Sunteți pe pagina 1din 30

440 JOURNAL OF ATMOSPHERIC AND OCEANIC TECHNOLOGY VOLUME 15

Worldwide Measurements of Directional Wave Spreading


GEORGE Z. FORRISTALL AND KEVIN C. EWANS
Shell International Exploration and Production, Rijswijk, the Netherlands
(Manuscript received 26 July 1996, in final form 13 May 1997)

ABSTRACT
The directional spreading of waves is important for both theoretical and practical reasons. Enough measure-
ments have now been made to draw conclusions about the behavior of wave spreading at sites in different
climatic regimes. The measurements presented here of the directional spreading of fetch-limited waves agree
in general with those of M. A. Donelan et al., but additional evidence is also found to support the conclusion
of I. R. Young et al. that the spreading function is bimodal at high frequencies. The spreading factor f is defined
to be the square root of the in-line variance ratio defined by R. E. Haring and J. C. Heideman. This spreading
factor gives an integrated measure of the degree of directional spreading in the wave spectrum and predicts the
reduction in the in-line particle velocities under waves due to direction spreading. The value of f is 1 for
unidirectional waves and 0.707 for omnidirectional waves. For fetch-limited conditions, f is essentially constant
at 0.906. Results from the Exact-NL wave model agree reasonably well with this value, but an operational third-
generation model produces directional spreading that is broader than the observations. The statistics of f are
calculated from thousands of hours of measurements from many sites around the world. The median value of
f is 0.880 for low latitude monsoon conditions and 0.867 for tropical cyclones. For extratropical storms, f
decreases with latitude. Regressions on the statistics for high waves give f 5 0.944 for latitude 368 and f 5
0.869 for latitude 728. The latitudinal dependence of f is caused by the facts that waves are generally more
broadly spread at sites close to the center of a storm and that storm tracks are concentrated at high latitudes.

1. Introduction The low-order trigonometric moments can also be


used to calculate measures of wave spreading at each
Following the publication of the Marine Board (1982)
frequency, and wave spreading is at least as important
report that urged the development of reliable and in-
as the mean wave direction for many applications. In
expensive systems for the measurement of directional
particular, the peak particle velocities under waves be-
wave spectra, a number of such instruments have be-
come available and widely used. The amount of direc- come smaller as the waves become more spread. The
tional wave data available for study has grown rapidly American Petroleum Institute’s (API 1993) recom-
in the last few years. mended practices take this reduction into account by
Much of the literature on directional wave measure- applying a ‘‘spreading factor’’ to the kinematics cal-
ments has concentrated on the question of how to extract culated from two-dimensional wave theories. The API
the highest directional resolution from the limited in- recommends using a spreading factor in the range 0.85–
formation available with standard instrumentation. For 0.95 for tropical storms and 0.95–1.00 for extratropical
most instruments, it is possible to compute only the first storms. One of the main purposes of this report is to
two trigonometric moments of the spreading function produce more specific values of the spreading factor for
at each frequency. The mean wave direction as a func- use in designing structures.
tion of frequency can be calculated from the first trig- We begin by showing how the spreading factor should
onometric moments, and most applications of direc- be calculated from directional wave measurements. A
tional measurements have used only these mean direc- theoretical development shows that the velocity reduc-
tions. For example, Cavaleri et al. (1991), Guillaume tion under waves due to directional spreading is equal
(1990), and Khandekar et al. (1994) compared wave to the square root of the in-line variance ratio defined
directions from numerical wave models to measure- by Haring and Heideman (1980). In contrast to the API,
ments. which recommends the use of a spreading factor ‘‘cor-
responding to the spectral peak period,’’ we find that
the spreading factor must be calculated by integrating
the velocity variances over the entire wave spectrum.
Corresponding author address: Dr. George Z. Forristall, Shell In-
ternational Exploration and Production, Postbus 60, 2280 AB Rijs-
This conclusion is confirmed by an analysis of the ve-
wijk, the Netherlands. locity measurements made under very high waves in the
E-mail: gforrist@euronet.nl North Sea made by Jonathan et al. (1994).

q 1998 American Meteorological Society


APRIL 1998 FORRISTALL AND EWANS 441

Fetch-limited waves under steady winds are the sim- Using linear wave theory, the wavenumber k m is related
plest case of wave generation. Measurements of the di- to the frequency f m by
rectional spectra of fetch-limited waves have been made (2p f m ) 2 5 gk tanh(k m d), (4)
by Mitsuyasu et al. (1975) and Hasselmann et al. (1980)
using pitch-and-roll buoys, by Donelan et al. (1985) where d is the water depth and g is the acceleration of
using an array of 14 wave staffs, and by Holthuijsen gravity. Again, from linear theory, the u and y com-
(1983) using stereophotography. Some of our measure- ponents of the wave particle velocity are given by

OOA
` `
ments are also in clearly identifiable fetch-limited con-
ditions that include higher waves than the earlier studies, u(x, y, z, t) 5 mn Q m (z) cosu n cosc mn (x, y, t)
m51 n51
and we compare the directional spectra in those cases
to the earlier measurements. We also compare the mea- (5)

OOA
sured directional spreading functions to those predicted ` `
by third-generation numerical wave models. y (x, y, z, t) 5 mn Q m (z) sinu n cosc mn (x, y, t),
The core of our study is the analysis of thousands of m51 n51

hours of directional wave measurements made at many (6)


sites around the world. The measurements were made
with a wide variety of instruments, but the consistency where
checks we have been able to make indicate that most coshk m z
of the instruments give the same value for the spreading Q m (z) 5 2p f m (7)
sinhk m d
factor in the same conditions. This again contradicts the
API (1993), which states that ‘‘pitch–roll buoys tend to gives the attenuation of the velocity with depth with z
significantly overestimate spreading, while directional measured upward from the bottom.
data from a two-horizontal-axis particle velocimeter are
thought to provide a good estimate of spreading.’’ We 1) COVARIANCES
believe that the API conclusion was largely based on The covariance functions between h, u, and y can be
the Wave Direction Measurement Calibration Project calculated by taking cross products among Eqs. (1), (5),
(WADIC) experiment (Allender et al. 1989). Our rean- and (6), and passing from summation to integration to
alysis of the WADIC data shows that in this particular produce

E E
case, the particle velocimeter underestimated spreading ` p
due to interference from its support structure.
C hh (t ) 5 cos2p f t df S( f, u) du, (8)
The statistics of the wave-spreading factor show sig- 0 2p

E E
nificant dependence on the location of the site and the
` p
wave height. In particular, a clear dependence on the
latitude of the site is evident at the sites exposed to C hu (t ) 5 Q( f ) cos2p f t df S( f, u) cosu du,
0 2p
extratropical storms. This dependence can be related to
physical factors in wave generation and propagation. (9)

E E
The report closes with recommendations of appropriate ` p
spreading factors for use in engineering design. C hy (t ) 5 Q( f ) cos2p f t df S( f, u) sinu du,
0 2p

2. Calculation of the spreading factor (10)


a. Definitions from linear wave theory
Expand the sea surface h as a Fourier series in the form
C uu (t ) 5 E
0
`

Q 2 ( f ) cos2p f t df E2p
p

S( f, u) cos 2u du,

OOA
` `
(11)
h(x, y, t) 5 cosc mn (x, y, t), (1)

E E
mn
m51 n51 ` p

where the phase function cmn is given by C yy (t ) 5 Q 2 ( f ) cos2p f t df S( f, u) sin 2u du,


0 2p

c mn (x, y, t) 5 k m x cosu n 1 k m y sinu n 2 2p f m t 1 f mn


(12)
(2)
and

E
and f m is the wave frequency, k m is the wavenumber, `
u n is the direction toward which the wave is propagating, C uy (t ) 5 Q 2 ( f ) cos2p f t df
and f mn is a phase angle. The Fourier coefficients A mn 0

E
are related to the one-sided directional spectrum of the
p
sea surface S by
3 S( f, u) sinu cosu du. (13)
S( f m , u n )D f Du 5 A mn
2
/2. (3) 2p
442 JOURNAL OF ATMOSPHERIC AND OCEANIC TECHNOLOGY VOLUME 15

2) CO- AND QUADSPECTRA are normally distributed with standard deviations a uu


and a y y .
The covariance function between, say, u and y can
also be calculated from the co- and quadspectra using
C uy (t ) 1) THE WAVE-SPREADING FACTOR

E E
` `
It is always possible to find a coordinate system (a,
5 c uy ( f ) cos2p f t df 1 q uy ( f ) sin2p f t df , b) in which the velocity components are statistically
0 0
independent—that is, a ab 5 0. The angle between the
(14) (x, y) and (a, b) coordinate systems is given by
so from inspection of Eqs. (8)–(13), the cospectra are 2a u2y
tan2u p 5 , (24)

E p a uu
2
2 a yy2

c hh ( f ) 5 S( f, u) du, (15) where u p is defined as the principal wave direction. In


2p
the new coordinate system, the velocity variances at zero

c hu ( f ) 5 Q( f ) E p

2p
S( f, u) cosu du, (16)
lag are given by

a aa
2
1 2
5 (a uu 1 a yy
2
)1r (25)

E p 2
c hy ( f ) 5 Q( f ) S( f, u) sinu du, (17) and
2p

E
1 2
p
a bb
2
5 (a uu 1 a yy
2
) 2 r, (26)
c uu ( f ) 5 Q 2 ( f ) S( f, u) cos 2u du, (18) 2
2p
where

c yy ( f ) 5 Q 2 ( f ) E 2p
p

S( f, u) sin 2u du, (19) 1 2


r 2 5 (a uu
4
2 a yy
2 2
) 1 (a u2y ) 2 . (27)

and The velocity distribution in the principal wave direction

c uy ( f ) 5 Q 2 ( f ) E2p
p

S( f, u) sinu cosu du, (20)


is normal with standard deviation a aa . If all of the wave
energy were in one direction, then the velocity distri-
bution would have a standard deviation a given by
and the quadspectra are zero.
a 2 5 a uu
2
1 a yy
2
5 E
0
`

Q 2 ( f ) df E
2p
p

S( f, u) du. (28)
b. Velocity distributions
Wave spreading thus reduces the rms velocity in the
The variance and covariance of the velocity com- principal wave direction by the factor
ponents at zero lag are defined as

E E
a aa
` p
f5 . (29)
a 5 C uu (0) 5
2
uu Q ( f ) df
2
S( f, u) cos u du,
2 a
0 2p
Since the distributions of the velocity components re-
(21) main Gaussian, the velocity at any probability level is

E E
` p reduced by the same factor. We define f to be the wave-
a yy
2
5 C yy (0) 5 Q 2 ( f ) df S( f, u) sin 2u du, spreading factor. It lies in the interval [1/Ï 2, 1]. We
0 2p will show that it is a robust and useful measure of the
overall directional spread of a sea state. Our spreading
(22) factor is the square root of the ‘‘in-line variance ratio’’
and defined by Haring and Heideman (1980), and our prin-

E
cipal wave direction is equal to their ‘‘dominant wave
`
direction.’’
a u2y 5 C uy (0) 5 Q 2 ( f ) df
0

E p 2) DISTRIBUTION OF MAXIMUM SPEED


3 S( f, u) sinu cosu du. (23) The peaks of the in-line velocity are defined as the
2p
absolute values of the maxima between two zero cross-
If the phases f mn are uniformly distributed as expected ings of the velocity component in the principal wave
for linear wave theory, then the velocity components direction. These peak values have a Rayleigh probability
APRIL 1998 FORRISTALL AND EWANS 443

