JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 109, COS016, doi:10.1029/20031C002073, 2004 On the distributions of wave periods, wavelengths, and amplitudes in a random wave field Delun Xu, Xiang Li, Lizhen Zhang, Ning Xu, and Hongmin Lu Physical Oceanography Labor, Oxsan Univer of China, Qineo, China Recsved 29 uly 2003; revise 29 February’ 200; sexed Mach 200; published 14 May 2008, [i] On the basis of a narrow-band Gaussian process, two theoretical probability density functions (PDFs) are derived by applying an approximate Hilbert transform to the process represented by a general form. The first PDF is for the joint distribution of the wave period ‘and amplitude p of sea waves. It has the same merits as the PDF previously derived of p and so facilitates its handling in theory and application in practice. In virtue of the advantage, a relatively simple conditional PDF of + assuming p is derived, which may be used to predict an arbitrarily defined characteristic conditional wave period assuming the wave height from the averaze wave period. The second PDF is for the joint distribution of wavelengths and amplitudes in a unidirectional Gaussian wave field with narrow spectrum, It has the same form as the first PDF but depends only on the parameter j defined by 1? = mona/m3 — 1, where ny is the nth moment of spectrum. From this joint PDF, a PDF for the distribution of wavelengths of sea waves is also derived. Numerical simulation and laboratory experiments of wind waves and their results are reported. It is shown that the PDFs derived in this paper give fairly good fits to the simulated data and suitably filtered laboratory data, INDEX TERMS: 0624 Electomagnetes: Guided waves, 3384 Meteorology and Atmospherie Dynamies: Waves and tides; 4263 Oceanography: General: Oeean prediction: KEVFORDS: joint probability of wave variables, Hilbert ansform, experiment Citation: Xu, D., X. Li, L. Zhang, N. Hu, and H. Lu (2004), On the distributions of wave periods, wavelengths and amplitudes in a random wave field, J. Geophys. Res., 109, COS016, doi:10.1029/20031C002073. 1. Introduction [2] The statistical distributions of periods, wavelengths and amplitudes of sea waves have long been interesting due to their great importance to marine science. For them many theoretical probability density functions (PDFs hereinafter) hhave proposed in the last centenary. [a] As early as in 1950s, Longuet-Higgins (1952, 1957] proposed the Rayleigh PDF for the distribution of sea-wave amplitudes, based on the envelope theory of Gaussian noise which had been developed in studying electric communi- cation [Rice, 1945]. The Rayleigh PDF of amplitudes has been widely employed for many theoretical and practical purposes in marine studies. Also in this period, Cartwright ‘and Longuet-Higgins [1956] derived a theoretical PDF of the maximum in a narrow-band Gaussian process. This PDF hhas also been applied in many fields of marine science. For example, it has been employed by Srokose [1986] to develop a simple model for the probability of wave breaki in deep water. [4] In the 1970s, many attentions were paid to the joint distribution of sea wave periods and amplitudes due to Copyright 2004 by the American Geophysical Union. biazonrTiyzoussco0203 C5016 requirements in marine science. Lindgren (1972) proposed an approximation to the joint PDF of wavelength and amplitude (defined as the differences in time and eight between a maximum and the following minimum) in a Gaussian process, but the resulting expression requires a great deal of computation to evaluate and involves high ‘moments of spectrum. Longuet-Higgins 1975] derived a PDF for th joint distibution of wave periods and ampli tudes, based on a narrow-band approximation applied tothe Tinear theory of Gaussian noise; is expression depends only cn the spectral with v, where v2 = mmyin ~ Twith my a8 the mth moment of spectrum. While giving 8 fairly good fit to sea wave data with a narrow spectrum, this PDF failed in accounting for the observed asymmetry in the distribution of wave petiods (eg, Goda, 1978 Chakrabarti and Coolley, 1977], Later, also based on a narrow-band Gaussian process, Cavanié ef al. [1976] proposed another joint PDF, which accounts successfully forthe asymmetry in the distibution of wave periods but involves the fourth ‘moment of spectrum, my, and therefore is inconvenient for practical purposes, for m, depends rather critically on the bichavior of spectrum at high frequencies and iti inexistent to currently used spectral models of sea waves, such a the wellknown JONSWAP spectrum and Pierson-Moscowitz spectrum. A lengthy and somewhat more accurate approx: 1 of 16cosor6 imation tothe joint PDF of wavelength and amplitude ina Gaussian’ process was given by Lindgren and Rychlik {1982}, but the resulting expression also depends on mg. [@) To overcome the difeuties existing in the above mentioned joint PDFs of wave period and amplitude, Longuet- Higgins [1983] improved the joint PDF in the work of Longuet-Higgins {1975} by introducing « normal- ization factor L(v), to take account of the fact that only Positive values of wave period should be considered. The Improved PDF is asymmetric in the distabution of wave periods and depends ony on v. Hosver, this improvement Js merely a mathematical teatment, which leads to a drawback that the resufing PDF predicts slightly non- Rayleigh distribution of amplitdes, rather than the well known Rayleigh one as itis under the linear theo Gavssian process, Although this drawback isnot sign for usual purposes, it