The Amalfi Coast consists of a steep mountain front (up to 1444 m a.s.l.) that rises abruptly from the Tyrrhenian Sea (Fig. 1). It is a rocky coast mostly formed by a pile of Mesozoic carbonate rocks, covered by Tertiary to Quaternary siliciclastic and pyroclastic units which have been tectonically uplifted since the lower Pleistocene. Bedrock rivers and channels deeply dissect the carbonate bedrock, forming a complex fluvial system characterized by small catchments that are very high relative to the base sea level. These rivers show a distinct seasonality and torrential behaviour, with main delivery areas into the adjacent marine shelf (Fig. 2; Esposito et al., 2004; Budillon et al., 2005; Violante, 2009; Violante et al., 2009).

During the last millennia this area has been repeatedly mantled by the pyroclastic products of the Somma-Vesuvio that create favourable conditions for volcaniclastic debris to generate rainfall-triggered mass flows and flash floods. The Plinian eruption, that destroyed the Roman cities of Pompeii, Stabiae and Herculaneum in AD 79, deposited up to 2 m of erosion-prone volcaniclastic material (Sigurdsson et al., 1985) on the steep coastal slopes, causing conditions of increased geomorphic instability.

Geologic evidences for rapid slope erosion following the Pompeii pyroclastic fall include alluvial reworked volcaniclastic sequences (locally called Durece) occurring as residual outcrops along narrow stream valleys (Cinque and Robustelli, 2009) and coastal fan deltas fed by small alluvial fans at the mouth of the main streams (Sacchi et al., 2009; Violante et al., 2009). The latter are composed of wedge-shaped coarse-grained alluvial deposits that thicken towards the sea and represent the subaqueous counterpart of small fans at the river mouths (Fig. 2, inset).

Pyroclastic air-fall tephra derived from the late Quaternary activity of Somma-Vesuvio still represent unstable sedimentary covers on top of the steep carbonate slopes of the Amalfi Coast. These deposits create conditions of elevated slope instability in conjunction with rainstorm events that frequently hit the Amalfi Coast. The slides are mostly shallow and very wide, extending all the way to the mountain ridge and crest, and largely belong to the categories of soil slip and debris/earth flow phenomena. In addition to rapid sediment transport along the valley flanks, landslide debris flowing from slope to streams produces fast-moving large debris torrents (flash floods) with significant catastrophic implications for local communities mostly living on alluvial fans at stream mouths. Here flood-prone streams have been artificially forced to flow underneath roads and squares to exploit the whole delta surfaces for urban development (Fig. 2).

Human interference plays an important role in the generation of flood hazard (e.g. Smith and Ward, 1998; Plate, 2002). In particular, the mismanagement and lack of maintenance of channel sections and the deforestation of water catchment areas can significantly increase the risk of flash floods. In the study area, extensive terracing through deforestation and use of rocks to build retaining walls has generated a cultural landscape strictly depending on the correct management of mountain slopes and channel beds. Interference by human actions has been documented for past floods in the Sarno-Quindici area (Calcaterra et al., 2003), located a few kilometres north of Atrani, which was hit by a destructive flooding event in 1998.

In this paper we analyse the 9 September 2010 rainfall event that hit the Amalfi Coast and its effects on the Dragone catchment and the village of Atrani. Direct field observations that include geological investigations and damage to property and infrastructures have been combined with meteorological and hydraulic/hydrological analyses. Reconstruction and recurrence of past events based on different historical sources and marine geophysical and geological data of the Dragone submerged delta have also been taken into account.

The Dragone catchment drains an area of 9.3 km2 along the steep coastal slopes of the Amalfi Coast. The basin develops in a north–south direction and is strongly asymmetric, with the eastern flank composed of a short and abrupt slope corresponding to a fault scarp and the western flank formed by four main sub-basins: Scalandrone, Nocelle, Senite, and S. Caterina. A fifth sub-basin, Frezzi, develops at the head of the Dragone stream (Fig. 3a). The drainage area rises up to 1420 m a.s.l. and cuts into Mesozoic limestone discontinuously mantled by Quaternary volcaniclastic and alluvial deposits. A low drainage frequency (5 km-2) and a sub-dendritic pattern characterize the hydrographic network with the main stream, the Dragone, discharging directly into the Tyrrhenian Sea. Slope analysis based on a 5 × 5 m cell size DTM (digital elevation model) indicates that the topographic gradient of the catchment area is arranged into two main slope classes, ranging from 15 to 35∘ and from 35 to 50∘ (golden/yellow and brown colours, respectively, in Fig. 3b). The mean slope is 30∘.

