storm clustering and classification for the port of ... · rethymno, as mentioned above, is located...

Water Utility Journal 20: 67-79, 2018. © 2018 E.W. Publications

Storm clustering and classification for the port of Rethymno in Greece

N. Martzikos*, V. Afentoulis and V. Tsoukala Laboratory of Harbour Works, School of Civil Engineering, National Technical University of Athens * e-mail: [email protected]

Abstract: Storm events are one of the most destructive natural hazards which affect low lying coastal areas and they are responsible for urban coastal flooding and damages in ports and coastal structures. Storm analysis and classification is fundamental for understanding these extreme and violent coastal processes especially nowadays when the frequency and the severity of coastal storms are increased due to climate change and climate variability. In this work an extended analysis of storm events in Rethymno, city of Crete Island, is attempted, for wave climate projections of 1960 to 2100, under the frames of PEARL (Preparing for Extreme And Rare events in coastaL regions), an undergoing EU funded project. Based on the methodology of Dolan-Davis and Saffir-Simpson for storm and hurricane classification, the aim of this work is firstly to present an integrated analysis of storm events and their thresholds thus enhancing the storm definition for this location and secondly to classify them by means of clustering methods. A storm is usually considered as a violent hydro-meteorological phenomenon which can cause damage to the surrounding coastal zone; it starts when the significant wave height exceeds a given minimum threshold and it remains above this for a certain duration. However, the thresholds for the wave height, the duration and the calm phase between two successive storms vary depending on the area of interest. Investigating extreme historical storm events and their future projections, the thresholds’ selection is examined, checking also the independence between of two successive events. Consequently, the definition of a storm is created for this region and the storm events are classified into five classes, depending on their severity. The classification is accomplished with cluster analysis based on the storm energy and the storm period. The different clustering algorithms and methods (hierarchical, k-means, partitioning around medoids and fuzzy clustering) are validated via appropriate indices and the results are presented, checked after all the wave propagation and the variation of significant wave height close to the port structures.

Key words: extreme events; storms; thresholds; clustering evaluation; classification

1. INTRODUCTION

Storm events which affect coastal areas are known in literature as coastal storms and nowadays they concentrate the interest of many researchers from different disciplines, as they are highly destructive for coastal environments and they are responsible for major economic and human loss (Godschalk et al., 1989; Ciavola and Coco, 2017). The destructiveness of these extreme events depends on their energy and their duration as well as on the bathymetry and the coastal zone’s exposure to winds with large fetches.

In a time when the climate is changing and the impacts of this shift are becoming more and more frequent, a considerable research in this field is being conducted and storms are attracting increasing attention. In the scientific literature, the formation of a storm and the storm definition were described by many authors (Waisse and von Storch, 2010; Harley, 2017a; Ciavolla et al., 2015; Boccotti, 2014). Without big divergences in these definitions many applications regarding storm modelling were presented (De Michele et al., 2007; Corbella and Stretch, 2012; Lin-Ye et al., 2016) although the definition of storm thresholds (Almeida et al., 2012; Armaroli et al., 2012; Trifonova et al., 2012; Del Rio et al., 2012), the extended study of their impacts (Harley et al., 2017b, Davidson et al., 2017; Grottoli et al., 2017; Plomaritis et al., 2017; Puig, 2016, 2014; Coco et al. 2014) and their classification (Mendoza et al., 2011; Rangel-Buitrago and Anfuso, 2011; Dolan and Davis, 1992, 1994) gather the bulk of relevant research in this field. Consequently, the ulterior motive of this knowledge is to be used in coastal zone management, for an integrated risk assessment with a view to reducing disaster risk, and to develop early warning systems (Martinez et

68 N. Martzikos et al.

al., 2017; Ferreira et al., 2017; Costas et al., 2015; Hissel et al., 2014). In this work an analysis of storm events and their thresholds is attempted with the main objective

to present a storm classification, such as Dolan-Davis and Saffir-Simpson (Dolan and Davis 1992, 1994) intensity scales for storms and hurricanes, according to its severity and through clustering methods. Trying in this way to enlighten the use of clustering methods in storm analysis and their validation which are rarely described in the relative literature.

The results of this work are part of the research that has been carried out in the context of PEARL, an undergoing EU funded project. The main goal of PEARL is to prepare the coastal communities for extreme and rare events, mainly based on past and future projections of extreme hydrometeorological events (Markopoulos et al., 2014). Six case studies from across Europe were examined by the project, one of which was the coastal area of Rethymno city on Crete island. As described in detail by Tsoukala et al. (2016), in the past few years many extreme storm events have hit Rethymno. Both urban flooding and wave overtopping cause serious damages in the stability of breakwaters, threatening the safety of human population and interrupting the coastal economic activity.

The structure of this work is the following: Section 2, presents the study area and the dataset analysis. In Section 3, the methodology for threshold selection and the evaluation of clustering methods are described. Section 4 includes the storm classification while the impacts of storm events are presented in Section 5 and the conclusions are discussed in Section 6.

2. DATA AND STUDY AREA

2.1 The study area

Rethymno, as mentioned above, is located on the North of Crete island in Greece in the Aegean Sea (see Figure 1) and is the third most populous urban area in Crete. It is generally exposed to strong N and NW winds with large fetches which are responsible for the development of great waves.

Figure 1. Regional description of study area, (a) the Crete Island in Greece, (b) the Rethymno City, the coast (c) and the port (d) of Rethymno.

