COASTAL

ENGINEERING
ELSEVIER Coastal Engineering 22 (1994) 201-215

On the theory of the joint probability of heights and

periods of sea waves
Constantine D. Memos
Department of Civil Engineering, National Technical University of Athens, Athens, Greece
(Received 24 July 1992; accepted after revision 24 June 1993)

Abstract

The determination of the joint probability density of wave heights and periods in the short-term
statistics of sea waves of any band-width is of great importance in modelling coastal wave processes,
environmental loading on floating vessels, etc. This research is based on known probability relations
as well as on simple theoretical considerations. Assuming Gaussian waves of any band-width, a set
of differential equations has been formulated giving the desired probability density function of wave
heights and periods. Non-symmetrical wave heights about the mean can be accommodated, while
some simplifying approximations are discussed. Numerical results are presented in a companion paper
appearing in this issue.

1. Introduction

Many coastal engineering problems, ranging from processes such as shoaling and wave
breaking to vessel motions, involve statistics of the sea surface associated with both wave
height and wave period. It is, therefore, quite clear that in a random sea better knowledge
of the joint probability of height and period would improve substantially the available
methods dealing with relevant coastal processes.
The theoretical joint distribution proved to be an intractable problem for the broad-band
sea, whilst some progress has been made for the narrow-band spectrum. The main investi-
gations of this problem resulted in the following four approximations: Longuet-Higgins
(1975), Cavani6 et al. (1976), Lindgren and Rychlik (1982) and Longuet-Higgins (1983).
In the first paper Longuet-Higgins developed previous results by Rice ( 1944, 1945) for
noise in electrical circuits and produced a joint density function for narrow spectra. He
assumed statistically independent phases of the constituent individual waves, thus producing
a linear Gaussian model.
Comparison of his result with visual observations of wave properties in a narrow banded

0378-3839/94/$07.00 © 1994 Elsevier Science B.V. All rights reserved

SSDI0378-3839 (93) E0025E
202 C.D. Memos / Coastal Engineering 22 (1994) 201-215

spectrum published by Bretschneider (1959) showed good agreement. However, the under-
lying assumption of narrow-banded frequency spectrum renders the result not good for
unfiltered records of surface elevation Goda (1978). Also, the expression for the probability
density function is symmetrical about a non-dimensionalized characteristic period. This
implies in effect that the correlation between wave height and period is zero, a deduction
not supported by real life data. In fact, there appears to be a strong tendency for individual
waves of small heights to have short periods, and so the joint probability density function
of wave heights and periods deviates from the symmetrical shape indicated by the above
result. A reformulation of this theory was provided by Longuet-Higgins (1983) by express-
ing differently the zero-upcrossing local wave period in order to get rid of the symmetry in
the resulting joint density. His new formulation, based on envelope theory, can simulate the
characteristics of observed waves, in such a way that short-period waves tend to have
smaller heights. The distribution agrees well with wave data taken in the North Atlantic
(Chakrabarti and Cooley, 1977) and in the Sea of Japan (Goda, 1978). Nevertheless, there
are still shortcomings in the theory, as for example the overall correlation coefficient between
wave heights and periods is slightly negative, which causes the significant period to be
shorter than the mean period, in contrast with the observed characteristics of sea waves
(Goda, 1985).
Both results of Longuet-Higgins have been used by various researchers for applications
or further elaboration. For example, Swift (1988) used both expressions to calculate cumu-
lative probability of peak wave forces on cylindrical elements. Tayfun (1990) used the
same technique to induce asymmetry as that of Longuet-Higgins and obtained an expression
for the joint density function of large wave heights and periods.
Cavani6 et al. (1976) define a wave period based upon the elevation and acceleration at
the crest, by assuming a sine curve there. The joint density of elevation and acceleration for
a narrow-band spectrum is given by Rice ( 1944, 1945) from which Cavani6 et al. derived
the joint density of wave heights and periods. The results agreed rather well with analysis
of 200 twenty-minute storm wave recordings in the North Sea. A practical problem with
this formulation lies in estimating the value of the spectral width parameter which depends
on the fourth moment of the variance spectrum. The values of this parameter are thus more
descriptive of the high frequency cut-off than of the sea state as a whole. In fact, for the
usual f - 5 high frequency tail, f = frequency, the fourth moment of the spectrum tends
slowly to infinity as the integration is taken to higher frequencies.
Comparisons of the distributions given by Cavani6 et al. and by Longuet-Higgins with
observations and with numerically derived data are given by Shum and Melville (1984)
and by Srokosz and Challenor (1987). They found in general good agreement in the region
of high waves of a narrow spectrum.
In the last of the previously mentioned papers Lindgren and Rychlik (1982) give approx-
imations to the theoretical probability density within bounds, regardless of bandwidth,
assuming a Gaussian sea surface. The solutions are based on a certain model process, and
can be applied to bi-modal spectra as well, but are not closed form relationships. They are
expressed in terms of wave height and period defined between a crest and the following
trough. This definition, including negative maxima and positive minima of the elevation, is
a serious drawback for engineering applications, since it allows a great deal of high frequency
components to distort both the wave height and period probability structures. As with
 C.D. Memos~Coastal Engineering 22 (1994) 201-215 203

