<p><span>Seismology allows continuous recording of the activity of gravitational instabilities whatever the context, and is therefore able to provide a tool for the study of the spatio-temporal evolution of the activity of gravity instabilities with a unique resolution. Due to the considerable fall in the costs of the means of acquiring seismological data and the increasing densification of global, regional and local networks observed in recent years, the amount of data to be processed is growing exponentially. Thus access to information is more and more complete but in return the volume of data to be processed becomes considerable. To analyze this volume of data and extract relevant information, it is necessary to develop automatic methods of identification of seismic sources and location to quickly build the most complete seismicity catalogs possible. </span></p><p><span>We present a new machine-learning based method for automatically constructing catalogs of gravitational seismogenic events from continuous seismic data. We have developed a robust and versatile solution, which can be implemented in any context where seismic detection of landslides or other mass movements is relevant. The method is based on spectral detection of seismic signals and the identification of sources with a machine learning algorithm. Spectral detection detects signals with a low signal-to-noise ratio, while the Random Forest algorithm achieves a high rate of positive identification of seismic signals generated by landslides and other seismic sources. The processing chain is implemented to operate in parallel in a high-performance data center, which allows years of continuous seismic data to be explored and a database of events to be rapidly built up. This solution is also deployed for near-real time seismicity catalogs construction in the framework of slow moving landslides monitoring done by the Observatoire Multidisciplinaire des Instabilit&#233;s de Versants (OMIV). Here we present the preliminary results of the application of this processing chain in different contexts, locally for the monitoring of slow-moving landslides (La Clapi&#232;re, Super-Sauze, S&#233;chilienne), and at the regional level for the detection of large landslides field (Alaska and Alps).</span></p>
