Earthquake Clusters Research Articles

Voronoi residual analysis of spatial point process models with applications to California earthquake forecasts

Many point process models have been proposed for describing and forecasting earthquake occurrences in seismically active zones such as California, but the problem of how best to compare and evaluate the goodness of fit of such models remains open. Existing techniques typically suffer from low power, especially when used for models with very volatile conditional intensities such as those used to describe earthquake clusters. This paper proposes a new residual analysis method for spatial or spatial-temporal point processes involving inspecting the differences between the modeled conditional intensity and the observed number of points over the Voronoi cells generated by the observations. The resulting residuals can be used to construct diagnostic methods of greater statistical power than residuals based on rectangular grids. Following an evaluation of performance using simulated data, the suggested method is used to compare the Epidemic-Type Aftershock Sequence (ETAS) model to the Hector Mine earthquake catalog. The proposed residuals indicate that the ETAS model with uniform background rate appears to slightly but systematically underpredict seismicity along the fault and to overpredict seismicity in along the periphery of the fault.

Maximum magnitude estimations of induced earthquakes at Paradox Valley, Colorado, from cumulative injection volume and geometry of seismicity clusters

The Paradox Valley Unit (PVU), a salinity control project in southwest Colorado, disposes of brine in a single deep injection well. Since the initiation of injection at the PVU in 1991, earthquakes have been repeatedly induced. PVU closely monitors all seismicity in the Paradox Valley region with a dense surface seismic network. A key factor for understanding the seismic hazard from PVU injection is the maximum magnitude earthquake that can be induced. The estimate of maximum magnitude of induced earthquakes is difficult to constrain as, unlike naturally occurring earthquakes, the maximum magnitude of induced earthquakes changes over time and is affected by injection parameters. We investigate temporal variations in maximum magnitudes of induced earthquakes at the PVU using two methods. First, we consider the relationship between the total cumulative injected volume and the history of observed largest earthquakes at the PVU. Second, we explore the relationship between maximum magnitude and the geometry of individual seismicity clusters. Under the assumptions that: (i) elevated pore pressures must be distributed over an entire fault surface to initiate rupture and (ii) the location of induced events delineates volumes of sufficiently high pore-pressure to induce rupture, we calculate the largest allowable vertical penny-shaped faults, and investigate the potential earthquake magnitudes represented by their rupture. Results from both the injection volume and geometrical methods suggest that the PVU has the potential to induce events up to roughly MW 5 in the region directly surrounding the well; however, the largest observed earthquake to date has been about a magnitude unit smaller than this predicted maximum. In the seismicity cluster surrounding the injection well, the maximum potential earthquake size estimated by these methods and the observed maximum magnitudes have remained steady since the mid­ 2000s. These observations suggest that either these methods overpredict maximum magnitude for this area or that long time delays are required for sufficient pore-pressure diffusion to occur to cause rupture along an entire fault segment. We note that earthquake clusters can initiate and grow rapidly over the course of 1 or 2 yr, thus making it difficult to predict maximum earthquake magnitudes far into the future. The abrupt onset of seismicity with injection indicates that pore-pressure increases near the well have been sufficient to trigger earthquakes under pre-existing tectonic stresses. However, we do not observe remote triggering from large teleseismic earthquakes, which suggests that the stress perturbations generated from those events are too small to trigger rupture, even with the increased pore pressures.

Study of Crustal Thickness and Poisson's Ratio in the Coastal Area of South China

AbstractThe crustal thickness (H) and Poisson's ratio (ν) are two important parameters for understanding the plate motion and the structure in the earth. We collected teleseismic receiver functions of earthquakes, which occurred from 2001 to 2008, from 75 permanent seismic stations of Guangdong and Fujian networks. In total, 74 estimations of H and ν were obtained by making use of the H‐κ method. Combining characteristics of the fault distribution, seismicity and topography, several conclusions were achieved below: (1) Crustal thickness ranges from 26 to 32 km, reducing from land to sea and from north to south. Poisson's ratio is 0.23∼0.28, increasing from the inland to sea. Variations of H and ν have positive and negative correlations with surface topography, respectively. (2) Fault activity corresponds to the block‐like and band‐shaped distributions of H and ν. Transitional variation zones of H and ν spread in accordance with the seismicity. Earthquake clusters and mantle upwelling are generally corresponding to the rapid increase of ν. (3) Crustal rocks in this area mainly consist of intermediate and acidic mineral. Crustal thicknesses are positively and negatively related to Poisson's ratio under actions of extensional tectonics offshore and thrusting structure in folded regions, respectively, revealing the diverse evolution models in different eras and districts.