2.2 The variables

The available dataset in this work were projections of the local wave climate (1960 to 2000) to the future (2000-2100) which were derived by SWAN wave model (Booij et al. 1999; Ris et al. 1999), that was set up and used in the context of the research project named, “Estimating the effects of Climate Change on SEA level and WAve climate of the Greek seas, coastal Vulnerability and

Water Utility Journal 20 (2018) 69

Safety of coastal and marine structures - CCSEAWAVS” (Prinos, 2014) and described thoroughly by Athanassoulis et al. (2015). As input, the wind data used for the hindcast and climate simulations were proceeded by using a dynamical down-scaling approach of the atmospheric simulations conducted with the regional model RegCNET (Pal et al., 2007), driven by hindcast carbon dioxide and other greenhouse gas emissions during 1960-2000, and integrating SRES-A1B emission scenarios for the period 2000-2100 (Vagenas et al. 2014; Velikou et al. 2014). At this point, a part of the surrounding maritime area of Rethymno is selected at 35°42'00.0"N and 24°30'00.0"E in order to be exposed to developed sea states. These simulations consisted of 3-hourly timeseries with the length of 409081 values of various wave parameters in deep waters, however only the significant wave height Hs (Eq. 1), the mean wave direction Dir (Eq. 2) and the peak wave period Tp were subsequently analysed.

The above wave parameters are calculated by SWAN (The SWAN team, 2017) as following:

4 ( , )sH E d dω θ ω θ= ∫∫ (1)

sinθΕ(ω,θ)dω d180Dir = arctancosθE(ω,θ)dω d

θπ θ

⎡ ⎤⎢ ⎥⎢ ⎥⎣ ⎦

∫∫

(2)

where ( , )E ω θ is the variance density spectrum, ω is the absolute radian frequency and θ is the direction. The peak period is obtained using the maximum of a parabolic fitting through the highest bin and two bins on either side of the discrete wave spectrum. The mean wave direction is measured counter clockwise, so it denotes where the waves are going to or where the wind is blowing to.

2.3 Data analysis

Regarding the data analysis, Figure 2 presents the direction of wave occurrence with associated significant wave height, where it is confirmed that intense waves with significant wave height greater than 2 m are developed mainly on the North direction and, as expected, a slight increase for the wave height is obvious for the period of 2000-2100 (Figure 2b). Furthermore, the descriptive statistics of dataset, with the maximum (max), the mean value (m), the standard deviation (s) and the coefficient of variation (cv) are presented in Table 1. In Figure 3 a density scatterplot shows the correlation of significant wave height (Hs) and the wave peak period (Tp) with a colour gradation according to the wave direction (Dir) for the dataset. The fact that extreme values of Hs and Tp do not occur simultaneously is clearly because of the existence of swell.

Figure 2. Direction wave occurrence with associated significant wave height at Rethymno, for the period 1960-2000 (a) and for the period 2000-2100 (b).


Table 1. The descriptive statistics of data.

Full Dataset 1960-2000 2000-2100 Hs (m) Tp (s) Hs (m) Tp (s) Hs (m) Tp (s)

max 7.68 13.57 6.31 11.82 7.68 13.57 m 0.81 4.75 0.80 4.71 0.82 4.76 s 0.64 1.63 0.63 1.63 0.64 1.63

cv 0.79 0.34 0.79 0.35 0.78 0.34

Figure 3. Scatter plot of significant wave height (Hs) and the wave peak period (Tp) according to the wave direction (Dir) and their marginal density histograms, for the period 1960-2100.

3. METHODOLOGY

3.1 Storm thresholds

In order to define a storm event in detail (see also Figure 4), the storm intensity, the thresholds of important variables (i.e. HS), the minimum duration, as well as the calm period between two consecutive storm events are needed (Lin-Ye et al., 2016; Corbella and Stretch, 2012). The optimum selection of thresholds is always a laborious procedure which, to some extent, is quite subjective.

Figure 4. Timeseries of significant wave height (Hs) of the available data and the description of storm event.


Storms are extreme and rare events, so a threshold for the significant wave height could be defined as a high percentile of the data set, as it has already been used in literature (Martzikos et al., 2017; Rangel-Buitrago and Anfuso, 2011). However, a stability check of the parameters of distribution for a range of thresholds, proposed by Coles (2001), provides a better way to select one (Bernardara et al., 2014). Having confirmed that timeseries of the significant wave height of our data follows Weibull distribution, then extreme value theory implies that generalized Pareto distribution adequately approaches the upper tail above a threshold. Thus, after fitting a generalized Pareto (Gilleland and Katz, 2016) for different thresholds, the parameters of scale (σ*) and shape (ξ) are estimated in order to identify when a further increase does not affect their values. The values of σ* and ξ according to the threshold range are shown in Figure 5. Taking into account thresholds of Hs at a range of 1.8 - 2.5 m which could describe extreme events in the Aegean Sea, a threshold of 2.13 m is a reasonable value while there is a stability before and after this for both parameters and confidence interval bars (vertical lines). On the other hand, the exceedances of this threshold of 2.13 m correspond to the 5% of the total wave data for the HS, so the 95th percentile of the data is defined as the new threshold for this analysis.

Figure 5. Stability check plots for shape and reparametrized scale parameters of generalized Pareto distribution for significant wave height of dataset at a range of thresholds.

