Séminaire en préambule des journées GFZ/ISTerre précédé par une introduction de Michel précédé par une introduction de Michel Campillo

Predicting seismic shaking and earthquakes impact in the era of big data

Fabrice Cotton

GFZ (German Research Centre for Geosciences) and University of Potsdam.

The amount of seismological data is increasing exponentially. In addition, new types of data (felt reports collected by smart phones, fiber optics records) are becoming available. These new data offer new perspectives for the analysis of the factors controlling seismic ground motions, the prediction of seismic shaking and the evaluation of the impact of earthquakes. We now have about ten years of experience in the use of machine learning methods for seismic motion prediction and I will first illustrate their evolution (from simple neural network methods to predict simple seismic vibration parameters to more complex methods to predict the 3D response of sedimentary basins, the temporal trace of seismic motion or follow the temporal evolution of ground-shaking characteristics). The potential (but also limitations) of these methods as well as efforts to improve their acceptability in the context of engineering seismology or to test/couple machine learning and numerical simulation methods will be discussed. Finally, I will give examples of the potential use of "non-seismological" data to record “strong” seismic shaking and evaluate the impact of large earthquakes (observations of fiber-optic recording of strong earthquakes and use of the felt reports collected by the smartphone application Lastquake for earthquake rapid response )