GNGTS 2017 - 36° Convegno Nazionale

GNGTS 2017 S essione 2.1 271 Incorporating Long-Range Dependence in the Seismic Process into Probabilistic Seismic Hazard Assessment S. Barani 1 , C. Mascandola 2 , E. Riccomagno 3 , D. Spallarossa 1 , D. Albarello 4 , G. Ferretti 1 , D. Scafidi 1 , P. Augliera 2 , M. Massa 2 1 Dipartimento di Scienze della Terra dell’Ambiente e della Vita, Università di Genova, Italy 2 Istituto Nazionale di Geofisica e Vulcanologia, Sezione di Milano, Italy 3 Dipartimento di Matematica, Università di Genova, Italy 4 Dipartimento di Scienze Fisiche della Terra e dell’Ambiente, Università di Siena, Italy Since the beginning of the 1980s, when Mandelbrot (1982) observed that earthquakes occur on “fractal” self-similar sets, many scientists have investigated the dynamical mechanism leading to self-similarities in the earthquake process. Interpreting seismicity as a self-similar process is undoubtedly convenient to bypass the physical complexities related to the actual process. Self- similar processes are indeed governed by power laws as a consequence of scale invariance. One of the fundamental scaling laws in seismology describes the magnitude-frequency distribution of earthquakes (Gutenberg and Richter, 1944). Such law arises from the self-similarity in the fracturing process (Mandelbrot, 1982; King, 1983; Frankel, 1991; Scholz, 2002). This property, along with the space-time clustering of seismicity (e.g., Kagan and Knopoff, 1980; Kagan and Jackson, 1991) and the mechanics of earthquake interaction (e.g., Stein, 1999; Scholz, 2002), implies that the process of accumulation and release of seismic strain has memory effects (e.g., Lomnitz, 1994). Periods of high release of seismic deformation will be more likely followed by years of higher-than-average seismic strain release. Conversely, seismically quiet periods will tend to be followed by quiet years. This behavior mimics that of fluctuating long-term water storage in reservoirs. Tectonic loading and seismic strain release play the roles of inflow and outflow, respectively. Hence, we expect a certain degree of correlation in time series of seismic strain release. In this work, we examine series of seismic moment release in Italy and worldwide through Hurst’s rescaled range ( R / S ) analysis (Hurst, 1951) in order to study whether and how the temporal scaling of seismic moment varies both spatially and temporally. This is achieved by examining space-time variations of the Hurst exponent ( H ). The Hurst exponent measures the level of correlation (memory) in time series. It takes values between 0 and 1, indicating a persistent behavior if H > 0.5 and an anti-persistent behavior if H < 0.5. A value close to 0.5 indicates a random process with no correlation and no dependence within the time series. To fulfill the scope of work, we analyze both historical time series for single sites and extended areas in Italy and instrumental data sets for specific seismic episodes occurred in the Italian Central Apennines in the last twenty years. Local H values are then compared and contrasted with those obtained for all of Italy and worldwide. Our analysis shows that seismicity is a memory process with a Hurst exponent H ≈ 0.87. Such a value is indeed indicative of long-range dependence (LRD). Moreover, we have found that H is substantially space- and time-invariant, thus proving the universal behavior in the process of seismic strain release with time. The previous findings clearly invalidate the assumption of seismicity as a Poisson process, which is widely used in conventional probabilistic seismic hazard analysis (PSHA). Hence, we developed a new probability model for earthquake occurrence that allows for LRD in the seismic process (Barani et al. , 2017). Unlike the Poisson model, our model allows for dependent events, such as foreshocks and aftershocks, thus avoiding the tricky phase of catalog declustering. The effectiveness of the model has been evaluated through a retrospective test. Specifically, we developed a 30-year forecast map for the Central Apennines representing the probability of exceeding a magnitude 5.5 earthquake since January 1987. Probabilities were then converted into the equivalent 30-year Poisson rate (e.g., Petersen et al. , 2007; Akinci et al. , 2009). Comparing the equivalent 30-year rate with the observed number of earthquakes shows