9789812774828_0009
TRANSCRIPT
-
8/18/2019 9789812774828_0009
1/148
Chapter 9
Real Ocean W aves
9.1. Introduction
Kinsman (1965, p. 386) is careful to point out that a valid specification of real
ocean waves m ust integrate the following three concepts: 1) Fourier and spec
tral analyses of random processes, 2) probability theories applied to stochastic
processes, and 3) hydrody nam ics. Techniques from the first two concepts that
do not depend on the physics of the hydrodynamic processes are available for
analyzing real ocean waves. However, only those techniques from concepts
that may be related rigorously to the physics of the hydrodynamics of real
ocean waves are reviewed.
The theoretical techniques reviewed are applicable to
stationary ergodic
processes and are limited strictly to short term statics. Isaacson and
M acKenzie (1981) give an excellent review of long term statistical and prob
abilistic techniques applied to real ocean waves. The significance of the
stationary ergodic
hypothesis is that the
ensemble
average
E[x(t\)]
at the
same time t\ shown in Fig. 9.1 of an infinite number of finite length time
series x\(t\),X2(t\),x-s(t\), ,*oo(^l) is equivalent to the temporal aver
age over all times shown in Fig. 9.2 of an infinitely long single time series
X\(t\),X\(t2),Xl{tl),....,X\(tooY,i-S;
E[x(t
n
)]=
lim —
I
R
xi(t)dt.
(9.1)
T
R
^OO 1R JO
719
W a v e s a n d W a v e F o r c e s o n
C o a s t a l a n d O c e a n S t r u c t u r e s D o w n l o a d e d f r o m w w w . w o r l d s c i e n t i f i c . c o m
b y M C G
I L L U N I V E R S I T Y o n 0 4 / 0 5 / 1
6 .
F o r p e r s o n a l u s e o n l y .
-
8/18/2019 9789812774828_0009
2/148
720 Waves and Wave Forces on Coastal and O cean Structures
Xjrft)
x
3
(0
-A/l/yAA^JjV/
x
2
(t)
xtf)
-X -
1
' /
*
/ /
Fig. 9.1. Two ensemble averages of finite length records
*,- (t
) at times
t\
and
ti
(Bendat and
Piersol, 1986).
Xrft)
Fig. 9.2. Temporal average of a single time series
x\ it)
of infinite length.
9.2. Fourier Analyses
Definitions of Fourier coefficients
Fourier coefficients are defined separately for
deterministic
and for
non-
deterministic
(or, equivalently,
random)
analyses.
W a v e s a n d W a v e F o r c e s o n C o a s t a l a n d O c e a n S t r u c t u r e s
D o w n l o a d e d f r o m w w w . w o r l d s c i e n t i f i c . c o m
b y M C G
I L L U N I V E R S I T Y o n 0 4 / 0 5 / 1
6 .
F o r p e r s o n a l u s e o n l y .
-
8/18/2019 9789812774828_0009
3/148
Real Ocean Waves
721
Deterministic:
Fourier coefficients are those coefficients that provide the
best least squares fit to the data.
Non-Deterministic: Fourier coefficients are those coefficients that explain
the contribution by each frequency to the total variance of
a
random process.
W hen both time and frequency are continuous, the Fourier transform pairs
are given by the following integrals:
i
r
+0
°
r)(t)
=
—f==.
/
F(a))exp±(i(ot)d(D,
(9.2a)
J—
c
'In j — oo
i
r
+
°°
F(a>) = - = t](t)exp^(icot)dt, (9.2b)
V2TT
J—OO
where u>
= 2nf =
radian frequency; where
F(co) = Fourier transform
of the time series
r){t);
and where the plus + sign in Eq. (9.2a) must be
paired with the negative — sign in Eq. (9.2b). Both choices for the ±
signs in the arguments of the exponential functions exp(«) in Eqs. (9.2)
may be found in the literature; and the placement of the normalizing con
stant
2n
is also arbitrary (Ligh thill, 1964). If the frequency / is given
in Hertz, the normalizing constant
2n
does not appear and Eqs. (9.2)
reduce to
/
+ 0 0
F(f)ex
V
±(i2nft)df,
(9.3a)
-00
/ + 0 O
r,(t)exp
T
(i2nft)dt. (9.3b)
-00
Data records that are continuous in time are termed
time series;
and data
records that are digitized to discrete values of time are termed
time sequences.
Modern Fourier analyses employ discrete/inite Fourier transform (FFT)
algorithms that are designed
to
take advantage of high speed digital compu ters.
Discretization of Eqs. (9.3) requires a finite record length TR and discrete
W a v e s a n d W a v e F o r c e s o n
C o a s t a l a n d O c e a n S t r u c t u r e s D o w n l o a d e d f r o m w w w . w o r l d s
c i e n t i f i c . c o m
b y M C G
I L L U N I V E R S I T Y o n 0 4 / 0 5 / 1
6 .
F o r p e r s o n a l u s e o n l y .
-
8/18/2019 9789812774828_0009
4/148
722 Waves a nd Wave Forces on Coastal and O cean Structures
values of both time and frequency; i.e.;
t ->• t„ = nAt, f ->• f
m
= mAf, co -» co
m
=
2nf
m
= InmAf,
(9.4 a-c)
where At(=T[t/N) and Af(=l/NAt) are constant fixed temp oral and
frequency intervals, respectively, of N total discrete values of time and fre
quency. A discrete Fourier transform pair may be approximated from
Eqs. (9.2) by
N-l
rj(n) = A / ^ F ( r a )e x p ± {ilizrimAf
At),
n = 0 , 1 , 2 , . . . ,N - 1,
m=0
(9.5a)
N-l
F(m) = At y ^ rj(n)exp ^(Unnm AtAf), m = 0,1,2,.. .,N — 1,
n=0
(9.5b)
where the total number of discrete time and frequency values are equal to ./V;
and where
the
FF T coefficients F(m) are complex-valued quantities.
The
mean
value of the discrete time sequence r\ (n) is given by the real-valued FFT coeffi
cient ^ ( 0 ) ; and the FFT coefficient
F(N/2)
is also real-valued at the Nyquist
or folding frequency
fy
=
(N/2)Af
in Eq. (9.5a). The discrete positive-
definite frequencies f
m
> 0 in Eq s. (9.2 and 9.3) are represented by the
indices 1
-
8/18/2019 9789812774828_0009
5/148
Real Ocean Waves
723
Eqs.
(9.5) did not permit zero indices and Eqs. (9.5) were given by only
positive-definite indices m, n > 0 according to
N
rj(n) = A / ^ F ( m ) e x p ± ( / 2 ; r ( M - l)(m - 1 ) A / A 0 , n = l,2,...,N
(9.6a)
N
F(m) = At^t](n)exp =F(/2jr(n - \)(m - \)AtAf), m =
1,2,...
,N.
(9.6b)
The mean value of r)(n) in Eqs. (9.6) is given now by the real-valued FFT
coefficient
F(l);
and the real-valued coefficient
F(N/2
+ 1) is now at the
Nyquist or folding frequency
/N = (N/2
+ 1 ) A / . The discrete positive-
definite frequencies
f
m
> 0 in Eq. (9.5a) now have the indices 2
-
8/18/2019 9789812774828_0009
6/148
724
Waves and Wa ve Forces on Coastal and Ocean Structures
that may be reduced to the following compact result:
-;\-z
N
SN —
\-z
N
, z # l
, z = l
If z is a complex-valued variable given by
z
= exp ±
{i2n(m
—
m)/N),
where m and m are integer constants; then Eq . (9.7a) becom es
N-l
S
N
= ^ exp[±i2jr(#n - m)/Nf
n=0
and Eq. (9.7b) is
SN =
1
—
exp ± \i2n{m
—
m)\
1
—
exp ± [ilTz{m — m)/N]
N, m
—
m =
0,
0 , 0 < m —
m < N
— 1.
(9.7b)
(9.7c)
(9.8)
(9.9a)
(9.9b)
(9.9c)
The normalizing constant CN may be determined by substituting Eq. (9.5a)
into Eq. (9.5b) with a careful change of dummy indices from
m
to
m
and
obtaining
N-l
F(m) =
AtJ2
n=0
N-l
A / y ^ F(m) exp ± (ilnnmAf At)
x exp
==
{ilnnm Af At)
N-l (N-l
= A ? A / ] T F (m) | ] T exp ± (ilnim - m)nAfAt)
m=0
N-l
l n = 0
N-l
= AtAf J^
F
^)
J ]
{ e x p ±
(
i2n
(™
~
m)AfAt)f.
(9.9d)
m = 0 n = 0
If
AtAf = N
1
; then the term in curly brackets {•}" = N (5^
m
by
Eqs.
