14 CHAPTER 1.

BASIC CONCEPTS

1.3 Surface waves

1.3.1 Solution for a single wave component in deep water

A plane wave can generally be described as a function

v = f (θ)

where θ = k · x − ωt + δ is a phase function. Here k is the wave number vector, ω

is the circular wave frequency (radians per second) and δ is a constant phase shift
(radians). The wave number vector defines the direction of wave propagation, and
the ratio between the wave frequency and wave number, that is c ≡ ω/|k| defines
the propagation speed, or phase speed of the wave.
The equations that describe surface wave motion can be derived in different ways.
We will here follow Gill (1982) and focus on the pressure perturbation in the
waves. We define this pressure perturbation as p̃, and take the pressure to be
p = p0 − ρgz + p̃. From the continuity equation we have

∇ · v = 0, (1.3.1)

and from the linearized equations, ignoring friction and the Coriolis force (we
assume that rotational effects are negligible for the wave motion since the wave
periods are so small, but, as we shall see later on, this is not quite correct for the
wave-induced drift) we have
∂v
ρ = −∇p̃. (1.3.2)
∂t
Taking the time derivative of (1.3.1) and combining it with (1.3.2), we find that
the perturbation pressure is governed by

∇2 p̃ = 0, (1.3.3)

hence p̃ is a harmonic function.

We are primarily interested in surface gravity waves in deep water and take the
water depth to be infinite. In practice we only need to require that the ratio be-
tween the wavelength λ and the local water depth H is such that λ/H � 1, but
the above assumption simplifies the analysis. Thus, as z → −∞, we require
that the pressure perturbation and the velocities vanish. At the surface z = η,
1.3. SURFACE WAVES 15

we have one kinematic boundary condition and one dynamic boundary condition.
The kinematic boundary condition simply expresses that the vertical velocity at
the surface is the same as the vertical velocity of the surface itself. To leading
order this becomes
∂η
w= , z = 0. (1.3.4)
∂t
The dynamic boundary condition expresses that the pressure at the surface is equal
to the atmospheric pressure p0 . Since we have p(z = η) = p0 − ρgη + p̃, we have

p̃ = ρgη, z = η. (1.3.5)

The perturbation pressure p̃ in a single wave component can be found postulating

a solution
p̃ = G(z) cos(kx + ly − ωt + δ). (1.3.6)
Inserting (1.3.6) into (1.3.3), and using the requirement that p̃ → 0 as z → −∞,
we obtain
p̃ = Aeκz cos(kx + ly − ωt + δ), (1.3.7)
√
where κ = k 2 + l2 is the wave number and A is a constant1 . We now assume
that the surface elevation is given by

η = a cos(kx + ly − ωt + δ), (1.3.8)

that is, in phase with the pressure perturbation. Here a is the amplitude of the wave
component. Combining (1.3.2), (1.3.4) and (1.3.8), we find for the integration
constant A:
aω 2
A=ρ . (1.3.9)
κ
The ratio between the wave height and the wave length is expressed by the wave
steepness, ε ≡ aκ. The propagation speed, or phase speed, of the waves is given
by c = ω/κ. Hence we can write the pressure perturbation in the waves as

p̃ = ρεc2 eκz cos(kx + ly − ωt + δ). (1.3.10)

We note that the pressure is linear in the wave steepness, quadratic in the phase
speed, and that it vanishes at a depth of O(κ−1 ), which is proportional to the
1
More general solutions can be found assuming G = G(z, t), in which case A can be a function
of time. It is also possible to allow G to be a function of all the spatial coordinates, assuming small
amplitude changes over the length and time scales given by the wave length and wave period.
16 CHAPTER 1. BASIC CONCEPTS

wavelength λ. The Stokes depth 2κ−1 is often used as a typical depth scale for
the influence of the waves. Recall, however, that we are here only considering
gravity waves in deep water. In shallow water (λ ≥ H) the wave motion extends
throughout the water column (e.g. Gill, 1982).
The next thing we need to do is to find the relation between ω and κ. Using the
dynamic boundary condition (1.3.5) we obtain

