Clim Dyn

DOI 10.1007/s00382-015-2659-1

Long‑term changes in the frequency, intensity and duration

of extreme storm surge events in southern Europe
Alba Cid1 · Melisa Menéndez1 · Sonia Castanedo1 · Ana J. Abascal1 ·
Fernando J. Méndez1 · Raúl Medina1 

Received: 22 January 2015 / Accepted: 16 May 2015

© Springer-Verlag Berlin Heidelberg 2015

Abstract  Storm surges are one of the major hazards in for the 50-year return level are also estimated, showing a
coastal regions; positive surge events are added to tidal lev- large spatial variability with relatively higher values along
els, increasing the risk of coastal flooding by extreme water the coast. A clear sensitivity of extreme storm surges to
levels. In this study, changes in the frequency (occurrence negative NAO index is detected, specifically in the west-
rate per year), intensity (magnitude of the extremes) and ern Mediterranean basin. Results show that negative NAO
duration of extreme storm surge events from 1948 to 2013 phases lead to an increase in the number of extreme events
are investigated using a non-stationary statistical model. To and also in their intensity.
fully model extremes, the time-dependent statistical model
combines the Generalized Pareto Distribution (GPD) for Keywords  Extreme events · Extreme value model ·
studying exceedances over the threshold, and the non- NAO · Non-stationary · Pareto–Poisson · Storm surge
homogeneous Poisson (P) process for studying the occur-
rence rate of these exceedances. Long-term trends and the
association between storm surges and the North Atlantic 1 Introduction
Oscillation (NAO) are represented in the model by allow-
ing the parameters in the GPD–P model to be time-depend- Coastal areas are specially vulnerable to climate vari-
ent. Different spatial patterns in the three analysed proper- ability and changes in sea level. Particularly, extreme sea
ties of storm surges are found in the Atlantic region and level events have immediate impacts on the coast (Storch
the Mediterranean Sea. The up to now uncharted regional and Woth 2008). During the past century, storm surge
patterns of storm surge duration show completely different events have caused considerable damage to properties
values between the Atlantic and the Mediterranean regions, and even loss of life. One of the worst natural disasters
being the duration of storms surges in the Atlantic two in Europe was the 1953 North Sea flood (Baxter 2005),
times longer than the duration in the Mediterranean. For which led to promote major studies on the strengthening
the last half century, we detect positive and negative spa- of coastal defences. Over the Mediterranean Sea, there are
tial trends in terms of intensity of storm surge but only sig- also numerous low pressure systems, a few of which may
nificant decreasing rates, of around 2 %, in the number of develop a dynamical evolution similar to the one of tropi-
extreme events per year. Regarding duration, we find posi- cal cyclones (medicanes) (see Cavicchia et al. 2014). Such
tive trends in certain Mediterranean areas, with durations of storms produce severe damage in the highly populated
extreme events increasing at a rate of 0.5–1.5 h/year. Values coastal areas surrounding the Mediterranean Sea. There-
fore, it is important in the general coastal planning and in
flood risk assessment, to try to foresee future storm surge
* Alba Cid impacts on the coast. These extreme events are generally
[email protected] driven by the combination of high tides and storm surges.
1 While astronomical tide is a deterministic component of
Environmental Hydraulics Institute “IH Cantabria”,
Universidad de Cantabria, C/ Isabel Torres n 15, PCTCAN, the sea level, storm surge depends on the atmospheric pres-
39011 Santander, Spain sure and wind; hence, in order to study long-term changes

13
 A. Cid et al.

in coastal risks, it is necessary to analyse this non-deter- the Aegean and Ionian seas, Butler et al. 2007 in the North
ministic variable from a statistical point of view. Sea, and Marcos et al. 2009 in the Mediterranean Sea).
Various statistical techniques have been used to study Both methods, the AMM and the r-largest were used by
changes in extreme sea level events. A basic approach is Haigh et al. (2014) to estimate extreme water levels around
the percentile time series analysis. This technique has pro- the coastline of Australia. Finally, the peak-over-threshold
vided interesting information about the different behav- (POT) approach has been recently applied to determine
iour of high water levels with respect to the mean sea level return water levels, as for instance, in northern Germany
(e.g. Woodworth and Blackman 2004), however, it does (see Arns et al. 2015); and specifically for storm surges, it
not allow the estimation of return water levels. Focusing was used by Tebaldi et al. (2012) along the US coasts, Hal-
on extreme value models, two familiar approximations in legatte et al. (2011) in Copenhagen-Denmark and Bernard-
extreme value theory are the Generalized Extreme Value ara et al. (2011) along the French Atlantic coasts.
(GEV) distribution from block maxima, and the Gener- The IPCC (Intergovernmental Panel on Climate Change)
alized Pareto (GP) distributions from the excesses over a concludes that climate change might affect the frequency,
threshold. A set of methods can be used for the selection intensity, and length of many extreme events, such as
of extreme sea levels. The classical annual maxima method floods and storms (Church et al. 2013), which would also
(AMM) has the disadvantage of gathering a small extreme lead to a non-stationary behaviour of storm surges. There-
sample that could disregard extremes in the remaining data. fore, we propose a time-dependent extreme model that
The limitation of AMM can be overcome by methods that combines the Generalized Pareto Distribution (GPD) for
select maxima from shorter time blocks than a year such studying exceedances over a threshold (in terms of inten-
as monthly maxima, the r-largest values within a year, or sity and duration), and the Poisson (P) process for study-
the independent peaks over a threshold. The use of monthly ing the occurrence rate of these exceedances. Note that a
maxima requires modelling the seasonal dependence for point-process approach in terms of extreme value models
the good estimation of return periods (see Méndez et al. is not commonly used to analyse variations of sea surface
2007) and the r-larger values requires defining a specific variables, and moreover, non-stationarity in extremes tends
number of extremes per period (Coles et al. 2001). These to be overlooked in the extreme value theory. Some excep-
techniques are, however, constrained to choosing the same tions can be found in the literature. The effect of temporal
number of maxima within time blocks, which can lead to dependence on the frequency of storm surge in the western
a misinterpretation of climatic changes in an environment Mediterranean basin is explored from a non-homogeneous
with a mean sea level rise and large inter-annual varia- Poisson model in Luceño et al. (2006). The methodology
tions. Taking this into consideration, in this work we have to analyse climate variations at different time scales from a
decided to define extreme events as the exceedances over time-dependent monthly GEV model in extreme sea levels
a high threshold. Two important issues that must be over- is described in Méndez et al. (2007). The seasonal, mean
come when using the peak over threshold approach are the sea level rise and astronomical modulations in high water
selection of the threshold and the minimum time span that levels are explored in Menendez et al. (2009) from a time-
will be required to assume the independence of consecu- dependent GEV as a potential extreme forecast tool. Differ-
tive storm events, since both affect the results in terms of ent time-dependent extreme models are applied in Menén-
the frequency and the exceedance estimates (see, e.g, Arns dez and Woodworth (2010) to a global tide-gauge dataset to
et al. 2013). Storm surge extremes have been selected from conclude that the highest water levels have been increasing
an already validated surge database (GOS database, see since the 1950s in most regions of the world mainly due to
Cid et al. 2014) that provides, for southern Europe, a long mean sea level rise, although higher regional extremes are
(66 year length) and spatially uniform set of data, which is also caused by interannual-decadal variations associated
adequate to analyse long-term changes in the storm surge with climate fluctuations. Non-stationary extreme models
extremes. have also been applied to storm surge water levels; Mar-
The above-mentioned statistical techniques have been cos et al. (2011) analysed the extension of GEV model to
applied lately to analyse extreme sea level events. For exam- r-largest values with several time-dependent terms in order
ple, Gräwe et al. (2012) tested different statistical methods to study changes in storm surges under projections of cli-
to estimate return periods on simulated storm surges. Mer- mate scenarios. Mudersbach and Jensen (2010) used an
rifield et al. (2013) use the AMM to investigate spatial pat- approach based on the GEV distribution and compared the
terns of the relative contributions from tides and non-tidal results with common stationary methods. Serafin and Rug-
residual components. The AMM is also the approach chosen giero (2014) modelled extreme events of total water levels
by Bernier and Thompson (2006) to calculate return levels (waves, tides, and non-tidal residuals) using non-stationary
in the north–west Atlantic. The r-largest method has been extreme value distribution that includes seasonality and cli-
widely used in European regions (e.g. Tsimplis 1997 in mate variability.

