In the first months of 1783, a catastrophic seismic sequence, with at least five main events, hit the Calabria region (southern Italy), provoking about 30 000 casualties and the destruction of over 200 villages (Guidoboni et al., 2007). Cascade effects were also reported, including tsunamis and diffuse landslide occurrences in the Apennines that form the backbone of the region (Tinti and Piatanesi, 1996; Graziani et al., 2006; Porfido et al., 2011).

The most tragic event took place during the night of 6 February, in the town of Scilla on the Tyrrhenian coast of Calabria (see map in Fig. 1). The day before, one of the strongest earthquakes of the sequence caused severe destruction in the village, forcing most of the population to find refuge on the wide beach of Marina Grande, west of the town. A strong aftershock in the night caused the collapse of the coastal flank of Mount Pacì in the sea, 1 km west of the Marina Grande beach. The ensuing tsunami killed more than 1500 people (Mercalli, 1906) and destroyed houses and churches (Tinti and Guidoboni, 1988; Graziani et al., 2006).

The huge death toll and the impression raised by the disaster induced the Bourbon government ruling the region at that time to finance and arrange several ad hoc surveys. Further, such a calamity was seen as a big challenge by the coeval international scientific community that was engaged in understanding the nature of earthquakes. Therefore, a number of additional investigations were carried out in the first few years after the tragedy by scholars and travelers from different countries (see notes about the mainshock in Guidoboni et al., 2007). The overall result of these observations and research activities that is relevant for our study is that a description of the tsunami source and of the tsunami consequences was made available for future generations (see the discussion in Zaniboni et al., 2016).

Figure 2Toponyms reported in Table 1 (yellow circles). The area swept by the 1783 landslide is delimited in red.

In this work, our attention is focused on the tsunami impact outside the surroundings of Scilla. The most relevant reported effects are listed in Table 1 and are mainly gathered from recent papers that quote historical sources (i.e., Graziani et al., 2006; Gerardi et al., 2008; the Italian tsunami catalog by Tinti et al., 2007). The last column includes, among others, the main coeval references. The affected locations are shown in Fig. 2.

From Table 1, one can observe that the effects of the tsunami were much less severe in the neighboring area than in Scilla. This is not surprising since it is typical of landslide–tsunamis to attenuate rapidly with distance (see, for example, Masson et al., 2006 and Harbitz et al., 2013, and references therein). However, remarkably the waves were seen in Calabria from Nicotera (point 1, Fig. 2), about 40 km northeast of Scilla, to Reggio Calabria (point 5, Fig. 2), 20 km to the southwest. On the other side of the Straits of Messina, the place closest to the source is the easternmost part of Sicily, which is Capo Peloro (point 6, Fig. 2), next to the village of Torre Faro (point 7). Here the tsunami was strong and disastrous. In addition to historical sources, the impact of high-energy tsunami waves in this area was confirmed by recent geological investigations carried out in a site called Torre degli Inglesi (Capo Peloro, point 6), about 40 m away from the modern shoreline, where a 15 cm thick sand deposit in a trench layer sequence was attributed to the 1783 event (Pantosti et al., 2008).

The numerical procedure and techniques used in this paper were developed by some of the authors and presented in previous applications (see, e.g., Tinti et al., 2011; Argnani et al., 2012; Zaniboni et al., 2013, 2014a, b, c; 2016; Ceramicola et al., 2014). Therefore, only a brief description will be given here, since the reader can refer to the quoted papers.

The first step of tsunami simulations is the assemblage of the computational grids; the area covered and the adopted grid resolution depend deeply on the goal of the simulation. In general, the higher the grid resolution, the more accurate the results are but also the heavier the related computational effort is. The tsunami simulations in Zaniboni et al. (2016) were concentrated in the town of Scilla, and the extent of the very local regular mesh was a few kilometers in length and width with 10 m spaced nodes (see Grid 1, in green, Fig. 4).

Figure 4Computational grids adopted for numerical simulations of tsunami propagation: Grid 1 (green) was used for the tsunami simulations by Zaniboni et al. (2016); Grid 2 (in red) was used in the first simulation of this paper; Grids 3 and 4 (in blue) are an extension of Grid 1 covering the target area of Capo Peloro. Coordinates are in UTM-WGS84 projection, zone 33^∘ N.

