Variability in storm climate along the Gulf of Cadiz: the role of large scale atmospheric forcing and implications to coastal hazards

In the context of increased coastal hazards due to variability in storminess patterns, the danger of coastal damages and/or morphological changes is related to the sum of sea level conditions, storm surge, maximum wave height and run up values. In order to better understand the physical processes that cause the variability of the above parameters a 44 years reanalysis record (HIPOCAS) was used. The HIPOCAS time-series was validated with real wave and sea-level data using linear and vector correlation methods. In the present work changes in the magnitude, duration, frequency and approach direction of the Atlantic storms over the Gulf of Cadiz (SW Iberian Peninsula) were identified by computing various storm characteristics such as maximum wave height, total energy per storm wave direction and storm duration. The obtained time-series were compared with large-scale atmospheric indices such as the North Atlantic Oscillation (NAO) and the East Atlantic pattern. The results show a good correlation between negative NAO values and increased storminess over the entire Gulf of Cadiz. Furthermore, negative NAO values were correlated with high residual sea level values. Finally, a joint probability analysis of storm and sea level analysis resulted in increased probabilities of the two events happening at the same time indicating higher vulnerability of the coast and increased coastal risks. The above results were compared with coastal inundation events that took place over the last winter seasons in the province of Cadiz.

level (Sebastiao et al., 2008). The output time-step was 3 hours for all the parameters. 156 From the above data set five stations were selected that cover the entire north coast of the 157 Gulf of Cadiz (Figure 1). From west to east the selected stations are: Faro, that represents 158 the most exposed part of the Gulf of Cadiz with a narrow shelf and a steep continental 159 slope; Huelva and Seville, located further to the east, at the widest part of the continental 160 shelf and partially sheltered from the west and north component winds by the Cape St. 161 Vincent; Cadiz station is located further to the southeast where the shelf width starts to 162 reduce and the coastline is more exposed to the Atlantic storms; finally the Zahara station

HIPOCAS data validation 170
An extensive validation exercise was undertaken by Mendez et al. (2006) between the 171 HIPOCAS wave data and wave buoy data collected around the Spanish coasts. However, 172 in order to optimize the results in the Gulf of Cadiz, a new correction was applied in the 173 present study that consisted in: (i) fitting the model wave height to observations focusing 174 mainly in the case of storm conditions; (ii) evaluating differences between model and 175 measured wave directions using a vector correlation approach (Kundu, 1976). The wave 176 height validation was evaluated by calculating the bias and the Brier Skill Score (BSS). 177 The latter parameter relates the variance of the difference between data and model with 178 the variance of the data. BSS=1 means perfect skill, BSS=0 means no skill (Roelvink et 179 al., 2009). The wave direction validation was evaluated based on the Kundu coefficient 180 and mean angle rotation (θ). Both wind waves and swell were analysed together since 181 they coexist during storm events and no spectral information was available. 182

183
In order to obtain a statistically independent wave height data set of storm conditions a 184 peak over threshold analysis (POT) was used for the period with simultaneous model and 185 observation data (Kamphuis, 2000). The above analysis allowed producing a correction 186 based on the peak values of each storm and was then applied to the entire data set. The 187 threshold value for the POT analysis was set as 1.5m wave height (a threshold value 188 proposed by the local authorities for civil protection), and a storm independence criteria 189 (time between two consecutive independent storms) was calculated based on the integral The agreement between the original (uncorrected) model data and buoy data, in both Faro 210 and Cadiz wave buoys, was tested for wave height and direction simultaneously using the 211 Kundu (1976) vector correlation approach. This correlation method produces a 212 coefficient between the directional wave heights of the two time series and the main 213 angle (θ) through which the first series would have to be rotated anticlockwise to match 214 the direction of the second series. Although overlapping of directional wave data between 215 the buoy of Cadiz and the HIPOCAS data only exists over part of 2001, the period is long 216 enough to cover both calm and stormy conditions. For the case of Faro the overlapping 217 period was much longer (1997)(1998)(1999)(2000)(2001). The correction coefficients calculated for the two 218 time series were not statistically different; hence a common correction equation was 219 derived for both sites and then applied to all the wave data. 220 221 Differences in wave propagation direction between the measured and modelled data are 222 presented against the significant wave height in monthly data corresponding to the HIPOCAS dataset were obtained from NOAA climate 281 centre and were calculated from the rotated EOF of the 500 hPa geopotential height. The 282 SCA and POL did not present any significant correlation; hence the results are not 283 presented. Significant differences between two correlation coefficients were tested using 284 the Fisher r-to-z transformation (Fisher, 1970). This method converts first each 285 correlation coefficient into a z-score. Then, making use of the sample size employed to 286 obtain each coefficient; these z-scores are compared.

Wave and Residual Sea Level Climate 306
The mean annual cycle for the corrected significant wave height (Hsc) and the associated 307 wave directions for the coastal area of the northern Gulf of Cadiz are presented in Figure  308 5a and 5b. The average wave heights over the area are higher during the winter months 309 and part of the autumn.  (Table 1). 390 391 For the case of EA the above parameters explain a small but statistically significant part 392 of the variability (Table 2)  In accordance with the correlation presented above between the storm variables and 480 NAO, the joint probability analysis was undertaken separately for positive and negative 481 NAO events. In general the ratio between storm events occurred during a positive NAO 482 and events occurred during a negative NAO phase is close to 1 for all sites (Table 3). For 483 NAO phases with an index higher/lower than +/-1 and +/-1.5 it can be seen that the 484 negative NAO phases present almost twice the events than the positive ones for the 485 central part of the Gulf (Seville and Huelva). On the contrary the two sites located at the 486 extremities of the Gulf of Cadiz (Faro and Zahara) do not show this pattern. This is partly 487 due to the strong easterly winds that can also create short-fetch storms for these areas, 488 such events are more frequent during positive NAO (Dorman et al., 1995). These events 489 are present in the wave record of Zahara due to the proximity to the Strait of Gibraltar 490 and in Faro due to the orientation of the coastline and the considerable easterly fetch. 491 Differences are not present in extreme NAO phases (-2 > NAO > +2) probably due to the 492 small number of events (Table 3). 493

494
The joint probability analysis for positive and negative NAO events with index 495 higher/lower than ±1.5 is presented in Figure 10