rayleigh 227
- ray 242
- relation to wave vector 67 M
- Rossby wave 157, 158
three-dimensional 65 mass conservation 19,151
- versus KD 47 maximum phase speed 27
-, vertical 67 mean
growth rate 215,217,219 - definition 231
-, maximum 219 - flow 209,221,231,235
Gulf Stream 180,231,247 - momentum 233
- potential vorticity, rate of change 232
H -, state 231
medium, homogeneous 2,9, 16
heat mode 137, 139-141, 189, 191
- conduction 243 - baroclinic 197,208
- nonreversible addition 19 - barotropic 190,197,208
specific 19 - channel 136
Hermite equation 199 - external, dispersion relation 85
- internal gravity 189
-, wave 86
- normal 79,117,227
incompressibility 72 -, definition 80
initial value problem 41,53,100,103,242 -, frequency 243
-, solution 49 -, problem 243
instability 225,228,229 - phase speed 139
- baroclinic 213,216 -, vertical 183
- barotropic 222 model, homogeneous 183
- conditions 221,227 momentum
-, source 221 - equation 19,23,93,95
interaction, wave-mean flow 231 -, horizontal 121,205
isopycnal 1, 207, 211 -, vertical 96
-, sloping 212 - mean 233
-, surface 221 -, zonal 234,236
-, wave flux 237
K motion
- adiabatic 20, 184
Kelvin -, equation 60
- mode 142,148,195,202 cooperative 1
- wave 133,136,142,144,148,152,154, cyclonic geostrophic 181
191,195,203,245 equation 19,91,92
- dispersion relation 209 -, linearized 92
-, superposition 144-146 -, rotating stratidied fluid 19
Index 257
- geostrophic 131, 183 -, method of 45