TABLE 1. Lower and upper bounds of the speed distribution for because they assumed that the phase angles of the en-
P(j, f) 5 0.001. velopes of the velocity components were both the same.
f j u /j The upper and lower bounds are reasonably close
0.80 0.847 together. For example, if we set
0.85 0.880
0.90 0.917
P l (j, 1) 5 P l (j l , f) 5 P u (j u , f) 5 0.001, (33)
0.95 0.958
we immediately have j l /j 5 f. A numerical solution
for j u gives the results in Table 1. The upper bound for
the peak velocity in 1000 peaks is thus 1%–5% greater
distribution. If we normalize the peak velocities S by a than the lower bound.
so that
j 5 S/a, (30) c. An example from the Tern platform in the North
then the probability that the normalized peak velocity Sea
in the principal direction is greater than j is given by The Tern oil production platform is located northeast
of the Shetland Islands in 167 m of water. The instru-
1 2
j2
P l (j, f ) 5 exp 2 . (31) mentation on the platform includes two wave height
2f 2 sensors and a Marsh–McBirney electromagnetic current
The maximum speed in a spread sea is not necessarily meter at 41-m depth below mean water level. Several
in the principal wave direction, but may be somewhat severe storms have been recorded by the instrumenta-
off that axis. Nolte and Hsu (1979) show that the dis- tion. Jonathan et al. (1994) give more details on the
tribution of maximum speed has a lower bound of P l (j, instrumentation and the waves in some of these storms.
f) and an upper bound of Figure 1 shows the speed distribution from 9 h of data
from 2300 UTC on 3 January 1993 to 0700 UTC on 4

1 2 1 2
f2 j2 January 1993. The hourly significant wave height during
P u (j, f ) 5 exp 2 this span ranged from 11.07 m to 12.73 m. This range
2f 2 2 1 2f 2
is just slightly greater than expected from natural sam-

11
2f 2 12
exp 12
2(1 2 f )2
12f 2
j 2 pling variability if the significant wave height was ac-
2
. 2
(32) tually constant.
The solid line in Fig. 1 shows the speed peak distri-
Forristall et al. (1978) mistakenly stated that Eq. (32) bution for unidirectional waves—that is, Eq. (31) is
was the distribution of speed rather than an upper bound evaluated with f 5 1.0. The variance and covariance

FIG. 1. Velocity distribution at Tern, 3–4 January 1993.


444 JOURNAL OF ATMOSPHERIC AND OCEANIC TECHNOLOGY VOLUME 15

of the velocity components at zero lag were computed give only the first few moments of the spreading func-
directly from the time series using an expression of the tion, as defined in the next section.
form

O u (t ).
N
1 1) TRIGONOMETRIC MOMENTS OF THE SPREADING
a uu
2
5 2
n (34)
N n51
FUNCTION

Then f was computed from Eqs. (25) to (29) for the Define the spreading function H( f, u) by
hourly data segments. For the 9 h of data, f ranged
S( f, u) 5 S( f )H( f, u), (35)
from 0.866 to 0.892, with an average of 0.874. The
dashed lines in Fig. 1 are the upper and lower bounds where

E
of the speed distribution calculated from Eqs. (31) and p
(32) using the average value of f. H( f, u) du 5 1. (36)
The velocities were normalized using Eq. (30) with 2p
a from the hour of the speed measurement. The speed
peaks, that is, the peaks of the modulus of the velocity Then following Longuet-Higgins et al. (1963), expand
vectors, were found for the same time intervals as the H( f, u) in the Fourier series:
in-line component of the velocity and used the same

5 O [a ( f ) cosnu 1 b ( f ) sinnu]6 .
`
normalization. The normalization eliminates any errors 1 1
H( f, u) 5 1
from imperfections in the current meter calibration or p 2 n51
n n

linear wave theory and isolates the effect of direction-


ality. (37)
The sample distribution of the in-line velocities Inserting this expansion into Eqs. (9)–(13) gives
agrees closely with the theoretical lower bound distri-
bution, that is, the in-line velocities are reduced from c hu ( f ) 5 Q( f )S( f )a1 ( f ), (38)
the value they would have in a unidirectional sea by the c hy ( f ) 5 Q( f )S( f )b1 ( f ), (39)
factor f. The sample distribution of the speed is slightly
higher than that for the in-line velocity and agrees close- 1
ly with the theoretical upper-bound velocity. c uu ( f ) 5 Q 2 ( f )S( f )[1 1 a2 ( f )], (40)
2
It would be appropriate to use the distribution for the
speed, given approximately by the upper-bound distri- 1
c yy ( f ) 5 Q 2 ( f )S( f )[1 2 a2 ( f )], (41)
bution, for the calculation of forces on a single pile. But 2
for calculation of the forces on a structure made up of
many members, the distribution of the in-line velocity and
is more appropriate, since the instantaneous directions 1
of the particle velocities will vary across the structure c uy ( f ) 5 Q 2 ( f )S( f )b2 ( f ). (42)
2
in directionally spread seas. For engineering purposes,
the reduction in velocity due to directionally spread seas The standard method of estimating a directional spec-
can thus simply be taken to be the factor f. trum from a wave staff and current meter or from a
pitch-and-roll buoy is to calculate the cospectra using
the fast Fourier transform and invert Eqs. (38)–(42) to
d. Calculation of f from the directional spectrum give a1 , a 2 , b1 , and b 2 , which can be thought of as the
The previous section has demonstrated that the re- first- and second-order trigonometric moments of the
duction in particle velocity due to directional spreading spreading function.
is described by the spreading factor f, which is cal- By definition, the variances and covariance of the
culated from the variance and covariance of the velocity velocity components are given by

E
components. The most direct method of computing the `
1
variance and covariance is from summation of the time a uu
2
5 Q 2 ( f )S( f )[1 1 a2 ( f )] df , (43)
series as in Eq. (34). In many cases, however, we may 2 0

E
wish to compute the spreading factor at a depth other `
than that at which velocity measurements were made, 1
a yy
2
5 Q 2 ( f )S( f )[1 2 a2 ( f )] df , (44)
or we may be given a directional spectrum calculated 2 0
by some other means than particle velocity measure-
ments, such as a pitch-and-roll buoy or a wave hindcast. and
In these cases, the variances and covariances can be
calculated from the integrals in Eqs. (21)–(23). We do
not need the complete details of the directional spectrum
a u2y 5
1
2 E
0
`

Q 2 ( f )S( f )b2 ( f ) df . (45)

to perform these integrations, which is fortunate since The spreading factor f may thus be calculated given
most methods of estimating the directional spectrum only the second trigonometric moments of the spreading
APRIL 1998 FORRISTALL AND EWANS 445

function. The factor Q 2 S in Eqs. (43)–(45) can be cal- force on a pile can then be calculated from Eqs. (43)–
culated either from the velocity measurements or from (45) with Q eliminated, using

E
measurements of wave elevation and Eq. (7). The latter `
method is generally preferable since velocity measure- 1
a uu
2
5 S( f )[1 1 a2 ( f )] df , (50)
ments are generally noisier than elevation measure- 2 0

E
ments. Furthermore, it allows the spreading factor to be
`
calculated at depths where no velocity measurements 1
were made. Using Eq. (7) also permits data from pitch- a yy
2
5 S( f )[1 2 a2 ( f )] df , (51)
2 0
and-roll buoys to be used in exactly the same way as
measurements from a current meter and wave staff. If and

E
the sensors are operating properly, either type of mea- `
surement will give the same values for the circular mo- 1
a u2y 5 S( f )b2 ( f ) df . (52)
ments as a function of frequency, and thus the same 2 0
value for the spreading factor.
In other words, the global spreading function is cal-
culated by integrating over the second trigonometric
2) THE SPREADING FACTOR INTEGRATED OVER moments of the spreading function over frequency,
DEPTH weighted by the power spectrum. This intuitively simple
definition also appears to be the most generally useful
Using Eqs. (43)–(45) gives the spreading factor ap-
definition for engineering use since it gives the force
propriate for calculating the reduction of velocity due
reduction on a drag-dominated structure. This definition
to spreading at one particular depth. For many purposes,
of the spreading function will be used for all of the
it is preferable to have a more global view of the spread-
calculations in the rest of this paper.
ing. The drag force on a cylinder is proportional to the
square of the velocity, so the total force F on a vertical
cylinder in deep water is proportional to the integral of 3) CIRCULAR MOMENTS
the velocity squared over depth:

E 0 The vector mean of the directional distribution u1 can


F} u 2 dz. (46) be calculated from the lowest Fourier coefficients as
2`
u1 ( f ) 5 tan 21 [b1 ( f )/a1 ( f )]. (53)
So from Eq. (5),
Then following Kuik et al. (1988), we define the cen-
F} E 0

1O O A
` `

Q m (z) cosu n cosc mn (x, y, t) 2


2

dz.
tered Fourier coefficients as

E
mn
2` m51 n51 2p

(47) m1 ( f ) 5 cos[u ( f ) 2 u1 ( f )]H( f, u) du, (54)


0
If we expand the square and interchange the summation
and integration, the integration will be over terms of the 5 [a12 ( f ) 1 b12 ( f )]1/2 , (55)
form

Fij } E 0

Q i (z)Q j (z) dz. (48)


m2 ( f ) 5 E
0
2p

cos2[u ( f ) 2 u1 ( f )]H( f, u) du, (56)

2`
5 a2 ( f ) cos2u1 ( f ) 1 b2 ( f ) sin2u1 ( f ), (57)

E
In the deep water limit, 2p

E 0 n1 ( f ) 5 sin[u ( f ) 2 u1 ( f )]H( f, u) du, (58)


Fij } (2p) f i f j (z)
2
e (k i1k j)z
dz 0

2`
5 b1 ( f ) cosu1 ( f ) 2 a1 ( f ) sinu1 ( f ), (59)
g fi fj
5 2 . (49) and

E
f i 1 f j2 2p

Since the energy containing part of the wave spectrum n2 ( f ) 5 sin2[u ( f ) 2 u1 ( f )]H( f, u) du, (60)
is over a fairly small range of frequencies, f i ø f j , and 0

