The study domain is in the north-east of the Iberian Peninsula and consists of the coastal zone along Catalonia and the river basins flowing into it, which are composed of the internal river basins of Catalonia and the Ebro lower river basin (Fig. 1). The coastline runs in the SE–NE direction and is bounded by the presence of two parallel mountain ranges located close to the sea: the littoral range (maximum altitude around 600 ma.s.l., metres above sea level) and the pre-littoral range (maximum altitude around 1800 ma.s.l.). The northern part of the region is limited by the Pyrenees, running from west to east, with altitudes greater than 2000 ma.s.l. Therefore, the region is prone to the development of flash floods and thunderstorms (Llasat et al., 2014b), both from a hydrological point of view (existence of many small torrential catchments) and from a meteorological point of view (i.e. orographic forcing of Mediterranean air masses) (Llasat and Puigcerver, 1992). In fact, the impact of mountains on the low-level wind circulation usually triggers convective instability and affects the pressure fields (Jansà et al., 2014).

The Catalan coastline extends about 600 km, of which ∼ 280 km corresponds to sedimentary coasts. The combination of the decrease in river sediment supplies, current level of urbanization and infrastructure development, and the natural littoral dynamics has led to an overall shoreline erosion during the last few decades (Jiménez and Valdemoro, 2019). From the perspective of coastal storms, the area is subjected to dominant NE–E extreme waves as well as secondary impacts from the S–SE (Mendoza and Jiménez, 2009; Bolaños et al., 2009; Mendoza et al., 2011). The NW Mediterranean is a microtidal environment with an astronomical tidal range of about 0.25 m. Meteorological tides are of low amplitude, reaching maximum recorded values of up to 0.5 m during favourable conditions (under low atmospheric pressure centres and landward-blowing winds), in such a way that they are much lower than the wave-induced Ru during coastal storms (Mendoza and Jiménez, 2009). The order of magnitude of the storm-induced coastal hazards in the study area can be seen in Mendoza and Jiménez (2009) and Bosom and Jiménez (2011). Although the coastal storm intensity has not significantly changed (e.g. Casas-Prat and Sierra, 2010), the wave action on a progressively narrowing coastline has resulted in a significant increase in coastal damages during the last few decades (Jiménez et al., 2012).

To perform an integrated study of the terrestrial (rainfall) and “coastal” (waves) compound events, the study region was divided into seven areas following previous studies on flash floods (e.g. Llasat et al., 2016), dividing the region into its main groups of natural catchments along the coast (Fig. 1). Each area is composed of several river catchments and/or groups of torrential catchments flowing to their corresponding coastal stretch.

Three main climatic datasets were used in this work to characterize the rainfall, coastal storms, and weather conditions. Rainfall was characterized using daily rainfall (P24h) data obtained from the Spanish Meteorological Agency (AEMET) database (Ramis et al., 2013), which includes records from 491 automatic weather stations (AWSs) in Catalonia (Fig. 2) covering (non-homogeneously) the period 1950–2015. The selection criteria for identifying records to be used in this analysis consisted of identifying those AWSs belonging to catchments in coastal regions with a homogeneous coverage of the 41-year period from 1973 to 2013, resulting in 69 case-study rain gauges (Fig. 2 and Table 1). Flood impacts were obtained from the INUNGAMA database, which contains all the flood events that have affected Catalonia since 1981 as well as all the catastrophic events since 1900 (Barnolas and Llasat, 2007; Llasat et al., 2016).

The wave data used were obtained from the hindcast Downscaled Ocean Waves (DOW) dataset (Camus et al., 2013), which was derived from the Global Ocean Waves dataset (Reguero et al., 2012). Data consisted of hourly values of hindcast wave conditions characterized by the significant wave height (Hs), wave period, and mean wave direction covering the same period as the rainfall data (1973–2013). The datasets were retrieved for 19 nodes located nearshore (about 20 m water depth), homogeneously covering the analysed basins along the coast (Fig. 1 and Table 1) to properly capture regional variations in the storm climates. Additionally, wave records during the Gloria storm in January 2020 were obtained from the SIMAR database from Puertos del Estado (http://www.puertos.es/es-es/oceanografia, last access: June 2020).

