**Roman to Middle Age Earthquakes Sourced by the 1980 Irpinia Fault: Historical, Archaeoseismological, and Paleoseismological Hints**

#### **Paolo Galli 1,2**


Received: 11 June 2020; Accepted: 22 July 2020; Published: 27 July 2020

**Abstract:** The Italian seismic compilations are among the most complete and back-in time extended worldwide, with earthquakes on record even before the Common Era. However, we have surely lost the memory of dozen strong events of the historical period, mostly in the first millennium CE. Given the lack of certain or conclusive written sources, besides paleoseismological investigations, a complementary way to infer the occurrence of lost earthquakes is to cross-check archaeoseismic evidence from ancient settlements. This usually happens by investigating collapses/restorations/reconstructions of buildings, the general re-organization of the urban texture, or even the abrupt abandonment of the settlement. Exceptionally, epigraphs mentioning more or less explicitly the effects of the earthquake strengthened the field working hypothesis. Here, I deal with both paleoseismological clues from the Monte Marzano Fault System (the structure responsible for the catastrophic, Mw 6.9 1980 earthquake) and archaeoseismological evidence of settlements founded in its surroundings to cast light on two poorly known earthquakes that occurred at the onset and at the end of the first millennium CE, likely in 62 and in 989 CE. Both should share the same seismogenic structure and the size of the 1980 event (Mw 6.9).

**Keywords:** Irpinia fault; historical earthquakes; archaeoseismology; paleoseismology

#### **1. Introduction**

As widely demonstrated by several works, the Irpinia fault—which is the popular name of the Monte Marzano Fault System (MMFS, [1])—was responsible in November 1980 for a devastating earthquake (Mw 6.9), which was accompanied by the longest surface faulting ever observed in Italy (>30 km), at least before the recent central Apennines event (October 2016, Mw 6.6). Early paleoseismological studies [2] claimed that the recurrence time for 1980-like characteristic earthquakes during the Holocene was approximately 2 kyr, which is a value clashing with both the frequent destructive seismicity of the area affected by the fault [3] and the high GPS-derived strain rate of the region (at least 2.9 mm/year [4,5]). This inconsistency was fixed through high-precision topographic leveling run across the compound fault scarp [6], and new paleoseismological trenches and pits opened on different fault segments of the 43 km long MMFS [7]. Results revealed that most of the destructive earthquakes that hit the upper Ofanto valley in the past two millennia (e.g., events of 1980, Mw 6.9; 1694, Mw 7.0; 1466, Mw 6.3, and many others; as shown in Figure 1) might be ascribed to the MMFS periodical activation, without the necessity of evoking other, unknown faults.

**Figure 1.** Macroseismic epicenters of Mw > 5.5 earthquakes in southern Irpinia (modified from [7]). Notwithstanding possible approximate location, earthquakes cluster mainly in the hanging wall of the Monte Marzano Fault System (MMFS). CF, Caggiano fault; UF, Ufita fault. Stars are the instrumental epicenters of 1980 (Mw 6.9; epicenter from [8]) and 1996 (Mw 5.1) events. Diverging triangles indicate crustal extension, as deduced by GPS analyses published by [4,5].

Macroseismic studies of recent normal faulting earthquakes in Italy showed that the spatial distribution of damage is strongly influenced by rupture directivity effects [9–14], explaining why the same seismogenic source might generate different shaking scenarios time after time. For instance, the 1980 rupture, nucleating from the SE tip of the MMFS toward NW [15], induced the northwestward damage distribution depicted by [16] (Figure 2D). Conversely, in the stronger 1694 earthquake, the rupture nucleated from the NW tip toward SE [7], as proven by the huge damage suffered by many villages located on the eastern side of the highest-intensity data points distribution (from now: HIDD; see Figure 2C).

This paper deals with two little-known earthquakes that were likely generated by the recurring rupture of the Irpinia fault. The first one occurred in the second half of the 1st century CE (Common Era), whereas the second occurred at the end of the 10th century (e.g., 989 CE). Both have left their destructive imprint in some archaeological settlement of the region, and both are mentioned in a few primary historical sources. Moreover, their signature can be read across the stratigraphy exposed within the paleoseismological trenches opened in the past years along the Irpinia fault or scanning the stepped profile of its compound scarp. By joining historical/epigraphical sources, archaeological evidence, and paleoseismic data, I have tried to provide more reliable seismogenic hypothesis and robust parameters for these events.

Here, I will refer to the MCS scale, which is the Mercalli–Cancani–Sieberg Macroseismic scale [17,18] adopted in all the seismic catalogues and works in Italy. Io and Is are epicentral and site intensity, respectively.

#### **2. Historical Seismicity of the MMFS Region**

The historical seismicity of the investigated area is among the strongest of the Apennines, both in terms of frequency and of maximum magnitude, with epicenters mainly concentrated in the MMFS hanging wall (Figure 1). As observed for the longest normal fault systems dissecting the Apennines [19], the largest events of Irpinia (Mw ≥ 6.9) were sourced by the contemporary rupture of the entire MMFS, whereas minor earthquakes occurred on single or grouped segments of this system [7].

Among the latter, the strongest was on January 14, 1466 (Figure 2B), striking the upper Ofanto and Sele Valleys, destroying the town of Conza (Is 9–10 MCS) and causing damage, collapses, and casualties in many other villages in the area, as witnessed by both archaeoseismic indications [20] and contemporary sources [21]. Although we have only 30 intensity points, the 1466 HIDD clearly falls in the hanging wall of the MMFS, roughly mimicking the 1694 and 1980 effects distribution (Figure 2C–D). Nonetheless, as the 1466 event had a lesser destructive impact, it was not sourced by the entire rupture of the MMFS, as happened in the two forthcoming, catastrophic earthquakes. By inverting its HIDD with the Boxer algorithm [22], its equivalent magnitude was rated as Mw 6.3 [7].

**Figure 2.** Distribution of effects induced by the 989, 1466, 1694, and 1980 earthquakes (MCS intensities from [7]). (**A**), 989 event; collapsing columns indicate coeval archaeoseismic evidence (see text). Asterisks, damaged villages quoted by contemporary sources. (**B**), 1466 event (data from [20,23]. (**C**), 1694 event; the arrows near Teora indicate the sector of the slope where contemporary accounts describe the opening of a long chasm, which was probably due to surface faulting [7]. (**D**), 1980 event; yellow circles in the background are 1980 aftershocks. Focal mechanism from [24]. MCS: Mercalli–Cancani–Sieberg Macroseismic scale.

During the 16th century, two damaging earthquakes had again their epicenters in the MMFS hanging wall. On March 29, 1517, the first one caused collapses and victims in Conza, and it was strongly felt from Ariano Irpino to Serre [20]. The second is part of a complex seismic sequence, which included three mainshocks occurring between July 30 and August 19, 1561. Castelli et al. [3] located one of these in the upper Ofanto Valley (Figure 1), where it caused heavy damage in several settlements. However, both the 1517 and 1561 events lack conclusive parameters, as their poor or confusing historical descriptions make it difficult to assess robust epicentral coordinates or magnitude.

Starting from the end of the 17th century, the region suffered the strongest seismic period of its long history. Two strong long-term foreshocks in 1680 (9 November) and 1692 (4 March; Figure 1) paved the way to the catastrophic September 8, 1694 earthquake (Mw 7.0), which caused ca. 6000 deaths, razing to the ground (i.e., Is 10–11 MCS) dozens of villages in the upper Ofanto Valley [25]. Even in 1694, the HIDD focuses on the MMFS hanging wall, with an abrupt intensity decrease in the footwall (Figure 2C). Coeval accounts described an impressive ground rupture formed in the Teora mountains, with a length of 10 Neapolitan miles (18.5 km), suggesting the occurrence of surface faulting along the northern sections of the MMFS, as confirmed by paleoseismological investigations [7].

Leaving aside the catastrophic earthquake of 29 November 1732, which also struck towns located in the upper Ofanto Valley, but with an epicentral area shifted further northwest (Figure 1; Ufita Fault), in 1853 and 1910, the MMFS hanging wall was affected by two other moderate earthquakes. The former (Mw 5.5) caused severe damage in the upper Sele Valley villages (e.g., Caposele and Calabritto [26]), whereas the latter (Mw 5.7) induced collapses and victims mainly in Calitri. The 1853 earthquake can be directly associated with the rupture of a segment of the MMFS, whereas the 1910 was likely sourced by an antithetic fault bounding the hanging wall to the north [7].

At the end, the 23 November 1980 earthquake (Mw 6.9; Io 10) hit the same villages already devastated in 1466 and 1694, destroying 75,000 buildings and severely damaging another 275,000 (Figure 2D). The death toll was 3000, the highest in Italy in the 20th century, after the 1908 (Messina) and 1915 (Fucino) events, with the total destruction of some towns such as Conza (30% of the victims; Is 11), which was abandoned and rebuilt in another place [16]. The physics of this earthquake have been investigated in several studies [27–30] that reconstructed the complex rupture process along both the main fault (mainshock, and 20-s sub-event) and the antithetic, SW-dipping fault (40-s sub-event). The rupture nucleated at ca. 10 km in depth [29] and caused impressive surface faulting across the Mount Marzano massif, especially east of the Sele Valley [7,24,31–34].

#### **3. The Little Known Earthquake of the 1st Century CE**

#### *3.1. Archaeoseismic Evidence from Volcei*

Robust proofs of multiple destructive events have been found in this ancient town, starting from the late 3rd century BCE [35]. The Roman municipium of Volcei (today Buccino) developed over a settlement existing since the Iron Age close to the Apennine watershed, between the Campania and Basilicata regions (Figure 1). The modern Buccino was heavily damaged (Is 8 MCS) by the 1980 Irpinia earthquake (Mw 6.9), with the successive reconstruction works providing an exceptional opportunity to rediscover the buried remains of the Roman town. As Buccino shares the same ruinous destiny as the other villages struck by the 1980, 1694, 1561, and 1466 events [3,23], the recognition of widespread, seismically induced effects to the ancient structures was not an unexpected discovery, and it allowed extending our information well behind the memory of the written Modern sources.

During the post-1980 reconstruction works, many indications of ancient collapses, including *butti* (i.e., stacks of archeological debris), fills, and the leveling of destroyed buildings were found everywhere in the Buccino underground. There was also evidence of restoring and/or rebuilding several Roman houses with architectural elements from previous buildings, such as architraves and epigraphs, which were recycled in the new constructions. Besides all these findings, a coeval epigraph explicitly mentions the restoration made after a collapse due to an earthquake. In the whole, all the indications point to a destructive event falling in the second half of the 1st century CE, as summarized in the forthcoming points (see Figure 3 for sites location), which account mostly for unpublished data collected and discussed with A. Lagi, who was formerly responsible for the Salerno Superintendence for the Buccino excavations, and with many other archaeologists who followed one another in the project.

**Figure 3.** Map of Buccino showing the main archaeological sites attesting the 1st century CE earthquake. Upper panel is the photomosaic of the epigraph of Otacilius Gallus, a far ancestor of mine, attesting the collapse of the *Caesareum* (photo by P.G.).

#### 3.1.1. Forcella Palace

Below this palace, which was built along the Roman Decumanus, the indications consist of a dumping grave containing domestic pottery (lamps and dishes) datable within the first half of the 1st century CE. As no sigillata chiara A pottery (early second half of 1st century CE) was found within the grave, the dumping age must fall at the onset of the second half of the 1st century CE.

#### 3.1.2. Castle

Over the southern side of the main 12th century Norman tower (i.e., the donjon, resting over the basement of a Roman temple), the excavations unearthed a broad, rubble fill, which was supported by a wall. The fill was rich in domestic material, bricks, tiles, and limestone masonry blocks, and it was lined upward by a raw concrete pouring. More importantly, it also contains sigillata italica and Africana chiara A pottery, the latter datable to the second half of the 1st century CE (maximum 60–70 CE). At the bottom of the fill, a coin of Emperor Tiberius (23–30 CE) provides a certain post quem term for the rubbles mass, which was thus leveled between the late 1st and the beginning of the 2nd century.

#### 3.1.3. Sotto San Nicola Street

In the southeastern slope of Buccino, another dumping grave of both building and domestic rubble has been found. It contains vases and lamps datable within the first half of the 1st century CE (i.e., sigillata italica, pareti sottili pottery), being hence coeval with the castle area fill.

#### 3.1.4. Amendola Square

In this place, along the Decumanus, three different dumping graves have been overlaid and sealed with a restoration floor; the infill material contains pottery shards with pareti sottili (early 1st century CE). Amongst all the pottery, the archaeologist found the relic of a pot (glilarium or vivarium in doliis) containing the skeleton of a dormouse (Glis glis), which was ready to be cooked. At least one of the buildings facing the Decumanus was restored in the 2nd century CE, when also a porticus with four pillars was added to the house. One of the pillars supported an Osco-Latin epigraph, likely recalling the restoration of a nearby vicum venerlum (a brothel?) in the 2nd century (G. Camodeca and A. La Regina, personal communications, 2006). Here, I sampled and dated some charred materials belonging to the wooden structure of the porticus, which was buried by the subsequent collapse of this building during the Early-High Middle Age (see next section). The calibrated age (110–330 CE, 2σ cal.; Table 1) fits the period of general restoration of the town, providing the ante quem term for the collapse.

**Table 1.** Radiocarbon ages of samples collected in the investigated area (AMS, Accelerator Mass Spectrometry and Radiometric ages by Beta Analytic Inc., Miami, FL, USA). 2σ calibration with software Calib 7.1 [36].


#### 3.1.5. Thermal Baths

Between the 1st and the 2nd century CE, the thermal baths were restored, and their orientation was changed, whereas the floors were completely renewed with different mosaics.

#### 3.1.6. Macellum

In the same period (1st–2nd century CE), in the area of the macellum, the two tholoi were dismantled, and the macellum itself was abandoned, whereas its remains were leveled and occupied by new workshops.

#### 3.1.7. Salimbene House

Below this house facing the Decumanus, the archaeologists found the remains of a 1st century BCE room ceiling that abruptly collapsed and was successively buried by other structures. By removing the fallen material, it was possible to observe that the incannucciato ceiling collapsed directly over the mortar floor of the room, where it also buried pottery shards of the 1st century CE (Figure 4). This clearly means that the room was in use when the collapse happened.

#### 3.1.8. Caesareum Temple

Here, in the same period (1st century CE), an opus caementicium cistern was built with the aim of supporting the damaged retaining wall of the temple. The severe damage suffered by this temple is also testified by the epigraph of Otacilius Gallus that will be hereafter described.