F ij ø g/2, a constant. Integration of the velocity squared 5 b2 ( f ) cos2u1 ( f ) 1 a2 ( f ) sin2u1 ( f ). (61)


over depth thus effectively removes the frequency de-
pendence in the factor Q so that the drag force is ef- Kuik et al. (1988) then demonstrate that the best defi-
fectively independent of wave frequency for a given nitions for the circular standard deviation s, the skew-
wave height. The spreading factor appropriate for the ness m 3 , and the kurtosis m 4 are, respectively,
446 JOURNAL OF ATMOSPHERIC AND OCEANIC TECHNOLOGY VOLUME 15

s ( f ) 5 {2[1 2 m1 ( f )]}1/2 , (62) Eq. (29). The variances can be conveniently calculated
by integrating the second trigonometric moments of the
2n2 ( f ) directional spreading function over frequency as given
m3 ( f ) 5 , (63)
{[1 2 m2 ( f )]/2} 3/2 in Eqs. (50)–(52).
The spreading factor defined here has the advantage
and that it can be calculated accurately from the information
6 2 8m1 ( f ) 1 2m 2 ( f ) produced by commonly available measuring systems
m4 ( f ) 5 . (64) such as pitch-and-roll buoys or electromagnetic current
s 4( f ) meters. Despite the fact that these systems do not have
Given s, m 3 , and m 4 or the central moments, the spread- very good directional resolving power and thus cannot
ing factor f can be found by calculating m 2 from Eq. give the details of the directional spreading function,
(64) and n 2 from Eq. (63). Then a 2 and b 2 can be found they do give its second moments and thus the spreading
by solving Eqs. (56) and (60) to get factor. If the waves are unidirectional, a pitch-and-roll
buoy will show that f 5 1.
a2 ( f ) 5 m 2 ( f ) cos2u1 ( f ) 2 n2 sin2u1 ( f ) (65) The spreading factor is robust in the face of calibra-
and tion errors in the measurements since it depends only
on the ratio of the in-line and total variances. Essentially
b2 ( f ) 5 m 2 ( f ) sin2u1 ( f ) 1 n2 cos2u1 ( f ). (66) all that is required is that the two axes of the measure-
Other definitions of the circular moments are possi- ment system have the same (linear) response. Since the
ble; for instance, those given by Mardia (1972) are definition in Eqs. (50)–(52) weights the integration by
the power spectrum, pitch-and-roll buoys and electro-
slightly different from Eqs. (62) through (64). Care
magnetic current meters will give the same spreading
should be taken to invert the equations that were actually
factor. This equality is in contrast to the results that
used to produce the circular moments.
would be obtained if the variances were calculated di-
It is somewhat surprising that the spreading factor
rectly from the time series given by the instruments.
cannot be found given only the standard deviation of The velocities would then be weighted by an additional
the spreading function. The problem is that s is based factor of f 2 , and the tilts would be weighted by f 4 . This
on the first trigonometric moments of the spreading weighting to the higher frequencies in the spectrum
function rather than the second moments, which actually would lower the spreading factor since the higher fre-
show the amount of energy in the principal wave di- quency waves generally have broader spreading.
rection. Given only s, it is probably most reasonable
to assume that the skewness is zero and the spreading
function is a normal distribution for which the kurtosis 3. Fetch-limited waves
m 4 ( f ) 5 3. Then Eqs. (62)–(64) give n 2 ( f ) 5 0 and We begin by considering the directional distribution
for the simplest case of wave generation: fetch-limited
3 seas under constant winds. The form of the directional
m 2 ( f ) 5 1 2 2s 2 ( f ) 1 s 4 ( f ). (67)
2 distribution for fetch-limited sea states has been the sub-
ject of a number of investigations, including the well-
The approximation in Eq. (67) has the disadvantage that known ones of Mitsuyasu et al. (1975), Hasselmann et
m 2 reaches a minimum of ⅓ at s 2 5 ⅔. The minimum al. (1980), and Donelan et al. (1985). Descriptions of
value of the spreading function is thus f 5 Ï ⅔ rather the directional distributions proposed in these studies
than Ͻ, as would be expected with omnidirectional are given in the following sections, but in view of the
spreading. Thus, it is probably better to express the fact that these distributions are all unimodal and there
spreading function as a normal distribution and make is evidence now that fetch-limited sea states may have
the approximation bimodal directional distributions at frequencies above

m2 ( f ) 5 E
2p
p

H( f, u) cos2u du
the spectral peak frequency, we include a section on a
bimodal distribution based on data presented by Ewans
(1995) and include this distribution in our comparisons.

E
State-of-the-art, third-generation wave models calculate
`

1 2
1 2u 2 the nonlinear wave–wave interaction source term ex-
ø exp cos2u du
sÏ2p 2`
2s 2 plicitly and do not rely on a specific empirical form of
the directional distribution. We include two such mod-
5 exp(22s 2 ). (68) els—Exact-NL and PHIDIAS—for comparison with the
empirical forms.

e. Summary a. Proposed directional distributions


The spreading factor f gives the fraction of the wave 1) MITSUYASU DISTRIBUTION
energy in the principal wave direction. It is calculated Mitsuyasu et al. (1975) measured directional wave
as the square root of the in-line variance ratio given by spectra with a cloverleaf buoy at open sea locations in
APRIL 1998 FORRISTALL AND EWANS 447

the Sea of Japan and the Pacific Ocean and in a bay on 

 12
4.06
f
the east coast of Japan. Meteorological data were col- 6.97 f , 1.05 f p ,
lected from a tending ship near each observation station. fp
s5  (72)
9.771 ff 2
They chose five datasets for estimating the directional m

distribution, with wind speeds ranging from 7 to 10 m f $ 1.05 f p ,


s21 and significant wave heights from 0.74 to 2.34 m.  p

The cloverleaf buoy enables the first four pairs of the where m has a dependence on wave age as follows:
coefficients in Eq. (37). However, the higher-order co-
efficients, which are available from the measurement of
1c 2
U10
the wave curvature, were not used because they were m 5 22.33 2 1.45 2 1.17 . (73)
thought to be inaccurate. Thus, the data used by Mit- p

suyasu et al. (1975) were the same as if the measurement The directional distribution defined by Eqs. (69), (72),
instrument was a pitch-and-roll buoy. and (73) will be referred to as the Hasselmann distri-
The Mitsuyasu distribution is based on the so-called bution in the remainder of this paper.
cosine2s form:

1 2
u2u 3) DONELAN DISTRIBUTION
H( f, u) 5 A(s) cos 2s , (69)
2 Donelan et al. (1985) report an analysis of data re-
where A(s) is a normalization factor to ensure Eq. (36) corded with an array of 14 wave staffs in Lake Ontario
is met, and u is the mean wave direction at frequency and a similar, scaled-down version in a large laboratory
f. Like u, the parameter s is a function of frequency. tank. The wave staffs were mounted on a tower 1 km
Based on their data, Mitsuyasu et al. (1975) proposed offshore in a water depth of 12 m. Meteorological data
the following parameterization for s: were also collected from the tower and with a buoy 11
km from the tower in deeper water. Eighty-five field
 recordings and seven laboratory recordings were used
 12
5
f
sp f , fp, in the analysis. Donelan et al. (1985) do not report the
fp absolute range of wind speed and significant wave
s5  (70)
s 1 ff 2
22.5
heights associated with their analysis dataset, but the
p f $ fp, field data were in the range 0.83 , U/c p , 4.6 and the
 p
laboratory data in the range 7.2 , U/c p , 16.5.
where sp is the value of s at the frequency of the spectral Based on the theoretical directional characteristics of
peak fp given by freely propagating, second-order Stokes wave groups,
Donelan et al. (1985) proposed the following directional
22.5 distribution:
1 2
U10
s p 5 11.5 , (71)
cp H( f, u) 5 0.5b sech 2b(u 2 u ), (74)
where U10 is the wind speed at 10 m above sea level where
and c p 5 g/2pf is the deep water phase speed at the

 1f2
1.3
spectral peak. The directional distribution defined by f
Eqs. (69)–(71) will be referred to as the Mitsuyasu dis- 2.61 0.56 , f / f p , 0.95,
fp
tribution in the remainder of this paper. 21.3

b 5 2.281 2


0.96 , f / f p , 1.6, (75)
f p
2) HASSELMANN DISTRIBUTION

Hasselmann et al. (1980) report an analysis of data 1.24 f / f p , 1.6.


recorded during the JONSWAP experiment. The direc-
tional wave data were collected with a pitch-and-roll The Donelan et al. (1985) dataset extended to only
buoy located in 22 m of water 52 km off the island of f / f p 5 1.6. Thus, the constant value of b 5 1.24 for
Sylt in the North Sea. Meteorological data were also frequencies greater than 1.6 was assumed.
collected at this site with a meteorological buoy located The directional distribution defined by Eqs. (74) and
27 km offshore in a water depth of 18 m. The dataset (75) will be referred to as the Donelan distribution in
chosen for analysis ranged in wind speed from 6.8 to the remainder of this paper.
15.0 m s21 and significant wave heights from 0.55 to
1.88 m. 4) DONELAN–BANNER DISTRIBUTION
The Hasselmann distribution is also based on the co-
sine2s form (69) with the following parameterization Based on high-frequency stereo photography, Banner
for s: (1990) concluded that b was not a constant at values
448 JOURNAL OF ATMOSPHERIC AND OCEANIC TECHNOLOGY VOLUME 15

of f / f p 5 1.6, as specified by Donelan et al. (1985), of the distribution are defined by two functions of the
and proposed that normalized frequency—one that models the angular sep-
aration of the peaks and another that models the shape
b 5 10{20.410.8393 exp[20.567 ln(( f / f p) 2)]}, f / f p . 1.6.
of the angular width parameter as follows:
(76)
Du 5 14.93 for f , f p ,
Thus, the Donelan–Banner distribution consists of the
Donelan distribution to f / f p 5 1.6 and the Banner def- 21

1 122
f
inition for b [Eq. (76)] for f / f p . 1.6. Du 5 exp 5.453 2 2.750 for f $ f p , (78)
fp

5) MAUI DISTRIBUTION
where Du is the angular separation, um2 2 um1, of the
peaks, and
Ewans (1995) reports an analysis of wave directional 27.929

12
measurements made with a Datawell WAVEC buoy f
moored near to the Maui-A platform, 32 km off the west s 5 11.38 1 5.357 for f , f p ,
fp
coast of the North Island of New Zealand, in a water
22

12
depth of 110 m. The wave measurements were accom- f
panied by wind measurements made with a Lambrecht s 5 32.13 2 15.39 for f $ f p . (79)
fp
anemometer installed on the platform at a height of 95
m above sea level. Thus, the angular separation is nonzero at all fre-
At this location, southeast wind events produce well- quencies but set equal to the fixed value of 14.938 for
defined, fetch-limited wind–seas, enabling a population f , f p . Functions (77), (78), and (79), constitute the
of 77 fetch-limited directional spectra to be established. symmetric double Gaussian distribution. These func-
The 77 spectra had significant wave heights ranging tions describe a distribution that is unimodal at and be-
between 0.54 m and 4.2 m, mean wave periods between low the peak frequency, but splits into two components
3.3 s and 6.9 s, and had associated vector average wind above the peak frequency, the angular separation of the
speeds ranging from 4.6 to 18.3 m s21 , and inverse wave peaks increasing with increasing frequency. We refer to
ages, U10 /c p , ranging from 0.70 to 1.4. this distribution as the Maui distribution in this paper.
Ewans estimated directional distributions of the 77
spectra using both maximum entropy and maximum 6) EXACT-NL
likelihood methods, and in both cases found the distri-
butions were unimodal at the peak frequency, consistent It has been proposed (e.g., Hasselmann et al. 1980;
with previously proposed distributions, but were bi- Donelan et al. 1985) that nonlinear wave–wave inter-
modal at frequencies above the spectral peak. This result actions may play an important part in determining the
supported those of Young et al. (1995), who observed wave directional distribution. The results of Banner and
bimodal spectra in wave data collected with a wave Young (1994) and Young et al. (1995) have confirmed
gauge array in Lake George, Australia. The lake had a this to be the case. In addition, Banner and Young
water depth of 2 m and the reported waves had inverse showed that the directional distribution predicted by a
wave ages in the range 1.7 , U10 /c p , 3; the Maui numerical model that gives a full solution to the non-
results thus extended the Young et al. observations to linear source term, the so-called Exact-NL model, is
open ocean conditions. unimodal at f / f p 5 1 but bimodal at higher frequencies,
Based on curve fits to the Maui data, Ewans has pro- and this prediction was generally supported by the field
posed a double-peaked form for the directional distri- observations reported in Young et al. (1995). We present
bution, based on a symmetric double Gaussian function results in this paper from a directional distribution, cal-
defined as follows: culated using such a numerical model, and refer to it as
the Exact-NL directional distribution.
H( f, u)

