The Role of the Ocean Observatories Initiative in Monitoring the Offshore Earthquake Activity of the Cascadia Subduction Zone

USAGE Permission is granted to copy this article for use in teaching and research. Republication, systematic reproduction, or collective redistribution of any portion of this article by photocopy machine, reposting, or other means is permitted only with the approval of The Oceanography Society. Send all correspondence to: info@tos.org or The Oceanography Society, PO Box 1931, Rockville, MD 20849-1931, USA. Oceanography THE OFFICIAL MAGAZINE OF THE OCEANOGRAPHY SOCIETY


INTRODUCTION
The largest and some of the most damag ing earthquakes in history have occurred in subduction zones, where one plate plunges beneath another plate.At the Cascadia subduction zone (Figure 1), which extends from northern California to Vancouver Island, the Juan de Fuca Plate is being subducted beneath the North American Plate.In several respects, Cascadia is an end member in the global spectrum of subduction zones (Wang and Tréhu, 2016).Because of the young age of the subducting plate (<15 million years), a relatively slow convergence rate (3-4 cm yr -1 ), and abundant sediment supply that blankets subducting oceanic crust, the Cascadia plate boundary fault is hotter than most subduction mega thrusts.It is also unusually free of recent earthquake activity (Figure 1).
Despite the low level of current earth quake activity, there is evidence for large (magnitude >8) historic and prehistoric earthquakes on the Cascadia megathrust.Tsunami records in Japan, coastal subsid ence, and oral histories indicate there was a magnitude 8.7-9.2 earthquake at the Cascadia subduction zone on January 29, 1700 (Atwater et al., 2015).Onshore and offshore paleoseismic data indicate the occurrence of 20 large earthquakes in the past 10,000 years, with interevent times ranging from 200 years to 1,200 years (Goldfinger et al., 2012).
The paucity of recent earthquakes can mean that the two plates are slip ping past each other without generating earthquakes or that they are firmly locked together, storing elastic strain that will be released in a future large earthquake, or some combination thereof.Thermal models (e.g., Hyndman and Wang, 1993;Cozzens and Spinelli, 2012) have been used to predict the boundary between the seismogenic zone, where the plates are cold and may be locked (tempera ture <350°-450°C), and a deeper zone of "episodic tremor and slip" (ETS), charac terized by slow slip observed in GPS data accompanied by seismic tremor (Rogers and Dragert, 2003).Figure 1 shows the position of the 450°C isotherm (short dashed line) and the updip position of ETS (short/long dashed line), which are approximately coincident and may indi cate the downdip limit of seismogenic slip in major subduction zone earthquakes (Hyndman, 2013;Wang and Tréhu, 2016).Note that the Cascadia seismo genic zone lies almost entirely offshore.
Global plate models (e.g., DeMets et al., 2010) predict that ~8 m of slip perpendicu lar to the margin has accumulated off cen tral Oregon since 1700.This slip deficit is large enough to generate a magnitude 8 or larger earthquake if it is released through sudden slip on the shallow (offshore) plate ABSTRACT.Geological and historical data indicate that the Cascadia subduction zone last ruptured in a major earthquake in 1700.The timing of the next event is currently impossible to predict, but recent studies of several large subduction zone earthquakes provide tantalizing hints of precursory activity.The seismometers at the Ocean Observatories Initiative (OOI) Slope Base and Southern Hydrate Ridge nodes are well placed to provide new insights into interplate coupling because they are located over a segment of the subduction zone that is nominally locked but that has been relatively active for more than a decade.Since their installation in 2014, 18 earthquakes with magnitudes up to 3.8 have been located by the Pacific Northwest Seismic Network between 44°N and 45°N in the region of the plate boundary thought to be accumulating strain.The OOI seismometers have also detected events that were not reported by the onshore seismic network.Noting that OOI data are available in real time, which is a necessary criterion for routine earthquake monitoring, and that the OOI seismometers generally have lower noise levels than campaignstyle ocean bottom seismometers, there would be significant benefit to adding seismometers to existing nodes that are not yet instrumented with seismometers.boundary (Tréhu, 2016).To evaluate the seismic hazard in Cascadia, it is import ant to understand how much of the poten tial slip is being stored, to be released in a future great subduction zone earth quake, and how much is being accom modated through aseismic slip and small earthquakes on the plate boundary or by internal deformation of the Juan de Fuca and North American Plates.Developing this understanding requires knowledge of where offshore earthquakes occur.Moreover, recent results from subduc tion zone earthquakes elsewhere show that at least some earthquakes are pre ceded by distinctive foreshock activity and/or slow slip within the seismogenic zone (e.g., Bouchon et al., 2013;Ito et al., 2013;Burgmann, 2014).As these precur sory patterns become better documented and modeled, their value for earthquake forecasting should increase.The objective of this paper is to deter mine the value of seismometers located at the Ocean Observatories Initiative (OOI) Slope Base (SB) and Southern Hydrate Ridge (SHR) nodes to the detec tion and location of earthquakes on the Cascadia margin.Through coincidence, the OOI SBSHR network is located near an anomalous segment of the margin that has been the site of persistent seismic ity in the magnitude 2-5 range since at least 1989 (Tréhu et al., 2015).Between November 4, 2014, when data from the OOI SBSHR network became avail able through the IRIS Data Management Center, and September 30, 2017, 18 earth quakes with magnitudes up to 3.8 were reported by the Pacific Northwest Seismic Network (PNSN) between 44°N and 45°N and 125.5°W and 124°W.In addition, from November 2014 to October 2015, the Cascadia Initiative ocean bottom seis mometer (OBS) network (Toomey et al., 2014) operated simultaneously with the OOI SBSHR network, providing the opportunity to compare signaltonoise ratios and evaluate the impact of potential expansion of the OOI seismic network.