The Dragone stream is 6.8 km in length with the terminal section covered by a roadway crossing the village of Atrani. The covered section of the water course has an input section of 3 × 9 m (h × w) that reduces to 1.80 × 5.50 m at the closing section, and has a total length of about 300 m. Runoff waters are regulated by concrete levees and check dams that extend over two-thirds of the hydrographic network.

At sea a submerged fan delta is present at the closing section of the Dragone catchment (Sacchi et al., 2009; Violante et al., 2009). The delta body is 0.2 km2 wide and reaches a maximum thickness of 25 m. It displays a generally conical morphology with a delta front slope of 30∘ and foreset inclination ranging from 15 to 30∘. This structure is composed of alluvial sequences that coincide with significant changes in the river activity during streamflow phenomena. Increases in fluvial sedimentary discharge are recorded as successive phases of delta growths with which temporary shoreline progradations are associated.

The detailed study of the internal stratigraphic architecture of the Dragone fan delta indicates various depositional phases following the main AD 79 alluvial crisis, possibly modulated by the interplay between the availability of loose pyroclastic covers and the varying erosional rates due to the climatic oscillations occurring in the last millennia (Fig. 4; Sacchi et al., 2009; Violante et al., 2009). The major change detectable in the Amalfi fan deltas occurred in the Early Medieval Cool Period (ca. AD 500–AD 800), that developed immediately after the Roman Warm Period. Further changes in the stratal patterns of the delta foresets, indicative of high streamflow activity, may be correlated with the Medieval Warm Period (ca. AD 900–AD 1100) and the Little Ice Age (ca. AD 1400–AD 1850).

A field survey was undertaken soon after the 9 September 2010 rainstorm in the Dragone catchment and village of Atrani. The observed rainstorm-induced geological effects include shallow landslides and sediment removal along channels, a temporary dam in the mid-lower section of the Dragone stream, the partial breaking up of the main road crossing the village of Atrani, and the deposition of a coarse terminal fan at the mouth of the Dragone stream (Fig. 5).

The field data indicate that slope erosion triggered by the 9 September 2010 rainstorm was predominantly linear. Sediment removal by linear erosive processes significantly engraved tributaries and the main stream up to a depth of 2 m (Fig. 5a). The displaced materials were mostly composed of pyroclastic deposits and solid wastes occurring along the channel beds, often above the hydraulic check dams. A large number of tree trunks of different sizes, used to build bridges across ditches and tributaries, were also included in the transported material.

Minor soil slips have been observed uphill on the high slope at Punta delle Castagne, a rock crest connecting the S. Eustacchio, S. Caterina, and Scalandrone watersheds (Fig. 5b) and along the western flank of the Frezzi watershed. Here displacement of the channel sides occurred both at channel heads and, locally, at the middle–upper reaches of small tributaries.

Apart from overspill deposits composed of white pumices observed at break slopes just below the watershed areas and along the middle reach of the Dragone stream, no significant aggradation was observed in the catchment area. Removal of materials from the stream bed was produced by a fast-moving debris torrent with high erosive capacity, transporting the mobilized materials all the way down to the coast. Here the Dragone stream is artificially forced to flow underneath the main road crossing the village of Atrani, resulting in siphoning and consequent breaking up of the above roadway (Fig. 5c). Once it reached the coastline, the hyperconcentrated flow engraved the Atrani beach and dumped the transported materials into the sea in the form of a coarse alluvial fan (Fig. 5e). This induced a shore progradation of about 30 m. The alluvial deposits entering into the sea were mostly composed of white-grey pumices including tree trunks, wastes, and rock boulders of different sizes reaching up to 1 m in diameter. Some cars parked along the main road were transported up to the beach area and beyond (Fig. 5e).

Despite the carbonate nature of the study area, there was no evidence of karst features that could have influenced the flooding mechanism. In fact the drainage of waters in the Amalfi Coast is mainly by relatively deep underground paths towards offshore springs located in correspondence of the main neotectonic lineaments (Celico et al., 2005). This may, at least in part, prevent direct interactions between groundwater and surface water during rainfall events as typically occurs in karst settings (Jourde et al., 2007; Gutierrez et al., 2014).

