Statistical methods for middle-term forecast of earthquake occurrences

Investigation into possible precursors of strong earthquakes constitutes a challenging research topic which is carried out mainly in two directions: the one based on the analysis of physical parameters and the one based on statistical methodologies. In the first, recent studies have shown significant correlation between major earthquakes and anomalies of different physical parameters measured in the atmosphere/ionosphere which cover time intervals of months.

On the contrary in this presentation we focus on the statistical modelling of the parameters that constitute an earthquake record in a catalog (location, time, magnitude) and we show that significant variations are observed in the months/years preceding a strong earthquake. In particular we consider the spatial distribution of a set of earthquakes and its temporal variations by modelling the area of Voronoi cells generated by the epicenters through a generalized Pareto (GP) distribution. Following the Bayesian paradigm we analyze the recent seismicity of the central Italy and we compare the posterior marginal likelihood of the most promising distributions in shifting time windows. We point out that the best fitting distribution varies over time and the trend of the GP distribution and of other distributions among the most studied in the literature converges to that of the exponential distribution a few months before the start of the preparatory phase to the main shock.