BACKGROUND
The Challenge of Recording Seismic Waves Offshore Earthquakes radiate seismic energy across a range of frequencies, with larger earthquakes generating relatively more lowfrequency energy.The seismic waves are recorded using sensors that typically record three orthogonal components of ground motion.Development of portable broadband sensors in the 1980s greatly expanded the ability of onshore seismol ogists to record seismic waves at many locations and across a wide frequency band, improving their ability to resolve Earth structure and earthquake source processes.Expansion of this capability to the ocean has been difficult because of the exceptionally low shear strength of marine sediments and the high noise levels generated by nonlinear inter actions and direct effects of ocean waves (e.g., Webb, 1998).Moreover, conven tional OBS designs include power, buoy ancy, timing control, and data recording in a single package.The resulting bulky package is more susceptible to narrow band frequency resonances and is more attractive to seafloor fauna, which dis turb the instrument, generating signals, termed "biobumps, " that can obscure local earthquakes.Many of these noise sources are potentially mitigated by OOI sensors because on the Cascadia mar gin they are buried in seafloor sedi ments and separated from the power and recording hardware.

The Challenge of Locating Offshore Earthquakes
An earthquake is located by determining the arrival times of the seismic waves it generates on a network of seismometers.Accurate determination of source latitude and longitude requires a good azimuthal distribution of seismic stations around an earthquake and knowledge of the velocity with which seismic waves travel through the rocks from the source to the receiver.Both requirements are problematic when only onshore data are available to locate small earthquakes on the shallow part of subduction zone plate boundaries, which are generally offshore.Moreover, the crustal thickness and velocity structure change dramatically across the continent ocean margin (Figure 2), which is gener ally not taken into account for routine earthquake monitoring.
To interpret the tectonic implications of earthquakes, it is important to deter mine their depth.A general rule of thumb for resolving depth from observed travel Red dots indicate the projected positions of the two magnitude 4.7-4.8earthquakes of 2004.The projected position of a subducted seamount (SS) based on magnetic anomaly data is also shown (Figure 3A).The basement rocks of the upper plate in this region are formed by mafic rocks of the Siletz terrane, anomalously thick oceanic crust that was accreted to North America ~50 million years ago and that is slowly moving northward and rotating relative to the core of North America, effectively acting as a microplate known as the Oregon Block (McCaffrey et al., 2007)  times is that the horizontal distance to the closest station should not exceed the depth.This condition is nearly always violated when land stations record off shore events.Offshore seismic stations are required to locate earthquake epicenters accurately, obtain better estimates of source depth, and decrease the earthquake detection threshold.Although conven tional OBSs (from which data are down loaded when the instrument is recovered) can be used to revise depth estimates post facto, realtime data are needed for rou tine earthquake monitoring.

