//
M.Sc. thesis
L.A.J. Fraza
t~;:
TU Delft
Committee:
Prof. dr. ir. J.A. Battjes
Abstract
Table of contents
Abstract
1 Introduction
1.1
1.2
4
5
5
5
1.3
1.4
Background
Development of the SW AN model
Aim of this study
Approach in this study
2 Theory
2.1
7
7
8
8
2.1.1
2.2
2.2.1
2.2.2
2.2.3
2.2.4
2.2.5
Wave theory
Action conservation equation
The SW AN model
Introduction
Discretisation
Implementation of the propagation schemes
Integration of the source terms
Implementation of the source terms
3.2
3.3
3.4
3.5
3.6
3.7
Introduction
Test
Model
Expected results
Results
Analysis
Conclusion
4.2
4.3
4.4
4.5
4.5.1
4.5.2
4.5.3
4.5.4
4.5.5
4.6
4.7
Introduction
Test
Model
Results
Analysis
Evolution of energy density
Influence of the time step
Influence of the limiter
No iterations per time step
Two iterations per time step
Comparison with other models
Conclusions
10
11
11
13
13
13
14
15
17
19
19
20
20
20
21
21
23
23
23
24
25
25
26
29
5 Moving Hurricane
5.1
5.2
5.3
5.4
5.5
5.6
Introduction
Test
Model
Results
Analysis
Conclusions
Introduction
Test
Model
Results
Analysis
Conclusions
Conclusions
Recommendations
30
30
30
31
31
35
35
36
36
37
37
37
41
42
43
43
44
8 References
45
9 Nomenclature
47
Appendices
1 Introduction
1.1 Background
Coastal constructions are subject to wave induced forces. For the design of
coastal or nearshore constructions it is important to assess the total wave
impact. Information on wave conditions is required for the estimation of beach
and dune erosion and for off and nearshore operations as well. To obtain
realistic estimates of wave conditions in coastal waters numerical wave models
are widely used. In these models the wave data at deep water are transformed
to the shore to obtain the wave climate in the coastal zone. Propagation,
generation, dissipation and nonlinear wavewave interactions are accounted
for in such models. A model recently developed especially for coastal regions is
the SWAN (acronym for .lmulating WAves liearshore) spectral wave model.
The development of spectral wave models started with the pioneering work of
Gelci (1956) and since then many models have been developed with
continuing extensions and improvements. Most wave models have been
developed for deep water conditions. Subjects that are of special relevance to
wave modelling are the propagation of waves, their generation by wind, their
nonlinear properties and their dissipation. The nonlinear properties have been
represented in wave models by first, second and thirdgeneration
formulations. In first generation models the nonlinear wavewave interactions
are not expressed explicitly. In second generation models these interactions
are highly parameterised, based on an assumed standard spectral shape. In
thirdgeneration models they are explicitly calculated, so that the energy
spectrum is computed without any a priori restrictions on the spectrum. The
WAM model of the WAMDI group (1988) is the first thirdgeneration ocean
wave model. The SWAN model is a thirdgeneration model as well.
For applications in coastal areas several models, firstly developed for deep
water, have been extended to shallow water by introducing depth dependent
propagation velocity and refraction (WAVEWATCH, WAM cycle 4). In these
deep water models the propagation of wave energy is integrated with use of
explicit numerical schemes. For stable integration using such schemes the time
step and geographical grid sizes are subject to the CourantFriedrichsLevy
stability criterion. On oceanic scales with geographical resolutions of a hundred
kilometre this leads to time steps of thirty minutes. For coastal applications with
relatively small grid sizes in geographical space (approximately hundred metre)
time steps of about ten seconds are required to meet the stability criterion (Ris,
1997). In addition the grids in coastal areas regularly contain up to ten times as
many grid points as oceanic models. The combination of many grid points and
small time steps increases the computational time excessively. These models
are therefore impractical for coastal applications.
The SW AN model is based on implicit numerical schemes for the integration of
the propagation terms. These schemes are unconditionally stable. The time
step and the geographical grid size can thus be chosen based on accuracy
rather than stability. This way the SWAN model can be economically used in
coastal regions.
2 Theory
2.1.1
a ( )+cxN(cr,e
a
)+cyNcr,8)+c,,N(cr,8
a (
a
)+c
a 8 Ncr,e
( )=s(cr,
e)
Ncr,8
at
ax
0y
ocr
08
cr
(21)
In this equation the first term in the lefthand side represents the local rate of
change of action density in time, the second and third term represent
propagation of action density in geographical space. Cx and Cy are the
propagation velocities in x and yspace respectively. The fourth term
represents shifting of the relative frequency due to variations in depth and
current. C" is the propagation velocity in crspace. The fifth term represents
refraction, either depth or currentinduced. C0 is the propagation velocity in 8space.
The term S(cr,e) on the righthand side is the energy source term. This term can
be written as the sum of separate terms, each representing a different type of
process:
(22)
Here S;n represents the generation of wave energy by wind. Sds represents the
dissipation of wave energy due to whitecapping, wavebottom interactions and
in very shallow water depthinduced wave breaking. Sn1 represents the transfer
of wave energy due to conservative nonlinear wavewave interactions (both
triad interactions Sn13 and quadruplet interactions Sn14).
2.2.1
a
a
a
E(cr,e)+cxE(cr,e
)+ cyE(cr,0)=
S(cr,e)
at
ox
Oy
(23)
In this chapter the discretisation in five dimensions is described first. Then the
difference schemes for the evolution in time and propagation in geographical
space are elucidated. Next the calculation of the source terms (the right hand
side of the equation) in SWAN is elucidated. The implementation of the source
term calculation in SWAN is described in the last paragraph of this chapter.
2.2.2
Discretisation
Both the action density balance equation that accounts for all previous given
processes and the energy density balance equation that is used in this study
are numerically integrated. To numerically integrate these equations in five
dimensions, they are discretised. The time step used in the SWAN model is
constant. The propagation and the source terms are integrated simultaneously
(using the same time step) in time. This is different than in some other models
(WAM, WAVEWATCH).
In geographical space the discretisation is based on a regular grid with
constant t:,.x and t:,.y. The discretisation in spectral space uses a constant
directional resolution /::,.B.
The discretisation in relative frequency space (o) is a bit more complex. The
frequency range of interest for wind wave forecasting lies between
approximately 0.25 and 2rr rad/s. The spectrum stretches to higher
frequencies. Energy in the high frequency range has influence on the low
frequency spectrum through wavewave interactions. Because it is to
expensive to calculate the spectrum in the entire range from 0.25 rad/s upward,
it is divided in two parts (see Figure 2a).
Figure 2a
Computational range of the
wave spectrum divided in
two ranges
Minimum frequency
Cutoff frequency
!
Frequency (rad/s)
prognostic part
diagnostic part
The first part is a prognostic part where the spectrum is explicitly calculated.
The second part is an analytical diagnostic part that is added at the high
frequency side of the prognostic part. This part is needed to calculate the
wavewave interactions across the cutoff frequency. The prognostic part in this
study lies between 0.25 rad/s and 2n rad/s. Between these frequencies the
energy density spectrum E(a, B) is represented by logarithmically distributed
discrete frequencies. This is done to allow for the efficient calculation of the
nonlinear wavewave interactions with the DIA of Hasselmann et al. (1985)
(see below). For frequencies lower than 0.25 rad/s the energy density is taken
to be zero. The level of the diagnostic spectral tail is determined by the energy
of the highest resolved frequency bin of the prognostic part. The shape of the
tail is prescribed according to:
(24)
Here Ohl is the cutoff frequency. The value of m has significant influence on the
results (Tolman, 1992a). The tail parameterisation is related to the physics and
beyond the scope of this study. In this study m = 5 and Ohl = 2n rad/s. Where
SWAN uses a fixed cutoff frequency other thirdgeneration models use a cutoff frequency that is based on variables as wind speed and mean or peak
frequency. The reason for the fixed cutoff frequency lies in the fact that SW AN
is developed for coastal areas.
In coastal areas (especially behind islands) swell can propagate while at the
same time a new wind sea develops. Such mixed sea states have different
characteristic frequencies. Using a selfscaling cutoff frequency for mixed sea
states may give inaccurate results. When the dynamic cutoff frequency is
scaled to swell, the cutoff frequency may be too low to properly describe a
locally generated wind sea.
