This action might not be possible to undo. Are you sure you want to continue?
GENERATION AND ANALYSIS
OF RANDOM WAVES
Zhou Liu and Peter Frigaard
3. udgave. januar 2001
Laboratoriet for Hydraulik og Havnebygning
Instituttet for Vand, Jord og Miljøteknik
Aalborg Universitet
Preface
Sea waves are the most important phenomenon to be considered in the design of coastal
and oﬀshore structures.
Every sailor has noticed that, when wind is blowing, there are a lot of large and small
waves propagating in many directions. Such waves are called shortcrested waves because
they do not have a long crest. Contrast to shortcrested waves, we have longcrested
waves, i.e. large and small waves moving in one direction. Even though there are some
research eﬀorts on shortcrested waves and their eﬀects on structures, longcrested waves
dominate today’s structure design. The book deals with longcrested waves. The contents
of the book is illustrated in the ﬁgure.
i
There are two analysis methods for time series of irregular waves, namely timedomain
and frequencydomain analysis, which will be dealt with in Chapter 1 and Chapter 2
respectively.
It should be stressed that, even though all contents in the book are related to sea waves,
they have broader applications in practice. For example, the extreme theory has also been
applied to hydrology, wind mechanics, ice mechanics etc., not to mention the fact that
spectral analysis comes originally from optics and electronics.
The book intents to be a textbook for senior and graduate students who have interest in
coastal and oﬀshore structures. The only prerequirement for the book is the knowledge
of linear wave theory.
Michael Brorsen, Associate Professor at the Hydraulic and Coastal Engineering Labora
tory, Aalborg University is gratefully acknowledged for the valuable comments.
ii
CONTENTS
Contents
1 Time series analysis I : Timedomain 1
1.1 Deﬁnition of individual wave : Zerodowncrossing . . . . . . . . . . . . . . 1
1.2 Characteristic wave heights and periods . . . . . . . . . . . . . . . . . . . . 2
1.3 Distribution of individual wave heights . . . . . . . . . . . . . . . . . . . . 3
1.4 Maximum wave height H
max
. . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.5 Distribution of wave period . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.6 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.7 Exercise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2 Timeseries analysis II: Frequencydomain 12
2.1 Some basic concepts of linear wave theory . . . . . . . . . . . . . . . . . . 12
2.2 Example of variance spectrum . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.3 Fourier series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.4 Discrete signal analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.5 Characteristic wave height and period . . . . . . . . . . . . . . . . . . . . . 23
2.6 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.7 Exercise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3 Windgenerated waves 26
3.1 Wave development and decay . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.2 SPMmethod . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.3 Standard variance spectrum . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.4 Introduction to shortcrested waves . . . . . . . . . . . . . . . . . . . . . . 37
3.5 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
3.6 Exercise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
4 Extreme wave height analysis 40
4.1 Design level: Return period and encounter probability . . . . . . . . . . . 40
4.2 General procedure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
4.3 Data sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
4.4 Candidate distributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
4.5 Fitting methods and procedure . . . . . . . . . . . . . . . . . . . . . . . . 45
4.6 Plotting position formulae . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
iii
CONTENTS
4.7 Fitting goodness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
4.8 Design wave height: x
T
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
4.9 Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
4.10 Sources of uncertainties and conﬁdence interval . . . . . . . . . . . . . . . 55
4.11 Physical consideration of design wave height . . . . . . . . . . . . . . . . . 59
4.12 Wave period . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
4.13 Water level . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
4.14 Multiparameter extreme analysis . . . . . . . . . . . . . . . . . . . . . . . 62
4.15 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
4.16 Exercise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
5 Wave generation in laboratory 65
5.1 Principle of wave generator . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
5.2 Bi´esel transfer functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
5.3 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
5.4 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
5.5 Exercise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
iv
1 TIME SERIES ANALYSIS I : TIMEDOMAIN
1 Time series analysis I : Timedomain
The recorded time series of the surface elevation of irregular waves can be studied by
either the timedomain or the frequencydomain analysis. These two analysis methods
will be described in the next two chapters, respectively.
1.1 Deﬁnition of individual wave : Zerodowncrossing
Individual wave is deﬁned by two successive zerodowncrossing points, as recommended
by IAHR (1986), cf. Fig.1.
Fig.1. Individual waves deﬁned by zerodowncrossing.
Fig.2 is an example of surface elevation recordings. The application of zerodowncrossing
gives 15 individual waves (N=15). In Table 1 the data are arranged according to the
descending order of wave height.
Fig.2. Application of zerodowncrossing.
Table 1. Ranked individual wave heights and corresponding periods in Fig.2.
rank i 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
H (m) 5.5 4.8 4.2 3.9 3.8 3.4 2.9 2.8 2.7 2.3 2.2 1.9 1.8 1.1 0.23
T (s) 12.5 13.0 12.0 11.2 15.2 8.5 11.9 11.0 9.3 10.1 7.2 5.6 6.3 4.0 0.9
wave no.
in Fig.2 7 12 15 3 5 4 2 11 6 1 10 8 13 14 9
1
1 TIME SERIES ANALYSIS I : TIMEDOMAIN
1.2 Characteristic wave heights and periods
Usually the surface elevation recording exempliﬁed in Fig.2 contains more than 100 indi
vidual waves. Which wave should be chosen as the design wave ?
Maximum wave: H
max
, T
Hmax
This is the wave which has the maximum wave height. In Table 1,
H
max
= 5.5 m T
Hmax
= 12.5 s
Maximum wave is chosen as the design wave for structures which are very important and
very sensitive to wave load, e.g. vertical breakwaters. Note H
max
is a random variable
with the distribution depending on the number of individual waves.
Highest onetenth wave: H
1/10
, T
H
1/10
H
1/10
is the average of the wave heights of the onetenth highest waves. T
H
1/10
is the
average of the wave periods associated with the onetenth highest wave.
Signiﬁcant wave: H
s
, T
s
or H
1/3
, T
H
1/3
Signiﬁcant wave height is the average of the wave heights of the onethird highest waves.
Signiﬁcant wave period is the average of the wave periods associated with the onethird
highest wave. In Table 1,
H
s
=
1
5
5
¸
i=1
H
i
= 4.44 m T
s
=
1
5
5
¸
i=1
T
i
= 12.8 s i is the rank no.
Signiﬁcant wave is most often used as the design wave. The reason might be that in old
days structures were designed based on visual observation of waves. Experiences show
that the wave height and period reported by visual observation correspond approximately
to signiﬁcant wave. Therefore the choice of signiﬁcant wave as design wave can make use
of the existing engineering experience.
Mean wave: H, T
H and T are the means of the heights and periods of all individual waves. In Table 1,
H =
1
15
15
¸
i=1
H
i
= 2.9 m T =
1
15
15
¸
i=1
T
i
= 9.25 s
Rootmeansqure wave height H
rms
In Table 1, H
rms
=
1
N
N
¸
i=1
H
2
i
=
1
15
15
¸
i=1
H
2
i
= 3.20 m
Wave height with exceedence probability of α%: For example H
0.1%
, H
1%
, H
2%
etc.
2
1 TIME SERIES ANALYSIS I : TIMEDOMAIN
1.3 Distribution of individual wave heights
Histogram of wave height
In stead of showing all individual wave heights, it is easier to use wave height histogram
which tells the number of waves in various wave height intervals. Fig.3 is the histogram
of wave height corresponding to Table 1.
Fig.3. Histogram of wave height.
Nondimensionalized histogram
In order to compare the distributions of wave height in diﬀerent locations, the histogram
of wave height is nondimensionalized, cf. Fig.4.
Fig.4. Nondimensionalized histogram of wave height.
When ∆(H/H) approaches zero, the probability density becomes a continuous curve. Ex
perience and theory have shown that this curve is very close to the Rayleigh distribution.
Roughly speaking, we say that individual wave height follows the Rayleigh distribution.
3
1 TIME SERIES ANALYSIS I : TIMEDOMAIN
Rayleigh distribution
The Rayleigh probability density function is
f(x) =
π
2
x exp
−
π
4
x
2
x =
H
H
(11)
The Rayleigh distribution function is
F(x) = Prob{X < x} = 1 − exp
−
π
4
x
2
(12)
Relation between characteristic wave heights
If we adopt the Rayleigh distribution as an approximation to the distribution of individual
wave heights, then the characteristic wave heights H
1/10
, H
1/3
, H
rms
and H
α%
can be
expressed by H through the manipulation of the Rayleigh probability density function.
H
1/10
= 2.03 H
H
1/3
= 1.60 H
(13)
H
rms
= 1.13 H
H
2%
= 2.23 H
Fig.5 illustrates how to obtain the relation between H
s
and H.
Fig.5. Relation between H
s
and H.
The Rayleigh distribution function given by H
s
instead of H reads
F(H) = 1 − exp
−2
H
H
s
2
(14)
4
1 TIME SERIES ANALYSIS I : TIMEDOMAIN
Individual wave height distribution in shallow water
Only in relatively deep water, the Rayleigh distribution is a good approximation to the
distribution of individual wave heights. When wave breaking takes place due to lim
ited water depth, the individual wave height distribution will diﬀer from the Rayleigh
distribution.
Stive, 1986, proposed the following empirical correction to the Rayleigh distribution based
on model tests but roughly checked against some prototype data
H
1%
= H
m
0
ln100
2
1
2
1 +
H
m
0
h
−
1
3
(15)
H
0.1%
= H
m
0
ln1000
2
1
2
1 +
H
m
0
h
−
1
2
where h is the water depth, H
1%
means the 1% exceedence value of the wave height
determined by zero downcrossing analysis, whereas the signiﬁcant wave height H
m
0
is
determined from the spectrum. The correction formulae are very useful for checking the
wave height distribution in small scale physical model tests, cf. Fig.6.
Fig.6. Comparison of the expression by Stive, 1986, for shallow water wave
height distribution with model test results. Aalborg University Hy
draulics Laboratory 1990 (from Burcharth 1993).
Klopmann et al. (1989) proposed a semiempirical expression for the individual wave
height distribution. Researches have also been done by Thornton and Guza (1983). Chap
ter 2 gives a more detailed discussion on the validity of the Rayleigh distribution, based
on energy spectrum width parameter.
5
1 TIME SERIES ANALYSIS I : TIMEDOMAIN
1.4 Maximum wave height H
max
H
max
is a random variable, whose distribution depends on the number of individual wave
hegiht, N, cf. Fig.7.
Distribution of H
max
The distribution function of X = H/H is the Rayleigh distribution
F
X
(x) = Prob{X < x} = 1 − exp
−
π
4
x
2
(16)
If there are N individual waves in a storm
1
, the distribution function of
X
max
= H
max
/H is
F
Xmax
(x) = Prob{X
max
< x} = ( F
X
(x) )
N
=
1 − exp
−
π
4
x
2
N
(17)
Note that F
Xmax
(x) can be interpreted as the probability of the nonoccurrence of the
event ( X > x ) in any of N independent trials. The probability density function of X
max
is
f
Xmax
(x) =
dF
Xmax
dx
=
π
2
N x exp
−
π
4
x
2
1 − exp
−
π
4
x
2
N−1
(18)
The density function of X and the density function of X
max
are sketched in Fig.7.
Fig. 7. Probability density function of X and X
max
.
1
A storm usually lasts some days. The signiﬁcant wave height is varying during a storm. However we
are more interested in the maximum signiﬁcant wave height in a short period of time. In practice, N is
often assumed to be 1000.
6
1 TIME SERIES ANALYSIS I : TIMEDOMAIN
Mean, median and mode of H
max
Mean, median and mode are often used as the characteristic values of a random variable.
Their deﬁnitions are given in Fig.8.
Mean xmean = E[X] =
+∞
−∞
xf
X
(x)dx
Median x
median
= x
F
X
(x)=0.5
Mode x
mode
= x
f
X
(x)=max
Fig. 8. Mean, median and mode of a random variable X.
By putting eqs (17) and (18) into the deﬁnitions, we obtain
(H
max
)
mean
≈
¸
ln N
2
+
0.577
√
8 ln N
¸
H
s
(19)
(H
max
)
mode
≈
ln N
2
H
s
(110)
Furthermore, (H
max
)
µ
, deﬁned as the maximum wave height with exceedence probability
of µ, cf. Fig. 9, is
(H
max
)
µ
≈
1
2
ln
¸
N
ln
1
1−µ
¸
H
s
(111)
Obviously (H
max
)
median
= (H
max
)
0.5
.
Fig. 9. Deﬁnition of (H
max
)
µ
.
7
1 TIME SERIES ANALYSIS I : TIMEDOMAIN
MonteCarlo simulation of H
max
distribution
The distribution of H
max
can also be studied by the MonteCarlo simulation. Individual
wave heights follow the Rayleigh distribution
F(H) = 1 − exp
−2
H
H
s
2
(112)
The storm duration corresponds to N individual waves.
1) Generate randomly a data between 0 and 1. Let the nonexceedence
probability F(H) equal to that data. One individual wave height
H is obtained by (cf. Fig.10)
H = F
−1
( F(H) ) = H
s
−ln(1 −F(H))
2
(113)
2) Repeat step 1) N times. Thus we obtain a sample belonging to the
distribution of eq (112) and the sample size is N.
3) Pick up H
max
from the sample.
4) Repeat steps 2) and 3), say, 10,000 times. Thus we get 10,000
values of H
max
.
5) Draw the probability density of H
max
.
Fig.10. Simulated wave height from the Rayleigh distribution.
8
1 TIME SERIES ANALYSIS I : TIMEDOMAIN
1.5 Distribution of wave period
It is summarized as
• There is no generally accepted expression for the distribution of wave period.
• The distribution of wave period is narrower than that of wave height.
• In practice the joint distribution of wave height and wave period is of great impor
tance. Unfortunately, Until now there is no generally accepted expression for the
joint distribution, even though there are some socalled scatter diagrams based on
wave recording. Such a diagram is valid only for the measurement location. An ex
ample of scatter diagrams is given in Chapter 4, section 12. The relation between H
s
and T
s
is often simpliﬁed as T
s
= αH
β
s
, e.g. in Canadian Atlantic waters α = 4.43
and β = 0.5 (Neu 1982).
• The empirical relation T
max
≈ T
1/10
≈ T
1/3
≈ 1.2 T (Goda 1985).
1.6 References
Burcharth H.F. , 1993. The design of breakwater. Department of Civil Engineering, Aal
borg University, 1993.
Dean, R.G. and Dalrymple, R.A. , 1991. Water wave mechanics for Engineers and sci
entists. Second printing with correction, World Scientiﬁc Publishing Co. Pte. Ltd.,
Singapore, 1991.
Goda, Y. , 1985. Random seas and design of marine structures. University of Tokyo
Press, Japan, 1985
IAHR. , 1986. List of sea state parameters. Supplement to Bulletin No. 52, 1986
Klopmann, G. and Stive, M.J.F. , 1989. Extreme waves and wave loading in shallow wa
ter. E & P Forum, Report No. 3.12/156, 1989.
Neu, H.J.A. , 1982. 11year deep water wave climate of Canadian Atlantic waters. Cana
dian Tech. Rept. of Hydrography and Ocean Sciences, 13.
Stive, M.J.F. , 1986. Extreme shallow water conditions. Delft Hydraulics, Intern Report
H533, 1986.
9
1 TIME SERIES ANALYSIS I : TIMEDOMAIN
1.7 Exercise
1) The application of the downcrossing method gives the following 21 indi
vidual waves.
wave wave wave
number height period
H (m) T (s)
1 0.54 4.2
2 2.05 8.0
3 4.52 6.9
4 2.58 11.9
5 3.20 7.3
6 1.87 5.4
7 1.90 4.4
8 1.00 5.2
9 2.05 6.3
10 2.37 4.3
wave wave wave
number height period
H (m) T (s)
11 1.03 6.1
12 1.95 8.0
13 1.97 7.6
14 1.62 7.0
15 4.08 8.2
16 4.89 8.0
17 2.43 9.0
18 2.83 9.2
19 2.94 7.0
20 2.23 5.3
21 2.98 6.9
Calculate H
max
, T
max
, H
1/10
, T
1/10
, H
1/3
, T
1/3
, H, T, H
rms
2) Prove H
2%
= 2.23 H
3) Explain the diﬀerence between H
1/10
and H
10%
.
4) Suppose individual waves follow the Rayleigh distribution. Calculate the
exceedence probability of H
1/10
, H
s
and H.
5) An important coastal structure is to be designed according to H
max
. The
signiﬁcant wave height of the design storm is H
1/3
= 10 m. The duration
of the storm corresponds to 1000 individual waves.
(1) Calculate (H
max
)
mean
, (H
max
)
mode
, (H
max
)
median
, (H
max
)
0.05
(2) Now suppose that the storm contains 500 individual waves. Calculate
(H
max
)
mean
, (H
max
)
mode
, (H
max
)
median
, (H
max
)
0.05
. Compare with the
results of (1).
(3) Use MonteCarlo simulation to determine (H
max
)
mean
, (H
max
)
mode
,
(H
max
)
median
, (H
max
)
0.05
10
1 TIME SERIES ANALYSIS I : TIMEDOMAIN
6) You have generated a linear wave in a laboratory ﬂume. The wave period is
1 second. You would like to measure the surface elevation in order to check
whether it is the same as desired.
(1) What is the consequence if sampling frequency is too high, e.g. 100 Hz ?
(2) What is the consequence if sampling frequency is too low, e.g. 1 Hz ?
(3) What is an appropriate sampling frequency ?
11
2 TIMESERIES ANALYSIS II: FREQUENCYDOMAIN
2 Timeseries analysis II: Frequencydomain
The concept of spectrum can be attributed to Newton, who discovered that sunlight can
be decomposed into a spectrum of colors from red to violet, based on the principle that
white light consists of numerous components of light of various colors (wave length or
wave frequency).
Energy spectrum means the energy distribution over frequency. Spectral analysis is a
technique of decomposing a complex physical phenomenon into individual components
with respect to frequency.
Spectral analysis of irregular waves is very important for structure design. For example,
in the oildrilling platform design where wave force plays an important role, it is of im
portance to design the structure in such a way that the natural frequency of the structure
is fairly far away from the frequency band where most wave energy concentrates, so that
resonance phenomenon and the resulted dynamic ampliﬁcation of force and deformation
can be avoided.
2.1 Some basic concepts of linear wave theory
Surface elevation
The surface elevation of a linear wave is
η(x, t) =
H
2
cos(ωt −kx +δ) = a cos(ωt −kx +δ) (21)
where H wave height
a amplitude, a = H/2
ω angular frequency, ω = 2π/T
T wave period.
k wave number, k = 2π/L
L wave length
δ initial phase
We can also deﬁne the observation location to x = 0 and obtain
η(t) = a cos(ωt +δ) (22)
The relation between wave period and wave length (dispersion relationship) is
L =
g T
2
2π
tanh
2πh
L
(23)
where h is water depth.
12
2 TIMESERIES ANALYSIS II: FREQUENCYDOMAIN
Wave energy
The average wave energy per unit area is
E =
1
8
ρ g H
2
=
1
2
ρ g a
2
(Joule/m
2
in SI unit) (24)
Variance of surface elevation of a linear wave
The variance of the surface elevation of a linear wave is
σ
2
η
= Var[η(t)] = E
¸
η(t) − η(t)
2
(E: Expectation)
= E [ η
2
(t) ]
=
1
T
T
0
η
2
(t) dt (T: wave period)
=
1
2
a
2
Superposition of linear waves
Since the governing equation (Laplace equation) and boundary conditions are linear in
small amplitude wave theory, it is known from mathematics that small amplitude waves
are superposable. This means that the superposition of a number of linear waves with
diﬀerent wave height and wave period will be
superposition wave 1 wave 2 · · · wave N
velocity potential ϕ = ϕ
1
+ ϕ
2
+ · · · + ϕ
N
surface elevation η = η
1
+ η
2
+ · · · + η
N
particle velocity u = u
1
+ u
2
+ · · · + u
N
dynamic pressure p = p
1
+ p
2
+ · · · + p
N
13
2 TIMESERIES ANALYSIS II: FREQUENCYDOMAIN
2.2 Example of variance spectrum
First we will make use of an example to demonstrate what a variance spectrum is.
Surface elevation of irregular wave
Fig.1 gives an example of an irregular wave surface elevation which is constructed by
adding 4 linear waves (component waves) of diﬀerent wave height and wave period. The
superposed wave surface elevation is
η(t) =
4
¸
i=1
η
i
(t) =
4
¸
i=1
a
i
cos(ω
i
t +δ
i
) (25)
Fig.1. Simulation of irregular waves by superposition of linear waves.
14
2 TIMESERIES ANALYSIS II: FREQUENCYDOMAIN
Variance diagram
In stead of Fig.1, we can use a variance diagram, shown in Fig.2, to describe the irregular
wave.
Fig.2. Variance diagram.
In comparison with Fig.1, the variance diagram keeps the information on amplitude (a
i
)
and frequency (f
i
, hence T
i
and L
i
) of each component, while the information on initial
phase (δ
i
) is lost. This information loss does not matter because the surface elevation of
irregular wave is a random process. We can simply assign a random initial phase to each
component.
Variance spectral density S
η
(f)
The variance diagram can be converted to variance spectrum, The spectral density is
deﬁned as
S
η
(f) =
1
2
a
2
∆f
(m
2
s) (26)
where ∆f is the frequency band width
2
, cf. Fig.3.
Fig.3. Stepped variance spectrum.
In reality an irregular wave is composed of inﬁnite number of linear waves with diﬀerent
2
we will see later that ∆f depends on signal recording duration. In the ﬁgure it is assumed that
∆f = 0.01Hz
15
2 TIMESERIES ANALYSIS II: FREQUENCYDOMAIN
frequency. Fig.4 gives an example of stepped variance spectrum. When ∆f approaches
zero, the variance spectrum becomes a continuous curve.
Fig.4. Continuous variance spectrum (wave energy spectrum).
Variance spectrum is also called energy spectrum. But strictly speaking, the energy
spectral density should be deﬁned as
S(f) =
1
2
ρ g a
2
∆f
(m
2
s) (27)
Construction of time series from variance spectrum
We can also construct time series of surface elevation from variance spectrum. In ﬁg.4
the known variance spectral density S
η
(f) is divided into N parts by the frequency band
width ∆f. This means that the irregular wave is composed of N linear waves
η(t) =
N
¸
i=1
η
i
(t) =
N
¸
i=1
a
i
cos(ω
i
t +δ
i
) (28)
The variance of each linear wave is
S
η
(f
i
) ∆f =
1
2
a
2
i
i = 1, 2, · · · , N (29)
Therefore the amplitude is
a
i
=
2 S
η
(f
i
) ∆f i = 1, 2, · · · , N (210)
The angular frequency is
ω
i
=
2π
T
i
= 2πf
i
i = 1, 2, · · · , N (211)
The initial phase δ
i
is assigned a random number between 0 and 2π. Hence by use of eq
(28) we can draw the timeseries of the surface elevation of the irregular wave which has
the variance spectrum as shown in Fig.4.
16
2 TIMESERIES ANALYSIS II: FREQUENCYDOMAIN
2.3 Fourier series
Conversion of irregular surface elevation into variance spectrum is not as simple as the
above example, where the linear components of the irregular wave are predeﬁned (cf.
Fig.1). We need to decompose the irregular wave into its linear components. First let’s
see how it can be done with a known continuous function x(t).
Fourier series
Fourier series is used to represent any arbitrary function
3
.
Fig.5. Arbitrary periodic function of time.
x(t) = a
0
+ 2
∞
¸
i=1
a
i
cos
2πi
T
0
t
+ b
i
sin
2πi
T
0
t
= 2
∞
¸
i=0
(a
i
cos ω
i
t + b
i
sin ω
i
t) (212)
where a
i
and b
i
are Fourier coeﬃcients given by
a
i
=
1
T
0
T
0
0
x(t) cos ω
i
t dt
b
i
=
1
T
0
T
0
0
x(t) sin ω
i
t dt
i = 0, 1, 2, · · · , ∞ (213)
Note a
0
=
1
T
0
T
0
0
x(t)dt and b
0
= 0.
3
Not all mathematicians agree that an arbitrary function can be represented by a Fourier series.
