Weather and Climate Extremes 5-6 (2014) 40–47

Contents lists available at ScienceDirect

Weather and Climate Extremes

journal homepage: www.elsevier.com/locate/wace

Statistical analysis of extreme ocean waves in Galle, Sri Lanka

T. Thevasiyani a,n, K. Perera b
a
Postgraduate Institute of Science, University of Peradeniya, Sri Lanka
b
Department of Engineering Mathematics, Faculty of Engineering, University of Peradeniya, Postgraduate Institute of Science, University of Peradeniya, Sri Lanka

ar t ic l e i nf o a b s t r a c t

Article history: Study of extreme wave heights is vital for design of coastal structures. The purpose of this study is to
Received 4 January 2014 estimate the wave heights for several return periods in order to use them in coastal constructions. Wave
Received in revised form height measurements collected from Galle, Sri Lanka were analyzed. Analysis was separately performed
4 July 2014
for sea, swell and overall waves, season wise. Peak Over Threshold method (POT) was used for the
Accepted 28 July 2014
sample selection. The possible range of threshold values were identified by using the Mean residual life
Available online 27 August 2014
plot and the specific threshold value was selected using the Generalized Pareto Distribution (GPD). Using
Keywords: the POT method, the GPD was fitted for the sampled data and the special type of GPD was identified
Peak Over Threshold method statistically. Further diagnostic plots were obtained to ensure the validity of the distribution. Return
Mean residual life plot
levels were calculated for several return periods and the confidence intervals were constructed for the
Generalized Pareto Distribution
return levels.
Return levels
Return period Exponential distributions were the best fitted distributions for south-west (SW) monsoon and
Extreme value analysis October–November (ON) season for sea wave heights, while the Beta distribution was the best fitted
distribution for swell wave heights. Pareto distribution fits well with the overall wave heights for south-
west monsoon and overall wave heights were fitted well with Beta distribution for October–November
(ON) and March–April (MA) seasons. Analysis was omitted for March–April (MA) season and December–
February (DF) season for sea and swell waves, while December–February (DF) season was omitted for
overall waves as they did not have significant extreme values. It was found that, the return levels for sea
wave heights were comparatively higher than the swell wave heights for all the seasons. When
designing the coastal constructions, the return levels of extreme wave heights in south-west monsoon
should be considered than the other seasons, and also comparatively the return levels of sea waves have
significant impact than the swell waves. Moreover return levels of overall wave heights also have to be
considered in designing.
& 2014 Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license
(http://creativecommons.org/licenses/by-nc-nd/3.0/).

1. Introduction Sri Lanka is an island encircled by Indian Ocean and this country
is an important place of the international ship path. Particularly,
Often, engineering construction design of ships, boats, ports, Colombo and Galle are considered as, central import and export
seawall and harbors are determined by the behavior of the ocean commercial harbors. There was a directional wave buoy instrument
waves. As a consequence of human settlement along the coastline installed at Galle harbor for the purpose of collecting the wave data
and of economic activities across the ocean, extreme events in 1989 February by Coast conservation Project. With this data
generated within the ocean have to be carefully considered. Design Scheffer et al. (1994) perform the EVA and obtained the return
structure of a construction will be designed with the probability of levels for SW monsoon of sea, swell and overall wave heights using
failure for a given or expected life time of the structure. Some the Weibull distribution. In this paper Generalized Pareto Distribu-
natural events such as, rainfall, flood and occurrence of storm tion (GPD) using Peak Over Threshold(POT) method was used to
events in sea states are not following bell-shaped normal distribu- estimate the distribution of the wave heights and using the
tions. In such a situation extreme events were placed in the tail of identified distribution, return levels were calculated.
the distribution. Those extreme points were identified as outliers Unusually larger waves are called extreme waves. Even though
in preliminary statistical analysis. But an outlier has to be checked the occurrences of the extreme waves are rare, it will make
carefully and proper method of analysis has to be carried out. severe damages on the coastal and off-shore structures. It is very
important to consider the extreme wave heights during the
design process of the marine structures. Occurrences of extreme
n
Corresponding author. Tel.: þ 94 0772255188. wave heights plays an important role in the construction of
E-mail address: [email protected] (T. Thevasiyani). coastal lines and offshore structures (Soares and Scotto, 2004;

http://dx.doi.org/10.1016/j.wace.2014.07.003
2212-0947/& 2014 Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/3.0/).
 T. Thevasiyani, K. Perera / Weather and Climate Extremes 5-6 (2014) 40–47 41

