Ocean Engineering 32 (2005) 1709–1725

www.elsevier.com/locate/oceaneng

Application of fuzzy inference system in the

prediction of wave parameters
M.H. Kazeminezhad, A. Etemad-Shahidi*, S.J. Mousavi
Structure and Hydrostructure Research Center, College of Civil Engineering, Iran University of Science
and Technology, P.O. Box 16765-163, Tehran, Iran
Received 18 June 2004; accepted 1 February 2005
Available online 2 April 2005

Abstract
Wave prediction is one of the most important issues in coastal and ocean engineering studies. In
this study, the performance of Adaptive-Network-Based Fuzzy Inference System (ANFIS) and
Coastal Engineering Manual (CEM) methods for predicting wave parameters were investigated. The
data set used in this study comprises of fetch-limited wave data and over water wind data gathered
from deep-water location in Lake Ontario. The data set of year 2002 was used to develop the ANFIS
models as wave predictor models. The data set of year 2003 was then used to test the developed
ANFIS models and also the CEM method. Results indicate that ANFIS outperforms the CEM
method in terms of prediction capability as the scatter index of predictions of ANFIS is less than that
of CEM method. In particular, the CEM method overestimates the significant wave height and
underestimates the peak spectral period, while ANFIS results in more accurate predictions.
q 2005 Elsevier Ltd. All rights reserved.

Keywords: Wave prediction; Fuzzy inference system; Adaptive-network; Coastal engineering manual; Fuzzy
clustering; Fetch-limited

1. Introduction

Wind induced waves are of great importance in coastal and offshore engineering
because of their high energy. Wave parameters are required for harbor design, coastal and
offshore structures design, sediment transport estimation and other coastal engineering

* Corresponding author. Tel.: C98 21 7896623; fax: C98 21 7454053.

E-mail address: [email protected] (A. Etemad-Shahidi).

0029-8018/$ - see front matter q 2005 Elsevier Ltd. All rights reserved.
doi:10.1016/j.oceaneng.2005.02.001
1710 M.H. Kazeminezhad et al. / Ocean Engineering 32 (2005) 1709–1725

works. In many applications, having long-term observed wave data is necessary. However,
in many situations there are no long-term measurements available and hence proper wave
prediction models for hindcasting is required.
During the past decades, several numerical computer models for wave prediction have
been developed. These models generally involve the solution of energy balance equation
in finite difference form throughout a grid placed over the water area, where active wave
generation is taking place. However, in this approach, preparation of meteorological data
and computer processing demand quite heavy work, making it unattractive for practical
use (Goda, 2003). Moreover, for preliminary or even final design in some cases, use of
these models is not justified due to economic point of view (Goda, 2003). Therefore, many
engineers tend to use simplified wave prediction methods. These methods can be useful
when quick, and low cost estimates are needed. They are reasonably accurate for simple
situations where local effects are small (US Army, 2003). These methods are usually based
on semi-empirical equations developed based on interrelationships between dimensionless
wave parameters.
Several simplified wave prediction methods have been presented in the literature, such
as SMB (Bretschneider, 1970), Wilson (Wilson, 1965), JONSWAP (Hasselmann et al.,
1973), Donelan (Donelan, 1980; Donelan et al., 1985), Shore Protection Manual (US
Army, 1984) and Coastal Engineering Manual (US Army, 2003). In all these methods,
wave height and period are determined using fetch length and meteorological data such as
wind speed and duration. In most of these simplified methods, it is assumed that wind and
wave directions are the same.
Some of these methods have been investigated in different situations. Bishop (1983)
showed the accuracy of the Donelan model is slightly superior to that of the JONSWAP
and SMB models. Bishop et al. (1992) showed that the Shore Protection Manual 1984
(SPM) method tends to over predict wave height and period. It should be noted that the
SPM model has been recently replaced by CEM method and it has not been compared with
other models yet.
Artificial neural networks (ANNs) are known as flexible modeling tools with
capabilities of learning the mathematical mapping between input and output variables
of nonlinear systems. Therefore, they have been recently used for wave prediction.
Ravikant (1999), Deo et al. (2001), Agrawal and Deo (2002) and Tsai et al. (2002) have
used ANNs for forecasting wave parameters.
In the past years, Fuzzy Inference System (FIS), which is based on expertise expressed
in terms of ‘IF–THEN’ rules, has been employed in different subjects. FIS can be used to
predict uncertain systems and its application dose not require knowledge of the underlying
physical process as a precondition. To the best knowledge of the authors, this paper
presents the first application of FIS coupled with ANNs to wave prediction problem.
However, it has been used in other water resources problems such as reservoir operation
(Russel and Camplell, 1996; Shrestha et al., 1996; Dubrovin et al., 2002; Ponnambalam
et al., 2003; Mousavi et al., 2005), climatic and metrological studies (Bardossy et al., 1995;
Galambosi et al., 1998) and other related fields.
The purpose of this study is to determine the accuracy of CEM method for estimation of
wave characteristics in a fetch-limited water body. Moreover, an Adaptive-Network-
Based FIS (ANFIS) (Jang, 1993), which is a combination of ANN and FIS, is also used to
 M.H. Kazeminezhad et al. / Ocean Engineering 32 (2005) 1709–1725 1711