13
Long-term changes in the frequency, intensity and duration of extreme storm surge events...

Fig. 1  Panel a shows the model domain (green box) and the location 2007, respectively. Measured data (blue line) versus modelled data
(black dot) where data are compared. b Storm surge comparison (m) (red line). Green circles encompass instants of high surge levels
at Vigo (top) and Barcelona (bottom) tide gauges for years 1998 and

The main goal of this work is the analysis of changes of trends, sensitivity to NAO and the estimation of 50-year
in extreme storm surges (i.e. changes in extreme sea levels return levels. Finally, a summary and a discussion are pro-
produced by the effect of wind stress and pressure gradients vided in Sect. 5.
over the sea surface). Following this purpose, a statistical
framework to study extreme surges is established. Extreme
events have been chosen using a POT technique, which 2 Database description
allows us not to limit the number of extremes to a certain
amount, but to consider all the events above a threshold. A Extreme events have been selected from a 66-year (1948–
spatial description of these storm surge extremes in terms 2013) high-resolution surge hindcast named GOS 1.1,
of intensity, duration and frequency is given. In a second which describes the sea level variation due to the effect
step, trends were studied by means of a time-dependent of wind stress and pressure gradients (Cid et al. 2014).
GPD–Poisson model, where time variability is included Although the database in Cid et al. (2014) ends in 2009,
by adding linear trends on the scale parameter of Pareto for this work GOS 1.1 has been updated until December
and on the Poisson parameter. The use of a time-depend- 2013. It has been performed for southern Europe using the
ent GPD–P model has enabled us to estimate long-term Regional Ocean Modelling System (ROMS) of Rutgers
trends, not only in terms of the magnitude of the extremes University. ROMS is a three-dimensional, free-surface, ter-
but also in the frequency. Specifically, significant trends in rain-following ocean model that solves the Reynolds-aver-
the intensity, frequency and duration of the extreme storm aged NavierStokes equations using the hydrostatic vertical
surge events, as well as the value of the 50-year return level momentum balance and Boussinesq approximation with
were estimated at every grid-point of the model domain. a split-explicit time stepping algorithm (Haidvogel et al.
Sensitivity of frequency and magnitude of extreme surges 2000; Shchepetkin and McWilliams 2005; Haidvogel et al.
to monthly NAO index was also studied. 2008). It uses a horizontal curvilinear Arakawa C grid and
Following this purpose, the next section describes briefly vertical stretched terrain-following coordinates. The model
the database used for selecting the extremes (Sect. 2). The domain (see green box on subplot at Fig. 1) encloses south-
methodology applied for defining and fitting the extremes ern Europe, including the Mediterranean Sea and the Atlan-
is explained in Sect. 3. Section 4 describes the extremes in tic area in southern Europe, with a horizontal resolution
terms of intensity, frequency and duration and also gives of 1/8◦ (∼14 km). GOS 1.1 was driven with hourly mete-
the results of implementing the statistical model, in terms orological data of wind and atmospheric pressure from the