However, all agree that if x(t) is a periodic function of time t, with period T
0
then x(t) can be expressed
as a Fourier series. In our case x(t) is the surface elevation of irregular wave, which is a random process.
if T
0
is large enough, we can assume that x(t) is a periodic function with period T
0
.
17
2 TIMESERIES ANALYSIS II: FREQUENCYDOMAIN
Physical interpretation
Now we say that the continuous function x(t) is the surface elevation of irregular wave
η(t), which can be expanded as a Fourier series.
η(t) = 2
∞
¸
i=0
(a
i
cos ω
i
t + b
i
sin ω
i
t)
= 2
∞
¸
i=0
(c
i
cos δ
i
cos ω
i
t + c
i
sin δ
i
sin ω
i
t)
=
∞
¸
i=0
2c
i
(cos δ
i
cos ω
i
t + sin δ
i
sin ω
i
t)
=
∞
¸
i=0
2c
i
cos(ω
i
t − δ
i
) (214)
That is to say, any irregular wave surface elevation, expressed as a continues function, is
composed of inﬁnite number of linear waves with
amplitude 2c
i
= 2
a
2
i
+b
2
i
period T
i
=
2π
ω
i
=
T
0
i
i = 0, 1, · · · , ∞ (215)
{a
i
, b
i
}, i = 0, 1, 2, · · · , ∞, are given in eq (213).
18
2 TIMESERIES ANALYSIS II: FREQUENCYDOMAIN
2.4 Discrete signal analysis
The measurement of surface elevation is carried out digitally. We do not have, neither
necessary, a continuous function of the surface elevation. In stead we have a series of
surface elevation measurement equally spaced in time, cf. Fig.6.
Fig.6. Sampling of surface elevation at regular intervals.
If the sampling frequency is f
s
, then the time interval between two succeeding points is
∆ = 1/f
s
. Corresponding to the total number of sample points N, the sample duration
T
0
= (N −1)∆. Thus we obtain a discrete time series of surface elevation
η
0
, η
1
, · · · , η
N−1
The Fourier coeﬃcients
(a
0
, b
0
), (a
1
, b
1
), · · · , (a
N−1
, b
N−1
)
can be obtained by Fast Fourier Transforms (FFT)
4
. That is to say, the irregular wave
surface elevation, expressed by digital time series, is composed of N linear waves
η(t) =
N−1
¸
i=0
η
i
(t) =
N−1
¸
i=0
2
a
2
i
+b
2
i
cos(ω
i
t +δ
i
) (216)
amplitude 2
a
2
i
+b
2
i
angular frequency ω
i
=
2πi
T
0
period T
i
=
2π
ω
i
=
T
0
i
frequency f
i
=
1
T
i
=
i
T
0
i = 0, 1, · · · , N −1 (217)
4
FFT is a computer algorithm for calculating DFT. It oﬀers an enormous reduction in computer
processing time. For details of DFT and FFT, please refer to Newland (1975)
19
2 TIMESERIES ANALYSIS II: FREQUENCYDOMAIN
Therefore we obtain the variance spectrum
frequency band width ∆f = f
i+1
−f
i
=
1
T
0
spectral density S
η
(f
i
) =
1
2
(amplitude)
2
∆f
=
2(a
2
i
+b
2
i
)
∆f
(218)
An example of variance spectrum is shown in Fig.7.
Fig.7. Variance spectrum (including aliasing).
Nyquist frequency f
nyquist
Nyquist frequency f
nyquist
is the maximum frequency which can be detected by the Fourier
analysis.
Fourier analysis decomposes N digital data into N linear components. The frequency of
each component is
f
i
=
i
T
0
i = 0, 1, · · · , N −1 (219)
The nyquist frequency is
f
nyquist
= fN−1
2
=
N−1
2
T
0
=
N−1
2
(N −1) ∆
=
1
2 ∆
=
f
s
2
(220)
where f
s
sample frequency
∆ time interval between two succeeding sample points, ∆ = 1/f
s
N total number of sample
T
0
sample duration, T
0
= (N −1) ∆
The concept of nyquist frequency means that the Fourier coeﬃcients { a
i
, b
i
}, i =
0, 1, · · · , N − 1, contains two parts, the ﬁrst half part (i = 0, 1, · · · , N/2 − 1) represents
20
2 TIMESERIES ANALYSIS II: FREQUENCYDOMAIN
true components while the second half part (i = N/2, N/2 + 1, · · · , N −1) is the folding
components (aliasing).
Fig.8 gives an example on aliasing after the Fourier analysis of discrete time series of a
linear wave.
Fig.8. Aliasing after Fourier analysis.
The solution to aliasing is simple: double S
η
(f
i
), i = 0, 1, 2, · · · , N/2−1 and let S
η
(f
i
), i =
N/2, N/2 + 1, · · · , N −1, equal to zero, cf. Fig.9. That is the reason why f
nyquist
is also
called cutoﬀ frequency. In doing so we are actually assuming that irregular wave contains
no linear components whose frequency is higher than f
nyquist
. This assumption can be
assured by choosing suﬃciently high sample frequency f
s
, cf. eq (220).
21
2 TIMESERIES ANALYSIS II: FREQUENCYDOMAIN
Fig.9. Variance spectrum after cutoﬀ (refer to Fig.7).
Taper data window
Fourier analysis requires that η(t) is a periodic function with period T
0
, it may be desirable
to modify the recorded time series before Fourier analysis, so that the signal looks like a
periodic function. The modiﬁcation is carried out with the help of taper data window.
The widelyused cosine taper data window reads
d(t) =
1
2
1 −cos
10πt
T0
0 ≤ t ≤
T0
10
1.004
T0
10
≤ t ≤
9T0
10
1
2
1 + cos
10π(t−
9T
0
10
)
T0
9T0
10
≤ t ≤ T
0
(221)
Fig.10. Taper data window.
22
2 TIMESERIES ANALYSIS II: FREQUENCYDOMAIN
2.5 Characteristic wave height and period
The variance spectrum, illustrated in Fig.11, says nothing about how high the individual
waves will be. Now We will see how to estimate the characteristic wave height and period
based on the variance spectrum.
Fig.11. Variance spectrum.
n order moment m
n
m
n
is deﬁned as
m
n
=
∞
0
f
n
S
η
(f) df (222)
The zero moment is
m
0
=
∞
0
S
η
(f) df (223)
which is actually the area under the curve, cf. Fig.11.
Spectrum width parameter and validity of the Rayleigh distribution
From the deﬁnition of m
n
, it can be seen that the higher the order of moment, the more
weight is put on the higher frequency portion of the spectrum. With the same m
0
, a
wider spectrum gives larger values of the higher order moment (n ≥ 2). LonguetHiggins
has deﬁned a spectrum width parameter
ε =
1 −
m
2
2
m
0
m
4
(224)
It has been proven theoretically that
spectrum width parameter wave height distribution
ε = 0 narrow spectrum Rayleigh distribution
ε = 1 wide spectrum Normal distribution
23
2 TIMESERIES ANALYSIS II: FREQUENCYDOMAIN
In reality ε lies in the range of 0.40.5. It has been found that Rayleigh distribution is a
very good approximation and furthermore conservative, as the Rayleigh distribution gives
slightly larger wave height for any given probability level.
Signiﬁcant wave height H
m
0
and peak wave period T
p
When wave height follows the Rayleigh distribution, i.e. ε = 0 , the signiﬁcant wave
height H
m
0
5
can theoretically be expressed as
H
m
0
= 4
√
m
0
(225)
In reality where ε = 0.4 − 0.5, a good estimate of signiﬁcant wave height from variance
spectrum is
H
s
= 3.7
√
m
0
(226)
Peak frequency is deﬁned as (cf. Fig.11)
f
p
= f
Sη(f)=max
(227)
Wave peak period (T
p
= 1/f
p
) is approximately equal to signiﬁcant wave period deﬁned
in timedomain analysis.
2.6 References
Burcharth, H.F. and Brorsen, M. , 1978. On the design of gravity structures using wave
spectra. Lecture on Oﬀshore Engineering, Edited by W.J.Graﬀ and P. Thoft
Christensen, Institute of Building Technology and Structural Engineering, Aalborg
University, Denmark.
Goda, Y. , 1985. Random seas and design of marine structures. University of Tokyo
Press, Japan, 1985
Newland, D.E. , 1975. In introduction to random vibrations and spectral analysis. Long
man, London, 1975.
5
H
m0
denotes signiﬁcant wave height determined from spectrum while H
s
or H
1/3
is signiﬁcant wave
height determined from timedomain analysis. They are equal to each other when wave height follows
the Rayleigh distribution.
24
2 TIMESERIES ANALYSIS II: FREQUENCYDOMAIN
2.7 Exercise
1) An irregular wave is composed of 8 linear components with
wave no. 1 2 3 4 5 6 7 8
wave height H (m) 5.0 4.3 3.8 3.6 3.3 2.8 2.2 0.3
wave period T (s) 10.3 12 9.4 14 7 6.2 5 3.3
The recording length is 20 seconds. Draw the variance diagram and variance
spectrum of the irregular wave.
2) Convert the variance spectrum obtained in exercise 1) into time series of
surface elevation.
3) Make a computer program to simulate the surface elevation of an irregular
wave which is composed of 8 linear components. Wave height and period
of each component are given in exercise 1). Suppose the sample frequency
is 3 Hz and the recording length is 500 seconds.
(1) Determine H
s
and T
s
by timedomain analysis.
(2) Compare the distribution of individual wave height with the Rayleigh
distribution.
(3) Calculate total number of linear components to be given by Fourier anal
ysis N, frequency band width ∆f, and the nyquist frequency f
nyquist
.
(4) Draw the variance spectrum of the irregular wave by FFT analysis. (only
for those who have interest.)
4) In reality where ε = 0.4 − 0.5, a good estimate of signiﬁcant wave height
from variance spectrum is
H
s
= 3.7
√
m
0
Try to ﬁnd out the principle of getting this empirical relation.
25
3 WINDGENERATED WAVES
3 Windgenerated waves
If a structure is to be built at the location where there is no direct wave measurement,
wave characteristics may be estimated by wind data.
Two simpliﬁed methods have been used to determine wave characteristics from a known
wind ﬁeld. The one is called SPMmethod (Shore Protection Manual, 1984), which is
the modiﬁcation of SverdrupMunkBretschneider method (SMBmethod). SPMmethod
gives signiﬁcant wave height (H
m
0
) and peak period (T
p
) in terms of wind ﬁeld
6
. The
other is called spectrummethod, which gives variance spectrum in terms of wind ﬁeld.
When required, a signiﬁcant wave height and peak period can be estimated from the
spectrum and the results will be the same as SPMmethod.
Besides these two simpliﬁed method, there are also numerical methods solving a diﬀeren
tial equation governing the growth of wave energy. This approach will not be discussed
in detail because the application of such models require specialized expertise.
3.1 Wave development and decay
Wind waves grow as a result of a ﬂux of energy from the air into the water. When the
wind velocity near the water surface exceeds a critical value of about 1 m/s, one can
observe water surface ripples of length 510 cm and height 12 cm.
The process of wave development is complex. First the windwave interaction transfers
wind energy to shorter waves. Then the wavewave interaction transfers energy in shorter
waves to energy in longer waves, thus resulting in the growth of longer waves.
Wind energy can be transferred to the waves only when the component of surface wind
in the direction of wave travel exceeds the speed of wave propagation. Waves begin to
decay when winds decrease in intensity or change in direction, or waves propagate out of
wind ﬁeld.
Therefore a change in wave energy depends on the transformation of the wind’s kinetic
energy into the wave energy, the transformation of wave energy at one frequency into wave
energy at other frequencies, the dissipation of wave energy into turbulence by friction,
viscosity and breaking, the advection of wave energy into and out of a region.
6
Notice that H
m0
is the signiﬁcant wave height determined from variance spectrum. In Shore Protec
tion Manual (1984) the peak period is denoted T
m
.
26
3 WINDGENERATED WAVES
3.2 SPMmethod
This method is presented in Shore Protection Manual (1984), edited by the US Army
Corps of Engineers, Coastal Engineering Research Center (CERC).
Involved parameters
1 Fetch (F in (m)): Fetch is the distance between the point of interest and
shoreline in the upwind direction. Because the fetches surrounding the wind
direction will inﬂuence the wind generated waves, SPM (1984) recommends
to construct 9 radials from the point of interest at 3degree intervals and to
extend these radials until they ﬁrst intersect shorelines. The fetch is equal to
the average of the length of these 9 radials, i.e.
F =
9
¸
i=1
F
i
9
(31)
2 Wind stress factor (U
A
in (m/s)): Wind stress is most directly related to wave
growth. The accurate estimation of vertical proﬁle of wind speed, and hence
wind stress, involves the airsea temperature diﬀerence, sea surface roughness
and friction velocity. In SPM (1984), all these factors are accounted for by
using U
A
a. Elevation. If the given wind speed is not measured at the 10 meter elevation,
the wind speed must be adjusted accordingly by
U
10
= U
z
10
z
1/7
for z < 20 m (32)
where U
10
and U
z
are wind speed at the elevation of 10 m and z m respectively.
27
3 WINDGENERATED WAVES
b. Location eﬀects. If wind speeds is estimated by visual observations on ships,
they should be corrected by
U = 2.16 U
(
7
9
)
s (33)
where U
s
is the shipreported wind speed in knots and U is the corrected wind
speed in knots.
If wind data over water is not available, but data from nearby land site are,
Fig.1 can be used to convert overland winds to overwater winds if they are
the result of the same pressure gradient and the only major diﬀerence is the
surface roughness
c. Stability correction. If the airsea temperature diﬀerence (∆T = T
air
−T
sea
)
is zero, the boundary layer is stable and wind speed correction is unnecessary.
If ∆T is negative, the boundary layer is unstable and wind speed is more
eﬀective in causing wave growth. If ∆T is positive, the boundary layer is
unstable and the wind speed is less eﬀective. Fig.2 gives the wind speed
ampliﬁcation factor (R
T
) due to airsea temperature diﬀerence. In the absence
of temperature information R
T
= 1.1 can be applied.
d. Durationaveraged wind speed. The wind speed is often observed and re
ported as the maximum shortdurationaveragedspeed. This should be con
verted to the wind speed averaged in an appropriate duration by
U
t
U
t=3600s
= 1.277 + 0.296 tanh
0.9 log
10
45
t
for 1s < t < 3600s (34)
U
t
U
t=3600s
= 1.5334 − 0.15 log
10
(t) for 3600s < t < 36000s (35)
where U
t
is the average wind speed in t seconds.
e. Windstress factor. The windstress factor is implemented in order to
account for the nonlinear relationship between wind stress and wind speed.
U
A
= 0.71 U
1.23
10
(36)
where U
10
is the wind speed at the height of 10 m over mean water level,
modiﬁed according to location and airsea temperature, and averaged over an
appropriate duration. It should be noted that the unit of U
10
and U
A
is m/s
because eq (36) has no unithomogenity.
3 Wind duration (t in (s)) and water depth (h in (m)).
28
3 WINDGENERATED WAVES
Fig.1. Ratio of wind speed over water (U
w
) to wind speed over land (U
L
)
(scanned from SPM 1984).
Fig.2. Ampliﬁcation factor accounting for the eﬀect of airsea temperature diﬀerence
(scanned from SPM 1984).
29
3 WINDGENERATED WAVES
We may expresses the signiﬁcant wave height and peak period in functional forms
H
m
0
, T
p
= f ( U
A
, F, t, h ) (37)
A dimensional analysis applied to eq (37) gives
g H
m
0
U
2
A
,
g T
p
U
A
= f
g F
U
2
A
,
g t
U
A
,
g h
U
2
A
(38)
where g = 9.81 (m/s
2
) is the gravitational acceleration.
Fetchlimited case
It is the situation where the wind has blown constantly long enough for wave heights at
the end of the fetch to reach equilibrium.
1 Deep water (
h
L
>
1
2
): The condition for deep water waves to be fetchlimited
is that the wind duration is longer than the minimum necessary duration t
min
,
given by
g t
min
U
A
= 68.8
g F
U
2
A
2/3
(39)
Signiﬁcant wave height and peak period under fetchlimited condition are
g Hm
0
U
2
A
= 0.0016
g F
U
2
A
1/2
g Tp
U
A
= 0.2857
g F
U
2
A
1/3
(310)
Eq (310) shows, a larger fetch gives a larger wave height and longer wave
period. But there is a limit, the socalled fully arisen sea. This wave condition
refers to the case where the waves have reached an equilibrium state in which
energy input from the wind is exactly balanced by energy loss. The fully arisen
sea occurs when
g F
U
2
A
≥ 23123 (311)
That is to say, eq (310) is valid up to
g F
U
2
A
= 23123. When
g F
U
2
A
> 23123, waves
become fullyarisen, and the signiﬁcant wave height and peak period are
g Hm
0
U
2
A
= 0.0016 ( 23123 )
1/2
= 0.243
g Tp
U
A
= 0.2857 ( 23123 )
1/3
= 8.134
(312)
30
3 WINDGENERATED WAVES
2 Transitional or shallow water (
h
L
<
1
2
): Waves feel the eﬀect of sea bottom.
Some part of wave energy dissipates due to bottom friction and percolation.
For the same wind speed and fetch, wave height will be smaller and wave
period shorter in comparison with deep water situation. SPM (1984) suggests
the following formulae
g H
m
0
U
2
A
= 0.283 tanh
¸
0.53
g h
U
2
A
3/4
¸
tanh
0.00565
g F
U
2
A
1/2
tanh
0.53
g h
U
2
A
3/4
¸
¸
¸
¸
¸
¸
g T
p
U
A
= 7.54 tanh
¸
0.833
g h
U
2
A
3/8
¸
tanh
0.0379
g F
U
2
A
1/3
tanh
0.833
g h
U
2
A
3/8
¸
¸
¸
¸
¸
¸
g t
min
U
A
= 537
g T
p
U
A
7/3
SPM (1984) calls the above formulae ’interim formulae’ because the modiﬁ
cation is ongoing in order to make the above formulae consistent with deep
water.
Durationlimited
It is the situation where the wind duration is shorter than the minimum necessary dura
tion.
There is no generally accepted formula. SPM (1984) suggests to make use of the formulae
for the fetchlimited situation. It proceeds as
1) Check out t < t
min
, i.e. duration limited
2) Replace t
min
by t in eq (39) and calculate the ﬁctional fetch F
3) Calculate H
m
0
and T
p
by eq (310) where the fetch is the ﬁctional fetch.
31
3 WINDGENERATED WAVES
Example Application of SPMmethod
Given Eight consecutive hourly observations of fastest mile wind speed U
0
=
20 m/s are observed at an elevation z = 6 m, approximately 5 kilometers
inland from shore. The observation site is at an airport weather station.
The airsea temperature diﬀerence is estimated to be −6
0
C.
Wanted H
m0
and T
p
for the fetch 100 kilometers at a deep water location.
Solution Fastest mile wind speed is the fastest wind speed, averaged over the
duration equal to the time needed for the fastest wind speed to travel 1
mile. 1 mile = 1609 m.
We proceed as follows
1. Elevation adjustment
U
10
= U
z
10
z
1/7
= U
0
10
6
1/7
= 21.5 m/s
2. Location adjustment
From Fig.1 it is found R
L
= 0.9, the wind speed is adjusted to
0.9 ×21.5 = 19.4 m/s.
3. Temperature adjustment
From Fig.2 it is found R
T
= 1.14, the wind speed is adjusted to
1.14 ×19.4 = 22.1 m/s.
4. Duration adjustment
The duration over which the fastest mile wind speed is averaged is actu
ally the time needed for the fastest mile wind speed to travel one mile.
t =
1609
22.1
= 72.8 s
i.e. the wind velocity of 22.1 m/s is the average velocity in 72.8 s, denoted
as U
t=72.8
. It should be converted to the average wind velocity in one
hour, because in this example the fastest mile wind speed is given on
hourly basis.
Ut=72.8
Ut=3600
= 1.277 + 0.296 tanh
0.9 log
10
45
72.8
= 1.22
U
t=3600
= U
t=72.8
/1.22 = 18.1 m/s
5. The wind stress factor is
U
A
= 0.71 (18.1)
1.23
= 25 m/s
32
3 WINDGENERATED WAVES
6. Type of wind wave
The given fastest wind speed indicates that wind is constant in 8 hours,
the minimum necessary wind duration is
t
min
= 68.8
g F
U
2
A
2/3
U
A
g
= 23688 s = 6.6 hours < 8 hours
Therefore it is fetchlimited condition.
Because
g F
U
2
A
= 1568 < 23123
it is not fully arisen sea.
7. H
m0
and T
p
are given by
H
m0
= 0.0016
g F
U
2
A
1/2
U
2
A
g
= 4.04 m
T
p
= 0.2857
g F
U
2
A
1/3
U
A
g
= 8.47 s
33
3 WINDGENERATED WAVES
3.3 Standard variance spectrum
PM spectrum
In 1964, W.J.Pierson and L.Moskowitz put forward, on the basis of a similarity theory
by S.A.Kitaigorodskii, some suggestions for deep water wave spectra for the sea state
referred to as fully arisen sea. This wave condition refers to the case where the waves
have reached an equilibrium state in which energy input from the wind is exactly balanced
by energy loss. The only variable is thus the wind velocity. It is important to emphasize,
that spectra of this type are only valid when the fetches are large enough to reach this
equilibrium.
Out of the three analytical expressions suggested by Pierson and Moskowitz, the one
below was found to give the best agreement with empirical wave data. This spectrum is
called PM spectrum.
S
η
(f) =
αg
2
(2π)
4
f
−5
exp
¸
−0.74
f
0
f
4
¸
(313)
α = 0.0081
f
0
= g (2πU
19.5
)
−1
U
19.5
: Wind speed, 19.5 m above mean water level
g : Gravitational acceleration
PM spectrum has been transformed to parameterized spectrum by H
s
= 4
√
m
0
and
T
p
= 1.4 T = 1.4
m
0
m
1
S
η
(f) =
5
16
H
2
s
f
4
p
f
−5
exp
¸
−
5
4
f
p
f
4
¸
(314)
Fig.3. Example of PM spectrum.
34
3 WINDGENERATED WAVES
JONSWAP spectrum
The Joint North Sea Wave Project (JONSWAP) was started in 1967 as a collaboration
among institutes in Germany, Holland, UK and USA. The objectives of the project was
originally partly to investigate the growth of waves under fetchlimited condition, and
partly to investigate wave transformation from sea to shallow water area. Simultaneous
measurements of waves and winds were taken at stations along a line extending 160 km
in a westerly direction from the island of Sylt in the Germany Bright.
During the processing of a large number of spectra corresponding to steady easterly wind,
the socalled JONSWAP spectrum was obtained
S
η
(f) =
αg
2
(2π)
4
f
−5
exp
¸
−
5
4
f
f
m
−4
¸
γ
exp
−
1
2σ
2
(
f
fm
−1)
2
(315)
where α = 0.076 x
−0.22
x = g F U
−2
10
f
m
=
3.5 g x
−0.33
U
10
σ = 0.07 f ≤ f
p
σ = 0.09 f > f
p
γ : Peak enhancement coeﬃcient
U
10
: Wind speed, 10 m above mean water level
The parameterized JONSWAP spectrum reads
S
η
(f) = α H
2
s
f
4
p
f
−5
γ
β
exp
¸
−
5
4
f
p
f
4
¸
(316)
α ≈
0.0624
0.230 + 0.0336 γ −
0.185
1.9 + γ
β = exp
−
(f − f
p
)
2
2 σ
2
f
2
p
σ ≈ 0.07 f ≤ f
p
σ ≈ 0.09 f ≥ f
p
γ: Peak enhancement coeﬃcient
35
3 WINDGENERATED WAVES
The JONSWAP spectrum is characterized by a parameter γ, the socalled peak enhance
ment parameter, which controls the sharpness of the spectral peak, cf. Fig.4. In the
North Sea the γ value ranges from 1 to 7 with the mean value 3.3.
Fig.4. Example of JONSWAP spectrum.