ω 2 = gκ. (1.3.11)

This is the dispersion relation that enables us to relate the wave frequency to the
wavelength. Note that (1.3.11) is a special case of the relation (1.2.14) that we
derived using dimensional analysis. For the phase speed we have
� �
g gλ
c=± =± . (1.3.12)
κ 2π

1.3.2 Mass, momentum and energy transport in surface grav-

ity waves

The energy in a wave component is the sum of the potential and kinetic energy.
The velocities can be obtained from the linearized momentum equation (1.3.2),
and the kinetic energy is
� η
ρ 2 1
Ek = {ũ } dz = ρga2 , (1.3.13)
−∞ 2 4
where we have used the dispersion relation (1.3.11). Setting the reference level at
the undisturbed surface z = 0, the potential energy is given by
� η
1
Ep = ρgz dz = ρga2 . (1.3.14)
0 4
We see that the kinetic and potential energies are equal. This is a general property
of surface waves and is usually referred to as the “equipartition principle”. The
total energy in the waves becomes Etot = Ek + Ep = (ρga2 )/2.
The transport of energy in the waves is a curious concept. We have already estab-
lished that a single wave component propagates with a speed c, but in reality the
sea state is the sum of many wave components with different phases, amplitudes,
1.3. SURFACE WAVES 17

periods, wave lengths and directions, which leads to some interesting results. The
classical way of introducing the energy transport in surface waves is to add two
components of nearly equal wave frequencies and wave numbers. Consider two
such wave components propagating in the positive x-direction. From standard
trigonometric identities we can write this sum as
a cos[(k − Δk)x − (ω − Δω)t) + a cos[(k + Δk)x − (ω + Δω)t) = (1.3.15)
2a cos(kx − ωt) cos(Δkx − Δωt).
The result is a wave of twice the amplitude 2a cos(kx − ωt) that is modulated by
the function cos(Δkx − Δωt). This function is in itself just like a single wave
component, but with a speed given by Δω/Δk. Since Δk and Δω are assumed
small, this wave component is much longer than both of the original components,
and with much longer period. Hence we can view the sum of the two nearly
equal wave components as one wave component with twice the amplitude and the
average wave number and wave frequency, which is enveloped by a much longer
wave. The energy in the waves are contained within this envelope, or wave group,
and hence advected with the propagation speed Δω/Δk. We now use the limits
Δk → 0 and Δω → 0 and define the group velocity
∂ω
cg = . (1.3.16)
∂κ
For deep water waves it is easy to show from (1.3.11) that cg = c/2. As the waves
propagate into shallower water, the ratio cg /c increases until we have cg = c in
the limit kH → 0 when the wave length is much larger than the local depth.
Static figures are not particularly effective for demonstrating how individual waves
propagate with the phase speed c, while at the same time the whole wave group
travels at a slower speed cg . There are some excellent animations on the Internet,
however, see for example the Wikipedia entry for the group velocity2 .
The trajectories of fluid parcels due to the waves can, as a first approximation, be
obtained by integrating the velocities in time:
�� � � �
x̃ = (x̃, ỹ, z̃) = ũ dt, ṽ dt, ṽ dt . (1.3.17)

Inserting for the wave velocities the fluid parcels describe closed trajectories. For
deep water waves propagating in the positive x-direction:
(x̃, z̃) = ae−kz (cos(kx − ωt), sin(kx − ωt)) , (1.3.18)
2
https://en.wikipedia.org/wiki/Group_velocity
18 CHAPTER 1. BASIC CONCEPTS