2.2.3
The terms for propagation in geographical and spectral space are obtained with
the linear wave theory. For a description of these terms see appendix A. The
numerical schemes implemented in SWAN to integrate the propagation in time
and geographical space are elucidated here. For a description of the numerical
schemes for the other two dimensions (o and 8) reference is made to Ris et al.
(1998). The numerical schemes that are used in SWAN have been chosen on
the basis of robustness, accuracy and economy.
In the SWAN model implicit upwind schemes are used for evolution in time and
propagation in geographical space. These schemes permit relatively large and
mutually independent steps.
Due to the fact that in geographic space the state in a grid point is determined
by the state in the upwave grid points (as defined by the direction of
propagation), the spectral space is decomposed in four quadrants. Except for
the interactions between the different quadrants due to refraction and nonlinear wavewave interactions, the calculations can be carried out
independently per quadrant. To properly account for the interactions between
the quadrants the calculations are carried out iteratively. In the nonstationary
calculations the results from the previous time step are assumed to be a
reasonable approximation for the next time step, so the number of iterations
per time step is expected to be small.
The corresponding discretisation of the energy balance equation in time is
where for brevity a and 8 are omitted from E(o,8):
(25)
(26)
In these discretisations it is the time level index, ix, iy, i0 , and i8 are grid counters
and Llt, Llx and Lly are the increments in time and geographical space
respectively. n is the iteration index.
In the tests elucidated in the main report of this study the iteration index n=1
unless stated differently. This way every time step is calculated only once and
calculations are no longer iterative. When the time step is so large that the
results of the previous time step are not a reasonable approximation of the next
time step a higher iteration index can be chosen. This way more iterations are
carried out per time step. Tests with higher iteration indices are expounded in
chapter four. In the other tests the iteration index is equal to one.
10
2.2.4
The term on the right hand side of the energy density balance equation
describes the source terms. As the SWAN model is a thirdgeneration model
the shape of the spectrum is computed explicitly taking nonlinear wavewave
interactions into account. A full description of the nonlinear wavewave
interactions is given by Hasselmann (1963a). This description consists of a sixdimensional Boltzmann integral. To solve this integral with use of an exact
method takes excessive computational time. An approximation of the exact
method that strongly reduces the computational time and retains the essential
physical properties of the integral has been developed by Hasselmann et al.
(1985). This is the discrete interaction approximation (DIA). The SWAN model,
as most other thirdgeneration models, uses the DIA to calculate the nonlinear
quadrupletwave interactions.
When large gradients in energy density in spectral space are present,
integration of the source term with the DIA can lead to numerical instabilities.
To suppress the development of these instabilities a limiter is used in the
SWAN model. Like the DIA this limiter is copied from the WAM wave model.
The limiter restricts the total change of energy density per iteration per spectral
bin. It is set at 10% (Tolman 1992a) of the (directionally independent) Phillips
(1957) equilibrium level (formulated in terms of energy density):
(27)
2.2.5
A full description of the physics of the source terms S;n, Scts and Sn1 as modelled
in SWAN is not given here. Reference is made to Ris et al. (1998). In this
paragraph only the numerical implementation of the source term is briefly
elucidated.
The source terms at the right side of the energy balance equation are
estimated at every iteration level. The numerical estimations of the source
terms in the SWAN model are either implicit or explicit. The discretisation of the
source terms in SWAN can de described as where for brevity a and 8 are
omitted from S(o,8):
(28)
Here ix, iy, i' and i6 are grid counters and the iteration index n* for the source
term is either n or n1. The terms "implicit" and "explicit" refer to the
approximation of the source terms within each iteration. In the "implicit"
approximation the source term at iteration level n depends on the energy at
iteration level n, where in the "explicit" approximation the source term at level n
depends on the energy at iteration level n1.
11
All energy source terms except the linear wave growth term can be described
by S=<PE, in which <P is a proportionality coefficient that depends on wave
parameters and energy densities of other spectral components. Because these
parameters and energy densities are only known at the previous iteration level,
1
this coefficient is determined at that level, <ff' The linear term of S;n is
integrated with an explicit scheme because it depends on known parameters
only.
The source terms wind input (exponential) and positive triadinduced transfer to
each wave component are integrated with an "explicit" scheme:
(29)
Both positive and negative quadrupletinduced transfer to each wave
component are integrated with this "explicit" scheme as well. Tolman (1992a)
has shown that when using such a scheme in combination with the limiter (see
below) the results are similar to results of calculations that use a more
expensive "implicit" scheme.
Negative source terms are integrated with an "implicit" scheme where Sn does
depend on En. The strongly nonlinear, negative source term of depthinduced
wave breaking at iteration level n is estimated with a linear approximation:
(210)
The other negative terms are integrated with a simpler and more efficient
1
1
approximation where (aS/aEt is replaced by (S/Et :
(211)
=<t>n lEn.
(212)
12
3.1 Introduction
The velocity at which wave energy propagates in geographical space has
significant influence on the performance of nonstationary models. As part of
the development of the nonstationary option the wave energy propagation
velocity in the SWAN model is examined. This study is a first test to enquire if
the energy propagation velocity in the SWAN model is in accordance with the
group velocity of waves.
3.2 Test
The propagation velocity of wave energy (group velocity) can be analytically
calculated with use of the linear wave theory. In the SWAN model the
propagation of energy in time and geographical space is numerically integrated
with use of first order upwind implicit schemes. The propagation velocity in the
model is not given as output and has to be determined from the results of a
propagation calculation. This velocity is different from the value as calculated
analytically due to discretisation errors in the numerical integration. The
numerical propagation velocity depends on the geographical grid sizes and the
time steps that are used for the integration (Vreugdenhil, 1989). In these tests
only the influence of the used time step on the results is examined.
To be able to calculate the propagation velocity of wave energy a front is
introduced in the wave energy level of the test model (see Figure 3a). The
propagation velocity of the front is then calculated. The propagation velocity of
wave energy depends on the wave frequency. Monochromatic waves are used
in this test so the front will not be stretched out due to varying energy
propagation velocity for various wave frequencies. To prevent energy from
being dissipated or dispersed to other frequencies or directions all source
terms are deactivated in these tests. The energy balance equation now
reduces to:
a CxE(cr,0)= 0
ata E(cr,0)+ ax
(31)
Here the xaxis points in the mean wave direction. In the direction normal to the
mean wave direction (y) there is no net wave energy transport.
13
3.3 Model
A deep water model is used to perform this test. It measures 30 kilometre in
length (mean wave direction) and 40 kilometre in width (normal to mean wave
direction). The model is so wide to reduce the influence from boundary
conditions at the short sides of the model. The longitudinal grid size a hundred
metre. This is based on grid sizes in coastal applications. The lateral grid size
is ten times larger to reduce the computational time. The resolution in 8space
is five degrees. To calculate monochromatic waves with a two second period
two spectral bins are defined around cr = n rad/s with L'lcr is n/50 rad/s.
Calculations are made for two hours using a ten minute time step. To examine
the influence of the time step, calculations are made for a one minute time step
as well.
Despite the use of monochromatic waves and the deactivation of the source
terms the shape of the front defined in the initial condition will alter due to
numerical diffusion. Besides the erroneous boundary conditions at the short
sides of the domain will influence the results. Numerical diffusion is very large
due to the implicit schemes in SWAN. As a result of the large diffusion, a part
of the wave energy leaves the domain at the downwind boundary when the
front has only propagated a part of the length of the model. Because it is
necessary for the calculation of the propagation velocity that all wave energy
remains within the computational domain, the initial conditions are chosen in
such a way that numerical diffusion is minimised. To do so short period waves
are used.
In the initial state wave period and directional spreading are equal at the
upwind boundary and in the domain. It consists of monochromatic waves with a
two second period and twenty degree directional spreading travelling into the
domain perpendicular to the boundary. The energy level Etot at the upwind
3 2
boundary in the initial state is 15.625*10" m /Hz whereas Etot within the model
3 2
is 6.15*10" m /Hz. The velocity at which the front between the unequal energy
levels propagates in geographical space is considered to be the propagation
velocity of the wave energy. To be able to compare the results of tests that use
different time steps the initial state is equal for both calculations.
For the calculation of the propagation velocity the location of the front is
determined at several points in time. The location of the front is calculated by
taking the negative spatial derivative of the energy level at these points in time
(Figure 3d and 3e). The centre of gravity of the area under the derivative is
taken to represent the centre of the front.
