On 24 August 2016, at 01:36:32 UTC, a Mw 6.0 earthquake originated at a depth of approximately 8 km underneath the relay zone between the overlapping Mt Vettore-Mt Bove normal fault (VBF) and the Mt Gorzano Fault (GF) in the Central Apennines of Italy (Figure 1
). The severe earthquake destroyed the town of Amatrice and many other villages, killing 299 people [1
]. The inversion of waveforms from strong motion accelerometers [2
] and numerical modeling of Differential Interferometry Synthetic Aperture Radar (DInSAR) measurements, which were integrated with structural geological data, suggest a seismogenic scenario characterized by a bilateral rupture propagating on the VBF and GF planes, conjoined at the base [1
Field surveys performed in the weeks following the earthquake allowed the primary ruptures, caused by surface faulting, along an ~5.8 km section of the southern part of the VBF to be mapped (Mt Vettore-Castelluccio di Norcia area, Figure 1
a); coseismic throws were found to range from 10 to 27 cm (an integral average of 12 cm) [1
]. Remarkably, no clear evidence of surface faulting was observed along the GF or along the northern part of the VBF, i.e., in the Mt Bove-Ussita area (Figure 1
a) highlighting that the coseismic rupture preferentially propagated from the hypocenter to the north along the southern VBF.
The VBF is a well-known active fault in the Central Apennines extensional belt with a 6.5 ≤ Mw ≤ 6.8 maximum expected magnitude [5
]. The 24 August earthquake was characterized by a NNW-SSE-striking focal mechanism that had normal kinematics (Figure 1
a) and was followed by a sequence of aftershocks with epicenters that were chiefly distributed in the southern part of the VBF hanging wall and northern part of the GF hanging wall [2
] (blue dots in Figure 1
Independent geological and seismological data confirm that the August 24 earthquake ruptured the VBF along a length of ~12 km [1
]. The coseismic deformation in the hanging wall of the ruptured fault section is inferred from DInSAR interferograms (Figure 1
] that show two NNW-SSE-striking coalescent depressions with a maximum displacement along the satellite’s ascending Line of Sight (LoS) of nearly 20 cm (blue lines in Figure 1
a). We confirmed the evidence by preparing a new coseismic interferogram using a pair of European Space Agency (ESA) Sentinel-1 images taken on 21 and 27 August 2016 (Figure S1
). The independent coseismic interferograms demonstrate that the northern boundary of the surface deformational pattern caused by the Mw 6.0 earthquake coincided with the northern tip of the coseismic rupture mapped in the field along the VBF (Figure 1
On 26 October 2016, at 19:18:05 UTC, a Mw 5.9 earthquake struck along the VBF, 25 km NNW of the epicenter of the 24 August event (Figure 1
a). On 28 and 29 October, we executed new field surveys along the VBF, and we found evidence of primary ruptures caused by surface faulting near the northern tip of the VBF, E of Ussita (Figure 1
a), with throws ranging from 8 to 15 cm [4
]. In the field, we did not observe evidence of new surface ruptures along the southern section of the VBF. The surface evidence of the earthquake was confirmed by a COMET interferogram [11
] that showed a subsiding synform striking NW-SE nearly parallel to the VBF and an associated 14 km-long surface faulting with a maximum displacement of ~18 cm (green lines in Figure 1
Four days later, on 30 October 2017, at 06:40:17 UTC, a Mw 6.5 earthquake (hereafter, mainshock) occurred along the VBF, with the epicenter between the two previous major earthquakes (Figure 1
b). As shown by DInSAR interferograms [10
] (black lines in Figure 1
b) and confirmed by field data [4
], the strong earthquake re-activated and ruptured nearly the entire surface trace of the VBF, for a total length 32 > L > 22 km, with local coseismic surface displacements reaching 222 cm and a maximum slip on the same segment of the VBF that ruptured on August 24 [1
To document the combined coseismic deformation produced by the Mw 5.9 and Mw 6.5 earthquakes, we prepared a coseismic interferogram using a pair of ESA Sentinel-1 images taken on 26 October and 1 November 2016 (Figure S2
). The interferogram showed a complex, ~30 km long and ~8-to-12 km wide, NW-SE-striking area of deformation W of the VBF. The interferograms (Figure 1
b and Figure S2
) showed a widespread offset of the fringes along the western slope of Mt Vettore, with surface deformational gradients up to 30 cm·km−1
. We estimate that the coseismic deformation exceeded 90 cm on the Castelluccio di Norcia plain (Figure 1
b), a value which is also confirmed by the analysis of local GNNS data [13
]. We emphasize, as the seismological and geological data confirm, that the three major earthquakes originated by ruptures nucleated on different sections of the VBF [1
] and that the 24 August Mw 6.0 and 26 October Mw 5.9 events can be considered to be the major foreshocks of the 30 October mainshock, so we refer to them as such throughout the manuscript.