indicating that the deep water wave trajectories are perfect circles with radii that
decay exponentially with depth. We will see below that this is only correct to first
order in the wave steepness.
It appears from (1.3.18) that there is no net transport of mass in the waves. Any
fluid parcel will return to its initial position after a wave period. We do know,
however, that surface waves (and many other types of waves) are associated with a
mean drift velocity. Consider any point in the fluid that is above the wave troughs,
that is z > −a. Under the wave crest, we have zero vertical velocity and maximum
horizontal velocity in the wave propagation direction. Under the wave trough, we
also have zero vertical velocity and maximum velocity opposite to the wave prop-
agation direction. Above the wave trough there is only air, however, which means
that there is no transport of water. At any specific point above the wave troughs,
the transport of water is therefore on average going in the wave propagation di-
rection, while below the wave troughs the net mass transport is zero.
The description above is based on our Eulerian description of the fluid motion.
It is more instructive, and also more relevant to oceanic transport problems, to
consider the wave-induced drift from a Lagrangian perspective. In the Lagrangian
description we label each fluid parcel and follow them in time, in contrast to the
Eulerian description in which we stay in fixed positions and note the velocities of
the fluid parcels that happen to flow by. It is clear that we do not need to consider
the somewhat awkward case "sometimes air, sometimes water" if we are using a
Lagrangian description of the motion. We are, after all, following individual fluid
parcels that maintain their material properties.
A direct Lagrangian approach has its merits, in particular when studying wave-
induced drift (Weber, 2019), but we will not use it here. We will instead use a
quasi-Lagrangian description introduced by Longuet-Higgins (1953a). The point
is to evaluate the Eulerian velocities at the positions traced by the fluid parcels.
That is, we write the Lagrangian velocity as
uL = u(x + x̃, t). (1.3.19)
The next thing we do is to make a first order Taylor expansion of the righthand-
side, such that
u(x + x̃, t) ≈ u(x, t) + x̃ · ∇u. (1.3.20)
Recall that the Eulerian velocity consists of mean, wave and turbulent parts so that
u = ū + ũ + u� . If we average over the wave cycle, we get from (1.3.20)
{u(x, t) + x̃ · ∇u} = ū + {x̃ · ∇ũ}. (1.3.21)
1.3. SURFACE WAVES 19

The latter part is called the Stokes drift velocity (Stokes, 1847), which we will
denote by ūS . For deep water waves we have

ūS = {x̃ · ∇ũ} = (ε2 c)e−2κz . (1.3.22)

The difference between the Eulerian and Lagrangian description of the Stokes drift
is illustrated in Fig. 1.2. We define the Lagrangian mean velocity ūL . Correct to
second order in the wave steepness, we now have

ūL = ū + ūS . (1.3.23)

A couple of comments are in place here. First of all, the Lagrangian mean ve-
locity is the drift velocity of any neutrally buoyant object that is in the water, and
which is sufficiently small so that inertia can be neglected. This is the case for
some types of nutrients, plankton and pollution, and approximately the case also
for many types of buoyant small objects (e.g. pelagic fish eggs or oil droplets),
although, as we will see later on, the vertical motion of such objects relative to
the water is very important for determining the horizontal mean drift because of
the high vertical shear close to the surface. Second, numerical ocean circulation
models are overwhelmingly using an Eulerian framework, solving for ū. Hence
the Lagrangian mean velocity that is so important for oceanic drift modeling will
need to be calculated using (1.3.23), which generally requires additional output
from a numerical wave prediction model for calculating ūS .
The total mean momentum in the waves is given by
� 0
Mw = ρUS = ρ ūS dz. (1.3.24)
∞

The wave field acts as a reservoir of mean momentum, and spatial and temporal
changes in the wave field is associated with momentum fluxes between the waves
and the atmosphere, and the waves and the mean oceanic flows, a subject that will
be discussed in more detail in Sec. 2.

1.3.3 Spectral representation of surface waves

By assuming that the wave field is stationary and homogeneous, ie, its statistical
properties are unchanging in time and space—clearly an unrealistic assumption—
and by furthermore assuming that the wave field decorrelates in a finite distance,
20 CHAPTER 1. BASIC CONCEPTS

Figure 1.2: The mean drift velocity in deep water surface waves with amplitude 1
m and a period of 6 s. The total transport, and hence the mean wave momentum,
is the same in both the Lagrangian and Eulerian descriptions. From Broström
et al. (2014).

we can formally employ the Fourier transform to define a power density spectrum
of a time series of the surface elevation. Less formally, and hopefully more un-
derstandably, what this means is that we can identify those waves (sinusoids to a
first approximation) that make up the wave field. In a one-dimensional sea (which
literally never occurs), we can simplify this picture further to those waves that
travel either to the left or to the right.

