Professional Documents
Culture Documents
Two alternativesseemto be available:(1) extendthe above efficientto use the Eulerian approach,in which the wave evo-
(phaseaveraged)approachof the energyor action balance lution is formulatedon a grid. This is essentiallythe technique
equationby addingthe requiredphysicalprocesses usingother that has been used for deep-oceanor shelf-seawave models
numericaltechniquesor (2) exploitthe alternativeapproachof suchas the Wave Model (WAM) [WAMDI Group,1988].All
phase-resolving modelsbasedon massand momentumbalance relevant processesare then readily included as sourcesand
equations.Such phase-resolving modelsare usuallybased on sinks in the basic equation. Drawbacksof this approachin
coastal waters are the absence of diffraction and the use of
Hamiltonian equations [e.g., Miles, 1981; Radder, 1992],
Boussinesq equations[e.g.,Peregrine,1966;Freilichand Guza, linear wave theory for wave propagation.The first drawback
1984;Madsenand SOrensen, 1992],or on the mild-slopeequa- impliesthat the area of interestshouldbe a few wave lengths
tion (e.g.,Berkhoff[1972]or its parabolicversion,e.g.,Radder awayfrom obstacleswith (near)verticalwalls (dependingon
[1979] and Kirby [1986]). For a recentreview,seeDingemans the short-crestedness of the waves[Booijet al., 1992]).This is
[1997]. These models reconstructthe sea surfaceelevation in often the casealongfairly open coastswith an occasionalsmall
island or breakwater in the far field, as opposedto confined
spaceand time while accountingfor sucheffectsas refraction,
situationssuchas in a harbor or directlybehind breakwaters.
diffraction, and, in some models, also triad and quadruplet
The seconddrawbackimpliesthat, for the model to be appli-
wave-waveinteractions.Dissipationprocessessuchas bottom
cable,nonlinearcorrectionsto linear wavepropagationshould
either be sufficientlywell representedby triad and quadruplet
•Nowat WL/Delft Hydraulics,
Delft, Netherlands.
wave-waveinteractionsor that they be dominatedby the gen-
Copyright1999 by the American GeophysicalUnion. erationor dissipationof the waves(theseprocesses can all be
Paper number 98JC02622. includedin a phase-averaged model). The latter is often the
0148-0227/99/98JC-02622509.00 case, even in the absenceof wind, when waves break in shallow
7649
7650 BOOIJ ET AL.: A THIRD-GENERATION COASTAL WAVE MODEL, 1
water. Since these limitations seem to be acceptablein many prototypicalthird-generationmodel that was developedon the
real field situationson a scaleof 20-30 km with a water depth basis of these formulations is the WAM model of the WAMDI
lessthan 20-30 m, the model to be developedin the present Group [1988] (also see Kamen et al. [1994]). Other third-
studyis a phase-averaged, Eulerian model (seeBattjes[1994] generationmodelsare the WAVEWATCH model of Talinart
for a reviewof similar considerations). [1991], the model of Li and Maa [1992], the Program for
Eulerian, phase-averagedmodelshave been used for waves Hindcastingof Wavesin Deep, Intermediate and ShallowWa-
in the deep oceanand in water of intermediatedepthsincethe ter (PHIDIAS) model of Van Vledderet al. [1994], and the
pioneeringwork of Gelci et al. [1956], and extensivework has Telemac Based Operational Model AddressingWave Action
been done with these models [e.g., Cardone et al., 1976; Computation(TOMAWAC) modelof Benoitet al. [1996].For
WAMDI Group, 1988;Kamen et al., 1994]. The propagationof shallowwater the formulations for the deepwater processes
wavesin thesemodelsis readily extendedto finite-depthwater need to be adapted and extended.This has been achieved,to
by introducing a depth-dependentpropagation speed and a someextent,in the abovethird-generationmodelswith (1) the
Eulerian representationof refraction[Piest,1965;Hasselmann useof the shallow-waterphasespeedin the expressions of wind
et al., 1973]. However, these modelsdo not computeall rele- input, (2) a depth-dependent scalingof the quadrupletwave-
vant physicalprocessesfor finite-depthwater, in particularly, waveinteractions,(3) a reformulationof the whitecappingin
not depth-inducedwave breaking and triad wave-waveinter- termsof wave numberrather than frequency,and (4) adding
actions,which, for coastal regions, can be rather important. bottom dissipation.All this will be also used in the SWAN
Moreover, all these models are based on explicit numerical model and supplementedwith formulationsfor (5) depth-
schemesfor propagationwhich are subject to the Courant inducedwave breakingand (6) triad wave-waveinteractions
criterion for numerical stability.For oceanicconditionsthis is that havebeen developedspecificallyfor SWAN (and similar
not a problem, but for coastalapplicationsit is unacceptable, models)by Eldeberkyand Battjes[1995] andEldeberky[1996].
as it leadsto very small time stepsin the computations(the As for implicit propagation schemes,one type of implicit
spatialstepin oceanmodelsis of the order of 100 km, and the propagationschemeis piecewisewave ray propagationfrom
correspondingtime step is typically30 min; in coastalmodels, one mesh of a computationalgrid to another [e.g., Young,
in 10-m water depth, the numbersare typically100 m and 10 s, 1988; Yamaguchi,1990;Benoit et al., 1996]. However, spatial
respectively).The situation is aggravatedby the fact that variationsin the driving fields (wind, bottom, and currents)
coastalmodelstypicallycontain 2-5 times as many geographic over a distanceof caAt are ignoredduringthe integration
grid points as ocean models. The application of explicit alongthesecharacteristics
(cgistheenergypropagationspeed,
schemesin coastalsituationswould therefore require about 2 and At is the integrationtime step). Sincethe grid is usually
orders of magnitudemore computationaleffort than in deep chosen
to properlyresolve
thesevariations,
cgAtshouldstillbe
water. The solutionto this problem is to useimplicit numerical smallerthan the meshkize.This is equivalentto the Courant
schemeswhich are unconditionally stable and therefore not criterion for explicit schemes(now for reasonsof accuracy
subjectto the Courant criterium.These two problems,of add- rather than stability), and this approachis therefore not a
ing depth-inducedbreaking and triad wave-waveinteractions reasonable alternative. The alternative that has been devel-
and of usingimplicit numericalschemes,will be addressedin oped for SWAN is an implicit propagationschemebased on
detail for the developmentof the model of the present study finite differences.
(SimulatingWavesNearshore(SWAN)). To validate both the propagationcharacteristicsand the
State-of-the-art formulations of the processesof wave gen- physicalprocessesof generation and dissipationin SWAN,
eration, dissipation, and wave-wave interactions in phase- computationalresultsare comparedwith (1) analyticalsolu-
averaged models are presently third generation. In first- tionsin academiccases,(2) laboratoryobservations, and (3)
generation models these physicalprocessesare not properly (generalized)field observations.The verificationof SWAN
represented.Generation is simulatedwith simple empirical with real field casesis describedin a sequelpaper [Riset al.,
expressions, and dissipation(whitecapping)is simulatedwith this issue].The paper is organizedas follows.The basicfor-
an assumeduniversalupper limit of the spectraldensities.The mulationsof the model are given in section2. The numerical
absenceof quadrupletwave-waveinteractionsis compensated implementationand the relatedvalidationsof the propagation
by enhancingthe wave growth [e.g., Ewing, 1971]. Second- schemesand of the sourceterm integrationare presentedin
generationmodelstry to remedythis for the localwind seaby section3. A discussion with conclusionsis given in section4.