predict significant wave height and peak spectral period. The ANFIS-based and CEM-
based forecasts of wave height and period are subsequently analyzed and compared.

2. Study area

The data set used in this study comprises of meteorological and wave data gathered in
Lake Ontario from April to November, 2002 and further from June to November, 2003.
The data set was gathered by National Data Buoy Center (NDBC) in station 45012 at 438
37 0 09 00 N and 778 24 0 18 00 W (Fig. 1), where water depth is 145 m. Maximum measured
period at the peak of spectrum, Tp, was 10.81 s and hence most of the observed waves were
deep-water waves.
Wind and wave data was collected using 3-m discus buoy. Wave data was collected for
20 min at 1 h intervals, at a sampling frequency of 2.56 Hz. Wind data was also collected
for 8 min at 1 h intervals, at a frequency of 1.28 Hz. The wind speed at buoy was measured
at a height of 5 m above the mean sea level. The buoy measured and transmitted
barometric pressure; wind direction, speed, and guest; air and sea temperature; and wave
energy spectra. Significant wave height, dominate wave period and average wave period
were derived from wave energy spectra. To measure the wave characteristics, the
accelerometers and inclinometers of the buoy measure the heave acceleration and the
vertical displacement of the buoy hull during the wave acquisition time. Then a fast
Fourier transform (FFT) is applied to the data by the processor of the buoy to transform the
data from the temporal domain into the frequency domain.

3. Wave prediction methods

In all semi-empirical methods, it is assumed that the generation of wind waves is

mainly a function of three parameters which are wind speed, fetch length, and wind
duration. In fetch-limited condition, wave parameters are function of wind speed and fetch
length but wind duration is required for determining the fetch-limited waves.

Fig. 1. Map of the Lake Ontario and location of NDBC buoy 45012.
1712 M.H. Kazeminezhad et al. / Ocean Engineering 32 (2005) 1709–1725

In all methods, the first step towards a wave forecast is locating a fetch. Fetch is an area
of the sea surface over which a wind with a reasonably constant direction and speed is
blowing. In the CEM (US Army, 2003), fetch is defined as a region in which the wind
speed and direction variations do not exceed 2.5 m/s and 158 from the mean, respectively.
It is also assumed that a coastline upwind from the point of interest limits the fetch. Based
on the SPM and Bishop et al. (1992), the fetch length for a certain direction was
determined by constructing 30 radials from the point of interest (at 18 intervals) and
extended them until they first intersect the coastline. Then fetch length was calculated as
arithmetic average of extended radials.
To determine the duration of winds, definition of constant wind presented in the CEM
and SPM was used. In this way, wind duration at ith hourly data point was considered to be
equal to number of preceding consecutive and acceptable hours which satisfies the
following conditions

jUi K Uj! 2:5 (1)


jDi K Dj! 15 (2)
where U and D are the average of preceding consecutive and acceptable hourly wind speed
and direction, respectively. Ui and Di are wind speed and direction at ith hourly data point.
The average of wind speeds and fetch lengths over the preceding consecutive and
acceptable hours were used in wave prediction methods (Bishop, 1983).

3.1. CEM wave prediction method

CEM method is a semi-empirical method that has been developed based on

interrelationship among dimensionless wave parameters. In this method, dimensionless
wave height and period have been defined using friction velocity. This method is
appropriate for simple situation where local effects are small (US Army, 2003). The over
water wind speeds which are used in the CEM formulas must be adjusted for height and
stability effects as follow.

