Professional Documents
Culture Documents
doi: 10.1093/gji/ggv397
GJI Seismology
SUMMARY
A non-linear, global-search, probabilistic, double-difference earthquake location technique is
illustrated. The main advantages of this method are the determination of comprehensive and
complete solutions through the probability density function (PDF), the use of differential arrival
times as data and the possibility to use a 3-D velocity model both for absolute and doubledifference locations, all of which help to obtain accurate differential locations in structurally
complex geological media. The joint use of this methodology and an accurate differential
time data set allowed us to carry out a high-resolution, earthquake location analysis, which
helps to characterize the active fault geometries in the studied region. We investigated the
recent microseismicity occurring at the Campanian-Lucanian Apennines in the crustal volume
embedding the fault system that generated the 1980 MS 6.9 earthquake in Irpinia. In order
to obtain highly accurate seismicity locations, we applied the method to the P and S arrival
time data set from 1312 events (ML < 3.1) that occurred from August 2005 to April 2011
and used the 3-D P- and S-wave velocity models optimized for the area under study. Both
manually refined and cross-correlation refined absolute arrival times have been used. The
refined seismicity locations show that the events occur in a volume delimited by the faults
activated during the 1980 MS 6.9 Irpinia earthquake on subparallel, predominantly normal
faults. We find an abrupt interruption of the seismicity across an SWNE oriented structural
discontinuity corresponding to a contact zone between different rheology rock formations
(carbonate platform and basin residuals). This barrier appears to be located in the area
bounded by the fault segments activated during the first (0 s) and the second (18 s) rupture
episodes of the 1980s Irpinia earthquake. We hypothesize that this geometrical barrier could
have played a key role during the 1980 Irpinia event, and possibly controlled the delayed times
of activation of the two rupture segments.
Key words: Seismicity and tectonics; Computational seismology; Continental tectonics:
extensional; Fractures and faults.
1 I N T RO D U C T I O N
The knowledge of the spatial distribution of microseismicity allows
the identification and geometrical characterization of active fault
structures (e.g. Got et al. 1994; Waldhauser & Schaff 2008), the
study of the regional stress field (e.g. De Matteis et al. 2012),
the characterization of the small-scale variability of faulting style,
stress and strength (e.g. Hardebeck 2006; Stabile et al. 2012), the
C
The Authors 2015. Published by Oxford University Press on behalf of The Royal Astronomical Society.
1821
Accepted 2015 September 15. Received 2015 September 14; in original form 2014 December 17
1822
G. De Landro et al.
2 METHODOLOGY
For absolute, non-linear, global-search location in 3-D models we use the methodology proposed by Lomax et al.
(2000), implemented in the NonLinLoc (non-linear location;
http://www.alomax.net/nlloc) software, where the complete probabilistic solution of the earthquake location problem is represented
by a posterior probability density function (PDF).
In order to perform dd earthquake locations, we propose a
methodology that optimizes the relative spatial x, y, z, and t (origin
time) coordinates for a set of hypocentres given a set of differential
phase arrival times measured at each station for multiple hypocentres. The global optimization process is based on the annealing Metropolis algorithm search (Mosegaard & Tarantola 1995;
Lomax et al. 2000; Lomax et al. 2009). It has a temperature parameter which allows searching using a large step-size in x, y, z, and t
for event perturbation initially, with a cooling, but not freezing
to an adaptive, nearly constant step size for later sampling. Thus
the later sampling becomes Metropolis sampling to define the PDF
for each hypocentre in x, y, z, and t. This Metropolis sampling uses
an adaptive step size for each event, which is increased slowly if
the acceptance rate for new, perturbed hypocentres for the event is
high (e.g. > 14 ) and decreased more rapidly if the acceptance rate is
low (e.g. < 14 ). This global optimization is much faster than a pure
Monte Carlo approach and likely faster, simpler and more thorough
than a cascading grid-search.
An important reason for using a non-linear, directed, Metropolistype stochastic search instead of a linearized approach, in addition to
the simple use of 3-D models (3-D traveltime fields) with arbitrary
complexity and parameterization, is the chance of better exploring
complicated cost functions with multiple minima, curved, irregular,
and non-parabolic valleys (see Supporting Information Fig. S11).
Though the extreme high dimensionality of the multievent inverse
problem (curse of dimensionality) does make this extremely difficult for all methods (including and sometimes especially linearized
methods). The disadvantage of this global search method relative
to a linearized one is that the computation time is certainly much
longer and the maximum number of differential times and events
that can be processed is much less.
In the model parameters space exploration, the algorithm seeks
for the parameter solution that maximizes the likelihood function
(based upon the misfit between measured and calculated differential
phase arrival times) while perturbing the hypocentre coordinates.
