Over 3000 local and regional earthquakes were recorded by a compact network of eight ocean bottom seismographs (500 m instrument spacing) from August 1996 to January 1997 in Middle Valley, a sediment-covered rift valley on the northern Juan de Fuca Ridge. Thirteen swarms of small-magnitude microearthquakes (−1.2 < Mw < 0.2) were detected beneath Dead Dog vent field, a major hydrothermal area in Middle Valley with exit fluid temperatures near 270°C. High precision relative positions for 304 events within swarms were determined using waveform cross-correlation techniques. The events were relocated into small, spatially distinct clusters. The intensity of the swarms is correlated with high heat flow with the largest swarm positioned 1.3 km beneath the Dead Dog vents. Smaller clusters of earthquakes are located up to hundreds of meters outside the vent field. The results suggest that the observed seismicity in the Dead Dog region is triggered by thermal strain (contraction) in the hydrothermal reaction zone as fluids extract heat from hot basement rock. Microearthquake swarms appear to be concentrated in regions where faulting has promoted seawater penetration through the sediment layer, cooling the crust, and yielding larger strain rates than those produced by seafloor spreading.
 Hydrothermal processes at ridge crests have been studied with a broad range of techniques because they play a controlling role in the thermal structure of the ridge, create a fascinating and unique biology, and have formed many of the important commercial ore deposits now on land. Hydrothermal circulation can penetrate several kilometers into oceanic crust, but our understanding of these systems is largely based on samples and specimens obtained from the seafloor. As a result, hydrothermal flow patterns within the crust are poorly understood.
 Thermal gradients within newly formed oceanic crust are among the largest naturally occurring on Earth, and reflect the close physical proximity between 0°C seawater and ∼1200°C basaltic magma. Simple calculations suggest that thermal strain generated by convective heat flow within these systems is several orders of magnitude greater than tectonic strain [Sohn et al., 1999]. The brittle deformation can occur when the rock is cooled below the rigidus (∼550°C), and when stresses exceed the strength of the host rock [Lister, 1974, 1983]. In this view, the hydrothermal reaction zone is a seismogenic zone for microearthquakes associated with contraction from thermal strain.
 A close association between fluid flow and microseismicity is commonly observed in subaerial geothermal systems. Seismicity at the Krafla [Arnott and Foulger, 1994] and Hengill-Grensdalur areas [Julian et al., 1997; Foulger, 1988a] in Iceland is triggered by thermal processes as fluids cool hot rock emplaced by volcanic activity [Foulger and Long, 1984; Foulger, 1988b; Arnott and Foulger, 1994]. Seismograph networks also record high levels of artificial seismicity associated with the injection of cold water into hot rock at the Geysers geothermal field, California [Stark, 1992]. Correlations between microseismicity and fluid injection times at the Geysers field indicate that the majority of earthquakes near the injection sites are induced by thermoelastic stresses (A. Mossop and P. Segall, Induced seismicity at the Geysers geothermal field, correlations and interpretations, manuscript in preparation). In these land-based studies, microearthquake hypocenters cluster in areas where faults form fluid pathways in the crust, providing a unique look at the plumbing networks of the geothermal systems.
 Midocean ridge crest ocean bottom seismometer (OBS) deployments near high-temperature vent fields also detect high levels of seismic activity [e.g., Riedesel et al., 1982; McClain et al., 1993; Sohn et al., 1995, 1999; Wilcock et al., 1999]. However, only in the last few years have marine seismograph surveys featured networks dense enough and deployments long enough to obtain accurate hypocenter locations and image subsurface structures comparable to subaerial geothermal studies [e.g., Sohn et al., 1999; Wilcock et al., 1999]. A network of nine seismographs on the East Pacific Rise (EPR) at 9°50′N monitored a hydrothermal cracking event consisting of a 3-hour swarm of 162 events. The swarm, located as a thin vertical column over the margin of the axial magma chamber, triggered a 7°C increase in vent exit fluid temperatures [Fornari et al., 1998; Sohn et al., 1999]. On the Endeavour segment of the Juan de Fuca Ridge, 15 seismometers detected thousands of local microearthquakes at 2–4 km depth beneath the Main, High Rise, Salty Dog, and Mothra vent fields [Wilcock et al., 1999].