In this work, our purpose is twofold: (i) to compute tsunami propagation over a wider area, involving tens of kilometers of the Calabrian and Sicilian coasts; and (ii) to compute detailed inundation in the specific area of Capo Peloro, located in front of Scilla. For the first goal, we have built a lower-resolution grid (50 m space step), denoted in Fig. 4 in red as Grid 2, by using three different databases: SRTM for topography (about 90 m spaced, Jarvis et al., 2008); EMODnet DTM, in its 2018 version, providing bathymetry every ∼160 m; and the nautical chart N.138 from Istituto Idrografico della Marina. For this latter case, we have digitized isolines with the aim of better characterizing the shallow water area, which was not adequately described by previous datasets. The resulting computational mesh covers an area of 30×27 km². As concerns the Calabrian coast, it runs from Bagnara Calabra, 10 km west of Scilla to Reggio Calabria, about 20 km southwestward. Regarding Sicily, the grid includes the city of Messina and 7–8 km of coast southward, as well as a piece of the northern coast facing the Tyrrhenian Sea via an extension of at least 10 km.

The choice of Capo Peloro as the target area for the grid resolution enhancement will be justified later, in the discussion of the simulation results obtained on the coarser domain. Figure 4 illustrates the extent of Grids 3 and 4 (how they differ from one another will be explained later as well) in blue that are actually a westward extension of Grid 1 with the same spatial resolution (10 m).

Before discussing the simulation results, it is worth providing some terminological specifications: with “wave height” or “wave elevation” offshore, we mean the difference between the offshore water top and the mean sea level. The “flow depth” is the height of the water with respect to the ground, onshore. Consequently, the onshore “wave height” is the sum of flow depth and local topographic level. Finally, with “run-up” and “inundation distance” we denote the maximum onshore ground elevation and the maximum onshore penetration (approximately normal to the shore) reached by the tsunami, respectively.

Figure 5Maximum tsunami elevation, computed for each point of Grid 2. Numbers represent the positions of the available observations (see Table 1). The respective toponyms are reported in the legend, together with the observed inundation distance (I) and run-up (R). The dashed blue rectangle marks the limits of the 10 m resolution Grids 3 and 4, zooming on the area of Capo Peloro, that differ from one another only in a small part of their topography. The purple boundary delimits the sliding surface of the 1783 Scilla landslide. For geographical coordinates, see Fig. 4.

5.1 Simulation over the 50 m spaced grid

Figure 5 shows the maximum water elevation, obtained by considering the maximum value of the sea surface height reached during the simulation time span in each node of Grid 2. This category of plots provides a general spatial pattern of the tsunami energy distribution that, in the intermediate- and far-field, depends more on the bathymetry than on the source type. Therefore, it allows one to distinguish the areas that are more prone to tsunami occurrence.

The area close to the slide shows, as expected, the highest water elevations with more than 8 m in Scilla and waves at least 4 m high (in red) affecting the surrounding coast for about 6 km. The wave height distribution illustrated in Fig. 5 also provides indications on the extension of the coast affected by 1.5 to 2 m (in yellow) and 1 m maximum wave (in green), resulting in approximately 10 and 20 km, respectively, along the Calabrian coast. Southwestward, in the Messina Strait, the maximum height reaches 0.5 m.

Moving to the coast of Sicily, especially north of the Messina Strait, some significant features are observed. First, a strong tsunami energy concentration can be noted towards the easternmost end of Sicily, named Capo Peloro, where waves higher than 6 m hit the coastline. Along the Tyrrhenian coast of Sicily other tsunami beams are clearly visible (in green, meaning at least 1 m height), the most interesting of which is the western one, affecting the coastal area close to the small village of San Saba. Overall, the simulation confirms the generally known feature that for landslide–tsunamis the wave height decays rapidly with distance.

When comparing the simulation results with the observed effects a general underestimation can be noticed: for example, in point 3 (Cannitello) of the map of Fig. 5 a run-up of 2.9 m has been reported, higher than the 1 m maximum height obtained in the simulation. The same holds for Messina (point 8) and Reggio Calabria (point 5), the most important towns in the Messina Strait, where the difference is by far bigger. Only the area of Capo Peloro (points 6 and 7) seems to fit the observations.