may bring about some difficulties or inconvenisnees in Finer theoretis! handling and prateal applications ofthe PDF. as will be pointed out subsequently. [6] Few theoretical PDFs have been proposed for the joint distribution of wavelengths and amplitudes and the distribution of wavelengths. This is perhaps because the wavelength of sea waves isa spatially two-dimensional ‘variable and its direct measurement is difficult. Sun [1988] derived a joint PDF of four variables: amplitude, frequency, and wave numbers (in x and y directions), based onthe ray theory of waves and a Fourier mode of thre-dimensional random wave field (an extension of the model for the Surfiee elevation at a fixed point resented by: Longuet- Higgins [1975]). The parameters in this joint PDF are expressed in terms of a detenminant eossting of the omelations between these variables. From this joint PDF. 2 joint PDF of wavelength and amplitude may be derived, but the relat analysis is dificult andthe resulting expres- sion depends on the directional spectrum of sea wave {7} The effect of nonlinearities on the statistical distribu- tions of sea wave variables has been investigated empiri- cally and theoretically by a numberof researchers [e.g Cox dnd Munk, 1954; Longuet- Higgins, 1963; Huang and Long, 1980; Huang et al, 1983; Tayfin, 1980; Srokose, 1986] However, these investigations concentrated upon the non- linear effect on the distributions of sea surface elevations and slopes, and scarcely involved the distributions of wave Periods, wavelengths and amplitudes [8] The purpose of this paper is to present an alternative theoretical joint PDF of wave period and amplitude in a narrow-band Gaussian process and « theoctcal joint PDF of wavelength and amplitde in a unidirectional Gaussian ‘wave field of narrow-band. In setion 2 we shall give the derivation ofthe Tormet, in which an approximate Hilbert transform wll be applied; the resting joint PDP has the Same merits as that inthe work of Longuet-Hligeins [1983] but predicts an exactly Rayleigh distribution of amplitudes. in aecordance with ths joint PDF, conditional distributions of wave periods assuming the amplitude will be examined in dtl. The derivation ofthe joint PDF of wavelength and amplitude willbe given in section 3 in which the aprox- imate Hilber transform will also be applied; the reeling joint PDF depends only on the parameter ji, where)? ‘mymain’ — 1. From this joint PDF, « PDF of wavelength depending only on the parameter will be derived. Section 4 ‘ill eport numerical simulation and laboratory experiments XU ET AL: DISTRIBUTIONS OF SEA WAVE VARIABLES. cosn1s performed in a wind-wave flume. Their results well support the theoretical PDFS derived in this paper. Some conclu sions will be drawn in the final section, 2. Joint Distribution of Wave Periods and Amplitudes 2.1, Joint Probability Density Function (PDF) [)] As in the work of Longuet-Higgins (1975, 1983], we bezin with the representation of the sea surface elevation (0 i the form. 0) s() = Reale At) = (ne? where © is the cartier frequency, and the phase ¢(t) and local amplitude p(?) are both real and slowly varying functions of the time ¢. Let (0 = 010) +a, @ It follows from equation (1) that sf) = Re ple 6) ‘We also adopt the narrow-band hypothesis, namely, a a) ‘where m, denotes the mth moment of the one-sided spectrum, se) of <(0), defined as mem [Fee 6) Longuet-Higgins [1957] has shown that the variance of the change rate j (2) (a dot denotes differentiation with respect to f, hereinafter) is proportional to v*, This implies that by the hypothesis (4), j(0) is negligible in comparison with (0 In this case the Hilbert transform of <(?), which is defined as, con8 [7 Dears © ca sao = tape 0 since the Hilbert transform is a quadrature transform [see, ©. Papoulis, 1984], fa Pe sin) (ha = +00)? (8) vo =aenl] ® (u)_ It is worthwhile noting that the Hilbert transform technique has been extensively and effectively employed to 20f 6 Arengoupyfatend) 395 Th20z/20/e2] uo Msg aullug lq "tsoUopU| Mand YEN AG €2020097E002/6¢01 01 /op/uo> Xopwsesaouuo'sndnGe/ sy wy PopeOlNMad "= 002 >2O2V9SI2 seu20p suowiuos anneaig s}aeudde a ha pauiano6 ae s;Ue Yo '98 o 50130) fse2q7 avo KAN Wo (e¥oNIpUe>-puE-sHAY/u0>K9 suonypuep pe sicosor6 measure the local wave variables from a wave record with narrow-spectrum [Melville, 1983; Bitner-Gregersen and Gran, 1983; Shum and MeWville, 1984; Hwang et al., 1989: Stansell and MacFanlane, 2002), [2] We further assume that <(0) is @ stationary Gaussian process. As £(), (0) and £(0 are all near transforms of <(), they are jointly stationary Gaussian and their joint PDF can therefore be expressed as F860 atm{ 5% Sh}. 0) where mom =&n = ha 6 ay Ry Riz Ra Res Ray Roa Ray Rw (2 Rea Rs Ru Ry Re Ry Ray and Dj is the residual subdeterminant of D with regard to Ry, with Ry defined as Ry = HONE, where the over line denotes the assembly average, namely, the expected value. (1 Calculating Rj, Dy and D in equation (10) from equations (3) and (7) and transforming f(s, & & §) into fp. 2s 2) (9ee Appendix A) we have the joint PDF of (0) and 20) 8) = ex — 5 (mz + mg? - 2m) }, ( F0.8) = orm esn{ Bgl +m? 20). 