Stansell, 2004). Soares and Scotto (2001) applied several para- data coverage rate was higher than the other seasons. The wave
metric models such as Weibull and f 3 to fit the long-term buoy was deployed in February 1989, and data were collected
distribution of significant wave heights. The results indicated every three hours for 5 and ½ years. Generally, in SW monsoon
that the accuracy of prediction depends on the behavior of the season, wave heights are higher than the other seasons in Galle, as
extreme wave height values. Covariate effects are often ignored it is located in south-west part of Sri Lanka. Unit of measurement
in practice because of complexity of extreme value modeling of wave height is in meters.
with covariates (Jonathan et al., 2008). Three types of waves have been separately analyzed:
GPD is the distribution for threshold excesses and the shape
parameter (ξ) is determining the tail behavior of the distribution  Sea waves: the waves generated under the influence of the
(Jonathan et al., 2008; Stansell, 2004). Rayleigh distribution under- wind within the wind field.
estimate the occurrence of extreme crest and over-estimate the  Swell waves: the waves move out from the area of generation
trough heights (Stansell, 2005). Buoy measurements are the and are no longer subjected significantly to wind action.
trustworthy data source and can be used in the analysis.  Overall waves: the combination of the sea and swell waves is
(Panchanga et al., 1998). However, coverage in data collection is called overall wave. Overall wave heights were calculated as
limited by the failures of the instrument and damage caused by follows:
climate change. qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Another vital aspect is the seasonality constraints in climate H overall ¼ H sea 2 þ H swell 2 , where H denotes the height of the
data. It was suggested that the year should be split-up into months waves.
or other sensible period by Carter and Challenor (1981). When In this study, sea, swell and overall waves were analyzed season
dealing with coastal management, analysis of flooding risk and the wise. Seasons are given below:
design of off-shore structures depend on the seasonal or monthly
characteristic in the return value calculation (Minguez et al., 2010).  December–February (DF) season.
When we omit the homogeneity and carry out the analysis, it will  March–April (MA) season.
give the unreliable results. This homogeneity can be ensured by  May–September (SW) monsoon season.
separation into carefully selected seasons (Monsoonal analysis)  October–November (ON) season.
(Minguez et al., 2010; Mendeza et al., 2008). Distribution of
extreme wave heights differs with the season in which it occurs
(Mackay et al., 2010; Soares and Scotto, 2001). Seasonality and 3. Methods
duration were contributed to accurate estimation (Mendeza et al.,
2008; Mackay et al., 2010). According to the data set samples were selected above a
Maximum individual wave heights are important in con- threshold value u and POT method was used to fit the GPD
structing offshore structure designs (CapitGo and Burrows, distribution. The Mean Residual Life plot (MRL plot) and fitted
1995; Alves et al., 2003). Peak Over Threshold (POT) method for GPD over a range of threshold values could be used for selecting
wave heights was used and it was fitted to GPD (Mackay et al., the specific threshold value. Appropriate specific threshold value
2010). Annual Maxima (AM) approach solves immediately some will be chosen by observing the stability of the GPD plot. Condi-
problems of the initial distribution method, however main tional distribution of excesses of a specific threshold u is deter-
difficulty of this method is, dealing with EVA with lack of enough mined by the probability density function, which is called
data for distribution fitting (Soares and Scotto, 2004). The AM conditional tail distribution. This conditional tail distribution can
method is not appropriate, because less number of annual be approximated using GPD, and the distribution function is given
maxima data points will not provide good estimates (Soares below.
and Scotto, 2001).  y  1=ξ
Various methods can be used for the threshold selection. Mean Gðy; σ u ; ξÞ ¼ 1  1 þ ξ
σ
residual life plot was suggested by Stansell (2004) to select the
specific cut-off value. Based on statistical theory Maximum Like- where σ is the scale parameter in the range of σ 4 0 and ξ is the
lihood Estimator (MLE) is considered as the best parameter shape parameter in the range of  1o ξ o 1
estimation method. Estimated parameters are unbiased and hav- Shape parameter of ξ in GPD is same as shape parameter of
ing relatively small variance (CapitGo and Burrows, 1995). There is Generalized Extreme Value (GEV) distribution:
a unique test of goodness-of-fit for GPD called “Bootstrap method  y  1=ξ
for GPD”. Villasenor-Alva and Gonzalez-Estrada (2009) performed 1þξ 40
σ
this test using the data obtained from Mexico City's ozone level. σ
EðyÞ ¼ 1  ξ, where ξ o 1.
Ultimate step of the analysis was the estimation of the return
Further using the sign of the shape parameter ξ; GPD will be
values with the given probability of occurrence, using the best
divided into three different types of distributions.
fitted extreme value distribution (CapitGo and Burrows, 1995).