13
 A. Cid et al.

SeaWind I hindcast (Menendez et al. 2014). The SeaWind I

dataset is a high-resolution (30 km) atmospheric dynami-
cal downscaling from NCEP global atmospheric reanalysis
focused on offshore surface wind fields. The ROMS model
was run in barotropic mode and the inverted barometer
effect was imposed at the open boundaries of the domain
(north and west edges), using a free surface Chapman con-
dition and 2D Flather condition for momentum. The GOS Fig. 2  Peak over threshold technique. Red dots are extreme events,
values over the 99.5 %. Each of the selected extremes occurs at a spe-
1.1 database was exhaustively validated with tide gauges cific time ti, has a specific duration of di and exceeds the threshold by
and satellite data, reaching high correlation coefficients and an amount of yi
small root mean square errors all over the domain, except
for three semi-enclosed areas (northern Adriatic, Gulf of
Gabes and Aegean Sea) where the hindcast overestimates in the region. The maxima sample data is then introduced
satellite measures. An example of the general agreement in the extreme model. Results of the extreme statistical
between measured and modelled data can be seen in Fig. 1 analysis in terms of number of events per year and duration
for Vigo and Barcelona tide gauges. The modelled surge will be shown in Sect. 4.
data (red line) are well correlated with the tide gauge meas-
ures (blue line), where statistics show a root mean square 3.2 The time‑dependent GPD–Poisson model
error of 6.5 cm and a correlation factor of 0.88 (see Cid
et al. 2014). A good agreement is also observed when refer- As already mentioned, the statistical model chosen to esti-
ring to extreme values, as can be seen from the data within mate long-term trends in the frequency (extremes occur-
green circles. Note that GOS 1.1 grid-points have a spatial rence rate), intensity (sea level reached above the thresh-
resolution of about 14 km and therefore they do not exactly old) and duration (number of hours where the level is above
represent in-situ values in shallow water depths at the coast the threshold) of storm surge events is a time-dependent
(i.e. tide gauge records). version of the Peak Over Threshold model (Méndez et al.
2006). This model combines the Generalized Pareto Distri-
bution (GPD) (see Eq. 1) for studying exceedances over a
3 Extreme value model threshold, and the Poisson Distribution (P) (see Eq. 2) for
studying the occurrence rate of exceedances.
3.1 Selection of the extreme events  x −1/ξ
G(x; σ , ξ ) = 1 − 1 + ξ (1)
Surge time series are extracted from the hourly database σ
at every grid cell of the domain. Following the Peak Over
n
Threshold (POT) technique, all data over the threshold P(n; ) = e− (2)
of 99.5 %, derived from the entire period, are considered n!
exceedances. The threshold selection is made taking into The basic idea is to choose a high threshold (u) and to study
account that the number of events above the threshold is the statistical properties of all the exceedances over u; spe-
sufficient not to violate the basis of the model (i.e. an aver- cifically, the number of events (n) every year, the amounts
age of around two events per year, following the experience by which the threshold is exceeded (threshold excesses, i.e.
of previous works, e.g. Menéndez et al. 2008) yi) and the duration over that threshold (di).
Figure  2 represents the applied method to characterise The number of events (n) over the level u in any given
storm surge extremes. Red dots indicate the peaks of two year follows a Poisson distribution with annual mean 
events (excesses yi and yi+1) that happen at time ti and ti+1 . The intensity and duration of the extremes will be the
; the duration of each extreme event is given by di and di+1 variables analysed by the GPD distribution. The com-
, respectively. The statistical model used to estimate trends bination of Poisson model for frequency and the GPD
requires the extremes to be independent. Thereby, the model for intensity (or duration) can be expressed as
independence between consecutive events is achieved by shown in Eq. 3.
imposing that extremes must be separated by at least 3 days   
x −1/ξ
in between. This value has been chosen taking into account H(x, n; σ , ξ , ) = exp − 1 + ξ (3)
σ
that the average lifetime of the cyclones in the Mediter-
ranean is 28 h and the typical values for the North Atlan- where  > 0 is the Poisson parameter (mean num-
tic are 3 days (Trigo et al. 1999). Therefore, the value of ber of events over a period of time), and σ > 0 and ξ
3 days is selected to ensure independence between storms (−∞ < ξ < ∞) are the scale and shape Pareto parameters,

13
Long-term changes in the frequency, intensity and duration of extreme storm surge events...

Fig. 3  Effect of the time-dependent parameters. Blue dots extreme ent scale Pareto parameter. c Time-dependent Poisson parameter. d
events; blue line scale Pareto parameter; green line Poisson parame- Both Pareto and Poisson parameters are time-dependent
ter; black line 50-year return level. a Stationary case. b Time-depend-

respectively. x corresponds to the threshold excess (or the where ν(t) represents the rate of occurrence of independent
duration) of the selected extreme events. events over the threshold every year. σ (t) accounts for the
For this purpose, the GPD–P model is run twice; in one mean value and the variance of the exceedances.
case the variable of the GPD distribution (x) is the sample The added value of this extreme model in the estimation
of the independent peaks (i.e. yi) and in a second run the of trends on return values is shown in Fig. 3. In this figure,
GPD variable is the sample of durations of the independ- blue dots represent the excesses of the selected extreme
ent events (i.e. di). In both cases the Poisson variable is the events for a grid cell located at the south-west of Cyprus
occurrence of the extreme events (i.e. ti). (31.5° in longitude and 34.23° in latitude), blue line corre-
The main idea of the approach in this work is to con- sponds to scale Pareto parameter, green line is the Poisson
sider that the exceedance probability of a certain extreme parameter and black line is 50-year return level estimated
event varies through time. Consequently, variability is as shown in Eq. 6.
represented in the model by allowing the parameters in
σ (t) 
the GPD–P to be time-dependent, using linear parametric 1 − (ν(t) T ) ξ

Y50 (t) = u − (6)
ξ
expressions for the scale Pareto parameter (Eq. 4) and the
Poisson parameter (Eq. 5). where σ (t), ξ and ν(t) are the GPD–P parameters; T is, in
this case, 50 years and Y50 (t) is the time-dependent 50-year
σ (t) = σ0 + σ1 (t − t0 ) (4) return level.
 tf
(t) = ν(t) dt ; ν(t) = ν0 + ν1 (t − t0 ) A visual inspection of the maxima sample (blue dots)
t0 (5) on Fig. 3 shows a slight increase in magnitude and a clear

13
 A. Cid et al.

decrease in the frequency of the events throughout the ana- where Cov(t) is the covariate defined by the monthly NAO
lysed period (1948–2013). Figure 3a shows the station- index.
ary approach, where the GPD–P parameters are constant
(u = 23 cm; σ̂t = σˆ0 = 5.4 cm; νˆ0 = 3.3 events/year). The 3.3 Statistical inference
estimated flat 50-year return level is 44.15 cm. The increase
in the magnitude of the peaks is depicted in Fig. 3b, where The maximum likelihood estimation (MLE) method is
a trend of 0.86 mm/year in σ̂t indicates that 42.5 cm is a chosen to estimate the GPD–P model parameters. If the
storm event of the 50-year period during the sixties whilst N exceedances x1, x2, ..., xN are observed at days t1, t2, ...
46 cm is the 50-year return level from the 2000s. Panel c tN over a T-year period, the log-likelihood function for the
shows the results if we only take into account time depend- time-dependent GPD–Poisson model is expressed as shown
ence in the occurrence rate of extreme events. A decrease in Eq. 9.
in the occurrence rate of events per year is detected, with
about 4 events/year until the seventies to within three N 
 
