Accepted Manuscript

Improving the location of induced earthquakes associated with an undergroundgas storage in the Gulf of Valencia (Spain)

Beatriz Gaite, Arantza Ugalde, Antonio Villaseñor, Estefania Blanch

PII: S0031-9201(16)30006-1DOI: http://dx.doi.org/10.1016/j.pepi.2016.03.006Reference: PEPI 5903

To appear in: Physics of the Earth and Planetary Interiors

Received Date: 13 November 2015Revised Date: 9 March 2016Accepted Date: 10 March 2016

Please cite this article as: Gaite, B., Ugalde, A., Villaseñor, A., Blanch, E., Improving the location of inducedearthquakes associated with an underground gas storage in the Gulf of Valencia (Spain), Physics of the Earth andPlanetary Interiors (2016), doi: http://dx.doi.org/10.1016/j.pepi.2016.03.006

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customerswe are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, andreview of the resulting proof before it is published in its final form. Please note that during the production processerrors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

1

Improving the location of induced earthquakes associated with an underground

gas storage in the Gulf of Valencia (Spain)

Beatriz Gaite*a, Arantza Ugaldea, Antonio Villaseñora, and Estefania Blanchb

a Institute of Earth Sciences Jaume Almera, ICTJA-CSIC, Lluís Solé i Sabarís, s/n,

Barcelona 08028, Spain.

b Observatori de l’Ebre, CSIC – URL, Horta Alta 38, Roquetes 43520, Spain.

Beatriz Gaite: bgaite@ictja.csic.es

Arantza Ugalde: augalde@icja.csic.es

Antonio Villaseñor: antonio@ictja.csic.es

Estefania Blanch: eblanch@obsebre.es

*Corresponding author: e-mail bgaite@ictja.csic.es, Tel. +34 4095410, Fax +34

934110012

2

Abstract

On September 2013, increased seismic activity was recorded near the CASTOR

offshore underground gas storage (UGS), in the Gulf of Valencia (Spain). According to

the reports by the Spanish Instituto Geográfico Nacional (IGN), more than 550 events

occurred during two months, the strongest having a magnitude of Mw=4.2 which took

place two weeks after the gas injection stopped. The low magnitude of the events (with

only 17 earthquakes having mbLg greater than 3), the lack of nearby stations, and the

inhomogeneous station distribution made the location problem a great challenge. Here

we present improved locations for a subset of 161 well recorded events from the

earthquake sequence using a probabilistic nonlinear earthquake location method. A new

3-D shear-wave velocity model is also estimated in this work from surface-wave

ambient noise tomography. To further improve the locations, waveform cross-

correlations are computed at each station for every event pair and new locations are

obtained from an inverted set of adjusted travel time picks. The resulting hypocentral

solutions show a tighter clustering with respect to the initial locations and they are

distributed in a NW-SE direction. Most of the earthquakes are located near the injection

well at depths of about 6 km. Our results indicate that the observed seismicity is closely

associated with the injection activities at the CASTOR underground gas storage and

may have resulted from the reactivation of pre-existing unmapped faults, located a few

kilometers below the reservoir.

Keywords Induced seismicity; nonlinear earthquake location; cross-correlation; 3-D

velocity model; earthquake multiplets.

3

Highlights

• We improve locations of induced seismicity at an offshore underground gas

storage.

• We compute a 3D crustal velocity model of the region to locate the events.

• The locations are improved through waveform cross-correlation.

• We report locations distributed in a NW-SE direction close to the injection well

at about 6 km in depth.

4

1 Introduction

Starting on September 5, 2013 and during two months, a tight cluster of more than 550

earthquakes with mbLg magnitudes ranging from 0.7 to 4.2 were located in the shallow

offshore area of the Gulf of Valencia off the eastern coast of Spain. This seismic activity

rate was extraordinary since, according to the IGN (Instituto Geográfico Nacional)

earthquake catalogue (http://www.ign.es, last accessed: April, 2015), only ten

earthquakes with magnitudes 2.0 or greater had been recorded in this area in the

previous thirty years, the largest were a mbLg 3.3 event that occurred on January 10,

1981 and a mbLg 3.1 on April 8, 2012 (Fig. 1). The sudden seismicity increase coincided

in time with the initial phase of the injection of the base gas in a newly developed

underground natural gas storage (UGS) facility (CASTOR) off the eastern Spanish

coast. Besides the temporal coincidence, the epicentral area of the earthquake sequence

was located close to the gas injection well. The base gas injection finished on

September 16, 2013 but the high seismicity rate lasted until the end of October.

According to the IGN reports, the strongest earthquake (Mw=4.2) took place on October

1, 2013, two weeks after the gas injection stopped, and it was followed by two Mw=4.1

events the day after. Seismic activity declined rapidly six weeks after the gas injection

finished.

Several earthquakes from this sequence were felt with EMS intensities II or III by the

population at the closest coastal villages (Alcanar, Vinarós, Benicarló, etc.) and caused

high social and media impact. According to the seismic hazard map of Spain (IGN

2013) horizontal peak ground acceleration values between 0.04g and 0.05g are expected

in the area for a 475-year return period. These levels are the lowest considered in the

current Spanish seismic regulations (NCSR-02 2003), but they do not take into account

5

the occurrence of seismicity induced by human activities. According to the IGN

historical macroseismic database, six earthquakes with maximum EMS intensity V have

occurred in this region prior to the year 1920. It is worth noting that the surrounding

region is densely populated and industrially developed, with singular infrastructures

such as nuclear power plants, and those of the chemical and oil industry. This situation

led the Spanish government to suspend the activities and put the project on hold (BOE

2014).

Accurate hypocentral locations are important to identify lineaments associated with

faults and other seismically active structures. However, in this case the location

uncertainties of the available seismic catalogue are significant due to the lack of seismic

stations in the epicentral area (Fig. 1). The nearest station is located inland,

approximately 26 km to the NW of the injection well. Moreover, the azimuthal gap of

the network geometry is high because of the lack of offshore seismic stations. This gap

is partially closed to the NE and to the SE at larger distances (65 and 160 km,

respectively). The reference velocity models used also affect the accuracy of the

earthquake locations, especially if the area is highly heterogeneous. In consequence, it

has been difficult to resolve the faults associated with the observed seismicity (e.g.,

Cesca et al. 2014).

