We perform a joint inversion of phase velocities from both earthquake and ambient noise induced Rayleigh waves to determine shear wave velocity structure in the crust and upper mantle beneath the Pacific Northwest. We focus particularly on the areas affected by mid-Miocene to present volcanic activity. The joint inversion, combined with the high density seismic network of the High Lava Plains seismic experiment and data from the EarthScope Transportable Array, provides outstanding resolution for this area. In Oregon, we find that the pattern of low velocities in the crust and uppermost mantle varies between the High Lava Plains physiographic province and the adjacent northwestern Basin and Range. These patterns may be due to the presence of the Brothers Fault Zone which separates the clockwise rotating northwest Basin and Range from the relatively undeformed areas further north. Further to the east, the Owyhee Plateau, Snake River Plain (SRP) and northeastern Basin and Range are characterized by high crustal velocities, though the depth extent of these fast wave speeds varies by province. Of particular interest is the mid-crustal high velocity sill, previously only identified within the SRP. We show this anomaly extends significantly further south into Utah and Nevada. We suggest that one possible explanation is lateral crustal extrusion due to the emplacement of the high density mafic mid-crustal sill structures within the SRP.
 Since the mid-Miocene, the Pacific Northwest U.S.A. (PNW) has been subjected to widespread volcanic activity that is not limited to ongoing Cascade arc volcanism (where eruptions began in the region of the present-day arc ca. 45–50 Ma [Madsen et al., 2006]). Voluminous flood basalt volcanism began ∼16.5 Ma with dike eruptions near Steens Mountain in southeastern Oregon. Following the Steens eruptions, the locus of volcanism migrated rapidly northward along a series of N-S rifts, culminating in the Columbia River flood basalts (CRB) with eruptive centers concentrated near the Idaho/Washington/Oregon border region [Camp and Ross, 2004] (Figure 1). Over the course of ∼1.5 Ma, over ∼220,000 cubic kilometers of basalt were erupted from the Steens, Chief Joseph, and Monument dike swarms [Camp and Hanan, 2008]. Silicic volcanic eruptions began just southwest of the Owyhee Plateau (OP) ∼16 Ma [Brueseke et al., 2007]. This silicic volcanism then migrated away from the OP in two directions. One track headed northeast to Yellowstone, roughly parallel to, and at the same rate as, North American plate motion [e.g., Smith et al., 2009]. The other progressed northwest to Newberry Caldera along the Oregon High Lava Plains (HLP) [e.g., Jordan et al., 2004].
 These silicic time progressive volcanic tracks have been the subject of a great deal of research, but consensus on their causes remains elusive. This is partly because of the range of tectonic factors that exist in the region, some or all of which may have played a role in their development. These tectonic factors begin with the potential inheritance of varied lithospheric structures due to the presence of terrane boundaries across the area. The location of the 87Sr/86Sr = 0.706 line (Figure 1) indicates that Proterozoic North America abuts against younger provinces along a roughly north-south trend adjacent to the Oregon-Idaho border, cutting to the northwest into Washington State and southwest into the Basin and Range [Fleck and Criss, 1985; Ernst, 1988]. Later accreted terranes identified in the area include the Blue Mountain Province [Dorsey and LaMaskin, 2007; Schwartz et al., 2010] and the Coast Range Basalt Province [Madsen et al., 2006], also referred to as the Siletzia terrane [Wells et al., 1998], which may extend east well into central and eastern Washington and Oregon [Schmidt et al., 2008; Humphreys, 2009; Schmandt and Humphreys, 2011; Gao et al., 2011]. In addition, active ongoing tectonic processes include the subduction of the Juan de Fuca and Gorda plates (remnants of the once much larger Farallon plate) and the formation and migration of the Mendocino Triple Junction as the San Andreas transform fault extended northward. Trench rollback along the Oregon/Washington coast is progressing at a rate of 35 mm/year relative to a Pacific hot spot reference frame [Schellart et al., 2008]. Basin and Range extension, which began in central Nevada ∼35 Ma, reached southern Oregon by ∼22 Ma [Scarberry et al., 2010]. Basin and Range extension today ends just south of the High Lava Plains physiographic province as evidenced by the decreasing amount of displacement along major fault lines in this area [Trench et al., 2012]. Concurrently, but independently, beginning at ∼7 Ma, the transtensional NW trending Brothers Fault Zone developed across much of the modern High Lava Plains physiographic province (Figure 1) [Jordan et al., 2004; Trench et al., 2012]. Geodetic studies indicate that much of central and eastern Oregon, Idaho, and northern California and Nevada exhibits block-like rotation about a pole near the Oregon/Washington/Idaho border, driven in part by Basin and Range extension, particularly in southern Oregon [e.g.,McCaffrey et al., 2007]. Trench et al.  suggest that the Brothers Fault Zone accommodates the greater rates of rotation of blocks south of the HLP from the comparatively stable blocks closer to the pole.