The variance of a monochromatic wave

Consider first a monochromatic wave whose elevation is η = a cos(kx − ωt + δ),

similar to Eq (1.3.8), but in one spatial dimension, x, only. The mean square
or variance (since the wave field is zero mean) is found by taking the temporal
average over one wave period T = 2π/ω, ie, by integrating η 2 over one wave
period T and dividing by T ,
�
1 T 2 2πt
2
{η } = a cos2 dt. (1.3.25)
T 0 T
1.3. SURFACE WAVES 21

Since we know [see e.g. Gradshteyn and Ryzhik 2007, Eq (2.513-11), p 154] that
� 2π
cos2 t dt = π (1.3.26)
0

it follows after an elementary variable substitution of u = 2πt/T which leads to

dt = T2π du
(the integral now goes from u0 = 0 to u1 = 2πT T
= 2π) that {η 2 } =
a2 /2. This demonstrates that the variance of the surface elevation is proportional
to the square of the amplitude for a monochromatic wave (a single sine wave). It
also shows that by multiplying the wave variance by ρg we get the total energy,
Etot , see Eqs (1.3.13)–(1.3.14).

Superposition of sinusoids—The Fourier series

Any discrete and evenly sampled time series η(ti ) where the measurements are
taken at times ti = (n − 1)Δt, n = 1, 2, ..., N and the length of the domain is T
can be represented exactly as a Fourier series of cosines,

N
a0 � � �
η(t) = + an cos 2πnt
T
− δ n . (1.3.27)
2 n=1

So a sum of (co)sines with different amplitudes, frequencies and phases can be

fitted to a finite set of measurements. As shown conceptually in Fig 1.3, a surpris-
ingly small number (eight here) of sine waves when superposed can create quite
complex-looking time series.

The autocovariance function

Unfortunately, an infinite time series has infinite energy and cannot be represented
as a Fourier series. To represent the variance of the different wave components
present on the sea surface, we look instead at the autocovariance of the surface.
We ask “How do water level measurements correlate with other water level mea-
surements separated by a length of time τ ?”. It seems natural that measurements
that are close in time are also strongly correlated (if you measure the peak of a
wave, you expect the next measurement a fraction of a second later to also be
22 CHAPTER 1. BASIC CONCEPTS

�������������

Figure 1.3: Eight sinusoids (blue) with different frequencies and phases δn super-
posed create a complex-looking time series (red).
1.3. SURFACE WAVES 23

high, likewise, a measurement in a trough will be followed by another measure-

ment low down). If we are not interested in the phase of the sine waves, only their
amplitudes, we can use the autocovariance function
N
�
Rηη (τ ) = η(tn )η(tn + τ ). (1.3.28)
n=1

This function will presumably die off for large τ since we don’t expect a wave
hundreds of wave periods later (or hundreds of wave lengths down-wave in space)
to be in phase with our measurement here and now. We say that waves decorrelate
in time and space. This makes the autocovariance function approach zero for large
time lags τ and its integral remain finite,
� ∞
|Rηη |2 dτ < ∞. (1.3.29)
−∞

In Fig 1.4 we see a time series of 20 minutes (and a zoom of the first 150 seconds
in Fig 1.5) sampled twice a second with a nadir-looking radar altimeter at the
Ekofisk field in the central North Sea. It is evident that the waves are irregular in
shape, but also that they have a dominant period of about 10 seconds. How will
measurements covary with other measurements taken at a later time? This is what
Eq (1.3.28) answers, and Fig 1.6 shows what the autocovariance of Fig 1.5 out to
time lags of 75 seconds is like. As expected, the autocovariance is strongest at
a time lag of zero. This simply means that measurements covary perfectly with
themselves, as of course they must! We can calculate the significant wave height
directly from the time series in Fig 1.5,
�
� N
�1 �
Hs = 4 � η(tn )2 . (1.3.30)
N n=1