- hydrostatic 120, 183 -, wave number 47,52
-, equations 183 -, variable 2
-, large-scale 119 plane, equatorial 193,204
- in ocean 119 plane wave 2,3,5,9
- incompressible 22 - dispersion relation 17, 77
- inviscid 22, 184, 190 - field of motion 37
- linear 22 - generalization 239
-, equations 243 - solution 26,33
- linearized 183 -, tree-dimensional 62
- low frequency 154, 160, 168, 205 - two-dimensional 73
- oscillatory 23 - wavelength 4
- quasi-geostrophic 167 Poincare
-, stratified 205 - mode 195,202
- vertical - wave 115,133,139,141,144,148,152-154,
- equation 61 159,194,201,204,245
- genesis 181 -, dispersion relation 193
- scale 119 -, minimum frequency 203
point
N - amphidromic 145
-, stationary 48
nondispersive wave equation 133 Poisson 41
nonlinearity 20,231,241 position, fixed 52
potential vorticity 107,110-113,122-124,127,
o 129,133,166,210,229,245
- adjustment problem 123
operator, Laplacian 26 -, ambient 169,173
oscillation 38, 61, 117, 215, 245 - conservation 112, 113, 122
- free 79 - equation 169,181,208,212,231,246
-, modes 241,243 - flux 233,236
- frequency 64, 109 -, gradient 210,221,222,228,229
- harmonic 109 - layer model 120
-, dispersion relation 64 - meridional gradient 210
- rate of change 151
p - Rossbywave 156,166,168,169,173
pressure 2,21,112
period 246 - distribution 103
-, definition 4 - disturbance 21
perturbation 34,72,92,94, 108, 109,120,183 - equation 112
- density 62,223,224 - field 34,37,113,143
- energy 93,94 - atmospheric 121
-, equation 221 - basic state 92
- initial spatial 56 - horizontal structure 186
- linear 221 - initial 113
- pressure 94, 224 -, scale 184
-, initial 113 - fluctuation 21
- steady 96 - geostrophic 113
- velocity 217,226 - gradient
-, wave energy 93 -, vertical 119,120,159
phase -, horizontal 159, 185
- constant 5, 6, 67
- decrease 4, 12 Q
- increase 4, 7, 11
- orientation 103 quantum mechanical
- scalar, gradient 10 -, oscillator 199
- speed 5,13,17,30,52,65,133,171 -, uncertainly principle 56
-, definition 5 Quasi-Biennial Oscillation 193
-, direction 170 quasi-geostrophic approximation 183
-, vertical 74 quasi-geostrophic potential vorticity equation
- stationary 46,47 206,209-211,221,246
-, condition 49 quasi-geostrophy 159
258 Index
R shear 94
- horizontal 221, 222
radiation - mean 222
- condition 91,100,102,103,106 -, vertical 221,225,226
- incoming 102 signal speed 2
- outgoing 101,102 simplification 119
- outward 102 singularity, stationary phase 54
ray sound, speed 21, 241
- equation 78, 79,240 -, water 28
-, path 79 specific entropy 240
reflection 67,69,72,179,243,247 sphericity factor 163
- of internal gravity wave 68 spin-down time 182
- at a steep slope 71 state
- at solid boundary 68 - basic 221
- in a open wedge 72 - geostrophic 123, l30
-, in a shallow wedge 71 - geostrophically balanced 129
- of Rossby wave 117, 247, 248 - mean 231
-, kinematics 72 -, of rest 59
relation steady
- dispersion 83 - circulation theory 180
- eigenvalue 83 - density perturbation 113
- Lagrangian, linearized 93 - flow 91
- thermal wind 217 - perturbation 96
resonance 242 - problem 100, 103
Reynolds stress 93-96,99,222,233,234 - solution 244,245
Rossby 149 -, state 100, 126
- adjustment problem 123 Stokes theorem 11
- deformation radius 125 strain tensor 19
-, mode 195 stratification 1, 79
-, equatorial 201 -, vertical 60
- number 160,161,163,166,168,171,205, stream function 69,97,173,178,218,235,246
234 -, geostrophic 174,206,207,222,223
-, velocity 175 streamline 99
-, wave 18,149,152,153,156,159,171,174, stress tensor 19
177,191,194,204,208,222,229,246,247 structure, vertical 183, 190, 191
- baroclinic 207 Sturm-Liouville eigenvalue problem 208
- barotropic 239 surface
- dispersion relation 155,156 - elevation, free 33
- energy 173 - gravity wave 20,60
- equation 16 - frequency 28
- frequency 154 - group velocity 28
- group velocity 157 - phase speed 28
- maximum frequency 156,203 - variables 24
- minimum period 156 -, vertical scale 60
- phase speed 155 - height field as function of time 51
- quasi-geostrophic 169,221 - isopycnal 73
- reflection 117,247,248 - pressure 241
- speed 201 - sloping 212
- spin-down 180 - wave
-, total energy 174 -, frequency 81
Rossby-gravity wave, mixed 202 -, phase speed 81
rotation 1,20,91,107,108,110,119, l31, l37, Sverdrup
152, 153 -, dynamics 180
-, large 140 -, flow, eastward 180
-, planetary 241 synoptic scale eddy 217
rotation mode 140
T
5
temperature, potential 19
scalar phase, gradient 10 thermocline 2, 119
Schroedinger equation 199 thickness, gradient 221
shallow water wave l33 tide 20
Index 259
time scale 107 - equation 169,181,208,212,231,246