 For this study, we invert Rayleigh wave phase velocity dispersion curves to obtain vertical shear wave velocity profiles at 20,691 different locations in map view. These one-dimensional velocity profiles are subsequently combined to create a three-dimensional shear wave velocity model. We define the map view locations at 0.1° intervals between 38° and 47°N, and −125° and −108°W. At each location, we compile a dispersion curve using the phase velocity maps published inWagner et al. (a two-plane wave approach) andHanson-Hedgecock et al. (an ambient noise approach). Included in these dispersion curves are the phase velocities at each of 21 periods analyzed in the aforementioned studies: 8, 10, 12, 15, 20, 25, 28, 30, 33, 35, 40, 45, 50, 58, 66, 77, 91, 100, 111, 125, and 143 s. Below we outline specifics for each of the steps in this process: 1) determining phase velocities at each point, 2) creating a 1-D shear wave velocity reference model, 3) adjusting the reference shear wave velocity model at each point to reflect changes in crustal thickness, and 4) inverting for shear wave velocity and evaluating regularization and resolution.
2.1. Determining Phase Velocities at Each Point
 The analysis of earthquake-induced Rayleigh wave phase velocities ofWagner et al.  was performed at periods between 25 and 143 s. The ambient noise inversions for Rayleigh wave phase velocities of Hanson-Hedgecock et al.  provide constraints on shorter period phase velocities (with periods ranging between 8 and 40 s). Within the overlapping range of 25–40 s., Wagner et al.  analyze four different periods: 25, 28, 33, and 40 s. Hanson-Hedgecock et al.  calculate phase velocities for six periods in this range: 25, 28, 30, 33, 35, and 40 s. In Figure 2, we plot the results of both the ambient noise and earthquake induced phase velocity inversions for the four overlapping periods (25, 28, 33, and 40). In general, there is very good agreement between the results from the two methodologies, though some differences are still seen. For this study, we need to assign a phase velocity for each period to each point in map view. The question then arises whether to choose one map over the other, or to average the two results. In the case of the phase velocities at 25 and 28 s, the standard deviation of the results reported in Wagner et al.  are markedly higher than for the better resolved longer periods. This is largely because there are fewer data available at shorter periods for Rayleigh waves induced by earthquakes due to the more rapid attenuation of higher frequencies over the long distances traveled by the teleseismic plane waves. Ambient noise induced Rayleigh waves do not have this limitation. In comparing the phase velocity maps of the two methodologies at 25 and 28 s, the primary difference between the two is in the smoothness of the model: the earthquake induced Rayleigh wave velocity maps show smaller scale structures that are not necessarily reliable and do not appear in the ambient noise maps. Similar differences are observed at 33 s but not at 40 s. We decided that at the shorter periods (25, 28, and 33 s), the ambient noise results were more conservative than the earthquake Rayleigh wave results, and therefore use the ambient noise maps for determining phase velocities at those periods. At 40 s, we could see no clear reason to prefer one map over the other, so for this period, the phase velocities were averaged at each point in map view.
2.2. Creating a 1-D Shear Wave Velocity Reference Model
 In order to calculate a reference 1-D shear wave velocity model, we first average the phase velocities at all points for each period to obtain an average observed dispersion curve (Figure 3a). As described above, we used the phase velocities from the ambient noise inversions for periods between 8 and 35 s, and averaged the results between the ambient noise results and the earthquake induced surface wave results for the 40 s phase velocities. For periods longer than 40 s, we use the earthquake induced Rayleigh wave phase velocity maps alone. The starting model for this inversion, modified from the reference model of Wagner et al. , consists of 6 crustal layers with a total crustal thickness of 35 km, and 12 upper mantle layers extending to 420 km depth (Figure 3b). Layer thicknesses increase with depth due to the broadening depth sensitivity of the longer period surface waves. Ideal resolution is generally above 150 km depth with some resolution between 150 and 250 km depth (see section 2.4 for further discussion). Additional layers of increasing thickness are included up to 420 km depth in order to accommodate the very broad sensitivity kernels of the longest period surface waves.