2. Materials and Methods
2.1. Fault-Slip Data Collection and Strain Markers
We performed a high-resolution sampling of the coseismic and intra-sequence ground deformations along the VBF (Figure 2
) using a digital mapping method based on GPS-integrated Fieldmove software (Move™, produced by Midland Valley Exploration Ltd 2018 Glasgow, Scotland UK, and Petroleum Expert Edinburgh, Scotland UK) installed on an Apple IPad-Pro. At each survey site, we collected data on the following: (i) the type of ground deformation, including ruptures of the main fault plane within the fault rock, ruptures at the contact between the fault plane and the soil/debris covering the slope, and subsidiary ruptures displacing the local Quaternary cover; (ii) the rupture attitude and associated displacement (i.e., the net slip or throw values); and (iii) any other kinematic indicator providing a true displacement vector.
During the fieldwork, at multiple sites along the VBF, we drew permanent strain-markers on the fault plane to record the position of the topography cut-off line in the hanging wall—at a given known date (Figure 2
d, Figure S3a,e
). We revisited the sites and used the different positions of the strain-markers to obtain temporal constraints to the bands characterizing the composite free faces recognized along the fault scarp. In several outcrops, the free faces typically showed three adjacent bands produced by different periods of exposure of unearthed rocks (Figure 2
b,d, Figure S3
We note that lithology played a role in strict control of the preservation of the composite free faces. In fact, the three bands described above were observed only on well cemented and polished fault mirrors set in unstratified Jurassic limestones of the Calcare Massiccio Fm. On the contrary, they were not detectable where the coseismic scarp originated in not cohesive slope debris or in the poorly cemented fault breccia derived from the basinal limestones of the Umbria-Marche stratigraphic succession.
Anyhow, in all the survey sites shown in Figure 2
a, the three bands referable to (a) 24 August coseismic slip, (b) 24 August to 29 October slip, and (c) 30 October coseismic slip were clearly recognizable and mapped with continuity of tens of meters, along the fault strike.
The field evidence allowed the Intra-sequence Ground Deformation (IGD) to be estimated, i.e., the deformation occurred after the 24 August foreshock and before the 30 October mainshock, along the southern portion of the VBF. Unfortunately, no data for this time span are available for the other fault segments because, during the intra-sequence period, our fieldwork was focused along the 24 August coseismic rupture and we were not aware that ground deformations were affecting a much wider portion of the hanging wall.
2.2. DInSAR Processing
Using DInSAR analysis [15
], we inferred the surface deformation patterns along the VBF for different periods. Here, we focused on the IGD detected between the 24 August Mw 6.0 and the 26 October Mw 5.9 foreshocks. According to data availability, we divided this timespan into two ~1 month-long periods—i.e., T1 and T2 (Table 1
To study IGD, we generated individual interferograms using the flow chart available in Sar Scape [17
]. We also elaborated coseismic interferograms and compared them with published interferograms prepared by the ESA’s SEOM Programme InSARap project [10
] for the 24 August and the 30 October earthquakes and by the COMET [11
], for the 26 October earthquake. Our new coseismic interferograms for the Mw 6.0 foreshock (Figure S1
) and the cumulated effects produced by the Mw 5.9 foreshock and Mw 6.5 mainshock (Figure S2
) confirm the results published by the InSARap project for the 24 August foreshock and 30 October mainshock [10
] and by COMET for the 26 October foreshock [10
] (Figure 1
). An independent interferogram prepared using the DIAPASON processing chain (https://terradue.github.io/doc-tep-geohazards/tutorials/diapason-iw.html
), a DInSAR suite developed by the French Space Agency (CNES) and maintained by TRE-Altamira, confirmed our T1+T2 intra-sequence interferogram. Interferograms covering periods of shorter lengths between June and August 2016 did not show any significant surface deformation in the study area before the onset of the seismic sequence. Similarly, surface deformations were not detected in the area from February to April 2017, i.e., after the melting of snow cover that had mantled part of the area since mid-November 2016, preventing DInSAR measurements from mid-November 2016 to February 2017 in most of the study area.