Case 1: ξ ¼ 0: Light tail distribution. Take the limit as ξ-0

then, Gðy; σ u ; ξÞ follows Exponential distribution with mean σ
2. Data
with the distribution function given below:
 y
Data collected from the CCD-GTZ Coast conservation project Gðy; σ u ; 0Þ ¼ 1  exp 
was used for the analysis. A directional and roll buoy WAVEC, σ
manufactured by DATAWELL B.V., Netherlands was used to collect
the wave measurements from deep sea water. It was installed Case 2: ξ 40: Long tail Pareto distribution, where 1  GðyÞ
8 km south off Galle harbor at about 70 m water depth. This is the decays at the same rate as y  1=ξ for large y. And the distribu-
average water depth of the relatively narrow continental shelf. tion function defined as follows:
Even though 70 m is not considered as deep water, buoy was GðyÞ ¼ 1  cy  α , where ξ ¼ ð1=αÞ.
installed there by considering the mooring requirement and Case 3: ξ o 0: This distribution has a finite upper end point at
financial and practical operational constraints. In SW monsoon  ðσ=ξÞ. It is called Beta distribution.
42 T. Thevasiyani, K. Perera / Weather and Climate Extremes 5-6 (2014) 40–47

4. Results

Analysis was carried out season wise. The EVA was not
performed for MA and DF seasons sea and swell waves, and also
for DF season overall waves, since they did not have significant
extreme wave heights.

4.1. Extreme value analysis for sea wave heights of SW monsoon

season

According to the mean residual life plot (Fig. 1), considering the
linearity, possible range of threshold values were selected as 2.00–
2.75 m. Further specific cut-off value was selected as 2.5 m using
fitted GPD for a range of threshold values, the value which showed
stable behavior of scale and shape parameters (Fig. 2).
Samples were selected above the threshold value of 2.5 m and
GPD was fitted with ξ ¼ 0:05 using POT method. To obtain the
special type of distribution of GPD, likelihood ratio test was
performed to check the null hypothesis, H0: ξ ¼0. And null Fig. 1. Mean residual life plot of sea wave heights for SW monsoon.
hypothesis was not rejected (p-value ¼0.184 0.05), which con-
firmed ξ ¼0. Further profile likelihood confidence interval was
constructed and it is (  0.021, 0.146), which also confirms that
ξ¼ 0. Thus, the special type of GPD, the Exponential distribution is
the best distribution for the SW monsoon sea waves. Further the
Bootstrap goodness-of-fit test for GPD (p-value ¼0.831 40.05)
confirmed the Exponential distribution. Linear behavior of the
probability plot (Fig. 3a), quantile plot (Fig. 3b) and also the
density plot (Fig. 3d) concludes the goodness of fit of the
Exponential distribution. Using this distribution, return levels
were calculated for several return periods with the confidence
intervals (Table 1). Also the return level plot was constructed for
several return periods (Fig. 3c).

4.2. Extreme value analysis for swell wave heights of SW monsoon

season

Threshold value was selected as 2.3 m using MRL plot and fitted
GPD over a range of threshold values and the samples were selected Fig. 2. GPD fit for a range of 50 threshold values from 2 m to 2.75 m of sea wave
above the threshold value. The sampling distribution fitted well with heights for SW monsoon.

