Imaging composite dike propagation (Etna, 2002 case)

Marco Aloisi,1 Alessandro Bonaccorso,1 and Salvatore Gambino1

Received 27 June 2005; revised 13 January 2006; accepted 16 February 2006; published 21 June 2006.

[1] Late on the night of 26 October 2002, a dike intrusion started suddenly at Mount Etna,producing intense explosive activity and lava effusion on the southern flank. Five to sixhours afterward, a long field of eruptive fractures propagated radially along thenortheastern flank of the volcano, producing marked variations at the permanent tiltnetwork. The dike propagation velocity was inferred by the associated seismicity. Wemodeled the temporal evolution of the continuously recorded tilt data, both during thevertical dike propagation on the high south flank on 26 October and during the radialpropagation along the northeast flank, between 27 and 28 October. The reproduction ofthe recorded tilt signal allowed us to describe the geometry and characteristics of the twodikes in greater detail than the previous static inversion. We deduced that the eruptionwas characterized by an unusual composite mechanism, clearly showing a transition froma nearly pure opening mode displacement to a mechanism characterized by an equallystrong normal dip-slip component and a smaller left lateral strike-slip component. In thisstudy we demonstrate the interaction between the final segment of the dike and apreexisting structure that was reactivated in response to the intrusion. We show that tilt andits modeling represent a powerful tool to verify and constrain dike intrusions in detail.

Citation: Aloisi, M., A. Bonaccorso, and S. Gambino (2006), Imaging composite dike propagation (Etna, 2002 case), J. Geophys.

Res., 111, B06404, doi:10.1029/2005JB003908.

1. Introduction

[2] Intrusions of magma as horizontally or verticallypropagating dikes are often observed and modeled inbasaltic volcanoes (e.g., Cervelli et al. [2002] in Hawaii,Yamaota et al. [2005] in Japan, and Fukushima et al. [2005]at Piton de la Fournaise).[3] Mount Etna (eastern Sicily, Italy) is one of the largest

and most active volcanoes of the world. Volcanic activity atMount Etna is principally focused at both summit cratersand lateral flanks. Persistent activity at summit craters ischaracterized by degassing, Strombolian explosions, lavafountaining and low effusion rate lava emissions. The flankeruptions are characterized by dike propagation associatedwith opening of eruptive fractures, generally oriented alongtwo main rift zones: the northeast (NE) rift and the south (S)rift (Figure 1).[4] Late on the night on 26 October 2002, a series of dike

intrusions began on Mount Etna, fifteen months after theeruption of July–August 2001. The onset of the intrusionswas heralded by only a few hours of premonitory seismicity.Eruptive fissures opened both on the south and northeastflanks of the volcano. A first dike ascended verticallythrough the volcano edifice in only a few hours and waslocated close to the 2001 eruption site [Aloisi et al., 2003].Similar to the previous 2001 eruption, the eruption of the

southern dike produced cinder cones at 2750 m, fed lavaflows on the S-SW sector, and was accompanied by firefountain activity for about two months. No eruptive fissurespropagated toward the lower part of the south flank, and thelava flows did not pose any risk for the villages on thesouthern flank.[5] A second dike propagated laterally to the northeast for

a few days. It produced explosive activity forming a line ofseveral cinder cones along the fissures from which lavaflows poured out [Andronico et al., 2005]. Therefore in thisflank the main problem was the lava flow, which came outat a shallower altitude of �2000 m above sea level (asl) andthreatened the village of Linguaglossa (Figure 1). Duringthe first days, there was concern that this eruptive fracturecould have propagated further and discharged magma in ahuge flank eruption. The continuous tilt signals showedmarked variations during the intrusion process [Aloisi et al.,2003] and also clearly showed when the intrusion stopped.[6] The second intrusive mechanism was accompanied by

a clear epicentral migration of earthquakes toward NErecorded in the first 24 hours, while compressive focalmechanisms mainly characterized the sector when theeruptive fissure stopped [Barberi et al., 2004]. The mainscientific aspect was to understand the dynamics andmechanisms of this dike emplacement.[7] Previously, ground deformation from GPS and tilt

data was modeled by inverting changes in ground positionsand tilts recorded between 26 and 27 October [Aloisi et al.,2003]. These changes were accumulated in a time intervalbefore and after the dike propagation. The inverse modelwas constrained to have two tabular dislocations. Theresults indicated a composite intrusion involving a tensile

JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 111, B06404, doi:10.1029/2005JB003908, 2006ClickHere

for

FullArticle

1Istituto Nazionale di Geofisica e Vulcanologia, Sezione di Catania,Catania, Italy.

Copyright 2006 by the American Geophysical Union.0148-0227/06/2005JB003908$09.00

B06404 1 of 13

vertical dike rising in the upper portion of the southern flank(with an opening component of �1.5 m) and a radialintrusion in the NE sector (with opening and strike-slipcomponents of �1.0 m and �0.5 m, respectively). Recently,several authors have modeled the propagation of the dikesusing continuous tilt data [Bonaccorso, 1998; Aoki et al.,1999; Battaglia and Bachelery, 2003]. In order to investi-gate the double dike intrusion, in this paper we employ amodeling technique which used all tilt data recorded duringthe time interval of the dike intrusion taking also inconsideration seismic features. We show that the methodprovides a powerful tool to constrain the geometry andcharacteristics of the intrusion process in greater detail withrespect to the previous static analysis.[8] The modeling shows that the 2002 eruption was

characterized by a new mechanism composed by the coex-istence of the two main different classes of intrusion, i.e.,both by a vertical intrusion separated from central conduitsin the S flank and by a shallow horizontal intrusionpropagating from craters area toward the NE flank.

2. Feeder Dikes Mechanism at Mount Etna[9] Mount Etna is one of the most active and best-

studied volcanoes in the world. In particular, during thelast 25 years extensive ground deformation surveying hasprovided important information on the eruption mecha-nisms, including the final transport within the volcanoedifice. Several studies on ground deformation havebeen conducted on the recent flank eruptions of 1981,1983, 1989, 1991–1993 and 2001 (for an overview, seeBonaccorso and Davis [2004]).

