Rocky Mountain evolution: Tying Continental Dynamics of the Rocky Mountains and Deep Probe seismic experiments with receiver functions

Transcription

1 JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 114,, doi: /2008jb005726, 2009 Rocky Mountain evolution: Tying Continental Dynamics of the Rocky Mountains and Deep Probe seismic experiments with receiver functions Eva-Maria Rumpfhuber, 1,2,3 G. Randy Keller, 2 Eric Sandvol, 4 Aaron A. Velasco, 1 and David C. Wilson 5,6 Received 26 March 2008; revised 3 January 2009; accepted 31 March 2009; published 11 August [1] In this study, we have determined the crustal structure using three different receiver function methods using data collected from the northern transect of the Continental Dynamics of the Rocky Mountains (CD-ROM) experiment. The resulting migrated image and crustal thickness determinations confirm and refine prior crustal thickness measurements based on the CD-ROM and Deep Probe experiment data sets. The new results show a very distinct and thick lower crustal layer beneath the Archean Wyoming province. In addition, we are able to show its termination at 42 N latitude, which provides a seismic tie between the CD-ROM and Deep Probe seismic experiments and thus completes a continuous north-south transect extending from New Mexico into Alberta, Canada. This new tie is particularly important because it occurs close to a major tectonic boundary, the Cheyenne belt, between an Archean craton and a Proterozoic terrane. We used two different stacking techniques, based on a similar concept but using two different ways to estimate uncertainties. Furthermore, we used receiver function migration and common conversion point (CCP) stacking techniques. The combined interpretation of all our results shows (1) crustal thinning in southern Wyoming, (2) strong northward crustal thickening beginning in central Wyoming, (3) the presence of an unusually thick and high-velocity lower crust beneath the Wyoming province, and (4) the abrupt termination of this lower crustal layer north of the Cheyenne belt at 42 N latitude. Citation: Rumpfhuber, E.-M., G. R. Keller, E. Sandvol, A. A. Velasco, and D. C. Wilson (2009), Rocky Mountain evolution: Tying Continental Dynamics of the Rocky Mountains and Deep Probe seismic experiments with receiver functions, J. Geophys. Res., 114,, doi: /2008jb Introduction [2] The Rocky Mountains remain an intriguing and complex mountain belt with a long geologic history that has yet to be completely unraveled [e.g., Karlstrom et al., 2005; Keller et al., 2005]. For example, relationships between variations in crustal thickness and topography are complex, and the nature of isostatic compensation thus remains unresolved. Furthermore, the geometry and the nature of the suture zone between the Archean Wyoming province and Proterozoic terranes to the south and the 1 Department of Geological Sciences, University of Texas at El Paso, El Paso, Texas, USA. 2 ConocoPhillips School of Geology and Geophysics, University of Oklahoma, Norman, Oklahoma, USA. 3 Now at ExxonMobil Upstream Research Company, Houston, Texas, USA. 4 Department of Geological Sciences, University of Missouri-Columbia, Columbia, Missouri, USA. 5 Jackson School of Geosciences, University of Texas at Austin, Austin, Texas, USA. 6 Now at U.S. Geological Survey, Hawaiian Volcano Observatory, Hawaii National Park, Hawaii, USA. Copyright 2009 by the American Geophysical Union /09/2008JB005726$09.00 relationships between Precambrian structures and younger features, such as Laramide uplifts that produced most of the basement outcrops, continue to be unclear (Figure 1). [3] The Continental Dynamics of the Rocky Mountains (CD-ROM) experiment utilized seismic arrays employing both earthquake (passive) and controlled sources to explore the seismic structure of the lithosphere along a north-south trending transect across the southern Rocky Mountains from central Wyoming to central New Mexico (Figure 1) [e.g., Karlstrom et al., 2005; Keller et al., 2005]. The goal of this project was to advance our understanding of the tectonic evolution of the Rocky Mountain region. Figure 1 shows the geologic structures and the experiment design for CD- ROM. The controlled source survey focused on crustal structure along the entire transect, while the passive arrays targeted the deep structure of the Cheyenne belt suture in Wyoming and Colorado and the Jemez lineament in Colorado and New Mexico. The objective of CD-ROM was to investigate the assembly of Laurentia in the Proterozoic [Karlstrom et al., 2005]. The location of the Deep Probe experiment [Gorman et al., 2002], whose goal was to investigate deep earth structure across the northern Rocky Mountain region, is also indicated in Figure 1. In our study, we focus on the data from the northern CD-ROM passive 1of17