The duration of a storm event is defined as the time period in which the significant wave height remains over a threshold (Boccotti, 2014). In this work, trying to have events which could be analysed, the duration is considered not be less than 9 hours, taking into account the smallest event to be consisted of 4 values.

The calm phase or the inter-arrival time (De Michele et al, 2007) between two consecutive storms is of great importance, assuming the investigation of independence of two events. For this purpose all the events with inter-arrival time not exceeding 48 hours are checked on their correlation, estimating Kendall's tau and Spearman's rho statistic test. The p-value should be lower than 0.05 in order to reject the null hypothesis of having dependent events. The results show that the correlation has no evident trend concerning the duration between two events, but the first low p-values are indicated for the calm phase of 18 hours. According to this threshold, the events which are developed at a time interval of less than 18 hours, are considered as the same storm event and they are unified with measurements smaller than threshold.

Based on the above analysis and for the specific area of Rethymno, a storm event is defined as an extreme wave event which has a significant wave height over the threshold of 2.13 m, for a minimum duration of 9 hours and a minimum inter-arrival time of 18 hours from the next event.

Regarding the direction, the data are sorted in three categories for North, Northeast and Northwest incident waves at the Rethymno coastal zone. The waves of these directions have high


frequencies of occurrence as shown in Figure 2, and additionally vertically incident waves which are considered of high-risk for the coastal zone (Allsop et al., 2008) are also responsible for the highest waves (greater of 2.13 m). It should be noted that the direction of a storm event is identified with the mean direction of each event, while the standard deviation and the coefficient of variation do not fluctuate greatly around the mean, so the average value could sufficiently represent the whole storm event, achieving in this way, a better overall view than the direction of the storm’s peak which is usually used in literature.

Finally, the storm intensity is approached here by the definition of storm energy (Eq. 3) as it was proposed by Dolan and Davis (1992, 1994) and was also used by Mendoza et al. (2011) and Kokkinos et al., (2014) for the Mediterranean Sea, where t1 – t2 denotes to storm duration. The whole procedure, which is followed for the determination of storm thresholds, is summarized in Figure 6.

E = Hs

2

t1

t2

∫ dt (3)

Figure 6. The flow chart of the procedure followed in the determination of storm thresholds.

3.3 Storm clustering

The aim of clustering is to divide data into groups with similar properties. In previous studies, hierarchical agglomerative cluster analysis was carried out with Ward's method (Kokkinos et al., 2014) or average linkage method (Dolan and Davis, 1992, 1994; Mendoza et al. 2011), using the euclidean distance between two objects and the energy content as a classification variable.

In this work the cluster analysis is based upon the energy and the average peak period, with the overall goal to associate a storm event with the most important variables (i.e. HS, duration and TP) and finally to classify storm events for the period of 1960-2000 (past climate) and for 2000-2100 (future climate). The classification is arranged into five classes, in agreement with Dolan-Davis Scale and the Saffir-Simpson Hurricane Wind Scale (Dolan and Davis, 1992) which are extensively used by many scientists and organizations (Kelman, 2013).


However, many studies in Data Science (de Morsier et al., 2015; Arbelaitz et al., 2013; Jain, 2010; Halkidi et al., 2001) have proposed a lot of indices in order to evaluate the clustering results and find the optimum method. An integrated analysis consequently is conducted for the purpose of clarifying the clustering procedure.

Before trying to find patterns in our data it is necessary to investigate if it is highly clusterable. The Hopkins statistic value (Banerjee and Dave, 2004) indicates, especially if it is lower than 0.5, that there are clusters in our data and so it is worth proceeding, which in this dataset is confirmed for both time periods and for all directions. Next for the validation of clustering methods, the connectivity, Dunn and silhouette (Sil.) indices are estimated. The connectivity (Handl, et al., 2005) indicates the degree of clusters’ association and should be closer to zero. Dunn index (Dunn, 1974) is intended to specify the density of clusters and how well-separated they are; Dunn value should be maximized. The silhouette sidth (Kaufman and Rousseeuw, 2005) is the mean of each silhouette value which estimates the degree of confidence in a specific cluster, values close to one indicate well clustered data without outliers. Correspondingly, many measures evaluate the stability of clustering results removing one element at each time and comparing the results. Thus, the average proportion of non-overlap (APN), the average distance (AD), the average distance between means (ADM), and the figure of merit (FOM) are also estimated as they were proposed and described by Datta and Datta, (2003), while small values for all of them are preferred. Note that, due to many indices which are examined, importance is given to a general view of each method, selecting in this way, the best method which on average meets all the conditions.

All the above estimations are carried out by R language (R Core Team, 2017) and mainly based on specific packages of Brock et al. (2008), Maechler (2016) and Kassambara and Mundt (2016). Here the typical cluster algorithms which are examined are hierarchical agglomerative clustering, k–means, fuzzy and partitioning around medoids (Pam), for all the clustering methods (i.e. Ward, single – complete and average linkage) and for two types of distance, euclidean and manhattan. The internal and stability clustering indices are thoroughly presented underneath in Table 2 and Table 3, for the storm events of North direction and for the period of 1960-2000. Authors would like also to highlight that a detailed description of clustering procedure and their methods is absent in this work, prompting the reader for an extensive study of basic books (e.g. Theodoridis and Koutroumbas, 2009) and references therein.