[10] At Mount Etna the final intrusion mechanisms of thevarious flank eruptions can be characterized in terms of dikeinjection and eruption. Two classes of final magma pene-tration are identified: (1) dikes that propagate vertically and(2) dikes that propagate horizontally (Figure 2) [Bonaccorsoand Davis, 2004].[11] The first class of dikes propagate vertically from

several km depth to break out at the surface bypassing thecentral conduit system. The dimensions of these dikes aregreater in height (around 2–3 km). Examples of this caseare the 1981 and 2001 eruptions. The main characteristics ofthese two eruptions were the fast emplacement of the dikes(3–4 months for the 1981 and only 4 days for the 2001case), and the small volume of lava output. In the 1981 casethe magma output was stored in this dike during the 3–4 months before the eruption [Bonaccorso, 1999] whichthen discharged about half its volume in the succeedingeruption. In the case of the July 2001 eruption, the emplace-ment was, by contrast, much faster and the volume of themodeled dike (�18 � 106 m3) was much smaller than theestimated erupted volume (�50� 106 m3). It is believed thatthe deeper source also supplied magma to the eruption[Bonaccorso et al., 2002].[12] The second class of dikes propagate from the summit

craters to the eruption point. These dikes supply the magmato the flanks from a deeper magma storage through thesummit conduit zone. Typically the dike depths are about0.5–1 km. Examples of this kind of event include the 1983,1989 and 1991–1993 eruptions. Those eruptions are pre-sumed to be connected to and fed by a deeper storagereservoir, since their lava output is much larger than

Figure 1. Map of Mount Etna with the permanent tiltnetwork and the 2002 lava flow fields. The south (S) andnortheast (NE) rift zones are schematically indicated. Thesummit area in the inset is enlarged in Figure 3. Of thevillages on the slopes of Mount Etna, only the villages ofNicolosi and Linguaglossa are shown.

Figure 2. Map with the location of the dikes modeled forthe recent flank eruptions (redrawn from Bonaccorso andDavis [2004]).

B06404 ALOISI ET AL.: IMAGING COMPOSITE DIKE PROPAGATION

2 of 13

B06404

Figure 3. Tilt signals recorded at (a) PDN station and (b) NE radial propagating intrusion. D1 and D2represent the upper and lower modeled segments. Diamonds and circles represent the position of themodel dike as it propagates. ‘‘A’’ represents the starting point of the model, ‘‘B’’ represents theorthogonal projection to PDN radial component, and ‘‘C’’ represents the end of fracturing. ‘‘Rad’’ and‘‘Tang’’ indicate the radial and the tangential components at the PDN tiltmeter, respectively.

B06404 ALOISI ET AL.: IMAGING COMPOSITE DIKE PROPAGATION

3 of 13

B06404

estimates of the magma stored in the dikes and the centralconduit [Bonaccorso and Davis, 2004]. For the 1983 and1991–1993 cases the volume of magma erupted was verylarge, �1000 � 106 m3, and �2400 � 106 m3, respectively.[13] It is noteworthy that all the previously mentioned

modeled eruptions did not involve the NE flank of MountEtna, and therefore the 2002 eruption represents the firstopportunity to monitor and model an intrusive processalong this flank. Furthermore, a previous case of twoeruptive fractures on two different flanks of the volcanoduring the same eruption is reported only for the 1879event, and therefore the 2002 eruption is an exceptional casewith the coexistence of the two different classes of intru-sion, i.e., the presence both of a vertical intrusion separatedfrom central conduits in the S flank and by a shallowhorizontal intrusion propagating from the crater area towardthe NE flank.

3. The 2002 Double Intrusion at South andNortheastern Rifts

[14] A multidisciplinary volcanological study of the2002–2003 Etna eruption is reported in Andronico et al.[2005]. The authors show that the magma erupted from theS fissure was the relatively undegassed volatile-rich magmafraction which drained the deep feeding system bypassingthe central conduits. Magma erupted from the NE fissurewas the partially degassed fraction normally residing withinthe central conduits and the shallow plumbing system.Therefore the eruption was characterized by an unusualcomposite mechanism [Aloisi et al., 2003], with a verticaldike rising in the southern flank, independently of thecentral conduits, and another dike propagating radially fromthe craters along the NE rift (Figure 1) and emptying themagma residing in the shallow plumbing system.[15] Despite the analytical modeling of geodetic data

related to recent (since 1981) eruptions (for a review, seeBonaccorso and Davis [2004]) the 2002–2003 eruptiongives us, for the first time, the opportunity to model anintrusive mechanism along the NE rift (Figure 1). TheNE rift extends � 7 km from Etna’s summit down to1500 m asl and contains a series of eruptive fissures thatstrike N to NE [e.g., Garduno et al., 1997; Neri et al., 2004].In the summit area, the fissures strike N and are almostcompletely covered by historic lava flows from the NEsummit crater. From 2500 m down to 1700 m asl the fissureforms aNE-striking system. The rift is linked to the Pernicanafault system, which strikes E-W and dissects the northeastflank of the volcano (Figure 1). The NE rift and the Pernicanafault system play an important role in the large eastwarddisplacements measured on the eastern flank representing thenorthern border of this unstable sector [e.g., Froger et al.,2001; Bonforte and Puglisi, 2003; Patane et al., 2005].[16] In the NE rift 2002 eruption, the dike propagated

for �2 km down-rift without erupting (D1 in Figure 3)[Andronico et al., 2005]. It erupted over a stretch of �4 kmfrom 2500 m asl to �1800 m asl, forming about 20 smallexplosive cones in about 25 hours (D2 in Figure 3). Theeruptive fissure (D2) was composed of three segments(Figure 3). The first strikes N20�–30�E between 2500–2300 m asl and is 1.2 km long. The second segment strikesN30�–45�E, is 0.7 km long, and contains a series of right-

stepping echelon fractures between 2300 and 2190 m asl.The third segment strikes N45�–65�E and extends 1.8 Kmto 1890 m asl. About 107 cubic meters of lava was eruptedfrom these factures in the following 7 days [Andronico etal., 2005]. The NE flank eruption, even if short-lived,produced severe fires in the woods, destroyed many touristinfrastructures and interrupted a main road.

