Advances in Water Resources 30 (2007) 22712282

www.elsevier.com/locate/advwatres

Long-term tidal level distribution using a wave-by-wave approach

Sonia Castanedo *, Fernando J. Mendez, Raul Medina, Ana J. Abascal
Ocean and Coastal Research Group, IH Cantabria, Universidad de Cantabria. Avda. de los Castros s/n. 39005, Santander, Spain
Received 28 November 2006; received in revised form 3 May 2007; accepted 7 May 2007
Available online 21 May 2007

Abstract
Tidal analysis is usually performed in the time domain by means of the decomposition of the time series of the free surface in a number
of harmonics, characterizing every single component along a shelf or inside an estuary. Although this kind of analysis has proven to be
very useful in numerous studies, when it comes to characterizing the tide statistically (i.e., the long-term sea level distribution) this
approach is inadequate. This paper presents a dierent approach. Instead of working with the complete time series, some statistical properties of the signal, such as the probability density function (pdf) of the tidal wave heights (TWH) are used. The tidal elevation (TE) pdf is
obtained by means of a statistical procedure that consists of the denition of the compound pdf as a function of the TWH pdf and the Ushaped pdf for the elevations of a single wave. In order to have an analytical representation of the probability density functions, the use
of kernel density functions is explored. An extension to account for asymmetries in the tidal elevations is also proposed. Both, the symmetric and the asymmetric models are applied to dierent tide gauge data along the Worlds coastline (symmetric and asymmetric positive and negative skewed ). The results show that the symmetric approach is capable of representing the TE pdfs for roughly symmetric
tides. However, in shallow areas where the distortion of the tide is more pronounced, the asymmetric model provides a better description
of the TE pdfs.
2007 Elsevier Ltd. All rights reserved.
Keywords: Tidal elevation; Tidal range; Statistical analysis; Tidal asymmetry; Nodal cycle; Probability density function; Kernel density model

1. Introduction
Knowledge of sea level uctuations in a bay or estuary
has been the key stone of many scientic studies and estuarine restoration and protection projects. Designing a
channel entrance with navigation criteria, establishing
ood defences, preventing shoreline erosion and, in general, any man-made alterations to these systems require
the determination of sea level. Sea level analysis in tidal
dominated embayments, has traditionally been performed
through the calculation and prediction of tides and
currents. Usually, this has been done decomposing a
tidal record into a number of harmonics constituents
*

Corresponding author. Tel.: +34 942 201810; fax: +34 942 20 18 60.
E-mail addresses: castanedos@unican.es (S. Castanedo), mendezf@
unican.es (F.J. Mendez), medinar@unican.es (R. Medina), ana-julia.abascal
@alumnos.unican.es (A.J. Abascal).
0309-1708/$ - see front matter 2007 Elsevier Ltd. All rights reserved.
doi:10.1016/j.advwatres.2007.05.005

[7,3,6,8,12,24]. However, the study of tidal records from a

statistical point of view is increasingly used in dierent
applications.
Due to the increasing awareness of the ecological importance of wetlands and coastal aquifers, environmental studies, guidelines or even political strategies aimed at the
restoration of these usually degraded habitats, are being
developed in several countries [4,25,17,14,13]. Frequency
and duration of ooding or soil saturation, water permanence and water regime are key factors in describing the
hydrology of these systems [22,15].
Moreover, studies of extreme water levels using the joint
probability of surge and tide require a proper determination of the probabilistic distribution of tidal elevations.
Also, the importance of tidal asymmetries in the transport
and accumulation of sediments in estuaries demands a statistical study of tide. Accordingly, the probability density
function (pdf) of the tidal elevation (hereafter, TE) is

2272

S. Castanedo et al. / Advances in Water Resources 30 (2007) 22712282

increasingly used to statistically describe the tide at a specic location [23,19,26].

In this paper, the statistical distribution of tidal levels,
hereafter also named long-term water level distribution, is
examined in more detail. A statistical approach is presented
to be applied in the study of areas with predominantly
semidiurnal tides. The novelty here is that, instead of working with the complete time series, we will make use of some
statistical properties of the signal. The main assumption is
that a free surface tidal record can be represented as a consecutive series of waves with dierent wave heights (tidal
range) and wave periods (tidal period). Therefore, although
the time scales are very dierent, a parallelism between the
short-term probabilistic description of the free surface in a
sea state (1 h) and the analysis of a tidal record (years) is
explored.
2. Tide gauge data
As an example of statistical analysis of tidal records,
Fig. 1 shows the TE pdfs, relative to the mean sea level
(MSL) for several semidiurnal tidal records located along
the Worlds coastline: Santander and Bonanza in Spain,
Newlyn and Sheerness in England, Delfzijl in the Netherlands, Bremen in Germany, New Brunswick in Canada,
Boston in USA and Heywood Shoal in Australia. Data