Table 2. Internal clustering indices for storm events of North direction for the period 1960-2000

Distance Euclidian Manhattan

Method Connectivity Dunn Sil. Connectivity Dunn Sil.

Hierarchical Agglomerative

Ward 12.58 0.00 0.58 7.81 0.00 0.60 Single 17.65 0.09 0.76 17.65 0.09 0.76

Complete 11.83 0.01 0.61 13.10 0.01 0.56 Average 16.80 0.03 0.63 16.16 0.02 0.63

K-means Ward 15.19 0.00 0.62 15.43 0.00 0.62 Other 23.29 0.00 0.63 23.28 0.00 0.63

Funny - 21.75 0.00 0.52 24.02 0.00 0.52 Pam - 18.00 0.00 0.57 16.62 0.00 0.57

Table 3. Stability clustering indices for storm events of North direction for the period 1960-2000

Distance Euclidian Manhattan

Method APN AD ADM FOM APN AD ADM FOM

Agglomerative

Ward 0.37 132.72 75.07 107.53 0.45 141.01 93.21 107.53 Single 0.01 226.62 30.78 135.58 0.01 227.11 30.78 135.58

Complete 0.26 140.44 74.72 108.09 0.38 156.69 97.36 108.09 Average 0.26 158.88 67.45 109.45 0.27 159.66 68.23 109.45

K-means - - - - - - - - - Funny - 0.34 137.36 70.03 113.96 0.35 138.00 70.72 113.96 Pam - 0.34 130.31 69.90 114.36 0.34 130.75 70.22 114.36


The results show that the optimum clustering algorithm is the hierarchical agglomerative, always contains the best combination of all examined indices. In addition, the Euclidean distance indicates, in every case, better results than the manhattan one, contrary to the clustering method which differs according to the direction of the storm events.

More specifically, for the period of 1960-2000, the storm events are clustered by: § Average linkage for N direction § Single linkage for NE direction § Ward method for NW direction. And for the period of 2000-2100 the storm events are clustered by: § Single linkage for N direction § Average linkage for NE direction § Ward method for NW direction.

4. STORM CLASSIFICATION

The use of foregoing clustering methods contributes to a well-developed storm classification for past and future wave climate and for each direction N, NE, NW. The storm events are classified into five classes depending on their severity, (I - Weak, II - Moderate, III - Significant, IV - Severe, V - Extreme). In Figure 7, a dendrogram represents this procedure of hierarchical clustering based on an agglomerative algorithm.

Figure 7. Dendrogram of storm clustering using hierarchical clustering algorithm for the storm events of North direction and the period of 1960-2000.

All the results for the period 1960-2000 are described below in Tables 4-6 in association with the wave height, the wave peak period and the duration for each storm class, the storm events for the period 2000-2100 are presented in Tables 7-9.

Table 4. Storm events and their characteristics for the period 1960-2000 North wave direction

Storm Class

Significant Wave Height [m]

Peak Period [s]

Energy [m2h]

Duration [h]

Events

max mean min max mean min max mean mean I 3.65 2.33 4.99 9.08 7.42 40.91 170.12 94.35 16.52 191 II 4.18 2.58 5.7 9.17 7.69 174.62 401.04 263.37 37.56 133 III 5.11 2.86 4.84 9.75 7.95 409.93 661.31 533.9 60.72 50 IV 5.26 3.13 5.54 10.36 8.16 698.18 965.24 784.4 74.21 19 V 6.31 3.56 6.13 10.82 8.49 1131.4 1862.63 1462.18 105 14


Table 5. Storm events and their characteristics for the period 1960-2000 Northeast wave direction

Storm Class


Peak Period [s]

Energy [m2h]

Duration [h]

Events

max mean min max mean min max mean mean I 2.82 2.25 5.31 7.74 7.15 40.8 127.88 76.59 14.5 24 II 2.9 2.38 5.85 9.45 7.45 149.38 234.08 181.77 31.29 14 III 3.7 2.63 5.83 8.81 7.62 255.35 347.81 282.86 38.62 8 IV 4.15 2.95 5.85 9.02 7.89 408.12 467 434.13 46.5 4 V 4.29 2.95 6.27 9.09 7.92 555.86 726.1 621.79 67.2 5

Table 6. Storm events and their characteristics for the period 1960-2000 Northwest wave direction

Storm Class


Peak Period [s]

Energy [m2h]

Duration [h]

Events

max mean min max mean min max mean mean I 3.02 2.26 6.56 8.97 7.77 40.2 78.39 55.94 10.29 14 II 3.2 2.36 5.83 9.15 7.76 90.33 109.6 98.35 17 6 III 3.55 2.6 5.76 8.87 7.71 127.6 163.33 141.89 19.5 6 IV 4.24 3.29 6.19 9.64 6.19 387.99 387.99 387.99 33 12 V 4.38 2.88 4.74 10.34 8.5 197.5 279.09 236 25.5 4

Table 7. Storm events and their characteristics for the period 2000-2100 North wave direction

Storm Class


Peak Period [s]

Energy [m2h]

Duration [h]

Events

max mean min max mean min max mean mean I 4.28 2.41 3.97 9.66 7.53 40.87 280.57 135.03 22.09 695 II 5.46 2.72 3.99 10.32 7.84 281.98 568.63 394.11 50.1 240 III 5.92 2.97 4.91 11.12 8.02 574.35 1161.92 800.52 83.97 109 IV 6.76 3.41 5.6 11.07 8.39 1216.63 2172.63 1515.15 119.4 20 V 7.68 3.78 6.62 11.3 8.62 2868.74 3852 3360.37 210 2

