A double branching model for earthquake occurrence

The purpose of this work is to put forward a double branching model to describe the spatiotemporal earthquake occurrence. The model, applied to two worldwide catalogs in different time-magnitude windows, shows a good fit to the data, and its earthquake forecasting performances are superior to what was obtained by the ETAS (first-step branching model) and by the Poisson model. The results obtained also provide interesting insights about the physics of the earthquake generation process and the time evolution of seismicity. In particular, the so-called background seismicity, i.e., the catalog after removing short-time clustered events, is described by a further (second-step model) branching characterized by a longer time-space clustering that may be due to long-term seismic interaction. Notably, this branching highlights a long-term temporal evolution of the seismicity that is never taken into account in seismic hazard assessment or in the definition of reference seismicity models for a large earthquake occurrence. Another interesting issue is related to the parameters of the short-term clustering that appear constant in a different magnitude window, supporting some sort of universality for the generating process. Copyright 2008 by the American Geophysical Union.