Standardization of data, trend removal and filtering Prior to calculating the frequency spectrum of the surface elevation ζ, various preliminary operations are usually applied to the data, i.e. data standardisation, trend removal and filtering (Massel, 1996). Filtering of wave data prior to a detailed analysis is desirable for various seasons. If we are particularly interested in wind-induced waves, the swell component should be filtered out. In the other hand, if we concentrate on the most energetic part of the wave spectrum only, the high frequency components with negligible energy should be removed by filtering. Digital filtering can be performed in either the time domain or the frequency domain To standardise the ζ values we present them in a non-dimensional form:

ξn =

ςn −ς , σς



1 N

n −1





For a stationary ergodic process, the quantity ς is an unbiased estimate of the true mean value. The unbiased estimate of the standard deviation of data ς n is given by:
2⎤ ⎡ 1 N σς = ⎢ ∑ ςn −ς ⎥ ⎣ N − 1 n −1 ⎦



1/ 2


Especially in the case of wave orbital velocity measurements in a coastal zone, in regions with a large tidal motion, removal of the spurious trend or low frequency components, with a wavelength longer than the record length, is usually required. The most common technique for trend removal is to fit a low-order polynomial to the data using the least square method. Thus, we assume that the original wave data {ς n } can be approximated by a polynomial of order K:

ς n = ∑ bk (n∆t )k n=1, 2,..., N.
~ k =0



A “least squares” fit provides a system of equations for unknown coefficients bk as (Bendat and Piersol, 1986):


∑ bk ∑ (n∆t )
k =0 n =1



k +m

= ∑ ς n (n∆t ) , m=0,1,2,...,K
m n =1



Assuming that K=1, we obtain: b0 = b1 = 2(2 N + 1)∑n =1 ς n − 6 ∑n =1 nς n

N ( N − 1)


12∑n =1 nς n − 6 ( N + 1)∑n =1 ς n

∆tN (N − 1)(N + 1)


The corrected time series ς n now becomes:

ς n (n∆t ) = ς n (n∆t ) − ∑ bk (n∆t )k
^ k =0



E.3. Calculation of frequency spectra by Blackman-Tukey method The frequency spectra are usually estimated by the Blackman-Tukey method, which is based on the Wiener-Khinchine theorem. The Blackman- Tukey procedure requires calculations in the following steps (Massel, 1996):
⎧^ ⎫ 1. Subtracting a mean value from digital data ⎨ς n ⎬ , trend removing and filtering ⎩ ⎭

(if necessary). 2. Calculating the autocorrelation function:
K (r∆t ) = 1 N −r
N −r ^ n =1

∑ ς n ς n+ r , r=0,1,2,...,m,



where r is called the lag number and m is the maximum lag number (m<N). Selection of the m value provides the optimum estimate for the autocorrelation function. A finite value of m implies that surface elevation ς ( t ) , at times t f m∆t are uncorrelated. 3. Suppression of the spectrum leakage using a window for the autocorrelation function. Such window tapers the autocorrelation function to eliminate the discontinuity at the end of function K. There are numerous such windows in use. A typical window is the Hanning window:
u hr = 1⎛ πr ⎞ ⎜ 1 + cos ⎟ . m⎠ 2⎝


The modified autocorrelation function becomes: 2

K (r∆t ) = K (r∆t )u hr .


4. Calculating the frequency spectral density by numerical integration of the autocorrelation function K (r∆t ) :
~ m −1 ~ ⎫ ∆t ⎧ ~ ⎛ πrk ⎞ ~ S (ω k ) = ⎟ + K (m∆t ) cos( πk )⎬ , ⎨ K (0 ) + 2∑ K ( r∆t ) cos⎜ π ⎩ ⎝ m ⎠ r =1 ⎭


for frequencies ω k = k∆ω =

Kπ , k=0,1,2,...,m. m∆t

⎡^ ⎤ The estimate S (ω ) describes the time average of in terms of ⎢ς ( t )⎥ in terms of its ⎣ ⎦ ⎛B frequency components laying inside the frequency band: ω − ⎜ e ⎝ 2 by the resolution bandwidth Be (rad/s). Equation (E.12) gives ⎞ ⎛B ⎟, ω + ⎜ e ⎠ ⎝ 2 ⎞ ⎟ , divided ⎠

1 m independent estimates 2

of the spectrum. The estimates separated by frequency increments smaller than correlated For a given bandwidth Be, the required maximum lag number m is:
m= 2π . B e ∆t

2 fc are m


The standard error ε of the spectrum estimation is usually presented as a function of the number of degrees of freedom n:

where n =

2 = n
2N m

m , N

(E.14) (E.15)

It should be noted that, for given N, when the maximum lag number m is small, then the error is also small. The minimum total record length t, required to achieve a desired error ε in terms of parameters, is given by:
t= 2N m ∆t = 2 . 2 ε Be ε