Fig. 3. (a) Probability plot, (b) quantile plot, (c) return level plot, and (d) density plot for sea wave heights of SW monsoon season.
 T. Thevasiyani, K. Perera / Weather and Climate Extremes 5-6 (2014) 40–47 43

Table 1
Return level comparison for SW monsoon and ON season.

Probability Return period Return level and confidence interval for SW monsoon season Return level and confidence interval for ON season

Sea wave height Swell wave height Sea wave height Swell wave height

0.200 05 4.12 (3.940,4.399) 2.82 (2.770,2.890) 2.60 (2.423,2.873) 2.11 (2.015,2.257)

0.100 10 4.44 (4.190,4.801) 2.90 (2.833,2.982) 2.82 (2.590,3.140) 2.21 (2.091,2.413)
0.066 15 4.63 (4.335,5.027) 2.92 (2.862,3.033) 2.95 (2.683,3.290) 2.26 (2.131,2.490)
0.050 20 4.77 (4.437,5.190) 2.95 (2.881,3.067) 3.04 (2.746,3.395) 2.30 (2.156,2.536)
0.020 50 5.22 (4.757,5.726) 3.00 (2.929,3.164) 3.32 (2.935,3.725) 2.41 (2.228,2.671)
0.010 100 5.58 (5.009,6.149) 3.03 (2.956,3.228) 3.53 (3.083,3.971) 2.49 (2.273,2.764)

Probability Plot Quantile Plot

0.8 2.9

Empirical
Model

2.7
0.4
2.5

0.0 2.3
0.0 0.2 0.4 0.6 0.8 1.0 2.3 2.5 2.7 2.9
Empirical Model

Return Level Plot Density Plot

3.2 3.0
Return level

2.0
f(x)

2.8
1.0
2.4
0.0
0.1 1 10 100 2.4 2.6 2.8 3.0
Return period (years) x
Fig. 4. (a) Probability plot, (b) quantile plot, (c) return level plot, and (d) density plot for swell wave heights for SW monsoon season.

the GPD with the shape parameter of  0.37. Likelihood ratio the GPD distribution is 0.11. Further profile likelihood confidence
test confirms that the shape parameter does not equal to zero interval was (0.012, 0.240). Likelihood ratio test was carried out
(p-value¼ 0.001o0.05). Profile likelihood confidence interval was (p-value ¼0.027o 0.05) and the data fits well with the Pareto
( 0.526,  0.181), which shows that, the shape parameter posi- distribution. Further, from the bootstrap goodness of fit test
tioned within the negative interval. Using Bootstrap goodness-of-fit (p-value ¼0.97040.05) and the AD test (test statistic value ¼1.421
(p-value¼ 0.77440.05) and Anderson Darling (AD) test (test statis- 40.757), confirms that the best fitted distribution is the Pareto
tic¼1.06540.757 table value), it was concluded that the shape distribution. Return levels for several return period using the
parameter is negative. Thus the data follows Beta type bounded identified distribution are in Table 2 and diagnostic plots are in
tail distribution. Using the distribution, return levels were calculated Fig. 5a–d.
for several return periods (Table 1) and diagnostic plots (Fig. 4a–d)
were constructed as well. Using the distributions return level of 4.4. Extreme value analysis for other seasons of sea, swell and overall
swell wave heights and sea wave heights were obtained for the same wave heights
return periods. And it is apparent that return levels of swell wave
heights are smaller than the sea wave heights (Table 1). Because, Threshold value for ON season for sea wave heights was 1 m by
swell waves are not affected by the wind action after the generation using MRL and fitted GPD and the shape parameter was  0.02.
of waves. So the recorded wave heights are smaller than the sea wave According to the likelihood ratio test (p-value ¼0.602 40.05) and
heights. the confidence intervals ( 0.098, 0.067) it was concluded that, the
sampled data follows Exponential distribution. Further, using
4.3. Extreme value analysis for overall wave heights of SW monsoon bootstrap test (p-value ¼0.528 40.05) and AD test (test statis-
season tic¼ 0.489 o0.757 table value) it was confirmed that the, fitted
distribution is the Exponential distribution. Using the fitted dis-
Using MRL plot and fitted GPD over a range of threshold values, tribution return levels (Table 1) were calculated and diagnostic
the selected threshold value was 2.9 m. The shape parameter of plots (Fig. 6a–d) were obtained.
44 T. Thevasiyani, K. Perera / Weather and Climate Extremes 5-6 (2014) 40–47

Table 2
Season wise return levels and confidence intervals for overall wave heights.

