Ocean Surface Waves-Stanislaw

December 11, 2012 17:27 World Scientific Book - 9in x 6in chelsea-BK06 bookmassel
162 Statistical Properties of Ocean Waves

Copyright 2013. World Scientific Publishing Company. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses
wind speed U and wind fetch X is due to Phillipss constant. Therefore

from Eq. (4.147) we obtain
u2 = 0.9680

c2 = 0.7375 . (4.148)

c2 /u2 = 0.7619

In Figs. 4.14 and 4.15, the comparison of the theoretical formulae (4.148)
with the experimental Cox and Munk (1954) data are illustrated for selected
wind fetches X = 10, 50, 100 km. The agreement is now much better than
in the case of the uni-modal directional spreading. However it should be
noted that in the Cox and Munk experiment, the exact values of the wind
fetches are not known.
4.4 Wave Height
4.4.1 Probability distribution of wave height for a narrow-

band spectrum
Besides the probability distribution for local sea surface deviations from
mean water level, random wave process is frequently characterised by am-
plitude statistics. Amplitude distributions are widespread in physics and
electronics. However, for surface ocean waves, the wave height probabil-
ity is also frequently used. Both these statistical characteristics will be
discussed below.
Wave amplitude can be detected in the simplest way under the as-
sumption of narrow-band spectrum, in which the positive and negative
maximum excursions of wave surface process are equal to each other, and
permitted under U.S. or applicable copyright law.
equal to wave amplitude. However, this is not usually the case for real
ocean surface waves. A typical record of ocean waves for a non-narrow-
band spectrum is given in Fig. 4.16. It is clear that the envelopes of the
process are not symmetrical curves. Therefore, the positive and negative
amplitudes are different. In order to overcome this uncertainty, in physical
oceanography and ocean engineering, the crest-to-trough excursion is used
and wave height is defined as a sum of the two values of the upper and
lower amplitudes. Only for a narrow-band spectrum, is the wave height
equal to twice the upper (or lower) amplitude.
EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 3/16/2016 10:37 PM via UNIVERSITY OF
WESTERN AUSTRALIA
AN: 564575 ; Massel, Stanislaw R..; Ocean Surface Waves : Their Physics and Prediction
Account: s3358796
October 17, 2012 11:24 World Scientific Book - 9in x 6in chelsea-BK06 bookmassel
4.4. Wave Height 163

Fig. 4.16 Definition of zero-downcrossing waves.
It should be noted that the definition of an individual wave height depends

entirely on the choice of the trough occurring before or after the crest.
In the following, we will adopt the recommendations of the International
Association for Hydraulic Research Working Group (IAHR, 1989). Thus,
a wave height is defined as zero-down-crossing wave height and the wave
trough occurs before the wave crest (Fig. 4.16).
Consider the unidirectional wave motion of a narrow-band frequency
spectrum at a given point
X
(t) = ai cos (i t i ) . (4.149)
i=1
Using a frequency p , which corresponds to the spectrum peak, we can
rewrite Eq. (4.149) as
(t) = Ac (t) cos (p t) As (t) sin (p t) , (4.150)
in which
Ac (t) =
P
i=1 ai cos [(i p ) t i ]
. (4.151)
P
As (t) = i=1 ai sin [(i p ) t i ]

Because of the narrowness of the process, the amplitudes Ac (t) and As (t)
are very slowly varying functions of time. Let us rewrite Eq. (4.150) in
terms of wave amplitude A(t) and phase (t)
(t) = A(t) cos [p t + (t)] , (4.152)
WESTERN AUSTRALIA
Account: s3358796

where
p
A(t) = A2c (t) + A2s (t), (4.153)
(t) = tan1 [As (t)/Ac (t)] (4.154)
and
Ac (t) = A(t) cos (t), As (t) = A(t) sin (t). (4.155)

It should be noted that, in the case of a narrow-band spectrum, the ampli-
tude A is equal to the amplitude of the wave envelope, which varies slowly
in time. In virtue of the Central Limit Theorem, Ac and As are Gaussian
processes with mean value equal to zero and variance 2 as
E A2c = E A2s = E 2 = 2 .

(4.156)
It can be shown that E [Ac As ] = 0. Thus, c and s are independent

variables and the two-dimensional probability density function becomes
" #
1 A2c + A2s
f2 (Ac , As ) = f (Ac ) f (As ) = exp . (4.157)
22 22
(Ac , As )
Eq. (4.157) and the Jacobian transformation J = = A, allows to
(A, )
represent f2 as a function of variables A and , i.e.
!
A A2
f2 (A, ) = f [Ac (A, ) , As (A, )] J = exp 2 . (4.158)
22 2
Finally, the one-dimensional probability densities for amplitude A and

phase are obtained through integration of Eq. (4.158), with respect to
phase and amplitude A, respectively

Z !
A A2
f (A) = f2 (A, ) d = 2 exp 2 (4.159)
2
and
!

1 A A2 1
Z
f () = 2 exp 2 dA = . (4.160)
2 0 2 2
Equation (4.159) represents the well-known Rayleigh distribution for wave

amplitude. Result (4.160) indicates that for a narrow-band process, the
phase is uniformly distributed in the range (, ).
WESTERN AUSTRALIA
Account: s3358796

For later convenience, we calculate the moments of wave amplitude

Z m+1 !
2
m A A m/2 m
E [A ] = Am = 2 exp 2 dA = 22 1+ ,
0 2 2
(4.161)
in which ( ) is a Gamma function (Abramowitz and Stegun, 1975). In
particular, for m = 2 we have
1 1
2 =
E[A2 ] = A2 . (4.162)
2 2
Therefore, the Rayleigh distribution takes the form
!
A2 A2

A 2A
f (A) = 2 exp 2 = exp . (4.163)
2 A2 A2
As was mentioned above, in physical oceanography and ocean engineering,
the wave height H is preferred to wave amplitude A. The assumption of
a narrow-band spectrum leads to the conclusion that H 2A. The trans-
formation of variable A to H in Eq. (4.163) gives the Rayleigh distribution
in terms of wave height H
!
H H2
f (H) = exp 2 (4.164)
42 8
or

1 1
f () = exp 2 , (4.165)
4 8
H
in which = .