2 array that extended from Rawlings, Wyoming to Steamboat Springs, Colorado (Figure 2), traversing the Cheyenne belt suture between the Archean Wyoming province and the Proterozoic Yavapai province [Karlstrom and Houston, 1984]. A specific goal of our receiver function analysis was to better constrain the lower crustal structure in the subsurface gap in ray coverage between the CD-ROM and Deep Probe experiments, which unfortunately occurs in the region of the Cheyenne belt suture. Since the primary focus of the previous receiver function studies of the CD-ROM passive array was the upper mantle structure [Dueker et al., 2001; Zurek and Dueker, 2005], we investigated crustal structure with the goal of integrating our receiver function results with controlled source data and other geophysical and geological constraints. We were able to image the region of suturing between Proterozoic terranes to the south and the Wyoming craton to the north in more detail than in previous studies. We also delineated a very strong midcrustal reflector along the Deep Probe profile north of the Cheyenne belt that is the top of an unusually fast and thick lower crustal layer that was probably underplated in the Proterozoic [Gorman et al., 2002]. 2. Regional Geologic Setting and Previous Geophysical Results [4] The Precambrian geological features in the region (Figure 1) include the Cheyenne suture that separates the Ga Wyoming Archean craton from the Ga Yavapai Proterozoic province. This suture constitutes a terrane boundary between an Archean continental nucleus and terranes that were accreted in the Proterozoic [e.g., Karlstrom and Bowring, 1988; Karlstrom et al., 2005]. The Jemez lineament is an alignment of Miocene and younger volcanic centers and geologic features that extends from east central Arizona to northeastern New Mexico (Figure 1) [e.g., Aldrich, 1986], and it aligns with the structural grain of the Precambrian basement in the region [e.g., Gilbert et al., 2007]. It also lies in an area thought to be a transition zone between the Yavapai and Mazatzal terranes [e.g., Whitmeyer and Karlstrom, 2007]. [5] The results of the CD-ROM controlled source seismic experiment show that the crust is about 50 km thick beneath the high topography of northern Colorado and thins southward toward New Mexico [Snelson et al., 2005; Levander et al., 2005]. The thinning of the crust beneath southern Wyoming suggested in earlier studies [e.g., Johnson et al., 1984; Snelson et al., 1998] is not required by the CD-ROM controlled source data set, but the integration of these data with the Deep Probe and gravity data indicate that it is present [Snelson et al., 2005]. The CD-ROM velocity models showed that the lowermost crust consisted of a layer with high velocities (up to 7.2 km/s) although the thickness and continuity of this layer varied between the models of Levander et al. [2005] and Snelson et al. [2005]. [6] From the passive source CD-ROM data, Zurek and Dueker [2005] calculated and stacked receiver functions using the common conversion point (CCP) technique with a focus on the lithospheric structure. The CCP images, combined with teleseismic tomography of the CD-ROM experiment [Dueker et al., 2001; Zurek and Dueker, 2005], showed that the Cheyenne belt region has a thick lithosphere (>150 km), which they interpreted to be underlain by the remnants of a Proterozoic oceanic slab fragment. In addition, Zurek and Dueker [2005] showed that differences between the Archean and Proterozoic lithosphere are observed in crustal thickness and subcrustal reflectivity. About 150 km to the west of the CD-ROM profile, the Deep Probe passive array showed that in the Proterozoic terrane to the south of the Cheyenne belt, the crust is also 50 km thick and thins beneath the Cheyenne belt to the north [Crosswhite and Humphreys, 2003] (Figure 2). Zurek and Dueker [2005] suggested that the Moho was imbricated just north of the Cheyenne suture and that the Proterozoic crust appears to be underthrusting the Archean crust. At the northern end of the CD-ROM passive array, Archean subcrustal layering was observed, and the depth of this layering increases northward, and this was interpreted to be related to 2.1 Ga rifting of the southern margin of the Wyoming craton [Zurek and Dueker, 2005]. [7] The CD-ROM and Deep Probe experiments share a common shot point (Figure 2) at the northern end of the CD-ROM profile. The results of the Deep Probe experiment show that north of the CD-ROM profile the crustal thickness increases