4. Tilt Network

[17] At present, the Mount Etna permanent tilt network(Figure 1) comprises nine biaxial instruments, from AppliedGeomechanics, Inc., installed in shallow boreholes at about3 m depth, at various locations on the mountain, and onelong-base instrument high on the eastern slope [Bonaccorsoet al., 2004]. The radial axis of each tiltmeter is directedtoward the crater, and a positive signal variation meanscrater-up tilting. The borehole tiltmeters have a nominalprecision in the order of �0.1 mrad. The real precision,however, is affected by temperature effects. At shallowdepths of 2–5 m thermoelastic effects introduce measurablenoise on both seasonal and daily timescales [e.g., Wyatt,1988; Dzurisin, 1992]. Temperature changes also affect thesensor response directly [Bonaccorso et al., 2004]. There-fore at shallow depth the real precision of the instrumentscan be considered in the order of 1 mrad. The long-baseinstrument is installed at the volcanological observatory ofPizzi Deneri (PDN), located at 2850 m asl on the north-eastern flank of Mount Etna volcano about 2 km away fromthe summit craters (Figure 3). The PDN observatory hastwo 80 m long orthogonal artificial tunnels, partially under-ground, where the tiltmeter was installed. The instrumenta-tion is positioned along these two tunnels, and is composedof two 80 m long orthogonal tubes filled with mercury.Vertical changes at tube ends are measured by optical lasersensors. This device has little sensitivity to temperaturechanges and provides a real sensitivity of 0.1 mrad. Dataloggers provide 48 samples/day (one sample every 30 min)for the borehole devices and 144 samples/day (one sampleevery 10 min) for the long-base instrument, includingacquisition of the two tilt components, air and groundtemperatures, and instrumental control parameters. The dataare transmitted to Catania via radio link.

5. Data

5.1. Seismicity

[18] A seismic swarm started on 26 October at 2025 UTpreceded and accompanied the formation of the eruptivefissure. During the first 3–4 hours, seismicity took place inthe southern-upper part of the volcano; successively a clearmigration of earthquake location from summit craters areatoward the NE Rift was observed (Figure 4).[19] We obtained distance temporal variations and veloc-

ity (Figure 4a) considering seismic activity recorded fromthe seismic network during the 27 October (Figure 4b). Thecatalogue earthquakes have been relocated with a three-dimensional (3-D) velocity model proposed by Aloisi et al.[2002] using the SimulPS12 code [Thurber, 1993], obtain-ing a significant improvement in location parameters, like-wise as made by Barberi et al. [2004]. We considered onlyearthquakes (58 of 64 events) recorded along the NE rift as

B06404 ALOISI ET AL.: IMAGING COMPOSITE DIKE PROPAGATION

4 of 13

B06404

a function of time and distance from NE Crater (Figure 4b).This analysis shows that the dike in the NE rift propagatedfaster in the upper portions of the rift (1.2 km/hr) than in thelower part, where the dike erupted (0.25 km/hr).[20] The 2002–2003 eruptive kinematics has been inves-

tigated by Barberi et al. [2004] using focal mechanismbased techniques. Generally, magma intrusion is accompa-nied by different types of focal mechanisms, as observed forexample at Hawaii [Thurber and Gripp, 1988] and at MountUsu [Matsumura et al., 1991]. During the Mount Etna2002 eruption strike-slip mechanisms dominated in theintrusive volume, while more compressive mechanismshave characterized the volume situated northeast of eruptivefractures where the intrusion stopped [Barberi et al., 2004](Figure 4b).

5.2. Tilt

[21] Different stations recorded tilt changes starting atdifferent times and also indicated varying durations. From2100 to 2400 UT on 26 October, only the tilt stations closeto the southern fracture (PDN, MDZ, CDV and MSC)showed small changes (Figure 5). Tilt accompanied thedike emplacement and the formation of the south eruptivefissures whose lava fountains were detected at �2300 UT[Andronico et al., 2005]. At 0010 UT on 27 October, largetilt variations started in the summit area and on the northern

flank at PDN, MSC and MMT stations (Figure 5) andaccompanied the first phase of the dike radial propagationfrom NE crater to opening of eruptive fracture at 2500 m[Andronico et al., 2005]. The beginning of dike intrusion isalso marked by a short-lived explosive activity evidenced at0028 UT at the base of the NE Crater [Branca et al., 2003;Del Negro et al., 2004]. Subsequently, the propagation ofthe intrusion along the NE Rift was revealed by the signalvariations at the northeastern stations (MNR and DAM).The tilt changes ended earlier at the stations in the southern(0730 UT) and western (1030 UT) flanks (Figure 5). At2300 UT on 27 October, tilt variations stopped at alltiltmeters.[22] In evaluating the intrusion along the south rift, we

considered the continuous data from southern boreholetiltmeters (CDV, MDZ, MSC and PDN); for the intrusionalong the northeast rift, we considered the continuous datafrom borehole tiltmeters on the northern flank (MMT,MNR, and DAM on Figure 1). These stations showedgreater deformation than those elsewhere on Mount Etna.We also used data from the PDN long-base station, whichshowed changes of about 150 mrad. Moreover, we consid-ered the interpolated signal at the MMT tiltmeter every10 min to use them in combination with PDN data, and thedata recorded at MNR and DAM tiltmeters (one sampleevery 30 min). We did not use MSC and MEG signalsbecause they were affected by strong coseismic noises

Figure 4. Earthquakes along the NE rift as a function oftime and distance (a) along the rift and (b) their location(black circles in map). Events recorded on 27 October butlocated outside the NE rift area and not considered (openblack circles). Open shaded circles show the 26 Octoberearthquakes, and shaded closed circles are the eventsrecorded during the successive 36 hours. Map also reportssome significant focal mechanisms (redrawn from Barberiet al. [2004]) of earthquakes recorded inside (strike slip) andoutside (dip slip) the intrusive volume. Shaded line indicatesthe NE 2002 eruptive fracture.

Figure 5. (a) Tilt signals recorded between 26 and 29October 2002. The shaded area shows the 2100–2400 UT(of the 26 October) time interval, in which only tilt stationsclose to the southern fracture (CDV, MSC, PDN, MDZ)showed (b) changes (zoomed). EC10 recorded smallvariations (1–2 mrad) that are not visible at this scale.SPC was not functioning.