The association of micro‐earthquake clusters with mapped faults in the Dead Sea basin

AbstractAbout 1100 micro‐earthquakes that occurred along the Dead Sea basin during the past 25 years express ongoing seismic activity along mostly obscured fault segments. We apply seismological and statistical methods in order to associate seismic activity with geological and geophysical data and to locate and characterize active fault segments in the basin; three regional 1‐D velocity models and the double‐difference method were used for the relocation of seismic events. Geostatistical analysis shows that in first order the micro‐earthquakes reflect faulting along segments of the main longitudinal fault while seismic quiescence along historically active zones indicates locked segments and stress concentration. Based on waveform similarity, spatial and temporal proximity, we define earthquake clusters that comprise about 30% of the relocated seismicity and two thirds of the total seismic moment. About 80% of the clusters are associated with N‐S trending faults. In two cases the clusters indicate a migration of the seismicity in time and space, one of which preceded the largest earthquake in our catalog (M5.2). In addition to the associated longitudinal activity, several clusters appear to represent faulting in east‐west orientation which is interpreted here as a secondary fault system.

Decreasing Rate of M>=7 Earthquakes in the Northern Hemisphere Since 1900

Online Material: Figures showing variation through time of magnitude of completeness. The preceding decade was characterized by the abrupt increase of the number of large earthquakes ( M w≥8), with the occurrence of three of the six strongest seismic events recorded in the period of instrumental seismology (2004 M 9.0 Sumatra–Andaman earthquake; 2010 M 8.8 Maule, Chile, earthquake; 2011 M 9.0 Tohoku, Japan, earthquake [Shearer and Stark, 2012]). Many authors analyzed seismicity back to 1900 to assess whether large earthquakes either cluster in time or follow a random distribution (Daub et al. , 2012, and references therein) but reached contradictory conclusions (Kerr, 2011). We considered a declustered global catalog of M ≥7 earthquakes that occurred between 1900 and 2013 for a different, although related, objective. We investigated the existence of significant trends in the yearly distribution of earthquakes and evaluated their possible dependence on the geography. This study tries to extend the analysis by Shearer and Stark (2012), who find that after 1900, worldwide M ≥7 earthquakes are not uniformly distributed in time, but instead are mainly concentrated in the first half of the twentieth century. We separately considered the earthquakes that occurred in the southern and northern hemispheres. We found the seismicity rate is almost stable in the south, whereas it is strongly decreasing in the north, where it reduced from about nine earthquakes per year to about five per year in just over a century. We also found the decrease is common to the more important seismic regions of the northern hemisphere. It is possible that the result is an artifact due to a biased estimate of magnitude, especially in the first half of the twentieth century, when the instrumental coverage was scarce and different magnitude scales were adopted. The problem was raised by Engdahl and Villasenor (2002) …

Foreshock activity related to enhanced aftershock production

AbstractForeshock activity sometimes precedes the occurrence of large earthquakes, but the nature of this seismicity is still debated, and whether it marks transient deformation and/or slip nucleation is still unclear. We here study at the worldwide scale how foreshock occurrence affects the postseismic phase and find a significant positive correlation between foreshock and aftershock activities: earthquakes preceded by accelerating seismicity rates produce 40% more aftershocks on average, and the length of the aftershock zone after 20 days is 20% larger. These observations cannot be reproduced by standard earthquake clustering models that predict the accelerating pattern of foreshock occurrence but not its impact on aftershock activity. This strongly suggests that slow deformation transients, possibly related to episodic creep, could initiate prior to the main shock and extend past the coseismic phase, resulting in compound ruptures that include a very long period (up to tens of days) component.

Nonstationary ETAS models for nonstandard earthquakes

The conditional intensity function of a point process is a useful tool for generating probability forecasts of earthquakes. The epidemic-type aftershock sequence (ETAS) model is defined by a conditional intensity function, and the corresponding point process is equivalent to a branching process, assuming that an earthquake generates a cluster of offspring earthquakes (triggered earthquakes or so-called aftershocks). Further, the size of the first-generation cluster depends on the magnitude of the triggering (parent) earthquake. The ETAS model provides a good fit to standard earthquake occurrences. However, there are nonstandard earthquake series that appear under transient stress changes caused by aseismic forces such as volcanic magma or fluid intrusions. These events trigger transient nonstandard earthquake swarms, and they are poorly fitted by the stationary ETAS model. In this study, we examine nonstationary extensions of the ETAS model that cover nonstandard cases. These models allow the parameters to be time-dependent and can be estimated by the empirical Bayes method. The best model is selected among the competing models to provide the inversion solutions of nonstationary changes. To address issues of the uniqueness and robustness of the inversion procedure, this method is demonstrated on an inland swarm activity induced by the 2011 Tohoku-Oki, Japan earthquake of magnitude 9.0.