CavaniE's result, the existence of the fourth moment of the spectrum is required, which
implies that the method can only be applied if a high frequency cut-off is specified. Also,
in Lindgren and Rychlik's paper (1982) a constraint is implicitly assumed, namely that
within every length of record equal to the maximum period, as defined above, falls either
none or one minimum. In other words, the ratio of the maximum to the minimum wave
period in any record is less than 2. The numerical evaluation of the result is based on
knowledge of the entire spectrum rather than on a spectral width parameter. Therefore, a
considerable amount of computer time is needed for carrying out the required numerical
integrations.
In an older work, Rice and Beer (1965) used a different approach to establish the
distribution of wave heights without employing the above mentioned constraint on the wave
period. However, their procedure did not include the necessary computational aspects.
In the following, an analytical formulation for the joint density of the general broad-band
problem is presented under justifiable assumptions. This is based on probability relations
and simple theoretical considerations. Assuming Gaussian waves of any band-width, a set
of differential equations has been formulated describing the probability density function of
wave heights and periods. Some simplifying approximations are also given. A novel feature
appears to be the capability of the developed model to cope with non-symmetrical wave
heights about the mean. Numerical results are presented in a companion paper (Memos and
Tzanis, 1994) showing good agreement with real life data.

2. Theoretical development

The usual assumption is used that the sea surface elevation x above the mean level at a
fixed point is well represented by a Gaussian process x ( t ) at times t for which the environ-
mental conditions are well established, so that the process be regarded as statistically
stationary. No restriction is made regarding the narrowness of the variance spectrum of the
underlying process. The definition of the crest-to-trough wave height H and the period T
between successive up- (or down-) crossings of the free surface with the mean level is
given in Fig. 1. In this figure the ergodic process x ( t ) at a fixed point in the ocean is
represented.
It should be noted that in engineering applications it is quite customary to deal with the
wave height H rather than with crest elevations ~, since the former is the physical quantity
more closely related to the loading of the coastal structures.
The wave height H is formally defined as

H = s ~, -s~2 s~2<0 (1)

where points 1 and 2 denote the maximum and minimum of the process respectively, within
the corresponding wavelength T. The pair (H,T) is thus uniquely determined for every up-
crossing of the mean level. In general it is assumed that ~:~4: I ¢2 I.
In the following, the aim will be to obtain an expression for the joint probability density
function Q ( T , H ) of wave heights and periods as defined previously.
204 C.D. Memos/Coastal Engineering 22 (1994) 201-215

Let us define the following:

