We used more than 40 000 S-receiver functions recorded by the USArray project to study the structure of the upper mantle between the Moho and the 410 km discontinuity from the Phanerozoic western United States to the cratonic central US. In the western United States we observed the lithosphere–asthenosphere boundary (LAB), and in the cratonic United States we observed both the mid-lithospheric discontinuity (MLD) and the LAB of the craton. In the northern and southern United States the western LAB almost reaches the mid-continental rift system. In between these two regions the cratonic MLD is surprisingly plunging towards the west from the Rocky Mountain Front to about 200 km depth near the Sevier thrust belt. We interpret these complex structures of the seismic discontinuities in the mantle lithosphere as an indication of interfingering of the colliding Farallon and Laurentia plates. Unfiltered S-receiver function data reveal that the LAB and MLD are not single discontinuities but consist of many small-scale laminated discontinuities, which only appear as single discontinuities after longer period filtering. We also observe the Lehmann discontinuity below the LAB and a velocity reduction about 30 km above the 410 km discontinuity.
Lithospheric plates, including thick old cratons, translate over thousands of kilometers over the viscous mantle. However, relatively little is still known about the internal structure of cratons and the transition between the craton and the convecting mantle. Even after more than half a century since the general acceptance of plate tectonic theory, the lithosphere–asthenosphere boundary (LAB) below cratons is still thought to be “elusive” (Eaton et al., 2009) and an additional velocity drop frequently observed in seismic data in the shallow cratonic lithosphere (mid-lithospheric discontinuity, or MLD) is referred to as “enigmatic” (Karato, 2012). These descriptions not only apply to the petrophysical properties of the LAB and MLD but are also a result of still inadequate seismological imaging. The lithosphere–asthenosphere system was originally a mechanical definition (Barrell, 1914) and is not a seismic definition. However, we are using the name LAB here for seismic velocity reductions observed near 200 km depth in cratons and near 100 km depth in oceans and Phanerozoic regions by tomography and receiver functions (e.g., Yuan and Romanowicz, 2010). Tomography is not directly sensitive to discontinuities, and therefore the transition from lithosphere to asthenosphere is derived from the velocity–depth functions or its vertical gradients (e.g., Yoshizawa, 2014). Only a few tomography studies observed in cratons a shallow low-velocity zone near 100 km depth, which could be related to the MLD (e.g., Lekic and Romanowicz, 2011).
The alternative view of the deep structure of cratons, independent of the lithosphere–asthenosphere model, defines a tectosphere with the keel of the cratons reaching down to about 400 km (Jordan, 1975). This model is based on different versions of the tomography method (see Jordan and Paulson, 2013, for a summary). The tectosphere may be decoupled from the convecting mantle by a low-viscosity layer directly above the 410 km discontinuity. A possibly related velocity reduction above the 410 is observed globally by Tauzin et al. (2010).
The LAB and MLD are mainly observed with receiver functions. Summaries of their global distribution are given by Rychert and Shearer (2009), Rychert et al. (2010), Fischer et al. (2010) and Kind et al. (2012). Since the MLD and in some places the cratonic LAB are relatively well observed with seismic body waves, both discontinuities must be relatively sharp. Rychert et al. (2007) deduced from P-receiver functions a sharpness of 11 km for the LAB at the east coast of the US. Li et al. (2007) obtained a sharpness of about 20 km in the western United States using S-receiver functions. Rychert et al. (2007) concluded that the sharpness of the LAB excluded temperature variation as the single cause of the LAB. Karato (2012) suggested the grain-boundary sliding model as the cause of the MLD and LAB. This model predicts a strong MLD and weak LAB below cratons. According to Yuan and Romanowicz (2010) the North American craton consists of two layers with different chemistry and anisotropy, with the upper layer reaching to about 100 km depth being the Archean lithosphere. Selway et al. (2015) reviewed the mechanisms which could cause the MLD.