Remarks on standard spectra
Actually wave spectra usually exhibit some deviations from these standard spectra. Con
cerning the spectrum of swell, the available information is insuﬃcient because many swell
records are contaminated by local wind waves. Ochi et al. (1976) presented a spectrum
which has two peaks, one associated with swell and the other with locally generated waves.
One of the few reports on pure swell spectra indicates that it can be approximately de
scribed by the JONSWAP spectrum with relatively larger γ value (Goda 1985). There
are still other standard spectra, e.g. Bretsneider (1959), Darbyshire (1952), Scott (1965),
Mitsuyasu (1971, 1972) and the ISSC spectrum.
Furthermore, the above mentioned spectra are onedimensional and valid only for deep
water. With respect to shallow water wave spectrum and directional spectrum, many
researches are ongoing.
36
3 WINDGENERATED WAVES
3.4 Introduction to shortcrested waves
The directional wave spectrum, S
η
(f, θ), is considered a product of the unidirectional
spectrum, S
η
(f), and a spreading function, D(f, θ), i.e.
S
η
(f, θ) = D(f, θ) ×S
η
(f) (317)
where f is the wave frequency, θ the angle of wave propagation. D(f, θ) must satisfy
π
−π
D(f, θ) dθ = 1 (318)
to assure identical wave energy in S
η
(f, θ) and S
η
(f), i.e. the same zero moment
m
0
=
∞
0
π
−π
S
η
(f, θ) dθ df =
∞
0
S
η
(f)
π
−π
D(f, θ) dθ
df
=
∞
0
S
η
(f) df (319)
An example of S
η
(f, θ) is given in Fig. 5.
Fig.5. An example of directional wave variance spectrum.
For simplicity, the spreading function D(f, θ) is treated as a function of only θ. There are
mainly two types of spreading functions, one using s as a spreading parameter and the
other using σ as a spreading parameter.
37
3 WINDGENERATED WAVES
The cosinepower spreading function proposed by LonquetHiggins is using s as a spread
ing parameter
D(θ) =
2
2s−1
π
Γ
2
(s + 1)
Γ(2s + 1)
cos
2s
θ − θ
m
2
, −π ≤ (θ −θ
m
) ≤ π (320)
where s is a spreading parameter, Γ the Gamma function and θ
m
the mean wave incident
angle. The equation was found to provide a reasonable ﬁt to measured ocean wave spectra
by LonquetHiggins, Cartwright and Smith (1961).
In Fig. 6 the spreading function D(θ) is plotted as a function of (θ − θ
m
) for various
values of the spreading parameter s.
Fig.6. The spreading function D(θ) by LonquetHiggins.
With the spreading parameter σ the spreading function is assumed to follow the Gaussian
distribution
D(θ) =
1
√
2π σ
exp
−
(θ −θ
m
)
2
2σ
2
, −π ≤ (θ −θ
m
) ≤ π (321)
It can be seen that σ is the standard deviation of the spreading function, which, in this
case, is angle in degree. Therefore σ is also called spreading angle.
38
3 WINDGENERATED WAVES
3.5 References
Burcharth, H.F. and Brorsen, M. , 1978. On the design of gravity structures using wave
spectra. Lecture on Oﬀshore Engineering, Edited by W.J.Graﬀ and P. Thoft
Christensen, Institute of Building Technology and Structural Engineering, Aalborg
University, Denmark.
Goda, Y. , 1985. Random seas and design of marine structures. University of Tokyo
Press, Japan, 1985
Sarpkaya, T. and Isaacson, M. , 1981. Mechanics of wave forces on oﬀshore structures.
Van Nostrand Reinhold Company, New York, ISBN 0442254022, 1981.
SPM , 1984. Shore Protection Manual. Coastal Engineering Research Center, Waterway
Experiment Station, US Army Corps of Engineers, 1984.
3.6 Exercise
1) The wind speed measured at the elevation of 5 m is U
5
= 20 m/s. Calculate
U
19.5
to be used in the PMspectrum.
2) Convert the fastest mile wind speed U
f
= 29 m/s to twentyﬁveminute
average wind speed U
t=25 min
.
3) Calculate H
m
0
and T
p
with
deep water situation, fetch 200 kilometers, wind speed at z = 5 m over
water surface is 20 m/s over 2 hours.
39
4 EXTREME WAVE HEIGHT ANALYSIS
4 Extreme wave height analysis
The design wave height is often represented by signiﬁcant wave height. Signiﬁcant wave
height is a random variable. It varies with respect to time and location. If a structure
is to be built in a location of sea where a longterm wave height measurement/hindcast
is available, the question an engineer must answer is: How to determine the design wave
height ?
Extreme wave height analysis gives the answer to that question, i.e. it is a method to
determine the design wave height, based on the importance of the structure (design level)
and the statistical analysis of a longterm wave height measurements/hindcast.
4.1 Design level: Return period and encounter probability
The design level is represented by return period or encounter probability.
Return period T
To deﬁne return period the following notations are used
X Signiﬁcant wave height, which is a random variable due to the statis
tical vagrancy of nature.
x Realization of X.
F(x) Cumulative distribution function of X, F(x) = Prob(X ≤ x).
t Number of years of observation of X.
n Number of observations in a period of t.
λ Sample intensity, λ = n/t.
Fig.1 illustrates the cumulative distribution function of X. The nonexceedence probabil
ity of x is F(x), or the exceedence probability of x is (1 − F(x)). In other words with
(1 − F(x)) probability an observed signiﬁcant wave height will be larger than x.
Fig.1. Cumulative distribution function of X.
If the total number of observations is n, The number of observations where (X > x) is
k = n ( 1 − F(x) ) = t λ ( 1 − F(x) ) (41)
40
4 EXTREME WAVE HEIGHT ANALYSIS
The return period T of x is deﬁned as
T = t
k=1
=
1
λ ( 1 − F(x) )
(42)
i.e. on average x will be exceeded once in every T years. x is also called Tyear event.
Encounter probability p
Based on the fact that on average x will be exceeded once in every T years, the exceedence
probability of x in 1 year is 1/T. Therefore
nonexceedence probability of x in 1 year Prob(X ≤ x) = 1 −
1
T
nonexceedence probability of x in 2 years Prob(X ≤ x) =
1 −
1
T
2
nonexceedence probability of x in L years Prob(X ≤ x) =
1 −
1
T
L
and the encounter probability, i.e. the exceedence probability of x within a structure
lifetime of L years is
p = 1 −
1 −
1
T
L
(43)
which in the case of larger T can be approximated
p = 1 − exp
−
L
T
(44)
Design level
Traditionally the design level for design wave height was the wave height corresponding
to a certain return period T. For example, if the design wave height corresponding to
a return period of 100 years is 10 m, the physical meaning is that on average this 10 m
design wave height will be exceeded once in every 100 years.
In the reliability based design of coastal structures it is better to use encounter probability,
i.e. the exceedence probability of the design wave height within the structure lifetime.
For example If the structure lifetime L is 25 years, the encounter probability of the design
wave height (10 meter) is
p = 1 −
1 −
1
T
L
= 22%
This means that this 10 m design wave height will be exceeded with 22% probability
within a structure lifetime of 25 years.
41
4 EXTREME WAVE HEIGHT ANALYSIS
4.2 General procedure
In practice engineers are often given a longterm signiﬁcant wave height measurement/hindcast
and required to determine the design wave height corresponding to a certain return period.
The general procedure to perform the task is:
1) Choice of the extreme data set based on a longterm wave height mea
surement/hindcast
2) Choice of several theoretical distributions as the candidates for the ex
treme wave height distribution
3) Fitting of the extreme wave heights to the candidates by a ﬁtting method.
If the least square ﬁtting method is employed, a plotting position formula
must be used
4) Choice of the distribution based on the comparison of the ﬁtting goodness
among the candidates
5) Calculation of the design wave height corresponding to a certain return
period
6) Determination of the conﬁdence interval of the design wave height in
order to account for sample variability, measurement/hindcast error and
other uncertainties
If structure lifetime and encounter probability are given in stead of return period, we can
calculate the return period by eq (43) and proceed as above.
If the followings will be discussed the procedures one by one.
4.3 Data sets
The original wave data are typically obtained either from direct measurements or from the
hindcasts based on the meteorological information. Most of the measurements/hindcasts
cover a rather short span of time, say less than 10 years in the case of direct measurements
and less than 40 years in the case of hindcasts.
In practice three kinds of extreme data sets have been used.
Complete data set containing all the direct measurements of wave height
usually equally spaced in time.
Annual series data set consisting of the largest wave height in each year of
measurements/hindcasts, cf. Fig.2.
Partial series data sets composed of the largest wave height in each individ
ual storm exceeding a certain level (threshold). The
threshold is determined based on the structure loca
tion and engineering experience, cf. Fig.2. It is also
called POT data set (Peak Over Threshold).
42
4 EXTREME WAVE HEIGHT ANALYSIS
Fig.2. Illustration of the establishment of annual series data set and partial series
data set.
The extreme data sets, established based on the original wave data, should fulﬁll the
following 3 conditions:
Independence There must be no correlation between extreme data. The annual
series data set and the partial series data set meet the indepen
dence requirement because the extreme data are from diﬀerent
storms.
Homogeneity The extreme data must belong to the same statistical population,
e.g. all extreme data are from windgenerated waves.
Stationary There must be stationary longterm climatology. Studies of wave
data for the North Sea from the last 20 years give evidence of
nonstationarity as they indicate a trend in the means. Average
variations exist from decades to decades or even longer period of
time. However, until more progress is available in investigating
longterm climatological variations, the assumption of stationary
statistics might be considered realistic for engineering purpose,
because the longterm climatological variation is generally very
weak.
The complete data set cannot fulﬁll the required independence between data. Goda
(1979) found correlation coeﬃcients of 0.3−0.5 for signiﬁcant wave heights (measurement
duration is 20 minutes and time interval between two succeeding measurements is 24
hours). Moreover, what is interesting in the case of design waves is the wave height
corresponding to a very high nonexceedence probability, i.e. the very upper tail of the
distribution. If the chosen distribution is not the true one, the very upper tail value will
be distorted severely because in the ﬁtting process the chosen distribution will be adjusted
to the vast population of the data. For these reasons the complete data set is seldom used.
Most engineers prefer the partial series data set over the annual series data set simply
because the former usually gives larger design wave height and hence, more conservatively
designed structures.
43
4 EXTREME WAVE HEIGHT ANALYSIS
4.4 Candidate distributions
Generally the exponential distribution, the Weibull distribution, the Gumbel (FTI) dis
tribution, the Frechet distribution and the Lognormal distribution are the theoretical
distributions which ﬁt the extreme wave data well.
Exponential F = F
X
(x) = P (X < x) = 1 −e
−(
x−B
A
)
(45)
Weibull F = F
X
(x) = P (X < x) = 1 −e
−(
x−B
A
)
k
(46)
Gumbel F = F
X
(x) = P (X < x) = e
−e
−
(
x−B
A
)
(47)
Frechet F = F
X
(x) = P (X < x) = e
−(
x
A
)
k
(48)
Lognormal F = F
X
(x) = P (X < x) = Φ
ln(x) − B
A
(49)
where X A characteristic wave height, which could be the sig
niﬁcant wave height H
s
or the onetenth wave height
H 1
10
or the maximum wave height H
max
, depending
on the extreme data set.
x Realization of X.
F Nonexceedence probability of x (cumulative fre
quency).
A, B, k Distribution parameters to9 be ﬁtted. In the log
normal distribution A and B are the standard devia
tion and the mean of X respectively.
Φ Standard normal distribution function.
No theoretical justiﬁcation is available as to which distribution is to be used. The author
have tried to ﬁt 7 sets of partial series data to all these distributions. These data sets
are real data representing deep and shallow water sea states from Bilbao in Spain, Sines
in Portugal, the North Sea, Tripoli in Libya, Pozzallo and Follonica in Italy and Western
Harbour in Hong Kong. The results show that the Weibull and the Gumbel distributions
provide the closest ﬁts. Therefore the following discussion is exempliﬁed with these two
distributions.
44
4 EXTREME WAVE HEIGHT ANALYSIS
4.5 Fitting methods and procedure
Four generally applied methods of ﬁtting the extreme data set to the chosen distributions
are the maximum likelihood method, the method of moment, the least square method
and the visual graphical method. The most commonly used methods are the maximum
likelihood method and the least square method.
Least square method
Eqs (46) and (47) can be rewritten as
X = A Y + B (410)
where Y is the reduced variate deﬁned according to the distribution function
Y = (−ln(1 −F))
1
k
Weibull distribution (411)
Y = −ln(−lnF) Gumbel distribution (412)
The ﬁtting procedure is summarized as the follows:
1) Rearrange the measured/hindcast extreme data (total number n) in
the descending order, (x
i
), i = 1, 2, · · · , n (X
1
=max).
2) Assign a nonexceedence probability F
i
to each x
i
by an appropriate
plotting position formula (cf. next section), thus obtaining a set of
data pairs, (F
i
, x
i
), i = 1, 2, · · · , n.
3) Calculate the corresponding y value by eq (411) or eq (412), thus
obtaining a new set of data pairs, (y
i
, x
i
), i = 1, 2, · · · , n.
4) Determine the regression coeﬃcients of eq (410) by
A =
Cov(Y, X)
V ar(Y )
B = X −AY
V ar(Y ) =
1
n
n
¸
i=1
y
i
−Y
2
Cov(Y, X) =
1
n
n
¸
i=1
(y
i
−Y )(x
i
−X)
Y =
1
n
n
¸
i=1
y
i
X =
1
n
n
¸
i=1
x
i
In the case of the Weibull distribution various k values are predeﬁned and A and B are
ﬁtted accordingly. The ﬁnal values of the three parameters are chosen based on the ﬁtting
45
4 EXTREME WAVE HEIGHT ANALYSIS
goodness.
Maximum likelihood method
The 2parameter Weibull distribution is
Weibull F(x) = 1 −e
−
x−x
A
k
(413)
where x
is the threshold wave height, which should be smaller than the minimum wave
height in the extreme data set. For unexperienced engineers several threshold values can
be tried, and the one which produces best ﬁt is ﬁnally chosen.
the maximum likelihood estimate k is obtained by solving the following equation by an
iterative procedure:
N + k
N
¸
i=1
ln (x
i
− x
) = N k
N
¸
i=1
(x
i
− x
)
k
ln (x
i
− x
)
N
¸
i=1
(x
i
− x
)
k
−1
(414)
The maximum likelihood estimate of A is
A =
¸
1
N
N
¸
i=1
(x
i
−x
)
k
¸
1/k
(415)
For the Gumbel distribution, the maximum likelihood estimate of A is obtained by solving
the following equation by an iterative procedure:
N
¸
i=1
x
i
exp
−
x
i
A
=
¸
1
N
N
¸
i=1
x
i
−A
¸
N
¸
i=1
exp
−
x
i
A
(416)
The maximum likelihood estimate of B is
B = A ln
N
N
¸
i=1
exp
−
x
i
A
−1
¸
¸
(417)
46
4 EXTREME WAVE HEIGHT ANALYSIS
4.6 Plotting position formulae
When the least square method is applied a plotting position formula must be chosen. The
plotting position formula is used to assign a nonexceedence probability to each extreme
wave height. The plotting position is of special importance when dealing with very small
samples.
The nonexceedence probability (F
i
) to be assigned to (x
i
), can be determined based on
three diﬀerent statistical principles, namely sample frequency, distribution of frequency
and order statistics, cf Burcharth et al (1994).
Mean, median and mode
The deﬁnition of mean, median and mode of a random variable X is given in the following
because they are involved in some of the plotting position formulae.
Take the Gumbel distribution as an example. The distribution function F
X
(x) and density
function f
X
(x) of a Gumbel random variable X reads
F
X
(x) = P(X < x) = e
−e
−
(
x−B
A
)
f
X
(x) =
dF
X
(x)
dx
(418)
The deﬁnition and value of the mean, the median and the mode are
Mean x
mean
= E[X] =
+∞
−∞
xf
X
(x)dx ≈ B + 0.577A (419)
Median x
median
= x
F
X
(x)=0.5
= B + 0.367A (420)
Mode x
mode
= x
f
X
(x)=max
= B (421)
Fig. 3. Mean, median and mode of the Gumbel random variable.
47
4 EXTREME WAVE HEIGHT ANALYSIS
Order statistics
Assume that a random variable X has a cumulative distribution function F
X
, and prob
ability density function f
X
, i.e.
F
X
(x) = P(X < x) (422)
Furthermore, assume n data sampled from X and arranged in the descending order, x
1
being the largest value in n data.
Here x
1
is one realization of the ordered random variable X
1
, deﬁned as the largest value
in each sample. The distribution function of X
1
is
F
X
1
(x) = P(X
1
< x) = (F
X
(x))
n
= (P(X < x))
n
(423)
F
X
1
(x) may also be interpreted as the probability of the nonoccurrence of the event
( X > x ) in any of n independent trials.
The density function of X, f
X
, and the density function of X
1
, f
X
1
, are sketched in Fig.4.
Fig. 4. f
X
: Density function of X. f
X
1
: density function of X
1
.
For other ordered random variables X
i
, i = 2, 3, · · · , n, The distribution functions
F
X
i
(x) can also be expressed as the function of F
X
(x), cf. ThoftChristensen et al.(1982).
48
4 EXTREME WAVE HEIGHT ANALYSIS
Plotting position based on sample frequency
This method is based solely on the cumulative frequency of the samples. The widely used
formula is the socalled California plotting position formula
F
i
= 1 −
i
n
i = 1, 2, · · · , n (424)
where x
i
Extreme data in the descending order (x
1
=max)
F
i
Nonexceedence probability of x
i
.
n Sample size, i.e. total data number.
The disadvantage of this plotting position formula is that the smallest extreme data x
n
cannot be used because F
n
= 0.
Plotting position based on distribution of frequencies
Assume that the random variable X has a cumulative distribution function F
X
. The
i’th highest value in n samples, X
i
, is a random variable, too. Consequently, F
X
i
(x
i
), the
cumulative frequency of x
i
, is a random variable, too. The philosophy of this method is to
determine the plotting position of x
i
via either the mean, the median or the mode of the
random variable F
X
i
(x
i
). The plotting position formula by this method is independent of
the parent distribution (distributionfree).
Weibull (1939) used the mean of F
X
i
(x
i
) to determine the cumulative frequency F
i
to be
assigned to x
i
Weibull F
i
== 1 −
i
n + 1
(425)
There is no explicit formula for the median of F
X
i
(x
i
). However, Benard (1943) developed
a good approximation
Benard F
i
≈ 1 −
i −0.3
n + 0.4
(426)
The plotting position formula based on the mode of F
X
i
(x
i
) has not drawn much attention,
because the chance of the occurrence of mode is still inﬁnitesimal even though mode is
more likely to occur than the mean and median,
49
4 EXTREME WAVE HEIGHT ANALYSIS
Plotting position based on order statistics
The philosophy of this method is to determine the plotting position of x
i
via the mean,
the median and the mode of the ordered random variable X
i
.
Fig. 5. Illustration of the determination of F
1
based on the mean, the median
and the mode of X
1
.
Plotting positions based on the mean value are distributiondependent and not explicitly
available. The best known approximations are
Blom F
i
= 1 −
i−3/8
n+1/4
Normal distribution (427)
Gringorten F
i
= 1 −
i−0.44
n+0.12
Gumbel distribution (428)
Petrauskas F
i
= 1 −
i−0.3−0.18/k
n+0.21+0.32/k
Weibull distribution (429)
Goda F
i
= 1 −
i−0.2−0.27/
√
k
n+0.20+0.23/
√
k
Weibull distribution (430)
The plotting position based on the median value of the ordered random variable is the
same as that based on the median value of distribution of frequency.
Summary on plotting position formulae
The choice of the plotting position formula depends on engineer’s personnel taste.
From the statistical point of view the plotting position formula based on the mean (un
biased) is preferred because the expected squared error is minimized. Rosbjerg (1988)
advocates the choice of the median plotting position formula (Benard formula) because
it is distributionfree and is based both on the distribution of frequency and the order
statistics. In practice the Weibull plotting position formula is most widely used.
50
4 EXTREME WAVE HEIGHT ANALYSIS
4.7 Fitting goodness
Normally several candidate distributions will be ﬁtted and the best one is chosen. The
linear correlation coeﬃcient, deﬁned as
ρ =
Cov(X, Y )
V ar(X) V ar(Y )
(431)
is widely used as the criterion for the comparison of the ﬁtting goodness. However, ρ is
deﬁned in the linear plotting domain (y, x), where the reduced variate y is dependent on
the distribution function. Therefore, the interpretation of this criterion is less clear.
With the ﬁtted distribution functions, the wave heights corresponding to the nonexceedence
probability of the observed wave heights can be calculated, cf. eqs (433) and (434). The
average relative error E, deﬁned as
E =
1
n
n
¸
i=1
 x
i,estimated
−x
i,observed

x
i,observed
(432)
is a good simple criterion with a clear interpretation. E = 5 % means that on the
average, the central estimation of wave height deviates from the observed wave height by
5 %. Obviously a smaller Evalue indicates a better ﬁtted distribution.
The statistical hypothesis test can also be used in the comparison of the ﬁtting goodness
(Goda et al. 1990)
4.8 Design wave height: x
T
The design wave height x
T
is the wave height corresponding to the return period T.
The Weibull and Gumbel distributions, eqs (46) and (47), are rewritten as
x = A(−ln(1 − F))
1
k
+ B Weibull distribution (433)
x = A(−ln(−ln(F))) + B Gumbel distribution (434)
Deﬁne the sample intensity λ as
λ =
number of extreme data
number of years of observation
(435)
51
4 EXTREME WAVE HEIGHT ANALYSIS
and employ the deﬁnition of return period T
T =
1
λ ( 1 − F )
or F = 1 −
1
λT
(436)
Inserting eq (436) into eqs (433) and (434), we get (now x means the wave height
corresponding to return period T, and therefore is replaced by x
T
)
x
T
= A
−ln(
1
λT
)
1
k
+ B Weibull distribution (437)
x
T
= A
−ln(−ln(1 −
1
λT
))
+ B Gumbel distribution (438)
where A, B and k are the ﬁtted distribution parameters.
Remark
Some students have a confusion between the shortterm distribution of individual wave
heights and the longterm distribution of extreme wave heights. The confusion can be
cleared by the following ﬁgure.
52
4 EXTREME WAVE HEIGHT ANALYSIS
4.9 Example
Delft Hydraulics Laboratory performed a hindcast study for the Tripoli deep water wave
climate and identiﬁed the 17 most severe storms in a period of 20 years. The ranked
signiﬁcant wave heights are listed in Table 2.
Table 2. Tripoli storm analysis
Signiﬁcant nonexceedence Reduced variate Reduced variate
rank wave height probability y
i
y
i
i x
i
(m) F
i
Gumbel Weibull (k = 2.35)
1 9.32 0.944 2.86 1.57
2 8.11 0.889 2.14 1.40
3 7.19 0.833 1.70 1.28
4 7.06 0.778 1.38 1.18
5 6.37 0.722 1.12 1.11
6 6.15 0.667 0.90 1.04
7 6.03 0.611 0.71 0.98
8 5.72 0.556 0.53 0.92
9 4.92 0.500 0.37 0.86
10 4.90 0.444 0.21 0.80
11 4.78 0.389 0.06 0.74
12 4.67 0.333 0.09 0.68
13 4.64 0.278 0.25 0.62
14 4.19 0.222 0.41 0.55
15 3.06 0.167 0.58 0.49
16 2.73 0.111 0.79 0.40
17 2.33 0.056 1.06 0.30
You are required to ﬁnd the design wave height which has 22% exceedence probability
within a structure lifetime of 25 years.
The steps in the analysis are as follows:
1) Calculate the sample intensity by eq (435) λ =
17
20
2) Calculate the return period by eq (43) T = 100 years
3) Assign a nonexceedence probability F
i
to each observed wave height x
i
according to the Weibull plotting position formula. Results are shown in
Table 2.
4) Choose the Weibull and the Gumbel distributions as the candidate distri
butions. Calculate the values of the reduced variate {y
i
} according to eqs
(411) and (412) respectively. For the Weibull distribution {y
i
} involves
the iterative calculation. {y
i
} of the two distributions are also shown in
Table 2.