In particular, the misfit and the solution likelihood are determined
by evaluating the dd equation proposed by Waldhauser & Ellsworth
(2000).
The equation describing the likelihood for a set of arrival-time
difference measures that concerns an event i with coordinates x, y,
z, and origin time t is
L i (x, y, z, t)
obs
cal
2
j
j
i
i
tk tk
tk tk
1
= exp
wk
2 k
2T 2
(1)
1823
perturbed
Li
, is evaluated to check if the perturbation improves the fit
to the data, in the sense of the Metropolis rule:
perturbed
Figure 1. Geological sketch map of the Campania-Lucania region integrated with seismogenic sources (DISS Working group 2010), location of seismic ISNet
stations (blue triangles) and INGV stations (turquoise triangles), main historical earthquakes with their focal mechanisms and aftershock zones (modified after
De Matteis et al. 2012).
1824
G. De Landro et al.
3 3 - D M I C R O E A RT H Q UA K E L O C AT I O N
I N T H E I R P I N I A A C T I V E FAU LT Z O N E
3.1 Geological setting
The Campania-Lucania region (Fig. 1) is located in the axial portion of Southern Apennines, an Adriatic-verging duplex system,
orogenically transported over the flexured southwestern margin of
the Apulia foreland (Patacca et al. 1990). This duplex system consists of a complex architecture of carbonate horsts deriving from
the Apulia Carbonate Platform (ACP), which is overthrusted by
rootless nappes. The belt is located between the Tyrrhenian backarc
basin to the west and with the Bradano foredeep to the east.
The ACP consists of 78 km thick Mesocenozoic carbonate sequence, which overlies Permotriassic clastic deposits (e.g. Verrucano Fm.; Roure et al. 1991). Plio-pleistocene terrigenous deposits
stratigraphically cover the flexed ACP in the eastern margin of the
Bradano Trough (Casnadei 1988). Moving westwards to the external zone of the belt, the ACP progressively dips below the rootless
nappes and is in turn involved in the folds and thrusts of the thrust
belt.
The orogenic stack overlying the ACP is formed by thrust sheets
coming from the deformation of the stratigraphic successions associated with the following main paleogeographic domains (Patacca
et al. 1992):
4 DISCUSSION
An innovative dd location method has been applied to the seismicity (1312 microearthquakes) recorded by 42 ISNet and INGV stations from August 2005 to April 2011 along the Campania-Lucania
Apennine chain (Southern Italy). The proposed methodology solves
the dd equations (Waldhauser & Ellsworth 2000) using the nonlinear, global-search, probabilistic location approach developed by
Lomax et al. (2000) instead of linearized approaches. Moreover,
the methodology gives the advantage of relocating the seismicity
1825
1826
(a)
G. De Landro et al.
1500'
1530'
1600'
(c)
1500'
1530'
4100'
4100'
1600'
(f)
300
200
100
0
0
1
2
MRP ErrHo (km)
1
2
MRP ErrZ (km)
200
100
4030'
A
0
20
(b)
0
km
10
0.0
20
(d)
(g)
10
Depth (km)
10
20
Depth (km)
Depth (km)
10
20
20
15.5
Longitude
16.0
15.0
15.5
Longitude
(e) A
16.0
10
20
(h) B
30 40 50 60
Number of events
70
80
20
C
15
60
40
20
0
20
25
10
10
10
10
20
30
40
Distance (km)
50
60
70
Num. of events
Depth (km)
10
20
20
10
0
10
Distance (km)
20
Figure 2. (a) Events of the ISNet bulletin located with hypo2000 and the 1-D velocity model. (b) Eastwest vertical section of the ISNet bulletin events. (c)
Double-difference earthquake location of seismicity from 2005 August to 2011 April, from MRP data (in grey) and from CRP data (in turquoise). Red lines are
the surface projection of the three fault segments that ruptured during the 1980 Irpinia earthquake (Pantosti & Valensise 1990). (d) Eastwest vertical section
of the seismicity. (e) Anti-Apenninic AA vertical section of seismicity and histogram of events number moving in the NWSE direction along the same profile.
