Assessment of vulnerability to storm induced flood hazard along diverse coastline settings

European coasts suffer notably from hazards caused by low-probability and high-impact hydrometeorological events. The aim of the study is to assess in probabilistic terms the magnitude of storm induced flooding hazard along Varna regional coast (Bulgaria, western Black Sea) and to identify susceptible coastal sectors (hotspots). The study is performed employing the Coastal Risk Assessment Framework (CRAF) developed within EU FP7 RISC-KIT project. It constitutes a screening process that allows estimation of Total water level was the chief property considered for calculation of coastal flooding hazard. It was estimated using Holman model (for sandy beaches) and EurOtop formulation (for artificial or rocky slopes). Resulting values were subjected to Extreme Value Analysis to establish that the best fitting distribution corresponds to Generalized Extreme Value distribution. Furthermore, hazard extents were modelled by means of bathtubbing or overwash estimation in order to form the flooding hazard indicator. Land use, social vulnerability, transport systems, utilities and business settings were considered as exposure indicators. Finally, potential risk was assessed by coastal indices following an index-based methodology, which combines hazard and exposure study found that the concentration of hotspots is highest in Varna Bay.


Introduction
Storms and related disasters are one of the most important phenomena producing coastal hazards and endangering human life and occupation. Recent and historic high-impact events (e.g. Xynthia [1], Ligurian Flash Floods [2] and 1953 North Sea storm surges [3]) have proved the damage that could be caused by marine hazards, which European coastal areas are exposed to. In western Black Sea, particularly large damage was caused by storms that occurred in the late 1970s and the early 1980s, although increased storm activity was also observed in the early 1990s [4][5][6]. Even if the past decade has seen a return to relatively calmer storm conditions, reducing vulnerability to storm impacts is not likely to be expected, especially considering severe storms during 2010-2012 [7][8][9].
Recent climatic studies have revealed that coastal risk is likely to increase in the future due to increase of frequency and intensity of coastal hazards, such as surgedriven floods, erosion and flash floods [10]. On the other hand, due to population growth in coastal areas ± already 40.8% of the EU population lives in coastal regions [11] ± risk is also likely to increase since the consequences will possibly increase as well.
For these reasons, storm-induced flooding has recently become a topic of increased scientific and political interest. Thus, the EU Floods Directive [12] identifies marine inundations and resulting negative geoecological and socio-economic consequences as a specific category representing major environmental threat for European coasts. Therefore, it requires Member States to assess if water sources and coastlines are subject to flood risk, to map the flood extent, assets and population at risk in vulnerable areas, and to take adequate and coordinated measures to reduce the flood risk.
Hence, the aim of the present study is to assess in probabilistic terms the magnitude of stormǦinduced flooding hazard along Varna regional coast and to identify coastal hotspots in support to coastal managers, decisionǦ and policyǦmakers. Results can be used for further high resolution risk modelling of the most vulnerable areas and for design of disaster risk reduction strategies aimed at increased coastal resilience to lowǦfrequency, high-impact hydroǦmeteorological events.
The study is performed employing the Coastal Risk Assessment Framework (CRAF) developed within EU FP7 project RISC-KIT (Resilience-Increasing Strategies for Coasts ± toolKIT) [13]. Briefly, this framework constitutes a screening process that allows delimitation of susceptible alongshore sectors (hotspots) by assessing relevant hazard intensities, hazard extents and potential UHFHSWRUV ¶ H[SRVXUH YXOQHUDELOLW\ ZLWKLQ WKHP 7KLV assumes sufficiently accurate reproduction of existing (or projected) spatial variability of both coastal morphology and hydroǦmeteorological forcing in terms of extreme storm events. The latter provides different combinations of boundary conditions that result in certain response to a hazard. Eventually, the variety of estimated responses (representing entire range of natural variability of coastal hazard for all tested conditions) gives possibility to perform a probabilistic analysis on the obtained coastal impact dataset [14,15]. The main outcome is the probability distribution function of a given hazard that allows hazard parameters of interest to be directly evaluated. Receptors taken into consideration in the framework are land use, population, transport, utilities and economic activities.
Finally, the approach in use results in assessment of potential risk through calculation of coastal indices following an existing index-based methodology. It combines several hazard and exposure indicators into a single index, thereby allowing a rapid comparison of coastal sectors vulnerability [16][17][18].
Index-based approach was applied extensively worldwide [e.g. [19][20][21][22]. Similar studies were performed for Varna regional coastline as well [23,24] The first one was ortophotogrametric digital surface model of 10 cm horizontal and vertical resolution available for two coastal areas: a strip located northward of Varna comprising several international seaside resorts (sectors 2-7) and Varna city coast (sectors [17][18][19][20][21][22][23][24][25][26][27][28][29]. The topography of areas outside those high resolution domains was obtained from 1:5 000 topographic maps by digitizing contour lines up to 30 m a.m.s.l. [25]. These sets were complemented by field measurements at several beaches situated within the study site and performed in 2010-2011. Shallow water bathymetry was generated using data from single-beam echo-sounder surveys. Thus, the resultant DEM has 1 m horizontal and 10 cm vertical resolution. For the study purposes, Varna regional coast was divided into 65 coastal sectors of average length ~ 1.1 km according to the predominant morphological setting: 33 of sectors are mainly sandy beaches (8 of them are marked with dune presence), 14 are cliffs of various height, and 11 are artificial (man modified), while 7 sectors are of mixed type -cliffs with adjacent small beaches. In Figure 1, coast types are shown as coloured strips corresponding to each sector, as strip width is selected purely for visual representation purposes.
Sectors were characterized at least by one representative average profile (average scenario) and one sensitive profile (worst scenario) as the total number of profiles was 159. Generally, they were selected depending on their slope. The worst scenario was defined by profiles with steeper slopes inducing larger run-up for the same forcing conditions, which is valid for continuously rising slopes, in particular. In case of lowlaying coastal topography (optionally with dune presence) scenarios were determined on the basis of overwash extent magnitudes. For average scenario, the range of profile slopes is 0.014÷0.391, while for worst one ± 0.037÷0.562.

