We explore volcanotectonic earthquake activity during the early stages of volcanic activity (1995–96) at Soufrière Hills volcano. We focus on several zones of temporally-confined seismic activity that comprise assemblages of small scale structural elements, >2–4 km distant from the volcano, at depths 2–4 km below sea level, and consider seismicity in relation to regional tectonics and heterogeneity of stiffness and strength as revealed by the SEA-CALIPSO tomography experiment. The clustered seismicity and relatively aseismic zones are interpreted to reflect a broad weakened tectonic zone of ESE trend that crosses Montserrat, and the ascent of a magmatic dike of NNE trend, which altered the stress distribution to promote localized fault movements and caused localized dilatation with changes in pore-fluid pressures, to either weaken or strengthen the rock mass depending on the local polarity of strain.
 The andesitic Soufrière Hills volcano (SHV) has been erupting intermittently since the phreatic activity of 18 Jul 1995, with a brief lava extrusion on 25 Sep 1995, and quasi-continuous magmatic extrusion commencing ∼15 Nov 1995. The structural setting on Montserrat is complex [Harford et al., 2002], the architecture of the magmatic system under SHV is imprecisely resolved [Voight et al., 2010], and aspects of seismicity have been conflicting and controversial [Aspinall et al., 1998; Gardner and White, 2002; Roman et al., 2008]. The early seismicity, recorded over a broad region in south Montserrat, gives the seismic expression of magma ascent after decades of quiescence, but a puzzling aspect is the odd localized distribution of clustered seismicity. We seek to improve understanding of this seismicity in relation to structures, heterogeneous media, and magma transport dynamics with associated force and fluid pore-pressure redistributions.
2. Tectonic and Magmatic Setting
 The dominant geologic structure is related to a ESE trending lineament (Figure 1), inferred as the Belham Valley Fault (BVF) [Harford et al., 2002], which has affected both onshore and offshore geomorphology. The BVF crosses the island and bounds uplifted and tilted fault blocks comprising Garibaldi Hill (GH), St. Georges Hill (SGH), and Roche's Bluff (RB) with 1-Ma old marine sediments (Figure 1). Related faults with WNW strike and some cross-faults bound the sides of GH and SGH (Figure 1). Thus the system of related faults is roughly about 5 km broad, and if extended south-eastward, passes through SHV where details are obscured by the volcanic deposits.
 Offshore, south of Montserrat, the Bouillante-Montserrat half-graben (BMF) extends SE towards Guadeloupe [Feuillet et al., 2001, 2002] and records transtension with sinistral slip. The regional shallow-plate stress field comprises a roughly westerly-directed maximum compression and a northerly minimum compression, compatible with sinistral slip on BMF, and rupture on the 16 March 1985, MW 6.3, Nevis earthquake [Feuillet et al., 2002]. Feuillet et al.  proposed that the andesitic domes on Montserrat were aligned within a pull-apart step between two oblique sinistral faults (BVF and BMF), and assigned volcanism to the extensional pull-apart. But the matter is more complex because a right-stepping sinistral system forms a contractional, restraining stepover, not a pull-apart.
 Evolving stress fields within the geometrically complex region of step-over are heterogeneous in contrast to the regional stress field [Cunningham and Mann, 2007], and localized extensional zones can develop within them. On Montserrat the local crustal stress states are also influenced by the evolving magma system. The magma reservoir top is at ∼5 km depth from mineral assemblages (data by Voight et al. ) and SEA-CALIPSO tomography [Shalev et al., 2010; Paulatto et al., 2010]. Dikes transport magma from the pressurized reservoir to higher crustal levels, and their orientations are controlled by the local axes of minimum compression. Feeder dikes with NW to NNW trend have been interpreted from deformational data in 1995–96 [Mattioli et al., 1998] and 1997 [Hautmann et al., 2009], but Roman et al.  proposed a dike of NE orientation that was stable throughout the eruption, so the question is in dispute. Strong evidence for dikes of both WNW and ENE trends exists for eruptive phases in 2004 and 2008 [Linde et al., 2010; B. Voight et al., Explosion dynamics from strainmeter observations, Soufrière Hills Volcano, Montserrat, W.I.: 2008–2009, submitted to Geophysical Research Letters, 2010].
