We evaluate the PSHA of the Campotosto region by modelling fault sources. The code used is Openquake (Pagani et al., 2014), which provides hazard curves for the desired Ground Motion Model (GMM), and uniform hazard response spectra (UHRS) for selected sites. The fault model is based on the 3-D fault system given in the Fault2SHA Central Apennines Database (CAD) (Faure Walker et al., 2020); rupture rates are parameterized by using the SHERIFS code (Chartier et al., 2019) that allows seismic ruptures involving multiple sections of a single fault or interactions between multiple faults.
The Fault2SHA code CAD (Faure Walker et al., 2020) brings together the knowledge of field geologists and seismic hazard modellers, and facilitates the definition of PSHA uncertainties.
The SHERIFS code requires the faults’ geometries and the slip rates to calculate rupture rates. Given a fixed shape of the magnitude-frequency distribution (MFD) set for the entire fault system (target MFD), its iterative procedure establishes all possible earthquake rupture scenarios in the area. At each iteration, a magnitude is randomly selected based on the target MFD, and an earthquake rupture scenario that generates that magnitude is selected. For this scenario, an increase in the slip-rate budget of the affected faults is converted into an earthquake rupture rate for the magnitude under consideration. At the beginning of the iterative process, only the shape of the target MFD is set without knowing its absolute value. The iterative computation continues until the fault slip-rate budget is exhausted. The code compares the modelled earthquake rates with the local earthquake catalogue. We use for this comparison the rates based on the declustered Parametric Catalogue of Italian Earthquakes (Rovida et al., 2019; 2020) and the completeness periods of Meletti et al. (2019).
We started by defining the limits of the fault system in an area of about 100 km around Campotosto Lake (Fig. 1). Since the Fault2SHA CAD (Faure Walker et al., 2020) database does not cover the northern part of the study area, we integrated the system with the individual seismogenic sources of the DISS 3.2.1 catalogue (DISS Working Group, 2018), collecting all the information needed by SHERIFS: the geometry (trace, dip and upper and lower seismogenic depth), the slip rate with uncertainties and the kinematics (normal, strike slip, reverse or rake) of the fault sections. We defined the length of the fault sections in order of the dimensions of the seismogenic depth, which was set at 10 km. Faults with a longer length are divided into smaller sections (however, smaller sections can break together). We have assumed that the seismicity of the complete fault system follows the form of a Gutenberg and Richter (1944) type distribution.
We have analysed the uncertainties associated with the fault slip rate, the magnitude scale parameters, and the b-value. In SHERIFS, the uncertainty limits are defined and explored based on the number of user-defined random samples, which in our case were set to 20, each representing a model for each branch of the logic tree (see the electronic supplement Fig. ES1). For each model, a slip-rate value is chosen uniformly within the uncertainty limits associated with each fault in the database; the scale law parameters are independently selected according to a Gaussian distribution within their error limits, and a b-value is set within the range considered by the user, in our case 0.9–1.1. The uncertainties of the two magnitude scale relationships (Wells and Coppersmith, 1994; Leonard, 2010) in estimating the Mmax are accounted for by random sampling of the two-scale relations.
In summary, to account for the epistemic uncertainties, we implemented a logic tree of 120 branches, as shown in Fig. 2: the fault model is unique, as described previously; two magnitude-rupture scaling relationships (W&C94, Wells and Coppersmith, 1994 and Leo10, Leonard, 2010) to calculate the maximum magnitude each fault can accommodate, based on the rupture area for normal faults; twenty random samples within the lower and upper bounds of a centred triangular distribution to explore the epistemic uncertainty of 1) the b-value of the target MFD for the fault network, 2) the maximum magnitude (Mmax) due to the uncertainty in the scaling law and 3) the slip rates of the faults. Also, we use the three GMMs with the highest weights selected by Lanzano et al. (2020) to compute the new Italian seismic hazard map: the Cauzzi et al. (2015), the Bindi et al. (2014), and the Bindi et al. (2011) calibrated on a global, European and Italian database, respectively.
To assess the seismic hazard of the Campotosto Lake area, we use the synthetic seismicity rates resulting from the SHERIFS’ computations as input for the Openquake code (Pagani et al., 2014), using the specific codes of Scotti et al. (2021). In this study, we did not analyze the influence of the different soils and considered the whole area as bedrock (Vs30 > 800m/s).
The most common result of a hazard analysis is the hazard curve, which represents the annual probability of exceeding certain shaking levels (Kramer, 1996) in a fixed period. From the hazard curves obtained for PGA and response spectra computed at different spectral ordinates (e.g. 0.1s, 1s…), it is possible to obtain the uniform hazard response spectra, which have the same probability of exceeding, at all frequencies, for a fixed return period. The Campotosto hazard curves for both the horizontal and vertical components are shown in Fig. 3a and 3b: the dispersion of the individual branches of hazard curves is linked to the different epistemic uncertainties analysed.
Figures 3c and 3d illustrate the horizontal UHRS for the 2475 and 475 years return periods, respectively, obtained from the different GMMs, while Figs. 3e and 3f refer to the vertical components, for which only the Bindi et al. (2011) model is available. The obtained mean and different quantiles spectra are also represented as the weighted average of the various individual spectra of the logic tree. When analysing the logic tree branches, a strong dispersion is noted in PGA, maximum spectral acceleration, and the corresponding period for the horizontal components. The numerical results and the shape of the curve are strongly influenced by the branches of the Cauzzi et al. (2015) model.
In a classical PSHA, all earthquakes contribute to building the seismic hazard of a site. In our analysis, we obtain the scenario that most affects the hazard of a site, expressed in terms of magnitude-distance pair, by disaggregating the probabilistic calculation. The disaggregation allows us to identify, among all the earthquakes considered to compute the seismic hazard of a region, the one that contributes most (McGuire, 1995).
For a 2475 years return period (see the electronic supplement Fig. ES2a), the magnitude-distance pair that most affects the hazard of a site within the first 10 km of distance is MW 6.75 ± 0.25, whereas for a 475 years return period (see the electronic supplement Fig. ES2b), it is MW 5.25 ± 0.25. To evaluate the contribution of each fault section to the hazard, we computed PGA curves for each rupture scenario at the site. The sum of all these hazard curves gives the total hazard at the site (Fig. 4a). The slip values of the fault sections included in the seismic hazard computation are shown in Fig. 4b.
The contribution of each section to the hazard of Campotosto is obtained by summing its hazard curves for each rupture scenario in which it is involved, normalizing the sum to the total hazard curve. Figure 5a shows, for each PGA level, the sections contributing to the probability of exceedance (POE) of that PGA level. In Fig. 5b, we map the contribution of the sections to hazard at the Campotosto site (PGA 0.816 g) and surrounding area, obtained for the 2475 years return period: the hazard is mainly influenced by the nearest sections around our study site (Laga, Assergi and Barisciano sections) with similar contribution, while, for lower PGA levels, the Laga sections become prevalent (Fig. 5a), even if they are not in the highest slip rate classes (Fig. 4b). This is due to the close distance of these sections to the site. This is a significant result obtained from mapping hazard based on faults, compared to seismotectonic zoning approaches that smooth out the hazard in large zones, ignoring local information on active faults.