 Using this starting velocity model, we invert the average dispersion curve to determine our reference 1-D shear wave velocity model using the method ofSaito . The resultant shear wave velocity model, and the predicted phase velocities for that model are shown in Figures 3b and 3a respectively. Regularization for this inversion is achieved by assigning a standard deviation to each phase velocity observation in the dispersion curve; for ambient noise phase velocities, we assume a standard deviation of 1% of the phase velocity at a given period. For periods of 40 s and longer, we use the average standard deviation from Wagner et al.  at that period (Figure 3a).
 For this study, we seek a reference shear wave velocity model that fits the observed phase velocities but makes as few assumptions as possible about existing lithospheric structure, especially with respect to the presence or absence of a fast mantle lid. A number of recent studies [e.g., Li et al., 2007; Abt et al., 2010; Levander and Miller, 2012] have analyzed variations in the depth to the lithosphere-asthenosphere boundary (LAB) across the western U.S. using both Ps and Sp receiver functions. These studies each show that the depth to the LAB (designated as the top of a region of lower shear wave velocities) varies between 40 and 140 km within our study area. The inversion described above does not produce a low velocity zone, but if we reduce the standard deviation of each of the phase velocities by an order of magnitude, the resultant shear wave velocity model (shown inFigure 3b) does show a subtle low velocity zone between 70 and 167 km depth. The overlying higher velocity lid has shear wave velocities below 4.25 km/sec. The phase velocities predicted for this velocity model are shown in Figure 3a. The fit of this less-damped model to the observed average phase velocities is slightly better than our more-damped model, but the differences are small compared to the average standard deviations at each period.
 In order to test the sensitivity of our results to the starting model, we repeat this inversion, this time using starting velocities from the TNA model of Grand and Helmberger which include a high velocity lid (4.4 km/sec) overlying a low velocity zone between 50 and 250 km depth. This inversion produces a low velocity zone starting at 45 km depth, with a maximum lid velocity of 4.29 km/sec. Despite the very different starting models, the result of this inversion is very similar to the lower-damping result calculated using theWagner et al.  starting model, and is nearly identical between 70 and 250 km depth. This suggests that the shear wave velocities across our study area in the upper most mantle are indeed on average substantially lower than those found by Grand and Helmberger . It also indicates that the depth to the LAB in our inversion results (70 km for the lower-damping model using theWagner et al.  starting model, 45 km for the result of the TNA starting model) is dependent on the starting model, and is therefore not well resolved.
 Given the inconsistency in the geometry of the low velocity zone, together with the limited improvement in fit to the observed average phase velocities for the results that do show a low velocity zone over the one which does not, we choose to use the more simple model that makes no assumptions about the depth to the LAB or low velocity zone. Also, as discussed in section 2.3, we need to be able to change crustal thickness at each point in map view while keeping our starting model as close to the reference model as possible. The absence of a mantle lid and low velocity zone greatly facilitates this process.
2.3. Crustal Thickness Adjustments to the Reference Shear Wave Velocity Model
 The next step is to invert the observed dispersion curves at each point in map view for the best fit shear wave velocity profile. Surface waves are generally poor at constraining Moho depth so we adjust the starting model at each point to reflect a priori constraints on crustal thickness (Figure 4). The Moho depths shown in Figure 4 were calculated from a combination of the receiver function results of Eagar et al.  and from the EARS catalog for regional Transportable Array stations [Crotwell and Owens, 2005] (see Wagner et al.  for details). This is the same Moho map used in Hanson-Hedgecock et al. , which in turn is a slightly smoothed version of the Moho map used in Wagner et al. .
 At each point, we first adjust the starting 1-D model slightly to take into account variations in crustal thickness. Our adjustment method preserves the original layer thickness parameterization as much as possible, meaning that the number of layers within the crust may vary but the number of layers between the surface and a given depth remains relatively constant. To accomplish this, we determine which layer contains the Moho at that point in map view based on the Moho depths shown inFigure 4. The part of the Moho-containing layer that lies above the Moho is added to the overriding layer thickness, and the part of that layer that is below the Moho is added to the layer below. The new thicknesses of the two layers are compared, and, in order to keep a constant total number of layers and to avoid overly thick layers, the thicker of the two is split in half. If layers are moved from crust to mantle, the new mantle layers are assigned the same velocities and density as the uppermost mantle layer in the starting model. If layers are moved from mantle to crust, the new crustal layers are assigned the same velocities as the deepest crustal layer in the starting model.