Figure 3a
Shape of the front of wave
energy after ten minutes
(one ten minute time
step).
16
14
12
N
I
"'.... 10
E
'?
LU
6
4
10000
20000
30000
Xcoordinate
14
1
kd ] xc=x=1.56m/s
1 gT
c 9 = +
[ 2 sinh 2kd
2 2rc
(32)
As a result of the directional spreading the group velocity is smaller than the
above calculated value. The propagation of wave energy in mean wave
direction is equal to c9 xcos8 for all wave components, where 8 = 0 is the mean
wave direction.
Figure 3b
Propagation component in
mean
wave
direction
(dotted)
and
directional
distribution (line).
directional s eading
pi
pi
Deviation from mean wave direction (radians)
o(e) = Bcosm 8
(33)
~
The value of Bis such that:
Jo(e)de = 1.
(34)
~
Combination of (33) and (34) gives:
r(~m+1)
.
1
2
B .1s given
as: B = / ; (
1
r m+2
2
1
) .
cosm 8
=,
(35)
( )
36
15
For directional spreading of 20 9 , m "' 7. Using (32) the group velocity now
becomes:
rr4.5
r5 x Cg = 0.939 x Cg =1.47m/s.
'\/TC  
x Cg =
cos 8d8
/;~
(37)
r4.5
(38)
c b.t
cr = _g_ ,
b.x
(39)
2
rcb.x
L
(310)
Table 31
Time step
Ten minutes
one minute
.llt
L'l.x
L
600s
60s
100 m
100 m
Approx. 15000 m
Approx. 7000 m
Cr
0.946
0.993
16
3.5 Results
Etot in the model is presented in this paragraph at ten minute intervals from 60
to 120 minutes after the start of the calculations. The results in the first hour are
not presented because initial conditions influence the results in this period. Etot
is shown for both the ten minute time step (Figure 3c) and the one minute time
step (Figure 3d) calculation.
Figure 3c
18
16
14
12
N'
I
N' 10
"'0
.,... 8
6
4
2
0
0
20000
30000
20000
30000
10000
Xcoordinate (m)
Figure 3d
18
16
14
N'
12
N' 10
"'0
w
6
4
2
0
0
10000
Xcoordinate (m)
17
Figure 3e
Negative spatial derivative
of wave energy at ten
minute intervals.
Time step ten minutes.
E3
,ob
<ll
>
E:<ll
"C
<ll
>
~
0)
~ 1
10000
20000
30000
20000
30000
Xcoordinate (m)
Figure 3f
Negative spatial derivative
of wave energy at ten
minute intervals.
Time step one minute.
<ll
>
.2: 2
(i;
"C
Q)
>
~
0)
~ 1
10000
Xcoordinate (m)
18
Figure 3g
1.001
  Dt = 1min (numerical)
 Dt= 1O min (numerical)
.. group velocity (analytical)
0.999
+~~~~~~
3600
4200
4800
5400
6000
6600
7200
Time (s)
3.6 Analysis
The numerically calculated propagation velocity is nearly equal to the analytic
results for both calculations. The numerical propagation velocity is higher than
the analytically calculated velocity. this is contrary to what was expected. The
roundoff mistakes are larger than the difference between the numerical and
the analytical velocity. The decrease of propagation velocity for the last two
time steps in the ten minute time step calculation is due to energy is leaving the
domain at the leeward boundary. The strong increase of the width of the front is
due to numerical diffusion. As expected the numerical diffusion is considerable
in the SWAN wave model. This is due to the first order upwind implicit scheme
that is used.
Contrary to what was expected the propagation velocity as calculated with a
ten minute time step is equal to the value that results from the one minute time
step computation. No explanation was found for this as a full analysis of the
propagation scheme is beyond the scope of this study, but the method used to
calculate the location of the front can be of influence.
3. 7 Conclusion
The velocity of wave energy in the current SWAN model shows a very good
performance. It equals the analytical group velocity in deep water cases. The
influence of the time step on the propagation velocity is smaller that was
expected from the preliminary analysis.
Although the calculated propagation velocity is good there is reason to
implement another numerical scheme. The numerical diffusion of the used
propagation scheme is very large. Even for the small sized model as used in
this test. The nonstationary option is among other things being developed in
order that the SW AN model can be used for large areas. For such large areas
the numerical diffusion of the implicit scheme as used in version (30. 75) is even
larger than in this test. In this study it is not tested if the numerical diffusion
significantly influences the results because only the propagation velocity is
looked at. At time of writing a thirdorder (less diffusive) implicit propagation
scheme is being implemented in the SW AN model.
19
4.1 Introduction
In this test the growth of wave energy in time is calculated in idealised
circumstances. A constant, uniform wind field and deep water conditions are
used. In these idealised circumstances wave energy will grow to an equilibrium
state where energy input by wind is balanced by dissipation due to
whitecapping. Such growth curves that are dependent only on time are
presented here to illustrate the performance of the SWAN model.
The growth curves are found to be very sensitive for the time discretisation.
The main attention in this test is therefore given to the influence of the time
discretisation on the evolution of significant wave height and mean or peak
frequency. To test the influence of the time discretisation several calculations
(with varying time step size) are performed. The growth curve that is least
influenced by the discretisation is then compared with that of other models.
4.2 Test
The evolution of the wave spectrum depends on the source terms for input,
dissipation and the nonlinear interactions. To separate the integration of the
source terms from the propagation terms homogeneous deep water conditions
are considered in this test. In the initial condition no wave energy is present in
the model. A linear wind input source term is activated so that energy will grow
for the initial condition given. The energy density balance equation now
reduces to:
(41)
The tests are done for three different time step sizes: 750, 150, and 30
seconds respectively. In this study the main attention is given to the calculation
of the evolution process. A 24 hour period is calculated. A fullgrown
(equilibrium) state is not reached in this period.
The results of the test calculated with the largest time step showed a restriction
of the growth rate compared to the small time step calculations (see figure 4a).
To allow for larger growth rates (see analysis) one calculation has been
performed where two iterations (iteration index n equal to three) are computed
per time step. The time step used in this calculation is 750 seconds.
20
4.3 Model
For these tests a large rectangle is modelled so that the results will not be
influenced by fetch. The length of the model (parallel to the wind direction) is
12000 kilometre. The grid size in wind direction is 250 kilometre. The width is
30000 kilometre to minimise the influence of the boundary conditions at the
short sides of the model on the results in the middle. To reduce the
computational time the grid size in this dimension is 2000 kilometre. The wind
speed used is U10 = 20 mis. Results are requested 10000 kilometre from the
upwind boundary and in the middle of the model relative to width. Fetch can
then be considered infinite.
The waves travelling in the wind direction (sector 90 to 90) are calculated
with a 10 resolution in 8space. Calculations are not performed outside this
sector. In frequency space 25 bins are defined where fm;n = 0.04 Hz and fmax =
1 Hz. The linear wind growth term of Cavaleri and Malanotte (1981) is activated
in these tests.
4.4 Results
The evolution of the significant wave height (figure 4a) and the mean frequency
(figure 4b) is given for three different time steps. From these results it can be
seen that the time step size strongly influences the results. The larger the time
step, the more the growth rate is restricted. The mean frequency decreases
correspondingly more slowly when a larger time step is used.
In figures 4c and 4d the results of a calculation with a 750 second time step
and two iterations per time step are presented. They are plotted against the
results of the 30 second time step (where n is equal to one still).
The development of the wave spectra in the first ten hours is presented in
appendix B. These spectra are from calculations with no iterations per time
step and either a 30 or a 750 second time step.
10
Figure 4a
Evolution of significant wave
height for different time
steps.
Dt=30s
Dt=150s
~Dt=750s
0
0
12
Time (hr)
18
24
21
0.35
~+ 
Figure 4b
Dt=30s
Dt=150s
0.3
~Dt=750s
g: 0.25
c:
ct!
Q)
0.2
LL
0.15
0.1
0
Figure 4c
12
18
24
Time (hr)
10
]:
(/)
::r:
4
:I
0
Dt=30 s
 Dt=750s; 2 iterations
12
18
24
Time (hr)
0.35
Dt=30 s
 Dt=750s; 2 iterations
Figure 4d
0.3
Evolution of mean wave
frequency
using
two
iterations per time step.
(Calculation using a 30
second time step and no
iterations for comparison.)
g: 0.25
c:
ct!