Table 8. Storm events and their characteristics for the period 2000-2100 Northeast wave direction

Storm Class


Peak Period [s]

Energy [m2h]

Duration [h]

Events

max mean min max mean min max mean mean I 2.99 2.3 5.69 8.19 7.23 41.38 149.92 87.36 15.88 68 II 3.54 2.53 5.5 8.44 7.5 154.91 283.47 218.33 32.65 34 III 3.82 2.67 5.71 8.98 7.57 299.64 431.42 362.35 48 16 IV 3.98 2.82 5.21 8.95 7.75 457.35 720.22 549.09 65.1 10 V 4.62 3.5 7.01 9.26 8.42 883.05 1045.14 964.09 73.5 2

Table 9. Storm events and their characteristics for the period 2000-2100 Northwest wave direction

Storm Class


Peak Period [s]

Energy [m2h]

Duration [h]

Events

max mean min max mean min max mean mean I 3.86 2.34 4.32 10.67 7.67 42.15 137.92 78.48 13.45 60 II 4.34 2.62 4.95 10.87 8.13 145.05 218.72 181.07 24.21 14 III 3.39 2.7 6.6 9.34 8.26 243.95 311.02 279.91 36.6 5 IV 3.66 2.90 7.96 9.12 8.75 - 530.32 - 60 1 V 4.14 3.24 7.10 9.67 8.63 - 1150 - 105 1

5. STORM IMPACTS

The connection of storm events according to their severity with their impacts is of interest to many researchers (Coco et al., 2014; Del Rio et al., 2012; Almeida et al., 2012). With the purpose of studying unfavourable sea states, three storm events of class V and one from class IV for North direction are selected and, for instance, only one of them is presented below in Figure 8 and Figure


9. The results extracted from the above-mentioned storm analysis have been utilized as initial conditions for the evaluation of wave propagation towards the shore. The methodology followed here is similar to that of Afentoulis et al. (2017a, b). Within this context, the precise simulation of the wave deformation in the coastal area of Rethymno, has been carried out using MIKE 21 model (DHI 2015a, b). The fully spectral formulation used in the model run is based on the wave action conservation equation, where the directional-frequency wave action spectrum is the dependent variable. The bathymetry data have been provided by Hellenic Navy Hydrographic Service, with a 15 m spatial resolution and they have been interpolated over an unstructured triangular mesh with a different level of resolution. The time series of wave climate has been applied in the northern boundary of the domain. The evaluation of waves approaching the wind-ward breakwater leads to the conclusion that the structures are exposed to high wave loads, the significant wave height of which reaches the value of 3.5 meters. Assessing the characteristics of the 84-hour storm event, in Figure 8 and Figure 9 it is depicted that the highest waves are mainly observed in the intermediate intervals of the simulation.

Table 10. Selected storm events and their characteristics which are examined for their impacts

Date Storm Class

Direction Energy [m2hr]

TP range [s]

HS range [m]

Duration [hr]

26-Feb-1965 V N 916.33 5.87-9.74 2.13-5.11 39 13-Feb-1995 V N 822.78 7.81-10.35 2.13-5.26 48 09-Feb-2047 IV N 1882 6.15-11.06 2.13-6.76 84 25-Jan-2091 V N 2868 7.37-11.30 2.13-7.67 99

Figure 8. Spatial distribution of significant wave height (Hs) by MIKE 21 SW for IV storm scenario in the middle of the run.

Figure 9. Spatial distribution of significant wave height (Hs) by MIKE 21 SW for IV storm scenario in the middle of the run for the near port area.


6. CONCLUSION

An integrated analysis of storm events at Rethymno was investigated in this work and a storm classification, by means of clustering methods, was also presented. The significant wave height and the inter-arrival period thresholds were estimated according to the extreme value theory with the stability method and the 95th percentile. It is worth mentioning that 95th percentile is not another rule of thumb for the storm definition, which could also be confirmed by other methods such as the stability check but, as it turns out, it is a reliable way to set thresholds and to identify rare and extreme events. Consequently, a new definition of storm event for the coastal zone of Rethymno was derived. Well-known clustering algorithms and their methods were validated using many indices which are generally utilized in Data Science community for the cluster results evaluation. All the above analysis and procedure should be taken into account by anyone who wants to cluster wave data in any specific area. Finally, numerical simulations could be used as a tool in order to associate the storm impacts with each class. Four extreme events of class IV and V were analysed and it was found that during a storm event the port structures are exposed to high wave loads, the significant wave height of which reaches the value of 3.5 meters.

ACKNOWLEDGMENTS

This work is part of Mr. Martzikos’ PhD thesis which is co-funded through the Action: «Enhancing the Human Research Potential through the Implementation of Doctoral Research» of the Operational Program «Human Resources Development, Education and Lifelong Learning» of NSRF 2014-2020, co-financed by the European Social Fund and the Greek Government. The research leading to these results has also received partial funding from the European Union Seventh Framework Programme (FP7/2007-2013) under Grant agreement n° 603663 for the research project PEARL (Preparing for Extreme And Rare events in coastaL regions). The research and its conclusions reflect only the views of the authors and the European Union is not liable for any use that may be made of the information contained herein.