(9.9b, c
and 9.5b) reduces to
F(m) = ( A f A / ) F ( m ) { A ^
m
)
(9.9e)
W a v e s a n d W a v e F o r c e s o n
C o a s t a l a n d O c e a n S t r u c t u r e s D o w n l o a d e d f r o m w w w . w o r l d s c i e n t i f i c . c o m
b y M C G
I L L U N I V E R S I T Y o n 0 4 / 0 5 / 1
6 .
F o r p e r s o n a l u s e o n l y .
-
8/18/2019 9789812774828_0009
7/148
Real Ocean Waves
725
that is possible only if
At • Af • N = I,
so that
Af A / = i
=
CN
(9.10)
and the normalizing constant CN is inversely proportional to N.
For example,
the FFT
algorithm
in the
symbolic software
MATHEMATICA™ employs the finite Fourier transform pair Eqs. (9.6),
pre-multiplies the complex-valued FFT coefficients F(m) by a normalizing
constant
CN = s/N
and selects the minus (—) sign for the exponential func
tion in Eq. (9.6a) and the positive sign (+) for the exponential function in
Eqs. (9.6b). With this convention, Eqs. (9.6) become
N
r](n)= ] [ ^ f l ( m ) e x p - ( i 2 j r ( n -
l)(m -
\)/N);
n = l,2,...,N,
(9.11a)
N
B(m) = */NF(m) =
^ rj(n) exp+(/27r(n - l)(ra - l)/N);
m = l,2,...,N. (9.11b)
Because Eq. (9.11a) applies the minus (—) sign for the argument in the
exponential function exp(«), the com plex-valued FFT coefficients B(m) are
expressed with a negative (—) sign for the phase by
B(m) = \B(m)\
exp -
ia(m),
(9.11c)
where a(m) = phase angle at the discrete frequency mAf. Figure 9.3 illus
trates the frequency domain representation of the amplitudes (m oduli) \B(m)\
in Eq. (9.11c).
To illustrate a simple numerical test that may be applied to determine
where a particular FFT algorithm places the norm alizing constant
CN,
synthe
size the following time sequence that consists of
a
non-zero m ean F(l); first
F(2)
and fourth
F(5)
harmonic components with positive and negative phase
W a v e s a n d W a v e F o r c e s o n
C o a s t a l a n d O c e a n S t r u c t u r e s D o w n l o a d e d f r o m w w w . w o r l d s
c i e n t i f i c . c o m
b y M C G
I L L U N I V E R S I T Y o n 0 4 / 0 5 / 1
6 .
F o r p e r s o n a l u s e o n l y .
-
8/18/2019 9789812774828_0009
8/148
726
Waves and Wave Forces on Coastal and Ocean Structures
\B(m)\
, I-J I i I U,J_I I .
t i >
i s
1
e • \ i | J « J . « " •
, M ,
-
m H H
" ' ' "
fff
Fig. 9.3. Frequency d oma in representation of the amplitudes of the FFT coefficients
\B{m)\.
1 1
0.5 ,
0
"sT -l
-2 -
1 1 1 .
i 1 j i
i
i i T •
"-• r r r r '
J- J - J- 4_
1 1 •
1 1 1 - 1
. • L L L » L
I • I • I I
1 -
» - H -i i
I
I I I
i * i i
1 4 7 10 13 1
n
Fig. 9.4. Discrete nondim ensional time sequence for
N =
16.
angles
a(2)
and a (5 ), respectively; and a total number of discrete sequences
N
= 16 = 2
4
(NOT E: «/w«ys
select N = 2
2M
where 2M must always be an
even integer so that if the FFT algorithm applies the square root J conven
tion for the normalizing constant
CN
then the normalizing constant CM will
always be a rational num ber). Consider the discrete time sequence
3 (Inn
TT
1 (litAn n
+
2
C
°
S
U ^ ~ 4
(9.12)
that is illustrated in Fig. 9.4. A program for synthesizing the normalized dis
crete time sequence in Eq. (9.12) by the FFT algorithm in
MATHEMATICA™
is listed below. The amplitudes
\B(m)\
of the complex-valued FFT coeffi
cients are illustrated in Fig. 9.5; and the phases aim) of the complex-valued
FFT coefficients are illustrated in Fig. 9.6. Note the change in sign of the
phases a(m) in Fig. 9.6 as a result of the sign convention defined for the
W a v e s a n d W a v e F o r c e s o n C o a s t a l a n d O c e a n S t r u c t u r e s
D o w n l o a d e d f r o m w w w . w o r l d s c i e n t i f i c . c o m
b y M C
G I L L U N I V E R S I T Y o n 0 4 / 0 5 / 1 6 .
F o r p e r s o n a l u s e o n l y .
-
8/18/2019 9789812774828_0009
9/148
Real Ocean Waves
727
5
4
~^~ 3
1
0
7 10
m
i
i
-r-
i
13
16
Fig. 9.5. Am plitudes of the FFT coefficients for the discrete time sequ ence
r)(ri)
in Fig. 9.4.
1 4
Fig. 9.6. FFT phase angles for the discrete time sequen ce shown in Fig. 9.4.
FFT coefficients in Eq. (9.11c). Table 9.1 lists both the expected amplitudes
\F(m) |
of the Fourier coefficients without regard for the normalizing constant
C
N
and the amplitudes \B(m)\ from the program MATHEMATICA™. By
dividing the values of the amplitudes
\B(m)\
from
MATHEMATICA™
by
the expected amplitudes \F{m)\ of
the
Fourier coefficients in Table 9.1 , it is
easy to obtain the normalizing constant CV = V 1 6 = 4. Note that in Table 9.1
that the negative-definite frequencies in the FFT algorithm are stored in the
discrete frequencies identified by the indices N /2 + 2
-
8/18/2019 9789812774828_0009
10/148
728 Waves and Wave Forces on Coastal and Ocean Structures
Table
9.1.
Determination of normalizing constant Q y and phase angles
a (m)
from MATHEMATICA™ FFT algorithm for N = 16 = 4
2
{terms in (•)
are the amplitudes and phases of the complex-conjugate FFT values}.
a(m)
m
1
2(16 )
3(15)
4 (14 )
5(13)
6(12)
7 (11)
8(10)
9
\F(m)\
1.0
0.75 (0.75)
0(0)
0(0)
0.25 (0.25)
0(0)
0(0)
0(0)
0
\B(m)\ = C
N
\F(m)\
4.0
3.0(3.0)
0(0)
0 (0 )
1.0(1.0)
0(0 )
0 (0 )
0 (0 )
0
C
N
= JN
4.0
4.0 (4.0)
- ( - )
- ( - )
4.0 (4.0)
- ( - )
- ( - )
- ( - )
-
(Radians)
n
-JT/4(+TC/4)
0(0)
0(0)
+ j r / 4 ( - 7 r / 4 )
0(0)
0(0)
0(0)
0(0)
of the positive-definite frequencies with indices 2
-
8/18/2019 9789812774828_0009
11/148
-
8/18/2019 9789812774828_0009
12/148
730 Waves
a nd
Wave
Forces on Coastal and O cean Structures
SameQ[finv==f]
(* OUTPUT FILE TO BE IMPORTED TO SPREADSHEET "QP RO " *)
SetDirectory["C:\lfn\"];
z=Table[ { t[[i]],f[[i]],absbn[[i]],phase[[i]]}, {i, 1,16}]
ColumnForm[z];»debugfft.prn
9.3.
Ocean Wave Spectra
Spectral representations of stationary ergodic random seas are required for
engineering analyses of random ocean waves. Several two-parameter theoret
ical spectral models have been applied to engineering applications along with
five-
and six-parameter spectral models. The parameters applied
in the
original
derivations of these spectral models are varied. Among the most commonly
applied parameters are wind speed U
w
, fetch length F , significant wave height
H
s
and period T
s
. For those two-parameter spectral densities, it is convenient
for engineering design
to
replace the original parameters with
the
zeroeth spec
tral
m oment mo (= to the variance
a
2
of the time series or the area under the
spectrum) and the frequency of the spectral peak /o or COQ.
The mean ( = p,\), the mean square value ( = \xi), the variance (=
a%
=
mo) and the standard deviation ( =
9
= lim — /
(x(t) - n\Ydt = 112- p\,
(9.14b)
T
R
^oo
TR
JO
W a v e s a n d W a v e F o r c e s o n
C o a s t a l a n d O c e a n S t r u c t u r e s D o w n l o a d e d f r o m w w w . w o r l d s
c i e n t i f i c . c o m
b y M C G
I L L U N I V E R S I T Y o n 0 4 / 0 5 / 1
6 .
F o r p e r s o n a l u s e o n l y .
-
8/18/2019 9789812774828_0009
13/148
Real O cean Waves
731
where £[•] = an ensemble averaging operator defined in
Eq.