Such discrepancies can be ascribed to the low resolution of the computational domain (50 m) not allowing one to describe the coastal zone and all the nonlinear effects prevailing in shallow water properly, such as the inland inundation. Further, low-resolution simulations often cannot account for tsunami wave amplification caused by resonances induced inside coastal basins, such as harbors (see, for example, Dong at al., 2010; Vela et al., 2010). The latter can explain, for example, the underestimations of the observations reported in Messina and Reggio Calabria.

These considerations push for the use of a more detailed computational grid for the sites of major interest, suggesting that the optimal approach is (1) to assess the general tsunami energy distribution in the first run of simulations carried out in a wider low-resolution domain and (2) to pick up the areas most exposed to tsunami occurrence (e.g., areas hit by energy beams) to perform higher-resolution investigations. With this approach in mind, the focus has been moved to the area enclosed in the dashed blue rectangle of Fig. 5, including Capo Peloro, where most of the devastating effects of the tsunami outside the source zone were reported (see Table 1, nos. 6 and 7).

Figure 6Area covered by Grids 3 and 4 employed to simulate the impact of the 1783 tsunami at Capo Peloro, the easternmost point of Sicily. The magenta contour delimits the strip swept by the landslide. Coordinates are in UTM-WGS84 projection, zone 33^∘ N.

5.2 First tsunami simulation in the Capo Peloro area

An additional computational grid (Grid 3) has then been built to simulate the tsunami propagation and inundation in this zone.

Actually, in this application, due to the proximity of the landslide to the target area, the choice was to build Grid  3 by extending the high-resolution grid (Grid 1, 10 m node spacing) used by Zaniboni et al. (2016) westward for the simulation of the local tsunami effects in Scilla. The relevant data for the Sicilian region was provided by the Civil Protection Department of Sicily and retrieved from the regional cartography service, SITR (Sistema Informativo Territoriale Regionale), in the form of a DTM describing the present topography of the coastal zone in high detail. As for bathymetric data, they have been retrieved from the same sources used for Grid 2. Grid 3 covers an area of 12.5 km (east–west) by 4 km (north–south), for a total of more than 500 000 nodes. As can be noticed from the map in Fig. 6, Capo Peloro presents a morphology that is radically different from the Calabrian coasts. In Calabria, coasts are steep, rapidly ascending to 400 m a.s.l., while in the area of Capo Peloro a wide lowland is found, extending for about 2 km×1 km with elevation on the order of 1 to 5 m. This area is now densely inhabited, with many houses facing the sea, especially along the southern coast. Another interesting feature is the presence of two brackish lakes, called Lago di Ganzirri and Pantano Piccolo (also known as Faro Lake), characterized by 6 and 29 m water depth, respectively. They are connected to the sea (and to each other) by narrow channels, built under British authority at the beginning of 19th century (Leonardi et al., 2009; Manganaro et al., 2011; Ferrarin et al., 2013). One of the most relevant observations reported in the historical reconstruction of the tsunami effects is that the basin of Pantano Piccolo was reached by sea water in 1783 (Minasi, 1785). This feature should be considered in the evaluation of the simulated tsunami accuracy.

Figure 7Propagation frames for the 1783 Scilla landslide–tsunami in the target area of Capo Peloro, eastern Sicily. Geographical coordinates are given in Fig. 6.

Figure 7 shows different frames of the tsunami propagation simulated by means of the code UBO-TSUFD, starting from the snapshot taken at 80 s. It is worth pointing out that the origin time coincides with the landslide motion initiation and recalling that, according to our simulation, the slide enters the sea between 10 and 30 s and settles at the final position around 110 s (see Zaniboni et al., 2016, for the accurate description of these results).

The landslide-generated wave heads towards Capo Peloro with a strong positive front 5 m high, directly caused by the slide entering the sea and characterized by an almost circular shape when moving in deep water. At t=100 s the wave begins changing form, due to the interaction with the platform characterizing the seabed east of Capo Peloro. When meeting shallow water, the tsunami is intensely decelerated and begins to deform. Frames at 100, 110, and 120 s display the subdivision of the wave into two fronts, north and south of the Capo Peloro end, hitting the coast obliquely. At 140 s the southern coast of the cape is reached by the tsunami, which floods the mainland by some tens of meters and then inundates the other parts. In the eastern extreme the water penetration is maximum, reaching 600–700 m distance (as visible from the 200 s frame). At the same time, the northern branch of the tsunami reaches the coastline, with a positive 2 m high front that tends to align to a direction parallel to the shoreline. The following oscillations constituting the train of waves (already visible behind the tsunami front in the first snapshots) do not produce relevant effects on the coast, apart from the cape, where a second relevant positive signal, exceeding 2 m, can be noticed at t=240 s.