Discharge and depth of flow downstream was probably increased by abrupt draining of a temporary dam reported by eyewitnesses in the lower section of the Dragone stream. Such damming was favoured by a narrow flow section and enhanced by a man-made structure built in the stream bed (Fig. 5d). The failure of temporary debris dams and the draining of ephemeral lakes have been described for different flood events that have occurred in the study area (Passerini, 1925; Penta et al., 1954). These phenomena can produce exceptional temporary discharges and highly destructive peak flows, reaching depths as high as 8–10 m (Larsen et al., 2001; Perez, 2001; Esposito et al., 2004; Violante, 2009).

The first days of September 2010 in the Mediterranean area were characterized by a drastic seasonal transition. The first depression from the Atlantic was favoured by the formation of a high-latitude anticyclone over Scandinavia. Such an anticyclone forced the oceanic perturbed airstream towards the medium European latitudes and the Italian peninsula. The unsettled air of Atlantic origin headed later for the low Tyrrhenian basin where very humid airstreams coming from south/south-west were forced by an area of low pressure centred in the northern Tyrrhenian Sea. This unstable scenario was fed by the Mediterranean Sea, which created conditions for a secondary area of low pressure to develop in the lower layers of the atmosphere.

The front was followed on 9 September 2010 by a far more organised instability in the southern Tyrrhenian basin with well-defined storm cells localized along the Campania, Sicily, and Calabria coastal areas. The high humidity associated with such a positive vorticity led to a mesoscale convective system (Fujita, 1986) over the southern Tyrrhenian Sea and southern Italy, as captured by a Meteosat image in the visible channel at 17:00 UTC (time hereafter expressed in UTC) on 9 September in Europe (Fig. 6a). The warm-humid flow associated with this system hit the Campania region, inducing a strong maritime thunderstorm between 10:00 and 22:00, with maximum intensity on the Lattari Mountains, as shown in the map of ground lightings (Fig. 6b).

The character of the convective system was conditioned by the Picentini and Lattari mountains that forced the air masses to follow the orography of the coastal range (see Fig. 1) and the positive anomalies of the coastal waters. Data from the Italian Oceanographic Network (ISPRA; Fig. 7) indicate that air temperature was lower than sea-surface temperature during the rainstorm, maintaining the thermodynamic conditions that allowed the convective system to sustain the storm cell.

Physical analysis of the rainstorm was based on the real-time data recorded by the rain gauges provided by the Centre for weather forecast and monitoring of the Campania regional authority (CEMPID, 2010; Table 1). On the coast the rainfall lasted 4 h, starting at 14:10 east of Salerno (Fig. 8). At the Ravello rain gauge, located along the watershed of the Dragone basin, the rain started at 15:50 and persisted for about 1 h (cumulative rainfall 80.8 mm) with a maximum rainfall intensity (19.4 mm over 10 min; Table  1) from 15:50 to 16:00. At the Agerola weather station, located about 5 km from the Dragone basin, the same event lasted about 1 h with an intensity rainfall peak of 157.2 mm h-1 from 16:00 to 16:10, showing a significant similarity with the cumulative rainfall of the Ravello gauge (see Fig. 8). This suggests that the Atrani event is consistent with a storm cell characterized by a very limited areal extent, as confirmed by the Meteosat visible image that shows a single cell with a very flat elliptical shape elongated in a NE–SW direction (inset in Fig. 6a).

The size of the storm cell that produced the Atrani event was calculated (1) by considering the 50 mm isohyet corresponding to the downdraught in the time span ranging from 16:00 to 16.50 (Fig. 9); and (2) on the basis of the number of pixels forming the storm cell visible in the Meteosat images at 17:00. In the first case the area was about 80 km2, while in the latter it ranged from 56 to 75 km2. Therefore, the Atrani storm cell can be ascribed to a mesoscale convective system (MCS) with a horizontal scale of about 20–30 km and duration of less than 1 h (lower bound of the MCS mesoscale β; Orlansky, 1975; Thunis and Borstein, 1996). These events typically occur in the Mediterranean Sea between April and September, with a higher frequency mostly concentrated during September (Lionello et al., 2006), and include multiple storm cells in different evolutionary stages (Morel and Senesi, 2002) that are strongly influenced by local orography.