Similarly to Eq. (4.161), the moments of wave height can be presented
as follows
m
E [H m ] = H m = 23m/2 m 1 + . (4.166)
2
In particular, the mean wave height H and the root-mean-square wave
height Hrms are obtained from Eq. 4.166, if m = 1 and m = 2, respectively
2
H = 2 , Hrms = 2 2 = H. (4.167)

Using Eq. (4.167), the distribution (4.164) can be rewritten as
H2

2H
f (H) = 2 exp 2 (4.168)
Hrms Hrms
WESTERN AUSTRALIA
Account: s3358796

Fig. 4.17 Comparison of various probability density functions for wave height with
experimental data Rayleigh distr (4.165); - - - - - modified Rayleigh distribu-
tion (4.197); crest-to-trough distribution (4.216).
or
f () = 2 exp 2 ,

(4.169)
H
in which = and
Hrms
" 2 #
H H
f (H) = exp . (4.170)
2 H2 4 H
The Rayleigh distribution (4.165) is compared with experimental data

recorded during a few hours of a tropical storm in the Coral Sea (see
Fig. 4.17).
It should be noted that the wave height is normalised by the standard

deviation m0 calculated from the spectrum. Two other distributions are
shown in the same Figure. Their origin and properties are described in
next section.
Although various wave height relations based on the Rayleigh distri-
bution can be found elsewhere (for example, Massel, 1989), here we only
discuss one specific wave height frequently used in oceanographic and ocean
engineering practice namely, the significant wave height Hs . The concept of
the significant wave was introduced by Sverdrup and Munk (1947). They
WESTERN AUSTRALIA
Account: s3358796

Fig. 4.18 Determination of significant wave height Hs .
defined wave height Hs as the average of the highest one-third of wave

heights. This wave height is close to the mean wave height estimated by
the naked human eye. Assuming a narrow-band wave spectrum, we first
calculate the threshold wave height H which has the probability of ex-
ceedance of 1/3 and corresponds to value H = H/ 3.0 (see Fig. 4.18).
Using Eq. (4.168) we have
Z " 2 #
2H H 1
Prob {H > H } = 2
exp dH = (4.171)
H Hrms Hrms 3
or
( " 2 #)

H 1
Z
d exp = . (4.172)
Hrms 3
Thus

H = ln3 Hrms = 1.048 Hrms. (4.173)
According to definition, the significant wave height corresponds to the cen-
ter of gravity of the shaded area in Fig. 4.18. The balance of moments with
respect to the origin yields
Z
1
Hs = H f (H)dH (4.174)
3 H
or

3
Hs = erf c ln 3 + ln 3 Hrms 1.416Hrms, (4.175)
2
WESTERN AUSTRALIA
Account: s3358796

where the complementary error function erf c(z) is given by (Abramowitz

and Stegun, 1975)
Z
2
erf c(z) = exp(t2 )dt = 1 erf (z) (4.176)
z
and
z
2
Z
erf (z) = exp(t2 )dt. (4.177)
0
Using the second relation of (4.167), the significant wave height can be
expressed as
Hs = 1.416 Hrms = 4.004 . (4.178)
The development of various reliable digital data recording and analysis

techniques (see Chapter 14) over the last 20 years has led to a fundamental
change in the way significant wave height is estimated. Except for Hs ,
being the mean of the highest one-third of wave height, significant wave
height is commonly estimated from gauge records using relation (4.178)
when standard deviation is calculated from the frequency spectrum S().
This significant wave height is usually noted as Hm0

Hm0 4.00 m0 . (4.179)
In other words, Hm0 is an energy based significant wave height, determined

as four times the square root of the area contained under the energy spec-
trum S() (IAHR, 1989). The value of Hm0 is approximately equal to Hs
except when water depth is small (see Section 4.4.5) or waves are very steep.
In situations where Hm0 is equal to Hs , either can be used. However, when
Hm0 differs from Hs , it cannot be used directly to estimate wave height

statistics.
The formula (4.171) can be generalised to evaluate the average of the
highest 1/N th wave heights H 1/N , based on the Rayleigh distribution, i.e.

N
H 1/N = erf c ln N + ln N Hrms . (4.180)
2
In Table 4.2 the values of H 1/N are listed for selected values of N .
Haver (1985) developed an empirical probability density function for
significant wave height in the form of combined log-normal and Weibull
distributions. The coefficients in this distribution are based on the 29 years
WESTERN AUSTRALIA
Account: s3358796

Table 4.2 Characteristic wave heights based on the Rayleigh distribution.
N H 1/N /Hrms H 1/N /H H 1/N /2 Remarks
500 2.680 3.023 7.580

200 2.503 2.823 7.078
100 2.359 2.662 6.671 Highest 1/100th wave
50 2.206 2.488 6.239
25 2.042 2.303 5.775
10 1.800 2.030 5.090
5 1.591 1.795 4.499
3 1.416 1.597 4.004 Significant wave
1 0.886 1.000 2.505 Mean wave
of observations in the northern North Sea. Thus we have

" #
2
1 (ln H s ln H s )
exp for Hs 3.25 m

2

2(ln Hs ) Hs

2(ln Hs )
f (Hs ) = " #
1
H Hs

s

exp for Hs > 3.25 m,
0 0

(4.181)
where
2

(ln Hs ) = 0.371, ln Hs = 0.801
. (4.182)
= 1.531, 0 = 2.713

2
The ln Hs and (ln Hs ) are the mean value and variance, respectively of
ln Hs , and and are the Weibull paramaters. The probability density
function f (Hs ) and corresponding cumulative distribution function F (Hs )
are shown in Fig. 4.19.
4.4.2 Influence of wave nonlinearity on wave height

distribution
4.4.2.1 Modification of the Rayleigh distribution
All statistical characteristics of wave height discussed in the previous section
are based on the assumption that the probability distribution of surface
displacement (t) is Gaussian. In fact, as was shown in Section 4.2.2,
the probability distribution of is only approximately Gaussian, having
some skewness and kurtosis. The non-Gaussian profile is asymmetric with
WESTERN AUSTRALIA
Account: s3358796

