Chinese-German Joint Symposium on Hydraulic and Ocean Engineering, August 24-30, 2008, Darmstadt

STUDY OF WAVE GROUP VELOCITY

CALCULATION FROM NONHOMOGENEOUS WAVE
FIELD BY SPATIO-TEMPORAL WAVELET
TRANSFORM
Li-Chung WU*, Laurence Zsu Hsin CHUANG*, Chia Chuen KAO* and Dong-Jiing Doong**
*
National Cheng Kung University, Tainan
**
National Taiwan Ocean University, Keelung

Abstract:

We here applied spatio-temporal wavelet transform to calculate wave group velocity from
inhomogeneous wave field and discussed its accuracy here. The relationship between wavelet and
wave group velocity results was presented. We revealed the practicability of spatio-temporal wavelet
transform in wave field analysis by using different kinds of simulated wave field. According to our
analyzing cases, we observed the calculated accuracy is influenced by the frequency resolution from
the wave field. At least continuous 64 images with the size of 128pixel×128pixel are capable of
obtaining adequate frequency resolution for wave group velocity calculation and getting high
accuracy results.

temporal wavelet transform, so as to understand the

I. INTRODUCTION practicability of the wavelet theory upon wave group
Ocean wave group velocity is the velocity at velocity calculations from the inhomogeneous wave
which a group of waves travels across the ocean. field.
More importantly, it is also the propagation velocity
of wave energy. Wave group velocity is always II. METHODOLOGY
significant factors in discussing both oceanography In this paper we used the spatio-temporal wavelet
and coastal engineering applications, as in the transform to detect wave group velocity from wave
studies on wave shoaling coefficient, wave energy field image sequences. Spatio-temporal wavelet
flux, long shore sediment transport equation, mild- transform is a particular time- or space-scale
slope equation and wave numerical mode. representation of signals that has found a wide
According to the wave theory, wave group velocity range of applications in physics, signal proceeding
can be estimated from the relationship among wave and mathematics.
frequency, wavenumber and water depth. In other The continuous wavelet transform is [1]:
words, wave spectrum analysis from wave field is a
key process to estimate wave group velocity.
Traditionally, the homogeneous 3D Fourier transform 1
S (b , τ , θ ; a , c ) = ψ b ,τ ,θ ;a ,c s ( x , t ) . (1)
was used to estimate the spectrum from Cψ
homogeneous wave field [2]. Owing to possible non-
homogeneous properties from nearshore wave field,
the method for analyzing non-homogeneous wave in which s (x , t ) = s(x, y, t ) is a three dimensional
field should be developed for obtaining reasonable signals of finite energy; the complex-valued function
wave group velocity. ψ b ,τ ,θ ;a ,c must satisfy the admissibility condition
Wavelet transform is recognized as a powerful tool
for non-stationary and also inhomogeneous signal.
Spatio-temporal wavelet transform is a branch of 2
wavelet theory; it is always focused on analyzing the ψˆ (k , ω )
∫
3
Cψ = (2π ) d 2 k dω < ∞ . (2)
signal both in time and space domain. It is workable R2 ×R 2
to apply spatio-temporal wavelet transform to k ω
calculate the speed of some objects by the speed
tuning parameter from the wavelet theory [1]. It where ψˆ is the Fourier transform of ψ , ω is the
should be feasible to calculate wave group velocity
by using the speed tuning parameter instead of angular frequency and k is the spatial frequency.
traditional Fourier spectrum analysis. The relationship between ψˆ and ψ b ,τ ,θ ;a ,c is defined
The aim of this research was to calculate wave as
group velocity from wave field images by spatio-