In this work we obtained high-precision earthquake locations for a subset of 161 well

recorded earthquakes in the sequence. We observe that the quality of the locations is

greatly conditioned by the uneven azimuthal geometry and large epicentral distances of

the available recording stations. Nevertheless, our results demonstrate that a major

improvement in location accuracy can be achieved with the use of an accurate three-

6

dimensional (3-D) velocity model and differential times among event multiplets

obtained through waveform cross-correlation. [Fig. 1]

2 Setting

The study area is located in the western margin of the Valencia Trough, which is a

NE-SW trending extensional basin characterized by low strain and low-to-moderate

seismicity. The seismic sequence was located offshore, in the Amposta Basin, close to

the Western, Central and Eastern Amposta faults (Fig. 1). The Western and Central

Amposta faults are 18 km and 35 km long, respectively, and dip 60º towards the SE.

The Eastern Amposta fault is 51 km long and dips 60º to the NW (Roca and Guimerà

1992; Perea 2006; Perea et al. 2012). The stress field in the north-western Valencia

Trough is characterized by a normal-faulting tectonic regime with a strike-slip

component (Schindler et al. 1998). The maximum horizontal stress is oriented NE-SW

in the northern portion of the Valencia Trough and N-S to the south of the Ebro Delta

which correlates well with the strike of the normal faults in the region (e.g., Schindler et

al. 1998). Based on historical and instrumental seismicity data, very few seismic events

have occurred in this area, all of them having low magnitudes (M ≤ 3.3) (Fig. 1).

There are several economical oil fields located within a radius of 50 km from the

Ebro Delta that belong to the Valencia Trough hydrocarbon province. Amposta was the

first offshore oil field, it was discovered in the 1970s and operated by Shell from 1973

to 1988, when it was abandoned. The depleted Amposta oil field is located 22 km off

the Spanish coast. It is an elongated, tilted horst feature, dipping towards the ESE at 20º

(Seemann et al. 1990). The western margin of the field is bounded by the Eastern

Amposta fault. Main reservoir rocks are fractured and intensely karstified Jurassic and

7

Lower Cretaceous limestones, overlain by Miocene sediments (Playà et al. 2010). The

CASTOR project was designed to use the Amposta structure as a submarine natural gas

storage facility (Batchelor et al. 2007). The gas was going to be transferred through a

pipeline from a land plant to the offshore platform and then injected in a 60 m water

depth well into the depleted Amposta oil field at ~1750 m depth (Fig. 2). [Fig. 2]

3 Data

Seismic activity in the area is continuously monitored by several regional networks

operated by different institutions: Ebre Observatory (network code EB), that operates

the monitoring network established for the CASTOR project; the IGN (network code

ES) is responsible for earthquake monitoring of the entire Spanish territory; the Institut

Cartogràfic i Geològic de Catalunya (ICGC) and the Institut d’Estudis Catalans (IEC)

(network code CA) (Fig. 1).

Although EB network detected more than 1100 earthquakes from September 5th

, 2013

to March 1st, 2014, our reference in this work was the IGN revised bulletin for the same

time period including 567 earthquakes that combines travel time readings for all the

stations in the region. All the selected earthquakes were recorded by at least 7 to 19

seismic stations. Fig. 3 shows an example waveform recorded at station ALCN for the

largest earthquake of the sequence, the Mw=4.2 event of October 1, 2013. [Fig. 3]

4 Development of a 3-D velocity model

4.1 Method

One dimensional (1-D) velocity models are commonly used for estimating earthquake

locations, especially for automatic, real-time, seismicity monitoring that requires rapid

8

computations. Nevertheless, the accuracy of seismic locations is closely related to the

velocity model used for its determination, especially if the area is highly heterogeneous.

The global 3-D crustal velocity model CRUST 1.0 (Laske et al. 2013) is available in the

whole region with a horizontal resolution of 1º. There are also several two-dimensional

(2-D) crustal seismic models from deep reflection and refraction profiles (e.g., Roca et

al. 2004; Torné et al. 1992; Vidal et al. 1998) that provide detailed crustal structures in

specific areas. The IGN has computed a 3D velocity model that extrapolates the results

of the 2D profiles. However, as far as we know, a published regional 3-D seismic

velocity model of the crust and upper mantle structure of the study area does not exist.

The introduction of Ambient Noise Tomography (ANT) (e.g., Shapiro et al. 2005) has

allowed the construction of surface wave dispersion maps through the inversion of

interstation dispersion measurements extracted from the cross-correlation of ambient

seismic noise. ANT has been widely applied to produce Rayleigh- and Love-wave

group and phase velocity maps at multiple scales across different regions of the world

(e.g., Yao et al. 2006; Villaseñor et al. 2007; Bensen et al. 2008; Yang et al. 2008;

Frontera 2009; Gaite et al. 2012).

In this study we computed the first 3-D shear-wave velocity model for the region

from joint inversion of group and phase velocity of fundamental mode Rayleigh wave.

Using the 2-D Rayleigh wave group and phase velocity maps at different periods from

Silveira et al. (2013) as input data, we first constructed the dispersion curves between 4

and 40 s period at each point on a 0.5º × 0.5º geographic grid. The 1-D shear wave

velocity model at each node was then obtained through joint inversion of the dispersion

curves following Gaite et al. (2014). We solve the forward problem using Herrmann

(1987) and invert using the algorithm simulated annealing (Goffe, 1996). We

9

parameterized the medium with three constant-velocity layers over a half-space. The

four velocities and three thicknesses of the model were perturbed in a wide searching

space. The misfit between data and synthetics was computed using the L2 norm and the

best-fitting model was obtained through iteration. We only allowed optimum models

with velocity increasing with depth. Combining all the 1-D estimates we obtained a

final 3-D crustal and uppermost mantle shear-wave velocity model that covers the

whole area from the crust down to 100 km in depth. The input tomography data has a

horizontal resolution of 100 km or better in this region (Silveira et al. 2013). The 1-D

model grid spacing restricts the better horizontal resolution to 0.5º (~55 km). [Fig. 4]

4.2 Characteristics of the 3-D model

Fig. 4 shows examples of dispersion curves and best fitting 1-D shear-wave velocity

models at four geographic locations. The IGN average velocity model for the Iberian

Peninsula is plotted as a dashed line for comparison. Variability in the models with

respect to the geological settings of the locations is observed in this figure, as the

presence of a thicker crust in the Pyrenees and the Iberian Range than in the Valencia

Trough and the Balearic Promontory (Fig. 1). The velocity differences between the

CRUST2.0 and the model of this study at a near position of CASTOR UGS are clearly

observed in computed fundamental mode Rayleigh-wave group and phase velocities at

long periods (T ≥ 20 s) where the computed model fits well the ANT data (Fig. 5). The

shallower than 20 km moho depth obtained is in coherence with the thin crust observed

from deep seismic reflection profiles across the Valencia shelf and close to CASTOR

UGS (e.g., Torné et al. 1992). Fig. 6 shows the geographical distribution of the misfit

10

values obtained from the inversion. It can be observed that the higher resulting misfit

values lay outside our study area (Fig. 1). [Fig. 5]

The horizontal layers of shear-wave velocity at different depths are plotted in Fig. 7.