Q)
LL
0.2
0.15
0.1
0
12
Time (hr)
18
24
4.5 Analysis
Several processes have influenced the results. Among these are the time
discretisation and the limiter (see Chapter two). An elucidation of the influence
of these different processes is given in this paragraph. The results are then
analysed taking the different processes into account.
4.5.1
Source terms and energy density are interdependent variables. The time
scales at which the energy density spectrum responds to the source terms and
vice versa, are relative to the different frequencies of interest (Komen at al.,
1994). The response time scales are relatively short for the high frequencies
and longer for low frequencies. The growth of wave energy in these tests starts
in the high frequency range. Short time scales then govern the growth process.
In these tests the rate of change of wave energy strongly reduces in time in the
high frequency range. The frequencies of interest (with larger wave energy
change rates) now shift to lower values. Therefore in time the interest shifts
towards larger time scales.
As a result of the stationary wind conditions used in this study the rates of
change of wave energy become very small in the high frequency range.
Because the spectral shape in the high frequency range depends on the
spectral shape of the still developing low frequency range a true equilibrium is
not reached. In these tests, with the used discretisation in frequency range and
spectral shape of the diagnostic part, a near equilibrium has developed. The
spectral shape at the high frequency range (in both developing and equilibrium
state) is still a point of discussion in wave modelling. Whether the shape of the
spectrum in the high frequency range of the prognostic part as calculated in
these tests is in accordance with the physics is not investigated in this study.
This is more a discussion on the performance of the DIA, also related with the
shape of the analytic tail.
4.5.2
The time discretisation will influence the numerically determined results in two
ways (Tolman, 1992a). First, errors occur due to the discretisation. Second,
time step sizes that are large relative to the time scale of the change of
calculated processes can lead to the activation of the limiter.
The disparity in response time scales within the frequency range is large in the
SW AN model as a results of the wide prognostic discretisation in frequency
space. The time step used for the integration of the source terms is constant for
all discrete frequencies. In the high frequency range with short response time
scales, this constant time step is relatively large where in the low frequency
range it is relatively small. When the time step size is larger than the time
scales of interest the integration will become unstable. The range where the
integration becomes unstable depends on the time step size. Decrease of time
step size decreases this range. Reduction of the time step size to such an
extent that the integration becomes stable in the entire prognostic range leads
to uneconomically small time steps.
For economic use the time steps are chosen small enough to calculate the
evolution of the energy density spectrum near the frequencies that contain
most energy. These time steps are so large that the integration in the high
frequency range becomes unstable. To allow for a stable integration despite
the use of these large time steps a limiter is used in SW AN and other third
generation models (WAM, WAVEWATCH). This limiter is activated when the
rate of change of action density becomes too large (see Chapter two).
23
Figure 4e
The response time scales
vary
for
different
frequencies.
The brackets show the
frequency range
where
response time scales are
smaller then the used time
step. In this picture time
step 1 is larger then time
step 2.
N'
::r::
.,.,. time step 1
'""~~
UJ
time step 2
Frequency (Hz)
O
Increase time scales
By restricting the maximum rate of change of energy density the limiter ensures
a stable integration. As a result the integration remains stable when time steps
are used that are too large to calculate the source term integration in the high
frequency range. Activation of this limiter will influence the results (see next
paragraph).
In these tests the influence of the time step decreases in time. This is due to
the stationary (wind) conditions. Once a near equilibrium state is reached in the
high frequency zone (as described in the previous paragraph) the rate of
change of wave energy tends to zero within this interval. The growth of wave
energy is now determined by processes in lower frequencies. Large time steps
can still lead to instabilities in the high frequency range. These instabilities at
their turn will still activate the limiter in the high frequency range. Activation of
the limiter in this range no longer influences the results because a near
equilibrium is reached.
4.5.3
The limiter restricts the rate of change per iteration. Activation of the limiter can
influence the results in two ways. Growth of energy can be restricted in one or
more spectral bins. By restricting growth in a number of bins, the shape of the
energy density spectrum will be altered.
Any restriction of change of energy density is a restriction of the evolution of
significant wave height, because the significant wave height increases with the
total energy density. Due to the time discretisation, the limiter is activated (if
activated) in the (relative) high frequency part of the prognostic range. This
results in restricted growth in this range and unrestricted growth in the low
frequency range. As a result the spectral shape alters, leading to a lower mean
frequency. This is a direct result of the activation of the limiter. By restricting the
maximum change of wave energy per bin per iteration in a sequence of
calculations the limiter also restricts the shift of wave energy to low
frequencies. This shifting of wave energy within the spectrum is one of the
main properties of the nonlinear wavewave interactions. The limiter does not
prescribe a solution because the change of wave energy is limited and not the
value of wave energy. The full grown (equilibrium) state, which is not presented
here, will therefore not be influenced.
24
In this test calculations have been carried out with either no or two iterations
per time step (n is equal to one or three). First the influence of the limiter in
case no iterations are calculated is discussed. Then the influence of the limiter
is elucidated in case more iterations are calculated per time step.
4.5.4
4.5.5
A possibility has been looked at to increase the growth rate in this study without
large increase of the computational time. To not restrict energy growth, two
solutions are possible. In the first option the time step depends on the growth
rate. Small time steps are then used only for large growth rates. This requires a
variable time step. This option is used in the WAVEWATCH model (where the
time step is variable in geographical space as well, LH =f(t,x,y)). The use of this
option for the SWAN model is not investigated in this study because this
requires a drastic change in the source code.
In the second option the growth rate is not restricted despite the use of large
time steps. To achieve this the number of iterations per time step is increased.
When extra iterations are needed at one point in geographical space, an extra
calculation is performed for the entire model (n = n(t)). Although restricted per
iteration, the growth is now no longer restricted per time step. To decrease the
computational time in comparison with the option where small time steps are
used the number of iterations only has to be increased when high growth rates
occur. A criterion has to be introduced which the number of iterations will be
based on. In this study only a first investigation is made to see if an increase of
the number of iterations leads to satisfactory results. If so, criteria can be
formulated and possibly implemented in the model. In the tests a fixed number
of two iterations and a 750 second time step have been used. The results of
this investigation are remarkable. With 2 iterations per time step (n is equal to
three) the growth rate of wave energy and the rate of shifting the mean
frequency increases strongly (see figure ()).
25
The time step size is large in this test, so the frequency range where the limiter
is activated is wide. The 'extra' growth now occurs in the low, unrestricted
frequency range. This explains the rapid decrease of mean frequency.
Because of the stationary conditions, the spectrum that is calculated develops
to the same equilibrium state as the noniterative calculations.
This option counteracts the influence of the limiter in this test. The 'extra'
growth increases the growth rate of wave energy at the start of the calculation.
When evolution of the spectrum in the high frequency range is of main interest
(here only at start of calculation) small time steps are still needed, as time
steps need to be smaller than the time scales of the process calculated.
g2Etot
Ufp
= ~ , fp = g
Where
u?
Cd(U
)= {
10
, t
gt
= LJ.
(42)
1.2875x10for
(0.8+0.065s/m x U10 )x 10 3 for
by Wu (1982) as:
(43)
The results of SWAN and all models investigated by the SWAMP group (except
EXACTNL) are presented in figures 4f and 4g.
26
.
Nonstationary calculat1ons
w ith the SWAN wave model
Figure 4f
Dimensionless gro wth curve
for
I
ve energy
d SW AMP
for tota wa
SWAN (line) an
{hatched).
104
E*
103
102
10
104
105
107
106
t*
Figure 4g
0.03
evolution
Dimensionle~s frequency
for
curve of pea
d SW AMP
SWAN (line) an
(hatched).
f*
p
0.01
0.004
104
107
10 5
t*
27
10
Figure 4h
Hs(m)
0
0
0.35
Figure 4i
0.30
f (Hz)
12
..
..
Time(hr)
10
24
. Wavewatch
0.25
\
\
.
\.
0.20
0.15
....
..................
0.10
0
...... '111...............
_ _ _ _ __J
............_
................................
12
Time (hr)
18
24
The comparison with the W AVEW ATCH model shows good resemblance for
the evolution of wave energy. The evolution of the mean frequency shows
significant difference between the models.
28
4.7 Conclusions
To calculate a wave spectrum explicitly based on third generation formulations,
the source terms are integrated up to high frequencies. The integration in the
high frequency zone requires small time steps as time scales, at which the
source terms respond to changing wave spectra and vice versa are short in this
range.