Recent Seismicity on the Central Oregon Margin
Figure 3A shows a map of all earth quakes from January 1, 1989, through September 3, 2017, with magnitudes ≥3 between 43.5°N and 45.5°N and 126°W and 123°W.The data are from the US Geological Survey Advanced National Seismic SystemComprehensive Earthquake Catalog (ANSSComCat; see Figure 1 caption), which includes events reported by the PNSN and other regional networks.While earthquakes are scat tered throughout this region, most seis micity has occurred in clusters labeled N, S A , and S B .The largest event in these clusters was a magnitude 4.8 earthquake in cluster S A in July 2004 and a mag nitude 4.7 earthquake in cluster N in August 2004 (Tréhu et al., 2008).These two events were large enough to generate lowfrequency energy within the noise notch (see below), permitting determina tion of the source mechanism and depth through moment tensor inversion of the waveforms recorded by regional seismic networks.From moment tensor inver sion, with additional support from travel time analysis of secondary seismic phases (pP, PmP, and SmS) observed on distant and regional seismic stations, Tréhu et al. (2008) concluded that these were low angle thrust earthquakes that occurred on or near the plate boundary.
When smaller earthquakes in these clusters are relocated (Tréhu et al., 2015, and references therein), either by deter mining their depth relative to the two larger earthquakes or by including OBS data when available, they define a surface dipping about 12° to the east (Figure 4), corresponding to the plate boundary iden tified in previous controlled source imag ing experiments (Figure 2).This activity may result from grinding of a buried sea mount against the seaward edge of crys talline basement (labeled "Siletz terrane" in Figure 2).Cluster S B was first identi fied as being distinct from S A when four nearly identical earthquakes with magni tudes of 2.8-3.4 occurred on January 25, 2013.The three clusters continue to pro duce earthquakes large enough to be detected by the onshore network at a rate of about six events per year.
In map view, earthquake locations obtained through these detailed stud ies are similar to those obtained through PNSN analyses, although the clus ters are better defined.The earthquake depths and their relationship to the plate boundary structure, however, are sig nificantly affected by the lack of off shore stations and by the simplified velocity models used in routine PNSN analysis (Figure 4).

RESULTS OF NEW ANALYSES ENABLED BY THE OOI SEISMOMETERS
To examine the impact of including OOI stations in routine operations and the potential of augmenting the offshore array by instrumenting additional exist ing OOI nodes with seismometers, we compare OOI and OBS noise levels and look at three "case studies" of earthquakes occurring since 2014.Figure 3B shows the locations of OOI nodes, Cascadia Initiative OBSs, land seismic stations, and velocity models used in this study.

