Mining Macroseismic Fields to Estimate the Probability Distribution of the Intensity at Site

Abstract

The analysis of the seismic attenuation is a prominent and problematic component of hazard assessment. Over the last decade it has become increasingly clear that the intrinsic uncertainty of the decay process must be expressed in probabilistic terms. This implies estimating the probability distribution of the intensity at a site Is as the combination of the distribution of the decay DI and of the distribution of the intensity I0 found for the area surrounding that site. We focus here on the estimation of the distribution of DI. Previous studies presented in the literature show that the intensity decay in Italian territory varies greatly from one region to another, and depends on many factors, some of them not easily measurable. Assuming that the decay shows a similar behavior in function of the epicenter-site distance when the same geophysical conditions and building vulnerability characterize different macroseismic fields, we have classified some macroseismic fields drawn from the Italian felt report database by applying a clustering algorithm. Earthquakes in the same class constitute the input of a two-step procedure for the Bayesian estimation of the probability distribution of I at any distance from the epicenter, conditioned on I0, where DI is considered an integer, random variable, following a binomial distribution. The scenario generated by a future earthquake is forecast either by the predictive distribution in each distance bin, or by a binomial distribution whose parameter is a continuous function of the distance. The estimated distributions have been applied to forecast the scenario actually produced by the Colfiorito earthquake on 1997/09/26; for both options the expected and observed intensities have been compared on the basis of some validation criteria. The same procedure has been repeated using the probability distribution of DI estimated on the basis of each class of macroseismic fields identified by the clustering algorithm.