U ÷ (0,T I ~t) the probability that only positive x's occur within the open interval
(t,t + T), given that s¢2~<x ~<~:t there.
U - (0,T I ~2) the probability that only negative x's occur in the interval (t,t+ T), given
that ~2 ~<x ~<~ there.
U(1,T I ~,~2) the probability that exactly one zero-crossing occurs in (t,t+ T), given
that ~2 ~<x <~~ there.
Po ÷ ( T I ~1)dT the probability that a zero occurs in (t + T,t + T + dT), given that the last
zero occurred at t and that 0 < x ~ ~:~ in (t,t+ T).
Po- (TI~2)dT the probability that a zero occurs in (t + T,t + T + dT), given that the last
zero occurred at t and that ~:2~<x < 0 in (t,t + T).
p~(T[ ~1,~2)dT the probability that a zero occurs in (t + T,t + T + dT), given that the one
before the last zero occurred at t and that ~2 ~<x ~<~j, in (t,t + T).
q~- (TI ~ ) dTd~ the joint probability that a zero occurs in (t + T, t + T + d T ) given that the
last zero occurred at t, and the local extreme occurs in ( ~ , ~ + d ~ ) given
that O<x~< ~ in ( t , t + T ) .
For sea waves U + can be assumed differentiable with respect to T. Therefore, one can
write:
OU + ( O,T) ~, )
U + (O,T+ dTI ~ ) - U ÷ (O,T I ~,) - dT (2)
0T

If by No we denote the number of zero-crossings per unit time, then E [ No] dT, E indicating
the expected value, is the probability of a zero occurring in the interval (t + T,t + T + dT).
Obviously the probability of finding more than one zero in an infinitely small interval is

Xl X X3 Xn

,t
T2 D \ l Tn /~n+l
\
t/ )

'i
J

T
)II

Fig. 1. Definition of wave characteristics.

 C.D. Memos/ Coastal Engineering 22 (1994) 201-215 205

negligible. Assume, also, that pk( TI ~1 ,~2) ~ 0 for k < 0. Then the following interval decom-
position
(t,t+ T+ dT) = (t,t+ T] + [t+ T,t+ T+ dT) (3)
applied to Eq. 2 can lead easily (McFadden, 1958) to the following recursion relationships
(see Appendix)
02U + (0,T I ~:1)
OT 2 -E[No]Po +(Tl~j) (4)

and
02U- (0,TI ~:2)
OT z -g[No]Po - (TI ~:2) (5)

Now, we denote by P0 (TI~l,~2)dT the probability

podT=po + (TI £~)dT+po - ( TI sC2)dT (6)
and provide by the conservation of total probability the following separation of U
Prob{0<x(s) ~<s¢1} Prob{£2 <~x(s) <0}
U(0,TI ~, ,~2) -- +
Prob{ ~:2<~x(s) ~<~ } Prob{~e <~x(s) <~~1 }
= U + (0,T I ~:j) + U - (0,TI ~z) (7)
where U(0,T] ~1,sc2) is the probability that no zero-crossings occur in (t,t + T), given that
~z ~<x ~<~:1 there, and t < s < t + T.
Combination of Eqs. 4, 5, 6 and 7 yields
02U(0,TI ~, ,~2)
OT 2 -E[No]Po(TI ~ ,•) (8)

Eq. 8 holds for the case where no zero-crossing occurs in the interval (t,t + T). The corre-
sponding relation for one such crossing in the same interval is (see Appendix)
O2U( I,TI ~:71,¢2)
0T 2 - E [ N 0 ] [p, (TI ~, ,~:2) -2po(Tl~,,~2)] (9)

Combination of Eqs. 8 and 9 gives

1 02
P~(TI~'~:2) E[No] 0T 2 [U(I'T~:~'I~:z) +2U(0'TI~I'~:2)] (10)

It is noted that the preceding analysis does not assume that successive intervals are
independent. Approximations of this kind will be referred to later on.
Now, define
q~(T,~,~z)dTd~d~2 the joint probability that a zero occurs in ( t + T , t + T + d T ) given
that the one before the last zero occurred at t, and the local extremes
occur in (~l,~l+d~:l), (~:2,£2+d~2) given that ~:2~<x~<£1 in
(t,t+ T).
The following relation can be easily shown (see Appendix)
206 C.D. M e m o s / Coastal Engineering 22 (1994) 2 0 1 - 2 1 5

ql( T,~:, ,~2) -- (11)

