GEOLOGICA CARPATHICA, FEBRUARY 2007, 58, 1, 89—95
www.geologicacarpathica.sk
Introduction
Southern Calabria is one of the most active seismic zones
in Europe. The Calabrian arc is a part of the Mediterranean
orogenic belt (Fig. 1) connecting the Southern Apennines
with the Maghrebian thrust belts (Haccard et al. 1972; Al-
varez 1976). In this tectonic arc the effects of Quaternary
activity are well known (e.g. Tortorici et al. 1995).
In southwestern Calabria, the Aspromonte and Serre
structures (Fig. 1) consist of a pile of thrust sheets, mainly
composed by gneisses and granites, which belong to the
uplifted metamorphic basement. To the east, the structures
gently slope towards the Ionian Sea, whereas their western
edge is steeper and is controlled by faulting (Amodio-Mo-
relli et al. 1976).
The Gioia Tauro Basin (Fig. 1) consists of about 600
meters of marine sediments (from Upper Pliocene to Lower
Pleistocene) mainly composed of several hundred meters
of clay, and about 100 m of regressive sands and conglom-
erates (Middle Pleistocene). This sequence is overlain by
alluvial fanglomerates and sands (from Middle to Upper
Pleistocene).
Detailed seismic studies in the area are strictly linked to
the accuracy of earthquake locations. From 1985 to 1994,
a local seismic network was operated by ENEL (the Italian
national electric company) with the main goal of monitor-
ing the seismicity in the Gioia Tauro Basin. More than
3500 earthquakes were located during this period.
In this paper, to better constrain the hypocenter location
for the Gioia Tauro Basin, we have defined a Minimum 1-D
model, by considering the a-priori P-wave velocity model
Improvement in earthquake location accuracy from
Minimum 1-D velocity model: an example from the Gioia
Tauro Basin (southwestern Calabria, Italy)
ROSARIO RAFFAELE and SEBASTIANO IMPOSA
Dipartimento di Scienze Geologiche, Universit
à di Catania, Corso Italia 57, I-95129 Catania, Italy;
raffaele@unict.it; imposa@unict.it).
(Manuscript received February 23, 2006; accepted in revised form June 22, 2006)
Abstract: We computed a 1-D velocity model in the Gioia Tauro Basin area (southwestern Calabria, Italy) by inverting
P-wave arrival times recorded by local seismic network consisting of eleven stations. For this purpose we used a data set
of 207 local earthquakes which were located with a minimum of 7 arrivals, a travel time residual (rms)
0 .3 s and an
azimuthal gap 180
º. This “Minimum 1-D velocity model” is complemented by station corrections which are influenced
by near-surface velocity heterogeneity. Using the new P-wave velocity model and the program HYPOELLIPSE, we
relocated the 207 selected events. Tests were carried out to verify the robustness of inversion results in order to
corroborate the conclusions drawn from our findings. The comparison between previous and present earthquake
locations has shown a significant improvement. The obtained “Minimum 1-D velocity model” may strongly help in the
future routine earthquake location in the area.
Key words: southern Calabria, Gioia Tauro Basin, earthquake location, minimum 1-D model.
of Bottari et al. (1982) and following the most relevant ap-
proach to this problem (Kissling et al. 1994). The obtained
model minimizes the average of rms (travel time residual)
for a set of well-located events, by computing a solution
for the coupled hypocenter and 1-D velocity model prob-
lem. Ignoring the coupling between hypocentral and ve-
locity parameters, during the location process, can
introduce errors in the hypocenter location, which de-
pends also on the assumed velocity structure (Kissling et
al. 1995).
This concept, which represents a first step towards more
detailed seismic studies, can be considered universal as we
can give examples not only from Gioia Tauro Basin and
from southern Italy (Chiarabba & Frepoli 1997; Musumeci
et al. 2003), but also from other areas of the world. In fact
one of the first extensions of the Minimum 1-D was ap-
plied to northwestern Italy (Kissling et al. 1995), but this
method was also used in areas such as northern Chile
(Husen et al. 1999), Costa Rica (Quintero & Kissling
2001), Romania (Popa et al. 2001), New Zealand (Sher-
burn & White 2005), northern Egypt (El-Hadid et al.
2005).