3. Seismic Data and Processing
 Stations used in our analysis are shown on Figure 1 and their locations listed in Table S1 of the auxiliary material. All seismic events were inspected by MVO and classified by type [Miller et al., 1998]. We examine events with sharp onsets, which includes volcanotectonic (VT) events and “impulsive” hybrids as defined by Miller et al. , but our emphasis is on VTs. Data processing and reduction methods are described in the auxiliary material. Our approach is based on standard location methods using station corrections, and differs from earlier analyses of catalog time picks primarily in the use of a velocity model with slower speeds appropriate for the southern part of the island [Rowe et al., 2004; Shalev et al., 2010], and the selection criteria used to identify higher quality events. We include all events from 28 Jul 1995 to 31 Oct 1996 in the location procedure, and use catalog time picks to fill previously reported waveform data gaps [Roman et al., 2008]. All source depths are referenced to sea-level.
 The representative focal mechanisms we present are for the largest magnitude events (MC ≥ 2.5) that had no inconsistent polarities and provided reasonable control on the faulting geometry. From 1534 VT earthquakes with A+ quality locations we derived 615 high quality fault-plane solutions. Of these, 394 (64%) have strike-slip or oblique strike-slip mechanisms (p-axis plunge < 45°). We use rose diagrams of p-axis azimuths (Figure S7) as a proxy to interpret local compression stresses [cf. Roman et al., 2008].
4. Results and Interpretation
 We discuss four zones/sequences in detail (see Figures 1, 2, and S1–S5) and describe the geometry and timing of the located events. The SGH-Z and NE-Z zones were previously noted by Aspinall et al. , and SGH was clearly active in previous volcano-seismic crises [Powell, 1938; Shepherd et al., 1971]. A third zone at Windy Hill (WH-Z) was identified by Gardner and White . We introduce a fourth zone that extends out from SHV (WNW-Z). We analyze these structures using the hypocenters from our study, crustal velocity data and our most reliable focal mechanisms. Our insights and interpretations differ from earlier views.
 Zone NE-Z extends to the NE near SHV (Figures 1, 2, and S2), and activity here began on 5 Aug 1995 at ∼3 km from the vent at depths 3–4 km, where several larger (MC ∼ 3.0) VT events form tight clusters which might represent small individual faults (Figures 2b and S2). Solutions for two large events (1,2) that occurred a few hours apart on 5 Aug (Figure 2b) both indicate dextral strike-slip motion suggestive of WSW–ENE compression, compatible with the regional stress field. Later that day events located further NE at greater depths (>5 km, low-quality locations). On 6 Aug, several events occurred at 2–3 km depth near the N and SW sides of the SHV vent (Figures 1 and S2). Overall, zone NE-Z is ∼4–5 km long, excluding offshore events. Beneath SHV the zone is at ∼2 km depth and declines toward the NE (Figure S2b). The clustered subsets of larger magnitude events may reflect separate structures with different strike. In support of this is a surface fault in the Long Ground area (Figure 1), revealed in late-1998 as a series of cracks of 068° trend on a tarmac-covered road. The data are thus consistent with a fault zone at this location, antithetic with respect to BVF. Mechanisms 1, 2 suggest its sense of motion (Figure 2b), and this is supported by the dominant ENE p-axis trends in NE-Z for 5 Aug (about two-thirds of events show strike-slip or oblique strike-slip). The northerly p-axis trends for 6 Aug correspond to the seismic cluster north of the vent, and suggest a different slip direction and likely a different structure. Thus we are unable to verify the p-axes results of Roman et al. [2008, Figure 8].
 Zone SGH-Z is located ∼3.5 km WNW of SHV, with activity on 12–13 Aug at depths 3 to 4 km (Figures 1, 2, and S3). Mechanisms for relatively large events (3,4) on 12 August (18.30h, 19.46h) in the mid- cluster suggest normal faulting with ∼WNW–ESE extension (Figure 2b). However two large events (5,6) occurring on the peripherals of this zone slightly later (20.18h, 20.21h) suggest normal faulting with a NE–SW sense of extension. Aspinall et al.  offered two “preliminary” mechanisms for this group, a strike-slip and a thrust mechanism that suggested NW–SE compression. We obtained a similar solution for their large thrust event on 13 Aug (not shown here due to our quality cut off). Our p-axis trends for strike-slip or oblique strike-slip events support NW-compression, as do results of Roman et al. , but only 36% of our solutions indicated this fault type. The variability in focal mechanisms makes an interpretation for this zone more complex than suggested by Aspinall et al.  or Roman et al. . Field outcrops are sparse but some show faults of NE orientations with normal slip sense, compatible with mechanisms 3, 4. Variations in the prevailing local stress conditions both temporally and spatially are likely, given the rapidly evolving magmatic system.