1) where A= (mm, — 2)". (sy [i] The local wave period x(t) is defined as bor x) as) ‘where (2) = (is the local radian frequency. Introducing the normalized amplitude R() and wave period 71): io (2m)"? (16) x0) ay XU ET AL: DISTRIBUTIONS OF SEA WAVE VARIABLES. cosn1s as in the work of Longuet-Higgins [1983], we have from equation (13) the joint PDF of R(®) and 71) j= 228(\son{-2}) oof SE [1+2(1-7)]}: a) ich es in xpression appara the int PDF in Sal R.T) 1 Mise Th as) However the numerical difference between f(R, 7) and fidR, T) is very slight, as can be seen from Figures 1a and 1b which show the contours of AR, T)/fmax and fuihR, Tifistmaxs respectively, for a sequence of values of Wi there fone Jets manu oF ade of the PDF found from the condition that 2 and “both vanish. {is}, We note thatthe diteeR2e betw&ln equations (18) and (19) is Because equation (18) i drcly derived fom a Gaussian process under the hypothesis v= << 1, whereas equation (19) is obtained by inoducing the normalization Factor 20) to impreve Longuetftgins [1973] PDF whose derivation involved the approximation ronstign tla It is this approximation that makes the resulting joint PDF symmetric in T [see also Shum and Mehille, 1984], though such a approximation is legitimate to order v and its effect on the PDF is slight in magnitude. [is]. To further compare equations (18) and (19) we refer to h(®) = 29{0) as the local wave height and define the normalized wave height as pote) =a (20) Combining equations (20) and (16) we have Ry, en Thus the PDFs (18) and (19) can be transformed into the forms @) Sil tt.t) = 3 0f 16 cArsqysuo/ sa) 295 Tyz0e/20/e2) uo Aig allup Kaa "s9UpU| Aoud YEN AG €L02003"E002/620,OL/OP/O> Ko AsesgHouquO'sgndnBe//sday wo PPK seu20p suowiuos anneaig ajar a ha pauiano6 a sie Yo '98 o 590130) fseq7 avo KUNA Wo (e¥oNIpUe>-pUE-sHAY/u0>f9 suonypuep pe sicoso16 XU ET AL: DISTRIBUTIONS OF SEA WAVE VARIABLES coso16 © a y a a v=02 | a, as v q| | | R15) 5 15| 1 1 z 05| os os| oes ees ae 1 1 Tr ° a y a os v=02 | ,, v al | 1 | R15] 45] 45] 1 4| 1 os) os os| aa 7? 7? 1 Tr r Figure 1. Contours of (a) (R, Mf and (b) fisAR, Tusa given by equations (18) and (19), respectively: /fnan and fiitfisiows take the values 0.1, 0.2, ‘outward, Clearly, the difference between lH, 7) and firAHt, manifests itself only by that between G(#Z, T; v) and L(). We note that G(H, F; v) results from the rigorous derivation based on a narrow-band Gaussian process, whereas L(V) isa normalization factor introduced t0 improve the PDF in the ‘work of Longuet-Higgins [1975] [ir] A striking difference between (R, 7) and fizAR, T) is that the PDF of R(®) predicted by ((R, 7) is exactly Rayleigh: £8) 2Rexp{—R} ey as itis under linear theory of Gaussian process [Longuer- Higgins, 1952, 1957), whereas that predicted by fisAR, 1) is slightly non-Rayleigh: JiR) = 2RL(0) exp RE ert Rv) L [eta were er) repens the well cm fnton angrier ay aso the numerical difference between f(R) and fi1dR) is slight. Hower uh pe ad lt fo tne Sd apply than fi,(R), especially in examining the conditional Ein fen pe asaning ie wave gh wil ese soy (as) exf(R/v) 2.2. Discussion [is] Traditionally the wave period and wave height in a ‘wave signal are defined as the time duration and maximum . 0.9, respectively, from the center height between two successive up-zero-crossings (such wave variables are refered to as the zero-rossing ones), whereas what we deal with here are the local variables, which in practice take values at points uniformly spaced at time intervals of 1. It is pertinent to inquire whether or under what condition the distribution of a zero-crossing variable is equivalent to that of its corresponding local Variable. We have adopted the narrow-band hypothes ¥ -puR-swuAy/ wor asu297 suouio> anness9 sygeudde an hq pawano6 ar suowypuop pue sia, aus 995 Tp202/20/82) Uo KEIQFaUIUG ROM I89VOPUT Ad EN KE EL0200€002/8201cosor6 8 4 v=02 Apel 6 Zs 3 3:pr1/0 4:p=1/100 05 1 15 2 os 1 T XU ET AL: DISTRIBUTIONS OF SEA WAVE VARIABLES. cosn1s vos Lp 2p S:p-M10 4:p-100 Figure 2. The conditional distributions of T assuming HM = Hy, ys, Hyao and Hyon given by equation 3) together with equation (34) for v = (a) 0.2, (b) 0.4 and (€) 0.6. new variables called the wavelength and amplitude which are defined as the differences in time and height between a ‘maximum and the following minimum, and gave some Tengthy approximations. This leads to that the resulting joint PDF depends on the parameter © involving ms, and requires a great deal of computation to evaluate. In fact, a pointed out by Longuet-Higgins [1983], observational evi- ddonces have suggested that further corrections of order \* are small and not very significant for practical applications. 2.3. Conditional Distribution of Wave Periods (20) Longuer-Higeins [1983] has partly examined the conditional distribution of T(@) in terms of fiyAT). As the present PDF (R, 7) predicts an exactly Rayleigh PDF of R(p), this makes it advantageous to further examining the conditional distribution and giving some useful results. For convenience, we shall adopt the PDF (7, 7) given in equation (22) that is equivalent to /(R, 7) in equation (18) [a1] The total PDF of #1(0) given by fl, 7) is SH) =F Htexp{- FH} (26) ‘The conditional PDF of T() assuming #0), TID, is found by per ay LED @ Insertion oequtos (3) nd 26) in quion (27) gies perian= 38, (tee{-ZE}) oof - (28) On the other hand, equation (26) gives the PDF of local ‘wave height A(0) = 2p(0) as : . Aoof-E} o and then the average wave height iis related to mg by rh) i [ar iyah = Cm)”. (0) [22] Lot, denote the value of wave height averaged over the p fraction highest waves, that is, t y= ~ p(t oh v= [som where fip is the p quantile, namely, such a value of wave hheight that the accumulative probability of waves with wave height exceeding the value is p: hs is well known as the significant wave height. Normalizing hy as G1) a Gem) wwe have from equation (28) the conditional