They display distinct patterns among different geological regions. The most striking

feature is the low crustal velocities (~2.5–2.7 km/s) at 5 km depth in the Valencia

Trough, indicating a thick sedimentary basin (Fig. 7a). Considering a vP / vS value of

1.75 (Vidal et al., 1998), we obtained equivalent values (vP converted to vS) of 2.9 km/s

for this area from deep 2-D reflection profiles. At 10 km depth, high velocities (~3.9

km/s) are noticeable offshore, in the NE Valencia Trough. In this area, geodynamical

models infer a backarc extension during the Western Mediterranean evolution (e.g.,

Doglioni et al. 1999). The thinner crust along the Algerian Basin and the Valencia

Trough is evidenced at 20 km depth by the high velocity values obtained (4.0-4.4 km/s),

in contrast with the low velocities observed onshore and in the Balearic Promontory. At

40 km depth, the lowest crustal velocities are obtained beneath the central Pyrenees.

Crustal roots reaching 50 km have been reported from seismic refraction profiles in this

part of the orogen (Gallart et al. 1981). [Fig. 6]

Fig. 8 shows a selection of vertical cross-sections with locations indicated on the map

in Fig. 8a. Fig. 8b shows a comparison of the shear-wave velocity model obtained in

this study with the lithospheric structure inferred from numerous deep seismic

refraction/reflection profiles along the TRANSMED II transect (e.g., Roca et al. 2004).

This profile spans from the Pyrenees to the South of the Algerian Basin (A-A’ in Fig.

8a). Despite the low sensitivity of surface-wave velocities to shear-wave vertical

variations, our results show minor differences with respect to the moho depths obtained

by Vidal et al. (1998), Roca et al. (2004), and Carballo et al. (2014) from the Pyrenees

11

to the Balearic Promontory. The upper-mantle velocities obtained in this study beneath

the Valencia Trough (~4.3 km/s) are lower than in the Balearic Promontory (4.4-4.5

km/s). Equivalent values of ~4.5 km/s were modelled by Carballo et al. (2014) in this

area. [Fig. 7]

Some important features of the shear-wave velocity model estimated in this study are

the crustal thinning of up to 15 km observed from different azimuths towards the

Valencia Trough and the low uppermost mantle velocities offshore Eastern Iberia. This

behaviour can be noticed in Fig. 8c and d along the B-B’ and C-C’ profiles. Therefore,

it is observed that the shear-wave velocity model obtained is in accordance with

previous studies showing imprints of the Neogene rifting around the Valencia Trough

affecting the mainland Iberia, (e.g., Zeyen et al. 1985). [Fig. 8]

5 Waveform cross-correlation

Phase readings still may be affected by lack of consistency resulting from path effects,

signal-to-noise conditions, and differences in the source radiation pattern within the

network, in spite of the careful, visual review of the phase arrival times performed at all

the recording stations by an experienced analyst. Earthquake location using inconsistent

travel times makes it difficult to relate location patterns to tectonic structures.

Waveform cross-correlation techniques allow to obtain time shifts among events

characterized by similar location, fault mechanism and propagation pattern (referred to

as multiplets) that can be used to achieve precise relative locations (e.g., Poupinet et al.

1984; Deichmann and García-Fernández 1992; Got et al. 1994; Schaff and Richards

2011). Alternatively, the measured differential times can be combined with the absolute

travel time readings to invert for an adjusted, more consistent set of picks that will

12

produce less scatter in the travel time residuals and the final event locations (e.g.,

Shearer 1997; Rowe et al. 2002, 2004).

To identify families of similar events in our earthquake catalogue, we computed

waveform cross-correlations on the full set of events recorded at the closest station

(ALCN), due to the sparse network and the decrease of quality records at further

stations. Selecting similar events at only one station carries a lack of constrain at

different azimuths that we tried to counterbalance being restrictive when defining the

multiplets. The process was performed using a time window of 20 s, starting 0.2 s

before the P-wave pick in the IGN revised catalogue. This window includes the P and S-

wave arrivals that increases restriction and confidence to identify similar events as done

in other studies for long event-to-station distances (e.g., Shaff and Richards 2004). To

improve the correlation coefficient of the time series, and after testing several frequency

bands, waveforms were filtered in a pass band of 1-10 Hz prior to their correlation.

Waveform similarity was quantified as the maximum value of the cross-correlation

coefficient function. To group events into multiplets, we used an algorithm based on

Maurer and Diechmann (1995). The efficiency of the algorithm depends on the choice

of various threshold values and their correct selection requires extensive testing, as

discussed in detail by these authors. We chose the following value parameters: 0.85 for

waveform similarity; 0.5 for the normalized scalar product of the correlation coefficient

matrix to identify the multiplets; and 1.0 for event association to a multiplet. With these

values, we detected 51 multiplets with at least three members that contain 46% of the

data of the reference catalogue (Fig. 9). The largest multiplet comprises 15 earthquakes.

Visual inspection of the results illustrates the waveform similarity between events from

the same multiplet and the differences with other families, even when the events

13

occurred within the same time period (Fig. 10). One may not expect that alignment of

some portions of records belonging to a multiplet will guarantee alignment of P wave

arrival. However, the multiplets we found show consistent waveforms at P, S and other

converted arrivals (Figure 10). This is strong evidence of the similarity on the

hypocenter location and the rupture mechanism. These observations support the idea of

the ability to obtain accurate adjusted P and S picks for each event within the multiplets

using windows that comprise P and S waves. [Fig. 9]

Similarly we computed waveform cross-correlations for other stations, within the

multiplets defined by station ALCN. In order to include the P- and S-wave arrivals in all

cases, the time window length had to change according to the epicentral distance. We

selected values of 20, 40, and 60 s long time-windows for distances shorter or equal

than 0.7º, 1.5º, and 2.0º, respectively, starting 0.4 s before the P-wave arrival catalogue

reading. Not all the events of the multiplets defined through station ALCN correlated

highly for all stations, so we only kept the information for pairs of events with cross-

correlation coefficients greater than 0.6. Otherwise, we retained the revised, manually

repicked onsets by the IGN analyst. [Fig. 10]