3.1.1. Wind speed adjustments

The CEM formulas have been developed based on measured wind speed at 10 m above
the sea surface. Therefore, the wind speed that is used in the CEM formulas must be
measured at 10 m above the sea surface. The observed wind speed at any level, z, should
be adjusted to 10 m level. The simple approximation for level adjustment given in the
CEM is

1=7
10
U10 Z Uz (3)
z
where U10 is the wind speed at 10 m above the sea surface (m/s); and Uz is the wind speed
at level z (m/s).
In the CEM, Eq. (3) is recommended for z near the 10 m (within z range of about
8–12 m) but in the SPM, this equation has been recommended for z less than 20 m.
 M.H. Kazeminezhad et al. / Ocean Engineering 32 (2005) 1709–1725 1713

Fig. 2. Amplification ratio, RT, accounting for effects of air–sea temperature differences (after CEM 2003).

Since wind speed has been measured at a height of 5 m above mean sea level, therefore,
the factor of 1.104 has been used for converting the measured wind speed to 10 m level.
According to the CEM, an adjustment for stability of the boundary layer may also be
needed. In the CEM, this adjustment is recommended for fetches longer than 16 km. Fig. 2
shows the stability correction factor where
U Z RT U10 (4)
and U is the stability compensated overeater wind speed at height of 10 m and RT is the
correction factor.

3.1.2. CEM formulas

This method can be used for predicting the significant wave height and the peak spectral
period in fetch-limited; duration limited and fully developed conditions. For accomplish-
ing fetch-limited condition, the wind duration must be greater than tmin, which is given as

X 0:67
tmin Z 77:23 (5)
U 0:34 g0:33
where X is the fetch length (m); and g is the gravitational acceleration (m/s2).
In the fetch-limited condition, the equations for predicting wave parameters are
 1=2
gHs K2 gX
Z 4:13 !10 (6)
u2 u2
 1=3
gTp gX
Z 0:651 2 (7)
u u
where Hs is the significant wave height (m); Tp is the peak spectral period (s); and u* is the
friction velocity (m/s).
1714 M.H. Kazeminezhad et al. / Ocean Engineering 32 (2005) 1709–1725

The friction velocity can be estimated as

u Z UðCD Þ1=2 (8)

where CD is the drag coefficient which is defined as:
CD Z 0:001ð1:1 C 0:035UÞ (9)

3.2. Fuzzy inference system

Fuzzy logic starts with the concept of a fuzzy set. Fuzzy set theory has been developed
for modeling complex systems in uncertain and imprecise environment. A fuzzy set is an
extension of a classical set whose elements may partially belong to that set. Suppose X is
the universe of discourse (input space) and its elements are denoted by x, then a fuzzy set A
in X is defined as a set of ordered pairs
A Z fx; mA ðxÞjx 2Xg (10)
where mA(x) is called the membership function (MF) of x in A. MF is a function that defines
how each element x in the input space is mapped to a membership value (or degree of
membership) between 0 and 1.
One of the most useful tools presented within the context of fuzzy set theory to deal
with nonlinear, but ill-defined, mapping of input variables to some output ones is what is
known as Fuzzy Inference System (FIS). FIS is a framework, which simulates the behavior
of a given system as IF–THEN rules through knowledge of experts or past available data
of the system. It is a process of how to map a set of given input variables to an output
variable using fuzzy logic. A fuzzy inference system is composed of five functional
blocks:

1. A rule base containing a number of fuzzy IF–THEN rules.

2. A database which defines the membership functions of the fuzzy sets used in the fuzzy
rules.
3. A decision making unit which performs the inference operations on the rules.
4. A fuzzification inference which transforms the crisp inputs into degree of match with
linguistic values.
5. A defuzzification interface which transforms the fuzzy results of the inference into a
crisp output.

Usually, the rule base and the database are jointly referred to as the knowledge base.
Generally, a fuzzy IF–THEN rule involves two parts. The first is IF part and the second is
THEN part which are called premise and consequent, respectively. The general form of a
fuzzy IF–THEN rule is as follows.
Rule: If x is A then z is B
Several types of FIS have been proposed in the literature (Lee, 1990). It is due to the
differences between the specification of the consequent part and the defuzzification
schemes. One of these types is the so called Takagi and Sugeno FIS (Sugeno, 1985) in
 M.H. Kazeminezhad et al. / Ocean Engineering 32 (2005) 1709–1725 1715

which the consequent variable of each rule is defined as a linear combination of input
variables. Then the final output is the weighted average of each rule’s output. For example,
a Sugeno FIS including two input variables x, y, one output variable f and two fuzzy rules
is as follows:

Rule 1: If x is A1 and y is B1 then f1 Z p1 xC q1 yC r1

Rule 2: If x is A2 and y is B2 then f2 Z p2 xC q2 yC r2

where pi, qi, and ri are the consequent parameters of ith rule. Ai and Bi are the linguistic
labels which are represented by fuzzy sets.
The so called firing strength or degree of fulfillment of a pair (x, y) to rule i, which
measures the degree to which that pair belongs to rule i, can be defined as
wi Z mAi ðxÞo mBi ðyÞ; i Z 1; 2 (11)

where mAi ðxÞ and mBj ðyÞ are membership functions of x and y in fuzzy sets Ai and Bi. ‘o’
denotes a fuzzy T-norm operator which is a function that describes a superset of fuzzy
intersection (AND) operators, including minimum or algebraic product. In this study
algebraic product was used as a T-norm operator. The final output of the system is the
weighted average of all rules outputs as
Pn
wi fi
Final output Z PiZ1 n (12)
iZ1 wi

In the significant wave height prediction problem in fetch-limited condition, x, y, and f

are wind speed, fetch length, and significant wave height, respectively. Therefore, the
fuzzy IF–THEN rules used may have the following form.
If wind speed is high and fetch length is large then
significant wave height Z pi !ðwind speedÞ C qi !ðfetch lengthÞ C ri
The parameters pi, qi, and ri are estimated using available input–output data. Similar
IF–THEN rules are used in the peak spectral period prediction problem.
There is no systematic way to know what type and shape of membership functions of
premise variables have the best performance in a defined FIS. An efficient way for doing
this is using an artificial neural networks (ANNs) model trained by input–output data. This
method is called Adaptive-Network-Based-Fuzzy Inference system (ANFIS) which is
explained in Section 3.2.1.

3.2.1. Combining neural nets and FIS

An Adaptive-Network-Based Fuzzy Inference System (ANFIS) (Jang, 1993) is a
Sugeno type FIS in which the problem of fine-tuning membership functions of premise
variables is carried out by a feed-forward neural network. ANFIS combines the advantages
of both neural networks (e.g. learning capabilities, optimization capabilities, and
connectionist structures) and fuzzy inference systems (e.g. human like ‘IF–THEN’ rule
thinking and ease of incorporating expert knowledge). The basic idea behind these neuro-
adaptive learning techniques is very simple. They provide a methodology for the fuzzy
1716 M.H. Kazeminezhad et al. / Ocean Engineering 32 (2005) 1709–1725

modeling procedure to learn information about a data set, in order to compute the
membership function parameters that best allow the associated FIS to track the given
input–output data. ANFIS is based on the premise of mapping a FIS into a neural network
structure so that the membership functions and consequent part parameters are optimized
using a hybrid learning algorithm. In this algorithm, parameters of the membership
functions are determined by a neural network back-propagation learning algorithm while
the consequent parameters by the least square method. Fig. 3 shows the structure of ANFIS
including two inputs x (wind speed), y (fetch length), and one output f (significant wave
height or peak spectral period) and two rules which were described in previous part. The
first step is the fuzzifying layer in which Ai and Bi are the linguistic labels. The output of
this layer is the membership functions of these linguistic labels. In other words, in this
step, the premise parameters are calculated. The second step calculates the firing strength
for each rule. The output of this step is the algebraic product of the input signals as can be
seen in Eq. (11). The third step is the normalized layer. Every node in this layer calculates
the ratio of the ith rule’s firing strength to the sum of all rules’ firing strength as:
wi
w i Z ; i Z 1; 2 (13)
w1 C w2
The output of every node in fourth layer is
wi fi Z wi ðpi x C qi y C ri Þ (14)
The fifth layer computes the overall output as the summation of all incoming signals,
which represents the results of significant wave height or peak spectral period as can be
seen in Eq. (12).
In this study, two ANFIS models were developed; the first one as a significant wave
height predictor and the second one as a peak spectral period predictor. For ANFIS
simulation, the data sets were divided into three groups. The first one was used as training
data, the second one as checking data, and the third one as testing data. The training and

Layer 1 Layer 2 Layer 3 Layer 4 Layer 5

x y
A1
x w1 ¯1
w
Π N w
¯ 1 f1
A2
f

B1 Π w2 N w w
¯ 2 f1
¯2
y
B2 x y

Adaptive node Fixed node

Fig. 3. ANFIS architecture.