Weather conditions were characterized by geopotential fields at 1000 hPa from the US National Center for Environmental Protection (NCEP) datasets. NCEP/NCAR Reanalysis I (1948–present) and NCEP/DOE Reanalysis II (1979–present) generated by the National Oceanic and Atmospheric Administration (NOAA) were used. In the first case, NCEP considered the same climate model that was initialized with different types of weather sources (Kalnay et al., 1996). The second version of the first reanalysis considers the starting point of the major satellite era, which implies that more observations included fewer errors in the resulting fields (Kanamitsu et al., 2002). To retrieve weather data from the NCEP/NCAR Reanalysis datasets, we used the RNCEP library of R-Cran (Kemp et al., 2012). The fields were collected for a spatial extent covering longitudes -25 to 29∘ and latitudes 30 to 64∘, with a spatial resolution of 2.5∘ × 2.5∘ (N = 13 × 15 grid) covering the period 1973–2013 with a temporal resolution of 6 h.

The general methodological framework adopted in this study consisted of the following steps (Fig. 3): i.

Compound events are identified. First, individual heavy rainfall and coastal storm episodes are identified at all rain gauges and coastal nodes. Then, compound events are defined by identifying dates upon which the two considered drivers co-occur along the territory. Extreme events with only “pure” coastal storms without rain or pure rain episodes without waves are discarded. Each compound event is characterized in terms of a representative Hs (the maximum value reached during the event) and daily precipitation (P24h) per coastal basin.

The first step consisted of identifying individual extreme events from selected rain and wave datasets (Fig. 1). This was done by applying the peak over threshold (POT) method to daily rainfall (P24h, in mm) and to significant wave height (Hs) as representative climate variables of the compounding drivers: heavy rainfall and coastal storms.

Following previous studies in the area (Barbería et al., 2014; Cortès et al., 2018), potential extreme rainfall episodes were identified using a P24h of 40 mm and a 3 d interval between consecutive events to identify independent episodes. Although the daily threshold is below the value used by some international projects such as MEDEX (Jansà et al., 2014 proposed 60 mm), local flash floods and urban floods can be produced when this precipitation falls within less than 2 h. The event is then characterized by the maximum recorded P24h value. A second threshold of 100 mm was used to flag severe extreme events to later differentiate them during the characterization assessments. This threshold has previously been used in the region when studying extreme rainfall associated with large riverine flooding (Gilabert and Llasat, 2018).

Coastal storms were identified using a double-threshold POT (see Sanuy et al., 2019). The 98th percentile of the Hs time series is used as the first filter to locate storm start and end times, which roughly correspond to Hs = 2 m, in agreement with Mendoza et al. (2011) for NW Mediterranean conditions. Then, an upper threshold given by the 99.5th percentile is applied to retain only extreme storms. This criterion results in class III storms according to the Mendoza et al. (2011) classification, which corresponds to the minimum required conditions to produce significant impacts on the coast (i.e. erosion and inundation; see Mendoza and Jiménez, 2009). In addition, a 3 d interval of fair-weather conditions between consecutive events was used to identify independent episodes.

The result of this step is a collection of heavy rainfall and wave storm individual episodes for all AWSs and wave nodes. Each episode is characterized by an initial and final date and the maximum values of P24h and Hs for rainfall and waves, respectively.

The second step consisted of identifying and characterizing compound events in each area along the coast (Fig. 2). First, we established the occurrence of a coastal (wave) storm event by comparing the storm initial time at all coastal nodes within a 24 h window. All storms within such a window were considered a single event, which is labelled with the same date, and they should correspond to a coastal storm propagating along the territory. Then, all rain gauges were surveyed to identify the occurrence of heavy rainfall episodes during the identified coastal (wave) storm and in the 3 d before. If no heavy rainfall is recorded at any rain gauge, the event is removed from the analysis, as it would correspond to a pure wave storm. On the contrary, if any station records an extreme P24h episode, the event is flagged as a compound event, with its date of occurrence being given by the earliest starting time of coastal storms at any node. Finally, each compound event is characterized by the maximum P24h and Hs values at the stations and nodes within each coastal area and during the event duration.