parameterizingtheseinteractions[e.g.,Young,1988] or by us-
ing a sea-state-andwind-dependentupper limit of the spectral 2. Model Formulation
densities[e.g.,Halthuijsenand De Boer, 1988] or by reducing
2.1. Introduction
the wavedescriptionto a few integralspectralparameters[e.g.,
Hasselmannet al., 1976]. Such models are usually supple- In SWAN the waves are described with the two-dimensional
mented with freely propagatingswell. In a third-generation wave action densityspectrum,even when nonlinearphenom-
model all relevantprocessesare representedexplicitlywithout ena dominate(e.g., in the surf zone). The rationalefor using
a priori restrictionson the evolutionof the spectrum.For the the spectrumin suchhighly nonlinear conditionsis that, even
developmentof thesemodels,Hasselmannet al. [1985]formu- in suchconditions,it seemspossibleto predictwith reasonable
lated the discreteinteractionapproximation(DIA) of the qua- accuracythisspectraldistributionof the second-ordermoment
druplet wave-waveinteractions.This first-principleapproach of the waves(althoughit maynot be sufficientto fully describe
of the quadrupletwave-waveinteractionspermitsthe genera- thewavesstatistically). The actiondensityspectrum N(rr, 0 ) is
tion of wavesby wind to be formulated on the basisof theo- considered rather than the energydensityspectrumE (rr, 0)
reticalwork of Miles [1957] and the empiricalwork of Snyderet since in the presenceof ambient currents, action density is
al. [1981]. For deep water the problemwas finally closedwith conserved whereasenergydensityis not [e.g.,Whitham,1974].
the formulation of whitecappingby Kamen et al. [1984]. The The independentvariablesare the relative frequencyrr (as
BOOIJ ET AL.' A THIRD-GENERATION COASTAL WAVE MODEL, 1 7651
Options indicatedas Wave Model (WAM) 3 and WAM 4 are the formulationsalso used in WAM cycle 3 and WAM cycle 4 models,
respectively.
Optionsthat are availablein SWAN are indicatedby x. Preferredoptionsin SWAN (default) are indicatedby xx.
observedin a frame of reference moving with the current and direction.The expressionfor the term.4 is due to Cavaleri
velocity)and the wavedirection0 (the directionnormalto the and Malanotte-Rizzoli[1981] with a filter to avoid growth at
wavecrestof eachspectralcomponent).The actiondensityis frequencieslower than the Piersonand Moskowitz[1964] fre-
equal to the energydensitydividedby this relative frequency, quency[Tolman, 1992a].Two optionalexpressions for the co-
N(•, 0) = •(•, 0)/•. efficientB are usedin SWAN. The first is taken from an early
versionof the WAM model (knownasWAM cycle3 [W.4MDI
2.2. Action Balance Equation Group,1988]).It is dueto Snyderet al. [1981],rescaledin terms
The evolution of the wave spectrumis describedby the of friction velocityU, by Komenet al. [1984]. The drag coef-
spectralactionbalanceequation,which, for Cartesiancoordi- ficientto relate U, to the drivingwind speedat 10 m elevation
nates,is [e.g.,Hasselmanne! al., 1973] U•o is taken from Wu [1982]. The secondexpressionfor B in
SWAN is taken from the most recent version of the WAM
O O O O O s
o-7N + •xx CxN + •yy CyN + •-•coN + •-•coN =• (1) model (knownasWAM cycle4 [Komenet al., 1994]).It is due
to Janssen[1991a],and it accountsexplicitlyfor the interaction
The first term on the left-handsideof (1) representsthe local between the wind and the wavesby consideringatmospheric
rate of changeof actiondensityin time, the.secondand third boundary layer effects and the roughnesslength of the sea
term representpropagationof action in geographicalspace surface.The corresponding setof equationsis solved(as in the
(withpropagation velocitiescx andCyinx andy space, respec- WAM model) with the iterative procedureof Mastenbroeket
tively). The fourth term representsshiftingof the relative fre- al. [1993].
quencydue to variationsin depthsand currents(with propa-
2.4. Dissipation
gationvelocityc• in crspace).The fifth term representsdepth-
induced and current-inducedrefraction (with propagation The dissipationterm of wave energyis representedby the
velocityco in 0 space).The expressions for thesepropagation summation of three different contributions: whitecapping
speedsare takenfrom linearwavetheory[e.g.,Whirham,1974; Sas,w(O-, 0), bottomfrictionSas,t,(o', 0), anddepth-induced
Dingemans,1997].The term $ [= S(cr, 0)] at the right-hand breakingSds,t,r(o-, 0). Whitecapping isprimarilycontrolledby
sideof the actionbalanceequationis the sourceterm in terms the steepnessof the waves.In SWAN, as in other presently
of energydensity,representingthe effectsof generation,dis- operatingthird-generationwavemodels,the whitecappingfor-
sipation,and nonlinearwave-waveinteractions.A brief sum- mulation is based on the pulse-basedmodel of Hasselmann
mary of the formulationsthat are used for the various pro- [1974], as adaptedby the WAMDI Group [1988]:
cessesin SWAN is given next, with an overviewin Table 1. k
Details of the relatively new formulationsfor depth-induced Sds,w(O',O)•---I-'•'•E(o-,0) (3)
breaking and triad wave-waveinteractionsare given in the
appendix.The more establishedformulationsfor the other where F is a steepnessdependentcoefficient,k is wave num-
processesare well describedin the followi.ngreferences. ber, and • and k denote a mean frequencyand a mean wave
number, respectively[cf. WAMDI Group, 1988]. Komen et al.
2.3. Wind Input [1984]estimatedthevalueof F by closingthe energybalanceof
Transferof wind energyto the wavesis describedin SWAN the wavesin fully developedconditions.This impliesthat this
with the resonancemechanismof Phillips[1957] and the feed- value dependson the wind input formulation that is used.
back mechanismof Miles [1957]. The correspondingsource Since two expressionsare used for the wind input in SWAN,
term for thesemechanismsis commonlydescribedas the sum two values for F are also used. The first is due to Komen et al.
of linear and exponentialgrowth: [1984](as in cycle3 of the WAM model). It is usedin SWAN
when the wind input coefficientB of Komen et al. [1984] is
Sin(O',O)=A -3-BE(o', O) (2) used.The secondexpressionis an adaptationof this expression
in whichA and B dependon wavefrequencyand directionand basedonJanssen[1991a](as in cycle4 of the WAM model [see
wind speed and direction. The effects of currents are ac- Janssen,1991b;Gantheret al., 1992]).It is usedwhenthe wind
countedfor in SWAN by usingthe apparentlocal wind speed input term B of Janssen[1991a] is used. Youngand Banner
7652 BOOIJ ET AL.: A THIRD-GENERATION COASTAL WAVE MODEL, 1
and Beji, 1992;Arcilla et al., 1994]. A discretetriad approxi- versionwas used in the second-generationHindcasting Shal-
mation (DTA) for collinearwaveswas subsequently obtained low Water Waves (HISWA) wave model of Holthuijsenet al.
by consideringonly the dominant self-selfinteractions.Their [1989]. They have been chosen on the basis of robustness,
formulationhas been verifiedwith flume observationsof long- accuracy,and economy.
crested,random waves breaking over a submergedbar [Beji Sincethe nature of the actionbalanceequationis suchthat
andBattjes,1993]and over a barredbeach[Arcillaet al., 1994]. the state in a grid point is determined by the state in the
The formulation appearedto be fairly successful in describing upwavegrid points, the most robustschemewould be an im-
the essentialfeaturesof the energytransfer from the primary plicit upwindscheme(in both geographicand spectralspace).