In the SWAN model large time steps can be used. The integration remains
stable for these large time steps due to the limiter that is used. When using
large time steps the limiter influences the integration, and the results are not
based only on the third generation formulations. Noniterative calculations
converge very slowly. The solution is lagging because growth and frequency
shift are restricted. To calculate uninfluenced nonstationary calculations time
steps have to be so small that the limiter is not activated. To prevent activation
of the limiter time steps have to be very small because in SW AN the energy
density equation is integrated up to a high frequency.
The possibility to use extra iterations per time step leads to satisfactory results
in this test. The restriction of energy growth by the limiter is counteracted. The
growth process at the start of the calculation (in the high frequency range) is
not properly calculated with large time steps using extra iterations however. To
calculate this process with small time scales, small time steps are needed. In
the next tests the calculations are noniterative, because growth rates are not
expected to be as large as in this test.
29
5 Moving Hurricane
5.1 Introduction
In the previous tests propagation and growth were investigated independently.
In this test these two processes are combined in the rather extreme
circumstances of a hurricane. Contrary to the previous test, wind conditions
vary in time at every grid point. The results of this test are compared with
results of the SAIL model (SWAMP 1985) because this is a good secondgeneration wave model.
5.2 Test
The test consists of a stationary hurricane wind field (see below) that is moving
over a rectangular model at a constant velocity. This way a nonstationary
process is defined because at every grid point of the model, the wind velocity
and direction are constantly changing. Because a constant wind field
propagating at a constant velocity is used, an equilibrium state will develop for
locations defined relative to the moving wind field. Within the twenty four hours
of this test an equilibrium is not reached, however.
Contrary to the test in the previous chapter wave energy in present in the
domain at the start of the calculation. Because time discretisation has large
influence on the results (see Chapter four), calculations are performed for
different time steps. The results of the evolution of significant wave height and
mean frequency as well as the entire wave field after twenty four hours, are
compared with the results of the SAIL model. The SWAN results will differ from
the SAIL results because SAIL is a second generation model, but the evolution
of wave energy is expected to show a similar tendency.
wind [m/s]
Hurricane field
Figure Sa
Hurricane wind field over
the model at start of the
calculation.
40
2.5
35
30
25
~1.5
.s>
20
15
10
0.5
5
0
0'..l.'
5
10
x[m]
15
20
5
x 10
30
5.3 Model
The hurricane wind field used in this test is equal to the field used in tests done
by the SWAMP group. In this field the maximum wind speed of 40 mis is
located at SO km northeast of the eye of the hurricane. The wind field
propagates over the model at 1Sm/s in positive y direction (North). The model
used is a 13SO km (x direction) by 2700 km (y direction) deep water rectangle.
The location of the eye of the hurricane is not explicitly given for the wind field
used in this test. At the start of the calculation the eye is located at
approximately x=690 km and y=840 km. The input grid size (~x. ~y) for the
wind field is 77.S km by 77.S km. The used resolution in geographical space is
37.S km by 37.S km, which is equal to the resolution used for SAIL. In
frequency space 2S bins are defined, where fm;n is 0.03SHz and fmax is 1 Hz.
The resolution in espace is 10. Apart from the wind field no boundary
conditions are given. The SAIL model uses a one hour time step, while the time
step used in the SW AN calculations is S and 1S minutes.
The initial condition that is given in the SW AN model consists of a fetch limited
JON SW AP spectrum that is calculated at every grid point. This spectrum is
computed from the local wind speed and direction. As fetch the average (over
the entire computational grid) of the spatial resolution is used. The formulations
for the calculation of the initial condition have been adapted because results
were nonsatisfying at first. The routine to calculate an initial condition
contained an error. The relation for dimensionless significant wave height and
peak frequency as given by Kahma and Calkoen is now used. The PiersonMoskovitz relation is used for limit values (see footnote). For SAIL the
parametric model of Ross (1976) was used to provide an initial condition. This
model defines the total energy and peak frequency at any point in a hurricane
from the wind speed and radial distance from the eye (SWAMP 198S).
Calculations have been performed with this initial condition and starting without
any energy present.
5.4 Results
First the development of the maximum significant wave height (Hsmax) and the
mean frequency are presented. Hsmax is the maximum significant wave height
within the entire computational domain at one point in time. Then the wave field
as calculated after twenty four hours is given. In addition a two dimensional
spectrum is given at Hsmax at this point in time.
The evolution of Hsmax and the mean frequency is presented in figures Sb and
Sc.The mean frequency that is presented is taken at Hsmax Results are given
for SWAN for both time steps. Besides two calculations with the SAIL model,
as presented by the SWAMP group are shown. The evolution of both
significant wave height and mean frequency for the SW AN model shows a
dependency on the time step size that is known from the previous test. The
time step size shows little influence on the state as developed after twenty four
hours.
The wave field after twenty four hours, as calculated by both models, are now
compared. The SWAN result is presented in figure (Sd).
Footnote
fp
31
Figure Sb
Evolution of Hsmax for SAIL
and SWAN.
20
Hs(m)
15
10
SAIL with initial condition
Figure Sc
Evolution of mean wave
frequency for SAIL and
SWAN.
12
Tirne(hr)
18
24
0.22
f(Hz)
SAIL with initial condition
0.13

 
0.04
12
nme(hr)
18
24
The result of SAIL was not entirely presented by the SWAMP group. Instead a
couple of parameters were presented that show the character of the calculated
wave field. The SAIL parameters presented in table (51) describe the wave
field of that calculation where the parametric model of Ross (1976) was used to
provide an initial condition. To generate these parameters for SWAN, the five
minute time step calculation was used.
32
Hs [m]
Wave field
Figure Sd
Wave field after twenty four
hours as calculated by
SWAN. Arrows show mean
wave direction
20
2.5
15
~1.5
.s>
10
0.5
0
0
5
10
x[m]
15
20
5
x 10
Comparison of both models after twenty four hours is based on these
parameters.
SWAN
SAIL
Hs maximum (m)
19
20
ESE
SE
Distance (km)
140
90
Distance Hs/2
500
375
0.08
0.07
NNW
NNE
Table 51
Results of SWAN and SAIL
after twenty four hours.
fmean
emean
at (Hs)max (Hz)
at (Hs)max
The results show a maximum significant wave height that is nearly equal to the
SAIL result. The mean wave frequency as calculated by SWAN is higher than
the SAIL result. The area that is influenced by the hurricane is larger in the
SWAN calculations than in the SAIL tests. This shows in the larger distance
from the eye of the hurricane to the location where the significant wave height
is reduced to fifty percent of the value of Hsmax
33
In figure (5e) The location of Hsmax relative to the eye of the storm is given for
both models, in combination with the mean wave direction. The arrows start at
the location where the significant wave height reaches its maximum value. The
arrows point in the direction of the mean wave direction, and the arrows are
scaled to the significant wave height.
Figure Se
Mean wave direction at
Hsmax after twenty four
hours for SAIL and SWAN.
Arrows start at location of
Hsmax and point in mean
wave direction. The length
is scaled to the significant
wave height
200
Y[km]
SCALE
Hs=10m
Eye~
SAIL
200
300
300
X[km]
In figure (5f) a two dimensional wave spectrum is given at the location where
the significant wave height reaches its maximum value at twenty four hours.
The arrow shows the local wind direction. Most wave energy is contained in
waves propagating north.
NORTH
Figure Sf
...
WEST
.:.:....
EAST
0.05
0.1
SOUTH
34
5.5 Analysis
The results will be analysed in this paragraph in the same order as they were
presented in the previous one.
The wave field develops to a steady state relative to the wind field. The
increase of the amount of energy within the entire model depends on the time
step. For the development of the wave spectrum at a single location the time
step is less important than in the previous test because wave energy can
increase due to propagation. The maximum wave height in the model depends
on the amount of wave energy in the rest of the model through the propagation
terms. The value of the maximum significant wave height is a reflection of the
total amount of energy within the model therefore. When the entire model is
looked at as a whole, propagation of wave energy does not contribute to the
growth of energy. In that case, energy can only growth due to wind input. Here
the influence of the time step is shown once more. Reduction of the time step
leads to an increase of the growth rate of the total amount of wave energy in
the model.