539
 Chinese-German Joint Symposium on Hydraulic and Ocean Engineering, August 24-30, 2008, Darmstadt

( (
ψ b ,τ ,θ ;a ,c (x , t ) = a −3 2ψ a −1c −1 3 r−θ x − b , a −1c 2 3 (t − τ ) . ) ) (3) According to the relationship between the wavelet
function ψ (x ) before and after scaling, shifting,
rotating, and speed turning, (3) is rewritten here:
In (3), b corresponds to the position of the
wavelet as it is shifted through the space domain. τ
is similar with b , corresponds to the location of the ψ b ,τ ,θ ;a ,c (x , t )
. (8)
wavelet as it is shifted through the time domain. θ is ( ( )
= a −3 2ψ a −1c −1 3 r−θ x − b , a −1c 2 3 (t − τ ) )
the function of rθ , which rotates the wavelet in
spatial coordinates [shown in (4)]. a is the scaling The relationship between the function ψ (x ) before
parameter, which is related to the spatial frequency and after scaling, shifting, rotating, speed turning in
of the space domain. c is the speed turning the space of Fourier transforms [6]:
parameter, the speed of the objects from the image
sequences could be detected by this parameter. It
was also used for calculating wave group velocity in ( )
ψˆ b ,τ ,θ ;a ,c k , ω
. (9)
our study.
( ()
= a 3 2 ⋅ e −i (b k +ωτ )ψˆ ac1 3 r−θ k , ac −2 3ω )
⎛ cos θ sin θ ⎞ The wave image for analyzing is always in a digital
r−θ = ⎜⎜ ⎟, 0 ≤ θ < 2π . (4)
⎝ − sin θ cos θ ⎟⎠ form. In order to use the framework of continuous
wavelets to analyze discretely sampled data, it is
Several wavelet functions have been proposed necessary to sample the analyzing wavelet [5].
before. For the sake of detecting wave direction from Consider the physical space series sampled with
wave field, one needs a wavelet which is sampling space ∆x m (Fig.1). The total amount of
directionally selective. Morlet wavelet function is one dimensional space for N x points, which will be
of the wavelet functions, and it is directionally defined as the number of sample points of the
selective. Due to Morlet wavelet function is often wavelet, is N x ∆x . If the total non-dimensional space
used for signal analysis and also ocean signal
analysis [3], it was used in our study. Spatio- length is 2 X , it is mapped for N x points. The
temporal Morlet mother wavelet function [6] is shown definitions of N t , T and ∆t are similar with N x , X
in (5). and ∆x . However, they mean the samples in time
domain. Considering the influence of discretely
sampled data, the relationship between wavelet
⎡ ⎤
ψ ( x , t ) = ⎢e (ik 0 ⋅ A ) ⋅ e − 0.5 A
2 2 2
−1
−1 − 0.5 A −1 x − 0.5 k 0
x x
−e ⋅e ⎥ parameters and ocean wave results is shown here
⎣⎢ ⎦⎥ . (5)
× ⎡e iω 0 t
⋅e − 0.5t 2
−e − 0.5t 2
⋅e − 0.5ω 02 ⎤ k 0 x XN t ∆t ω n
⎢⎣ ⎥⎦ c= ⋅ . (10)
ω0TN x ∆x k nx
Morlet mother wavelet function in Fourier space is
shown: Due to wave celerity ( v ) is the ratio of angular
frequency and wavenumber, (10) could be written
as:
⎡ −0.5 Ak − k 2 − 0.5⎛⎜ Ak + k 0 ⎞⎟ ⎤
2 2

ψˆ (k , ω ) = ⎢e 0
−e ⎝ ⎠⎥
⎢ ⎥. (6) ω0TN x ∆x
⎣ ⎦ v= ⋅c . (11)
k 0 x XN t ∆t
− e − 0.5(ω +ω0 ) ⎥
⎡ −0.5 ω −ω0 2 2 2
⎤
× ⎢e
⎣ ⎦
c is the speed turning parameter from wavelet
The parameters k 0 and ω 0 from (5) and (6) are function, it is non-dimensional. v is the wave
celerity, its unit is m/sec. As shown in (11), T = X if
the central locations of mother wavelet function in
Fourier space of space and time domain. A is an k 0 x = ω0 . Equation (11) can be simplified:
usual anisotropy matrix, and ε ≥ 1 :
∆x ⋅ N x
v= ⋅c
⎡ε −0.5
0⎤ ∆t ⋅ N t . (12)
A=⎢ ⎥. (7)
⎢⎣ 0 1⎥⎦
The equation for calculating the wave group
Equation (3) shows the analyzing result from wave velocity was shown in (13).
field by spatio-temporal wavelet transform, it is not
the wave group velocity really. It is necessary to
1⎛ 2kd ⎞
derive the relationship between wavelet result and Cg = ⎜1 + ⎟v
wave group celerity. 2 ⎜⎝ sinh( 2kd ) ⎟⎠ . (13)
1⎛ 2kd ⎞⎛ ∆x ⋅ N x ⎞
= ⎜⎜1 + ⎟⎜ ⎟c
2 ⎝ sinh( 2kd ) ⎟⎠⎜⎝ ∆t ⋅ N t ⎟
⎠