2.3. Measurements of Length of the Deforming Ground
We visually inferred the northernmost and southernmost points of ground deformation within the VBF hanging wall, in the T1, T2 and T1+T2 interferograms (Figure 3
and Figure 4
, black dots with white bars).
Adopting a conservative approach, we identified points where deformation was detectable, provided that these were included in areas of high coherence (Figure S4
This latter condition was also confirmed by the displacement fields produced on high-coherence areas located south and north of the VBF. Figure 5
and Figure 6
shows parts of the LoS displacement field where the coherence is high in the northernmost (Figure 5
b and Figure 6
b) and southernmost (Figure 5
c and Figure 6
c) portions of the VBF (compare to Figure S4
In these reconstructions, there are some uncertainties in the quantitative estimate of the deformation for a few reasons, including some possible local contributions from the atmosphere, which cannot be exactly estimated with the available data.
Some differences between the patterns shown in Figure 5
and Figure 6
could also be due to the different LoS. In fact, Figure 5
was obtained from a satellite in ascending mode, with a shot direction from west to east, which is almost perpendicular to the VBF, while Figure 6
was obtained from a satellite in descending mode and a SE-NW shot direction, at an angle with respect to the VBF. The two acquisitions can intercept different components of the whole displacement.
Moreover, it must be borne in mind that, as demonstrated in [4
], the 24 August and 30 October coseismic displacement vectors (which are also coherent with the long-term kinematics) vary sensibly, along the VBF trace, as regards the slip direction and the amount of net slip.
This observation can contribute to explaining the differences in the values of motions observed along the fault during the intra-sequence phase (compare the graphs of Figure 5
and Figure 6
). In fact, the along-strike variable kinematics (dip-slip to oblique-dextral and locally oblique-sinistral) could significantly affect the length of the displacement components detected along the LoS, in the different orbits.
Anyhow, we carried out the interpretation of the interferograms only where the change of phase seems to be atypical in shape for atmospheric disturbances and topographic effects.
The crossing profiles represent deformation measurements that, although they could be influenced by local effects, including gravitational movements, show that qualitatively the behavior is the same along all the traces.
It is noteworthy that in both sample areas (Figure 5
b,c and Figure 6
b,c), the LoS displacement field pattern shows a step or a slope between the footwall and hanging wall across the mapped fault traces (Figure 5
and Figure 6
lower graphs) also at great distances from the points where we measured slips in the field (survey sites shown in Figure 2
We used the distance between the northernmost and southernmost point with detectable deformation to infer the Length of the Deforming Ground (LDG) during the T1+T2 period, i.e., the fault parallel length of the surface deformation that occurred after the 24 August Mw 6.0 foreshock and before the combined effects of the 26 October Mw 5.9 foreshock and the 30 October Mw 6.5 mainshock.
2.4. Procedure for Estimating the Fault Rupture Area
Comparing the along-fault extent of the IDG with the distribution of the coseismic ruptures observed after the 30 October mainshock [4
], we suggest the intra-sequence LDG is a good proxy for the expected coseismic Surface Rupture Length (i.e., LDG ≈ SRL), which is known to correlate with the maximum expected earthquake magnitude.
The seismological data in the literature based on relocated hypocenters [1
], and our own field data (Figure 2
, Figure S3
), agree in showing a dip-angle of the VBF in a range between α = 50° and α = 60°. We, therefore, used the LDG and the above dip-angle values (considered as end-members) to estimate the expected coseismic fault Rupture Area (RA, in km2
). We also hypothesized two possible depths for the coseismic fault rupture, D = 8 km, corresponding to the hypocentral depth of the 24 August 2016, Mw 6.0 earthquake—assuming the hypocenter coincides with the deepest part of the seismogenic source—and D = 11 km—the depth above which most of the seismicity occurred after the Mw 6.0 event [9
]—corresponding to the inferred base of the seismogenic layer. The two alternative depths can be considered on the basis of the recent literature concerning the major instrumental earthquakes that have affected the Central Apennines extensional belt, such as the 1997 Umbria-Marche and the 2009 L’Aquila earthquakes, and associated sequences [19
]. In both these cases, in fact, the fault-slip patterns, obtained by inverting GPS, DInSAR and strong motion data [23
], show that the hypocenters of the mainshocks lay on the lower part of the coseismic rupture, or near its deeper boundary, which nearly corresponds to the base of the seismogenic layer below this part of the Apennine chain.