Noise Levels
Considerable effort has been devoted to understanding the characteristics of seis mic noise in the ocean to improve record ings of teleseismic earthquakes on broad band OBSs (Webb, 1998).Broadband seismometers are best deployed in bore holes (Montagner et al., 1994;Collins et al., 2001), but where they are not available, shallow burial is preferred to emplacement on the seafloor because it ensures good ground coupling and shields the seismometer from ocean currents.All seismic stations of the SBSHR net work are buried.The broadband stations at SHR and SB were placed beneath the seafloor following an approach described by Romanowicz et al. (2003).A section of PVC pipe was inserted into the sedi ments to form a caisson, the sediments were evacuated from inside, the seis mic sensor was enclosed in a spheri cal housing and placed at the bottom of the caisson so as not to touch the walls (online Supplementary Figure S1), and the caisson was covered with glass beads flush with the seafloor.The three short period sensors at SHR are contained in narrow titanium cylinders that were buried by inserting them into a groove gouged by the arm of a remotely operated vehicle (Supplementary Figure S2); the groove was then filled with glass beads.
Figure 5A compares noise power spec tra from the vertical channel of the two OOI broadband seismometers with those from a station located 20 km inland (BABR) and from three autonomous sea floor OBSs from the Cascadia Initiative experiment.From 0.1 Hz to 3 Hz, the noise spectra are dominated by micro seism peaks, which have higher ampli tudes at the seafloor stations compared to the land station because they are gen erated by nonlinear interactions of ocean waves (Webb, 1998).At frequencies immediately below the microseism peak, there is a lownoise notch on all stations except for J25C.This is the signal band used for inversion of waveform data to obtain moment tensor estimates.At the seafloor stations, the noise level increases at lower frequencies because the pressure perturbations from infragravity waves (longwavelength ocean waves) reach the seafloor.In shallow water, the infragrav ity wave noise extends to higher frequen cies and impinges on, and in the case of station J25C eliminates, the lownoise notch.Processing techniques can remove this noise from the vertical channel using data from the horizontal channels or a coincident seafloor pressure sensor (e.g., Crawford and Webb, 2000).
From 3 Hz to 20 Hz, the noise lev els onshore and at three of the five sea floor stations, including the two OOI sta tions, are remarkably similar (Figure 5A).This signal band is needed to pick travel times of P and Swaves from local earth quakes.Figure 5B shows noise levels at 5 Hz recorded by Cascadia Initiative and OOI seismometers as a function of water depth.For the Cascadia Initiative OBSs, the noise levels are quite scattered, but increase markedly at shallower depths.This likely reflects poor coupling of the sensors to the seafloor, noise induced by components of the OBS package in the presence of strong ocean currents, and frequent impulsive signals due to bio logical activity, which can number in the thousands per day at water depths <500 m (Williams et al., 2010).Burying the seismometers, as was done for the OOI instruments, shields the sensors from currents and fauna, leading to lower noise levels at higher frequencies.

Modeling the Impact of Offshore Stations on Detection Levels Using Site-Specific Noise Characteristics
Figure 6 shows the results of applying the method of McNamara et al. (2016) to develop a series of magnitude detec tion threshold maps for different seismic Circle radius is proportional to magnitude.Events in red were relocated by Tréhu et al. (2015); dark gray circles show PNSN locations for these events.Light gray circles are catalog events that were not relocated either because they were too small or they occurred prior to 2000 when there were too few stations (e.g., larger events west of 124.8°W).Relocations done for this study are in orange.Clusters labeled N, S A , and S B are discussed in the text and shown in map view in Figure 3A.Gray arrows (from Tréhu et al., 2015) and dashed lines (this study) illustrate relocated depths for selected events.network configurations by combining sitespecific spectral noise characteris tics with simple earthquake source and wave propagation models.If the ground motion at a station from a model earth quake exceeds the noise model at any fre quency, we infer that the station detects the earthquake.Detections at four sta tions result in a network detection and hence a locatable earthquake.To investi gate potential improvements in detection capability that might result from adding offshore stations to the current onshore seismic network, we used modal noise profiles derived from OOI and Cascadia Initiative stations, which represent the most likely noise levels at hypothetical sites.We also use modal noise values for PNSN onshore stations.These maps represent a bestcase sce nario because a signal just above noise does not guarantee an automatic detec tion; in practice, signal must exceed noise by some (difficult to quantify) amount.However, the method is con sistent and objective, and thus useful for comparing network detection per formance.Figure 6A shows the detec tion threshold from the onshore seis mometers and current OOI SBSHR and Axial Seamount (see Wilcock et al., 2018, in this issue) networks using noise pro files computed for a randomly selected month.The detection threshold is very low onshore and shows how clusters of stations can further reduce the detection threshold locally in regions of particular interest.Offshore, the detection thresh old increases smoothly with increasing distance from the network, with locally lower magnitude detection thresholds in the immediate vicinity of the OOI net works.Figure 6B shows how an ambitious offshore network of buried seismome ters would impact the detection thresh old, assuming noise levels characteris tic of OOI sites and a station distribution   similar to the Cascadia Initiative network along the Cascadia subduction margin.This exercise indicates that improve ment in the detection threshold for off shore earthquakes along the continental margin is decreased, although a signifi cant decrease in the magnitude detection threshold would require a large invest ment.However, it is important to note that this analysis considered only detect ability.The impact of offshore stations on the resolution of earthquake source depth, a parameter that is important for understanding the tectonic and hazard implications of seismicity, is discussed in the next section.It also does not include the impact of Tphases on detections, as discussed in the penultimate section.