peak of the spectrumto the superharmonics. A slightlydiffer- The adjective"implicit" is usedhere to indicatethat all deriv-
ent version(the lumpedtriad approximation (LTA)) waslater ativesof actiondensity(in t, x, or y) are formulatedat one
derived by Eldeberky[1996]. Details on this LTA as used in computational
level,it, ix, or iy, exceptthe derivativein the
SWAN are providedin the appendix. integration dimensionfor which also the previousor upwave
level is used (time and x or y, dependingon the directionof
3. Model Implementation and Validation propagation).For sucha schemethe valuesof the time and
3.1. Introduction
space steps At, Ax, and Ay are mutually independent.An
The integration of the action balance equation has been implicit schemeis also economicalin the sensethat such a
implementedin SWAN with finite difference schemesin all scheme is unconditionallystable. It permits relatively large
five dimensions(time, geographicspace,and spectralspace). time stepsin the computations(much larger than for explicit
These are first describedand validatedfor the propagationof schemesin shallowwater). Severalyearsof experiencein using
the waveswithout sourceterms. Then the implementationof the HISWA model have shownthat for coastalregionsa first-
the source terms is described and validated. order upwind differenceschemein geographicspaceis usually
In SWAN, time is discretizedwith a simple constanttime accurateenough.However, this experience,togetherwith test
step At for the simultaneousintegrationof the propagation computationswith SWAN, has also shown that in spectral
and the sourceterms.This is different from the time discreti-. space a higher accuracythan that of a first-order upwind
zation in the WAM model or the WAVEWATCH model, schemeis required. This has been achievedby supplementing
where the time stepfor propagationis different from the time the schemewith a second-ordercentralapproximation(more
stepfor the sourceterms.Geographicspaceis discretizedwith economicthan a second-orderupwind scheme).For SWAN
a rectangulargrid with constantresolutionsAx and Ay in the therefore implicit upwind schemesin both geographic and
x and y direction, respectively.The spectrumin SWAN is spectralspacehave been chosen,supplementedwith an im-
discretized with a constant directional resolution A0 and a plicit central approximationin spectralspace.
constantrelative frequencyresolutionAo-/o-(logarithmicfre- The fact that in geographicspacethe state in a grid point is
quencydistribution). For reasonsof economyan optionis avail- determinedby the statein the upwavegrid points(as defined
ableto computeonlywavecomponents travelingin a predefined by the directionof propagation)permits a decompositionof
directionalsector(0min ( 0 ( 0max, e.g.,thosecomponents that the spectralspaceinto four quadrants(eight octantswould be
travelshoreward withina limiteddirectionalsector).The discrete an alternative).In eachof the quadrantsthe computations can
frequenciesare defined between a fixed low-frequencycutoff be carried out independentlyfrom the other quadrants,except
and a fixed high-frequencycutoff (the prognosticpart of the for the interactions between them due to refraction and non-
spectrum).For thesefrequenciesthe spectraldensityis uncon- linear sourceterms (correspondingto boundaryconditions
strained.Belowthe low-frequency cutoff(typically,fmin = 0.04 betweenthe quadrants).The wavecomponentsin SWAN are
Hz for field conditions)the spectraldensitiesare assumedto correspondingly propagatedin geographicspacewith the first-
be zero. Above the high-frequency cutoff (typically,1 Hz for order upwind schemein a sequenceof four forward marching
fieldconditions)a diagnosticf-mtail is added(thistail is used sweeps(one per quadrant). To properly account for the
to computenonlinearwave-waveinteractionsat high frequen- boundaryconditionsbetweenthe four quadrants,the compu-
ciesandto computeintegralwaveparameters).The reasonfor tations are carried out iteratively at each time step. The inte-
usinga fixed high-frequencycutoff rather than a dynamiccut- gration in time is a simple backwardfinite difference, so that
off frequencythat dependson the wind speedor on the mean the discretizationof the actionbalanceequationis (for positive
frequency,as in the WAM model, is that in coastalregions,
propagationspeeds,includingthe computationof the source
mixedseastateswith rather different characteristicfrequencies
termsbut ignoringtheir discretization)
may occur. For instance, a local wind may generate a very
youngseabehindan island,totallyunrelatedto (but superim- N ..... N 't-• [cxN],x- [cxN],•_•
+
posedon) a simultaneously occurringswell.In suchcasesa At
tx,ty,t
•,t o
Ax
iy,i•,to
dynamiccutoff frequencymay be too low to properlyaccount
for the locally generated sea state. On the basis of physical
arguments,the valueof m (the powerin the aboveexpression + ZXy ix
....
io
of the spectraltail) shouldbe between4 and 5 [e.g.,Phillips,
(1 - v)[Co•N],•+•
+ 2V[Co•N],•-(1 + •)[Co•N],•-•
1985].In SWAN, m - 4 if the wind inputformulationof Komen +
2Ao'
et al. [1984] is used (WAM cycle3) and m = 5 if the wind tx,ty,tO
wherei t is the time levelindex;ix, iy, irr, and io are grid needstherefore to be solvediteratively until some break-off
counters;and At, zLr, Ay, Ao-, and A0 are the incrementsin criteria are met.
time, geographicspace,and spectralspace,respectively.The In the field casesof the presentstudy(thispaperandRis et
iterative nature of the computationis indicatedwith the iter- al., this issue],the break-offcriteria are as follows'in more
ation index n (the iteration index for the sourcetermsn* is than 97% of the submergedgrid pointsthe changein signifi-
equalto n or n- 1, dependingon the sourceterm; seebelow). cantwaveheight [Hs = 4X/-•o,wherem,•= ,• o-'•E(o-) do-]
Becauseof theseiterations,the schemeis also approximately from one iteration to the next is less than 3% or 0.03 m and,
implicitfor the sourceterms (seesection3.3.1). For negative also,the changein the mean relativewave period (Tmo1 --
propagationspeeds,appropriateplus and minus signsare re- 2rrmo/m•) is lessthan 3% or 0.3 s. For laboratorycases,
quired in (6). stricter criteria are used.
The coefficientsv and r• determinethe degreeto which the The numberof iterationsfor caseswith wave generationby
schemein spectralspaceis upwind or central.They thus con- wind is typically5-15 in the presentstudy(stationarycasesin
trol the numericaldiffusionin frequencyanddirectionalspace, whichthe iterationsstartfrom totallycalm sea).For the aca-
respectively.A value of v = 0 or r• = 0 corresponds to central demic cases,stricter criteria were used to verify the conver-
schemeswhich have the largestaccuracy(numericaldiffu- genceto the proper analyticalsolutions(e.g., the break-off
sion •- 0). A value of v = 1 or r• = 1 corresponds to upwind criterion for the significantwave height was 0.1%) and the
schemes, which are somewhat more diffusive and therefore number of iterationswas typically30-40. For all thesecom-
lessaccuratebut more robust.In the presentstudyall compu- putationsthe convergence wasmonotonic,althoughextending
tations are carried out with v - r• = 1/2. If large gradientsof the computationsbeyond the break-off criteria showedthat
the actiondensityin frequencyspaceor directionalspaceare occasionally, the solutioncontinuedwith fluctuationsthat were
present,numericaloscillations
can arise(especially
with the somewhat smaller than the break-off criterion. To reduce the
centraldifferenceschemes),resultingin negativevaluesof the numberof iterationsin stationarymodewith wind generation,
actiondensity.In eachsweepsuchnegativevaluesare removed thecomputations thatareshownin thispaperandin the sequel
from the two-dimensionalspectrumby setting these values paper [Riset al., thisissue]startwith a reasonablefirstguessof
equal to zero and rescalingthe remainingpositivevaluessuch the wavefield (a "quickstart"basedon the second-generation
that the frequency-integrated actiondensityper spectraldirec- sourceterms of Holthuijsenand De Boer [1988], adaptedfor
tion is conserved.The depth derivativesand currentderivatives shallowwater). It reducesthe numberof iterationsmentioned
in the expressions of c• and c 0 are calculatedwith a first-order abovetypically
bya factorof 2 (i.e.,threeto seveniterations).