(f) Comparison between horizontal (left) and vertical (right) errors of absolute location (light grey) and dd locations for the MRP data set (top, dark grey) and
CRP data set (bottom, turquoise). (g) Histogram of the events as function of depth. (h) Earthquake cross-section along the profiles BB and CC indicated in
the map in panel (c).
by using the same 3-D P- and S-wave velocity model (and by considering station elevations) adopted for the absolute location of events,
which is a crucial condition for more reliable and comprehensive
dd locations in complex media such as the area investigated in this
study. Indeed, even if the dd technique allows for the minimization of errors due to un-modelled velocity structures, it has been
shown (e.g. Michelini & Lomax 2004) that the use of an inappropriate 1-D velocity model will, in most cases, lead to bias and error
in the dd locations. Therefore, the combination of the proposed
methodology with cross-correlation differential times allowed us
to perform a robust, high-resolution study for exploring new insights concerning the distribution and the geometry of the analysed
microseismicity.
Actually, the investigated crustal volume is cut by an NW
SE striking normal fault system along the axis of the Apenninic
chain, and to an approximately EW striking, strike-slip fault system, oblique with respect to the main trend of the belt. It has to be
noted that these fault systems correspond with those responsible for
the 1980 MS 6.9, and the 19901991 ML 5.2 and ML 4.7 earthquakes
(De Matteis et al. 2012).
The recent background microseismicity along the chain axis identifies the hanging wall volume delimited by the main fault and its
antithetic one responsible for the 1980 MS 6.9 Irpinia earthquake.
The present-day low-magnitude seismicity follows the same pattern
of the 1980 event aftershocks, with a main elongation parallel to the
15.0
Depth (km)
km
10
Depth (km)
4030'
1827
Figure 3. (a) Comparison of the dd location of seismicity with the topography of the Apulian Carbonate Platform top (modified after Improta et al. 2003). (b)
Comparison of tomographic cross-sections reported in map in panel (a) and the earthquake occurred in this area.
1828
G. De Landro et al.
Zollo 1989). This discontinuity has, thus, formed a barrier that has
resisted and slowed down the evolution of the rupture SE, forcing
it to change its geometry and restarted with a delay of about 20
s along a low-angle fault deeper, probably where the fracture has
found again the carbonate platform (Bernard & Zollo 1989).
5 C O N C LU S I O N S
D ATA A N D R E S O U R C E S
(1) The method, implemented in the NLDiffLoc code, has the advantage to obtain from one hand comprehensive solutions through
the PDF using as data the differential traveltimes and, on the other
hand, is implemented taking into account of complex media requiring a 3-D velocity models.
(2) This method was applied to relocate the microseismicity in
the Campania-Lucania region inverting MRP and CRP P- and Swave arrival-times, and using the high-resolution, 3-D Vp and Vs
velocity model of the area recently determined by Amoroso et al.
(2014a).
(3) The relocated seismicity shows a diffuse distribution in the
NWSE direction along the Appeninic chain for the Irpinia area
and the seismicity along the chain does not occur on a single fault
but in a volume delimited by the faults activated during the 1980
Irpinia earthquake. This is consistent with previous observations
and seismicity analyses (De Matteis et al. 2012; Matrullo et al.
2013; Amoroso et al. 2014a).
(4) The distribution of relocated seismicity and the tomographic
image show the presence of a rheological discontinuity due to the
contact between the carbonate units (APC) and the basinal facies
(LB) that have different rheological behaviour in to the chain stress
regime. This discontinuity is underlined by a significant reduction
in the rate of events in the zone corresponding to a low of the
Apulian Platform filled by the deposits of the LB.
AC K N OW L E D G E M E N T S
This work has been partially funded by Project SHEERSHale
gas Exploration and Exploitation induced Risks (GA N.640896
H2020-LCE-2014-2015/H2020-LCE-2014-1), and Project EPOS
IP (GA N.676564H2020-INFRADEV-2014-2015). The authors
wish to thank the editor Yehuda Ben-Zion and the reviewers Amir
Allam and Jean-Luc Got for their valuable comments which have
been contributed to improve the quality of the manuscript.
REFERENCES
Allam, A.A. & Ben-Zion, Y., 2012. Seismic velocity structures in the Southern California plate-boundary environment from double-difference tomography, Geophys. J. Int., 190, 11811196.
Amato, A. & Selvaggi, G., 1993. Aftershock location and P-wave velocity
structure in the epicentral region of the 1980 Irpinia earthquake, Ann.
Geofis., 36(1), 315.
Figure 4. Schematic geological interpretation. The positions of the interfaces are deduced from both tomographic model and hypocentre-relocated events. The
name of the lithologies and the position of main topographic references are also indicated and supposed from Improta et al. (2003). The crosses indicate the
rough position of the retrieved hypocentres. A dashed line is used to represent the hypothetical barrier/ fault scarp.