It would be difficult to show the existence of the MLD in Phanerozoic regions since here the LAB would also be expected at a similar depth. Therefore it is interesting to study the structure of the MLD and LAB at continental collision zones as in the western US. Abt et al. (2010) and Kumar et al. (2012a, b) observed in almost the entire United States only a shallow negative discontinuity near 100 km depth. This discontinuity varies to some extent in depth but nowhere does it reach 200 km depth. In the western United States and at the east coast this signal was called the LAB while in the central cratonic United States it was called the MLD by Abt et al. (2010), and the LAB by Rychert and Shearer (2009) and Kumar et al. (2012a, b). Levander and Miller (2012) have mapped the Phanerozoic LAB in the western US. More detailed regional studies in the western United States are published by Rychert et al. (2005, 2007), Li et al. (2007), Hopper et al. (2014), Hansen et al. (2013), Foster et al. (2014) and Lekic and Fischer (2014). Rare observations of the cratonic LAB near about 200 km depth have been obtained by Foster et al. (2014) in the US. Similar observations in other cratons have been obtained in Canada (Miller and Eaton, 2010), Scandinavia (Kind et al., 2013) and South Africa (Sodoudi et al., 2013).
The structure of the mantle lithosphere in western North America was formed by the collision of the Farallon plate with the Laurentia craton and was first resolved by tomographic studies. The subducted Farallon plate is visible in the upper and lower mantle, even below the eastern United States (Grand, 1994; Schmandt and Lin, 2014). The collision with the Precambrian North American craton about 50 million years ago during the Laramide orogeny tore and broke the Farallon plate. An example is the “big break” in the western United States of Sigloch et al. (2008) and Sigloch (2011) and references therein, or the vertical high-velocity “curtain” near the longitude of Yellowstone (Schmandt and Humphreys, 2011). The boundary of the craton follows approximately the Rocky Mountain Front (see for example Yuan et al., 2011, and Zheng and Romanowicz, 2012).
We use the S-receiver function technique (meaning S-to-P conversions) to image seismic discontinuities between the crust–mantle boundary (Moho) and the seismic discontinuity at 410 km depth (see, e.g., Yuan et al., 2006, or Kind et al., 2012, for a description of the technique). The receiver function method determines the response of the Earth structure below a seismic station. Teleseismic waves arriving at a station are scattered by the underlying discontinuities, causing conversions and multiple reflections, which lead to images of the layered structure below the station.
The most important step in receiver function processing is stacking of many seismic traces in order to enhance the weakly converted waves. The simplest approach is to align many records for a given station with respect to a main phase, for example S, after amplitude and sign normalization, and sum these traces. The summation is performed for each component separately (vertical, radial and transverse). We rotated the components by theoretical back azimuth and incidence angle of the S phase and obtained approximately the P, SV and SH response of the medium below the station. As the travel-time moveout between the main phase and the scattered phase depends on epicentral distance, this kind of summation can, without a moveout correction, only be done in narrow epicentral distance windows (Shearer, 1991). A distance moveout correction permits summation over larger distance ranges (Yuan et al., 1997). However, a velocity model is required. Since the moveout correction can only be made for one type of phase at a time (for example P-to-S conversions, or multiples), signals traveling with different slownesses will be canceled by stacking.
Map of North America with seismic stations (triangles) from USArray
(
Traditionally, deconvolution is used to equalize (and approximately remove) the source signals of the different earthquakes before stacking. For example, in P-receiver functions, a window around the P signal on the vertical component is used to deconvolve the radial component. We should also mention that, theoretically, deconvolution is not required in the receiver function technique. Summation of plain seismograms leads to similar results (see Kumar et al., 2010; Bodin et al., 2014). Here we applied deconvolution since we obtained better signal-to-noise ratios. Another important step in receiver function processing is the migration from the time domain into the depth domain using a known velocity model. For migration, the seismic energy is back-projected along the ray path within a given model and stacked, assuming that the energy is distributed in the Fresnel zone (Jones and Phinney, 1998; Kosarev et al., 1999; Yuan et al., 2006). The one-dimensional IASP91 reference model is used for moveout correction and migration. Both moveout correction and migration are relatively insensitive to the model used.
We use S-receiver functions in the present study, which have a significant advantage over P-receiver functions for upper mantle studies. In S-receiver functions, the direct conversions arrive before the S signal while the crustal multiples arrive after the S signal. In P-receiver functions, both direct conversions and multiples arrive after the P signal. Multiples in P-receiver functions frequently overwhelm direct conversions and make it difficult to identify the true structure. However, S-receiver functions have other problems which need to be considered. In the next section we include a discussion of some problems of the interpretation of the wave field of S precursors using the example of USArray data.
