Statistical description of wave parameters
Because of the random nature of natural waves, a statistical description is normally always used. A fair approximation of the observed distribution of wave heights is given by the Rayleigh distribution. Statistical wave parameters are often calculated based on this distribution. The most commonly used variables in coastal engineering are described below.
Contents
Most commonly used variables in coastal engineering
Significant wave height
An example of a wave record representative for a certain sea state is shown in Fig. 1. The significant wave height, [math]H_s[/math], is the mean of the highest third of the waves; instead of [math]H_s[/math] the notation [math]H_{1/3}[/math] is also often used. [math]H_s[/math] represents well the average height of the highest waves in a wave group. The significant wave height can also be computed from the wave energy. For nonbreaking waves it appears that [math]H_s \approx H_{m0} = 4 [{\overline E} / (g \rho)]^{1/2}, [/math] where [math]H_{m0}[/math] is the spectral significant wave height. An explanation, definitions and formulas are given in appendix A.
Mean wave period
The mean wave period, [math]T_m[/math], is the mean of all wave periods in a timeseries representing a certain sea state.
Peak wave period
The peak wave period, [math]T_p[/math], is the wave period with the highest energy. The analysis of the distribution of the wave energy as a function of wave frequency [math]f=1/T[/math] for a timeseries of individual waves is referred to as a spectral analysis. Wind wave periods (frequencies) often follow the socalled JONSWAP or PiersonMoskowitz spectra (see appendix B). The peak wave period is extracted from the spectra. As a rule of thumb the following relation can be used, see Fig. 5^{[1]}:
[math]T_p \approx 5 \sqrt{H_{m0}}. \qquad (1) [/math]
Mean wave direction
The mean wave direction, [math]\theta_m[/math], is defined as the mean of all the individual wave directions in a timeseries representing a certain sea state.
Description of wave conditions
These various wave parameters are often calculated from continuous or periodic timeseries of the surface elevations; typically the parameters are calculated once every one or three hours, whereby a new discrete timeseries of the statistical wave parameters is constructed. This timeseries is thereafter analyzed statistically to arrive at a condensed description of the wave conditions as follows:
 Wave height distribution represented by [math]H_s[/math] vs. percentage of exceedance. This often follows a Weibulldistribution (see appendix A and the example in Fig. 3);
 Directional distribution of the wave heights, which is often presented in the form of a wave rose (see appendix B and the example in Fig. 4);
 Scatter diagram of [math]T_p[/math] vs. [math]H_s[/math] (example in Fig. 5).
Analyses of extreme wave conditions are performed on the basis of max. wave heights in single storm events or on the basis of annual max. wave heights. These analyses are often presented as exceedance probability vs. wave heights, see Fig. 6 for an example.
Further reading
 Mangor, K., Drønen, N.K., Kaergaard, K.H. and Kristensen, S.E. 2017. Shoreline Management Guidelines. DHI Water and Environment, 451pp. ^{[2]}
 Coastal Engineering Manual, part II, chapter 1. US Army Corps of Engineers (USACE), 2008
Appendix A: Rayleigh distribution
The wave contribution to the ocean sea level [math]\eta(x,y,t)[/math] at a certain location [math]x,y[/math] is generally a superposition of a large number [math]n[/math] of random waves with amplitudes [math]a_j[/math], radial frequencies [math]\omega_j[/math] and random phases [math]\phi_j[/math], originating from different nearby and remote regions. This superposition can be represented by
[math]\eta=Re[\sum_{j=1}^{n} a_j \exp(i\omega_j t + i\phi_j)]. \qquad (A1)[/math]
The statistical distribution of wave heights was derived by LonguetHiggins (1952)^{[3]} under a few specific conditions: (a) the random numbers [math]a=\sum_{j=1}^{n} a_j \cos\phi_j, \, b=\sum_{j=1}^{n} a_j \sin\phi_j[/math] are statistically independent and normally (Gaussian) distributed; (b) the radial frequencies [math]\omega_j[/math] of the random waves are grouped in a single narrow band around a central frequency [math]\omega[/math] such that [math]\omega_j \omega_j'/ \omega \lt \lt 1[/math] for each [math]j, j'[/math]. Under this last condition the expression (A1) may be approximated for the time interval [[math]\pi / \omega \lt t\lt \pi / \omega[/math]] by