 Zone WH-Z (Figures 1, 2, and S4) is located 2–3 km NNW of SHV, with activity 8–10 Sep 1995. The largest events form a south-plunging trend at 2–4 km depth. Hypocenter locations suggest a branching structure with WNW trend (Figure 1). Most events have focal mechanisms consistent with a N–S fault (Figure 2b). The events shown (7,8) occurred an hour apart on 8 Sep (22.09h, 23.10h) and suggest a NE–SW trending tension axis (concordant with the peripheral events at SGH), with oblique motion on a N–S plane (sinistral, E-block down). We examined the waveforms and verified that P polarities of these nearby events changed at nearby stations. The p-axes for other events (about 70% strike-slip or oblique strike-slip) are directed WNW (Figure S7) and support the WH focal mechanisms shown. In comparison we find similarly oriented p-axes for the SHV cluster (radius <1 km), for events prior to 15 Nov (about 65% strike-slip or oblique strike-slip).
 On 18–22 Nov 1995, just as surface dome growth initiated, an intriguing zone of seismicity extended WNW from SHV (WNW-Z in Figure 2a inset; Figure S5). The sequence occurred hours after activity concentrated beneath the SHV vent, which suggests that the station constellation was sufficient to locate seismicity in this region (see Animation S1). We do not observe a systematic migration of seismicity. The sequence began with shallow diffuse seismicity in a region ∼2 km long, at ∼1–2 km depth adjacent to SHV. On 19 Nov, activity extended ∼4 km towards SGH. On 20 Nov, activity was concentrated <2 km from SHV, with scattered hypocenters to 4 km. Most events are small and of B, C location quality. By 22 Nov seismicity had reduced but two larger (A+ quality) events occurred, one at each end of an extended region that now crossed the SHV, but >2 km deep. The event near SGH showed dextral strike-slip movement with WNW- extension (Figure 2a inset). The deepest events in this sequence extend to ∼3 km. The p-axes for the SHV cluster for VT events prior to 15 Nov are also oriented WNW (Figure S7).
5. Heterogeneous Media, Strain, and Fluid Pressure
 We interpret the eruption-onset VT seismicity as consistent with rock stress perturbations induced by ascending magma. Such perturbations are superposed on an ambient stress field, and the relative strengths and orientations of the component stress states will vary spatially, and temporally, to influence patterns of resulting VT seismicity [Roman and Heron, 2007]. To assess the role of spatial variations in mechanical properties, we compare our locations with the 3D tomography model from Shalev et al.  in Figure 2, which shows the A+ quality, MC ≥ 2 events for depths <2.75 km plotted overtop of P-wave velocity contours on a map-view slice at depth 2.5 km b.s.l., and the same quality/magnitude events for events > 2.75 km on a velocity slice at 3 km depth. P-wave velocity correlates with both the modulus and strength of the rockmass. The first-order feature is the large region of relatively low P-wave speeds along the SW region of the island (Figures 1 and 2), which we interpret as volcaniclastic flank deposits with geothermal alteration. SHV volcano sits in a saddle separating regions of relatively high P-wave speeds centered beneath Centre Hills (CH), and South Soufrière Hills. Although resolution is limited, the saddle opens in the direction of the distal NE-Z zone, suggesting a relatively weak, faulted rockmass; this zone includes the first structure activated seismically in the time period for which we have relatively good event locations. The deeper seismicity (Figure 2b) concentrates mainly along regions of higher velocity gradient (gradients may correlate with structure), although WH-Z occurs within a N–S prong of high-Vp extending south from CH, likely structurally controlled. Vp gradients south of CH, extending from GH to SHV, are approximately parallel to BVF and probably reflect this structure.