The coupled hypocenter-velocity model problem
The travel-time of a seismic wave is a non-linear func-
tion of both hypocenter parameters and seismic velocities
sampled along the ray paths between stations and hypo-
center. This dependence on hypocenter parameters and
seismic velocities is called the coupled hypocenter-veloc-
90
RAFFAELE and IMPOSA
ity model problem (Crosson 1976; Kissling 1988; Thurber
1992). It can be linearized and written in matrix notation
(Kissling et al. 1994) as:
t = Hh + Mm+ e = A d + e
where t = vector of travel time residuals; H = matrix of
partial derivatives of travel time with respect to hypocen-
tral parameters; h = vector of hypocentral parameter adjust-
ments; M = matrix of partial derivatives of travel times
with respect to model parameters; m = vector of velocity
parameter adjustments; e = vector of travel time errors, in-
cluding contributions from errors in measuring the ob-
served travel times (t
obs
), errors in calculated travel times
(t
calc
)
due to errors in station locations, use of the wrong
velocity model and hypocentral coordinates, and errors
caused by the linear approximation; A = matrix of all par-
tial derivatives; d = vector of hypocentral and model pa-
rameter adjustments.
In a standard earthquake location, the velocity parame-
ters are fixed to a-priori values, and the observed travel
times are minimized by perturbating the four hypocentral
parameters.
Neglecting the coupling between hypocenter and ve-
locity parameters during the location process, however,
can introduce systematic errors. Precise hypocenter loca-
tions and error estimate, therefore, request the simulta-
neous solution of both velocity and hypocentral
parameters. Then the correct hypocentral coordinates can
be reliably achieved by solving the coupled hypocenter-
Fig. 1. Seismotectonic sketch of the Calabria arc. The location of the Gioia Tauro Basin is marked by the black ellipse.
91
1-D VELOCITY MODEL: IMPROVEMENT IN EARTHQUAKE LOCATION ACCURACY (ITALY)
velocity model problem, rather than alternating indepen-
dent hypocenter and velocity adjustment steps (Thurber
1992; Kissling et al. 1994).
The obtained Minimum 1-D model represents a velocity
model that reflects the a-priori information and that leads
to a minimum average of rms values for the best selected
earthquakes used in the inversion.
Each velocity layer of the Minimum 1-D model is the
weighted area averaged over all rays in the data-set within
that depth interval. To account for lateral variations in the
shallow structure, station corrections are included in the
1-D inversion process.
The general applicability of the concept of the Mini-
mum 1-D model and its performance for high-precision
earthquake location have been tested and documented by
relocating mine blasts and shots. In particular, no signifi-
cant and systematic shift in hypocenter locations is ob-
served after a 3-D inversion when using the Minimum 1-D
model as the initial reference model (Kissling 1988). Thus,
also in the Gioia Tauro Basin area, the Minimum 1-D
model concept is the most appropriate for uniform high-
precision earthquake location, outperforming any velocity
model based on a-priori information.
Earthquake data and inversion procedure
In the Gioia Tauro region, a local seismic network cov-
ering an area of about 1200 km
2
(Fig. 2), was operated
from April 1985 to April 1994. Three seismic stations
(DLV, PDL and CST) had 3-component seismometers,
while eight had only a vertical one. Each station was
equipped with short period Mark L4C seismometers with a
natural frequency of 1 Hz. The data were sampled with a
sampling frequency of 153.8 Hz, and an anti-alias filter
with a corner frequency of 30 Hz was applied. Data were
transmitted continuously to the acquisition center, located
near the CGT station, by UHF radio and stored in trigger
mode on magnetic tape.
Since all instruments were operated in continuous
mode, to detect possible events a simple software trigger
using LTA/STA (Long Term Average/Short Term Aver-
age) ratio 8 was used. The records were grouped, convert-
ed in .EVE format and stored on CD-ROM. Identifying
first arrival times of P- and S-waves on waveforms was
done with the W
INHYPO
package (Raffaele 2004). It allows
a comfortable data processing including earthquake loca-
tion using the
HYPOELLIPSE
program (Lahr 1989).
During the operating period of the network 3539 events
were located (Fig. 2) with a 1-D a-priori velocity model
(Bottari et al. 1982). The weights of the time readings were
assigned with respect to the estimated errors (Table 1).
Most of the P-wave arrivals were assigned a weight of 0 as
they could be read with a picking accuracy of a few tens of
milliseconds. For S-waves the usual weight was 2, corre-
sponding to a reading error of about 0.2 s. The number of
observations per class of weight is shown in Table 2.
The approach used here for the simultaneous inversion
of a 1-D velocity structure into two steps. During the first
step, the code
HYPOELLIPSE
(Lahr 1989) routinely relo-
Fig. 2. Map of the study area. Epicenters are represented by circles for located 3539 events. Seismic stations are represented by trian-
gles. The surface evidence of faults is reported with continuous lines.