This is sometimes referred to as H4rms because it is four times the standard devia-
tion (or root-mean-square of a zero-mean process). This shows that the significant
wave height is “4σ”, which is quite far from the “average” wave height. This is
easy to see in Fig 1.5 where the distance between the red lines is the significant
wave height. In fact, we can read off the variance of the time series at time lag
zero in Fig 1.6. Knowing that the significant wave height is four standard devia-
tions of the water√level, and seeing that the variance is approximately 1.67 m2 , we
find that Hs ≈ 4 1.67 ≈ 5.2 m in this case.
24 CHAPTER 1. BASIC CONCEPTS

��������������������������������������������������
���������������������������

��

��

��

��

� ��� ��� ��� ��� ���� ����

�
���������������������

�
� ��� ��� ��� ��� ���� ����
��������

Figure 1.4: A 2 Hz 20-minute time series of the water elevation as measured by a

SAAB radar altimeter from approximately 30.5 m height. In grey (upper panel) is
shown the distance to the sea surface (what the radar altimeter measures), and in
blue (lower panel) the elevation above a mean sea level (the inverse of the zero-
mean time series). Also shown as two red lines is the significant wave height,
Hs .
1.3. SURFACE WAVES 25

������������������������� ������������������������
� �

� �

� �
���������������������

� �

� �

� �

� �

� �

� �
� �� ��� ��� � ��� ��� ��� ���
�������� ������������

Figure 1.5: A closer look at the first 150 seconds (left) and a histogram of the
water elevation measurements throughout the 20 minutes depicted in Fig 1.5. It is
evident from the histogram that Hs envelops a very high portion (about 95%) of
the measurements, as expected as Hs = 4 std(η), ie, the red lines are located as
±2 std(η).
26 CHAPTER 1. BASIC CONCEPTS

��������������������������������
���

���
�������������������

���

���

���

���

���

� �� �� �� �� �� �� ��
������������

Figure 1.6: The autocovariance function of the time series in Fig 1.5.
1.3. SURFACE WAVES 27

More to the point for our discussion of waves of different periods is the observa-
tion that the first “trough” in our autocovariance function occurs at a time lag of
about 5 seconds. This means that the strongest negative correlation between mea-
surements is found at a separation of 5 seconds. This means that measurements
separated by this time lag disagree the most, which is what we expect after half a
wave period since a measurement at a wave crest is the opposite (high) of a wave
trough. Moving on to the first peak of our autocovariance function we luckily see
that it occurs after a time lag of about 10 seconds. This is precisely what we would
expect following our visual inspection of Fig 1.5. The measurements that are sep-
arated by about 10 seconds are mostly (but not perfectly—note the irregular shape
of the waves in Fig 1.5) in phase.
Now imagine that we take the integral of the square of the autocovariance func-
tion in Fig 1.6. It seems to be trailing off for large time lags. In fact, it will
approach zero, although not perfectly. However, if we accept that the remaining
weak covariance at long time lags is just the product of random coincidence, we
can assume that the true covariance of the wave field decays to zero. This means
that we have a function which, like we anticipated in Eq (1.3.29), is bounded, ie,
not infinite when integrated to infinite time lags. Such a function has a frequency-
domain counterpart which can be found using the Fourier transform,
� ∞
G(f ) = Re−i2πf τ dτ. (1.3.31)
−∞

We have now dropped the subscripts on R for brevity. This integral is generally
complex, but is real for even (symmetric) functions. Conveniently, the autoco-
variance R(τ ) of a time series is even, which can most easily be understood by
imagining that we correlate observations that are separated by a negative time lag.
Clearly it shouldn’t matter whether we compare the first N − k measurements
1, 2, ..., N − k with the last N − k measurements, k + 1, k + 2, ..., N or vice versa.
This proves that a negative time lag τ = −kΔt should give the same result as the
positive time lag kΔt and the autocovariance function must be symmetric about
zero time lag, ie, even.