A formal evaluation of the joint probability of wave height and period can be provided
by Eqs. 10 and 11, since U can be expressed through the joint density of the underlying
process. In fact for a Gaussian process U(0,T] ~,,~2) can be written as

U(0,TI{:,,~O= lira U.(0;t, . . . . . t. lsC,,~f2) (12)

where t = t~ < t 2 < . . . < t. = t + T and in the limit this partition is such that the maximum
spacing between successive t's tends to zero;

u,,(o;t, ..... t. I ~ , ~ )
#i #i 0 0

f p(x~ ..... x,)dx~...dx,,

0 ,~-,,----~ 0
?l n

p(xj . . . . . x,) is the n-order joint probability density function of the Gaussian process.
Similarly, U( 1,T[ ~,~2) can be written in the following manner

U(1.T]£,.£2) = lim U . ( 1 ; t , . . . . . t. 1~.~2) (14)

n ~w~

Denoting by tp the time when the zero-crossing occurs within the interval (t,t + T) and
choosing the partition of this interval in such a way that tt, belongs to the partition points ti,
we may write

U , ( I ; h . . . . . t, I~:,,~c2) = ~ up(l;tl ..... t, [~c,,~2) (15)

p= 1

where

u,,(1;t, ..... t. I ~,,~2)=

#1 #1 0 0

17
I(f I1 -- p
..... .....
(16)

The factor 2 in Eq. 16 takes account of both up- and down-crossing modes of the process.
Using Eq. 1 we put

Q(T,H) = i q~(T'~'~2) ~ = ~ + H d~2 E<0, IEI << 1 (17)

--H

which, in view of Eq. 11, becomes

 C.D. M e m o s / Coastal Engineering 22 (1994) 2 0 1 - 2 1 5 207

Q(T,H) =
-H
i 0 2p I LT,~:I ,~2)
Oscl0~2 e, =~+H d~:2 E<0, IEI << 1 (18)

It is known that for a normal process with zero mean the joint density can be expressed
as

p(xl ..... xn)-- (2~.)./21An11/z exp -(1/2) Moxix ~ (19)

i,j= 1

where M o the (]-element of the inverse of the covariance matrix A..

The expected value of the frequency of zero-crossings No is given for a normal process
by
E[No] = ( 1/ 7) (m2/mo)1/2 (20)
where mi denotes the i-th moment of the spectral density of the process.
Eqs. 10 to 20 provide a formal solution to the problem of estimating the joint density
Q(T,H).

3. Approximations

3.1. The symmetric extremes

A simplification of the preceding formulation is produced if we assume that the extremes

of the process that define the wave height are symmetrical about the mean level. Putting
£1 = 1£21 = ~:= H/2, the following relations can be easily obtained from the corresponding
equations of the previous chapter. Thus we have
1 ~2
p l ( T I + H/2) = E[No] 0T 2 [U(I'TI + H/2) + 2U(0,TI + H/2) (10')

where the primed equations correspond to the unprimed ones of the general case.
Similarly we get (see Appendix)
02pl(T I + H / 2 )
q~(T, + H / 2 ) = - 4 i)HZ (11')

and
H/2 H/2

Un(O;tl ..... tn)l+H/2)=21...Ip(xl ..... Xn)dXl...dx n (13')

o ~ o
n

while Eq. 16 is modified only as regards the limits of integrations.

Finally, since H is fixed in space with regard to the mean level, we have
Q(T, H) =ql(T, +_H/2) (17')
208 C.D. Memos/Coastal Engineering 22 (1994) 201-215

3.2. Independent intervals

So far, the successive intervals between zeros have not been assumed statistically inde-
pendent. If, however, these intervals are assumed independent the probability density
P0( TI ~1,~2) can be determined for a Gaussian sea through the following integral equation
(McFadden, 1958).
T

po(TI E~,~z) = W(TI El , 6 ) -- f W(ll El,6) .po(T-l)dl (21)