upwind scheme. In nonstationary mode a reasonablefirst guessper time stepis
For stationaryconditions,SWAN can be run in stationary availablefrom the previoustime stepand the numberof iter-
mode. Time is then removedas a variable,but the integration ationsis expectedto be small(lessthan 4). If no iterationsare
(in geographicspace)is still carriedout iteratively.The prop- usedin nonstationarymode (as in mostother phase-avetaged
agation schemeis still implicit, as the derivativesof action wave models),the computationsof propagationare still im-
density(inx ory) at thecomputational
level(ix or iy, respec- plicit and thereforestill unconditionallystable.
tively) are formulated at that level, exceptin the integration The boundaryconditionsin SWAN, both in geographic
dimension(x ory, dependingon the directionof propagation), spaceand spectralspace,are fully absorbingfor waveenergy
where the upwavelevel is alsoused.The valuesof Ax and Ay that is leavingthe computationaldomain or crossinga coast-
are therefore still mutually independent. line.The incomingwaveenergyalongopengeographic bound-
To explainthe abovenumericalsolutiontechniquein terms aries needsto be prescribedby the user. For coastalregions
of matrix solutions,first ignore the decompositionin quad- suchincomingenergyis usuallyprovidedonlyalongthe deep-
rants. The propagationof the wavesin both geographicand water boundaryand not along the lateral geographicbound-
spectralspacewould then be describedwith one large basic aries(that is, the spectraldensitiesare assumedto be zero).
matrix that canbe solvedin severalways.Removingrefraction, This impliesthat sucherroneouslateral boundaryconditions
frequencyshifting,and nonlinear sourceterms from this basic are propagatedinto the computationalarea. The affectedar-
matrixpermitsa matrixsolutionwith a Gauss-Seidel technique easare typicallytriangularregionswith the apexat the corners
[e.g.,Goluband vanLoan, 1986]in whichthe matrixis decom- betweenthe deepwaterboundaryand the lateral boundaries,
posedin four sections(the abovefour directionalquadrants), spreadingtowardshoreat an angleof 30øto 45ø (for wind sea
whichare eachsolvedin one step(superconvergence). Restor- conditions)on eithersideof the deepwatermeanWavedirec-
ing refractionand frequencyshiftingto the matrixrequiresthe tion(lessforswellconditiohs;
thisangleisessentially
halfthe
solutionof a submatrixfor each geographicgrid point. If no totalwidth of the directionaldistributionof the incomingwave
currentsare presentand the depth is stationary,this is readily spectrum).For this reason,the lateral boundariesshouldbe
done with a Thomasalgorithm[e.g.,Abbottand Basco,1989] sufficientlyfar away from the area of interest to avoid the
(ca = 0 and the submatrixis a simpletridiagonalmatrix). If propagationof this error into the area of interest.
currentsare presentor the depthis not stationary,the subma- 3.2.2. Tests of propagation. Ideally, the propagation
trix is a band matrix. It is solvedwith an iterative Incomplete scheme in SWAN should be free of numerical diffusion. How-
Lower and Upper Triangular Matrix Decomposition-Bi- ever,for longpropagationdistances (oceanicscales)the diffu-
Conjugate Gradient Stabilized (ILU-BiCGSTAB) method sion for the first-order implicit schemesof SWAN is fairly
[Vuik, 1993; Van der Vorst, 1992]. Restoringrefraction and large;but, sinceSWAN is primarilya coastalwavemodelwith
frequencyshiftingalso introducescoefficientsin each matrix shortpropagationdistances,the integrativeeffect of the diffu-
section(directionalquadrant)that causedependencybetween sionseemsto be acceptable.A full analysisof the diffusionin
the matrix sections.The samehappenswhen nonlinearsource both geographicand spectralspaceis fairly complicatedand
terms are added to the matrix. The basic matrix as a whole beyondthe scopeof this study(the diffusioncoefficientsin-
BOOIJ ET AL.: A THIRD-GENERATION COASTAL WAVE MODEL, 1 7655
Ys(m) O (o)
+ + SWAN 30
- - analytical
1.25
25
ß• 20
(o)
-30
0.75
-35
0.50 , , -40
where the wave direction is calculated with Snell's law. This other spectralcomponents.Sincetheseare only knownat the
analyticalsolutionis shownin Figure 3, wherethe resultsof the previousiteration leveln - 1, the coefficient4>is determinedat
SWAN computationsare also shownin termsof (significant) that iterationlevel:4>= qbn-•
waveheight(sameresolutionsand long crestedness as in the For positivesourceterms (wind input and the triad and
abovepropagationtest). The agreementwith linear theory is quadrupletwave-waveinteractionsif positive)the integration
again good.The errors are lessthan 0.1% in significantwave is generallymore stableif an explicitformulationis used(i.e.,
heightand lessthan 0.25ø in direction(for depthslarger than thesource
termdepends
onE n- • andnotonE •) ratherthan
0.05 m). an implicitformulation(i.e., the sourceterm dependsalsoon
E•). The explicitformulationfor thesesourcetermsin SWAN
3.3. Generation, Wave-Wave Interactions, and Dissipation is therefore
3.3.1. Implementation of source terms. The numerical
estimationsof the sourceterms in SWAN are essentiallyim- Sn• qbn-lE
•-1 (13)
plicit. This is achievedwith explicitor implicit approximations For reasonsof economythis explicitapproximationis alsoused
of the sourceterms,which in the limit of a large number of for the formulationof the quadrupletwave-waveinteractionsif
iterations(seesection3.2.1) are alwaysimplicit.In actualcom- negative.This is consideredreasonablesinceTolman [1992a]
putations,final convergenceis obviouslynever achievedand has shownthat usingan explicitformulationin combination
the estimationsof the source terms are therefore, strictly with a limiter (see below) givesresultssimilar to those of a
speaking,only approximatelyimplicit. In the following the more expensiveimplicit scheme(this implicit formulationis
adjectives"explicit"and "implicit"refer to the approximations also optionallyavailablein SWAN; in the WAM model it is
of the source terms within each iteration. indicatedas the semi-implicitscheme[W.4MDI Group, 1988;
The linear growth term .4 is independentof integralwave Komenet al., 1994]).
parametersand the energydensityand canthereforebe readily For negativesourcetermsthe integrationis generallymore
computed.All other sourceterms depend on energydensity, stableif. an implicit schemeis used.The stronglynonlinear,
and they canbe describedasa (quasi-)linearterm: S = qbE,in negativesourceterm of depth-inducedwave breakingat iter-
which 4>is a coefficientthat dependson (integral)wave pa- ationleveln is accordingly estimatedwith a linear approxima-
rameters(e.g., Etot, 0', k, o-, k, etc.) and action densitiesof tion:
Hs(m) + + SWAN
analytical
] 0(ø) 30
3
20
0 , , , -10
0 1000 2000 3000 x (m) 0 1000 2000 3000 x (m)
103
K&C
] SWAN
(WAM
3+)
..........
Wilson
•_;
.;;:.;;:';Z""'•..••
,r,-1I 'x"•',
'Z.....