We obtained the data from the open-access IRIS archive in Seattle, Washington
(
We have inspected manually over 200 000 records from events with magnitudes
greater than 5.7 within the epicentral distance range of 60–85
Epicenters of 1102 earthquakes used in our study. Black triangle marks the center of the network used. Black circles with labels indicate epicentral distances from the center. Not all earthquakes have been recorded at all sites since the USArray stations were moved every 2 years.
Finally more than 40 000 records have been selected for our study. Next we
will discuss the wave field of the S precursors of all available data in
different graphical presentations. The data in most of the following figures
are low-pass-filtered with 8 s corner period. In Fig. 3a all traces are shown
as a function of the epicentral distance. The traces are summed within
0.5
Display of binned S-receiver function traces along a
There are obviously several seismic phases visible in Figs. 3–5. Seismic phases marked red (positive) are caused by a discontinuity with downward increasing velocity. Blue phases (negative) mark downward decreasing velocity. All phases converted from the direct S phase to P are marked with black labels in Figs. 3–5. These are the conversions from the Moho, two blue phases labeled LVZ1 and LVZ2, the conversion from the discontinuity at 410 km depth (marked “410”), a red discontinuity marked “Lehmann” and another blue discontinuity following closely the 410 signal and marked “LVZ410”. The Moho is not the focus of our study. The LVZ410 is observed in P- and S-receiver functions by a number of authors in different parts of the world (e.g., by Schaeffer and Bostock, 2010, in northwestern Canada; by Vinnik et al., 2010, in California and globally by Tauzin et al., 2010). It is interpreted by the presence of water causing partial melt. Jordan and Paulson (2013) discuss the role of this discontinuity for decoupling the thick continental tectosphere (which extends below the LAB) from the convecting mantle. The Lehmann discontinuity is widely considered as the bottom of the asthenosphere. A global study of the Lehmann discontinuity is given by Gu et al. (2001).
Display of binned S-receiver functions as a function of the
back azimuth (see Fig. 2) of the source of each record, independent of
epicentral distance. Only stations on the cratonic part of the United States (east of
110
Theoretical travel-time curves of S precursors at the above mentioned discontinuities are marked by dashed black lines in Fig. 3a. They are computed using the IASP91 global reference model with three negative discontinuities (LVZ1, LVZ2 and LVZ410 at 90, 170 and 380 km depth, respectively) and one positive discontinuity (Lehmann at 270 km) added. The Moho is set at 35 km depth. LVZ1 and the 410 are clearly observed in Figs. 3–5. The LVZ2 and the Lehmann discontinuity are better observed in Fig. 4.
Additional seismic phases, besides the ones discussed so far, are only visible
in Fig. 3a but not in Figs. 4 and 5. Theoretical travel-time curves have been
computed to explain these phases. They are marked in green. We see clear
negative SKS660p and ScS660p phases from the 660 km discontinuity cutting
through all other phases prior to the S arrival. These phases are strong at
68–73
Theoretical seismograms (vertical component) computed with the reflectivity method for comparison with the observed data in Fig. 3a. The strong SPn phase after S agrees well with the data in Fig. 3a. A disagreement between computed and observed seismograms is the source-side Pn in front of the S signal (this figure) which is not in the data (Fig. 3a). This is a phase traveling as P along the surface on the source side and is continuously radiating S waves downward. At the receiver side these phases are converted back to P and travel again horizontally along the uppermost mantle and are observed on the vertical component. These phases are not observed in the real data probably because of heterogeneities in the real Earth.
Theoretical seismograms for different structures at the source and
receiver sides.
There have been concerns that higher-order P multiples could influence S precursors on the P component (Bock, 1994; Wilson et al., 2006). However, we do not see such phases in the complete observed precursor wave field in Fig. 3a. The reason is probably that the upper mantle of the Earth is too heterogeneous to permit the efficient propagation of higher-order P multiples. In Fig. 6 we have computed S precursors for the one-dimensional IASP91 Earth model (vertical component). We used the version of the reflectivity method by Kind (1985), which is an extension of the original method by Fuchs and Müller (1971) for different source and receiver structures. This version avoids P multiples in the theoretical seismograms, which cause a high noise level in front of the S phase making it difficult to obtain good S precursors. An approximated spike was used as the source time function. The strong SPn and SsPmp phases after S agree well with the data in Fig. 3a. A disagreement between computed and observed seismograms is the source-side Pn in front of the S signal (Fig. 6), which is not in the data (Fig. 3a). This is a phase traveling as P along the surface on the source side and continuously radiating S waves downward. At the receiver side these phases are converted back to P and travel again horizontally along the surface. These phases are not observed in the stacked S-receiver functions (Fig. 3a) because of stacking of records from many regions with different structures and source depths. The same phases marked green in the data (Fig. 3a) are also marked green in the theoretical seismograms in Fig. 6. Note that the IASP91 model has neither an upper-mantle low-velocity zone, the Lehmann discontinuity, nor the negative discontinuity above the 410. Therefore these phases are not computed.