2.4. Inverting for Shear Wave Velocity: Evaluating Regularization and Resolution
 Using this starting shear wave velocity model and the phase velocities at that point in map view, we invert for a 1D shear wave velocity profile at that point. Regularization of these inversions is achieved by assigning a standard deviation to each phase velocity measurement. The goal is to find a regularization scheme that results in velocity deviations that vary gradually with depth but still allow significant enough variations to fit the observed phase velocities well [e.g., Yang et al., 2008a]. While standard deviations are not available for the phase velocity maps from Hanson-Hedgecock et al. , they were provided for the phase velocity maps of Wagner et al. . In order to take advantage of this information, we used the following regularization scheme for the models shown here: for periods longer than 40 s, we use a value equal to half of the standard deviation of the phase velocity at that point as published in Wagner et al. . For shorter periods, we take the lesser of 1% of the phase velocity or the value used for standard deviation for the 40 s phase velocity. This scheme was determined by testing a variety of regularization models, including constant values for standard deviation across all periods, and values weighted solely by the absolute phase velocities at each period. We chose this particular regularization method because it produced velocity deviations that varied gradually with depth, but still allowed for strong velocity gradients where required to fit the observed phase velocities. This regularization also takes into account the varying resolution of phase velocities at different points in map view.
 We evaluate how well our 3D shear wave velocity model fits the observed phase velocity maps by calculating predicted phase velocities at each point, and calculating the RMS misfit at each point, averaged over all periods. The result of this test can be seen in Figures 5. For most of the study area, errors are less than 0.02 km/sec, which compares favorably with the misfits of previous studies [e.g., Yang et al., 2008b]. We can also assess the sensitivities of our inversions to depth by looking at the diagonal of the resolution (R) matrix for each point in map view. In an ideal inversion, a perfectly resolved layer with no trade-offs with other layers would have a diagonal value of 1. Lower values on the diagonal imply trade-offs with adjacent layers. We plot the diagonal of the R matrix for each inversion (i.e., each point in map view) inFigure 6a, along with the mean and standard deviation for each layer. Layers 1–4 and 9–18 have constant depths and thicknesses. Layers 5–8 vary somewhat in depth and thickness due to varying Moho depths. This results in an increase in the range of diagonal R values, depending on whether the layer above and below the Moho ended up being thicker or thinner than the reference model layer. Thinner layers are more likely to trade off with adjacent layers than thicker layers, resulting in a lower value on the diagonal of the R matrix. Mean R-values are generally above 0.2 for depths between 5 and 140 km depths, but then decrease at 140 km depth to between 0.1 and 0.15. We observe another decrease at 250 km depth to R values between 0.01 and 0.07. R-values are also sensitive to the standard deviations assigned to the phase velocities. The R-values for layers 4 (15–20 km), 12 (107–137 km), 13 (137–187 km), and 16 (250–320 km) are shown inFigures 6b–6e. These show the highest resolution in areas within the High Lava Plains deployment, and lower resolution at the periphery where the phase velocities are less well constrained and have higher standard deviations in Wagner et al. .
 The prominent low velocities in both the crust and upper mantle associated with the Yellowstone Caldera at shallow depths, and with the larger Yellowstone/Snake River Plain at lower crustal and upper mantle depths are similar to those imaged by Wagner et al.  and Hanson-Hedgecock et al. , though some differences exist. Notably, while Hanson-Hedgecock et al.  found that very low velocities associated with the Yellowstone Caldera persisted nearly vertically through the full thickness of the crust, we image some deviation in the magnitude and location of the lower crustal low velocity anomaly (Figure 10). Given that the phase velocities for periods shorter than 40 s are the same as those used in the earlier study, the difference must be due to the addition of phase velocity information at periods 40 s and longer. The results of the present study are more consistent with previous results which also image a transition from slow to fast crustal material directly below the Yellowstone Caldera [e.g., Stachnik et al., 2008].