As a result of this process, a compound event is identified herein when a heavy rainfall episode at any station along the coast occurs simultaneously, or in rapid succession (within a 3 d interval), with extreme waves at any location along the coast. Therefore, an identified compound event at a given time may present different characteristics along the costal basins (hereafter also named areas or sectors): areas in which rainfall and wave extreme events simultaneously co-occur, areas with only one extreme component (either rain or waves), and areas without any extreme episodes. Then, each compound event is characterized along the coast (in each sector) as follows: (i) multivariate (simultaneous rainfall and wave episodes); (ii) spatially compounding (SC) rain, where local extreme conditions correspond to rainfall; and (iii) spatially compounding (SC) waves, where local extreme conditions correspond to storm waves. The classification intends to classify the event as it is experienced in each basin. Thus, in the face of a compound event (regional scale), there will be basins that experience it as multivariate (both components co-occur) and basins that experience it as spatially compounding. In the second case, the basin may be receiving only rain (SC-rain) or only waves (SC-waves). According to our definition of a compound event, there will always be a co-occurrence of the two components at the regional scale.

To investigate the correlation between the magnitude of both components of the compound event, P24h and Hs, across all areas, we used the Spearman ρ coefficient (e.g. Genest and Favre, 2007); this is defined as the Pearson correlation coefficient between the variable ranks (Eq. 1). The correlation of the wave magnitude (Hs) at each coastal area with the P24h rainfall at each of the nine areas is calculated as follows: 1ρ=cov(rgHs,rgP24h)σrgHsσrgP24h, where rgHs and rgP24h are the ranks of Hs and P24h, respectively, cov is the covariance, and σ is the standard deviation.

During the development of a compound event, different weather types can be present; in this study, we therefore use corresponding pressure fields (geopotential height at 1000 hPa) at the closest time to the date–time assigned to each event (Sect. 3.3). Weather conditions were then classified by applying a correlation-based method (Yarnal, 1993; Wu et al., 2018) to the 1000 hPa geopotential height field. The method consists in obtaining map patterns using the Pearson product-moment correlation (rxy, Eq. 2) to depict the degree of similarity of spatial structures between pairs of gridded data (i.e. the map typing focuses on the positions of high- and low-pressure centres, rather than their magnitudes). All maps were extracted at the closest time to the beginning of the compound event, as defined in Sect. 3.4 (i.e. the beginning of the coastal storm).

First, all maps are normalized via Zi = (zi-z¯)/σz, where Zi represents the number of positive or negative standard deviations from the mean at each grid cell i, zi is the original value at grid point i, and z¯ and σz are the mean and standard deviation of the N = 13 × 15 grid point values. Once normalized, each map is compared with all other maps using 2rxy=∑i=1Nxi-x¯yi-y¯∑i=1Nxi-x¯2yi-y¯2, where xi and yi represent the normalized value at each of the N points of the pair of maps being compared, and x¯ and y¯ are the corresponding means across the N-point grids. A pair of maps is considered similar when rxy ≥ rt, where rt is a correlation threshold. Different sources of subjectivity exist in choosing the value of rt (Yarnal et al., 1993), which usually depends on a balance between the number of identified patterns and the number of dates (events) remaining without classification.

The process was applied as described by Wu et al. (2018). The first date of the reference is used as the key day, and all maps are compared to create the first class. Then, all classified dates are removed, and another date from the non-classified pool is used as the second key day. A value of rt = 0.2 was used in this step, which led to four weather-type candidates. Then, a second comparison between each individual map and the average of each group candidate was performed to ensure that maps that are similar to more than one type are classified into the group corresponding to a maximum rxy. This led to a final three-group classification with a mean rxy of 0.64–0.7 per group.