1829
1830
G. De Landro et al.
S U P P O RT I N G I N F O R M AT I O N
Additional Supporting Information may be found in the online version of this paper:
Figure S1. Test with the 3-D velocity model: maximum distance
5 km versus 10 km. The synthetic arrival times were inverted to
calculate the initial absolute locations with NLLoc. For the dd location, we built the differentials arrival times considering all possible
combinations of events with maximum distances of 5 and 10 km.
The figure shows comparison between the dd location results of the
synthetic events using the 3-D velocity model and considering 5 km
(blue) and 10 km (green) of maximum distance to calculate the differential times. Red stars indicate the reference positions of events.
Grey triangles are the ISNet seismic stations. (a) Seismicity distribution in plane. (b) Eastwest vertical section of the seismicity. (c)
Distance in plane between the calculated and the true position (left)
and distance between the calculated and the true depth of the events
(right). (d) Horizontal (left) and vertical (right) locations errors. (e)
Root mean square (rms) distribution.
Figure S2. Spatial location error distribution for synthetic events
considering the maximum distance of 5 km. (a) Confidence ellipsoid
at 68 per cent projections in the XY plane. (b) Confidence ellipsoid
at 68 per cent projections in the XZ plane. (c) Confidence ellipsoid at
68 per cent projections in the YZ plane. (d) Zoom of the confidence
ellipsoids in the zone with higher events density. (e) Zoom of the
confidence ellipsoids in the zone with lower events density.
Figure S3. Plot of events interconnection. (a) Considering 5 km
maximum distance, every event has a maximum of eight connections and at least one single connection. With this condition the obtained locations have mean distance from the true position of about
200 m in plane and 200 m in depth (Fig. S1c, blue). (b) Considering 10 km maximum distance, every event has at least seven
connections and a maximum of 16 connections. With this condition
the obtained locations are significantly improved, in fact, the mean
distance is about 30 m in plane and 25 m in depth (Fig. S1c, green).
Thus the minimum number of connections to constrain the position
of a single event is about 10, equal to the number of connections of
the central events of the finer grid. (c) In the case of real locations, for
the MRP data set with the use of the sections every event has on
average 40 connections, whereas for the CRP data set with the use
of 5 km as the maximum distance every event has on average 15
connections (see Fig. S12b).
Figure S4. Test with the 1-D velocity model: absolute versus dd
locations. The 3-D model synthetic arrival times were inverted with
the 1-D velocity model to calculate the initial absolute locations
with NLLoc. For the dd location, we built the differential arrival
times considering all possible combinations of events with maximum distance of 5 and 10 km, and then locate considering the
1-D velocity model (Matrullo et al. 2013). (a) Comparison between
the absolute initial location (blue) and the dd final location of the
synthetic events (green) with the 1-D velocity model. Red star indicate the reference positions of events. Grey triangles are the ISNet
seismic stations. (a) Seismicity distribution in plane. (b) Eastwest
vertical section of the seismicity. (c) Distance in plane between the
calculated and the true position. (d) Distance between the calculated
and the true depth of the events. (e) rms distribution.
Figure S5. (a) Comparison between absolute initial location (blue)
and dd final location (green) of the synthetic events with the 3-D
velocity model. Red stars indicate the events reference positions.
Grey triangles are the ISNet seismic stations and red stars are the
reference positions of events. (a) Seismicity distribution in plane.
(b) Eastwest vertical section of the seismicity. (c) In plane (XY)
and vertical (Z) distance between the locations and the true position
of events. (d) Horizontal (XY) and vertical (Z) location errors. (e)
rms distributions.
Figure S6. Spatial location error distribution of synthetic
events considering data with noise. (a) Confidence ellipsoid at
68 per cent projections in the XY plane. (b) Confidence ellipsoid
Poupinet, G., Ellsworth, W.L. & Frechet, J., 1984. Monitoring velocity variations in the crust using earthquake doublets: an application to the Calaveras Fault, California, J. geophys. Res., 89(B7), 57195731.
Rigo, A., Lyon-Caen, H., Armijo, R., Deschamps, A., Hatzfeld, D., Makropoulos, K., Papadimitriou, P. & Kassaras, I., 1996. A microseismic
study in the western part of the Gulf of Corinth (Greece): implications for large-scale normal faulting mechanisms, Geophys. J. Int., 126,
663688.
Roure, F., Casero, P. & Vially, R., 1991. Growth processes and melange
formation in the Southern Apennines accretionary wedge, Earth planet.
Sci. Lett., 102, 395412.
Rowe, C., Aster, R., Brochers, B. & Young, C., 2002. An automatic, adaptive
algorithm for refining phase picks in large seismic data sets, Bull. seism.
Soc. Am., 92, 16601674.
Rubin, A.M., Gillard, D. & Got, J.L., 1999. Streaks of microearthquakes
along creeping faults, Nature, 400, 635641.