(9.1);
p(»)
= the
probability density function (pdf) for the random variable (•) ; and 7> = a tem
poral record that is infinitely long in
Eq.
(9.1). Data are frequently normalized
by subtracting the mean \x\ and then dividing by the standard deviation a
x
of
x(t)
in order to obtain the following zero-mean , unit variance data record:
=
* ( O - M I W
( 9 1 4 c )
Ox
The
Wiener-Khinchine Fourier transform pair
is
similar
to the
Fourier series
transform pair from Eqs. (9.2) but relates the covariance function C
xy
(x) to
the two-sided spectral density function G
xy
(co) (Bendat and Piersol, 1986)
according to
1 r+°°
C
xy
(r) = —— / G
xy
(oo)exp ±(ioor)dco, (9.15a)
V2TT
J—
oo
i r+°°
G
xy
(co)
= —= I C
xy
(r)exTp^(ia)T)dr,
(9.15b)
V2TT J-OO
where G
xy
(co) = a complex-valued, two -sided cross-spectral density function
that may be expressed as
G
xy
(co) = C
xy
(co) ± iQ
xy
(a>), (9.15c)
where C
xy
(oo) = coincident spectral density function and Q
xy
(co) = quadra
ture spectral density function. Alternatively, the complex-valued, two-sided
cross-spectral density function
in Eq.
(9.15c) may be expressed
as
an amplitude
and a phase by
G
xy
(co) - \G
xy
(oo)\exp ±ia
xy
(co), (9.15d)
where the amplitude
\G
xy
(oo)\
and phase
a
xy
(co)
are computed from
G
xy
(oo)\ = JC^Jco) + Q lJco), a
xy
(co) =
arctan
Q xy (CO)
C
xy
(00) _
(9.15e,f)
W a v e s a n d W a v e F o r c e s o n
C o a s t a l a n d O c e a n S t r u c t u r e s D o w n l o a d e d f r o m w w w . w o r l d s c i e n t i f i c . c o m
b y M C G
I L L U N I V E R S I T Y o n 0 4 / 0 5 / 1
6 .
F o r p e r s o n a l u s e o n l y .
-
8/18/2019 9789812774828_0009
14/148
732
Waves
and
Wave
Forces on Coastal and Ocean Structures
A real-valued coherence function may be computed by
2
\G
xy
{co)\
2
YI
y
(w ) = \ )
= area l-valu ed, two-sided spectral density function for the time
series £(/)• A real-valued one-sided spectral density function
S
m
{co)
for the
time series r) (t) may be computed from the two-sided spectral density function
G
xy
{(o) by
S
m
{a>)=2G
m
(a>)U{a))
(9.15h)
where U(a>) = the Heaviside step function in Eq. (2.1) in Sec. 2.2.2. One
sided spectral
density
values
S
nri
(a>
m
) m ay be computed for the discrete radian
frequency
co
m
from two-sided, com plex-valued discrete FFT
amplitudes
| B
m
\
in Eq. (9.11c) by
2\B
m
\
2
\B
m
\
2
S
m
(a )
m
) = ' " ' = —^Ndt, (9.15i)
IjcdfC^i TTCN
where Eq. (9.10) has been substituted for ^f and where
CN
= th e FFT nor
malizing constant defined in Eq. (9.10) for the FFT coefficients computed by
MATHEMATICA™. Similarly, values may be computed from Eq. (9.1 5i)for
the random wave simulations in Sec. 9.6 by
2\B
m
\
2
S^icom) = ^ - ^ (9.15J)
where CN = the FFT normalizing constant defined in Eq. (9.10) for the FFT
coefficients computed by
MATHEMATICA™.
One-sided spectral density functions
S
m
(•) m ay be expressed as functions
of the independen t variables (• ) of radian frequencies a>(=27tf), or of cycles-
per-second (cps) frequencies / , or of wave periods
T,
or of scalar wave
numbers k{=2iz/X) or of vector wave num bers k. In order to determine the
relationship between spectral densities expressed with different independent
variables, equate the differential area under each spectral density in a small
incremental interval of the independent variable according to
-S^mdT =
S
nr}
{(o)da>
= Sr,„(f)df = S
nn
{k)dk = S
nri
{k)dk. (9.16a)
W a v e s a n d W a v e F o r c e s o n C o a s t a l a n d O c e a n S t r u c t u r e s
D o w n l o a d e d f r o m w w w . w o r l d s c i e n t i f i c . c o m
b y M C G
I L L U N I V E R S I T Y o n 0 4 / 0 5 / 1
6 .
F o r p e r s o n a l u s e o n l y .
-
8/18/2019 9789812774828_0009
15/148
Real Ocean Waves
733
For co
= 2nf,
the transformation Jacobians required for Eq. (9.16a) may be
determined from
dco
= litdf = -^rdT = -—d T =
-27tf
2
dT; (9.16b)
T
z
2n
so that, accordingly, for the independent variables
T, f
and co
7
T
CO
S
m
(T) =
f
2
S
m
(f)
= —
S
m
(a>),
(9.16c)
lie
S^f) = 2nSr,
n
(eo).
(9.16d)
Transformations between frequency / and wave num ber
k
spectra require the
linear dispersion equations (4.15) in Chapter 4.3 given by
co
2
=
( 2 T T / )
2
=
gktanhkh;
(9.16e)
and the corresponding Jacobian transformation from Eq. (9.16e) is given
by Eq. (4.60c) in Chapter 4.5
where CG = t h e wave group velocity. The Jacobians from Eq. (9.16f) for the
deep-
and shallow-water approximations, respectively, for Eq. (9.16e) are
dco co dco co /—- ,„ , ̂ , s
TT =^r> ^ =
T
= Vgk, (9.16g,h)
UK deep-water ZICQ UK shallow-water K
where the deep-water wave num ber ko
= co
2
/g.
Correlation-covariance definitions
There do not appear to be consistent definitions with respect to the time
series C
xy
{x) (Kinsm an, 1 965). The function C
xy
(r) is defined as a cross-
correlation
function if the times series
x(t)
and
y{t)
are scaled by their
means
[i\
and standard deviations
at
in accordance with Eq. (9.14c) such
that each time series has zero-mean /xi = 0 and unit variance
of
= 1.
If the time series
x{t)
and
y(t)
of record length
TR
are not scaled by
Eq. (9.1 4c), then the function
C
xy
( r ) is defined as a
cross-covariance
function.
W a v e s a n d W a v e F o r c e s o n
C o a s t a l a n d O c e a n S t r u c t u r e s D o w n l o a d e d f r o m w w w . w o r l d s
c i e n t i f i c . c o m
b y M C G
I L L U N I V E R S I T Y o n 0 4 / 0 5 / 1
6 .
F o r p e r s o n a l u s e o n l y .
-
8/18/2019 9789812774828_0009
16/148
734
Waves and Wave Forces on Coastal and O cean Structures
The two-sided spectral density function
G
xy
(a>)
is always defined as a cross-
spectral density function regardless of whether or not the time series x(t)
and
y(t)
are scaled in accordance with Eq. (9.14c). If the two time series
x(t)
and
y(t)
are identical, then Eq. (9.15a) is defined as an
auto-covariance
(or
-correlation)
and Eq. (9.15b) the
auto-spectral
density
function.
Historically, the covariance (or correlation) function
was
computed in order
to efface the randomness from a time series of a random process in order to
expose the invariant statistical anatomy of the process (Wiener 1964, p. 6).
The cross-covariance and cross-correlation functions are computed from time
series x(t) and y{t) by
1 fT
R
/2
C
xy
(x) = lim — / x(t)y(t + r)dt, | r | < oo, (9.17a)
TR^OO 1R J-T
R
/2
T
R
-+oo 1
R
J-T
R
/2
T
R
/2
C
xx
(r)= lim — / x(t)x(t + r)dt, (9.18a)
r
f i
^ o o
1
R
J-T
R
/2
c
? x & ( T ) =
T
l m i — / —
2
dt, ( 9 . 1 8 b )
W a v e s a n d W a v e F o r c e s o n
C o a s t a l a n d O c e a n S t r u c t u r e s D o w n l o a d e d f r o m w w w . w o r l d s
c i e n t i f i c . c o m
b y M C G
I L L U N I V E R S I T Y o n 0 4 / 0 5 / 1
6 .
F o r p e r s o n a l u s e o n l y .
-
8/18/2019 9789812774828_0009
17/148
Real Ocean Waves
735
where
Eqs.