PDF of 710) assuming (0) = H, @) Further, H, can be expressed as @ function of p by inserting equation (29) in equation (31) and using equations (30) and w= a"afe-<["]} as has been given in the work of Longuet-Higgins [1982] Several typical results given by equation (34) areas follows: Hh, = 1 Hys = 1.898.119 = 2.082, Hyg = 2.683. (38) [23] Figure 2 shows the conditional distributions of T assuming H = Hy, Hy, Hio and Hyg given by equations (33) and (34) for v = 02, 04 and 0.6, espec- tively Itmay be of interest for ocean enginezting and others to find a relationship between a characteristic conditional 5 of 16 Arengoupyfatend) Opruo> ropmfsesgnousuorsgndnGe/ acy wos popeoqunod "002 22OZ9SI2 ne yo fn Jo 34 40} Asa a0 Ka Uo (evo; PueD-PuR-swuAy/ ora suowypuop pue sia, aus 995 Tp 20/20/82) Uo KEIQFaUIUG ROM I89VOPUT Ad EN KE EL0200DE002/EEO1 asu297 suouio> anness9 sygeudde an hq pawano6 arcosor6 XU ET AL: DISTRIBUTIONS OF SEA WAVE VARIABLES. cosn1s Figure 3. Counters of Ty computed from equation (41) together with equations (33) and (34) for (left) v= 04 and (right) v= 056. wave period and the average wave period 7, so as to Predict the former fiom 7. We shall define two Kinds of Characteristic conditional wave periods and derive. such relationships in terms of (71) shown in equation (33) Teil Fist, we define the charsteristc conditional wave petiod ty by =f tetyer [seiner G6) where fi) is the conditional PDF of + assuming. t= hp and ‘pis Such a value of (0) that the accumulative probability of waves with (2) ~ hp and 7(0) > To is q In other words, tyy isthe wave period averaged over the g fraction longest” waves with wave heights equaling. the identical value h,, It is easy to see fiom equation (36) that ‘when 79 ~0. In this ease m= [eine o” is the wave period averaged over al of the waves with wave hheight equaling J, Inthe extreme case, p = I and g = 1, we have [25] _Longuet-Higgins (1975] has shown that for a narrow- ‘band Gaussian process, + can be approximated as nm 8) so that the normalized wave period Tle) can be rewritten TW) 9) Normalizing ty and 7p as (40) wwe have ve Lf meiner one an a= frriayer Les] This is the relationship between Zp and + expressed in terms of fU7\H,). Inserting equations (33) and (34) into equation (41) and numerically evaluating the integrals we obtain values of Tyy for different p,q and v. Figure 3 shows contours of Ty, for-v = 0.4 and 0.6. The relation between Fyn and ng, can be found from r, Tne Troe (a) [27] It may be interesting to discuss another kind of characteristic conditional wave periods, 7, defined by are | 4 =o. (a3) Gof 6 cArsqysuo/ sa) prod pm dserguauquo'sgndnBe/ Pedy wos Pap pn yo fn Jo 34 40} Kia a0 Ki Uo {eUo%Pue>-puR-swuAy/ wor suowypuop pue sia, aus 995 Tp 20/20/82) Uo KEIQFaUIUG ROM I89UODUT Ad EN KE EL0200€002/EE01 asu297 suouio> anness9 sygeudde an hq pawano6 arcosor6 « v=02 on o ve04 oa * v=08 ™ “for or + Figure 4. 7,, versus p given by equation (44) together with equations (33) and (34) for (a) v = 0.2, (b) 0.4, and () 0.6 In other words, 7, is the most probable wave petiod of the ‘waves with wave heights equaling J The normalization counterpart of equation (43) is a) where (as) Inserting equations (33) and (34) into equation (44) and ‘numerically computing the differential we obtain values of for different p and v, which are shown in Figure 4. 24, Distribution of Wave Periods [2s]. Integration of equation (22) with respect to H over 0.] Figure 5 shows the distributions given by /(7) and JuskD), respectively, for a sequence of v. As can be seen from the figure, the numerical difference between f7) and JisdT) is very slight for all values of v. [0] The PDF of the wave period + derived from equation (46), namely, the dimensional version of equation (46) is s(n) = (48) in mr? — drm + Am) It is eagy to see that (1) is asymmetric and j=) = 0 when. ‘7 = 0. We shall provide some numerical simulation and laboratory experimental evidences for equation (48) in section 4, 3. Joint Distribution of Wavelengths and ‘Amplitudes 3.1, Joint PDF of Wavelength and Amplitude [st], Fora unidirectional random wave field, we represent the surface elevation in the direction of wave progression in the form slaye) = Rep(annea (9) We assume that the wave field is stationary in r and homogenous in x. This allows us to confine concem to the statistics of wavelength at a given 1 In this case, equation (49) can be simplified as sla) = Repay", (60) which is the same in form as equation (3) except for the substitution of ¢ with x [(2] Still we adopt the narrow-band hypothesis. As is similar to equation (4), here this hypothesis is characterized by eae (si) where M, is the mth moment of the wave mumber spectrum, (B) (one-sided) of (2), defined as = [even (2) By this hypothesis p,(x) is negligible in comparison with (2) (@ subscript x denotes differentiation with respect to x hereinafter), so that the Hilbert transform of <(x) defined as wt [7Benwirt, is T0f 6 cArsqysuo/ sa) 295 Tyz0e/20/e2) uo Aig allup Kaa "s9upU| Aoud YEN AB €L02003"E002/620,0,/IOP/O> Ka AsesgHouquO'sgndnBe//sday WOH Pap seu20p suowiuos anneaig ajar a ha pauiano6 a sie Yo '98 o 590130) fseq7 avo KUNA Wo (e¥oNIpUe>-pUE-sHAY/u0>f9 suonypuep pe sicosor6 XU ET AL: DISTRIBUTIONS OF SEA WAVE VARIABLES. cosn1s » Figure 5. Comparison between (a) f07) and (b) fA 7). ‘can be pertinently approximated as (2) = Imp(aje (4) by reason similar to that of equation (7). It follows from ‘equations (50) and (54) that os) = (2) + 2a)" (55) ee] «6 ss) = aan 2 0) As &(9), s(t) and (2) are all finear