Seismic Activities and Earthquake Potential in the Tibetan Plateau

AbstractThe Tibetan Plateau is a region with intensive tectonic deformation and high level earthquake activity. On the basis of historical records, there have been 18 earthquakes with magnitude over 8 and more than 100 events with magnitude between 7 and 7.9 in the plateau. All of these events occured either along plate boundary of the Himalaya or along boundaries among the intra‐plate fault blocks as well as sub‐blocks. There have been three phases of earthquake clustering in the Tibetan Plateau, 1920 to 1937, 1947 to 1976, and 1995 to present, since instrumental seismic records began in 1900. Each phase of the cluster is characterized by events of magnitude over 8 and a number of events of magnitude over 7 to form an earthquake series. The three phases are the Haiyuan‐Gulang earthquake series in 1920s to 1930s, the Zayu‐Damxung earthquake series in 1950s to 1970s, and the Kunlun‐Wenchuan series since the beginning of this century. Each earthquake series also has its major active region of occurrences. For example, the major active region of occurrence is associated with the newest phase of earthquake series is the Bayan Har fault block. The episodes of seismic activity in the Tibetan Plateau coincide with the high level activities of global mega‐earthquake (Mw ≥ 8) activity. The Kunlun‐Wenchuan earthquake series concurs with the global earthquake activity since 2001 suggesting geodynamic connections between them. Based on careful comparative studies, we suggest the current phase of earthquake activity continues today, giving that the global mega‐earthquakes (Mw8∼9) occur frequently along plate boundaries and major intra‐plate earthquakes (Mw7∼8) occur along boundaries of crustal fault block. Through studies of relationships between global and regional earthquake activities and of temporal and spatial distribution of the earthquakes in the Tibetan Plateau, this paper summarized general tendencies of current earthquake activity, assesses recent earthquake potential, and proposes suggestions for intensive earthquake monitoring.

Diverse rupture modes for surface-deforming upper plate earthquakes in the southern Puget Lowland of Washington State

Earthquake prehistory of the southern Puget Lowland, in the north-south compressive regime of the migrating Cascadia forearc, reflects diverse earthquake rupture modes with variable recurrence. Stratigraphy and Bayesian analyses of previously reported and new 14C ages in trenches and cores along backthrust scarps in the Seattle fault zone restrict a large earthquake to 1040–910 cal yr B.P. (2σ), an interval that includes the time of the M 7–7.5 Restoration Point earthquake. A newly identified surface-rupturing earthquake along the Waterman Point backthrust dates to 940–380 cal yr B.P., bringing the number of earthquakes in the Seattle fault zone in the past 3500 yr to 4 or 5. Whether scarps record earthquakes of moderate (M 5.5–6.0) or large (M 6.5–7.0) magnitude, backthrusts of the Seattle fault zone may slip during moderate to large earthquakes every few hundred years for periods of 1000–2000 yr, and then not slip for periods of at least several thousands of years. Four new fault scarp trenches in the Tacoma fault zone show evidence of late Holocene folding and faulting about the time of a large earthquake or earthquakes inferred from widespread coseismic subsidence ca. 1000 cal yr B.P.; 12 ages from 8 sites in the Tacoma fault zone limit the earthquakes to 1050–980 cal yr B.P. Evidence is too sparse to determine whether a large earthquake was closely predated or postdated by other earthquakes in the Tacoma basin, but the scarp of the Tacoma fault was formed by multiple earthquakes. In the northeast-striking Saddle Mountain deformation zone, along the western limit of the Seattle and Tacoma fault zones, analysis of previous ages limits earthquakes to 1200–310 cal yr B.P. The prehistory clarifies earthquake clustering in the central Puget Lowland, but cannot resolve potential structural links among the three Holocene fault zones.