B06404 ALOISI ET AL.: IMAGING COMPOSITE DIKE PROPAGATION

5 of 13

B06404

related to particularly energetic earthquakes [Aloisi et al.,2003]. The long-baseline instrument at PDN was notaffected by shaking [e.g., McHugh and Johnston, 1977;Wyatt, 1988].[23] We did not use the data from the Etna continuous

GPS network because the stations were planned to recorddaily sessions of about 8 hours and therefore the datawere not available during the entire investigation period(Figure 6).[24] The PDN radial tilt component, recorded at 0140 UT

(Figure 3a), constrained the position of the propagatingdike at that time (point ‘‘B’’ in Figure 3b). From theory[Bonaccorso and Davis, 1993], a reversal in tilt in thedirection parallel to dike strike can occur when a dike passesthe point where a tilt station projects onto the plane of thedike (see Figure 7). It is noteworthy that the observed tiltreverse pattern of the PDN radial component (Figure 3a)matches the model prediction (Figure 7). In order to selectthe predicted curve (Figure 7) that best reproduces the tiltsignal during dike propagation (Figure 3a), we searched thespace of parameters for a propagating tabular dislocation[Okada, 1985]. The results were tested with a chi-squareanalysis (see next paragraph).[25] Many observers reported the transition between the

two different phenomenon styles (from dry propagation toexplosive/effusive activity) at about 0200–0230 UT on27 October [Branca et al., 2003]. Considering the positionand time of point B from tilt data and a dike propagationwith speed of ca 1.2 km/hr from the seismicity, we inferred

a transition between upper and lower segment (point ‘‘T’’ inFigure 3b) at about 0210 UT on 27 October. Moreover, thechange in the slope of the tilt record for the PDN tangentialcomponent at 0210 UT (Figure 3a) is another indication of avariation in the dike characteristics (point ‘‘T’’ inFigure 3a). This change in style of dike propagation didnot happen instantaneously and was characterized by astrong explosive activity and occurrence of the most ener-getic earthquakes (five events with 3.3 � Md � 4.2 between0158 and 0250 UT).

6. Modeling

[26] We model the signals recorded from 2100 UT on26 October to 2300 UT on 27 October, assuming a com-posite mechanism that includes a vertical uprising dike inthe upper southern flank and a radially propagating intru-sion in the NE sector. A separate analysis for the two dikeswas performed. We observe that the NE lateral intrusionprovoked the main contribution to the tilt variation on thestations in the northeastern sector. Aloisi et al. [2003]inferred that the NE intrusion was the primary cause ofthe recorded deformation pattern, while the southern dikegave only a minimal contribution, mainly in the S sector. Adike intrusion is likely accompanied by the emptying of amagma chamber [e.g., Segall et al., 2001] which wouldproduce a deflating effect. At Mount Etna, a possible 1–4 km below sea level (bsl) depth magma storage wasinferred during the 1993–2000 recharging period from

Figure 6. Example of a GPS coordinates signal recorded at ETDF station located in the upper southernflank, indicating the limited acquisition session (8 hours every day) configured for the Etna continuousGPS monitoring network at the time of the 2002 eruption. The modeling was obtained using thecontinuous tilt data covering the entire period of the intrusions.

B06404 ALOISI ET AL.: IMAGING COMPOSITE DIKE PROPAGATION

6 of 13

B06404

seismic and deformation studies [Patane et al., 2003;Bonaccorso and Davis, 2004; Bonaccorso et al., 2005].However, in our study we assume that the possible empty-ing of a magma storage represented a second-order effect. Infact, during the first days the eruption was characterized by

small volumes of lava, so we have not considered the smalldeflation effects of the possible 3–6 km depth source(below surface) in spite of the large variations caused bydikes propagation. Moreover, this aspect is also supportedby the fact that the selected tilts (Figure 5) showed radialcomponents up which are not compatible with a primaryeffect of a deflation phase.[27] In order to reproduce the tilt signals during dike

propagation, we searched the space of parameters for apropagating tabular dislocation [Okada, 1985]. The resultswere tested with a chi-square analysis. The Okada model isdescribed by 10 parameters: coordinates of the top centre(X, Y, Z), dimensions of the structure (length and width),

Figure 7. Tilt from a rectangular elastic opening modedislocation at the surface of a half-space. The tilt expected isin the direction parallel to a vertical propagating tensile forfive different depths of the crack. The depths from thebottom to the top are 300, 450, 600, 800, and 1000 m,respectively. The tilt station is 1000 m from the crack, andits projection on the crack is Xp. The crack starts itspropagation 3.5 km from Xp and propagates for 7 km. In theexample the crack depth is assumed constant, thecrack width is assumed equal to depth, and the tensileopening equals 1 m (redrawn from Bonaccorso and Davis[1993]).

Figure 8. Modeled and recorded tilt patterns at thedifferent stations in the upper and southern sectors duringthe first intrusion in the southern flank.

B06404 ALOISI ET AL.: IMAGING COMPOSITE DIKE PROPAGATION

7 of 13

B06404

orientation (strike and dip) and displacements of the dislo-cation (strike slip, dip slip, opening).

6.1. South Modeling

[28] From 2100 to 2400 UT on 26 October, tilt changedon stations close to the south fracture area, exclusively.Moreover, at about 2300 UT on 26 October, Andronico etal. [2005] report violent phreatomagmatic explosions andformation of dense ash and lapilli columns on the highS flank. So, to image the southern flank intrusion, we modelthe 10 min sampled signals recorded from 2100 to 2330 UTon 26 October at the tiltmeters CDV, MDZ, MSC and PDN.A grid search was performed to explore the unknownparameters for a tabular tensile dislocation propagatingvertically in the upper southern flank. We assumed that at2330 UT the dike reached the surface right at the beginningof the eruption. The depth where the intrusion started at2100 UT was included in the grid search. The dike prop-agated vertically from this depth toward the volcano sur-face. We assumed that Z changes linearly with time.Moreover, we suppose that, during the vertical propagationof the dike, all model parameters remained constant in timeexcept for the Z top coordinate. Also with the aim ofverifying the formerly estimated parameters for the southernintrusion by Aloisi et al. [2003], the ranges used in the gridsearch were: X (from 499,000 to 501,000 m; step 500 m); Y(from 4,175,000 to 4,177,000 m; step 500 m); width (from1.0 to 4.0 km; step 0.5 km); strike (from 0� to �20�; step�5�); dip (from 80� to 89�; step 3�); length (from 1 to 3 km;step 1 km); strike slip (fixed to 0 m); dip slip (fixed to 0 m);opening (from 0.8 to 2 m; step 0.4 m). Recorded andexpected tilt are reported in Figure 8. Table 1 reports theestimated model parameters. We inferred the best fit for anear vertical tensile dike intrusion, 1 km long, uprising fromabout 1.3 km depth below the sea level (Table 1) toward thevolcano surface with an opening component of �1 m. Theaverage misfit for each component is very small (Table 2).The inferred model refines the formerly estimated parame-ters [Aloisi et al., 2003].