 To map regions of hydrothermal cracking using seismicity, we deployed a small array of eight OBSs near Dead Dog vent field, a major area of active venting in Middle Valley, Juan de Fuca Ridge. To locate and analyze small microearthquakes with shallow hypocenters, the OBS array featured interelement spacing of a few hundred meters to ensure that seismic phases from each event would be recorded at every instrument. The array recorded 1220 local microearthquakes within 5 km of the vents over a 145-day period. Hypocenter locations with standard errors on the order of a few hundred meters were obtained for 480 of these events using a grid search algorithm, with travel times computed throughout a 3D model of the area [Golden, 2000]. Waveform cross-correlation and relative relocation techniques were used to exploit waveshape similarity and constrain the relative hypocenter locations to within a few tens of meters. The microearthquakes are located beneath the Dead Dog vents, and are correlated with a broad area of anomalously high seafloor heat flow. The results from this study link the observed seismicity with hydrothermal processes, and we use this information to develop a model of the deep hydrothermal circulation and thermal structure beneath Dead Dog vent field.
2. Geological Setting
 Middle Valley is a sediment-covered rift graben on the northernmost portion of the Juan de Fuca Ridge (Figure 1, inset), a rise crest with a full spreading rate of 6 cm/yr situated approximately 300 km off the northwest coast of the United States. Middle Valley was the primary spreading center in the region until ∼10 ka, when the rate of tectonic extension began to decrease, and magmatic activity ceased. Spreading activity on this part of the northern Juan de Fuca Ridge is presently accommodated at the adjacent West Valley [Davis and Villinger, 1992]. Normal faults mark the eastern and western boundaries of Middle Valley (Figure 1), and a prominent west-facing normal fault is located near the valley center, separating the central rift graben from a shallow basement bench to the east [Davis and Villinger, 1992]. This axial valley morphology is characteristic of intermediate and slow spreading centers [e.g., Macdonald and Luyendyk, 1977; Karsten et al., 1986; Kong et al., 1988].
 Middle Valley is bounded to the north by the Sovanco fracture and the Nootka fault zones (Figure 1, inset), marking a tectonically complex, seismically active triple junction between the Pacific, Explorer, and Juan de Fuca plates. The split between the Explorer and Juan de Fuca plates is perpetuated as young, thin oceanic lithosphere is subducted at different rates along the Cascadia margin only a few hundred kilometers east of the ridge system [Riddihough, 1977]. Faulting in this unstable region yields high levels of seismic activity. Land-based seismograph stations operated by the Geological Survey of Canada (GSC) and marine OBS surveys of the area have typically recorded moderate-magnitude earthquakes (M > 3 for GSC stations, and 1 < M < 3 for OBSs) along the Sovanco and Nootka fault zones [Wahlstrom and Rogers, 1991; Hyndman and Rogers, 1981]. The seismicity distributions depict both the Sovanco and Nootka boundaries as diffuse fault zones rather than single transform faults.
 Heat and fluid flux at Middle Valley are controlled by a thick layer (up to 2 km) of turbidite sediment that blankets the volcanic basement [Davis and Villinger, 1992]. The turbidite layer is characterized by low permeability and thermal conductivity, and it effectively insulates the hot crustal rock from the water column. Hydrothermal recharge may be focused into areas where permeable basement is exposed at the seafloor, or regions where faults provide permeable pathways through the sediment [Davis and Fisher, 1994; Stein and Fisher, 2001].
 Active hydrothermal discharge in Middle Valley occurs at Bent Hill and Dead Dog, two distinct sites located along the shallow basement bench to the east of the central rift graben. In this region, the crust beneath the turbidites is composed of an interlayered sequence of basaltic sills and sediment, unlike the entirely extrusive layer of a standard ridge crest structure [Davis and Villinger, 1992]. Discharge at Bent Hill has created a massive sulfide mound that is 120 m thick, indicating that fluid temperatures once reached an excess of 400°C. Current exit fluid temperatures at this site are roughly 265°C [Davis and Fisher, 1994]. Dead Dog vent field is located ∼4 km west of Bent Hill, and is delineated by a 400 × 800 m2 region of high side scan acoustic backscatter near a west-facing normal fault [Davis and Villinger, 1992]. There are no extensive sulfide deposits associated with Dead Dog, and it appears that the vents have been continuously discharging moderate temperature fluids (270°C) since their formation [Davis and Fisher, 1994].