Extreme event definition
The study follows response approach [26] making use of long-term water level and wave data to estimate flooding hazard parameters ± wave run-up, total water level and overwash. This implies definition and selection of extreme events and generation of storm surge and wave time series for each coastal sector.
Extreme events were determined using wave hindcast data obtained by means of nested WAM-SWAN wave model train with maximum resolution of 400 m and covering 57-year period   [23,27]. The wave models were forced with wind data originated from regional atmospheric model REMO [28]. Selected grid points for extraction of wave climate time series are shown in Figure 1. Storm surge data covering the same time-span consist of tide gauge daily measurements at bay conditions (Varna Bay) and at open shore.
Peak over threshold (POT) analysis was used to identify the individual coastal storms using threshold of 2 m for significant wave height, which should be exceeded for a time-span longer than 18 hours. The threshold for significant wave height was selected since it represents the 99 percentile, while duration was set on the basis of measured morphological impact of short-lived storms of equal intensity at two beaches within the study area. The impact was considered important in terms of thresholds established in [29]. Thus, 144 storm events were selected, each represented by surge level, significant wave height, peak wave period, mean wave direction and storm duration.

Hazard intensity
Once the forcing along the coast has been defined, resultant hazard intensities were assessed using empirical models applicable to each type of coast presented at the study site. Herein, the TWL, represented as a superposition of storm surge level and wave-induced runup, was the chief property considered for calculation of coastal flooding hazard. In case of sheltered areas such as ports and canals only surge level component was taken into account. Three empirical models were considered for calculation of wave run-up, namely Stockdon model  Data analysis showed that Stockdon model underestimated the wave run-up for the entire range of slopes ( Figure 2). The N & H model underpredicted considerably the TWL for mild slopes (WDQȕ < 0.15), while for larger slopes the opposite was valid. Holman model best coped with reproduction of observed TWL for mild slopes, but also tended to increasingly overestimate it as the beach slope grows. Obviously, the model results are problematic for reflective conditions but, on the other hand, there was a very small difference between estimates obtained with Holman and N & H models for those profiles. Based on this analysis, Holman model was preferred for application on 105 beach profiles with beach face slopes ranging between 0.01 and 0.26.
Holman model predicts the run-up magnitude Ru 2% as [31]: where H s is significant wave height in deep water and ߦ is Iribarren number. For artificial and rocky slopes, EurOtop formulation was employed [33]: where ߛ is berm influence factor, ߛ -surface roughness factor and ߛ ఉ -influence factor of wave direction. EurOtop model was applied on 54 rocky and artificial profiles with slopes ranging between 0.04 and 0.56.
The TWLs for all storm events that satisfied the POT criteria were calculated for each coastal sector. Resulting values were subjected to Extreme Value Analysis to establish the best fitting distribution to be Generalized Extreme Value (GEV) distribution ( Figure 3). Hazard intensities relevant to return periods (T r ) of 20, 50 and 100 years were further analyzed since they were considered benchmark values by most of the regional stakeholders [34]. Therefore, hazard indicator (i h ) was calculated for above mentioned return periods and for both average and worst scenarios along Varna regional coast. Herein, map representation and hotspot identification were done based on the worst scenario with T r = 100 years.