53
4 EXTREME WAVE HEIGHT ANALYSIS
5) Fit data (y
i
, x
i
) to eq (410) by the least square method and obtain the
distribution parameters:
Weibull, k = 2.35, A = 5.17, B = 0.89
Gumbel, A = 1.73, B = 4.53
The ﬁtting of the data to the Gumbel and the Weibull distributions is shown
in Fig. 6.
6) Compare the goodness of ﬁtting according to the value of the average rela
tive error E, eq (432)
E = 4.72 % for the Weibull distribution ﬁtting
E = 6.06 % for the Gumbel distribution ﬁtting
Because of a clearly smaller Evalue the Weibull distribution is taken as the
representative of the extreme wave height distribution
7) Calculate the wave height corresponding to a return period of 100 years
X
100
by eq (437) x
100
= 10.64 m
Fig. 6. Fitting to the Gumbel and the Weibull distributions and comparison.
54
4 EXTREME WAVE HEIGHT ANALYSIS
4.10 Sources of uncertainties and conﬁdence interval
Sources of uncertainties
The sources of uncertainty contributing to the uncertainty of the design wave height are:
1) Sample variability due to limited sample size.
2) Error related to measurement, visual observation or hindcast.
3) Choice of distribution as a representative of the unknown true long
term distribution
4) Variability of algorithms (choice of threshold, ﬁtting method etc.
5) Climatological changes
The uncertainty sources 1) and 2) can be considered by numerical simulation in the
determination of the design wave height.
Wave data set contains measurement/hindcast error. Measurement error is from mal
function and nonlinearity of instruments, such as accelerometer and pressure cell, while
hindcast error occurs when the sealevel atmospheric pressure ﬁelds are converted to wind
data and further to wave data. The accuracy of such conversion depends on the quality
of the pressure data and on the technique which is used to synthesize the data into the
continues wave ﬁeld. Burcharth (1986) gives an overview on the variational coeﬃcient C
(standard deviation over mean value) of measurement/hindcast error.
Visual observation data should not be used for determination of design wave height be
cause ships avoid poor weather on purpose. With the advance of measuring equipment
and numerical model, generally C value has been reduced to below 0.1.
Table 1. variational coeﬃcient of extreme data C
Methods of Accelerometer Horizontal radar Hindcast Hindcast Visual
determination Pressure cell by SPM other
Vertical radar
Variational Coe. C 0.050.1 0.15 0.120.2 0.10.2 0.2
Conﬁdence interval of design wave height x
T
We use an example to demonstrate how the conﬁdence interval of the design wave height
is determined. The gumbel distribution curve in Fig.7 is obtained by ﬁtting Tripoli
signiﬁcant wave height to Gumbel distribution by the least square ﬁtting method and the
Weibull plotting position formula.
55
4 EXTREME WAVE HEIGHT ANALYSIS
Fig.7. Design wave height.
If the design level for design wave height is a return period of 100 years, i.e. T = 100, the
design wave height is x
100
= 12.2 m.
If other uncertainties, e.g. sample variability, is included, the design wave height x
100
becomes a random variable. The distribution of the design wave height x
100
, which is
usually assumed to follow the normal distribution, can be obtained by numerical simula
tion to be described in the next section, cf. Fig.7. In order to account sample variability,
a conﬁdence band is often applied. For example, the design wave height is 14.8 m which
corresponds to the 90% onesided conﬁdence interval, cf. Fig.7.
Numerical simulation
To exemplify the discussion, it is assumed that the extreme wave height follows the Gum
bel distribution
F = F
X
(x) = P (X < x) = exp
−exp
−(
x −B
A
)
(439)
where X is the extreme wave height which is a random variable, x a realization of X, A
and B the distribution parameters.
Due to the sample variability and measurement/hindcast error, the distribution parameter
A and B become random variables,
In order to account the sample variability and measurement/hindcast error, a numerical
simulation is performed as explained in the followings.
A sample with size N is ﬁtted to the Gumbel distribution. The obtained distribution
parameters A
true
and B
true
are assumed to be the true values.
56
4 EXTREME WAVE HEIGHT ANALYSIS
1) Generate randomly a data between 0 and 1. Let the nonexceedence
probability F
1
equal to that data. the single extreme data x is
obtained by (cf. Fig.8)
x = F
−1
X
(F
1
) = A
true
[−ln (−ln F
1
)] + B
true
(440)
2) Repeat step 1) N times. Thus we obtain a sample belonging to the
distribution of eq (439) and the sample size is N.
3) Fit the sample to the Gumbel distribution and get the new esti
mated distribution parameters A and B.
4) Calculate the wave height x
T
corresponding to the return period T
by eq (438)
5) Repeat steps 2) to 4), say, 10,000 times. Thus we get 10,000 values
of x
T
.
6) Choose the wave height corresponding to the speciﬁed conﬁdence
band.
In order to include the measurement/hindcast error the following step can be added after
step 1). This step is to modify each extreme data x generated by step 1), based on the
assumption that the hindcast error follows the normal distribution, cf. Fig.8
1
∗
) Generate randomly a data between 0 and 1. Let the nonexceedence
probability F
2
equal to that data. the modiﬁed extreme data x
modiﬁed
is obtained by
x
modiﬁed
= x + C x Φ
−1
(F
2
) (441)
where Φ is the standard normal distribution and C is the coeﬃcient
of variation of the measurement/hindcast error. C ranges usually
from 0.05 to 0.1.
Fig.8. Simulated wave height taking into account measurement/hindcast error.
57
4 EXTREME WAVE HEIGHT ANALYSIS
Example
Again the Tripoli deep water wave data is used as an example to demonstrate the deter
mination of the design wave height and the inﬂuence of sample variability.
By ﬁtting the extreme data to Gumbel distribution we obtain the distribution parameters
A = 1.73 and B = 4.53, cf. Fig.9. The design wave height corresponding to a return
period of 100 years is 12.2 m.
Fig.9. Simulated distribution of x
100
(sample variability).
If sample variability is included, the design wave height x
100
becomes a random variable.
The distribution of the design wave height x
100
can be obtained by numerical simulation,
cf. Fig.9. In order to account sample variability, an 80% conﬁdence band is often applied.
In the case of wave height estimate, onesided conﬁdence interval is preferred over two
sided conﬁdence interval because the lower bound of the conﬁdence band is of less interest.
Therefore, the design wave height is 14.8 m which corresponds to 90% onesided conﬁdence
interval.
58
4 EXTREME WAVE HEIGHT ANALYSIS
4.11 Physical consideration of design wave height
Wave breaking
The design wave height must be checked against wave breaking condition. Wave breaking
occurs due to wave steepness (Stokes wave theory) or limited water depth (Solitary wave
theory). Based on laboratory and ﬁeld observations, many empirical formulae for wave
breaking condition have been proposed, e.g. Goda (1985).
Structural response characteristics
The choice of design wave height depends not only on the structure life time, but also on
the character of the structural response.
Fig.10 indicates as an example the diﬀerences in armour layer damage development for
various types of rubble structures. The ﬁgure illustrates the importance of evaluation of
prediction and conﬁdence limits related to the estimated design wave height, especially in
case of structures with brittle failure characteristics. To such cases a lower damage level
must be chosen for the mean value design sea state. The ﬁgure is illustrative. In reality
also the conﬁdence bands for the damage curves should be considered.
Fig.10. Illustration of typical armour layer failure characteristics for
various types of rubble mound structures (Burcharth 1993).
59
4 EXTREME WAVE HEIGHT ANALYSIS
4.12 Wave period
There is no theory to determine the design wave period corresponding to the design wave
height obtained by the extreme analysis, due to the complexity and locality of the joint
distribution between wave height and wave period.
Fig.11 shows examples of scatter diagrams representing the joint distribution of signiﬁcant
wave height, H
s
, and mean wave period, T
m
, and still water level, z, respectively. The
numbers in the scatter diagrams are the number of observations falling in the correspond
ing predeﬁned intervals of H
s
, T
m
and z.
Fig.11. Scatter diagrams signifying examples of joint distributions of
H
s
and T
m
, and H
s
and water level, z.
In practice, several wave periods within a realistic range are simply assigned to the design
wave height to form the candidates of the design sea state conditions. Then by theoretical
consideration and/or laboratory investigation, the one which is most dangerous is chosen.
DS449 gives the range of peak wave period
130 H
s
g
< T
p
<
280 H
s
g
(442)
60
4 EXTREME WAVE HEIGHT ANALYSIS
4.13 Water level
The sea water level is aﬀected by the following eﬀects:
1) Astronomical eﬀect: Tides generated by the astronomical aspect is the best
understood due to their extreme regularity and the simplicity of observa
tions. At a site without any previous tidal records usually one or a few
month of recording will be suﬃcient to analyze the astronomical eﬀect on
the water level. The astronomical tidal variations can be found in the Ad
miralty Tide Tables.
2) Meteorological eﬀect: In shallow water the water level is also aﬀected by
the meteorological eﬀects, namely,
i) Barometric: The higher barometric pressure causes a lower water level
and vise versa.
ii) Wind: Strong wind creates a setup of the water level on the downwind
side and a setdown on the upwind side.
It is diﬃcult to determine the meteorological eﬀect on the water level. If
water level records are available for a long period of time, the meteorological
eﬀect can be isolated from the astronomical eﬀect and subjected to the
extreme analysis in order to establish the longterm statistics of the water
level. If such records are not available, numerical models can, using wind
and/or barometric chart, give reliable results.
3) Earthquake
The water depth read from the Chart Datum is the one corresponding to the Lowest
Astronomical Tide, which is the lowest tide level under the average meteorological con
ditions, cf. Fig.12, which gives also the widely used terminology and abbreviation of the
various sea water levels.
The extreme analysis should be performed on both the high water level and the low water
level. Based on the established longterm statistics is given the design low water level and
the design high water level.
Fig.12. Water depth.
61
4 EXTREME WAVE HEIGHT ANALYSIS
4.14 Multiparameter extreme analysis
A sea state should be characterized at least by some characteristic values of wave height
(e.g. H
s
), wave period (e.g. T
m
), the wave direction, and the water level, because these
four parameters are the most important for the impacts on the structures. Of importance
is also the duration of the sea state and sometimes also the shape (type) of the wave
spectrum.
When more sea state parameters have signiﬁcant inﬂuence on the impact on the structure
considerations must be given to the probability of occurrence of the various possible
combinations of the parameter values.
Burcharth (1993) proposed the following principle for multiparameter extreme analysis:
For the general case where several variables are of importance but the correlation coeﬃ
cients are not known the best joint probability approach would be to establish a longterm
statistics for the response in question, e.g. for the runup, the armour unit stability, the
wave force on a parapet wall, etc.
If we assume that the variables of importance are H
s
, T
m
, α (wave direction) and z (water
level) then by hindcasting or/and measurements several data sets covering some years can
be established
(H
s,i
, T
m,i
, α
i
, z
i
) , i = 1, 2, . . . n
For each data set the response in question is either calculated from formulae or determined
by model tests. If for example runup, R
u
, is in question a single variable data set is
obtained
(R
u,i
) , i = 1, 2, . . . n
The related longterm statistics can be established by ﬁtting to a theoretical extreme
distribution (extreme analysis).
62
4 EXTREME WAVE HEIGHT ANALYSIS
4.15 References
Benard, L.R. , 1943. Statistical analysis in hydrology. Trans. Am. Soc. Civ. Eng., 108,
pp 11101160
Burcharth, H.F. , 1986. On the uncertainties related to the estimation of extreme environ
mental conditions. Proceeding of Seminar on Uncertainties Related to the Design
and Construction of Oﬀshore Jacket Structures, Copenhagen, 1986, Published by
Danish Society of Hydraulic Engineering
Burcharth, H.F. , 1993. The design of breakwaters. Department of Civil Engineering,
Aalborg University, Denmark, 1993
Burcharth, H.F. and Zhou Liu , 1994. On the extreme wave height analysis. Proceedings
of HYDROPORT’94, Yokosuka, Japan, 1921 October, 1994
Cunnane, C. , 1978. Unbiased plotting positions – a review. J. Hydrology, 37, pp 20522.
Goda, Y. , 1979. A review on statistical interpretation of wave data. Port and Harbour
Research Institute, 18(1), 1979
Goda, Y. , 1985. Random seas and design of marine structures . University of Tokyo
Press, Japan, 1985
Goda, Y. , 1988. On the methodology of selecting design wave height. Proc. 21st Int.
Conf. on Coastal Engr., Spain.
Goda, Y., Kobune, K. , 1990. Distribution function ﬁtting for storm wave data. Proc.
22nd Int. Conf. on Coastal Engr., The Netherlands.
Le Mehaute, B. and Shen Wang , 1984. Eﬀects of measurement error on longterm wave
statistics. Proceedings of the 19th International Conference on Coastal Engineering,
Houston, USA, 1984.
Liu, Z. and Burcharth, H.F , 1996. Design wave height related to structure lifetime. Pro
ceedings of the 25th International Conference on Coastal Engineering, Orlando,
USA, 1996.
Rosbjerg, D. , 1985. Estimation in partial duration series with independent and depen
dent peak values. Journ. of Hydrology, 76, pp 183195.
Rosbjerg, D. , 1988. A defence of the median plotting position. Progress Report 66,
ISVA, Technical University of Denmark
Ross, S.M., , 1987. Introduction to probability and statistics for engineers and scientists.
John Wiley & Sons, Inc., 1987, PP 245305. ISBN 0471608157.
ThoftChristensen, P. and Michael J. Baker , 1982. Structural reliability theory and its
application. ISBN 3540117318, SpringerVerlag, 1982.
Weibull, W. , 1939. A statistical theory of strength of material. Ing. Vet. Ak. Handl.
(Stockholm), 151.
63
4 EXTREME WAVE HEIGHT ANALYSIS
4.16 Exercise
1) The design wave height for Sines breakwater in Portugal is the signiﬁcant
wave height corresponding to 100 years return period.
The hindcast study of Sines breakwater wave climate gave the following 17
severest storms in the period of 19701985:
H
s
in meter
12.0 10.8 10.7 10.2 10.1 9.8 9.6
9.3 9.3 9.0 8.8 8.1 7.8 7.7
7.3 6.9 6.3
Fit the data to Gumbel distribution by the least square method and the
maximum likelihood method and calculate the design wave height.
2) Taking into consideration sample variability, Use the MonteCarlo simula
tion to draw the probability density function of the design wave height and
calculate the upper bound of the design wave height corresponding to 90%
conﬁdence.
64
5 WAVE GENERATION IN LABORATORY
5 Wave generation in laboratory
The importance of wave generation in laboratory is due to the fact that we cannot describe
wave phenomenon (formation, transformation and especially breaking) and wavestructure
interactions (wave force, runup, overtopping etc.) purely by mathematics, and hence
model tests play an important role in the design of coastal and oﬀshore structures.
5.1 Principle of wave generator
Fig.1 illustrates the basic concept of wave generators. The input signal is the time series
of voltage to be sent to servo mechanism. At the same time the servo mechanism receives
information on the position of wave paddle through the displacement censor (feedback).
After the comparison of the input signal with the paddle position, the servo mechanism
sends a control signal to the valve of hydraulic pump, which converts the output of the
hydraulic pump into the movement of the wave paddle.
In stead of hydraulic pump, electric servomotor, directcurrent motor and hydraulic pulse
pump have also been applied. The wave paddle shown in Fig.1 is called pistontype wave
paddle. Another popular one is hingedtype wave paddle. In reality there are over 20
types of wave paddles.
Fig.1. Outline of the wave generator at AaU, Denmark.
65
5 WAVE GENERATION IN LABORATORY
5.2 Bi´esel transfer functions
Bi´esel transfer functions express the relation between wave amplitude and wave paddle
displacement (Bi´esel et al. 1951).
Formulation
Under the assumption of irrotational and incompressible ﬂuid, the velocity potential pro
duced by paddle movement is formulated in Fig.2.
Fig.2. Formulation of boundary value problem.
In Fig.2 the equations express:
0. Laplace equation. Basic equation for potential ﬂow.
1. All water particles at the free surface remain at the free surface (kinematic boundary
condition). Free surface is at constant pressure (dynamic boundary condition).
2. The water accompanies the wave paddle. The horizontal velocity of water particle
is the same as the paddle. The time series of paddle movement is
e(z, t) = x(z, t)
paddle
=
S(z)
2
sin(ωt) (51)
where S(z) is the stroke of the paddle, cf. Figs 4 and 5.
3. The bottom is impermeable.
4. The propagating wave is of constant form.
66
5 WAVE GENERATION IN LABORATORY
Nearﬁeld and farﬁeld solution
By solving the boundary value problem the surface elevation in the generated wave ﬁeld
is
η(x, t) = c · sinh(kh) cos(ωt −kx) +
∞
¸
n=1
c
n
sin(k
n
h) e
−kn·x
sin(ωt) (52)
where c, c
n
and k
n
are coeﬃcients depending on the paddle type, paddle cycling frequency
and water depth.
The ﬁrst term in eq (52) expresses the surface elevation at inﬁnity, by Bi´esel called the
farﬁeld solution, while the second term is the nearﬁeld solution. In general only the far
ﬁeld solution is interesting because the amplitude of a linear wave should not change with
location. Fortunately, the “disturbance” from the nearﬁeld solution will in a distance of
12 water depth from the wave paddle be less than 1% of the farﬁeld solution, cf. Fig.3.
Fig.3. Wave amplitude and phase of the generated wave ﬁeld relative to the farﬁeld
solution. Water depth = 0.7 m and wave period = 0.7 sec
Fig.3 shows that the farﬁeld surface elevation is phaseshifted
π
2
relative to the displace
ment of the wave paddle. However, because the initial phase of the surface elevation will
not change wave properties, the paddle movement is often written as in phase with the
surface elevation.
The Bi´esel transfer function, i.e. the amplitude relation between wave and paddle, is
obtained by the farﬁeld solution
c · sinh(kh) =
H
2
(53)
The Bi´esel transfer functions for the two most popular wave paddles are given in the
followings.
67
5 WAVE GENERATION IN LABORATORY
Bi´esel Transfer Function for pistontype paddle
S(z) = S
0
H
S
0
=
2 sinh
2
(kh)
sinh(kh) cosh(kh) +kh
(54)
Fig.4. Bi´esel Transfer Function for pistontype paddle.
Bi´esel Transfer Function for hingedtype paddle
S(z) =
S
0
h
· (h +z), S
0
= S(z = 0)
H
S
0
=
2 sinh(kh) (1 −cosh(kh) +kh sinh(kh))
kh (sinh(kh) cosh(kh) +kh)
(55)
Fig.5. Bi´esel Transfer Function for hingedtype paddle.
68
5 WAVE GENERATION IN LABORATORY
5.3 Examples
Calibration of wave paddle
Before generating waves the wave paddle should be calibrated in order to obtain the
calibration coeﬃcient of the paddle. The calibration of wave paddle is performed by
sending a signal which increases gradually from 0 to 1 voltage in one minute, and then
measuring the wave paddle displacement, cf. Fig.6.
Fig.6. Signal to be sent to wave paddle for calibration.
Modiﬁcation of signal
In order to avoid a sudden movement of wave paddle, the signal should be modiﬁed by a
data taper window. Fig. 7 illustrates the principle of the linear data taper window.
Fig.7. Linear data taper window.
69
5 WAVE GENERATION IN LABORATORY
General procedure
The most important aspect of wave generation is the preparation of the input signal
corresponding to the variance spectrum of design wave. Fig.8. illustrates one of simple
methods.
Fig.8. Preparation of input signal.
Of course the recording of the generated wave is necessary so that it can be checked
whether the variance spectrum of the generated wave is close to that of design wave,
according to Murphy’s law
7
.
7
Murphy’s law: Anything, which might go wrong, will go wrong.
70
5 WAVE GENERATION IN LABORATORY
Example : Linear wave generation
Generate a linear wave by pistontype wave paddle.
Wave height H = 0.1 m
Wave period T = 1.5 s
Water depth h = 0.4 m
1) The surface elevation of the linear wave is
η(t) = a cos(ωt +δ) =
H
2
cos(ωt +δ)
where the angular frequency ω = 2π/T = 4.2 s
−1
, The initial phase δ is
given a random number between 0 and 2π.
2) Convert the time series of surface elevation into the time series of piston
movement by the help of Biesel transfer function.
The Bi´esel transfer function for the pistontype wave paddle
B =
H
S
0
=
2 sinh
2
(kh)
sinh(kh) cosh(kh) + kh
= 0.80
The time series of the piston movement is
e(t) =
S
0
2
cos(ωt +δ) =
H
2B
cos(ωt +δ)
3) Convert the time series of piston movement into the time series of voltage
to be feeded to the pistontype wave paddle.
volt(t) = C e(t)
where C is the calibration coeﬃcient of the wave paddle.
4) Modify the time series of the voltage by the linear data taper window in
order to avoid sudden movement of the wave paddle.
5) Sample the data from the modiﬁed time series of the voltage at f
sample
= 50
Hz and send the signal to the wave paddle.
71
5 WAVE GENERATION IN LABORATORY
Fig.9. Preparation of input signal for linear wave.
72
5 WAVE GENERATION IN LABORATORY
Example : Irregular wave generation
Generate irregular wave by pistontype wave paddle according to JONSWAP spectrum
with H
s
= 0.1 m, T
p
= 1 s, water depth h = 0.4 m and peak enhancement coeﬃcient
γ = 3.3.
JONSWAP spectrum
S(f) = α H
2
s
f
4
p
f
−5
γ
β
exp
−
5
4
fp
f
4
α ≈
0.0624
0.230 + 0.0336 γ − (
0.185
1.9 + γ
)
β = exp
−
(f − fp)
2
2 σ
2
f
2
p
σ ≈ 0.07 f ≤ f
p
σ ≈ 0.09 f ≥ f
p
γ : peak enhancement coeﬃcient
1) Draw the JONSWAP spectrum with the speciﬁed parameters.
Note S
η
(f
p
) = 0.00193 (m
2
s).
2) Divide the spectrum evenly into N parts in the interval (f
start
, f
stop
).
To ensure accuracy usually
N ≥ 50 S
η
(f
start
) ≤ 0.01 S
η
(f
p
) S
η
(f
stop
) ≤ 0.01 S
η
(f
p
)
For the sake of simplicity in this example
N = 7 f
start
= 0.6 Hz f
stop
= 2.0 Hz
The frequency band width
∆f =
f
stop
−f
start
N
= 0.2 Hz
That is to say, the irregular wave is composed of 7 linear waves. The surface
elevation of the irregular wave is
η(t) =
7
¸
i=1
η
i
(t) =
7
¸
i=1
a
i
cos(ω
i
t +δ
i
)
73
5 WAVE GENERATION IN LABORATORY
3) Determine the angular frequency ω
i
, amplitude a
i
and initial phase δ
i
of
each linear waves.
frequency f
i
= f
start
+ i ∆f −
∆f
2
angular ω
i
=
2π
T
i
= 2πf
i
variance S
η
(f
i
) ∆f =
1
2
a
2
i
amplitude a
i
=
2 S
η
(f
i
) ∆f
i = 1, 2, · · · , 7
S(f
i
) is calculated from the JONSWAP spectrum. The initial phase δ
i
is
assigned a random number between 0 and 2π.
4) Convert the time series of surface elevation into the time series of piston
movement by the help of Biesel transfer function.
