Rockfalls predictability in the Séchilienne corridor (France).

A. Schmid, J.-R. Grasso, A. Helmstetter

ISterre, OSUG, Université de Grenoble

We use 5 years of OMIV observatory data on the Séchilienne rockslide to quantify the predictability of rockfalls time of occurrence within the Séchilienne corridor (so-called Ruines de Séchilienne). The four data sets we use are displacements of a few Séchilienne rockslide benchmarks and extensiometers, rainfall history from the neighboring Vizille weather station, time series of micro-earthquakes located within the rockslides (Ml>-2) and time series of rockfalls (20.000 events), as detected by seismic signals. We choose as prediction targets the 100 largest rockfalls that occurred between 2007 and 2012. First, for each of the four observables, we characterize the average forerunner patterns using superposed epoch analysis, which increases the signal/noise ratio that exists before a single event. Second we compare the predictive power of the four observables individually, using the error diagram framework. On average, there is an increase in rockfall rate a couple of days before rockfall occurrence, with a poor dependence to rockfall size. This pattern is not well robust through time when using micro-earthquake time series. When it is resolved on 2007-2009 period it vanishes on the 2010-2012 period, when the rockfall rate increases. Conversely the rainfall rate increases significantly above noise level a few days before the target rockfall occurrences on the 2010-2012 period whereas it is not the case on the 2007-2009 period. The displacement values of benchmark on the rockslide are the weaker forerunners to rockfall occurrence. In this first step study we resolve the predictability increase when decreasing the sampling time window from daily to 1/4 hours, respectively.