[m]
Fig. 4.19 Probability dnsity function f (Hs ) and cumulative distribution function F (Hs )
according to Haver (1985).
respect to the mean water surface. Usually the crests are sharper and
narrower, and the troughs are shallower and longer. Thus, quadratic and
higher-order interactions between component waves exist, and the resulting
surface displacement is not a Gaussian process.
In this section we examine the influence of the wave field nonlinearities
on the statistics of wave heights. In general, the departure from a Gaussian
process of surface displacement is not very large and surface displacement
(t) can be treated as a quasi-Gaussian. Thus, it is useful to explore first
the application of the GramCharlier series approach to wave height distri-
bution. The narrow-band wave process can be represented as a harmonic

signal with randomly modulated amplitude and phase, which are slowly
varying in time [see Eq. (4.152)]
(t) = A(t) cos [p t + (t)] , (4.183)
where amplitude A(t) is given by Eq. (4.153). It is quite acceptable to
assume that surface displacement (t) is differentiable in time
d(t)
= (t) A(t) p sin [p t + (t)] , (4.184)
dt
in which we neglected the time derivatives of slowly varying amplitude A(t)
and phase (t). Using Eq. (4.153) we rewrite Eq. (4.183) as follows
(t)
= Ac (t) sin (p t) As (t) cos (p t) . (4.185)
p
WESTERN AUSTRALIA
Account: s3358796

Solving Eqs. (4.150) and (4.185), with respect to Ac (t) and As (t), gives

(t)
Ac (t) = (t) cos (p t) sin (p t)

p

. (4.186)
(t)
As (t) = cos (p t) (t) sin (p t)

p
Thus, amplitude A(t) becomes (Tikhonov, 1966)
v !2
u
u (t)
A(t) = t 2 (t) + . (4.187)
p
If process (t) is Gaussian, the derivative (t) is also Gaussian. By virtue

of the linearity of Eq. (4.186), Ac (t) and As (t) will be Gaussian too.
The joint probability density function f2 (Ac , As ) can be obtained by
using the two-dimensional Hermite polynomials (Kuznetsov et al., 1960).
Then, transforming the function f2 (Ac , As ) to the joint probability den-
sity function of amplitude and phase f2 (A, ) and integrating it against ,
yields the probability density function for wave amplitude (Bitner, 1980).
Provided that the peak-to-trough wave height H is approximately equal
to the double amplitude, Bitner obtained the probability density function
f (H) as follows
" #( "
H H2 H4
f (H) = exp 1 + (b40 + b04 + 2b22 ) +
42 82 10244
#)
H2 1
(b40 + b04 + 2b22 ) + (b40 + b04 + 2b22 ) , (4.188)
322 8
in which
b40 = c4 3, b04 = s4 , b22 = c2 s 1, (4.189)
Ac (t) Ac (t) As (t) As (t)

c = s = . (4.190)

In particular, mean wave height H is

1
H = 2 1 (b40 + b04 + 2b22 ) . (4.191)
64
However, no suitable parameterisation technique is available for evaluation
of the coefficients b40 , b04 and b22 . Bitner (1980) determined these coeffi-
cients numerically using experimental data.
WESTERN AUSTRALIA
Account: s3358796

In general, in deep water, the corrections in the mean wave height H, due
to coefficients b40 , b04 , and b22 are not significant. This conclusion is in
agreement with the LonguetHiggins (1980) suggestion that the Rayleigh
distribution for non-Gaussian random waves is still applicable when the
root-mean-square wave height in Eq. (4.168) is properly chosen. Particu-
larly, he used the probability density function f (H) in the form
H2

2H
f (H) = 2 exp 2 , (4.192)
Hrms Hrms
where
2
2 1
Hrms = 82 1 2 < 82 , (4.193)
8 2
in which is the lower-order bandwidth parameter. From curve-fitting the
Forristall (1978)
data from the Gulf of Mexico, LonguetHiggins obtained
Hrms = 1.85 2 < 2 2 .
A simple argument that the denominator in Eq. (4.192) should be
smaller than 82 was given by Vinje (1989). Assuming that a wave train
is a superposition of a carrier wave with slowly varying amplitude A, fre-
quency and phase , and a small random perturbation 1 (t), we obtain
(t) = A(t) cos [(t) + (t)] + 1 (t), (4.194)
in which is a small parameter. It should be expected that the maxima

and minima of the surface elevation (t) approximately coincides with the
maxima and minima of the carrier wave. Therefore, the wave height H is
simply the difference of the surface elevations at time t (wave crest) and

t = tmax + (wave trough). To determine the probability distribution for

H we define the Gaussian process X(t) as

X(t) = (t) t + . (4.195)

Now wave height H can be considered as the maximum of the process X(t).
The wave height H will have approximately the Rayleigh distribution with
the root-mean-square value Hrms
h i
Hrms = 4 K (0) K , (4.196)

where K ( ) is the auto-correlation
function of (t). For narrow-banded
processes, the value K corresponds approximately to the minimum

WESTERN AUSTRALIA
Account: s3358796

of K ( ); so K [K ( )]min . Finally, the normalised probability

density distribution function f (H) becomes
2

f () = exp , (4.197)
4 (1 rmin ) 4 (1 rmin )
where rmin denotes the global minimum of the normalised auto-correlation
function rmin = [K ( )]min /K (0). For the very narrow-banded process
rmin 1 and the probability density function (4.197) becomes identical
with the Rayleigh distribution function (4.165).
The distribution (4.197) was also obtained by Naess (1984). In his
derivation, the normalised auto-correlation function rmin was calculated at
the time as = Tp /2, where Tp denotes the period corresponding to the
spectrum peak frequency.
It should be noted that LonguetHiggins value of Hrms = 1.85 2 cor-
responds to rmin = 0.711. Using the auto-correlation function K( ), cor-
responding to the PiersonMoskowitz spectrum, results in rmin = 0.653.
A similar calculation for the JONSWAP spectrum gives 0.73 and 0.8,
for peakedness = 3.3 and = 7.0, respectively.
The probability distribution function (4.197) was shown in Fig. 4.17.
The minimum rmin of the normalised auto-correlation function was calcu-
lated using the experimental data (rmin = 0.59). Thus, the denominator
in Eq. (4.197) is 4 2 (1 rmin ) = 6.36 2 . Note that for the same experi-
mental data, the LonguetHiggins formula (4.193) gives the value 7.19 2 .
It should be noted that because the envelopes A are symmetric, they
cannot be used to describe the nonlinear crest or trough statistics. However,
in order to estimate the influence of the second-order of the nonlinearities
on the probability density functions of peaks and troughs of the Stokes
waves we apply the perturbation technique as follows

