The increasing degree of human activity in the last decades has significantly modified and placed a great pressure on coastal regions. The impact and vulnerability of these regions are expected to increase in the future, due to the different effects of climate change, such as the global sea level rise and the changes in the regional storm climate and the atmospheric circulation patterns. Storms are one of the natural agents that cause the most dramatic and long-lasting changes in coastal zones, despite the fact that their immediate action takes place over short time periods [1
]. Thus, in order to provide a reliable scenario of the future impacts of climate change on coastal areas at a regional scale, it is important to look and understand how storm climate has evolved over the past decades, by providing a realistic analysis of the variability and trends that are associated with those extreme events. Previous studies, based on observational data [2
], altimetry measurements [3
], or modelled hindcast [5
] convey a consistent picture of increasing trends in wave storminess at the northern Atlantic above the 50° parallel, and decreasing trends at mid and southern latitudes. By its latitude, the Bay of Biscay area is located in the transition between both trends, being a pivotal sector in the North Atlantic basin, which might explain the contrasted results provided by several sources [9
Waves are mostly generated by the action of the wind, and they dynamically link the atmospheric circulation, represented, in mid-latitudes, by extratropical cyclones, and the sea surface. Although it is widely accepted that large wave storms in the Atlantic Ocean are a direct consequence of the passage of extratropical cyclones, the details of this relationship at the regional scale need to be clarified, since it is not known whether the magnitude of the event is more closely related with the track or the depth of the forcing cyclones. Additionally, these cyclones usually propagate along preferred trails, so-called storm tracks, and the phases of enhanced or reduced storminess along European Atlantic coastlines are modulated by hemispheric-scale circulation patterns, known as teleconnections [14
The examination of the temporal variability of the storm climate is a previous, yet a key question, to understanding the degree of exposure to present and future coastal storms for a particular seaside region. Consequently, the aims of this study are first, to understand the relationship between extreme wave events along the Bay of Biscay area and the extratropical cyclones, determining how the atmospheric circulation influences the characteristics of wave storms. Secondly, we will analyze trends in storminess and connect them with the temporal variability of large-scale-climate patterns, such as the main teleconnections that affect the Atlantic area and the coastal regions of Western Europe.
2. Study Area, Material and Methods
The study area is the south-easternmost corner of the Bay of Biscay, off the Spanish coast (Figure 1
). This sector is characterized by a rugged but diverse coastline: cliffs and rocky coasts are dominant, but sandy beaches and wetlands are also present. At the present, most of the coast is classified as slightly erosive, although scattered signs of retreat are evident [17
]. The coastal municipalities concentrate approximately 75% of the population and contribute up to 80% of the regional GDP. As a consequence, this region is strongly affected by extreme events, and the storm events are of great economic impact [20
Data used in this research come from several sources (Table 1
). Observed wave records (significant wave height (Hs
), wave period (Tp
), and wave direction (θ)) were obtained from the directional Bilbao-Vizcaya buoy (3.09° W; 43.64° N, mooring depth 870 m), belonging to the REDEXT network of Puertos del Estado. This buoy provided three-hourly values from 1991 to 1996, and hourly values onwards; thus, in order to work with a homogeneous temporal sampling, only the data from 1997 to 2015 were analyzed. Moreover, the original time series were filtered with a 6 h moving average for consistency with the temporal resolution (6-hourly) of the model outputs. The latter consists of a hindcast wave dataset, provided by Dr. Gillaume Dodet and was obtained from the third-generation spectral wave model WAVEWATCH III, forced with the 6-hourly wind fields of the NCEP (National Centers for Environmental Prediction) /NCAR (National Center for Atmospheric Research) Reanalysis 1 project, from January 1948 to March 2015 (for details regarding the modeling approach, please see Masselink et al. [22
]). Only significant wave height was retrieved from the hindcast wave dataset.
Although hindcast wave data are useful to for analyzing long-term trends and large spatial patterns, particularly in deep waters, the coarse resolution of the original wind fields prevents the reproduction of the complete spectrum of the observed wave variability. Consequently, observed values were preferred for deriving a classification for coastal storms, based on the measured wave characteristics, whereas the long-term wave variability was analyzed using the closest grid point to the buoy from the hindcast output. Nevertheless, to assess the model skill, simulated significant wave height values were compared to the observed record through the calculation of some statistics. Moreover, some derived storm parameters were also evaluated for both datasets (namely, the number of storms and the duration).
