A new probabilistic seismic hazard model, called Modello di Pericolosità Sismica 2019 (MPS19), has been recently proposed for the Italian territory, as a result of the efforts of a large national scientific community. This model is based on 11 groups of earthquake rupture forecast inputs and, particularly, on 5 area-source seismogenic models, including the so-called MA4 model. Data-driven procedures were followed in MA4 to evaluate seismogenic parameters of each area source, such as upper and lower seismogenic depths, hypocentral-depth distributions, and nodal planes. In a few cases, expert judgement or ad hoc assumptions were necessary due to the scarcity of data. MA4 consists of 20 seismicity models that consider epistemic uncertainty in the estimations of the completeness periods of the earthquake catalogue, of maximum magnitude values and of seismicity rates. In particular, five approaches were adopted to calculate the rates, in the form of the truncated Gutenberg–Richter frequency–magnitude distribution. The first approach estimated seismicity rates using earthquakes located in each area source, while the other approaches firstly calculated the seismicity rates for groups of areas considered tectonically homogeneous and successively partitioned in different ways the values to the area forming each group.

The results obtained in terms of seismic hazard estimates highlight that the uncertainty explored by the 20 seismicity models of MA4 is at least of the same order of magnitude as the uncertainty due to alternative ground motion models.

A recent project led by the Seismic Hazard Centre (Centro di Pericolosità Sismica, CPS) of the Italian Istituto Nazionale di Geofisica e Vulcanologia (INGV) and funded by the Italian Civil Protection Department produced a new time-independent probabilistic seismic hazard assessment (PSHA) model for Italy, called Modello di Pericolosità Sismica 2019 (MPS19; Meletti et al., 2021). The model consists of a suite of earthquake rupture forecasts (ERFs) and ground motion models (GMMs), described in Visini et al. (2021) and Lanzano et al. (2020), respectively, that are based on updated and new data acquired in the last decade after the release of the current reference Italian seismic hazard model in 2004–2006 (MPS04, Modello di Pericolosità Sismica 2004; Stucchi et al., 2011), which is currently adopted as seismic input in the Italian building code (NTC, 2018).

In particular, MPS19 consists of 564 alternative seismic hazard models (i.e. logic-tree branches) obtained by combining 11 groups of ERFs, each made by a different number of sub-models (for a total of 94 ERFs) to consider the epistemic uncertainty inside each group, with a set of six GMMs (three for active shallow crustal regions, two for subduction zones and one for volcanic areas). In terms of seismic source typologies, 5 groups of ERFs out of 11 consider area sources: 2 are based on smoothed seismicity calculated on a grid of points; 2 combine faults sources with background seismicity; and 2 derive earthquake rates from geodetic data over a grid of points. The ERFs are based on updated and new historical, geological, geodetic and palaeoseismological datasets collected over the last 15 years, since the realization of MPS04, for the Italian territory and its conterminous areas.

In this paper we describe one of the five area-source ERFs that is the so-called MA4 model (i.e. area-source model no. 4, modello ad aree no. 4 in Italian), based on seismogenic zoning ZS16 (zonazione sismogenetica 2016). ZS16 represents the evolution of previous area-source models proposed in the last 30 years as unique area-source input in seismic hazard assessment performed in Italy; these models are all based on the same seismotectonic approach to seismogenic zoning described by Meletti et al. (2000).

The ZS16 of the MA4 model incorporates a number of different parameters for
each defined area source: (a) geographical boundaries, (b) top and bottom
depth of the seismogenic layer, (c) hypocentre distribution, and (d) style of
faulting. For each area source of ZS16, MA4 uses five alternative
frequency–magnitude distributions, providing the annual rates of all
earthquakes with

In the following, we first briefly introduce the input data considered for developing the ZS16 and the MA4 model and then describe the methods used to define the geometry of area sources and to estimate, for each of them, the top and bottom depth of the seismogenic layer, hypocentre distribution, style of faulting, and annual rates of earthquake occurrence. Finally, seismic hazard estimates computed using the MA4 model are presented and discussed.

