On August 21, 2017, a Md 4.0 earthquake struck the volcanic island of Ischia (Gulf of Naples, Italy), producing heavy damage and two casualties at Casamicciola (Azzaro et al., 2017), locality known for having been destroyed by the famous 1883 event with a life toll of 2300 victims. The macroseismic features of the 2017 earthquake as well as the historical ones, are those typical of the seismicity in volcanic areas: damage affecting small areas, impressive intensity attenuation within very short distances, high epicentral intensities vs moderate magnitudes. Evaluating the ground motion attenuation in these geologic conditions is still a point of issue. The numerous relationships adopted in tectonic domains cannot be used since they produce an excessive overestimation of the expected shaking at a site or, conversely, underestimation of the epicentral intensity when calculated from the instrumental magnitude. For these reasons, it is necessary to tackle the problem with studies specific for each volcanic zone, since they show an extreme variability in the characteristics of the source and propagation of the seismic energy in the shallow crust. A first deterministic approach to the analysis of the macroseismic attenuation in the Italian volcanic districts was performed by Azzaro et al. (2006), who found specific attenuation trends for Etna and Ischia showing the highest decay of intensity with the epicentral distance (ÎI 4 in 20 km). A probabilistic approach to model the attenuation of the macroseismic intensity was later applied to Mt. Etna volcano (Azzaro et al., 2013; Rotondi et al., 2016), exploiting the huge amount of data available from the local historical earthquake catalogue. This procedure allowed estimating the probability distribution of the intensity at a site (Is) conditioned on the epicentral intensity of the earthquake (I0) and the epicentre-site distance through a binomial-beta model (Zonno et al., 2009). In this paper, we apply the same probabilistic approach to model the attenuation of macroseismic intensity at Ischia, a major issue in the seismic disaster prevention dramatically highlighted by the last 2017 earthquake. The application of this procedure at Ischia requires an ad-hoc calibration of some parameters through a reference macroseismic dataset, which is characterised by a seismic catalogue extremely poor for the island. The most updated release of the Parametric Catalogue of Italian Earthquakes (CPTI15, Rovida et al., 2016) reports twelve events with I0 â¥ VI-VII MCS but only three of them have a number of intensity data points suitable to model the intensity decay (Nip â¥10 as for Etna). Therefore, in order to limit the uncertainties due to poorly constrained parameters we selected the 1828, 1881 and 1883 earthquakes; in addition, we used the 2017 event. In all, 78 intensity data spread through the island of Ischia are available for the analysis. The procedure to estimate the parameters of the binomial probability distribution is reported in Rotondi et al. (2015). In short, the procedure consists in i) summarizing the set of epicentre-site distances for each decay value ÎI of the macroseismic fields with fixed I0 through some statistical summaries (e.g. mean, quantiles), ii) collecting this information in a matrix and applying a hierarchical agglomerative clustering method to identify classes of fields with similar attenuation trend. For each class, a beta-binomial probabilistic model was estimated, following a Bayesian approach, so as to obtain the probability distribution p(Is | I0, d) of the intensity Is at any site, conditioned on I0 and the epicentre-site distance d. In the light of the most recent macroseismic database DBMI15 (Locati et al., 2016), the procedure has been repeated starting from the detection of the isoattenuation classes, and then updating the probability distributions p(Is | I0, d). The graphical comparison among the attenuation of the Ischia earthquakes and the one of the isoattenuation Italian classes confirmed the well-known strong attenuation of volcano seismicity; this issue can be overcome by applying a scaling factor k=20 which makes the attenuation trends comparable. This means that what is felt at distance d in the case of Italian earthquakes in class B, is similar to what is felt at distance d/20 in the volcanic district of Ischia. Consequently, the suitably modified probability distributions of that class was used as prior information on the decay at Ischia; then, the intensity data points of the 1828, 1881, 1883 earthquakes were used to update the parameter estimates and the corresponding distributions p(Is). The seismic scenario of the 2017 earthquake has been forecast and compared with the really observed one. The backward validation of the results also includes the comparison with simulations from a deterministic approach. The entire procedure is implemented into a software to generate probabilistic shake maps expressed in terms of macroseismic intensity. This application may also represent a practical tool for the INGV data acquisition centres for obtaining real-time seismic scenarios. Finally, the matrix of values of the predictive probability function of the intensity at site are used, together with observed intensities, to obtain local probabilistic seismic hazard maps for the island, based on the site approach (SASHA code, D'Amico and Albarello, 2008).