significantly and include an anomalously thick and high-velocity layer in the lower part of the crust [Snelson et al., 1998; Gorman et al., 2002]. This thick lower crustal layer (LCL) underlies the entire Wyoming province and continues northward into the Medicine Hat Block in Montana in the Deep Probe velocity models. The termination of this layer is expected to be north of the Cheyenne Belt in southern Wyoming, but south of the common shot point of the two experiments. The crustal thickness was modeled to be 45 km beneath the Cheyenne belt and increased to a maximum of 60 km beneath central Wyoming [Gorman et al., 2002]. Owing to the large distance between shots across Wyoming and Montana, the control on the structure of the LCL is limited, and the seismic phases that define the LCL are variable. However, the unusually thick crust from central Wyoming across Montana into southernmost Alberta is a required result of these analyses. 3. Receiver Function Estimation [8] The northern CD-ROM passive seismic transect included 24 broadband IRIS/PASSCAL seismographs spanning 240 km with a station spacing of about 10 km (Figure 2), and recorded for 1 year (June 1999 to June 2000). For our receiver function analysis, we selected earthquakes with magnitudes >5.5 and distances between 15 and 97. [9] The teleseismic receiver function technique represents a straightforward method to extract information on the structure of the lithosphere in the vicinity of the recording station [e.g., Phinney, 1964; Helmberger and Wiggins, 1971; Langston, 1979]. Using deconvolution, the technique essentially strips off the P wave energy except for the direct arrival, and leaves enhanced P-to-S conversions and reverberations on the resulting waveform. The vertical component seismogram from the teleseismic distance range behaves like a pulse-like time function convolved with the instrument response, with only minor later arrivals. The first-order information about the crustal structure under a station can be derived from the radial receiver function, which is dominated by P-to-S converted energy from major 2of17

3 Figure 1. Map of the Continental Dynamics Rocky Mountain (CD-ROM) experiment (modified from Snelson et al. [2005]). The controlled source seismic line (black), its according shot points, and the northern and southern passive arrays (gray diamonds) are shown. Yellow background indicates Precambrian and white background Proterozoic age; outcrops of Precambrian basement are shown in pink. The Deep Probe experiment is outlined in blue. The gray box indicates the area shown in Figure 2. velocity discontinuities in the crust and upper mantle. These converted phases from interfaces with sharp velocity gradients, such as the crust-mantle boundary (Moho), define the structure and velocities in the vicinity of the recording station [Langston, 1979]. [10] We employed the receiver function technique using an iterative deconvolution of Ligorria and Ammon [1999], which represents an extended version of the frequency domain division into the time domain. This approach calculates an estimated misfit between the horizontal component seismogram and an iteratively updated receiver function, allowing for a robust characterization of the resulting receiver function. We selected a fit of 80% or higher between the estimated and observed waveform for our receiver function calculations. This quantitative approach of determining the quality of the receiver functions was followed by a careful manual inspection. We identified 348 receiver functions, which were included in the next stage of processing. [11] The first set of receiver functions were computed using the iterative time deconvolution with Gaussian width (Ga) factors of 5, 2.5, and 1, which is equivalent to applying low-pass filters with cutoff frequencies of 2.4, 1.2, and 0.5 Hz, respectively [Ligorria and Ammon, 1999]. Figure 3 3of17

4 Figure 2. Index map of the region of the CD-ROM north passive array. The controlled source seismic line (black circles) and the according shot points (red) as well as the passive stations (blue diamonds) are indicated. The Deep Probe and seismic experiment is shown in yellow. shows examples of radial and transverse receiver functions for station N17 (Figure 2) using a Ga factor of 5 (Figures 3a and 3b) and a Ga factor of 1 (Figures 3c and 3d). Figures 3a and 3b show higher frequency content due to the higher cutoff frequency. The Moho P-to-S converted phase (Ps) is visible at 7 s after the direct P wave arrival; however owing to the high frequency content on the seismogram, it is not obvious. Furthermore, the crustal multiples PpPs and PsPs+PpSs cannot be explicitly identified