(9.18a, b) are always real-valued functions, symm etric about
x
= 0
and equal to the variance
0%
of x(t) for r = 0; i.e.,
Cxx(0) 1
=
}a
x
2
] (9 8c )
lCfc
&
(0)J 1 1 J (9.18d)
Unless x(t) is a strictly periodic time series, Eq. (9.18b) is also proportional
to the mean ii\ (x) for r
—
> ±00; i.e.,
j
y/C
xx
{±oo)
I _ j ^ j ^ J (9.18e)
I V ^ ( ± ° ° ) J ~ 1 ° J (9.18f)
The first analyses of random data computed the cross- (or auto-) covariance
(or correlation) function from the time series by Eqs. (9.17a, b or 9.18a, b) and
then applied the Wiener-Khinchine Fourier transform Eq. (9.15b) to obtain
the two-sided spectral density function. Modern analyses of digitized dis
crete time sequences (t
n
= ndt, Eq. (9.4a)) employ the discrete finite Fourier
transform (FFT)
(f
m
— mdf,
Eq. (9.4b)) to compute the complex-valued
discrete FFT coefficients
F
x
(m )
and
F
y
{m)
of the discrete time sequences
x(n)
and
y(n);
and then apply these discrete coefficients to com pute either
the discrete cross-covariance
C
xy
{n)
(or cross-correlation
C
xy
{n))
function or
the two-sided
G
xy
(m)
(or one-sided S
xy
(m)) cross-spectral density functions.
A comparison of these two methods for obtaining discrete spectral estimates
from an FFT algorithm is illustrated
in
Fig. 9.7 where the notation F F T implies
the forward transform from Eq. (9.11b); and the notation F F T
- 1
implies the
inverse transform from Eq. (9.11a).
To illustrate how either a two-sided G
xy
(m) or a one-sided
S
xy
(m )
cross-
spectrum may be com puted by either of the two paths in Fig. 9.7 by a discrete
FFT form of the Wiener-Khinchine Fourier transform pair in Eqs. (9.15a, b),
a random time series
rj{t)
of six cosine waves is synthesized from the ampli
tudes and phase angles listed in Table 9.2 and is illustrated in Fig. 9.8a. The
amplitudes
A
m
,
phase angles
a
m
and discrete frequencies
co
m
= lizmdf
for
each of the six cosine wave are summarized in Table 9.2. Continuous time
and frequency are discretized by Eqs. (9.4a, b) for application of an FFT algo
rithm where dt = 0.2 sec,
AT
= 64 and df = 1/Ndt = 0.078125 Hz. The
auto-correlation function C^(x) for the normalized random time series f (/ )
computed from x (t) = rj (t) by Eq. (9.14c) m ay be computed from the discrete
W a v e s a n d W a v e F o r c e s o n
C o a s t a l a n d O c e a n S t r u c t u r e s D o w n l o a d e d f r o m w w w . w o r l d s
c i e n t i f i c . c o m
b y M C G
I L L U N I V E R S I T Y o n 0 4 / 0 5 / 1
6 .
F o r p e r s o n a l u s e o n l y .
-
8/18/2019 9789812774828_0009
18/148
736
Waves and Wave Forces on C oastal and Ocean Structures
{x(n),y(n)}
C„
{n) " F
x
(m),F
y
(m)
G
xy
(m),S
xl
,(m)
Fig. 9.7. Com parison of method s for compu ting spectra by FFT and by covariance functions.
Table 9.2. Parameters for the six cosine waves for
the random time sequence in Figs. 9.8a and b (d t =
dx
= 0.2
sec, N =
64,
df = 1/Ndt =
0.078125 Hz, MI (?) = 0 ft and tr
2
= 64ft
2
).
frequency m
5
1
9
11
13
15
a>
m
(rads/ sec)
2.4544
3.4361
4.4179
5.3996
6.3814
7.3631
A
m
(f t )
2.0
4.0
8.0
6.0
2.0
2.0
a
m
(rads)
1.0192
2.0579
5.2495
5.3168
2.6336
0.6556
Fourier coefficients F
f
(m ) by the FFT coefficients B
m
for f (n) following the
horizontal path in the middle of Fig. 9.7 according to
C
ff
(T„) = |F
f
( m ) |
2
= - ^ - , (9.1 9a)
where Civ=the FFT normalizing constant defined in Eq. (9.10) for the
complex-valued FFT coefficients
B
m
computed by
MATHEMATICA™.
The two-sided discrete amplitude spectrum
\G^(m)\
in Eq. (9.15d) may be
computed from discrete FFT coefficients B
m
by
\Gu(m)\ = \F
(
(m)\
2
= ¥^-, (9.19b)
W a v e s a n d W a v e F o r c e s o n
C o a s t a l a n d O c e a n S t r u c t u r e s D o w n l o a d e d f r o m w w w . w o r l d s
c i e n t i f i c . c o m
b y M C G
I L L U N I V E R S I T Y o n 0 4 / 0 5 / 1
6 .
F o r p e r s o n a l u s e o n l y .
-
8/18/2019 9789812774828_0009
19/148
Real Ocean Waves
737
0 2 4 6 8 10 12 14
t
(sec)
Fig. 9.8a. Random time sequence r){t
n
) of six cosine waves in Table 9.2 where t
n
=
n(0.2) sec.
JJJ..
JJJ..
JJJ..
JJJ..
JJJ.
JJJ..
JJJ..
JJJ,
JJJ..
JJJ,.
JJJ..
JJJ.
_LLL
_LLL
.LLL
-LLL
.LLL
JJJ.
.LLL
.LLL
.LLL
.LLL
.LLL
J.1L
J-LL
J-LL
AIL
J-LL
JJ.L
J 1 L
J 1 L
J L L
AIL
ALLALL
_
L
L
L
L
L
L
_UJ.
JJJ.
JJJ.
JJJ.
JJJ.
JJJ.
JJJ.
JJJ.
JJJ
.LLL
.LLL
.LLL
.LLL
.LLL
.LLL
.LLL
.LLL
.LLL
.LLL
.LLL
.LLL
—J,
AAL
AIL
AIL
AAL
J-LL
J 1 L
J-LL
i l l
J L L
J 1 L
J 1 L
J 1 L
-J—
JJJ.
-UJ.
.UJ.
JJJ.
-UJ.
-UJ,
-UJ-
-UJ_
-UJ-
_UJ_
-UJ.
.LLL
.LLL
LLL
LLL
LLL
LLL
LLL
LLL
LLL
LLL
LLL
-UJ*
0 8 16 24 32 40 48 56 64
m
Fig. 9.8b. Two-sided discrete amplitude spectrum computed from FFT coefficients for six
cosine waves in Table 9.2 for f
m
= m/Ndt Hz.
where
CM
=the FFT normalizing constant defined in Eq. (9.10) for the FFT
coefficients computed by
MATHEMATICA™.
The two-sided discrete amplitude spectrum |G
w
( m ) | for the time series
rj(t)
computed from FFT coefficients
B
m
by Eq. (9.19b) is illustrated in
Fig. 9.8b. The symmetry about the Nyquist or folding frequency
m — N/2 +
1 = 33 in Fig. 9.8b of the two-sided discrete amplitude spectrum
\G
m
(m)\
is
a consequence of the negative-definite frequencies being represented by the
discrete frequency interval
N/2 + 2 < m < N
— 1.
The mean and standard deviation of
the
random time series
r](t)
synthesized
from the parameters in Table 9.2 are ii\{r)) = Oft and o
r]
=
8.0ft,
respec
tively. The random time series r](t) is normalized by this mean ii\(,rf) and
standard deviation
-
8/18/2019 9789812774828_0009
20/148
738
Waves and Wave Forces on Coastal and Ocean Structures
2 4 6 8 10 12 14
t
(sec)
Fig. 9.9a. Normalized random time sequence f (t
n
) of the six cosine waves in Table 9.2 where
0.25
_ 0.2
10.15
to. 0.1
0.05
0
H-4 +
'4 4 +
43±
-1-4 +
lOT
TO
E d :
4-14-
L I J - l L U -
i-l-ti-i—n-t-
TTT44-HI4T
- t -H- + 4-4-4+4-
xrt
+ 4-h
xnr
-4 +
3 1
44 +
trnnTrriTT
uremic 431
B J '
i t t i ± t J i t
1-4 +
331
-U +
111
0 8 16 24 32 40 48 56 64
m
Fig. 9.9b. Two-sided discrete amplitude spectrum computed from FFT coefficients for the
normalized time sequence
r (t)
in Table 9.2 for
f
m
= m/Ndt
Hz.
by
Eq.
(9.19b) and is illustrated in Fig. 9.9b. This illustrates the computational
procedure for computing the two-sided discrete amplitude spectrum \Grr(m)\
following the FFT path on the right side of
Fig.
9.7.
The two paths illustrated in Fig. 9.7 for computing the two-sided discrete
amplitude spectrum \Grr(m)\ for the normalized random time sequence f (t
n
)
in Fig. 9.9a will both be followed in order to demonstrate their differences.