6.2. Northeast Modeling

[29] In order to image the dike intrusion on the north-eastern flank, we model the signals recorded from 0010 to

2300 UT on 27 October at the tiltmeters PDN, MMT,MNR, DAM. Fissures in the upper part of the system (aboutfrom craters to 2400 m elevation) did not erupt lava,whereas those in the lower part (about from 2400 m to1900 m) did [Branca et al., 2003]. Therefore we performedseparate analyses for these two main parts of the dikeintrusion (Figure 3). Moreover, the projection onto theEarth’s surface of the top coordinates (X, Y) along thestructure plane is known by the fissure position on the ground[Branca et al., 2003]; therefore, we did not consider theseparameters in the grid search. We assumed that all modelparameters remained constant in time during the horizontalpropagation of the dike, except for the X, Y top coordi-nates, moving linearly with time along the ground fissureposition.[30] The upper part of fissure system (see D1 in Figure 3)

was modeled using a tabular dislocation propagating forabout 3.05 km from the summit with incremental steps of235 m (distance between diamonds in Figure 3b) along aline striking N20.7�E. The value of 235 m is calculated bydividing the upper segment length by the number ofsamples (13) recorded at the tilt stations from 0010 to0210 UT on 27 October (one every 10 min). As a prelim-inary step, we explored the unknown parameters with largesteps, also evaluating the formerly estimated parametersfor the NE intrusion by Aloisi et al. [2003]. Subsequently,we performed the following detailed grid search: Z (from0.0 to 1.0 km; step 0.2 km); width (from 1.6 to 3.0 km; step0.2 km); dip (from 75.0� to 105.0�; step 5.0�); strike slip(from 0.0 to 1 m; step 0.25 m); dip slip (from 0.0 to 1 m;step 0.25 m); opening (from 3 to 4 m; step 0.1 m). The valuerange of 3–4 m was inferred for the opening from thethickness of the dike exposed in the wall of a pit crater ofthe eruptive fissure [Branca et al., 2003].[31] The lower part of the dike was modeled using a

tabular dislocation with an incremental length changes of76 m (distance between circles in Figure 3b), propagatingfor about 3.19 km, along three segments striking fromN31.0�E to N59.8�E (see D2 in Figure 3). The value of76 m is calculated by dividing the lower segment length bythe number of samples (42) recorded at the tilt stations from0230 to 2300 UT on 27 October (one every 30 min). In thelower part, a sample rate of 30 min is adequate to reproducethe measured tilt variations, given the slower fracturepropagation with respect to the upper part. The ranges usedin the search were: Z (from 0.0 to 0.5 km; step 0.1 km);width (from 1.6 to 3.0 km; step 0.2 km); dip (from 44.0� to89.0�; step 5.0�); strike slip (from 0 to 1 m; step 0.25 m);

Table 1. Estimated Model Parameters and Uncertainties (South-

ern Intrusion)a

Parameter Value

X, m 500,426 ± 528Y, m 4,176,459 ± 368Strike, deg �8.0 ± 6.7Length, km 1.0 ± 0.2Width, km 3.1 ± 0.7Dip, deg 84.8 ± 3.3Strike slip, m 0.0 (fixed)Dip slip, m 0.0 (fixed)Opening, m 1.0 ± 0.3

aThe reference surface has been assumed at 1.8 km, which is the meanaltitude of the modeled tilt stations. So, the dike intrusion rose toward thevolcano surface (about 1.8 km above sea level.) for 3.1 km (the inferredwidth) starting from 1.3 km below the sea level. The source parameter errorhas been calculated as the standard deviation in the class of all the solutionsthat satisfy the c2 test for our degrees of freedom within the significancelevel of 5%. The dike propagated vertically, at constant velocity, reachingthe volcano surface (Z = 0).

Table 2. Average Misfit for Each Component (Southern Intrusion)

Component Misfit,a mrad

PDN radial 0.2PDN tangential 0.3CDV radial 0.1CDV tangential 0.4MSC radial 0.4MSC tangential 1.7MDZ radial 0.5MDZ tangential 0.1

aThe misfit was considered as the standard deviation between observedand expected data.

B06404 ALOISI ET AL.: IMAGING COMPOSITE DIKE PROPAGATION

8 of 13

B06404

Figure 9. Modeled and recorded tilt patterns at the different stations in the NE flank. The shaded areaindicates the transition between the upper dry segment of the dike and its lower part with explosive andeffusive activity. The change in style of dike propagation did not happen instantaneously, and thetransition was characterized by a strong explosive activity and occurrence of the most energeticearthquakes.

B06404 ALOISI ET AL.: IMAGING COMPOSITE DIKE PROPAGATION

9 of 13

B06404

dip slip (from 1.5 to 3.5 m; step 0.25 m); opening (from 2 to3.5 m; step 0.25 m).[32] Recorded and expected tilt are reported in Figure 9;