 M.H. Kazeminezhad et al. / Ocean Engineering 32 (2005) 1709–1725 1717

checking data were used for estimating membership function parameters and controlling
the possibility of falling model into the over fitting problem, respectively. After
developing those ANFIS models, the testing data was used for validating the developed
ANFIS models. Since it is important that the ANFIS is kept as fast and efficient as possible,
a subtractive clustering method (Chiu, 1994) was used to estimate the number of clusters
and cluster centers in the data set including wind speed, fetch length, and significant wave
height or peak spectral period. This helps find an initial FIS in which the number of fuzzy
rules is manageable. Subsequently ANFIS was used to tune this initial FIS.

3.3. Selecting fetch-limited data points

In order to select the fetch-limited data points, wind duration was first calculated using
Eqs. (1) and (2). Then the values of tmin were calculated using Eq. (5) and data points with
wind duration greater than tmin, were selected. Moreover, in order to select important
waves, all data points with measured significant wave height greater than 0.3 m, and
measured peak spectral period greater than 3 s were selected (Bishop, 1983). In this way,
from 9287 hourly data points, 470 fetch-limited hourly data points were selected for
further processing, of which 270 hourly data points belong to year 2002 and 200 hourly
data points to year 2003.

4. Results and discussion

To evaluate the performance of the CEM wave prediction method, significant wave
heights (Hs) and peak spectral periods (Tp) have been predicted using wind data related to
year 2003. The comparison between observed and predicted significant wave heights and
peak spectral periods are shown in Fig. 4(a) and (b), respectively. As can be seen, the Hs
predictions are scattered relative to the measured Hs, but they are not significantly biased
while the Tp predictions are biased.
For statistical comparison of predicted and observed wave parameters, bias and scatter
index (Lin et al., 2002) were used. The bias is shown by mean error (ME) and the scatter
index (SI) is defined as the root mean square error (RMSE) normalized by the mean of
observed values of the reference quantity as follows

1 XN
bias Z ðP K Oi Þ (15)
N iZ1 i
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
u
u1 X N
RMSE Z t ðO K Pi Þ2 (16)
N iZ1 i

RMSE
SI Z !100 (17)
average observed value
where Oi is an observed value, Pi is a predicted value and N is the number of observations.
1718 M.H. Kazeminezhad et al. / Ocean Engineering 32 (2005) 1709–1725

Predicted significant wave height (m)

4

0
0 1 2 3 4 5
(a) Observed significant wave height (m)
10
Predicted peak spectral period (s)

0
0 2 4 6 8 10
(b) Observed peak spectral period (s)

Fig. 4. Comparison between observed and predicted wave parameters by CEM method.

Table 1 shows the error statistics of calculated wave parameters and also the average
values of predicted and observed wave parameters. It can be seen that the CEM method
over predicts the significant wave heights (biasZ0.1 m) and under predicts the peak
spectral periods (biasZK0.64 s) in the studied case. In addition, the scatter index for
predicted Hs (SIZ32.5%) is larger than that of predicted Tp (SIZ20.9%).
The accuracy of simplified wave prediction models depend on the fetch length
calculation, the validity of the methods to arrive at an ‘adjusted wind speed’ and
Table 1
Statistics of observed and predicted wave parameters by CEM model

Wave parameters Average observed Average predicted SI (%) Bias

values values
Hs (m) 1.15 1.25 32.5 0.10
Tp (s) 4.62 3.98 20.9 K0.64
 M.H. Kazeminezhad et al. / Ocean Engineering 32 (2005) 1709–1725 1719

the empirical formulas used to relate wave parameters to the wind speed (Bishop et al.,
1992). These issues are discussed as follows.
We assumed that fetches are limited to a coastline upwind from the point of interest.
Therefore, assumed fetch length may be greater than real fetch length that leads to
overestimating the significant wave height and peak spectral period. Results indicate that
the CEM method overestimates the significant wave height but underestimates the peak
spectral period. Therefore, the errors cannot be due to the fetch length calculation.
To adjust the wind speed, generally the wind speed profile including stratification effect
should be used which is given by (US Army, 1984)