92
RAFFAELE and IMPOSA
Table 1: Weight of the arrival time readings with respect to the
reading errors.
Table 2: Number of observations per class for the 3539 events.
Fig. 3. Block diagram of the inversion procedure to compute the
Minimum 1-D model.
cates the events using the a-priori 1-D model. Then, the re-
localized events are selected following the criteria ex-
posed in the section 4. In the second step, a 1-D inversion
is performed to compute simultaneously hypocenter pa-
rameters, 1-D velocity model and station delays using the
best events. The aim of this step is to constrain the 1-D
Minimum model, this new velocity model should con-
strain hypocenter parameters that give the global mini-
mum of the cost function (Kissling 1988).
An outline of the inversion procedure for the computa-
tion of the 1-D Minimum velocity model is presented in
Fig. 3.
Minimum 1-D model and station delays
Because uncertainties in hypocenter locations introduce
instabilities in the inversion process (Kissling et al. 1994),
the computation of the Minimum 1-D model for the area
under study was carried out after a careful filtering of data
with respect to their quality. We considered solely events
for which at least 7 arrival times were available, with a
“gap” 180º and a rms
0.3 s. The resulting data-set con-
sists of 207 well-locatable events with 1555 P- and 402
S-onset readings.
The selected events are then inverted by using the pro-
gram Velest (Kissling 1995) to calculate the adjustments
of P-wave velocities and station corrections. As the num-
ber of readable S-wave arrival onsets was restricted, we for-
sake inverting S-wave velocities. In particular, the steps
that lead to the final model consist of twelve inversions, of
one iteration, where the output files of the previous inver-
sion, as regards to the P-wave model, the hypocenter loca-
tions and station corrections, become input files to the
following one. For the S-wave model the input file was de-
rived from the P-wave model assuming V
p
/ V
s
= 1.73.
Therefore S-wave readings were used only to better con-
strain the focal depths.
Following the method proposed by Kissling et al.
(1994), as the layer depths are kept fixed, we have begun
with a large number of 2 km thick layers and combined
layers for which velocities converged to similar values
during the inversion process. The inversion process was
Fig. 4. Starting 1-D velocity model (dotted line) from Bottari et al.
(1982) and computed Minimum 1-D velocity model (solid line).
stopped when the earthquake locations, stations delays
and velocity values did not further vary significantly.
The Minimum 1-D velocity structure is shown in Fig. 4,
together with the a-priori model of Bottari et al. (1982).
P-wave velocity is 4.90 km/s in the upper layer, while in
the deeper layers the P-wave velocity is respectively of
5.20, 6.05, 6.25 and 6.40 km/s up to 7.7 km/s below
33 km. Below 25 km the starting value of velocity remains
unchanged, as few illuminating ray-paths were available.
Station corrections must be considered an integral part
of the velocity model and reflect the heterogeneous near-
surface structure (which may otherwise not be adequately
represented by a 1-D model) (Haslinger 1998). The station
delays are given as relative values with respect to the ref-
erence station DLV. They support the validity of the ve-
locity model as it can be related clearly to the general
93
1-D VELOCITY MODEL: IMPROVEMENT IN EARTHQUAKE LOCATION ACCURACY (ITALY)
Fig. 5. Distribution of seismicity located by Velest after the 1-D inversion process and station corrections of the Minimum 1-D velocity
model. Negative delays correspond to the true velocities faster than the model. E-W and N-S cross-sections show the focal depth for the
207 selected events. The surface evidence of faults is reported with continuous lines.
Fig. 6. Comparison in rms between the location with the routinely
used model and re-location with the computed 1-D model, and
station correction for the 207 selected events. Note the significant
decrease in rms value obtained using the 1-D model.
near-surface conditions inferred from geological evidence.
They show a ‘0’ or positive value in correspondence of
thick sediments and negative values for the others stations
outside the Gioia Tauro Basin, associated with outcropping
metamorphic rocks of the Aspromonte and Serre structures
(Figs. 1 and 5).
Results and discussion
In order to test whether the computed model improves the
routine earthquake location, we relocated the selected data
with the
HYPOELLIPSE
program, and using the computed
Minimum 1-D model and the station corrections. Fig. 6
shows the rms differences between the location using the
model of Bottari et al. (1982) and our optimum 1-D model.
With the optimum 1-D model the rms travel time residuals
were reduced to over 85 % compared to the starting model.
The quality of relocated events significantly improves be-
cause of the considerable reduction of travel time residuals.
