Quantification des perturbations non gaussiennes dans le signal simique continu

International audience; Several methods classically exist for capturing perturbations (earthquake, explosion...) in the continuous seismic signal. Most of them, like the famous STA/LTA, detect seismic events by taking advantage of the impulsive nature of their waveform compared to the background noise. In contrast, very few are able to point out long period transient signals (typically greater to 5 s period). However, these long-period waveforms often exist in the continuous seismic signal due to technical problems (as the so called glitches) or to external forcings (e.g. pressure drops, acoustic waves). We present here a novel method that quantifies the strength of the gaussianity of the seismic signal in a given time window. It relies on the basic assumption that the ground motion oscillations balance around an equilibrium position. After centering/sorting operations, the computation of the distance between the 1st and the 4th quarters and a reference curve enables to produce two statistic parameters, called α+ and α- which, in turn, give an asymmetry parameter and an overall gaussian modification estimator. Using a classical sliding window strategy, we show that this method is able to quantify the Gaussianity of various types of smooth perturbations in a noisy background signal, unlike classical estimators such as the skewness. Some examples using both terrestrial and martian seismic data are shown. Finally, based only on statistical thresholds we propose a new way to detect non-Gaussian perturbations.