The MA4 ERF is based on the area-source approach for defining seismicity parameters. The choice operated in this work was to update previous zonings designated for previous seismic hazard assessment in Italy. Since 1990, several seismogenic zonings have been released, adopting the seismotectonic approach described in Meletti et al. (2000). In the first half of the 1990s, the first zoning adopted in a seismic hazard computation for the whole national territory was ZS4 (zonazione sismotettonica vers. 04, Meletti et al., 2000, shown in Fig. 1), used for the PS4 (pericolosità sismica vers. 04) hazard model (Slejko et al., 1998). ZS4 was delineated interpreting the seismicity in terms of tectonic regimes and, as a second-order criterion, the spatial variation in seismicity; the model was constituted by 80 area sources.

At the beginning of the 21st century, an updated version of the area-source zoning (ZS9; Meletti et al., 2008, shown in Fig. 1) was defined for the elaboration of the MPS04 seismic hazard model (Stucchi et al., 2011). ZS9 resulted from modifications, merges and eliminations of the numerous areas delineated in the previous zoning of ZS4, as well as from the introduction of new ones. The goal of ZS9 was to build a model consistent with new data collected at the time of its development.

Since most of that knowledge was considered still reliable during the development of ZS9, this later was built without introducing substantial novelties as regards the general kinematic framework on which ZS4 was based. In some cases, groups of area sources of ZS4 were merged on the basis of the characteristics of the kinematic domain to which each of the area sources was attributed. In the meantime, the geometry of the area sources was modified according to the changed seismotectonic knowledge. Most importantly, in ZS9 area sources were designed strictly enveloping the seismogenic sources that were at that time known and defined in the DISS database (Database of Individual Seismogenic Sources; Basili et al., 2008). In ZS4, on the contrary, the areas extended over the known seismogenic sources, including regions where faults were not mapped, according to what was thought to be a cautionary criterion. We should consider that, for PSHAs using an area-source model, the seismicity rates computed using faults and earthquakes located inside the area are equally spaced in a grid of point sources where each point has the same seismicity occurrence properties (i.e. rate of events generated). The arbitrary increase in the surface of some of the area sources of ZS4, therefore, led to a reduction in the hazard estimate of PS4 in those areas. ZS9 was then developed by constraining the geometry of the area sources to the location of seismogenic faults and historical and instrumental earthquakes, avoiding arbitrary extensions of the dimensions of the area sources. Figure 1 shows the ZS9 model, consisting of 36 area sources, together with ZS4.

The development of ZS16 was driven by the choice to update the area-source model of ZS9 only where new data suggest different interpretations. To summarize, as shown in Fig. 1, ZS16 updates ZS9 and constitutes the base for the MA4 ERF. MA4, along with other ERF models, was used as one of the inputs of MPS19 (Meletti et al., 2021; Visini et al., 2021).

Area sources for PSHA represent regions with seismicity spatially uniform in
terms of earthquake occurrence rates, maximum magnitude, expected rupture
mechanism and so on. In our model, mapped active faults played a major role
in defining the boundaries of the area sources; however we integrated
geological data with historical and instrumental seismicity, as well as with
geophysical data, including geodetic strain field, maximum horizontal stress
(Sh

To determine the boundaries and the seismic parameters of the area sources, we collected and analysed several seismotectonic datasets (Fig. 2), some of which were compiled in the framework of MPS19 (Meletti et al., 2021) to be used as common inputs for the development of all the ERFs. Among these datasets, we used a historical earthquake catalogue (Catalogo Parametrico dei Terremoti Italiani, version 1.5, hereinafter CPTI15; Rovida et al., 2016, 2020); an instrumental earthquake catalogue (Gasperini et al., 2016; Lolli et al., 2020); version 3.2.1 of the Database of Individual Seismogenic Sources (DISS 3.2.1; Basili et al., 2008; DISS Working Group, 2018); a harmonized GPS velocity model for the Mediterranean area (Devoti et al., 2017); and other geological and geophysical data, available for specific regions and for the whole territory, as described in the following. Data used to draw boundaries of area sources of ZS16, earthquake catalogues and a brief description of the area sources are available on request from the corresponding author.

Main datasets

CPTI15 v1.5 lists 4389 earthquakes with moment magnitude

Although different methods for identifying mainshocks are available in the literature, within the MPS19 project the widely used and tested procedure by Gardner and Knopoff (1974) with the space and time windows defined therein was selected. The procedure resulted in a catalogue of 3353 mainshocks, corresponding to 76 % of the whole CPTI15, which was used in all the ERFs of MPS19.