1
 
xi

lx,t (θ) = log ν(ti ) − (ti ) − log σ (ti ) − 1 + log 1 + ξ
events/year nowadays. This decrease in the frequency of i=1
ξ σ (ti )
extreme events also affects the 50-year return level, which (9)
results in 44.5 cm during the sixties and a slightly lower where xi is the threshold excess (or the duration over
value (43.5 cm) at the end of the 2000s. Figure 3d high- the threshold) on day ti and θ is a vector of regression
lights the time dependent analysis taking into account sig- parameters.
nificant trends in both, Pareto and Poisson parameters. The Statistical significance of the time-dependent terms
combination of an increase in the magnitude of extreme compared with the stationary case, is estimated according
events and a decrease in the occurrence rate leads to a rise to likelihood ratio test which relates the deviance function
of the 50-year return value through time. Therefore, Fig. 3 (see Eq. 10) to the Chi squared distribution (see Coles et al.
evidences the importance of using a point-process approach 2001).
since resulting trends may have opposite signs depend-  
ing on the feature (magnitude/frequency) allowed to be Ds = 2 l(θ a ) − l(θ b ) (10)
time-dependent.
In order to investigate if the estimated trends are partly where l(θ a ) and l(θ b ) are two log-likelihood functions that
due to interannual-decadal fluctuations, the statistical rela- differ only in the number of regression parameters con-
tionship between climate indices and storm surge extremes tained in θ.
is studied. A linear term is added to the scale parameter of Grid points where the analysed trends or the NAO covar-
Pareto (Eq. 7) and the Poisson parameter (Eq. 8) to esti- iate are not significant above the 90 % confidence level,
mate the sensitivity of the storm surge exceedances to will show a white dot in the maps of obtained results.
climate patterns. The North Atlantic Oscillation index
(monthly NAO index, available from the Climate Research
Unit http://www.cru.uea.ac.uk/cru/data/nao) was chosen 4 Results
to be used as a covariate in the model since its phase and
strength has been widely linked to storm surges in Europe. 4.1 Descriptive statistics of the storm surges
Positive NAO phases lead to an increase in the storms mag-
nitude and frequency towards north–east Europe, while The spatial characteristics of the storm surge in southern
negative NAO phases lead to a shift and weakening of the Europe can be described from the aforementioned variables
storms towards southern Europe. This causes a negative of the extreme model, i.e. the value of the 99.5 % thresh-
correlation between the sea level and the NAO index along old, the mean number of events per year and the duration
the Mediterranean and eastern North Atlantic (e.g. Calafat of each selected event over and above the threshold. Fig-
et al. 2012; Woolf et al. 2003). Most of the English Chan- ure 4 shows these estimated variables from the latter half of
nel is weakly and negative correlated to NAO, but from the the twentieth century to the present and for each grid point
Straits of Dover up to the North Sea, strong and positive of the GOS dataset. Figure 4a shows the threshold value
correlations are found (e.g. Haigh et al. 2010; Woodworth above which we have considered surge events as extremes.
et al. 2007). As can be seen, there is a latitudinal gradient; values of the
99.5 percentile are higher than 30 cm at latitudes above 40◦
σ (t) = σ0 + σ2 Cov(t); (7) N and decrease gradually to the south, reaching values of
 tf
around 10 cm nearby the Canary Islands. It is also notice-
(t) = ν(t) dt ; ν(t) = ν0 + ν2 Cov(t); able how surge values are greater throughout the continen-
t0 (8) tal shelf due to the wind set-up. This is clearly visible at the

13
Long-term changes in the frequency, intensity and duration of extreme storm surge events...

Fig. 4  Descriptive statistics of storm surge extremes. a Value of the 99.5 % threshold (cm). b Mean number of events per year (extremes/year).
c Mean of the duration over the threshold (hours)

13
 A. Cid et al.

French coast, where its wide shelf originates surge levels because some uncertainties in the GOS dataset were found
of above 35 cm. At the Mediterranean sea, the 99.5 percen- during the validation process (Cid et al. 2014).
tile is larger at the west basin, and the largest values are Trends in frequency are directly given by the non-
located at five semi-enclosed areas (Gulf of Lion, Gulf of homogeneous Poisson parameter (Fig. 5b). In this work,
Gabes, Adriatic Sea, Gulf of Sirte and Aegean Sea). Fig- it is expressed as a percentage of change in the number of
ure 4b depicts the mean number of extreme events per year; extreme events per year. Apart from a positive trend in the
as can be seen, storm surge events over the 99.5 percentile occurrence rate of events assessed in the south of Greece
are usual in semi-enclosed geographical areas every year. (increase of 1.5 %), only statistically significant decreases
More than five events per year over the threshold are found are obtained, mainly in semi-enclosed areas at the south-
in the five mentioned Mediterranean semi-enclosed areas ern Mediterranean Sea. The largest decreases are found in
plus the western and easternmost Mediterranean areas the middle-south Mediterranean (i.e. Gulf of Sirte and Gulf
(i.e. the Alboran Sea and the eastern Levantine Mediterra- of Gabes), at diminution rates in the number of events per
nean basin). The Atlantic part of the domain experiences year of around 2 %. Negative trends of about 1 % are found
less extreme events every year, with values of around two in the Alboran Sea, south of Sicily and Levantine Mediter-
events/year. Figure 4c represents the mean duration of the ranean basin, and slight decreases are detected in the north
extreme events. There are clearly two different patterns Adriatic and Aegean Sea.
regarding this variable: the Atlantic and the Mediterranean. Trends in the duration of extreme events are showed in
Storm surges at the Atlantic part of the domain have a dura- Fig.  5c. Results indicate a general decrease in the Atlan-
tion of around 1 day while the duration at the Mediterra- tic and a general increase in the Mediterranean. The sig-
nean is much smaller, with maximum values of 15 h in the nificance of these trends is mainly relevant in the eastern
deepest region of the Mediterranean basin. The occurrence Mediterranean basin. In particular, positive rates of 1.5 h/
rate and duration maps of storm surges have an inverse spa- year are located at the south of Italy and Greek coasts. Less
tial relationship. This behaviour suggests that the relative pronounced trends are found along the Italian Mediterra-
small spatial and temporal scale of the meso-scale atmos- nean coast, with increasing values in the duration of storm
pheric circulation eddies, added to the semi-enclosed char- surges of rates of 0.5 h/year. Similar values are reached at
acter of some coastlines, limit the duration of storm surge Levantine basin, Alboran Sea and along the coasts of Tur-
events and make them more frequent. key and Egypt.
Summing up, (i) northern regions show larger mag- Overall, some relevant contributions from the analysis
nitudes of high storm surges; (ii) coasts exposed to the of trends for the last half century are: (i) except for some
Atlantic Ocean received extreme events of large duration; specific regions, most of the Atlantic and Mediterranean
(iii) the semi-enclosed areas in the Mediterranean region southern Europe does not show significant increases or
receive the largest and most frequent storm surge events. decreases in extreme storm surges. It may indicate that cli-
mate variations at lower time scales (seasonal, interannual)
4.2 Long‑term trends in frequency, intensity are the main drivers of changes in extreme storm surges.
and duration of extreme storm surges (ii) Only south Greek coasts show an increase of all ana-
lysed properties of extreme storm surge. It can be related to
Once the GPD–P model is applied, we are able to estimate the clear significant increases of the magnitude and dura-
long-term trends. Figure 5 depicts the estimated trends in tion of extreme events in the easternmost Mediterranean
frequency, intensity and duration. White dots represent basin and may indicate an intensification or changes in cir-
areas where trends have a confidence level below the 90 %. culation patterns of cyclone activity in this region. (iii) The
Trends in magnitude (see Fig. 5a, mm/year) are shown estimated trends for the last half century reveal a slight but
through the linear trend estimated in the time-dependent clear decrease in magnitude, occurrence, and duration of
50-year return level. Except for the northern Adriatic, extreme storm surge events in the north Adriatic. A detailed
north–west Atlantic, and Gabes and Sirte gulfs, a gen- analysis should be taken for estimations of changes in Ven-
eral increase is estimated in most of the analysed domain. ice lagoon and surroundings due to the complex bathym-
These increases are, however, significant only in the Medi- etry and coastline of this area.
terranean Levantine basin, south Greek coast and north–
west Genoa gulf, with trends of around 1 mm/year, and the 4.3 Relationship between NAO and extreme storm
Alboran Sea and south coast of Sicily, with increases of surges
about 0.5 mm/year. Negative trends of 2 mm/year are esti-
mated at the Gulf of Gabes and Sirte and the north Adriatic. Most of the Atlantic and Mediterranean southern European
Special caution, however, related to these negative trends regions do not show significant long-term trends in extreme
must be taken in the north Adriatic and Gulf of Gabes storm surges. One explanation for this behaviour could