-, advective 160 - flux 233,236
topography 99 - gradient 210,221,222,228,229
-, force on 102 - layer model 120
-, potential vorticity 162, 169 - meridional gradient 210
trajectory 39,40, 119,225,240 -, rate of change 151
-, Lagrangian equation 39 -, relative 109,111,163,165,167,205,245
transformation, adiabatic 241 -, gradient 169,221
transport variables 152
trapping w
-, vertical scale 209
-, zone 194 wave
turning point 79 - acoustic 9,240
- amplitude 2,92
v -
-
concept 1
crest 3
variable - conservation 13
-, dimensional 163 - force diagram 63
-, dynamical 34 - number 6, 13
-, non-dimensional 161 -, tilt 101
velocity - cyclone 1
- advective 212 - deepwater 27
- field 23,39,112,125, 173 - definition 1
- fluid element 68 - depth of influence 35
- geostrophic 164, 170,206,232 - dispersive 52
-, zonal 126 - effect on mean field 231
- horizontal 34,119,120,184,205 - energy 178, 222
-, divergence 134 - change 37
-, geostrophic 113 - downward 102
-, vector 34 - flux 37,95
initial 124 - propagation equation 36
- meridional -, source 218
-, residual 235 - equation, disturbance 122
- perturbation 217 - equatorial 193,204
- potential 23,33,36,39 -, dispersion diagram 202
-, equation 41 -, eigenfunctions 200
-, solution 43 - field 33
- profile 127 - energy equation 72
- residual 234-236 - first order description 2
-, mean meridional 234 -, incident 68
-, mean vertical 234 -, oscillating 112
- vector 135 -, pressure 99
- vertical 35,73,120,129,184,206 - free 26,33
- component 34 - frequency 2
- Laplace equation 62 - function, equation 16
- residual 234 - generation, theory 20
-, scale 61 - gravity 1,9
- zonal 126 - guide 194
viscosity 19,20,182 - in ocean 20
-, second 19 - incident 68,69, 179
vortex line 109 -, energy flux 179
vorticity 23,96,108,109,122,243 - interaction 1
- budget 205 - internal 59,79
- components 243 -, normal mode problem 243
- cyclonic 180 -, structure 59
- equation 62, 122, 151, 205 - Kelvin 133,136,142,144,148,152,154,
- planetary 163, 205 191,195,203,245
-, gradient 169,210 -, equatorial 195-197
- potential 107,110-113,122,124,127, - kinematics 2
129,133,166,210,229,245 - light, equation 28
-, ambient 169, 173 - linear theory 91
-, conservation 112,113,122 - local energy 174
260 Index
- low frequency 180 - slowly varying 9

- mean flow 209,232 -,small amplitude 2
- mode 195 - solution 148
- motion 159 - sound, equation 28
- nature of 133 - spectrum, initial 53
- nondispersive 27, 133 - stable 218,219
-, equation 133 - standard equation 28
-, solution 29 - surface gravity 20, 26, 79
- number 33,47,50,52, 133,240 - frequency 28
- constant 50 - group velocity 28
- generalized definition 50 - phase speed 28
- incident 178 -, variables 24
- non-dimensional 82 -, synoptic scale 216
- propagation 50 - theory 1
- refelcted 178 -, basic 1
- spectrum 53 - topographic 208
- vector 247 -, bottom trapped 212
- on a string 28 - train 50, 54
- packet 16,17,56,68 - unstable 218,219
-, energy, averaged 39 - variable phase, definition 2
-, Fourier amplitude 57 - vector 2,3,4,66,67,105,170,171,176
-, one-dimensional 17 - conservation 69
- pattern 6 - direction 4,175
- period 4,9 - frequency 7
- plane 2,3,5,9 - incident 71
-, three-dimensional 26 - magnitude 3
-, wavelength 4 - orientation 63, 65
- Poincare 115,133,139,141,144,148, - rate of change 14
152-154,159,194,201,204,245 - reflection 69,71
-, equatorial 192-195 -, relation to frequency 7
- potential vorticity flux 232 - velocity of element 2
- pressure 113 -, wavelength 3, 9
-, field 34 wavelength 4,34,52,60
- propagation, vertical 91 -, definition 4
- properties, radiative 91 wedge of instability 225
- radiation 96,99, 103 western boundary current 180
- reflected 179 wind
-, energy flux 179 - thermal
- reflection 69 -, equation 21 0
- Rossby 16, 18 -, relation 217,235
-, equatorial 194-196,201 -, stress 195
- self-excited 218,227 WKB Theory 75
- shallow water 27,133
- signal y
-, propagation 41
-, speed 2 Yanai wave 202-204
he first to know
with the new online notification service
SpringerAlert
You decide how we keep you up to date on new publications:
• Select a special ist field within a subject area
• Take your pick from various information formats
• Choose how often you'd like to be informed
And receive customised information to suit your needs
Register
nOW
1I".l~I"' w.r;lt,.de/alert
..
and then you are one click away