540
 Chinese-German Joint Symposium on Hydraulic and Ocean Engineering, August 24-30, 2008, Darmstadt

In addition to (13), we can derive C g by using (14)

where d is the water depth, g is the acceleration of
gravity.

Cg =
dω d
=
( gk tanh (kd ) )
dk dk . (14)
g tanh (kd ) + gkd ⋅ (sec h(kd ))2
=
2 gk tanh (kd )

Figure 2. Simulated irregular wave field image and its

corresponding bathymetry.
Figure 1. Correspondence between sampled points of the
physical quantity and the sampled points of the wavelet The spectra S (ω , k ) from different locations of
irregular wave field are shown in Fig. 3. The energy
III. VERIFICATION BY SIMULATED WAVE FIELD peak of the spectrum moves to high wavenumber
gradually if the analyzing location is close to shallow
To examine the validity of the new technique water area. The curve in every spectrum is the
presented above and to test the algorithm of dispersion relation curve. Due to the influence of
computing the spatio-temporal wavelet transform, water depth, the curve would be change in different
the following numerical simulations of wave field locations. We observed the energy from every
images were performed and discussed. spectrum was consistent with its dispersion relation
Ocean waves in Mother Nature are often irregular. curve. It meant the relationship between ω and k
In addition to the case of regular waves, we also from the wavelet results were consistent with the
tried to analyze irregular wave field by spatio- linear wave dispersion theory.
temporal wavelet transform. Irregular wave field The calculated wave group velocity from different
simulation is based on the directional wave sprctrum locations were analyzed (Fig.4), the differences
S (ωn , θ m ) which can be represented by simulation between theory and calculated values were also
[4]. similar with the regular wave field analysis. We
observed the errors were obvious in Location1 and
N /2 M Location6 from the wave field. It was also observed
η (t , x , y ) = ∑ ∑
n = − N 2 m =1
S (ω n , θ m )∆ω∆θ this phenomenon in the case of regular wave field
analysis. We discussed this issue in the following
. (15) section.
[ (
× cos nωt − k mn x x − k mn y y − ε mn )]
We simulated the irregular wave field by above-
mentioned methods. Wave field with 2.5 m
significant wave height, 45° dominate wave direction,
and 8 sec mean wave period were simulated. In this
case, spreading parameter smax was set up as 25.
The simulated irregular wave field is shown in Fig. 2.
Figure 3. The wave group velocity calculated results from
different locations of nonhomgeneous wave field by wavelet
transform.

541
 Chinese-German Joint Symposium on Hydraulic and Ocean Engineering, August 24-30, 2008, Darmstadt

Figure 4. The spectra from different locations of irregular and

non-homogeneous wave field.