O `
1 7) WAM
5
Ï8p s ( f ) k52`
Wave hindcasting is now generally accepted as a valid
3 [exp(21/2[(u 2 u m1 ( f )22pk)/s ( f )] 2 ) approach to producing a long-term wave database from
which design wave criteria can be derived. Third-gen-
1 exp(21/2[(u 2 u m2 ( f )22pk)/s ( f )] 2 )], eration models such as the WAM model (Hasselmann
(77) et al. 1988) are now widely used for both wave hind-
casting and forecasting, but for computational efficien-
where s ( f ) is the angular width and is a measure of cy, the nonlinear interactions source term is approxi-
the spreading of each component, and u m1 ( f ) and u m2 ( f ) mated in these operational models. While there is plenty
are the locations of the peaks, centered at equal angles of evidence (e.g., Hasselmann et al. 1988) that this does
on each side of the mean wave direction. The parameters not seriously compromise the accuracy with which the
APRIL 1998 FORRISTALL AND EWANS 449

FIG. 2. Circular rms spreading as a function of frequency for the model distributions.

significant wave height is predicted, there has been no with the Mitsuyasu and Hasselmann distribution was
systematic evaluation of the spreading factor, and it is due to the measurements with three-component systems,
therefore of interest to compare the shape of the direc- such as the pitch-and-roll buoys used by Hasselmann et
tional spreading associated with such models with Ex- al. (1980) and Mitsuyasu et al. (1975), not providing as
act-NL and with the proposed empirical distributions. much resolution as the 14-element spatial array used by
We present fetch-limited runs of the PHIDIAS model, Donelan et al. (1985). The analysis reported by Young
which is an implementation of the WAM algorithm (van (1994) demonstrates how progressively increasing the
Vledder et al. 1993). number of measurement elements progressively increas-
es the resolution of the directional distribution, but the
circular rms spreading parameter s is a function only
b. Circular rms spreading of a1 and b1 and should be estimated as well with a
1) COMPARISON OF THE PROPOSED DISTRIBUTIONS well-calibrated pitch-and-roll buoy as any spatial array
with three or more elements. The Maui data support
The circular rms spreading, s ( f ), defined in Eq. (62) this. The dotted line in Fig. 2 is the Maui distribution
is a useful measure of spreading. It does not require any and the plus symbols give the average of the 77 Maui
a priori assumption on the shape of the directional dis- spectra. Except for estimates at frequencies less than the
tribution, and it is often available from hindcast studies peak frequency, the Maui data indicate lower spreading
and wave directional measurement programs. A plot of than the Donelan distribution, which was derived from
s ( f / f p ) for each proposed distribution is given in Fig. a 14-element array. Both the Exact-NL and PHIDIAS
2. All curves show a spreading minimum near the spec- distributions are broadly consistent with the empirical
tral peak frequency ( f / f p 5 1.0) and increasing spread- distributions, but the Exact-NL spreading is lower than
ing at higher and lower frequencies. The Mitsuyasu and the PHIDIAS spreading at higher frequencies.
Hasselmann distributions are given for an inverse wave The Maui parameterization and data are also consis-
age of 0.8 and 1.2. The Mitsuyasu curves reflect the tent with the other distributions but show rather less
large range of directional spreading with wave age that spreading at the peak and at higher frequencies. A sim-
occurs for that distribution; the Hasselmann curves ilar plot given by Ewans (1995) but with the Maui data
show only a variation at frequencies above the peak grouped into inverse wave age categories indicated that
with wave age; while the Donelan and Donelan–Banner the fetch-limited Maui data were not dependent on wave
curves show the absence of any variation of wave age, which is consistent with the observations of Don-
spreading with wave age and also generally lower elan et al.
spreading values than the Mitsuyasu and Hasselmann For practical purposes it is often useful to refer to the
distributions. Donelan et al. (1985) speculated that the integrated circular rms spreading, s, of the sea state
lower spreading of their distribution when compared defined by
450 JOURNAL OF ATMOSPHERIC AND OCEANIC TECHNOLOGY VOLUME 15

FIG. 3. Integrated circular rms spreading as a function of wave age for the model distributions.

s 5 Ï2(1 2 m1 ), (80) only an Exact-NL calculation of s for one value of the


wave age, but the results of Banner and Young (1994)
where m1 5 Ï a 1 b1 , and a1 and b1 are the weighted
2 2
1
indicate that s decreases slightly as wave age increases.
integrals of the first two Fourier coefficients, defined as
follows: Their results also show that the details of the source

E
function parameterization have a significant influence
`
on the spreading. In contrast, the PHIDIAS model pro-
a1 5 E( f )a1 ( f ) df . duces a slight increase in s over the limited range of
0
wave ages for which the calculations were made.

b1 5 E 0
`

E( f )b1 ( f ) df .
2) DIRECTIONAL SPREADING FACTOR

In the case of the Mitsuyasu, Hasselmann, Donelan–


The curves of f( f / f p ) for the proposed distributions
Banner, and Maui distributions, the omnidirectional fre-
are plotted in Fig. 4. They demonstrate similar behavior
quency spectrum, E( f ), was assumed to take the JON-
to the circular rms spreading curves, with the exception
SWAP form, with parameters as defined by Hasselman
et al. (1973)—that is, with the Phillips parameter and of the sidelobes about the peak that occurs in some of
peak frequency defined by the wind speed and fetch, the curves. These sidelobes result from the contribution
and the peak shape parameters set equal to the JON- to the variance along the principal wave direction by
SWAP constants. wave energy traveling at angles greater than 908 to the
Plots of s for each of the distributions and the Maui principal wave direction. Physically, f is directly pro-
data are given in Fig. 3. The effects seen in the fre- portional to the ratio of the wave variance in the direc-
quency-dependent circular rms spreading s ( f / f p ) are tion of the principal wave direction, var p , to the total
also seen in Fig. 3. The Mitsuyasu values show a large wave variance; so wave energy traveling opposite to the
range, decreasing with increasing wave age; the Has- principal wave direction also contributes to var p . Thus,
selmann curve shows a much smaller range, but still a while this wave energy is traveling at large angles rel-
gradual decrease with wave age; and the Donelan and ative to the principal wave direction, and will have a
Donelan–Banner curves show no dependency on wave high circular rms spreading accordingly, it causes var p
age, with the Donelan distribution giving the slightly to increase and hence f to increase.
lower constant value of 27.38. The Maui parameteriza- Figure 4 shows that although the distributions are
tion gives a constant value of 27.68, which is slightly similar in shape, their values differ significantly. The
larger than the Donelan values, but less than the Banner, Exact-NL, PHIDIAS, and Hasselmann distributions
Exact-NL, Hasselmann, and PHIDIAS curves. We have have similar values in the vicinity of the spectral peak
APRIL 1998 FORRISTALL AND EWANS 451

FIG. 4. Spreading factor as a function of frequency for the model distributions.

( f / f p 5 1), but are significantly lower than the Donelan defined by Hasselmann et al. (1973), in the same way
and Maui distributions. as for the calculation of s.
The calculation of forces for engineering purposes Plots of f as a function of wave age for each of the
requires the directional spreading factor integrated over distributions and the Maui dataset are given in Fig. 5.
the spectrum, as defined in Eqs. (50)–(52). To perform Some interesting contrasts are evident when this figure
this, the omnidirectional frequency distribution was as- is compared with Fig. 3. In terms of f, the Maui dis-
sumed to take the JONSWAP form, with parameters as tribution shows less directional spreading by compari-

FIG. 5. Integrated spreading factor as a function of wave age for the model distributions.
452 JOURNAL OF ATMOSPHERIC AND OCEANIC TECHNOLOGY VOLUME 15

son with the Donelan–Banner and Exact-NL distribu- second moments of the spreading function, a properly
tions, which is opposite to that indicated by s. The functioning three-element system measures it as accu-
different behavior for the two measures of spreading is rately as a spatial array with many elements.
due to differences in the shapes of the directional spread- On the basis that nonlinear wave–wave interactions
ing functions. control the directional spreading, it would appear that
The effects seen in the frequency-dependent s ( f ) are the Donelan–Banner and Maui distributions give more
also seen in Fig. 5. The Mitsuyasu values show a large realistic estimates of the directional spreading values
range—from the theoretical minimum of 0.707 to the for fetch-limited sea states. For these formulas there is
theoretical maximum of 1.0. The Hasselmann curves no dependence on wave age and, therefore, under fetch-
show a much smaller range, but still a gradual increase limited conditions, a sea state with a short fetch and,
with wave age, with values ranging from 0.85 to 0.87. for example, a significant wave height of 1 m has the
The Donelan–Banner, Donelan, and Maui curves show same directional spreading factor as the sea state further
no dependency on wave age and have values of 0.903, down the fetch with a 10-m significant wave height. For
0.907, and 0.906, respectively. The Maui data have a the particular case of fetch-limited waves, we thus pro-
mean value of 0.910. The Exact-NL calculation gives pose that a good estimate for the directional spreading
a spreading factor of 0.887, nearly as large as the recent factor is 0.906, the average of the Donelan–Banner and
measurements, but the spreading factors from the PHID- Maui measured values.
IAS 3G model are significantly lower than the mea-
surements.
4. Data sources and processing
c. Discussion The number of directional wave measurements has
greatly increased over the last few years. There are now
Young (1994) cites reports of investigations into the
enough good quality observations to make the calcu-
development of the directional spectrum under fetch-
lation of meaningful statistics for different storm types
limited conditions using a spectral model with a full
and intensities possible. Figure 6 shows the sites where
solution to the nonlinear source term. These studies con-
the datasets used in this study were collected. The sites
clude that the directional spreading is almost completely
are all on the continental shelf or slope, but the water
controlled by nonlinear wave–wave interactions, with
depths are generally greater than 100 m, so that refrac-
atmospheric input playing only a very minor role. This
tion or shoaling should have little effect on the waves.
suggests that any dependence of directional spreading
The sites cover a wide range of geographical areas from
on wave age should be small. Young (1994) points out
the Tropics to the Arctic.
the small range of wind–sea wave ages involved in the
Tables 2, 3, and 4 give more details of the datasets.
Mitsuyasu data, bringing the accuracy of the strong de-
The tables are divided into locations dominated by ex-
pendence on wave age in the Mitsuyasu formulation into
tratropical storms, tropical storms, and a tropical area
question. The strong wave age dependence of the Mit-
(northwest Borneo) under the influence of seasonal
suyasu spreading formulation is not supported by other
monsoon winds. The first column of the tables gives the
measurements or by the numerical wave model exper-
names of the sites and the next two columns give their
iments, and Young concludes that the directional spread-
locations. The time of the measurements is given in the
ing distributions are more likely to conform to the Don-
form of months per year. For most sites, the time is
elan or Hasselmann formulations.
given as a span of several months or years, but the data
The Hasselmann and Mitsuyasu formulations are in
at some sites, particularly for tropical storms, were col-
reasonable agreement over the parameter range common
lected over a few days in the month indicated. Most of
to both measurement sets, but both distributions are
the sites had nearly continuous records at a 1- or 3-h
broader than the Donelan formulation and also the Maui
recording interval. The instruments and data types are
results. Young (1994) performs numerical experiments
explained in the following sections.
that show that improvements in the spectral resolution
available from three element systems, such as a pitch-
and-roll buoy, can be achieved with spatial arrays that a. Instruments
have more than three elements. This result is reasonably
well known and is given as an explanation for the Do- A wide variety of instruments have been used to mea-
nelan et al. (1985) results producing a narrower spread- sure the directional properties of waves. In shallow wa-
ing than the Mitsuyasu and Hasselmann distributions. ter, arrays of pressure transducers or wave staffs have
However, the Maui data were recorded using a pitch- been popular. Satellite and radar sensors can also mea-
and-roll buoy, and they show directional spreading val- sure directional wave spectra. All of the data considered
ues that are narrower than the Hasselmann and Mitsuy- in this report, however, came from in situ sensors in
asu formulations and are in fact in better agreement with deep water. Some of these sensors were mounted on
the Donelan–Banner distribution. Since our definition deep water oil production platforms and others were on
of the directional spreading factor depends only on the floating buoys.
APRIL 1998 FORRISTALL AND EWANS 453

FIG. 6. Locations of wave-spreading data are shown as the filled circles.