First, the auto-correlation function Crr(t) is computed by Eq. (9.19a) from
the FFT coefficients B
m
for the normalized random time sequence £(?„) and
is illustrated in Fig. 9.10a. Algorithms for constructing covariance or corre
lation functions from FFT coefficients are given by Brigham (1974, p.206,
Fig.
13-6a) or by Bendat and Piersol (1986, Chapter 11.6.2, pp. 406-407).
Note in Fig. 9.10a the symmetry about T = 0 of the auto-correlation function
Crr (T)
in accordance with Eq. (9.17d); and the limiting values of the auto
correlation function Crr (T) ~ 0 = the mean /AI(£) as r ->• ±7>2dt = 6.4
sec in accordance with Eq. (9.18f). Next, the two-sided discrete amplitude
W a v e s a n d W a v e F o r c e s o n C o a s t a l a n d O c e a n S t r u c t u r e s
D o w n l o a d e d f r o m w w w . w o r l d s c i e n t i f i c . c o m
b y M C G
I L L U N I V E R S I T Y o n 0 4 / 0 5 / 1
6 .
F o r p e r s o n a l u s e o n l y .
-
8/18/2019 9789812774828_0009
21/148
Real Ocean Waves
739
- 6 - 4 - 2 0 2 4 6
T (sec)
Fig. 9.10a. Au to-correlation function for normalized time sequence f (?„) in Table 9.2.
0.25
-H4-
0.2
FmF -"
fc
444-
; 0 . 1 5 : m
0.1 --H-I-
: B I
0.05
444*^4-
H 4 - H 4 -
^ 4 -
- 1 ^ * 1 .
-|—i*r« -jp r-i -T-T—I"
Q ^ . . i | r 5 y h » | T r T
: P 3 : : c q :
bzfci l t d : t t t
- t - H - - n - i - - n - t - - i - n -
t c t l t c a i t z l i t
-H -
rp:
1 1 1 : 3 1 1 : 3 1 1
-t-i-t--t-t-t--i-t-t-
4 J -H4
41- 1 - 4
iIBIWi
• l -H-
4 H - 4 +
t + t
i n n t t n n i r
4
:
S
-144
-144-
nrx
. . . . . . . J -H -
m c d u m t a r
' - + 4 * 4 4 4
: •
0 8 16 24 32 40 48 56 64
m
Fig. 9.10b. Two-sided discrete amp litude spectra for the auto-correlation function C jf (r ) com
puted from Table 9.2 by the FFT coefficients from f (r){ »»} and by the Wiener-Khinchine
Fourier transform pair {xx}.
function
\G^(m)\
is then computed from the FFT coefficients of the auto
correlation function
C^(t) by Eq.
(9.19b) following the path on the lower left
side in Fig. 9.7. Second, the two-sided discrete am plitude spectrum |
G^
{m) \
is computed from the FFT coefficients of
the
normalized time sequence £ (/)
by Eq. (9.19b) following the path on the lower right in Fig. 9.7. Both of these
two-sided discrete amplitude spectra \G^(m)\ are compared in Fig. 9.10b.
The non-deterministic definition for Fourier coefficients in Sec. 9.2 is
illustrated in Fig. 9.10b. Both of the two-sided discrete amplitude spectra in
Fig. 9.10b have unit variance; but contributions to the variance by the ampli
tudes from each frequency are very different. The discrete spectrum computed
from the FFT coefficients for f (f) by the path on the right side in Fig. 9.7 are
limited
to
only
the
six discrete frequencies in Table 9.2. In contrast, the d iscrete
W a v e s a n d W a v e F o r c e s o n
C o a s t a l a n d O c e a n S t r u c t u r e s D o w n l o a d e d f r o m w w w . w o r l d s
c i e n t i f i c . c o m
b y M C G
I L L U N I V E R S I T Y o n 0 4 / 0 5 / 1
6 .
F o r p e r s o n a l u s e o n l y .
-
8/18/2019 9789812774828_0009
22/148
-
8/18/2019 9789812774828_0009
23/148
Real Ocean Waves
741
then a dimensionless generic four-parameter one-sided wave density spectrum
S( Q)
may be expressed by the product of the two dimensionless functions
F\ (Q ) and F2(fi) that are defined by
m mour
p
'
F i(f l ) = — , F
2
(Q) = exp - « " « ,
and that may be com bined according to the following product
— < oo. . (9.23e)
m
Dim ensionless values of
F \ (Si)/A, F
2
(fi) and S( fi) are illustrated in Fig. 9.11
and demonstrate that
F\{Q )/A
controls the spectral behavior at frequencies
higher than the dim ension less spectral peak frequency £2 > £2o =
a)/mo =
1
and that
F
2
(Q )
controls the spectral behavior at frequencies lower than the
dimensionless spectral peak frequency
Q
< £2o = «/
-
8/18/2019 9789812774828_0009
24/148
742
Waves and Wave Forces on Coastal and Ocean Structures
defined by
mour
n
Jo \m J mo/m
'
/0, (9.25a)
q \ q J
where T(«) = Gam ma function defined by Eq. (2.6a) in Chapter 2.2.5 and
the dimensionless zeroth moment mo = 1. For each dimensionless spectral
moment
n
in Eq. (9.25a), there corresponds a dimensionless characteristic
radian wave frequency defined by (Vanmarcke, 1983, Chapter 4 .1 , Eq. (4.1.4))
On =
(^)
l/
"
=
(-J^-Y"
=
(m,)
1
'-, »>0. ( 9 . 2 5 b )
m I \mozrr"
Spectral peak frequency
a>o
The dimensional spectral peak frequency u>o may be computed from
Eq. (9.23e) by
— ^ = 0, n = «
0
= - = 1 (9.26)
ail coo
so that the dimensional characteristic radian wave frequency parameter
m
may
be defined by COQ as
m =
coo
(-)
(9.27)
where (p/q)
l
^
q
is the multiplicative constant noted following Eq s. (9.20).
The dimensionless generic parameter
A
in Eq. (9.23e) may be replaced by
W a v e s a n d W a v e F o r c e s o n
C o a s t a l a n d O c e a n S t r u c t u r e s D o w n l o a d e d f r o m w w w . w o r l d s
c i e n t i f i c . c o m
b y M C G
I L L U N I V E R S I T Y o n 0 4 / 0 5 / 1
6 .
F o r p e r s o n a l u s e o n l y .
http://mozrr/http://mozrr/http://mozrr/http://mozrr/
-
8/18/2019 9789812774828_0009
25/148
Real Ocean Waves
143
Eq. (9 .25a) wi th
n =
0 and wi th the d ime nsionless var iance of the t ime
series mo = 1.0 (or, equivalently, the area und er the dim ensio nless spec trum
5 ( f i ) ) ;
i.e.,
A = - (9 .28)
r « / > - i )A? )
and the d imensional character is t ic radian wave frequency
m
from Eq. (9 .27)
so that Eq . (9 .23e) becomes
S(Q
0
) =
jP/o
q(.(j>/9)-D r ((p - \)/q)
u>
0 <
SIQ =
— < oo .
coo
Q
0
p
exp
—
| — Q
0
q
(9 .29a)
A dimensional form of Eq. (9 .29a) for arbitrary values of the exponents
p
and
q
may be obtained by mul t ip ly ing Eq. (9 .29a) by
mo/m
wi th
m
defined by
Eq. (9 .27) to obtain
S(co,mo,coo,p,q) =
1
(iq + \-p)lq)
m
0
p ( d - p ) / « ) co
0
r((p - \)/q) \ o
x exp
(9 .29b)
_ im
q\co
A number of theoret ical wave spect ra l densi ty funct ions have exponent
values of
p —
5 and
q =
4 . The parameters of several of these theoret ical
wave spectral density functions may be converted to the parameters of mo and
coo
to obtain a generic d imensional two-parameter spect ra l densi ty funct ion
given by
S(co,
m o,
coo,
P = 5 , q = 4) = 5
mo
(coo
COQ \
co
exp
5
/coo
~4\~co~
(9 .29c)
and that are tabula ted in Table 9 .5 . A d im ensio nless plo t of the gene ric spectral
density function Eq. (9 .29c) is shown in Fig. 9 .12.
It is not an easy task to dete rm ine the spe ctral peak freque ncy
coo
f rom me a
sured wave data because of the variabil i ty in real spectral est imates obtained
from dim ensiona l FFT algori thm s. An es t imate of the spectra l pea k frequency
may be computed from FFT coefficients in a best least-squares sense by
W a v e s a n d W a v e F o r c e s o n
C o a s t a l a n d O c e a n S t r u c t u r e s D o w n l o a d e d f r o m w w w . w o r l d s
c i e n t i f i c . c o m
b y M C G
I L L U N I V E R S I T Y o n 0 4 / 0 5 / 1
6 .
F o r p e r s o n a l u s e o n l y .