2.5. Evaluation of the Expected Earthquake Magnitude
To estimate the expected magnitude of a possible earthquake originated on the VBF after the 24 August foreshock, we used two empirical relationships linking the (expected, in our case) fault rupture area (RA, in km2
) to the expected earthquake magnitude, including M = 0.47 + 0.98 log(RA) [26
] and M = 0.40 + log(RA) [27
]. Both relationships are suited for extensional tectonic domains characterized by “slow” faults that have a slip rate < 1 cm·yr−1
], which is the case of the Central Apennines extensional belt.
Over the last 20 years, two seismic sequences comparable to the 2016 sequence shook the Central Apennines active extensional belt: (i) the 1997 Umbria-Marche sequence [19
] and (ii) the 2009 L’Aquila sequence [21
]. Both sequences originated on sets of NNW-SSE striking normal faults and were characterized by mainshocks with focal depths in the range of 8–11 km. For the Umbria-Marche sequence, culminating in the 26 September 1997 Mw 6.1 Colfiorito earthquake, field data [34
] and DInSAR analyses [23
] are insufficient to accurately document the post-seismic surface deformation. For the L’Aquila sequence, climaxing at the 6 April 2009 Mw 6.3 earthquake [24
], two terrestrial laser scanner measurements performed along and near the fault surface rupture documented a post-seismic surface deformation of ~2.7 cm during the ~4-month period from 8 to 124 days after the Mw 6.3 earthquake [41
]. Advanced DInSAR analyses confirmed a cumulative surface deformation of up to −4 cm with a temporal evolution showing a clear exponential decay within about 102
To study and analyze the deformations that occurred during the 2016 Central Italy sequence, we preferred DInSAR [15
] (i.e., construction of single interferograms from pairs of SAR images) to advanced DInSAR [42
] (i.e., construction of time series of the deformation using a large set of images) because of the limited number (eight or nine) of ESA Sentinel-1 images available between the 24 August Mw 6.0 and 26 October Mw 5.9 foreshocks and for the short time period we investigated [45
Intra-sequence interferograms are more difficult to interpret than coseismic because the magnitude of the deformation is smaller [16
] and the noise eventually introduced by the atmosphere is difficult to quantify. In fact, our interpretation is based on the deformation patterns, while the quantitative values of deformations obtained after the phase unwrapping are to be considered as deformation scenarios in the case of no disturbances. We produced all the possible interferograms coupling all the available images acquired in T1+T2 along the two orbits and decided to interpret only those that apparently showed a less evident atmospheric contribution [16
]. We did not interpret those parts in which disturbances are present and clearly detectable, e.g., the central part of Figure 3
c. Furthermore, the asynchronous acquisition (either in sensing day or timing) of the images in ascending and descending mode (Table 1
) strengthens the assumption that the similar patterns observed on both orbits are not significantly affected by atmospheric disturbances but can represent phase changes related to surface deformations.
and Figure 6
show some differences, and this could be due to the different LoS. In Figure 5
, the satellite is in ascending mode, with a west-east direction shot almost perpendicular to the Mt Vettore fault, while in Figure 6
the satellite is in descending mode and the shot is east-west directed and at an angle with respect to the Mt Vettore fault. As a matter of fact, the two acquisitions are consistently able to intercept different components of the whole displacement.