The Importance of Close Stations for Determining Earthquake Depth: Case Study of the June 25-26, 2015, Earthquakes
The PNSN catalog includes two magni tude 2.5 earthquakes on June 25 and 26 with locations in cluster S A and nominal source depths of about 30 km.To relocate these events, we picked P and S arrival times from seismic stations within about 120 km of the epicenter, including the OOI stations, Cascadia Initiative OBSs, and onshore stations.Figure 7A com pares data from broadband OOI station HYS14 in the SHR network to data from onshore broadband station BABR.Both stations were located about 60 km from the earthquake and show clear, impulsive P and Swave arrivals, and the signal tonoise ratio is similar in the 1-20 Hz frequency band.Although not shown here, signaltonoise ratios are also sim ilar at SB (broadband station HYSB1) and on shortperiod stations HYS1113 in the SHR network.
Figure 7B compares the Pwaves recorded on HYSB1 and HYS14 to Pwaves recorded on several Cascadia Initiative OBSs, including J25D, which was the station closest to the epicenter.Although P and S first arrivals are impul sive on J25D, the signaltonoise ratio at distances >50 km is quite variable.While this may be due in part to the radiation pattern of the earthquake, it may also be due to effects of soilstructure inter action, bottom current activity, and other noise sources that affect OBSs more strongly than the buried OOI seismome ters, as discussed in the previous section.Note that although FC03D is closer to the source than HYS14, and J17D is closer than HYSB1, the noise levels are lower at the OOI site than at the Cascadia Initiative site for both of these station pairs.
Earthquakes were located in one dimensional velocity models.All stations were normalized to sea level by applying station corrections based on the veloc ity of the uppermost layer in the model (2.5 km s -1 ) to correct for different sta tion elevations.This results in station corrections that range from −1.23 s for OBS station J10D to 1.15 s for onshore station B032 and is equivalent to remov ing the Coast Range and filling the ocean with sediment.
Results for the event on June 25, 2015, at 20:25 based on two different one dimensional Pwave velocity models were compared.See Supplementary Figure S3 for details.The first model was extracted from a twodimensional velocity model of the continental margin derived from controlledsource imaging at the posi tion marked OR3 in Figure 3B.The sec ond model is appropriate for the Coast Range (CR on Figure 3B) and is similar to the model used by PNSN to locate earth quakes in this region.For both models, we assumed a ratio of Pwave to Swave velocity of 1.77, which is a typical value for crustal rocks, although it is low for marine sediments.To test the sensitivity  of the solution to source depth, we deter mined the bestfitting source latitude, longitude, and origin time for a series of fixed source depths and compared the misfit for the different velocity models and various subsets of data.
The effect of velocity model uncer tainty on the source latitude and lon gitude is small.All solutions are within about 2 km of each other (and of the epicenter reported by PNSN), which is consistent with the nominal horizon tal uncertainties.The bestfit source depth, however, depends strongly on the velocity model, with model OR3 lead ing to shallow source depths (<15 km) and model CR leading to deeper depths (>30 km).Moreover, the misfit for OBS station J25D, located ~17 km from the epicenter, is small only for depths of 8-13 km independent of the velocity model used.Addition of seismometers to existing OOI nodes that do not cur rently have them (Figure 3B) would be of great benefit for improving the depth resolution when locating earthquakes along this anomalously active segment of the margin.