An initial shorter version of the paper has been presented at the 10th World Congress of the European Water Resources Association (EWRA2017) “Panta Rhei”, Athens, Greece, 5-9 July, 2017.

REFERENCES

Afentoulis, V., Kragiopoulou, E., Skarlatou, E., Moschos, E., Lykou A., Makropoulos, C., Tsoukala, V., 2017. Coastal Processes Assessment Under Extreme Storm Events Using Numerical Modelling Approaches. Environ. Process.; 4(3): 731-747. https://doi.org/10.1007/s40710-017-0253-8

Afentoulis, V., Mohammadi, B., Tsoukala, V., 2017. The investigation of form and processes in the coastal zone under extreme storm events-the case study of Rethymno, Greece. 19th EGU General Assembly, EGU2017, proceedings from the conference held 23-28 April, 2017 in Vienna, Austria: pp.7077.

Allsop, W., T. Bruce, T.A. Pullen, J.W. Van der Meer., 2008. Direct hazards from wave overtopping-the forgotten aspect of coastal flood risk assessment? in: 43rd Defra Flood and Coastal Management Conference, 1-3 July 2008, Manchester University. ID code: 215.

Almeida, L.P., Vousdoukas, M.V., Ferreira, O., Rodrigues, B.A., Matias, A., 2012. Thresholds for storm impacts on an exposed sandy coastal area in southern Portugal, Geomorphology; 143–144: 3–12. https://doi.org/10.1016/j.geomorph.2011.04.047

Arbelaitz, O., Gurrutxaga, I., Muguerza, J., Pérez, J., Perona, I., 2013. An extensive comparative study of cluster validity indices. Pattern Recognition; 46(1): 243-256, http://dx.doi.org/10.1016/j.patcog.2012.07.021

Armaroli, C., Ciavola, P., Perini, L., Calabrese, L., Lorito, S., Valentini A., Masina, M., 2012. Critical storm thresholds for significant morphological changes and damage along the Emilia-Romagna coastline, Italy. Geomorphology 143-144 :34-51. https://doi.org/10.1016/j.geomorph.2011.09.006

Athanassoulis, G.A., Belibassakis, K.A., Gerostathis, Th.P., Kapelonis, Z.G., 2014. Application of SWAN wave model for climatic simulation of sea condition at coastal areas of the Mediterranean. In: 6th Panhellenic Conf. Coastal Zones Manage. Improvement, 24-27 November 2014, Athens, Greece: 345-364.

Banerjee, A., Dave, R., 2004. Validating clusters using the Hopkins statistic. IEEE International Conference On Fuzzy Systems (IEEE Cat. No.04CH37542), http://dx.doi.org/10.1109/fuzzy.2004.1375706

Bernardara, P., Mazas, F., Kergadallan, X., L., Hamm, 2014. A two-step framework for over-threshold modelling of environmental extremes. Nat. Hazards Earth Syst. Sci.; 14: 635-647. https://doi.org/10.5194/nhess-14-635-2014, 2014


Boccotti, P., 2014. Wave mechanics and wave loads on marine structures. Elsevier Inc. Booij, N., Ris, R.C., Holthuijsen, L.H., 1999. A third-generation wave model for coastal regions: 1. Model description and

validation. Journal of Geophysical Research; 104(C4): 7649. https://doi.org/10.1029/98JC02622 Brock, G., Pihur, V., Datta, S., Datta, S., 2008. clValid: An R Package for Cluster Validation. Journal of Statistical Software; 25(4):

1-22, http://www.jstatsoft.org/v25/i04/. Ciavola, P., Coco, G., 2017. Coastal Storms: Processes and Impacts. Wiley-Blackwell Ciavola, P., Ferreira, O., van Dongeren, A., de Vries, J., Armaroli, C., Harley, M., 2015. Prediction of Storm Impacts on Beach and

Dune Systems: 227-252. In: Quevauviller P., (Ed.), Hydrometeorological Hazards: Interfacing Science and Policy, John Wiley & Sons.

Coco, G., Senechal, N., Rejas, A., Bryan, K., Capo, S., Parisot, J. et al., 2014. Beach response to a sequence of extreme storms. Geomorphology; 204: 493-501, http://dx.doi.org/10.1016/j.geomorph.2013.08.028

Corbella, S., Stretch, D.D., 2012. Predicting coastal erosion trends using non-stationary statistics and process-based models. Coastal Engineering; 70: 40-49. https://doi.org/10.1016/j.coastaleng.2012.06.004

Costas, S., Ferreira, O., Martinez, G., 2015. Why do we decide to live with risk at the coast?. Ocean & Coastal Management; 118: 1-11, http://dx.doi.org/10.1016/j.ocecoaman.2015.05.015

Datta, S., Datta, S., 2003. Comparisons and validation of statistical clustering techniques for microarray gene expression data. Bioinformatics; 19(4): 459-466. https://doi.org/10.1093/bioinformatics/btg025

Davidson, Μ., Turner I., Splinter, K., Harley M., 2017. Annual prediction of shoreline erosion and subsequent recovery. Costal Engineering; 130: 14-25. https://doi.org/10.1016/j.coastaleng.2017.09.008

De Michele, C., Salvadori, G., Passoni, G., Vezzoli, R., 2007. A multivariate model of sea storms using copulas. Coastal Engineering; 54: 734-751. https://doi.org/10.1016/j.coastaleng.2007.05.007