[math]\eta \approx Re[ \exp(i \omega t) \sum_{j=1}^{n} a_j \exp(i\phi_j)] \equiv \frac{1}{2} H \, Re[\exp(i \omega t+i \phi)] , \quad H = 2 \sqrt{a^2 + b^2} . \qquad (A2)[/math]
A wellknown mathematical theorem^{[4]} states that the length of a vector with Gaussian distributed components follows the Rayleigh distribution. In this case the vector length is the wave height [math]H[/math] and the components are the random numbers [math]2a, 2b[/math]. The Rayleigh probability density function [math]p_R(H)[/math] for the wave height [math]H[/math] reads:
[math]p_R(H) = \Large\frac{2H}{H_{rms}^2}\normalsize \exp\Large (–(\frac{H}{H_{rms}})^2)\normalsize . \qquad (A3)[/math]
The root mean square wave height (also called mean energy wave height) [math]H_{rms}[/math] is related to the average wave energy [math]\overline E[/math]:
[math] H_{rms}^2 = \int_0^{\infty} p_R(H) H^2 dH =\Large\frac{8}{g \rho}\normalsize \overline E. \qquad (A4)[/math]
The average wave energy is defined as [math]\overline E = g \rho \, m_0 \equiv g \rho \overline{(\eta(t)\overline{\eta})^2}[/math], where [math]g[/math] is the gravitational acceleration, [math]\rho[/math] the seawater density and where [math]\; \overline{…} \; [/math] designates the average over a period much longer than the characteristic wave periods.
The average wave height [math]\overline H[/math] is related to the root mean square wave height [math]H_{rms}[/math] by
[math]\overline H= \int_0^{\infty} p_R(H) H dH = \Large\frac{\sqrt{\pi}}{2}\normalsize H_{rms} \approx 0.89 H_{rms} .\qquad (A5)[/math]
The cumulative Rayleigh distribution (probability of wave height [math]\lt H[/math]) is given by
[math]P_R(H)=\int_0^H p_R(H')dH' = 1\exp\Large ((\frac{H}{H_{rms}})^2)\normalsize .\qquad (A6) [/math]
Assuming that wave heights are Reynolds distributed, relations can be derived between different wave parameters that are often used in practice. For the derivation of the significant wave height [math]H_s \equiv H_{1/3}[/math] (the mean of the highest 1/3part of the waves) we first determine the lowest height of the highest 1/3part of the waves, [math]H_3[/math], from the condition [math]P_R(H_3)=2/3[/math], yielding [math]H_3=H_{rms} \sqrt{\ln(3)}[/math]. The significant wave height is then related to the root mean square wave height by
[math]H_s=\Large\frac{\int_{H_3}^{\infty} p_R(H)HdH}{\int_{H_3}^{\infty} p_R(H)dH }\normalsize = 3 \, \int_{H_3}^{\infty} p_R(H)HdH \approx 1.6 \overline H = 1.42 H_{rms}. \qquad (A7)[/math]
From Eqs. (A4) and (A7), it follows that the significant wave height [math]H_s[/math] is related to the average wave energy [math]\overline E[/math] by
[math]H_s \approx H_{m0} \equiv 4 \Large \sqrt{ \frac{\overline E}{g \rho}}\normalsize \equiv 4 \sqrt{m_0}. \qquad(A8)[/math]
Extreme wave heights can be derived from the Rayleigh distribution in a similar way. For example, the mean of the 1% highest waves is given by
[math]H_{1/100} \approx 1.52 H_s . \qquad (A9)[/math]
The Rayleigh distribution has been derived under fairly restrictive conditions ((a) and (b)). Since the Rayleigh distribution does not put a limit on the wave height, it allows for unrealistic high waves. When comparing with wave height statistics obtained from field observations, it appears that the Rayleigh distribution tends to overestimate wave heights. This overestimation is greatest (more than 8%) for high longperiod waves. More elaborate distributions have been proposed in the literature that generally provide a better fit to the observations^{[5]}^{[6]}. The Weibull distribution is one of the proposed alternatives to the Rayleigh distribution. The Weibull distribution reads ^{[7]} :
[math]p_W(H)=\Large\frac{m}{\lambda}(\frac{H}{\lambda})^{(m1)}\normalsize \exp\Large (–(\frac{H}{\lambda})^{m}) \normalsize . \qquad (A10)[/math]
The Rayleigh distribution corresponds to the Weibull distribution for [math]m=2, \; \lambda=H_{rms}[/math]. The Weibull distribution has an additional parameter ([math]m[/math]) that allows suppression of the highest waves for [math]m\gt 2[/math] and an optimum adjustment to the observed wave data.