One feature of the activated southern Ordos block: the Ziwuling small earthquake cluster

Abstract: Small earthquakes (Ms > 2.0) have been recorded from 1970 to the present day and reveal a significant difference in seismicity between the stable Ordos block and its active surrounding area. The southern Ordos block is a conspicuous small earthquake belt clustered and isolated along the NNW direction and extends to the inner stable Ordos block; no active fault can match this small earthquake cluster. In this paper, we analyze the dynamic mechanism of this small earthquake cluster based on the GPS velocity field (from 1999 to 2007), which are mainly from Crustal Movement Observation Network of China (CMONOC) with respect to the north and south China blocks. The principal direction of strain rate field, the expansion ratefield, the maximum shear strain rate, and the rotation rate were constrained using the GPS velocity field. The results show that the velocity field, which is bounded by the small earthquake cluster from Tongchuan to Weinan, differs from the strain rate field, and the crustal deformation is left-lateral shear. This left-lateral shear belt not only spatially coincides with the Neo-tectonic belt in the Weihe Basin but also with the NNW small earthquake cluster (the Ziwuling small earthquake cluster). Based on these studies, we speculate that the NNW small earthquake cluster is caused by left-lateral shear slip, which is prone to strain accumulation. When the strain releases along the weak zone of structure, small earthquakes diffuse within its upper crust. The maximum principal compression strees direction changed from NE-SW to NEE-SWW, and the former reverse faults in the southwestern margin of the Ordos block became a left-lateral strike slip due to readjustment of the tectonic strees field after the middle Pleistocene. The NNW Neo-tectonic belt in the Weihe Basin, the different movement character of the inner Weihe Basin (which was demonstrated through GPS measurements) and the small earthquake cluster belt reflect the activated southern margin of the Ordos block, which was generated through readjustment of the tectonic strees field after the middle Pleistocene.

Application of cluster analysis based on waveform cross-correlation coefficients to data recorded by ocean-bottom seismometers: results from off the Kii Peninsula

Waveform cross-correlation coefficients (CCs) have often been used to investigate clustering of earthquakes. Such techniques, however, have rarely been applied to waveform data recorded by ocean-bottom seismometers (OBSs). The data recorded by some OBSs are strongly influenced by site effects such as the existence of unconsolidated sediment layers. The resulting waveforms tend to have a monotone frequency so that the calculated CCs stay artificially high, and false delay times are obtained at certain multiples of the period. This effect also varies from place to place. To overcome such problems, CC measurements were first performed using multiple time windows with different lengths in order to reject unstable results. Station-specific CC thresholds were then objectively determined based on the CC distributions. This method was applied to seismic measurements obtained by OBSs deployed off the Kii Peninsula from 2003 to 2007 in order to identify characteristic hypocenter patterns. Two types of earthquake clusters were found. The first occurred within the oceanic mantle, and the component events were linearly aligned along the NNE direction. Such earthquakes are considered to occur on preexisting faults due to dehydration of the serpentinized mantle. The second type of cluster consisted of interplate earthquakes that occurred at the southern tip of the Kii Peninsula. These are thought to be the result of stress caused by local structural heterogeneity of the dense rock body.

A geological explanation for intraplate earthquake clustering complexity: The zeolite-bearing fault/fracture networks in the Adamello Massif (Southern Italian Alps)

Interconnected networks of faults and veins filled with hydrothermal minerals such as zeolite are widespread in many orogenic terrains. These fractures commonly form at relatively low temperatures (e.g. <200 °C) late in the tectonic history and represent significant phases of fluid flow and mineralisation during exhumation. Zeolite-bearing fractures spatially associated with the Gole Larghe Fault Zone in the Southern Italian Alps are preserved along an interconnected network of variably orientated pre-existing structures. They show evidence of repeated episodes of hydraulic tensile fracturing and small magnitude (total offsets <5 m) shear displacements. We use geological observations and Coulomb stress modelling to propose that repeated seismogenic rupturing of larger offset faults led to local stress transfer and reactivation of widely distributed smaller pre-existing structures in the wall rocks. The differing orientations of the pre-existing features within what is assumed to have been a single regional stress field led to the simultaneous development of reverse, strike-slip and extensional faults. The kinematic diversity and cyclic nature of the hydraulically-assisted deformation suggest that the mineralised fracture systems represent a geological manifestation of intraplate micro-earthquake clusters associated with fluid migration episodes in the upper crust. Our observations highlight the role of crustal fluids and structural reactivation during earthquakes.