The Bi´esel transfer function for the pistontype wave paddle
B
i
=
H
i
S
0,i
=
2 sinh
2
(k
i
h)
sinh(k
i
h) cosh(k
i
h) + k
i
h
i = 1, 2, · · · , 7
The time series of the piston movement is
e(t) =
7
¸
i=1
S
0,i
2
cos(ω
i
t +δ
i
) =
7
¸
i=1
H
i
2B
i
cos(ω
i
t +δ
i
)
i 1 2 3 4 5 6 7
f
i
(Hz) 0.7 0.9 1.1 1.3 1.5 1.7 1.9
T
i
(s) 1.4 1.1 0.9 0.8 0.7 0.6 0.5
L
i
(m) 2.5 1.7 1.2 0.9 0.7 0.5 0.4
ω
i
(/s) 4.4 5.7 6.9 8.2 9.4 10.7 11.9
k
i
(/m) 2.6 3.6 5.0 6.9 9.1 11.6 14.5
Sη(f
i
) (m
2
s) 0.00007 0.00079 0.00103 0.00036 0.00021 0.00012 0.00007
a
i
(m) 0.0052 0.0178 0.0203 0.0119 0.0092 0.0070 0.0055
H
i
(m) 0.0103 0.0356 0.0406 0.0239 0.0183 0.0141 0.0110
δ
i
1.2 0.7 0.3 2.5 6.1 4.3 4.1
B
i
1.00 1.36 1.69 1.90 1.98 2.00 2.00
5) Convert the time series of piston movement into the time series of voltage
to be feeded to the pistontype wave paddle.
volt(t) = C e(t)
where C is the calibration coeﬃcient of the wave paddle.
6) Modify the time series of the voltage by the linear data taper window in
order to avoid sudden movement of the wave paddle.
74
5 WAVE GENERATION IN LABORATORY
7) Sample the data from the modiﬁed time series of the voltage at f
sample
= 50
Hz and send the signal to the wave paddle.
Fig.10. Preparation of input signal for irregular wave.
75
5 WAVE GENERATION IN LABORATORY
Remarks
For more complicated wave generation method, including active wave absorption and 3
D wave generation, reference is made to Frigaard et al. (1993). All these aspects have
been implemented in a userfriendly software package named PROFWACO (Frigaard et
al. 1993). With respect to wave generation techniques we are proud of the fact that the
Hydraulic & Coastal Engineering Laboratory of Aalborg University is one of the leading
institutes in the world.
5.4 References
Dean, R.G. and Dalrymple, R.A. , 1991. Water wave mechanics for Engineers and sci
entists. Second printing with correction, World Scientiﬁc Publishing Co. Pte. Ltd.,
Singapore, 1991.
Frigaard, P., Hgedal, M. and Christensen, M. , 1993. Wave generation theory. Hydraulic
& Coastal Engineering Laboratory, Department of Civil Engineering, Aalborg Uni
versity, Denmark, 1993.
Frigaard, P., Hgedal, M. and Christensen, M. , 1993. PROFWACO, Users Guide. Hy
draulic & Coastal Engineering Laboratory, Department of Civil Engineering, Aal
borg University, Denmark, 1996.
Goda, Y. , 1985. Random seas and design of marine structures. University of Tokyo
Press, Japan, 1985
5.5 Exercise
1) Estimate the minimum power of motor needed for generating a linear wave
with Wave height H = 0.20 m, wave period T = 1.5 s, water depth h = 0.4 m,
wave ﬂume width B = 1.5 m.
2) What is the minimum distance between the wave gauge and wave paddle ?
3) Explain the importance that the wave paddle can produce suﬃciently large
wave height. List the factors which limit the maximum wave height obtain
able in a wave ﬂume.
4) Make a computer program for examples given in lecture.
76
Preface
Sea waves are the most important phenomenon to be considered in the design of coastal and oﬀshore structures. Every sailor has noticed that, when wind is blowing, there are a lot of large and small waves propagating in many directions. Such waves are called shortcrested waves because they do not have a long crest. Contrast to shortcrested waves, we have longcrested waves, i.e. large and small waves moving in one direction. Even though there are some research eﬀorts on shortcrested waves and their eﬀects on structures, longcrested waves dominate today’s structure design. The book deals with longcrested waves. The contents of the book is illustrated in the ﬁgure.
i
There are two analysis methods for time series of irregular waves, namely timedomain and frequencydomain analysis, which will be dealt with in Chapter 1 and Chapter 2 respectively. It should be stressed that, even though all contents in the book are related to sea waves, they have broader applications in practice. For example, the extreme theory has also been applied to hydrology, wind mechanics, ice mechanics etc., not to mention the fact that spectral analysis comes originally from optics and electronics. The book intents to be a textbook for senior and graduate students who have interest in coastal and oﬀshore structures. The only prerequirement for the book is the knowledge of linear wave theory. Michael Brorsen, Associate Professor at the Hydraulic and Coastal Engineering Laboratory, Aalborg University is gratefully acknowledged for the valuable comments.
ii
CONTENTS
Contents
1 Time series analysis I : Timedomain 1.1 1.2 1.3 1.4 1.5 1.6 1.7 Deﬁnition of individual wave : Zerodowncrossing . . . . . . . . . . . . . . Characteristic wave heights and periods . . . . . . . . . . . . . . . . . . . . Distribution of individual wave heights . . . . . . . . . . . . . . . . . . . . Maximum wave height Hmax . . . . . . . . . . . . . . . . . . . . . . . . . . Distribution of wave period . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 1 2 3 6 9 9
Exercise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10 12
2 Timeseries analysis II: Frequencydomain 2.1 2.2 2.3 2.4 2.5 2.6 2.7
Some basic concepts of linear wave theory . . . . . . . . . . . . . . . . . . 12 Example of variance spectrum . . . . . . . . . . . . . . . . . . . . . . . . . 14 Fourier series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17 Discrete signal analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19 Characteristic wave height and period . . . . . . . . . . . . . . . . . . . . . 23 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24 Exercise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25 26
3 Windgenerated waves 3.1 3.2 3.3 3.4 3.5 3.6
Wave development and decay . . . . . . . . . . . . . . . . . . . . . . . . . 26 SPMmethod . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27 Standard variance spectrum . . . . . . . . . . . . . . . . . . . . . . . . . . 34 Introduction to shortcrested waves . . . . . . . . . . . . . . . . . . . . . . 37 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39 Exercise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39 40 . . . . . . . . . . . 40
4 Extreme wave height analysis 4.1 4.2 4.3 4.4 4.5 4.6 Design level: Return period and encounter probability General procedure
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
Data sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42 Candidate distributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44 Fitting methods and procedure . . . . . . . . . . . . . . . . . . . . . . . . 45 Plotting position formulae . . . . . . . . . . . . . . . . . . . . . . . . . . . 47 iii
. . . . . . . . . . . . . . . . . . . . . . . . 64 5 Wave generation in laboratory 5. . . . . . . . . . . . . . .11 Physical consideration of design wave height . . . . . . . . . . . .4 5. . . . . . . . . . . . . . . . . . . .3 5. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65 Bi´sel transfer functions . .13 Water level . . . . . . . . . . . .5 65 Principle of wave generator . 51 Design wave height: xT . . . . . . . . . . . . . . . . . . . . . . . . 66 e Examples . . . 55 4. . .16 Exercise . . .7 4. . . . . . . . . . . . . . . 63 4. . . . . . . . . . . . . . . . . . . 53 4. . . . . . . . . . . . . .10 Sources of uncertainties and conﬁdence interval . . . . . . . . . . . . 51 Example . . . 61 4. . . . . . . . . .15 References . . . . . . . 76 Exercise . . . . . . . . . . . . . . . . . . . . . . . . . . . 62 4. . .12 Wave period . . . . . .2 5. . . . . . . . . . . . . . . 60 4. . . . . . . . . . . . . . . . .9 Fitting goodness . . . . . . . . . . . . . . . . 69 References . . . . . . . . . . . . . . . . . . . . . . 76 iv . . .CONTENTS 4. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59 4. . . . .14 Multiparameter extreme analysis . . . .1 5. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .8 4. . . . . . . . . . . .
in Fig. respectively. Ranked individual wave heights and corresponding periods in Fig.4 8. Fig. The application of zerodowncrossing gives 15 individual waves (N=15).5 2 4. rank i H (m) T (s) wave no. These two analysis methods will be described in the next two chapters.3 10 2.8 13.8 6. Individual waves deﬁned by zerodowncrossing. 1.1 4. Fig.9 7 12 15 3 5 4 2 11 6 1 10 8 13 14 9 1 .6 13 1.2 12. cf.9 8 2.9 11. Table 1.0 15 0.8 15.1.9 5. Fig.1.0 4 3.1 11 2.8 11.2. as recommended by IAHR (1986).3 14 1. In Table 1 the data are arranged according to the descending order of wave height.23 0.0 9 2.2 6 3.2 1 5.1 TIME SERIES ANALYSIS I : TIMEDOMAIN 1 Time series analysis I : Timedomain The recorded time series of the surface elevation of irregular waves can be studied by either the timedomain or the frequencydomain analysis.3 10.2 7.1 Deﬁnition of individual wave : Zerodowncrossing Individual wave is deﬁned by two successive zerodowncrossing points.7 9.2. Fig.5 12.2 12 1.2 5 3.9 11. Application of zerodowncrossing.2 is an example of surface elevation recordings.0 3 4.5 7 2.
g.1 TIME SERIES ANALYSIS I : TIMEDOMAIN 1. Hrms = 1 N N Hi2 = i=1 1 15 15 Hi2 = 3. H1% .44 m i=1 Ts = 1 5 5 Ti = 12.8 s i=1 i is the rank no. In Table 1. Signiﬁcant wave is most often used as the design wave. 2 . vertical breakwaters. Highest onetenth wave: H1/10 . Signiﬁcant wave period is the average of the wave periods associated with the onethird highest wave.20 m i=1 Wave height with exceedence probability of α%: For example H0. Which wave should be chosen as the design wave ? Maximum wave: Hmax . The reason might be that in old days structures were designed based on visual observation of waves. In Table 1. Hs = 1 5 5 Hi = 4.1% .2 contains more than 100 individual waves.5 s Maximum wave is chosen as the design wave for structures which are very important and very sensitive to wave load. Signiﬁcant wave: Hs . TH1/3 Signiﬁcant wave height is the average of the wave heights of the onethird highest waves. Therefore the choice of signiﬁcant wave as design wave can make use of the existing engineering experience. Note Hmax is a random variable with the distribution depending on the number of individual waves. Mean wave: H. TH1/10 is the average of the wave periods associated with the onetenth highest wave. H = 1 15 15 Hi = 2.9 m i=1 T = 1 15 15 Ti = 9. Ts or H1/3 . THmax This is the wave which has the maximum wave height.5 m THmax = 12. H2% etc.2 Characteristic wave heights and periods Usually the surface elevation recording exempliﬁed in Fig. Hmax = 5. TH1/10 H1/10 is the average of the wave heights of the onetenth highest waves. In Table 1. Experiences show that the wave height and period reported by visual observation correspond approximately to signiﬁcant wave. e. T H and T are the means of the heights and periods of all individual waves.25 s i=1 Rootmeansqure wave height Hrms In Table 1.
the probability density becomes a continuous curve. Fig. cf. Histogram of wave height. we say that individual wave height follows the Rayleigh distribution. Experience and theory have shown that this curve is very close to the Rayleigh distribution. When ∆(H/H) approaches zero. Nondimensionalized histogram of wave height. Nondimensionalized histogram In order to compare the distributions of wave height in diﬀerent locations. Fig. Roughly speaking. Fig.4.1 TIME SERIES ANALYSIS I : TIMEDOMAIN 1.3.4. Fig. the histogram of wave height is nondimensionalized.3 Distribution of individual wave heights Histogram of wave height In stead of showing all individual wave heights. 3 . it is easier to use wave height histogram which tells the number of waves in various wave height intervals.3 is the histogram of wave height corresponding to Table 1.
03 H H1/3 = 1. (13) (12) Fig. Relation between Hs and H.5 illustrates how to obtain the relation between Hs and H.5.23 H Fig.60 H Hrms = 1. The Rayleigh distribution function given by Hs instead of H reads F (H) = 1 − exp −2 H Hs 2 (14) 4 . then the characteristic wave heights H1/10 . H1/3 . Hrms and Hα% can be expressed by H through the manipulation of the Rayleigh probability density function. H1/10 = 2.1 TIME SERIES ANALYSIS I : TIMEDOMAIN Rayleigh distribution The Rayleigh probability density function is f (x) = π π x exp − x2 2 4 x= H H (11) The Rayleigh distribution function is π F (x) = Prob{X < x} = 1 − exp − x2 4 Relation between characteristic wave heights If we adopt the Rayleigh distribution as an approximation to the distribution of individual wave heights.13 H H2% = 2.
cf. Researches have also been done by Thornton and Guza (1983). whereas the signiﬁcant wave height Hm0 is determined from the spectrum. 5 .6. Fig. The correction formulae are very useful for checking the wave height distribution in small scale physical model tests. for shallow water wave height distribution with model test results.1% = Hm0 1+ Hm0 h 1 −2 where h is the water depth. (1989) proposed a semiempirical expression for the individual wave height distribution. Stive. Comparison of the expression by Stive.6. proposed the following empirical correction to the Rayleigh distribution based on model tests but roughly checked against some prototype data H1% = Hm0 ln100 2 ln1000 2 1 2 Hm0 1+ h 1 2 −1 3 (15) H0. Klopmann et al. based on energy spectrum width parameter. Aalborg University Hydraulics Laboratory 1990 (from Burcharth 1993). 1986. the Rayleigh distribution is a good approximation to the distribution of individual wave heights. 1986. Chapter 2 gives a more detailed discussion on the validity of the Rayleigh distribution. H1% means the 1% exceedence value of the wave height determined by zero downcrossing analysis. the individual wave height distribution will diﬀer from the Rayleigh distribution. Fig.1 TIME SERIES ANALYSIS I : TIMEDOMAIN Individual wave height distribution in shallow water Only in relatively deep water. When wave breaking takes place due to limited water depth.
1 TIME SERIES ANALYSIS I : TIMEDOMAIN 1.7. N is often assumed to be 1000.7. Fig. 1 6 . The signiﬁcant wave height is varying during a storm. 7. A storm usually lasts some days. The probability density function of Xmax is fXmax (x) = dFXmax dx π π N x exp − x2 2 4 π 1 − exp − x2 4 N −1 = (18) The density function of X and the density function of Xmax are sketched in Fig. Xmax = Hmax /H is FXmax (x) = Prob{Xmax < x} = ( FX (x) )N = π 1 − exp − x2 4 N (16) the distribution function of (17) Note that FXmax (x) can be interpreted as the probability of the nonoccurrence of the event ( X > x ) in any of N independent trials. However we are more interested in the maximum signiﬁcant wave height in a short period of time. whose distribution depends on the number of individual wave hegiht. N . Distribution of Hmax The distribution function of X = H/H is the Rayleigh distribution π FX (x) = Prob{X < x} = 1 − exp − x2 4 If there are N individual waves in a storm1 . Fig.4 Maximum wave height Hmax Hmax is a random variable. In practice. Probability density function of X and Xmax . cf.
7 . 8. Their deﬁnitions are given in Fig. is (Hmax )µ ≈ 1 N Hs ln 1 2 ln 1−µ (111) Obviously (Hmax )median = (Hmax )0. 9.577 ≈ Hs + √ 2 8 ln N (19) (Hmax )mode ≈ ln N Hs 2 (110) Furthermore. median and mode of Hmax Mean.5 Mode xmode = x fX (x)=max Fig. deﬁned as the maximum wave height with exceedence probability of µ. cf.1 TIME SERIES ANALYSIS I : TIMEDOMAIN Mean. Fig. median and mode of a random variable X.8. Deﬁnition of (Hmax )µ . median and mode are often used as the characteristic values of a random variable. Mean. +∞ Mean xmean = E[X] = −∞ xfX (x)dx Median xmedian = x FX (x)=0. 9. By putting eqs (17) and (18) into the deﬁnitions. we obtain (Hmax )mean ln N 0.5 . Fig. (Hmax )µ .
10) H = F −1 ( F (H) ) = Hs − ln(1 − F (H)) 2 (113) 2) Repeat step 1) N times. Thus we obtain a sample belonging to the distribution of eq (112) and the sample size is N . Individual wave heights follow the Rayleigh distribution F (H) = 1 − exp −2 H Hs 2 (112) The storm duration corresponds to N individual waves. One individual wave height H is obtained by (cf. 3) 4) 5) Fig. 10. Repeat steps 2) and 3). Simulated wave height from the Rayleigh distribution. Pick up Hmax from the sample. Let the nonexceedence probability F (H) equal to that data. say. 8 .000 values of Hmax . 1) Generate randomly a data between 0 and 1. Draw the probability density of Hmax . Thus we get 10.1 TIME SERIES ANALYSIS I : TIMEDOMAIN MonteCarlo simulation of Hmax distribution The distribution of Hmax can also be studied by the MonteCarlo simulation. Fig.000 times.10.
J. The design of breakwater. and Dalrymple.A. Y. 1991.5 Distribution of wave period It is summarized as • There is no generally accepted expression for the distribution of wave period.5 (Neu 1982). An example of scatter diagrams is given in Chapter 4. 1. Unfortunately. Intern Report H533. 1993. . 1993. . M. Stive. Report No. Aalborg University. 3. 1982. The relation between Hs β and Ts is often simpliﬁed as Ts = αHs . Canadian Tech.2 T (Goda 1985).J.6 References Burcharth H. M. . Pte. University of Tokyo Press. 1986. R. 52. Extreme waves and wave loading in shallow water.43 and β = 0. . • The empirical relation Tmax ≈ T1/10 ≈ T1/3 ≈ 1. section 12. 11year deep water wave climate of Canadian Atlantic waters. Until now there is no generally accepted expression for the joint distribution.F. Goda. 9 . Department of Civil Engineering. Random seas and design of marine structures.g. Dean. Supplement to Bulletin No. . 1985. of Hydrography and Ocean Sciences. Neu. Second printing with correction.1 TIME SERIES ANALYSIS I : TIMEDOMAIN 1. H. G. 1986. • The distribution of wave period is narrower than that of wave height.G. in Canadian Atlantic waters α = 4.A. 1989. 1986 Klopmann. even though there are some socalled scatter diagrams based on wave recording. e. . Ltd. • In practice the joint distribution of wave height and wave period is of great importance. E & P Forum. World Scientiﬁc Publishing Co. List of sea state parameters. R. 1985 IAHR. Water wave mechanics for Engineers and scientists. 13. ..F. Extreme shallow water conditions. 1989. Singapore. 1986. Japan. Rept.F.12/156. Delft Hydraulics. and Stive.J. Such a diagram is valid only for the measurement location. 1991.
H1/3 . (Hmax )0. (Hmax )mode . The duration of the storm corresponds to 1000 individual waves.94 2.52 2.05 (2) Now suppose that the storm contains 500 individual waves.2 7.05 2.05 4. (Hmax )median .0 6.08 4. Calculate (Hmax )mean . T1/10 .3 5.7 Exercise 1) The application of the downcrossing method gives the following 21 individual waves. (Hmax )median .3 4.03 1.3 6.0 5.9 Calculate Hmax .0 8.00 2.58 3. (3) Use MonteCarlo simulation to determine (Hmax )mean .1 8.89 2. (Hmax )mode .87 1. Hrms 2) Prove H2% = 2. T1/3 .2 6.98 6. Calculate the exceedence probability of H1/10 .9 7. (Hmax )0.05 . H.2 8.20 1.0 9. (Hmax )mode .0 9. Tmax .05 10 .90 1.2 8. The signiﬁcant wave height of the design storm is H1/3 = 10 m. (Hmax )0.97 1.83 2.4 4.62 4.43 2. 5) An important coastal structure is to be designed according to Hmax . 4) Suppose individual waves follow the Rayleigh distribution. (Hmax )median . (1) Calculate (Hmax )mean .37 4.1 TIME SERIES ANALYSIS I : TIMEDOMAIN 1.95 1.4 5.23 H 3) Explain the diﬀerence between H1/10 and H10% .6 7.9 11. Compare with the results of (1). T . H1/10 .54 2.3 11 12 13 14 15 16 17 18 19 20 21 1.0 7. wave number wave height H (m) wave period T (s) wave number wave height H (m) wave period T (s) 1 2 3 4 5 6 7 8 9 10 0.23 2. Hs and H.
e. (1) What is the consequence if sampling frequency is too high. You would like to measure the surface elevation in order to check whether it is the same as desired.1 TIME SERIES ANALYSIS I : TIMEDOMAIN 6) You have generated a linear wave in a laboratory ﬂume.g.g. The wave period is 1 second. 1 Hz ? (3) What is an appropriate sampling frequency ? 11 . e. 100 Hz ? (2) What is the consequence if sampling frequency is too low.
For example. 2. k = 2π/L wave length initial phase (21) We can also deﬁne the observation location to x = 0 and obtain η(t) = a cos(ωt + δ) The relation between wave period and wave length (dispersion relationship) is L = g T2 2πh tanh 2π L (23) (22) where h is water depth. Spectral analysis is a technique of decomposing a complex physical phenomenon into individual components with respect to frequency.1 Some basic concepts of linear wave theory Surface elevation The surface elevation of a linear wave is η(x. ω = 2π/T wave period. a = H/2 angular frequency. Spectral analysis of irregular waves is very important for structure design. in the oildrilling platform design where wave force plays an important role. based on the principle that white light consists of numerous components of light of various colors (wave length or wave frequency). 12 . t) = where H a ω T k L δ H cos(ωt − kx + δ) = a cos(ωt − kx + δ) 2 wave height amplitude. wave number. it is of importance to design the structure in such a way that the natural frequency of the structure is fairly far away from the frequency band where most wave energy concentrates. Energy spectrum means the energy distribution over frequency.2 TIMESERIES ANALYSIS II: FREQUENCYDOMAIN 2 Timeseries analysis II: Frequencydomain The concept of spectrum can be attributed to Newton. so that resonance phenomenon and the resulted dynamic ampliﬁcation of force and deformation can be avoided. who discovered that sunlight can be decomposed into a spectrum of colors from red to violet.
This means that the superposition of a number of linear waves with diﬀerent wave height and wave period will be superposition velocity potential surface elevation particle velocity dynamic pressure ϕ η u p = = = = wave 1 ϕ1 η1 u1 p1 + + + + wave 2 ϕ2 η2 u2 p2 + + + + ··· ··· ··· ··· ··· + + + + wave N ϕN ηN uN pN 13 . it is known from mathematics that small amplitude waves are superposable.2 TIMESERIES ANALYSIS II: FREQUENCYDOMAIN Wave energy The average wave energy per unit area is E = 1 1 ρ g H 2 = ρ g a2 8 2 (Joule/m2 in SI unit) (24) Variance of surface elevation of a linear wave The variance of the surface elevation of a linear wave is 2 ση = Var[η(t)] = E η(t) − η(t) 2 (E: Expectation) = E [ η 2 (t) ] = = 1 T 1 2 T 0 η 2 (t) dt (T: wave period) a2 Superposition of linear waves Since the governing equation (Laplace equation) and boundary conditions are linear in small amplitude wave theory.
14 .1 gives an example of an irregular wave surface elevation which is constructed by adding 4 linear waves (component waves) of diﬀerent wave height and wave period.1. Simulation of irregular waves by superposition of linear waves. The superposed wave surface elevation is 4 4 η(t) = i=1 ηi (t) = i=1 ai cos(ωi t + δi ) (25) Fig.2 Example of variance spectrum First we will make use of an example to demonstrate what a variance spectrum is.2 TIMESERIES ANALYSIS II: FREQUENCYDOMAIN 2. Surface elevation of irregular wave Fig.
to describe the irregular wave. while the information on initial phase (δi ) is lost.1. cf.2. The spectral density is deﬁned as Sη (f ) = 1 2 a 2 ∆f (m2 s) (26) where ∆f is the frequency band width2 . This information loss does not matter because the surface elevation of irregular wave is a random process. shown in Fig. Fig. the variance diagram keeps the information on amplitude (ai ) and frequency (fi . Fig. hence Ti and Li ) of each component. Variance spectral density Sη (f ) The variance diagram can be converted to variance spectrum. Variance diagram. Fig.3.2 TIMESERIES ANALYSIS II: FREQUENCYDOMAIN Variance diagram In stead of Fig. In reality an irregular wave is composed of inﬁnite number of linear waves with diﬀerent we will see later that ∆f depends on signal recording duration. In the ﬁgure it is assumed that ∆f = 0.1.01Hz 2 15 .2. In comparison with Fig. Stepped variance spectrum. we can use a variance diagram.3. We can simply assign a random initial phase to each component.