1) HUV (WAVE STAFF AND CURRENT METER) tured by Marsh-McBirney, Inc. Baylor inductive wave
staffs were used except for the measurements at Tern,
Since their development fast response current meters
where an EMI laser was used.
have been used with wave staffs to estimate the direc-
tional wave spectrum. Electromagnetic current meters
were first used by Bowden and White (1966) in the 2) NORWAVE
intertidal zone at the mouth of the Mersey River. The Norwave is a 2.75-m buoy with a Datawell Hippy
Acoustic current meters, pressure transducers, and la- 120-A pitch-and-roll sensor. It was manufactured by
ser wave staffs were mounted on the Edda platform in Bergen Ocean Data. In the comparisons by Allender et
the central North Sea as a part of the comprehensive al. (1989), this buoy gave broader directional spreading
WADIC experiment described by Allender et al. (1989). than most of the other instruments. This directional
WADIC also included a thorough comparison of the broadening was apparently due to electronic noise that
pitch-and-roll buoys that were available at the time of periodically appeared in the signal. We suspect that this
their experiment in 1985. Allender et al. (1989) com- noise was due to some problem with the electronics on
bined data from the array of five lasers and five pressure this particular deployment of the buoy. Nevertheless,
cell/current meter triplets to form their ‘‘best estimate we will see that the wave-spreading factors observed at
data set (BEDS).’’ Due to instrument malfunctions, most Haltenbanken with this type of buoy were lower than
estimates used only three lasers. Directional spreading those at any other site.
at low frequencies was calculated from the current me-
ter/pressure cell triplets and the spreading at high fre-
quencies was calculated from the laser array. 3) WAVEC
All of the other HUV measurements listed in Tables The WAVEC is a 2.5-m buoy manufactured by Da-
2 and 3 used electromagnetic current meters manufac- tawell, again with a Datawell Hippy 120-A sensor.

TABLE 2. Extratropical storm sites, negative latitudes, and longitudes in the Southern Hemisphere and western hemisphere, respectively.

Site Latitude Longitude Time Instrument Data type


Nordkappbanken 728009 318009 5/88–5/90 Norwave Fourier
Nordkappbanken 728009 318009 5/90–5/92 WAVEC Fourier
Haltenbanken 658509 78349 3/80–3/88 Norwave Fourier
Tern 618169 08559 1/92–1/93 HUV Time series
North Cormorant 618149 18099 1/91–5/95 WAVEC Fourier
Fulmar 568309 28079 11/81 HUV Time series
WADIC 568289 38069 9/85–1/86 HUV,buoys Fourier
Auk 568249 28039 1/91–5/95 WAVEC Fourier
46050 44.68 2124.58 11/91–6/92 3–m discus Fourier
44025 40.38 273.28 4/91–6/92 3–m discus Fourier
Maui 2398369 1738239 10/86–11/87 WAVEC Time series
44001 38.48 273.68 9/90–4/91 3–m discus Fourier
44015 37.58 273.48 9/90–4/91 3–m discus Fourier
46042 36.88 2122.28 4/88–6/92 3–m discus Fourier
44014 36.68 274.88 10/90–6/92 3–m discus Fourier
454 JOURNAL OF ATMOSPHERIC AND OCEANIC TECHNOLOGY VOLUME 15

TABLE 3. Tropical cyclone sites.

Site Latitude Longitude Time Instrument Data type


SP62 298059 2888449 9/74, 8/75 HUV Moments
Buccaneer 288539 2948429 9/73 HUV Moments
Cognac 288479 2898039 9/79 HUV Moments
EI331 288149 2918439 9/74, 8/75 HUV Moments
Bullwinkle 278539 2908549 8/92, 10/95 HUV Time series
North Rankin 2198359 1168089 2/91, 2/92 DWR Time series

Wave-spreading estimates from it were similar to the widths measured by the buoy and the platform agreed
BEDS estimate in the Allender et al. (1989) dataset, reasonably well. The platform measurements had slight-
although there were relatively few data points in the ly wider spreading.
comparison since the buoy capsized several times in the O’Reilly et al. (1996), however, found that the inte-
WADIC experiment. The mooring systems for the buoy grated circular rms spreading measured by an NDBC
have since been improved, and the data return for the buoy was significantly different from the spreading mea-
WAVEC deployments listed in Tables 2 and 4 was quite sured by the pressure transducer array on the Harvest
good. platform. The buoy measurements were biased 68 higher
than the measurements from the pressure transducer ar-
ray. This bias is roughly equivalent to the spreading
4) DIRECTIONAL WAVERIDER
factor measured by the buoy being biased 0.035 too
The directional waverider (DWR) is also manufac- low.
tured by Datawell, but it uses a three-axis accelerometer
instead of a pitch-and-roll sensor. The signal processor b. Data types
on the buoy performs a double integration of these mea-
surements to produce the north and east displacements The columns labeled ‘‘data type’’ in Tables 2–4 show
of the buoy in the wave orbits along with its heave. the form of the data that was input to our analysis. In
Barstow (1991) found that the data from the DWR com- some cases, the data were available in several different
pared well to that from the WAVEC. O’Reilly et al. forms. In those cases, we used the form of the data that
(1996) compared a DWR to measurements from an array had the least processing done to it. For example, if time
of six pressure transducers mounted on the Harvest plat- series of pitch and roll or horizontal velocity was avail-
form off California and found excellent agreement in able, we calculated the Fourier coefficients of the
the circular rms spreading integrated over the swell band spreading function from them and the spreading factor
from 0.06 to 0.14 Hz. Most of their measurements were from the Fourier coefficients. If the Fourier coefficients
in a long period swell with significant wave heights less were available, as on the CD-ROMs published by the
than 4 m. NDBC, they were used. Finally, if only the moments of
the spreading function were given, they were converted
to Fourier coefficients.
5) 3-M DISCUS (NATIONAL DATA BUOY CENTER) The length of almost all of the data records was be-
The 3-m discus buoy is the standard hull used by the tween 20 and 30 min. A few of the datasets, including
U.S. National Data Buoy Center (NDBC) for directional the Fulmar and Tern data as well as the hurricane data
wave measurements. The pitch-and-roll sensor in the from the Gulf of Mexico, had data records 1 h long.
buoy is a Datawell Hippy 40 Mark II. Anctil et al. (1993)
compared the measurements from one of these buoys 5. Results
with wave staff and current meter measurements on the
Bullwinkle platform. The sea states during the com- a. Extratropical storms
parison were low, with H s always less than 2 m. For Figures 7–11 are representative scatterplots of the
significant wave heights greater than 1 m, the directional spreading factor f plotted against the significant wave

TABLE 4. North coast of Borneo.

Site Latitude Longitude Time Instrument Data type


Kinebalu 58279 1148469 2/91–11/92 WAVEC Fourier
Samarang 58269 1148469 8/92–2/94 DWR Fourier
M Field 58229 1128119 7/92–3/94 DWR Fourier
Baram 48459 1138529 5/88–8/90 WAVEC Fourier
Baram Delta 48459 1138529 11/92–2/93 DWR Fourier
Baronia 48409 1138409 11/93–3/94 DWR Fourier
APRIL 1998 FORRISTALL AND EWANS 455

FIG. 7. Scatterplot of the directional spreading factor f against significant wave height at Auk.
The curves give the median, 75% and 90% exceedence levels, and maximum value of f for 0.5-
m-wide wave height bins.

FIG. 8. Scatterplot of the directional spreading factor f against significant wave height at North
Cormorant. The curves give the median, 75% and 90% exceedence levels, and maximum value of
f for 0.5-m-wide wave height bins.
456 JOURNAL OF ATMOSPHERIC AND OCEANIC TECHNOLOGY VOLUME 15

FIG. 9. Scatterplot of the directional spreading factor f against significant wave height at
Nordkappbanken. The curves give the median, 75% and 90% exceedence levels, and maximum
value of f for 0.5-m-wide wave height bins.

FIG. 10. Scatterplot of the directional spreading factor f against significant wave height at buoy
46042. The curves give the median, 75% and 90% exceedence levels, and maximum value of f
for 0.5-m-wide wave height bins.
APRIL 1998 FORRISTALL AND EWANS 457

FIG. 11. Scatterplot of the directional spreading factor f against significant wave height at buoy
46050. The curves give the median, 75% and 90% exceedence levels, and maximum value of f
for 0.5-m-wide wave height bins.

height. Each dot represents one data sample that is gen- agate; so at a distant location from the source, the di-
erally 20–30 min in length. Data for significant wave rectional spreading factor could be higher than the value
heights less than 2.0 m are not plotted in most of the of 0.906 for fetch-limited seas even if the waves at the
figures because of the huge number of data points re- source were produced under perfect fetch-limited con-
quired. The scatter in the values of the spreading factor ditions. Indeed, some active wind–sea states could be
is very large, particularly for low wave heights. The expected to include swell components that have a prin-
values of f range from close to the theoretical minimum cipal wave direction coincident to the wind–sea com-
of 0.707 to near the maximum of 1. ponent; this scenario would also produce directional
What accounts for the variability in the values of the spreading factors larger than 0.906.
directional spreading factor of the observed sea states? The orography of the location can also be expected
Clearly, the directional spreading will increase if the to influence the possible values of the directional spread-
wind direction is varying quickly in either space and ing. For example, a coastline near to a location may act
time. Rapidly turning winds will produce low spreading as a filter to some wave directions and tend to decrease
factors as the high-frequency waves follow the change the amount of spreading at the location for particular
of direction of the wind while the direction of the low- mean wave directions. Shallow water effects can also
frequency waves lags behind (e.g., van Vledder and be expected to affect the value of the spreading. For
Holthuijsen 1993). Thus, a sea state at a location ex- example, a decrease in spreading would be expected at
periencing a local wind–sea associated with a fast trav- progressively closer locations to the coast for a shoaling
eling tropical cyclone would be expected to have a lower bathymetry, while a complicated seabed topography
directional spreading factor than a sea state at a location might increase the spreading at some locations. The
in a long and steady fetch of an extratropical cyclone. measurements discussed in this paper were made in wa-
Very low values of the spreading factor can also be ter deep enough and far enough from coastlines that
produced when waves from two or more storm systems these effects should be small. They should, however, be
combine. Even if the individual modes of the spectrum considered carefully for any coastal locations.
are relatively narrow, the spreading factor for the com- The lines in Figs. 7–11 show statistics of the data
bined sea will be small if the modes propagate in dif- points calculated over 0.5-m bins in significant wave
ferent directions. height. The median value, 75 and 90 percentile values,
Conversely, as a wave field propagates outside the and maximum are shown. The median values increase
influence of the wind, the various directional compo- significantly with increasing wave height, while the 75
nents would be expected to radially spread as they prop- and 90 percentile lines are more nearly constant. These
458 JOURNAL OF ATMOSPHERIC AND OCEANIC TECHNOLOGY VOLUME 15

FIG. 12. Median values of the directional spreading factor f as a function of significant wave
height for the sensors deployed in the WADIC experiment.