  
U z z
Uz ¼ ln Kj (18)
k z0 L

where k is the Von karman’s constant, approximately equal to 0.4; z0, roughness height of
the surface; j, universal similarity function; L, Obukov stability length. Algebraic forms
for j and L can be found in the ACES Technical Reference (Leenhnecht et al., 1992).
Eq. (3), which is used in the CEM, is an approximation for level adjustment. Regarding
Eqs. (6)–(9), it can be seen that both significant wave height and peak spectral period
magnitudes are proportional to wind speed. If the CEM method errors presented in
Tables 1 and 2 were only due to the adjusted wind speed, then both predicted significant
wave height and peak spectral period should be overestimated or underestimated.
However, results indicate that the CEM method overestimates the significant wave height
and underestimates the peak spectral period. Therefore, the observed errors are not due to
the wind speed adjustment and they are probably due to the equations used in the CEM.
This is in line with the results obtained by Bishop et al. (1992) in Lake Ontario where they
found that the SPM method overestimated the wave parameters.
The other prediction model developed is ANFIS model. The data set of year 2002 was
used as training and checking data. Two hundred data points were used as training and 70
data points as checking data. After training, all 200 data points of year 2003 were used for
testing the model to verify the accuracy of the predicted values of wave parameters. In
other words, wind data related to year 2003 was used to predict the wave parameters and
also to determine the accuracy of the model developed.
Using training data, including wind speed and fetch length as input variables and
significant wave height as output variable, a FIS along with using a subtractive clustering
method was developed for significant wave height prediction. This FIS was then used as an
initial FIS for ANFIS modeling. ANFIS finds the best function mapping the input
variables, i.e. wind speed and fetch length, to output variable, i.e. significant wave height,

Table 2
Statistics of observed and predicted wave parameters by ANFIS model

Wave parameters Average observed Average predicted SI (%) Bias

values values
Hs (m) 1.15 1.08 17.6 K0.06
Tp (s) 4.62 4.49 10.8 K0.13
1720 M.H. Kazeminezhad et al. / Ocean Engineering 32 (2005) 1709–1725

Degree of membership
1
0.8
0.6
0.4
0.2
0
6 7 8 9 10 11 12 13 14 15
Initial Membership Function of Wind Speed (m/s)
Degree of membership

1
0.8
0.6
0.4
0.2
0
6 7 8 9 10 11 12 13 14 15
Final Membership Function of Wind Speed (m/s)

Fig. 5. Initial and final MF of wind speed for significant wave height prediction.

in fetch-limited condition. Fig. 5 shows the initial and final membership functions of wind
speed variable. There is obviously a considerable change in the shape of membership
functions after training. Fig. 6 shows the initial and final membership functions of fetch
length variable. It is seen that the change in the shape of membership functions of this
variable is not as significant as what is seen for wind speed variable. The trained FIS was
then used to predict the significant wave height using data pairs of year 2003 (testing data),
not used in training procedure. These predicted values are compared with observed data to
Degree of membership

1
0.8
0.6
0.4
0.2
0
4 6 8 10 12 14
Initial Membership Function of Fetch Length (m) x 104
Degree of membership

1
0.8
0.6
0.4
0.2
0
4 6 8 10 12 14
Final Membership Function of Fetch Length (m) x 104

Fig. 6. Initial and final MF of fetch length for significant wave height prediction.
 M.H. Kazeminezhad et al. / Ocean Engineering 32 (2005) 1709–1725 1721

Predicted significant wave height (m)

4

0
0 1 2 3 4 5
Observed significant wave height (m)

Fig. 7. Comparison between observed and predicted significant wave height by ANFIS model.

see how well the ANFIS model performs. Fig. 7 shows the comparison of observed
and predicted values. As can be seen from this figure, ANFIS has performed quite well in
predicting the significant wave height.
Using a similar procedure, another ANFIS model was developed for prediction of peak
spectral period. The initial and final membership functions of this procedure are shown in
Figs. 8 and 9. Fig. 8 shows the changes of the final membership function of wind speed
after training. Fig. 9 shows the initial and final membership function of fetch length.
Degree of membership

1
0.8
0.6
0.4
0.2
0
6 7 8 9 10 11 12 13 14 15
Initial Membership Function of Wind Speed (m/s)
Degree of membership

1
0.8
0.6
0.4
0.2
0
6 7 8 9 10 11 12 13 14 15
Final Membership Function of Wind Speed (m/s)

Fig. 8. Initial and final MF of wind speed for peak spectral period prediction.
1722 M.H. Kazeminezhad et al. / Ocean Engineering 32 (2005) 1709–1725

Degree of membership
1
0.8
0.6
0.4
0.2
0
4 6 8 10 12 14
Initial Membership Function of fetch length (m) x 10 4
Degree of membership

1
0.8
0.6
0.4
0.2
0
4 6 8 10 12 14
Final Membership Function of fetch length (m) x 10 4

Fig. 9. Initial and final MF of fetch length for peak spectral period prediction.