= a cos + a2 k cos2 , (4.198)
in which a is amplitude and is a phase = kx t + . The peaks of
the wave profile are given by
c = a + a2 k, (4.199)
or in non-dimensional form
c = + 2 k, (4.200)
in which c = c / and = a/ .
WESTERN AUSTRALIA
Account: s3358796

Fig. 4.20 Probability density function for crests and troughs of nonlinear waves.
Using the method of the transformation of random variables we obtain the

probability density function of c as (Tung and Huang, 1985)
1/2
1 + (1 + 4c k)
f (c ) =
2 k(1 + 4c k)1/2
( )
1 h i2
1/2
exp 2 2 1 + (1 + 4c k) . (4.201)
8 k
The function (4.201) is shown in Fig. 4.20 for k = 0.10. The Rayleigh
distribution was plotted in this figure for comparison to demonstrate the

effect of nonlinearity of wave motion on the probability density function.
The probability density function of the troughs of (t) may be obtained
in a similar way to produce the final result as
1 (1 4t k)1/2
f (t ) = 1/2

2 k(1 4t k)
( )
1 h i2
1/2
exp 2 2 1 (1 4t k) , (4.202)
8 k
in which = |t |/ and t is an ordinate of wave trough. Function f (t )
is also plotted in Fig. 4.20. Nonlinear waves have a higher probability of
WESTERN AUSTRALIA
Account: s3358796

occurrence of high crest values than linear waves and lower probability of
occurrence of high trough values.
4.4.2.2 Crest-to-trough wave height distribution

The Rayleigh distribution and its modifications were evaluated under the
assumption that upper and lower wave envelopes coincide with crests and
troughs. This is essentially relevant to the narrow-banded spectrum when
the envelopes vary slowly in time (see Fig. 3.2). When the spectrum is non-
narrow banded, the envelopes are not the symmetric curves with respect to
mean value. Therefore, it may be more appropriate to consider the wave
height as the sum of the highest-crest (Ac ) and the lowest-trough (At )
amplitudes in a zero-upcrossing wave
H = Ac (t) + At (t + ). (4.203)
Both amplitudes are separated by the time interval . We expect that time
is approximately equal to T /2, where T is a mean wave period. The
(+) ()
probability density function of two normalised amplitudes A1 and A2
separated by time was given by Rice (1945) (see also Ochi, 1990)
12 + 22

1 2 1 2 r0
f2 (1 , 2 | ) = I0 exp , (4.204)
1 r02 1 r02 2 (1 r02 )
in which
At (t) Ac (t + )
1 = , 2 = , (4.205)
m0 m0
r0 ( ) is a correlation coefficient between wave amplitudes (strictly speaking
between squares of the amplitudes), separated by time interval and
r02 ( ) = 20 ( ) + 20 ( ), (4.206)
in which

1
Z
0 ( ) = S() cos [( ) ] d, (4.207)
m0 0

1
Z
0 ( ) = S() sin [( ) ] d. (4.208)
m0 0
The I0 ( ) is a modified Bessel function of zero order (Abramowitz and

Stegun, 1975).
Note that for any frequency spectrum S(), which is symmetric with
respect to frequency , the coefficient 0 ( ) = 0 for all .
WESTERN AUSTRALIA
Account: s3358796

Let us assume that spectrum S() is extremely narrow ( 0)

S() = m0 ( ) , (4.209)
in which ( ) is the delta function. Substituting Eq. (4.209) into Eqs. (4.207)
and (4.208) we obtain 0 ( ) = 1 and 0 ( ) = 0; thus, r0 = 1. It can be
shown that the same result can be obtained when assuming | | 0 for an
arbitrary spectrum.
When r0 1, because of the high correlation between 1 and 2 , the
two-dimensional probability density function f2 (1 , 2 ; ) takes the form
(Middleton, 1960)
f2 (1 , 2 ; ) f (1 ) (1 2 ) , (4.210)
which corresponds to the Rayleigh distribution when r0 1 ( 0). For
other extreme cases, when , the correlation coefficient r0 0, and
so
f2 (1 , 2 | ) = f (1 ) f (2 ) . (4.211)
Thus, amplitudes 1 and 2 are statistically independent.
Let us normalise the wave height in Eq. (4.204) using variance m0 ; then
= 1 + 2 and 1 = 2 , (4.212)

where = H/ m0 ; the 1 and 2 are given by the Eq. (4.205).
The probability density function for takes a form of the convolution
integral (Tayfun, 1981)
Z
f ( | ) = f2 ( 2 , 2 ; ) d2 , (4.213)
0
in which = T /2, where T is a wave period. Allowing the wave period T
to take all possible values from 0 to , the probability density function of
crest-to-trough wave height becomes

Z Z
T
f () = f (T ) f2 2 , 2 ; = d2 dT, (4.214)
0 0 2
in which f (T ) is a probability density function of the wave period.
Calculation of the probability density function in (4.214) is complicated.
To simplify the problem, Tayfun (1981) assumed that the spectrum is suffi-
ciently narrow and is concentrated around the mean frequency . Then, the
probability density function of the wave period has non-zero values around
the mean period T only, and Eq. (4.214) becomes
Z
T
f () = f2 2 , 2 ; = d2 . (4.215)
0 2
WESTERN AUSTRALIA
Account: s3358796

Substitution of Eq. (4.204) into Eq. (4.215) gives

" #
2
( x) + x2

1 ( x) x r0
Z
f () = ( x) x I0 exp dx.
1 r020 1 r02 2 (1 r02 )
(4.216)
The first two moments of distribution (4.216) are

E[] = 2 (4.217)
and
2 3 4

E 2 = 4 1 +

1++ + + + ... , (4.218)
4 4 4 64
in which = r02 /4.