Figure 8Maximum water inundation at Capo Peloro for the high-resolution Grid 3 (western portion). The black squares mark the cells inundated in the simulation with the low-resolution Grid 2. The purple symbols represent the positions of historical buildings: Torre degli Inglesi, the site where tsunami deposits were recognized and associated with the 1783 event (Pantosti et al., 2008) and Torre Bianca and which is partially buried by sand deposits (Bottari and Carveni, 2009). For geographical coordinates, see Fig. 6.

The maximum water elevation for each node of the western portion of Grid 3 is illustrated in Fig. 8. Some of the features already observed in the 50 m grid (Grid 2; see Fig. 5) are confirmed, such as the maximum wave height concentration in the easternmost zone of Capo Peloro. Here, in correspondence with the edifice named Torre degli Inglesi (“English Tower”, referring to a building realized during British rule, marked in black in Fig. 8), some tsunami deposits were recognized and associated with the 1783 event (Pantosti et al., 2008). The simulation shows the maximum water elevation exactly for this spot, exceeding 6 m, and the maximum inland penetration, more than 700 m. Furthermore, here one can notice the main discrepancy between the low-resolution Grid 2 and the high-resolution Grid 3 simulations: in the former case, only a few inland cells (purple in Fig. 8) are inundated (corresponding to an inundation distance of 150 m), while in the latter the inland penetration is larger by far. For the remaining coastal stretch, the water flooding is limited to the scale of tens of meters or less, showing little difference between the two simulations. It is very relevant to stress that, concerning the small lake of Pantano Piccolo, both simulations exclude the possibility that it is reached by the tsunami, contrary to what reported in the historical reconstructions.

The transition from Grid 2 to Grid 3 produced interesting results and precious insights on the tsunami effects in the target area. Yet, as already noted, one of the most relevant features (i.e., the inundation of the lake of Pantano Piccolo) was not reproduced, nor can the uncertainties naturally connected with the simulation of natural events like landslides and tsunamis be invoked. Indeed, to generate a tsunami able to penetrate further the flat area beyond Torre degli Inglesi and to cover the distance of more than 2 km separating this from the lake, a much larger landslide would be needed, which goes against the geological evidence. Another possibility would be that the tsunami finds a penetration route from another direction, for example from the north. In fact, the strip of land between the basin and the sea is narrow (around 150 m) and is characterized by 3 to 5 m topography, which is again really unlikely to be overrun by the simulated tsunami.

A possible alternative solution comes from the considerations of the coastal morphology of this area. In a historical study of this zone, Randazzo et al. (2014) gave evidence of the strong variability of the coastal stretch around Capo Peloro, due to the action of atmospheric agents. The dataset used to build Grid 3 refers to the present ground level that could have experienced great changes since the 1783 event. Based on the historical investigations by Bottari et al. (2006) and Bottari and Carveni (2009), we know that in Pantano Piccolo hundreds of ships took shelter during the early Roman period (5th century BCE) and that a small harbor probably also existed, meaning the presence of direct access from the sea. In the following centuries the channel was filled with sediments, but a look at the topography clearly shows that the lower ground area starts from the northeastern corner of the lake (as also evidenced by the modern 2 m isoline taken from Bottari et al., 2006) and is the best candidate for the channel that was connecting the lake to the sea more than 2000 years ago. Currently, the access is closed by a 3–4 m sand dune on the coast that can be reasonably associated with progressive sedimentation due to wind and storms and removed, in order to reconstruct a tentative 1783 morphology of the area.