BNs are statistical tools based on acyclic graph theory and Bayes' theorem (Pearl, 1988; Jensen, 1996) and have demonstrated their versatility and utility in efficiently combining multiple variables to predict or characterize system behaviour (e.g. Gutierrez et al., 2011; Plant et al., 2016; Beuzen et al., 2018). The BN is used here to assess the probabilistic relationship between each type of meteorological forcing (Sect. 3.5) and their associated effects (multivariate or spatially compounding) and driver intensities (Hs and P24h) at each of the different basins.

Figure 4 shows the BN structure, the considered variables, and the variable discretization. The arrows depict the parent–child relationships; i.e. the results will be presented in terms of the probability of given values of Hs, P24h, and impact type conditioned to the different possible combinations of the synoptic case and area. The training dataset consisted of 1260 variable combinations resulting from the previous assessment of 140 compound events in nine different areas.

During the analysed 1973–2013 period, 225 coastal storms and 605 heavy rainfall episodes affecting at least one of the considered areas were identified. It must be considered that the real number of rainfall events would be higher, since convective localized episodes are under-represented in the dataset used. From this total, 140 episodes can be classified as compound events in which wave storms are accompanied by heavy rainfall in any area along the coast. This means that 62 % of coastal storms and 23 % of heavy rainfall episodes can be labelled as compound events and that the probability of having such compounding conditions is larger under coastal (wave) storms. The average frequency of occurrence along the Catalan coast during the study period was 3.4 compound events per year, without presenting any statistically significant trend during the 41 years analysed (Fig. 5).

Figure 6 and Table 2 show the spatial distribution of the 140 identified compound events along the Catalan coast according to their typology as locally recorded: multivariate (simultaneous rainfall and wave storm episodes in the same area), SC-waves (a solo wave storm episode in a given area with simultaneous heavy rainfall co-occurring in a different area), and SC-rain (a solo rainfall episode in a given area with a simultaneous wave storm co-occurring in a different area). In addition, the local absence of extreme conditions for both drivers was retained.

The results show that, in the presence of a compound event along the Catalan coast, areas with the highest probability of experiencing a multivariate event are in the northernmost part, Girona N, and the Lower Ter–Tordera basins (with an occurrence frequency of about 55 %) followed by the southernmost end in the lower Ebro basin and delta (with an occurrence frequency of about 40 %). On the other hand, although the areas located at the central part of the coast show a non-negligible probability (about 20 %–30 % of recorded events) of experiencing multivariate events, they are dominated by the presence of spatially compound events with the local presence of wave storms (about 40 %–50 % of recorded events). These results would indicate that, in the study area, when a regional compound event occurs, wave storms are the “spatially dominant” driver, with all areas along the coast having a probability greater than 60 % of having local wave storms (either multivariate or SC-waves). On the other hand, areas presenting a high probability (> 60 %) of having rainfall extremes (either multivariate or SC-rain) during regional compound events are restricted to the two northernmost areas and the southernmost one (58 %); the central part of the coast (Areas 3 to 6) presents relatively low probabilities of experiencing extreme rainfall (35 %–42 %). All percentages given are relative to the total number of identified compound events.

Rainfall events in the terrestrial areas which correspond to the Ter and Llobregat upper basins (areas 2b and 4b) are filtered with P24h > 100 mm to assess their potential effects at the coastal fringe; fewer than 5 % of cases reach those precipitation levels in combination with extreme waves at the coast (Table 2).

Once the probability of occurrence of compound event was analysed at different areas along the coast, the correlations among the magnitude of the climatic drivers (rainfall and waves) needed to be determined. Figure 7 shows the computed Spearman ρ coefficient by correlating the waves (Hs) in a given area (marked with a diamond in the figure) with rainfall (P24) at all areas along the coast.