on the radial receiver function of Figure 3a. Using a lower value for Ga, as seen in Figures 3c and 3d, the crustal multiples can be identified. The crustal converted phase Ps occurs at 7 s after the direct P wave, the crustal multiple PpPs occurs at 20 s, and the multiples, PpSs+PsPs with a negative pulse, which are somewhat more difficult to identify, occur at 26 s. Lateral heterogeneities, such as big lateral changes in velocity, a steep dip in the reflector of interest, or anisotropy in the vicinity of the recording station would cause a significant portion of energy to be recorded on the transverse component seismogram [e.g., Langston, 1979]. The low energy content on the transverse component receiver function (Figures 3b and 3d) compared to the radial component seismograms (Figures 3a and 3c) indicate that there are no major lateral heterogeneities in the immediate vicinity of the receiver that need to be considered. [12] Since we have a limited number of receiver functions, we used the deconvolution misfit from the iterative approach [Ligorria and Ammon, 1999] followed by a careful visual inspection for assessing receiver function quality. We determined the quality of the receiver functions on a station-by-station basis. We identified the main criteria for our qualitative selection as (1) a clear P wave pulse, (2) a Ps converted phase that can be identified, and (3) consistency throughout a set of receiver functions per station. Single waveforms, which were inconsistent with the data set or had high noise content, were excluded from the data set. For example, a strong negative pulse after the direct P wave arrival indicates a negative impedance contrast, such as from a low velocity zone in the vicinity of the recording station. Thus, it would need to be consistent throughout the set of waveforms for one station, or at least be azimuthdependent, to be included in further analysis. Since none of the stations show a negative pulse consistently throughout all waveforms, we singled out the waveforms with a strong negative pulse and eliminated them from the data set. Figure 4 shows an example of a sorted data set for station N17, which includes 36 out of 45 available radial receiver 4of17

5 Figure 3. Examples of receiver functions for station N17 of the CD-ROM passive experiment. The receiver functions are calculated for an event with a back azimuth of 315 and a distance of (a) Radial receiver function (rf) with a Gaussian width factor (Ga) of 5, (b) transverse receiver function for a Ga of 5, (c) radial rf for a Ga of 1, and (d) transverse rf for a Ga of 1. The value for Ga enhances the visibility of the converted phases. functions for the later processing. The seismograms are sorted by source-receiver distance (D, in degrees). The P wave arrival is clearly visible, and the Ps converted phase can also clearly be identified at 7 s after the direct P wave arrival. The relatively late arrival time of the Ps converted phase can be attributed to the thick crust underneath the northern area of the CD-ROM profile. The crustal multiples PpPs can be identified at 20 s after the direct P wave arrival, and the less clear PsPs+PpSs should occur at 25 s after the direct P wave arrival. A prominent feature can be seen in both Figures 3 and 4 at 14 s after the direct P wave arrival. However, it is only visible on station N17 and missing on neighboring stations, thus it has been attributed as a local anomaly. In addition, Figure 4 also shows a coverage map of the selected events for station N Stacking and Bootstrap Error Estimation [13] We employed the receiver function stacking technique introduced by Zhu and Kanamori [2000], which estimates the crustal thickness and a V p /V s ratio based on the radial receiver function. This technique provided us with a first look at the Moho depths and V p /V s ratios along the northern passive transect on a station-by-station basis. The results are valid within in the vicinity of each recording station. Assuming no lateral velocity heterogeneities exist, the seismic ray will only be dependent on the vertical slowness, whereas the horizontal slowness p remains constant. The time separation between the Ps converted wave and the direct P wave obtained from receiver functions (t Ps ) can then be used to estimate crustal thickness (H), given the average crustal velocity V P, and the constant ray parameter p of the incident wave [e.g., Gurrola et al., 1994]. [14] The trade-off between the thickness and the crustal velocities presents an ambiguity that can be reduced by using the arrival times of later multiple phases t PpPs and t PsPs+PpSs, which provide additional constraints to both V p /V s and