### Seismic scenarios and hazard assessment in the island of Ischia (Neapolitan volcanic district, Italy): a probabilistic approach based on macroseismic intensity data

#####
*R Rotondi;E Varini*

##### 2018

#### Abstract

On August 21, 2017, a Md 4.0 earthquake struck the volcanic island of Ischia (Gulf of Naples, Italy), producing heavy damage and two casualties at Casamicciola (Azzaro et al., 2017), locality known for having been destroyed by the famous 1883 event with a life toll of 2300 victims. The macroseismic features of the 2017 earthquake as well as the historical ones, are those typical of the seismicity in volcanic areas: damage affecting small areas, impressive intensity attenuation within very short distances, high epicentral intensities vs moderate magnitudes. Evaluating the ground motion attenuation in these geologic conditions is still a point of issue. The numerous relationships adopted in tectonic domains cannot be used since they produce an excessive overestimation of the expected shaking at a site or, conversely, underestimation of the epicentral intensity when calculated from the instrumental magnitude. For these reasons, it is necessary to tackle the problem with studies specific for each volcanic zone, since they show an extreme variability in the characteristics of the source and propagation of the seismic energy in the shallow crust. A first deterministic approach to the analysis of the macroseismic attenuation in the Italian volcanic districts was performed by Azzaro et al. (2006), who found specific attenuation trends for Etna and Ischia showing the highest decay of intensity with the epicentral distance (ÎI 4 in 20 km). A probabilistic approach to model the attenuation of the macroseismic intensity was later applied to Mt. Etna volcano (Azzaro et al., 2013; Rotondi et al., 2016), exploiting the huge amount of data available from the local historical earthquake catalogue. This procedure allowed estimating the probability distribution of the intensity at a site (Is) conditioned on the epicentral intensity of the earthquake (I0) and the epicentre-site distance through a binomial-beta model (Zonno et al., 2009). In this paper, we apply the same probabilistic approach to model the attenuation of macroseismic intensity at Ischia, a major issue in the seismic disaster prevention dramatically highlighted by the last 2017 earthquake. The application of this procedure at Ischia requires an ad-hoc calibration of some parameters through a reference macroseismic dataset, which is characterised by a seismic catalogue extremely poor for the island. The most updated release of the Parametric Catalogue of Italian Earthquakes (CPTI15, Rovida et al., 2016) reports twelve events with I0 â¥ VI-VII MCS but only three of them have a number of intensity data points suitable to model the intensity decay (Nip â¥10 as for Etna). Therefore, in order to limit the uncertainties due to poorly constrained parameters we selected the 1828, 1881 and 1883 earthquakes; in addition, we used the 2017 event. In all, 78 intensity data spread through the island of Ischia are available for the analysis. The procedure to estimate the parameters of the binomial probability distribution is reported in Rotondi et al. (2015). In short, the procedure consists in i) summarizing the set of epicentre-site distances for each decay value ÎI of the macroseismic fields with fixed I0 through some statistical summaries (e.g. mean, quantiles), ii) collecting this information in a matrix and applying a hierarchical agglomerative clustering method to identify classes of fields with similar attenuation trend. For each class, a beta-binomial probabilistic model was estimated, following a Bayesian approach, so as to obtain the probability distribution p(Is | I0, d) of the intensity Is at any site, conditioned on I0 and the epicentre-site distance d. In the light of the most recent macroseismic database DBMI15 (Locati et al., 2016), the procedure has been repeated starting from the detection of the isoattenuation classes, and then updating the probability distributions p(Is | I0, d). The graphical comparison among the attenuation of the Ischia earthquakes and the one of the isoattenuation Italian classes confirmed the well-known strong attenuation of volcano seismicity; this issue can be overcome by applying a scaling factor k=20 which makes the attenuation trends comparable. This means that what is felt at distance d in the case of Italian earthquakes in class B, is similar to what is felt at distance d/20 in the volcanic district of Ischia. Consequently, the suitably modified probability distributions of that class was used as prior information on the decay at Ischia; then, the intensity data points of the 1828, 1881, 1883 earthquakes were used to update the parameter estimates and the corresponding distributions p(Is). The seismic scenario of the 2017 earthquake has been forecast and compared with the really observed one. The backward validation of the results also includes the comparison with simulations from a deterministic approach. The entire procedure is implemented into a software to generate probabilistic shake maps expressed in terms of macroseismic intensity. This application may also represent a practical tool for the INGV data acquisition centres for obtaining real-time seismic scenarios. Finally, the matrix of values of the predictive probability function of the intensity at site are used, together with observed intensities, to obtain local probabilistic seismic hazard maps for the island, based on the site approach (SASHA code, D'Amico and Albarello, 2008).I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.