transforms of the Gaussian process s(x), they are jointly Gaussian. Their joint PDF can thus be expressed as where 91 = S92 = & ys = Se and Yu Go Dade] if = 1.2.34, (58) with Ry defined as Ry = HUE. (59) [63] Through a derivation parallel to that given in Appendix A, we obtain from equation (57) the joint PDF of pla) and tx) rein{-$ (dh + Mot amie) pee) =—— Soe) = Sapa T= MM, (1) [4] The instantancous wave number A(x), wavelength C0) and wave height h(x) are defined as k(x) = ola), (62) day = 2% (63) 0 te : 09) = 200, wo and their normalized versions are defined as rea as « Me) ; 1) =F idg/M ss) an) =) (rv) After a derivation parallel to that of equation (22) we have from equation (60) the joint PDF of B(x) and L(x) Arengoupyfatend) Opruo> ropmfsesgnousuorsgndnGe/ acy wos popeoqunod "002 22OZ9SI2 ne yo fn Jo 34 40} Asa a0 Ka Uo (evo; PueD-PuR-swuAy/ ora suowypuop pue sia, aus 995 Tp 202/20/82) Uo KEsQFaUIUG ROM I89UOPUT Ad EN KE EL02001E002/E201 asu297 suouio> anness9 sygeudde an hq pawano6 arcosor6 As is easy to see, this expression is formally the same as {Alf, T) given in equation (22) and their difference manifests itself only in implications of the variables and parameter. We shall not show the contours of 4B, L) once again. {i} Integration of equation (68) with respect to L over 0< L Ka AsesgHouquO'sgndnBe//sday WOH Pap seu20p suowiuos anneaig ajar a ha pauiano6 a sie Yo '98 o 590130) fseq7 avo KUNA Wo (e¥oNIpUe>-pUE-sHAY/u0>f9 suonypuep pe sicosor6 XU ET AL: DISTRIBUTIONS OF SEA WAVE VARIABLES. cosn1s &) Figure 6. Comparisons between the numerically simulated distributions and their corresponding theoretical distribution, (a) R(0), (b) TW), and (c) the joint distribution of R() and 7), digitizing S(.) (double-sided) truncated at 18 us, and the random numbers wih distribution of N(O, 1) is ereated by computer to close with the sequence H() and then obtain (3) trough equation (78),™=0,1,2,....N ~ 1(V=2x 108), From S(0,) and IM(o,) we abiain’Z(.,) through equation (77) and then the simulated signal s(@), = 0, 1, 2,...,N— I(r 0.025) though IFFT. The local amplitude (Ata) and local period +(%) are measured from s(t) by the Hilbert transform technique, which will be outlined in the following subsection. Figure 6 shows the distribution histograms of R(@ and 7(), the joint distribution counters of -R() and T(), and ther corresponding theoretical PDF given by equations (24), (46), and (18), respectively. As ean be seen from Figures 6a, 6b, and 6c, the agreements between simulated distributions and theoretical ones are well satis- factory, as expected 42. Laboratory Experiment I [4s] This experiment is performed in a covered flume of Physical Oceanography Laboratory, Ocean University. of China. The flume is 65 m long, 1.5 ma wide and 1.5 m high, “The waves in the flume are generated by stable and uniform winds that are produced by a wind-blower system joined with a rectifying channel, The water in the flume was 0,75 m ddeop during the experiment [4s] By a capacitance-type wave gauge the surf elevation (()) is measured at a point with fetch = 30 m and at U= 7,9 and 11 ms-!, where the fetch is refereed to 10 of 16, cArsqysuo/ sa) prod pm dserguauquo'sgndnBe/ Pedy wos Pap pn yo fn Jo 34 40} Kia a0 Ki Uo {eUo%Pue>-puR-swuAy/ wor suowypuop pue sia, aus 995 Tp 20/20/82) Uo KEIQFaUIUG ROM I89UODUT Ad EN KE EL0200€002/EE01 asu297 suouio> anness9 sygeudde an hq pawano6 arcosor6 the distance from the wind issue to the point, and U denotes the reference wind speed. Three records, each being S min long, are obtained and sampled at Ar = 0.02 s. The number of sample, N, is 15000 for each record, fo] The local wave period (6), m= 1, 2, 2 Ny is computed from ((1,) by applying the Hilbert transform technique that has been extensively and effectively employed to measure the local wave variables, such as the local wave height, period and phase, from a wave recon with narrow specieum (see section 1). This technique is briefly outlined inthe following [is] In accordance with the properties of the Hilbert transfom fee, Papoulis, 1984], itis easy to show fiom equation (6) that 2F{)} w>0 PUMA) = ) ° uso where WO =) +80. (so) From equations (79) and (80) it is seen that the Hilbert transform £(2) can be computed form the record C(0) through the Fourier transform and its inversion, The FFT algorithm ‘makes the computation highly efficient. The local amplitude pls) and phase (0) are given by equations (8) and (9), respectively, and the local wave period 7() is given by an = Ta (a) [4s] To make the wave records close to the condition of | the narrow-band hypothesis adopted in deriving the present PDFs, we band pass the records by a method that is an improvement of the Fourier series method presented by Longuet-Higgins (1984) for band passing a wave record. In ‘order to move out, from a record, the sinusoidal components, ‘with frequencies outside the range (uy, wiv), where n! and ni are two numbers such that 0 < uy < wigs < waa Wit ya as the Nygiust frequency, we change equation (79) into ° we (HO) = (2) 2F{UN} otherwise It is easy to show thatthe real and imaginary pants of 1) are the filtered record (0) and its Hilbert transform €(), respectively, namely, () w= ro) &() = Im (0) andthe local amplitude and wave period of (are given by a) = (P+ (84) Qn #O= (ss) [at] XU ET AL: DISTRIBUTIONS OF SEA WAVE VARIABLES. wi - a a eee ° Figure 7. (a) A typical section of the original record and its: Iocal “amplitude computed by the Hilbert transform technique. (b) The corresponding filtering record and its local amplitude