 Results from Ocean Drilling Program (ODP) drilling at Middle Valley (Legs 139 and 169) suggest a prominent basement edifice beneath Dead Dog perpetuates venting at the site. The structure is a small seamount that was buried by 250 m of sediment during the Pleistocene [Shipboard Scientific Party, 1992b]. Langseth and Becker  and Stakes and Franklin  have postulated that the volcanic center beneath Dead Dog emplaced the basalt sills in the eastern portion of the valley. Long-term hydrothermal convection in the region is most likely driven by heat trapped in the basaltic basement beneath the sediment blanket.
 The Dead Dog area generates an elongate positive heat flow anomaly with a north-south orientation [Davis and Villinger, 1992]. Basement topography beneath Dead Dog focuses hydrothermal fluid flow through the locally thinner sediment cover [Davis and Fisher, 1994], and seafloor gravity data suggest the sediments in this high heat flow area have been highly lithified by hydrothermal processes [Ballu et al., 1998]. Direct hydrothermal fluid discharge at the vent field is regulated by an indurated sediment cap roughly 30 m below the seafloor, which appears to result in a secondary level of hydrothermal circulation in the upper 30 m of sediment [Stein et al., 1998].
 Although the Middle Valley hydrothermal fields are well characterized in comparison with most deep-sea systems, several fundamental questions have not been addressed. In particular, the pattern of fluid flow in the basement is essentially unconstrained, as is the depth of the hydrothermal reaction zone. In addition, recharge rates estimated at the eastern boundary fault zone near Site 855 (25 m3/yr) are too small to match vent field fluid fluxes [Davis and Fisher, 1994], and the bulk sediment permeabilities appear to be too small (10−17 m2) to account for the difference [Stein and Fisher, 2001]. Thus the major seawater source regions feeding the Dead Dog circulation system are not completely understood, and the subsurface fluid pathways defining the lateral and vertical extent of the hydrothermal reaction zone in the basement remain unknown.
 The OBSs were free-fall deployed from the R/V Wecoma in August 1996. Each instrument was equipped with a three-component, 1 Hz natural period geophone (Mark Products L-4), a hydrophone, and 6 Gbytes of disk storage space. The vertical and hydrophone channels were each sampled at 128 Hz, whereas the horizontal channels were not sampled to conserve disk space and maximize the duration of the experiment. Each OBS recorded continuous data on the vertical and hydrophone channels for 145 days, until January 1997. Timing was maintained within each OBS using a temperature compensated crystal (Seascan Co.) with timing corrected using the drift measured over the deployment.
 The OBS seafloor locations were precisely navigated by means of a shipboard acoustic survey of the instrument transponders. The joint inversion method of Creager and Dorman  was used to calculate station locations from acoustic ranges and GPS ship positions. OBS location standard errors are less than 15 m.
 Seismographs were positioned around two permanent ODP boreholes, Sites 858 G and 857 D, prior to ODP Leg 169. Scientific goals of the drilling leg included a revisit to the Middle Valley drill sites, and a further study of hydrothermal circulation in the area. The OBS array was deployed to monitor background seismicity associated with Dead Dog vent field and record any microearthquakes induced by man-made perturbations to the hydrothermal system during ODP Leg 169 borehole operations.
 The experiment featured a compact configuration of eight instruments on the seafloor, with approximately 500 m distance between sensors chosen to best investigate microearthquakes associated with the ODP operations at Sites 858 G and 857 D during Leg 169. The reduced interelement spacing was crucial for locating small-magnitude microearthquakes in the shallow crust because it allowed waveforms associated with small seismic moments to be recorded at all of the stations. Although no microearthquakes were detected that were clearly associated with ODP operations during Leg 169, the OBS array did record high levels of natural seismicity beneath the Dead Dog vents, providing an opportunity to investigate the hydrothermal system.