The results show that, in general, the correlation between the intensity of local wave storms and rainfall across the territory (measured as peak values of Hs and P24h during the event) decreases from north to south, following the observed trend in the dominance of multivariate events. The highest correlation value (ρ = ∼ 0.65) was obtained for multivariate events in the northernmost area (Fig. 7), suggesting a strong link between locally simultaneous wave storms and rainfall. The connection between these drivers extends southwards in such a way that the correlation between the wave storm intensity in Area 1 with simultaneous rainfall episodes in the adjacent Area 2 is of the same order of magnitude. The correlation progressively decreases southward as we compare it with the rainfall recorded in the central basins, but, in all cases, the correlation is statistically different from zero. Notably, a value of ρ > 0.4 is obtained in the southernmost sector, where the presence of multivariate events is higher than in the central basins.

When the intensity of wave storms recorded in Area 2 is correlated with rainfall across the territory during compound events (Fig. 7), a similar behaviour than that in Area 1 is observed, although with lower correlation values. As we progressively move to the south, the correlation between the intensity of the local wave storms and P24h at any area consistently decreases to very low values or, directly, they are statistically uncorrelated. On a regional scale, the central basins have the lowest values of ρ, suggesting an independence of storm waves and intense rainfall events.

Weather conditions during the 140 identified compound events were classified in three different synoptic types (Fig. 8). Synoptic type 1 conditions prevail during 42.1 % of cases and are characterized by the presence of lower pressures north-west of the Iberian Peninsula over the Atlantic Sea and higher pressures in the central Mediterranean. The deep low in the north-western part of the Iberian Peninsula and the strong anticyclone over central Europe favour a strong pressure gradient and consequently induce intense winds from the south (Llasat, 1987). This type of situation usually creates a mesoscale structure when the deep low and strong anticyclone impinge over the Pyrenees range, known as an orographic dipole, with a mesoscale high over Catalonia that modifies the synoptic pressure field and creates an eastern component of the wind that favours the entrance of warm and wet air. At the same time, the mountain range triggers potential instability and develops convective systems and heavy rainfall (Trapero et al., 2013; Llasat et al., 2014b).

Synoptic type 2 (18.6 % of cases) is characterized by the presence of a depression in the south-eastern Iberian Peninsula and an anticyclone in the north-west, which creates a strong pressure gradient and N–NE winds. This kind of meteorological situation is more effective in generating sea storms than it is in generating heavy rainfall. Synoptic type 3 conditions (39.3 % of cases) are similar to those of type 2, with a deep low in the southern Iberian Peninsula and an anticyclone to the north-east. The main difference is that this anticyclone is placed over the centre of Europe in this type; this pattern gives rise to a strong E–SE wind at low levels.

The BN was used to calculate the probability of occurrence of multivariate and spatially compounding (wave or rainfall) events in different areas given the different synoptic types (Fig. 9). The BN was also used to calculate the probability of exceedance of significant thresholds of Hs and P24h for each type of compound event at the different areas along the coast to assess the intensity of each contributing component (Fig. 10).

Under a meteorological forcing generated by synoptic type 1, which is the most likely to occur, the probability of occurrence of multivariate events anywhere in the territory is the lowest compared to that associated with the other types (Fig. 9). Moreover, when multivariate events occur, they are concentrated in the northernmost part of the coast (areas 1 and 2). In the rest of the coast, the dominant event is SC-waves. Generated wave storms present the smallest Hs values along the territory, without inducing extreme storms (Hs > 5.5 m) (Fig. 10). On the other hand, the probability of SC-rain events is higher than in the other types, especially in the Ebro basin. Moreover, it is the most likely to exceed a P24h of 100 mm.

Synoptic type 2 episodes are more prone to create Levante (winds blowing from the east) situations, during which the probability of occurrence of multivariate events increases, especially in the northern and southern areas, while in the central part of the coast, multivariate and SC-waves are equally probable (Fig. 9). The severity of coastal storms (Hs) increases, especially at the northern and southern ends, where extreme storms (Hs > 5.5 m) have been recorded (Fig. 10). The probability of rainfall episodes with P24h > 40 mm increases in the same areas with respect to type 1, whereas the frequency of the most intense episodes (P24h > 100 mm) decreases, except for Area 1 (Fig. 10).