This is especially relevant for shallowwater waves, which are truncated due to depthinduced wave breaking (see Breaker index). Because of this truncation, the random numbers [math]a[/math] and [math]b[/math] in Eq. (2) are not Gaussian distributed; the wave height therefore does not follow a Rayleigh distribution. For this situation, alternative distributions have been proposed, for example by Battjes and Groenendijk (2000)^{[8]}. According to this study, a Weibull distribution with [math]m=3.6[/math] should be used above a certain threshold, [math]H_{tr}[/math] (threshold for depthinduced wave breaking). This implies that the relationships (A7A9) are not valid in shallow water. For example, if [math]H_{tr}\lt H_s[/math] (i.e. very shallow water), Eq. (A9) should be replaced by ^{[8]}
[math]H_{1/100} \approx 1.28 H_s . \qquad (A11)[/math]
Flume experiments of shallowwater wave transformation show that the value of [math]m[/math] is not constant but varies over the surf zone slope (gradual increase followed by decrease^{[9]}).
Appendix B: Frequency spectrum
A wave record can further be characterized by its frequency spectrum. The energy density spectrum of a sea state is generally designated by [math]E(f)[/math]. The total energy is given by
[math]\overline E =\int_0^{\infty} E(f)df . \qquad (B1)[/math]
The wave frequency spectrum can be determined from a wave record [math]\eta(t)[/math] by using a Fourier transform as follows: The wave energy averaged over a period [math][T/2 \lt (t t_0)\lt T/2] [/math] is given by [math]\overline{E}=\frac{g \rho}{T} \int_{T/2}^{T/2} (\eta(tt_0)  \lt \eta\gt )^2 dt[/math], where [math]\lt \eta\gt [/math] is the mean value. Inserting in this expression the Fourier development of [math](\eta(tt_0)  \lt \eta\gt )[/math] gives
[math]\overline{E}=\sum_{k=1}^{\infty} E(f_k) \Delta f , \quad f_k=k \Delta f, \; \Delta f = \frac{1}{T} , \quad E(f_k) \Delta f =\frac{g \rho}{8} H_k^2 , \quad H_k = \frac{4}{T} \int_{T/2}^{T/2} (\eta(tt_0)\lt \eta\gt ) e^{2 i \pi f_k t} dt . [/math]
The wave frequency spectrum can also be determined by modelling the windinduced wave field in a large source area. Empirical formulas have been established for fully developed wave fields under constant wind stress. For deep water without fetch restriction, it is recommended to use the adapted PiersonMoskowitz frequency distribution [math]E_{PM}[/math] ^{[10]}:
[math]E_{PM}(f) = \alpha_{PM} \Large\frac{\overline E}{f_p}(\frac{f_p}{f})^4 e^{(\frac{f_p}{f})^4}\normalsize . \qquad (B2)[/math]
For the average wave energy [math]\overline E[/math] and the peak frequency [math]f_p[/math] the following empirical expressions are found:
[math]\overline E \approx 0.005 \rho g^{1} U_{10}^4\; , \; f_p \approx 0.123 g U_{10}^{1} \; ,[/math]
where [math]g[/math] is the gravitational acceleration and [math]U_{10}[/math] the wind velocity at 10 m above the sea surface. For the coefficient [math]\alpha_{PM}[/math] a usual value is [math]\alpha_{PM} \approx 3.26[/math].
For fetchlimited seas, the spectrum is more strongly peaked around the peak frequency. For this situation, the adapted empirical JONSWAP spectrum can be used. It has the form
[math]E_{JWP} = \alpha_{JWP} \Large\frac{\overline E}{f_p}(\frac{f_p}{f})^4 e^{(\frac{f_p}{f})^4}\normalsize \gamma^\delta \;, \quad \delta =e^{\Large\frac{1}{2}\Large(\frac{(f/f_p)1}{\sigma})^2}\normalsize , \qquad (B3)[/math],
where the parameters [math]\alpha_{JWP}, \gamma, \sigma, f_p[/math] depend on the wind velocity and the fetch length and should be fitted to the wave data. For [math]\gamma=1[/math] the PiersonMoskowitz and JONSWAP spectra are the same. The most usual value is [math]\gamma=3.3[/math].