-
8/18/2019 9789812774828_0009
26/148
744
Waves
and
Wave
Forces on Coastal and Ocean Structures
mju>
a
1 . 9
n
K-
n
i
i
[.../.
-4
i
1
\
1
0 0.5 1 1.5 2 2.5 3
G5/W„
Fig. 9.12. Dimensionless two-parameter generic spectral density function for
p
= 5 and
q=4.
applying the linear Taylor differential correction method (Marquardt, 1963).
Because the dimensional FFT coefficients are indexed to discrete frequen
cies m, the method may be defined with discrete dimensional radian wave
frequencies a)
m
= mAco = m2nAf for the dimensional FFT algorithm
from Eq. (9.5b). A dimensional mean-square error e~ between a dimensional
measured
spectral density estimate SM(W) computed from the dimensional
two-sided discrete FFT coefficients at discrete frequencies m by Eq. (9.15i)
and the dimensional theoretical generic spectral density
S(m)
computed by
Eq. (9.29c) may be defined by
M
C
2
J2 [SM("0
-
5(m)]
,
(9.30)
1 =
1
M
c
-M
s
m—Ms
where
Ms =
the starting index for the first significant dimensional FFT
coefficient and Mc = the index of the cut-off frequency above which the
dimensional FFT coefficients are negligible in the measured spectral density
SM(»») -
By restricting the linear Taylor differential correction method to only
those few frequencies in the vicinity of the estimated value of COQ , the algorithm
is very efficient. The generic dimensional theoretical spectral density S(m)
from Eq. (9.29c) may be expanded in a Taylor series about the dimensional
spectral peak frequency u>o by
dS(m)
S(m) = S(m) +
dcoo
-
-
8/18/2019 9789812774828_0009
27/148
Real Ocean Waves
745
gives
SCOQ =
9 < w o
j : (S
M
(m )
- S(m))
m=Ms
(9.32)
An initial estimate for
J
0
+
8 1 .
(9.33)
The iterations are terminated when the corrections 8a)
J
Q
are stable and
acceptably small (10~
6
,say). Note that the theoretical spectrum S(m) from
Eq. (9.29c) must be recomputed after each iteration because of the newly com
puted value of o from data, the dimensional characteristic radian wave frequency m in the
dimensionless generic four-parameter spectrum in Eq . (9.23e) may be replaced
Table 9 .3 . Sum mary of least -squares fit to Hurr icane Car la da ta for
MQ
—
Ms
= 305
and N = 4096 (Hudspe th , 1975) .
R e c o r d N o .
06885 /1
0 6 8 8 6 / 1
06886 /2
06887 /1
Init ial col.
[rad/sec]
0 .52
0.50
0.50
0.50
Final
Q
[rad/sec]
0 .4990
0 .5187
0.4847
0 .5199
Final
-
8/18/2019 9789812774828_0009
28/148
746
Waves
a nd
Wave
Forces on C oastal and Ocean Structures
by the dim ensional average or mean radian w ave frequency
cb
or period T that
may be computed easily from the spectral moments in Eqs. (9.25) for
n =
1
by (Vanmarcke, 1983)
m\
{mQ Tu)m\ „ \ i J
co
= — =
— — — =
mmi = w—)
—-Y =
mA
2
\(p,q)
(9.34a)
m
0
(m
0
)(m
0
= 1) H ^ )
from Eq. (9.25a) and where
Aijip, q)
=
)
q
.{ .
(9.34b)
A dimensional characteristic radian wave frequency
m
and dimensionless
frequency ratio Q may be replaced by
OJ
m = coAn(p,q), fit = —, (9.35a,b)
OJ
where A12 is the multiplicative constant noted following Eq. (9.20) and
Eq. (9.23e) becomes
qA\
2
{p,q) (A
n
(p,q)\
q
CO
S(Q )= ' " " ' " e x p - "-?'
1
' , 0s
radian wave fre
quency may be defined as
mi
/
(mom
2
)m
2
_ .— .- __ .
a>
z
=Q )s = J— = J - T — — = mjmi (9.37a)
V mo V (
m
o)(mo = 1)
W a v e s a n d W a v e F o r c e s o
n C o a s t a l a n d O c e a n S t r u c t u r e s
D o w n l o a d e d f r o m w w w . w o r l d s c i e n t i f i c . c o m
b y M C G I L L U N I V E R S I T Y o n 0 4 / 0 5 / 1 6 .
F o r p e r s o n a l u s e o n l y .
-
8/18/2019 9789812774828_0009
29/148
Real Ocean Waves
1A1
and the dimensionless seco nd-mo ment m,2 from Eq. (9.25a) with the constant
A defined b y Eq. (9.28) for
mo
= 1 as
m
2
= ) \[ =
A31
(p,q), Vz = ^- (9.37b,c)
and Eq. (9.23e) becomes
•S("z) = —7 rv
TZ v
e x p - '
'(V)
« r
v «?
0 < ^
z
= - ^ < o o . (9.38)
Vanmarcke (1983) defines a generic characteristic wave frequency
Q k
as
i/k
" (
mA
m
0
/
«jfc = — • (9 -3 9 )
9 .3 .2 .
Wave and Spectral Parameters Computed from Spectral
Moments m
n
All o f the variab les in this section Sec. 9 .3 .2 are dimensional variables; and ,
consequently,
th e
tilde
( • )
notation applied
in
Sec. 9.3.1
is not
applied here
to denote dimensional variables. Dimensional spectral density moments
are
computed from
a
dimensional one-sided spectrum
S
m
{a))
b y
/•OO
m
n
= I
co
n
S
nr)
(oo)da). (9.40)
Jo
Many of the wave
a n d
spectral parameters that are computed below are eval
uated
for the
dim ension al generic two- param eter sp ectrum
S
m
{ma,a>o,co,
p
= 5,q =
4) in Eq. (9.29c); consequently, th e first four dim ens ion less
spectral moments computed b y a dimensionless E q . (9.40) are summarized
in Table
9.4
where
th e
Gamma function
F(») is
defined
b y E q .
(2.6a)
in
Chapter 2.2.5.
Spectral shapes
Dimensional spectral moments
m
n
computed from
E q .
(9.40)
are
functions
of the shape o f the dimen sional spectral density S
nn
{co). These dimensional
W a v e s a n d W a v e F o r c e s o n
C o a s t a l a n d O c e a n S t r u c t u r e s D o w n l o a d e d f r o m w w w . w o r l d s c i e n t i f i c . c o m
b y M C G
I L L U N I V E R S I T Y o n 0 4 / 0 5 / 1
6 .
F o r p e r s o n a l u s e o n l y .
-
8/18/2019 9789812774828_0009
30/148
748
Waves and Wave Forces on Coastal and Ocean Structures
Table 9.4. Summ ary of dimensionless spectral moments
m
n
for a dimen-
sionless generic two-parameter spectrum S^imo,ci)o,a>,p = 5,q = 4 )
where m
n
=
m
n
/o,5),p = 5,
q = 4) in Eq. (9.29c) for the first five dimensionless spectral mom ents
n =
0 - 4 in Table 9.4.
Spectral bandwidth parameters
e,q and
v
A dimensionless spectral bandwidth parameter e that may be applied to com
pute some extreme value statistics for a Gaussian process (vide., Sec. 9.4) is
the following (Cartwright and Longuet-Higgins, 1956):
e
2
= 1
2
-, (9.41)
where the dimensional spectral density moments m
n
are computed from
Eq. (9.40). For a narrow-banded spectrum, e -> 0 and the maximum values
for a stochastic process that is represented by a narrow-banded spectrum are
Rayleigh distributed (vide., Sec. 9.4.2). For a broad-banded spectrum,
e
-» 1
and the maxim a values for a stochastic process that is represented by a broad-
banded spectrum are Gaussian distributed (vide., Sec. 9.4.1). For the generic
two-parameter dimensional spectral density 5^^(mo,a>0 P = 5,g = 4) in
Eq. (9.29c), Eq. (9.41) is
rHm_
=
r ( i ) r (0)
W a v e s a n d W a v e F o r c e s o
n C o a s t a l a n d O c e a n S t r u c t u r e s
D o w n l o a d e d f r o m w w w . w o r l d s c i e n t i f i c . c o m
b y M C G I L L U N I V E R S I T Y o n 0 4 / 0 5 / 1 6 .
F o r p e r s o n a l u s e o n l y .