from a world of geoscience information!
Come and visit Springer's Geoscience Online
Library
Springer
Lecture 21
Wave-Mean Flow Interaction
We have been considering the dynamics of waves in this course and have remarked
several times on the linearization restriction we have normally placed on the dynam-
ics to make progress, and we have skirted rather completely the role of nonlinearity
on the dynamics of the waves themselves. It is a difficult subject.
At the same time, even small amplitude waves, for which linear theory might be a
good first approximation, can have an effect on the mean state of the medium through
which the waves are propagating. If the waves have small amplitude, we would antici-
pate that since the fluxes of momentum and density by the waves are of O( amplitude)2,
the effect on the mean will be similarly small. That doesn't mean that alteration is
unimportant or uninteresting, and the calculation of that change often can give in-
sight into how the waves can have an effect on the mean fields in which they are em-
bedded. The role of waves in altering the mean is clearly of importance in questions
concerning the general circulation or even current systems of smaller scales, e.g.,
coastal currents.
How then can we calculate the effect of waves on the mean field? This, too, is a very
complicated subject, and in this lecture we will just touch on a special case but one
which is both revealing and often used as a model for more general situations. We will
consider the effects of waves on the mean for low-frequency large-scale motions gov-
erned by quasi-geostrophic dynamics. Even with these restrictions, the issue is com-
plicated, and we will simplify further by considering mean states that correspond to
zonal flows that are functions of y, z and t but that are independent of x. The waves, of
course, will be functions of all three spatial variables. We will define the mean by the
spatial average
x
p= ~ Jpdx (21.1)
x~= -x
by an average in x where p is any dependent variable. It would appear that this defini-
tion is more suitable for atmospheric flows, and it is certainly true that the discussion
that follows came first from the meteorological literature, but one can imagine strong
currents that are nearly zonal such as the Gulf Stream after separation, the equatorial
currents, etc. for which this is at least a sensible first approach.
The governing equation for the full system of waves and the mean flow is the po-
tential vorticity equation, which in quasi-geostrophy is

232 Lecture 21 . Wave-Mean Flow Interaction
oq + u oq + v oq = 0 (2 1.2 a)
ot OX oy
U=-If/y (21.2b)
v =If/x (21.2c)
q=V21f/+~(fi OIf/)+j3y (21.2d)

OZ N 2 OZ
We will assume that the wave field is periodic in x with no mean so that the x-aver-
age of any variable associated with the wave field will be zero. Furthermore, the x-aver-
age of the geostrophic meridional velocity must itself be zero, if the flow is periodic
in x or independent of x.
We can therefore write all variables as a mean plus a wave part:
P=P+P' (21.3a)
p'=o (21.3b)
Noting first that the pv equation can be written
oq + ouq + ovq = 0 (21.4)

ot ox oy
since the geostrophic flow has zero horizontal divergence, an x-average of the
v
pv equation yields, using = 0,
oq __ ~(Vlql) (21.5)
at'- oy
The rate of change of the mean potential vorticity is given entirely in terms of the
potential vorticity flux of the waves. Using the above equation, assuming the pv flux is
known, one can simply calculate Ii. Since the variable part of Ii satisfies the relation
with the stream function
- 0 (fi OW) (21.6)

q=lf/yy+oz N2 0z
it is with appropriate boundary conditions possible to invert to find li/, once Ii is known.
In this sense, the problem of wave mean flow interaction is straightforward. First, one
calculates the linear wave field. Then, one finds the average flux of pv in the wave field.
Its meridional divergence yields the change in the mean pv. Once computed, that, with
the inversion of the elliptic problem for stream function in terms of q, completes the
specification of the change in the mean. Note that the forcing of that change is due
entirely to the pv flux in the waves.
Lecture 21 . Wave-Mean Flow Interaction 233
While the problem is formally complete, the approach leaves two important issues
unclear. First, what can we say, a priori, about the pv flux in the wave field? Second, is
there a more direct and physically intuitive way we can understand how the waves al-
ter the mean momentum and density distribution other than the inversion of the above
n
equation relating stream function and potential vorticity?
Let's take up the second question first. We note that
y' q' +"+[;:

= vi, ?'= Y',-U', (21.7a)
v't"'=v'v'
~ x -v'u' y =-v'u' y =-(v'u') y (21.7b)
where we have repeatedly used the fact that p;, = 0 for any variable P. With the geo-
strophic and hydrostatic approximations, this allows the pv wave flux to be written
v'q'=-(~'u')y -(v'p' gfo 2 )