Thus, the mean wave height is identical with that resulting from the
Rayleigh distribution. The series representation (4.218) of the normalised
root-mean-square wave height for r0 1( 0) gives E 2 = 8, which
is also the Rayleigh limit.
When 0 r0 1, Eq. (4.218) gives 2.672 Hrms / 2 2; this means
that the root-mean-square value of zero-crossing wave heights predicted by
Eq. (4.216) is smaller than that given by the classical Rayleigh distribution.
The distribution (4.216) is illustrated in Fig. 4.21 for r0 = 0.5 and
r0 = 0.8. The Rayleigh distribution is included for comparison. Both
distributions resulting from Eq. (4.216) predict lower probabilities for small
and high waves. On the other hand, the wave heights close to the mean,
i.e. when = 2 2.5, are more likely to occur according to Eq. (4.216)
than according to the Rayleigh function. The smaller r0 value produces the
higher probability of occurrence of such waves.
Comparison of the distribution (4.216) with experimental data is also

shown in Fig. 4.17. Experimental data in this figure corresponds to the
correlation coefficient r0 in the range 0.57 r0 0.60 and band-width
parameter in the range 0.36 0.38.
4.4.3 Probability distribution of large wave heights

A probability distribution of large wave heights is of special interest for
ocean engineers. However, the application of distribution (4.216) is impeded
by the necessity of numerical integration which requires evaluation of the
Bessel function I0 .
When we restrict our attention to the higher waves,
say for E [] = 2 and use the asymptotic expansion of the Bessel
WESTERN AUSTRALIA
Account: s3358796

Fig. 4.21 Comparison of various theoretical probability density functions for large wave
heights Rayleigh distribution (4.165); crest-to-trough distribution for r0 =
0.5 and r0 = 0.8 (4.216); Tayfuns distribution for large wave heights for r0 = 0.5
and r0 = 0.8 (4.219).
function I0 ( ) for larger values, the probability density function f () can

be presented in the much simpler closed form (Tayfun, 1990)
1 r02 2

f () = 1 + exp . (4.219)
2[2r0 (1 + r0 )]
1/2 4r0 2 4 (1 + r0 )
A typical comparison of Eq. (4.219) with exact distribution (4.216) is shown
in Fig. 4.21 for r0 =0.5 and r0 = 0.8. It should be noted that for normalised
wave height 2 2.5, both distributions (4.216) and (4.219) are
almost the same.

It should be noted that the significant wave height calculated on the

base of Eq. (4.219) is smaller than the classical value 4.004 m0 for all
practical values of r0 (0.5 r0 1.0). In particular, for r0 = 0.5 the

significant wave height is equal to 3.75 m0 , and then approaches almost

linearly to 4.004 m0 for r0 = 1.
The corresponding probability of exceedance of a given value , when
E[] is
Z 1/2
1 r02 1 + r02 2

F1 () = f ()d = 1 + exp .
4r0 2 2r0 4 (1 + r0 )
(4.220)
WESTERN AUSTRALIA
Account: s3358796

Fig. 4.22 Probability of exceedance a given normalised wave height Rayleigh

distribution (4.165); crest-to-trough distribution for r0 = 0.5 and r0 = 0.8 (4.216);
distribution for large wave heights (4.220).
The function F1 () for r0 = 0.5 and 0.8 is illustrated in Fig. 4.22. For
comparison, the Rayleigh distribution, which represents the limit r0 1,
was added. A good agreement was found when comparing the asymptotic
probability of exceedance (4.220) with experimental data (Forristall, 1978)
and numerical simulation (Forristall, 1984; Osborne, 1982).
A reliable estimation of the probability exceedance of high waves is fun-
damental in the design of structures and in the planning of operational and

maintenance activities. Many experiments in the field, as well as simulation
exercises, indicate that the Rayleigh distribution over-predicts the probabil-
ity of the highest waves, and hence errors grow towards the low probability
trail of the distribution. This can be seen in Fig. 4.23, in which the Rayleigh
distribution is compared with the modified distribution (4.197), and with
the experimental data from the Coral Sea.
4.4.4 Probability distribution of extreme wave heights

Design criteria for coastal and offshore structures are based on occur-
rence probabilities of extreme wave conditions. For example, the classical
WESTERN AUSTRALIA
Account: s3358796

Fig. 4.23 Comparison of various probabilities of exceedance with experimental data

Rayleigh distribution; modified Rayleigh distribution; experimental
data.
Rayleigh probability density distribution (4.164) decreases exponentially as

the wave height becomes large. The maximum wave height Hmax in a sam-
ple depends fundamentally on the number of waves N in the sample. Be-
cause the sampling of waves is random, the maximum wave height is a ran-
dom variable too, with its own probability density fHmax (Hmax ). Let us
normalise the maximum wave height as follows max = Hmax /[2(2m0 )1/2 ].
1/2
The quantity 2(2m0 ) approximate the root-mean-square wave height of
the stationary record, so that it is a convenient choice as the characteris-

tic wave height scale. Assuming that the occurrence of wave height is an
independent event and that the wave spectrum is narrow-banded, Longuet
Higgins (1952) showed that the probability density function for normalised
highest wave max in a stationary record of N waves is
fmax (max ) = N f (max ) [F (max )]N 1 (4.221)
and
N
Fmax (max ) = [F (max )] , (4.222)
where f (max ) is the probability density distribution of individual wave
height, and F (max ) is a corresponding cumulative distribution as follows
f (max ) = 2max exp max 2 ,