In the whole, the archaeological data evidence a general, abrupt discontinuity within the urban texture of the Roman Volcei, followed by a reconstruction phase focused between the 1st and 2nd century CE. The great abundance of domestic pottery, tiles, bricks, and stones in the dumping graves, summed to the existence of leveled rubble fills, are the proof of contemporary, extensive building collapses in the town. Moreover, the discovery of the pot with the dormouse, ready to be cooked when it was buried under the rubble, and the Otacilius' epigraph, attesting the collapse of the *Caesareum*, are conclusive proofs concerning the occurrence of this event. Last but not least, a further clue evocating the tragedy caused by this earthquake comes from the funerary monument of Gresia Tertia, which is located only 10 km SE to Volcei. Here, archaeological investigations unearthed an epigraph datable within the 1st century CE, where an *infelix mater* (a desperate mother) cries over the death of her family, namely all the four sons and the father. Even if the cause is not declared, the simultaneous decease of five persons in the same family could really be related to the collapse of their house.

**Figure 4.** Simultaneous collapse of the *incannucciato* ceiling (unit 24) and of the plaster (25) over the mortar floor (30) of the inhabited room. Unit 21 is instead the foundation of a medieval wall that was carved within the Roman rubble. For the record, this medieval wall collapsed due to a further earthquake, and its relics were found inside an adjacent room (photo by P.G.).

#### *3.2. Archaeoseismic Evidence from Compsa*

The municipium of Compsa (today Conza, Figure 1), similar to the nearby Volcei, starting from the 1st century BCE, flourished throughout the whole Imperial period, surviving to the Late Antiquity decline without dramatic urbanistic breaks [35]. However, despite the monumental buildings of the 1st century BCE being renewed under Emperor Augustus at the beginning of the 1st century CE [37,38], the archaeologists have found sparse indications of further works made just after a little interval, consisting of the reworking of the *Capitolium*, its *podium,* and its staircase. Moreover, they noticed the restoration of other new buildings surrounding the *forum*, with the demolition of structures built only a few years before, as the columns of the *porticus* or the wall of the *sacellum* (F. Soriano, personal communications, 2020). Thus, whereas the architectural improvement of the monumental buildings during the Augustus period is typical in almost all the Roman towns of the time, the rebuilding of the same structure only a few years after might suggest the occurrence of a destructive event, such as an earthquake. In this case, the archaeologist placed it just after the first half of the 1st century CE.

#### *3.3. Paleoseismic Evidence from MMFS Trenching*

Besides the archaeological proofs of this earthquake, there exists also geological evidence of its occurrence and size. As aforementioned, the authors of [6,7] performed both high-precision topographic leveling across the MMFS scarp and new paleoseismic trenches across different segments of the master fault. About 60 topographic profiles revealed the presence of at least 4 clusters of retreated scarp edges (e.g., up-slope: migrating inflection points: [6]), representing as many surface faulting events (Figure 5). These match 4 paleoseismic episodes recognized within trenches, the age of which was successively defined by the AMS dating of numerous key samples collected there.

**Figure 5.** Horizontal versus vertical separation values measured through high-precision topographic profiles across the compound fault scarp of Monte Marzano faults (vertical values are cumulated). The different clusters provide evidence for at least four major surface-faulting events. The older the rupture, the higher the vertical and horizontal separation (mod. from [6]).

In detail, the fourth cluster was dated in the trenches T1 and T2 of [7]. These excavations (Figure 6) provided robust evidence of slope debris offset that occurred just before 80–310 CE, which is a time matching the oldest event recognized also in T3, where this was dated between 540–390 BCE and 540–650 CE. The event also fits event 2 in [2], which occurred between 620 and 230 CE. Considering all the above, the preferred age of [7] was bracketed within the early Imperial Period (1st–2nd century CE).

**Figure 6.** View of trenches T1–T2 during the excavation along the >30◦ dipping slope of Monte Valva-Marzano (summer 2010). Note, in the right panel, the 1980 surface offset, still visible after 30 years. The blow-up of T2 (left panel) focuses on the tectonic wedges formed in the 989 and 1694 surface faulting (see details in [7]) (photos by P.G.).

#### *3.4. Historical Sources*

In a strict sense, historical accounts concerning the effects of this earthquake in Irpinia do not exist. Leaving aside the intriguing epigraph of Gresia Tertia and the ambiguous Osco-Latin epigraph in Amendola square, the only certain information is provided by the epigraph of Otacilius Gallus. This was carved on an architrave (Figure 3), and it recalls the collapse of the *Caesareum*, which is a temple built only one century before, around 50–60 BCE. The text is:

#### *OTACILIVS EX TESTAMENTO OTACILI GALLI PATRIS CAESAREVM*/ *[TERRAE MOTV] CONLAPSVM P(ecVnia) [S(Va) R(estitVit)]. CVIVS OPER[IS] DEDICATIONE*/ *[DEDIT DECVRIONIBVS] (sestertios) XXX, AVGVSTA[L]IBVS (sestertios) XX, VICANIS (sestertios) XII, VX[ORIBVS]*/ *DECVRIONVM (sestertios) XVI, AVGVSTALIVM (sestertios) VIII, VICANORVM (sestertios) IIII*

which roughly means: "Otacilius, according to the will of his father, rebuilt by its own the Cesareum which was destroyed by an earthquake ... etc.". If the integration of the missing text (square-brackets: [39]) is correct, the *Caesareum* was destroyed by an earthquake that occurred before the end of the 1st century CE, and it was then restored by Otacilius in the 2nd century (G. Camodeca, personal communication, 2006).

#### **4. The 989 CE Earthquake**

#### *4.1. Historical Sources*

This is the first earthquake explicitly quoted in the Irpinia area, and it was already reported by the first Italian editors of seismic catalogues, as Giannozzo Manetti, Colanello Pacca, and Marcello Bonito [40–42]. Manetti mentions the collapse of several buildings in Benevento and Capua, as in Ariano, Ronza, and Conza, where many people and the Bishop died. In turn, Pacca places the earthquake in 982 in the Naples Kingdom, describing the effects in Benevento, Capua, Ariano, Frigento, Conza, and Ronza. Bonito places it in 989, reporting texts and sources that he painstakingly collected. In short, the primary sources to which these three authors refer are the Annales Beneventani [43], the Chronica Monasterii Casinensis [44], and the Chronicon of Romualdo Guarna [45]. The exact year of the earthquake has been fixed to 989 by [46], who cross-checked other certain historical episodes to which the event can be chronologically referred. Nevertheless, on the basis of the only historical descriptions, its epicentral parameters were not univocally defined. The most interesting information derives from Leone Ostiense, the author of [44], who was writing in the Montecassino Monastery and likely used first-hand sources, which are lost today. In few words, he informs us that the earthquake struck Vipera (this toponym still exists in the Istituto Geografico Militare maps near Benevento) and Benevento, where 150 died and 15 towers collapsed, whereas in Capua, many houses fell down, and the bells rung. Ariano Irpino and Frigento were almost completely destroyed, as was Conza, where many inhabitants died together with the Bishop, while Ronza (a settlement close to Conza) was totally razed to the ground. Leaving aside the inconsistent news from the very far town of Capua (where bell towers just ring while houses are collapsing), on the basis of all the news, [20] assigned Is 10–11 MCS to Ronza, 10 MCS to Conza, and 9–10 MCS to Ariano and Frigento, while in Benevento, the most important and populated town of the region, the number of casualties may suggest at least an Is 8–9 MCS intensity.

#### *4.2. Archaeoseismic Evidence from Compsa*

Here, the archaeological excavation performed after the 1980 earthquake by the former Superintendency of Salerno, Avellino, and Benevento unearthed vast areas of the Roman and Medieval town. As a matter of facts, buildings and tombs of Byzantine age (6th century) before, and Langobardic (7th century) then, occupied the Forum area, always resting in phase with the Roman town levels. A proto-cathedral was also built over the Roman Capitolium, with a staircase connecting the church to the forum pavement [47–49]. The archaeoseismic survey conducted with the archaeologists during the excavation in 2003 allowed collecting indications of a strong break in the Early Middle Age history of Compsa, which was namely between the 10th and 11th century. After this break, for the first time after its foundation, the town was not rebuilt, respecting the inherited urbanistic Roman texture. An extended level of rubble, containing pottery shards and coins of the 10th century, buried the limestone slabs of the *Forum* and all its inscriptions [50]. The first cathedral, erected on the *Capitolium*, collapsed and was later rebuilt close to the *Forum*, although it was rotated with respect to the axes of the Roman–Early Middle Age town. The new apse occupied a part of the previous *Forum*, covering the leveled strata of destruction. Here, I made a radiocarbon dating of a huge charred fragment sampled

within the collapse levels, which provided an age of 660–970 CE (2σ cal. Age; Table 1), representing the *post quem* term of the earthquake destruction (Figure 7). Unfortunately, this new *basilica*, consecrated in 1122 [51], will be soon destroyed by the forthcoming 1466 earthquake, and then again in 1694, 1732, and definitely 1980 [20].

**Figure 7.** Rubble and abandonment layers excavated in the Forum area of Compsa, over the limestone slabs (below the booklet). The radiocarbon dating of a charred wood (660–970 CE; 2σ cal. age) provided the *post quem* term for the destruction of the Early Middle Age town in the 989 earthquake (photo by P.G.).

#### *4.3. Archaeoseismic Evidence in Volcei*

Here, the evidence of the 989 event is represented by the synchronous and total collapse of the buildings excavated below Amendola Square and in other neighboring *insulae*. Moreover, it is witnessed by the general abandonment of the surviving Late Antiquity buildings, which were still inhabited during the Langobardic period, and by the new urban topography that, as in Conza, drifts apart from the Roman imprint, assuming a concentric path around the new castle. Actually, in Volcei, it is difficult to provide univocal age brackets for this earthquake, mainly because of the paucity of the Early Middle Age pottery. However, archaeoseismic clues are constrained between the 7th–8th and the 12th century, even if I cannot exclude the occurrence of multiple events within this time span. The crude set of indications can be summarized as follows (Figure 3 for location).

#### 4.3.1. Amendola Square

The excavations have revealed the synchronous collapse of all the buildings surviving since the Late Roman times. Below the rubble, it has been possible to read the history of these houses, with the different redistricting of each room during times, the wall restorations, the floors overlapping, and the doorstep reutilization. The collapse affected all the masonry walls, the pillars of the *porticus*, and the roofs, which have been found all directly overlaying the floors (Figure 8). This catastrophic collapse also definitely buried the *Decumanus,* which was still in use and well maintained at least during the 7th century, as testified by the materials found in the ditches. At first glance, the collapse also killed a small sheep, which was hit and buried by the rubble on the road *basoli* (see the arrow in Figure 8). I sampled and obtained an AMS collagen dating of 1034–1214 AD (2σ cal.) from the sheep bones. Inside the porticated building, beside the collapse of the wall plasters of the *incannucciato* ceiling, tiles, and masonry, it was possible to observe some walls and the four bricks/pillars that fell away from the road, burying *bande rosse* pottery (used all along the Early Middle Age).

**Figure 8.** Amendola Square insula. View looking west of the amazing, total, and simultaneous collapse of the porticated house over the *Decumanus* before the removal of the roof (reddish material). The arrow points to the limestone *basoli*, where the sheep was found (photo by P.G.).

#### 4.3.2. Salimbene House

An Early Middle Age cobble wall was founded inside the fill burying the Roman buildings (Figure 4). The pottery shards inside the foundation trench are the same as in Amendola Square, i.e., it contains *bande rosse* pottery. Therefore, this wall—which successively collapsed over a nearby room—might suggest the onset of the reconstruction of the Early Middle Age Buccino after and over the earthquake rubble.

#### *4.4. Other Settlements*

In order to enrich the framework of effects distribution related to this earthquake, I collected further indications concerning damage to ancient buildings (see Figure 2A) with the assistance of M. Rotili (Naples University; personal Communication, 2002).

#### 4.4.1. Frigento

Here, [52] described the burying of the right apse of the Langobardic Mother Church, which collapsed and was successively leveled prior to the early 11th century. The rebuilding of the church is dated between the 11th and 12th century, when the new structures were founded over the walls of the collapsed church, or directly over the infill of its ruins.

#### 4.4.2. Montella

In the Angevin Park, [53] found the 13th–14th century boundary wall built over the leveling of the collapsed 9th century wall. In turn, other remains of the 9th century wall were still laying outside the newer wall, burying tombs of the 7th–8th century.

#### 4.4.3. Rocca San Felice

Here, [53] hypothesizes the destruction and successive reconstruction of the 7th–8th fortified structure because of the 989 earthquake. This is also attested by the 10th–12th century *donjon* that was built over the leveling of the Langobardic structures.

#### 4.4.4. Sant'Angelo dei Lombardi

The same author [53] claims for the 989 earthquake shaking in order to explain the destruction of the Langobardic fortified settlement, which was rebuilt in the 11th–12th century. As well as the restoration and reinforcing of the previous damaged wall, the Norman *donjon* was erected over the leveling of the Langobardic structures, as in Rocca San Felice. Moreover, it is worth noting that the right apse of the 11th–12th century cathedral was founded over the leveling of the old, collapsed boundary wall.

#### *4.5. Paleoseismic Evidence from MMFS Trenching*

In the trenches opened by [7] and along the fault-scarp profiles discussed in [6], between the signatures of the 1st century CE and the 1694 earthquakes, these authors have found the clear evidence of a Middle Age surface faulting (Figure 5). In trenches T1–T2 (Figure 6), they placed it contemporary or slightly after 720–970 CE (age of the infilling colluvium at the bottom of a coseismic tectonic wedge), and well before 1190–1270 CE and 1260–1390 CE (age of the units sealing offset levels). The same event was also found in T8, where it occurred after 685–892 CE.

#### **5. Discussion and Conclusions**

An unexpected and tragic event, such as the Mw 6.9, 1980 Irpinia earthquake, allowed rediscovering the buried relics of the Roman *municipium* of Volcei (Buccino) and of Compsa (Conza). The amazing discovery is that below the ruins of the 1980 earthquake, the archaeologists have found a palimpsest of constructions/collapses/reconstructions attributable to as many seismic events that, each time, have partly allowed the freezing of the buildings history below their own rubble. Although sometimes, the unraveling of this tangled skein of construction/destruction events is a very hard task, with the guide and help of the archaeologists, it has been possible to collect several indications from different sectors of the towns that suggest the occurrence of different earthquakes striking Volcei/Buccino and

Compsa/Conza in the past, and in particular, one in the 1st century CE and one in the 10th century, i.e., the 989 earthquake.