Synoptic type 3 is similar to type 2 in that it also represents Levante situations but is characterized by the marked south–north pressure gradient, which results in significant windstorms and strong waves. Consequently, they present similar overall probabilities of occurrence of multivariate events along the coast, with type 3 presenting a larger probability of multivariate occurrence at the northern and southern extremes and a higher frequency of SC-waves, especially in the central areas (Fig. 9). In terms of intensity, wave storms recorded under synoptic type 3 present the highest probability of exceeding Hs > 3.5 m in all areas and the highest probability of extreme waves (Hs > 5.5 m), which is restricted to the two northernmost areas (Fig. 10). On the other hand, the distribution of probability of P24h exceeding 40 or 100 mm is very similar to type 2, since in both cases there is an incidence of a warm and humid air mass from the east.

Differences in weather patterns result in events with different characteristics and, consequently, impacts throughout the territory. To put the potential consequences of these events in the context of risk management, the impact of selected events recorded in the study area under the different synoptic types is illustrated with information gathered from after-event press coverage and the INUNGAMA and PRESSGAMA databases (Llasat et al., 2009, 2014a; Jiménez et al., 2012). Synoptic conditions during each analysed event are shown in Figs. 11 and 12, with maps being extracted following the criteria described in the methodological framework. Table 3 shows the maximum Hs and P24h recorded in each area along the coast during each event. Two representative events were chosen for types 1 and 3, as they occur twice as frequently as type 2 events do, which are only represented here by one event.

Between 6 and 8 November 1982, a compound event generated under a type 1 synoptic situation (Fig. 11) took place along the Catalan coast. From a meteorological point of view, the event was dynamically forced, as it unfolded in the prefrontal and frontal zones of a strong Atlantic baroclinic storm, although the Pyrenees played a relevant role by triggering deep convection. The largest contribution of humidity was from the Atlantic (mainly tropical and subtropical regions but also from the north), with relevant additional input from the western Mediterranean (Insua-Costa et al., 2019). This was a very extensive episode of heavy rain affecting Portugal, Spain, Andorra, and France. Catastrophic flash floods and landslides occurred in the Upper Llobregat basin (Area 4.b) and Upper Ter basin (Area 2.b) (Puigdefàbregas, 1983; Llasat, 1987), where nearly 342 and 556 mm were recorded in less than 24 and 72 h, respectively. The Llobregat River (Areas 4.a and 4.b) recorded a peak flow of 1600 m3s-1 near its mouth when its average discharge was 328 m3s-1.

The main peak rainfall and wave conditions recorded during the event along the Catalan coast are shown in Table 3. As can be seen, although waves exceeded storm threshold conditions along the entire coast, their values were relatively low, with only the northernmost sector presenting severe storm conditions according to the Mendoza et al. (2011) classification. Due to this, coastal-storm-induced damages were relatively low and were limited to some stretches at Costa Brava (Areas 1 and 2.a), where waves induced minor damage to some marina facilities and caused overtopping at some beach waterfronts. Some beaches in the Maresme region (Area 3) were also affected, with extensive erosion and overtopped promenades. On the other hand, the rainfall-induced damage was extensive and very important, with 14 casualties and EUR 1033 million (adjusted for 2020) of private flood damages paid by the Insurance Compensation Consortium (the Spanish public re-insurance company, CCS) as a consequence of the floods in Catalonia (throughout the entire territory and not only in coastal areas). In summary, although this was a multivariate event at many basins, the most important and relevant damage was caused by rainfall-induced floods (Fig. 12).