A map of relocated events by the program V
ELEST
is
shown in Fig. 5. The distribution of seismicity suggests
that most of the seismic events are clustered in space and
time. One large cluster of about fifty events occurred on
September 11, 1992 at a depth of ca. 10 km, lying in the
area of the station STC. Most of the earthquakes are con-
fined in the upper 25 km of the crust. They occurred along
existent weaknesses zones, such as the Cittanova fault
(Figs. 1 and 5).
To verify if the results are robust performed a stability
test on the selected events as proposed in the user’s guide
of the V
ELEST
code (Kissling 1995).
For this purpose, before the inversion process, we ran-
domly varied, the starting hypocenter locations by an
amount of up to 8 km in E-W, N-S and Z directions. This
94
RAFFAELE and IMPOSA
Fig. 7. Hypocenter stability test. Grey squares: coordinate differ-
ence between randomized input and Minimum 1-D locations.
Black triangles: difference after inverting with the randomized in-
put data. For more details, see text.
process provides a check for a bias in the hypocenter loca-
tions and for the stability of the solution to the coupled
problem. If the proposed Minimum 1-D velocity model de-
notes a robust minimum in the solution space, no signifi-
cant changes in hypocenter locations are expected (Husen
1999). We used the disturbed starting locations in the joint
1-D inversion and examined the differences between the so-
lutions found with the original data and the ones where dis-
turbed starting locations were used. Five performances of
this experiment were carried out and the maximum differ-
ence, between the solutions was taken as an evaluation of
stability (Fig. 7). In fact, all events have been relocated
close to their original position, indicating that the hypo-
center locations were not systematically biased.
A second stability test was carried out, as suggested by
Haslinger (1998), maintaining fixed the final hypocenter
coordinates of the 207 inverted events, and repeating the
inversion process using initial velocity models with high-
er or lower velocities with regard to the Minimum 1-D
model. The stable convergence to the Minimum 1-D mod-
el indicates that the inferred velocity model is an adequate
1-D approximation of the upper 25 km of the crust (Fig. 8).
Concluding remarks
In the Gioia Tauro Basin, data from a 9-year running
seismic network were used to compute the one dimension-
al P-wave velocity model (Fig. 4) and corresponding sta-
tion delays corrections (Fig. 5), by minimizing P-wave
residuals for a data-set of selected events.
Two tests have been performed to check both the hypo-
center locations and the Minimum 1-D model repeating the
inversion process, maintaining fixed either the obtained ve-
locity model or the final earthquake locations. If the pro-
posed Minimum 1-D velocity model denotes a robust
minimum in the solution space, no significant changes in
velocity and hypocenter locations are expected demonstrat-
ing the stability of the solution to the coupled problem. In
the first test all hypocenters were more or less relocated
back to their original positions, while in the second test the
results indicate that the Minimum 1-D model is an adequate
approximation of the crust above 25 km depth.
In general, the location of the epicentres suggests a rela-
tion of the seismicity to the main tectonic lines, such as
the Cittanova fault, which have a NE-SW trend. Most of
the earthquakes are confined to the upper 25 km of the
crust and a significant earthquake cluster (ca. 50 events
occurred on September 11, 1992) has been recorded, at a
depth of ca. 10 km (Fig. 5).
Finally, we explained that the application of the method
by Kissling et al. (1994) is able to calculate a new Mini-
mum 1-D velocity model. This model represents an im-
provement for routine earthquake location with respect to
other 1-D velocity models based on actual a-priori infor-
mation of the basin area.
Acknowledgments:
We thank M. Bielik, J.P. Fourno, H.
Langer and G. Milano for their thoughtful criticisms. We
Fig. 8. Stability tests with high and low input velocity models. In-
put models and Minimum 1-D model shown as solid lines, output
models shown as dashed lines.
95
1-D VELOCITY MODEL: IMPROVEMENT IN EARTHQUAKE LOCATION ACCURACY (ITALY)
are also grateful to an anonymous reviewer, who gave
valuable hints and suggestions for the improvement of the
manuscript. R. Raffaele benefit from a PhD fellowship by
the University of Catania. S. Imposa was supported by
grants (Fondi di Ateneo 2003 and 2004) from the Univer-
sity of Catania.
References
Alvarez W. 1976: A former continuation of the Alps. Bull. Geol.
Soc. Amer. 87, 891—896.