1st century CE earthquake—In Volcei, many archaeological evidence indicate a strong, destructive event that occurred in the second half of the 1st century AD, as constrained well by the age of the huge amount of pottery materials found by the archaeologists, which is supported also by the AMS dating of wooden structures. Conversely, in Compsa, this event was not clearly identified due to the lack of the Roman stratigraphy, which was mostly removed by the early post-1980 works. However, here archaeologists agree that the 1st century BCE *Forum* was newly rebuilt in the early 1st century CE [39,50], whereas it was again restored and reworked only a few years after (F. Soriano, personal communication, 2020), i.e., contemporary to the well documented Volcei reconstruction. This earthquake, which is unknown to the seismological compilations in this region, can be related to the one found by means of paleoseismological analyses across the Mount Marzano Fault System [7], where it was dated a time before 80–310 CE. Alongside the archaeoseismic and paleoseismic indications, this event is recorded by an epigraph that, although incomplete, clearly mentions the collapse of the *Caesareum* of Volcei. Given the lack of classical literary sources for the Irpinia area, a very speculative hypothesis concerning its age may be suggested by the concurrence of the so-called Pompei earthquake in 62 CE. Bearing in mind that the strongest earthquakes sourced by the MMFS, as well as by many others Apennines faults [54], always induced high-intensity effects in the surrounding of Naples (e.g., Is 7 MCS in both 1694 and 1980 events, with isolated collapses and casualties; see [55]), it could be possible that the damage reported by the historical sources in ancient Pompei, Naples, Ercolano, and Nocera might represent the far-field effects of the same earthquake that razed to the ground Compsa and Volcei. It is worth remembering that Seneca in the *Naturales quaestiones* [56], besides describing the effects in the aforementioned towns, also stresses that the earthquake "*Campaniam* ... *magna strage vastavi*" (destroyed the Region Campania with many casualties) and "*non desiit enim assidue tremere Campania*" (Campania continued trembling), likely alluding that damage was spread over a vast area. If this were true, according to the theory of elastic stress transfer, one could also tentatively hypothesize that this Apennine earthquake was the one that triggered the forthcoming 79 CE Vesuvius eruption. This is suggested by some of the case histories presented and discussed in [57], although more robust hypotheses were discussed in [58].

989 CE earthquake—Differently from the 1st century earthquake, Compsa was explicitly quoted among the towns destroyed by the 989 event. In turn, even in this case, its archaeological evidence is less detailed than in Volcei, as most of the first excavations after the 1980 earthquake were made without a systematic description, documentation, and collection of archaeological materials. However, the available data undoubtedly suggest that an earthquake occurred at the end of the 10th century (i.e., post 660–970 CE, before 1122) causing the almost total destruction of Compsa, which was successively abandoned and then slowly rebuilt and repopulated. Besides Compsa, other settlements of the region show archaeoseismic evidence of contemporary destruction and reconstruction, such as Frigento (also mentioned by the historical sources), Montella, Rocca San Felice, and Sant'Angelo dei Lombardi, which were all heavily struck by the 1980 event as well.

In Volcei, there are many indications of coeval destruction around the end of the 10th century, although the univocal dating of collapses is problematic. Indeed, whereas the pottery shards involved and buried by the collapses predate the onset of the second millennium, the AMS dating of the sheep buried under the rubble would indicate a slightly later age. Considering the uncertainty that often arises from collagen dating, the simplest hypothesis is that an event occurred between the uppermost and lowermost boundary of the two terms, i.e., around 1000 AD, a time fitting the October 25, 989 earthquake. However, this framework is complicated by the presence of a 16th century historical source who mentions that an earthquake occurred at the times of Pope Callistus II (1119–1124), the destructive effects of which were still visible in Buccino in the "*horti...nella parrocchia de S Maria Sollitta*..." ( ... in the gardens of St Mary Sollitta church [59]). The period of Pope Callistus fully matches the AMS age of the sheep squashed under the rubble, although it is not consistent with the time span

suggested by the pottery. Nevertheless, considering the high frequency of earthquakes occurrence in this region [3,20], we cannot exclude that more than one event hit Buccino just before and after 1000 CE, cumulating damage and favoring the collapse of the highly vulnerable Middle Age buildings. As a matter of fact, in one of the houses below Amendola Square, the archaeologists unearthed a small lime furnace that was operating at the time of the collapse. The furnace overlays a thin abandonment level on the floor, suggesting that men were working inside an uninhabited house [33]. Thus, an attractive hypothesis, which makes no claims to being conclusive, is that while works were in progress for repairing the damage of the 989 earthquake, another event caused the complete collapse of the buildings around 1120 CE and the definitive burying of the ancient *Decumanus*.

As well as in the case of the Roman event, paleoseismic results along the MMFS support the presence of a Middle Age surface faulting. This was dated as contemporary or slightly after 720–970 CE, and well before 1190–1270 CE. Last but not least, the rupture of the MMFS in this period matches the distribution of the settlements destroyed by the 989 earthquake, which all fall in the hanging wall or in the surroundings of the Irpinia fault.

As a concluding remark, I would like to stress that the results summarized in this short review represent some intriguing case histories in the investigation of ancient earthquakes. The concurrent presence of historical sources, archaeoseismic indications, and paleoseismic evidence of the same earthquake have been rarely observed not just in Italy, but also worldwide. The contribution of each discipline strengthens the parametrization of the earthquake, both in terms of age, location, and energy released. The probable association with the seismogenic source of the 1980 earthquake suggests that both the 1st century CE and 989 CE events had a comparable magnitude (Mw 6.9), whereas the age brackets provided by the archaeoseismic and paleoseismic results suggest the challenging hypothesis that the older event might be the one that in 62 CE damaged Pompei, anticipating and, dubiously triggering, the famous Vesuvius eruption in 79 CE.

**Funding:** This research received no external funding.

**Acknowledgments:** I am grateful to all the archaeologists who shared with me all their findings, discussing with me the hypotheses presented in this paper. In particular, I am indebted to Adele Lagi, who spent a lot of time explaining me the secrets of all the stones of Volcei. I wish to thank also Gabriella Colucci Pescatori, Enzo di Giovanni, Pierfrancesco Talamo, Fiammetta Soriano and Marcello Rotili.

**Conflicts of Interest:** The author declares no conflict of interest.

#### **References**


© 2020 by the author. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

### *Article* **Insights into Mechanical Properties of the 1980 Irpinia Fault System from the Analysis of a Seismic Sequence**

**Gaetano Festa \*, Guido Maria Adinolfi, Alessandro Caruso, Simona Colombelli, Grazia De Landro, Luca Elia, Antonio Emolo, Matteo Picozzi, Antonio Scala, Francesco Carotenuto, Sergio Gammaldi, Antonio Giovanni Iaccarino, Sahar Nazeri, Rosario Riccio, Guido Russo, Stefania Tarantino and Aldo Zollo**

> Physics Department "Ettore Pancini", University of Naples Federico II, 80126 Napoli, Italy; guidomaria.adinolfi@unina.it (G.M.A.); alessandro.caruso@unina.it (A.C.); simona.colombelli@unina.it (S.C.); grazia.delandro@unina.it (G.D.L.); luca.elia@unina.it (L.E.); antonio.emolo@unina.it (A.E.); matteo.picozzi@unina.it (M.P.); antonio.scala@unina.it (A.S.); francesco.carotenuto2@unina.it (F.C.); sergio.gammaldi@unina.it (S.G.); antoniogiovanni.iaccarino@unina.it (A.G.I.); sahar.nazeri@unina.it (S.N.); rosario.riccio@unina.it (R.R.); guido.russo2@unina.it (G.R.); stefania.tarantino@unina.it (S.T.); aldo.zollo@unina.it (A.Z.)

**\*** Correspondence: gaetano.festa@unina.it; Tel.: +39-081-675248

**Abstract:** Seismic sequences are a powerful tool to locally infer geometrical and mechanical properties of faults and fault systems. In this study, we provided detailed location and characterization of events of the 3–7 July 2020 Irpinia sequence (southern Italy) that occurred at the northern tip of the main segment that ruptured during the 1980 Irpinia earthquake. Using an autocorrelation technique, we detected more than 340 events within the sequence, with local magnitude ranging between −0.5 and 3.0. We thus provided double difference locations, source parameter estimation, and focal mechanisms determination for the largest quality events. We found that the sequence ruptured an asperity with a size of about 800 m, along a fault structure having a strike compatible with the one of the main segments of the 1980 Irpinia earthquake, and a dip of 50–55◦ at depth of 10.5–12 km and 60–65◦ at shallower depths (7.5–9 km). Low stress drop release (average of 0.64 MPa) indicates a fluid-driven initiation mechanism of the sequence. We also evaluated the performance of the earthquake early warning systems running in real-time during the sequence, retrieving a minimum size for the blind zone in the area of about 15 km.

**Keywords:** earthquake seismology; microseismicity; seismic techniques; seismotectonics

#### **1. Introduction**

Seismic sequences are a useful tool to shed light on fault mechanics and geometry, on the chemical and physical processes occurring on faults and ultimately to illuminate the preparatory phase of large earthquakes. For the goal, diverse dense multi-disciplinary observation infrastructures were created around faults that can potentially host large destructive earthquakes, to understand how those faults slip, what is the seismic and aseismic balance during strain release, and what is the role of fluids in earthquake production.

The study of microseismicity can provide mechanical constraints on the faults, through accurate earthquake location and source parameter computation. Earthquake location can benefit from measurements of P and S wave arrival times and back-azimuth estimation. In dense networks, absolute locations of microearthquakes can reach a sub-kilometric accuracy, using global-search techniques in 3D velocity media tailored for the area [1,2]. When using relative location techniques, such as double differences [3], the accuracy can be pushed down to a decametric scale [4], when relative arrival time determination is performed through cross-correlation [5]. With this resolution, earthquake location can delineate the geometry of faults [3,6] and following event space and time evolution, it is possible to infer the role of the fluids in seismicity migration [7,8].

**Citation:** Festa, G.; Adinolfi, G.M.; Caruso, A.; Colombelli, S.; De Landro, G.; Elia, L.; Emolo, A.; Picozzi, M.; Scala, A.; Carotenuto, F.; et al. Insights into Mechanical Properties of the 1980 Irpinia Fault System from the Analysis of a Seismic Sequence. *Geosciences* **2021**, *11*, 28. https:// doi.org/10.3390/geosciences11010028

Received: 16 November 2020 Accepted: 3 January 2021 Published: 5 January 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

Source parameters are usually inferred through the fit of a specific rupture model with a simplified description of the observations, such as the displacement amplitude spectrum [9]. Rupture evolution is synthesized in few macroscopic parameters, such as the seismic moment, the source size, and the average stress drop. Large epistemic uncertainties in source parameter estimation come from the medium description, which is usually assumed as homogeneous [10]. This problem can be mitigated either using smaller magnitude earthquakes as empirical Green's functions [11] or computing a non-parametric, data-driven attenuation function [12]. Source parameters provide insights on the stress released during earthquakes and on the earthquake size, enabling the understanding of the role of fluids in earthquake production [13] and the recognition of possible forcing aseismic mechanisms in sequence generation [14].

Characterization of the seismic activity during sequences can constrain fault geometry and mechanics in the vicinity of the swarm location. This information can be crossed with tomographic models in velocity and attenuation, to get a picture of fluid-filled domains and infer the role of the pore pressure in the stress distribution and release at depth [15].

While in most cases seismic sequences end into background seismicity, sometimes acceleration of the cumulative seismic moment release during sequences can yield a large magnitude event, especially for interplate events [16]. Capability to recognize foreshocks before the occurrence of a large event is fundamental for risk monitoring and reduction. There is a large debate in the seismological community on precursory seismicity before a large event and contrasting results have been obtained in indicating the occurrence of foreshock sequences as a persistent feature of moderate to large events (see for California [17,18]). Nevertheless, when a sequence clearly evolves to a large event, foreshocks emerge only when high-resolution seismic catalogs are available, with a magnitude of completeness much smaller than the one presently available for standard monitoring networks. Thus, advanced techniques for event detection are required to improve the catalog, such as the template matching [19] or autocorrelation algorithms [20].

In this study, we have applied some of these advanced techniques to analyze a seismic sequence that occurred in the area of the 1980, M 6.9, Irpinia earthquake. Since 2005, in this region, INFO (Irpinia Near Fault Observatory) operates ISNet—the Irpinia Seismic Network—with the goal of monitoring the seismicity evolution and its relationship with the underlying fault system [21]. ISNet is a dense, high-dynamic range seismic network of 31 stations, covering an area of about 100 × 70 km<sup>2</sup> along the Campania-Lucania Apennine chain, surrounding the fault system that generated the 1980 Irpinia earthquake [22]. The seismic stations are deployed within two imaginary concentric ellipses, with the major axis parallel to the Apennine chain. The mean distance between stations in the outer ellipse is about 20 km, while the distance between the two ellipses is about 10 km. The average inter-station distance within the inner ellipse is less than 10 km. All stations are equipped with a strong-motion accelerometer and a weak motion sensor, the latter being a short-period, a broadband velocimeter or an accelerometer with lower full-scale. Stations of ISNet provide real-time data with controlled delay at the ISNet control center, enabling the network to be the backbone infrastructure for earthquake early warning systems [23] and for near real-time computation of regional ground-shaking maps.

We used ISNet data to investigate the mechanical properties of the Rocca San Felice seismic sequence, that started on 3 July 2020 and lasted for 4 days. In this study, we reviewed the seismotectonic setting of the area, then we applied advanced techniques for event detection and characterization, to enlarge the automatic catalog and to provide accurate locations, based on double-difference techniques. We thus computed source parameters and focal mechanisms to infer the stress release and its main directions, the size of the events, and to discriminate the rupture plane. Finally, we analyze the ground motion and the performances of earthquake early warning systems, as real-time risk reduction tools.