from Spanish tidal gauges are obtained from Puertos del

Estado (Ministerio de Fomento, Spain) and the rest of
the data are extracted from X-tide package (http://
www.aterco.com/xtide/). Table 1 summarizes the main
characteristics of each tide gauge location.
To obtain these distributions, long tidal series are
needed (at least 18.6 years to include nodal modulation)
[10,5]. If a sea level record of this length is not available,
harmonic analysis is usually applied to the tidal signal, taking into consideration the nodal corrections. According to
this, Fig. 1 shows the corresponding TE pdfs associated to
tide level records at 0.05 h intervals over a 19-year period
for each location. The chosen locations exhibit dierent
tidal ranges (varying between 3.5 m at Bonanza and more
than 8.0 m at New Brunswick) and pdf shapes. Eight of
the nine pdfs present a bimodal distribution with the most
probable levels at mean high water neaps and mean low
water neaps. Note that, while in Newlyn and New Brunswick it can be assumed that these two peaks are symmetric
with regards to the MSL, in the rest of locations the tide
asymmetry is more evident. Moreover, in Bonanza and
Delfzijl, where the shallow water eects become important,
the asymmetry can be also observed in the lowest and highest astronomical tide level. Also it is interesting to observe
that these two locations have the skewed distribution peaking on a dierent side of the MSL. The TE pdf at Heywood

Fig. 1. TE pdfs at (a) Santander in Spain, (b) Newlyn in England, (c) Delfzijl in the Netherlands, (d) Bonanza in Spain, (e) Bremen in Germany, (f) Heywood
Shoal in Australia, (g) Boston in USA, (h) New Brunswick in Canada, and (i) Sheerness in England. Tidal level above mean sea level. Data from Spanish
tidal gauges are obtained from Puertos del Estado (Ministerio de Fomento, Spain); rest of data: from X-tide package (http://www.aterco.com/xtide/).

S. Castanedo et al. / Advances in Water Resources 30 (2007) 22712282

2273

Table 1
Location of the tide gauges
Tide gauge

Country

Longitude

Latitude

Location

Santander
Newlyn
Delfzijl
Bonanza
Bremen
HeywoodShoal
Boston
New Brunswick
Sheerness

N Spain
SW England
NE The Netherlands
SW Spain
N Germany
NW Australia
E USA
E Canada
SE England

3.79W
5.54W
6.93E
6.35W
8.72W
124.05E
71.05W
66.87W
0.74E

43.46N
50.1N
53.3N
36.8N
53.12N
13.47S
42.35N
45.05N
51.44N

Inside the Santander Bay

Seaward side of the South Pier in Newlyn harbour
Left bank of the river Ems estuary
Inner part of the Gualdalquivir river inlet
Bremen harbour in the Weser river
Heywood Shoal
Boston harbour
Matthews Cove (Bay of Fundy)
On a small pier in the Sheerness docks

Shoal is dierent from the others because it presents a single peak.

3. Statistical analysis of time evolving tidal waves
3.1. Wave-by-wave analysis
Given a wave record, the customary practice in wave climate analysis is to utilize the zero-upcrossing or the zerodowncrossing methods as the standard techniques for
dening waves [9]. In this study, we adopt the zero-upcrossing method to dene wave heights and periods. Using this
technique, the distance between the two adjacent zeroupcrossing points denes the tidal wave period, TWP,
and the vertical distance between the highest and lowest
points between the adjacent zero-upcrossing points is
dened as the tidal wave height, TWH. Fig. 2 shows the
application of the zero-upcrossing method to a specic
tidal record (Santander tidal gauge).
In this work, each tidal wave obtained as shown in Fig. 2,
is characterized by means of the following parameters: H is
the TWH, T is the TWP. If we call zw(t) the TE elevations
corresponding to a particular tidal wave, we can dene
the following magnitudes: a = max[zw(t)] is the wave crest
amplitude, b = jmin[zw(t)]j is the wave trough amplitude
R tT
(note that the TWH is H = a + b), m T1 t zw t dt is
the mean sea level elevation associated to the particular
q