Table 3 reports the model parameters. In the upper segment,a shallow near-vertical dislocation with a strong openingcomponent of �3.4 m and a normal dip-slip component of�0.5 m (Table 3) can explain the recorded deformationpattern (Figure 9). The fit is good with an average misfit ofthe different tilt components ranging from about 2 to 12 mradand an average misfit normalized to the maximum recordedsignal change ranging from 5% to 25% (Table 4). Forthe lower segment, a dislocation dipping 44� to the east(Table 3) with �3 m of opening, �3 m of normal dip slip,and �0.8 m of strike slip provide a good fit (Figure 9). Theaverage misfit of the different tilt components ranges fromabout 6 to 11 mrad (Table 4). The average misfit normalizedto the maximum recorded signal change ranges from 8% to30% (Table 4). An exception is represented by the radialcomponent of DAM station, whose expected signal does notreproduce well the recorded one. To estimate the sourceparameter errors (Table 3), we calculated the deformationeffects of all possible solutions in a denser grid andconducted a chi-square test. Moreover, we took into accountthe class of all the solutions that satisfy the chi-square testfor our degrees of freedom within the significance level of5%. The error on the source parameter is the standarddeviation of the parameter values inside this statistical class.These errors indicate how well the parameter is constrainedfor the type of model chosen and for the available recordeddata. We obtained small parameter errors (Table 3). Our

model cannot explain all the details in the observationsbecause we use a simple tabular dislocation model [Okada,1985] in a simple elastic half-space without topography andheterogeneity. Therefore the uncertainty (standard devia-tion) is more than the instrumental error and the remainingmisfit is because the model of a simple rectangular crack inhomogeneous half-space cannot completely explain thedetails in the observations. A complete 3-D sketch map ofthe modeled sources is shown in Figure 10.

7. Discussion

[33] The recorded ground deformation pattern was con-sistent with the action of a complex mechanism of intrusionon both the southern and the northeastern flank of thevolcano. On the southern upper flank a near vertical tensiledike intrusion, 1 km long, uprising from about 1.3 km bsltoward the volcano surface with an opening component of�1 m explains the recorded deformation pattern. We obtainthat the model parameters estimated in this work (Table 1)represent an interesting refinement of the formerly inferredones [Aloisi et al., 2003]. In fact, now the dike position fitsthe observed ground fissures better. A smaller length (1 kmrespect to 1.6 km) and a greater width (3.1 km respect to1.8 km) were obtained; these new values are more realisticbecause the inferred starting depth for the dike intrusion(1.3 km bsl; see Table 1) is coherent with the depth of the1–4 km depth bsl magma storage inferred from seismic anddeformation studies [Patane et al., 2003; Bonaccorso andDavis, 2004; Bonaccorso et al., 2005]. Therefore wehypothesize that the intrusion started due to an overpressurein this magma storage and rose vertically through thevolcanic edifice toward the surface with a velocity of about1.2 km/hr. The southern vertical dike, emplaced in the samepathway of 2001 eruption, gave only a minimal contributionto the observed deformation; instead, the northeastern radialintrusion was the primary cause of the recorded deformationpattern. Previous work modeled the static offsets, recordedbetween 26–27 October before and after the dike propaga-tion, at the permanent GPS and tilt networks [Aloisi et al.,2003]. The authors used a double source dislocation modelthat took into account the static cumulative effect of the twointrusions that occurred on the volcano in the first days oferuption. Their modeling did not consider the evolution ofthe phenomena and obtained a large width (4.6 km) thatrepresents an unusual value for a lateral propagating dikeintrusion.

Table 3. Estimated Model Parameters and Uncertainties (North-

eastern Intrusion)a

Parameter Upper Segment Lower Segment

Z, km 0.3 ± 0.1 0.2 ± 0.1Width, km 2.1 ± 0.4 2.5 ± 0.2Dip, deg 101.2 ± 5.3 44.0 ± 0.5Strike slip, m 0.0 ± 0.03 0.81 ± 0.20Dip slip, m 0.55 ± 0.37 3.16 ± 0.28Opening, m 3.39 ± 0.30 3.31 ± 0.18

aThe reference surface has been assumed at 2100 m, which is the meanaltitude of the modeled tilt stations. The source parameter error has beencalculated as the standard deviation in the class of all the solutions thatsatisfy the c2 test for our degrees of freedom within the significance levelof 5% (see text). The dike propagated horizontally, at constant velocity,along the ground fissure position; therefore the X and Y top coordinates areknown.

Table 4. Average Misfit and Value Normalized to the Maximum Recorded Signal Change for Each Component

(Northeastern Intrusion)

Component

Upper Segment Lower Segment

Misfit, mrad Signal Change, % Misfit, mrad Signal Change, %

PDN radial 7.6 5.5 11.1 8.3PDN tangential 11.5 12.4 11.7 15.7MMT radial 2.9 21.0MMT tangential 1.7 25.0MNR radial 10.2 26.2MNR tangential 6.3 30.0DAM radial 30.2 65.0DAM tangential 7.4 25.4

B06404 ALOISI ET AL.: IMAGING COMPOSITE DIKE PROPAGATION

10 of 13

B06404

[34] The authors suspected that this large width resultedfrom the marked eastward displacement on the east flank[e.g., Froger et al., 2001] which amplified the horizontaldisplacements recorded from the GPS stations located in thelow eastern sector. The scientific debate about the causes ofthe large deformative effects measured on the eastern flankof the edifice is still in progress. Some authors hypothesizethat the gravitational effect plays an important role in theeastward motion of the flank [e.g., Bonforte and Puglisi,2003]. Other authors assume that the large measured dis-placements affecting this highly fractured flank of theedifice may also be explained as a stress redistributionbetween Etnean volcano-tectonic and regional structures,in response to magma storage inflation and/or dike intru-sions [e.g., Patane et al., 2005]. The tilt data that weconsidered in modeling allowed us to follow the effects ofthe propagating dike along the NE rift almost without theeast flank movement contribution because we used tiltstations located in the upper flank. The interaction mecha-nism between magmatic activity (magma storage inflationand/or dike intrusion) and East flank movement is still notclear. Our results suggest that the 2002–2003 eruptionstarted due to an overpressure in the magma storage inferredfrom seismic and deformation studies [Patane et al., 2003;Bonaccorso and Davis, 2004; Bonaccorso et al., 2005]. Theeruption events, especially the NE intrusion as suggested bythe modeling, encouraged the slip on Etna’s eastern flank asshown by the GPS displacement recorded on this flankduring the weeks following the eruption onset. This con-

clusion agrees with the results obtained by Walter et al.[2005] who studied the stress transfer and described amutual interaction between volcanic activity and flankmovement.[35] The NE dike width (2.1–2.5 km), modeled with our