N (210) The angular frequency is ωi = 2π = 2πfi Ti i = 1. N (211) The initial phase δi is assigned a random number between 0 and 2π. Fig. Hence by use of eq (28) we can draw the timeseries of the surface elevation of the irregular wave which has the variance spectrum as shown in Fig.4.4 the known variance spectral density Sη (f ) is divided into N parts by the frequency band width ∆f .4 gives an example of stepped variance spectrum. Variance spectrum is also called energy spectrum. 2. Continuous variance spectrum (wave energy spectrum). 2. Fig. But strictly speaking. the energy spectral density should be deﬁned as S(f ) = 1 ρ 2 g a2 ∆f (m2 s) (27) Construction of time series from variance spectrum We can also construct time series of surface elevation from variance spectrum.4. In ﬁg. When ∆f approaches zero. N (29) Therefore the amplitude is ai = 2 Sη (fi ) ∆f i = 1. · · · . the variance spectrum becomes a continuous curve.2 TIMESERIES ANALYSIS II: FREQUENCYDOMAIN frequency. 2. · · · . · · · . 16 . This means that the irregular wave is composed of N linear waves N N η(t) = i=1 ηi (t) = i=1 ai cos(ωi t + δi ) (28) The variance of each linear wave is Sη (fi ) ∆f = 1 2 a 2 i i = 1.
with period T0 then x(t) can be expressed as a Fourier series.2 TIMESERIES ANALYSIS II: FREQUENCYDOMAIN 2. ∞ (213) x(t)dt and b0 = 0. However. Fourier series Fourier series is used to represent any arbitrary function3 . 3 17 . · · · . Arbitrary periodic function of time. Not all mathematicians agree that an arbitrary function can be represented by a Fourier series. where the linear components of the irregular wave are predeﬁned (cf. 1. which is a random process. In our case x(t) is the surface elevation of irregular wave. all agree that if x(t) is a periodic function of time t.1). Fig.5. 2. ∞ x(t) = a0 + 2 i=1 ai cos 2πi t T0 + bi sin 2πi t T0 ∞ = 2 i=0 (ai cos ωi t + bi sin ωi t) (212) where ai and bi are Fourier coeﬃcients given by ai = bi = Note a0 = 1 T0 1 T0 1 T0 T0 0 T0 0 T0 0 x(t) cos ωi t dt x(t) sin ωi t dt i = 0.3 Fourier series Conversion of irregular surface elevation into variance spectrum is not as simple as the above example. if T0 is large enough. We need to decompose the irregular wave into its linear components. First let’s see how it can be done with a known continuous function x(t). Fig. we can assume that x(t) is a periodic function with period T0 .
2 TIMESERIES ANALYSIS II: FREQUENCYDOMAIN Physical interpretation Now we say that the continuous function x(t) is the surface elevation of irregular wave η(t). which can be expanded as a Fourier series. 18 . expressed as a continues function. · · · . bi }. i = 0. is composed of inﬁnite number of linear waves with amplitude period 2ci = 2 a2 + b2 i i Ti = 2π ωi = T0 i i = 0. ∞. ∞ η(t) = 2 i=0 (ai cos ωi t + bi sin ωi t) ∞ = 2 i=0 ∞ (ci cos δi cos ωi t + ci sin δi sin ωi t) = i=0 ∞ 2ci (cos δi cos ωi t + sin δi sin ωi t) = i=0 2ci cos(ωi t − δi ) (214) That is to say. 2. ∞ (215) {ai . · · · . 1. 1. are given in eq (213). any irregular wave surface elevation.
cf. ηN −1 The Fourier coeﬃcients (a0 . η1 . b0 ). neither necessary.4 Discrete signal analysis The measurement of surface elevation is carried out digitally. b1 ). is composed of N linear waves N −1 N −1 η(t) = i=0 ηi (t) = i=0 2 a2 + b2 cos(ωi t + δi ) i i (216) amplitude angular frequency period frequency 4 2 a2 + b 2 i i ωi = Ti = fi = 2πi T0 2π = Ti0 ωi 1 = Ti0 Ti i = 0. Fig. then the time interval between two succeeding points is ∆ = 1/fs . Corresponding to the total number of sample points N . Fig. For details of DFT and FFT. a continuous function of the surface elevation. ···. expressed by digital time series. It oﬀers an enormous reduction in computer processing time. Thus we obtain a discrete time series of surface elevation η0 .6. Sampling of surface elevation at regular intervals. We do not have. bN −1 ) can be obtained by Fast Fourier Transforms (FFT)4 .2 TIMESERIES ANALYSIS II: FREQUENCYDOMAIN 2. In stead we have a series of surface elevation measurement equally spaced in time. 1. (aN −1 . ···.6. the irregular wave surface elevation. the sample duration T0 = (N − 1)∆. N − 1 (217) FFT is a computer algorithm for calculating DFT. If the sampling frequency is fs . · · · . That is to say. please refer to Newland (1975) 19 . (a1 .
Fig. 1. · · · .2 TIMESERIES ANALYSIS II: FREQUENCYDOMAIN Therefore we obtain the variance spectrum frequency band width ∆f = fi+1 − fi = 1 T0 = 2(a2 + b2 ) i i ∆f (218) spectral density Sη (fi ) = 1 (amplitude)2 2 ∆f An example of variance spectrum is shown in Fig. Fourier analysis decomposes N digital data into N linear components. contains two parts. Variance spectrum (including aliasing). · · · . T0 = (N − 1) ∆ The concept of nyquist frequency means that the Fourier coeﬃcients { ai . N − 1.7. The frequency of each component is fi = i T0 i = 0. bi }. Nyquist frequency fnyquist Nyquist frequency fnyquist is the maximum frequency which can be detected by the Fourier analysis. the ﬁrst half part (i = 0.7. i = 0. N/2 − 1) represents 20 . ∆ = 1/fs total number of sample sample duration. 1. · · · . N − 1 (219) The nyquist frequency is fnyquist = f N −1 = 2 N −1 2 T0 = (N − 1) ∆ N −1 2 = 1 fs = 2∆ 2 (220) where fs ∆ N T0 sample frequency time interval between two succeeding sample points. 1.
That is the reason why fnyquist is also called cutoﬀ frequency. The solution to aliasing is simple: double Sη (fi ). Aliasing after Fourier analysis.8 gives an example on aliasing after the Fourier analysis of discrete time series of a linear wave. cf. In doing so we are actually assuming that irregular wave contains no linear components whose frequency is higher than fnyquist .9.8. 21 . Fig. N/2 + 1. 2. · · · . eq (220). N − 1. · · · . equal to zero. Fig. Fig. i = N/2. This assumption can be assured by choosing suﬃciently high sample frequency fs . N/2−1 and let Sη (fi ). N − 1) is the folding components (aliasing). N/2 + 1. 1. · · · .2 TIMESERIES ANALYSIS II: FREQUENCYDOMAIN true components while the second half part (i = N/2. cf. i = 0.
7).2 TIMESERIES ANALYSIS II: FREQUENCYDOMAIN Fig. The modiﬁcation is carried out with the help of taper data window. 22 .10.004 0≤t≤ T0 10 T0 10 9T0 10 ≤t≤ (221) 1 + cos 10π ( 9T0 t− 10 ) T0 9T0 10 ≤ t ≤ T0 Fig. it may be desirable to modify the recorded time series before Fourier analysis.9. Variance spectrum after cutoﬀ (refer to Fig. The widelyused cosine taper data window reads d(t) = 1 2 1 2 1 − cos 10πt T0 1. Taper data window Fourier analysis requires that η(t) is a periodic function with period T0 . so that the signal looks like a periodic function. Taper data window.
11. n order moment mn mn is deﬁned as mn = ∞ 0 f n Sη (f ) df (222) The zero moment is m0 = ∞ 0 Sη (f ) df (223) which is actually the area under the curve.2 TIMESERIES ANALYSIS II: FREQUENCYDOMAIN 2. says nothing about how high the individual waves will be. a wider spectrum gives larger values of the higher order moment (n ≥ 2). With the same m0 . illustrated in Fig. Fig. Fig. Now We will see how to estimate the characteristic wave height and period based on the variance spectrum.11. the more weight is put on the higher frequency portion of the spectrum.11. Spectrum width parameter and validity of the Rayleigh distribution From the deﬁnition of mn . LonguetHiggins has deﬁned a spectrum width parameter ε = 1 − m2 2 m0 m4 (224) It has been proven theoretically that spectrum width parameter ε=0 ε=1 narrow spectrum wide spectrum wave height distribution Rayleigh distribution Normal distribution 23 . it can be seen that the higher the order of moment. cf.5 Characteristic wave height and period The variance spectrum. Variance spectrum.
1978.Graﬀ and P. Edited by W.4 − 0. as the Rayleigh distribution gives slightly larger wave height for any given probability level. H.40. On the design of gravity structures using wave spectra. and Brorsen.6 References Burcharth. M. Aalborg University. It has been found that Rayleigh distribution is a very good approximation and furthermore conservative. Longman.J.7 √ m0 (226) Peak frequency is deﬁned as (cf. ε = 0 . University of Tokyo Press. Random seas and design of marine structures. .5. Goda. In introduction to random vibrations and spectral analysis. Lecture on Oﬀshore Engineering. a good estimate of signiﬁcant wave height from variance spectrum is Hs = 3. ThoftChristensen. They are equal to each other when wave height follows the Rayleigh distribution. D. . Hm0 denotes signiﬁcant wave height determined from spectrum while Hs or H1/3 is signiﬁcant wave height determined from timedomain analysis. i. London.F. 1985 Newland.2 TIMESERIES ANALYSIS II: FREQUENCYDOMAIN In reality ε lies in the range of 0. Signiﬁcant wave height Hm0 and peak wave period Tp When wave height follows the Rayleigh distribution. 2. Fig. Y. Institute of Building Technology and Structural Engineering.5. 5 24 .e. the signiﬁcant wave height Hm0 5 can theoretically be expressed as Hm0 = 4 √ m0 (225) In reality where ε = 0. Denmark.E.11) fp = f Sη (f )=max (227) Wave peak period (Tp = 1/fp ) is approximately equal to signiﬁcant wave period deﬁned in timedomain analysis. 1975. 1985. 1975. . Japan.
(2) Compare the distribution of individual wave height with the Rayleigh distribution.3 12 3 3. 3) Make a computer program to simulate the surface elevation of an irregular wave which is composed of 8 linear components.4 − 0.7 m0 Try to ﬁnd out the principle of getting this empirical relation.3 2 4.8 6.3 The recording length is 20 seconds.8 9. (3) Calculate total number of linear components to be given by Fourier analysis N .6 14 5 3. (only for those who have interest.2 7 2.5. and the nyquist frequency fnyquist . Draw the variance diagram and variance spectrum of the irregular wave.0 10. (1) Determine Hs and Ts by timedomain analysis.4 4 3. 2) Convert the variance spectrum obtained in exercise 1) into time series of surface elevation. Suppose the sample frequency is 3 Hz and the recording length is 500 seconds.2 TIMESERIES ANALYSIS II: FREQUENCYDOMAIN 2.2 5 8 0.) 4) In reality where ε = 0. wave height H (m) wave period T (s) 1 5. 25 .3 3.3 7 6 2. (4) Draw the variance spectrum of the irregular wave by FFT analysis. a good estimate of signiﬁcant wave height from variance spectrum is √ Hs = 3. frequency band width ∆f . Wave height and period of each component are given in exercise 1).7 Exercise 1) An irregular wave is composed of 8 linear components with wave no.
viscosity and breaking.1 Wave development and decay Wind waves grow as a result of a ﬂux of energy from the air into the water. The other is called spectrummethod. which is the modiﬁcation of SverdrupMunkBretschneider method (SMBmethod). This approach will not be discussed in detail because the application of such models require specialized expertise. 1984). Then the wavewave interaction transfers energy in shorter waves to energy in longer waves. Two simpliﬁed methods have been used to determine wave characteristics from a known wind ﬁeld. Wind energy can be transferred to the waves only when the component of surface wind in the direction of wave travel exceeds the speed of wave propagation. When the wind velocity near the water surface exceeds a critical value of about 1 m/s. Besides these two simpliﬁed method. Therefore a change in wave energy depends on the transformation of the wind’s kinetic energy into the wave energy. the dissipation of wave energy into turbulence by friction. The process of wave development is complex. one can observe water surface ripples of length 510 cm and height 12 cm. The one is called SPMmethod (Shore Protection Manual. the advection of wave energy into and out of a region. 3. In Shore Protection Manual (1984) the peak period is denoted Tm . there are also numerical methods solving a diﬀerential equation governing the growth of wave energy. Notice that Hm0 is the signiﬁcant wave height determined from variance spectrum. First the windwave interaction transfers wind energy to shorter waves. which gives variance spectrum in terms of wind ﬁeld.3 WINDGENERATED WAVES 3 Windgenerated waves If a structure is to be built at the location where there is no direct wave measurement. Waves begin to decay when winds decrease in intensity or change in direction. When required. 6 26 . thus resulting in the growth of longer waves. wave characteristics may be estimated by wind data. or waves propagate out of wind ﬁeld. the transformation of wave energy at one frequency into wave energy at other frequencies. SPMmethod gives signiﬁcant wave height (Hm0 ) and peak period (Tp ) in terms of wind ﬁeld 6 . a signiﬁcant wave height and peak period can be estimated from the spectrum and the results will be the same as SPMmethod.
The fetch is equal to the average of the length of these 9 radials. edited by the US Army Corps of Engineers. Coastal Engineering Research Center (CERC). 27 . i. Elevation. SPM (1984) recommends to construct 9 radials from the point of interest at 3degree intervals and to extend these radials until they ﬁrst intersect shorelines. involves the airsea temperature diﬀerence. If the given wind speed is not measured at the 10 meter elevation.e. all these factors are accounted for by using UA a. 9 Fi F = i=1 9 (31) 2 Wind stress factor (UA in (m/s)): Wind stress is most directly related to wave growth. sea surface roughness and friction velocity. Because the fetches surrounding the wind direction will inﬂuence the wind generated waves. In SPM (1984).2 SPMmethod This method is presented in Shore Protection Manual (1984). The accurate estimation of vertical proﬁle of wind speed. and hence wind stress. Involved parameters 1 Fetch (F in (m)): Fetch is the distance between the point of interest and shoreline in the upwind direction.3 WINDGENERATED WAVES 3. the wind speed must be adjusted accordingly by U10 = Uz 10 z 1/7 for z < 20 m (32) where U10 and Uz are wind speed at the elevation of 10 m and z m respectively.
If wind data over water is not available. 3 Wind duration (t in (s)) and water depth (h in (m)). and averaged over an appropriate duration. If ∆T is positive.3 WINDGENERATED WAVES b. Windstress factor.23 UA = 0.9 log10 45 t for 1s < t < 3600s (34) = 1.5334 − 0.277 + 0.296 tanh 0.15 log10 (t) for 3600s < t < 36000s (35) where Ut is the average wind speed in t seconds. If ∆T is negative. 28 . the boundary layer is unstable and wind speed is more eﬀective in causing wave growth. Location eﬀects. they should be corrected by U = 2. It should be noted that the unit of U10 and UA is m/s because eq (36) has no unithomogenity.1 can be applied. the boundary layer is stable and wind speed correction is unnecessary. the boundary layer is unstable and the wind speed is less eﬀective. The windstress factor is implemented in order to account for the nonlinear relationship between wind stress and wind speed.2 gives the wind speed ampliﬁcation factor (RT ) due to airsea temperature diﬀerence. d. 1. If wind speeds is estimated by visual observations on ships.71 U10 (36) where U10 is the wind speed at the height of 10 m over mean water level. The wind speed is often observed and reported as the maximum shortdurationaveragedspeed.16 Us 9 (7) (33) where Us is the shipreported wind speed in knots and U is the corrected wind speed in knots. This should be converted to the wind speed averaged in an appropriate duration by Ut Ut=3600s Ut Ut=3600s = 1. In the absence of temperature information RT = 1. modiﬁed according to location and airsea temperature.1 can be used to convert overland winds to overwater winds if they are the result of the same pressure gradient and the only major diﬀerence is the surface roughness c. Durationaveraged wind speed. Fig. Stability correction. Fig. but data from nearby land site are. e. If the airsea temperature diﬀerence (∆T = Tair −Tsea ) is zero.
Fig. Ratio of wind speed over water (Uw ) to wind speed over land (UL ) (scanned from SPM 1984).1. Ampliﬁcation factor accounting for the eﬀect of airsea temperature diﬀerence (scanned from SPM 1984).2.3 WINDGENERATED WAVES Fig. 29 .
This wave condition refers to the case where the waves have reached an equilibrium state in which energy input from the wind is exactly balanced by energy loss. .3 WINDGENERATED WAVES We may expresses the signiﬁcant wave height and peak period in functional forms Hm0 . But there is a limit. waves UA UA become fullyarisen. a larger fetch gives a larger wave height and longer wave period. eq (310) is valid up to g 2 = 23123. t. the socalled fully arisen sea.0016 = 0. Tp = f ( UA .2857 ( 23123 ) 1/3 g Tp UA = 8.2857 g F 2 UA g F 2 UA 1/2 1/3 (310) Eq (310) shows. 1 h Deep water ( L > 1 ): The condition for deep water waves to be fetchlimited 2 is that the wind duration is longer than the minimum necessary duration tmin . where g = 9.243 (312) = 0. given by g tmin = 68. When g 2 > 23123.0016 ( 23123 )1/2 = 0.134 30 . and the signiﬁcant wave height and peak period are g H m0 2 UA = 0. Fetchlimited case It is the situation where the wind has blown constantly long enough for wave heights at the end of the fetch to reach equilibrium.81 (m/s2 ) is the gravitational acceleration. The fully arisen sea occurs when gF ≥ 23123 2 UA (311) F F That is to say. .8 UA g Hm U2 0 A g Tp U A gF 2 UA 2/3 (39) Signiﬁcant wave height and peak period under fetchlimited condition are = 0. F. h ) A dimensional analysis applied to eq (37) gives g Hm0 2 UA g Tp = f UA gF 2 UA gt UA gh 2 UA (38) (37) .
53 g F 2 UA 3/4 g h 2 UA 1/2 g Tp = 7. Durationlimited It is the situation where the wind duration is shorter than the minimum necessary duration.0379 g F 2 UA tanh 0. 31 . i.54 tanh 0.3 WINDGENERATED WAVES 2 h Transitional or shallow water ( L < 1 ): Waves feel the eﬀect of sea bottom. SPM (1984) suggests to make use of the formulae for the fetchlimited situation. SPM (1984) suggests the following formulae g Hm0 2 UA = 0.e. duration limited 2) Replace tmin by t in eq (39) and calculate the ﬁctional fetch F 3) Calculate Hm0 and Tp by eq (310) where the fetch is the ﬁctional fetch. It proceeds as 1) Check out t < tmin .833 UA gh 2 UA 3/8 tanh 0. For the same wind speed and fetch. There is no generally accepted formula. 2 Some part of wave energy dissipates due to bottom friction and percolation.53 gh 2 UA 3/4 tanh 0. wave height will be smaller and wave period shorter in comparison with deep water situation.00565 tanh 0.283 tanh 0.833 3/8 g h 2 UA 1/3 g tmin = 537 UA g Tp UA 7/3 SPM (1984) calls the above formulae ’interim formulae’ because the modiﬁcation is ongoing in order to make the above formulae consistent with deep water.
4.8 .1 m/s. The observation site is at an airport weather station.5 = 19. 1 mile = 1609 m.1 m/s is the average velocity in 72. The airsea temperature diﬀerence is estimated to be −60 C. Location adjustment From Fig.8 = 1. It should be converted to the average wind velocity in one hour.e.14. the wind speed is adjusted to 1.1) 1. Fastest mile wind speed is the fastest wind speed.22 = 18.1 m/s 5.8 s.8 Ut=3600 = 1. approximately 5 kilometers inland from shore. Hm0 and Tp for the fetch 100 kilometers at a deep water location. Duration adjustment The duration over which the fastest mile wind speed is averaged is actually the time needed for the fastest mile wind speed to travel one mile. averaged over the duration equal to the time needed for the fastest wind speed to travel 1 mile. the wind velocity of 22.8 s 22. Temperature adjustment From Fig. 3.9.9 log10 45 72.4 = 22. denoted as Ut=72. because in this example the fastest mile wind speed is given on hourly basis.2 it is found RT = 1.71 (18. The wind stress factor is UA = 0. Elevation adjustment U10 = Uz 10 z 1/7 Wanted Solution = U0 10 6 1/7 = 21.277 + 0.1 it is found RL = 0.22 Ut=3600 = Ut=72.5 m/s 2.1 i.9 × 21.3 WINDGENERATED WAVES Example Given Application of SPMmethod Eight consecutive hourly observations of fastest mile wind speed U0 = 20 m/s are observed at an elevation z = 6 m.4 m/s. t = 1609 = 72.14 × 19.296 tanh 0. Ut=72.8 /1.23 = 25 m/s 32 . the wind speed is adjusted to 0. We proceed as follows 1.
04 m 1/3 Tp = 0.2857 UA g = 8.6 hours < 8 hours Therefore it is fetchlimited condition. the minimum necessary wind duration is tmin = 68.8 gF 2 UA 2/3 UA g = 23688 s = 6.0016 gF 2 UA gF 2 UA 1/2 2 UA g = 4. Because gF 2 = 1568 < 23123 UA it is not fully arisen sea. Type of wind wave The given fastest wind speed indicates that wind is constant in 8 hours.3 WINDGENERATED WAVES 6.47 s 33 . Hm0 and Tp are given by Hm0 = 0. 7.
74 Sη (f ) = 4 (2π) f α = 0.5 m above mean water level g : Gravitational acceleration PM spectrum has been transformed to parameterized spectrum by Hs = 4 Tp = 1. Out of the three analytical expressions suggested by Pierson and Moskowitz.3 Standard variance spectrum PM spectrum In 1964. W.5 : Wind speed. The only variable is thus the wind velocity.5 )−1 U19.Moskowitz put forward. 34 . on the basis of a similarity theory by S. that spectra of this type are only valid when the fetches are large enough to reach this equilibrium.A. This spectrum is called PM spectrum.4 T = 1.3. αg 2 −5 f0 f exp −0. It is important to emphasize.4 m0 m1 4 (313) √ m0 and 5 2 4 −5 5 Hs fp f exp − Sη (f ) = 16 4 fp f 4 (314) Fig.0081 f0 = g (2πU19. some suggestions for deep water wave spectra for the sea state referred to as fully arisen sea.Pierson and L. 19.Kitaigorodskii.J. This wave condition refers to the case where the waves have reached an equilibrium state in which energy input from the wind is exactly balanced by energy loss. Example of PM spectrum.3 WINDGENERATED WAVES 3. the one below was found to give the best agreement with empirical wave data.
09 : f ≤ fp f > fp Peak enhancement coeﬃcient Wind speed.07 = 0. UK and USA. During the processing of a large number of spectra corresponding to steady easterly wind.33 U10 = 0. and partly to investigate wave transformation from sea to shallow water area.5 g x−0.9 + γ β = exp − σ ≈ 0. The objectives of the project was originally partly to investigate the growth of waves under fetchlimited condition. the socalled JONSWAP spectrum was obtained αg 2 −5 5 f exp − Sη (f ) = 4 (2π) 4 = 0.09 f ≤ fp f ≥ fp γ: Peak enhancement coeﬃcient 35 . 10 m above mean water level U10 : The parameterized JONSWAP spectrum reads 2 4 Sη (f ) = α Hs fp f −5 γ β exp − 5 4 fp f 4 (316) α ≈ 0.076 x−0.185 1.07 σ ≈ 0.22 −2 = g F U10 f fm −4 γ exp − 1 2σ 2 ( ff −1) m 2 (315) where α x fm σ σ γ = 3.3 WINDGENERATED WAVES JONSWAP spectrum The Joint North Sea Wave Project (JONSWAP) was started in 1967 as a collaboration among institutes in Germany.0336 γ − (f − fp )2 2 2 σ 2 fp 0. Simultaneous measurements of waves and winds were taken at stations along a line extending 160 km in a westerly direction from the island of Sylt in the Germany Bright. Holland.0624 0.230 + 0.