Figure 9(a) Shaded-relief map of the current topography of Capo Peloro, with a 2 m isoline (dashed red line) obtained from Bottari et al. (2006) showing the most likely connection between the lake and the sea, just east of Torre Bianca. The corrections are shown in green and blue when negative (meaning “digging” with respect to the present ground level) and in yellow and red when positive, meaning increased ground elevation. (b) Contour plot of Grid 4, with the low topography from the northeastern corner of the lake to the sea. In green, the new 2 m isoline is evidenced. With the corrected topo-bathymetry we have built, Grid 4 has the same areal coverage as Grid 3. For geographical reference, see Fig. 6.

Another element that needs to be considered is the existence, on the narrow coastal stretch separating Pantano Piccolo from the Tyrrhenian Sea, of a partially buried tower named Torre Bianca (“White Tower”, known also as “Mazzone”; see Figs. 8 and 9 for the location), which now is filled with sand up to the first floor (Bottari and Carveni, 2009). Yet, in the 18th century, this building served as a storehouse, so it was probably fully available and unburied, meaning that the previously cited assumption of removing 3–4 m of sediments (corresponding to the average height of one floor) in this area is perfectly reasonable.

Keeping all these considerations in mind, a rearrangement of Grid 3 has been done resulting in a new mesh, the Grid 4, with the same spatial extension. Fig. 9a shows a shaded-relief map of the present topography and bathymetry and the corrections that have been done to reconstruct the likely morphology of 1783. Notice again that in Fig. 9 only the western portions of Grids 3 and 4, covering the target area, are reported, since the source area remained unchanged. Such reconstruction is based on hypotheses and some evidence. For example, the two small channels now connecting Pantano Piccolo to the sea did not exist in 1783, since they were excavated during the following British rule. For this reason, they have been “filled” in Grid 4. The most relevant changes, regarding the area between the northeastern corner of Pantano Piccolo and the Torre Bianca site, is the removal of 1 to 5 m of the surface ground layer. The highest amount of “digging” is located close to the coast, where the sand dune currently closing the possible access of water (and presumably filling the first floor of Torre Bianca) has been leveled. Morphological changes in other areas of the cape are much less realistic since they would entail lowering of high topography that is less intensely affected by natural agents such as storms, winds, or sea waves.

Figure 10Propagation frames of the 1783 landslide-generated tsunami in the western part of Grid 4. Notice the water propagation inside the lake. Geographical coordinates are given in Fig. 6.

The tsunami simulation in Grid 4 coincides with the simulation in Grid 3 for the first 200 s, as is shown in Fig. 10. The main differences start from the 220 s frame, when the coastal area east of Torre Bianca begins to be inundated. The following frames show that the water reaches more than 1 m height, channeling through the modified area and reaching the lake between 240 and 260 s. After that, a small positive wave propagating inside the basin can be noticed, 10 to 20 cm high, with some reflections. The adopted correction to the topography, then, facilitates the tsunami flooding of the Pantano Piccolo lake.

Figure 11(a) Maximum water elevation for the western part of Grid 4. Notice that tsunami penetrates also into the lake. (b) Maximum flow depth for each topographic node. For geographical coordinates, see Fig. 6.

The map with maximum water elevation (Fig. 11a) is analogous to the one of Fig. 8. The main difference comes from the modified zone, where the water climbs up to 2.5 m close to the coast and 1 m just before entering the lake. Inside the basin, the maximum wave elevation reaches 40 cm in the western part but less in the central area, which is deeper. Figure 11b displays the height of the water column characterizing the flood, known usually as flow depth. The plot shows that the northern coast is affected by higher flow depth, reaching 2 m almost everywhere , with the maximum located at the Torre degli Inglesi (more than 5 m). The flood, just before entering Pantano, is characterized by a water column of about half a meter.

In this paper we have performed a deep analysis of the 1783 tsunami, extending the results of a previous study where attention was restricted to the area of the highest impact, i.e., Scilla and surroundings. Here, we have simulated the tsunami in a broader region and we can draw the following remarks. An interesting, yet unexpected, feature that we found is the strong tsunami beam reaching the village of San Saba, located along the northern coast of Sicily, 10 km to the west (see Fig. 5). There is no historical account of a tsunami impact here. But this can be easily explained by considering that the tsunami probably only affected the beach and passed unobserved. As for the Calabrian coast, we observe that the simulated wave height rapidly decreases with distance from the tsunami origin, as expected for landslide-induced tsunami events. In addition to the positive results highlighted above, we also have to point out as a drawback that in the region covered by the 50 m resolution grid (Grid 2) a general underestimation of the observed wave height is noted, which has to be ascribed to an insufficient description of the local, shallow bathymetry and of the coastal topography features. The conclusion is that 50 m computational domain is suitable to assess the general pattern of the tsunami energy distribution, highlighting the areas where the highest waves concentrate, although this is inadequate to provide good coastal tsunami data for the 1783 case.