0

where

W(T I E~, 6 ) = E P-(TI El'6)

n=O

Now, function W can be expressed through the corresponding normalized autocorrelation

function p(T I~l,E2) of the normal process. This function is defined, as usually, by the
following relations

p(T) =r(T)/r(O)

r( T) - r( T I ~ , , 6 ) = lim x(tl E,,6)x(t+ TI E1,6)dt

_.g

where r(T) is the autocorrelation function of the process x(T[ E l , ~ 2 ) '

It should be noted in this respect that this latter process can retain its Gaussian character
by a suitable selection of the extremes ~1, E2 depending on the variance of the elevation of
the free surface. This general remark is also valid throughout the numerical elaboration of
the theoretical results.
Assuming now a zero-mean process and making use of the relevant result by Rice (1945),
one can arrive at the following expression of W.
Ll( 1 __p2) -3/2
W=(1/'n')[(I/H) + tan- 1H] [ - - p " ( O ) ] 1/2 (22)

where
H=Ll(L22 Li2)-1/2
LI =p"( 1 --p 2) q._pp,2

and
L2 = - p"(O) ( 1 - p2) _ p,2

the dependence on (T I ~l,E2) being understood, and primes denote differentiation with
respect to T.
 C.D. Memos / Coastal Engineering 22 (1994) 201-215 209
Eq. 21 can be solved numerically and through Eq. 8 U(0,T[ ~:~,~2) can be evaluated.
By assuming symmetrical extremes as previously, we can write Eq. 8 as follows
1 02U(0,TI + H / 2 )
po(T I + H / 2 ) = E[No] OT z (8')

Now define qo( T, ± H / 2 ) as the probability that a zero-crossing occurs in (t + T,t + T + dT)
given that the zero occurred at t, and - H I 2 <<.x <~H I 2 in (t,t + T). Then we get in a manner
similar to that of Eq. (11')

O2po(TI +n/2)
qo(T,+__H/2) = - 4 OH2 (23)

Hence,
4 04U(T[ ± H / 2 )
qo(T,±H/2)= - (24)
E[No] 8T2~H 2

Eq. 12, modified for the symmetrical extremes, coupled with Eq. 13' can be used for the
evaluation of U.

3. 3. Quasi- independent intervals

If we assume now that a given interval is independent of the sum of the ( 2 m + 2 )

immediately preceding intervals for all values of m, m = 0,1,2 . . . . . then one can arrive at
the following integral equation for Po (McFadden, 1958).

tl(1--p2)-3/2 {[ ~-1 ( ~1- +

po = 2[_p,,(0)]~/2 tan -~
H)+ 2] (25)

where (*) denotes the convolution in the form of the second term of the right member of
Eq. 21, and the dependence on (T[ ~:~,~:2) being understood throughout.
The first term in the right member of Eq. 25 has been given by S.O. Rice as an approxi-
mation topo(T) when Tis not too large. For Ganssian sea, Eq. 25 can be solved numerically
for several power spectra. Again, in order to obtain p~, which is of crucial importance in
solving our problem, we have to go back to Eq. 10 and substitute 02U(0,T[ ~:1,~2)/oT2
through Eq. 8.
It has been shown (McFadden, 1958) that the assumption of quasi-independence, in
contrast with that of statistical independence, is an acceptable one when the high-frequency
cut-off is not extremely sharp.

3.4. The non-Gauss±an sea

The previous theoretical analysis assumed that the waves are a Gauss±an random process.
That is, the statistical distribution of the wave profile is assumed to follow the normal
210 C.D. Memos / Coastal Engineering 22 (1994) 201-215

probability law in which only two parameters - - the mean value and the variance - - define
the distribution. Since the mean value is taken zero, the probability density can be established
by knowing the variance of the surface elevation, which can be evaluated from the frequency
spectrum. However, it has been suggested (e.g. by Goda, 1985) that due to wave non-
linearities the statistical distribution of the surface elevation is distorted from the normal.
This distortion is usually described by the skewness and kurtosis defined as:

skewness ~ _ ~,~
1 N1 ~N ( x i - ~ ) 3 (26)
i=1

1 1 N
kurtosis /32- 4 ~ (xi- ~)4 (27)
~:rms / i=l

The normal distribution has the values v//3~= 0 and/32 = 3 but higher values have been
observed in nature. The positive value of the skewness suggests that the deviations of the
wave crests from the mean level are greater than those of the wave troughs. This is indeed
a known feature of sea waves, predicted by theories of finite amplitude waves.
A better statistical description of the surface elevation can be provided by modification
of the normal distribution through series of the cumulants-generating type (Longuet-Hig-
gins, 1963) or the more conventional Gram-Charlier type expansion. This latter series are
expressed using Hermite polynomials F, G defined by higher correlations of the normalized
random variables xi. The modified joint density is written as (Kamp6 de Frriet, 1972)

g(xl ..... x.) =p(x I ..... X.) Z AU . . . . . . . 'riFt ' . . . . . . . "n(Xl . . . . . X.v/) (28)
t'l q-.,J,l'n = 0

where

Ao ..... o ~ 1

A,. 1...... . = 0 for v~+...+v,=lor2

1
A,.~ ..... , . o - E ~t G(X, . . . . . .,.,,
,., ..... . )]
t'l!...v,[

Another approach to describing the non-Gaussian waves is to assume a random sequence

of Stokes waves. Tayfun (1980) derived a probability density for non-linear waves based
on Stokes waves of second order along with the narrow-band assumption.
In the formulations derived in the previous chapter the Gaussian density p has been
assumed in Eq. 19. However, the limits of integration in Eqs. 13 and 16 allow for wave
heights non-symmetrical about the mean, including thus the more realistic cases of higher
order waves. It was thought at this stage that the slight discrepancy between Gaussian and
non-Gaussian sea would be mostly covered by the technique of the previous chapter, which,
although not based on the properties of the normal process, approximates better the real
situation.
 C.D. Memos~Coastal Engineering 22 (1994) 201-215 211

4. Numerical results

The general solution with ~:j 4: l~:2l as well as the symmetrical case have undergone
numerical evaluation. The main algorithm was based on appropriate geometrically con-
structed wavelets, which were manipulated by combinatorial analysis. The procedure aimed
at handling numerically the results obtained in the preceding theoretical development and
in particular those of Eqs. 12 to 16. The whole numerical evaluation is presented in a
companion paper to the present one, where the results are also discussed.

5. Conclusion

The problem of the joint distribution of wave heights and periods has been tackled
theoretically. The formulation of equations is valid for wave spectrum of any band-width,
an improvement over existing solutions. The numerical evaluation revealed some interesting
aspects of the problem, which are reported in a companion paper appearing in the present
issue.

Acknowledgments

This research was carried out mainly during a 6-month subbatical leave at Imperial
College, London. My thanks are due to Dr. K. Anastasiou and Prof. P. Holmes for useful
discussions and the support they provided during my stay at the College. Also, the comments
of the reviewers J. Fenton, A. Naess, D. Basco are gratefully acknowledged.

Notation

C positive integer
COV covariance
e basis of natural logarithms
exp exponential
E[.] expected value
H wave height
i,],k positive integers
mi ith moment of spectral density
n number of columns in a waveform
iV,, zero-crossings per unit time
?- autocorrelation function
rij correlation coefficient
T wave period
X surface elevation
skewness squared
~2 kurtosis
212 C.D. Memos / Coastal Engineering 22 (1994) 201-215

mpl covariance matrix

crest elevation
trough elevation
P normalized autocorrelation function
O -2 variance
4, covariance function

Appendix

Proof of Eqs. 4 and 5

Consider the following events

E~: There are only positive points in the interval (t,t + T) and ~2 ~<x ~<~j there.
E2: There are only positive points in (t,t + T + dT) and ~2 ~<x ~<~ in (t,t + T).
E3: There are only positive points in (t,t+T) and there is one crossing point in
( t + T , t + T + d T ) and ~2~<x~<~j in (t,t+T).
Taking probabilities we may write
Pr{El } = Pr{E2 } + Pr{E3 }
But,