the crustal thickness [e.g., Gurrola et al., 1994; Zhu and Kanamori, 2000]. Using and stacking multiple events helps to increase the signal-to-noise ratio (SNR), which may be caused by background noise, scattering from crustal heterogeneities, and P-to-S conversions from other velocity discontinuities. This stacking technique [Zhu and Kanamori, 2000] weights each phase and plots the stacked phases as a gridded image, which reaches a maximum when all three phases are stacked coherently with the correct estimate for the crustal thickness H and the V p /V s. The main advantages of this grid-search-based technique are that a large number of receiver functions can be processed without the need of manually picking Ps arrival times and that the stacking results in an enhancement of the SNR and a suppression of lateral variations in the vicinity of the recording station. We employed this technique to derive an average single layer crustal model including the crustal thickness H and V p /V s (k) Crustal Multiples [15] The first set of stacks was calculated for all stations of the northern CD-ROM transect using the radial receiver functions computed with a Ga factor of 5. We chose an average P wave velocity of 6.5 km/s, which was determined from a velocity model of the controlled source seismic data set [Snelson et al., 2005]. We extracted 1D velocity profiles at each shot point location, and calculated the average crustal velocity from these profiles. Figure 5 shows examples of the resulting gridded images for selected stations. The Ps converted phase is clearly mapped as a broad swath, but a lack of clear crustal multiples using a Ga factor of 5 is immediately evident. The multiples PpPs and PpSs+PsPs are of particular importance in this technique, since they serve as constraints for the crustal thickness H and V p /V s ratio. A unique solution can only be obtained when two branches cross with the one of the converted Ps phase. Hence, the values for H and V p /V s become ambiguous and nonunique in the case of an absence of clear multiples. [16] A possible cause for weak crustal multiples can be strong crustal heterogeneity in the vicinity of the recording station or a change in the character of the Moho such as a velocity gradient, Moho steps or a large dip. These effects would lead to a failure of the mathematical basis of the technique, which includes the assumptions that (1) the velocity field gradually increases with depth, (2) little or no lateral changes are present, and (3) the interfaces are presumably flat. However, in the presence of strong lateral 5of17

6 Figure 4. (left) Selected data set for station N17 (Colorado) showing 36 out of 45 available selected radial receiver functions for the following processing. The receiver functions are sorted by sourcereceiver distance (vertical axis, D in degrees). The direct P wave arrival is followed at 0 s, followed by a clear Ps converted phase at 7 s; the multiples are less obvious, but visible. It is clear that the amplitude of the Ps Moho converted phase increases with decreasing distance, consistent with a relatively flat Moho. (right) Coverage map of the 36 selected seismograms. The triangle indicates the recording station; the stars represent the selected earthquakes in distances between 15 and 97. heterogeneities in the vicinity of the recording station, a large part of the seismic energy should appear on the transverse receiver functions, which is not the case for the CD-ROM northern transect (Figure 3). [17] Since the lack of clear multiples is not due to unusually complex crustal structure, its origin must stem from a different geological cause. A closer look at the CD-ROM velocity models derived from the refraction profile [e.g., Snelson et al., 2005; Levander et al., 2005] reveal a layer of high velocities in the lower crust, the socalled 7.xx layer, which is found immediately above the Moho. As a result, the Moho is an area of gradual velocity increase, rather than a sharp boundary. Hence in order to sample the Moho as a boundary, the receiver functions needed to be recalculated using a much lower value for the Gaussian width (Ga) factor. This leads to longer wavelengths in the resulting receiver function waveforms, which makes it possible to better image the Moho as a boundary but with increased depth uncertainty. [18] Figure 6 shows synthetic receiver functions based upon two different simple crustal input models. The first velocity-depth model includes a 40 km thick crust with a P wave velocity of 6.5 km/s, and has a sharp velocity increase to 8 km/s at the Moho. The second model shows a similar model, however the P wave velocity changes gradually from 6.5 