In order to point out another possible source of disturbing S precursors, we have computed theoretical seismograms similar to Fig. 6 for different crustal models at the source and receiver sides (see Fig. 7). A narrower slowness integration window was used, which excludes source-side Pn. There is a 50 km thick crust at the source side in Fig. 7a and no crust is included at the source side in Fig. 7b. Receiver-side crust is in both cases 40 km thick. The rotated L component is shown, which carries practically only P energy. We see ScS and SKS conversions at the 410 and 660 discontinuities, which cross the Sp conversions at the Moho and at the 410. In Fig. 7c ray paths of Smp with a conversion at the receiver-side Moho (Fig. 7b) and of PmSmp with an additional P-to-S conversion at the source-side Moho (Fig. 7a) are shown. The PmSmp phase is visible in Fig. 7a as a negative precursor of the Moho conversion, which might be mistaken in the real data as a S-to-P conversion at a negative discontinuity below the Moho underneath the station. In receiver functions, however, this phase is reduced due to summation of many events with different source depths and source-side structures. Care should be taken if single seismograms are used because it seems impossible to identify all phases uniquely in these cases.
Depth-migrated west–east S-receiver function profiles (see location
map at the top of the figure). The width of each profile is 2
The same data and same correlation marks as in Fig. 8 but summed
along broader south–north and west–east profiles. The profile width and
orientation is shown in the location maps at the top of the figure.
Region
To image the topography and structure of such discontinuities in the study area we plotted in Figs. 8 and 9 several depth-migrated profiles and in Figs. 10 and 11 several profiles in the time domain. The width of the profiles is two degrees or more of latitude or longitude. The profiles cannot be chosen much narrower because the number of traces available for summation would be too small. Each of the traces in Figs. 10 and 11 was obtained by summation of several hundred traces. The depth domain profiles are chosen along great circles and the time domain data along latitude and longitude. The traces in the time domain profiles (Figs. 10 and 11) are selected by the location of their S-to-P piercing points at 200 km depth. To a first approximation the precursor time may be multiplied by a factor of 10 to obtain the corresponding depth. The IASP91 model is used for depth migration in Figs. 8 and 9. The first and perhaps most important task of the data interpretation is the recognition of patterns in the data. We have chosen variable width of the profiles in the different figures to show the variability of the correlations in dependence on the profile width. In the mantle lithosphere below the Moho we observe several clearly correlatable blue signals (velocity decrease downward). We interpreted these signals as being caused by the LAB in the western United States (marked with magenta lines) and the MLD in the central United States (marked with yellow lines) and the cratonic LAB in the eastern United States (marked with green lines). In the following we discuss the topography of each of these negative discontinuities in the mantle lithosphere.