The variance density spectrum

The quantity G(f ) in Eq (1.3.31) is known as the power spectral density, or in our
parlance, the variance density spectrum. We are now ready to calculate it from
our time series in Fig 1.5.
28 CHAPTER 1. BASIC CONCEPTS

First, the discrete Fourier transform of the time series is constructed,

� N −1
Δt �
η̂(fk ) = η(tn ) exp (2πikn/N ) (1.3.32)
N n=0

The Fourier coefficients are generally complex, and there are as many frequencies
as there are points in the time series,but half of them are negative, −fk = f−k , k =
0, 1, ..., N/2. To get to the power spectral density, which is what is needed for the
variance density spectrum, we must now find the squared modulus of the positive
and negative frequencies, which is a real number,

F (fk ) = |η̂(fk )|2 + |η̂(−fk )|2 . (1.3.33)

This is called the periodogram method and the variance density spectrum of our
time series in Fig 1.5 is shown in Fig 1.7. The variance density spectrum shows
us how the variance is distributed over frequencies. To better understand what is
meant by this, let’s return to 1.3.25 By integrating the spectrum we get the total
variance of the time series,
� ∞
E= F (f )df. (1.3.34)
0

This should then equal var(η). This is formally known as Parseval’s theorem
which here means that the total wave energy of the spectrum is the same as the
average wave energy of the time series.
Let us now look at how the variance density is related to the variance of a single
sine wave (1.3.25). If our spectrum consisted of this one frequency band, the
variance density spectrum is simply
1
Fk = a2k /Δfk , (1.3.35)
2
and since there is only one frequency, k = 1. As we add more frequency bands,
the superposition principle tells us that they can be added to each other without
any further ado and no interaction will take place (we’ll see later that this is a half-
truth). Since each frequency band accounts for a certain amount of the variance,
it must mean that the sum of the variance of all the frequency bands must equal
the total variance,
N N/2
1� 2 �
2
{η } = a = Fk Δfk . (1.3.36)
2 k=0 k k=0
1.3. SURFACE WAVES 29

���������������������������������������������
��
��������������

�� ��������������
������������������������

��

��

��

��

�
��� ��� ��� ��� ��� ���
��������������

Figure 1.7: The variance density spectrum of the time series in Fig 1.5.
30 CHAPTER 1. BASIC CONCEPTS

As Δf → 0, the summation (1.3.36) approaches the integral

� ∞
2
{η } = F (f ) df. (1.3.37)
0

Spectral moments and integrated parameters

The second method for computing the wave height is the spectral estimate
√
Hm0 = 4 E. (1.3.38)

The reason it is given this strange subscript is that the spectral “moment” of order
n is defined as � ∞
mn ≡ f n F (f ) df. (1.3.39)
0
The zeroth moment m0 is then simply the integral E of the variance density spec-
trum (1.3.37), which we know from Eq (1.3.36) is the total wave variance. Its
square root multiplied by 4 is then “4 rms”.

Wave energy balance equation

The spectral representation and the superposition principle allows us to treat the
water surface as a sum of (weakly interacting, more later) Fourier components,
each advancing with a phase speed given by the dispersion relation (1.3.11) and
whose energy advances with the group speed (1.3.16). This means that we can see
the wave energy balance as an advection equation,
∂F
+ ∇ · (cg F ) = Sin + Sds + Snl . (1.3.40)
∂t

Source terms

This equation is immensely important in wave modelling. It is at the heart of every

spectral wave model. The variance (often referred to as energy) density is a func-
tion F = F (x, y, t, f, θ). So it varies with location, time, wave frequency f and
direction of propagation θ. Note also that although we refer to it as the energy bal-
ance equation, the energy density is strictly speaking ρgF . To see this, start with
1.3. SURFACE WAVES 31