A Test of the Broadband Data from the OOI Seismometers:
The April 15, 2017, Earthquake The magnitude 3.8 event in subcluster S B on April 15, 2017, was large enough to generate longperiod waves suitable for moment tensor inversion.Results of that analysis are shown in Figure 8. Data at periods of 12-20 s and velocity model OR3 were used for the inversion.Stable moment tensor inversions near coasts, where the microseismic noise is strong, require using seismic waves in the frequency band of the lownoise notch (Figure 5A), which are not excited strongly enough to rise above back ground noise levels for earthquakes with magnitudes less than about 3.2.
Figure 8A shows several examples of the data and the waveforms predicted for the bestfitting solution, which has a depth of 12 km, and for a solution with a depth of 36 km, which is close to the depth in the PNSN catalog (38 km). Figure 8B shows the focal mechanism (type of faulting in the source region) of the bestfit solu tion along with azimuthal data coverage.Figure 8C shows the residual for the inver sion as a function of depth and the focal mechanism for the solution at each depth.If the OOI data are not included, the mis fit decreases by ~0.05 s at each depth and the focal mechanism and source depth are not significantly different.The increased misfit when OOI data are included is likely due to higher noise levels on the horizontal components of the OOI broad band seismometers.Although inclu sion of the OOI data did not significantly affect the moment tensor solution in this case, this exercise demonstrated that the OOI broadband data are useful for low frequency waveform modeling.
We conclude that this earthquake was a thrust event near the plate boundary.The fault dip appears to be ~45° on either an eastdipping or westdipping plane.The focal mechanism is similar to that of a magnitude 3.8 earthquake in cluster S A on March 20, 2012 (Tréhu et al., 2015), and somewhat different from that of the two larger earthquakes in 2004 (Tréhu et al., 2008).These results suggest a scenario in which the subducted seamount is act ing as an asperity (lock on the fault) that is generating earthquakes around its edges or in the lower crust of the upper plate (Wang and Bilek, 2011).The implications of these observations for estimates of  Cascadia earthquake hazard are unclear, and continued monitoring of this region with improved capability to resolve the source depth and mechanism is needed.

Events Detected by the OOI Network That Are Not Reported by Onshore Networks
We visually scanned all data collected from January 1 to August 15, 2017, at stations HYS1114 and HYSB1 to deter mine if any local signals were recorded across the SBSHR network that were not reported by the PNSN.Visual scans are a preliminary step toward identification of potential "templates" for automated searching of the data for additional earth quakes (Morton and Bilek, 2015) and other seismic signals.The dense network of stations at the summit of SHR (inset in Figure 3B) was designed, in part, to detect possible seismic tremor or low frequency earthquakes generated by pulses of fluid motion associated with known methane vents.
We detected two small events in 2017, one on January 4 and one on April 18, that were clearly recorded across the OOI array and were not reported by the land network because they were not clearly recorded on enough stations to trigger a detection (Supplementary Figure S4).These events have been tentatively located ~50 km northwest of station HYSB1 (Figure 3B) and may represent earthquakes in the sub ducting plate as it approaches the defor mation front.A number of earthquakes in this general region have been reported in the past, including a magnitude 5.8 event in 1973 (Spence, 1989).The OOI seismic stations at SB, SHR, and Axial Seamount have the potential to decrease the detec tion threshold and improve the accuracy of earthquake locations within this part of the Juan de Fuca Plate, improving knowl edge of the tectonics of this region and the relationship between intraplate deforma tion and subduction.
Another type of event recorded on the OOI SBSHR network, but not onshore, is energy from submarine earthquakes that couples into the SOFAR channel and produces a packet of energy that follows the P (primary) and S (secondary) waves.These are generally known as T (ter tiary) phase.Figure 9 shows aftershocks associated with a swarm of earthquakes (including two events with magnitudes of 5.8 and 5.9) on the Blanco Transform Fault on June 1, 2015.In the hour follow ing the magnitude 5.8 event, the ANSS ComCat catalog includes six earthquakes (the smallest of magnitude 3.9).During that same time period, Tphases from more than 20 smaller amplitude events were detected on the OOI network.Future work includes determination of magnitudes based on Tphase amplitudes (e.g., Dziak, 2001) to increase the magni tude range available for detailed analysis of swarm characteristics, which can pro vide clues to in situ stress state and earth quake dynamics.