Hazard extent
The hazard extents were determined using two approaches depending on the slope of the hinterland areas. In case of continuously rising slopes bathtub approach was DSSOLHG ³%DWKWXEELQJ´ RU HTXLOLEULXP flood mapping) is a simplified method that is often utilized to re-draw coastal flood zones. It is essentially horizontal spreading of flood elevation data to areas of lower elevation [35]. For low-laying hinterland with dune presence, flood extent was calculated following [36]. If water level does not reach dune crest elevation, flood extent is the horizontal projection of the TWL, equal to TWL/WDQȕ. When water level exceeds the dune crest elevation, the extent is a sum of the horizontal distance from still water level to the dune crest, and the dune overwash extent, calculated by [37]: where h(x) is flow depth on the back barrier slope as a function of distance x, h c -flood depth over the beach crest, u c -flow velocity at the dune crest, and a is proportionality constant for infiltration.

Exposure indicators
Five types of exposure indicators are used to measure the relative exposure of different receptors to coastal flooding, namely: Land Use; Population; Transport systems; Utilities; Business settings [14]. Each of the five resultant indicators are ranked from 1 to 5: non-existent or very low (1), low (2), moderate (3), high (4) and very KLJK UHIOHFWLQJ WKH GHJUHH RI HDFK UHFHSWRU ¶V H[SRVXUH to flooding hazard.
The Land Use exposure indicator (i exp-LU ) reflects two components for each coastal sector: the exposed surface and an associated importance value for each selected land use. Calculation of i exp-LU within the inundated area of each coastal sector is done by the formula: where n is number of land use classes; S is surface in m 2 of each land use class; V is importance (weight) value assigned to each land use class (from 1 to 10). The final output is summation of all the areas per inundated zone in a coastal sector. For the present study the available land use classes were identified on the basis of Bulgarian Law on spatial planning [38]. The Population exposure indicator (i exp-POP ) measures the relative exposure of different communities along the coast. As such, the indicator is similar to Social Vulnerability Index [39], which assesses the relative vulnerability of different areas to long-term health and financial recovery from an event, i.e. it considers the socio-economic characteristics of the areas exposed to certain hazards. In the present study the Social Flood Vulnerability Indicator (SFVI) [40] was adopted in FRPELQDWLRQ ZLWK UHVXOWV IURP WKH SURMHFW ³5HJLRQDO profiles ± LQGLFDWRUV RI GHYHORSPHQW´ [41]. This approach was preferred because it was established that exact or closely similar to SFVI indicator of social vulnerability is not yet created for Bulgaria. The SFVI for Varna regional coast was adopted following [15] lowest available level of disaggregation ± municipality. Data on variables were transformed into percentages of the total population of each municipality, analyzed that they are not normally distributed, hence, subjected to additional normalization [40], and finally standardized as Z scores. Thus processed variables were put into a general equation to calculate the SVI for each municipality [15]: where W j is weight of each variable, n ± number of variables and C j ± average of the variables. Information about the following three indicators was collected by means of desktop research, field surveys and stakeholders interaction in order to map and to value each element of the systems.

Value
Rank Description Presence of transport network with local/regional importance 4 High High density and multiple networks (train, road, airport) of local importance or regional importance 5 Very high High density and multiple networks (train, road, airport) of National or International importance Table 1.

Ranking values for i exp-TS
To construct the indicator for Transport systems (i exp-TS ) for each coastal sector a 5-step approach proposed in [15] was applied. Data on terrestrial transport networks (roads and railroads) location and relative importance (capacity and use) of their assets (links and nodes) was gathered. To assess the exposure of transport systems within the inundated area of each coastal sector road classification of Bulgaria [42] was taken into account and following the rules presented in Table 1 the exposure indicator i exp-TS was derived.