10 1 . +++P&M
"/111"
'"': i•o "l -t,• • ,.,..,•
10 -1
ß .,
lO 3
lO -1
10 1 -
10-2_
10-1_
showsthat the agreementis good between the SWAN results increasesonly 5% when the WAM cycle 3 formulationsare
on the one hand and the expressions of Kahma and Calkoen used.This is consistentwith the findingsof Tolman[1992a].To
[1992]and the limit valuesof Piersonand Moskowitz[1964]on achieve a better agreement with the Piersonand Moskowitz
theother.However, forveryshortfetches (X* < 104,say)the [1964] limit values,the coefficientCdissin the expressionfor
model with linear growthadded overestimatesthe total energy whitecappingof Janssen[1991b] (Cdiss-- 4.5, Janssen's nota-
and, correspondingly, underestimatesthe peak frequencyas tion) wasretunedwith the resultsthat Cdiss-- 3.0 (remarkably
comparedwith the data of Wilson[1965].This overestimation, closeto the value of 2.6 as initially usedby Janssen[1991a]).
respectively,underestimation,at very smallfetchesis ascribed The requiredPiersonand Moskowitz[1964] limit valueswere
to the linear growthterm A becauseat very short fetchesthe thusobtained(seeFigure 5), but (1) the total energyis signif-
low frequenciesshowrelativelyhigh levelsof energy(incom- icantlyoverestimatedat very shortfetchesand underestimated
patible with the shape of the JONSWAP spectrum[Hassel- at larger fetches,(2) the peak frequencyis correspondingly
mann et al., 1973]).A possiblereasonis that the expressionof significantlyunderestimatedat short fetches(but not at very
Cavaleriand Malanotte-Rizzoli[1981] is usedfor all spectral large fetches),and (3) spuriousoscillationsare introducedat
components (above the Pierson and Moskowitz [1964] (very)largefetches.In view of thisratherpoor agreement,only
frequency),whereasit is strictlyvalid only near the resonance the WAM cycle3 formulationsare used in the following.
frequencies(varyingwith spectraldirection). Removingthis Figure 6 showssomeof the spectracomputedwith SWAN in
linear growth improves the agreement over a much larger the abovecasewith WAM cycle3 formulations.The overshoot
range of fetches, as shown in Figure 4. For this reason, the is not as pronouncedas in observations(e.g., during JON-
linear growthis omitted in the following(but not in the first- SWAP [Hasselmann et al., 1973]) or computedwith exactcom-
guessstart of the stationarymode of SWAN; in the nonsta- putationsof the quadrupletwave-waveinteractions[Hassel-
tionary mode a low initial sea state is assumed). mann and Hasselmann,1981; SWAMP Group, 1985]. This
As noted earlier, the results of Youngand Banner [1992] result is similar to that of the WAM model with its cycle 3
suggestthat some effectsare to be expectedof usinga fixed formulations[seeKomenet al., 1984].
cutoff frequencyin SWAN rather than a dynamiccutoff fre- To test the implementationof the triad wave-waveinterac-
quencyas in the WAM model. However,the aboveagreement tions (the LTA formulationof Eldeberky[1996]), considera
with the observations suggestsno suchsensitivity in SWAN (at one-dimensional casein whichthe wavespropagatefrom deep
least not for the WAM cycle3 formulations).A possibleex- water over a submergedbar without breaking but with the
planationis that Youngand Banner[1992] usedother approx- generationof a significant,secondaryhigh-frequency peak in
imationsfor the wind input and for computingthe quadruplet the spectrum.This situationhasbeen observedBeji and Battjes
wave-wave interactions (Yah [1987] and Resio and Petrie [1993] in a flume (Figure 7). In the SWAN computationsthe
[1991], respectively). observedspectrumis imposedat the first wavegauge1 (i.e., a
The deepwatergrowth curvesof SWAN as computedwith JONSWAP spectrumwith a significantwave height of H s =
the WAM cycle 4 formulations (without the linear growth 0.02 m and a peakfrequency of fp = 0.50 Hz). The long-
in Figure crestedness
term)for a windspeedof U1o= 20 m s-1 areshown issimulated
asbeforewitha cos
søø(0) directional
5. In contrastto the resultswith the WAM cycle3 formula- distribution,and the computationsare carried out without
tions, these results do not agree well with the generalized quadrupletwave-waveinteractions(see section2.5). It was
observations. Again, this may be due to the useof a fixedcutoff verified that the computationalresultsare insensitiveto bot-
frequencyrather than a dynamic cutoff frequency.One indi- tom friction and depth-inducedbreakingin the computations.
cationof thisis that changingthe cutofffrequencyin the WAM The computationalresultsshownin Figure 8 agreewell with
cycle4 formulationsfrom 1 to 1.5 Hz increasesthe significant the observations.However, the evolution of the secondary
waveheightin the fully developedsituationby 16%, whereasit peak is fairly insensitivefor the value of the proportionality
BOOIJ ET AL.' A THIRD-GENERATION COASTAL WAVE MODEL, 1 7659
E (m2/I•)
, -.................
4.1x105 5
............ 9.1x10
6
............ 2.0x10
75 - 6
............ 4.8x10
6
6.8x10
50 -
I
25 -
l
/
I
0
d,J
!
.,/•
.... ! -oø !
coefficientO/EBin the expressionof Eldeberky[1996] (see the observationsand the computedresultsis good for both
(A7)), whichwasvariedbetween0.125and0.5 (seeFigure8). formulationswith nearly identicalresults.The constantvalue
Thisisprobablydueto the factthatthe shape'of the observed of 3' = 0.73 is usedin the following.
spectraover the bar is practicallyan equilibriumshapefor To test bottom friction, considerthe generationof wavesin
triad wave-waveinteractionsin the expressionof Eldeberky an idealized situation in shallow water. The situation is the
[1996](the interactionsin the LTA formulationare zerowhen sameas the aboveidealizeddeepwaterwind generationsitua-
the spectrallevel at the base frequencyis twice the spectral tion, exceptthat the water depthd is limited and constant.The
levelof its secondharmonic).On the otherhand,the evolution computationscan be comparedwith the expressionsof the
of the primary peak is sensitiveto variationsin O/EBand the CoastalEngineering ResearchCenter(CERC) [1973]and Young
resultssuggestthat, in this respect,a compromiseis obtained and Verhagen[1996a] and the envelopeof observationsre-
with aEB = 0.25. This value is therefore used in all following viewedby Holthuijsen[1980];seeFigure 10, where the dimen-
computations. sionless
parameters
are• = g2Etot/Ulo
4 , •p = fpUlo/gand
To testthe implementationof depth-inducedbreaking,con- • = #d/U•o.Theexpression
forthepeakfrequency of CERC
sider a one-dimensionalcasein which the wavespropagate [1973] is obtainedhere by assumingthat the peak period is
from deep water into shallowwater to the beach acrossa equalto the significantperiodasusedin the CERC expression
bar-troughprofile with violent wave breaking.This situation (thisseemsproperasthe spectrumis ratherpeakedin shallow
has been observedby Battjesand Janssen[1978] in another water[Youngand Verhagen, 1996b]).The SWAN computations
flume (Figure 9). The incidentwavespectrumin this experi- in these idealized conditions are carried out with the formula-
ment is a JONSWAP spectrumwith significantwave height tionsof WAM cycle3 adaptedfor shallowwaterwith either(1)
0.20 m and a peak frequency0.53 Hz. The long crestedness is the JONSWAP bottom friction coefficient(Hasselmannet al.
simulated
asbeforewitha cos
søø(0) directional
distribution, [1973]withShottom
-- 0.067m2 s-3 asproposed
byBouwsand
andfor the samereasonasabove,the computationsare carried Komen[1983]for wind seaconditions)or (2) the Collins[1972]
out withoutquadrupletwave-waveinteractions(but with triad model(withCf = 0.015 assuggested
byCollins[1972]or the
wave-waveinteractions).Again,bottomfrictionwasverifiedto Madsenet al. [1988] model with the bottom roughnesslength
be insignificant.The resultsof the computationsare given in scaleKN in the expressionfor the friction factorf,• of Jonsson
Figure9 for the constantbreakerparameter3,= 0.73 of Battjes [1980];KN = 0.05 m as proposedby Tolman [1991]). The
and Stive[1985] and the (clipped)variablebreakerparameter computationalresultsare shownin Figure 10. Theseresultsare
3,of Nelson[1987, 1994] (also (A5)). The agreementbetween very similar for all three bottom friction models, and they
waves 1 2 3 4 5 6 7 stations
. • I I I I I } I
)
0.5
0.0
station 4 station 5
1.0
0.5
0.0
0.0
0.5 1. 1.5 2. 0. 0.5 1. 1.5 2.
f (Hz) f (Hz)
fp * ß ß Hasselmann
etal.