At every point in time, the hurricane generates waves that propagate in all
directions. In the direction of translation of the hurricane these waves are
superimposed. Within the hurricane, waves that travel parallel to the hurricane
path are generated east of the eye. The location where Hs will reach its
maximum value is located east of the eye. Because most energy is contained
in waves that propagate more slowly than the hurricane itself, Hsmax is lagging,
i.e. is located south east of the eye. This shows in figure (Sd) for SWAN and for
the SAIL model in table (S1). For most other models tested by the SWAMP
group, this result is found as well.
A twodimensional spectrum is shown in figure (Sf). Most energy is contained in
waves propagating north. This is equal to the direction of movement of the
hurricane. The local wind direction has shifted the mean wave direction. The
mean wave direction at the location where the significant wave height reaches
its maximum value in the SAIL model is different (see figure Se), but the local
wind direction is different as well, because Hsmax is located elsewhere.
The area affected by the hurricane is larger in the SWAN model. The location
where maximum wave height occurs in SWAN is located more east of the eye
at a larger distance than in SAIL. The distance from the eye where the
significant wave height is reduced by fifty percent is different for both models.
5.6 Conclusions
Now that propagation is accounted for, wave energy at a single grid point can
increase due to propagation of wave energy. The influence of the time step on
the growth rate reduces.
The wave height as calculated with the SWAN model is comparable with the
SAIL results. The location of Hs max and the mean wave direction at that
location differ for both models. However, the models are totally different in the
approach to calculate the wave spectrum. The shifting of the location where the
significant wave height reaches its maximum value that was calculated by the
other models is computed by SWAN as well.
The mean frequency is higher in SWAN. than in SAIL. A similar result was
found in the previous test as well.
35
6.1 Introduction
Figure 6a
In the last test of this study the performance of the SWAN model is examined in
a real case. In the first days of December 1989 a storm hit the area around the
island of Malta in the Mediterranean sea. It was called Gorbush storm after
President Gorbachev and President Bush who had a meeting on a ship just off
the coast of the island on 2 December. This storm is used as input for the test.
The results of SWAN are compared with results of the WAM (WAMDI Group,
1988) wave model and with measurements near Malta and Sicily. An
impression of the wind field at one point in time is given in figure (6a). The six
locations where results of WAM are available are also presented in this figure.
wind [m/s]
20
20
'
'
... '
,, '
..........

,,
....
15
' ' '
' '
....
......
....
'
\
\
'
\.
.....
,......,10
E
''
......
....
\.
' 
15
......
10
>.
.t
5
t
'
'
...
,,,..
I
I
ti'
.....
,,.
....
'
'
5
0
0
0.5
1.5
x [mJ
2.5
3.5
x 10
36
6.2 Test
In the previous tests the mean frequency as calculated by the SWAN model
was higher than the results of other models (WA VEWATCH, SAIL). The first
test in this chapter (see below) shows that the mean frequency calculated by
SWAN is higher than the WAM results as well. Attention is therefore given to
the mean frequency.
To restrict the computational domain in spectral space, SWAN uses a fixed cutoff frequency (see Chapter two). The WAM model, being developed for oceanic
scales, uses a flexible cutoff frequency. This and the different numerical
propagation scheme are the main differences between the two models. To test
if the large difference in mean frequency is due to the value of the cutoff
frequency, a second calculation has been performed. In this calculation the cutoff frequency is scaled down.
6.3 Model
The WAM model, being developed for ocean scales, uses a spherical grid in
geographical space. The grid size for this test is half a degree in both longitude
and latitude. The calculations are performed for an area that measures from
longitude 6W to 36E and from latitude 30N to 46N. The SWAN model uses
a Cartesian grid (flat plane).The grid size in y direction (northsouth) in SWAN
is 55560 meter. The grid size in x direction (eastwest) is 44950 metre (half a
degree at 36 latitude). The Cartesian grid in the SWAN model is used as input
grid for the bottom file, the wind file and for the computations. The number of
grid points is equal for both models (85x33). A discretised spectrum of 26
frequencies in geometric progression has been used, from 0.25 rad/s to 2n
rad/s. In directional space a 10 resolution is used. The time step used in this
study is 20 minutes as was done in the W AM computations. The water depth is
equal in both WAM and SWAN.
The wind field used for both calculations is the limited area version (LAM,
resolution T333) of the ECMWF spectral global model. Output data from the
LAM model are available at 3hour intervals. Figure (6a) shows the wind field at
18 UTC 2 December. Refraction is deactivated in these tests because the
steep slopes of the Mediterranean sea caused instabilities (see below). In the
WAM calculations refraction was also not accounted for.
6.4 Results
The first calculation became unstable at 0 UTC 2 December at Mazara
(location 3), just south of Sicily. This was ascribed to the steep slopes of the
Mediterranean sea in combination with the large grid sizes. From Mazara these
instabilities propagated into the model. The instabilities resulted in extreme
wave heights and wave periods (see appendix ()). A more stable implicit
upwind difference scheme for refraction was then used. This difference scheme
is optional in SWAN. With this more robust scheme the calculations still
became unstable however. Refraction was then deactivated. The cause of the
instabilities is not further investigated in this study.
37
The results of the test without refraction are compared with observations and
the results of the W AM model. Observations are available at three locations
(2,4 and 6) and results of WAM are available at six locations south of Italy (see
figure 6a). In this paragraph some of the results are elucidated (figures 6b,6c
and 6d). In appendices D and E an overview of the results at all six locations is
given.
The significant wave height and the mean frequency at Catania as observed
and as calculated by both models is given in figures 6b and 6c. Where the
significant wave height is comparable for both models the mean wave
frequency differs significantly.
Figure 6b
Evolution of Hs at Catania,
calculations
and
both
measurements.
SWAN
WAM
Observations
Cf)
Time (hourday)
Figure 6c
:0:
"O
0
~
Cl.
<13
Q)
:E
3
2
SWAN
WAM
0
,
cb
c\J
cb
C\J
C\J
cb
C\J
c\J
C\J
cb
C')
Time (hourday)
38
Figure (6d) where the results of both models are compared at Malta, shows the
same tendency.
4.5
0.4
SWAN
WAM
3.5
0.3
'N
~
:[2.5
(/)
:r:
0.2
Q)
lL
1.5
0.1
SWAN
WAM
0.5
0
<b
0
~
c\i
C\J
cO
C\J
C\J
<b
c\i
C\J
cO
C')
C')
C')
<b
c\i
C')
cO
time (hourday)
C\J
<b
c\i
cO
C\J
<b
C\J
C\J
cO
c\i
C')
C')
C')
C')
<b
c\i
cO
Figure Gd
Evolution of Hs (left) and
Fmean (right) at Malta, for
SWAN and WAM.
The mean frequency as calculated by SWAN is much higher than the WAM
results. (up to 50% at Catania at 0 UTC 4 December (SWAN 0.254Hz and
WAM 0.169 Hz)) Regrettably there are no spectra available of the WAM model
at Catania. The onedimensional spectrum is available at Malta at 0 UTC 3
December. The spectra of WAM and SWAN are given in figure (6e). The mean
frequencies of both spectra are given as well. The mean frequency at Malta at
0 UTC 3 December differs strongly for SWAN and WAM. At the other locations
mean frequency is higher as well (see appendix E).
Figure Ge
Onedimensional
wave
spectrum at Malta at 0 UTC
3 December For SW AN and
WAM.
11
15
/\
I\
I I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
0.1
SN.AN
I
I
\
I
I
\
WAM
Fmean SN.AN = 0.169 Hz
\
I
Fmean WAM
\
=0.132 Hz
0.2
0.3
0.4
0.5
Frequency (.Hz)
39
The large difference in mean wave frequency between the two models is
remarkable. Both are third generation models, and the DIA used in SWAN to
calculate the nonlinear wavewave interactions originates from the W AM
model. The influence of the cutoff frequency, that is differently defined in both
models (see above) is now investigated. To do so, the cutoff frequency has
been adapted in one test run. Where in the standard run the frequency range
reaches up to 1 Hz, in this second run the cutoff frequency has been down
scaled to 0.5 Hz.
In W AM the cutoff frequency is defined as:
(61)
Here fmean is the mean frequency and fpM is the PiersonMoskowitz peak
frequency. For the WAM calculations fmax was set at 0.492 Hz. Using fmax, the
mean frequency and the local wind speed, the cutoff frequency for WAM is
calculated at Catania (location 4). In figure (6f) the value of the cutoff
frequency for the WAM calculations is given. It is remarked here that these
values are calculated from the results.