-
8/18/2019 9789812774828_0009
31/148
Real Ocean Waves
749
where r ( 0 ) = oo in Table 9.4. Consequently, all of the generic two-parameter
dimensional spectra
S
vr)
{m,Q,coQ,CL>,p
— 5, q = 4) that are tabulated in
Table 9.5 may be interpreted as being dimensional broad-banded spectra;
i.e., e —> 1. However, even though the dimensionless spectral bandwidth
parameters e,
q
and
v
discussed here are useful for estimating certain sta
tistical quantities, the effects of the variability in the spectra computed from
measured realizations limit the interpretation that these measured spectra are
broad-banded for engineering applications. The effects of this variability in the
realizations from these spectra are evaluated by the Hilbert transform and the
envelope function for engineering applications to damage estimates for rubble
mound breakwaters in Sec. 9.5. The fourth spectral moment m.4 is undefined
by Eq. (9.40) and in Table 9.4 for the generic two-parameter dimensional
spectra S(mo,coo,co,p = 5,q = 4) in Eq. (9.29c). In addition, m.4 is also
non-converging when computed numerically from data. For this reason, the
spectral bandwidth parameter e may not be computed reliably in engineer
ing applications from data; but it does continue to serve a useful purpose for
evaluating the theoretical distributions of the maximum values in time series
analyses (vide., Cartwright and Longuet-Higgins, 1956, Sec. 4, and Rice,
1954).
Extremal statistics may also be computed from the dispersion of the
wave frequency spectrum about the central spectral wave frequency from the
following alternative dimensionless spectral bandw idth parameter q (Vanmar-
cke,
1972):
2
, mi
q
2
= 1
1
— ,
(9.43a)
mom2
where the bold q in Eq. (9.43a) should not be confused with the italic q
in the exponent in Eq. (9.21). Because the spectral bandwidth parameter q
in Eq. (9.43a) does not require the fourth spectral moment ra4,q
2
may be
computed reliably from data. Vanmarcke (1983, Chapter 4) determines that
extreme values are Poisson distributed. For the generic two-parameter spec
trum
S(mo,(Oo,(o,p = 5,q =
4 ) , q
2
from Eq. (9.43a) with the dimensional
spectral mom ents ra„ from Table 9.4 is given by
7 r
2
( 3 / 4 )
q
2
= 1 - ^ = 0.152787. (9.43b)
Jn
W a v e s a n d W a v e F o r c e s o n
C o a s t a l a n d O c e a n S t r u c t u r e s D o w n l o a d e d f r o m w w w . w o r l d s
c i e n t i f i c . c o m
b y M C G
I L L U N I V E R S I T Y o n 0 4 / 0 5 / 1
6 .
F o r p e r s o n a l u s e o n l y .
-
8/18/2019 9789812774828_0009
32/148
Table 9.5. Conversion of two-param eter theoretical spectra to generic spectral param eters
m Q , a>o
a n (
l £/(•) = Heaviside step function.
SPECTRUM S(u>)
m
0
a>0
S(mo, C O Q , a>)
S-M-B: {H
s
,co
s
) (Bretschneider 1 958, 1966)
1.618f(^)
5
exp [-1.03(f)
4
]
Neumann {C , U
w
] (1943) (Kinsman, 1965)
(¥)
-)a> "exp '•\u
w
w) ]
P-N-J: Pierson, etal. {a,U
w
} (1955)
(vide., Pierson and M oskowitz, 1 964 )
^
e x
P
- 0 . 7 4
\UuCo) ]
ISSC{H
s
,
0
+ 0.26) -U(co-
co
0
- 1.65)}
WALLOPS
IP,p,coo)
(Huang, etal., 1981)
(^r-p[-f(^)
4
]
0.3932 tf
0
2
0.9528&).
3 C JT
64
\
Q.11U5)
0.71ft>
7
ft)o
ft)o
5^( f)
5
exp[- e)
4
]
24/|^(t)
6
ex
P
[-3(^)
2
]
5^(^)
5
exp[-I(f)
4
]
5^( f)
5
ex
P
[-I(^)
4
]
5 ^ e )
5
e x p [ - f e )
4
]
3.424mo exp
( a ) - a ) p )
[ 0 . 0 6 5 ( a ) - w o -
A l
s
8
£
a
s
a.
55
SPECTRUM S(f)
mo
/o
S(m ) « P -
(ff W a v e s a n d W a
v e F o r c e s o n C o a s t a l a n d O c e a n S t r u c t u r e s D o w n
l o a d e d f r o m w w w . w o r l d s c i e n t i f i c . c o m
b y M C G I L L U N I V E R S I T Y o n 0 4 / 0 5 / 1 6 .
F o r
p e r s o n a l u s e o n l y .
http://uuco/http://uuco/http://uuco/
-
8/18/2019 9789812774828_0009
33/148
Real Ocean Waves
751
Longuet-Higgins (1975) defines a spectral narrowness parameter v that
may be related to the Vanmarcke param eter q according to
v2 =
mom _
1 =
/momA
q 2
{gM)
m\ \ m\ ]
Goda dimensionless spectral peakedness parameter Q
p
Goda (1970) identifies a parameter Q
p
that correlates with wave groups in
Sec.
9.5 and is defined for dimensional one-sided spectral densities S
m
(f) by
2 r
0 0
Q P = —,\
fSl(f)df.
(9.45)
vv
The Goda spectral peakedness parameter Q
p
appears to be less sensitive to
the cutoff frequency that is required in order to compute spectral moments
m
n
from data and also less sensitive to wave nonlinearities than the spectral
narrowness parameter v in Eq. (9.44). The generic two-parameter spectrum
S
m
(mo, coo, co,p = 5,q
= 4) in Eq. (9.29c) may be converted from radial fre
quencies co = 2n f to Hertzian frequencies / by equating differential spectral
areas from Eq. (9.16d) according to
S(f)df = S(co)dco = S((o)2 7tdf,
S(f) = 2TZ S(CO).
For the generic four-parameter spectrum
S^ (mo, coo,
-
8/18/2019 9789812774828_0009
34/148
752
Waves
and
Wave
Forces on Coastal and O cean Structures
For the dimensional generic two-parameter spectrum S^ (mo, U>Q , CO, 5,4)
in Eq. (9.29c), Q
p
is
Q
P
= h (9.47)
that is approximately equivalent to
Q
p
for a broad-banded Gaussian white
noise spectrum (Goda, 1985).
9.3.3. M ulti-Param eter Theoretical Spectra
All of
the
variables in this section Sec. 9.3.3 are
dimensional
variables; and,
consequently, the tilde (•) notation applied in Sec. 9.3.1 is not applied here
to denote dimensional variables. Multi-parameter theoretical spectra include
both variance-preserving variable shape spectra and multiple peak (bi-modal)
spectra.
Goda-JONSWAP variance-preserving spectrum
The dimensional variance-preserving Goda-JONSWAP one-sided wave spec
trum is (Goda, 1985 or Chakrabarti, 1987)
e x p [ - ( / V o )
2
/ (2 r ,
2
/
0
2
) ]
;
(9.48a)
where
„ _ 0.0624
_
0.230
+ 0 . 0 3 3 6 / - 0.1 85(1 .9 + y ) -
1
'
r
a
= 0.07 i f / < /
0
; r
b
= 0.09 if / > /< ,,
and where 1 ,p = 5,q = 4) in Eq. (9.29c)
Snri(f) = a *
Hi
/o
exp
- 1 . 2 5
(9.48b)
(9.48c,d)
W a v e s a n d W a v e F o r c e s o n C o a s t a l a n d O c e a n S t r u c t u r e s
D o w n l o a d e d f r o m w w w . w o r l d s c i e n t i f i c . c o m
b y M C G
I L L U N I V E R S I T Y o n 0 4 / 0 5 / 1
6 .
F o r p e r s o n a l u s e o n l y .
-
8/18/2019 9789812774828_0009
35/148
Real Ocean Waves
753
derived in Sec. 9.3.1 if
Eq.
(9.29c) is converted from radian frequency
co
to
Hertzian frequency / by Eq. (9.16d) and
H
s
= 4^/mo
from Table 9.6. When
the spectral peakedness parameter y = 10, Eq. (9.48a) is a relatively narrow-
banded spectrum. Exam ples of these two extreme values for y are applied to
rubble m ound breakw aters in Sec. 9.5 in order to evaluate the effects of spec
tral shapes on wave groups and the corresponding damage to rubble mound
breakwaters. Goda (1998) modified the variance preserving coefficient a in
Eqs.
(9.48); but these modifications do not appear to be a good as the variance
preserv ing coefficient
a*
in Eqs. (9.48).
OCHI-HUBBLE six-parameter (bi-modal) spectrum
Ochi-Hubble (1976) derive a theoretical six-parameter (bi-modal) wav e spec
trum consisting of both low and high frequency peaks that results in a double
peaked spectrum. Each of these two frequency components require three
param eters: viz., a significant w ave height
H
Sj
,
a modal peak frequency /o
7
,
and a shape (or peakedness) parameter
Xj
where the subscript
j =
1 for the
low frequency components and j = 2 for the high frequency components.