features of the statistics can be explained by the fact The spreading factors from the BEDS data are much
that most of the low values of the spreading factor occur higher than those from any of the buoys or the wave
for low wave heights. The maximum per bin decreases staff array. BEDS was created using the current meters
for the highest wave heights because there are fewer for the low-frequency Fourier components and the wave
samples per bin. staff array for the high-frequency components. The
agreement between the spreading factors from BEDS
and current meter 2 in Fig. 12 indicates that the current
1) THE WADIC DATASET
meter data dominated the spreading estimates since the
The WADIC experiment (Allender et al. 1989) was low-frequency partition contained most of the wave en-
particularly important because it was the most thorough ergy. The large differences between the two platform
side-by-side comparison of directional wave sensors based estimates of the wave spreading prompted us to
ever made. Figure 12 shows the median values of the take a close look at the raw time series of the mea-
directional spreading factor as a function of wave height surements.
for most of those sensors. BEDS is constructed from Figure 13 shows 20 min of velocity measurements
the current meters and laser wave staffs, CM2 is the made by current meter number 2 at 15 m below mean
current meter at 15-m depth, and Array is the array of sea level at 1400 UTC 2 November 1985. Each point
laser wave staffs. All of the buoys measured pitch and in the plot is the tip of the horizontal projection of the
roll and differed mainly in the shape of the surface- velocity vector at a 2-Hz sampling rate. Plots of wave
following float. The Marex buoy was 3 m in diameter, particle velocities displayed in this way usually show
the Seatex Wavescan buoy and the Bergen Ocean Data bilateral symmetry since the velocities under wave
Norwave were 2.75 m in diameter, and the Endeco troughs are roughly equal and opposite to the velocities
Wavetrack was 0.76 m. under the crests. A scatterplot of the u and y components
The mean values of the spreading estimated from the of the velocity thus has an elliptical shape. The WADIC
various sensors differ considerably. These differences data, however, are not elliptical and, for the cloud of
are consistent with the comparisons in the scatterplots points to become elliptical, many more points would
given by Allender et al. (1989). The spreading factors have to be added to the left of the origin. There thus
calculated from the Endeco buoy are clearly too small, appears to be a distinct and systematic deficit of veloc-
possibly because of relatively high noise and problems ities in the negative u direction. Most of the other storm
with the mooring line. The data from the Norwave buoy wave data plots from WADIC that we examined exhib-
were corrupted by regularly occurring noise, which bi- ited similar velocity deficits, always in the same direc-
ases the spreading factor to the low side. tion. The problem was most evident when the waves,
APRIL 1998 FORRISTALL AND EWANS 459

measurements were made 12 km from the location of


the Auk buoy, and the Tern current meter measurements
were made 13 km from the North Cormorant (NC) buoy.
Only a few hours of the Fulmar and Tern measurements
were analyzed, but they were during very high waves.
Figure 14 compares the average values of the spreading
factor calculated from the instruments at these sites. The
Auk WAVEC agrees very well with the Fulmar current
meter and wave staff (‘‘huv’’) and the NC WAVEC
agrees very well with the Tern current meter and wave
staff.
The Fulmar and Tern current meters do not appear to
have had as much of a problem with platform interfer-
ence as the WADIC current meters. The Fulmar meters
were mounted on a taut wire loop stretched between the
deck and a horizontal platform member. The nearest
vertical platform member was about 10 m from the cur-
rent meters. Interference from the structure was thus
greatly reduced at the expense of some motion of the
mounting system. The Tern current meters were mount-
ed on vertical stand-offs about 3 m above horizontal
FIG. 13. Horizontal velocities measured by current meter 2 (215
framing members. Although the horizontal member pre-
m) at WADIC at 1400 UTC 2 November 1985. Each plotted point sumably affected the vertical component of the wave
is the tip of a velocity vector sampled at 2 Hz by the acoustic current orbital velocities, it apparently had little effect on the
meter. horizontal velocities measured by the current meters.
Scatterplots of the measurements at both Fulmar and
Tern in the form of Fig. 13 have the anticipated bilateral
and thus the orbital velocities and displacements, were symmetry.
large.
There is a ready physical explanation for the missing
points in the negative u direction since the current me- 3) LATITUDINAL DEPENDENCE OF THE SPREADING
FACTOR
ters were mounted on a space frame tower hung off the
side of the Edda platform. The tower was in the positive Figure 14 also shows a clear difference in the average
u direction in the local coordinate system, so flow to- spreading factors measured near Auk and those mea-
ward negative u was in the wake of the tower, and ve- sured near North Cormorant. The measurements must
locities in that direction were reduced. reflect a real difference between the wave climates at
The interference from the tower has a strong effect the two sites since the same type of buoy was used at
on the estimate of the wave-spreading factor. Rough Auk and North Cormorant. Furthermore, the measure-
calculations indicate that if the crescent-shaped data ments made with the current meters at Fulmar and Tern
cloud in Fig. 13 were filled in to make an ellipse, the during different years show almost the same difference
spreading factor would decrease from about 0.95 to between the sites.
0.90, and thus come into reasonable agreement with the Inspection of plots similar to Figs. 7–11 for all of the
slope array and most of the buoy measurements made measurement sites revealed further differences between
at the same time. Because of this flow interference with the statistics of the spreading factors at those sites. To
the current meter measurements in WADIC, we have our surprise, we found that these differences correlated
chosen to ignore the BEDS and current meter wave- very well with the absolute value of the latitude of the
spreading estimates in our data summaries and use the sites. Figure 15 shows the median: 75% and 90% values
union of the Array, Marex, and Wavescan data to rep- of the spreading factor for significant wave heights over
resent the wave spreading during WADIC. 5 m plotted as a function of the latitude of the mea-
surement site. Each measurement site is plotted as a
separate set of points. The straight lines in the figure
2) OTHER COMPARISONS OF BUOY AND CURRENT
are simple linear regressions through the data points.
METER ESTIMATES
The data points show an almost monotonic decrease
It is possible to compare the statistics of wave-spread- with increasing absolute value of latitude, despite the
ing measurements made with buoys and current meters fact that the sites are in both Northern and Southern
at two other locations in the North Sea, although the Hemispheres and on both east and west margins of
individual measurements cannot be compared since they oceans. Some of the sites are in semienclosed seas, and
were made at different times. The Fulmar current meter others have full exposure to open ocean over at least
460 JOURNAL OF ATMOSPHERIC AND OCEANIC TECHNOLOGY VOLUME 15

FIG. 14. Comparison of average spreading factors for buoys and current meters at sites in the
North Sea. Auk is close to Fulmar and North Cormorant is close to Tern. The buoys were WAVECs,
and huv denotes estimates from a wave staff and current meter.

FIG. 15. Statistics of spreading factors as a function of latitude. Only significant wave heights
greater than 5 m have been included in the statistics. The straight lines are simple linear regressions
through the data points.
APRIL 1998 FORRISTALL AND EWANS 461

FIG. 16. Median spreading factor for significant wave heights greater than 5 m at buoys 46050
and 46042 off the west coast of the United States, plotted as a function of wave age.

one semicircle. Yet the most important geographical fac- too surprising since no correlation was found in the
tor in determining the directional spreading of waves measurements for fetch-limited waves discussed above.
seems simply to be the latitude of the site. We also tried an explicit calculation of the swell frac-
Most of the data points at lower latitudes in Fig. 15 tion of the wave energy by calculating the Pierson–
come from measurements made by the NDBC 3-m dis- Moskowitz (PM) spectrum for the wind speed at the
cus buoys. If we accept the conclusion from the com- wave measurement site and ascribing all the energy in
parisons of O’Reilly et al. (1996) that the spreading the measured spectrum above the PM curve to swell.
factor in these measurements is biased about 0.035 too There was a tendency for high spreading factors when
low, then the decrease in spreading factor with latitude the swell fraction was very high, but the overall cor-
would be even stronger. In our analysis, however, we relation between swell fraction and spreading factor was
accept all of the measurements at face value, partly be- very poor.
cause the spreading factors from the 3-m buoys are al- Bimodal spectra are likely to have low spreading fac-
ready so high. Many of them are, in fact, higher than tors since there is a good chance that the different modes
any of those measured at the Harvest platform by have different directions of travel, but few of the di-
O’Reilly et al. rectional spectra we examined in detail were obviously
We investigated a number of possible reasons for the bimodal. Figure 17 shows two fairly typical examples
variation of the average spreading factor in high waves from a storm in January 1993 in which a 963-mb low
from site to site. It seems reasonable that low-latitude tracked north of Scotland to northern Norway. The con-
sites would receive more wave energy in the form of tour plots of directional spectra were plotted from a
swell from distant storms than high-latitude sites. Gen- cosine power-law parameterization of the spreading
erally, the wave age defined as c p /U10 is a good measure function at each frequency. Linear contour intervals
of wave development. Figure 16 shows median spread- were chosen to emphasize the energy-containing part of
ing factors at two buoys off the west coast of the United the spectra.
States plotted as a function of wave age. Only high North Cormorant was closer to the center of the storm
waves with significant heights greater than 5 m are in- and the waves were higher there than at Auk, which is
cluded in the statistics. There is some tendency for located farther south in the central North Sea. Both spec-
young waves to have slightly smaller spreading factors, tra show some evidence of bimodality in frequency, but
but the spreading at the two sites is still different for there is no sign of any swell trains propagating in dif-
the same value of wave age. The lack of correlation ferent directions from the wind sea in either of the spec-
between the wave age and the spreading factor is not tra. The spectra do not, in fact, appear different in any
462 JOURNAL OF ATMOSPHERIC AND OCEANIC TECHNOLOGY VOLUME 15

it does farther from the storm center. As the storm


moves, the wind direction at a fixed site also changes
more rapidly when the site is close to the center. The
results of these effects are difficult to see individually,
but a correlation between distance to the storm center
and the spreading function can be found.
To study the effect of storm distance on spreading,
we constructed Fig. 18 by studying the details of several
storms that had maximum significant wave heights over
5 m on the Pacific coast of the United States and in the
North Sea. The average spreading factor for the 12-h
period surrounding the maximum H s in these storms was
calculated. The distance to the storm center at the start
of the 12-h period was found by examining the synoptic
charts in the Global Historical Fields CD-ROM (Na-
tional Climatic Data Center 1994a). North Cormorant
is north of Auk and buoy 46050 is north of 46042, so
the storm centers are likely to be closest to North Cor-
morant and 46050, but there is considerable overlap in
the distances. For example, one storm did pass close to
46042 and produced waves with a low spreading factor
there. The correlation between the distance to the center
of the low (note the logarithmic scale) and the average
spreading factor at the peak of the storm is clear, al-
though there is considerable scatter.
The spreading factor tends to decrease with decreasing
distance from the storm center. For this fact to produce
the observed decrease of the spreading factor with latitude,
there must be more storms at high latitudes than at low
latitudes. This fact is commonly observed, but in order to
FIG. 17. Contour plots of directional wave spectra from North quantify it, we counted the number of storms in the Global
Cormorant and Auk during a storm in January 1993. The contours Tropical/Extratropical Cyclone Climatic Atlas CD-ROM
are autoscaled to the maximum energy density and plotted at linear
intervals. Higher contours are shown as darker lines.
(National Climatic Data Center, 1994b) that passed
through the longitude band between 08 and 108W. We
selected this longitude band instead of a band 108 to the
obvious way except that the NC waves are spread more east over the North Sea since the storm tracks in the North
broadly at every frequency, leading to a smaller spread- Sea did not appear to be complete in the atlas. Figure 19
ing factor. shows the number of storms that passed through 58 boxes
We made an automated search for bimodality by cal- of latitude between 08 and 108W from November through
culating the ratio of the average wave period to the peak January. The cumulative number of storms with minimum
period of the spectrum. The spreading factor showed no pressures lower than 950, 960, and 970 mb are shown.
correlation with this ratio. We also plotted the spreading
The points are plotted with an abscissa in the middle of
factor against wind speed, wind direction, wave direc-
the latitude box—for example, the points at 62.58 give the
tion, and the variation in wind direction over several
hours. None of these factors explained the difference in number of storms that passed between 608 and 658. The
wave spreading between the measurement sites. number of storms peaks at about 658N. This agrees quite
The one factor that did display some influence on the closely with the minimum in the spreading factor at Hal-
spreading factor was the distance traveled by the waves. tenbanken (658509N) as shown in Fig. 15.
Swell should become more tightly focused in both fre- We conclude that the spreading factors for waves gen-
quency and direction when it travels a long distance, erated by extratropical storms are relatively low when the
but the wave age does not, in general, give a good center of the storm passes near a site. The average spread-
measure of the distance traveled by waves. There are ing factor thus decreases at high latitudes, reaching a min-
other factors related to the relative locations of the storm imum at the latitude where storm tracks are most concen-
centers and the measurement sites that could also influ- trated. While low spreading factors near the center of a
ence the wave spreading. The isobars around a low have storm are intuitively reasonable, a detailed understanding
greater curvature closer to the storm center, so the wind of the physical processes involved awaits controlled ex-
direction changes more rapidly over the fetch there than periments with a carefully verified hindcast model.
APRIL 1998 FORRISTALL AND EWANS 463

FIG. 18. Average spreading factors for high waves at two sites in the North Sea and off the
Pacific coast of the United States as a function of the distance to the center of the storm.