The developed model was then used to predict the peak spectral period using data pairs of
year 2003 (testing data). The comparison of observed and predicted peak spectral period
can be seen in Fig. 10. The figure shows that the prediction accuracy is acceptable.
Table 2 shows the error statistics of significant wave height and peak spectral period
calculated by ANFIS models and also the average values of predicted and observed wave
parameters. As can be seen, the ANFIS models underestimate both significant wave
heights (biasZK0.06 m) and the peak spectral periods (biasZK0.13 s) in the studied

10
Predicted peak spectral period (s)

0
0 2 4 6 8 10
Observed peak spectral period (s)

Fig. 10. Comparison between observed and predicted peak spectral period by ANFIS model.
 M.H. Kazeminezhad et al. / Ocean Engineering 32 (2005) 1709–1725 1723

Table 3
Statistics of predicted significant wave height by CEM and ANFIS models in different height ranges

Wave height Number of data ANFIS CEM

range (m)
SI (%) Bias (m) SI (%) Bias (m)
0.3!Hs!1.0 120 17.8 0.0 41.1 0.20
1.0!Hs!2.5 71 17.0 K0.18 27.0 0.04
HsO2.5 9 6.8 0.07 19.5 K0.68

Table 4
Statistics of predicted peak spectral period by CEM and ANFIS models in different period ranges

Wave height Number of data ANFIS CEM

range (m)
SI (%) Bias (m) SI (%) Bias (m)
3!Tp!5 138 10.1 K0.03 16.2 K0.41
5!Tp!7 53 10.7 K0.29 22.2 K0.96
TpO7 9 11.3 K0.83 26.4 K2.1

case. However, it can be seen from Tables 1 and 2 that the accuracy of ANFIS models is
better than that of the CEM model. Table 3 shows the accuracies of the ANFIS and CEM
models in predicting the Hs in different height ranges. As shown, in all range of Hs, the
scatter index of the ANFIS model in wave height prediction is less than that of the CEM
method. Moreover, the errors of both methods in prediction of Hs decrease as wave height
increase. Table 4 presents the accuracy of both methods in prediction of Tp in different
wave period ranges. As seen, in all ranges of Tp, the accuracy of ANFIS model is better
than that of the CEM method. Furthermore, the CEM method error in predicting the Tp
increases as the Tp increases while that of the ANFIS model is nearly constant.

5. Summary and conclusions

In this study, the performance of ANFIS and CEM methods for predicting the wave
parameters in fetch-limited condition were investigated. Two ANFIS models were
developed to predict wave height and period. For this purpose, the data set of year 2002 of
Lake Ontario was used as training and checking data. Then the data set of year 2003 of
Lake Ontario was used as testing data to evaluate the ANFIS models and also CEM
method. Results indicate that the errors of ANFIS models in predicting wave parameters
are less than those of the CEM method. The scatter index of the CEM method in predicting
significant wave height is 32.5% and it reduces to 19.5% for higher wave while the scatter
index of ANFIS model in prediction of significant wave height is 17.6% and it reduces to
6.8% for higher wave. The scatter index of CEM method in predicting peak spectral period
is 20.9% while the scatter index of ANFIS model in peak spectral period prediction is
10.8%. In addition, the CEM model error in predicting peak spectral period, Tp, increases
while Tp increases and the scatter index and bias increase to 26.4% and K2.1 s,
1724 M.H. Kazeminezhad et al. / Ocean Engineering 32 (2005) 1709–1725

respectively. Results show that the proposed ANFIS modeling approach outperforms the
CEM method in terms of prediction capability and the adaptive-network FIS approach is a
viable tool for modeling the wave parameters.

