Tsunami generation and propagation: modelling of the 1908 Messina straits case

Many potential sources, co-seismic, underwater landslides, or a combination of those, have been proposed and simulated in numerical models to explain the catastrophic tsunami that followed the 1908 earthquake in Messina, destroying large costal areas in Calabria and Sicily, in southern Italy. However, there are still significant discrepancies between the proposed mechanisms of tsunami generation and the observations of coastal impact made in the weeks and months following the event, and hence no real consensus yet on the actual sources for the tsunami. Another factor for this situation has been the idealizations made in many numerical models of tsunami generation, particularly from underwater landslides. The objective of the research is to perform numerical simulations with state-of-the-art tsunami generation and propagation models, using the best possible bathymetric and topographic data, to re-evaluate the hypothesis of a dual source, co-seismic and a landslide (itself triggered by the earthquake, possibly with some delay), suggested by many. Such a source would have generated two sets of tsunami waves, mpacting the coast at different locations and time, as is supported by many observations. According to various authors who have studied the surface deformation on the two river banks of the Messina Straits, the fault responsible for the 1908 tsunami is located in the middle of the Straits and is not directly visible from the surface. From the geometry of some faults proposed in the literature, the vertical component of the co-seismic displacement was calculated with Okada’s (1985) method and used to initialize tsunami propagation from the earthquake, and show that a seismic source alone could not have generated the 1908 tsunami as it was observed. Afterword, the dual source hypothesis formulated by Favalli et al. (2009) was simulated. Results of these simulations were systematically compared to all the available observations of coastal inundation, runup, and arrival time in order to obtain useful elements leading to the formulation of a new hypothesis. Furthermore, it was demonstrated the inconsistency between the simulation results and the historical observations, as well as some assumptions about the dynamic of the event. At the same time, by using historical information collected by researchers at the time, and in particular the tsunami arrival time on the coast, it was possible, by a process of backward tracing of the tsunami waves, to identify the likeliest location of additional sources required to explain the tsunami coastal impact, which have been associated in the literature with underwater landslides. Because the generated wave trains are comprised of both longer and shorter waves, the latter associated with landslides, the fully nonlinear and dispersive long wave model FUNWAVETVD (Shi et al., 2012; Kirby et al., 2013; Grilli et al., 2015) is used to simulate tsunami propagation, in a series of nested grids of increasingly fine resolution, by one-way coupling. TVD refers to the “Total Variation Diminishing” shock-capturing algorithm that is used to more accurately simulate breaking waves and the associated dissipation and coastal inundation. For landslide tsunami generation, it was used the three-dimensional (sigma-coordinate) non-hydrostatic model NHWAVE (Ma et al., 2012), which solves the incompressible Navier-Stokes equations in conservative form. Both models are parallelized using the domain decomposition technique, the Message Passing Interface (MPI) protocol with non-blocking communication for data communication between processors, that allows for the modelling of large grids in a reasonable time.