The attenuation of macroseismic intensity in Italy: A probabilistic approach to seismic scenarios and hazard assessment

A new Italian seismic hazard map is going to be developed in the framework of an agreement between the Italian National Institute of Geophysics and Volcanology (INGV) and the Department of Civil Protection, to update current map released in 2004. This project involves many national institutions with the aim of delivering a largely shared probabilistic seismic hazard assessment, based on current knowledge and on established methodologies. We present our contribution to this project as for the development of a new Italian hazard map in terms of macroseismic intensity, a measure of earthquake severity which is worthy of consideration in Italy. A large and accurate Italian Macroseismic Database is published and periodically updated by INGV: the latest version, DBMI15, has been released in July 2016, covers the time‐window 1000‐2014 and contains 122,701 macroseismic data points related to 3,212 earthquakes (Locati et al., 2016; doi: http://doi.org/10.6092/INGV.IT-DBMI15). Italy has a high level of vulnerability that is not expected to decrease drastically over time because of the great historical, artistic and monumental heritage spread throughout the territory; for this reason, here macroseismic data represent important reference knowledge and clear added value to instrumental data. Rotondi and Zonno (Ann. Geophys., 2004) proposed to estimate the probability distribution of the intensity at a site, conditioned on the epicentral intensity and on the epicentre-to-site distance, by using a beta-binomial model. The choice of the binomial distribution is predicated on respecting as far as possible the ordinal nature of the intensity scale. During the 2012-2013 EU project Projects on Preparedness and Prevention “UPStrat-MAFA” (Grant Agreement No. 230301/2011/613486/SUB/A5), we applied this probabilistic model to macroseismic fields of different European seismic regions with the general aim of contributing to implement common strategies to forecast damage scenarios from macroseismic fields and to assess seismic hazard in different European countries. In the present study, first we update the estimated beta-binomial model based on the latest database DBMI15 under the assumption of isotropic decay. Then we propose how to include the estimated probabilistic attenuation model in the calculation of the seismic hazard map for areal sources. We select a subset of 538 macroseismic fields of good quality from DBMI15 to be used as learning set (related to earthquakes since 1500, with epicentral intensity at least V, and having at least 40 felt reports). Then we group macroseismic fields with similar attenuation in classes by applying the hierarchical agglomerative clustering method known as Ward method (Kaufman and Rousseeuw, Wiley, 1990). The best performance, according to the agglomerative coefficient, is obtained when four attenuation classes are assumed. The four classes contain 362, 108, 47, and 21 macroseismic fields, respectively, and are denoted by A, B, C, and D in decreasing order of steepness (from the steepest attenuation trends to the flattest ones). We perform a Bayesian analysis of the beta-binomial model fitted to each of the four attenuation classes and for each epicentral intensity in the range from V to XI. As for epicentral intensities X and XI, we notice that there is no data in classes C and D and lack of information in class B. According to three validation criteria (logarithmic scoring rule, log-odds score and logarithmic discrepancy), we check that the assignment of few macroseismic fields to class B is poorly supported. Therefore, a unique decay trend on national scale is reasonably assumed for both X and XI epicentral intensity, corresponding to that obtained for class A. As for epicentral intensities V-IX, keeping in mind the goal of setting an attenuation model for areal sources, it is not possible to clearly identify regions in which an attenuation class prevails over others. Therefore, the model for the attenuation decay is defined as a mixture of the estimated binomial distribution of each class, weighted by the proportion of macroseismic fields that are assigned to that class. Nevertheless, we exclude the case of a unique nationwide mixture model because, by applying the Kruskal-Wallis test, we found that there is significant difference in the spatial distribution of the four attenuation classes along the Italian peninsula. Due to the absence of earthquake epicentres of class D in central Italy, we decided to divide the Italian territory into three polygonal regions, roughly corresponding to northern, central, and southern Italy. By applying again the Kruskal-Wallis test to each region, no significant differences are observed in the spatial distribution of the attenuation classes within each region. To conclude, a mixture model for each epicentral intensity from V to IX and for each of the three polygonal regions (northern, central, and southern Italy) is proposed. Weights associated to classes A, B, C, and D are 0.54, 0.23, 0.16, and 0.07 in northern Italy, 0.76, 0.18, 0.05, and 0.01 in central Italy, and 0.68, 0.21, 0.08, and 0.03 in southern Italy.