FIG. 19. The number of storms that passed through boxes of 58 lat between 08 and 108 W from
November through January. The cumulative number of storms with minimum pressures less than
950, 960, and 970 mb is shown.
464 JOURNAL OF ATMOSPHERIC AND OCEANIC TECHNOLOGY VOLUME 15

FIG. 20. Spreading factors measured in tropical cylcones. The data points are averages over 0.5-
m bins in the individual storms, and the ‘‘average’’ curve is the average of the plotted points in
the same wave height bins.

b. Tropical cyclones would be interesting and valuable, but it is outside the


scope of this study. For our purposes, we wish to get a
Since tropical cyclones are relatively rare, we have
general measure of spreading in tropical cyclones by
much less data for them than for extratropical storms.
appropriate averaging of the data. To give the same
Each of the sites listed in Table 3 was under the influence
weight to the data from each storm regardless of the
of tropical cyclones for only a few hours. The storms
that affected the sites in the Gulf of Mexico were Delia number of hours of measurements, we first averaged all
(1973), Carmen (1974), Eloise (1975), Frederic (1979), of the data for each particular storm in 0.5-m wave
Andrew (1992), and Opal (1995). The Australian storms height bins. These averages for the individual storms
were Daphne (1991), Ian (1992), Bobby (1995), Frank are shown as the data points in Fig. 20. In most cases
(1995), Jacob (1996), and Olivia (1996). the averages were over only a few measurement inter-
The amount of data available varied widely from vals. The ‘‘average’’ curve in the figure is the average
storm to storm. For some of the early hurricanes in the of the plotted points in the same wave height bins. The
Gulf of Mexico, only a few hours of data surrounding spreading factors for significant wave heights greater
the storm peak were available, whereas continuous mea- than 4.5 m are somewhat smaller than those for waves
surements for several days with multiple buoys were below 4.5 m, and the spreading factors for waves heights
made in the recent Australian storms. There was a wide greater than 9 m appear to be still somewhat higher, but
variety of storm strengths and locations of the mea- these high waves occurred in only a few measurement
surement stations with respect to the storm track. All intervals in one storm. Any tendency for the spreading
of these variables affect the wave spreading. Before the factor to increase for very high waves thus appears to
local wind increases significantly, most of the wave en- be very slight. The highest waves in a tropical storm
ergy is likely to be swell with a fairly high spreading are experienced when the storm makes its closest ap-
factor. As the storm winds increase, so does the wind proach to the measurement site. The wind and locally
sea, but there is also likely to be significant swell energy generated seas are then at their highest, but there is still
still coming from the storm center. These bimodal spec- a significant amount of swell traveling in a different
tra have a lower spreading factor. We found that the direction than the local seas. The local wind direction
spreading factors in tropical storms were often lowest also turns relatively rapidly at that time. These factors
several hours after the highest waves were recorded, combine to keep the spreading factor relatively low dur-
showing that the sea state was most confused then. ing high waves.
A thorough analysis of all of the factors that cause The average of spreading factors for all of the data
changes in the spreading factor during a tropical cyclone points with wave heights greater than or equal to 5.0 m
APRIL 1998 FORRISTALL AND EWANS 465

FIG. 21. Spreading factors measured in by the WAVEC at Baram off the northwest coast of
Borneo.

in Fig. 20 is 0.8571. The average for waves above 7.0 with the maximum significant wave height barely over
m is 0.8666, and the average for the few waves over 3 m. The median values of the spreading factor are
9.0 m is 0.8820. The average spreading factor for the essentially flat with wave height, while the higher per-
highest waves in each storm is 0.8529. These values are centiles show some tendency to decrease with increasing
just slightly less than for sites in the central North Sea, wave height.
which is somewhat surprising since it has usually been Median values of the spreading function for all of the
thought that waves are much more broadly spread in sites off Borneo are shown in Fig. 22. All of the datasets
tropical cyclones. The range of the average spreading are remarkably similar to each other, with only the mea-
factors from different storms is quite large. This vari- surements from the M Field standing out as including
ability is caused by the different combinations of sea higher waves and having slightly lower values of the
and swell in the different quadrants of a tropical storm. spreading factor. While all of the measurement sites
A reasonably conservative reading of the data would were in relatively deep water at least 50 m deep, the M
be to use 0.8666 in force calculations for extreme waves Field was considerably farther from the shore than the
in force calculations in tropical cyclones. This is at the other stations, about 200 km as opposed to 50 km.
lower end of the range suggested in API RP2A-LFRD. Different instruments were used in different years at
two of the sites, so it is possible to make some com-
parison of the instrument response. The comparison
c. Monsoon conditions must be statistical since the instruments were not at the
A large amount of directional wave data collected off same site at the same time, but the apparent stability of
the northwest coast of Borneo was available for this the climate still makes the comparison valuable. The
study. The sites are near the equator, and tropical cli- Baram measurements were made with a WAVEC, while
matic conditions prevail throughout the year. Winds and the Baram D measurements were made with a Direc-
waves are dominated by the progression of the Asian tional Waverider. Similarly, the Samarang 1 measure-
monsoon. The southwest monsoon generally begins in ments were made with a WAVEC and the Samarang 2
May or June and extends through October. Then the measurements were made with a Directional Waverider.
northeast monsoon builds from the northern end of the In both cases, the median values of the spreading factor
South China Sea and northeast or north winds and waves are very similar, which supports the conclusion that
prevail over the area during the winter months. these two wave buoys based on different principles give
Figure 21 shows the spreading factors measured at the same results.
one representative site, Baram, using a WAVEC buoy. These sites appear to be so similar that the most rea-
It is immediately obvious that the wave climate is mild, sonable characterization of their spreading factors is
466 JOURNAL OF ATMOSPHERIC AND OCEANIC TECHNOLOGY VOLUME 15

FIG. 22. Median spreading factors measured at all of the sites off the northwest coast of
Borneo.

simply to put all of the data together and calculate its of the energy at all frequencies. If the spreading factor
statistics. The results are given in Table 5. The median at a particular depth is needed, then the spectrum should
spreading value of 0.8800 is slightly less than the fetch- be weighted by the attenuation factor derived from lin-
limited value. The difference between the median and ear wave theory. For calculating the drag force on a
90% values is small, which presumably reflects the rel- single pile or jacket structure, it is appropriate to simply
atively uniform wave climate in this area. weight the second moments by the wave spectrum.
The spreading factor can be calculated from the in-
formation given by any of the systems used to measure
6. Conclusions and recommendations
the directional properties of waves. Essentially all that
The square root of the in-line variance ratio defined is needed is the ratio between the in-line and total vari-
by Haring and Heideman (1980) gives the ratio between ance of the wave energy, so it is not sensitive to the
the in-line component of the velocity under directionally absolute calibration of the instrument. Systems capable
spread waves to the velocity under unidirectional waves of high-resolution definition of the directional spectrum
in a random sea. It is thus the part of the wave kinematics are not needed. NOAA 3-m discus buoys, for example,
factor defined in API RP2A-LFRD that is due to direc- measure spreading factors as high as 0.98 when they
tional spreading, and we have defined it as the wave- are in highly focused swell.
spreading factor, f. Wave particle velocity measurements Good evidence that the directional wave measurement
made during a period of high waves at Tern in the North systems that are now in general use give accurate mea-
Sea showed that the in-line component of the velocity surements of the wave-spreading factor comes from the
distribution was indeed reduced by the factor f. consistency of the measurements that we have exam-
It is important to note that the spreading factor is ined. Measurements of fetch-limited waves using a
calculated by integrating the second trigonometric mo- WAVEC buoy off the coast of New Zealand gave almost
ments of the directional spreading function over the exactly the same value of the spreading factor as the
wave spectrum since the velocity variances are the sum high-resolution wave staff array of Donelan et al.
(1985). The average values of the spreading factor mea-
sured by current meters and buoys at nearby sites in the
TABLE 5. Statistics of the wave-spreading factor for monsoon North Sea agree very well with each other, although the
waves.
measurements were made in different years. The av-
Median 0.8800 erage values from WAVEC and Directional Waverider
75% 0.8945 buoys deployed off the coast of Borneo also agree very
90% 0.9070
well with each other. Finally, except for two buoys with
APRIL 1998 FORRISTALL AND EWANS 467

obvious noise problems and current meters subject to tematic variation of the spreading factor with wave
interference from their mounting support, the systems height. For engineering design, the overall median value
tested in the WADIC experiment were in agreement. of 0.8800 can thus be used with confidence.
There are relatively few hours of tropical cyclone data
a. Fetch-limited waves available, but the existing data does not show any strong
dependence on wave height or site. We thus recommend
Several empirical descriptions of the directional using the average value of 0.8666 for waves over 7.0
spreading function for fetch-limited sea states are avail- m in engineering calculations. This value of the spread-
able in the literature. Of these, that of Donelan et al. ing factor is at the lower end of the range suggested in
(1985) was based on the highest resolution measurement API RP2A-LFRD.
system—a 14-element wave staff array. Our buoy mea- The situation for extratropical storms is more com-
surements at Maui generally agree with Donelan’s re- plicated. There appear to be real variations in the spread-
sults and show narrower spreading than the earlier work ing factor with location and significant wave height. The
of Mitsuyasu et al. (1975) and Hasselmann et al. (1980). median spreading factor increases with significant wave
In agreement with Donelan et al., we find no evidence height at many of the sites. It would certainly be possible
that the directional spreading function depends on wave to fit a curve through the data, which could then be used
age. The measurements give an average spreading factor to estimate the median spreading for a given wave
of 0.906 for fetch-limited seas. height. But finding the spreading factor for the waves
The directional spreading function derived from the with a return period of 100 years or more that are used
Maui data is bimodal at frequencies somewhat above as design criteria would involve an extrapolation from
the spectral peak. These observations confirm those of the existing data. This extrapolation would eventually
Young et al. (1995), who found similar behavior using give a spreading factor greater than 1, which is physi-
a wave gauge array in Lake George, Australia. They are cally impossible.
also in agreement with high accuracy integrations of the We believe that more reasonable estimates of the
nonlinear interaction term in the wave evolution equa- spreading factor for very high waves can be made by
tion. Bimodal directional propagation may have impli- examining the characteristics of the distribution of mea-
cations for wave hindcast models. sured spreading values. The median value of the spread-
Precise (Exact-NL) integrations of the wave evolution
ing factor increases with wave height largely because
equation give spreading functions that are similar to,
there are many occurrences of low spreading factors in
but slightly broader than the Donelan and Maui func-
low confused seas. The curves giving the 75% and 90%
tions. The approximate integration used in the opera-
levels of the distribution are much flatter than the me-
tional third-generation WAM model, however, gives a
spreading factor of about 0.85, which is considerably dian, and the three tend to merge at high wave heights.
broader than the observations. The probability distribution function of f is thus steeper
for higher wave heights. This is illustrated explicitly in
Fig. 23, which shows the distribution function of f for
b. Latitudinal variation in wave spreading different wave height ranges at Auk. As the wave height
The thousands of measurements we have examined increases, the distribution functions become steeper, so
show great scatter in their calculated spreading factors, that there is less difference between, say, the 50% and
but also large and systematic differences in the statistics 90% levels. In fact, the curves appear to be approaching
of the factor from site to site. At sites exposed to ex- a nearly constant value through the middle percentages
tratropical storms, there is a clear tendency for the with short tails at high and low probabilities. It thus
spreading factor to be lower, that is, for the waves to seems reasonable to take the 90% probability level of
be more widely spread, at higher latitudes. This ten- the spreading factor as the limit to which the spreading
dency is related to the facts that there are more storm factor will converge for the highest waves at a site. For
tracks at high latitudes and that the spreading factor statistical stability, the 90% level can be calculated over
tends to be smaller when the site is close to the storm all waves greater than some moderately high value, for
center. Waves from distant storms are less spread be- example, 5 m.
cause swell becomes more focused as it travels and be- The statistics of the spreading factor show substantial
cause the wind direction changes less rapidly far from differences from site to site. The best site-specific in-
the center of a low pressure system. formation will come from a long series of measurements
at that site. For sites near those we have analyzed, the
figures in Table 6 can be used. The table gives the sta-
c. Recommended values for engineering design work tistics for significant wave heights greater than 5 m at
The statistics of the spreading factor in low-latitude each site.
monsoon conditions and in tropical cyclones appear to The correlation of spreading factor with latitude is so
be rather simple. The statistics at all of the monsoon strong that we believe it can be used to predict the
sites are quite similar, and there appears to be no sys- spreading factor at other locations. The regression lines
468 JOURNAL OF ATMOSPHERIC AND OCEANIC TECHNOLOGY VOLUME 15