method and smaller than the formerly inferred one [Aloisi etal., 2003], indicates that the radial intrusion propagatedalong a shallower portion of the NE flank. Moreover, thisstudy allows us to describe how the geometry of thedislocated structure changed during the intrusive process(Figure 10). In the upper segment, the opening of a shallownear-vertical crack can explain the tilt signals recordedduring the first hours. This indicates that a nearly pureopening mode displacement accompanied dike propagationin the NE rift. The lower part of fissure system, whichintersects the upper part of the Pernicana fault, is modeledby an opening dislocation with an equally strong normaldip-slip component and with a smaller left lateral strike-slipcomponent. These parameters are consistent with the senseof slip on the Pernicana fault [e.g., Garduno et al., 1997].The modeled width in the lower segment is larger than thevalue obtained for the upper segment, also in agreementwith the spatial distribution of the seismicity (Figure 10). Inthe upper segment, the lateral intrusion started along anearly vertical plane, with a high velocity (about 3 km in2 hours), with the dike propagating radially from thesummit area to the northeast. In the lower segment, theradial intrusion decreased in velocity by roughly an order ofmagnitude (to about 3 km in 20 hours) and explosive-

Figure 10. Three-dimensional sketch map showing the modeled sources obtained from the dynamicinversion of tilt data and seismicity recorded during the intrusion propagation.

B06404 ALOISI ET AL.: IMAGING COMPOSITE DIKE PROPAGATION

11 of 13

B06404

effusive activity started. Propagation took place along aplane with a dip of �44�. A dip of 44� is compatible withthe value estimated by an analytical inversion of Etna GPSdiscrete network data from July 2002 to July 2003(M. Palano, personal communication, 2003). A dippingvalue of 42� was deduced from leveling data recorded from1980 to 1997 along a segment of the Pernicana fault, nearour lower segment [Obrizzo et al., 2001]. We deduce thatthe dike slowed as a result of an interaction with thePernicana fault also evidenced by compressive focal mech-anism (Figure 4b). In this frame, the interaction with thispreexisting structure might explain the larger dislocation inthe lower part of the rift. The dike reactivated the Pernicanafault and the large recorded deformation pattern of the eastflank occurred in response to the intrusive event. So, stresschange associated with dike intrusions is one of the basicmechanisms responsible for the kinematics of the east flank[see also Obrizzo et al., 2001; Walter et al., 2005].[36] Our analysis of the 2002 eruption confirms that tilt

can play a very useful role in detecting ground variationsrelated to magma intrusions and eruptive fissure propaga-tion. Moreover, our analysis that accounts for dike propa-gation allows us to understand the intrusion process betterthan previous static inversions [Aloisi et al., 2003]. For thefirst time in the historical period and in the modeling ofrecent eruptions, the results show that the 2002 eruption wascharacterized by the coexistence of the two main differentclasses of intrusion at Mount Etna, represented by a verticalintrusion separated from central conduits in the S flank andby a shallow horizontal intrusion propagating from thecrater area toward the NE flank. Moreover, again for thefirst time at Mount Etna volcano, different continuous datasets (seismicity for propagation velocity and tilt data fordike parameters) have been used together to infer how adike propagates on the NE flank of Mount Etna.

[37] Acknowledgments. We thank Stephen Martel (Department ofGeology and Geophysics, University of Hawaii) and another anonymousreferee for their reviews leading to substantial improvements of a firstversion of the manuscript. We thank the Associate Editor, James Gardner,and two anonymous reviewers for their constructive comments. We aregrateful to many INGV colleagues for the productive discussions on thisstudy, particularly S. Branca, G. Puglisi, M. Mattia, D. Palano, andD. Andronico. We are indebted to the technical staff of the GroundDeformation Unit of INGV-CT, who ensure the regular working of themonitoring networks. We also thank S. Conway for correcting andimproving the English of this paper.

ReferencesAloisi, M., O. Cocina, G. Neri, B. Orecchio, and E. Privitera (2002), Seis-mic tomography of the crust underneath the Etna volcano, Sicily, Phys.Earth Planet. Inter., 134, 139–155.

Aloisi, M., A. Bonaccorso, S. Gambino, M. Mattia, and G. Puglisi (2003),Etna 2002 eruption imaged from continuous tilt and GPS data, Geophys.Res. Lett., 30(23), 2214, doi:10.1029/2003GL018896.

Andronico, D., et al. (2005), A multi-disciplinary study of the 2002–03Etna eruption: Insights into a complex plumbing system, Bull. Volcanol.,67, 314–330.

Aoki, Y., P. Segall, T. Kato, P. Cervelli, and S. Shimada (1999), Imagingmagma transport during the 1997 seismic swarm off the Izu Peninsula,Japan, Science, 286, 927–930.

Barberi, G., O. Cocina, V. Maiolino, C. Musumeci, and E. Privitera (2004),Insight into Mt. Etna (Italy) kinematics during the 2002–2003 eruptionas inferred from seismic stress and strain tensors, Geophys. Res. Lett., 31,L21614, doi:10.1029/2004GL020918.

Battaglia, J., and P. Bachelery (2003), Dynamic dyke propagationdeduced from tilt variations preceding the March 9, 1998, eruption ofthe Piton de la Fournaise volcano, J. Volcanol. Geotherm. Res., 120,289–310.

Bonaccorso, A. (1998), Evidence of a dyke-sheet intrusion at Strombolivolcano inferred through continuous tilt, Geophys. Res. Lett., 25, 4225–4228.

Bonaccorso, A. (1999), The March 1981 Mt. Etna eruption inferred throughground deformation modelling, Phys. Earth Planet. Inter., 112, 125–136.

Bonaccorso, A., and P. M. Davis (1993), Dislocation modelling of the 1989dike intrusion into the flank of Mount Etna, Sicily, J. Geophys. Res., 98,4261–4268.

Bonaccorso, A., and P. M. Davis (2004), Modeling of ground deformationassociated with recent lateral eruptions: Mechanics of magma ascent andintermediate storage at Mt. Etna, in Mt. Etna: Volcano Laboratory, Geo-phys. Monogr. Ser., vol. 143, edited by A. Bonaccorso et al., pp. 293–306, AGU, Washington, D. C.

Bonaccorso, A., M. Aloisi, and M. Mattia (2002), Dike emplacement fore-running the Etna July 2001 eruption modeled through continuous tilt andGPS data, Geophys. Res. Lett. , 29(13), 1624, doi:10.1029/2001GL014397.