R tT
2
tidal wave, and s T1 t zw t  m dt is the standard
deviation of the elevations. Here, a, b are referenced to the
local MSL and m is referenced to the overall MSL. Note that,

for a symmetrical tide dened as zw(t) = (H/2)cos(2pt/T),

p
a = b = 0.5H, s 2H =4 0:3535H and m = 0. However,
for a distorted tidal wave, a = aH, b = bH, s will have a
dierent value than that of a symmetrical wave, s = rH,
and m will have a non-zero value, m = lH. These dimensionless parameters, a, b, r and l, characterize every single
wave and will be used later in this work.
3.2. Empirical TWH probability density function
To illustrate the proposed method, some of the examples previously shown in Fig. 1 are analyzed. Series of
TWH and the corresponding TWP are obtained applying
the upcrossing method to a suciently long sea level
record. To do this, harmonic analysis is applied to the tidal
signals using the T_TIDE Harmonic Analysis Toolbox
[18]. As the authors recommended for time series longer
than one year but less than 18.6 years, the tidal records
were handled in a year-by-year analysis. Doing this, tidal
level records at 0.05 h intervals over a period of 19 years
were obtained.
Useful information about essential tide features at a specic location can be obtained from the TE pdf analysis.
Fig. 3 shows the sea level (left panels), the empirical joint
probability density function of TWH and TWP, pHT(H, T),
(central panels) and the marginal distribution of TWH,
pH(H), (right panels) calculated for each site. Observing
the tidal elevation time series helps us detect the dierent
characteristics of the tidal envelope in each location. For
example, at Bremen the reduced dierences between the
spring tide and neap tide induce the narrow shape of the

Fig. 2. Example of the application of the up-crossing method to a tidal record (Santander tidal gauge, N Spain).

2274

S. Castanedo et al. / Advances in Water Resources 30 (2007) 22712282

Fig. 3. Tidal elevation time series (left panels), joint probability density function of tidal wave height and tidal wave period (central panels) and marginal
empirical distribution of tidal wave height (right panels) for the tide at Newlyn in England, Delfzijl in the Netherlands, Bremen in Germany and Heywood
Shoal in Australia.

TWH pdf. On the contrary, at Heywood Shoal the neap

tide amplitude has a minimum value of almost zero resulting in a wide distribution of the TWH function. This is due
to the similar amplitude of the two main constituents
(M2 = 1.3 m, S2 = 0.97 m) producing a 14.8-day group
with an almost zero-amplitude node. Tide elevation pdf
at Delfzijl shows the asymmetry that was also appreciated
in Fig. 1. Also note the complex and bounded behavior of
H suggesting the need for a non-parametric model to
obtain a continuous description of the pdf. Regarding the
TWH pdfs, Newlyn has a minimum value around 1 m
(neap tides) and a maximum value around 6 m (spring
tides), being 4 m the most probable value of the tidal range
and, at Delfzijl, the minimum and maximum values are
approximately 2 m and 4 m, respectively, being 3 m the
most probable value of the tidal range.
On the other hand, the empirical joint probability density function of TWH and TWP, pHT(H, T), also provides
quite interesting information related to TWH. As can be
seen, the tidal period is not always dominated by the M2
component (12 h, 25 m  12.42 h). Periods vary between
12 and 13 h depending on the weight of the dierent contributions of the semidiurnal and diurnal tidal constituents. Note that the joint pdf for Heywood Shoal shows a
higher dispersion of tide periods than the rest of sites.
Tides in this area are semidiurnal but with a form number,
[F = (K1 + O1)/(M2 + S2)], of 0.206 indicating a mixedsemidiurnal regime at this location.

3.3. Estimated TWH probability density function using

kernel density function
Parametric models (e.g. Gaussian distribution) allow an
easy description of the pdf using a small number of parameters. However, and due to the complicated shapes detected
in Fig. 3, the denition of the pdfs using a non-parametric
model is preferred. To obtain this, the application of kernel
density functions is explored.
Kernel density functions (kdf) have been widely used in
statistical literature. Recently, Athanassoulis et al. [1] proposed a general kernel density model (KDM) for the analytic
representation of the pdf of any metocean parameter, provided that a representative histogram is available. Following
the mentioned work, an estimated probability density function of TWH, fH(H), is calculated using kernel density functions. The univariate pdf is represented by the KDM as


n
X
1
H  Hi
fH H
DHpH H i K
1
h
h
i1
where the positioning parameter Hi is made to coincide
with the center of the cells of the histogram frequency,
pH(Hi), DH is the bandwidth of the bars of the histogram,
DHpH(Hi) is the relative frequency of the cells of the TWH
histogram partition, K is the kernel estimator which stands
for smooth, non-negative functions, h is a parameter controlling the bandwidth of each K function and n is the number of classes of the histogram.