The various methods of evaluating the storminess of a coastal region can be summarized in two main approaches, both used in this paper: (1) statistical and (2) climatological [23
]. The statistical approach objectively quantifies coastal storm variability by applying a procedure that includes the definition and identification of individual storms, the choice of the parameters to characterize them, and the classification of the storms. Since so far there is no universally accepted definition for storm, a location-specific criterion that simplifies the extraction and comparison of storminess indices across several datasets was applied [24
]. As a result, a storm occurs when the Hs
values are greater than the 95th percentile of the daily maximum Hs
(4.3 m) for both datasets, provided that, at least one record exceeds the 99th percentile of the daily maximum Hs
(6.2 m). In order to identify statistically independent storms, the interval between two consecutive storms (storm peak to storm peak) should not be less than 36 h; otherwise, they are regarded as the same storm. Once a storm is identified, several indices are obtained to characterize its relevant properties: duration (number of records exceeding the aforementioned 95th percentile, assuming a minimum duration of 12 h, equaling the mesotidal cycle of the sea level along the coast of northern Spain); Hs
, and wave direction during the storm peak; and the Storm Power Index (SPI) [25
]. This latest index provides an approximation of the intensity, and consequently, the potential hazards that are induced by a storm. It is defined by the following equation:
is the significant wave height and td
is the duration of the storm in hours.
The atmospheric environment associated with wave storms (synoptic climatological approach) was examined by two complementary approaches. The first followed an Eulerian approach, which considers the passing of air particles through fixed points, and uses the surface pressure values around a selected region to quantify the atmospheric conditions by the estimation of several atmospheric indices (e.g., zonal (U) and meridional (V) wind components, and the strength of the geostrophic flow and the vorticity [21
]). The second approach used a Lagrangian methodology, based on the tracking of extratropical cyclone pressure minimums and the calculation of some output variables [26
] like, among others, the date (at 6-h intervals), the position (latitude and longitude) of each cyclone, and sea level pressure—in hPa—at the cyclone center (depth). The distance from the buoy to the center of each cyclone (in km) was derived through the Haversine’s formulae. A cyclone activity index (CAI) [27
] was calculated from a square box of 10° size centered at 5° W, 55° N, as Iy
, where C
is the number of cyclones observed during each winter in year y
, and L
is the temporal average of the local sea level pressure Laplacian. In both cases, the raw pressure values used as input were extracted from NCEP/NCAR Reanalysis 1 [28
]. When pairing storms with atmospheric information for the region of interest, we followed an environment-to-circulation approach [29
]: first, storms were identified, and after that, atmospheric configurations associated with the extreme local wave storms were subsequently determined. To do so, observed storms from the period of 1997 to 2015 (characterized by their values of Hs
, wave direction at the storm peak, and total SPI released during the whole episode) and simultaneous atmospheric variables (local sea level pressure, zonal and meridional wind component, vorticity, distance, and depth of the cyclones) were arranged together and subjected to a cluster analysis to identify and group them into homogeneous sets. To minimize the large differences in variance, a prior standardization was performed to the original variables, and Ward’s hierarchical algorithm was used, which avoids the “snowball effect” upon small samples [30
Monthly values of the most active patterns for the north Atlantic region (North Atlantic Oscillation—NAO, Eastern Atlantic—EA, Eastern Atlantic/Western Russia—EA/WR, and Scandinavian—SCAND) were downloaded from the Climate Prediction Center (http://www.cpc.ncep.noaa.gov/products/precip/CWlink/daily_ao_index/teleconnections.shtml
) and the West Europe Pressure Anomaly (WEPI) index were used to assess the connection between wave storms, atmospheric parameters, and large scale circulation patterns. The West Europe Pressure Anomaly (WEPI) index was calculated as the difference between sea level pressure records of Valentia (Ireland) and Santa Cruz de Tenerife (Canary Islands, Spain) [31
]. All monthly records were averaged into seasonal values. Quantification of the relationships was made using the non-parametric Spearman rank correlation coefficient, which is recommendable for data that is not normally distributed. The statistical significance of the long-term trends in storminess and atmospheric variability was revised by the means of robust, non-parametric methods, the Sen’s slope estimator [32
] and the Mann-Kendall test, to figure out the change per unit time of the observed trends, as well as the sequential version of the Mann-Kendall test [33
], to detect change points in long-term time series. In this version, a normalized Kendall tau was calculated either from the beginning (end) of the time series, conforming to a progressive u(t) or a retrograde u’(t) series; points where both series crossed themselves were considered to be approximate potential trend turning points. When either the progressive or the retrograde values exceeded certain confidence limits before and after the crossing points, the trend was considered to be significant at a pre-established significance level. Autocorrelations were computed for all the time series at varying time-lags, checking for randomness. As all lag-1 serial correlation coefficients were statistically not significant, there was no need to pre-white the data; thus, the statistical tests described above were applied to the original time series [34
Since storminess is a key issue in coastal erosion and climate change studies, characteristics and temporal evolution of wave storms along the Spanish coast of the Bay of Biscay area are the main subject of our analysis. The first task was accomplished using observed data from Bilbao-Vizcaya buoy, covering the period 1997–2015; meanwhile, a hindcast database, spanning from 1948–2015, was used for the second task after checking the reliability of the latter.