computed by the FFT filtered method. The record is measured at U= 9 ms" and fetch = 30 m. nf}. ag As the improved method can be employed to fast and efficiently band pass a wave record by using the FFT algorithm, it is refereed to as the FFT filtering method, ‘experiment the passband of filtering is chosen where He 03, << 204, (sn) Where up denotes the peak frequency of the spectrum estimated from a record, Figure 7 gives an example of band passed record; Figure 7a shows a typical section of the original record measured at U= 9 ms! and fetch = 30-m, and its local amplitude computed by the Hilber transform technique described in equations (79), (80), and (8), while Figure 7b shows the corresponding filtered record and its local amplitude computed by the FFT filtering method with the passband taken to be 03) < ws < 2.04, [si], For convenience in reporting the experiment, we first introduce some denotations: ((,), the original record mea- sured at U=9 ms! and fetch ~ 30 m in this experiment; iq) the filtered record computed from C(t) by the FFT filtering method with the passband chosen to be 0.3, << 2.0 py) and v() the local amplitude and period of C(,) computed by the Hilbert transform technique; i) and 7), the local amplitude and period of (tq) computed from (,) by the FFT filtered method: mand fi, the mth moments of the spectrum estimated from ((f,) and C(t), respectively: Rls) = phbn2rg)"; Rtg) =P CaM)" Ty) = eC Qrrmoimy): Ty) = HON alin). v= (mamalm — 1) rift tn, — 1)"; Wty) = pty): Ht) = 2p); and hy the averaged local wave height computed from ((¢,) and Guy), respectively; fs and hys, the significant’ wave heights computed from, (¢,) and C(t), respectively; Hs = dnysihy and Hiya = hySh [se] The results computed from the three records mea- sured in the experiment are quite similar. To save spa only those computed from the record measured at. U 9 ms”! and fetch = 30 m will be shown as examples in the following. The parametric values computed from the record ane listed in Table 1 [ss] Figures 8a and 8b respectively show the experimen- tal join distribution counters of RG) and T(t) and of Re 1 of 16 cAvemauvo/sca) 995 Tyz0e/20/¢2) uo sq alu Aaa ersoUOpU| AOad YEN AG *£L020097E002/601 01/IOp/ud> fapwAsesgousuo'sqndnGe//say wo popeDwuMod > Oe >2O2e9SI2 seu20p suowiuos anneaig s}aeudde a ha pauiano6 ae s;Ue Yo '98 o 50130) fse2q7 avo KAN Wo (e¥oNIpUe>-puE-sHAY/u0>K9 suonypuep pe sicosor6 ‘Table 1. Parametric Values Computed From the Revord Measured a U=9 me and Parameter ‘aus 256 20 2s be Bs as to 416 153, and T(,), and their corresponding theoretical counters siven by equation (18). As is seen, the distribution mode in Figure 8a is somewhat different from that in Figure 8b This is because b and v have diffrent values, as ean be seen from Table | [ss] Figure 9_ shows the histograms of conditional distributions of 7(,) assuming fy and of 7) assuming Hyss and their corresponding theoretical curves given by exuation (33). Figure 10 shows the histograms. of #4) and of 7(), and their eomesponding theoretical curves aiven by equation (48). [ss] From Figures 8, 9, and 10 we see the common features: (1) The theoretical distributions agree, o a great extent, withthe empirical distributions, and 2) The agree- rents between the theoretical and empirical distribution curves computed from th filtered record (fare better than those computed from the original record {(,), as expected 43. Laboratory Experiment I [se] This experiment is performed in the same flume as that described in section 4.2. The surface elevation C(1) is 25 XU ET AL: DISTRIBUTIONS OF SEA WAVE VARIABLES. cosn1s measured at fetch = 30 m simultaneously by two wave probes being 7.5 cm apart along the fume. Two pairs of synchronous records are obtained at U’ = 9 and 11 ms Each of these records is 10 min long and sampled at Ar = 0.05 s (N = 12000). All of these records are band passed by the FFT filtering method with the passband chosen to be 0.343, < w < 2.0u, As the passband is so narrow and the distance between the two probes is so small, the corresponding condition of the band passed records considered to be close to the condition assumed in deriving the theoretical PDF (73), [57] The instantaneous wavelength Nip )_m= 1, 2,-5.1, are computed from a pair of records according tothe formula MMbeis) = EGs3) ~ ellaix + Ax/2) — eC — Bx] mah de W (38) ee sa) wan] where x is regarded as a position parameter, A(t; x) ‘prs the taneous wave number and {Ce Sh Hier tanto ofthe ctr lt) ind cet [9 9) This formula has been frequently employed to compute the instantaneous wavelength from a pair of wave records with narrow spectra measured at suitably separated two points [eg., Melville, 1983; Hwang et al, 1989; Stansell and ‘MacFanlane, 2002). in the present experiment, x refers t0 the point with fetch = 30 m and Ax is 7.5 em. [ss] Still we shall merely show the results computed from the records measured at U = 9 ms~! and fetch = 30 m as examples in the following. Figure Ia shows the histogram, Figure 8. Comparison between the empirical joint distribution counters (solid curves) and its conresponding theoretical counters (dashed curves), (a) The joint distribution of R(,) and T(t) and its corresponding theoretical distribution. (b) The joint distributions of RU) and 7(,) and its corresponding ‘theoretical distributions, 12 of 16, cArsqysuo/ sa) prod pm dserguauquo'sgndnBe/ Pedy wos Pap pn yo fn Jo 34 40} Kia a0 Ki Uo {eUo%Pue>-puR-swuAy/ wor suowypuop pue sia, aus 995 Tp 20/20/82) Uo KEIQFaUIUG ROM I89UODUT Ad EN KE EL0200€002/EE01 asu297 suouio> anness9 sygeudde an hq pawano6 arcosor6 XU ET AL: DISTRIBUTIONS OF Si STM) a ay a ‘5 THis EA WAVE VARIABLES. “a, tO D Figure 9. (a) The histogram of 7) 9 and its comesponding theoretical curves, (b) The histograms of Mig)|Hys and its corresponding theoretical curves. The Figure 10. (a) The histogram of 7(j.) and ts correspondit and its corresponding theoretical curve. The theoretical curves are given by equation (48). 