 The crust and uppermost mantle throughout most of southeastern Oregon are dominantly characterized by low shear wave velocities [e.g., Moschetti et al., 2007; Lin et al., 2009; Moschetti et al., 2010a; Pollitz and Snoke, 2010; Wagner et al., 2010; Obrebski et al., 2011; Yang et al., 2011; Hanson-Hedgecock et al., 2012]. Variations exist in the locations of the strongest anomalies with depth. Within the crust (Figures 7c–7eand cross section B-B′ inFigure 9), the low velocity anomaly is most prominent in the northernmost Basin and Range, primarily south of the High Lava Plains physiographic province. This low velocity anomaly extends from mid crustal depths to the Moho, and is most pronounced due north of the California/Nevada/Oregon border. In this area we observe shear wave velocity deviations of up to 8%, and absolute shear wave velocities in the lower crust below 3.4 km/sec. The area within the High Lava Plains physiographic province tends to have somewhat less pronounced crustal shear wave anomalies, especially in areas with the most post-CRB volcanic activity (Figure 9). At upper mantle depths (Figures 8a–8d), by contrast, the low velocity anomaly is slightly more pronounced under the High Lava Plains physiographic province, whereas the northwestern Basin and Range region exhibits a less prominent low velocity anomaly. The upper mantle low velocity anomaly along the main volcanic track of the High Lava Plains appears most pronounced between 50 and 100 km depth, shallowing from east to west toward Newberry Volcano, where the HLP low velocity anomalies merge with low velocity structures associated with arc volcanism (Figure 8and cross section A-A′ inFigure 9).
 In contrast to the low crustal velocities observed within the High Lava Plains and northwest Basin and Range, we observe a high velocity anomaly within the crust that extends from the western margin of the Owyhee Plateau east to the Wyoming border and Wasatch fault (Figures 7a–7e). The north-south extent of this anomaly ranges from the northernmost extent of the western Snake River Plain south into northern Nevada and Utah. The depth extent of this anomaly varies, depending on location. Along the SRP, the anomaly does not reach the surface and does not extend below ∼25 km depth (Figure 7eand cross section C-C′ inFigure 10). Within the Owyhee Plateau, the high velocity anomaly is particularly pronounced at 10 – 20 km depth (Figures 7a–7c), but also extends down to the Moho (Figures 7e and 8aand cross section C-C′ inFigure 10). Areas south of the OP and SRP also show this high velocity anomaly at various depths. At 10 – 15 km depth, high velocities are seen extending into the Lake Bonneville region in northwestern Utah. Below 20 km depth, this anomaly broadens across northern Nevada, although the highest amplitude anomalies remain in northeasternmost Nevada and northwestern Utah. The area in northwestern Utah and northeastern Nevada with the thickest and most prominent high velocity anomaly corresponds spatially with a regional long-wavelength Bouguer gravity high (Figure 7f) [Kucks, 1999; DeNosaquo et al., 2009]. The D′-D cross-section inFigure 10 shows a profile across the gravity anomaly map from Figure 7f. This cross-section shows a striking spatial correlation between the topographic lows of the SRP and Great Salt Lake Basin area with relative Bouguer gravity highs. In between the two is an area of higher topography with somewhat lower gravity anomalies. In the seismic cross-section, this region represents the transition from the shallower, thinner high velocity anomaly within the SRP to the thicker, deeper high velocity anomaly further south. It is also underlain by the strong mantle low velocity anomaly that extends the length of the YSRP.
 In this paper, we will focus on two major features: the low velocity structures associated with the High Lava Plains (Figures 7–9), and the high velocity crustal anomaly that we observe beneath the SRP, Owyhee Plateau, and parts of northern Utah and Nevada (Figures 7, 8, and 10). Both of these observations are roughly consistent with most previously published velocity models, but the details and tectonic implications of each have not previously been fully investigated.
4.1. The High Lava Plains Low Velocity Anomaly
 Two aspects of the low velocity anomaly across southeastern Oregon are particularly striking: (1) the most pronounced low velocities in the lower crust lie south of the main trend of HLP volcanism; and (2) the most pronounced low velocities in the uppermost mantle are not located below the crustal low velocities, but rather to the north beneath the HLP physiographic province. These results are not entirely unexpected. The uppermost mantle anomalies observed are similar to those in Wagner et al. . The very low shear wave velocities in the crust immediately above the Moho in the northwestern Basin and Range are consistent with the results of Hanson-Hedgecock et al. . These results are also consistent with receiver functions from this area that have mid-crustal negative polarity arrivals consistent with a decrease in velocity with depth [Eagar et al., 2011]. Receiver function H-κ stacking indicates unusually high Poisson's ratio in this area as well [Eagar et al., 2011] which are also consistent with our observed low shear wave velocities.