Since our interest is to highlight the relationship between coastal storms and atmospheric forcing, we first applied a statistical procedure to identify and characterize storms, and secondly, a clustering to group them into different types. The attributes subject to grouping did not exclusively include local wave parameters, but also atmospheric variables that reproduced large scale processes. Such procedures allowed us to recognize two types of storms in response to extratropical cyclones that follow different tracks. Sea states corresponding to each class are mainly distinguished by the wave period and wave energy; distant cyclones produce higher wave periods, more westerly waves, and more energetic conditions (swell conditions) than closer cyclones (mixed swell and wind sea states). The current approach differs from other methodologies applied to identify and characterize energetic wave conditions. One example is the use of unsupervised clustering algorithms to group wave data into a few sea-state modes, each of them being associated with contrasted values for wave height, period, and direction [38
]. Other approaches [40
] that focus on the potential effect of storms upon coastlines, categorized storms into several classes on the basis of the energy content for each storm, but the link between wave storminess and large scale circulation is less evident than in our approach.
Subsequently, the long-term evolution of storminess, quantified by the SPI calculated from hindcast data, was compared with the evolution of the large scale circulation. We have not found any significant trends, although substantial interannual and decadal variability is present. The most active period of storminess spanned the second half of the 1970s and the 1980s; the decades following that maximum experienced a reduced activity, as well as in the initial years of the time series.
Those stormy seasons do not coincide with studies on storminess along other Atlantic regions of the Iberian Peninsula; rather, the Bay of Biscay area shares a common temporal behavior with northerly latitudes. The most intense stormy seasons in the Gulf of Cadiz area were 2010, 1963, and 1996, when cyclones followed a southerly-displaced track, which was related to the negative phases of the NAO [42
]. Storm variability along the northern coast of Spain closely resembles the storm activity north of 55° N: low during the 1950s, increased through the 1960s, and peaking in the early 1990s, before decreasing until 2014 [44
]. Such a trend was part of a decadal variability which, at secular time scales, does not show a trend in storm numbers.
An interesting feature is the occurrence of the exceptional winter of 2013–2014, at the end of the period of analysis, which offered record-breaking numbers of storms, wave energy, cyclonic activity, and morphological impacts on the coast of France and the UK [48
]. Analysis is underway on the effects of those storms during the winter of 2014, which also offer a picture of extraordinary damage along the northern coast of Spain in both the natural environment, and on infrastructure and equipment, the restoration of which was worth around €70 million [20
Finally, we have confirmed the relevance of the recent WEPI patterns for wave storms, Energy, and related cyclonic activity along the northern coast of the Iberian Peninsula; however, the correlation values that are obtained are lower in comparison with the French coastline [31
]. The underlying reason is that many storms that hit the Spanish coast need a more northerly component of the large scale flow (Cluster 2), while the French coast is open to storms with a zonal component (Cluster 1 [50
]). In the case of other “classical” teleconnections patterns, the EA/WR pattern is most relevant for the occurrence of wave storms along the northern coast of Spain, mostly linked to Cluster 2; the negative sign of the correlation between the seasonal value of the EA/WR index and the SPI reflects the association between wave storms and negative tropospheric height anomalies over Western Europe. The EA pattern, related to Cluster 1, as well as the NAO pattern (Cluster 2) does not show significant correlations with wave storminess. In that sense, these results are not in total agreement with previous publications, since, for example. Dodet et al. [5
] found weak to strong correlations between winter NAO indices and the annual 90th percentile of wave heights, while Charles et al. [10
] and Le Cozannet et al. [14
] also found a strong and direct correlation between the NAO and EA indices, and high-energy waves. Again, we think that the orientation of the coast in both sides of the Bay of Biscay can explain such a disagreement.
Coastal erosion depends not only on the energy released by the storms, derived from significant wave height, but also from other attributes, such as the storm duration, the clustering of successive storms, or the incidence angle with which the wave trains hit against the coastline. Climate scenarios at the basin scale show that global warming might lead to an eastward extension of the North Atlantic storm track [51
]. However, such evolution will offer substantial regional variability due to a complex association of processes working at different spatial and temporal scales. For example, an intensification and northeastward shift of strong wind core in the North Atlantic Ocean will induce a clockwise shift of winter swell directions along the Bay of Biscay [53
]. Such changes would impact the coastal dynamics, modifying the rates of accretion or erosion, by reducing the long-shore sediment flux.