 During the recording period, the OBS array detected 3646 local and regional events, the majority of which exhibit clear P and S arrivals on all instruments (Figure 2). P phases are generally distinguished in the data set by impulsive arrivals with a frequency near 32 Hz. By comparison, S phases are near 8 Hz, have amplitudes up to five times the size of P phases, and usually arrive several seconds behind the P wave depending on the range of the microearthquake to the OBS and the seismic velocity structure. Shear to compressional wave conversions (SP) from sharp boundaries in the seismic velocity structure are also present in the data. In Middle Valley, the mode conversions are created at the basement/sediment interface. Compressional to shear wave conversions (PS) were also occasionally observed. In addition, water wave phases (P waves reflected off the sea surface) are found in the OBS records, but they arrive within the high-amplitude, low-frequency shear arrivals and are difficult to distinguish.
 Initially, P and S arrival times were picked by hand for each local microearthquake and each station. Most events exhibited 12–16 pickable arrivals out of a possible total of 16. Earthquake hypocenters and errors were computed using the 3D grid search algorithm of Sohn et al. . Travel times were estimated by tracing synthetic rays through compressional and shear velocity models of the Middle Valley area spanned by the dashed box in Figure 1. Travel times were defined at nodes every 100 m in the x direction, 200 m in the y direction, and every 100 m in the z direction. The picks for each event were weighted by their relative uncertainties and then used to find the grid point with the minimum weighted RMS residual. Subsequently, a finer grid was constructed around this minimum point with nodes spaced every 50 m in the x and y directions, and spaced every 25 m in the z direction. Travel times within the finer grid were linearly interpolated, and the point with the minimum RMS residual was chosen as the event hypocenter.
 The compressional and shear velocity models were constructed by extending 1D basement profiles beneath an overlying sediment layer of variable thickness and constant velocity. Sediment thickness variations were based on values provided by Davis and Villinger , except near Sites 857 D and 858 G, where higher-resolution ODP drilling results were implemented [Shipboard Scientific Party, 1992a, 1992b]. The sediment layer P velocity was defined as 1970 m/s. This value is based on correlations between ODP well logs and multichannel seismic reflection data [Rohr and Groschel-Becker, 1994]. The 1D compressional velocity profile for the underlying basement was based on the Endeavour segment model of Cudrak and Clowes . The profile was modified in the upper 600 m of basement to fit results provided by a small seismic refraction data set from this region of Middle Valley [Golden, 2000]. The sediment shear velocity was constrained at 420 m/s by delay times between SP conversions and S arrivals in this data set. Shear velocities in the upper 600 m of basement were based on values of 800 m/s for the shallow crust [Christeson et al., 1997] and velocities in the deeper basement were derived from the compressional velocity profile assuming VP/VS = 1.85.
5. Microearthquake Results
 Out of 3646 local and regional earthquakes recorded during the 5-month experiment, roughly 70% occurred in swarms with seismicity rates of at least 2.5 events/hr. Thirteen distinct swarms were observed at random intervals beneath the Dead Dog field (over 1000 microearthquakes), with sizes ranging from just over 10 events in 4 hours (swarm 12), to over 300 events in 2 days (swarm 4). Three microearthquake swarms were detected outside of the Dead Dog area (Figure 4). These swarms are much larger than those associated with the vent field, with several hundreds of events and time durations of over a week.
 Hypocenter locations for 480 microearthquakes beneath the Dead Dog field were obtained via grid search localization, with a mean RMS travel time residual of 0.013 s and a standard error of 0.023 s (Figure 5a). The majority of the hypocenters are concentrated at 1–2.5 km depth, directly beneath the vents (Figure 5b). Hypocenters are also scattered to the north, east, and west of the vent field and deepen to ∼3 km away from the vent field. During the OBS survey, no microearthquakes were observed near the southern end of the OBS array. A number of epicenters estimated to be >3 km north of the vent field could not be reliably located due to their large distance from the instruments. Approximate locations based on relative P and S arrival times for these swarms (numbers 6, 7, 10, and 13) are depicted in Figure 5a.
 Gridded travel time residuals about each hypocentral point were used to compute a 3D error surface representing the 1σ limit for each event [Wilcock and Toomey, 1991; Sohn et al., 1998]. Error bars in the x, y, and z directions were computed from these error surfaces. The error bars represent hypocentral misfit caused by discrepancies in the seismic velocity model and inaccuracies in picking the event arrival times. Errors in hypocentral depth tend to increase as microearthquakes are located further from the seismometer array (Figure 5b), which is a common feature of local seismic networks.