#### **2. Seismotectonic Setting**

#### *2.1. Structural Setting of Southern Apennines*

The structure of the Southern Apennines is associated with the Meso-Cenozoic tectonic processes that involved the African and European plates and, in accordance with different tectonic phases of rifting, drifting, and shortening, deformed the Corsica-Sardinia and Adriatic-Apulian forelands [24,25]. The backbone of the mountain belt is characterized by E-NE verging duplexes geometries and out-of-sequence thrusting due to orogenic contraction that was active since upper Eocene-Oligocene Miocene up to late Pliocene [26]. Meso-Cenozoic successions of Adriatic-Apulian-African passive margin, deposited as carbonate platforms with interposed pelagic basins, were involved in folding and thrusting according to imbricated structural units detached from their crystalline basement, as documented by seismic profiles obtained since 1980s during hydrocarbon exploration [27]. Migrating eastward over the Apulian foreland, these sheets were tectonically overthrusted by internal units, deposited originally above the oceanic crust of Tethys. Considering the lithostratigraphic relationships, from top to bottom, units are grouped in: (1) post-orogenic intramontane basin units of marine, terrestrial, and volcanic origin, deposited during Plio-Pleistocene or Holocene in the Adriatic-Bradanic foredeep; (2) syntectonic top-thrust basin successions formed during the progressive shortening toward east; (3) orogenic wedge tectonic units involved in the NE-verging overthrusting from upper, internal domains (Tethyan oceanic crust or Adriatic-Apulian continental crust) to the lower, external domains (Apennine carbonate platforms with inter-basins pelagic units); (4) Apulian carbonates, buried, deformed, and overthrusted in the inner belt and undeformed in the outcropping foreland [25,28]. During the Quaternary, the Southern Apennine thrust belt was dissected by NW-SE oriented normal faults that accommodated an extensional tectonic phase, according to a stress field with the axis of maximum extension coaxial to the axis of maximum compression of Apennines belt (SW-NE trend [24,29,30]). In fact, contractional and extensional deformations took place simultaneously in different sectors of the Apennines with both E-migrating fronts at a similar rate of ~4 cm/year.

#### *2.2. Seismogenic Normal Faults*

In the Southern Apennines, normal faulting corresponds to a significant deformation process in terms of regional size of structures, morphotectonics, active displacement, and seismogenesis. Normal faults with NW-SE striking regulate the active tectonics in the thrust belt, accommodating an extension of 3–5 mm/year as evidenced by surface geology, borehole breakout, and available fault plane solutions of earthquakes [31,32]. Earthquakes up to X-XI MCS intensity struck the Southern Apennine chain, indicating that this sector is one of the highest seismic hazard areas of the Mediterranean region, with segmented, seismogenic structures capable of generating up to M 7 earthquakes [33,34] (Figure 1). These earthquakes occur principally in the axial sector of the Apennine chain with recurrence periods ≥1000 years and depths down to 10–15 km in the upper crust, in a 30–50 km wide belt that follows the orographic divide [35]. As proposed in different models [36–38], normal faults that dissect Apennines show an evolutionary trend from young, high-angle planar faults, seated in the upper crust and characterized by small extensional strains, to mature, listric faults reaching the crystalline basal detachment with high amount of extensional strain. A consequence of progressive E-migration of the extensional front is that normal faults show a different aging and degree of evolution from mature faults of the inner sector of the Apennine chain (to W, toward the Tyrrenian margin) to young faults superimposed on the accretionary wedge along its axial sector (to E, gradually decreasing toward the foreland). Large seismic events are believed to be associated to younger normal faults connected to the extensional front and, in the eastern sector, to fluids reaching an over-pressure condition at shallow depths due to the low permeability of the crust that inhibits their circulation [34].

**Figure 1.** (**a**) Epicentral map of the earthquakes (green circles) recorded by Irpinia Seismic Network (ISNet, red triangles) from 2008 to 2020 (http://isnet-bulletin.fisica.unina.it/cgi-bin/isnet-events/ isnet.cgi). The analyzed seismic sequence is highlighted with violet circles. The yellow and orange stars refer to the epicentral location of the 1980, M 6.9, and of the 1996, M 4.9 earthquakes. Historical seismicity is shown with black squares (I0 ≥ 6–7 MCS). Seismogenic sources related to the Irpinia fault system are indicated by orange rectangles; potential sources for earthquakes larger than M 5.5 in surrounding areas are indicated in grey (Database of Individual Seismogenic Sources, DISS, Version 3.2.1). Focal mechanism solutions for four instrumental earthquakes are reported (1) Ml 3.0, 3 July 2020, 16:14, Rocca San Felice sequence, from this study; (2) Ml 3.0, 3 July 2020, 16:19, Rocca San Felice sequence, from this study; (3) 1980, M 6.9, Irpinia earthquake [40]; (4) 1996, M 4.9 earthquake [57]. The area plotted in Figure 4 is identified by a dashed rectangle. (**b**) Histograms of magnitude and (**c**) depth for the microseismicity inside ISNet. The red arrow points to the depth above which 90% of earthquakes occur.

#### *2.3. 1980 Irpinia Earthquake*

The 1980, M 6.9, Irpinia earthquake was the most destructive, instrumental earthquake of the Southern Apennines that occurred along NW-SE trending normal faults. This event is characterized by a complex rupture process involving multiple fault segments according to (at least) three different nucleation episodes delayed each other of 20 s. Several models, characterized by different geometries and locations of activated fault segments, have been proposed to explain its complex rupture [39–42]. Retrieved seismic moment ranges between 2.4 × 1019 Nm and 3.0 × <sup>10</sup><sup>19</sup> Nm, the fault dip activated by the mainshock

ranges between 53◦ and 63◦ while the strike ranges between 305◦ and 33◦ [43]. Location of the event epicenter and surface projection of the three main fault segments are shown in Figure 1.

Kinematic models [41] have proposed the nucleation of the mainshock (0 s) along 30-km-long, NE-dipping, Mount Marzano and Picentini fault segments, with the first subevent activated at 18 s toward a SW, low angle, NE-dipping fault of 20 km length and a second subevent nucleated at 39 s in the Ofanto basin area, along a SW-dipping, antithetic fault. The mainshock initially ruptured at the north tip of the Mount Marzano segment and propagated bilaterally along two NW-SE striking faults. Toward south, it completed the rupture of the Marzano segment and toward north it progressed along the Mount Picentini segment.

The 1980 Irpinia faulting model has been also constrained on the evidence of coseismic surface ruptures [42], and shows that the event activated three main fault segments, along a 38-km-long, NE-dipping scarp with an average N128◦ direction and one main antithetic blind fault, SW-dipping, that was not capable to reach the surface due to its smaller size, leaving a ground deformation anomaly at NE of mainshock, near the Ofanto river. The NEdipping fault rupture corresponds to three main strands separated by gaps and anomalies in the continuity of surface faulting identified as: (1) the Cervialto scarp (Mount Picentini, strike of 125◦); (2) the Mount Marzano-Valva scarp (strike of 135◦) connected southward with the Mount Carpineta scarp (strike ranging between 110◦ and 135◦), and (3) the San Gregorio Magno scarp and related scarplets in and close to the Pantano basin (strike of 120–130◦). The Mount Cervialto and Mount Marzano–Mount Carpineta segments have been correlated to the 0 s event (mainshock), the Pantano of San Gregorio Magno segment has been correlated to the 20 s subevent; and the antithetic Ofanto fault segment has been associated with the 40 s subevent. The three segments are represented on the geological map of Figure 2.

Finally, relocated aftershocks and 3D velocity models have provided new constraints about the 1980 Irpinia source model [44,45]: (1) the main rupture started at 10 km depth, on a 60◦ dipping plane, in a high velocity region, associated with stiff Apulian carbonates and propagated upward along the softer Meso-Cenozoic succession; (2) most of the aftershocks are spread in a volume delimited by the NE-dipping, Marzano-Cervialto normal fault segment and the SW-dipping, antithetic normal fault, associated to the 40 s event; (3) seismicity depth is confined in the first 12 km of the upper crust and mainly concentrated beneath the Marzano-Valva fault segment; (4) largest magnitude aftershocks clustered between the fault segments activated at 0 s and 20 s, near their tips; (5) the Marzano segment is bounded by two clusters of seismicity likely separated by a lithological discontinuity, acting as a seismic barrier, where the stress of the main ruptures was concentrated (the northern Sele Valley and the southern tip of the Carpineta fault segment). In particular, the Marzano-Cervialto normal fault segments have been interpreted as two asperities separated by a low-strength zone, referred to as the Sele barrier [41,42]. New tomographic studies [34] confirm that the mainshock nucleated in basement, below the Apulian carbonates, and propagated in the high Vp/high-Vp/Vs region associated with fractured, water-saturated carbonates of the Apulian platform domain, in agreement with well-data (S. Gregorio Magno 1, [25]) and magnetotelluric surveys [46].

**Figure 2.** Geological and structural map of the Irpinia region. Here, surface traces (red lines) and sources (orange boxes, Database of Individual Seismogenic Sources, DISS, Version 3.2.1) of faults activated during the 23 November 1980, M 6.9, Irpinia earthquake (yellow star) are shown. Key: 1 = Cervialto fault segment; 2 = Marzano-Carpineta fault segment; 3 = San Gregorio fault segment. The seismic sequence is highlighted with violet circles and stars; the orange star refers to the epicentral location of the 3 April 1996, Ml 4.9 earthquake.

#### *2.4. Recent Seismicity of the Irpinia Region*

After the 1980 event, no large earthquakes have struck the area. In 1996, a seismic sequence with a mainshock of M 4.9 took place inside the epicentral area of the 1980 earthquake. Present-day low-magnitude seismicity (Ml < 3.5) occurs mainly in the first 15 km of the crust showing fault plane solutions with normal and normal-strike slip kinematics, indicating a dominant SW–NE extensional regime in agreement with the inversion of fault-slip data of the Irpinia region [32,47–49]. The background low magnitude seismicity appears to be spread into a large volume, and the related stress field is closely linked with the major fault segments activated during the 1980 Irpinia earthquake. In addition, microseismicity seems to be controlled by high pore pressure of water-saturated Apulian carbonates within a fault-bounded crustal volume [15,50]. Location of microseismicity epicentres is shown in Figure 1.

Several studies pointed out a strong relationship between seismicity and high-fluid pressure evidenced by the same location of crustal seismicity and CO2 degassing areas along the Apennines [38,51]. In the Irpinia area, underneath Mount Forcuso, tomographic images reveal a low-Vp/Vs dome-shaped body, 20 km long and 15 km wide, located between 6 km and 11 km depth [34]. This spot is interpreted as a pressurized CO2-rich rock volume, filled below the Apulian platform carbonates by fluid-rich mantle melts intruded into the crust. This anomaly correlates with high heat flow values (100–215 mW/m2) observed along the Mount Forcuso antiform and with geochemical data. This area is

characterized by a huge amount of nonvolcanic CO2-rich gas emission [52,53], the main expression of which occurs in the Mefite d'Ansanto degassing site, where deep, mantlerelated fluids are released through the active faults, as indicated by geochemical data.

Deep CO2-rich fluids may play a key role in the seismogenesis in the Southern Apennines [15,34,38]. Fluids migrating upward along lithospheric faults can be stored in reservoirs under the Apulian carbonates, sealed by Triassic anhydrites, reaching overpressure conditions that can periodically trigger large normal faulting earthquakes. The influence of high-pressure CO2 in the nucleation of large earthquakes has been also observed in the Central Apennine sector, for the 1997 Colfiorito and the 2009 L'Aquila events [54,55]. The same mechanism has been invoked for the nucleation of the 1980 Irpinia earthquake, being the initiation patch located in the basement, under the Apulian carbonate, at the top of a pressurized CO2 reservoir (low-Vp/Vs anomaly, [34]).

#### **3. Data**

The data used for the sequence analysis have been provided by ISNet. The available stations are shown in Figure 1, together with the epicenters of the seismicity covering the period 2008–2020. We used automatic event detections performed by the software Earthworm [56] to identify the seismic sequence. Specifically, when more than three events occur within 10 km and a couple of hours in time, an indicator light is turned on at the network control center and few days of continuous data before and after the declaration of the sequence are extracted for automatic and visual inspection, followed by machine and human controlled procedures to eventually extract additional events. For this sequence, the automatic system identified 43 events, starting from a Ml 1.9 event, declared on 3 July 2020, at 09:31 and ending with a Ml 1.2 event, that occurred on 6 July 2020, at 16:55. The two largest events in the sequence (Ml 3.0) occurred on 3 July 2020, at 16:14 and 16:19. All these events have been manually inspected, providing revised phase pickings (372 P arrival times and 208 S arrival times), locations, and local magnitude estimations.

During the manual revision of data, the near real-time software INERTIA [22] also provided moment magnitude estimations and ground motion maps. This information is made promptly available at the ISNet bulletin webpage (http://isnet.unina.it/). Continuous waveforms, event data and related information (picking, location, magnitude) were also used as starting point for further analysis as detailed in the rest of this paper. In Figure 3, we show an example of waveforms for a Ml 2.8 event of the sequence, that occurred on 5 July 2020, at 15:18. In the section, stations are ordered for increasing distance from the event epicenter. We can recognize the complexity of the waveforms as the distance increases and the diverse duration and frequency content at some stations, owing to site effects and instrumental filters.

**Figure 3.** Example of seismic data, represented for the Ml 2.8 event that occurred on 5 July 2020, at 15:18. In the section, stations are ordered from the top to the bottom for increasing epicentral distance. In the coda of the main event, we can recognize smaller magnitude aftershocks.

#### **4. Refined Seismic Catalog**

We scanned the continuous helicorder for 4 days starting from 3 July to improve the detection of microearthquakes during the sequence. We used the autocorrelation algorithm FAST (Fingerprint and Similarity Thresholding) [20,58] with the twofold aim of including in the catalog events featuring low signal-to-noise ratio and diminishing the magnitude of completeness for the sequence.

FAST is based on a locality-sensitive hashing algorithm [59] and performs a computationally efficient similarity search, scanning the signal spectrograms, computed in broad frequency bands, with the aim of detecting similar earthquake waveforms. In contrast with template matching, FAST does not require any template for the detection. This feature enables the technique for the detection of microevents during short and spatially constrained sequences.

Since our goal is the search for very-low magnitude events, we applied the technique to the velocimetric records at the five stations closest to the sequence centroid (RSF3, LIO3, NSC3, AND3, MNT3, Figure 4a), limiting our analysis to the vertical component. The distance of the stations from the centroid ranges from 3 to 17 km. We computed the spectrograms, filtering the records between 1 and 20 Hz and separating the continuous waveforms in moving windows having 6.0 s length with a 0.2 s lag. The squared modulus of the Fourier transform of each time window represents a column in the spectrogram, from which we extracted overlapping fingerprints having 32 samples with a lag of 5 samples. To declare a sequence of consecutive and correlated fingerprints as a candidate event, we fixed uncorrelated permitted gaps at a single station to 3 s, this time being roughly related to the maximum difference between S and P arrival times. Detection was declared if uncorrelated gaps at couples of stations had length below 3 s, mimicking maximum P wave arrival time differences across the selected stations. A sensitivity test was performed to determine the best set of the other parameters (number of hash tables, number of hash functions, number

of votes, minimum detection threshold; see [58]) minimizing the false negative events. An event was finally declared when detected at least at 2 stations.

