A0619
Title: Spatio-temporal clustering and classifying of seismic events in Chile
Authors: Orietta Nicolis - Universidad Andres Bello (Chile) [presenting]
Abstract: Chile is one of the most seismic countries in the world, especially due to its particular position above a subduction zone where the Nazca tectonic plate dives down under the South America plate. This movement is responsible for a great number of seismic events, some of these with great magnitudes. Clustering seismic events allow for the characterization of the seismicity process besides defining the spatial region and the temporal window of the seismic events associated with a mainshock. The resulting clusters can be then used in classification models for identifying those events which may be precursors of great earthquakes. A new methodology for automatically clustering and classifying seismic events is proposed. First, a spatio-temporal DBSCAN (ST-DBSCAN) algorithm with a variable semi-supervised $\epsilon$, which maximizes the distance to the nearest $n-$th earthquake, is proposed. Then, a graph neural network considering the spatial and temporal correlations among clusters is trained for classifying events into three categories: foreshocks, mainshocks and aftershocks. The proposed methodology is applied to the Chilean catalogue of seismic events. Three big clusters representing the earthquakes with a magnitude major than 8.0 on the Ritcher scale are taken for validation. Finally, the results are compared with those obtained by other traditional methods.