To define the seismogenic layers and the depth distributions of the
earthquakes in the area sources, we also considered an instrumental
catalogue with homogeneous

To collect a representative dataset useful to define the styles of faulting
of each area source, we started from the Italian CMT (centroid moment tensor) dataset (Pondrelli et
al., 2006; Italian CMT dataset available at

Active faults played an important role in defining the boundaries of the area sources; to this aim we consulted databases referring to a different scale of resolution, from the national to local scale, to include both the general seismotectonic picture and the details in the boundaries.

The DISS database (Basili et al., 2008; DISS Working Group, 2018) is a fundamental product for interpreting the relationships between faults and earthquakes in Italy. DISS 3.2.1 contains 127 individual seismogenic sources (defined as a simplified and three-dimensional representation of a fault plane; individual seismogenic sources are assumed to exhibit “characteristic” behaviour with respect to rupture length/width and expected magnitude), 188 composite seismogenic sources (defined as simplified and three-dimensional representations of crustal faults containing an unspecified number of seismogenic sources that cannot be singled out, Fig. 2c), 35 debated seismogenic sources and 3 subduction zones. All sources are based on geological/geophysical data and cover the whole Italian territory and portions of adjacent countries and seas.

At the national scale, we also considered the Structural Model of Italy (CNR, P. F. GEODINAMICA, 1990) and the seismotectonic model by Meletti et al. (2000). The latter was used as a guide for identifying homogeneous domains of active tectonics in Italy.

In some regions, we integrated the above datasets with data from local detailed geological–structural investigations to define the boundaries of the area sources, for example: Delacou et al. (2004) and Sue et al. (2007) for northwestern Italy; Collettini and Barchi (2002), Boncio et al. (2004), Papanikolaou and Roberts (2007), Lavecchia et al. (2007a), Faure Walker et al. (2010, 2012), Visini (2012), Tesson et al. (2016), and Valentini et al. (2017) for central and southern Italy; and Lavecchia et al. (2007b), Catalano et al. (2010), Billi et al. (2010), Visini et al. (2010) and Mastrolembo et al. (2014) for Sicily.

As a proxy for evaluating the thickness of the crust and defining zones with similar seismogenic thickness, we used the Moho maps by Solarino and Cassinis (2007) and Di Stefano et al. (2011) and the heat flow maps by Della Vedova et al. (2001).

We also considered the regional strain rate fields for the Mediterranean
area derived from GPS data (Devoti et al., 2017) and the maximum horizontal
stress Sh

MA4 is based on seismogenic zoning ZS16, which represents the evolution of previous area-source models proposed in the last 30 years as unique area-source input in seismic hazard assessment performed in Italy. The criteria for defining ZS16 are described in Sect. 4.1; methodologies for the calculation of parameters of the area source, useful for PSHAs, are described in Sect. 4.2 (top and bottom depth of the seismogenic layer), Sect. 4.3 (hypocentre distribution) and Sect. 4.4 (style of faulting). MA4 also models seismicity rates for each area source of ZS16 (Sect. 4.4) and includes epistemic uncertainties in the assessment of the completeness intervals of the earthquake catalogue (Sect. 4.4.1), maximum magnitude (Sect. 4.4.2) and alternative frequency–magnitude distributions (Sect. 4.4.3).

Although area sources are widely used for national and international PSHA,
there are no standard objective approaches for defining their boundaries. We
acknowledge the criteria defined in previous studies (e.g. Giardini, 1999;
Meletti et al., 2008; Wiemer et al., 2009; Vilanova et al., 2014; Danciu et
al., 2018) to set guidelines for the delineation of area sources in order to
describe the correlation between active faults, earthquakes and other
geophysical inputs. To update the existing reference national zoning scheme ZS9, we applied the following criteria.

Start from the area sources of the ZS9 model.

Be consistent with the general background delineated by the geodynamic model proposed by Meletti et al. (2000); i.e. an area source should belong to a unique tectonic zone (active shallow crustal, volcanic or subduction zone in the specific Italian case).

Incorporate all recent advances in the understanding of the active tectonics of the territory and in the distribution of seismogenic sources modelled in the DISS 3.2.1 database and other active fault compilations at the national and regional scale (see Sect. 3.3). In particular, for defining area-source boundaries that primarily follow the surface projection of mapped active faults, an area source should not interrupt a normal or reverse fault system unless major differences are observed (changes in stress orientation and/or changes in crustal depth); for strike-slip faults, boundaries should be parallel to the strike of the faults, and the area source should contain the faults.