a single sine wave and add the potential and kinetic energy, Eqs (1.3.13)+(1.3.14)
= ρga2 /2 = ρgF Δf . The latter is known from Eq (1.3.35).
Temporarily setting the S terms on the RHS to zero (ie, no wind input, no wave
breaking and no nonlinear interaction between the wave components), we see that
the left hand side is simply an advection balance, ie, the amount of wave energy
that comes rolling in on the group velocity minus the amount the goes out (this
is what the divergence operator calculates) must equal the local rate of change of
wave energy density. In other words, if more energy enters than leaves, the energy
level must go up. The right hand side is where the physics lies. The source terms
represent wind input (in), dissipation through wave breaking (white capping) and
other processes (ds) and nonlinear interaction between Fourier components. This
is where the weak interaction comes in. If surface waves were perfectly linear
this term would not exist. Then various Fourier components would travel across
the ocean, completely oblivious of other wave components. This is not the case,
however, and waves do interact, albeit very weakly. This weak interaction is all
the same enough to cause wave energy to “propagate” in the spectrum toward
lower and higher frequencies, away from the central area (the peak frequency)
where the wind feeds energy into the wave field (parameterized here through Sin ).

Wave action

If the wave is riding on a current, the Doppler effect will lead an observer standing
on the shore to see a different frequency ω = σ + k · u. Here, σ = 2πf is the in-
trinsic circular frequency which an observer riding on the current would observe.
This invalidates Eq (1.3.40). Wave energy is not conserved in the presence of a
slowly changing current because energy is transferred between the wave field and
the mean current. Fortunately, the wave action density

N ≡ F/σ (1.3.41)

does what the wave energy density cannot: it is conserved in the presence of
currents,
∂N
+ ∇ · ((cg + u)N ) = 0 (1.3.42)
∂t
(again in the absence of source terms). The wave action has a number of interest-
ing properties and is reminiscent of the Planck constant in quantum physics. The
energy density is proportional to F = σN —this can be seen from the definition
32 CHAPTER 1. BASIC CONCEPTS

(1.3.41). We now make a statement which we will prove over the following two
sections: The wave pseudo-momentum density is proportional to
F/c = F k/ω = N k. (1.3.43)
Here we have reverted to using ω because we have zero currents, in which case
σ = ω.
The action balance equation Eq (1.3.42) with source terms is written
∂N �
+ ∇ · ((cg + u)N ) = Si /ω. (1.3.44)
∂t i

If we now multiply Eq (1.3.44) by ρgk and integrate it we get

� ∞ � ∞ �� ∞ k
∂(N k)
ρg dω + ρg ∇ · ((cg + u)N k) dω = ρg Si dω.
0 ∂t 0 i 0 ω
(1.3.45)
Here, the i sums over the different source terms.
The relation between the wave momentum as defined through the Stokes drift in
(1.3.2) and our definition of wave momentum here is best explained as follows.
The Stokes transport equals the mean wave momentum, as shown in Eq (1.3.24).
For a single sine wave in intermediate water depth (Phillips, 1977), the Stokes
drift is
cosh(2k(z + h))
uS (z) = a2 ωk , (1.3.46)
2 sinh2 (kh)
where h is the water depth. In deep water, where the general dispersion relation
reduces to (1.3.11), ie where ω 2 = gk tanh kh ≈ gk when kh → ∞ (short waves
in deep water), this simplifies to
uS (z) = ωka2 e2kz . (1.3.47)
Since the amplitude is related to the spectrum as a2 /2 = F (ω)dω, we see that the
Stokes drift of a spectrum of deep-water waves is
� ∞
uS (z) = 2 ωke2kz F (ω) dω. (1.3.48)
0

The Stokes transport is the vertical integral of the Stokes drift,

� 0 � ∞ � ∞
US = 2 2kz
kωe F (ω) dω dz = ωF (ω) dω. (1.3.49)
−∞ 0 0
1.3. SURFACE WAVES 33

Substituting ω = gk/ω for deep-water waves (1.3.11), we can now write

� ∞
k
US = g F (ω) dω. (1.3.50)
0 ω
We now finally get to the end of this by noting that the wave momentum defined
in Eq (1.3.24) is � ∞
k
Mw = ρUS = ρg F (ω) dω, (1.3.51)
0 ω