10-3_
SWAN•
.o .o Madsen
Collinset al.
10-4
lO0
10-5
• 50Om
ad 0.Sm 4. Discussion and Conclusions
A third-generation wave model, called Simulating Waves
Figure 11. Bathymetryof Lake George (Australia) and lo- Nearshore(SWAN), has been developedfor wave computa-
cationsof the eight observationstations(stars). tionsin coastalregions(shallowwaterwith ambientcurrents).
7662 BOOIJ ET AL.: A THIRD-GENERATION COASTAL WAVE MODEL, 1
0.2 -
0.50
0.1 -
0.25
0.0
Hs(m)
UlO- 10.4m/s
0.75
0.4 -
0.3 - 0.50
0.2 -
0.25 -
0.1 -
Hs(m) f•Otz)
UlO- 15.2m/s
0.6 - 0.75 -
+
0.4 - 0.50 -
0.2 - 0.25 -
Figure12. Observed
andcomputed
(left)significant
waveheight
Hsand(right)peakfrequencyfp
in nearly
idealgeneration
conditions
at the eightstations
in LakeGeorgefor threeselected
cases(U•o = 6.4, 10.4,
and15.2m s-• fromnortherly
directions).
SWANresults,
shown
at allstations,
arein some
cases
(nearly)
identical to the observations.
"*"'-'- fP * withB&J
10-3_ o without
10.4
10-5
10-6 gert
........ CERC I
10'7 I •,, [0-1 I I
10 -2 10-i 100 d 10-2 10-1 100 d
Figure 13. SameasFigure10,but for computations with the JointNorthSeaWaveProject(JONSWAP)
bottomfrictionformulationof Hasselmann et al. [1973],with bottom-induced
breaking(spectral
versionof
BattjesandJanssen[1978])activatedand deactivated in the SWAN computations.
BOOIJ ET AL.: A THIRD-GENERATION COASTAL WAVE MODEL, 1 '7663
This has been achievedby (1) extendingthe formulationsof [Youngand Perhagen,1996a]showeda slightpreferencefor the
the WAM wavemodelfor deepwater and intermediate-depth JONSWAP formulation of Hasselmannet al. [1973].
water [WAMDI Group, 1988;Komenet al., 1994] by adding SWAN seemsto representfairly well the state of the art in
formulationsfor depth-inducedwavebreakingand triad wave- coastal wave modeling (within the class of linear, phase-
wave interactionsand (2) by usingunconditionallystablenu- averagedmodelswith nonlinearsourcesand sinksbut without
mericaltechniquesthat are particularlysuitedfor small-scale, diffraction).In the aboveacademiccasesthe model performs
shallow-water,high-resolutioncomputations.SWAN thus ac- well, indicatinga proper numericalimplementation.However,
countsfor the propagationof waveenergy(linear wave theory these tests are rather limited in their parameter range, and
for nonstationarywater depthsand currents),while the pro- further academic testing is still necessary.More important,
cessesof generationand dissipationare accountedfor with perhaps,are testsin real field conditions.A first seriesof such
third-generationformulations(windgeneration,whitecapping, field testsis describedin a sequelpaper [Riset al., this issue].
triad and quadrupletwave-waveinteractions,bottom friction, Since SWAN is a third-generationmodel in which all pro-
and depth-induced wavebreaking).The formulationfor triad cessesof generation,propagation,and dissipationare explicitly
wave-waveinteractionsand depth-induced wavebreaking[El- modeled, it offers the opportunityto absorbfuture basic de-
deberky,1996;EMeberkyand Battjes,1995]are new for thistype velopmentsin the understandingof thesephysicalprocesses.
of wave model. SWAN does not account for diffraction. The model is well suited for this sinceit is strictlyand logically
The numerical schemes that are used in SWAN are also new modular (e.g., each sourceterm in its own subroutine).To
for thistypeof model.They are implicitrather than explicitas encourageits use for suchdevelopmentsand for operational
in other wave models such as the WAM model. The basic purposes,SWAN hasbeenreleasedin the publicdomain(see
reasonfor this choiceis that implicitschemespermit relatively acknowledgments).
large time steps(limited only by accuracy).This is advanta-
geousin high-resolution situationswhere the time stepwould
otherwise(for explicitschemes)be 1 or 2 orderssmaller(and Appendix
require correspondingly more computationaleffort). In sta- The completeexpressions for depth-induceddissipationand
tionarymode,time is removedfrom the model as an indepen- the triad wave-wave interactions that are used in SWAN are
dent variable, allowing even more economic computations. given herein.
The propagationschemein SWAN is fairly diffusive,but since
the model is intendedto be usedon smallscalesonly (coastal A1. Depth-Induced Wave Breaking
regionswith horizontalscalesof lessthan25 km), the effectsof
this are deemed to be acceptable. To model the energy dissipationin random wavesdue to
Propagationtestswith shoalingand refractionin deep and depth-inducedbreaking,the bore-basedmodel of Battjesand
shallow water with and without ambient currents show excel- Janssen[1978]is used.The meanrate of energydissipationper
lent agreementwith analyticalsolutions.In idealizeddeepwa- unit horizontalarea due to wave breakingSds,br,to t is ex-
ter wind generationconditions,the computationalresultsof pressedas
SWAN with the WAM cycle 3 formulationsfor wind input,
quadrupletwave-waveinteractions,and whitecappingagree Sds,br,
tot
-- --• OI
BjQ
b• H}, (A1)
very well with the expressions of Kahtnaand Calkoen[1992],
the data compiledby Wilson[1965], and the limit values of in which O/Bj-- 1 in SWAN and Q b is the fractionof breaking
Piersonand Moskowitz[1964]. This is not the case for the wavesdeterminedby
resultswith the WAM cycle 4 formulations[Komenet al.,
1 - Qb Etot
1994].This is possiblydue to the differentchoiceof frequency
that separatesthe prognosticpart of the spectrumfrom the lnQ• =-8 Hm2, (A2)
high-frequency diagnosticpart. A testfor the triad wave-wave in whichHm is the maximumwave height that can existat the
interactionsshowsgoodagreementwith the laboratoryobser- givendepth and 6-is a mean frequencydefinedas
vations of Befi and Batties [1993]. However, this test is not
totallyconclusive in the sensethat the modelresultsare fairly
insensitiveto largechangesin the proportionalitycoefficientof •r = Et•• rrE(o-, 0) drr cO. (A3)
the expression of triad wave-waveinteractionsthat was used
[Eldeberky,
1996].Fortunately,the exactvalue of this coeffi-
cient seems to be irrelevant for the wave evolution in the cases Extendingthe expressionof Eldeberkyand Batties[1995] to
considered in this study. A similarly good agreement is include the spectraldirections,the dissipationfor a spectral
achievedin the test for depth-inducedbreaking where the componentper unit time is calculatedwith
computationalresultsare comparedwith laboratoryobserva- Sds,br,
tot
tionsof Battiesand Janssen[1978].This test indicatesa slight O)= - --EtotE(rr,
Sds,br(O', 0) (A4)
preferencefor a constantratio of maximumindividualwave
heightover depth (•, = 0.73). The maximumwave height H m is determinedwith H m --
A bottom friction test, based on generalized fetch- •,d, in which •, is the breakerparameter(the steepness depen-
independent,shallow-waterobservations (dimensionless rep- dencyin the Battjesand Janssen[1978]modelis ignoredhere,
resentation)and with all of the aboveprocesses plus bottom as this steepnesseffect is assumedto be representedby the