In order to evaluate the hydrological response of the Dragone basin at the stream mouth, a semi-distributed rainfall-runoff model has been used. For this purpose, each of the five sub-basins of the Dragone stream (see Fig. 3a and Table 2) were analysed by Soil Conservation Service (SCS) dimensionless unit hydrograph rainfall-runoff transformation, and SCS curve number (CN) loss method (USDA SCS, 1986a) implemented in HEC-HMS (USACE HEC, 2010).

The SCS dimensionless unit hydrograph procedure is one of the most well-known methods for deriving synthetic unit hydrographs, especially for small basins. The rainfall-runoff model is based on the lag time parameter, that is the time interval between the centroid of the effective rainfall hyetograph and the peak of discharge (Table 3; USDA SCS, 1986b; Singh, 1989).

The effective precipitation depth, according to the SCS-CN procedure, is Pe=P-Ia2P-Ia+SifP>Ia0ifP≤Ia. P is the total precipitation depth; Pe is the depth of excess precipitation or direct runoff; Ia is the “initial abstraction” or the amount of rain before the runoff starts, which infiltrates or is intercepted by vegetation; and S is the potential maximum soil moisture retention during the runoff.

The value of S for a given soil is related to the curve number (CN) which is a function of the hydrologic soil-cover complexes as (if S is expressed in mm) CN=25 400S+254. In the case of the Dragone basin, the adopted CN value is 66, corresponding to forest-like cover (woods–grass combination), considering hydrological soil group B (moderate infiltration) with antecedent moisture condition II (average) and Ia = 0.2 S.

The lag time tlag has been estimated from the concentration time tc calculated by the SCS formula (Chow et al., 1988): tlag=0.6×tc=0.6×0.00227×L0.81000CN-90.7i-0.5[hour]. L is the hydraulic watershed length expressed in metres (m) and i the mean basin slope as a percentage ( %). Figure 10 reports the results of the hydrological model where the estimated peak discharge of the clear water is about 65 m3 s-1 occurring around 17:00.

Due to the lack of stream gauge data, we analysed the hydraulic response of the Dragone catchment to the 9 September 2010 rainstorm event and the sediment transfer during the flood on the basis of (a) morphology of the Dragone basin, (b) pluviometric data, (c) field evidences, (d) homemade videos and photos, and (e) eyewitness accounts.

According to local eyewitness accounts, the peak flow of the Atrani flood event occurred between 16:50 and 17:10, which is about 40 min later than the hyetograph centroid, at about 16:25, while the flood duration ranged from 40 to 60 min (Fig. 11). In the Atrani urban area the covering of the Dragone stream caused the flood wave to be split into two different currents: a main flow underground, below the road level (Via Dei Dogi), along a closed section varying from 3(h) m × 9(w) m at the input section to 1.80(h) m × 5.50(w) m at the closing section; and a second flow at the surface, along the road, constrained by almost continuous man-made structures and buildings 5.5 m away from each other (see Fig. 5c). The maximum depth of the upper flow was about 1 m, i.e. 3.5 m on average above the stream bed.

The highest flow velocity for both flows was estimated from amateur videos by tracking particles transported by the flood through the analysis of video frames with an approach similar to that used in the particle image velocimetry technique (Fig. 12; Raffel et al., 2007). Peak discharge was obtained by multiplying estimated flow velocity and the flow section.

The estimated flow velocity along Via Dei Dogi is about 3–4 m s-1 and, consequently, the peak discharge is approximately 20 m3 s-1. For the underground flow the estimated peak velocity is about 6–7 m s-1, while the peak discharge is on the order of 60 m3 s-1, taking into account that the flow fills roughly 80 % of the closed section. Then the total estimated peak discharge is 80 m3 s-1 (water + sediment; see Fig. 11), which is 15 m3 s-1 more than the estimated clear water peak discharge (65 m3 s-1) due to sediment load (Qs). Similar values for total peak discharge (98.4 m3 s-1) have been reported by Ciervo et al. (2014), who used a dedicated numerical code for the hydraulic modelling of the 9 September 2010 Atrani flash flood.