Archaeological evidence for a destructive earthquake in Patras, Greece

Oriented collapse of columns, large-scale destruction debris and temporary abandonment of the area deduced from an archaeological excavation provide evidence for a major (intensity IX) earthquake in Patras, Greece. This, and possibly a cluster of other earthquakes, can be derived from archaeological data. These earthquakes are not included in the historical seismicity catalogues, but can be used to put constraints to the seismic risk of this city. Patras was affected by a cluster of poorly documented earthquakes between 1714 and 1806. The city seems to be exposed to risks of progressive reactivation of a major strike-slip fault. A magnitude 6.4 earthquake in 2008 has been related to it. This fault has also been associated with a total of four events in the last 20 years, a situation reminiscent of the seismic hazard at the western edge of the North Anatolian Fault.

Stress Drops of the 1997–1998 Colfiorito, Central Italy Earthquakes: Hints for a Common Behaviour of Normal Faults in the Apennines

Stress drop estimates of moderate-magnitude earthquakes in the Umbria–Marche region, in the northern Apennines, exhibit a large scatter. For the two Mw 5.7 and 6.0 main shocks of 26 September 1997 near Colfiorito, several papers resulted in stress drop estimates of 20 MPa, but values as low as 2–3 MPa were proposed as well. Also for the largest aftershocks (Mw > 4), estimates spread from < 1 MPa up to values ten times larger. We have critically revisited methods and data used in the literature. We have specifically faced the trade-off between source and propagation effects, as we believe that it is responsible for a part of the large scatter. To keep this trade-off under control, we have applied a methodology that combines the best fit of both source spectra after Empirical Green’s Function (EGF) deconvolution and observed ground motion spectra, finding that the results of the two different data sets converge independently at the same solution. We have used ground motions observed in the Colfiorito basin, where an accelerograph and a co-located seismological broad-band station recorded three clusters of earthquakes in a broad magnitude interval (1.7 ≤ Mw ≤ 6.0). We have found that the mainshock–aftershock sequences result in stress drops of 2–5 MPa at Mw ≥ 5.6, with an average tendency to decrease at smaller magnitudes where stress drop variability increases. These findings confirm the source scaling recently assessed through Empirical Green’s Function deconvolution for another well-monitored seismic sequence of normal-faulting earthquakes, which struck the city of L’Aquila in the central Apennines in April 2009. The similar scaling law of the two areas suggests common mechanisms of stress release for the shallow normal faults in the Apennines. The propensity of smaller earthquakes to increase in variability, with a tendency toward smaller stress drops, may reflect an effect of fault strength heterogeneities for smaller size ruptures.

Earthquake Preparedness Should Not Fluctuate on a Daily or Weekly Basis

Operational earthquake forecasting (OEF) is the practice of continual updating and dissemination of physics‐based short‐term (days) probabilities for the occurrence of damaging earthquakes. Although fully appreciating the noble intention of OEF and the scientific merits of the seismicity analyses it employs, we are concerned that its wide promotion may lead the public to believe that earthquake preparedness can fluctuate at timescales of days or weeks. Scientists who support OEF do recognize the importance of long‐term (decades and centuries) forecast. Jordan (2013) points out that long‐term forecasting can “guide earthquake safety provisions of building codes, performance‐based seismic design, and other risk‐reducing engineering practices, such as retrofitting to correct design flaws in older buildings.” However, they also believe that “properly done, short‐term forecasting complements long‐term seismic‐hazard analysis in promoting earthquake preparedness” and that “the age of OEF has arrived” (Jordan, 2013). Although to some extent short‐term forecasting is a worthwhile research subject, we question the claim that it should and can be made operational. A typical OEF warning would announce that the probability of an earthquake of certain magnitude happening in a given region over the next few days has increased from negligibly small values (such as 0.001%) to a larger value (such as 5%; Jordan and Jones, 2010). The probability gain of several orders of magnitude is thought to be useful for communicating risk to the public and enhancing preparedness. We have two concerns over this type of warning. First, warnings of this type send and reinforce a wrong message, that is, if scientifically estimated probability is very low, such as 0.001%, the public can afford to be less prepared for damaging earthquakes. OEF is based largely on analyses of earthquake clustering, or potential foreshock sequences, but the majority of damaging earthquakes are not preceded by …

Compositional change of seismic event sequences in focal zones during preparation of strong earthquakes

Abstract: The change of the structure of seismic event sequences in focal area in preparation period of strong earthquakes has been studied. Findings show that weak earthquake clustering in time increases in earlier stage of forming of focal area. However, spatial clustering of seismic events rises at the latest stage.