S. Castanedo et al. / Advances in Water Resources 30 (2007) 22712282

Martinez and Martinez [16] discuss the eciency of

several kernels and they conclude that the choice of a kernel depends on computational considerations or on the
amount of dierentiability required in the estimate. Following this, in this work the triweight kernel has been
used
(
35
1  x2 3 jxj 6 1
Kx 32
2
0
jxj > 1
Similar results were obtained using the triangular or the biweight kernels. It is important to note that these kernels are
left and right bounded, which is an important requirement
of the TWH distribution.
Fig. 4 presents the univariate TWH for the Santander
case. The KDM (fH(H)) is plotted against the empirical
pdf (pH(H)). The bandwidth h has been determined by
means
of an iterative procedure, trying to minimize
R1
2
f
H
 pH H  dH . As seen, the behaviour of the anaH
0
lytic function is adequate, providing an easy description of
the TWH pdf for further calculations.
Following a similar procedure, a KDM for the bivariate
distribution of TWH and TWP, fHT(H, T) could be
obtained. As this is not the main aim of this work, we refer
to Athanassoulis et al. [1] work to see the details of the
extension of the KDM to a bivariate variable.

2275

3.4. Symmetric TE probability density function

For a simple sinusoidal tidal wave dened as zw(t) =
(H/2)cos(2pt/T), where H/2 is the wave amplitude, H is
the TWH, and T is the TWP, the TE pdf is given by
1
fs z;H p
p 0:25H 2  z2
1
p ; 0:5H 6 z 6 0:5H 3
p 0:5H  z0:5H z
where H is known. This pdf is characterized by two sharp
peaks at high and low water, with density of probability
innity at jzj ! 0.5H (Fig. 5). This is due to the mathematical nature of the wave shape, presenting a maximum at
high water (z = 0.5H) and a minimum at low water
(z = 0.5H), that results in a local concentration of probability near the extremes, and consequently in the symmetrical U-shaped probability density.
Eq. (3) is valid for a particular value of H. However, in
the probability domain, the TWH can be considered as a
random variable H whose pdf is fH(H). Therefore, we could
obtain the TE pdf, associated to the 18.6-year time series
using the following compound pdf:
Z 1
fz z
fs z; H fH H dH
4
0

Fig. 4. Univariate kernel density model for tidal wave height (fH (H)) (solid line). Comparison with the empirical pdf, pH(H) (bars), at Santander. For this
particular case, DH = 0.1 m and h = 0.17 m.

2276

S. Castanedo et al. / Advances in Water Resources 30 (2007) 22712282

Fig. 5. (a) Symmetric wave prole; (b) tidal elevation pdf associated to (a).

Roughly speaking, for every tidal level z, the density of

probability is a weighted summation of all possible
TWHs, being fH(H) the weight function.
Fig. 6 shows some examples of the behaviour of fz(z) for
dierent theoretical probability densities for the TWH.
Results are expressed in a dimensionless form. Dierent
shapes (uniform, symmetric triangle, asymmetric triangle,
triweight kernel) and tidal ranges (between 0 and 1 and
covering or not the zero-amplitude TWH) are considered
for the TWH pdfs. We prefer to show these examples with

bounded pdfs since it is more realistic than the Gaussian

distribution used by Walden and Prescott [23] in the denition of the envelope of the tide level excluding the eect of
the M2 component. In all these cases, the resultant pdf is
zero-meaned and symmetric, since fs(z, H) is symmetric.
Fig. 6a and b show examples of the uniform distribution, with the corresponding TE pdfs bounded in
jzj = 0.5. This is obvious as the maximum tidal range is
H = 1. More interestingly, the lower tidal range conditions
the location of the peaks of density, corresponding to the

Fig. 6. Theoretical dimensionless tidal wave height pdfs (left) and associated tidal elevation pdfs (right). The labels are represented in the (i) and (j) cases.