Another significant type 1 event occurred in 2011, starting on 2 November and lasting until 7 November 2011 (Fig. 11). It mainly affected Catalonia (Spain) and Liguria (Italy). In the first region, the maximum cumulative rainfall was 326 mm (close to Area 2.b, Table 3), while the maximum in 24 h was 203 mm (close to Area 4.b, Table 3). It produced a flood in the Muga River (Area 1), with a peak discharge of 378 m3s-1 near the mouth (on 1 November the flow was 0.7 m3s-1) (Llasat et al., 2014a). Between 2 and 8 November, the CCS paid EUR 2.1 million (adjusted for 2020) for damages produced by the sea storm and EUR 458.8 million (adjusted for 2020) for damages produced by floods in insured assets. Meteorological features showed the presence of a trough at 500 hPa associated with a synoptic frontal wave that evolved into a mesoscale depression along the Catalan coast on 6 November. This situation favoured the entrance of very warm and wet air from the south-east over Catalonia and humidity advection from the Atlantic.

It should be noted that both events presented similar characteristics, with the most important contribution to damage being induced by rainfall. On the other hand, combined/compound effects were scarcely reported in just few areas, where they locally induce a moderate-to-high impact.

An historical compound event generated under type 2 conditions in the area occurred in November 2001 (Fig. 12), when a thermal orographic low over the African plateau interacted with an upper-level trough and developed a strong cyclone that moved toward the Mediterranean Sea over Algiers following a northward trajectory, creating torrential rainfalls and floods in Algiers (more than 700 deaths). When the cyclone reached the sea, the low reached its mature state (Fita et al., 2006; Genovés and Jansà, 2002). The depression was observed at all atmospheric levels, and there was also a zone of high pressure that was located to the north-west of the Peninsula with a pressure higher than 1035 hPa that contributed to the strong wind and pressure gradients. The deep low continued its trajectory to the NE and affected Catalonia, giving rise to a windstorm with a recorded maximum wind speed of 170 kmh-1 (Port-Bou, Area 1). In contrast to the previously described type 1 situation, this event was characterized by very high Hs values along the entire coast (Table 3), exceeding the threshold for severe storms according the Mendoza et al. (2011) storm classification in nearly all areas (except Area 5) and for extreme storms in the northernmost areas (1 and 2.a). Thus, compounding conditions were wave-dominated, with the rainfall being moderate (although exceeding the 40 mm threshold for P24h) except in the northernmost Area 1, where the P24h was higher than 100 mm. In addition, snowfall in the northern part of the region and severe weather (a tornado in Montgat, Area 3, and hailstorms in Tarragona N and S, Areas 5 and 6, respectively) were also observed. Although many civil protection interventions due to floods and wind action occurred mostly in the central part of the study area, most of the incurred damage was due to coastal (wave)-storm-induced hazards. Thus, the entire coastal zone was severely affected from south to north as the storm propagated along the coast (Fig. 13). In the southernmost part (Area 7), the Ebro Delta plain was extensively flooded, while the beaches were severely eroded. This resulted in significant damage to rice fields and existing infrastructure. Along the entire coast, many ports and marinas were significantly overtopped, with the breakwater of the port of Barcelona being damaged. In some stretches, such as Barcelona (Area 4.a) and some parts of Maresme (Area 3), many beaches were fully eroded, with their promenades being directly exposed to wave action; the coastal railway along Maresme was also impacted. In the northernmost part of the study site (Areas 1 and 2.a), coastal flooding occurred in several municipalities due to massive overtopping of beaches and promenades. In summary, although this was a compound event, the most important and relevant damages were caused by coastal (wave)-storm-induced hazards (Fig. 12).