13 0f 16, theoretical curves are given by equation (33). ing theoretical curve. (b) The histogram of 1(,) cosn1s 0 (suonpu0>-pur-swmy/wo>rayieiqyouve/ Psi) Ye yo '98 Jo sn 30} Asean aio fo suomspuo pue su ous 295 Ty 202/Z0/E2) Uo esq UIUO KOM "esauapul asa EN Ae 102009" 002/6201 01/op/ua> Nays Saesqlouue'sendnbe//sany wos popeclimog s) vO0? 20229612cosor6 fa Figure 11. XU ET AL: DISTRIBUTIONS OF SEA WAVE VARIABLES. cosn1s a) Histogram of ) indirectly measured from two synchronous wave records: (a) from the filtered records and (b) fom the original records. The solid and dashed curves represent the distributions tziven by the present PDF (73) and the empirical Rayleigh PDF (56), respectively. of Mt; x) computed from the filtered records and its corresponding theoretical curve given by equation (73) in ‘which the values of m, are estimated from one of the filtered records, while Figure’Ib shows the histogram of Ny; x) computed from the original records and its corresponding theoretical curves given by equation (73) in which the values of m, are estimated from one of the original records ‘Also shown in Figures 11a and 1b are the corresponding Rayleigh curves given by the empirical PDF 40-3 elt (0) where the averaged instantaneous wave height is taken as S ‘This PDF has been employed to derive the well-known Brotschneider spectrum of sea waves [Bretschneider, 1959]. ‘Ascan be seen, Figure Ia the theoretical curve gives fairly _g004 fit fo the histogram, as expected, while in Figure Ib the fits quite poor. For the latter our explanation is that, on one hhand, both the phase functions (jy; x — Ax/2) and (ty: + Av?) computed from the original records of wind waves are rapidly fluctuated, as can be seen from Figure 7, so that they ccan not be expected to result in well reasonable X(y; x) and that, on the other hand, the condition assumed in the derivation of equation (73) deviates fiom the condition of laboratory wind waves. We attribute the anomalous rising at the left end of the histogram in Figure IIb to the ‘measurement method and the chosen distance (7.5 em) between the two probes. It is worth mentioning that if the distance was chosen to be less than 7.5 em, the two wave probes would considerably interfere each other. As can also be seen, the empirical Rayleigh distribution substantially deviates from the histogram both in Figures [1a and Lb. {s9] As well known, itis difficult to estimate mg from a ‘wind-wave record because the spectrum estimated from the record has ail tail that makes the computed m, so large that the resulting = is always close to 1. To overcome the difficult ‘we have to perform the time-smoothing [see, e.., Glazman, 1986] on both the original and filtered records: wok [ene ® where T is the smoothing scale. This time-smoothing is equivalent to multiplying the spectrum estimated from a record by the smoothing finetion WoT) = fin(wt/2)/(0T/2)P, (2) The smoothing scale Tis chosen o be the Taylor microscale, namely, 7 = (rigim:)"®, as suggested by Glazman [1986] The values of m,, ji, and \ computed from the smoothed records are listed in Table 2. These values, as parametric ‘ones, have been used in computing the theoretical curves from equation (73). 5. Conclusions [eo] We have derived the PDF (18) or (22) for the joint distribution of wave heights and periods and the PDF (68) for the join distibution of wave heights and wavelengths. While based also on a Gaussian process under the narow- spectrum hypothesis the derivation of equation (18), unlike that in the work of Longuet-Higgns [1975] nor that of Longuet-Higgins (1983), 18 in virtue of an approximate Hilbert transior that is legitimate to this hypothesis and has been widely employed to measure local wave variables from a wave record with narrow spectrum. The joint PDF (18) has the following advantages: (1) iis almost the same in magnitude as the joint PDF in the work of Longuet- ‘Higgins [1983] that has been shown to give a good ft to 14 0f 16, Arengoupyfatend) Opruo> ropmfsesgnousuorsgndnGe/ acy wos popeoqunod "002 22OZ9SI2 ne yo fn Jo 34 40} Asa a0 Ka Uo (evo; PueD-PuR-swuAy/ ora suowypuop pue sia, aus 995 Tp 202/20/82) Uo KEsQFaUIUG ROM I89UOPUT Ad EN KE EL02001E002/E201 asu297 suouio> anness9 sygeudde an hq pawano6 arcosor6 Table 2. Values of mg and X Computed From the Smoothed ‘Wave Records 146 10 OSH 874 Original eoond 02107 Xt 0383340 Fiteed second O11 t sea-wave data with a narrow spectrum: (2) depending only fon the parameter v the present joint PDF is considerably simpler in form than that in the work of Longuet-Higgins [1983] and produces an exactly Rayleigh distribution of ‘wave heights rather than the slightly non-Rayleigh distri- bution as in the work of Longuer-Higgins [1983]. This simpler solution has led to the relatively simple PDF (33) for the conditional distribution of wave periods assuming the wave height and to the relatively simple PDF (48) for the distribution of wave period. From the former we have predicted several typical conditional statistics of sea-wave periods, which may be useful to scientific and engineering applications. [1] The derivation of the joint