Incorporate information derived from the investigation of the most recent seismic sequences that struck Italy after the compilation of ZS9, namely the 2009 L'Aquila, 2012 Emilia and 2016 Amatrice–Norcia sequences.

Be consistent with the spatial pattern of seismicity depicted by the CPTI15 earthquake catalogue. Area sources whose borders were drawn using mapped active faults (point c) should not cross spatial clusters of earthquakes which are attributed to the same faults. As earthquake locations, both macroseismic and instrumental, are affected by uncertainties, a spatial shift between the possible causative faults and the epicentre can occur.

Consider for the definition of the boundaries the pattern of seismicity,
focal mechanisms, geodetic strain field, Sh

Account for the variation in the style of faulting and tectonic regime with depth; therefore multiple area sources can overlap on the volume domain.

Cover the entire Italian territory, as required by MPS19.

Seismogenic depths for each area source of ZS16 were estimated using the instrumental catalogue by Gasperini et al. (2016). In particular, we assumed the upper and lower limits of the seismogenic layer as corresponds to the 5th and 95th percentiles of the depth distribution of the earthquakes inside each area (e.g. Boncio et al., 2009; Stucchi et al., 2011), and we modelled the depth distributions of hypocentres with the peaks of unimodal and bimodal distributions that best approximate the observed values.

To estimate these values, we first removed the earthquakes with fixed
hypocentral depth (i.e. 0, 5 or 10 km), which represents

For depths within the top and the bottom of the seismogenic layer, we
computed the modal values, standard deviation and log likelihood of the unimodal
and bimodal distributions that best fit the observed values (see Fig. 3
for an example). We evaluated and compared the AIC (Akaike information
criterion) index of the unimodal and bimodal distributions to select the
best model for the hypocentral-depth distribution of each area. In the case of
a unimodal distribution, we used the modal value as representative of the
hypocentral depth, while for bimodal distributions, we assigned weights to
both modal values by using their mixing proportion value in the bimodal
distribution. To evaluate the stability of the results with respect to the
number and the magnitude of the considered events, we calculated the upper
and lower seismogenic depths and the modal values of the distributions for
different minimum magnitudes (from

Hypocentral-depth distributions (grey bars) for different threshold magnitudes (reported on top of each panel along with the number of considered earthquakes) for area source no. 24. Black lines correspond to the 5th and 95th percentiles, assumed to be the upper and lower seismogenic depths (round value); blue curve and line represent the unimodal distribution and its modal value; red curves and lines represent the bimodal distribution and its two modal values. Solid lines indicate the best model between uni- and bimodal distributions; dashed lines indicate the other model. The right panel shows the depth ranges of the composite seismogenic sources (CSSs) of DISS 3.2.1 inside the area. The depth of the Moho, from Solarini and Cassini (2007) and Di Stefano et al. (2011), is approximately 30–35 km.

In the Mt Etna region, we assigned earthquakes with hypocentral depth

Pondrelli et al. (2020) defined the criteria to parametrize the styles of
faulting of expected earthquake ruptures and to evaluate their
representativeness in each area source. Using available seismic moment
tensors for relevant events (

Expected style of faulting for each area source (modified from Pondrelli et al., 2020). Plain circles represent random seismic sources. White circles represent 100 % random, while black, purple and yellow circles correspond to reverse, normal and strike-slip random sources, respectively. Colours for cumulative focal mechanisms follow the same criteria. Focal mechanisms with a grey background or plain circles with darker colours are the sources for deeper layers. Black numbers are the percentages of contribution to the final sources when their sum is the expected style of faulting.

To estimate the expected seismicity rates of each area source, we adopted a
time-independent (i.e. Poisson) model for earthquake occurrence. We assumed
that the distribution of the earthquake magnitudes follows the truncated
Gutenberg–Richter (TruncGR; Ordaz, 2004) model that has three parameters:

Two independent sets of completeness time intervals for the CPTI15 catalogue were defined according to (i) the historical approach of Stucchi et al. (2004, 2011) and (ii) the statistical method proposed by Albarello et al. (2001).