 Combined with observed high heat flow [Blackwell and Richards, 2004] and low resistivity [Patro and Egbert, 2008] throughout the area, one possible explanation for the very low velocities in the lower crust of the northwestern Basin and Range is the presence of small degrees of partial melting. For crustal compositions, an 8% reduction in shear wave velocities would correspond to a melt fraction of ∼3% [Watanabe, 1993; Takei, 2000; Caldwell et al., 2009], assuming the entire velocity anomaly is due to partial melt alone. This would suggest the presence of up to 3% partial melting in the lower crust beneath the northwestern Basin and Range, which is roughly consistent with the 2% found by Eagar et al. based on the high Vp/Vs ratio determined using receiver function H-κstacking. If this anomaly is indeed due to partial melting, the location of this lower crustal partial melting is notably not centered on the area in the High Lava Plains with the most significant amount of post-flood-basalt volcanism.Eagar et al.  suggest this may be evidence that these represent zones of “undrained” partial melt in the crust, whereas the area further north along the HLP physiographic province had erupted most or all of the partial melt previously present in the crust.
 One possible explanation for why the melting was able to erupt further to the north but not within the northern Basin and Range is the Brothers Fault Zone (BFZ), which may have acted as a conduit for the magma to escape. The BFZ began to develop sometime between 7.5 and 5.7 Ma, synchronously with, but separately from Basin and Range extension further south [Jordan et al., 2004; Trench et al., 2012]. This roughly coincides with a change in relative plate motions as suggested by Jordan et al. , resulting in a block rotation of much of the study area about a pole located near the Oregon/Washington/Idaho border [e.g., McCaffrey et al., 2007]. The NW trending faults of the BFZ delineate a small circle about this pole, separating Basin and Range extension from the relatively undeformed block closer to the pole. The spatial correlations between the BFZ, the main volcanic trend of the HLP, and the upper mantle low velocity zone, together with the lack of correlation with the most profound lower crustal low velocity anomalies suggest two possible hypotheses: (1) The BFZ formed in this location because of a pre-existing zone of weakness, seen today as a low velocity zone in the uppermost mantle, resulting in a focusing of volcanism along the HLP; or (2) the formation of the BFZ along the northern margin of the Basin and Range resulted in an area of shearing, weakening, and/or heating in the uppermost mantle, as indicated by the low velocities and apparent lack of mantle lithosphere observed today. The pattern of volcanism along the HLP indicates four major pulses of volcanic activity between 7.8 and 2 Ma [Trench et al., 2012]. These volcanic events may have produced a temporary localized weakening of the crust which allowed the BFZ to develop in that area. However, after melt extraction, the residual more mafic crust would result in an increase in crustal strength [Brace and Kohlstedt, 1980; Trench et al., 2012], perhaps contributing to the westward migration of silicic volcanism along the HLP. Long et al. suggest that the temporal pattern of the HLP silicic volcanism can be explained by patterns of trench roll-back and flow around the southern edge of the subducted slab. They argue that trench roll-back, together with a steepening of the subducted slab, would have initiated ∼20 Ma, resulting in a pulse of upwelling that produced the mid-Miocene flood basalts and thinning of the mantle lithosphere across much of the area. Ongoing trench retreat, together with toroidal flow around the southern edge of the subducted slab produced a W-to-E flow field that helps to explain both the migration of HLP volcanism and the presence of large magnitude E-W oriented SKS splitting [Long et al., 2009]. While these processes also produced small degrees of partial melting in the lower crust in areas south of the HLP, the quantities of magma were sufficiently small, or conduits to the surface were lacking, such that magma extraction did not occur, and the lower crust in this region remains today as a profoundly low velocity zone in the northwestern Basin and Range.