(4.223)
WESTERN AUSTRALIA
Account: s3358796

Table 4.3 Normalised mean maximum wave heights.
N Rayleigh Rayleigh Crest/trough Markov Finite depth
distr. modif. distr. chain H/h = 0.25 H/h = 0.5
200 2.41 2.22 2.24 2.37 2.02 1.65

1800 2.83 2.61 2.61 2.82 2.27 1.79
3600 2.95 2.73 2.75 2.94 2.35 1.82
5000 3.01 2.78 2.81 3.00 2.38 1.84
10000 3.13 2.88 2.91 3.11 2.45 1.88
20000 3.24 2.98 3.01 3.22 2.51 1.98
50000 3.38 3.11 3.14 3.35 2.59 1.95
100000 3.48 3.20 3.24 3.45 2.65 1.98
Table 4.4 Normalised the most probable maximum wave heights.
N Rayleigh Rayleigh Crest/trough Markov Finite depth
distr. modif. distr. chain H/h = 0.25 H/h = 0.5
200 2.32 2.16 2.18 2.29 1.96 1.64

1800 2.76 2.52 2.57 2.74 2.24 1.76
3600 2.88 2.64 2.68 2.86 2.32 1.80
5000 2.92 2.72 2.74 2.92 2.34 1.83
10000 3.04 2.80 2.84 3.04 2.40 1.84
20000 3.16 2.92 2.94 3.16 2.48 1.88
50000 3.30 3.04 3.08 3.30 2.56 1.92
100000 3.40 3.12 3.17 3.40 2.60 1.96
F (max ) = 1 exp max 2 ,

(4.224)
where = H/Hrms and max = Hmax /Hrms .
Therefore we obtain the probability density fmax (max ) for the highest
wave in N waves in the form

N 1
fmax (max ) = 2N max exp 2max 1 exp 2max

. (4.225)
The normalised mean extreme wave heights H/Hrms and the most probable
extreme wave heights H/Hrms for various sample sizes are listed in the
second columns of Tables 4.3 and 4.4.
In Fig. 4.24 the functions f (max ) and fmax (max ) for three sample
sizes (N = 200, 1800, 3600) are shown. For large max , the calculation of
the most probable value simplifies considerably. Hence from Eq. (4.223) we
have
df 2
Fmax (max ) max + (N 1) [fmax ] = 0 (4.226)
dmax
WESTERN AUSTRALIA
Account: s3358796

Fig. 4.24 Probability density function of extreme wave heights for three sample seizes.
The Rayleigh distribution is given for comparison.
or
1 22max
exp 2max =

. (4.227)
1 2N 2max
When max 1, Eq. (4.227) yields
exp 2max = N.

(4.228)
Therefore, the most probable value Hmax becomes

Hmax = ln N Hrms + O(ln N )3/2 . (4.229)
Closer approximations to max can be found by applying Newtons method
starting with a value given by Eq. (4.227).
It is widely acknowledged that the Rayleigh distribution does not reflect
a measured distribution of the more extreme waves from field data. Equa-
tion (4.168) over-predicts the probabilities of the higher waves in a record,
and the error increases toward the low-probability tail of the distribution
(see Fig. 4.22). Numerous reasons for the observed over-prediction have
been suggested (Forristall, 1978, 2000; LonguetHiggins, 1980; Tayfun,
1981, 2004; Naess, 1984; Myrhaug and Kjeldsen, 1986; Massel and Sobey,
2000).
WESTERN AUSTRALIA
Account: s3358796

Statistical estimates of extreme wave heights, that are based on the Rayleigh
distribution, will be even less likely to follow field observations. For exam-
ple, Myrhaug and Kjeldsen analysed statistically about 25, 000 single storm
waves recorded on the Norwegian continental shelf. The measured extreme
wave heights were smaller than those calculated from the Rayleigh distri-
bution. Tayfun (2004) explored further the least upper bound distribution
of the nonlinear wave crests and developed the theoretical expressions for
statistics of the largest crest in N waves. These provide a rational basis for
exploring the nature and occurrence of freak waves (see Chapter 8). Sobey
et al. (1990) compared the highest wave heights Hmax in a standard 20
minutes record from tropical cyclone Victor in March 1986 on Australias
North West Shelf with those computed from Eq. (4.229). Observations
during tropical cyclone Victor data show a systematic over-prediction of
order 10% by the Rayleigh distribution. The probable causes of the ob-
served discrepancy between measured and calculated extreme wave heights
are likely to be finite spectral width, correlation between successive wave
maxima, nonlinearity of profile and asymmetry of crest and trough of the
extreme waves, limited water depth, wave breaking, errors associated with
instrumentation, and others.
In Section 4.4.2 it was shown that finite spectral width can be incor-
porated into the Rayleigh distribution with properly chosen root-mean-
square wave height [see Eq. (4.193)]. The mean and the most probable
normalised extreme wave heights for various sample sizes, based on the
modified Rayleigh distribution (4.192), are shown in the third column of
Tables 4.3 and 4.4. The value = 0.45 was used in calculations. A com-
parison of the second and third columns suggests that finite spectral width
reduces the highest wave in a sample.
Strictly speaking, the above analysis of the extreme wave heights is valid
only for random sampling. However, when the bandwidth of the spectrum
goes to zero, there must be some correlation between members of the sam-
ple. Thus, waves with a narrow-band spectrum cannot be treated as inde-
pendent. There are several ways of including the correlation of neighboring
maxima of a Gaussian process (t). In fact, in Section 4.4.2 we applied
the correlation between adjacent wave crest and trough, and developed the
probability density function of the crest-to-trough excursion (4.214). Using
Eq. (4.216) in Eq. (4.221), the mean maximum wave heights and the most
probable maximum wave heights were calculated and listed in the fourth
column in Tables 4.3 and 4.4. For numerical calculation, the JONSWAP
spectrum with peak frequency p = 0.8 rad/s and correlation coefficient
WESTERN AUSTRALIA
Account: s3358796