The historical approach determines the complete intervals analysing the
local history of a set of sample localities. Based on this knowledge, the
years from which it is unlikely that earthquake effects of a given
intensity are not recorded in the local historical sources were determined.
The catalogue can be considered complete for earthquakes of the same
epicentral intensity (

The statistical completeness intervals were assessed using the procedure of
Albarello et al. (2001) for the same macroregions defined for the
historical approach. The method is very sensitive to the number of
earthquakes considered in each area and magnitude bin, and to ensure the
stability of the results, we selected magnitude bins of 0.46

Plot of time vs. completeness magnitude, defined according to both the historical and statistical approach, in the six macroareas shown in Fig. 5a. Grey circles represent the earthquakes in CPTI15. Hist: historical approach, Stat: statistical approach.

The two methods, although with differences due to their assumptions, provide
comparable results within each macroarea. The difference in the number of
events in the complete periods arises from the intervals assessed with the
statistical approach for low-magnitude bins, which contain more events than
the highest ones, which are longer than those obtained with the historical approach. On
the contrary, the historical approach determines longer complete periods for
the highest-

In conclusion, also taking into account the declustering procedure mentioned above (Sect. 3.1), the catalogues used for calculating annual rates of earthquake occurrences within the MA4 model contain 1800 and 1888 events, obtained, respectively, with the historical and the statistical completeness assessment.

For the definition of the maximum magnitude we used the estimates provided
for MPS19, described in Visini et al. (2021), which were based on the
estimate of the maximum observed earthquake in the earthquake record from
CPTI15. The Italian area was divided into 18 tectonic domains (Fig. 5b), and
the earthquakes listed in CPTI15 were assigned to them, according to their
location. Based on the average error in magnitude estimates for earthquakes
occurring before and after 1980, a minimum value of uncertainty for the

To calculate annual rates of earthquake occurrences for the active shallow
crustal areas, we first excluded from the CPTI15 the events belonging to the
southern Tyrrhenian subduction (i.e. those located in that area with
hypocentral depth larger than 40 km), and we imposed a minimum of 10
earthquakes and at least 2 non-empty classes of magnitude (0.1 bin size) in
each area source to derive stable

We used five different approaches to calculate the seismicity rates for the area sources, which are based on two different assumptions.

The first assumption is that

The second assumption is that

In each macroarea, we compared the observed seismicity rates of
occurrence of the macroarea with the ones of the area sources. We compared
seismicity rates of magnitudes ranging between

To build the TruncGR of each area source in a macroarea, we used

For each area source in a macroarea, the TruncGR uses

For each area source in a macroarea, the TruncGR was computed using

For area nos. 19, 20 and 25, we successively partitioned the five

Figure 7 shows an example of the frequency–magnitude distributions
calculated directly for area source nos. 24 and 36 and for macroarea nos. 6 and 11,
respectively. Whereas in areas of a relatively
high rate of seismicity (as for the extensional areas in the central Apennines, area source no. 24), evaluations of

Example of the frequency–magnitude distributions calculated for area source (AS) nos. 24 and 36. The frequency–magnitude distributions of the macroarea (MA) they belong to are also shown. The area source and the macroarea are shown in Fig. 5b.

Seismic hazard was calculated over the whole Italian territory (including sites located within 5 km outside the borders), adopting the MA4 seismogenic model. For this purpose, 52 area sources were used, considering area source nos. 19, 20 and 25 in the form shallow and deep and discarding area source no. 49 (Mt Etna). In fact, MPS19 defined additional ad hoc ERFs for three specific regions: (a) the Mt Etna volcanic area, which replaces the ERF for area source no. 49; (b) the subduction shallow interface seismicity and deep intraslab seismicity of the Calabrian Arc (spanning from the Ionian Sea to the southern Tyrrhenian Sea across the Calabria region); and (c) the seismogenic sources located outside the area of the CPTI15 catalogue (see Meletti et al., 2021; Visini et al., 2021).

Alternative choices and interpretations about the key elements were embedded
in a logic-tree structure (Kulkarni et al., 1984; Coppersmith and Youngs,
1986; Senior Seismic Hazard Analysis Committee, 1997), which is the
conventional tool to capture the epistemic uncertainty associated with the
input elements of a PSHA model. The adopted logic tree, shown in Fig. 8,
consists of a first branching level accounting for the two alternative
evaluations of the catalogue completeness time intervals described in
Sect. 4.4.1, i.e. one based on historical information (Stucchi et al.,
2004, 2011) and one on the statistical approach by Albarello et al. (2001).
Then, a second branching level considers the two alternative sets of maximum
magnitude

The logic-tree scheme adopted in this study. ZS16 is the
seismogenic zoning. The completeness time intervals for the CPTI15 catalogue
were defined according to the historical approach of Stucchi et al. (2004, 2011) and the statistical method of Albarello et al. (2001).