**Figure 4.** (**a**) Selected stations for the FAST (Fingerprint And Similarity Thresholding) analysis: the dots represent the events automatically detected by ISNet. (**b**) Gutenberg–Richter (GR) analysis for the automatic (blue circles) and FAST (red filled dots) catalogs. For this latter case, the GR is also reported with the b-value estimation. The dashed lines mark the magnitude of completeness for the two catalogs. (**c**) Cumulative moment release according to the two catalogs. The black arrows indicate the first event automatically detected by ISNet (Ml 1.9) and the two main events of the sequence (Ml 3.0).

Scanning the continuous waveforms with FAST, we retrieved 342 events; the new catalog also includes all the 43 events automatically detected by the ISNet procedures. From a visual inspection of the traces, we retrieved only one false positive detection due to a teleseismic event.

For the computation of the magnitude, we imposed that all the events are co-located with the sequence centroid (40.94◦ N–15.15◦ E); the local magnitude was then estimated through the ratio of the maximum amplitudes with the events detected and located by the ISNet procedures.

We investigated the magnitude-frequency distribution for the refined catalog, comparing it with the ISNet automatic catalog (Figure 4b). We estimated a magnitude of completeness of Ml 0.2, nearly one unit smaller than that retrieved for the whole seismic network (Ml 1.1, [60]).

We also reported the b-value estimate from the refined catalog as b = 0.75 ± 0.04. When performing this computation on the automatic detections, we retrieved b = 0.70 ± 0.11, which is compatible with the FAST estimate, but shows larger uncertainty. Nevertheless, it is worth to note that some events having magnitude larger than 1.2 have been missed by the automatic procedures, since some of these events fall in the coda of the previous ones. The cumulative seismic moment release (Figure 4c) is not significantly different for the two catalogs yielding an equivalent total magnitude of 3.5. However, we can recognize a foreshock sequence of 23 events preceding the two largest magnitude events in the

sequence, that started about 9 h before the first automatic detection (Ml 1.9, 3 July 2020, 09:31).

#### **5. Accurate Earthquake Location**

The absolute locations of the sequence events have been obtained using the NLLoc software [1]—http://www.alomax.net/nlloc—that implements a non-linear, global-search probabilistic location in 3D velocity models.

For double-difference (DD) locations, we applied NLDiffLoc [48], a location tool included in the NLLoc software, that allows to perform DD locations by inverting the differential times through a probabilistic, non-linear approach. We also used the tool Loc2ddct, which allows to calculate differential times; initial absolute location of events and corresponding differential travel-times were used as inputs for DD locations. The DD algorithm performs an optimized exploration of the model parameters space using the annealing Metropolis algorithm [1] seeking a solution that maximizes the likelihood function. The latter is based on the misfit between measured and calculated differential phase arrival times. During the exploration, the algorithm computes the posterior probability density function (PDF), which represents the complete solution of the earthquake location problem. The significance and uncertainty of the solution, i.e., the maximum likelihood point, cannot be assessed independently of the complete PDF. Indeed, Gaussian or normal estimators, such as the expectation value and covariance matrix, can be obtained from samples of this function. These estimators can describe location uncertainty in the case of a (non-) linear PDF with a single maximum and an ellipsoidal form [1]. In addition, the software can handle 3D velocity models, with arbitrary complexity and parameterization, required in the case of crustal volumes with strong lateral variations and irregular topography [61].

We analyzed 903 traces at 21 stations from the 43 best recorded events. Within the manual picking, a weight factor inversely proportional to the uncertainty on arrival time picking has been assigned. In this way, we obtained an average of 14 P and S arrival times for each event, ranging from a minimum of 5 phases to a maximum of 31 phases. We processed the first arrival times with NLLoc using the 3D P- and S-wave velocity models optimized for the Irpinia area through an iterative, linearized, tomographic approach in which the P and S arrival times are jointly inverted for earthquake location and velocity determination [62].

For the DD locations, we performed a selection on the events based on the absolute location quality (i.e., at least 8 phases and an azimuthal gap <230◦), keeping only 36 events for the analysis. The absolute arrival times of the selected events are combined to obtain about 4000 P and S differential times used as inputs for the DD location. Since the sequence is clustered, we did not impose a maximum distance between event couples.

The absolute locations show events distributed along the NW-SE direction, with an extension of the pattern of about 4–5 km. The depth of the events ranges from 5 to 15 km. The horizontal and depth errors are within 750 m for most of the events. The rms of the resulting absolute locations is within 0.2 s for almost all the events.

From the absolute location, we achieved the final DD locations (Figure 5). After relative location, the events appear more clustered with an extension of the pattern covering a size of 2–3 km and a clear alignment in the NW-SE direction (Figure 5a). The event depth ranges between 7.5 and 12.5 km (Figure 5b). In a section view, these events are distributed according to a direction inclined of about 45–55◦ with respect to the horizontal plane, with a NE immersion (Figure 5c,d). Horizontal and vertical location errors are within one hundred meters (Figure 5e–g) and the rms is within 0.06 s for most of the events. Since the location errors are estimated analyzing the likelihood function in the vicinity of the best solution, the resulting errors could be underestimated in the case of multiple maxima in the PDF. However, in our case, a DD location rms of 0.06 s indicates that the hypocenters are located within a sphere of about 300 m radius, if considering a P-wave velocity of 5.5 km/s at 8–10 km of depth [62]. Since the rms also accounts for the uncertainty on the origin time,

the effective uncertainty on spatial coordinates could be smaller, coherently with estimates from the posterior PDF.

**Figure 5.** Double-difference (DD) locations. (**a**) DD location of 36 events of the sequence located with NLDiffLoc in a 3D velocity model. The color of the event hypocenters is associated with time of occurrence (from yellow to red) and the size to their magnitude. The stars represent the two Ml 3.0 events. For these events, we also reported the focal mechanism. (**b**) East– west vertical section of the events. (**c**,**d**) Cross-section projection of events along the profiles AA and BB , respectively, indicated in the map of panel (**a**). (**e**) Histogram of horizontal location errors. (**f**) Histogram of vertical location errors. (**g**) Histogram of DD location rms. We reported the mean and the standard deviation of the rms distribution.

#### **6. Source Parameters**

We chose the Brune source model [63] to infer source parameters (seismic moment, earthquake size, stress drop, and radiated energy released during the seismic event), which corresponds to an instantaneous pulse applied to the fault. We applied the SPAR (Source PARameters estimator) technique [64], which inverts the observed displacement spectra relying on a probabilistic framework based on the conjunction of states of information between data and model spaces. The inversion strategy allows to jointly retrieve source parameters along with their uncertainties and to investigate the between-parameter correlations. In this approach, theoretical Green's functions are evaluated in a simplified model accounting for both geometrical spreading and anelastic attenuation. We assumed a constant, frequency independent quality factor, which corresponds to the median value Q = 226 retrieved for the area [65]. For the Irpinia region, it has been demonstrated that a frequency independent quality factor is a model preferable to a frequency dependent quality factor following a power law [65]. Beyond the uncertainty estimation, the shape of the posterior PDF is also used to evaluate the quality of the retrieved estimations and eventually to discard unconstrained solutions. This technique has been efficiently tested and validated on the main events (M > 4.0) of the 2016–2017 central Italy sequence [64].

In Figure 6a, the displacement amplitude spectra have been plotted for the Ml 3.0 event occurred on 3 July 2020, at 16:14. The several curves refer to the diverse stations at which the parameters are constrained during the inversion. For this event, we estimated an average moment magnitude of Mw = 2.91 ± 0.02, a corner frequency of fc = 5.0 ± 0.2 Hz, and a spectral fall-off γ = 2.42 ± 0.04. Source parameters have been estimated for 36 events

in the sequence, the same events for which we got accurate double-difference locations. In Figure 6b, we represent the retrieved corner frequencies as a function of the moment magnitude (red points). Despite the large scattering in the data, we recognize that larger magnitude events (Mw > 2.0) show larger stress drops (between 1 and 10 MPa). At smaller moment magnitudes (1.0 < Mw < 2.0), the stress drop is on average smaller, with most of the events having a stress drop between 0.1 and 1 MPa. However, about one-third of the events in this magnitude range shows a stress drop larger than 1 MPa, excluding a saturation of the corner frequency at small magnitude. In Figure 6b, we also represent the corner frequencies as a function of the moment magnitude retrieved for ~720 events occurred in the Irpinia, with magnitude between 1.0 and 3.5 (black points) [65]. We found that our results are very consistent with the findings of [65]. Assuming self-similarity, we retrieved an average stress drop of 0.64 MPa. It is worth to note that estimations of corner frequency for the smallest magnitude events (M < 1.5) may be biased due to limited bandwidth when inverting the spectra. For events with Ml > 1.5, we also retrieve consistent estimates of the moment magnitude as compared to the local magnitude.

**Figure 6.** (**a**) Displacement spectra for the main event (Ml 3.0, 3 July 2020, 16:14) of the sequence; for this event, we estimated an average moment magnitude of Mw = 2.91 ± 0.02, a corner frequency of fc = 5.0 ± 0.2 Hz, and a spectral fall-off of γ = 2.42 ± 0.04. The intra-event variability on the seismic moment and corner frequency ranges over one order of magnitude. The red arrows point to the mean values of the parameters. (**b**) Plot of the corner frequency as a function of the moment magnitude (this study—red points; seismicity of the Irpinia region from [65]—black points). Straight lines individuate the curves along which the stress drop is constant.

From the computation of the corner frequencies, we also retrieved the source radius for each of the analyzed events. In Figure 7a, we show the source radius as a function of the moment magnitude: we found that the radius ranges in the interval 150–400 m for the largest magnitude events in the sequence and between 30 and 60 m for an Mw 1.5 event. Looking at the double difference locations, we see that the spatial extension of the events in the sequence is of the same order of magnitude (hundreds of meters) as the size of the largest magnitude events, indicating that the regions fractured by the different events are contiguous and stress transfer is likely to be the main mechanism for event production.

We finally estimated the radiated seismic energy, relying on a time domain estimator [21], which is grounded on the computation of the squared velocity in the time domain integrated over the S-wave time window [66], after its correction for attenuation along the path. The attenuation model has been determined considering ~2300 local earthquakes recorded by INFO over the last ten years [21].

The velocity integral is computed considering band-pass filtered signals between 0.5 and 40 Hz, and a time window starting 0.1 s before the S-wave onset and ending at different percentages of the cumulated energy as a function of the source to site distance R: (i) 90% when R < 25 km; (ii) 80% when 25 km < R < 50 km; (iii) 70% when R > 50 km. In addition, we imposed a minimum time window length of 5 s and a maximum time window length of 20 s. The radiated energy is finally obtained by averaging the estimates over several

recording stations (minimum of 3 stations; the sum of signal to noise ratio for the three components ≥200). In Figure 7b, we represent the radiated energy as a function of the seismic moment. The energy-to-moment scaling observed for the sequence (red squares) is in good agreement with values estimated for the Irpinia seismicity (gray points). The radiated energy trend with seismic moment is also consistent with an average stress drop in the sequence smaller than the value assumed by Kanamori as global average (from 2 to 6 MPa) while the steeper slope suggests a possible deviation from self-similar behavior.

**Figure 7.** (**a**) Source radius as a function of the moment magnitude for the events in the sequence. The source radius ranges from about 50 m for a Mw 1.5 event to about 250 m for a Mw 3.0 event. (**b**) Scaling of the radiated energy as a function of the seismic moment. The red points correspond to events of this sequence, the gray points represent background events recorded by ISNet.

#### **7. Focal Mechanism Solutions**

We computed 21 fault plane solutions from the inversion of P-wave polarities for earthquakes with local magnitude ranging between 1.2 and 3.0. Focal mechanism solutions are calculated from the inversion of P-wave polarities using FPFIT code [67] for all the events showing at least 5 P-wave polarities. The two main Ml 3.0 events show a similar normal-fault kinematics with a minor strike-slip component. The nodal planes have NW-SE trending and a dip of about 50–60◦. The remaining solutions show a common normal faulting style with a minor and variable strike-slip component.

We classified the fault plane solutions according to the plunge of P- and T-axes to derive the tectonic regime in which the seismic sequence originated. As shown in Figure 8, most of the solutions (~66%) belong to a pure normal fault regime while few solutions (~34%) belong to a normal strike-slip regime. In agreement with this classification, Pplunges range between 45◦ and 85◦ and T-plunges range between 0◦ and 85◦. Moreover, T-axis orientations show an azimuth ranging between 25◦ and 97◦, in accordance with the regional stress tensor calculated for the Irpinia region [47].

We calculated the composite focal mechanism using the data of the two main earthquakes (Ml 3.0), as shown in Figure 8. The normal fault kinematics of the solution with NE-SW trending nodal planes (320, 55, −120; 185, 45, −54) is compatible with the fault plane solutions calculated for single earthquakes and well fits all the available polarity data. Despite the uncertainty of the solutions due to the small number of available data and the small size of the events, a good agreement across the solutions is reached. A large group of seismic stations show the same type of P-wave polarity highlighting a similar rupture kinematics and fault plane geometry during the evolution of the seismic sequence. In the lower panel of Figure 8, all the available polarities data are plotted on the composite focal mechanism solution obtained from the inversion of P-wave polarities for the two Ml 3.0 earthquakes of the sequence. Polarities from six stations (LIO3, MNT3, RDM3, SALI, SSB3, VDS3) are the same during the whole sequence, and 4 stations (AND3, CLT3, COL3, RSF3) show a dominant polarity. Only two stations (NSC3 and SNR3) display a large variability, not always consistent with composite fault plane solutions. The results are summarized in

Table 1. Discrepancies and variability can be explained with the station location on the focal sphere, close to nodal planes: if uncertainties associated with location and focal mechanism are considered, a little rotation of the rupture plane can justify the polarity variation.

**Figure 8.** (**a**) Fault plane solutions computed for 21 earthquakes of the sequence. Earthquakes with Ml 3.0, 2.5 ≤ Ml < 3.0, and 2.0 ≤ Ml < 2.5 are highlighted in red, orange, and yellow, respectively. (**b**) Composite focal mechanism solution obtained from the inversion of P-wave polarities for the two main events. (**c**) Fault plane solution classification according to the plunge of P- and T-axes with specific tectonic regimes (Legend: NF, normal fault; NS, normal-oblique; SS, strike-slip; TF, thrust fault; TS, thrust oblique; U, unknown). The number of earthquakes (color bar) is counted in bins of 15◦ × 15◦.


**Table 1.** 235 P-wave polarity data available for the whole seismic sequence. Data are organized by seismic stations.

Focal mechanism solutions of the main earthquakes show the activation of a NW-SE striking fault structure with 50–60◦ dip, in agreement with seismogenic sources of the Irpinia region. Despite the low magnitude of the analyzed earthquakes, fault plane solutions reveal a normal fault tectonic regime, consistent with the regional stress field. Considering the spatial distribution of the hypocenters, the NE-dipping nodal plane can be assumed as the preferential one along which the seismic sequence originated, showing the same orientation of the adjacent seismogenic fault segment activated during the 1980 Irpinia earthquake.