6 Earthquake location

6.1 Method

For locating the earthquakes, we used the NonLinLoc software package (NLL) (Lomax

et al. 2000; http://alomax.free.fr/nlloc/, last accessed: September, 2014). This method

provides a probabilistic solution to the location problem by means of a non-linear

earthquake location technique. The location algorithm is based on the formulation

proposed by Tarantola and Valette (1982). The travel-times between each station and all

14

nodes of a 3-D spatial grid are calculated by means of the finite-difference algorithm of

Podvin and Lecomte (1991) that solves the eikonal equation. To compute the complete,

probabilistic solution in terms of the Probability Density Function (PDF) in 3-D for the

hypocenter location, we used here the Equal Differential Time (EDT) likelihood

function that is more robust in the presence of outliers that may exist in the original data

set than the classic, least-squares, L2 norm (Lomax et al. 2000; Font et al. 2004). In a

heterogeneous velocity model, the EDT formulation involves the intersection of many

deformed hyperbolic surfaces that are independent of the earthquake origin time

estimate. The Oct-Tree, importance-sampling algorithm used by NLL provides a stable

method to estimate the maximum likelihood hypocenter and the PDF from the irregular

EDT topology. NLL also estimates a compatible origin time for the maximum

likelihood hypocenter. NLL can be applied for 1-Das well as for complex 3-Dvelocity

models.

6.2 Location of the seismic sequence using a 3-D velocity model

The results shown in Fig. 7 and 8 indicate high crustal heterogeneity of the area that will

affect the seismic-wave travel times coming from different azimuths. This is not

considered when using 1-D velocity models to solve the earthquake location problem.

The use of an accurate velocity model may help to obtain better locations in spite of the

distant and inhomogeneous azimuthal station distribution of the region. A constant

obtained using the Wadati method was assumed for P-wave arrivals in this

area. For both P- and S-wave velocities we generated a local 601 × 601 × 63 km 3-D

model, interpolated linearly from the original model onto a 1 km grid and centred at the

injection well.

/ 1.71P S

v v =

15

Locations obtained from the IGN catalogue readings using the 3-D velocity model

and the NLL software are plotted in Fig. 11. Comparative locations using NLL for two

different 1-D velocity models are also plotted in this figure. The 1-D velocity models

used are the IGN velocity model for the Iberian Peninsula (Mézcua and Martínez

Solares 1983) and an approximation of the 2-D velocity model of Vidal et al. (1998)

obtained from marine seismic reflection profiles. The selected hypocenter locations

correspond to 161 earthquakes with magnitudes mbLg ≥ 2.0 and have azimuthal gaps less

than 180º. The density scatter plots represent the uncertainties due to the geometry of

the network, measurement errors of the observed arrival times, and errors in the

calculation of theoretical travel times. The RMS travel time residuals are high. The

mean and standard deviation of the RMS values of all the events are 0.74±0.19 s. The

epicentral locations plotted in Fig. 11 do not show any linear alignment, although a

reduction of the scatter in the locations is noted, especially in depth. [Fig. 11]

Fig. 12 plots the relative differences in the locations between the 3-D velocity model

and the two 1-D velocity models shown in Fig. 11. It can be observed that the

relocations using the new three-dimensional velocity model produced a maximum shift

of the epicenters of 9 km to the NW, with respect to using the IGN average velocity

model for the Iberian Peninsula. The relocations with respect to using the 1-D adapted

velocity model of Vidal et al. (1998) produced maximum shifts of 4-6 km to the NW

and the SE. In this study we have compared the locations computed with three different

velocity models and we have obtained different event distributions in depth. This

vertical distribution seems to be related to the depth and magnitude of the velocity

contrast between the sediment layer and the basement of each model. In all cases a

considerable number of event locations are deeper than this velocity contrast. This fact

16

may indicate that the events are occurring more probably in the basement than in the

sedimentary layers. [Fig. 12]

6.3 Relocation of multiplets

For each event pair of a multiplet, the time lag of the cross-correlation function peak

value represents the differential time between the waveforms. Then, we solved for a set

of adjusted peaks for each multiplet and station using the differential times obtained

from waveform cross-correlation together with the travel time picks of the reference

earthquake catalogue following Shearer (1997). Finally, to correct for a possible overall

bias in the preliminary picks, we aligned the waveforms within each multiplet on their

adjusted picks and stacked the traces. Then, we manually picked the stacks and

corrected the adjusted picks accordingly (e.g., Rowe et al. 2004). The improvement in

the consistency of the travel time picks can be observed in the example plotted in Fig.

13. [Fig. 13]

Figure 14 plots the final locations resulting from applying NLL with the 3-D velocity

model to the new set of travel time picks inverted through cross-correlation. A reduction

of the scatter in the locations is noted, especially in depth, compared with the locations

using the original travel time readings (Fig. 11c). The mean and standard deviation of

the RMS values for the correlation adjusted locations is 0.45 ± 0.17 s. The average RMS

is reduced a 40% from the RMS of all events. The horizontal location average error is

reduced from 2.1 ± 1.6 km to 1.5 ± 0.7 km in the direction of maximum error. The error

ellipse strike average is 121 ± 20 º, that might be influenced by the station distribution.

The azimuthal gap of this data subset ranges between 88º and 147º. Most of the

multiplets had associated events with mbLg < 3. Only two events with magnitudes 3.0

17

and 3.2 were associated to a multiplet and they are located between 5 km and 6 km

depth beneath the injection well, as highlighted in Fig. 14. [Fig. 14]

7 Discussion

Injection tests at CASTOR UGS were performed from June to August, 2013, without

causing a seismic activity increase. Nevertheless, the continuous injection of base gas

that took place from September 5th to 16th caused a direct seismic response. More than

1,100 earthquakes were detected by the Ebro Observatory in the weeks following the

start of gas injection at the depleted Amposta oil field, located 22 km offshore the

eastern Spanish coast. Among these, 567 earthquakes with mbLg magnitudes ranging

from 0.7 to 4.2 were located by the IGN at shallow depths close to the gas injection

well. The seismic response to fluid injection at CASTOR was very short (on the time

scale of hours), since the first detected earthquake (mbLg=1.1) occurred on September 5,

2013, the same day of the beginning of injection. The largest observed magnitudes,

however, were experienced two weeks after the gas injection stopped. In this region, the

low background historical seismicity suggests a direct correlation of seismicity with

injection. The relationship between fluid injection into the subsurface and the triggering

of seismicity has been known for several decades (see Verdon 2014 for references).