S. Castanedo et al. / Advances in Water Resources 30 (2007) 22712282

minimum amplitudes of the TWH (jzj = 0.1 and jzj = 0.3

for H = 0.2 and H = 0.6, respectively).
The slopes of the TE pdfs depend on the concentration
of probability of the TWH pdf. For example, Fig. 6f shows
for an asymmetrical triangular distribution of TWH (with
a maximum of density of probability at H = 0.8), that the
maximum slope is obtained in jzj = 0.4. Therefore, the
higher the density of probability in H, the higher the slope
in the TE pdf in jzj = H/2. Logically, this result is due to
the U-shaped pdf for a particular sinusoidal wave that concentrates a large amount of density of probability near the
lower and upper bounds.
The location of the peaks of density is a combination of
the dierent TWH, mainly governed by the neap tides, but
it is not strictly dened by the lower TWH. These most
probable levels are usually associated to a mean neap tide
[19]. For the theoretical TWH pdfs presented in Fig. 6,
the levels that correspond to the position of the peaks
depend on the shape of the TWH pdf. The two peaks
degenerate into just one peak in Fig. 6g, due to the zeroamplitude limit for the TWH. This unimodal pattern
(example g) is usually associated to diurnal tides. In fact,
the main factor responsible for this behaviour in diurnal
tides is the similar amplitudes that usually present the O1
and K1 components, which form a 13.66-day group with
near zero-amplitude nodes.
Also note that the peaks are more pronounced (d and j)
when the TWH pdf is narrower. Logically, the asymptotic
case would correspond to a Dirac Delta TWH pdf. For this
case, we obtain Eq. (3) with both peaks having density of
probability innity.
Therefore, after this descriptive analysis covering dierent cases, one can deduce the shape of the TE pdf as a function of the TWH pdf, as well as the TWH pdf as a function
of the TE pdf.
Note that in the aforementioned analysis, we have
avoided the dependence on the TWP. It is clear that the
TE pdf depends also on the distribution of the periods
and we should consider a weighted summation depending on the value of the period and using the joint distribution of TWH and TWP, fHT(H, T), in the following way:
R1 R1
Tfs z; H fHT H ; T dH dT
fz z 0 R 01 R 1
5
TfHT H ; T dH dT
0
0
Some tests were performed using Eq. (5) but the improvement was negligible, thus suggesting that making the model

2277

more complicated is not worthwhile. Nevertheless, we

should be cautious with this fact if we wanted to use this
statistical approach to analyze, for instance, tide-induced
magnitudes such as the residual sediment transport [20,11].
3.5. Asymmetric TE probability density function
3.5.1. Parameterization of the asymmetry
The previous analysis shows that using the compound
pdf for symmetric waves may be a useful tool to estimate
the tide long-term water level distribution at a specic location. However, observing the examples shown in Fig. 1, it
can be seen that some of the TE pdfs exhibit an evident
asymmetry. The occurrence of tidal asymmetry is usually
associated to the generation of overtides (M4) in shallow
water areas [7]. Nevertheless, it has been recently demonstrated that the interaction (phase-locked patterns) of some
astronomical tidal constituents, as for example K1 and O1
with M2, or M2 and S2 with MSf, may also cause tidal
asymmetry (see i.e., [20,11,26]).
In order to evaluate the asymmetry shown by the TE
pdfs, the record of crest and trough amplitudes (ai and bi,
respectively), the standard deviation (si) and the mean level
(mi) associated to every wave i for the 19-year time series
have been obtained. As an example, Fig. 7 shows the scatterplot of these magnitudes versus the corresponding TWH
(Hi) for Santander. The magnitudes are presented in
dimensionless form, divided by every single Hi, thus representing (ai, bi, ri and li). These plots help us to detect
asymmetry, if a 5 0.5, b 5 0.5, r 5 0.3535 or l 5 0.
The analysis of the scatterplots reveals that the TE pdf at
the Santander tidal gauge is slightly distorted 
a 0:49;
 0:51; r
 0:355; l
 0. Note that a + b = 1, since
b
H = a + b. These mean values are also plotted as solid
lines. Therefore, as pointed out in Fig. 1, the TE pdf at
the Santander gauge is non-symmetric (skewed to the left
and with a more pronounced peak of density for the positive values).
3.5.2. Asymmetrical wave probability density function
Following the approach described in Section 3.4, it
would be highly desirable to be able to obtain a pdf for
every single wave that considers the asymmetry detected
previously in the TE pdf. Basically, the objective is to nd
an analytic representation of a pdf that accounts for the
asymmetry (dierent positive and negative bounds and,

Fig. 7. Horizontal axis: Tidal wave height, H (m). Vertical axis: crest (a) and trough (b) amplitude, standard deviation (r) and mean water level (l) as a
 0:51; r
function of H. Data: Santander tidal gauge. Solid line: mean value of the parameters a 0:49; b
 0:355; l
 0.

2278

S. Castanedo et al. / Advances in Water Resources 30 (2007) 22712282

consequently, dierent density of probability in the negative and positive part of the function). Moreover, the pdf
should be expressed in terms of easily obtained parameters,
such as the crest and trough amplitudes, and the standard
deviation of a single wave.
It has been found that the beta pdf is an adequate
approximate function to describe the above requirements
[2]. This pdf is dened on the interval [b + m, a + m]
and can be expressed as
fB z; a; b; s; m

z b  m

q1

a  z m

r1

a bqr1 Bq; r

where q and r are parameters that must be greater than

zero and B is the beta function. The parameters q and r
are expressed as
q

bab  s2
;
a bs2

aab  s2
a bs2

p
Note that for a symmetric wave (a = b = 0.5H, s 2H =4
and m = 0), the parameters q = r = 0.5, and the beta pdf
reduces to the one given in Eq. (3).
The performance of the beta pdf is as follows: one of the
most frequently observed causes of tide asymmetry is the
generation of compound tidal constituents and higher harmonics (overtides) of the principal constituents due to nonlinear eects in shallow waters, that is, the generation of the
M4 overtide and its interaction with the semi-diurnal lunar
tide, M2. Let us assume that the TE for these two harmonics is expressed as
zw t AM 2 cos2pt=T M 2  /M 2 AM 4 cos4pt=T M 4  /M 4