References

Agrawal, J.D., Deo, M.C., 2002. On-line wave prediction. Mar. Struct. 15, 57–74.
Bardossy, A., Duckstein, L., Bogardi, I., 1995. Fuzzy rule-based classification of atmospheric circulation patterns.
Int. J. Climatol. 15, 1087–1097.
Bishop, C.T., 1983. Comparison of manual wave prediction models. J. Waterway Port Coastal Ocean Eng., ASCE
109 (1), 1–17.
Bishop, C.T., Donelan, M.A., Kahma, K.K., 1992. Shore protection manual’s wave prediction reviewed. Coastal
Eng. 17, 25–48.
Bretschneider, C.L., 1970. Wave forecasting relations for wave generation. Look Lab, Hawaii 1 (3).
Chiu, S.L., 1994. Fuzzy model identification based on cluster estimation. J. Intell. Fuzzy Syst. 2 (3), 234–244.
Deo, M.C., Jha, A., Chaphekar, A.S., Ravikant, K., 2001. Neural networks for wave forecasting. Ocean Eng. 28,
889–898.
Donelan, M.A., 1980. Similarity theory applied to the forecasting of wave heights, periods and directions. In:
Proceedings of Canadian Coastal Conference. National Research Council of Canada, pp. 47–61.
Donelan, M.A., Hamilton, J., Hui, W.H., 1985. Directional spectra of wind-generated waves. Philos. Trans. R.
Soc. Lond. A315, 509–562.
Dubrovin, T., Jolma, A., Turunen, E., 2002. Fuzzy model for real-time reservoir operation. J. Water Resour. Plan.
Manage. 128 (1), 66–73.
Galambosi, A., Duckstein, L., Özelkan, E., Bogardi, I., 1998. A fuzzy rule-based model to link circulation
patterns, ENSO and extreme precipitation. In: Haimes, Y.Y., Moser, D., Stakhiv, E.Z. (Eds.), Risk-Based
Decision Making in Water Resources VIII. ASCE Press, New York, pp. 83–102.
Goda, Y., 2003. Revisiting Wilson’s formulas for simplified wind-wave prediction. J. Waterway Port Coastal
Ocean Eng., ASCE 129 (2), 93–95.
Hasselmann, K., Barnett, T.P., Bouws, E., Carlson, H., Cartwright, D.E., Enke, K., Weing, J.A., Gienapp, H.,
Hasselmann, D.E., Kruseman, P., Meerburg, A., Muller, P., Olbers, K.J., Richter, K., Sell, W., Walden, W.H.,
1973. Measurements of wind-wave growth and swell decay during the Joint North Sea Wave Project
(JONSWAP). Deutsche Hydrograph, Zeit, Erganzung-self Reihe, A 8 (12).
Jang, J.-S.R., 1993. ANFIS: adaptive-network-based fuzzy inference systems. IEEE Trans. Syst. Man Cybern. 23
(3), 665–685.
Lee, C.C., 1990. Fuzzy logic in control system: fuzzy logic controller—part I and part II. IEEE Trans. Syst. Man
Cybern. 20 (2), 404–435.
Leenhnecht, D.A., Szuwalski, A., Sherlock, A.R., 1992. Automated Coastal Engineering System Technical
Reference. Version 1.07. Coastal Engineering Research Center, Waterways Experiment Station, Vicksburg,
MS.
Lin, W., Sanford, L.P., Suttles, S.E., 2002. Wave measurement and modeling in Chesapeake bay. Continental
Shelf Res. 22, 2673–2686.
Mousavi, S.J., Ponnambalam, K., Karray, F., 2005. Reservoir operation using a dynamic programming fuzzy
rule-based approach. Journal of Water Resources Management, in press.
Ponnambalam, K., Karray, F., Mousavi, S.J., 2003. Optimization of reservoir systems operation using soft
computing tools. Fuzzy Sets Syst. 139 (2), 451–461.
Ravikant, K., 1999. Estimation of Wave Heights from Wind Speeds using Neural Networks. MTech dissertation.
Indian Institute of Technology, Bombay.
Russel, S.O., Camplell, P.E., 1996. Reservoir operating rules with fuzzy programming. J. Water Resour. Plan.
Manage. 122 (3), 165–170.
Shrestha, B.P., Duckstein, L.E., Stokhin, Z., 1996. Fuzzy rule-based modeling of reservoir operation. J. Water
Resour. Plan. Manage. 122 (4), 262–269.
 M.H. Kazeminezhad et al. / Ocean Engineering 32 (2005) 1709–1725 1725

Sugeno, M., 1985. Industrial Applications of Fuzzy Control. Elsevier Ltd. New York.
Tsai, C.P., Lin, C., Shen, J.N., 2002. Neural network for wave forecasting among multi-stations. Ocean Eng. 29,
1683–1695.
US Army, 1984.Shore Protection Manual, fourth ed., 2 vols. US Army Engineer Waterways Experiment Station,
US Government Printing Office, Washington, DC.
US Army, 2003. Coastal Engineering Manual, Chapter II-2, Meteorology and Wave Climate. Engineer Manual
1110-2-1100. US Army Corps of Engineers, Washington, DC.
Wilson, B.W., 1965. Numerical prediction of ocean waves in the north Atlantic for December 1959. Deutsche
Hydrographische Z. 18 (3), 114–130.