 Seismic moments were estimated from long period levels of body wave (S and P) amplitude spectra at each station [Brune, 1970; Hanks and Thatcher, 1972; Hildebrand et al., 1997; Sohn et al., 1998]. The seismic moment for each event is an average of the moments obtained for all eight instruments. Local magnitudes were computed from the seismic moments using the relation: log10M0 = 16 + 5Mw [Lee and Stewart, 1981]. Magnitudes for the Dead Dog microearthquakes range from −1.2 to 0.2. Focal plane solutions were not determined because the directional coverage in this experiment was relatively sparse.
 Epicenter locations for the three swarms located outside of the Dead Dog area are not well constrained since they are well outside the OBS network. Instead, approximate locations were estimated from relative first arrivals at each instrument and delay times between P and S phases. The swarms occur along the major fault zones in Middle Valley (Figure 1), such as the Nootka Fault which intersects the rift from the east roughly 12 km north of Dead Dog, the western boundary fault, and the central fault which marks the eastern edge of the primary rift graben. The large central fault swarm occurred roughly 8 km to the north of the OBS array. No earthquakes were generated along this fault south of Dead Dog vent field. P and S arrivals were clipped on the OBS records for a majority of the earthquakes associated with these swarms. However, a lower magnitude bound of Mw > 1 was estimated using the long-period spectral levels of the smallest-amplitude events and ranges to the instruments. These local earthquakes exhibit much larger seismic moments and a more episodic chronology than the events under Dead Dog. Large tectonic events are common for the northern terminus of the Juan de Fuca Ridge [e.g., Wahlstrom and Rogers, 1991; Hyndman and Rogers, 1981].
6. Relative Relocations
 Microearthquakes from each swarm located beneath the vent field form distinct groupings (Figure 5a), and waveforms for most events within any given swarm closely resemble each other (Figure 2). This suggests that the seismicity within a swarm is generated by a single-source mechanism at a common location, and that the implementation of relative relocation methods may yield more accurate relative hypocenter locations. We applied the cross-correlation/relative relocation technique of Shearer  to each Dead Dog swarm separately. Waveform cross-correlation was utilized to quantify waveform similarity and yield differential travel times at each instrument for each correlated event pair. Compressional and shear wave arrivals were cross-correlated independently for each microearthquake pair, using 0.4 and 0.6 s windows around each arrival, respectively. A minimum average correlation coefficient of 0.7 was required to define a “correlated” pair (“doublet”), and the corresponding peaks of the correlation functions for each OBS defined differential compressional and shear travel times. The differential times were then used to calculate new, relative hypocenter locations for each doublet using a grid search routine. A small grid of relative compressional and shear travel times was constructed about the centroid of the original, absolute hypocenter locations for each swarm. The grid spanned a 2 × 2 × 1.5 km3 area in the x, y, and z directions, with lateral spacing of 100 m and vertical spacing of 50 m. Travel times between grid points were linearly interpolated. Once relative hypocenter locations were obtained for each correlated pair, the entire group of differential locations was inverted to compute a final set of relative relocations. Standard (1σ) errors for each relocated event were estimated using bootstrap methods.
 The relative location method successfully relocated 304 out of 480 total microearthquake hypocenters beneath Dead Dog, corresponding to eight out of the nine originally located swarms (Figures 5c and 5d). Relative errors are greatly reduced from those associated with the original locations, and are typically on the order of tens of meters with the exception of swarms 8 and 9. Swarms 8 and 9 are contemporaneous with seismicity on the Nootka fault zone, which degrades the data quality of those swarms (c.f., Figures 3 and 4). Consequently, none of the five located swarm 8 events correlated sufficiently, and the original, absolute hypocenter locations are depicted in Figures 5c and 5d. Waveform cross-correlation using swarm 9 events did yield correlation coefficients above 0.7, but the relative errors are larger than average, just over 100 m in some instances.
 The relocated hypocenters form discrete pockets and columns, with the two largest clusters (swarms 4 and 11) located directly beneath the Dead Dog vents (Figures 5c and 5d). Smaller clusters surround the vent field to the north, east, and west. Relocated swarms deepen as they extend northward from the vent field, yielding a sloped seismogenic zone depicted in Figure 5d. Inaccuracies in our 3D structural model will lead to systematic errors in the depth of the swarms and hence in the slope of this ramp because the events are outside of the small array, but the relative locations between events should be stable. The primary control on locations is derived from the time difference between the P and S wave arrivals at each instrument.