13
Long-term changes in the frequency, intensity and duration of extreme storm surge events...

Fig. 5  Spatial distribution of the long-term trends. White dots rep- obtained from the time-dependent Poisson parameter and expressed
resent areas where the statistical significance of the trends is lower as percentage by using the mean number of events per year. c Trends
than 90 %. a Trends in intensity (mm/year), obtained from the linear in duration (hours/year), obtained from the second run of the GPD–P
fitting of the yearly 50-year return level. b Trends in frequency (%), model, using the duration of the extremes as the GPD variable

13
 A. Cid et al.

Fig. 6  a Sensitivity of the extremes intensity to NAO (cm/unit index). b Sensitivity of the extremes frequency to NAO (events per year/unit
index). White dots represent areas where the sensitivity is lower than 90 %

be that extreme storm surges have been more affected by winter season. A mainly negative relationship is found
inter-annual and decadal variability than by climate vari- both in terms of intensity and frequency, which means that
ations at longer time scales. This was also described by extreme storm surges are more intense and more frequent
Wahl and Chambers (2015) for the United States coastline, when NAO phase is negative. This is an expected result
where they identify regions with considerable multidecadal since negative NAO phases are characterised by a shift of
extreme sea level variations unrelated to changes in mean storms tracks towards lower latitudes, and by the low pres-
sea level. sure system over the Azores area being higher than during
Since storm surges are directly forced by changes in the positive NAO phases, which leads to an increase in the sea
zonal circulation of the atmosphere and meridional wind level related to the inverse barometer effect. The opposite
components, the relationship between the main coupled occurs from the northernmost English Channel to northern
ocean-atmosphere mode of variability in the area, the North European Seas, where positive NAO phases lead to higher
Atlantic Oscillation (NAO), and extreme storm surges is sea levels due to more intense storm surge events (Wakelin
investigated. et al. 2003; Woodworth et al. 2007; Yan et al. 2004).
Here, we use the monthly NAO to analyse the relation- Although sensitivity to NAO is generally negative, there
ship between this climate index and extreme surges. Note is a relevant difference of NAO influence patterns in mag-
that we focused on NAO conditions whenever the selected nitude and frequency of surge extremes. As can be seen
extreme surge events happened and not only during the from Fig. 6, where white dots represent areas with less

13
Long-term changes in the frequency, intensity and duration of extreme storm surge events...

Fig. 7  a Current 50-year return level (cm). b Shape pareto parameter (non-dimensional)

than 90 % of statistical significance, most southern Europe in the western Mediterranean basin is partly due to incom-
shows a significant influence of negative NAO phase in ing low pressure systems from the Atlantic through the
the occurrence of extreme events (see Fig. 6b) while sen- Gulf of Biscay and Iberian peninsula (Trigo et al. 1999)
sitivity of extremes magnitude is below 90 % in most of and because the cyclones often remain on the belt-shaped
the Atlantic basin and some areas of the Mediterranean Sea coastal area from the Alboran Sea through the Gulf of Lion
(see Fig. 6a). The difference in the statistical significance and the Gulf of Genoa to the Adriatic Sea (Bartholy et al.
of NAO influence in the properties of the extreme events 2009).
may be explained in terms of the local-regional scale; the Figure  6 has, therefore, proven that extreme events in
magnitude of the extremes is more affected by the orog- southern Europe (their magnitude and their occurrence rate)
raphy or the complex coastline than the occurrence of the are significantly affected by NAO, with its negative phase
event. The highest teleconnections to NAO are found in being one of the main drivers of the extreme storm surge
the western Mediterranean basin. Gulfs of Lion and Genoa events, especially in the western Mediterranean basin.
and northern Adriatic are the areas where the magnitude of
extreme surges is more affected by the NAO index; Alboran 4.4 50‑year return level
Sea and Gulf of Gabes are the areas where the frequency
of extremes is more affected by NAO index. These results Extreme value models enable the estimation of very unu-
can be explained by the fact that the occurrence of cyclones sual extreme events from a relatively short data period.