km/s at 30 km to 8 km/s at the crust-mantle interface at 40 km depth (i.e., a 10 km thick region with a velocity gradient). The waveform for the gradient model shows significantly lower amplitudes and longer wavelengths for the crustal phases, compared to its sharp equivalent. As a result of the low amplitudes, the mapping of H and V p /V s may be difficult, especially in the presence of high noise on an actual data set. Furthermore, the longer wavelengths and resulting broader waveforms of the multiple phases may be overprinted with high-frequency noise in the presence of a gradual velocity change at the mapped interface. Using a much lower value for the Ga factor solves these problems by filtering the high-frequency noise and enhancing the PpPs and PsPs+PpSs multiples. [19] To illustrate this effect, Figure 7 shows H-k stacks computed using the synthetic receiver functions of Figure 6. The columns of Figure 7 represent the three different weights used for the seismic phases Ps, PpPs, and PsPs+PpSs in the calculation. The assigned weights for each of the three phases are arbitrary, however their sum must be 100%. Different authors select different weighting factors for each phase. For example, Zhu and Kanamori [2000] suggest that the weight for Ps should be greater than the sum of the weights for PpPs and PsPs+PpSs to balance the contributions for the three phases. The default values in their study are 70, 20, and 10%. They also state that Ps has the highest signal-to-noise ratio, therefore it should be given 6of17

7 Figure 5. H-k stacks for selected stations along the northern CD-ROM transect using a Gaussian value of 5.0. The Ps converted phase is clearly visible as a broad swath; however, the crustal multiples are very unclear in these stacks. Thus, the results for H and V p /V s become ambiguous. a higher weight than the other two. Al-Damegh et al. [2005] suggest that the most suitable weighting to balance the contribution for each phase is 40, 35, and 25% for Ps, PpPs, and PsPs+PpSs, respectively, based upon the signal/noise ratio. [20] In our study, we assigned weights based on the prominence of each of the three phases on the actual receiver functions from the early visual inspection. In addition, the weights can be used as a tool for quality control. If the signal-to-noise ratio on the data set is favorable, the use of different weights will not have a significant impact on the resulting values of crustal thickness and V p /V s. On the other hand, if varying the weights yields significantly different values for the estimates of crustal thickness and V p /V s, the receiver functions used for the calculation should be rechecked for data consistency, and the values for H and V p /V s must be used with caution. [21] Figures 7a 7f show the stacking results for the synthetic receiver function resulting from a sharp velocity interface at the Moho (Figure 6, left), where a Ga factor of 5 was used for the computation of Figures 7a 7c and a Ga factor of 1 for Figures 7d 7f. All three seismic phases are distinct and can be clearly identified using a Ga factor of 5. The same phases are less obvious with a Ga factor of 1 and their branches are much wider owing to the lower filter parameter. However, the estimates of H = 40 km and V p /V s of 1.73 correctly reproduce the input model for both Ga factors. The main difference lies in the confidence interval of the result, which is much larger for a lower Ga factor, due to the broader wavelength of the phases. [22] Figures 7g 7l of show H-k stacks for synthetic receiver functions calculated with a gradual velocity increase at the crust-mantle interface. It is immediately evident that the results for both a Ga factor of 5 (Figures 7g 7i) and a Ga factor of 1 (Figures 7j 7l) are not reproducing the correct values for the crustal thickness and the V p /V s ratio. Instead, the technique maps an intermediate value between the onset of the gradual velocity interface at 30 km and its cessation at 40 km. This is an important finding for the interpretation of the stacking results since, in the case of an existing gradual 7of17