4.2. The Crustal High Velocities of the Snake River Plain/Owyhee Plateau/Great Salt Lake
 The broad extent of the high velocity crustal anomaly in our model is similar to anomalies seen in crustal velocity maps of a number of previous tomographic studies [e.g., Pollitz and Snoke, 2010; Yang et al., 2011]. The notable exception is Moschetti et al. [2010a, 2010b], though other more recent works by co-authors do show this feature [e.g.,Yang et al., 2011]. The principle difference between Moschetti et al. [2010a, 2010b] and other studies is the inclusion of Love waves and radial anisotropy. This might suggest that our observed fast shear wave velocities in the upper mantle are due to radial anisotropy, but the results of Moschetti et al. [2010b] do not show a corresponding high radial anisotropy anomaly in their upper mantle models in this area. Evidence for the presence of high velocity material in the crust across the area can also be seen in other seismic observables. The high velocity anomaly directly above the Moho (Figure 7e) near the OP is reminiscent of Eagar et al. [2011, Figure 10], which shows the amplitudes of Ps converted phases from Moho depths: the areas in which we observe high supra-Moho velocities exhibit particularly low Ps converted phase amplitudes in the Eagar study, consistent with a decreased seismic impedance contrast across the Moho due to elevated lower crustal velocities, as well as perhaps a sharply dipping Moho.
 The Owyhee Plateau (OP) itself is seen in our models as having high velocities from the surface to the Moho. A sharp boundary between the OP and areas to the west can be seen in Moho thicknesses, as the OP has a distinctly thicker crust than the adjacent HLP [Eagar et al., 2011]. Unlike the surrounding areas, the OP experienced comparatively little volcanism since the mid Miocene, with most of the volumetrically significant volcanism occurring along the western and northern margins [Shoemaker, 2004]. Hanson-Hedgecock et al.  suggest that this anomaly could be due to Precambrian lithosphere that was further depleted during the eruption of the Columbia River flood basalts. We note that unlike adjacent areas (discussed below), the OP is characterized by comparatively high elevations. This is consistent with the increased crustal thickness in the area, but may limit the amount of dense mafic restite that could present in the mid and lower crust without isostatically reducing the surface topography.
 The portion of the high velocity anomaly that has been most investigated is the mid-crustal feature within the bounds of the Snake River Plain. In our models, this anomaly is clearly not just limited to the SRP, but extends to the south into northwestern Utah where it thickens substantially (Figure 9, cross section D-D′). Previous studies have focused solely on the portion of this anomaly located within the SRP by investigating cross-sections that either entirely or mostly exclude the region of the high velocity anomaly we observe south of the SRP. For example,Peng and Humphreys found evidence for a high velocity mid-crustal layer that does not extend beyond the margins of the SRP. The seismic array used, however, was deployed in a transect north and east of the Wasatch fault, beyond the easternmost extent of our observed high velocities. Their work builds on the earlier reflection/refraction study along a similar transect bySparlin et al. who found evidence for a high velocity, high density mid-crustal layer localized within the margins of the SRP. Similar results have been found in subsequent seismic studies [e.g.,Schutt et al., 2008; Stachnik et al., 2008], but these also only investigate the portion of the Snake River Plain northeast of the Wasatch Fault.
 The most common explanation for the mid-crustal SRP high velocity anomaly is a gabbroic sill or series of sills [Shervais et al., 2006] left behind by the voluminous silicic eruptions along the SRP [Sparlin et al., 1982; McQuarrie and Rodgers, 1998; Peng and Humphreys, 1998; Schutt et al., 2008; Stachnik et al., 2008; DeNosaquo et al., 2009; McCurry and Rodgers, 2009; Rodgers and McCurry, 2009]. This high density structure is responsible for the isostatic subsidence of the SRP relative to the surrounding Basin and Range [McQuarrie and Rodgers, 1998]. However, a number of studies have also suggested that some amount of crustal material must have been extruded out of the SRP into adjacent areas. McQuarrie and Rodgers  modeled the structure of the crust in this region by looking at isostatically induced flexure along the northern margin of the SRP. Their results indicate that isostatic compensation of a mid crustal high density structure occurs in the lower crust rather than the upper mantle, accommodated by lateral extrusion of crustal material to areas north and south of the SRP. A similar conclusion is reached by McCurry and Rodgers  and Rodgers and McCurry  who use petrological and kinematic constraints respectively to argue for the extrusion of restite and densified crust into areas adjacent to the SRP. Yuan et al.  argue that crustal material has been extruded, in their case to the north, eminating from the northeastern end of the SRP near the Heise caldera field. Their conclusions are based on receiver function derived crustal thicknesses that indicate increased Moho depths, not just along the SRP volcanic track, but also to the north in the area described above. DeNosaquo et al. investigated the mid-crustal sill structure by combining gravity data with constraints from seismic imaging and seismicity patterns among others. They focus on the same cross section asPeng and Humphreys that does not intersect the southern extension of our high velocity anomaly but does however include the northeasternmost corner of the gravity high. Based on the extent of their observed gravity anomaly, they argue for a slight widening of the proposed sill structure caused by lateral extrusion of the mid-crustal sill accommodated by normal faults present in the northern Basin and Range. They also attribute a southward deflection of the seismogenic parabola surrounding the YSRP to this widened sill, arguing that the seismicity is the product of flexure induced by sill-related isostatic subsidence.