Probability Return Return level and confidence interval for SW Return level and confidence interval for ON Return level and confidence interval for MA
period monsoon season season season

0.200 05 4.38 (4.165,4.729) 2.79 (2.749,2.883) 2.11(2.033,2.222)

0.100 10 4.72 (4.414,5.149) 2.86 (2.815,2.975) 2.19 (2.104,2.343)
0.066 15 4.93 (4.560,5.409) 2.89 (2.845,3.023) 2.24 (2.141,2.412)
0.050 20 5.09 (4.664,5.602) 2.91 (2.864,3.055) 2.27(2.165,2.449)
0.020 50 5.62 (4.998,6.257) 2.96 (2.910,3.142) 2.37 (2.234,2.558)
0.010 100 6.06 (5.316,6.800) 2.99 (2.935,3.197) 2.43 (2.277,2.632)

Fig. 5. (a) Probability plot, (b) quantile plot, (c) return level plot, and (d) density plot for overall wave heights of SW monsoon.

For ON season of swell wave heights, threshold value was Finally, for MA season of overall wave heights the threshold was
selected as 1.3 m using the same methods. And the shape selected as 1.2 m using the same methods. The shape parameter of
parameter of GPD was  0.13. Using likelihood ratio test GPD was obtained as 0.14. Confidence interval was (0.205,
(p-value ¼0.031 o0.05) and profile likelihood confidence interval  0.058). Using likelihood ratio test (p-value¼0.001o0.05), boot-
(  0.223, 0.012) fitted distribution was identified is the Beta strap goodness of fit test (p-value¼0.31240.05) and AD test (test
distribution. Further, Bootstrap method (p-value ¼0.583 40.05) statistic¼1.36940.757) special type of the distribution was identi-
and AD test (test statistic ¼1.882 40.757) also confirmed that fied as the Beta distribution. Finally, diagnostic plots (Fig. 9a–d) were
the fitted distribution is the Beta distribution. After the identifica- obtained and return levels also calculated (Table 2) using the
tion of the distribution, return levels were calculated for several identified distribution.
return periods (Table 1) and diagnostic plots were constructed
(Fig. 7a–d).
Similarly, For ON season of overall wave heights, the threshold 5. Conclusions
value was obtained as 2.1 m. Sampling distribution fitted to
GPD with the shape parameter of  0.42. Using profile likelihood Sampled data fitted well with Exponential distribution for
confidence interval (  0.564,  0.249) and likelihood ratio test SW monsoon and ON season for sea wave heights. For swell
(p-value ¼0.000 o0.05) it was concluded that the best distribution wave heights, Beta distribution was the best fitted distribution
is the Beta distribution. Further, Bootstrap test (p-value ¼0.481 4 for SW and ON seasons. The DF and MA seasons were omitted
0.05) and AD test (test statistic ¼1.222 40.757) also ensured that from the analysis for sea as well as swell wave heights, since
the sampling distribution is the Beta distribution. Using the Beta there were no extreme wave heights observed in these seasons.
distribution return levels (Table 2) were obtained and diagnostic Since, Galle is located in the south-west part of Sri Lanka, the
plots (Fig. 8a–d) were constructed. effects due to the other seasons are comparatively less than SW
 T. Thevasiyani, K. Perera / Weather and Climate Extremes 5-6 (2014) 40–47 45

Probability Plot Quantile Plot

0.8 2.5

Empirical
Model
2.0
0.4
1.5

0.0 1.0

0.0 0.2 0.4 0.6 0.8 1.0 1.0 1.5 2.0 2.5 3.0
Empirical Model

Return Level Plot Density Plot

7
6
2.0
Return level

f(x)
4
3 1.0
2
1 0.0
0.1 1 10 100 1.0 1.5 2.0 2.5 3.0
Return period (years) x
Fig. 6. (a) Probability plot, (b) quantile plot, (c) return level plot, and (d) density plot for sea wave heights of ON season.

