Modeling Tsunamis Generated by Submarine Landslides at Stromboli Volcano (Aeolian Islands, Italy): A Numerical Benchmark Study

Abstract

We present a benchmark study aimed at identifying the most effective modeling approach for tsunami generation, propagation, and hazard in an active volcanic context, such as the island of Stromboli (Italy). We take as a reference scenario the 2002 landslide-generated tsunami event at Stromboli simulated to assess the relative sensitivity of numerical predictions to the landslide and the wave models, with our analysis limited to the submarine landslide case. Two numerical codes, at different levels of approximation, have been compared in this study: the NHWAVE three-dimensional non-hydrostaticmodel in sigma-coordinates and theMultilayer-HySEA model. In particular, different instances of Multilayer-HySEA with one or more vertical discretization layers, in hydrostatic and non-hydrostatic formulation and with different landslide models have been tested. Model results have been compared for the maximum runup along the shores of Stromboli village, and the waveform sampled at four proximal sites (two of them corresponding to the locations of the monitoring gauges, offshore the Sciara del Fuoco). Both rigid and deformable (granular) submarine landslide models, with volumes ranging from 7 to 25 million of cubic meters, have been used to trigger the water waves, with different physical descriptions of the mass movement. Close to the source, the maximum surface elevation and the resulting runup at the Stromboli village shores obtainedwith hydrostatic and non-hydrostaticmodels are similar. However, hydrostatic models overestimate (with respect to non-hydrostatic ones) the amplitude of the initial positive wave crest, whose height increases with the distance. Moreover, as expected, results indicate significant differences between the waveforms produced by the different models at proximal locations. The accuratemodeling of near-field waveforms is particularly critical at Stromboli in the perspective of using the installed proximal sea-level gauges, together with numerical simulations, to characterize tsunami source in an early-warning system. We show that the use of non-hydrostatic models, coupled with a multilayer approach, allows a better description of the waveforms. However, the source description remains the most sensitive (and uncertain) aspect of the modeling. We finally show that non-hydrostatic models, such as Multilayer-HySEA, solved on accelerated GPU architectures, exhibit the optimal trade-off between accuracy and computational requirements, at least for the envisaged problem size and for what concerns the proximal wave field of tsunamis generated by volcano landslides. Their application and future developments are opening new avenues to tsunami early warning at Stromboli.