where T M 2 is the period of the semidiurnal lunar tide, Aj

and /j (j = M2, M4) are the amplitudes and phase lags,
respectively. We have obtained the empirical pdf (histogram) for the TE corresponding to Eq. (8) and compared
it with the approximate analytical solution given by Eq.
(6). In the cases analyzed, we assume AM 2 1, /M 2 0,
AM 4 0:05  0:15 and /M 4 0 90 180 . Fig. 8 reveals
the good behaviour of the beta pdf in all the cases, being
able to describe the asymmetrical U-shape (non-symmetrical bounds and non-symmetrical mass of probability in the
negative and in the positive levels).
3.5.3. Compound asymmetric TE probability density
function
Once we have parameterized the asymmetry in terms of
useful parameters and used the beta pdf to approximate
this asymmetry, we can apply the compound pdf as
Z 1
fz z
fB z; aH ; bH ; rH ; lH fH H dH
9
0

Fig. 9 shows some theoretical examples for the triweight

TWH pdf, fH(H), shown in case (h) of Fig. 6, applying
Eqs. (9) and (6). The parameters used are indicated in the
gure. As seen, the shapes of the resulting theoretical TE
pdfs are skewed depending on the value of the set of
parameters (a, b, r, l). Note that a > 0.5, and consequently
b < 0.5, produces a positive skewness in the TE pdf (longer
right tail and more pronounced peak of density for negative values). On the contrary, when a < 0.5(b > 0.5), the
TE pdf presents a negative skewness (longer left tail and
more pronounced peak of density for the positive values).

Fig. 8. Theoretical beta pdfs. Dots: empirical pdf (histogram) for the tidal elevation corresponding to Eq. (8). Solid line: approximate analytical solution
given by Eq. (6), beta pdf.

S. Castanedo et al. / Advances in Water Resources 30 (2007) 22712282

2279

Fig. 9. Theoretical dimensionless asymmetric tidal elevation pdfs.

3.6. Shallow water eects: non-linear tidal distortion

4. Application to tide gauge data

The astronomical tide is strongly distorted during its propagation from oshore into a shallow tidal basin. This distortion is due to two mechanisms: (1) Energy dissipation of the
fundamental constituents (M2, S2) by friction and (2) transfer
from the fundamental constituents to the overtides by nonlinear advection, non-linear continuity and friction [21]. As
mentioned, this generation of overtides is the most frequently
observed cause of tidal asymmetry. Moreover, the higher the
ocean tidal height, the higher the tide distortion [7].
To investigate this, in Fig. 10 the Bonanza dimensionless
tidal wave parameters, (a, b, r, l), as a function of TWH,
H, are presented.
Comparing these results with those presented in Fig. 7,
it can be observed that assuming constant parameters for
the Santander tide is a quite good approximation. However, at Bonanza the tidal wave parameters strongly
depend on the TWH. According to this, and in order to
better represent the TE pdf in shallow water areas, the
wave parameters have been expressed as a function of the
TWH by means of a linear t,

In this section, the developed models are applied to tide

gauge data. Specically, the application of the symmetric
model to some tidal gauges (Newlyn, Boston, Bonanza
and Delfzijl) is shown in Fig. 11. We compare the empirical
TE pdfs, pz(z), with the model TE pdfs, fz (z), resulting
from Eq. (4), assuming that every single wave has the symmetric U-shape pdf of Eq. (3). As seen, the agreement
between pz (z) and fz(z) for the Newlyn and Boston cases
is clearly quite good. The lower and upper bounds are perfectly modelled and the modelled peak of density coincides
very accurately with the empirical TE pdf peaks. However,
the empirical TE pdf at Bonanza is not well-tted by the
proposed model and the model performance at Delfzijl is
worse. This is directly related to the symmetry or asymmetry of the pdfs. While the model given by Eqs. (3) and (4) is
capable of representing the TE pdfs for roughly symmetric
tides, the assumption of symmetry is not adequate for sites
such as Bonanza and Delfzijl where the distortion of the
tide is much more pronounced.
We have applied the methodology presented in Eq. (9)
to the tidal gauges analyzed in Fig. 11. These results are
displayed in Fig. 12. The average parameters for every case
 r
; l
 are included in the caption. As was expected,
a; b;
due to the symmetry present in the Newlyn and the Boston
pdfs, no improvements are appreciated in these cases compared with the symmetric model showed in Fig. 11. However, one can see that, in Bonanza, the model improves
the results obtained previously and that the distortion is
better modelled using the asymmetric model. Note that at
Delfzijl, where the TE pdf deformation is much more pro-

aH a1 H a2 ;
rH r1 H r2 ;

bH b1 H b2 ;
lH l1 H l2

10

where (ai, bi, ri, li), {i = 1, 2} are the best t parameters.