Probability Plot Quantile Plot

0.8 2.0
Empirical
Model

0.4 1.6

0.0 1.2
0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0 2.2
Empirical Model

Return Level Plot Density Plot

3
Return level

2.5
2
f(x)

1
1.5
0
0.1 1 10 100 1.2 1.4 1.6 1.8 2.0 2.2
Return period (years) x
Fig. 7. (a) Probability plot, (b) quantile plot, (c) return level plot, and (d) density plot for swell wave heights of ON season.
46 T. Thevasiyani, K. Perera / Weather and Climate Extremes 5-6 (2014) 40–47

Fig. 8. (a) Probability plot, (b) quantile plot, (c) return level plot, and (d) density plot for overall wave heights of ON season.

Fig. 9. (a) Probability plot, (b) quantile plot, (c) return level plot, and (d) density plot for overall wave heights of MA season.

monsoon season. When we consider the overall wave heights, in SW monsoon than other seasons and it was higher in sea
which is the combination of sea and swell wave heights, Beta waves than swell waves.
distribution was the best fitted distribution for ON and MA According to the results it can be concluded that the return
seasons, while the Pareto distribution was the best fitted levels of extreme wave heights in the SW monsoon season and the
distribution for SW monsoon season. Return levels are higher sea waves should be considered for designing coastal constructions.
 T. Thevasiyani, K. Perera / Weather and Climate Extremes 5-6 (2014) 40–47 47

Moreover return levels of overall wave heights should be taken into Alves, Jose Henrique G.M., Young, Ian R., 2003. On estimating extreme wave heights
consideration for the coastal structure designs. using combined Geosat, Topex/Poseidon and ERS-1 altimeter data. Appl. Ocean
Res. 25 (4), 167–186.
Jose, A., Villasenor-Alva, Gonzalez-Estrada, Elizabeth, 2009. A bootstrap goodness of
References fit test for the generalized pareto distribution. Comput. Stat. Data Anal. 53 (11),
3835–3841.
Minguez, R., Menendez, M., Mendez, F.J., Losada, I.J., 2010. Sensitivity analysis of
CapitGo, R., Burrows, R., 1995. Wave predictions based on scatter diagram data.
time-dependent generalized extreme value models for ocean climate variables.
A review. Adv. Eng. Softw. 23 (1), 37–47.
Carter, D.J., Challenor, P.G., 1981. Estimating return values of wave height. Institute Adv. Water Resour. 33 (8), 833–845.
of Oceanographic Sciences (Report no. 116). Stansell, Paul, 2004. Distributions of freak wave heights measured in the North Sea.
Mackay, Edward B.L., Challenor, Peter G., Bahaj, AbuBakr S., 2010. On the use of Appl. Ocean Res. 26 (1–2), 35–48.
discrete seasonal and directional models for the estimation of extreme wave Stansell, Paul, 2005. Distributions of extreme wave, crest and trough height
conditions. Ocean Eng. 37 (5–6), 425–442. measured in the North Sea. Ocean Res. 32 (8–9), 1015–1036.
Mendeza, Fernando J., Menendeza, Melisa, Lucen, Alberto, Medinaa, Raul, Graham, Jonathan, Philip, Ewans, Kevin, Forristall, George, 2008. Statistical estimation of
Nicholas E., 2008. Seasonality and duration in extreme value distributions of extreme ocean environments: the requirement for modelling directionality and
significant wave height. Ocean Eng. 35 (1), 131–138. other covariate effects. Ocean Eng. 35 (11–12), 1211–1225.
Soares, C. Guedes, Scotto, M., 2001. Modelling uncertainty in long-term predictions Scheffer, H.J., Fernando, K.R.M.D., Fittschen, T.,1994. Directional wave climate study
of significant wave height. Ocean Eng. 28 (3), 329–342. – south-west coast of Srilanka. Research report.
Soares, C. Guedes, Scotto, M.G., 2004. Application of the r largest-order statistics for Panchanga, Vijay, Zhaob, Liuzhi, Demirbilek, Z., 1998. Estimation of extreme wave
long-term predictions of significant wave height. Coastal Eng. 51 (5–6), heights using GEOSAT measurements. Ocean Eng. 26 (3), 205–225.
387–394.