 We applied a perturbation test to each relocated cluster independently to determine if its shape is required by the data or a reflection of an error surface in the inversion. For each swarm, differential compressional and shear travel times were randomly perturbed for each correlated event pair. Maximum perturbation amplitudes were constrained to be less than one standard deviation of the relative travel time differences for all event pairs in a given swarm. The relocation method was then applied to the randomly modified relative travel times.
 Artifacts from the relocation method (e.g., network geometry, grid spacing) are accentuated in the perturbed results. The perturbation test indicated that microearthquake clusters aligned with the OBS array (which is linear in the north-south direction) tend to be smeared in the east-west direction. This bias widens the shapes of swarms 2, 4, and 11 in Figure 5c. In addition, groups of relocated hypocenters positioned well outside of the array can be stretched in the vertical direction. This effect produces artificially vertical configurations for swarms 1, 2, and 3 in Figures 5c and 5d. However, the vertical extent of microearthquake clusters located closest to the instruments (i.e., 4, 5, and 11) is required by the data.
7. Source of the Vent Field Seismicity
 The results from our local earthquake survey lead us to believe that hydrothermal processes generated the seismicity observed beneath Dead Dog vent field. This inference is based on the spatial and temporal patterns of the observed seismicity, the correlation of seismicity with heat flow measurements, b-value estimates, and the geology of Middle Valley. Here we review this evidence, and use our results to deduce fluid flow patterns in the basement beneath the Dead Dog vents.
 The spatial and temporal characteristics of the microearthquakes in this study are similar to events recorded near the Bio9/P and Tube Worm Pillar/Y hydrothermal fields at 9°50′N on the EPR [Sohn et al., 1999]. Seismicity beneath both the Middle Valley and EPR sites is characterized by brief and fairly intense swarms of small-magnitude (generally Mw < 0), densely clustered, vertically aligned, repeatable events. The simplest way to explain this combination of parameters is with seismic triggering from contraction of cooled rock in regions of high thermal stress in the hydrothermal reaction zone. The other conceivable source mechanisms, magmatism and tectonism, are much more difficult to reconcile with the observations.
 Middle Valley is essentially an extensional graben, and the sediments and crust are cut by a variety of normal faults [Rohr and Schmidt, 1994]. The normal faults may play some role in the microearthquakes observed, but the vertical columns and small pockets of microearthquakes beneath Dead Dog vent field do not define fault-like structures. We are unable to determine focal mechanisms for the events given the limited array and emergent character of many of the arrivals.
 The locations of the seismic swarms beneath Dead Dog are correlated with results from seafloor heat flow studies of Middle Valley (Figure 6a). Thermal gradient and thermal conductivity have been measured at 550 sites spaced a few hundred meters apart using probes equipped with multiple-thermistor arrays [Davis and Villinger, 1992]. Dead Dog vent field is denoted by a positive heat flow peak of 24 W/m2, and the associated region of high heat flow (>1.0 W/m2) extends roughly 1 km to the south and 4 km to the north of the vents creating a linear feature on the contour map. The seismicity beneath Dead Dog is concentrated within this heat flow anomaly, and the epicenters follow its north-south orientation. The vigor of seismicity is also roughly correlated with the heat flow values. The two largest swarms are located beneath the heat flow peak, and the smaller swarms scatter northward within regions of reduced seafloor heat flow anomalies.
 The frequency-magnitude distribution of earthquakes beneath Dead Dog vent field produces a local b-value of 1.49, which reflects the nearly constant seismic moment release of the observed events. The b-values are commonly used as a basis for distinguishing tectonic and volcanic events, with values less than one being characteristic of tectonism, and values greater than 1.3 being characteristic of volcanism and large thermal strain [Warren and Latham, 1970; Wiemer and McNutt, 1997].
8. Model of Middle Valley Hydrothermal Flow
 If the seismicity beneath Dead Dog is predominantly triggered by thermal strain in the reaction zone, then the microearthquakes and their hypocentral distributions delineate the depth extent of hydrothermal circulation beneath the vent field (Figures 6b and 6c). We therefore surmise that the reaction zone beneath Dead Dog is oriented as a ramp that shallows to the south toward the high-temperature vents from a maximum depth of about 2 km to the north. The absence of seismicity more than 1 km south of Dead Dog is difficult to interpret given the relatively short period of observation, but the events are definitely biased to the north of the vents. The majority of the fluid discharged at Dead Dog vent field may be drawn into the system locally, via the faults to the north of the vent field.