And, although they are not frequent, there exist cases of induced seismicity by gas

storage operations (e.g., Kwee, 2012; Kraaijpoel et al. 2013). A failure on a fault can be

triggered by fluid injection through the reduction of the effective normal stresses caused

by pore pressure increase in the reservoir. However, changes in the local stress field can

propagate and trigger a seismic event at faults located within 20 kilometers away from

the injection area (e.g., Verdon 2014). In order to identify the structures associated with

18

the induced or triggered shocks, accurate hypocentral locations are required. But this

task is limited in our case by the actual geometry of the monitoring networks, lacking

offshore seismic stations in the injection area and with the nearest station located inland,

at 26 km from the CASTOR platform. Being aware of these difficulties, we have

explored the limits of the possible location accuracy using the available data.

The 3-D shear-wave velocity model computed from ambient noise tomography

describes the crustal structure from the seismic sequence region to each seismic station,

thus generating more accurate travel times than using a 1-D velocity model for the area.

The obtained model images well the tectonic heterogeneity of the region and the main

features revealed previously from seismic reflection and refraction profiles (Fig. 8).

Absolute hypocenter locations using NonLinLoc for a 3-D shear-wave velocity model

and a constant

ratio of 1.71 yield a NW-SE orientation of the epicenters

distribution. Most of the earthquakes are located around the injection well in a region

with a horizontal extension of about 14 km and depths between 0 and 15 km (Fig. 11).

The hypocentral locations plotted in Fig. 11 evidence the relationship between injection

and seismicity, however, they do not show a clear linear alignment that can help in the

correlation of the epicenters with the local faults.

The waveform cross-correlation approach, as applied in this work, has allowed to

obtain more homogeneous travel time picks at the expense of reducing the number of

phase-arrival readings but producing less scatter in the final absolute event locations, as

observed in Fig. 14. In return, additional earthquakes with magnitudes less than 2 could

be accurately located. Results show that the seismic sequence is still distributed in a

NW-SE region, with an approximate maximum horizontal extension of 7 km. Most of

the events are located at a depth of about 6 km. [Fig. 15]

/P S

v v

19

It is observed that almost all the events associated to multiplets have magnitudes mbLg

< 3. Only two events with mbLg > 3 were associated to a multiplet and both belong to

multiplets of events occurred after the gas injection finished (Fig. 10c). Figure 15 shows

a detailed view of the locations of the largest earthquakes of the sequence (mbLg > 3).

Most of the events are distributed to the SE of the CASTOR platform and have depths

around 6 km, including the strongest earthquake of Mw=4.2. The earthquakes located to

the NW occur deeper, at 8-10 km. The NW-SE distribution of the sequence as well as

the computed depths differ from the results of Cesca et al. (2014) who found a NNE-

SSW distribution and a depth range of 0-4 km. These authors estimated absolute and

relative locations and moment tensors for 11 events in the region with mbLg greater than

3.3 and proposed two possible rupture scenarios: one involving a low-angle fault plane

striking roughly parallel to the Eastern Amposta Fault and dipping to the SE or a system

of subvertical faults oriented NW–SE. The velocity models used in the location

procedures may be responsible for some of the observed differences. Whereas Cesca et

al. (2014) used the average velocity model extracted from the CRUST 2.0 model

(Bassin et al. 2000), in this work we have developed a new 3-D velocity model for the

region. The locations obtained in this study would be consistent with different deep

faults, although the events cannot be associated to specific ones due to the hypocentral

location uncertainty and the lack of information of the area at these depths.Different

studies based on ESCI deep seismic reflection profiles (Gallart el al. 1997) across the

northern Valencia Trough show normal fault systems to the deep upper-crust (e.g., Roca

and Guimerà 1992; Vergés and Sàbat 1999). Vidal el al. (1998) could not probe the

normal listric detachment in a deep upper-crust identified by Roca and Guimerà (1992),

but observed high lateral changes of reflectivity at the lower crust. Across the Valencia

20

shelf and closer to Amposta Basin, the north-eastern part of profile 819 from VALSIS

multichannel seismic profile (Watts et al. 1990) shows faults at two way time (TWT) of

2 s (Fig. 2 Mauffret et al. 1992), deeper than the CASTOR UGS. [Fig. 16]

Analyzing the obtained multiplets we can extract information on the temporal

evolution of the seismic sequence. The computed multiplets consist mainly of events

that occurred close in time, separated from tens of minutes to a few days (e.g., multiplet

8 in Fig. 10a). However, there are some exceptions such as multiplets whose events

occurred even more than 20 days apart. We also found that some earthquakes occurred

during the gas injection period are similar to events happened once the injection was

stopped (e.g., multiplet 6 in Fig. 10b). In this regard, the temporal evolution of the

earthquake locations does not show a clear trend of seismicity migration (Fig. 16).

Locations using the adjusted phase time readings are less scattered and they are

distributed around the platform location at about 6 km in depth.

8. Conclusions

The 2013 seismic sequence occurred in the offshore region of the Eastern Iberian

Peninsula seems to be closely associated with the injection activities at the CASTOR

underground gas storage facility. Although the long event-to-station distance and the

poor azimuthal coverage limited the accuracy of the earthquake location results, the

obtained hypocenter distribution pattern and the focal mechanisms determined in

previous works indicate that the induced seismicity in the region may have resulted

from the reactivation of pre-existing unmapped faults, located few kilometres below the

reservoir. This conclusion is also supported by the fact that the largest events, reaching a

magnitude of Mw=4.2, occurred two-weeks after the gas injection stopped. The injection

21

activities have been suspended at the CASTOR site and the seismic activity has tapered

off, but it continues 2 years later, although at a much lower level and no significant

earthquakes have been detected in the region since the end of April 2014.

22

Acknowledgments

We thank the Instituto Geográfico Nacional, Observatori de l’Ebre, Institut Cartogràfic i

Geològic de Catalunya, and the Institut d’Estudis Catalans for providing waveform and

catalogue data for this study. We gratefully acknowledge J. V. Cantavella for providing

helpful information on the IGN earthquake locations. We also thank the IGN analyst, R.

Cerdeño, who manually repicked the initial catalogue data used in this work. We thank

M. Ruiz for the information provided on the multiplet detection method. This research

was partially funded by project MISTERIOS (CGL2013-48601-C2-1-R).

Role of the funding source

The funding source is not involved in the study design; in the collection, analysis and

interpretation of data; in the writing of the report; and neither in the decision to submit

the article for publication.

Conflict of interest

The authors declare that they have no conflict of interest.

23

References

Bassin C, Laske G, Masters TG (2000) The current limits of resolution for surfacewave

tomography in NorthAmerica. EOS, Trans Am Geophys Un 81(48), Fall Meet Suppl

Abstract S12A-03.