Seismic hazard was calculated for rock-site conditions (i.e. EC8 site
category A or Vs

Figure 9 illustrates the spatial distribution of mean peak ground
acceleration (PGA) values obtained by applying the weighting scheme in
Fig. 8, for 10 % and 2 % probabilities of exceedance in 50 years, also
referred to as 475- and 2475-year return periods (RPs), respectively. The map for
PGA at 10 % probability of exceedance in 50 years (Fig. 9a) shows the
highest hazard estimates (PGA

Maps of mean values of PGA at 10 %

Hazard curves for PGA for the three selected cities. The curves represent the mean hazard level (black line), the hazard resulting from each of the 60 branches (realizations, grey lines), and the uncertainties expressed through the 16th and 84th percentiles (pct, red lines). The legend in the centre panel refers to all panels.

UHS for 10 % (lower spectra, bright colours) and 2 % (upper spectra, pale colours) probability of exceedance in 50 years for the three selected cities. The mean spectra and the 16th and 84th percentiles are reported. The legend in the first panel refers to all panels.

The map for PGA at 2 % probability of exceedance in 50 years (Fig. 9b)
shows that PGA

Figures 10 and 11, respectively, show hazard curves for PGA and uniform hazard spectra (UHS) for spectral periods from 0.1 to 4 s for the cities of Milan, L'Aquila and Syracuse (see locations in Fig. 9), chosen for exemplificative purposes. Figure 10 illustrates the variability in the expected ground motions in PGA, showing the mean hazard level (black line), the hazard curves resulting from each of the 60 realizations (grey lines) and the uncertainties expressed through the 16th and 84th percentiles (red lines). Figure 11 show the mean and the 16th and 84th percentiles of UHS for 10 % and 2 % probability of exceedance in 50 years.

Figure 12 shows the spatial distribution of the coefficient of variation
(CoV) of PGA values for 10 % and 2 % probabilities of exceedance in 50 years. Recalling that GMMs were weighted and the 60 branches do not have the
same weights, we calculated the CoV as the weighted standard deviation
divided by the weighted mean. CoV

Spatial distribution of the coefficient of variation (CoV) of PGA
values for 10 %

The CoV values in Fig. 12 contain uncertainty related to both the ERF and
the GMMs. We then investigated the relative contribution of both components
of epistemic uncertainty at three selected localities. For each site, hazard
curves for PGA were plotted by distinguishing the 3 groups composed of 20 realizations that use the same GMM for active shallow crustal regions. In
Fig. 13, the three groups are shown by light-red lines for the realizations
that adopt Bindi et al. (2011), and light-green lines and light-blue lines are for
those using Bindi et al. (2014) and Cauzzi et al. (2015), respectively.
Figure 13 also shows the mean probability of exceedance (POE) in 50 years
for the three groups. It can be seen that most of the seismic hazard curves from
the different GMMs are overlapping, and the uncertainty due to the GMMs is
of a similar order of magnitude as the uncertainty related to the ERF. For
Milan and L'Aquila, the mean curves of the three groups overlap in a wide range
of POE, especially for POE

Hazard curves for PGA for the three selected cities. Panels

To focus on the relative contributions of ERF and GMM uncertainties, in
Fig. 14 we illustrate the POE for a series of levels of PGA, by
distinguishing the branches that used a particular GMM. In particular, in
Fig. 14, the POE for each level of PGA shown is normalized to the
maximum, in order to have all values ranging from values

Relative contributions of epistemic uncertainty in ERF and GMM
hazard estimates. Branches that use the same GMM are shown with three colours,
red, green and blue. The

The MA4 seismogenic model in Italy represents a challenge for the construction of an area-source-based model that takes into account the variety of seismotectonic environments, which include spatial and depth variations in the main style of faulting. The MA4 model is based on a seismotectonic zoning (ZS16), defined according to a list of criteria specifically defined to this purpose. However, we cannot exclude the fact that the proposed zoning still contains some a priori bias or, simply, some area sources could not reflect the actual tectonics. Objective criteria to delineate area sources with a quality ranking of the basic data would be an additional step (e.g. Wiemer et al., 2009; Vilanova et al., 2014).