Figure 6f
1.2
Value
of
the
cutoff
frequency for SW AN and
WAM.
N'
b 0.8
 SWAN 1 Hz
  SWAN 0.5Hz
wAM (scaled)
()'
c
a:>
g.
0.6

1?
8 o.4
         
~
0.2
T"
<b
<b
c\J
cb
C')
C')
<b
time (hourday)
The results of SWAN with the down scaled cutoff frequency are shown for
Catania (location 4) in figure (6h). The WAM results of figure (6c) are presented
once more for comparison (frequency instead of period). In this figure it shows
that down scaling of the cutoff frequency decreases the mean frequency. The
influence of the down scaling of the cutoff frequency decreases with
decreasing mean frequency. From results not presented here it follows that the
significant wave height is hardly influenced by reduction of the cutoff
frequency.
40
Figure 6g
0.45
0.40
SWAN (0.5Hz)
  SWAN (1Hz)
wAM (scaled)
0.35
0.30
N'
~ 0.25
c:
Cl)
Q)
E 0.20
LL
0.15
0.10
0.05
0.00
Ol
')I
C')
Ol
N
c\J
Ol
')I
.,....
C')
C')
cb
Ji
.,....
di
.,....
cb
cf,
c\J
~
.,....
N
C')
C')
cb
Ji
'<!"
Time (hourday)
6.5 Analysis
The wave energy calculated is comparable for both models. The high mean
frequency as calculated by SWAN implies that in the SWAN results a larger
part of the computed wave energy is located in the high frequency range.
In SWAN the computational range stretches into the high frequency zone that
contains little wave energy. This is due to the fixed cutoff frequency. This part
of the high frequency range is now calculated where in WAM it consists of an
analytic tail. The high frequency part of the spectrum contains more energy
when it is calculated then when it consist of an analytic tail. From figure (6e) it
follows that the high frequency range in SW AN contains more energy than in
WAM because the significant wave height is equal for both models and the
mean frequency differs significantly. Besides the peak value of wave energy is
smaller in SWAN as well.
The shape of the high frequency range that is calculated results from the DIA
that is used to calculate the nonlinear wavewave interactions. The shape can
also be influenced by the limiter, because it is known from chapter four that the
influence of the limiter is largest in this range. Down sizing of the cutoff
frequency reduces the calculated mean frequency. Reducing the cutoff
frequency does not remedy the high mean frequency completely. From figure()
it follows that for comparable values of the cutoff frequency SW AN calculates
a higher mean frequency than WAM still. Besides the mean frequency
calculated with SWAN with a 0.5 Hz cutoff frequency, is higher than the
measurements still.
41
6.6 Conclusions
Figure 6h
Wave
field
at
the
Mediterranean sea at 18
UTC 2 December.
When using SWAN for large grid sizes and steep bottom slopes, integration in
directional space may become unstable. Although it is not further investigated
in this study this should be looked at and remedied before releasing the nonstationary option or a new stationary version.
The calculation of the significant wave height in the SW AN model shows good
comparison with the WAM model. The mean wave frequency as calculated by
the SWAN model is high. This is partly a result of the high cutoff frequency
used in these calculations. Because of the high value of the cutoff frequency
the spectrum is calculated up to a high frequency. In this calculated high
frequency interval more wave energy is contained than in WAM where the
wave energy in the same interval is not calculated but prescribed with use of an
analytic tail. The mean frequency as calculated by SWAN is not only high in
comparison with WAM but in comparison with measurements as well.
Although the difference in computational range can not explain the large
difference between both models, the value of the cutoff frequency influences
the calculated mean frequency. Down sizing of the cutoff frequency leads to a
reduction of the mean frequency.
Hs [m]
3.5
20
3
15
2.5
2
5
1.5
1
0.5
5
0
0
0.5
1.5
2
x [m]
2.5
3.5
x 10
42
7 Conclusions and
Recommendations
7 .1 Conclusions
From the tests done in this study several conclusions can be drawn concerning
the performance of the currently available nonstationary option (version
30.75). The conclusions concern different aspects of the SWAN model.
The implicit numerical scheme for propagation in geographical space that is
used in SW AN shows a good performance with respect to the propagation
velocity of wave energy. The numerical diffusion of the used propagation
scheme is very large however. The nonstationary option is, among other
things, being developed in order that the SWAN model can be used for large
areas. For such large areas the numerical diffusion of the implicit scheme as
used in version (30. 75) is too large.
When using SW AN for large grid sizes and steep bottom slopes, integration of
the action density balance equation can become unstable. It was found in the
last test of this study that when large grid sizes are used in combination with
steep slopes the integration in directional space can become unstable, even
when the optional upwind implicit scheme is used. In this study it has not been
investigated what was the cause of the instabilities.
The evolution of the wave spectrum is influenced by a limiter when time steps
are used that are large relative to the time scale of the evolution process. The
evolution time scales vary within the wave spectrum. The influence of the
limiter consists of restriction of growth of wave energy and restriction of shift of
wave energy within the spectrum. To calculate a wave spectrum explicitly
based on third generation formulations, the source terms are integrated in the
entire frequency range of interest. The integration of the nonlinear wavewave
interactions in the high frequency zone requires small time steps. The time
scales, at which nonlinear wavewave interactions respond to changing wave
spectra and vice versa are short in this range. When too large time steps are
used the integration becomes unstable. To ensure stable integration the limiter
is activated and thereby returns values that are not based on the action density
balance equation. In the tests of the Gorbush storm (a real case) the influence
of the limiter is negligible.
The possibility to use extra iterations per time step to increase the growth rate
when using large time steps leads to satisfactory results. The restriction of
energy growth by the limiter is counteracted.
The mean frequency as calculated by the SW AN model is high. It is higher than
the results of the other models that SW AN is compared with in this study
(WAVEWATCH, SAIL and WAM) and higher than measurements for the
Gorbush storm. The mean frequency of the calculated spectrum is influenced
by the cutoff frequency. Downscaling of the cutoff frequency leads to a
reduction of the mean frequency. In SWAN the cutoff frequency is set at a high
value to account for mixed sea states. Besides it is fixed. The difference in
computational range in spectral space cannot fully explain the difference
between the SW AN and the W AM wave model.
43
7.2 Recommendations
Although the influence of the high diffusion of the used numerical scheme has
not been investigated, implementation of a different numerical scheme is
recommended. Because the computational structure of SWAN is based on
upwind implicit schemes, such a scheme is required. To decrease numerical
diffusion a higher order implicit scheme can be implemented.
The cause of the instabilities that arose as a result of steep slopes in
combination with large grid sizes has to be investigated before releasing the
nonstationary option or a new stationary version.
The calculation of the spectrum in the high frequency range of the prognostic
part (approx. 0.31 Hz) is not satisfactory when calculated with economic time
steps (influence limiter, high mean frequency). To decrease the mean
frequency either a scaling cutoff frequency can be implemented (as is done in
WAM and W AVEW ATCH) or the calculation of the spectral shape in this range
must be improved. Because stable integration of the nonlinear source terms in
the high frequency range is impossible when using large time steps, calculation
in this range need be parameterised. This way the spectral shape will be based
on physical and not on numerical processes.
When more iterations are calculated per time step the restriction due to the
limiter is counteracted. The question when more iterations are needed and how
many iterations need to be calculated has not been answered in this study. A
criterion has to be introduced that will start the iteration process and that the
number of iterations will be based on.
44
8 References
Dell'Osso L.,Bertotti L., Cavaleri L., 1992: The Gorbush Storm in the
Mediterranean Sea: Atmospheric and Wave Simulation, Monthly Weather
Review, 120, No. 1, 7790
Hasselmann, K. et al., 1973: Measurements of windwave growth and swell
decay during the Joint North Sea Wave Project (JONSWAP), Dtsch. Hydro/gr.
Z. Supp/,, 12, AB
Kahma, K.K. and C.J. Calkoen, 1992: Reconciling discrepancies in the
observed growth of windgenerated waves, J. Phys. Ocreanogr., 22, 13891405
Komen G.J., Cavaleri, L., Donelan, M., Hasselmann, K., Hasselmann, S. and
P.A.E.M. Janssen, 1994: Dynamics and Modelling of Ocean Waves,
Cambridge University Press, 532 p.