Different characteristic wave periods can be derived from the wave spectrum: the significant wave period [math]T_{01}[/math], the mean wave period [math]T_{02}[/math] and the mean energy period [math]T_E \equiv T_{m1,0}[/math]. They are given by the expressions
[math]T_{01} = \Large\frac{\int_0^{\infty} E(f)df}{\int_0^{\infty} E(f)fdf }\normalsize, \quad T_{02} = \Large \sqrt{\frac{\int_0^{\infty} E(f)df}{\int_0^{\infty} E(f) f^2 df }}\normalsize, \quad T_E \equiv T_{m1,0} = \Large\frac{\int_0^{\infty} E(f) f^{1} df}{\int_0^{\infty} E(f)df }\normalsize \; .\qquad (B4) [/math]
The peak frequency [math]f_p=T_p^{1}[/math] is related to the mean energy period [math]T_E[/math]. Fitting the PiersonMoskowitz distribution (B2) to field data yields [math]T_E/T_p \approx 0.85 [/math]; fitting the JONSWAP distribution (B3) yields [math]T_E/T_p \approx 0.9 [/math]. The ratio [math]T_E/T_p[/math] can also be derived directly from field data.
The periods defined by Eq. (B4) can be misleading when the frequency spectrum has multiple peaks corresponding to different kinds of waves, for example shortperiod sea waves, longperiod swell waves and infragravity waves with the period of wave groups. In this case, the separate periods can be determined by integrating the spectrum over frequency bandwidths associated with the different kinds of waves. An analysis of hindcasted wave data for the US Atlantic and Pacific coasts ^{[11]} yielded an overall value of [math]T_E/T_p = 0.81  0.85 [/math]. Considering separately wind wavedominated data and swelldominated data, the resulting values were [math]T_E/T_p = 0.85 0.88[/math] for wind waves and [math]T_E/T_p = 0.93 – 0.97[/math] for swell waves.
In an irregular wave field, waves may come from different directions. The wave incidence direction is an important parameter for sediment transport in the coastal zone. Waves originating from different areas may have different spectra. The directional spread of incoming waves for a particular wave frequency can be represented by a distribution function [math]D(f,\theta)[/math], where [math]\theta[/math] is the wave incidence angle. We then have
[math]\overline E \equiv \int_0^{\infty} \int_0^{2 \pi} S(f, \theta) df d \theta \equiv \int_0^{\infty} \int_0^{2 \pi} E(f) \, D(f, \theta) df d \theta \; ,\quad \int_0^{2 \pi} D(f, \theta) d \theta =1 \; .\qquad (B5) [/math]
The directional wave spectrum [math]S(f, \theta)[/math] can be derived from directional wave buoys. In practice, it is often obtained by numerical modelling of the wave field in the major source area.
References
 ↑ Mangor, K., Drønen, N.K., Kaergaard, K.H. and Kristensen, S.E. 2017. Shoreline Management Guidelines. DHI Water and Environment, 451pp.
 ↑ https://www.dhigroup.com/upload/campaigns/ShorelineManagementGuidelines_Feb2017.pdf
 ↑ LonguetHiggins, M.S. 1952. On the statistical distribution of sea waves. Journal of Marine Research 11: 245266
 ↑ https://en.wikipedia.org/wiki/Rayleigh_distribution
 ↑ Casas‐Prat, M. and Holthuijsen, L. H. 2010. Short‐term statistics of waves observed in deep water. J. Geophys. Res. 115, C09024
 ↑ Karmpadakis, I., Swan, C. and Christou, M., 2020. Assessment of wave height distributions using an extensive field database. Coastal Engineering 157, 103630
 ↑ https://en.wikipedia.org/wiki/Weibull_distribution
 ↑ ^{8.0} ^{8.1} Battjes, J.A. and Groenendijk, H.W. 2000. Wave height distributions on shallow foreshores. Coastal Engineering 40: 161182
 ↑ Xu, J., Liu, S., Li, J. and Jia, W. 2021. Experimental study of wave height, crest, and trough distributions of directional irregular waves on a slope. Ocean Engineering 242: 110136
 ↑ The Rock Manual. The use of rock in hydraulic engineering (2nd edition).CIRIA. London, 2007
 ↑ Ahn, S. 2021. Modeling mean relation between peak period and energy period of ocean surface wave systems. Ocean Engineering 228, 108937
Please note that others may also have edited the contents of this article.