Value
Rank Description Presence of utilities networks/assets with local/regional importance 4 High High dense and multiple utilities networks/assets of local importance or regional importance 5 Very high high dense and multiple utilities networks/assets of national or international importance Table 2. Ranking values for i exp-UT The indicator for Utilities (i exp-UT ) for each coastal sector was obtained according to methodology proposed in [15]. Information on Utilities providing essential services, e.g. water, electricity, telecom networks and emergency centers, was collected. The assessment was done following the rules presented in Table 2, which led to derivation of i exp-UT .

Value
Rank Description Moderate Local or regional economic activities 4 High Regional importance 5 Very high National or international importance Table 3. Ranking values for i exp-BS Exposure indicator for Business settings (i exp-BS ) was derived for each coastal sector following a 6-step approach proposed in [15]. Information on different business settings, including location of assets and their relative importance (input, output and number of businesses) was collected. Taking into account the gathered materials and following the rules of Table 3 the exposure indicator i exp-BS was obtained.

Coastal index
The Coastal Index (CI) combines hazard intensities and related exposure vulnerability of selected receptors, thus, permitting evaluation of potential flood risk, which coastal sectors are subjected to. It was calculated using the formula [14]: with i h standing for hazard indicator and i exp ± overall exposure indicator. The resulting values were ranked from 1 to 5.

Flooding hazard evaluation
The coastal flooding hazard indicator (i h ) was calculated using the TWL for each sector. However, due to the wide variety of existing morphological settings, this hazard parameter cannot be considered sufficient to obtain comparable results on its own. Hence, the estimation included the areas exposed to inundation regardless of land use classes. They were categorized in ten sets and each set was assigned a weight value, ranging from 0.05 to 1. Then, a parameter Weighted TWL, merging hazard intensities and thus categorized hazard extents, was formulated. Accordingly, larger weights were assigned to those values of TWL that inundated broader hinterland areas. Subsequently, those values were ranked using the method of maximum breaks [43] in attempt to consider individual data values and group those that are similar. To this end, data sample was ordered from low to high, the differences between adjacent values were computed, and the largest of these differences served as class breaks. The ranking intervals are listed in Table 4.   Validation approach represents a comparison of measured flood extent and the reconstructed one corresponding to the TWL obtained in compliance with herein presented methodology. Besides, taking into account EVA results, return periods were assigned to the flooding hazard (for average and worst scenarios). Thus, comparison between real and estimated inundation was possible, testing the ability of CRAF to predict hazard intensities and correctly estimate the hazard indicator. Since the hazard intensity of 2012 storm comes towards CRAF T r = 20 event, presented results could be considered as validation of hazard indicator for that specific return period ( Figure 5).

Validation of flooding hazard
As it can be seen, in case of continuously rising slopes (Varna central beach), average and worst scenarios give quite similar extents that shows an excellent agreement with measurements. On the other hand, for low-laying hinterland (Asparukhovo beach), measurements show almost perfect concurrence with the worst scenario overwash extent, while for the average scenario it is halfway narrower than the measured one, which is evidently due to underestimation. On that account, indicators relevant for the worst scenario are illustrated in this paper.

Exposure evaluation
To calculate the Land Use exposure indicator i exp-LU the available land use classes [38] were merged to form more common classes in order to facilitate the valuation process. The assignment of importance value (1 to 10) to each class was done according to information gathered GXULQJ VWDNHKROGHUV ¶ LQWHUYLHZV 7KH JHQHUDOL]HG ODQG XVH classes and assigned importance values are presented in Table 5.