Considering this parameterization, the compound pdf
for asymmetric wave remains:
Z 1
fz z
fB z; aH H ; bH H ; rH H ; lH H fH H dH
0

11

Fig. 10. Horizontal axis: Tidal wave height, H (m). Vertical axis: Crest (a) and trough (b) amplitude, standard deviation (r) and mean water level (l) as a
function of H. Data: Bonanza tidal gauge. Solid line: linear t.

2280

S. Castanedo et al. / Advances in Water Resources 30 (2007) 22712282

Fig. 11. Comparison between the empirical tidal elevation pdfs (dots) and the model pdfs (solid line) resulting from Eq. (4) assuming that every single
wave has the symmetric pdf of Eq. (3).

Fig. 12. Comparison between the empirical tidal elevation pdfs (dots) and the model pdfs (solid line) resulting from Eq. (9) assuming that every single wave
 0:503; r
 0:51; r
 0:351; l
 0, Boston: a 0:49; b
 0:35; l
 0,
has the a symmetric pdf of Eq. (6). Mean parameters at Newlyn: a 0:497; b
 0:49; r
 0:57; r
Bonanza: 
a 0:51; b
 0:356; l
 0:0045 and Delfzijl: a 0:43; b
 0:352; l
 0:0014.

S. Castanedo et al. / Advances in Water Resources 30 (2007) 22712282

2281

Fig. 13. Comparison between the empirical pdfs (dots) and the asymmetric model pdfs resulting from Eq. (11) assuming that every single wave has the
asymmetric pdf of Eq. (6). Dashed line: using mean parameters. Solid line: using linear parameters.

 0:57; r
 0:352; l
 0:0014, the
nounced, 
a 0:43; b
improvement is much more evident although the model
performance is not as good as in Bonanza.
Therefore, it seems that this simple approach in dening
 r
; l
 is
the asymmetry in terms of average parameters 
a; b;
good enough in most of the cases. However, as seen in
Fig. 12 some discrepancies can be detected if the tide distortion becomes important.
Fig. 13 shows the comparison between the empirical TE
pdf and that obtained with Eq. (11) and the parameterization given by Eq. (10) for the Bonanza and Delfzijl tidal
gauges. Although no clear improvement is obtained in
the Delfzijl case, the t becomes better for the Bonanza
curves. In this case, the peak of density coincides more
accurately with the empirical TE pdf peaks and the tails
are perfectly predicted by the model. Thus, the above analysis shows that TE pdfs in shallow areas can be calculated
by the proposed method even when the tide asymmetry is
important.
5. Conclusions
A statistical approach has been presented to obtain the
tidal elevation (TE) pdf at a specic site. The proposed
method explores the parallelism between the short-term
probabilistic description of the free surface in a sea state
and the analysis of a tidal record. The TE pdf is calculated
by means of a compound pdf expressed in terms of the tidal
wave heights (TWH) pdf and the U-shaped pdf for the elevations of a single wave. Using tidal records along the
Worlds coastline, it has been demonstrated that this
approach predicts quite accurately the TE pdf for the cases
where the tidal asymmetry is not determinant.
An extension to account for asymmetries in the tidal elevations is also proposed. The asymmetry is dened in terms
of easy wave parameters such as the dimensionless crest (a),
trough, (b), mean sea level elevation (l) and standard deviation of the elevations (r). The results show that the pro-