Compared to the deformation rates measured along the fault that ruptured during the 2009 L’Aquila earthquake (Paganica—Mt Stabiata fault [36
]), the slip along the VBF during the 2016 intra-sequence period (T1 + T2)—estimated to be up to 6.2 cm during a 2-month period—appears too large to be considered only a post-seismic after-slip of the Mw 6.0 foreshock, even if, theoretically, a contribution to the deformation by this latter process cannot be excluded [46
]. It is noteworthy that the only currently available estimates of the 24 August after-slip [47
] suggest that it was close to zero in the VBF section along which we collected our displacement data in the field (compare Figure 2
a to Figure 9
b in [47
Additionally, our DInSAR analyses do not show a progressive reduction in the surface deformational rate in the T1+T2 period along the VBF (Figure 3
and Figure 4
), which is known to occur during the post-seismic phase of large earthquakes [48
] and was documented following the 2009 L’Aquila earthquake [41
]. Conversely, in the Mt Vettore-Castelluccio di Norcia area, where the surface deformation was greater, the temporal evolution of the IGD exhibited an area increase, and during the T2 period (before the Mw 5.9 and Mw 6.5 events), it continued to expand northward along the VBF, towards the Visso-Ussita area (Figure 3
b,d and Figure 7
Note that the VBF and Paganica-Mt. Stabiata faults show similar geometrical features (strike, average dip-angle, depth of detachment, and thickness of the seismogenic layer inferred by the hypocenter distributions [3
]) and displace at the surface, with comparable lithologies. In particular, in their northern trace, they both offset the Umbria-Marche Meso-Cenozoic carbonates, whereas in their southern sections, they displace softer sediments (Late Miocene turbidites and Pleistocene alluvial deposits).
We conclude that the different spatial-temporal patterns of the evolution of the post-seismic deformations show the different behaviors of the two sequences, with the 2009 L’Aquila post-seismic deformation rapidly decreasing and the 2016 Mw 6.0 foreshock post-seismic deformation increasing and widening northward up to the mainshock.
The observation that since the T1 period, and even more during the T2 period, the IGD was largely independent of the coseismic deformation caused by the 24 August foreshock suggests that it cannot be merely attributed to after-slip following the earthquake and that, conversely, it is a consequence of the increasing and spreading strain along the VBF and in the adjacent rock volume. The aforementioned observation fits the estimates of the coseismic stress change induced along the VBF by the 24 August foreshock. In fact, according to [14
], during the intra-sequence period, a region of highly positive Coulomb stress existed along the full length at the base of the VBF and this stress pattern appears to have contributed to the preparation of the Mw 6.5 earthquake. Ultimately, we interpret the displacements of the IGD in the VBF hanging wall block as an indicator of the forthcoming re-activation of the VBF that happened during the Mw 5.9 and Mw 6.5 earthquakes.
Based on the observations and interpretations presented in this work, we suggest that during a seismic sequence with a large, early “foreshock” that produces significant surface deformation (earthquakes with Mw ≥ 5.5 and focal depth <15 km, satisfying the condition in the Central Apennines of Italy), measurement of the deformation around the active fault that generated the earthquake can be used to evaluate the potential occurrence along the fault of new earthquakes of a comparable or larger magnitude than that of the foreshock. Our analysis of the 2016 seismic sequence in the Central Apennines sets the conditions for the application of this interpretive framework.
First, an earthquake E1 of magnitude M1 has to occur along an active fault, producing a surface deformation that is sufficiently large to be detected, e.g., through DInSAR analyses. Second, the maximum expected magnitude Mmax of an earthquake along the active fault has to be greater than M1, indicating that E1 released only part of the fault seismogenic potential. Third, during a short period after E1 (e.g., a few days to several weeks for the Central Apennines extensional belt) the post-seismic surface deformation along and around the active fault has to expand to affect larger volumes of rock (and a larger surface area) around the fault.
Where these conditions are met, we maintain that the possibility of a new earthquake with Mw up to ~Mmax along the active fault is high. The magnitude of the expected earthquake can be estimated from appropriate scaling dependencies linking the earthquake magnitude to the fault rupture area [26
]. The latter can be determined considering the length of the deforming ground (LDG), a proxy for the expected coseismic surface rupture length, and by obtaining the geometry (i.e., slope, depth) of the active fault from geological and seismological data. In the case of the Central Italy seismic sequence, the 30 October mainshock was Mw 6.5, well in the range (± 0.2) of 6.4 ≤ Mw ≤ 6.7 estimated by applying the aforementioned reasoning (Section 3.4
). We conclude that using LDG measurements obtained from intra-sequence interferometric data, confirmed by field evidence, the magnitude of the 30 October earthquake could have been anticipated with a narrow error range.
We conclude that our proposed interpretative framework to aid in making inferences regarding the evolution of active seismic sequences suggests that the ongoing seismicity and the associated progressing surface deformation could lead to stronger earthquakes. If confirmed by other earthquake sequences in the Central Apennines extensional belt, or in other similar extensional seismotectonic environments, the proposed framework could represent a significant step towards the prediction of some destructive earthquakes.