The proportional hazard model as applied to the CSEP forcasting area in Italy

Abstract

This study presents the strategies adopted to modify the Proportional Hazard Model to fit the requirements for forecasting testing within the Collaboratory Study for Earthquake Predictability (CSEP) experiment. The model was originally proposed to study the spatiotemporal distribution of M 5.5+ seismicity in Italy, through two spatial models: a regular grid, and a seismotectonic zonation. A prospective 10-year-forecast test has already been ongoing since 2005, and the results are available on the internet (http://earthquake.bo.ingv.it). For that test, we have reported the probability maps of M5.5+ earthquakes for the next 10 years for the two spatial models. As the original model is time-dependent, it is updated every year, and also immediately after the occurrence of a target event, e.g., Mw 5.5. Although this prospective test is continuing and the model updates probabilities that are different from those of the CSEP experiments, we argue that a full evaluation of the model can only be achieved through this CSEP testing, where the performances of different models are compared using the same rules and tests. The major modification we have introduced into our model is the simulation of the expected numbers of events in the exposure time Dx. This is performed considering the probability that an event occurs in Dx, and evaluating the change this will cause in the expected number of events. This procedure is also implemented for the first and second generation of aftershocks. © 2010 by the Istituto Nazionale di Geofisica e Vulcanologia.