13
 A. Cid et al.

This is often referred as return values and return periods, onshore winds (values of around 35 cm). The mean number
estimations required to design defence structures and to of events per year is mainly lower at the Atlantic part of the
quantify risk. domain than at the Mediterranean coasts (2 vs. 4–5 events/
Figure  7 displays the estimated results for the present year). Regarding the duration of the extremes over the
50-year return storm surge level, together with the esti- threshold, there are clearly two different areas; higher val-
mated shape parameter of the extreme value distribution ues at the Atlantic with mean durations of 24 h, and lower
for the study region. The shape parameter determines the values at the Mediterranean Sea with maximum durations
upper tail of the distribution and consequently describes the of 15 h. This is because, as pointed out by Trigo et al.
behaviour of the most unusual extreme events. In general, (1999), Mediterranean lows are, in general, less intense and
negative values are found, indicating a bounded upper tail. are associated with smaller spatial scales and with shorter
This can be a consequence of one of two reasons: either life cycles than Atlantic synoptic systems. Therefore, dura-
regularly-occurring intense extreme events (e.g. the north– tions of storm surges are also smaller in the Mediterra-
west Iberian Peninsula and the Gulf of Lion), or the scarce nean than in the Atlantic. An inverse relationship between
occurrence of extreme values (e.g. north Atlantic African the duration and the number of extreme events has been
shelf and Ionian Sea). There are also regions where the found. This is, extremes last longer offshore, while close
shape parameter is close or equal to zero (e.g. northern to the coast, and specially along semi-enclosed Mediterra-
Adriatic and eastern Biscay Gulf), indicating an exponen- nean coasts, extreme events are shorter but present a higher
tial behaviour of the upper tail and, therefore, higher mag- occurrence rate every year.
nitudes for the most unusual extreme events. Regarding trends, results show negative patterns
Values for the current 50-year return level were calcu- throughout the domain when studying the frequency of
lated, as shown in Eq. 6, using the non-stationary Pareto– storm surge events (mean values between 1 and 1.5 %),
Poisson model. According to the 99.5 % threshold spatial except for the south coast of Greece where the number of
pattern represented in Fig. 4a, the largest 50-year return events per year tends to increase at rates of 1.5 %. There
values are found at the northern Adriatic, Aegean Sea, are more areas with positive trends in frequency but they
Gulfs of Gabes and Sirte, the Gulf of Lion and along the do not reach the 90 % of significance. Regarding intensity,
Egyptian coast, reaching more than 75 cm. A similar north– both positive and negative trends are found; intensity of
south spatial gradient is detected, although it is less pro- storm surges has increased mainly at the easternmost Medi-
nounced. In contrast, the 50-year return level provides less terranean basin (1 mm/year) while decreasing rates at the
intense values at the north–west coast of the Iberian Penin- coast are mostly reduced to semi-enclosed areas such as the
sula and larger magnitudes at Alboran Sea and the eastern- Gulf of Sirte ( 2 mm/year), Gulf of Gabes and northern
most Mediterranean basin. Adriatic. With respect to duration, two different trend pat-
terns are present: positive in the Mediterranean and nega-
tive in the Atlantic, although at coastal areas significant
5 Summary and discussion trends are only positive within a range of 0.5–1.5 h/year. In
summary, most of the southern European region does not
In this study, a storm surge hindcast (GOS database) span- show significant trends in extreme storm surges; an excep-
ning from 1948 to 2013 and a time-dependent Pareto–Pois- tion to this are regions as the Alboran Sea, gulfs of Gabes
son model are used to characterise and analyse climate and Sirte, northern Adriatic and the easternmost Mediter-
variations in intensity, frequency and duration of extreme ranean. The spatial patterns of the trends in duration indi-
storm surge events. Non-stationarity is included into the cates an increase in the Mediterranean and a decrease in the
extreme model by allowing time-dependent parameters Atlantic but the mean duration of the extremes is higher in
through covariate NAO and linear trends. the Atlantic than in the Mediterranean. It is worth noting
The climatological analysis of not only the magnitude of that decreasing trends found in the northern Adriatic for the
the maximum extreme values, but also their duration and three analysed variables are in agreement with results from
frequency of occurrence throughout the time, has allowed a local studies as for example Trieste, where Raicich (2003)
better characterisation of the extreme storm surges. A gen- found that during the period 1939–2001, strong surges
eral north to south gradient is found over the entire domain became less frequent. The period of study is a key factor
with respect to the selected 99.5 % threshold value (approx- when studying trends; in areas where the inter-annual or
imately from 30 to 10 cm at the Atlantic and from 30 to 20 decadal variability is of great importance, the sign of trends
at the Mediterranean). This spatial pattern is obviously cor- may even change depending on the temporal coverage of
related to the wind speed over this area, as can be checked study. For instance, trends in magnitude around the Alboran
from Menendez et al. (2014). Higher storm surge values Sea switch from negative (not shown) to positive (Fig. 5a)
are located in coastal areas due to water piling induced by when the study period is 1948–2009 or 1948–2013,