Furthermore.3 WINDGENERATED WAVES The JONSWAP spectrum is characterized by a parameter γ. 36 . Bretsneider (1959). which controls the sharpness of the spectral peak. Remarks on standard spectra Actually wave spectra usually exhibit some deviations from these standard spectra. Scott (1965). Example of JONSWAP spectrum. the socalled peak enhancement parameter. There are still other standard spectra. one associated with swell and the other with locally generated waves. One of the few reports on pure swell spectra indicates that it can be approximately described by the JONSWAP spectrum with relatively larger γ value (Goda 1985). In the North Sea the γ value ranges from 1 to 7 with the mean value 3. Fig. Fig.g.4.4. Ochi et al. the above mentioned spectra are onedimensional and valid only for deep water. e.3. the available information is insuﬃcient because many swell records are contaminated by local wind waves. Concerning the spectrum of swell. Mitsuyasu (1971. Darbyshire (1952). cf. many researches are ongoing. With respect to shallow water wave spectrum and directional spectrum. 1972) and the ISSC spectrum. (1976) presented a spectrum which has two peaks.
is considered a product of the unidirectional spectrum. θ) dθ = 1 (318) to assure identical wave energy in Sη (f. There are mainly two types of spreading functions. i. θ). θ) is given in Fig.5. θ). D(f. θ) × Sη (f ) (317) where f is the wave frequency. θ) must satisfy π −π D(f.3 WINDGENERATED WAVES 3. θ) = D(f. the spreading function D(f. θ) and Sη (f ). one using s as a spreading parameter and the other using σ as a spreading parameter. θ) is treated as a function of only θ. Sη (f ).4 Introduction to shortcrested waves The directional wave spectrum. D(f. i.e. the same zero moment m0 = ∞ 0 ∞ 0 π −π Sη (f. Sη (f. 5. θ the angle of wave propagation. For simplicity.e. θ) dθ df = ∞ 0 π Sη (f ) −π D(f. and a spreading function. Sη (f. Fig. An example of directional wave variance spectrum. 37 . θ) dθ df = Sη (f ) df (319) An example of Sη (f.
is angle in degree. With the spreading parameter σ the spreading function is assumed to follow the Gaussian distribution D(θ) = √ 1 (θ − θm )2 . in this case.3 WINDGENERATED WAVES The cosinepower spreading function proposed by LonquetHiggins is using s as a spreading parameter D(θ) = θ − θm 22s−1 Γ2 (s + 1) . 38 . cos2s π Γ(2s + 1) 2 −π ≤ (θ − θm ) ≤ π (320) where s is a spreading parameter. The spreading function D(θ) by LonquetHiggins.6. which. The equation was found to provide a reasonable ﬁt to measured ocean wave spectra by LonquetHiggins. Γ the Gamma function and θm the mean wave incident angle. In Fig. 6 the spreading function D(θ) is plotted as a function of (θ − θm ) for various values of the spreading parameter s. Cartwright and Smith (1961). Therefore σ is also called spreading angle. Fig. exp − 2σ 2 2π σ −π ≤ (θ − θm ) ≤ π (321) It can be seen that σ is the standard deviation of the spreading function.
6 Exercise 1) The wind speed measured at the elevation of 5 m is U5 = 20 m/s. Van Nostrand Reinhold Company. H. 1985. University of Tokyo Press. 1981. Lecture on Oﬀshore Engineering. Denmark. Random seas and design of marine structures. fetch 200 kilometers. US Army Corps of Engineers. Coastal Engineering Research Center. 1985 Sarpkaya. ISBN 0442254022.5 References Burcharth. 2) Convert the fastest mile wind speed Uf = 29 m/s to twentyﬁveminute average wind speed Ut=25 min . Aalborg University. Y. 1984. Waterway Experiment Station. 1981. M. 1978.5 to be used in the PMspectrum. 3. Japan. 3) Calculate Hm0 and Tp with deep water situation.J. Mechanics of wave forces on oﬀshore structures. New York. On the design of gravity structures using wave spectra. ThoftChristensen. T. Goda. and Brorsen. Shore Protection Manual. 1984. .Graﬀ and P. 39 . SPM .3 WINDGENERATED WAVES 3. M. wind speed at z = 5 m over water surface is 20 m/s over 2 hours. Institute of Building Technology and Structural Engineering. .F. Calculate U19. Edited by W. . and Isaacson.
The nonexceedence probability of x is F (x). Number of observations in a period of t. or the exceedence probability of x is (1 − F (x)). It varies with respect to time and location. which is a random variable due to the statistical vagrancy of nature. In other words with (1 − F (x)) probability an observed signiﬁcant wave height will be larger than x. Cumulative distribution function of X. Fig. If a structure is to be built in a location of sea where a longterm wave height measurement/hindcast is available. The number of observations where (X > x) is k = n ( 1 − F (x) ) = t λ ( 1 − F (x) ) 40 (41) .1. Signiﬁcant wave height is a random variable.1 Design level: Return period and encounter probability The design level is represented by return period or encounter probability. it is a method to determine the design wave height. i. Cumulative distribution function of X. λ = n/t. Sample intensity. Number of years of observation of X. Return period T To deﬁne return period the following notations are used X Signiﬁcant wave height.4 EXTREME WAVE HEIGHT ANALYSIS 4 Extreme wave height analysis The design wave height is often represented by signiﬁcant wave height. x F (x) t n λ Realization of X. the question an engineer must answer is: How to determine the design wave height ? Extreme wave height analysis gives the answer to that question. If the total number of observations is n. Fig. based on the importance of the structure (design level) and the statistical analysis of a longterm wave height measurements/hindcast.1 illustrates the cumulative distribution function of X.e. F (x) = Prob(X ≤ x). 4.
the exceedence probability of x in 1 year is 1/T . For example If the structure lifetime L is 25 years.e. Therefore nonexceedence probability of x in 1 year nonexceedence probability of x in 2 years nonexceedence probability of x in L years Prob(X ≤ x) = 1 − Prob(X ≤ x) = Prob(X ≤ x) = 1 T 1 T 1 T 2 L 1 − 1 − and the encounter probability.e. 41 . In the reliability based design of coastal structures it is better to use encounter probability. i.4 EXTREME WAVE HEIGHT ANALYSIS The return period T of x is deﬁned as T = t = 1 λ ( 1 − F (x) ) (42) k=1 i. For example. the physical meaning is that on average this 10 m design wave height will be exceeded once in every 100 years. the exceedence probability of x within a structure lifetime of L years is p = 1 − 1 − 1 T L (43) which in the case of larger T can be approximated p = 1 − exp − Design level Traditionally the design level for design wave height was the wave height corresponding to a certain return period T . x is also called T year event.e. i. the exceedence probability of the design wave height within the structure lifetime. if the design wave height corresponding to a return period of 100 years is 10 m. Encounter probability p Based on the fact that on average x will be exceeded once in every T years. on average x will be exceeded once in every T years. the encounter probability of the design wave height (10 meter) is p = 1 − 1 1 − T L L T (44) = 22% This means that this 10 m design wave height will be exceeded with 22% probability within a structure lifetime of 25 years.
It is also called POT data set (Peak Over Threshold). composed of the largest wave height in each individual storm exceeding a certain level (threshold).2. 42 . The threshold is determined based on the structure location and engineering experience. Fig. If the followings will be discussed the procedures one by one. Complete data set Annual series data set Partial series data sets containing all the direct measurements of wave height usually equally spaced in time. 4. say less than 10 years in the case of direct measurements and less than 40 years in the case of hindcasts.2 General procedure In practice engineers are often given a longterm signiﬁcant wave height measurement/hindcast and required to determine the design wave height corresponding to a certain return period. cf. In practice three kinds of extreme data sets have been used.3 Data sets The original wave data are typically obtained either from direct measurements or from the hindcasts based on the meteorological information. a plotting position formula must be used 4) Choice of the distribution based on the comparison of the ﬁtting goodness among the candidates 5) Calculation of the design wave height corresponding to a certain return period 6) Determination of the conﬁdence interval of the design wave height in order to account for sample variability.2. cf. If the least square ﬁtting method is employed. Most of the measurements/hindcasts cover a rather short span of time. we can calculate the return period by eq (43) and proceed as above. The general procedure to perform the task is: 1) Choice of the extreme data set based on a longterm wave height measurement/hindcast 2) Choice of several theoretical distributions as the candidates for the extreme wave height distribution 3) Fitting of the extreme wave heights to the candidates by a ﬁtting method. measurement/hindcast error and other uncertainties If structure lifetime and encounter probability are given in stead of return period. consisting of the largest wave height in each year of measurements/hindcasts. Fig.4 EXTREME WAVE HEIGHT ANALYSIS 4.
However. e. Average variations exist from decades to decades or even longer period of time. Moreover.4 EXTREME WAVE HEIGHT ANALYSIS Fig. should fulﬁll the following 3 conditions: Independence There must be no correlation between extreme data. For these reasons the complete data set is seldom used. Studies of wave data for the North Sea from the last 20 years give evidence of nonstationarity as they indicate a trend in the means. Illustration of the establishment of annual series data set and partial series data set. i. 43 . The annual series data set and the partial series data set meet the independence requirement because the extreme data are from diﬀerent storms. Goda (1979) found correlation coeﬃcients of 0. what is interesting in the case of design waves is the wave height corresponding to a very high nonexceedence probability.5 for signiﬁcant wave heights (measurement duration is 20 minutes and time interval between two succeeding measurements is 24 hours). until more progress is available in investigating longterm climatological variations. Most engineers prefer the partial series data set over the annual series data set simply because the former usually gives larger design wave height and hence. because the longterm climatological variation is generally very weak. all extreme data are from windgenerated waves.3 − 0. The complete data set cannot fulﬁll the required independence between data. If the chosen distribution is not the true one. more conservatively designed structures. established based on the original wave data.2. the very upper tail of the distribution. Homogeneity The extreme data must belong to the same statistical population. the assumption of stationary statistics might be considered realistic for engineering purpose. The extreme data sets. Stationary There must be stationary longterm climatology. the very upper tail value will be distorted severely because in the ﬁtting process the chosen distribution will be adjusted to the vast population of the data.g.e.
the Frechet distribution and the Lognormal distribution are the theoretical distributions which ﬁt the extreme wave data well. In the lognormal distribution A and B are the standard deviation and the mean of X respectively.4 Candidate distributions Generally the exponential distribution. which could be the signiﬁcant wave height Hs or the onetenth wave height H 1 or the maximum wave height Hmax . the North Sea. Pozzallo and Follonica in Italy and Western Harbour in Hong Kong. the Weibull distribution. The author have tried to ﬁt 7 sets of partial series data to all these distributions. x F A.4 EXTREME WAVE HEIGHT ANALYSIS 4. Nonexceedence probability of x (cumulative frequency). depending 10 on the extreme data set. 44 . Exponential F = FX (x) = P (X < x) = 1 − e−( Weibull F = FX (x) = P (X < x) = 1 − e−( −e x−B A ) k (45) x−B A ) (46) Gumbel F = FX (x) = P (X < x) = e − x−B A ( ) (47) Frechet F = FX (x) = P (X < x) = e−( A ) x k (48) Lognormal F = FX (x) = P (X < x) = Φ where X ln(x) − B A (49) A characteristic wave height. These data sets are real data representing deep and shallow water sea states from Bilbao in Spain. The results show that the Weibull and the Gumbel distributions provide the closest ﬁts. Distribution parameters to9 be ﬁtted. Sines in Portugal. Realization of X. the Gumbel (FTI) distribution. Standard normal distribution function. B. k Φ No theoretical justiﬁcation is available as to which distribution is to be used. Therefore the following discussion is exempliﬁed with these two distributions. Tripoli in Libya.
thus obtaining a set of data pairs. n (X1 =max). i = 1. n. The ﬁnal values of the three parameters are chosen based on the ﬁtting 45 . the method of moment. the least square method and the visual graphical method.5 Fitting methods and procedure Four generally applied methods of ﬁtting the extreme data set to the chosen distributions are the maximum likelihood method.4 EXTREME WAVE HEIGHT ANALYSIS 4. (yi . 2. · · · . (xi ). xi ). thus obtaining a new set of data pairs. Assign a nonexceedence probability Fi to each xi by an appropriate plotting position formula (cf. (Fi . The most commonly used methods are the maximum likelihood method and the least square method. 2. X) = (yi − Y )(xi − X) n i=1 Y = 1 n yi n i=1 X= 1 n xi n i=1 In the case of the Weibull distribution various k values are predeﬁned and A and B are ﬁtted accordingly. · · · . xi ). X) V ar(Y ) B = X − AY 2 3) 4) V ar(Y ) = 1 n yi − Y n i=1 1 n Cov(Y. · · · . i = 1. 2. Least square method Eqs (46) and (47) can be rewritten as X = AY +B where Y is the reduced variate deﬁned according to the distribution function Y Y = (−ln(1 − F )) k = −ln(−lnF ) 1 (410) Weibull distribution Gumbel distribution (411) (412) The ﬁtting procedure is summarized as the follows: 1) 2) Rearrange the measured/hindcast extreme data (total number n) in the descending order. Calculate the corresponding y value by eq (411) or eq (412). i = 1. n. next section). Determine the regression coeﬃcients of eq (410) by A= Cov(Y.
Maximum likelihood method The 2parameter Weibull distribution is Weibull F (x) = 1 − e − x−x A k (413) where x is the threshold wave height. and the one which produces best ﬁt is ﬁnally chosen. which should be smaller than the minimum wave height in the extreme data set. the maximum likelihood estimate k is obtained by solving the following equation by an iterative procedure: N N N +k i=1 ln (xi − x ) = N k i=1 (xi − x ) ln (xi − x ) i=1 k N −1 (xi − x ) k (414) The maximum likelihood estimate of A is A = 1 N N i=1 1/k (xi − x ) k (415) For the Gumbel distribution. the maximum likelihood estimate of A is obtained by solving the following equation by an iterative procedure: xi xi exp − A i=1 N 1 = N N i=1 xi − A N i=1 exp − xi A (416) The maximum likelihood estimate of B is B = A ln N N i=1 exp − xi A −1 (417) 46 . For unexperienced engineers several threshold values can be tried.4 EXTREME WAVE HEIGHT ANALYSIS goodness.
median and mode of the Gumbel random variable. can be determined based on three diﬀerent statistical principles.5 = B + 0. the median and the mode are Mean xmean = E[X] = +∞ −∞ xfX (x)dx ≈ B + 0. Mean.4 EXTREME WAVE HEIGHT ANALYSIS 4. The nonexceedence probability (Fi ) to be assigned to (xi ). Take the Gumbel distribution as an example. The distribution function FX (x) and density function fX (x) of a Gumbel random variable X reads FX (x) = P (X < x) = e−e − x−B A ( ) fX (x) = dFX (x) dx (418) The deﬁnition and value of the mean. distribution of frequency and order statistics. median and mode The deﬁnition of mean. 3. median and mode of a random variable X is given in the following because they are involved in some of the plotting position formulae.367A = B fX (x)=max Fig. cf Burcharth et al (1994). namely sample frequency. Mean. The plotting position formula is used to assign a nonexceedence probability to each extreme wave height.6 Plotting position formulae When the least square method is applied a plotting position formula must be chosen.577A (419) (420) (421) Median xmedian = x Mode xmode = x FX (x)=0. The plotting position is of special importance when dealing with very small samples. 47 .
fX1 . fX1 : density function of X1 .e. 3.(1982). i = 2.4 EXTREME WAVE HEIGHT ANALYSIS Order statistics Assume that a random variable X has a cumulative distribution function FX . FX (x) = P (X < x) (422) Furthermore. fX . The distribution function of X1 is FX1 (x) = P (X1 < x) = (FX (x))n = (P (X < x))n (423) FX1 (x) may also be interpreted as the probability of the nonoccurrence of the event ( X > x ) in any of n independent trials. cf. are sketched in Fig. ThoftChristensen et al.4. 48 . fX : Density function of X. The distribution functions FXi (x) can also be expressed as the function of FX (x). deﬁned as the largest value in each sample. assume n data sampled from X and arranged in the descending order. x1 being the largest value in n data. Fig. 4. · · · . and the density function of X1 . and probability density function fX . The density function of X. Here x1 is one realization of the ordered random variable X1 . n. i. For other ordered random variables Xi .
Benard (1943) developed a good approximation Benard Fi ≈ 1 − i − 0.3 n + 0. Weibull (1939) used the mean of FXi (xi ) to determine the cumulative frequency Fi to be assigned to xi Weibull Fi = = 1 − i n+1 (425) There is no explicit formula for the median of FXi (xi ).e. Sample size. n Extreme data in the descending order (x1 =max) Nonexceedence probability of xi . too. FXi (xi ). the cumulative frequency of xi . · · · . 2. i. Consequently. the median or the mode of the random variable FXi (xi ). The philosophy of this method is to determine the plotting position of xi via either the mean. too. However. Plotting position based on distribution of frequencies Assume that the random variable X has a cumulative distribution function FX . (424) The disadvantage of this plotting position formula is that the smallest extreme data xn cannot be used because Fn = 0. 49 . The widely used formula is the socalled California plotting position formula Fi = 1 − where xi Fi n i n i = 1. because the chance of the occurrence of mode is still inﬁnitesimal even though mode is more likely to occur than the mean and median. total data number. The i’th highest value in n samples. Xi .4 (426) The plotting position formula based on the mode of FXi (xi ) has not drawn much attention.4 EXTREME WAVE HEIGHT ANALYSIS Plotting position based on sample frequency This method is based solely on the cumulative frequency of the samples. is a random variable. The plotting position formula by this method is independent of the parent distribution (distributionfree). is a random variable.
3−0.4 EXTREME WAVE HEIGHT ANALYSIS Plotting position based on order statistics The philosophy of this method is to determine the plotting position of xi via the mean.18/k n+0. 5.44 n+0. In practice the Weibull plotting position formula is most widely used. Rosbjerg (1988) advocates the choice of the median plotting position formula (Benard formula) because it is distributionfree and is based both on the distribution of frequency and the order statistics. The best known approximations are Blom Gringorten Petrauskas Fi = 1 − Fi = 1 − Fi = 1 − i−3/8 n+1/4 i−0. the median and the mode of X1 .20+0. Illustration of the determination of F1 based on the mean.32/k √ i−0. From the statistical point of view the plotting position formula based on the mean (unbiased) is preferred because the expected squared error is minimized. Fig.21+0.2−0.12 Normal distribution Gumbel distribution Weibull distribution Weibull distribution (427) (428) (429) (430) i−0. Summary on plotting position formulae The choice of the plotting position formula depends on engineer’s personnel taste. 50 .27/ √ k n+0.23/ k Goda Fi = 1 − The plotting position based on the median value of the ordered random variable is the same as that based on the median value of distribution of frequency. Plotting positions based on the mean value are distributiondependent and not explicitly available. the median and the mode of the ordered random variable Xi .
The linear correlation coeﬃcient.4 EXTREME WAVE HEIGHT ANALYSIS 4. Obviously a smaller Evalue indicates a better ﬁtted distribution. Y ) V ar(X) V ar(Y ) is widely used as the criterion for the comparison of the ﬁtting goodness. are rewritten as x = A (−ln(1 − F )) k + B Weibull distribution x = A (−ln(−ln(F ))) + B Gumbel distribution 1 (433) (434) Deﬁne the sample intensity λ as λ = number of extreme data number of years of observation (435) 51 . cf. eqs (433) and (434). the wave heights corresponding to the nonexceedence probability of the observed wave heights can be calculated. where the reduced variate y is dependent on the distribution function. The Weibull and Gumbel distributions.observed  n i=1 xi. eqs (46) and (47). x). The statistical hypothesis test can also be used in the comparison of the ﬁtting goodness (Goda et al. However. deﬁned as 1 n  xi. With the ﬁtted distribution functions.7 Fitting goodness Normally several candidate distributions will be ﬁtted and the best one is chosen. ρ is deﬁned in the linear plotting domain (y.estimated − xi.observed (431) E = (432) is a good simple criterion with a clear interpretation. the central estimation of wave height deviates from the observed wave height by 5 %. 1990) 4. E = 5 % means that on the average. The average relative error E.8 Design wave height: xT The design wave height xT is the wave height corresponding to the return period T . the interpretation of this criterion is less clear. deﬁned as ρ = Cov(X. Therefore.
B and k are the ﬁtted distribution parameters. Remark Some students have a confusion between the shortterm distribution of individual wave heights and the longterm distribution of extreme wave heights. The confusion can be cleared by the following ﬁgure. and therefore is replaced by xT ) xT = 1 A −ln( λT ) 1 k +B +B Weibull distribution Gumbel distribution (437) (438) xT = A −ln(−ln(1 − 1 )) λT where A. we get (now x means the wave height corresponding to return period T .4 EXTREME WAVE HEIGHT ANALYSIS and employ the deﬁnition of return period T T = 1 1 or F = 1 − λ(1 − F ) λT (436) Inserting eq (436) into eqs (433) and (434). 52 .
389 0.33 nonexceedence probability Fi 0.06 6.111 0.55 0.03 5. Calculate the values of the reduced variate {yi } according to eqs (411) and (412) respectively.80 0.667 0. The steps in the analysis are as follows: 1) 2) 3) Calculate the sample intensity by eq (435) λ = 17 20 Calculate the return period by eq (43) T = 100 years Assign a nonexceedence probability Fi to each observed wave height xi according to the Weibull plotting position formula.778 0.92 0. 4) 53 . {yi } of the two distributions are also shown in Table 2.71 0.86 2.333 0.06 0.86 0.72 4.11 1.74 0.67 4. The ranked signiﬁcant wave heights are listed in Table 2.278 0.57 1.222 0.64 4.90 4.79 1.28 1.14 1. Choose the Weibull and the Gumbel distributions as the candidate distributions.78 4.98 0.19 7. For the Weibull distribution {yi } involves the iterative calculation.944 0.32 8.056 Reduced variate yi Gumbel 2.41 0.73 2.19 3.4 EXTREME WAVE HEIGHT ANALYSIS 4.11 7.500 0.18 1. Results are shown in Table 2.15 6.06 2.40 1.37 0.38 1.167 0.70 1.889 0. Table 2.611 0.9 Example Delft Hydraulics Laboratory performed a hindcast study for the Tripoli deep water wave climate and identiﬁed the 17 most severe storms in a period of 20 years. Tripoli storm analysis Signiﬁcant wave height xi (m) 9.556 0.68 0.92 4.12 0.444 0.06 Reduced variate yi Weibull (k = 2.53 0.04 0.40 0.58 0.722 0.90 0.30 rank i 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 You are required to ﬁnd the design wave height which has 22% exceedence probability within a structure lifetime of 25 years.49 0.37 6.21 0.25 0.62 0.35) 1.833 0.09 0.
Compare the goodness of ﬁtting according to the value of the average relative error E. k = 2. A = 1. xi ) to eq (410) by the least square method and obtain the distribution parameters: Weibull. B = 0.17. A = 5. eq (432) E = 4.73.53 The ﬁtting of the data to the Gumbel and the Weibull distributions is shown in Fig. 6. Fitting to the Gumbel and the Weibull distributions and comparison. 54 .89 Gumbel. 6.06 % for the Gumbel distribution ﬁtting Because of a clearly smaller Evalue the Weibull distribution is taken as the representative of the extreme wave height distribution Calculate the wave height corresponding to a return period of 100 years X 100 by eq (437) x100 = 10.4 EXTREME WAVE HEIGHT ANALYSIS 5) 6) 7) Fit data (yi .35.64 m Fig. B = 4.72 % for the Weibull distribution ﬁtting E = 6.