posed model provides an adequate description of the TE

pdfs even when tidal distortion is important.
Acknowledgements
The work is partially funded by the projects ENE200408172 and CGL2005-05365/CLI from the Spanish Ministry
of Education and Science and the Project Wave climate
duration along the Spanish littoral from the Spanish Ministry of Public Works. Sonia Castanedo and Fernando J.
Mendez are indebted to the Spanish Ministry of Education
and Science for the funding provided in the Ramon y Cajal Program. Ana Abascal was sponsored by the Marcelino
Botin Foundation. The authors wish to thank Puertos del
Estado (Spanish Ministry of Public Works) and X-tide
package (http://www.aterco.com/xtide/) for the Spanish
tidal gauges data and the rest of tidal data respectively.
References
[1] Athanassoulis GA, Belibassakis KA. Probabilistic description of
metocean parameters by means of kernel density models. 1. Theoretical background and rst results. Appl Ocean Res 2002;24:120.
[2] Deodatis G, Graham-Brady L, Micaletti R. A hierarchy of upper
bounds on the response of stochastic systems with large variation of
their properties: random variable case. Probab Eng Mech
2003;18:34963.
[3] DiLorenzo JL. The overtide and ltering response of small inlet/bay
systems. Lecture notes on coastal and estuarine studies. In: Aubrey
DG, Weishar L, editors. Hydrodynamics and sediment dynamics of
tidal inlets, vol. 29. 1988.
[4] ERA: The estuarine restoration act 2000. http://era.noaa.gov/htmls/
era/era_act.html [accessed 01/2006].
[5] Foreman MGG. Manual for tidal heights analysis and prediction.
Pacic Marine Science Report 77-10, Institute of Ocean Sciences,
Patricia Bay, Victoria, BC, 1977. 57 pp.
[6] Foreman MGG, Henry RF. The harmonic analysis of tidal model
time series. Adv Water Resourc 1989;12(3):10920.
[7] Friedrichs CT, Aubrey DG. Non-linear tidal distortion in shallow
well-mixed estuaries: a synthesis. Estuar Coast Shelf Sci
1988;27:52145.

2282

S. Castanedo et al. / Advances in Water Resources 30 (2007) 22712282

[8] Friedrichs CT, Aubrey DG. Tidal propagation in strongly convergent

channels. J Geophys Res 1994;99:332136.
[9] Goda Y. Random seas and design of maritime structures. In: Liu
Philip L-F, editor. Advanced series on ocean engineering, vol.
15. Singapore: World Scientic; 2000.
[10] Godin G. The analysis of tides. Toronto: University of Toronto
Press; 1972, ISBN 0802017479.
[11] Hoitink AJF, Hoekstra P, van Maren DS. Flow asymmetry associated with astronomical tides: implications for the residual transport
of sediment. J Geophys Res 2003;108(C10):3315. 13-1/13-8.
[12] Kreeke J, Iannuzzi RA. Second-order solution for damped cooscillating tide in narrow canal. J Hydraul Eng 1998;24:125360.
[13] Leggett, DJ, Cooper, N, Harvey, R. Coastal and estuarine managed
realignment-design issues. CIRIA; London. ISBN 0860176282. 2004.
[14] Li L, Barry DA, Stagnitti F, Parlange JY, Jeng DS. Beach water table
uctuations due to spring-neap tides: moving boundary eects. Adv
Water Resourc 2000;23:81724.
[15] Li H, Jiao JJ. Inuence of the tide on the mean watertable in an
unconned, anisotropic, inhomogeneous coastal aquifer. Adv Water
Resourc 2003;26:916.
[16] Martinez WL, Martinez AR. Computational statistics handbook with
MATLAB. Chapman & Hall/CRC; 2002.
[17] Niedowski, NL. New York State Salt Marsh Restoration and
Monitoring Guidelines. http://www.dec.state.ny.us/website/dfwmr/
marine/smguide.html>; 2000 [accessed 02/2006].

[18] Pawlowicz R, Beardsley B, Lentz S. Classical tidal harmonic analysis

including error estimates in MATLAB using T_TIDE. Comput
Geosci 2002;28:92937.
[19] Pugh D. Changing sea levels. Eects of tides, weather and
climate. Cambridge University Press; 2004.
[20] Ranasinghe R, Pattiaratchi C. Tidal inlet velocity asymmetry in
diurnal regimes. Cont Shelf Res 2000;20:234766.
[21] Speer PE, Aubrey DG. A study of non-linear tidal propagation in
shallow inlet/estuarine systems. Part II. Theory. Estuar Coast Shelf
Sci 1985;21:20724.
[22] Tiner RW. Wetland indicators: a guide to wetland identication,
delineation, classication and mapping. CRC Press LLC; 1999,
ISBN 0873718925.
[23] Walden AT, Prescott P. Statistical distributions for tidal elevations.
Geophys J Roy Astronom Soc 1983;72:22336.
[24] Walton TL. Setup and setdown in tidal bays and wetlands. Estuar
Coast Shelf Sci 2002;55:78994.
[25] WFD Directive 2000/60/EC of the European Parliament and of the
Council establishing a framework for the Community action in the
eld of water policy. <http://ec.europa.eu/environment/water/waterframework/index_en.html>; 2000 [accessed 02/2006].
[26] Woodworth PL, Blackman DL, Pugh DT, Vassie JM. On the role of
diurnal tides in contributing to asymmetries in tidal probability
distribution functions in areas of predominantly semi-diurnal tide.
Estuar Coast Shelf Sci 2005;64(23):23540.