De Morsier, F., Tuia, D., Borgeaud, M., Gass, V., Thiran, J., 2015. Cluster validity measure and merging system for hierarchical clustering considering outliers. Pattern Recognition; 48(4): 1478-1489, http://dx.doi.org/10.1016/j.patcog.2014.10.003

Del Rio, L., Plomaritis, T.A., Benavente, J., Valladares, M., Ribera, P., 2012. Establishing storm thresholds for the Spanish Gulf of Cádiz coast. Geomorphology; 143–144: 13–23. https://doi.org/10.1016/j.geomorph.2011.04.048

DHI, 2015a. MIKE 21 and MIKE 3 Flow Model FM Hydrodynamic Module DHI, 2015b. MIKE 21 Wave Modelling- MIKE 21 SW – Spectral Waves FM Dolan, R., Davis, R.E., 1992. An intensity scale for Atlantic coast northeast storms. J. Coast. Res.; 8(4): 40-853,

http://www.jstor.org/stable/4298040 Dolan, R., Davis, R.E., 1994. Coastal storm hazards. J. Coast. Res.; (SI 12): 103-114. http://www.jstor.org/stable/25735593. Dunn, J.C., 1974. Well separated clusters and fuzzy partitions. Journal on Cybernetics; 4: 95-104.

https://doi.org/10.1080/01969727408546059 Eastoe, E., Koukoulas, S., Jonathan, P., 2013. Statistical measures of extremal dependence illustrated using measured sea surface

elevations from a neighbourhood of coastal locations. Ocean Eng.; 62: 68-77. https://doi.org/10.1016/j.oceaneng.2013.01.002 Ferreira, O., Viavattene, C., Jiménez, J.Α., Bolle, A., das Neves, L., Plomaritis, T., McCall, R., van Dongeren, A.R., 2017. Storm-

induced risk assessment: evaluation of two tools at the regional and hotspot scale. Coastal Engineering; (In Press). https://doi.org/10.1016/j.coastaleng.2017.10.005

Gilleland, E., Katz, R., 2016. extRemes 2.0: An Extreme Value Analysis Package in R. Journal of Statistical Software; 72(8): 1-39. doi:10.18637/jss.v072.i08

Godschalk, D.R., Brower, D.J., Beatley T., 1989. Catastrophic Coastal Storms: Hazard Mitigation and Development Management. Duke Press Policyc Studies.

Grottoli, E., Bertoni, D., Ciavola, P., 2017. Short-and medium-term response to storms on three Mediterranean coarse-grained beaches. Geomorphology; 295: 738-748. https://doi.org/10.1016/j.geomorph.2017.08.007

Halkidi, M., Batistakis, Y., Vazirgiannis, M., 2001. On Clustering Validation Techniques. Journal of Intelligent Information Systems; 17: 107, doi:10.1023/A:1012801612483

Handl, J., Knowles, K., Kell, D., 2005. Computational cluster validation in post-genomic data analysis. Bioinformatics; 21(15): 3201-3212. https://doi.org/10.1093/bioinformatics/bti517

Harley MD, 2017a. Coastal Storm Definition. In: Ciavola, P., Coco, G., (Ed.), Coastal Storms: Processes and Impacts, First Edition, Wiley Blackwell: 1 – 22.

Harley, M.D., Turner, I.L., Kinsela, M.A., Middleton, J.H., Mumford, P.J., Splinter, K.D., Phillips, M.S., Simmons, J.A., Hanslow, D.J., Short, A.D., 2017b. Extreme coastal erosion enhanced by anomalous extratropical storm wave direction. Scientific Reports; 7(1); Article number 6033. https://doi.org/10.1038/s41598-017-05792-1

Hissel, F., Morel, G., Pescaroli, G., Graaff, H., Felts, D., Pietrantoni, L., 2014. Early warning and mass evacuation in coastal cities. Coastal Engineering; 87: 193-204. https://doi.org/10.1016/j.coastaleng.2013.11.015

Jain, A. K., 2010. Data clustering: 50 years beyond K-means. Pattern Recognition Letters; 8: 651-666, http://dx.doi.org/10.1016/j.patrec.2009.09.011

Kassambara, A., Mundt, F., 2016. factoextra: Extract and Visualize the Results of Multivariate Data Analyses. R package version 1.0.3., https://CRAN.R-project.org/package=factoextra

Kaufman, L., Rousseeuw, P., 2005. Finding Groups in Data: An Introduction to Cluster Analysis (1st ed.). Hoboken: John Wiley & Sons, Inc., Chapter 2: 68.

Kelman, I., 2013. Saffir-Simpson Hurricane Intensity Scale. In: Bobrowsky, P. T., 2013. Encyclopedia of Natural Hazards. Springer Netherlands: 882-883, http://dx.doi.org/10.1007/978-1-4020-4399-4_306

Kokkinos, D., Prinos, P., Galiatsatou, P., 2014. Assessment of coastal vulnerability for present and future climate conditions in coastal areas of the Aegean Sea. In: Paper Presented at the 11th Inter-national Conference on Hydroscience & Engineering: Hydro-Engineering for Environmental Challenges. http://vzb.baw.de/e-medien/iche-2014/PDF/15%20Mini-Symposium%20Impacts%20of%20Climate%20Change/15_04.pdf.