friction active in the model, was inconclusive as to the best whitecappingprocess).In the literaturethisbreakerparameter
formulationfor the bottomfrictiondissipation[Hasselmann et •, is often a constantor it is expressedas a functionof bottom
al., 1973;Collins,1972;Madsenet al., 1988].Further testswith slopeor incidentwavesteepness [see,e.g.,Galvin,1972;Batties
fetch-limited, shallow-water observations in Lake George and Janssen,1978;Battjesand Stive,1985;Arcilla and Letnos,
7664 BOOIJ ET AL.: A THIRD-GENERATION COASTAL WAVE MODEL, 1
1990; Kaminskyand Kraus, 1993; Nelson, 1987, 1994]. Since with7' = 2rr/6-.Thetriadwave-wave
interactions
areignored
SWAN is locally defined, the dependencyon incident wave for Ur < 0.1. The interaction coefficient J is taken from
steepness cannotbe used.Instead,the other two options(con- Madsenand SOrensen
[1993]:
stantvalue or bottom slopedependent)are usedto determine
the value of the breaker parameter. k2•/2(gd
+ 2c2•/2)
Battjes and Janssen[1978], who describe the dissipation
model, use a constantbreaker parameter, based on Miche's
criterion, of 3/= 0.8. Battjesand Stive [1985] reanalyzedwave
J= (gd+ -•-2gd3k2o
kod 2rr2d2
.- •- ) (All)
data of a number of laboratory and field experimentsand
Acknowledgments.We wish to acknowledgethat in termsof basic
found values for the breaker parameter varying between 0.6
concepts,we standon the shouldersof the WAM Group, from whom
and 0.83 for differenttypesof bathymetry(plane,bar trough, we gleanedmany of our ideas.This group is too large to mention all
and bar), with an averageof 0.73. From a compilationof a members(they are listedbyKomenet al. [1994]),but we want to thank
large number of experiments,Kaminsky and Kraus [1993] Gerbrand Komen of the Royal Netherlands MeteorologicalInstitute
found breaker parametersin the range of 0.6 to 1.59, with an and Klaus Hasselmannof the Max-Planck-Institutf/Jr Meteorology,
Hamburg,for their inspirationalrole. We alsowant to thankKlausand
averageof 0.79. Nelson [1987, 1994, 1997] also (re-)analyzed Suzanne Hasselmannfor their permissionto use the code of the
laboratoryand field wave data and found breaker parameters discreteinteractionapproximationof the quadrupletwave-waveinter-
in the rangeof 0.55 (for horizontalbottoms)to 1.33 (for very actions.It wassomewhatmodifiedby Hendrik Tolman of the National
steep slopes).He suggestedthe followingexpressionfor the OceanicandAtmosphericAdministration(United States),who shared
these modifications with us. We thank him for this and also for the
breakerparameterfor a slopingbottom (note that in the work
many discussions that we have had with him on the intricaciesof
of Nelson [1994, equation (4)] the constant0.88 is missingin numerical wave modeling. We were very fortunate to have accessto
the expressionof 3/ (R. Nelson, personal communication, the original data of the Lake George experimentof Ian Young and
1996): LouisVerhagen, Universityof New SouthWales, Canberra,Australia,
whom we herebythank for their generosity.JurjenBattjesof the Delft
3/= Hm/d - 0.55 + 0.88 exp [-0.012 cot (/3)], (A5) Universityof Technologygave us the benefit of his enthusiasticsup-
port and his well-foundedsuggestions in the courseof this study.We
in which/3is the localbottom slopein the meanwavedirection. thank him for this. We also thank him and Yasser Eldeberky for
This expressionis basedon direct observationsof the ratio of sharing their findings on triad wave-waveinteractionsin the early
stagesof their work on this subject.We are grateful to Cees Vuik of
the observed maximum individual wave height over local Delft University of Technology, whose source code of the ILU-
depth. However, it does not fit the data that Nelson used CGSTAB solveris usedin SWAN. With the sponsorshipof the Office
particularlywell for 0.01 </3 < 0.1 [Nelson,1987,Figure 4]. In of Naval Research(United States),undergrantN00014-97-1-0113(PR
SWAN the expression is thereforecut off (clipped)at/3 = 0.01 number97PR02231-00),and the Ministry of Transport,PublicWorks
(maximumvalue for 3/ = 0.81). The bottom slope /3 in this and Water Management(Netherlands),SWAN hasbeen releasedin
the publicdomain(http://swan.ct.tudelft.nl).
expressionis estimatedin SWAN in the mean wave direction
with a first-orderupwindscheme.Sinceno informationseems
to be availablefor negativebottom slopes(e.g., behind sand References
bars),the constantvalue of 3/= 0.73 is usedfor suchslopes.In
SWAN both the constantvalue of 3/= 0.73 (the averagevalue Abbott, M. B., and D. R. Basco,ComputationalFluid Dynamics,425
pp., JohnWiley, New York, 1989.
of Battjesand Stive[1985, Table 1]) and the (clipped) slope- Abreu, M., A. Larraza, and E. Thornton, Nonlinear transformation of
dependentexpressionof Nelson [1987] are available. directional wave spectra in shallow water, J. Geophys.Res., 97,
15,579-15,589, 1992.
A2. Triad Wave-Wave Interactions Arcilla, A. S., and C. M. Lemos, Surf-ZoneHydrodynamics, 310 pp.,
Cent. Int. de Mdtodos Numer. en Ing., Barcelona,Spain, 1990.
The lumpedtriad approximation(LTA) of Eldeberky[1966], Arcilla, A. S., J. A. Roelvink, B. A. O'Connor, A. J. H. M. Reniers, and
whichis a slightlyadaptedversionof the discretetriad approx- J. A. Jimenez,The Delta flume '93 experiment,paper presentedat
CoastalDynamicsConference'94, Am. Soc.of Civ. Eng., Barcelona,
imation of Eldeberkyand Battjes [1995] is used in SWAN in Spain, 1994.
each spectraldirection: Banner,M. L., and I. R. Young, Modellingspectraldissipationin the
evolution of wind waves,I, Assessmentof existingmodel perfor-
S,3((r, 0)= S,;)3(o-,
0)+ S,)3(o
-, 0) (A6) mance,J. Phys.Oceanogr.,24, 1550-1571, 1994.
Battjes,J. A., and S. Beji, Breakingwavespropagatingover a shoal,in
with
Proceedings of 23rd InternationalConferenceon CoastalEngineering,
pp. 42-50, Am. Soc.of Civ. Eng., New York, 1992.
S•)3(o
-, 0) = max{0, CrEB2rrCC•J
2 [sin(t)l Battjes,J. A., and J.P. F. M. Janssen,Energy lossand set-updue to
breakingof randomwaves,in Proceedings of 16th InternationalCon-
ß[E2(rr/2, 0)- 2E(rr/2, O)E(rr, 0)]} (A7) ferenceon CoastalEngineering, pp. 569-587, Am. Soc.of Civ. Eng.,
New York, 1978.
S•)3(o-,0)--2S•33(2o-, 0) (A8) Battjes, J. A., and M. J. F. Stive, Calibration and verification of a
dissipationmodel for random breaking waves,J. Geophys.Res.,
in which a•B is a tunable proportionalitycoefficient.The bi- 90(C5), 9159-9167, 1985.
phase/3 is approximatedwith Battjes,J. A., M. Isaacson,and M. Quick (Eds.), Shallowwater wave
modelling,in Proceedings of Waves-Physical and NumericalModel-
ling, pp. 1-24, Univ. of B.C., Vancouver,1994.
/3= - 2-+ 2-tanh •rr (A9) Beji, S., and J. A. Battjes,Experimentalinvestigationof wave propa-
gation over a bar, CoastalEng., 19, 151-162, 1993.
with Ursell number Ur Benoit, M., F. Marcos, and F. Becq, Development of a third-
generation shallow-waterwave model with unstructuredspatial
meshing,in Proceedings of 25th InternationalConferenceon Coastal
!7 Hs•2 Engineering,pp. 465-478, Am. Soc. of Civ. Eng., New York, 1996.
