Model Space Exploration for Determining Landslide Source History from Long-Period Seismic Data
Résumé
The seismic signals generated by two large volcanic debris avalanches (Montserrat, Lesser Antilles, 1997 and Mount St. Helens, USA, 1980) and a large rock-ice avalanche (Mount Steller, USA, 2005) have been analyzed. For the two debris avalanches, given the times and locations of such landslides, their signals were recorded by only a few seismic stations. Moreover, these signals cover only a very narrow frequency band and include considerable noise. The Mount Steller, on the contrary, was precisely recorded. For each event, the source mechanism (i.e., point force) has been determined by waveform inversion using at most two broadband seismic stations. The resulting force is very difficult to interpret in terms of landslide characteristics. A Monte-Carlo inversion was therefore performed by imposing a simple force model associated with the landslide, based on the schematic view of an accelerating/decelerating mass traveling down the slope. The best parameter set of the force model was then found by minimizing misfits and maximizing correlations between data and synthetic signals. This model appears to contain the minimum degree of complexity required to well reproduce the seismic data. We detail here the method for the Montserrat debris avalanche and then present it’s validation on the well studied Mount St. Helens debris avalanche and the well recorded Mount Steller rock-ice avalanche. The horizontal and vertical components of the resulting force have different source time functions. The best force model compares well with the force obtained by waveform inversion. Finally, this simple force model was interpreted using analytical and empirical relations derived from the sliding block model, granular flow model and landslide studies. This made it possible to estimate the order of magnitude of the mass, flow duration and direction, initial topography slope, mean velocity and travel distance of the avalanches. For these three avalanches, the calculated characteristics are consistent with former studies.