Mei, C.C. 1983: The applied dynamics of ocean surface waves, Wiley, New
York, 740 p.
Phillips, O.M., 1957: On the generation of waves by turbulent wind, J. Fluid
Mech., 2, 417445
Phillips, O.M., 1977: The dynamics of the upper ocean, 2nd edition, Cambridge
University Press, 336 p.
Pierson, W.J. and L. Moskowitz, 1964: A proposed spectral form for fully
developed wind seas based on the similarity theory of S.A. Kitaigorodskii, J.
Geophys. Res., 69, No.24, 51815190
Ris, R.C., 1997: Spectral modelling of wind waves in coastal areas,
Communications on Hydraulic and Geotechnical engineering. Delft Univ. of
Tech., ISSN 01696548, Rep. No. 974
Ris, R.C., Booij, N., Holthuijsen, L.H., PadillaHernandez, R. and IJ.G.
Haagsma, 1998: SWAN user manual version 30.75, Delft University of
Technology, Department of Civil Engineering, The Netherlands
SWAMP Group, 1985: Ocean wave modelling, Plenum Press, New York and
London
Tolman, H.L., 1989: The numerical model WAVEWATCH: a third generation
model for the hindcasting of wind waves on tides on shelf seas.
Communications on Hydraulic and Geotechnical engineering. Delft Univ. of
Tech., ISSN 01696548, Rep. No. 892
45
46
9 Nomenclature
roman letters:
phase velocity
group velocity or propagation velocity of wave energy
relative velocity of propagation
absolute propagation velocity of wave energy in xdirection
absolute propagation velocity of wave energy in ydirection
propagation velocity of wave energy in 8direction
propagation velocity of wave energy in adirection
timeaveraged water depth
energy density
total wave energy per unit area
total wave energy nondimensionalised in terms of U
minimum and maximum discrete frequency, respectively
mean frequency
peak frequency
peak frequency nondimensionalised in terms of U
peak frequency nondimensionalised in terms of U10
acceleration of gravity
Hs
Hsmax
Cg
Cr
Cx
Cy
Cs
Ca
d
E
Etot
E*
fmin, fmax
fmean
fpeak
fp *
fp
Hix,iy
is,io
k
m
m0,m1
N
Nmax
n
r
s
Sds
S;n
Sn1
Tmo1
t
t*
U10
U*
x,y
Greek letters:
0PM
l::.x, b.y
1::.8,1::.a
47
Appendix A
II
Appendix B
Ill
Appendix C
IV
Appendix D
Appendix E
VI
Appendix F
VIII
Appendix A
The expressions for the propagation speeds in the action conservation
equation i.e., dx!dt, dy/dt, dcr/dt and d0/dt can be obtained from the kinematics
of a wave train, using the linear wave theory (e.g., Whitham , 1974; Mei, 1983):
dx 1 [ 1
2kd ] crk x U
Cx =ett=2 + sinh2kd k2+ x
dy 1 [ 1
2kd ]crky
Cy =ett=2 + sinh2kd k2+Uy
c
cr
ca= de
dt
 om
d
 =  + C 'V x y
dt at '
where
k ocr
k ak
C==
II
Appendix B
60
 2 hours
)(4 hours
<> 6 hours
Shours
r10 hours
50
40
30
20
10
0.1
0.3
0.2
60
Evolution of wave spectrum
using a seven hundred and
fifty second time step
 2 hours
~4 hours
0 6 hours
~a hours
+10 hours
50
40
30
20
10
0.1
0.2
0.3
Ill
Appendix C

30
  
25

        
SWAN
 WAM
20
'E
;; 15
I
10
5
0
0
20
40
60
80
100
120
30
25
SWAN
20
'.:
0 15
E
t
10
20
40
60
80
100
120
Evolution of significant wave height (top) for SWAN and WAM at Mazara
(Location 3). At Time=66 hour (0 UTC 2 December) the SWAN calculation
becomes unstable. The mean wave period is shown in the bottom figure.
IV
Appendix D
Comparison of SW AN with
WAM and measurements
for significant wave height
at Crotone.
Comparison of SW AN with
WAM and measurements
for mean wave frequency at
Crotone.
SWAN
WAM
Observations
7
6
::
15
~
Q.
c:
C1l
Ql
SWAN
WAM
Observations
0
..
..
cb
..
c{i
..
00
cb
00
C')
Time (hourday)
Appendix E
4.5
4.5
4
3.5
SWAN
WAM
I
:E.
SWAN
WAM
3.5
3
2.5
(/)
:i:
2.5
2
1.5
1.5
0.5
0.5
1
O+~<
,...
c.b
00
3.5

4.5
SWAN
WAM
3.5
~
:E.
2.5
:E.
0.5
0.5
0 +~<
00
time (hourday)
4.5
SWAN
WAM
SWAN
WAM
4
3.5
3
3
(/)
ID
o+~;
N
N
N
N
M
M
~
M
~
~ 00 6 ,...
N ,...
00 6 ID N 00 6
4.5
:i:
SWAN
WAM
time (hourday)
2
1.5
3.5
2.5
1.5
time (hourday)
time (hourday)
4.5
W N 00
2.5
(/)
:i:
2.5
2
1.5
1.5
0.5
0.5
1
O+~'
~
~
~
N
N
N
N
~
M
M M
~
N ~ 6 ~ N 00 6
N 00 6
time (hourday)
,...
c.b
N
,...
00 6
N 00
W ,...
N 00
time (hourday)
Significant wave height for both SWAN and WAM calculations at six locations
south of Italy. Locations one to six from top left to bottom right.
Location 1 is Sirte at 17E and 34.5N. Location 2 is Malta at 14.5E and 36N.
Location 3 is Mazara at 12.5E and 37.5N. Location 4 is Catania at 15.5E and
37.5N.
Location 5 is Jonio at 18.5E and 37.5N. Location 6 is Crotone at 17.5E and
39N.
VI
0.4
0.4
0.3
0.3
N'
N'
0.2
0.2
CD
CD
LL
LL
0.1
0.1
SWAN
SWAN
WAM
WAM
C\J
(".)
(".)
(".)
(".)
.q
cb
<b
cb
C\J
C\J
cb
C\J
0.4
0.3
0.3
N'
N'
0.2
0.2
CD
CD
LL
LL
0.1
0.1
SWAN
WAM
~
ID
N 00
ID
N 00
ID
N 00
SWAN
WAM
~
C\J
ID N 00 6
C\J
C\J
C\J
(".)
N 00 6
(".)
(".)
(".)
0.4
0.4
0.3
0.3
N'
N'
0.2
0.2
CD
CD
LL
LL
0.1
SWAN
0.1
SWAN
WAM
WAM
.q
W N 00 6
Mean wave frequency for both SWAN and WAM calculations at six locations
south of Italy. Locations one to six from top left to bottom right.
Location 1 is Sirte at 17E and 34.5N. Location 2 is Malta at 14.5E and 36N.
Location 3 is Mazara at 12.5E and 37.5N. Location 4 is Catania at 15.5E and
37.5N.
Location 5 is Jonio at 18.5E and 37.5N. Location 6 is Crotone at 17.5E and
39N.
VII
Appendix F
89120300 AT CATANIA
37.50N 15.50E
a>..~30"W.._~
.....lor:.
..
...
........
.... ! .. :1::...
..
t .. ...
.. .
. ... .
......
....
.................: .
.. ........
. . . :~.~~~\:.: . .
.........
SO"!
./
:~.
. ...
. ......t1!9..~~'1.Q
i ...........:........
Gii
:~. f~
. .. ... .
j ....
.. .. . : ......::....... . .:
.:.... .. .....
: ..... ....... ....
.,
t ...... . . . .
~ .
..
..f1~3.7.
_:.
...
. ...
..
\
. .
. v.o
. :s
'50'[
11~
NORTH
.. . .
. .
WEST :
:.:.:
EAST
0.1 .
0.2
. .
0.3
SOUTH
VIII
b)
89120300 AT MALTA
...
.
..
..
. .... ....
30"W
309!
..........
~1 .......... .
~
..:i..
IJ
36.00N 14.SOE
0
."......
..
......
:;:
.,
'
...
".,
'.,~
: t
~
.........
;
'....
"11 :
..
"}i$4. t':/
. .
...
.. ~!P<).6
NORTH
..
EAST
WEST '
" ~
.
.
. .
. 0.2
...
: 0.3
SOUTH
IX