#### **8. Ground Motion**

The ground motion characteristics were investigated through the main properties of peak values both in terms of acceleration and velocity. A first rough estimation of the frequency content associated with ground motion records can be obtained evaluating the ratio of peak ground acceleration (PGA) to peak ground velocity (PGV) [68,69]. We reported an average ratio value of 4.1 ± 1.6 g/ms−<sup>1</sup> which allows to classify the records in the class of high acceleration—low velocity. Such a high ratio suggests that the records are mainly characterized by a short duration, high predominant frequencies, and narrow-band spectra [70].

We also compared the peak values to the ground motion prediction equation (GMPE) inferred for the Irpinia region [71]. The comparison for the Ml 3.0 event of the sequence (3 July 2020, 16:19) is shown in Figure 9a. We observed a good agreement between predictions and observations; these latter are almost all included within one standard deviation in the predictions of the GMPEs.

**Figure 9.** (**a**) Comparison between observed peak ground values and the GMPEs (Ground Motion Prediction Equations) [71] for the Ml 3.0, 3 July 2020, 16:19 earthquake. Lines identify the mean GMPE, and the mean plus/minus one standard deviation. Open circles represent peak observations. (**b**) Shakemap® computed for instrumental intensity, for the same event. The red star corresponds to the event epicenter.

We also present, in Figure 9b, the Shakemap® [72] for the same event, in terms of instrumental intensity, automatically generated at ISNet. The intensity distribution is slightly elongated toward the south direction, with respect to the earthquake epicenter. The maximum instrumental intensity has been estimated to be III, which is consistent with reports provided by the INGV.

#### **9. Early Warning Analysis**

During the sequence, two earthquake early warning (EEW) systems, PRESTo and SAVE, were operating at INFO. PRESTo [23] is an open-source software platform for regional (network-based) EEW which integrates algorithms for real-time data collection, event detection, rapid earthquake location, magnitude estimation, and real-time ground motion prediction in the area of interest.

For the current setting at the INFO, the alert in PRESTo is released if at least five stations have detected the event, independently of the estimated event magnitude or shaking intensity in the area.

SAVE is an on-site, P-wave-based EEW approach [73] which has been conceived to operate either with a single station (i.e., a single sensor located at the target site) or with a set of co-located seismic nodes within a small area around the target to protect. SAVE processes the vertical component of both accelerometers and (broadband) velocimeters, and predicts the expected ground shaking at the recording site issuing a local alert level, together with a qualitative assessment of the earthquake magnitude and source-to-site distance, based on measurements on the early portion of the P-wave.

Unfortunately, during the early days of the sequence, some stations of the network suffered from a temporary failure of the communication system. This resulted in tremendous delays of data transmission and in the partial loss of a few recorded data, which prevented EEW systems from correct operation. For this reason, we could not evaluate the performance of the systems for the whole sequence, but each system automatically detected a smaller number of events, depending on the availability of real-time recorded data.

PRESTo detected a total of 21 events of the sequence, but only 10 of these events (with local magnitude between 1.4 and 2.8) did not suffer for real-time data communication problems and had real-time EEW estimates. The list of the detected events is available in Table 2. For each detected event, the table shows the comparison between the local (Ml) magnitude (as computed by INFO) and the estimated magnitude by PRESTo at the first alert. The difference between the first PRESTo estimate and the local magnitude is also reported in the table (ΔM). From the analysis of the PRESTo outputs, we found that the first magnitude estimate is available on average 3.9 s after the first P wave detection at the network and at the same time, the average difference between the estimated magnitude and the bulletin one is 0.1 unit.


**Table 2.** List of events detected by PRESTo. The event is identified by its ID, as shown in the ISNet bulletin (http://isnet-bulletin.fisica.unina.it/cgi-bin/isnet-events/isnet.cgi), the local magnitude, the first estimate from PRESTo and the difference between the two magnitudes.

Figure 10a shows the results of PRESTo during the Ml 2.7, event, that occurred on 4 July 2020 at 12:34, while Figure 10b shows the evolution of real-time estimates of magnitude, location, and number of data providing information to the system. As it can be seen from the plot, the first estimates of location and magnitude are available at about 5 s from the origin time, using 5 stations. After 2 s, the estimates of location (epicenter and depth) and magnitude converge and stabilize to the real values. We computed the available lead-time at the main cities of the Campania region, as the difference between the theoretical arrival time of the S-wave and the time of the first alert release. The lead-times are reported in Table 3. The lead-time ranges from 5.8 s at the city of Avellino (~25 km from the event epicenter) to 17.5 s at Naples (~70 km away from the epicenter).


**Table 3.** Theoretical lead-time. The table shows the epicentral distance and the theoretical lead-time for the main cities of the Campania region.

Finally, for this event, we theoretically estimated the radius of the blind zone (i.e., the area where the S-wave arrives before the alert release) which turned out to be around 15 km from the epicenter.

As for SAVE, at the time of the sequence, the system was running at 3 stations of the network: RSF3, COL3, and AVG3. The station RSF3, which is the closest to the sequence, detected 24 events, COL3 detected 11 events, while no events were detected at the farthest station AVG3.

Figure 11a shows a screenshot of the same Ml 2.7 event analyzed by PRESTo at RSF3. For this event, using 3 s of P-wave, the on-site EEW system estimated a local intensity of III, and no warning was declared at that station. For all the detected events, the magnitude and distance estimates were not available, due to the low signal-to-noise ratio that prevented from the use of the average period of the P-wave as a proxy for the earthquake magnitude. Instead, for all the events, SAVE predicted the expected ground motion intensity at each site. Figure 11b,c shows the cumulative performance in terms of lead-time and predicted vs. observed intensity for RFS3 (Panel b) and COL3 (Panel c).

Intensity estimates differ from the real ones of 0 or 1 unit. For the detected events, the available lead-time at the considered stations ranges between 1 to 6 s.

Finally, at the time of the sequence, the ISNet EWapp [74] was under testing by a limited number of users. The smartphones were distributed over the area, but the number was too small to perform a reliable statistical analysis. However, we can report here that the ISNet EWapp received the alerts provided by PRESTo and correctly predicted the expected intensity at each site. For all the events, the predicted intensity never exceeded the threshold of intensity IV at the location of the smartphones.

**Figure 10.** Performance of PRESTo during the Ml 2.7 event that occurred on 4 July 2020 at 12:34. (**a**) Screenshot of PRESTo during the event. (**b**) Real-time estimates of magnitude and location, and the number of available data as a function of time. From top to bottom, the figure shows: magnitude, epicentral error, depth error, number of stations with available 2 s of S wave (2S), 4 s of P wave (4P), 2 s of P wave (2P), and number of available P-wave picks.

**Figure 11.** Performance of SAVE. (**a**) Screenshot of SAVE during the Ml 2.7 event that occurred on 4 July 2020, at 12:34, recorded at RSF3. (**b**) Performance of SAVE at RSF3 and (**c**) COL3. Performances are shown in terms of lead-time (left side) and difference between predicted and observed intensities.

#### **10. Discussion and Conclusions**

In this study, we have analyzed a seismic sequence, detecting more than 340 events with magnitude in the range −0.5 < Ml < 3. The sequence occurred at the northern tip of the main segment of the 1980, M 6.9, Irpinia earthquake, at the boundary of the Monte Cervialto portion of the fault. Double difference locations and focal mechanisms agree in recognizing that the sequence ruptured an asperity along NE-dipping plane, with a dip angle ranging between 50◦ and 60◦. Despite large uncertainties in the smallest magnitude event locations, the hypocenter distribution delineates a structure dipping at 50–55◦ at depths of 10 to 12 km and becoming steeper at shallower depths (between 7.5 and 9 km), with a dip angle of 60–65◦, eventually being a portion of a listric fault. Strike and dip directions are consistent with the ones of the first fault segment that ruptured during the 1980 earthquake [42], indicating that this asperity is just ahead of the northern endpoint of the 1980 event, on the continuation of that segment. Most of the events within the sequence occurred at depths between 10.5 and 11.5 km, indicating that most of the slip was released at a depth comparable with that of the 1980 hypocenter [40]; according to tomographic models of the area [15,34], most of the events either developed in the basement or at its top with few events rupturing the upper Apulian carbonates. This is a common feature of the sequences that occurred in this region [4].

The size of the asperity, inferred from earthquake location is of the order of 800–900 m, corresponding to an approximate source radius of 400–450 m (Figure 5). On the other hand, the cumulative seismic moment results in an equivalent seismic event with moment magnitude of 3.5. If we estimate the source radius for this latter event, using the average stress drop retrieved from the analysis of source parameters, we get a value fully consistent with the extension inferred from earthquake location. When representing on the fault plane the seismic events within their own size, most of the events look contiguous, indicating that the sequence ruptured a single patch along the fault plane. Looking at the inter-event distance (median value of 370 m and standard deviation of 340 m), as compared to the size of main events, we argue that the dominant triggering mechanism within the sequence is the dynamic and static stress transfer, that allows the nucleation of individual events in the sequence.

From the stress drop analysis, we also recognize a strong heterogeneity in the stress release. Despite the large uncertainty in the evaluation of the stress drop, we reported a dual behavior for the earthquakes in the sequence: largest magnitude events featured a stress drop of 1–10 MPa, while stress drop of most of the small magnitude events ranges in the interval 0.1–1 MPa. Those values are fully consistent with the estimates retrieved for the seismicity of the whole region [65]. The largest stress drop values in the sequence are comparable with the stress drop estimated for the 1980 earthquake [75], while smaller values of stress drop in the area are ascribed to high pore-pressure that locally decreases the normal stress [65]. This variability has been observed also in the computation of the apparent stress over a much larger time interval and at the scale of the whole network [21].

It is worth to note that the retrieved stress drop is based on the estimation of the corner frequency and seismic moment from the inversion of the observed displacement spectra, fixing the path correction due to anelastic attenuation to the value found in [65]. To check the robustness of the stress drop estimates, we performed synthetic tests changing the value of the quality factor. Specifically, we have investigated the hypothesis that small events have the same stress drop as the largest magnitude events in the sequence, but their stress drop appears smaller because of an incorrect correction for path effects. Retrieving a stress drop one order of magnitude smaller than the input value at a source-receiver travel time of 5 s, requires an average quality factor as low as 130. This average value is very unlikely in the quality factor distribution represented in [65], when five or more estimates of the source parameters are available. Alternatively, an apparent stress drop at small magnitudes could be due to a complex, frequency dependent quality factor. This hypothesis cannot be rejected a-priori and requires further modeling and testing with respect to the analysis performed in [65], which is beyond the scope of this study.

The sequence started with a series of cracklings with M < 1 events and small stress drops, possibly indicating a fluid induced sequence initiation promoted by low normal stress and local fault lubrication. After the occurrence of many of those small magnitude events, the stress accumulated in the main asperity has reached the yield strength, releasing a large amount of stress in the two mainshocks of the sequence. In the later stage of the sequence, we reported again small stress drop events, some of which also occurred at few kilometers away from the sequence centroid, in the Apulian carbonates layer; these events could be related to local fluid diffusion, activating small pre-existent fractures prone to rupture.

Since the sequence occurred close to the main segment that ruptured during the 1980 earthquake, we may question why this sequence did not nucleate a large earthquake similar to the 0 s event of the Irpinia earthquake. This may be related to the fact that the sequence does not occur on the prolongation of the Monte Cervialto segment, but on a subparallel fault, or the Monte Cervialto fault has a stepover, with a geometrical discontinuity that prevented the jump on the main segment activated during the 1980 earthquake. Additionally, it is possible that the endpoint of the Monte Cervialto fault is a mechanical barrier, with large yield strength not overcome after stress release during the sequence. Investigating the geometrical details of these faults at such a small scale requires the development of new tools that will allow location and characterization of those events of small magnitude (M < 0.5) buried in the noise, and here detected with autocorrelation techniques.

Despite missing some information at the decametric scale, detection of events with Ml < 1, featuring low signal to noise ratio, is very useful in characterizing the time evolution of the sequence and in identifying potential foreshock activity. This sequence has indeed started 16 h before the main events, with initial seismic activity mainly characterized by small cracklings with Ml < 0.5. Thus, only such high-resolution catalogs, extracted with automatic autocorrelation techniques, enable to catch the foreshock initial phase of the sequence, as also pointed out by other studies [17]. It is worth to note that we performed an offline analysis of the events, with autocorrelation techniques running after the occurrence of the main events of the sequence. The near real-time analysis looking at the continuous data before the occurrence of the mainshocks merits further investigation, which is beyond the scope of this study. Within the improved catalog, we also reported a b value of b = 0.70, which is significantly different from the one related to the background seismicity of the area (b = 0.93, [60]). This decrease in the b-value can be related to an increase of the differential stress [76], associated with fluid pressure.

Since the sequence occurred in a sector where the network density is higher and the EEW systems worked at least for a part of the sequence, analysis of the system performance can highlight advantages and limits of actual systems. Specifically, we can individuate a sort of "minimum size" of the blind zone for the regional system, which has been estimated to be 15 km. This size represents the actual limit of regional EEW systems, within which no actions can be activated at the occurrence of an earthquake. This poses also a lower limit in magnitude (Ml < 6), below which EEW is not a viable risk reduction tool, because the blind zone almost superimposes with the area damaged by the earthquake. Onsite systems can provide a still positive, albeit very small lead-time for targets within the blind zone of the regional system, as also shown for the 2016–2017 central Italy sequence [77]. A contraction of the blind zone can be also obtained by estimating the earthquake size on an expanding P-wave time window, using for instance, the shape of the logarithm of the peak displacement in the time domain (LPDT curves [78]). When applied to this sequence, the average discrepancy between the predicted moment magnitude and the value provided by SPAR is 0.36, which is slightly larger than the average uncertainty on the magnitude estimate. This technique also provides a first rough estimate of the stress drop of the event, modeling the LPDT curve with a displacement function characterized by symmetric triangular function. For the sequence, the predicted average stress drop is 0.9 MPa, which is close to the estimate retrieved from spectral analysis.

**Author Contributions:** Conceptualization, G.F.; data curation, L.E., F.C. and R.R.; formal analysis, G.F., G.M.A., A.C., S.C., G.D.L., L.E., A.E., M.P. and A.S.; methodology, G.F.; software, L.E., M.P. and F.C.; supervision, A.Z.; validation, S.G., A.G.I., S.N., G.R. and S.T.; writing—original draft, G.F. and G.M.A.; writing—review & editing, A.C., S.C., G.D.L., A.E., M.P., A.S., S.G., A.G.I., S.N., G.R. and A.Z. All authors have read and agreed to the published version of the manuscript.