 The maximum depth of seismicity, and hence our inferred hydrothermal reaction zone, is ∼1.5 km below the surface of the basement. The downward velocity (u) of a hydrothermal cracking front into basement can be estimated using the water penetration theory of Lister . Using parameters appropriate for Dead Dog we obtain
where Q is the heat flow generated in the basement rock (1 W/m2) [Stein and Fisher, 2001], ρ is crustal density (2900 kg/m3), cρ is the specific heat capacity of the crust (1000 J/kg °K), T1 is the initial hot rock temperature (1500°K), and Tw is the hot fluid temperature (573°K). Stein and Fisher  use seafloor heat flow data to estimate Q by summing all conductive and advective heat flow outputs within a 260 km2 area surrounding Dead Dog vent field.
 If we assume that the cracking surface has migrated downward with a constant velocity of 1 cm/yr over the life span of hydrothermal circulation in this portion of Middle Valley, the total time to penetrate 1.5 km of basement (inferred from swarm depths) is 150,000 years. This calculation provides a maximum age limit (because the heat flow in this area was probably higher than 1 W/m2, when Middle Valley was actively spreading during the Pleistocene (A. Fisher, personal communication), but is consistent with the models of Langseth and Becker  and Davis and Fisher . These models suggest that Dead Dog is a very old hydrothermal system (i.e., hundreds of ka) that was created as magmatic activity ceased and rapid sedimentation immediately insulated the volcanic center, setting up moderate vent field temperatures (270°C) that have remained relatively constant.
 The cracking front velocity estimate of 1 cm/yr is about two orders of magnitude slower than theoretical values for unsedimented hydrothermal systems [Lister, 1982]. The result, however, is consistent with observations of hydrothermal seismicity at the adjacent Endeavor segment of the Juan de Fuca Ridge. Although there are lava flows in the North Endeavour Valley [Karsten et al., 1990] that are more recent than the turbidite sediments that blanket Middle Valley [Davis and Villinger, 1992], the depth of seismicity beneath the North Endeavour is ∼1–2 km deeper [Wilcock et al., 1999] than seismicity beneath Dead Dog. The depth difference emphasizes the insulating effect of the turbidites in Middle Valley, a feature that is absent at the Endeavour vent field areas. The thick sediment layer surrounding Dead Dog inhibits seawater recharge into the hydrothermal system, limiting the rate of cracking front propagation and ultimately shoaling the maximum depth of any microearthquake activity.
 Microearthquake activity from August 1996 to January 1997 beneath Dead Dog vent field in the Middle Valley of the Juan de Fuca Ridge is dominated by 13 seismic swarms. Hypocentral estimates for 480 of these events were obtained with a grid search method, and 304 of these hypocenters were relocated using waveform cross-correlation. Relocated events map into discrete clusters, with an apparent deepening of events moving away from the vent field to the north. The character of the seismicity is most compatible with a thermal mechanism, which we interpret as thermal strain resulting from the cooling of hot rock by hydrothermal fluids. The microearthquakes may therefore delineate the hydrothermal reaction zone during our experiment, placing it about 1–2 km directly under the vents. This suggests that the reaction zone is propagating downward at approximately 1 cm/yr, which is in agreement with calculations based on the extraction of heat by hydrothermal processes. This propagation rate is extremely slow and likely reflects the effect of the low-permeability turbidite sediment cover on heat flow within the Dead Dog system.
 This study illustrates the ability of local seafloor seismic networks to record small-magnitude microearthquakes associated with hydrothermal vent fields. It further suggests the utility of marine seismological methods for constraining the nature of fluid convection in oceanic crust.
 We thank Jacques Lemire and Tom Deaton for engineering and maintenance of the OBSs, Peter Shearer for assistance with the relative microearthquake relocations, Russ Johnson, Mark McDonald, Valerie Ballu, Wayne Crawford, and the officers and crew of the R/V Wecoma for help with the instruments at sea, and Jo Griffith for illustration support. This research was supported by the National Science Foundation (OCE 95-21282).