Batchelor JA, Sherley BW, Klepacki D, Lewis RE, Jensen RE (2007) Validating the

integrity of the Amposta structure for gas storage offshore Spain. Proceedings of the

Offshore Europe, September 4–7, Aberdeen, Scotland, UK, Society of Petroleum

Engineers, SPE 108247:80-95. doi:10.2118/108247-MS.

Bensen GD, Ritzwoller MH, Shapiro NM (2008) Broadband ambient noise surface

wave tomography across the United States. J Geophys Res 113:B05306.

doi:10.1029/2007JB005248.

Bensen GD, Ritzwoller MH, Yang Y (2009) A 3-D shear velocity model of the crust

and uppermost mantle beneath the United States from ambient seismic noise.

Geophys J Int 177(3):1177-1196. doi: 10.1111/j.1365-246X.2009.04125.

Berteussen KA (1977) Moho depth determination based on spectral ratio analysis of

NORSAR long-period P waves. Phys Earth Planet Inter 31:313-326.

doi:10.1016/0031-9201(77)90006-1.

BOE (2014) Boletín Oficial del Estado, Real Decreto-ley 13/2014, de 3 de octubre, por

el que se adoptan medidas urgentes en relación con el sistema gasista y la titularidad

de centrales nucleares. BOE-A-2014-10059 Sec I, pp. 79012-79024. ISSN: 0212-

033X (in Spanish).

Carballo A, Fernández M, Torné M, Jiménez-Munt I, Villaseñor A (2014) Thermal and

petrophysical characterization of the lithospheric mantle along the northeastern Iberia

geo-Transect. Gondwana Research. doi: 10.1016/j.gr.2013.12.012.

24

Deichmann N, García-Fernández M (1992) Rupture geometry from high-precision

relative hypocentre locations of microearthquake clusters, Geophys J Int 110(3):501-

517. doi: 10.1111/j.1365-246X.1992.tb02088.x.

Doglioni C, Fernández M, Gueguen E, Sàbat F (1999) On the interference between the

early Apennines-Maghrebides backarc extension and the Alps-Betics orogen in the

Neogene Geodynamics of the western Mediterranean. Boll Soc Geol It 118:75-89.

Font Y, Kao H, Lallemand S, Liu CS, Chiao LY (2004) Hypocentre determination

offshore of eastern Taiwan using the Maximum Intersection method. Geophys J Int

158:655-675.

Fremont MJ, Malone SD (1987) High precision relative locations of earthquakes at

Mount St. Helens. Washington, J Geophys Res 100:9865–9880.

doi: 10.1029/JB092iB10p10223.

Frontera T. (2009) Estudis de soroll sísmic ambiental amb registres de sismòmetres de

fons marí. Tomografia d’ones superficials al NE de la Península Ibèrica. PhD Thesis

pp 266 (In catalan).

Gaite B, Iglesias A, Villaseñor A, Herraiz M, Pacheco JF (2012) Crustal structure of

Mexico and surrounding regions from seismic ambient noise tomography. Geophys J

Int 188:1413–1424. doi:1/j.1365-246X.2011.05339.x.

Gaite B, Villaseñor A, Iglesias A, Herraiz M, Jiménez-Munt I (2015) A 3-D shear

velocity model of the southern North America and the Caribbean plates from ambient

noise and earthquake tomography. Solid Earth 6:271-284. doi:10.5194/se-6-271-

2015.

Gallart J, Banda E, Daignières M (1981) Crustal structure of the Paleozoic Axial Zone

of the Pyrenees and transition to the North Pyrenean Zone. Ann Geophys 37:457-480.

25

Gallart J, Vidal N, Estévez A, Pous J, Sàbat F, Santiesteban C, Suriñach E, ESCI-

Valencia Trough Gorup (1997) The ESCI Valencia Trough vertical reflection

experiment: a seismic image of the crust from the NE Iberian Peninsula to the

Western Mediterranean. Rev Soc Geol España 8(4):401-415.

Goffe WL (1996) SIMANN: A Global Optimization Algorithm using Simulated

Annealing. Studies in Nonlinear Dynamics & Econometrics. 1(3):1558-3708.

doi:10.2202/1558-3708.1020.

Got JL, Fréchet J, Klein FW (1994) Deep fault plane geometry inferred from multiplet

relative relocation beneath the south flank of Kilauea. J Geophys Res 99(B8):15375–

15386. doi: 10.1029/94JB00577.

IGN, Instituto Geográfico Nacional (2013) Actualización de mapas de peligrosidad

sísmica de España 2012. Centro Nacional de Información Geográfica (CNIG) Ed.,

Madrid, pp 267.

Kraaijpoel D, Nieuwland D, Dost B (2013) Microseismic monitoring and subseismic

fault detection in an underground gas storage. Fourth Passive Seismic Workshop:

Optimizing Development of Unconventional Reservoirs, EAGE PS19. Amsterdam,

The Netherlands.

Kwee J (2012) Microseismicity in the Bergermeer gas storage field, MSc Thesis. Univ.

Utrecht, The Netherlands.

Laske G, Masters G, Ma Z, Pasyanos M (2013) Update on CRUST1.0 - A 1-degree

Global Model of Earth's Crust. Geophys Res Abstracts 15. Abstract EGU2013-2658.

Lomax A, Virieux J, Volant P, Berge C (2000) Probabilistic earthquake location in 3D

and layered models: Introduction of a Metropolis-Gibbs method and comparison with

26

linear locations. In: Thurber CH, Rabinowitz N (eds.) Advances in Seismic Event

Location, Kluwer, Amsterdam, pp 101-134.

Mauffret A, Maillard A, Pascal G, Torné M, Buhl P, Pinet B (1992) Long-listening

multichannel seismic profiles in the Valencia Trough (Valsis 2) and the Gulf of Lions

(ECORS): a comparison. Tectonophysics 203:285-304.

Maurer H, Deichmann N (1995) Microearthquake cluster detection based on waveform

similarities, with an application to the western Swiss Alps. Geophys J Int 123(2):588-

600. doi: 10.1111/j.1365-246X.1995.tb06873.x.

Mezcua J, Martinez Solares JM (1983) Sismicidad del Area Ibero-Mogrebi. Instituto

Geográfico Nacional Publ. 303, Madrid, pp 189.

NCSR-02 (2003) Earthquake resistant building code: general part and construction.

Ministerio de Fomento, Spain, pp 96.

Perea H (2006) Falles actives i perillositat sísmica al marge nordoccidental del solc de

València. PhD Thesis, Universitat de Barcelona, Barcelona pp 382.

http://hdl.handle.net/10803/1919. Accessed 22 April 2015.