**Funding:** Data: data products and service provision of this research are funded by MUR, Ministero Università e Ricerca, through the project EPOS-Italia, and by DPC, Dipartimento di Protezione Civile, through a collaboration agreement with the University of Naples Federico II. Part of the research has been founded by the national project PRIN FLUIDS, grant number 20174 × 3P29.

**Institutional Review Board Statement:** Not Applicable.

**Informed Consent Statement:** Not Applicable.

**Data Availability Statement:** Publicly available datasets were analyzed in this study. This data can be found here: http://isnet-bulletin.fisica.unina.it/cgi-bin/isnet-events/isnet.cgi.

**Acknowledgments:** We are grateful to three anonymous reviewers who contributed to improve the manuscript. The geological map of Figure 2 has been redrawn from the map of [79].

**Conflicts of Interest:** The authors declare no conflict of interest. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or in the decision to publish the results.

#### **References**


### *Article* **Assessing Current Seismic Hazards in Irpinia Forty Years after the 1980 Earthquake: Merging Historical Seismicity and Satellite Data about Recent Ground Movements**

**Aldo Piombino 1,\*, Filippo Bernardini <sup>2</sup> and Gregorio Farolfi <sup>1</sup>**

	- filippo.bernardini@ingv.it

**Abstract:** Recently, a new strain rate map of Italy and the surrounding areas has been obtained by processing data acquired by the persistent scatterers (PS) of the synthetic aperture radar interferometry (InSAR) satellites—ERS and ENVISAT—between 1990 and 2012. This map clearly shows that there is a link between the strain rate and all the shallow earthquakes (less than 15 km deep) that occurred from 1990 to today, with their epicenters being placed only in high strain rate areas (e.g., Emilia plain, NW Tuscany, Central Apennines). However, the map also presents various regions with high strain rates but in which no damaging earthquakes have occurred since 1990. One of these regions is the Apennine sector, formed by Sannio and Irpinia. This area represents one of the most important seismic districts with a well-known and recorded seismicity from Roman times up to the present day. In our study, we merged historical records with new satellite techniques that allow for the precise determination of ground movements, and then derived physical dimensions, such as strain rate. In this way, we verified that in Irpinia, the occurrence of new strong shocks—forty years after one of the strongest known seismic events in the district that occurred on the 23 November 1980, measuring Mw 6.8—is still a realistic possibility. The reason for this is that, from 1990, only areas characterized by high strain rates have hosted significant earthquakes. This picture has been also confirmed by analyzing the historical catalog of events with seismic completeness for magnitude M ≥ 6 over the last four centuries. It is easy to see that strong seismic events with magnitude M ≥ 6 generally occurred at a relatively short time distance between one another, with a period of 200 years without strong earthquakes between the years 1732 and 1930. This aspect must be considered as very important from various points of view, particularly for civil protection plans, as well as civil engineering and urban planning development.

**Keywords:** Irpinia; seismic hazard; earthquake; strain rate; GNSS; InSAR

#### **1. Introduction**

This study is based on the analysis of a fine-scale ground velocity map of Italy determined by the fusion of Global Navigation Satellite Systems (GNSS) with synthetic aperture radar interferometry (InSAR) data derived from satellites [1]. The dataset derives from a period of observation between 1990 and 2012. The InSAR dataset is part of the "Piano Straordinario di Telerilevamento" (Special program for Remote Sensing, promoted by the Italian Ministry of Environment). Due to the quasi-polar orbit of the satellites, space-borne InSAR observations can only determine the East–West (E–W) and Up–Down (U–D) components of the movement of persistent scatterers. However, there are millions of scatterers that are unreachable, due to the fact that only a few hundred GNSS stations exist. The North–South (N–S) component is provided by a *C*<sup>2</sup> continuous bi-cubic interpolation function that is well suited to interpolate sparse GNSS stations displaced inside

**Citation:** Piombino, A.; Bernardini, F.; Farolfi, G. Assessing Current Seismic Hazards in Irpinia Forty Years after the 1980 Earthquake: Merging Historical Seismicity and Satellite Data about Recent Ground Movements. *Geosciences* **2021**, *11*, 168. https://doi.org/10.3390/ geosciences11040168

Academic Editors: Sabina Porfido, Giuliana Alessio, Germana Gaudiosi, Rosa Nappi, Alessandro Maria Michetti and Jesus Martinez-Frias

Received: 26 October 2020 Accepted: 30 March 2021 Published: 7 April 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

the study site and surrounding areas. To do this it uses a hierarchical structure at different refinement levels.

The fusion of GNSS with InSAR is a method based on the calibration of InSAR with GNSS measures derived from permanent stations and survey campaigns [2,3]. The results are a coherent fine-scale ground velocity map with a spatial resolution that is unreachable using the previous velocity field maps determined with the GNSS technique alone. By using this technique, Farolfi, Piombino, and Catani [1] provided new information about the complex geodynamics of the Italian peninsula and thanks to the high spatial resolution of the ground movements map, identified interesting patterns of small areas with respect to the surrounding ones. Moreover, their work confirmed the division of peninsular Italy into two sectors, with opposed E–W components of movement in the Stable Europe Frame (Figure 1). This has been depicted by older studies based only on GNSS station movements ([4] and references therein): the western block (Tyrrhenian) is moving westward, while the eastern one (Adriatic) shows an eastward movement. The relative motion of these blocks implies their divergence; the effect of which is represented by the numerous currently active normal fault systems along the central and southern Apennines which are close to the border between these two sectors—and the associated seismicity.

**Figure 1.** Map of the East component of the ground velocity field of the Italian Peninsula, derived from Global Navigation Satellite Systems (GNSS) and synthetic aperture radar interferometry (InSAR) during more than two decades of observation (1990–2017). The area of the Central Apennines presents major earthquakes from 1990 to present day. From the figure above, it is clear that the main seismogenic areas are linked to the boundary that divides the two blocks with opposite E–W components of velocity.

The Apennine chain, an approximately linear belt hosting the most rapidly slipping normal faults, and the most damaging earthquakes, are coincident with the areas in which the morphological surface height, when averaged on a horizontal scale of tens of kilometers, is greatest [5]. In this area, the first studies based on the relative movements of the GNSS stations have already determined a medium value of a ca. 3 mm/a extension, linked to the differential movements between the two blocks. This also allows the emplacement of melt intrusions along deep-rooted faults [6]—the last occurrence of this kind probably triggered the 2013/2014 Matese seismic swarm [7]—and the widespread emission of deep-originated CO2 [8]. This regime is dissecting the former Cenozoic east-verging thrust belt related

to the west-dipping subduction of the Apulian lithosphere [9]. This compressive regime ended at 650 ka in the middle Pleistocene [10].

The E–W component of InSAR movements [1] has also confirmed the frame depicted by [11], in which the Ortona–Roccamonfina is not a single lineament, but a 30 km wide deformation channel: this channel is characterized by prevalent west-directed velocities in the stable Europe frame, nested in the Adriatic eastward-moving block.

The vertical component of the InSAR data highlights the current general uplift occurring in most of Southern Italy, even if this uplift is lower than in the Central Apennines (especially in the "Abruzzo Dome" [1]), confirming a wealth of the geological literature. Conversely, few areas show subsidence, mainly because of human groundwater exploitation. In this frame, the highest uplift values of the whole Southern Apennines—exceeding 1.8 mm/a—are present in the chain segment between Benevento and Potenza. This area of higher-than-surroundings uplift roughly corresponds to the Irpinia sector, in a belt just west of the Campania–Puglia border. Thus, it is possible to call this area the "Irpinian Dome" (Figure 2). The Ufita and Marzano faults represent the surface traces of the two different patterns of the East–West ground velocity component (Figure 3 (top)).

**Figure 2.** (**Top**) Map of the distribution of persistent scatterers (PS) (red points) and the GNSS permanent stations (white points) involved in the detection of ground surface movements of the study area. The main geodynamic features are represented in the background of the map: the Irpinian Dome is the cyan area and the Ufita and Marzano faults are represented with black hashed lines. (**Bottom**) Map of the main seismic events (black circles) that occurred from 1466 to 2017 with the main towns highlighted (dark blue squares).

**Figure 3.** (**Top**) The East component of the ground velocity field of the Irpinia–Sannio area with the main earthquakes of M ≥ 3 occurred since 1900. For the earthquakes of M ≥ 4, the label represents the year of the occurrence. (**Bottom**) Map of the vertical component of the ground velocity in Irpinia with the main earthquakes that occurred from 1466 to 2017. The main towns are drawn and labeled with dark blue squares, and the main faults (see Figure 2 (top)) are represented with black hashed lines.

The uplifting area is divided into two different parts and, between them, exists a narrow corridor of lower uplift <1 mm/a, (Figure 3 (bottom)). It is interesting to note that this corridor is placed near the epicentral areas of the 1930 and 1980 earthquakes.

The geographical axis of the Irpinian dome is placed east of the main NE-dipping faults, on the surface projection of the hanging wall. Any useful information of the N–S component of the ground movement can be detected by InSAR satellites because their quasipolar orbits only make the detection of vertical and E–W velocity components possible.

#### **2. Relationship between Strain Rate and Earthquakes**

The strain rate provides a measure of the superficial deformation, and for this reason, is useful information for studying and analyzing geodynamics. Many authors have produced strain rate maps of the Italian territory using GPS station data. In the last decade, for example, Riguzzi et al., (2012) [12] estimated the strain rate, using the GPS velocity solution, of the Italian area—provided by Devoti et al., (2011) [13].

Palano (2015) [14] carried out an analysis of the stress and strain-rate fields of Italy. He performed a comparison of GPS inferred strain-rate data and 308 stress datasets interpolated at each node of a regular grid.

Montone and Mariucci (2016) [15] provided an updated present day stress map for the Italian territory combining seismicity, data retrieved from a breakout analysis in deep wells, and fault data. Starting from this base Mastrolembo and Caporali (2017) [16] presented a direct comparison of the principal horizontal directions of stress and strain-rate directions of extension, estimated at the position of each stress measurement in their data set. For this, they used GPS data coming from over 500 stations distributed on the Italian peninsula, however, they did not provide a general map.

This work instead benefits from a new fine-scale strain rate field of the whole continental Italy and Sicily (Figure 4) [17], determined from the surface ground movements map obtained by the satellite InSAR observations between 1990 and 2012 [1]. The twodimensional velocity gradient tensor is calculated by applying the infinitesimal strain approach [18,19] with a grid of 20 km × 20 km. The known horizontal incremental velocity vector *Vi* of the *i*-vertex polygon is defined as:

$$V\_i = A\_i + \frac{\partial V\_i}{\partial \mathbf{x}\_j} \mathbf{x}\_j = A\_i + t\_{ij} \mathbf{x}\_j \tag{1}$$

where *Ai* is the unknown velocity at the origin of the coordinate system, *xj* is the position of the station, and *tij* is the displacement gradient tensor. Following the tensor theory, we separated the second-rank tensor into a symmetric and an anti-symmetric tensor. Then, *tij* can be additively decomposed as follows:

$$t\_{ij} = \frac{\left(t\_{ij} + t\_{ji}\right)}{2} + \frac{\left(t\_{ij} - t\_{ji}\right)}{2} = \left.e\_{ij} + \omega\_{ij}\right. \tag{2}$$

The symmetric and anti-symmetric parts of the infinitesimal strain rates can be associated with the infinitesimal strain *eij* and rotation *ωij* tensors. Principal strains *e*1,*e*<sup>2</sup> were computed as:

$$
\epsilon\_1, \epsilon\_2 = \frac{1}{2} \left( \epsilon\_{\bar{i}\bar{i}} + \epsilon\_{\bar{j}\bar{j}} \right) \pm \frac{1}{2} \sqrt{\left( \epsilon\_{\bar{i}\bar{i}} - \epsilon\_{\bar{j}\bar{j}} \right)^2 + 4\epsilon\_{\bar{i}\bar{j}}^2} \tag{3}
$$

and the horizontal second invariant of the strain rate (*SR*) tensor was also evaluated as the scalars and is presented in Figure 4:

$$SR = \sqrt[2]{\varepsilon\_1^2 + \varepsilon\_2^2} \tag{4}$$

The determination of the second invariant of the strain rate provides important additional information to support the analysis of the geodynamics and the earthquake distribution of the study area. A recent study [17] based on the analysis of the seismic events that have occurred since 1990 in the Italian peninsula, shows that the probability of earthquakes occurring is linked to *SR* by a linear correlation. More specifically, the probability that a strong seismic event will occur doubles with the doubling of *SR*. Then, the *SR* is used as an independent and quantitative tool to spatially forecast seismicity.

The results of this study agree with these former studies, especially for the detection of the high strain rate along the Central and Southern Apennines axis and in Northern Sicily [19].

**Figure 4.** Map of the horizontal strain rate field of the Italian peninsula, determined by an infinitesimal approach from the horizontal velocity field derived from GNSS and InSAR during more than two decades of observation (1990–2017). Main earthquakes that occurred from 1990 to 2017 are represented on the map.

This new theory, based on observables, identifies significant earthquake (M > 5.5) prone areas with high strain rate areas. It gives a new perspective for the interpretation of recent earthquakes and this theory also predicts the events that occurred after the observation period of the study. For example, in the outer side of the Alps, the strongest earthquakes recorded in the past decades occurred in just two of the four areas of this sector showing high strain rate value: the M 5.3 20 December 1991 Graubunden [20] and the M 5.3 22 March 2020 Zagabria earthquake [21]. The others selected areas located around the cities of Marseille and Innsbruck. Furthermore, after 1990 in Italy, all earthquakes with M > 5.1 and a hypocenter less than 15 km deep only occurred in areas showing a high strain rate: 1997, 2009 and 2016 Central Apennines seismic sequences, the 2012 Emilia earthquakes, and the 2013 Lunigiana earthquake (on the surface of the Po Plain the *SR* is lower than in the buried and seismic Apenninic units because of its attenuation in the plastic Neogene sedimentary cover).

Outside these areas no shallow significant earthquakes occurred until 1990, even though strong events occurred after 1940, such as Friuli (the M 6.5 6 May 1976), Western Sicily (the Mw 6.4 15 January 1968), and Valais (the M 6.1 25 January 1946). In addition to these high strain rate areas that have been hit by strong earthquakes, there are others that, while showing high values of this value, have not been hit by relevant earthquakes since 1990. In recent years, only areas characterized by high strain rates have been affected by significant earthquakes, therefore, it is not unreasonable to empirically hypothesize that significant seismic events of the next decades have a greater chance of occurring only in the areas characterized by high strain rates. The year 1990 is taken as a milestone because, after beginning the survey in 1991, the former earthquakes do not influence the data.