Pr{E2 } - P r { E , } = U + (O,T+ dT I ~ ) - U + (O,T I ~,) -

° u + (0'TI ~J')-dT
~T
and

Pr{ E3 } = E[ No] dT ~ Po +( 11 ~,)dl

1¢
T

is the joint probability of finding one crossing point in ( t + T,t+ T + d T ) and none in
(t,t + T ) , while 0<x~< SOlin (t,t+T).
Combination of the last three equations yields
02U + (0,TI ~t)
~)T2 =E[No]Po +(T[~,) (4)

Similarly we derive the pair Eq. 5.

Proof of Eq. 9

Consider the following events.

E4: There is exactly one zero-crossing in (t,t + T) and ~2 ~<x ~<~ there.
Es: There is exactly one zero-crossing in (t,t + T + dT) and ~2 ~<x ~<~1 in (t,t + T).
E6: There is exactly one zero-crossing in (t,t+T) and one zero-crossing in
( t + T , t + T + d T ) , and ~2 ~<x~<~ in (t,t+ T).
 C.D. Memos / Coastal Engineering 22 (1994) 201-215 213

E7: There is either none or one zero-crossing point in ( t , t + T ) and one in

( t + T,t+ T + d T ) , and (2<~x<~(j in (t,t+ T).
Now, E5 can be divided into the following two alternatives:
E~t ~: There is exactly one zero-crossing in (t,t + T) and none in (t + T,t + T + dT), and
~2~<x~< ~ in ( t , t + T ) .
E~ 2): There is no zero-crossing in (t,t + T) and one in ( t + T,t + T + dT), and se2~ x ~<se~
in ( t,t + T).
Taking probabilities we have

Pr{ E 4 } = Pr{ E5 (1 ~} _ Pr{ E6 }

Pr{ E 5 } = er{ E5 ~ ~~ } + er{ E 5 ~2~}

Subtracting,
Pr{ E5 } - Pr{ E4 } = Pr{ E5 ~2~} + Pr{ E~ }

But as with Eq. 2 we can write

Pr{E5 } - Pr{E4 ] = ~U( 1,T I sel,~:2)dT

OT
Also,
Pr{ E6 } = Pr{ E7 } - er[ E5 ~2)}

while

Pr{ E7 } = E[ No]dT ~ p, (/1 se~,see)d/

T

The event E5~2~ can be further divided into two alternatives covering the positive and
negative values ofx respectively. Thus

Pr{Es~2)}=E[No]dTL)~( Pg (/I ~:,)d/+ Po (/I ~:2)dl

T T

Combination of the above relations yields finally the expression

zc c~

T T

which leads easily to Eq. 9.

Proof of Eq. I1

From the definition of pl it is clear that

214 C.D. Memos/ CoastalEngineering22 (1994)201-215

p~( TI ~ + d~l ,~2)dT-pj (T I ~ ,~2) d T = Pr{ a zero-crossing occurs in (t + T,t + T+ dT)

g i v e n that the one before that last zero occurred at t and m a x XE(~:I, ~ + d~2) g i v e n that
~2 ~<x in (t, t + T) } - q ~ + ( T , ~ , ~2) d T d ~ , say. Then,

q~- (T, ~1, ~2 +dsC2) d~:l--q~- (T, sol, so2) d T d ~ l = Pr{a zero-crossing occurs in ( t + T,
t + T + dT) g i v e n that the one before the last zero occurred at t, and the local extremes occur
in ( ~ , ~ + d ~ ) and (~2, ~2 + d ~ ) in the interval (t, t + T ) } =q~(T, ~ , ~2) d T d ~ d~2

C o m b i n a t i o n o f the previous two relations yields

Up, ( TI ~, ,~¢z)
q, (T,~, ,~2) - (11)

Proof of Eq. 11'