Each of the two three-parameter spectral peaks may be combined into the
following single double peaked, one-sided spectral density function:
'nv
(/) = i E
(^-f
H
j=i
x exp
r(kj)
2TT/O,
f
4A.J + 1
4A./ + 1
f
(9.49a)
Ai = 2.72,
X
2
=
1.82exp(0.0277/,) = 1 .82exp(0 .108v^o) .
(9.49b,c)
The dimensional significant wave heights H
s
. in each of the two frequency
regions j may be replaced by the total variance of the wave spectrum
according to
Hi
H's
HU]
• = i
-«i
+• H ^ =
1 6 m
0
Rfrf]
= 16mo-
(9.50a)
(9.50b,c)
W a v e s a n d W a v e F o r c e s o n C o a s t a l a n d O c e a n S t r u c t u r e s
D o w n l o a d e d f r o m w w w . w o r l d s c i e n t i f i c . c o m
b y M C G
I L L U N I V E R S I T Y o n 0 4 / 0 5 / 1
6 .
F o r p e r s o n a l u s e o n l y .
-
8/18/2019 9789812774828_0009
36/148
754 Waves
and
Wave
Forces on Coastal and O cean Structures
f/foi
Fig. 9.1 3. Hinged wavemaker laboratory simulation of Ochi-Hubble 6 parameter wave spec
trum (| = raw unsmo othed spectral density estimates; = smoothed estimate; and =
theoretical spectral density function).
In order to compare laboratory or field spectra with Eqs. (9.49), it is
convenient to scale Eq. (9.49a) by m o /2 ^ /o , and obtain
Snnif/foj) _
m o / 2 7 T /
0 l
4((4A
1
+ l)/4)*l f, , (
H
\̂
2
\ (X-\
r ( M ) j
i +
U J J \foJ
, 4 /
0 l
«4A
2
+l)/4)*2 J Oh.}
2
]
1
(J-\
+
/o
2
r(x
2
) 1
1 +
\n
n
)
I
\fo
2
)
(4A, + 1)
exp
- (4X2+1)
-*[-m(*r]
(9.51)
An exam ple of a laboratory simulation of random waves by Eq. (9.51 ) in
a 2D w ave channel (vide., Fig. 5.1 in Chapter 5.1) by a hinged wavem aker at
the O. H. Hinsdale-Wave Research Laboratory at Oregon State University in
the USA is shown in Fig. 9.13.
The parameters applied in the laboratory simulation in
Fig.
9.13 w ere Ai
—
2.72,^2 = 1.8,/o
2
//o, - 3 . 5 , /
0 l
= 1/7Hz,
fo
2
=
1/2Hz,
H
S2
/H
S1
=
1.0
and mo = 0.01 1m
2
.
9.3.4. Spectral Directional Spreading Fun ctions
The ocean wave spectra reviewed in Sec. 9.3 and tabulated in Table 9.5 are
unidirectional spectra. The Wave Project I and II hurricane wave force records
W a v e s a n d W a v e F o r c e s o n C o a s t a l a n d O c e a n S t r u c t u r e s
D o w n l o a d e d f r o m w w w . w o r l d s c i e n t i f i c . c o m
b y M C G
I L L U N I V E R S I T Y o n 0 4 / 0 5 / 1
6 .
F o r p e r s o n a l u s e o n l y .
-
8/18/2019 9789812774828_0009
37/148
Real Ocean Waves
755
reviewed in Chapter 7.6.6 identify in-line and resultant force coefficients C
m
and
Cd
•
This
distinction
is due
to the directionality of
the
random waves passing
the instrumented offshore platform. The spectral models that incorporate this
directionality of random waves are reviewed briefly (vide., Borgman, 1969a
for an extensive review)
A fundamental method for incorporating wave directionality is to assume
a separation of variables model according to
~S
m
(eo, 9) = S
m
{o))D(e,
n,-), (9.52a)
D (0 ,n . - ) = % ^ , (9.52b)
where
S
m
(a>)
=
the unidirectional one-side spectra models reviewed above
and D(0,Tli) = a directional spreading function that may depend on sev
eral empirical parameters n, and/or radian wave frequencies
w
and that must
satisfy the variance-preserving constraint required by
/ .
71
D(9,Tli)d9 = l
(9.52c)
n
so that the variance of the time series of random w aves may be com puted from
m
0
= tf=
/
S
nJ]
{(o)D{e,Ui)dedo.
(9.52d)
Dirac delta distribution for uni-directional
wave
spectra
For the special situation where the random waves propagate uniformly in the
same direction, the spreading function may be represented by the Dirac delta
distribution in Chapter 2.2.3 that is given by
S(9-9
0
)
D(e,Yli)= \ °\
(9.53)
27T
where
0
= the direction of propagation of the unidirectional wave spectrum.
W a v e s a n d W a v e F o r c e s o
n C o a s t a l a n d O c e a n S t r u c t u r e s
D o w n l o a d e d f r o m w w w . w o r l d s c i e n t i f i c . c o m
b y M C G I L L U N I V E R S I T Y o n 0 4 / 0 5 / 1 6 .
F o r p e r s o n a l u s e o n l y .
-
8/18/2019 9789812774828_0009
38/148
756 Waves and Wave Forces on Coastal and Ocea n Structures
Cosine raised to even integer powers
Mitsuyasu (1971) proposed a fetch-limited directional wave spreading
function given by
D(9, n,o
C(s) =
C(s)
cos
2s
o-e
0
-n < 9
<
n
C(s)cos
2
'
s
(0-9o)
2
(2s-l)
r
2
( j +
j )
it r ( 2 s + i ) '
n
<
2
_
< f
C(s) =
1 l\s + l)
(9.54a)
(9.54b)
(9.54c,d)
where 9Q
=
principal direction of propagation of the unidirectional wave spec
trum,
the
param eters n,- = J and 5 = empirically determined integer constants
that control the amount of angular spreading about the principal direction
#o; and the Gam ma function T(«) is defined by Eq. (2.6a) in Chapter 2.2.5 .
The parametric dependency on the empirical parameter s of the w idth of the
spreading function is illustrated in Fig. 9.1 4.
When the integer parameters s = s = 1, then T
2
(2) = 1, T( 3/2 ) =
*Jn72, r(3) = 2 and Eqs. (9.54) reduce to (Borgman 1969a and 1972b)
'9-9
0
'
0(0,n,-)
s>
f
=1
=
C(s =
1) cos"
-it < 9 < n,
n
Cis =
1)
cos
z
(9 - do),
-
-z
<
9 <
TC
1 2
C(s = 1) = - , C(s = l) = -= 2C(s = 1).
it n
(9.55a)
(9.55b)
(9.55c,d)
-180-120 -60 0 60 120 180
(9-e„) (deg)
Fig. 9.14. Parametric dependency of cos
2i
(fl
— 9Q )
spreading function on the shape
parameters .
W a v e s a n d W a v e F o r c e s o n C o a s t a l a n d O c e a n S t r u c t u r e s
D o w n l o a d e d f r o m w w w . w o r l d s c i e n t i f i c . c o m
b y M C G I L L U N I V E R S I T Y o n 0 4 / 0 5 / 1
6 .
F o r p e r s o n a l u s e o n l y .
-
8/18/2019 9789812774828_0009
39/148
-
8/18/2019 9789812774828_0009
40/148
758
Waves and Wave Forces on Coastal and O cean Structures
1.4i
1.2
^ 1
CD
0.8
Q 0 . 6
0.4
H
0.2
0
— j _
— i
_ _ _ u .
—f_
« = 8
—f-
— h a = 4
— h
___;_.
s^^s
— r -
-\—
/
//"
~ ^
~3T^
y
a
- 1
a =
10-i-j--
NT~
M "
- ^v\
afO
__L
a =
6
. J _ l _
- ; - r -
- H -
i
- t
-fi-
i
— J — i —
H ^ S f ^
-90 -60 -30 0 30 60 90
(6-e
0
) (deg)
Fig. 9.1 5. Param etric dependency of circular normal spreading function on the shape
parameter a.
to the shape parameter
5
in Eqs. (9.54) by
2 a r c c o s f l - — } = 4 ar cc os [( 0. 5)
1 /2
*]. (9.58b)
Figure 9.15 illustrates the parametric dependency on the shape parameter a of
the circular normal directional wave spreading function.
SW OP w ave spreading function
A wave directional spreading function derived from stereo photographs of
ocean waves is the SWOP model (Stereo
Wave
Observation Project, Cox
and Munk , 1954a, b) given by
D(6, &) = - •
it
1 + I 0.50 + 0.82 exp
+ 0.32 exp
-*©'
K I
0)'
cos46>
cos 20
(O =
U
5
'
it
(9.59a)
(9.59b)
where Us = wind speed measu red at 5 m (16.4 ft) above the sea surface.
W a v e s a n d W a v e F o r c e s o n C o a s t a l a n d O c e a n S t r u c t u r e