Amodio-Morelli L., Bonardi G., Colonna V., Dietrich D., Giunta
G., Ippolito F., Liguori V., Lorenzoni S., Paglionico A., Per-
rone V., Piccarreta G., Russo M., Scandone P., Zanettin-
Lorenzoni E. & Zuppetta A. 1976: The Calabria-Peloritani Arc
in the Apennine-Maghrebian orogen. Mem. Soc. Geol. Ital.
17, 1—60 (in Italian).
Bottari A., Caccamo D., Cefali F. & Neri G. 1982: Influence of the
velocity model for the hypocentral parameters determination
of the seismic events on southern Tyrrhenian. Istituto Interna-
zionale di Vulcanologia, C.N.R., Open-File Report 2, 82 (in
Italian).
Chiarabba C. & Frepoli A. 1997: Minimun 1D velocity model in
Central and Southern Italy: a contribution to better constrain
hypocentral determination. Ann. Geofisica, XL, 4, 937—954.
Crosson R.S. 1976: Crustal structure modelling of earthquake data.
Simultaneous least squares estimation of hypocenter and ve-
locity parameters. J. Geophys. Res. 81, 3036—3046.
EL-Hadid S., Tealeb A.A., EL-ATA A.S.A.Abu., Taha Y. & EL-
khrepy S. 2005: Initial reference velocity model using local
earthquake data in Dahshour area, Northern Egypt. NRIAG. J.
Geophys. 4, 1, 1—21.
Haccard D., Lorenz C. & Grandjacquet C. 1972: Essai sur
l’évolution tectnogénétique de la liason Alpes-apennines (de la
Ligurie
à la Calabre). Mem. Soc. Geol. Ital. 11, 309—341.
Haslinger F. 1998: Velocity structure, seismicity and seismotecton-
ics of Northwestern Greece between the Gulf of Arta and
Zakynthos. Ph.D. Thesis, Swiss Federal Institute of Technolo-
gy of Zürich, 1—158.
Husen S. 1999: Local Earthquake tomography of a convergent
margin, North Chile – a combined on- and offshore study.
Ph.D. Thesis, University of Kiel, 1—148.
Husen S., Kissling E., Fluhel E. & Asch G. 1999: Accurate hypocen-
tre determination in the seismogenic zona of the subduction
Nazca Plate in northern Chile using a combined on/offshore
network. Geophys. J. Int. 138, 687—701.
Kissling E. 1988: Geotomography with local earthquake data. Rev.
Geophys. 26, 659—698.
Kissling E. 1995: Velest User’s Guide. Int. Report, Inst. Geophys.,
ETH Zurich, 1—26.
Kissling E., Ellsworth W.L., Eberhart-Phillips D. & Kradolfer U.
1994: Initial reference models in local earthquake tomogra-
phy. J. Geophys. Res. 99, 19635—19646.
Kissling E., Solarino S. & Cattaneo M. 1995: Improved seismic ve-
locity reference model from local earthquake data in North-
western Italy. Terra Nova 7, 528—534.
Lahr J.C. 1989: Hypoellipse/Version 2.0: A computer program for
determining local earthquake hypocentral parameters, magni-
tude, and first motion pattern. U.S. Geol. Surv., Open-File Re-
port 89, 116, 1—81.
Musumeci C., Di Grazia G. & Gresta S. 2003: Minimum 1-D veloc-
ity model in Southeastern Sicily (Italy) from local earthquake
data: an improvement in location accuracy. J. Seismology 7,
469—478.
Popa M., Kissling E., Radulian M., Bonjer K.P., Enescu D., Dragan
S. & CALIXTO Research Group 2001: Local source tomogra-
phy using body waves to deduce a minimum 1D velocity
model for Vrancea (Romania) zone. Romanian Report in
Physics 53, 519—536.
Quintero R. & Kissling E. 2001: An improved P-wave velocity ref-
erence model for Costa Rica. Geofis. Int. 40, 3—19.
Raffaele R. 2004: 3D crustal structure from local earthquake data in
the Gioia Tauro Basin (Calabria-Italy). Ph.D. Thesis, Univer-
sit
à degli Studi di Catania, 1—118.
Sherburn S. & White R.S. 2005: Crustal seismicity in Taranaki,
New Zealand using accurate hypocentres from a dense net-
work. Geophys. J. Int. 162, 494—506.
Thurber C.H. 1992: Hypocenter-velocity structure coupling in local
earthquake tomography. Phys. Earth. Planet. Inter. 75, 55—62.
Tortorici L., Monaco C., Tansi C. & Cocina O. 1995: Recent and
active tectonics in the Calabrian arc (Southern Italy). Tectono-
physics 243, 37—55.