Lin-Ye, J., Garcia-Leon, M., Gracia, V., Sanchez-Arcilla, A., 2016. A multivariate statistical model of extreme events: An application to the Catalan coast. Coastal Engineering; 117: 138-156. http://dx.doi.org/10.1016/j.coastaleng.2016.08.002

Maechler, M., Rousseeuw, P., Struyf, A., Hubert, M., Hornik, K., 2016. cluster: Cluster Analysis Basics and Extensions. R package version 2.0.4.


Makropoulos, Ch., Tsoukala, V.K., Belibassakis, K., Lykoy, A., Chondros, M., Gougoura, P., 2015. Managing flood risk in coast cities through an integrated modelling framework supporting stake-holders' involvement: the case study of Rethymno. In: E-proceedings of the 36th IAHR World Conference, 28 June-3 July, The Hague, the Netherlands.

Martinez G., Armaroli, C., Costas, S., Mitchell D.Harley, M.D., Paolisso, M., 2017. Experiences and results from interdisciplinary collaboration: Utilizing qualitative information to formulate disaster risk reduction measures for coastal regions. Coastal Engineering; (In Press). https://doi.org/10.1016/j.coastaleng.2017.09.010

Martzikos, N., Lykou, A., Makropoulos, C., Tsoukala, V., 2017. Cluster analysis and classification of storm events at Rethymno. European Water; 57: 57-62.

Mendoza, E.T., Jiménez, J.A., Mateo, J., 2011. A coastal storms intensity scale for the Catalan sea (NW Mediterranean). Nat. Hazards Earth Syst. Sci.; 11(9): 2453-2462, http://dx.doi.org/ 10.5194/nhess-11-2453-2011.

Pal, J., Giorgi, F., Bi, X., Elguindi, N., Solmon, F., Rauscher, S. et al., 2007. Regional Climate Modeling for the Developing World: The ICTP RegCM3 and RegCNET. Bulletin Of The American Meteorological Society; 88(9): 1395-1409, http://dx.doi.org/10.1175/bams-88-9-1395

Plomaritis, T.A., Ferreira, Ó., Costas, S., 2017. Regional assessment of storm related overwash and breaching hazards on coastal barriers. Coastal Engineering (In Press). https://doi.org/10.1016/j.coastaleng.2017.09.003

Puig, M., del Río, L., Plomaritis, T., Benavente, J., 2014. Influence of storms on coastal retreat in SW Spain. Journal of Coastal Research; 70 (sp1): 193-198. https://doi.org/10.2112/SI70-033.1

Puig, M., del Río, L., Plomaritis, T., Benavente, 2016. Contribution of storms to shoreline changes in mesotidal dissipative beaches: case study in the Gulf of Cádiz (SW Spain). Natural Hazards and Earth System Sciences; 16 (12); 2543-2557. 10.5194/nhess-16-2543-2016

Prinos, P., 2014. Climate change effects on the Greek seas and coastal areas - the research project THALIS-CCSEAWAVS. In: 6th Panhellenic Conference on Coastal Zones Management and Improvement, 24-27 November 2014, Athens: 315-324.

R Core Team (2017). R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. https://www.R-project.org/.

Rangel-Buitrago, N., Anfuso, G., 2011. An application of Dolan and Davis (1992) classification to coastal storms in SW Spanish littoral. J. Coast. Res.; SI 64: 1891-1895.

Ris, R.C., Holthuijsen, L.H., Booij, N., 1999. A third-generation wave model for coastal regions: 2. Verification. J. Geophys. Res.; 104 (C4): 7667-7681, http://dx.doi.org/10.1029/1998JC900123.

The SWAN team, 2017. User Manual. SWAN Cycle III version 41.20; http://swanmodel.sourceforge.net/ Theodoridis, S., Koutroumbas, K., 2009. Pattern recognition (4th ed.). London: Academic Press. Trifonova, E.V., Valchev, N.N., Andreeva, N.K., Eftimova, P.T., 2012. Critical storm thresholds for morphological changes in the

western Black Sea coastal zone. Geomorphology 143-144: 81-94. https://doi.org/10.1016/j.geomorph.2011.07.036 Tsoukala, V., Chondros, M., Kapelonis, Z.G., Martzikos, N., Lykou, A., Belibassakis, K., Makropoulos, Ch., 2016. An integrated

wave modelling framework for extreme and rare events for climate change in coastal areas - the case of Rethymno-Crete, Oceanologia; 58 (2): 71-89, doi:10.1016/j.oceano.2016.01.002

Vagenas, C., Anagnostopoulou, C., Tolika, K., 2014. Climatic study of the surface wind field and extreme winds over the Greek seas. In COMECAP 2014. Abstracts and e-contributions. 12th International Conference on Meteorology, Climatology and Physics of the Atmosphere, Heraklion 28–31 May 2014. Hereklion, Crete: Crete University Press: 283–288.

Velikou, K., Tolika, K., Anagnostopoulou, C., Tegoulias, I., Vagenas, C., 2014. High resolution climate over Greece: assessment and future projections. In: 12th International Conference on Meteorology, Climatology and Atmospheric Physics (COMECAP 2014), Heraklion.

Weisse, R., von Storch, H., 2010. Marine Climate and Climate Change Storms, Wind Waves and Storm Surges. Springer, Berlin.

storm clustering and classification for the port of ... · rethymno, as mentioned above, is located...

Documents