PDF (68) is also in virtue of the approximate Hilbert transform but based on 4 unidirectional and homogeneous Gaussian wave field with narrow spectrum. It has been found that for such a ‘wave field the joint PDF of wave height and wavelength depends only on the parameter ju = (mgmgim3 — 1)", From this joint PDE we have obtained the PDF (73) of ‘wavelength, which is pethaps the first theoretical PDF proposed for the distribution of wavelengths of sea {o] ‘The numerical simulation experiment has shown the PDFs (18) or (22), (24) or (26) and (46) give well satisfactory fits to the data simulated with a narrow-band Wallops spectrum as target. The experiments of laboratory ‘wind waves have shown that the PDFs (18), (33) and (48) give firly good fits to their corresponding. histo- ‘grams and counters resulting form the band passed wave records and also that these PDFs agree to a great extent with the histograms and counters resulting from the original records of wind waves. The experiments have also shown that the agreement between the PDF (73) and its corresponding histograms is faily good for the filtered records but is rather poor for the original records, as shown in Figure 11}. We attribute the disagreement to the difference in conditions as well as the limitation of the ‘method described in equation (83) by which we measured the wavelength in the experiment Appendix A: Derivation of f(o, )) {x]_In accordance with the properties of the Hilbert transform of @ stationary process [see, eg, Papoulis, 1984], itis easy to show Ry in equation (12) t0 be Rua = Ray = Ris = Rot = Roa = Rea = Ros = Res Ry =Rn=m, Ru=Ru=m, Ro =R Ry = Ray (ay XU ET AL: DISTRIBUTIONS OF SEA WAVE VARIABLES. cosn1s It follows from equations (12) and (Al) that D = (moms — mt) = 32 (a2) and then from equation (10) that 1 1 88am as +m} (aa) malo? +€2) + mo(e? +2) The joint PDF of p, 9, 6 and @ is found by As658) Soeib9) “S680 oe (ay From equations (3) and (7) we have £0) = U0 coset) — 0 in 0) & = Hsin 0) + HEL e081) and then A688) 7 Hoe.59) ) Thus Se exo {= 35 [mut oli 4 -amre}} (a6) Integration of the above joint PDF with respect to @ from 26 to +50 and to 2 from 0 to 2 results in equation (13). [4] Acknowledgments. We ae very grate for he sponsorships of| cur work proved by the National Natal Sones Foundation of China Ur pant 40276006 and by the National 863 program under grant 2on2ah-a33100. References Biner Grapes, B. Mand S. Gran (1989), Local proper of ea waves Kayndzeglauquo'sandnBe//sdiy wos papeoKimog 's> yoo? 2z0zz9siZ fq poussno6 are sapiue yo ‘25 Jo sano} Kreuq7 suo fa.) wo (suoRpUOD-pue-swuDy/wo> seu20h suowiuon anneaig sygeadecosor6 Coastal Engineering, Sydney, Australia, pp. 227-246, Am. Soe. of| Cis Eng. Reson Vi Huang. NE. and'S. R, Long (1980), An expsrimental study of the rlceclevation probabil deribtion and statistics of wind generated wes, J Fl Mk, 101, 179-200 ‘Huang. N. ER. Long, CC. Tung. ¥. Yuan, and 1. . Bliven (1981), 1 uiiedtworparamcir wave spccual mal Tora general sea sta 2 Fld Mock, 70, 199-300. igng. NE. 8.'R Long. C-C. Tung, ¥. Yann and LF liven (1983), 'N'non-Gaisiansiatsical model Tor surface elevation of nonlineat random wave field J. Geophys Re. 88.7597 7606. yang, BA. D. Xo, and Wu (1089), Breaking of wind-generted ‘eaves: Mensrements and charters J hid Mech, 202, 177-200 Lindgren, G- (1972, Wavelength apd ampliude in Gaussian noise, Ad “ppl Proas, 4, S—08 Lindaren, Gand 1, Rychik (1982), Wave characteristic distribution for (Gaussian Waves--Wavelenath, amplitude nd stepacs, Ocsun En. 3 sna ‘Loaguet Higgins, M. S. (1952), On the statistical distbution of the ight of en waves, Mar Res, 11(3), 248-266 Longuct-Higgns, M.S. (1957), The statistical analysis of a random moving sce, Phis. Trans, Soe London 299, 521-38). ‘Longue Higins, M.S. (1963), The eff of no ieartes on statistical “suibuows in the dheaty oP Sea waves J Fhdd Mech, 16, 489480 Languet Hsing, MS" (1995), On the joint ditto of the peas ‘and amplitudes of sex waves. J. Gopi. Res. 1), 2688 2004 Longoetiggins, MS. J 989), On the joint dstbuton of wave periods ‘and amps ima random wave fic, Pro R. Soe London 389, aoe XU ET AL: DISTRIBUTIONS OF SEA WAVE VARIABLES. cosn1s Longs Higgins, M.S. (1984), Statistical properties of wave groupe Taea random sea sats, Phot Duns. R Soe" London 4, 312,219" 26 “Mali, WK. (1983), Wave modulation and breakdown, J: Fld Mech, 728, 489. 50, Paponlis, A (1984), Prohabiliy, Random Variables, and Stochasti ‘Process, 2 ed, 386 pp. McG Hil, New York Ric, SO. (1949, Mathematical of random Wass, Bel Sy. Teck 24, 86 156 Shu K.T, and W. K, Mebille (1988), Estimates of the joint statities ‘tampa and periods of san waves using an ileal ansfirm Technique, J. Geoplos Ree, SACH), 6489" 6175 Srokosr, MA. (19) On the proba of wae Breaking in dsep wate, “Z-Pivs. Oceano? 16, 02388 Siansll,P,and C MacFnlane (2002), Experimental investigation of wave "peaking ertria based on wave phase speeds. Phys. Oceano, 533), Tas Sun, F (1988), Staisical distbuion of variables of 3D ocean waves (in ‘Chinese, Chin Se, Ser 4,301 S08 Taylan, M, A. (1980), Naor noninea ssa wave, J. Geophys Res "5G, TAS 1552, “Ta, D. Xu, N. Xu, and L. Zhang, Physical Oceanography Laboratory, Ocson University of China, Gingday 265003, Chin, (Gudela@maiLoucedca) 16 0f 16, io sgndnbe//sainy wou pepeolumog 5) 002 220229Si2 “ian seu20f suowiuon anes s}aeudde a ha pauiano6 a stu Yo '98 o sn 30) f2eA97 ao Aa WO (s¥ON;pUO>-puE-sHAY/wo>fa|WKIE:AOOYO/H) suowypuop pue sia, aus 995 [p202/20/E2 uo fatsQraurug Aan I9UOPU Aala NEN Aa £L02003"E008/620101/OP/ue>ay