 The shallow (0–2 km b.s.l. (Figure S6)) hybrid activity under the SHV vent is concentrated within the faster material portion of the saddle (Figure 2a). The hybrids at SHV correlate with magma transport, and their crude northerly trend is perhaps suggestive of a dike. In contrast the small VT events in zone WNW-Z cross SHV and extend out into the embayment of relatively low P-wave speed and by correlation, weak rock (Figure 2a inset). The WNW-Z zone could reflect a zone of structural weakness responding to forced magma ascent in the uppermost part of the conduit system; this view is supported by the VT p-axis trends for <15 Nov. What controlled the enigmatic discrete-cluster seismicity at SGH, prominent not only in the current eruption but also in volcano-seismic crises in 1933–37 and 1966–67 [Powell, 1938; Shepherd et al., 1971]? Why was deep seismicity not continuous between SGH and SHV, and why did similar clusters not form in other sectors encircling the volcano on S and E? These are puzzling issues, and toward their resolution we propose that the isolated seismicity has been controlled by a combination of factors, including 1) the pre-existing structures largely related to the BVF; 2) zones of relative weak rockmass as indicated by tomography; 3) ambient stresses that reflect regional tectonics, topography, and residual stresses influenced by stress history; and 4) stress perturbations from an evolving magma conduit system, with very high pressures and deformations occurring prior to eruption. The localization of the clusters NW of the vent could reflect factors 1 and 2 above, in concert with dike-induced changes in hydraulic pressures, and the consequential decrease of frictional resistance on faults [Elsworth and Voight, 1995]. Evidence for saturated fractured media around SGH includes surface hot springs and hot water in boreholes [Chiodini et al., 1996], and a recent MT study indicates good correlation between low velocity zones and low resistivity at 1–4 km depth [Shalev et al., 2010].
 Water level or pressure head changes in relation to intrusions have been observed previously, e.g., artesian pressures in a 1565-m drillhole during the 1973 Heimaey fissure eruption, 1.7 km normal to the feeder dike [Bjornsson et al., 1977]; excess fluid pressures in geothermal wells near the Krafla, Iceland, fissure eruptions in 1977 and 1978 [Elsworth and Voight, 1992]; water level rises of 37 m in a 370-m well 2 km from the summit of Usu volcano in 1977–78 [Watanabe, 1983]. A theory of dike intrusion in a saturated porous solid is developed by Elsworth and Voight . Fluid pressure changes are proportional to the product of the volumetric strain, the undrained bulk modulus, and the two Skempton coefficients, such that with contractional strain, transient suprahydrostatic fluid pressures are generally produced. The AIRS borehole in Montserrat, ∼1 km from SGH, revealed rapidly-developed (∼several minutes) shallow volumetric strains >0.1 microstrain in 2004 and 2008 [Linde et al., 2010; Voight et al., submitted manuscript, 2010], reflecting rapid pulses of dike pressurization. Their models show a broad elliptic field of large contractive strain (at dike level, ∼5 microstrain for 10 MPa driving pressure, Young's modulus 5 GPa) perpendicular to dike strike, developed beyond a depth-dependent nodal distance of >1 km. Equivalent water level head changes are of order ∼2 m per microstrain in fractured rocks [Roeloffs and Linde, 2007]. In the near field, and farther out along dike strike, the strains are expansive, and in these regions induced pore-water pressures are negative and the rockmass strengthened. Thus the role of magma intrusions is twofold, to alter the distribution of stresses tending to cause fault movements [e.g., Roman, 2005], and to cause changes in pore fluid pressures that can either weaken, or strengthen, the rockmass in selective sectors, depending on the polarity of strain [Elsworth and Voight, 1995].
 These considerations, in combination with geologic heterogeneity as revealed by tomography, help to understand the clustering of earthquakes around SHV. A field of contractive strain overprinting SGH and WH is consistent with an intruding dike of roughly NNE trend; this is also roughly compatible with the northerly-aligned hybrids of Figure 2a and our p-axis trends for faulting under SGH, WH, and SHV (Figure S7). Thus we roughly support Roman et al.  on a dike orientation for 1995, but not their conclusion that this orientation remained stable [cf. Linde et al., 2010; Voight et al. submitted manuscript, 2010]. The lack of continuous deep seismicity in 1995 between SGH and the SHV is explained by a change in dilatation and pore pressure polarity (contraction at SGH, expansion near the vent). The lack of a similar cluster to the SSE is explained by stiffer, stronger rockmass (tomography) and reduced volumetric strains. The general lack of seismicity >4 km distance is explained by weaker stress perturbation, and reduced volumetric strains. The faulting in the outer NE-Z is compatible with control by the regional stress field.
 We thank NSF for project support, and V. Bass, A. Miller, S. Loughlin, R. Luckett, W. Aspinall for assistance with data.