On 26 and 27 December 2008 (type 3, Fig. 12), a very intense coastal storm affected the Catalan coast and was accompanied by strong winds, snow, and rain. The surface synoptic situation by 26 December was characterized by a pronounced anticyclone in northern Europe extending from Ireland to Russia, centred in Denmark, and a low-pressure area with two clear centres over Catalonia and Valencia and another located over the Azores. Along the Catalan coast, there was strong wet advection from the south-east due to the strong low–high dipole. This situation caused a very high pressure gradient that produced strong advection from the east and south-east of Catalonia. As a result, very high waves and rainfall were recorded along the study area, with the highest values being reached in the northern half of the coast (Areas 1, 2.a, 3, and 4.a). In addition to tangible damage, four fatalities occurred during the episode, three of which were associated with wave action and the fourth occurring from a flood in the Muga River (Area 1 and Table 3). Extreme wave impacts along the coastline induced significant damage, with extensive sediment losses in the beaches, promenades overtopped, and damage to infrastructure (Fig. 13). This occurred especially in the northern part of the coast (Areas 1 and 2.a), where waves reached values typical of extreme storms according to the Mendoza et al. (2011) classification. This was one of the most important recorded coastal storms, with observed impacts also on nearshore ecosystems (e.g. Sánchez-Vidal et al., 2012). On land, the strong wind uprooted a large number of trees and cut off the electricity and telephone lines. The Fabra Observatory in the city of Barcelona recorded a maximum wind gust of approximately 85 kmh-1. During the event, notable snowfall at low altitudes, some landslides, and a tornado in Platja d'Aro (Area 2.a) occurred. Many roads and train lines were cut off due to heavy snowfall and flooding of the tracks near the sea. The Ministry of the Environment allocated an equivalent of EUR 21.6 million (adjusted for 2020) to different municipalities on the Catalan coast to carry out emergency work, with the aim of repairing the damage caused by the waves on beaches and coastal infrastructure. The government granted aid packages to several municipalities and fisherfolk of more than EUR 0.72 million (adjusted for 2020).

Recently, the severe storm Gloria took place in the Catalan Sea in January 2020 (Fig. 12), with record-breaking events occurring in all areas for both wave heights and rainfall (Table 3). It started as a small superficial depression (about 600 km in diameter) located in the central part of the North Atlantic Ocean, which was increasing while moving eastward. By 18 January, the storm had nearly doubled in size, while the Azores High was being reinforced further south. On 19 and 20 January, the high-pressure zone moved to the north of the Gloria storm, leaving the latter over the south of the Iberian Peninsula. Between 20 and 23 January, both areas were well defined in the form of a dipole creating a strong pressure gradient that gave rise to very intense winds while favouring the entry of maritime air over the study area (Berdalet et al., 2020). The synoptic pattern (on 20 January +00:00 UTC, closest to the coastal storm start time) corresponded to a type 3 synoptic event. Wave heights recorded during the peak of the storm along the Catalan coast reached record maximum values, and they were accompanied by the presence of a moderate storm surge, reaching values of ∼ 0.5 m in the southernmost area (Amores et al., 2020; Jiménez, 2020). Wave impacts produced significant erosion at the beaches, with massive overtopping and flooding of low-lying areas such as the Ebro delta, waterfronts, and marinas as well as structural damage in some coastal groins and port breakwaters (Jiménez, 2020). The extreme coastal storm was accompanied by very intense rainfall and thunderstorms throughout the territory, reaching record values from the last seven decades, which significantly contributed to flooding along some coastal plains, occasional cut-offs and damage to roads and railways. The Department of Interior of the Government of Catalonia responsible for civil protection services activated three emergency plans for risk management: INUNCAT (flash floods, river floods, and coastal floods), NEUCAT (snowfalls, significant above 600 m), and VENTCAT (wind, extreme gusts lasting about 48 h, with a maximum of 144 kmh-1). These hazards caused four casualties (and 10 more in the Balearic Islands and Valencia) and extensive damage throughout the territory (Fig. 13), with a preliminary evaluation of payments to be covered by the Spanish public re-insurance company (CCS) rising up to about EUR 51 million. Inversions to rebuild port infrastructures affected by the storm are estimated to be about EUR 17.4 million; about EUR 6 million was budgeted by MITECO to repair damage in the public coastal domain, and damage due to floods in the river margins and flood plains was estimated at EUR 42 million by the Catalan Water Agency (ACA).

Notably, in agreement with the obtained results, both type 2 and 3 events (Fig. 12) are more severe than type 1 in terms of coastal storms. Heavy rainfall produced local floods and dangerous discharges in ephemeral rivers during both type 2 and type 3 events. Nonetheless, type 3 events are potentially the most compounding and intense, with Gloria 2020 being a perfect example of possible extreme impacts (e.g. Canals and Miranda, 2020; ICGC, 2020).