#### **3. Stain Rate in Irpinia**

Irpinia is one of the main areas of the core of the Central and Southern Apennines chain. The differential movements between the two blocks, in which the Italian peninsula is divided imply a medium strain rate of 50 nstrain/a [5]. Here, the main fault systems are the Ufita, Monte Marzano, and Caggiano faults [22] (Figure 2 (top)). However, deformation is also linked to faults with a highly different orientation, well constrained in the historical record. For this issue, it is interesting to note that the focal mechanism solutions of the 1930 and 1962 earthquakes are significantly different from the kinematics of the typical large earthquakes that occurred along the crest of the Southern Apennines. Instead, these are well-fitted by the Mw 6.9 23 November 1980 earthquake, caused by predominant normal faulting along NW–SE-striking planes. The fault linked to the Mw 6.7 23 July 1930 earthquake is blind and its magnitude and focal mechanism are debated ([23] and references therein). Many focal mechanisms have been proposed, from a "classical" NW– SE to an ESE–WNW striking plane. These belong to an array of oblique dextral slips on the EW-trending planes crossing the whole Southern Apennines which is dissecting the orogen in various contiguous sectors. The level of the transcurrent component is debated as well. However, the effects of the earthquake presented in [24] fit better with a NW–SE striking fault.

The 1962 sequence is composed of three different shocks at 18:09, 18:19, and 18:44 UTC, the second being the most destructive (Io IX MCS, Mw 6.1, [25]). Additionally, identification of the faults responsible for these earthquakes is difficult because of the lack of reported surface faulting. Only in 2016 was a reliable focal mechanism produced [25] with two solutions: dominant strike-slip rupture along a north-dipping, E–W striking plane, or along a west-dipping, N–S striking plane. Its depth is still controversial, varying between 7 and 35 km. Therefore, the focal mechanism solutions of the 1962 earthquakes are significantly different from the kinematics of the typical large earthquakes occurring along the crest of the Southern Apennines, well-fitted instead by the Mw 6.9 23 November 1980 earthquake, caused by predominant normal faulting along NW–SE-striking planes.

Irpinia is one of the areas in Italy showing a higher strain rate (Figures 4 and 5) during the 1991–2011 InSAR survey: currently north of it, in the Sannio sector, the strain rate is at a low level with a value of 20 nstrain/a 10 km north of Benevento. However, to the SE of Benevento, the value increases to 35 nstrain/a in less than 30 km at Grottaminarda, reaching the highest levels (48 nstrain/a) 15 km south of the epicenter of the 23 November 1980 earthquake. Therefore, Irpinia is still currently one of the areas with a higher strain rate in Italy, with values always >32 nstrain/a, and showing a maxima over the hanging wall of the Monte Marzano fault system. The southward strain rate dramatically drops to 35 nstrain/a near Polla. However, while north of Irpinia along the chain axis the value drops rapidly under 30 nstrain/a, the southward values remain above this value for much longer, up to the Pollino line (the border between Central Apennines and Calabria–Peloritani arc). In the picture of the EW-trending lithospheric faults dissecting the Apenninic orogen, these sudden strain rate drops north and south of Irpinia can be related to different strain rate conditions occurring in the adjacent sectors.

**Figure 5.** Map of the horizontal strain rate field of Irpinia derived from GNSS and InSAR during more than two decades of observation (1990–2017). The main earthquakes that occurred from 1466 to 2017 are represented on the map. The main towns of the area are drawn and labeled with dark blue squares, and the main faults (see Figure 2 (top)) are represented with black hashed lines. Since the strongest shallow events that occurred inland in Italy from 1990 to today are placed only in areas characterized by high strain rates [17], the high strain rate detected in Irpinia implies—from a theoretical point of view—a scenario where a new strong earthquake seems more likely. This can be somehow counterintuitive, because this area hosted most of the strongest earthquakes in southern Italy after 1908, in 1930, 1962, and 1980: only the Mw 6.4 1968 Belice and the Mw 6.0 1978 Patti gulf events (both in Sicily) reached similar magnitudes [26]. Only in the NE Sicily 1978 earthquake was the strain rate as high as in Irpinia. Therefore, from this point of view, we can hypothesize that in Irpinia, the probability of a new strong event is still very high.

#### **4. The Historical Record of Earthquakes in Irpinia**

Additionally, historical seismicity can allow this—somehow unexpected—statement, given the time intervals between Irpinian earthquakes. "Irpinia" is a historical–geographical area of southern Italy, located in the Campania region, approximately corresponding to the territory of the current province of Avellino, which in turn, largely recalls the historic province of Principato Ultra of the Kingdom of Naples.

The Irpinia area is one of the most seismically active sectors of the entire Italian territory. The seismogenic belt that runs along the Apennine chain, in fact, crosses the northern and eastern part of the province of Avellino, where strong earthquakes have frequently occurred over centuries.

The most important historical seismic events are placed in the hanging wall of the Monte Marzano fault system [22]. If we take a polygon with vertices at the coordinate points 41.314◦ N, 14.971◦ E; 41.105◦ N, 14.874◦ E; 40.739◦ N, 15.352◦ E; 41.056◦ N, 15.574◦ E (depicted in dark red in Figure 6 (bottom)), corresponding to the Apennine seismic belt site of the major historical and instrumental seismicity, the parametric catalog of Italian earthquakes CPTI15 [26] reports about twenty earthquakes with magnitude Mw ≥ 5.0, starting from the year 1000 (see Table 1). Of these, seven have a Mw between 6.0 and 6.8. It must be said that the catalog can be considered complete, for the strongest events (Mw ≥ 6.0) only for the last 400 years, namely from 1620 up to today [27]. From the

diagram in Figure 6 (top), it can be seen that until the end of the 17th century, the seismic history of the Irpinia sector is largely incomplete and poorly documented. This, obviously, is not because there were no earthquakes at all, but because only little and partial historical information about that area for those ancient periods exists today. Only a couple of earthquakes are known (in 1466 and 1517) to have occurred in this period, plus two events before the year 1000, which occurred in the year 989 and 62 CE [28]; thus, outside the reference window of the historical catalog. Both these events originated from the monte Marzano Fault [29]. The earthquake of 5 December 1456 [30] was deliberately not taken into consideration in the present study, because it is a complex event that affected a very large area of southern Italy, causing damage from Puglia to Abruzzo, and whose epicenter is not well located nor defined. Probably, that earthquake was made up of several shocks that occurred in different sectors of the central–southern Apennines a few days apart, and Irpinia was only one of the several areas that were struck [28].

**Table 1.** List of the main Irpinia earthquakes (Mw > 5.0) extracted from the CPTI15 catalog [26]. For the description of the various parameters see this catalog at https://emidius.mi.ingv.it/CPTI15- DBMI15/index\_en.htm (accessed on 30 January 2021) As attested by Rovida et al. [27], this historical record can be considered complete since 1620 for M 6.0+ earthquakes.


A lack of seismic events in the historical record for a given area can be due to the following reasons:


The seismic history of Irpinia is better documented, starting from the end of 1600, and as minor events (4.0 ≤ Mw < 5.0) can be considered well documented only starting from the end of the 19th century (Figure 6)

**Figure 6.** (**Top**) Seismic history of the Irpinia area as from the CPTI15 catalog [26]. The dark red polygonal area shows the Irpinian main seismic belt, the coordinates of its vertices are: 41.314◦ N, 14.971◦ E; 41.105◦ N, 14.874◦ E; 40.739◦ N, 15.352◦ E; 41.056◦ N, 15.574◦ E. Such a historical record is well documented only for the last 400 years (namely since 1620 [27]), whereas for the previous centuries it is poorly known, with long timespans lacking information. The figure also shows that inside the catalog completeness span-time of 400 years (since 1620), there are a couple of evident clusters of strong earthquakes (Mw ≥ 6.0) which are 200-years apart. The first, between 1694 and 1732, when three M 6.5+ events occurred over a period of 38 years. The second, between 1930 and 1980, when three M 6.0+ events occurred over a period of 50 years. (**Bottom**). Map of the main historical seismic events of Irpinia from 1466 until 1982, reported in Table 1. The main towns of the area are drawn and labeled with dark blue squares, and the main faults (see Figure 2 (top)) are represented with black hashed lines.

From its seismic history it can also be seen that, over the 400 year time-span of seismic catalog completeness for M ≥ 6.0 events, in Irpinia, the strongest earthquakes (Mw ≥ 6.0) tend to group over time, spaced from long phases characterized by lower and less frequent

seismicity (Figure 6 (top and bottom)). At the turn of the seventeenth and eighteenth centuries, over a period of 40 years, Irpinia was affected by four damaging earthquakes, three of which occurred in just 10 years (1692, 1694, 1702, and 1732). Of these, the ones that occurred in 1694 (considered as a sort of twin of the 1980 earthquake), in 1702 and in 1732 were large events of Mw > 6.5. Each of these caused extensive destruction over large areas and many casualties. Another cluster of strong earthquakes is the one that hit the sector in the twentieth century, between 1930 and 1980 (three events with Mw ≥ 6.0 over a period of 50 years). So, Irpinia belongs to the belt of very high seismic hazard running along the Central and Southern Apennines (Figure 7).

**Figure 7.** Map of seismic hazards in Irpinia (see [32]) and adjoining areas (colors in the background), derived mostly from the historical seismic records, as shown by the overlapping of strong seismic events within the map itself. The main towns of the area are drawn and labeled with dark blue squares, and the main faults (see Figure 2 (top)) are represented with black hashed lines.

It is unlikely that in the 200-year time-span between 1732 and 1930 there were large (M 6.0+) earthquakes in the Irpinia area, since these are not present in the historical record. In the same time interval, not just "minor" earthquakes are well documented in the very same area (i.e., the 1741 Mw 5.4, 1794 Mw 5.3, and 1853 Mw 5.6 Irpinian events; see Table 1, Figure 5), but strong events are also well known to have struck other adjacent Apennine areas (the 1805 Mw 6.7 Matese earthquake; those of 1851 Mw 6.5 and 1857 Mw 7.1 in Basilicata [26]). Therefore, it can be assumed that the historical seismicity of Irpinia has been characterized by periods of intense activity, with strong earthquakes over a few years or decades, interspersed with long periods of minor-to-moderate activity, with earthquakes of magnitude lower than 6.0.

The spatio-temporal clustering of earthquakes in the Southern Apennines is well documented in the scientific literature. By comparing the number of earthquakes on record in the last five to seven centuries, with the number implied by slip-rates on active normal faults averaged over 18 kyrs in the Southern Apennines, Papanikolaou and Roberts [33] demonstrated that the long history of earthquakes in the Italian Apennines may indeed contain evidence for earthquake clustering. In particular, according to Papanikolaou and Roberts [33], Irpinia and northern Basilicata show a very high number of earthquakes and this indicates that this area may be in a temporal earthquake cluster phase. Meanwhile, the sector located slightly further south, up to the Pollino massif, could be in a temporary anti-clustering process.

The strain rate map in Farolfi et al. [17], in which the Irpinia–Basilicata sector is characterized by a much higher strain rate than the Pollino sector (and the intermediate sector, Vallo di Diano, shows intermediate values), fits well with these results.

In the Central and Northern Apennines, earthquake clustering is known to exist. For example, Tondi and Cello [34] observed a time interval of ca. 350 years among the beginning of seismic clusters in the Central Apennines Fault System. The current sequence, which started in 1997 and continued with the events of 2009 and 2016, has arrived on time if we consider that two main seismic clusters in the past began in the years 1349 and 1688 [35]. In the Northern Apennines, a major seismic crisis occurred between 1915 and 1921 [36], while in this area, the historical record before 1915 is composed of a few destructive events [26]. Additionally, the two-year period 2012–2013 showed a high level of activity, not only in the area of the Emilia seismic sequence, but also in the Garfagnana sector, accompanied by a high strain rate.

Currently, the most likely explanation for seismic clustering is the "stress transfer" between faults [36] and references therein, due to coseismic movement rearranging the Coulomb failure stress on other nearby faults [37]. However, this explanation falls short when there is the occurrence of an isolated, single event (such as the Mw 6.1 6 November 1599 Valnerina, and the Mw 6.4 13 January 1832 Valle Umbra earthquakes) that do not trigger a level of Coulomb stress transfer, resulting in strong earthquakes on other neighboring faults.

For other researchers, there is a sort of "domino effect" between the crustal blocks that make up the Apennines [38].

In conclusion, we suppose that the deformation rate value, as described in Farolfi et al. [18], represents a *conditio sine qua non* for the occurrence of strong earthquakes (M > 5.5). This hypothesis was corroborated by the observation that all the strongest shocks of the last three decades in the Italian territory are located in areas characterized by a high rate of deformation [17].

#### **5. Conclusions**

In the last twenty years, the main shallow earthquakes (depth ≤ 15 km) in Italy and the Alps have occurred only in some of the horizontal strain rate zones, as depicted by Montone and Mariucci [15]. Meanwhile, the strain rate is currently low in other areas affected by recent earthquakes that occurred before the 1990–2012 survey, such as Belice (1968) and Friuli (1976). These areas are also where the higher seismic events from 1915 to now have occurred. The area of the 1915 Marsica earthquake also shows lower-than-surroundings strain rate values, such as in the Central and Eastern sections of the Northern Apennines (in the Western sector, higher seismicity barely corresponds to a slightly higher strain rate). In this picture, the high strain rate level indicates that the scenario of a new strong shake in Irpinia is not unlikely. Additionally, the historical record is in agreement with this, given the short temporal distance between strong (M6+) seismic events in Irpinia during the 400 years of catalog completeness (i.e., from 1620 to present), and a long 200-years period without M6+ seismic events occurring between 1732 and 1930.

Moreover, by merging historical seismicity and InSAR satellite data, we think that, in the future, a hypothesis related to the following scenarios should be explored:


**Author Contributions:** Conceptualization, A.P., F.B.; historical and geographical investigation, F.B., A.P.; methodology, software, validation, formal analysis, data analysis, G.F.; investigation, resources, writing—original draft preparation, writing—review and editing, visualization, F.B., A.P. and G.F.; supervision, project administration, A.P.; funding acquisition, F.B. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research received no external funding.

**Data Availability Statement:** The Parametric Catalogue of Italian Earthquake CPTI15 (version 3.0) is free and available at: https://emidius.mi.ingv.it/CPTI15-DBMI15/description\_CPTI15\_en.htm (accessed on 30 January 2021). As to any other dataset or maps here provided in this study, they can be available upon request.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


*Article*