As an example, the debate on “large” vs. “small” areas concerns subjective
choices. Small areas are designed to capture changes in seismicity at the
local scale (e.g.

The ZS16 seismotectonic zoning was recently also adopted in the new European Seismic Hazard Model (ESHM20) as the reference area-source model for Italy (Danciu et al., 2021).

To determine the activity rates of area sources from earthquake data, we
initially used the concept of macroarea to evaluate some parameters, such as
the maximum magnitude and the

MA4 explored a range of sources of epistemic uncertainty and is constituted by 20 alternative ERFs that consider different options in terms of completeness time intervals, maximum magnitude and earthquake rate estimation. We observed that the uncertainty related to this set of ERFs is comparable to or even higher than the uncertainty related to alternative GMMs. Although this observation is not generalizable to other ERFs, it contributes to the discussion on the relative importance of ERFs and GMMs in the overall uncertainty affecting seismic hazard estimates. The GMMs adopted in MPS19 were selected according to statistical criteria and elicitation procedures (Lanzano et al., 2020). We did not test the single ERFs (or branches) of MA4 with statistical procedures, but the mean seismic rates and the associated uncertainty were positively checked against observations (i.e. number of earthquakes occurred in the last centuries) before entering MPS19 (see Meletti et al., 2021, for details). In our opinion, however, trimming ERFs according to their performance against observations contributes to reducing the epistemic uncertainty but could result in a selection of models that is only guided by the earthquake occurrence realization observed in the past.

In this study, we presented a new seismogenic model for Italy, called MA4, that is part of a community-based effort that led to the development of a new seismic hazard model for Italy (MPS19; Meletti et al., 2021). MPS19, in fact, involved more than 150 Italian researchers at various stages of the project, and many of them were involved in the building of their earthquake rupture forecast (ERF). The finally adopted 11 groups of ERFs were composed of a number of alternatives that explore the epistemic uncertainty in seismicity modelling (see Visini et al., 2021). In this framework, MA4 represents 1 of the 5 area-source seismicity models included in the set of ERFs.

The MA4 model is based on an update (ZS16) of the previous seismogenic zoning ZS9 (Meletti et al., 2008) adopted by the current Italian reference seismic hazard model MPS04 (Stucchi et al., 2011). The new seismogenic zoning consists of 48 active shallow crustal area sources and 2 area sources corresponding to the Campanian and Mt Etna volcanic districts (the last of which was not used for computation here shown).

We used five different approaches to calculate the expected seismicity rates
for the area sources, based on two different assumptions. The first assumption
is that the truncated Gutenberg–Richter relation parameter

In conclusion, MA4 consists of 20 seismicity models that consider epistemic uncertainty in the estimations of the completeness periods of the earthquake catalogue, of maximum magnitude values and of seismicity rates. The results show that the uncertainties explored by the 20 different seismicity models is at least of the same order of magnitude of the uncertainty due to the different GMMs. We encourage future studies to account for such uncertainty propagation.

Data used to draw boundaries of area sources of ZS16, the shapefile of ZS16, earthquake catalogues and a brief description of the area sources are available from the corresponding author on reasonable request.

The supplement related to this article is available online at:

FV wrote the draft of the paper, prepared the codes for the analysis and ran the hazard calculations. All the authors contributed to designing the boundaries of ZS16, parametrizing the MA4 model, revising seismic hazard results and writing the final version of the manuscript.

The contact author has declared that none of the authors has any competing interests.

Publisher’s note: Copernicus Publications remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

We acknowledge the willingness of Michele Simionato to help us with OpenQuake and Francesco Martinelli for the development of the server and cluster for OpenQuake. This study has benefited from funding provided by the Italian Presidenza del Consiglio dei Ministri – Dipartimento della Protezione Civile (DPC). This paper does not represent DPC official opinion and policies.

This research has been supported by the Dipartimento della Protezione Civile (Convenzione B1 DPC – INGV).

This paper was edited by Oded Katz and reviewed by Julian Garcia-Mayordomo and two anonymous referees.