Earthquake clustering and possible fault interactions across Cook Strait, New Zealand, during the 1848 and 1855 earthquakes

The 1848 and 1855 earthquakes provide examples of clustered earthquake sequences involving several faults. Historical evidence indicates that fault rupturing and ground deformation on both sides of Cook Strait occurred over periods of several days during the October 1848 earthquakes on the Awatere, Ohariu and possibly the Wellington faults, with shallowing in Lambton Harbour, Wellington. The January–February 1855 earthquakes occurred on the Wairarapa, Wharekauhau, Nicholson Bank, Vernon, Awatere and Needles faults. The pattern of faulting suggests the existence of distinct seismogenic zones across Cook Strait, within which mainshock triggering of aftershock displacement on major faults can occur.

Fractal analysis and yule statistics for seismic prediction based on 2009 L’Aquila earthquake in Italy

Earthquakes represent a five-dimensional set of events: time, three space dimensions, and magnitude. To study the clustering between the space dimensions and the magnitude, a comparative analysis of epicenter and magnitude distributions was performed between the earthquake time series in Italy and around the L’Aquila region. The b value analysis reveals a slightly stronger seismic activity in the L’Aquila region compared with the Italian seismicity. Analysis of epicenter distribution shows that earthquakes in the L’Aquila region tend to fill the plane more than the overall distribution. Fractal dimension of hypocenter distribution for the L’Aquila region is approximately twice the b value, fulfilling the relation D = 2b. To investigate the relation between time and magnitude, an original interpretation was given for the variation of the bY value based on the Yule statistics. This approach reveals the relationship between earthquakes happened to propagate the existing faults and events creating new faults. The application of bY value analysis to the L’Aquila region highlights that clustering of earthquakes may be precursors of large earthquakes, and shows a reliable approach to predict violent earthquakes.

Transformation to equivalent dimensions—a new methodology to study earthquake clustering

A seismic event is represented by a point in a parameter space, quantified by the vector of parameter values. Studies of earthquake clustering involve considering distances between such points in multidimensional spaces. However, the metrics of earthquake parameters are different, hence the metric in a multidimensional parameter space cannot be readily defined. The present paper proposes a solution of this metric problem based on a concept of probabilistic equivalence of earthquake parameters. Under this concept the lengths of parameter intervals are equivalent if the probability for earthquakes to take values from either interval is the same. Earthquake clustering is studied in an equivalent rather than the original dimensions space, where the equivalent dimension (ED) of a parameter is its cumulative distribution function. All transformed parameters are of linear scale in [0, 1] interval and the distance between earthquakes represented by vectors in any ED space is Euclidean. The unknown, in general, cumulative distributions of earthquake parameters are estimated from earthquake catalogues by means of the model-free non-parametric kernel estimation method. Potential of the transformation to EDs is illustrated by two examples of use: to find hierarchically closest neighbours in time–space and to assess temporal variations of earthquake clustering in a specific 4-D phase space.

Three-dimensional seismic velocity structure and earthquake relocations at Katmai, Alaska

We invert arrival time data from local earthquakes occurring between September 2004 and May 2009 to determine the three-dimensional (3D) upper crustal seismic structure in the Katmai volcanic region. Waveforms for the study come from the Alaska Volcano Observatory's permanent network of 20 seismic stations in the area (predominantly single-component, short period instruments) plus a densely spaced temporary array of 11 broadband, 3-component stations. The absolute and relative arrival times are used in a double-difference seismic tomography inversion to solve for 3D P- and S-wave velocity models for an area encompassing the main volcanic centers.The relocated hypocenters provide insight into the geometry of seismogenic structures in the area, revealing clustering of events into four distinct zones associated with Martin, Mageik, Trident-Novarupta, and Mount Katmai. The seismic activity extends from about sea level to 2km depth (all depths referenced to mean sea level) beneath Martin, is concentrated near 2km depth beneath Mageik, and lies mainly between 2 and 4km depth below Katmai and Trident-Novarupta. Many new features are apparent within these earthquake clusters. In particular, linear features are visible within all clusters, some associated with swarm activity, including an observation of earthquake migration near Trident in 2008. The final velocity model reveals a possible zone of magma storage beneath Mageik, but there is no clear evidence for magma beneath the Katmai-Novarupta area where the 1912 eruptive activity occurred, suggesting that the storage zone for that eruption may have largely been evacuated, or remnant magma has solidified.