ID code
Land use classes Weight value %  Table 5. Types of merged land use classes, associated weight values and relative proportion (%) within inundated zones, worst scenario, T r = 100 The exposure indicator i exp-LU was calculated by equation (4) for each coastal sector within the inundated zones and ranked from 1 to 5 by data classification method ± maximum breaks (Slocum, 1999). Table 5 also presents the relative proportion of land use classes for worst scenario, T r = 100.
As stated previously the exposure indicator for population i exp-POP was formed as a combination of the tailored SVI and results from [41]. Firstly, the SVI was estimated for the five coastal municipalities. Then, its values were reclassified in a scale from 1 to 5. The results showed that the most vulnerable are the municipalities of Avren and Byala having ranks of 5 and 3, respectively. The low number of initial data suggests that they are not normally distributed and despite the transformation methods applied the results lacked representativeness since calculations were not performed for all 265 municipalities in Bulgaria. Supplementary information about socio-economic development of the entire Varna district in comparison to other districts in Bulgaria is presented in [41]. That research provides a snapshot of the socio-economic conditions in Bulgarian districts as of mid-2015, as well as their development since 2000. All 28 districts are clustered into 8 types and classified from µYHU\ JRRG ¶ WR µYHU\ SRRU ¶ VRFLR-economic condition. According to this report, Varna district falls into the class RI µJRRG ¶ VRFLR-economic conditions. This additional information allowed for a new reclassification of the SVI YDOXHV LQ D VFDOH IURP WR µYHU\ ORZ ¶ µORZ ¶ DQG µPHGLXP ¶ ZLWK UDQNLQJ LQWHUYDOV GHWermined by the method of equal intervals. Based on the new ranking Varna region coastal municipalities were assigned the following ranks representing the social vulnerability of the population: Aksakovo ± 1, Varna ± 1, Avren ± 3, Dolni Chiflik ± 1 and Byala ± 2. Unfortunately, this approach does not allow assignment of a specific value for each coastal sector; therefore, a single value for all sectors within a given municipality was used. The proposed ranking suggests difficulty in claiming that the available data are representative for the exposed population along the coastline, particularly for sparsely populated coastal areas in municipalities of Aksakovo and Avren.
For qualitative assessment of Transport systems and derivation of i exp-TS road classification of Bulgaria [42] was taken into account and different road/railroad levels were ranked according to the rules in Table 1. Ranking values vary between 1 (local roads) and 5 (national high ways and railroads).
To assess Utilities the necessary information on location of assets for telecommunication and power supply within coastal hinterland (if any) were gathered during several field surveys and their presence (exposure to hazard) were taken into account during ranking. The same approach was used with respect to utilities in port and industry zones. Based on expert judgment of company managing the water networks (supply and sewage systems) in Varna district none of major assets are vulnerable to coastal flooding, since the water supply and sewage pipes are deeply underground and assets like water sources or pumping stations are located far from the coastline. Nevertheless, the presence of drainage pipes for ravine waters within the beaches was considered. As for emergency no large hospitals or medical complexes are located in the close vicinity to coastline, but field surveys localized the presence of small medical centers situated within beach domains and that also was taken into account during the ranking.
Based on [15] two business settings were identified within study site domain: 1) Beach frontage urban area (in the city of Varna) and seaside resort and 2) Port and related commercial and industrial zones. Within the first category fall the international resorts Euxinograd, St. Constantine & St. Helena, Sunny day Co., Holiday Club Riviera and Golden sands located to the north of Varna city, as well as complexes Kamchiya and Long Beach (Shkorpilovtsi) situated to the south of cape Galata. Within the second category fall the industrial zone and the port complex situated deep into Varna Bay.
Due to the wide variety of existing coastal settings, location of different business assets and the fact that the inundation extents with few exceptions are predominantly within beach areas, the assignment of values to i exp-BS ZLWKLQ HDFK FRDVWDO VHFWRU ZDV GRQH DFFRUGLQJ WR DXWKRUV ¶ best judgment based on the information on different businesses presence gathered during several field surveys, desktop research and the rules in Table 3.
Finally, the ranks of all five exposure indicator were combined via equation (6) to form the overall exposure indicator i exp . In order to classify the resulting data values the equal ranking intervals were used (Table 6). Very high - Table 7. Ranking of the overall exposure indicator i exp Obviously, ranking does not exceed the value of 4 and even so only 8 sectors are ranked as µKigh ¶ 7hey are concentrated mostly within the inner curve of Varna Bay comprising port and industry facilities with regional, national and international significance.

Coastal Index evaluation
Coastal index values obtained by equation (7) were ranked following the method of equal intervals. The ranking intervals and assigned ranks are presented in Table 8. 3.0 < 5 Table 8. Coastal Index ranks and ranking intervals Colour mapping of the Coastal Indices along Varna regional coast is presented in Figure 6.