2 0. such as accelerometer and pressure cell.2 Conﬁdence interval of design wave height xT We use an example to demonstrate how the conﬁdence interval of the design wave height is determined. Visual observation data should not be used for determination of design wave height because ships avoid poor weather on purpose. Wave data set contains measurement/hindcast error.4 EXTREME WAVE HEIGHT ANALYSIS 4.7 is obtained by ﬁtting Tripoli signiﬁcant wave height to Gumbel distribution by the least square ﬁtting method and the Weibull plotting position formula. Burcharth (1986) gives an overview on the variational coeﬃcient C (standard deviation over mean value) of measurement/hindcast error. With the advance of measuring equipment and numerical model. variational coeﬃcient of extreme data C Methods of determination Accelerometer Pressure cell Vertical radar 0.15 0. The accuracy of such conversion depends on the quality of the pressure data and on the technique which is used to synthesize the data into the continues wave ﬁeld. 5) Climatological changes The uncertainty sources 1) and 2) can be considered by numerical simulation in the determination of the design wave height. 2) Error related to measurement. C 0.1 Horizontal radar Hindcast by SPM Hindcast other Visual Variational Coe. 55 .1. visual observation or hindcast.120. Table 1. Measurement error is from malfunction and nonlinearity of instruments. 3) Choice of distribution as a representative of the unknown true longterm distribution 4) Variability of algorithms (choice of threshold. The gumbel distribution curve in Fig. ﬁtting method etc.10 Sources of uncertainties and conﬁdence interval Sources of uncertainties The sources of uncertainty contributing to the uncertainty of the design wave height are: 1) Sample variability due to limited sample size.050. while hindcast error occurs when the sealevel atmospheric pressure ﬁelds are converted to wind data and further to wave data. generally C value has been reduced to below 0.10.2 0.
Fig. Design wave height.g. T = 100. is included. 56 . For example. Fig. the design wave height is 14.e. The obtained distribution parameters Atrue and Btrue are assumed to be the true values. it is assumed that the extreme wave height follows the Gumbel distribution F = FX (x) = P (X < x) = exp −exp −( x−B ) A (439) where X is the extreme wave height which is a random variable. Numerical simulation To exemplify the discussion. the design wave height is x100 = 12. x a realization of X. sample variability. can be obtained by numerical simulation to be described in the next section.7. If the design level for design wave height is a return period of 100 years. the distribution parameter A and B become random variables. In order to account sample variability. A sample with size N is ﬁtted to the Gumbel distribution. Due to the sample variability and measurement/hindcast error. i. a conﬁdence band is often applied. cf. The distribution of the design wave height x100 . cf. e.7. In order to account the sample variability and measurement/hindcast error.8 m which corresponds to the 90% onesided conﬁdence interval. a numerical simulation is performed as explained in the followings. A and B the distribution parameters. the design wave height x100 becomes a random variable. which is usually assumed to follow the normal distribution.2 m.4 EXTREME WAVE HEIGHT ANALYSIS Fig. If other uncertainties.7.
000 times. Fig. Simulated wave height taking into account measurement/hindcast error. 10. Let the nonexceedence probability F2 equal to that data. Fit the sample to the Gumbel distribution and get the new estimated distribution parameters A and B.1. Calculate the wave height xT corresponding to the return period T by eq (438) Repeat steps 2) to 4).000 values of xT . the modiﬁed extreme data xmodiﬁed is obtained by xmodiﬁed = x + C x Φ−1 (F2 ) (441) where Φ is the standard normal distribution and C is the coeﬃcient of variation of the measurement/hindcast error. 57 . In order to include the measurement/hindcast error the following step can be added after step 1).4 EXTREME WAVE HEIGHT ANALYSIS 1) Generate randomly a data between 0 and 1. Thus we obtain a sample belonging to the distribution of eq (439) and the sample size is N .8) −1 x = FX (F1 ) = Atrue [− ln (− ln F1 )] + Btrue (440) 2) 3) 4) 5) 6) Repeat step 1) N times. Choose the wave height corresponding to the speciﬁed conﬁdence band.8 1∗ ) Generate randomly a data between 0 and 1. say. cf. the single extreme data x is obtained by (cf. Thus we get 10. This step is to modify each extreme data x generated by step 1).8.05 to 0. based on the assumption that the hindcast error follows the normal distribution. Let the nonexceedence probability F1 equal to that data. Fig. Fig. C ranges usually from 0.
58 . In order to account sample variability. the design wave height is 14. cf.53. Fig.9.9. onesided conﬁdence interval is preferred over twosided conﬁdence interval because the lower bound of the conﬁdence band is of less interest. an 80% conﬁdence band is often applied. By ﬁtting the extreme data to Gumbel distribution we obtain the distribution parameters A = 1.73 and B = 4. Fig.4 EXTREME WAVE HEIGHT ANALYSIS Example Again the Tripoli deep water wave data is used as an example to demonstrate the determination of the design wave height and the inﬂuence of sample variability. Simulated distribution of x100 (sample variability).8 m which corresponds to 90% onesided conﬁdence interval. cf. In the case of wave height estimate. The distribution of the design wave height x100 can be obtained by numerical simulation. The design wave height corresponding to a return period of 100 years is 12. If sample variability is included.9. the design wave height x100 becomes a random variable. Fig. Therefore.2 m.
The ﬁgure is illustrative. Fig.4 EXTREME WAVE HEIGHT ANALYSIS 4. 59 .10 indicates as an example the diﬀerences in armour layer damage development for various types of rubble structures. In reality also the conﬁdence bands for the damage curves should be considered. Wave breaking occurs due to wave steepness (Stokes wave theory) or limited water depth (Solitary wave theory). Based on laboratory and ﬁeld observations. Goda (1985). The ﬁgure illustrates the importance of evaluation of prediction and conﬁdence limits related to the estimated design wave height.11 Physical consideration of design wave height Wave breaking The design wave height must be checked against wave breaking condition. Illustration of typical armour layer failure characteristics for various types of rubble mound structures (Burcharth 1993). especially in case of structures with brittle failure characteristics. but also on the character of the structural response.10. Fig.g. e. To such cases a lower damage level must be chosen for the mean value design sea state. many empirical formulae for wave breaking condition have been proposed. Structural response characteristics The choice of design wave height depends not only on the structure life time.
and still water level. Tm and z. and Hs and water level. due to the complexity and locality of the joint distribution between wave height and wave period. Fig.11 shows examples of scatter diagrams representing the joint distribution of signiﬁcant wave height. z. Tm .12 Wave period There is no theory to determine the design wave period corresponding to the design wave height obtained by the extreme analysis. and mean wave period. Then by theoretical consideration and/or laboratory investigation. DS449 gives the range of peak wave period 130 Hs < Tp < g 280 Hs g (442) 60 . In practice. several wave periods within a realistic range are simply assigned to the design wave height to form the candidates of the design sea state conditions. Hs . The numbers in the scatter diagrams are the number of observations falling in the corresponding predeﬁned intervals of Hs . Fig.11.4 EXTREME WAVE HEIGHT ANALYSIS 4. z. respectively. the one which is most dangerous is chosen. Scatter diagrams signifying examples of joint distributions of Hs and Tm .
12. 61 . which is the lowest tide level under the average meteorological conditions. Fig.13 Water level The sea water level is aﬀected by the following eﬀects: 1) Astronomical eﬀect: Tides generated by the astronomical aspect is the best understood due to their extreme regularity and the simplicity of observations. namely. cf. At a site without any previous tidal records usually one or a few month of recording will be suﬃcient to analyze the astronomical eﬀect on the water level. give reliable results. 2) Meteorological eﬀect: In shallow water the water level is also aﬀected by the meteorological eﬀects. If water level records are available for a long period of time. numerical models can. the meteorological eﬀect can be isolated from the astronomical eﬀect and subjected to the extreme analysis in order to establish the longterm statistics of the water level. Based on the established longterm statistics is given the design low water level and the design high water level. which gives also the widely used terminology and abbreviation of the various sea water levels. If such records are not available.4 EXTREME WAVE HEIGHT ANALYSIS 4. Fig. The astronomical tidal variations can be found in the Admiralty Tide Tables.12. Water depth. The extreme analysis should be performed on both the high water level and the low water level. It is diﬃcult to determine the meteorological eﬀect on the water level. i) Barometric: The higher barometric pressure causes a lower water level and vise versa. ii) Wind: Strong wind creates a setup of the water level on the downwind side and a setdown on the upwind side. 3) Earthquake The water depth read from the Chart Datum is the one corresponding to the Lowest Astronomical Tide. using wind and/or barometric chart.
If for example runup.g. wave period (e. Ru . When more sea state parameters have signiﬁcant inﬂuence on the impact on the structure considerations must be given to the probability of occurrence of the various possible combinations of the parameter values.i . . zi ) . i = 1. the wave direction.i . etc. Of importance is also the duration of the sea state and sometimes also the shape (type) of the wave spectrum. 2. . Tm . and the water level. . α (wave direction) and z (water level) then by hindcasting or/and measurements several data sets covering some years can be established (Hs. 2.i ) . Burcharth (1993) proposed the following principle for multiparameter extreme analysis: For the general case where several variables are of importance but the correlation coeﬃcients are not known the best joint probability approach would be to establish a longterm statistics for the response in question. . n For each data set the response in question is either calculated from formulae or determined by model tests. for the runup. αi . Hs ).g. Tm ). n The related longterm statistics can be established by ﬁtting to a theoretical extreme distribution (extreme analysis). the armour unit stability. is in question a single variable data set is obtained (Ru. the wave force on a parapet wall.g. If we assume that the variables of importance are Hs . .14 Multiparameter extreme analysis A sea state should be characterized at least by some characteristic values of wave height (e. because these four parameters are the most important for the impacts on the structures. e. 62 . i = 1.4 EXTREME WAVE HEIGHT ANALYSIS 4. . Tm.
University of Tokyo Press.F.. Structural reliability theory and its application. Design wave height related to structure lifetime. Japan. K. Proceedings of HYDROPORT’94. Introduction to probability and statistics for engineers and scientists. 1986. H. Copenhagen. Inc. 1939. Vet. Proc. H. . Statistical analysis in hydrology. Conf. Soc. . 1982.. Unbiased plotting positions – a review. .M.. 1994 Cunnane. Department of Civil Engineering. 1984. 1988.F. The design of breakwaters. 1984. 1987. 1993. and Burcharth. Trans. Y. and Michael J. 151. Rosbjerg. Proceeding of Seminar on Uncertainties Related to the Design and Construction of Oﬀshore Jacket Structures. ISBN 3540117318. Journ. ThoftChristensen. Orlando. John Wiley & Sons. Distribution function ﬁtting for storm wave data. 1996. Le Mehaute. 1994. Published by Danish Society of Hydraulic Engineering Burcharth. 1990.. 1979. and Zhou Liu . Aalborg University. Handl. 22nd Int. Y. Eng. (Stockholm). 1921 October. On the extreme wave height analysis. Am. ISVA. 21st Int. A defence of the median plotting position. 1985 Goda. Progress Report 66. Proc. Hydrology. C.R. PP 245305. Ing. Houston. Yokosuka. Kobune. 76. B. USA. Ak. 1978. Spain. SpringerVerlag. Rosbjerg. The Netherlands. pp 20522. 1985. 37. . . . Conf. Civ. Denmark. A statistical theory of strength of material. On the uncertainties related to the estimation of extreme environmental conditions. 1987. 1988. Random seas and design of marine structures . Japan. Eﬀects of measurement error on longterm wave statistics. H. P. of Hydrology. Port and Harbour Research Institute. Technical University of Denmark Ross. on Coastal Engr.F. 18(1). Goda. .4 EXTREME WAVE HEIGHT ANALYSIS 4. . 1979 Goda. USA. On the methodology of selecting design wave height. S. D. Proceedings of the 25th International Conference on Coastal Engineering. Baker . Weibull. pp 183195. 63 . Z.15 References Benard. 108. 1993 Burcharth. Proceedings of the 19th International Conference on Coastal Engineering. Y. 1943. 1986. A review on statistical interpretation of wave data. Y. 1985. pp 11101160 Burcharth.. L. D. . . . 1996. Estimation in partial duration series with independent and dependent peak values. Goda. . J.F .. ISBN 0471608157. 1982. and Shen Wang . H. on Coastal Engr. Liu. W.
3 9. 2) Taking into consideration sample variability.1 7.3 7.3 10.7 9.2 10.0 9.16 Exercise 1) The design wave height for Sines breakwater in Portugal is the signiﬁcant wave height corresponding to 100 years return period.8 10. Use the MonteCarlo simulation to draw the probability density function of the design wave height and calculate the upper bound of the design wave height corresponding to 90% conﬁdence.8 9.8 7.4 EXTREME WAVE HEIGHT ANALYSIS 4. 64 .9 6.0 6.1 9.6 8.3 10.8 8.7 Fit the data to Gumbel distribution by the least square method and the maximum likelihood method and calculate the design wave height. The hindcast study of Sines breakwater wave climate gave the following 17 severest storms in the period of 19701985: Hs in meter 12.
overtopping etc. Outline of the wave generator at AaU.1 Principle of wave generator Fig.1. runup. transformation and especially breaking) and wavestructure interactions (wave force. At the same time the servo mechanism receives information on the position of wave paddle through the displacement censor (feedback). the servo mechanism sends a control signal to the valve of hydraulic pump.) purely by mathematics.1 illustrates the basic concept of wave generators. 5.1 is called pistontype wave paddle. directcurrent motor and hydraulic pulse pump have also been applied. The wave paddle shown in Fig. After the comparison of the input signal with the paddle position. In reality there are over 20 types of wave paddles. Denmark. In stead of hydraulic pump. which converts the output of the hydraulic pump into the movement of the wave paddle.5 WAVE GENERATION IN LABORATORY 5 Wave generation in laboratory The importance of wave generation in laboratory is due to the fact that we cannot describe wave phenomenon (formation. Fig. and hence model tests play an important role in the design of coastal and oﬀshore structures. The input signal is the time series of voltage to be sent to servo mechanism. electric servomotor. 65 . Another popular one is hingedtype wave paddle.
t) = S(z) sin(ωt) 2 (51) paddle where S(z) is the stroke of the paddle. The propagating wave is of constant form. The water accompanies the wave paddle. 1951).2 the equations express: 0. Figs 4 and 5. Fig. t) = x(z. Free surface is at constant pressure (dynamic boundary condition). the velocity potential produced by paddle movement is formulated in Fig. The time series of paddle movement is e(z.5 WAVE GENERATION IN LABORATORY 5. Laplace equation. The bottom is impermeable. 2. The horizontal velocity of water particle is the same as the paddle. 3. Formulation of boundary value problem. Basic equation for potential ﬂow. 1.2.2 Bi´sel transfer functions e Bi´sel transfer functions express the relation between wave amplitude and wave paddle e displacement (Bi´sel et al. cf. In Fig.2. 4. 66 . All water particles at the free surface remain at the free surface (kinematic boundary condition). e Formulation Under the assumption of irrotational and incompressible ﬂuid.
5
WAVE GENERATION IN LABORATORY
Nearﬁeld and farﬁeld solution By solving the boundary value problem the surface elevation in the generated wave ﬁeld is η(x, t) = c · sinh(kh) cos(ωt − kx) +
∞ n=1
cn sin(kn h) e−kn ·x sin(ωt)
(52)
where c, cn and kn are coeﬃcients depending on the paddle type, paddle cycling frequency and water depth. The ﬁrst term in eq (52) expresses the surface elevation at inﬁnity, by Bi´sel called the e farﬁeld solution, while the second term is the nearﬁeld solution. In general only the farﬁeld solution is interesting because the amplitude of a linear wave should not change with location. Fortunately, the “disturbance” from the nearﬁeld solution will in a distance of 12 water depth from the wave paddle be less than 1% of the farﬁeld solution, cf. Fig.3.
Fig.3. Wave amplitude and phase of the generated wave ﬁeld relative to the farﬁeld solution. Water depth = 0.7 m and wave period = 0.7 sec Fig.3 shows that the farﬁeld surface elevation is phaseshifted π relative to the displace2 ment of the wave paddle. However, because the initial phase of the surface elevation will not change wave properties, the paddle movement is often written as in phase with the surface elevation. The Bi´sel transfer function, i.e. the amplitude relation between wave and paddle, is e obtained by the farﬁeld solution c · sinh(kh) = H 2 (53)
The Bi´sel transfer functions for the two most popular wave paddles are given in the e followings.
67
5
WAVE GENERATION IN LABORATORY
Bi´sel Transfer Function for pistontype paddle e S(z) = S0 H 2 sinh2 (kh) = S0 sinh(kh) cosh(kh) + kh (54)
Fig.4. Bi´sel Transfer Function for pistontype paddle. e
Bi´sel Transfer Function for hingedtype paddle e S(z) = S0 · (h + z), h S0 = S(z = 0)
2 sinh(kh) (1 − cosh(kh) + kh sinh(kh)) H = S0 kh (sinh(kh) cosh(kh) + kh)
(55)
Fig.5. Bi´sel Transfer Function for hingedtype paddle. e
68
5
WAVE GENERATION IN LABORATORY
5.3
Examples
Calibration of wave paddle Before generating waves the wave paddle should be calibrated in order to obtain the calibration coeﬃcient of the paddle. The calibration of wave paddle is performed by sending a signal which increases gradually from 0 to 1 voltage in one minute, and then measuring the wave paddle displacement, cf. Fig.6.
Fig.6. Signal to be sent to wave paddle for calibration. Modiﬁcation of signal In order to avoid a sudden movement of wave paddle, the signal should be modiﬁed by a data taper window. Fig. 7 illustrates the principle of the linear data taper window.
Fig.7. Linear data taper window.
69
5
WAVE GENERATION IN LABORATORY
General procedure The most important aspect of wave generation is the preparation of the input signal corresponding to the variance spectrum of design wave. Fig.8. illustrates one of simple methods.
Fig.8. Preparation of input signal. Of course the recording of the generated wave is necessary so that it can be checked whether the variance spectrum of the generated wave is close to that of design wave, according to Murphy’s law7 .
7
Murphy’s law: Anything, which might go wrong, will go wrong.
70
2 s−1 .4 1) The surface elevation of the linear wave is η(t) = a cos(ωt + δ) = H cos(ωt + δ) 2 where the angular frequency ω = 2π/T = 4. Sample the data from the modiﬁed time series of the voltage at fsample = 50 Hz and send the signal to the wave paddle. 2) Convert the time series of surface elevation into the time series of piston movement by the help of Biesel transfer function.5 m s m Water depth h = 0.5 WAVE GENERATION IN LABORATORY Example : Linear wave generation Generate a linear wave by pistontype wave paddle.80 = S0 sinh(kh) cosh(kh) + kh The time series of the piston movement is e(t) = 3) S0 H cos(ωt + δ) = cos(ωt + δ) 2 2B Convert the time series of piston movement into the time series of voltage to be feeded to the pistontype wave paddle. The Bi´sel transfer function for the pistontype wave paddle e B = 2 sinh2 (kh) H = 0. 71 . volt(t) = C e(t) where C is the calibration coeﬃcient of the wave paddle. 4) 5) Modify the time series of the voltage by the linear data taper window in order to avoid sudden movement of the wave paddle. The initial phase δ is given a random number between 0 and 2π. Wave height Wave period H = 0.1 T = 1.
Preparation of input signal for linear wave.5 WAVE GENERATION IN LABORATORY Fig. 72 .9.
0624 0. fstop ).4 m and peak enhancement coeﬃcient γ = 3.0336 γ − (f − fp )2 2 2 σ 2 fp 0. The surface elevation of the irregular wave is 7 7 η(t) = i=1 ηi (t) = i=1 ai cos(ωi t + δi ) 73 . JONSWAP spectrum 2 4 S(f ) = α Hs fp f −5 γ β exp − 5 4 fp f 4 α ≈ 0.00193 (m2 s). Note Sη (fp ) = 0.2 Hz N That is to say.0 Hz The frequency band width ∆f = fstop − fstart = 0.07 σ ≈ 0.3. Tp = 1 s.230 + 0. the irregular wave is composed of 7 linear waves.185 ( 1.5 WAVE GENERATION IN LABORATORY Example : Irregular wave generation Generate irregular wave by pistontype wave paddle according to JONSWAP spectrum with Hs = 0.01 Sη (fp ) Sη (fstop ) ≤ 0.6 Hz fstop = 2. water depth h = 0.01 Sη (fp ) For the sake of simplicity in this example N =7 fstart = 0.9 + γ ) β = exp − σ ≈ 0.1 m. To ensure accuracy usually N ≥ 50 Sη (fstart ) ≤ 0. 2) Divide the spectrum evenly into N parts in the interval (fstart .09 f ≤ fp f ≥ fp γ : peak enhancement coeﬃcient 1) Draw the JONSWAP spectrum with the speciﬁed parameters.
0183 6.6 0.9 8.0178 0.9 0.6 0.0119 0.5 1.7 5.5 WAVE GENERATION IN LABORATORY 3) Determine the angular frequency ωi .0356 0.00079 0.4 9.36 3 1.5 10.0 0.0406 0.0052 0.2 1.9 14. · · · .0141 4.69 4 7 i=1 Hi cos(ωi t + δi ) 2Bi 5 1. frequency angular variance fi = fstart + i ∆f − ωi = 2π Ti ∆f 2 = 2πfi 1 2 a 2 i Sη (fi ) ∆f = i = 1.9 1. 7 e(t) = i=1 i fi Ti Li ωi ki Sη (fi ) ai Hi δi Bi (Hz) (s) (m) (/s) (/m) (m2 s) (m) (m) 1 0.0092 0.7 11.90 0.3 0.1 1. 4) Convert the time series of surface elevation into the time series of piston movement by the help of Biesel transfer function. The initial phase δi is assigned a random number between 0 and 2π.1 0.0070 0.1 2. · · · .00021 0.5 0.9 1.98 6 1. 74 .6 S0. 2.7 3.00007 0.1 1.00012 0.0239 2.2 6.00103 0. The Bi´sel transfer function for the pistontype wave paddle e Hi 2 sinh2 (ki h) = Bi = S0. 6) Modify the time series of the voltage by the linear data taper window in order to avoid sudden movement of the wave paddle.4 2.0203 0.i sinh(ki h) cosh(ki h) + ki h The time series of the piston movement is 7 i = 1.6 0.9 0.7 0. 7 amplitude ai = 2 Sη (fi ) ∆f S(fi ) is calculated from the JONSWAP spectrum. amplitude ai and initial phase δi of each linear waves.00 1.5 0.7 1. volt(t) = C e(t) where C is the calibration coeﬃcient of the wave paddle.00 5) Convert the time series of piston movement into the time series of voltage to be feeded to the pistontype wave paddle.7 0.8 0.00036 0. 2.1 0.7 1.4 2.5 4.9 5.5 0.2 6.3 2.0055 0.00007 0.3 1.00 7 1.0103 1.0110 4.i cos(ωi t + δi ) = 2 2 0.7 9.4 11.
5 WAVE GENERATION IN LABORATORY 7) Sample the data from the modiﬁed time series of the voltage at fsample = 50 Hz and send the signal to the wave paddle. 75 .10. Preparation of input signal for irregular wave. Fig.
. Singapore.20 m. 1993.5 m. Random seas and design of marine structures. Hgedal. Department of Civil Engineering.. Ltd. Aalborg University. and Christensen. reference is made to Frigaard et al. and Dalrymple. 1991. Denmark. . With respect to wave generation techniques we are proud of the fact that the Hydraulic & Coastal Engineering Laboratory of Aalborg University is one of the leading institutes in the world. Denmark. All these aspects have been implemented in a userfriendly software package named PROFWACO (Frigaard et al. Wave generation theory. 4) Make a computer program for examples given in lecture..5 WAVE GENERATION IN LABORATORY Remarks For more complicated wave generation method. Department of Civil Engineering. Users Guide. . Japan. List the factors which limit the maximum wave height obtainable in a wave ﬂume. including active wave absorption and 3D wave generation.5 s. P. 2) What is the minimum distance between the wave gauge and wave paddle ? 3) Explain the importance that the wave paddle can produce suﬃciently large wave height. . 1993. 1985 5. Y. Frigaard. World Scientiﬁc Publishing Co. 1993. . and Christensen. M. Hydraulic & Coastal Engineering Laboratory. 1985. 5. 1993). 76 . Second printing with correction.4 m. PROFWACO. M. wave ﬂume width B = 1. M.4 References Dean. R. Frigaard. (1993). Water wave mechanics for Engineers and scientists.G. 1991. Hgedal. Aalborg University.5 Exercise 1) Estimate the minimum power of motor needed for generating a linear wave with Wave height H = 0.A. water depth h = 0. Goda. P. Pte. M. Hydraulic & Coastal Engineering Laboratory. R. wave period T = 1. University of Tokyo Press. 1996.