 We propose that a lateral extrusion of high density, high seismic velocity material from the central Snake River Plain may be an explanation for our observations in northwestern Utah and northeastern Nevada. The adjacent areas of the SRP, including the Bruneau-Jarbidge and Twin Falls volcanic centers, produced several times the volume of silicic magmatism observed at Yellowstone [Perkins and Nash, 2002; Bonnichsen et al., 2008], and presumably also produced correspondingly more gabbroic residual. No difference in crustal thickness or in the thickness of the mid-crustal high velocity zone within the SRP is seen, however, consistent with mass removal due to lower crustal flow. The area of the northeastern Basin and Range due west of the Wasatch fault is known to have undergone a great deal of extensional thinning, beginning in the late Cretaceous [Wells and Hoisch, 2008] and continuing into the Cenozoic [Egger et al., 2003] (Figure 4). This extension could have helped accommodate the influx of additional material, similar to, but on a larger scale than what was proposed by DeNosaquo et al. . This influx of high density material could account for the relatively low topography across the area, and may explain the relative quiescence of the seismic parabola surrounding the SRP west of the Wasatch fault [Anders et al., 1989]. While thinner crust might be used to explain lower topography as well, we note that similarly shallow or shallower Moho depths are seen across much of the Basin and Range (Figure 4) but the area of low topography is restricted to the area with the increased Bouguer gravity anomaly. Future modeling of the observed gravity and seismic anomalies is required to determine the density, volume, composition, and origin of this high velocity crustal material.
 Our inversions for the detailed shear wave velocity structure of the crust and upper mantle in the Pacific Northwest show low velocities in the crust and upper mantle in the High Lava Plains, and high velocities in the crust across the Owyhee Plateau, Snake River Plain, and northeastern Basin and Range. The anomalies in southeastern Oregon indicate that the lower crust in the northwestern Basin and Range may contain up to 3% partial melt, whereas the adjacent High Lava Plains physiographic province that has experienced more volcanism since the mid-Miocene shows no such anomaly. This may be due to the presence of the Brothers fault zone, which allowed partial melts to reach the surface in this area.
 The crustal high velocity anomalies observed within the Owyhee Plateau, Snake River Plain, and northeastern Basin and Range vary in depth and intensity. The Snake River Plain high velocities are constrained to the mid-crust, whereas the Owyhee Plateau and areas of northwestern Utah have high velocities that extend up to, and perhaps even into, the mantle. The high velocity crustal anomaly in the northeastern Basin and Range, particularly in northwestern Utah, is well correlated with a regional Bouguer gravity high. This high velocity crustal anomaly may be due to lateral flow of mafic restites from the Snake River Plain into this very extended region.
 Data for this project come in part from the High Lava Plains seismic experiment, funded by NSF Continental Dynamics EAR-0507248 (MJF) and EAR-0506914 (DEJ), as well as from EarthScope's USArray Transportable Array (http://earthscope.org). LSW was supported in part through NSF grant EAR-0809192. We would like to thank the wonderful people at the IRIS PASSCAL Instrument Center and at the IRIS Data Management Center for their ongoing support. We would also like to thank Jenda Johnson, Steven Golden, and the many people who helped install and service the 118+ stations in this deployment. Also thanks to Allen Glazner and John Bartley for their helpful suggestions on Basin and Range tectonics. Finally, we wish to thank the many ranchers and other landowners who generously and freely agreed to host seismic stations for the HLP experiment on their property. We would also like to thank Brandon Schmandt and one anonymous reviewer for their helpful comments and suggestions.