Scrocca, D., Carminati, E. & Doglioni, C., 2005. Deep structure of the
southern Apennines, Italy: thin-skinned or thick-skinned?, Tectonics, 24,
TC3005, doi:10.1029/2004TC001634.
Sen, M.K. & Stoffa, P.L., 1995. Global Optimization Methods in Geophysical
Inversion, Elsevier, 281 pp.
Stabile, T.A., Satriano, C., Orefice, A., Festa, G. & Zollo, A., 2012. Anatomy
of a microearthquake sequence on an active normal fault, Sci. Rep., 2,
doi:10.1038/srep00410.
Stabile, T.A., Iannaccone, G., Zollo, A., Lomax, A., Ferulano, M.F., Vetri,
M.L.V. & Barzaghi, L.P., 2013. A comprehensive approach for evaluating network performance in surface and borehole seismic monitoring,
Geophys. J. Int., 192(2), 793806.
Suarez, G., Monfret, T., Wittlinger, G. & David, C., 1990. Geometry of subduction and depth of the seismogenic zone in the Guerrero gap, Mexico,
Nature, 345, 336338.
Tarantola, A., 2005. Inverse Problem Theory and Methods for Model Parameter Estimation, SIAM.
Tarantola, A. & Valette, B., 1982. Inverse problems = quest for information,
J. geophys. Res., 50, 159170.
Thurber, C., Roecker, S., Zhang, H., Baher, S. & Ellsworth, W.,
2004. Fine-scale structure of the San Andreas fault zone and location of the SAFOD target earthquakes, Geophys. Res. Lett., 31(12),
doi:10.1029/2003GL019398.
Valoroso, L., Chiaraluce, L., Piccinini, D., Di Stefano, R., Schaff, D. &
Waldhauser, F., 2013. Radiography of a normal fault system by 64 000
high-precision earthquake locations: the 2009 LAquila (central Italy) case
study, J. geophys. Res., 118, 11561176.
Waldhauser, F. & Ellsworth, W.L., 2000. A double-difference earthquake
location algorithm: method and application to the northern Hayward Fault,
California, Bull. seism. Soc. Am., 90, 13531368.
Waldhauser, F. & Schaff, D.P., 2008. Large-scale relocation of two
decades of Northern California seismicity using cross-correlation
and double-difference methods, J. geophys. Res., 113, B08311, doi:
10.1029/2007jb005479.
Wessel, P. & Smith, W.H.F., 1998. New, improved version of the Generic
Mapping Tools Released, EOS, Trans. Am. geophys. Un., 79, 579.
Zhang, H. & Thurber, C.H., 2003. Double-difference tomography: the
method and its application to the Hayward fault, California, Bull. seism.
Soc. Am., 93, 18751889.
Zollo, A., Orefice, A. & Convertito, V., 2014. Source parameter scaling and radiation efficiency of microearthquakes along the Irpinia
fault zone in southern Apennines, Italy, J. geophys. Res., 119,
doi:10.1002/2013JB010116.
at 68 per cent projections in the XZ plane. (c) Confidence ellipsoid at 68 per cent projections in the YZ plane. (d) Zoom of the
confidence ellipsoids in the zone with higher events density. (e)
Zoom of the confidence ellipsoids in the zone with lower events
density.
Figure S7. (a) Comparison between the dd traveltime residuals
of the initial absolute location (green) and the final dd location
(blue) for the MRP data set. (b) Comparison between the horizontal
(right) and vertical (left) location errors. (c) Comparison between
rms. Residual histograms of final dd location are improved in terms
of narrower traveltimes residual distribution, in terms of rms that
decreases on average of 0.05 s, and in terms of location errors that
decrease on average of about 0.5 km.
Figure S8. (a) Comparison between the dd traveltime residuals of
the initial absolute location (green) and the final dd location (blue)
for the CRP data set. (b) Comparison between the horizontal (right)
and vertical (left) location errors. (c) Comparison between rms. The
residual histograms of final dd location are improved both in terms
of narrower traveltime residual distribution, in terms of rms that
decreases on average of 0.1 s, and in terms of location errors, that
decrease on average of about 0.5 km.
Figure S9. Histograms of normalized likelihood function for four
different events of the MRP data set located with the 3-D velocity
model.
Figure S10. Plot of the normalized near-ellipsoidal likelihood
function of an event of the MRP data set located in the 3D velocity model. (a) Normalized likelihood function in 3-D.
(b) Likelihood function section in the latitudelongitude plane
around the maximum likelihood depth. (c) Likelihood function section in the latitudedepth plane around the maximum likelihood
longitude.
1831