Bonaccorso, A., O. Campisi, G. Falzone, and S. Gambino (2004), Contin-uous tilt monitoring: A lesson from 20 years experience at Mt. Etna, inMt. Etna: Volcano Laboratory, Geophys. Monogr. Ser., vol. 143, editedby A. Bonaccorso et al., pp. 307–320, AGU, Washington, D. C.

Bonaccorso, A., S. Cianetti, C. Giunchi, E. Transatti, M. Bonafede, andE. Boschi (2005), Analytical and 3D numerical modeling of Mt. Etna(Italy) volcano inflation, Geophys. J. Int., 163, 852–862, doi:10.1111/j.1365-246X.2005.02777.x.

Bonforte, A., and G. Puglisi (2003), Magma uprising and flank dynamicson Mount Etna volcano, studied using GPS data (1994–1995), J. Geo-phys. Res., 108(B3), 2153, doi:10.1029/2002JB001845.

Branca, S., D. Carbone, and F. Greco (2003), Intrusive mechanism of the2002 NE-Rift eruption at Mt. Etna (Italy) inferred through continuousmicrogravity data and volcanological evidences, Geophys. Res. Lett.,30(20), 2077, doi:10.1029/2003GL018250.

Cervelli, P., P. Segall, F. Amelung, H. Garbeil, C. Meertens, S. Owen,A. Miklius, and M. Lisowski (2002), The 12 September 1999 UpperEast Rift Zone dike intrusion at Kilauea Volcano, Hawaii, J. Geophys.Res., 107(B7), 2150, doi:10.1029/2001JB000602.

Del Negro, C., G. Correnti, R. Napoli, and A. Vicari (2004), Volcanomag-netic changes accompanying the onset of the 2002–2003 eruption ofMt. Etna (Italy), Earth Planet. Sci. Lett., 229, 1–14.

Dzurisin, D. (1992), Electronic tiltmeters for volcano monitoring: Lessonsfrom Mount St. Helens, in Monitoring Volcanoes: Techniques and Stra-tegies, edited by J. W. Ewert and D. A. Swanson, U.S. Geol. Surv. Bull.,1966, 69–83.

Froger, J. L., O. Merle, and P. Briole (2001), Active spreading and regionalextension at Mount Etna imaged by SAR interferometry, Earth Planet.Sci. Lett., 187, 245–258.

Fukushima, Y., V. Cayol, and P. Durand (2005), Finding realistic dikemodels from interferometric synthetic aperture radar data: The February2000 eruption at Piton de la Fournaise, J. Geophys. Res., 110, B03206,doi:10.1029/2004JB003268.

Garduno, V. H., M. Neri, G. Pasquare, A. Borgia, and A. Tibaldi (1997),Geology of the NE-Rift of Mount Etna (Sicily, Italy), Acta Vulcanol., 9,91–100.

Matsumura, S., T. Ohkubo, and I. Masajiro (1991), Seismic swarm activityin and around the Izu Peninsula preceding the volcanic eruption ofJuly 13, 1989, J. Phys. Earth, 39, 93–106.

McHugh, S., and M. J. S. Johnston (1977), An analysis of coseismic tiltchanges from an array in central California, J. Geophys. Res., 82, 5692–5698.

Neri, M., V. Acocella, and B. Behncke (2004), The role of the Pernicanafault system in the spreading of the Mt. Etna (Italy) during the 2002–2003 eruption, Bull. Volcanol., 66, 417–430.

Obrizzo, F., F. Pingue, C. Troise, and G. De Natale (2001), Coseismicdisplacements and creeping along the Pernicana fault (Etna, Italy) inthe last 17 years: A detailed study of a tectonic structure on a volcano,J. Volcanol. Geotherm. Res., 109, 109–131.

Okada, Y. (1985), Surface deformation due to shear and tensile faults inhalf-space, Bull. Seismol. Soc. Am., 75, 1135–1154.

Patane, D., C. Chiarabba, P. De Gori, and A. Bonaccorso (2003), Magmaascent and the pressurization of Mt. Etna’s volcanic system, Science, 299,2061–2063.

Patane, D., M. Mattia, and M. Aloisi (2005), Shallow intrusive processesduring 2002–2004 and current volcanic activity on Mt. Etna, Geophys.Res. Lett., 32, L06302, doi:10.1029/2004GL021773.

Segall, P., P. Cervelli, S. Owen, M. Lisowski, and A. Miklius (2001),Constraints on dike propagation from continuous GPS measurements,J. Geophys. Res., 106, 19,301–19,317.

Thurber, C. H. (1993), Local earthquake tomography: Velocity and VP/VS-theory, in Seismic Tomography: Theory and Practice, edited by H. M.Iyer and K. Hirahara, pp. 563–583, CRC Press, Boca Raton, Fla.

B06404 ALOISI ET AL.: IMAGING COMPOSITE DIKE PROPAGATION

12 of 13

B06404

Thurber, C. H., and A. E. Gripp (1988), Flexure and seismicity beneath thesouth flank of Kilauea volcano and tectonic implications, J. Geophys.Res., 93, 4271–4278.

Walter, T. R., V. Acocella, M. Neri, and F. Amelung (2005), Feedbackprocesses between magmatic events and flank movement at Mount Etna(Italy) during the 2002–2003 eruption, J. Geophys. Res., 110, B10205,doi:10.1029/2005JB003688.

Wyatt, F. K. (1988), Measurements of coseismic deformation in southernCalifornia: 1972–1982, J. Geophys. Res., 93, 7923–7942.

Yamaota, K., M. Kawamura, F. Kimata, N. Fujii, and T. Kudo (2005), Dikeintrusion associated with the 2000 eruption of Miyakejima Volcano,Japan, Bull. Volcanol., 67, 231–242.

�����������������������M. Aloisi, A. Bonaccorso, and S. Gambino, Istituto Nazionale di

Geofisica e Vulcanologia, Sezione di Catania, Piazza Roma 2, I-95123Catania, Italy. (aloisi@ct.ingv.it)

B06404 ALOISI ET AL.: IMAGING COMPOSITE DIKE PROPAGATION

13 of 13

B06404