North of about 46
North of about 46
The MLD is marked yellow in the figures and it is visible in most profiles in
the central United States near 100 km depth where it is expected. It is poorly seen
south of about 37
The signals we interpreted as cratonic LAB near 200 km depth are marked
green in Figs. 8, 9 and 10. These signals are relatively weak in some of the
narrow profiles in Fig. 8. In the wider profiles in Fig. 9c, f and g, the
cratonic LAB signals are clearer. In the time domain data in Fig. 10
(33–38
In Fig. 11 the same data are shown as in Fig. 10, except that no filter is
applied apart from the deconvolution. The data are much shorter period than
in Fig. 10. If a discontinuity is sharp we expect that its signal also
becomes sharper if a shorter period filter or no filter is applied. This seems not to
be the case for all the upper mantle discontinuities observed in this study
such as the western and cratonic LAB and MLD. The discontinuities seem to
dissolve in a sequence of sharper discontinuities with a very scattered
appearance with sometimes a relatively short correlation length. A kind of
scattered lamella structure with stepwise decreasing velocity could cause
such signals. This means that these discontinuities are in reality transition
zones with stepwise decreasing velocities. A lamella structure of the upper
mantle has been observed in long-range controlled source profiles for a long
time, for example in Phanerozoic western Europe (see Kind, 1974) and in the
cratons of northern Eurasia (see Mechie et al., 1993) and North America (see
Thybo and Perchuc, 1997). No phases are marked in Fig.11 in order not to bias
the reader. Generally the same phases as in Fig. 10 are also visible in
Fig. 11, however, as groups of scattered signals. The signal with the largest
amplitudes is the western LAB, especially in Fig. 11 (33–38
West–east S-receiver function profiles in the time domain. The latitude of the distribution of piercing points in 200 km depth is marked in each panel. The data are filtered with an 8 s low-pass filter. Correlated phases are marked with the same colors as in Figs. 8 and 9, except that here black represents the LAB of the Juan de Fuca plate.
The same profiles as in Fig. 10 but with no filter applied.
Visualization of the interfingering of the colliding Farallon LAB in
the west with the Laurentia MLD in the east. The cratonic LAB is left out of
the figure. The profiles in Figs. 9a, b, c and 10 (48–50, 41–43, and
33–38
How do our observations agree with earlier seismic images of the mantle
lithosphere in the western and central US? Levander and Miller (2012) also
used S-receiver function data from USArray in the same area. They also
observe the break in the LAB along the Sevier thrust belt. However they
interpret the deep velocity drop east of the Sevier thrust belt as the
cratonic LAB of the Laurentia plate. They do not report on the MLD and a
west-dipping structure between the Rocky Mountain Front and the Sevier
thrust belt. Hopper et al. (2014) observed in the Yellowstone region the same
break in the mantle lithosphere, also using S-receiver functions. They
observed east of Yellowstone a faint shallow MLD but no west-dipping
structure. Hansen et al. (2013) observed the LAB below the Colorado Plateau
at 100–150 km depth and east of it at 150–200 km depth. This agrees with
our LAB observations along 38
Sodoudi et al. (2013) found in South Africa MLD signals in S-receiver
functions showing azimuthal anisotropy. They observed a sign change of the
S-to-P converted signal from about 85 km depth with a periodicity of
180
Tauzin et al. (2010) observed a nearly global velocity reduction at about 350 km depth without any relation to surface tectonics. The cause of this low-velocity layer is thought to be partial melt caused by dehydration (Bercovici and Karato, 2003). We observed a similar discontinuity in the USArray S-receiver functions (see, e.g., Figs. 4 and 9). Jordan and Paulson (2013) suggested that a low-viscosity zone directly above the 410 discontinuity decouples the thick cratonic tectosphere from the flowing mantle. Looking at Fig. 9 it seems that the strongest signals of the phase LVZ410 are observed in the western United States where the LAB occurs at about 100 km depth. The signal is weaker in the cratonic United States (Fig. 9b, c, g). This means that the velocities directly above the 410 are lower in Phanerozoic regions than in cratonic regions of the US. This supports the tomographic results summarized by Jordan and Paulson (2013) of the cratonic tectosphere having higher velocities down to the 410 km discontinuity.
We have been able to image with S-receiver functions the major seismic
discontinuities in the upper mantle below large regions of the western and
central United States. In the upper 200 km we see complex structures of
the western LAB and the MLD (Fig. 12). The east-dipping LAB interferes with
the west-plunging MLD in a complicated manner. We interpret these structures
as being caused by the continental collision of the Farallon plate and the
Laurentia plate. The MLD appears to be deformed in this collision in a
similar way as could be expected for a shallow non-cratonic LAB. This could
mean that the Archean lithosphere of the craton (Yuan and Romanowicz, 2010)
was deformed during the collision with the Farallon plate. The western LAB
dips partly far to the east to the mid-continental rift system, where it
could be mistaken for the cratonic LAB. The deep cratonic LAB near 200 km
depth is weakly observed at the eastern end of the considered area (from
about 90–82
This research was supported by the Deutsche Forschungsgemeinschaft. The
facilities of the IRIS Data Management System, and specifically the IRIS Data
Management Center, were used for access to waveform and metadata required in
this study (