F r o m the definition ofp~ it is clear that

p~ IT[ ( H / 2 ) + d ( H / 2 ) , - ( H / 2 ) - d ( H / 2 ) ] d T - p ~ ( T I +_H/2)dT = P r { z e r o - c r o s s i n g
occurs in ( t + T, t + T + d T ) g i v e n that the one before the last zero occurred at t, and
m a x xE(H/2, H / 2 + d ( H / 2 ) ) min x e ( - H / 2 , - H / 2 - d ( H / 2 ) ) in (t, t + T ) } =
ql (T, +_H/2 )dTd( H/2 )d( - H/2 ). H e n c e

~)2p~ ( T I + H / 2 )
q~(T, _ _ _ H / 2 ) = - 4 5H 2 (11')

References

Bretschneider, C.L., 1959. Wave variability and wave spectra for wind-generated gravity waves. Tech. Memo.
118, U.S. Beach Erosion Board, Washington, DC, 192 pp.
Cavani6, A., Arhan, M. and Ezraty, R., 1976. A statistical relationship between individual heights and periods of
storm waves. In: 1st Int. Conf. on Behaviour of Offshore Structures, BOSS '76, Trondheim. The Norwegian
Institute of Technology, Trondheim, Vol. 2, pp. 354-360.
Chakrabarti, S.K. and Cooley, R.P. 1977. Statistical distribution of periods and heights of ocean waves. J. Geophys.
Res., 82: 1363-1368.
Goda, Y., 1978. The observed joint distribution of periods and heights of sea waves. In: Proc. 16th Int. Conf. on
Coastal Eng., Sydney, pp. 227-246.
Goda, Y., 1985. Random Seas and Design of Maritime Structures. Univ. of Tokyo Press, Tokyo, 323 pp.
Kamp6 de F6riet, J., 1972. Sur une loi de probabilit6 du champ des vitesses d'un fluide dans un 6tat voisin de
l'6quilibre statistique. Symp. Mat. Ist. Naz. Alta Mat., IX: 273-298.
Lindgren, G. and Rychlik, I., 1982. Wave characteristic distributions for Gaussian waves - - wave-length, ampli-
tude and steepness. Ocean Eng., 9(5): 411-432.
Longuet-Higgins, M.S., 1963. The effect of non-linearities on statistical distributions in the theory of sea waves.
J. Fluid Mech., 17(3) : 459-480.
Longuet-Higgins, M.S., 1975. On the joint distribution of the periods and amplitudes of sea waves. J. Geophys.
Res., 80( 18): 2688-2694.
Longuet-Higgins, M.S., 1983. On the joint distribution of wave periods and amplitudes in a random wave field.
Proc. R. Soc. London A, 389: 241-258.
 C.D. Memos / Coastal Engineering 22 (1994) 201-215 215

McFadden, J.A., 1958. The axis-crossing intervals of random functions - - II. IRE Trans. Inf. Theory, (March):
14-24.
Memos, C.D. and Tzanis, K., 1994. Numerical results of the joint probability of heights and periods of sea waves.
Coastal Eng., 22: 217-235.
Rice, J.R. and Beer, F.P., 1965. On the distribution of rises and falls in a continuous random process. J. Basic
Eng., 87: 398--404.
Rice, S.O., 1944. The mathematical analysis of random noise. Bell Syst. Tech. J., 23: 282-332.
Rice, S.O., 1945. The mathematical analysis of random noise. Bell Syst. Tech. J., 24: 46--156.
Shum, K.T. and Melville, W.K., 1984. Estimate of the joint statistics of amplitudes and periods of ocean waves
using an integral transform technique. J. Geophys. Res., 89 (C4): 6467-6476.
Srokosz, M.A. and Challenor, P.G. 1987. Joint distributions of wave height and period: a critical comparison.
Ocean Eng., 14(4): 295-311.
Swift, R.H., 1988. Discrete wave approach to wave force statistics. J. Waterw., Port, Coastal Ocean Eng., 114(6) :
686--699.
Tayfun, M.A., 1980. Narrow-band non linear sea waves. J. Geophys. Res., 85(C3): 1548-1552.
Tayfun, M.A., 1990. Distribution of large wave heights. J. Waterw., Port, Coastal Ocean Eng., 116 (6): 686-699.