The area that is of the highest interest is Capo Peloro, the easternmost cape of Sicily, where the most damaging effects besides Scilla were recorded, including a deep tsunami penetration and the death of 26 people, which is the only death toll of the tsunami away from the source. Thus, this zone has been selected as a good target for a grid refinement. A higher-resolution computational domain (Grid 3, 10 m spaced) has then been assembled, based on the present morphology. The tsunami simulation on Grid 3 shows that the inundation distance in the area of maximum impact changes radically, passing from 150 m found for Grid 2 to more than 600 m (Fig. 8), in full agreement with the observations. This feature clearly shows the improvement brought by the computational grid adjustment.

However, the tsunami does not reach the Pantano Piccolo lake, which contrasts sound historical accounts (see Minasi, 1785). This misfit calls for simulation changes that can be done via three possible options: (i)  producing a stronger tsunami; (ii) using a better resolution, or (iii) changing the basic topo-bathymetry dataset, especially in the target area. We have excluded option 1, which is to use Grid 3 and to produce a stronger tsunami by using a larger volume landslide, since the landslide geometry is well constrained by geological considerations of the scar morphology, the identified offshore deposit, and the good agreement with observations of the tsunami run-up height in the Scilla beaches computed through Grid 1 (see Zaniboni et al., 2016). We have also further ruled out option (ii), which is to use grid spacing finer than 10 m, since this resolution was enough for the target area, as mentioned before, and also because it was enough to fit the 600 m observed inundation of Capo Peloro with Grid 3. So the only option left was to keep the same resolution and to build another set, Grid 4, which is based on opening a way to the tsunami access through an ancient channel that was a pathway to a well-protected inner harbor in Roman times, and which is presently filled with sediments and obstructed by a coastal sand dune (Fig. 9). The signature of this channel is still recognizable today in a linear topographic depression (only 2 m a.s.l.) going from the northeastern corner of the lake to the sea. We believe that Grid 4, allowing the tsunami to reach the lake, represents reasonably well the Capo Peloro topography at the time of the 1783 tsunami occurrence, also based on the existence of Torre Bianca building, whose present status reinforces the plausibility of sand deposit removal along the coast. The tsunami reached the lake through the channel either because there was no coastal dune, as assumed in Grid 4, or because the dune was lower than today and was overcome and/or demolished by the incoming tsunami.

Concerning original raw data, topography has been retrieved from the SRTM database, available at http://srtm.csi.cgiar.org/ (Jarvis et al., 2008); bathymetry has been obtained from the EMODnet database (https://doi.org/10.12770/18ff0d48-b203-4a65-94a9-5fd8b0ec35f6, EMODnet Bathymetry Consortium, 2018), integrated with nautical chart no. 138 of the Istituto Idrografico della Marina (© I.I.M Genova, 1993). The computational grids for the landslide and tsunami simulations have been assembled by re-elaborating and matching the different datasets. The simulation code for landslide motion and tsunami propagation is written in Fortran77 and Fortran90, respectively, and is developed and maintained by the authors. Researchers and scientists interested in the data and code can contact the authors.

FZ reconstructed the landslide scenario, performed the landslide simulation, arranged the tsunami computational grids, prepared the initial draft of the paper, and managed the later phases of publication. GP, who tests and maintains the tsunami simulation code, helped prepare the tsunami computational grid and contributed to the paper improvement. GG contributed to the landslide scenario definition and to the paper review and editing. MAP retrieved relevant information on historical reports reconstructing the tsunami effects and contributed to the landslide scenario definition. AA contributed to computational grid preparation and revised the paper. ST supervised the whole work and reviewed the paper drafts.

The authors declare that they have no conflict of interest.

This paper was edited by Ira Didenkulova and reviewed by Marinos Charalampakis, Amos Salamon, and one anonymous referee.