N
(21.8)
Po z
The potential vorticity flux is therefore the divergence in the y-z-plane of the
vector
j3=-u'v'j-v'p'~k (21.9)
PON 2
where j and k are unit vectors in the y- and z-directions, respectively. The vector j3 is
the Eliassen and Palm (EP) flux vector. Its horizontal component is the meridional
wave flux of zonal momentum per unit mass, or equivalently the Reynolds stress, while
its vertical component is, aside from a factor, the horizontal meridional density flux by
the waves. The direction of j3 in the y-z-plane gives us an immediate sense of whether
the meridional pv flux is due to momentum or density fluxes. For example, in the Eady
problem, the resulting unstable waves would have a purely vertical EP vector.
Note that the mean pv equation is simply
(&=_ av .j3 (21.10)

at ay
This is not apparently a great advance over our previous formulation, but it again
emphasizes the point that the change in the mean fields will be due entirely (with the
appropriate analysis of boundary effects skipped over here) to the divergence of the
EP vector.
This somewhat new formulation is of great assistance when we examine the x-av-
erage of the momentum equation itself. If we take the x-average of the x-momentum
equation and remember that the x-average of the geostrophic v is zero, we obtain
ali _!ov=-(u'v')y (21.11)

at
Note that on the left-hand side of the equation, there is an x-averaged meridional
velocity. From our earlier scaling exercise, we recognize that this in an order Rossby
number velocity that remains in the equation, because it is multiplied by the relatively
large Coriolis parameter and is thus of the same order as the (weak) acceleration of
the 0(1) geostrophic velocity. The adiabatic equation when x-averaged is
ap _wPON2 =_(~'p')y (21.12)

at g
A superficial glance at these equations appears to suggest that the mean zonal
momentum is actually only altered by the Reynolds stress provided by the wave field,
while the change in the mean density is associated with the x-averaged wave flux of
perturbation density. If this were the case, it would not be consistent with our earlier
view that it is the wave pv flux that is responsible for the change of all quantities in
the mean state. How can we resolve that apparent discrepancy?
It is important to note that the eddy fluxes as written drive not just the mean u and p
but also the x-averaged v and w, i.e., the mean circulation in the y-z-plane. Indeed, in
principle it is possible that the wave fluxes might produce a balancing meridional cir-
culation with no change in the mean zonal velocity and density. There are problems
where that is the case, and we shall shortly see how we can predict that. So, the above
formulation is not quite a precise enough picture.
We can attempt to deal with the possibility mentioned above by splitting the mean
vertical velocity into a part that may be balanced by the wave flux of density plus a
residual circulation, which we will indicate with an asterisk, i.e., we write
w=~(~'P'~+w* (21.13)
ay 2 PON
and we define w* as the residual mean vertical velocity (residual in the sense that
it is the mean vertical velocity after having accounted for what may be the purely wave
driven part. In terms of which, the adiabatic equation becomes
ap- -w*~=O
N 2
(21.14)
at g
In this formulation, the change of the mean density field is due entirely to the ver-
tical velocity in the residual meridional circulation. If there had been a non-adiabatic
source term for density on the right-hand side of the density equation, it would be the
residual vertical velocity that would balance that heating or cooling term in the steady
state for the mean. Experience has shown that in the presence of the time varying wave
fields, it is the residual velocities that most closely resemble the Lagrangian pathways
of the fluid in the meridional plane.
We would also like to define a residual mean meridional velocity, and here we
have to be a bit careful. When the x-average of the continuity equation is taken, we
have, in terms of original variables,
ily +wz =0 (21.15)

That is, the mean circulation is nondivergent in the y-z-plane. If we are substitut-
ing the residual circulation velocities v* and w* for the x-averages of v and w, we would
like to make sure that v* and w* also satisfy the same divergence-free condition. That
suggests defining
v =-(::~ Jz +v* (21.16)
for then it follows that
v*y +w*z =0 (21.17)
If this definition for v* is used in the x-averaged momentum equation,
au -fiv*=-(u'v') -(gp'v'J =v·p (21.18)