IV. DISCUSSION Figure 5. The diagrammatic sketch of the wave field analysis by
We discussed the factors which affect the wavelet transform.
accuracy of wave group velocity calculation by
applying spatio-temporal wavelet transform. In addition to the influence of image edge, the size
Observing the results of Fig.4, the errors are more of wave field images in time and space domain also
conspicuous in Location1 and Location6 from the effect the wave group velocity calculation. According
wave field than other locations. It can be explained to (13) and (14), wave group velocity ( C g ) is related
by Fig.5. As shown in (1), the wavelet transform of to speed turning parameter ( c ), wavenumber ( k )
the image function is seen as the inner product
between wavelet function and image function (wave
( )
and spectra [ E k x , k y and E (ω , θ ) ]. Wavenumber is
field function). For the analyzing locations which are calculated from the wavenumber spectrum E k x , k y . ( )
close to the edge of image, wavelet function would In other words, accurate speed turning parameter
be incomplete due to the cut off by the image edge and spectrum are the keys to calculate wave group
(Fig. 5b). This is the reason that influences the velocity precisely. It was revealed that speed turning
accuracy of wave group velocity calculation. parameter is related to wave frequency and
wavenumber. For the accuracy of the spectra, it is
also related to the resolutions of wave frequency and
wavenumber. According to discrete mathematics,
the resolutions of wave frequency and wavenumber
are influenced by the size of wave field images in
time and space domain [ N t and N x ]. In order to
speed up the calculation of wavelet transform, we

542
 Chinese-German Joint Symposium on Hydraulic and Ocean Engineering, August 24-30, 2008, Darmstadt

applied the numerical techniques to the calculations. accuracy results by spatio-temporal wavelet
Hence, the numbers of continuous wave field images transform.
must be power-of-two sizes. We discussed the
calculated accuracy from different numbers of ACKNOWLEDGMENT
continuous wave field images by wavelet transform. This study is supported by National Science
The calculated location was chosen far from the Council (NSC) of Taiwan, Republic of China under
wave field image edges for the sake of avoiding the the project number NSC97-2218-E-006-010. The
influence of these edges. As shown in Fig.6, the authors would like to express their great thanks.
normalized errors are more less than 5% if the
numbers of wave field images are larger than 64. REFERENCE
[1] Antoine, J.-P., Murenzi, R., Vandergheynst, P., Twareque Ali, S.,
2004. Two-Dimensional Wavelets and their Relatives. Cambridge
university press, United Kingdom.
[2] Borge, J. C. N., Reichert, k., Dittmer, J. , 1999. Use of nautical
radar as a wave monitoring instrument. Coastal Engineering 37,
331–342.
[3] Chien, H., Kao, C. C., Chuang, Z. H. L., 2002. On the
Characteristics of Observed Coastal Freak Waves. Coastal
Engineering Journal 44, 301-319.
[4] Goda, Y., 1999. A Comparative Review on the Functional Forms
of Directional Wave Spectrum. Coastal Engineering Journal 41(1),
1-20.
Figure 6. The calculated accuracy from different numbers of [5] Jordan, D., Miksada, R. W., Powers, E. J., 1997. Implementation
continuous wave field images by wavelet transform. of the continuous wavelet transform for digital time series
analysis. Rev. Sci. Instrum. 68(3), 1484-1494.
As shown in (14), it is possible to calculate wave [6] Mujica, F. A., 1999. Spatio-temporal continuous wavelet
group velocity directly by wavenumber and water transform for motion estimation. School of Electrical and
depth. Hence, we also discussed the optimal size of Computer Engineering Theses and Dissertations Georgia Tech
wave field image under the condition of 64 Theses and Dissertations, Georgia Institute of Technology,
Georgia, USA.
continuous wave field images. It showed that the
calculated accuracy was related to the size of wave
field image. The normalized errors are less than 5 %
if the size of image is larger than 128 pixel × 128
pixel (Fig.7).

Figure 7. The calculated accuracy from different size of wave

field images by wavelet transform.

V. CONCLUSIONS
In this study, we tried to apply spatio-temporal
wavelet transform to calculate wave group velocity
from inhomogeneous wave field and discussed its
accuracy. From the results of simulated wave field
analysis, we observed the wave results were
consistent with the inputting wave conditions of the
simulated irregular wave field. The energy
distribution from the spectra was also consistent with
wave dispersion relation. It was revealed the
practicability of spatio-temporal wavelet transform in
wave field analysis.
We discussed the influence of the number and the
size of wave field image sequences upon the
accuracy of wave group velocity by applying spatio-
temporal wavelet transform. We revealed that at
least continuous 64 wave field images with the size
of 128pixel × 128pixel is capable of obtaining high

543