Perea H, Masana E, Santanach P (2012) An active zone characterized by slow normal

faults, the northwestern margin of the Valencia trough (NE Iberia): a review. J Iber

Geol 38:31-52. doi:10.5209/rev_JIGE.2012.v38.n1.39204.

Playà E, Travé A, Caja MA, Salas R, Martín-Martín JD (2010) Diagenesis of the

Amposta offshore oil reservoir (Amposta Marino C2 well, Lower Cretaceous,

Valencia Trough, Spain). Geofluids 10:314-333. doi: 10.1111/j.1468-

8123.2009.00266.x.

Podvin P, Lecomte I (1991) Finite difference computation of traveltimes in very

contrasted velocity models: a massively parallel approach and its associated tools.

27

Geophysical Journal International 105:271-284. doi: 10.1111/j.1365-

246X.1991.tb03461.x.

Poupinet G, Ellsworth WL, Frechet J (1984) Monitoring velocity variations in the crust

using earthquake doublets: an application to the Calaveras Fault, California. J

Geophys Res 89:5719–5731. doi:10.1029/JB089iB07p05719.

Roca E, Guimerà J (1992) The neogene structure of the eastern Iberian margin:

structural constraints on the crustal evolution ofthe Valencia Trough (Western

Mediterranean). Tectonophysics 203: 203-218. doi: 10.1016/0040-1951(92)90224-T.

Roca E, Frizon De Lamotte D, Mauffret A, Bracène R, Vergés J, Benaouali N,

Fernández M, Muñoz JA, Zeyen H (2004), TRANSMED Transect II. In: Cavazza W,

Roure F, Spakman W, Stampfli GM, Ziegler PA (eds) The TRANSMED Atlas-The

Mediterranean region from crust to mantle. Springer, Berlin Heidelberg New York.

ISBN:3-540-22181-6.

Rowe CA, Aster RC, Borchers B, Young CJ (2002) An automatic, adaptive algorithm

for refining phase picks in large seismic data sets. Bull Seismol Soc Am, 92:1660–

1674. doi:10.1785/0120010224.

Rowe CA, Thurber CH, White RA (2004) Dome growth behavior at Soufriere Hills

Volcano, Montserrat, revealed by relocation of volcanic event swarms, 1995–1996. J

Volcanol Geotherm Res 134:199-221. doi:10.1016/j.jvolgeores.2004.01.008.

Sabra KG, Gerstoft P, Roux P, Kuperman WA, Fehler MC (2005) Surface wave

tomography from microseisms in Southern California. Geophys Res Lett,

32(14):L14311. doi:10.1029/2005GL023155.

28

Shapiro NM, Campillo M, Stehly L, Ritzwoller MH (2005) High-resolution surface-

wave tomography from ambient seismic noise. Science 307(5715): 1615-1618. DOI:

10.1126/science.1108339.

Schaff DP, Richards PG (2004) Repeating seismic events in China. Science

303(5661):1176-1178. doi: 10.1126/science.1093422.

Schaff DP, Richards PG (2011) On finding and using repeating seismic events in and

near China. J Geophys Res 116. doi:10.1029/2010JB007895.

Shearer P (1997) Improving local earthquake locations using the L1 norm and

waveform cross-correlation application to the Whittier Narrows, California aftershock

sequence. J Geophys Res 102:8269–8293. doi: 10.1029/96JB03228.

Schindler A, Jurado MJ, Muller B (1998) Stress orientation and tectonic regime in the

northwestern Valencia Trough from borehole data. Tectonophysics 300:63-77. doi:

10.1016/S0040-1951(98)00234-0.

Seemann U, Pümpin VF, Casson N (1990) Amposta oil field. In: AAPG Treatise of

Petroleum Geology, Atlas of oil and gas fields A-017, pp 1–20.

Silveira G, Días NA, Villaseñor A (2013) Seismic imaging of the western Iberian crust

using ambient noise: Boundaries and internal structure of the Iberian Massif.

Tectonophysics 589:186-194. doi: 10.1016/j.tecto.2012.12.025.

Tarantola A, Valette B (1982) Inverse Problems=Quest for information. Journal of

Geophysics, 50(3):159–170.

Torné M, Pascal G, Buhl P, Watts AB, Mauffret A (1992) Crustal velocity structure of

the Valencia Trough (western Mediterranean), Part I. A combined refraction/wide-

angle reflection and near-vertical reflection study. Tectonophysics 203(1):1-20.

doi:10.1016/0040-1951(92)90212-O.

29

Verdon J.P. (2014) Significance for secure CO2 storage of earthquakes induced by fluid

injection. Environ. Res. Lett. 9, 064022, 10pp. doi: 10.1088/1748-9326/9/6/064022.

Vergés J, Sabat F (1999) Constrains on the Neogene Mediterranean kinematic evolution

along a 1000 km transect from Iberia to Africa. In: Durand B, Jolivert L, Horváth F,

Séranne M (eds) The Mediterranean Basins: Tertiary extension within the Alpine

Orogen. Geol. Soc. of London, Special Publications 156:63-80.

Vidal N, Gallart J, Dañobeitia JJ (1998) A deep seismic crustal transect from the NE

Iberian Peninsula to the western Mediterranean. Journal of Geophysical Research

103(B6):12381-12396. doi: 10.1029/98JB00076.

Villaseñor A, Yang Y, Ritzwoller MH, Gallart J (2007) Ambient noise surface wave

tomography of the Iberian Peninsula: implications for shallow seismic structure.

Geophys Res Lett 34:L11304. doi:10.1029/2007GL30164.

Watts AB, Torné M, Buhl P, Mauffret A, Passcal G, Pinet B (1990) Evidence of deep

reflections from extended continental crust beneath the Valencia Trough, Western

Mediterranean. Nature 348(6302):631-635.

Yang Y, Ritzwoller MH, Levshin AL, Shapiro NM (2007) Ambient noise Rayleigh

wave tomography across Europe. Geophys J Int 168:259–274. doi: 10.1111/j.1365-

246X.2006.03203.x.

Yao H, van der Hilst RD, de Hoop MV (2006) Surface-wave array tomography in SE

Tibet from ambient seismic noise and two-station analysis-I, Phase velocity maps.

Geophys J Int 166:732–744.

Zeyen HJ, Banda E, Gallart J, Ansorge J (1985) A wide angle reconnaissance survey of

the crust and upper mantle in the Celtiberian Chain of eastern Spain. Earth Planet Sci

Lett 75:393-402. doi: 10.1016/0012-821X(85)90182-7.