CONCLUSIONS
The OOI seismometers contribute signifi cantly to the ability of the regional seis mic network in the Pacific Northwest to monitor the temporal evolution of seismic activity.Offshore sensors are critical for determining earthquake source depths, which are important for understanding Cascadia interplate dynamics.The OOI data are available immediately and can be incorporated into routine locating proce dures, unlike data from traditional ocean bottom seismometers.Moreover, data quality is comparable to that recorded on coastal stations, unlike data from ocean bottom seismometers, for which local effects of seafloor currents, biological activity, and OBSsediment coupling dif ferences due to different package designs can lead to considerable variability in the effective signaltonoise ratio.Although the magnitude detection threshold based only on P and Swaves is signifi cantly decreased primarily in the imme diate vicinity of the stations, analysis of Tphases, recorded only on offshore sta tions, promises to decrease the detection levels for earthquakes occurring through out the region.Expansion of the seismic network to the three active nodes on the central Cascadia margin that are not pres ently equipped with seismometers would improve depth resolution for events along a segment of the margin that has been  anomalously seismically active for the past several decades when compared with seg ments to the north and south.Significant expansion of the cabled network would be necessary to provide equivalent capability along the entire margin.

FIGURE 2 .
FIGURE 2. Interpreted cross section through the Cascadia subduction zone near latitude 44.65°N.The velocity model of Gerdom et al. (2000) is overlain by a geologic interpretation.The red line represents the plate boundary (dashed where the modeled plate boundary temperature is >450°C).Red dots indicate the projected positions of the two magnitude 4.7-4.8earthquakes of 2004.The projected position of a subducted seamount (SS) based on magnetic anomaly data is also shown (Figure3A).The basement rocks of the upper plate in this region are formed by mafic rocks of the Siletz terrane, anomalously thick oceanic crust that was accreted to North America ~50 million years ago and that is slowly moving northward and rotating relative to the core of North America, effectively acting as a microplate known as the Oregon Block(McCaffrey et al., 2007).Seaward of the Siletz terrane, the upper plate is composed of folded and faulted accreted sediments and slope sediments captured in basins formed through deformation and uplift of the margin.The seaward edge of the Siletz terrane is overlain by a forearc basin (FB).The red bar along the base of the figure indicates the region shown in Figure4.
FIGURE 2. Interpreted cross section through the Cascadia subduction zone near latitude 44.65°N.The velocity model of Gerdom et al. (2000) is overlain by a geologic interpretation.The red line represents the plate boundary (dashed where the modeled plate boundary temperature is >450°C).Red dots indicate the projected positions of the two magnitude 4.7-4.8earthquakes of 2004.The projected position of a subducted seamount (SS) based on magnetic anomaly data is also shown (Figure3A).The basement rocks of the upper plate in this region are formed by mafic rocks of the Siletz terrane, anomalously thick oceanic crust that was accreted to North America ~50 million years ago and that is slowly moving northward and rotating relative to the core of North America, effectively acting as a microplate known as the Oregon Block(McCaffrey et al., 2007).Seaward of the Siletz terrane, the upper plate is composed of folded and faulted accreted sediments and slope sediments captured in basins formed through deformation and uplift of the margin.The seaward edge of the Siletz terrane is overlain by a forearc basin (FB).The red bar along the base of the figure indicates the region shown in Figure4.