Assuming that active sediment removal occurred in a time span of 40÷ 60 min, the calculated sediment volume mobilized during the event is Vs=Qs×t=15×(40÷60)2×60=18 000÷ 27 000m3. This value is in good agreement with the volume estimation of the sediments deposited in the form of an alluvial fan delta at the Dragone mouth and on the street and square that cover the stream path (Fig. 13). Comparison of bathymetric data collected soon after the flood event with an older data set owned by IAMC-Napoli enabled the measurement of the sediment volume that flowed into the sea, at about 14 000 m3 (Fig. 13). On land, sediment thickness along Via Dei Dogi and Umberto I square reached depths averaging 0.5 m, while beach aggradation was about 1 m, with a volume of sediment accumulation of about 7000 m3. Therefore, the estimated volume of the sediment transported to the terminal section of the Dragone stream is about 21 000 m3. Taking into account the additional volumes (estimated at about 20 % of the measured volume) removed by the sea currents, or related to dispersal of finer sediments at sea, a total volume of about 25 000 m3 can be obtained.

Sediment volume can be expressed as sediment bulking factor (BF; Gusman et al., 2009) that defines the ratio between the peak flood discharge QB and clear water discharge (Qw): BF=QBQw=Qw+QsQw. In terms of sediment load, bulking factor can be also expressed as BF=11-CV100. CV is the volume concentration of sediment.

In the case of the 9 September 2010 flood event the estimated BF to the flood peak is 1.2 (5) that leads to a sediment concentration on the order of 20 % in volume (6). This value is close to the lower limit of hyperconcentrated flow (Costa, 1988; Jakob and Hungr, 2005) and it corresponds to a flow in which peak discharge is comparable to that of a clear water flood, and velocities are similar to those of water during a flood (Hungr et al., 2001; Pierson, 2005). This is also confirmed by amateur videos that show a very turbulent flow.

Historical records are an important source of information on floods and fundamental to establish a reliable flood return frequency. The potential of historical data is based on their value as repositories of the magnitude and characteristics of flood events by indicating the level of damage, the number of victims, as well as the type of the induced geological effects (e.g. Calcaterra and Parise, 2001; Tropeano and Turconi, 2004; Marchi and Tecca, 2006). This is particularly true for the Campania region, an area with a long history and much documentation, where historical research is crucial for hydrogeological risk assessment in the perspective of proper land use planning, damage mitigation, and reduction of human fatalities.

In the study area the systematic analysis of more than 3000 documents, including documentary and bibliographic sources, as well as newspapers at the state archives in Naples and Salerno (Esposito et al., 2003, 2004; Porfido et al., 2009), complemented by data attained from scientific papers and national and international projects (Guzzetti et al., 1994; Guzzetti and Tonelli, 2004), allowed the time–space distribution and the characteristics of flood events that have occurred since 1540 to be reconstructed. The quality and completeness of the various sources were evaluated and carefully analysed in their historical context, to obtain the best information rather than the best data set quality (Calcaterra and Parise, 2001; Barriendos et al., 2003). Eighteen events, excluding the one discussed in the present paper, were identified and characterized on the basis of (a) the distribution of the flooded areas, (b) distribution of damaged localities, (c) duration and timing of the event, and (d) number of casualties. The historical information shows that most of the events (13 events) took place at the transition between summer and autumn, four events were in winter, while only one was recorded during the spring season (Table 4).

The intensity and the duration of the rainfall, the level and distribution of damage of man-made structures, the number of victims, and the induced geological effects have been considered to distinguish major flash floods from minor flash floods (Table 4; Casas Planes et al., 2003; Llasat et al., 2005; Barnolas and Llasat, 2007). In particular, shoreline progradation has been considered as a key element for major flash floods as it results from significant bed load transport and hyperconcentrated streamflows. Based upon the above assumptions, the following 6 flood events out of the total of 18 have been classified as major flash floods.

October 1540: not much information is available for this event, but it is reported as “the great Atrani flood”. This indication, along with the occurrence of severe damage and extensive landslides and inundation, allows it to be classified as a flash flood.

Detailed field surveys and measurements, along with information from eyewitnesses and home-made videos, proved to be critical for reconstructing and modelling the flash flood that affected the village of Atrani on 9 September 2010. The collected data were combined with meteorological and historical analyses and marine geophysical surveys in order to reconstruct the physical features of the flood event and to evaluate the recurrence of flash floods in the study area. The main results can be summarized as follows.

The rainstorm that generated the Atrani event lasted 4 h and was strongly conditioned by the local orography and positive thermic anomalies of the coastal waters during the warm season.