Ur=8X/2-rr
2 d2 (A10) Berkhoff,J. C. W., Computationof combinedrefraction-diffraction, in
BOOIJ ET AL.: A THIRD-GENERATION COASTAL WAVE MODEL, 1 7665
mental study,Civil. Eng. Trans.29, pp. 157-161, Inst. of Eng. Aust., of a spectralwind-wavemodel in shallowwater, in Proceedings of
Barton, 1987. 24th InternationalConferenceon CoastalEngineering,pp. 761-774,
Nelson,R. C., Depth limited wave heightsin very flat regions,Coastal Am. Soc. of Civ. Eng., New York, 1994.
Eng., 23, 43-59, 1994. Vincent, C. L., J. M. Smith, and J. Davis, Parameterization of wave
Nelson, R. C., Height limits in top down and bottom up wave envi- breaking in models, in Proceedingsof International Symposium:
ronments,CoastalEng., 32, 247-254, 1997. Waves--PhysicalandNumericalModelling,vol. 2, editedby M. Isaac-
Peregrine,D. H., Long waveson a beach,J. Fluid Mech., 27, 815-827, son and M. Quick, pp. 753-762, Univ. of B.C., Vancouver,1994.
1966. Vuik, C., Solution of the discretizedincompressibleNavier-Stokes
Phillips,O. M., On the generationof wavesby turbulentwind, J. Fluid equationswith the GMRES method,Int. J. Numer. MethodsFluids,
Mech., 2, 417-445, 1957. 16, 507-523, 1993.
Phillips,O. M., The Dynamicsof the UpperOcean,336 pp., Cambridge WAMDI Group, The WAM model--A third generationoceanwave
Univ. Press, New York, 1977. predictionmodel, J. Phys.Oceanogr.,18, 1775-1810, 1988.
Phillips,O. M., Spectraland statisticalpropertiesof the equilibrium Weber, S. L., Surface gravity waves and turbulent bottom friction,
rangein wind-generatedgravitywaves,J. Fluid Mech., 156, 505-531, Ph.D. thesis, Univ. of Utrecht, Utrecht, Netherlands, 1989.
1985. Weber, S. L., Bottom friction for wind sea and swellin extremedepth-
Pierson,W. J., and L. Moskowitz,A proposedspectralform for fully limited situations,J. Phys.Oceanogr.,21, 149-172, 1991a.
developedwind seasbasedon the similaritytheory of S. A. Kitaig- Weber, S. L., Eddy-viscosityand drag-law models for random ocean
orodskii,J. Geophys.Res.,69(24), 5181-5190, 1964. wave dissipation,J. Fluid Mech., 232, 73-98, 1991b.
Piest,J., Seegangsbestimmung und Seegangsrefraktion in einem Meer Whitham, G. B., Linear and Nonlinear Waves,636 pp., John Wiley,
mit nichtebenemBoden; eine theoretischeUntersuchung(in Ger- New York, 1974.
man), Dtsch.Hydrogr.Z., 18, 67-74, 1965. Wilson, B. W., Numerical prediction of ocean waves in the North
Radder, A. C., On the parabolic equation method for water-wave Atlantic for December 1959, Dtsch.Hydrogr.Z., 18, 114-130, 1965.
propagation,J. Fluid Mech., 95, 159-176, 1979. Wu, J., Wind-stress coefficients over sea surface from breeze to hur-
Radder, A. C., An explicit Hamiltonian formulation of surfacewaves ricane,J. Geophys.Res.,87(C12), 9704-9706, 1982.
in water of finite depth, J. Fluid Mech., 237, 435-455, 1992. Yamaguchi, M., A numerical model for refraction computationof
Resio,D., and W. Perrie, A numericalstudyof nonlinearenergyfluxes irregular waves due to time-varyingcurrents and water depth, in
due to wave-waveinteractions,I, Methodologyand basicresults,J. Proceedingsof 22ndInternationalConference on CoastalEngineering,
Fluid Mech., 223, 609-629, 1991. pp. 205-217, Am. Soc. of Civ. Eng., 1990.
Ris, R. C., L. H. Holthuijsen, and N. Booij, A third-generationwave Yan, L., An improved wind input source term for third generation
model for coastalregions,2, Verification,J. Geophys.Res.,this issue. oceanwave modelling,Sci.Rep. WR 87-8, R. Neth. Meteorol. Inst.,
Roache, P. J., ComputationalFluid Dynamics,446 pp., Hermosa,Al- De Bilt, 1987.
buquerque,N.M., 1972. Young, I. R., A shallowwater spectralwave model, J. Geophys.Res.,
Shemdin, P., K. Hasselmann, S. V. Hsiao, and K. Herterich, Non- 93(C5), 5113-5129, 1988.
linear and linear bottom interaction effects in shallow water, in Young, I. R., and M. L. Banner, Numerical experimentson the evo-
TurbulentFluxesThroughthe Sea Surface,WaveDynamicsand Pre- lution of fetch limited waves, paper presented at Int. Union of
diction,NATO Conf Ser.,vol. V(1), 347-372, 1978. Theor. and Appl. Mech. (IUTAM), Sydney,Australia, 1992.
Snyder, R. L., F. W. Dobson, J. A. Elliott, and R. B. Long, Array Young, I. R., and G. P. van Vledder, A review of the central role of
measurementof atmosphericpressurefluctuationsabove surface nonlinear interactions in wind-wave evolution, Philos. Trans. R. Soc.
gravitywaves,J. Fluid Mech., 102, 1-59, 1981. London, Ser. A, 342, 505-524, 1993.
SWAMP Group, OceanWaveModelling,256 pp., Plenum,New York, Young, I. R., and L. A. Verhagen, The growth of fetch limited waves
1985. in water of finite depth, 1, Total energyand peak frequency,Coastal
Taylor, P. A., and R. J. Lee, Simple guidelinesfor estimatingwind Eng., 29, 47-78, 1996a.
speedvariationsdue to small-scaletopographicfeatures,Climatol. Young, I. R., and L. A. Verhagen, The growth of fetch limited waves
Bull., 18, 3-32, 1984. in water of finite depth, 2, Spectral evolution, Coastal Eng., 29,
Thornton, E. B., and R. T. Guza, Transformation of wave height 79-99, 1996b.
distribution,J. Geophys.Res.,88(C10), 5925-5938, 1983. Young, I. R., and L. A. Verhagen, The growth of fetch limited waves
Tolman, H. L., A third-generationmodel for wind waves on slowly in water of finite depth, 3, Directional spectra, CoastalEng., 29,
varying,unsteadyand inhomogeneous depthsand currents,J. Phys. 101-121, 1996c.
Oceanogr.,21, 782-797, 1991.
Tolman, H. L., Effects of numericson the physicsin a third-generation N. Booij and L. H. Holthuijsen(correspondingauthor), Facultyof
wind-wavemodel, J. Phys.Oceanogr.,22, 1095-1111, 1992a. CivilEngineering,Delft Universityof Technology,
Stevinweg1, 2628CN,
Tolman, H. L., An evaluation of expressionsfor the wave energy Delft, Netherlands.(N.Booij@ct.tudelft.nl;
L. Hølthuijsen@ct'tudelft'nl)
dissipation due to bottom friction in the presence of currents, R. C. Ris, WL/Delft Hydraulics, Rotterdamseweg185, 2629 HD,
CoastalEng., 16, 165-179, 1992b. Delft, Netherlands.(Røeland'Ris@wldelft'nl)
Van der Vorst, H. A., Bi-CGSTAB: A fast and smoothlyconverging
variant of Bi-CG for solutionof non-symmetriclinear systems,SIAM
J. Sci. Star. Cornput.,13, 631-644, 1992. (ReceivedJune 16, 1997;revisedMarch 3, 1998;
Van Vledder, G. P., J. G. de Ronde, and M. J. F. Stive, Performance acceptedMay 29, 1998.)