FIGURE 3 .
FIGURE 3. (A) Seismicity in the central Cascadia subduction zone.Dark gray circles mark earthquakes relocated as discussed by Tréhu et al. (2015); light gray circles are epicenters from the Pacific Northwest Seismic Network (PNSN) catalog prior to 2015 that have not been relocated; red dots denote earthquakes reported by PNSN from January 2015 through September 2017.Dot sizes are proportional to magnitude.N, S A , and S B indicate clusters of seismicity discussed in the text.The blue circle marks a subducted seamount inferred based on magnetic anomaly data(Tréhu et al., 2015).(B) Ocean Observatory Initiative (OOI) nodes and Cascadia Initiative (CI) and PNSN seismic stations used for this study.The inset shows the configuration of the four-station OOI array at Southern Hydrate Ridge (SHR); those labeled HYS11-13 are short period seismometers, and HYS14 is a broadband seismometer.Broadband seismometer HYSB1 is at node Slope Base (SB).White lines show the locations of two-dimensional velocity models discussed byGerdom et al. (2000); stars locate one-dimensional velocity models used in this study.Velocity models are shown in Supplementary FigureS5.

FIGURE 4 .
FIGURE 4. Cross section showing earthquake depths compared to the plate boundary in Figure2.Circle radius is proportional to magnitude.Events in red were relocated byTréhu et al. (2015); dark gray circles show PNSN locations for these events.Light gray circles are catalog events that were not relocated either because they were too small or they occurred prior to 2000 when there were too few stations (e.g., larger events west of 124.8°W).Relocations done for this study are in orange.Clusters labeled N, S A , and S B are discussed in the text and shown in map view in Figure3A.Gray arrows (fromTréhu et al., 2015) and dashed lines (this study) illustrate relocated depths for selected events.

FIGURE
FIGURE 5. (A) Velocity power spectra for background noise on the vertical broadband channels of selected PNSN (BABR), CI (J25C, J19D, M16D), and OOI (HYS14, HYSB1) sites, with water depths indicated.Spectra were calculated with one-hour data windows over one year; median values are plotted.(B) Median noise levels at 5 Hz as a function of instrument depth for all Cascadia Initiative OBSs deployed in 2014-2015 and for the OOI stations.Color corresponds to sites shown in (A).

FIGURE 7 .
FIGURE 7. Waveforms for the July 25, 2016, earthquake in cluster S A .Data have been high-pass filtered at 1 Hz to remove microseismic noise.(A) Comparison of three-component broadband waveforms for stations HYS14 and BABR.(B) P-waveforms recorded on the vertical component of OOI and OBS broadband seismometers ordered with source-receiver distance (17-93 km) increasing from top to bottom.Waveforms are aligned on the P-arrival pick.
Time (s) Relative to the First Arrival at Each Station HYS14

FIGURE 8 .
FIGURE 8. Moment tensor inversion for the April 15, 2017, magnitude 3.8 earthquake.(A) Examples of observed (black lines) long period waveforms (12-20 s period) compared to waveforms predicted for the preferred solution at 12 km depth (red dashed lines) and at 36 km depth (green dashed lines).In addition to the two OOI stations, 15 onshore broadband stations were used to obtain this solution.(B) Best-fit source mechanism, including a minor non-double couple component that is likely due to noise, has a nearly horizontal compression axis oriented at 275° and indicates thrust faulting on a plane dipping ~45°.Triangles on the outside of the focal sphere indicate azimuth of stations used.(C) Normalized variance and corresponding best-fit double couple solution as a function of assumed source depth.Red and green mechanism diagrams correspond to red and green seismograms in (A).

FIGURE 9 .
FIGURE 9. T-phases recorded on the vertical component of the OOI SB-SHR network from a swarm of earthquakes on the Blanco Transform Fault that occurred on June 1, 2015.The red dashed lines show T-phases from earthquakes in the ANSS Comprehensive Earthquake Catalog.Time and magnitude are indicated.The pink arrows show T-phases interpreted to be from additional, smaller earthquakes.