FIG. 23. Probability distribution of spreading factors measured at Auk. Each curve is the sample
distribution for a measurement with significant wave heights between the cell lower limit in the
legend and the next higher cell limit.

in Fig. 15 are given in Eqs. (81)–(83), where l is the ing factor for use in the design of space frame structures
latitude in degrees: are
f50 5 0.9948 2 0.00237|l|, (81) R low-latitude monsoons: f 5 0.880,
R tropical cyclones: f 5 0.867, and
f75 5 1.0144 2 0.00230|l|, (82) R extratropical storms for 36 , |l| , 72: f 5 1.0193
and 2 0.00208|l|.
f90 5 1.0193 2 0.00208|l|. (83)
d. The use of hindcast models
These equations should not be used outside the range
of 36 , |l| , 72 in which the measurements used to Many design criteria are now based on hindcasts of
derive them were made. It does, however, seem reason- waves made by numerical models. Hindcasts have the
able that the spreading factor could be very close to 1.0 advantage of providing a very long time series of data
at low-latitude sites where the wave climate is totally for extreme value analysis at a small fraction of the cost
dominated by swell, such as in the Gulf of Guinea. of an extended measurement program. Using the direc-
In summary, our recommended values of the spread- tional spectra produced by these hindcasts would offer

TABLE 6. Spreading factors for significant wave heights greater than 5 m at extratropical sites.

Site Latitude (deg) 50% 75% 90%


44014 36.6 0.8959 0.9177 0.9276
46042 36.8 0.9253 0.9444 0.9564
44015 37.5 0.8988 0.9239 0.9428
44001 38.4 0.9226 0.9452 0.9548
Maui 239.6 0.8901 0.9211 0.9390
44025 40.3 0.8757 0.8978 0.9112
46050 44.6 0.8863 0.9157 0.9358
Auk 56.4 0.8803 0.8994 0.9127
WADIC 56.5 0.8781 0.8942 0.9048
North Cormorant 61.2 0.8368 0.8669 0.8891
Haltenbanken 65.8 0.8029 0.8308 0.8564
Nordkappbanken 72.0 0.8437 0.8675 0.8850
APRIL 1998 FORRISTALL AND EWANS 469

the distinct advantage of giving simultaneous values of Donelan, M. A., J. Hamilton, and W. H. Hui, 1985: Directional spectra
of wind-generated waves. Philos. Trans. Roy. Soc. London, Ser.
wave height, mean direction, and spreading factor. First, A, 315 509–562.
however, the accuracy of the spreading factors produced Ewans, K. C., 1995: Observations of the directional spectrum of fetch-
by the models must be verified, which is an exercise limited waves off New Zealand. Preprints, Fourth Int. Wave
that is beyond the scope of this report. We believe that Hindcasting and Forecasting Workshop. Banff, AB, Canada,
such a verification study would be very valuable, but Federal Panel on Energy Research and Development, 263–277.
Forristall, G. Z., E. G. Ward, V. J. Cardone, and L. E. Borgman, 1978:
based on our limited results for fetch-limited waves, we The directional spectra and kinematics of surface gravity waves
suspect that the spreading factors produced by third- in Tropical Storm Delia. J. Phys. Oceanogr., 8, 888–909.
generation wave models will be smaller than those ob- Guillaume, A., 1990: Statistical tests for the comparison of surface
served by instruments. Study of hindcast and observed gravity wave spectra with application to model validation. J.
Atmos. Oceanic Technol., 7, 551–567.
spreading factors may then lead to improvements in the Haring, R. E., and J. C. Heideman, 1980: Gulf of Mexico rare wave
algorithms used in wave models. return periods. J. Pet. Technol., 265, 35–47.
Hasselmann, D., M. Dunckel, and J. A. Ewing, 1980: Directional
Acknowledgments. This study would not have been wave spectra. J. Phys. Oceanogr., 10, 1264–1280.
Hasselmann, K., and Coauthors, 1973: Measurements of wind-wave
possible without the enormous efforts that were made growth and swell decay during the Joint North Sea Wave Project
to collect the wave data we have studied. We do not (JONSWAP). Dtsch. Hydrogr. Z., Suppl. A, 8, 1–95.
know the names of all of the individuals involved, but Hasselmann, S., and Coauthors, 1988: A third-generation ocean wave
we do have a good idea of the amount of work they prediction model. J. Phys. Oceanogr., 18, 1775–1810.
did. Thank you. Holthuijsen, L. H., 1983: Observations of the directional distribution
of ocean-wave energy in fetch-limited conditions. J. Phys.
Ian Leggett provided the data from Auk, North Cor- Oceanogr., 13, 191–207.
morant, and Haltenbanken; Chris Graham provided the Jonathan, P., P. H. Taylor, and P. S. Tromans, 1994: Storm waves in
data from Borneo; Phil Jonathan provided the data from the northern North Sea. Seventh Int. Conf. on Behavior of Off-
Tern; and Mike Vogel and Charlie Gutierrez provided shore Structures, Massachusetts Institute of Technology, Boston,
MA, 481–494.
the data from Bullwinkle. Steve Buchan sent us the Khandekar, M. L., R. Lalbeharry, and V. Cardone, 1994: The per-
Tropical Cyclone Bobby data from North Rankin. Ste- formance of the Canadian spectral ocean wave model (SSOWM)
phen Barstow provided the WADIC data and discussed during the Grand Banks ERS-1 SAR wave spectra validation
the instrumentation system with us. Werner Hoekstra experiment. Atmos.–Ocean, 32, 31–60.
Kuik, A. J., G. Ph. van Vledder, and L. H. Holthuijsen, 1988: A
processed much of the data. Ian Young sent us the results method for the routine analysis of pitch-and-roll buoy wave data.
from the Exact-NL model and Gerbrant van Vledder J. Phys. Oceanogr., 18, 1020–1034.
sent the results from his PHIDIAS version of the WAM Longuet-Higgins, M. S., D. E. Cartwright, and N. D. Smith, 1963:
model. We profited from conversations on the interpre- Observations of the directional spectrum of sea waves using the
tation of directional wave data with Leo Holthuijsen. motions of a floating buoy. Ocean Wave Spectra, Prentice Hall,
111–136.
Peter Tromans showed us why the integrated drag force Mardia, K. V., 1972: Statistics of Directional Data. Academic Press,
on a pile is independent of wave frequency for a given 353 pp.
wave height. Chris Shaw and Tom Tucker made helpful Marine Board, 1982: Proceedings of the National Research Council
comments on earlier drafts of the manuscript. Workshop on Wave Measurement Technology. National Re-
search Council, 248 pp.
Mitsuyasu, H., F. Tasai, T. Suhara, S. Mizuno, M. Ohkuso, T. Honda,
and K. Rikiishi, 1975: Observations of the directional spectrum
REFERENCES of ocean waves using a cloverleaf buoy. J. Phys. Oceanogr., 5,
750–760.
Allender, J., and Coauthors, 1989: The WADIC project: A compre- National Climatic Data Center and Fleet Numerical Meteorological
hensive field evaluation of directional wave instrumentation. and Oceanography Detachment, 1994a: Global Historical Fields,
Ocean Eng., 16, 505–536. NCDC, Ashville, NC, CD-ROM.
Anctil, F., M. A. Donelan, G. Z. Forristall, K. E. Steele, and Y. Ouellet, , 1994b: Global Tropical/Extratropical Cyclone Climatic Atlas.
1993: Deep-water field evaluation of the NDBC-SWADE 3-m NCDC, Ashville, NC, CD-ROM.
discus directional buoy. J. Atmos. Oceanic Technol., 10, 97–112. Nolte, K. G., and F.-H. Hsu, 1979: Wave particle velocity statistics
API, 1993: Recommended Practice for Planning, Designing and Con- in a random sea. Proc. ASCE Conf. on Structural Safety and
structing Fixed Offshore Structures—Load and Resistance Fac- Reliability, Tucson, AZ, American Society of Civil Engineers,
tor Design. American Petroleum Institute, 225 pp. 1–23.
Banner, M. L., 1990: Equilibrium spectra of wind waves. J. Phys. O’Reilly, W. C., T. H. C. Herbers, R. J. Seymour, and R. T. Guza,
Oceanogr., 20, 966–984. 1996: A comparison of directional buoy and fixed platform mea-
, and I. R. Young, 1994: Modeling spectral dissipation in the surements of Pacific swell. J. Atmos. Oceanic Technol., 13, 231–
evolution of wind waves. J. Phys. Oceanogr., 24, 1550–1561. 238.
Barstow, S. F., and T. Kollstad, 1991: Field trials of the Directional van Vledder, G., and L. H. Holthuijsen 1993: The directional response
Waverider. Proc. First Int. Offshore and Polar Engineering of ocean waves to turning winds. J. Phys. Oceanogr., 23, 177–
Conf., Edinburgh, United Kingdom, ISOPE, 55–63. 192.
Bowden, K. F., and R. A. White, 1966: Measurements of the orbital Young, I. R., 1994: On the measurement of directional wave spectra.
velocities of sea waves and their use in determining the direc- Appl. Ocean Res., 16, 283–294.
tional spectrum. Geophys. J. Roy. Astron. Soc., 12, 33–54. , L. A. Verhagen, and M. L. Banner, 1995: A note on the bimodal
Cavaleri, L., L. Bertotti, and P. Lieonello, 1991: Wind wave cast in directional spreading of fetch-limited waves. J. Geophys. Res.
the Mediterranean Sea. J. Geophys. Res., 96, 10 739–10 764. 100, 773–778.

S-ar putea să vă placă și