at 0 y 11
/"0
N2
z
Thus, now the forcing term due to the wave flux in the x-momentum equation is
simply the divergence of the EP vector, or as we have seen, the wave pv flux. This is a
promising advance, since we anticipate that the changes in the mean fields are given
entirely in terms of the pv flux. Note, however, that the divergence of the EP vector
drives not only the time derivative of the mean zonal velocity but also the mean re-
sidual meridional velocity. How can we sort out one from the other?
The mean density equation, as derived above is
a- N2
~-w*~=O (21.19)
at g
Let's take advantage of the thermal wind relation as applied to the mean flow, i.e.,
.(- gp
JOu z = y (21.20)
Po
Take the z-derivative of the mean momentum equation and the y-derivative of the
mean adiabatic equation to obtain
2
- fo v *z +w *y N
2 a V· p-
= fo az (21.21)
Since the residual velocities are nondivergent in the y-z-plane, they can be written
in terms of a stream function:
v*=-Xz' w*=Xy (21.22)
which automatically satisfies the continuity equation for the residual velocities. This
in turn leads to the elliptic problem for the stream function:
N2 1 a -
Xyy fi +Xzz= fo az V·p (21.23)
Therefore, if the EP vector is known, this elliptic problem (with appropriate bound-
ary conditions) can be inverted to find X(y,z) and the residual velocities. Note again
that it is determined entirely in terms of the divergence of the EP vector. Once the re-
sidual circulation is known, the x-momentum equation yields the change in the mean
zonal momentum and the mean density. Of course, this inversion is no simpler than
the inversion of the original x-averaged pv equation:
aq =-~(v'q') (21.24a)
at ay
with
a(fi alfiJ
q = lfiyy + az l N Z az (21.24b)
which can be obtained by taking the y-derivative of the x-momentum equation and the
z-derivative of the density equation, which is nothing more than a re-derivation for the
mean fields of the potential vorticity equation. This more indirect approach has the
conceptual advantage of showing in detail how the mean field changes as a consequence
of the wave fluxes. It does not change, indeed it emphasizes, the fact that the change
comes about only due to fluxes by the waves of potential vorticity. What then, return-
ing to our first question, can we say a priori about the wave flux of potential vorticity?
If the wave amplitude is small so that the waves satisfy linear pv dynamics, we could
suppose that we would calculate the wave field from the linear equation:
a ' u -.!L+
-.!L+ a' v'-aq =Diss(q') (21.25)
at ax ay
Here I have added on the right-hand side of the equation a dissipation term, of arbi-
trary form, for potential vorticity assuming only that it is linear in q' and has zero x-aver-
age. We shall shortly see why this might be an interesting addition to the dynamics.
To find the meridional pv flux, we multiply the above perturbation equation by q'
and average in x to obtain
,-,aq
ay
v q -=q'D'lSS (q')- aq,z/2
_.
at
(21.26)
The potential vorticity flux, when x-averaged, is therefore proportional to the av-
erage increase with time of the variance of the wave pv and to the correlation of the
pv with its own dissipation. For steady, inviscid waves, both terms will be zero and the
wave pv flux will vanish. In this case, it follows immediately that there will be no change
in the mean zonal velocity or density fields due to the waves. The mean residual circu-
lation will be zero. There can be a mean Eulerian circulation
- (gV'p'J
V=- 2 and (21.27)
poN z
d (-,,)
W=-Vp g
--2 (21.28)
dy PON
so that the wave fluxes of momentum and density yield only a balancing Eulerian v
and w but no change in the zonal velocity or its supporting density field. We saw some-
thing like this when we looked at the steady internal wave field radiated by the inter-
action of a current with a rippling topography, and we argued that the only change in
the current would occur at the front of the radiating wave field where the time depen-
dence of the wave envelope would be strong. We see here a similar situation for geo-
strophic flow. This was first noticed, with some expression of amazement, by Charney
and Drazin (1961) in their pioneering paper on the propagation of planetary waves
from the troposphere into the upper atmosphere. They carefully calculated the wave
field and its effects on the mean field and found the effect was zero. Since that time, a
good deal of effort has gone into sharpening the theory to describe in detail the role
of dissipation and time dependence in describing how the waves can affect the mean.
A good example of this is found in a very nice paper by Edmond, Hoskins, and McIntyre
(1980). The resulting theory is by now rather vast, and further discussion is beyond
the scope of this course.
Further efforts to develop the theory for more oceanographically pertinent situa-
tions attempt to replace the zonal average (not terribly apt for the ocean) with a time
average. The resulting equations are complex, and it is still hard to see clear concep-
tual progress.