r0 = 0.73 was applied. Again, the smaller value of normalised mean and
the most probable maximum wave heights were obtained when compared
with the Rayleigh distribution.
Another way to include the correlation of particular sample elements
is to impose the Markov chain conditions on the sequence A1 , A2 , . . . , AN
of maxima (Naess, 1984). This means that correlation exists only between
neighboring maxima and is zero for further maxima. Epstein (1949) as-
suming that individual waves follow a stationary Markov process obtained
for cumulative distribution function for the largest values in samples of size
N in the form (Massel and Sobey, 2000)
N 1
[F2 (1max , 2max )]
Fmax (max ) = , (4.230)
[F1 (max )]N 2
where
F2 (x, y) = Prob [xi x, xi+1 y] (4.231)
and
Z
F1 (x) = Prob [xi x] = F2 (x, y)dy, (4.232)
0
in which x and y being dummy variables for max , and subscripts i and
i + 1 identifying consecutive values in the stationary Markov process.
Naess (1984) further assumed that the individual waves correspond with
consecutive maxima of a stationary Gaussian process with mean zero and
variance = m0 and obtained
(1 2 + )N 1
Fmax (max ) = N 2
(4.233)
(1 )
and
N 2
(1 2 + )
fmax (max ) = 2max N 1
M, (4.234)
(1 )
in which

1
M = 2(N 1)(1 ) (N 2)(1 2 + ), (4.235)
1 + rH

2
and = exp 2max and = exp 1+r 2

H
max .
Equation (4.234) is referred to as the Epstein/Naess model for the
highest wave. The results
of calculations for the mean JONSWAP spec-
trum ( = 3.3) with r T 0.5 are listed in the fifth columns of Tables 4.3
and 4.4. The Markov chain correlation results in extreme values which are
WESTERN AUSTRALIA
Account: s3358796

smaller than those predicted by the classical Rayleigh distribution. How-

ever, the difference is rather small ( 12%).
The influence of water depth on the probability density function of wave
height will be discussed in the next section. Here we only conclude that
in shallow water areas the extreme wave heights are also smaller (see the
sixth and seventh columns in Tables 4.3 and 4.4).
An inspection of Tables 4.3 and 4.4 shows that all examined mechanisms
provide values of extreme wave heights smaller than those resulting from
the Rayleigh distribution. The differences in the mean and in the most
probable wave heights usually are small; less than 10%. However, in the case
of limited water depth, extreme wave heights are much smaller. In a real
situation of severe sea, many mechanisms mentioned above superimpose
and contribute to the final result, which is difficult to predict. The extreme
values resulting from the Rayleigh distribution can be regarded as an upper
limit of other theoretically predicted or experimentally observed values.
Confirmation of any model for the distribution of the highest wave in
a stationary sea state is a difficult issue. Natural climate variability guaran-
tee that there will be sufficient stationary sea state field data to construct
only very coarse relative frequency histograms. An alternative and direct
prediction for the distribution of the highest wave has been establish from
random simulation. The theoretical basis of the methodology is the Gaus-
sian random wave model for an irregular sea state, whose statistical origin
in the theory of Gaussian random noise is shared by the theoretical dis-
tributions. For sea states described by the JONSWAP spectral form, the
simulation predictions for the distribution of the highest wave in N waves
is presented in approximate analytical form
" 2 #
1 1/2 max Q2
fmax (max ) = exp , (4.236)
Q Q
2
where = erf (0.5) 0.227, and Q is Q2 Q1 for Q Q2 , but Q3 Q2
for Q > Q2 . The quartiles approximately follow the empirical curve
Qn = an + bn ln N + cn ln2 N (4.237)
Coefficients an , bn , cn are weakly dependent on the spectral peak enhance-

ment factor . They are listed in Massel and Sobey (2000) paper for = 1
and 3.3.
WESTERN AUSTRALIA
Account: s3358796

4.4.5 Probability distribution of wave height in finite

water depth
In Section 4.2.4 it was shown that the probability distribution of the sur-
face displacement in finite water depth, and especially in shallow water
areas, is not Gaussian. The sea bottom imposes extra constraints on wave
motion, which departs further from the simple linear superposition model.
Although the sea surface is very asymmetric about the still water level and
the vertical lines passing through the wave crests, the influence of the non-
linear motion is not so strong. This conclusion is in agreement with Thorn-
ton and Guza (1983) experimental data, which showed that wave height,
even within the surf zone, is reasonably well described by the Rayleigh
distribution, which contradicts the findings of Bitner (1980) under similar
conditions. Bitner used the distribution (4.188) based on two-dimensional
Hermite polynomials.
Basically we can use the Rayleigh distribution or its modifications given
in Section 4.4.2, to describe the wave height distribution in finite water
depth. However, none of these distributions explicitly contains explicitly
the water depth as a parameter. Glukhovskiy (1966) has developed an
extension of the Rayleigh distribution for finite water depth in the form
1 + n 2

b H 1n H 1 n
f (H) = exp a , (4.238)

H H H
in which
2a
a= ; b= . (4.239)
n 1n
4 1+
2
The coefficient n is a ratio of mean wave height and water depth and ranges
from 0 to 0.5. When n = 0 (deep water), the Glukhovskiy distribution
coincides with the Rayleigh distribution (4.170). The upper limit, n = 0.5,
corresponds to the offshore limit of the surf zone when the probability
density function becomes
3 " 4 #
4a1 H H
f (H) = exp a1 , (4.240)
H H H
3/2
(2)
where a1 = 0.654.
4 1 + 2 2
WESTERN AUSTRALIA
Account: s3358796

Let us now normalise the probability distribution (4.238) using the root-
mean-square wave height Hrms . Hence, we have
Z
2
Hrms = H 2 f (H)dH (4.241)
0
or
Hrms = H, (4.242)
where Z
3n
2
1/2
= b x 1n exp a x 1n dx (4.243)
0
or 1/2
Z
+1
= b x exp (a x ) dx , (4.244)
0
in which = 1/(1 n). After integrating in Eq. (4.241) we obtain (Grad-
shteyn and Ryzhik, 1980)
1/2
b (1+ 2 ) 2
= a 1+ , (4.245)

2
in which ( ) is a gamma function. For n = 0, = 1.1284, while for

n = 0.5. Equation (4.245) yields 1.047.
Substituting Eqs. (4.242) and (4.245) into Eq. (4.238) gives the proba-
bility density function for normalised wave height = H/Hrms in the form
f () = b 1 exp [a ] . (4.246)
The probability density function f () is given in Fig. 4.25 for parameter n
varying from 0 to 0.5 at a step of 0.1.
In two limiting cases the function (4.246) simplifies as follows:
for n = 0 (deep water)
f () = 2 exp 2 ,

(4.247)
and for n = 0.5 (offshore of surf zone)
a1 4
f () = a1 3 exp
(4.248)
4
or
f () = 2.055 3 exp 0.514 4 .