13
Long-term changes in the frequency, intensity and duration of extreme storm surge events...

respectively. Therefore, the analysis of linear trends is a of the extremes affect the value of the 50-year return
simplistic approximation and it is totally dependent on the level but also does the frequency with which the extremes
analysed time period. The Alboran sea is an example of an occur. We have also used the North Atlantic Oscillation
area with a strong interannual-decadal variability. index as a covariate in the statistical model, addressing
To test if this interannual-decadal variability of storm the influence of the North Atlantic Oscillation index into
surges is related to the North Atlantic Oscillation, we have the long-term trends. The results found in this work can
studied the response of sea level to the NAO index. This be of great help in the estimation of the total water level
relationship has been analysed previously at different time (tide + surge + wave set up) in coastal areas: extremes of
scales, most of them focused on north–west Europe (e.g. storm surges together with wave height extremes, can be
Haigh et al. 2010; Woodworth et al. 2007). Studies analys- a valuable information in the assessment of the flooding
ing a bigger domain (e.g. Woolf et al. 2003), found that risk.
sea level is generally higher around north–west Europe in
NAO positive winters, while it is generally lower around Acknowledgments  The authors thank Puertos del Estado (Spanish
Ministry of Public Works) for providing the tide gauge records. This
the Azores and southern Europe. We have found an inverse work has been partially funded by the Spanish Ministry of Economy
(negative) connection between NAO and extremes, both in and Competitiveness under the research project PLVMA (TRA2011-
terms of intensity and frequency, consistent with the strong 28900) and the FP7 European project CoCoNet (287844). We also
and negative correlation found by Calafat et al. (2012) thank the reviewer Dr. Ivan Haigh and an anonymous referee for their
valuable comments and suggestions to improve the quality of this
between tide gauge records and NAO when studying sea paper.
level decadal variability. For southern Europe and specifi-
cally for the atmospheric contribution to sea level, Gomis
et al. (2008) found a negative correlation between NAO and References
surge monthly means. Marcos et al. (2009) also obtained
negative correlations between winter NAO and storm surge Arns A, Wahl T, Haigh I, Jensen J, Pattiaratchi C (2013) Estimating
values for the same domain. The significant connection extreme water level probabilities: a comparison of the direct
found in this work between NAO and extremes proves that methods and recommendations for best practise. Coast Eng
81:51–66. doi:10.1016/j.coastaleng.2013.07.003, http://www.
NAO contributes to changes in the extreme storm surges. sciencedirect.com/science/article/pii/S0378383913001270
Therefore, the latest tendency of the NAO index to posi- Arns A, Wahl T, Haigh ID, Jensen J (2015) Determining return water
tive phases since the 80’s, could lead to negative trends in levels at ungauged coastal sites: a case study for northern Ger-
the magnitude and the occurrence rate of extremes, as we many. Ocean Dyn 65(4):539–554. doi:10.1007/s10236-015-
0814-1, http://link.springer.com/10.1007/s10236-015-0814-1
have found here. Hence, in this study we provide details of Bartholy J, Pongrácz R, Pattantyús-Ábrahám M (2009) Analyz-
the relationship of the North Atlantic Oscillation with the ing the genesis, intensity, and tracks of western Mediterranean
extreme storm surges at an homogeneous spatial domain cyclones. Theor Appl Climatol 96(1–2):133–144. doi:10.1007/
and for both, magnitude and occurrence of extreme events. s00704-008-0082-9
Baxter PJ (2005) The east coast Big Flood, 31 January–1 February
Other indexes as the Mediterranean Oscillation Index 1953: a summary of the human disaster. Philos Trans Ser A Math
(MOI), Western Mediterranean Oscillation (WEMO), East Phys Eng Sci 363(1831):1293–1312. doi:10.1098/rsta.2005.1569
Atlantic pattern (EA), East Atlantic/Western Russian (EA/ Bernardara P, Andreewsky M, Benoit M (2011) Application of
WR), Atlantic multidecadal oscillation (AMO), could regional frequency analysis to the estimation of extreme storm
surges. J Geophys Res Oceans 116(C2):C02008. doi:10.1029/2
also have an influence in the Atlantic and the Mediterra- 010JC006229
nean basin. Thus, a further study including these climatic Bernier NB, Thompson KR (2006) Predicting the frequency of storm
indexes should be carried out. surges and extreme sea levels in the northwest Atlantic. J Geo-
The study of the 50-year return level has shown a gen- phys Res Oceans 111(C10):C10009. doi:10.1029/2005JC003168
Butler A, Heffernan JE, Tawn Ja, Flather Ra, Horsburgh KJ (2007)
eral north to south gradient at both the Atlantic and Medi- Extreme value analysis of decadal variations in storm surge
terranean basins. Some exceptions to this are found along elevations. J Mar Syst 67(1–2):189–200. doi:10.1016/j.
coastal areas in the Mediterranean African coast, reaching jmarsys.2006.10.006
values up to 70 cm for instance at the Gulf of Lion or Lib- Calafat FM, Chambers DP, Tsimplis MN (2012) Mechanisms of dec-
adal sea level variability in the eastern North Atlantic and the
yan and Egyptian coasts. A similar spatial pattern for the Mediterranean Sea. J Geophys Res Oceans 117(C9):C09022. doi
50-year return level was obtained by Marcos et al. (2009) :10.1029/2012JC008285
where they also found higher values along the Tunisian and Cavicchia L, von Storch H, Gualdi S (2014) Mediterranean tropical-
Egyptian coasts. like cyclones in present and future climate. J Clim 27(19):7493–
7501. doi:10.1175/JCLI-D-14-00339.1
Finally, in this study, we have presented a statisti- Church J, Clark P, Cazenave A, Gregory J, Jevrejeva S, Levermann A,
cal framework (GPD–P time-dependent model) to study Merrifield M, Milne G, Nerem R, Nunn P, Payne A, Pfeffer W,
extreme storm surge events, and we have used it to esti- Stammer D, Unnikrishnan A (2013) Sea level change. In: Cli-
mate long-term trends; showing that not only the intensity mate Change 2013: The Physical Science Basis. Contribution of