Dr. Joseph Pedlosky (Auth.) - Waves in The Ocean and Atmosphere - Introduction To Wave Dynamics-Springer-Verlag Berlin Heidelberg (2003) PDF

Uploaded by

Copyright:

Available Formats

Dr. Joseph Pedlosky (Auth.) - Waves in The Ocean and Atmosphere - Introduction To Wave Dynamics-Springer-Verlag Berlin Heidelberg (2003) PDF

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

Dr. Joseph Pedlosky (Auth.) - Waves in The Ocean and Atmosphere - Introduction To Wave Dynamics-Springer-Verlag Berlin Heidelberg (2003) PDF

Uploaded by

Copyright:

Available Formats

Waves in the Ocean and Atmosphere

Springer-Verlag Berlin Heidelberg GmbH

Dr. Joseph Pedlosky

Library of Congress Cataloging-in-Publication Data Applied For

Bibliographic information published by Die Deutsche Bibliothek

Cover Design: Erich Kirchner, Heidelberg

Printed on acid-free paper - 32/3140 - 5432 10

2 Kinematic Generalization ........................................ .................... 9

3 Equations of Motion; Surface Gravity Waves .................................... 19

4 Fields of Motion in Gravity Waves and Energy .................................. 33

5 The Initial Value Problem ............................................................ 41

6 Discussion of Initial Value Problem (Continued) ............................... 53

7 Internal Gravity Waves .............................................. . ................ 59

8 Internal Waves, Group Velocity and Reflection ................................. 67

9 WKB Theory for Internal Gravity Waves ......................................... 75

10 Vertical Propagation of Waves:

11 Rotation and Potential Vorticity .................................................. 107

12 Large-Scale Hydrostatic Motions ...................................... 119

13 Shallow Water Waves in a Rotating Fluid;

14 Rossby Waves ............ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 149

15 Rossby Waves (Continued), Quasi-Geostrophy ............................... 159

16 Energy and Energy Flux in Rossby Waves ...................................... 173

18 Equatorial Beta-Plane and Equatorial Waves ... .............................. 193

19 Stratified Quasi-Geostrophic Motion and Instability Waves .............. 205

20 Energy Equation and Necessary Conditions for Instability ............... 221

21 Wave-Mean Flow Interaction ..................................................... 231

Problems ............................................................................... 239

References ............................................................................. 249

Index ..................................................................................... 253

J. Pedlosky, Waves in the Ocean and Atmosphere

• u = the characteristic velocity of the fluid element in the wave, and

l{J(x,t) = l{J(Xi,t) = ReAei(Rox-rot) (1.2)

• Re implies that the real part of the following expression is taken.

We can define the variable phase of the wave eas

kiXi - rot = constant ( 1.5)

In two dimensions, for example, these will be the lines

R· x - wt = kjxj + kzxz - wt == kx + ly - wt (1.6)

~=~(k.x.-wt)=k. aXj =k-c5.. =k. (1.8)

i.e., the magnitude of the wave vector. Then

What is the speed of movement of the line of constant phase

O=kjXj -OJt=Ks-OJt (1.14)

{At constant e, de = 0 = Kds - mdt, so that ds / dt = w/ K}

d8= 0 = kdx - wdt or (1.l6a)

Thus also, the increase in phase along the wave vector is

---+ I\, .---

J. Pedlosky, Waves in the Ocean and Atmosphere

VrfJ =AeiOVe +ol L: J (2.3)

We define (guided by our experience with the plane wave):

K= ve local spatial increase of phase (2.4a)

-W= aB local increase of phase with time (2.4b)

Now consider the same increase calculated on curve C2 :

The difference between them is

it follows by definition that

anAB = m(A) _ m(B)

m(A) = rate of decrease of phase at point A (see Fig. 2.4)

Phase increasing in space

Fig. 2.4. The movement of the phase through the interval AB

We may think of this as a statement of the conservation of wave crests. Namely,

0)= Q(kpXi,t) (2.12)