(4.249)
The distribution is thus, more symmetric and narrow if the water depth
decreases. Moreover, the most probable wave height shifts towards higher
values. In Table 4.5, some characteristic wave heights are listed for param-
eter n varying from 0 to 0.5. All ratios, namely Hrms /H, Hs /Hrms and
Hs /H decrease with n (i.e. as waves approach the surf zone). Due to nar-
rowing of the probability density f (), the dependence of the characteristic
values on parameter n also weakens. The normalised mean and the most
probable maximum wave height based on the Glukhovskiy distribution are
included in Tables 4.3 and 4.4 for n = 0.25 and n = 0.5.
WESTERN AUSTRALIA
Account: s3358796

Table 4.5 Characteristic wave heights for Glukhovskiys

distribution.
n Hrms /H Hs /Hrms Hs /H
0.0 1.1284 1.4213 1.601

0.1 1.1127 1.4067 1.565
0.2 1.0961 1.3730 1.505
0.3 1.0792 1.3530 1.460
0.4 1.0625 1.3174 1.400
0.5 1.0465 1.2572 1.316
In fact the Glukhovskiy distribution is very similar to another distribution

which is frequently used in engineering and oceanography, i.e. the Weibull
distribution (Massel, 1996a; Ochi, 1998) developed on empirical grounds

1
f () = exp , (4.250)

where is normalised wave height, and are the distribution parameters.
If = H/ , = 2, and = 8, Eq. (4.250) is identical with the Rayleigh
distribution (4.165). The distribution (4.250) is frequently used for fitting
Fig. 4.25 The Glukhovskiys probability density function for n ranging from 0 to 0.5
at a step 0.1.
WESTERN AUSTRALIA
Account: s3358796

Ocean Surface Waves-Stanislaw

Hochgeladen von

Dokumentinformationen

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Copyright:

Verfügbare Formate

Ocean Surface Waves-Stanislaw

Hochgeladen von

Copyright:

Verfügbare Formate

December 11, 2012 17:27 World Scientific Book - 9in x 6in chelsea-BK06 bookmassel

162 Statistical Properties of Ocean Waves

wind speed U and wind fetch X is due to Phillipss constant. Therefore

4.4 Wave Height

4.4.1 Probability distribution of wave height for a narrow-

4.4. Wave Height 163

Fig. 4.16 Definition of zero-downcrossing waves.

It should be noted that the definition of an individual wave height depends

164 Statistical Properties of Ocean Waves

(t) = tan1 [As (t)/Ac (t)] (4.154)

Ac (t) = A(t) cos (t), As (t) = A(t) sin (t). (4.155)

It can be shown that E [Ac As ] = 0. Thus, c and s are independent

Finally, the one-dimensional probability densities for amplitude A and

phase and amplitude A, respectively

Equation (4.159) represents the well-known Rayleigh distribution for wave

4.4. Wave Height 165

For later convenience, we calculate the moments of wave amplitude

166 Statistical Properties of Ocean Waves

The Rayleigh distribution (4.165) is compared with experimental data

4.4. Wave Height 167

Fig. 4.18 Determination of significant wave height Hs .

defined wave height Hs as the average of the highest one-third of wave

168 Statistical Properties of Ocean Waves

where the complementary error function erf c(z) is given by (Abramowitz

Hs = 1.416 Hrms = 4.004 . (4.178)

The development of various reliable digital data recording and analysis

In other words, Hm0 is an energy based significant wave height, determined

Hm0 differs from Hs , it cannot be used directly to estimate wave height

4.4. Wave Height 169

Table 4.2 Characteristic wave heights based on the Rayleigh distribution.

N H 1/N /Hrms H 1/N /H H 1/N /2 Remarks

500 2.680 3.023 7.580

of observations in the northern North Sea. Thus we have

are shown in Fig. 4.19.

4.4.2 Influence of wave nonlinearity on wave height

170 Statistical Properties of Ocean Waves

bution. The narrow-band wave process can be represented as a harmonic

4.4. Wave Height 171

If process (t) is Gaussian, the derivative (t) is also Gaussian. By virtue

b40 = c4 3, b04 = s4 , b22 = c2 s 1, (4.189)

Ac (t) Ac (t) As (t) As (t)

172 Statistical Properties of Ocean Waves

(t) = A(t) cos [(t) + (t)] + 1 (t), (4.194)

in which  is a small parameter. It should be expected that the maxima

H we define the Gaussian process X(t) as

4.4. Wave Height 173

of K ( ); so K [K ( )]min . Finally, the normalised probability

waves we apply the perturbation technique as follows

174 Statistical Properties of Ocean Waves

Using the method of the transformation of random variables we obtain the

distribution was plotted in this figure for comparison to demonstrate the

4.4. Wave Height 175

4.4.2.2 Crest-to-trough wave height distribution

The I0 ( ) is a modified Bessel function of zero order (Abramowitz and

176 Statistical Properties of Ocean Waves

Let us assume that spectrum S() is extremely narrow ( 0)

crest-to-trough wave height becomes

4.4. Wave Height 177

Substitution of Eq. (4.204) into Eq. (4.215) gives

in which = r02 /4.

Comparison of the distribution (4.216) with experimental data is also

4.4.3 Probability distribution of large wave heights

178 Statistical Properties of Ocean Waves

function I0 ( ) for larger values, the probability density function f () can

(t) = A(t) cos [(t) + (t)] + 1 (t), (4.194)

in which is a small parameter. It should be expected that the maxima