31

Figure captions

Fig. 1 Seismicity map of the eastern Iberian Peninsula according to the IGN seismic

catalogue. Triangles indicate the location of the seismic stations from the networks

compiled in this study. Filled triangles are analyzed stations. Empty triangles indicate

stations that were not used for various reasons (their position is close to another

station, they were deployed after the seismic sequence, etc). The red square indicates

the location of CASTOR underground gas storage (UGS). The W, C and E denote the

Western, Central and Eastern Amposta faults, respectively.

Fig. 2 Sketch of the Amposta structure along the WNW-ESE seismic profile of

Seemann et al. (1990). The vertical scale is plotted in two-way traveltime. The

approximate location of the Castor well (dashed line) is shown together with the

original oil-water contact (OOWC) at 1,940 m in depth. The yellow area is the rough

location of the gas reservoir.

Fig. 3 Example of vertical-component waveform recorded at the short-period station

ALCN. The Mw=4.2 event that took place on October 1, 2013 is shown.

Fig. 4 Examples of 1-D shear-wave velocity models obtained at four nodes of the grid

located in the (a) Pyrenees, (b) Iberian Range, (c) Valencia Trough, and (d) Balearic

Promontory. The black lines are the optimum inverted velocity models and the

corresponding dispersion curves. The error bars indicate the misfit. The grey squares

and dots are the input phase and group, respectively, dispersion velocities from ANT.

The grey lines are the accepted models whose misfits are smaller than or equal to two

times the misfit of the best fitting. The IGN, 1-D shear-wave velocity model for the

Iberian Peninsula is indicated with a dotted line.

32

Fig. 5 1-D shear-wave velocity models close to CASTOR underground gas storage and

their computed fundamental mode Rayleigh-wave dispersion curves. Solid lines show

results from this study and dotted lines from CRUST 2.0 model (Bassin et al. 2000).

The grey dots are phase and group dispersion velocities from ANT from Silveira et al.

(2013) inverted to compute the shear-wave velocity model of this study.

Fig. 6 Geographical distribution of the shear-wave velocity inversion misfit. Structural

boundaries are indicated as thin gray lines. The study area is indicated with a dashed-

line box.

Fig. 7 Horizontal cross-sections of shear-wave velocity at different depths: (a) 5 km, (b)

10 km, (c) 20 km, and (d) 40 km. Structural boundaries are indicated as thin black

lines. The red square indicates the location of the CASTOR UGS.

Fig. 8 (a) Map showing the main geologic features of the region and the location of

profiles A-A’, B-B’ and C-C’. Bal. P. denotes Balearic Promontory, C.C.R. Catalan

Coastal Ranges and the red square the CASTOR underground gas storage facility; (b)

A-A’ vertical cross-section of shear-wave velocity model. The crustal bodies and

moho depth modeled by Carballo et al. (2104) are denoted as purple lines. The red

and black lines are the moho depths computed by Roca et al. (2004) and Vidal et al.

(1998), respectively; (c) B-B’ vertical cross-section, and (d) C-C’ vertical cross-

section. The grey line in (b), (c) and (d) denotes the moho depth computed in this

study as the depth of maximum increase of velocity.

Fig. 9 (a) Event pairs with cross-correlation coefficients (CCC) greater than 0.85

obtained at station ALCN; and (b) Selected multiplets. The gray area marks the gas

injection period.

33

Fig. 10 Examples of multiplets observed at ALCN station during and/or after the

injection activities. All the events have magnitudes lower than 3 except for one

earthquake with magnitude 3.2 which is plotted in (c).

Fig. 11 Hypocentral locations of 161 events obtained using NLL by means of the Oct-

Tree algorithm in combination with the EDT likelihood function for: (a) the IGN

velocity model; (b) the velocity model of Vidal et al. (1998) from marine seismic

reflection profiles; and (c) the 3-D velocity model of this study. The shear-wave

velocity models are plotted. In the 3D case, the 1-D shear wave velocity model

obtained for a grid node close to the injection platform is plotted. The gray dots are

the location PDFs. The Eastern, Central and Western Amposta faults are plotted. The

nearest seismic stations are plotted with triangles. The white square indicates the

location of the injection well.

Fig. 12 Blue lines indicate relocation vectors from 1-D velocity model locations to the

3-D velocity model’s. The IGN 1-D model for the Iberian Peninsula (top) and the 1-D

adapted velocity model Vidal et al. (1998) from marine seismic reflection profiles

(bottom).

Fig. 13 Comparison of the recordings at station VAN2 aligned at (a) the original picks

and (b) the adjusted picks for one multiplet. The left panels show a zoom of the first

1.4 s from the records in panels on the right.

Fig. 14 Hypocentral locations of 116 events associated as multiplet obtained using NLL

for a 3-D velocity model and the adjusted picks obtained from waveform cross-

correlation. The gray dots are the location PDFs. The Eastern, Central and Western

Amposta faults are plotted. The nearest seismic stations are plotted with triangles.

34

The white square indicates the location of the injection well. Two events with

magnitudes 3.0 and 3.2 are highlighted in green.

Fig. 15 Horizontal (top) and vertical (bottom) cross-sections with the location of the

earthquakes with magnitudes greater than 3. Open circles are the locations obtained

with NLL using the 3-D velocity model. Green solid circles represent events with

adjusted picks. Earthquakes with magnitudes greater than 4 are highlighted in blue.

The gray dots are the location PDFs. The black square indicates the location of the

injection well. The Eastern Amposta fault system is mapped in gray at the surface

(Quaternary Active Faults Database of Iberia,

http://www.igme.es/infoigme/aplicaciones/qafi/. Accessed September 2014) and at

1,700 meters in depth (Batchelor et al. 2007).

Fig. 16 Spatio-temporal distribution of seismicity. Gray circles are the locations

obtained with NLL using the 3-D velocity model. Red circles represent locations

using the adjusted picks. Latitude and longitude of the CASTOR well are plotted as a

reference. Injection activities finished on September 16, 2013.

36

Figure 1

37

Figure 2

38

Figure 3

39

Figure 4

40

Figure 5

41

Figure 6

42

Figure 7

43

Figure 8

44

Figure 9

45

Figure 10

46

Figure 11

47

Figure 12

48

Figure 13

49

Figure 14

50

Figure 15

51

Figure 16

52

Highligths

• A 3-D crustal velocity model for Western Mediterranean is obtained.

• Some similar events occurred during and after the gas injection.

• The relocated seismic sequence is distributed in a NW-SE direction.

• Most of the events are located around 6 km depth.