13
 A. Cid et al.

Working Group I to the Fifth Assessment Report of the Intergov- buoy measurements. Geophys Res Lett 35(22):L22,607. doi:10.1
ernmental Panel on Climate Change, pp 1137–1216 029/2008GL035394
Cid A, Castanedo S, Abascal AJ, Menéndez M, Medina R (2014) A Menendez M, Mendez FJ, Losada IJ (2009) Forecasting seasonal to
high resolution hindcast of the meteorological sea level com- interannual variability in extreme sea levels. ICES J Mar Sci
ponent for Southern Europe: the GOS dataset. Clim Dyn 43(7– 66(7):1490–1496. doi:10.1093/icesjms/fsp095
8):1–18. doi:10.1007/s00382-013-2041-0 Menendez M, García-Díez M, Fita L, Fernández J, Méndez FJ, Gutié-
Coles S, Bawa J, Trenner L, Dorazio P (2001) An introduction to sta- rrez JM (2014) High-resolution sea wind hindcasts over the
tistical modeling of extreme values, vol 208. Springer Series in Mediterranean area. Clim Dyn 42(7–8):1857–1872. doi:10.1007/
Statistics, London s00382-013-1912-8
Gomis D, Ruiz S, Sotillo MG, Álvarez Fanjul E, Terradas J (2008) Ma Merrifield, Genz AS, Kontoes CP, Marra JJ (2013) Annual maxi-
Low frequency Mediterranean sea level variability: the contri- mum water levels from tide gauges: contributing factors and
bution of atmospheric pressure and wind. Glob Planet Change geographic patterns. J Geophys Res Oceans 118(5):2535–2546.
63(2–3):215–229. doi:10.1016/j.gloplacha.2008.06.005 doi:10.1002/jgrc.20173
Gräwe U, Burchard H (2012) Storm surges in the Western Baltic Sea: Mudersbach C, Jensen J (2010) Nonstationary extreme value analy-
the present and a possible future. Clim Dyn 39(1–2):165–183. sis of annual maximum water levels for designing coastal struc-
doi:10.1007/s00382-011-1185-z tures on the German North Sea coastline. J Flood Risk Manag
Haidvogel DB, Arango HG, Hedstrom K, Beckmann A, Malanotte- 3(1):52–62. doi10.1111/j.1753-318X.2009.01054.x, http://doi.
Rizzoli P, Shchepetkin AF (2000) Model evaluation experiments wiley.com/10.1111/j.1753-318X.2009.01054.x
in the North Atlantic Basin: simulations in nonlinear terrain- Raicich F (2003) Recent evolution of sea-level extremes at Tri-
following coordinates. Dyn Atmos Oceans 32(3–4):239–281. este (Northern Adriatic). Cont Shelf Res 23(3–4):225–235.
doi:10.1016/S0377-0265(00)00049-X doi:10.1016/S0278-4343(02)00224-8
Haidvogel DB, Arango H, Budgell WP, Cornuelle BD, Curchitser E, Serafin KA, Ruggiero P (2014) Simulating extreme total water levels
Di Lorenzo E, Fennel K, Geyer WR, Hermann AJ, Lanerolle L, using a time-dependent, extreme value approach. J Geophys Res
Levin J, McWilliams JC, Miller AJ, Moore AM, Powell TM, Oceans 119(9):6305–6329. doi10.1002/2014JC010093, http://
Shchepetkin AF, Sherwood CR, Signell RP, Warner JC, Wilkin doi.wiley.com/10.1002/2014JC010093
J (2008) Ocean forecasting in terrain-following coordinates: Shchepetkin AF, McWilliams JC (2005) The regional oceanic mod-
formulation and skill assessment of the Regional Ocean Mod- eling system (ROMS): a split-explicit, free-surface, topography-
eling System. J Comput Phys 227(7):3595–3624. doi:10.1016/j. following-coordinate oceanic model. Ocean Model 9(4):347–
jcp.2007.06.016 404. doi:10.1016/j.ocemod.2004.08.002
Haigh I, Nicholls R, Wells N (2010) Assessing changes in extreme Storch H, Woth K (2008) Storm surges: perspectives and options.
sea levels: application to the English channel, 1900–2006. Cont Sustain Sci 3(1):33–43. doi:10.1007/s11625-008-0044-2
Shelf Res 30(9):1042–1055. doi:10.1016/j.csr.2010.02.002 Tebaldi C, Strauss BH, Zervas CE (2012) Modelling sea level rise
Haigh ID, Wijeratne EMS, MacPherson LR, Pattiaratchi CB, Mason impacts on storm surges along US coasts. Environ Res Lett
MS, Crompton RP, George S (2014) Estimating present day 7(1):014,032. doi:10.1088/1748-9326/7/1/014032
extreme water level exceedance probabilities around the coast- Trigo IF, Davies TD, Bigg GR (1999) Objective climatology of
line of Australia: tides, extra-tropical storm surges and mean sea cyclones in the Mediterranean region. J Clim 12(6):1685–1696.
level. Clim Dyn 42:121–138. doi:10.1007/s00382-012-1652-1 doi:10.1175/1520-0442(1999)012<1685:OCOCIT>2.0.CO;2
Hallegatte S, Ranger N, Mestre O, Dumas P, Corfee-Morlot J, Her- Tsimplis M (1997) Extreme sea-level distribution and return periods
weijer C, Wood RM (2011) Assessing climate change impacts, in the Aegean and Ionian Seas. Estuar Coast Shelf Sci 44(1):79–
sea level rise and storm surge risk in port cities: a case study 89. doi:10.1006/ecss.1996.0126
on Copenhagen. Clim Change 104(1):113–137. doi:10.1007/ Wahl T, Chambers DP (2015) Evidence for multidecadal vari-
s10584-010-9978-3 ability in US extreme sea level records. J Geophys Res Oceans
Luceño A, Menéndez M, Méndez FJ (2006) The effect of tempo- 120(3):1527–1544. doi10.1002/2014JC010443, http://doi.wiley.
ral dependence on the estimation of the frequency of extreme com/10.1002/2014JC010443
ocean climate events. Proc R Soc A Math Phys Eng Sci Wakelin SL, Woodworth P, Flather RA, Williams JA (2003)
462(2070):1683–1697. doi:10.1098/rspa.2005.1652 Sea-level dependence on the NAO over the NW Euro-
Marcos M, Tsimplis MN, Shaw AGP (2009) Sea level extremes in pean Continental Shelf. Geophys Res Lett 30(7):1403.
southern Europe. J Geophys Res Oceans 114(C1):1–16. doi:10. doi10.1029/2003GL017041, http://doi.wiley.
1029/2008JC004912 com/10.1029/2003GL017041
Marcos M, Jordà G, Gomis D, Bn Pérez (2011) Changes in storm Woodworth PL, Blackman DL (2004) Evidence for systematic
surges in southern Europe from a regional model under cli- changes in extreme high waters since the mid-1970s. J Clim
mate change scenarios. Glob Planet Change 77(3–4):116–128. 17(6):1190–1197. doi:10.1002/joc.761
doi:10.1016/j.gloplacha.2011.04.002 Woodworth PL, Ra Flather, Ja Williams, Wakelin SL, Jevrejeva S
Méndez FJ, Menéndez M, Luceño A, Losada IJ (2006) Estimation of (2007) The dependence of UK extreme sea levels and storm
the long-term variability of extreme significant wave height using surges on the North Atlantic Oscillation. Cont Shelf Res
a time-dependent Peak Over Threshold (POT) model. J Geophys 27(7):935–946. doi:10.1016/j.csr.2006.12.007
Res Oceans 111(C7):C07,024. doi:10.1029/2005JC003344 Woolf D, Shaw A, Tsimplis M (2003) The influence of the North
Méndez FJ, Menéndez M, Luceño A, Losada IJ (2007) Analyz- Atlantic Oscillation on sea-level variability in the North Atlantic
ing monthly extreme sea levels with a time-dependent GEV region. doi:10.1080/10236730310001633803
model. J Atmos Ocean Technol 24(5):894–911. doi:10.1175/ Yan Z, Tsimplis MN, Woolf D (2004) Analysis of the relationship
JTECH2009.1 between the North Atlantic oscillation and sea-level changes in
Menéndez M, Woodworth PL (2010) Changes in extreme high water northwest Europe. Int J Climatol 24(6):743–758. doi:10.1002/
levels based on a quasi-global tide-gauge data set. J Geophys joc.1035
Res Oceans 115(C10):1–15. doi:10.1029/2009JC005997
Menéndez M, Méndez FJ, Losada IJ, Graham NE (2008) Variability
of extreme wave heights in the northeast Pacific Ocean based on

13