8 Figure 6. Synthetic receiver functions calculated for simple crustal models. The input velocity models include (left) a sharp velocity boundary and (right) a gradual velocity increase above the Moho. velocity gradient above the Moho interface, the results for H and V p /V s are biased toward lower values compared to a sharp boundary at the same depth. In addition, Figure 7 illustrates that when the receiver function multiples are transformed to H-k space, they undergo more wavelength compression than the Ps phase. Thus, using a lower-frequency cutoff enhances the visibility of the reverberations Bootstrap Error Estimates [23] The same set of receiver functions used in the H-k stacking technique was utilized in the slant stacking approach by Al-Damegh et al. [2005]. This technique is similar to the grid-search-based imaging of Zhu and Kanamori [2000]; however, the uncertainties are estimated in a bootstrap approach [Efron and Tibshirani, 1991; Chevrot and van der Hilst, 2000]. Figure 8 shows examples of the resulting images for station N17 after application of the slant stacking technique of Al-Damegh et al. [2005]. Figure 8 illustrates the gridded images in the H V p /V s domain, and the black diamond represents the best estimate for H and V p /V s based upon the complete data set. The crosses correspond to the bootstrap results, each of which represents a random subset of the data and its according estimate for H and V p /V s.a total of 100 bootstrap calculations were completed, which is a typical number for this type of analysis [Efron and Tibshirani, 1991]. Both estimates for the crustal thickness in Figures 8 are 49 km and V p /V s = The 100 bootstrap error analysis yields uncertainties of ±0.54 (±0.02) km for the crustal thickness and ±0.015 (0.014) for the V p /V s in Figure 8. The bootstrap results show a very tight area around the actual estimate, which indicates good quality receiver functions, and reliable estimates for H and V p /V s, and results in the low values of uncertainties for this particular station Images [24] Figure 9 shows the H-k stacking results for the actual data set of station N17 using the approach by Zhu and Kanamori [2000], and Table 1 summarizes the parameters used for computation and the results. A total of 36 radial receiver functions were used for the stacks (Figure 4). The three rows in Figure 9 represent three different Ga factors (1, 2.5, and 5) of the input receiver functions, and the columns correspond to different weights used for the crustal phases. The estimates for H and V p /V s are demonstrated in Table 1. The multiples are clearly visible for a Ga of 1 but are less obvious for higher values of the Ga factor. Increasing the values for the weighting factor of the crustal reverberations PpPs and PsPs+PpSs boosted their significance in the stack, and resulted in a more stable estimate for H and V p /V s. The fact that the calculations for various input parameters deliver matching results for H and V p /V s indicates that the selected receiver functions were of good quality and the computations were stable. [25] Table 2 summarizes all results of both stacking approaches for the complete data set. Identical weighting factors for the three seismic phases were used for both calculations and can be found in the second column of Table 2. The estimates of crustal thicknesses are very similar with both techniques (and average of 50 ± 2 km for the Zhu and Kanamori [2000] approach and ±6 km for the Al-Damegh et al. [2005] approach. The average V p /V s ratio is on the order of ± 0.1 (Table 2), the values for the Poisson s ratio s are shown as well. Some stations show anomalous results, especially for the V p /V s ratios, which can 8of17

9 Figure 7. H-k stacks for synthetic receiver functions shown in Figure 6. The rows show varying Gaussian width factors (Ga); the columns indicate different weights used for the calculation. (a f) Stacking results for the synthetic receiver function using a sharp velocity interface at the Moho (Figure 6, left), and a Ga factor of 5 was used for the computation of Figures 7a 7c and a Ga of 1 for Figures 7d 7f. All three crustal phases are distinct and can be clearly identified using a Ga of 5. The same phases are less obvious using the lower Ga parameter of 1 (Figures 7d 7f). (g l) H-k stacks for synthetic receiver functions calculated with a gradual velocity increase at the crust-mantle interface. It is immediately evident that the results for both a Ga of 5 (Figures 7g 7i) and a Ga of 1 (Figures 7j 7l) are not reproducing the true values for the crustal thickness and the V p /V s ratio. be attributed to either an inadequate number of receiver functions or insufficient data quality and were therefore removed from the calculation of the average V p /V s. [26] The main differences are found in the estimated uncertainties. The bootstrap uncertainty estimates are an excellent indicator of the quality of the result. However, some recording stations had a low number of waveforms available and yielded 10 or fewer receiver functions (Table 2). This causes the bootstrap error estimate to be less reliable, since smaller statistical samples are available, and the random subsets of the data repeat themselves. Therefore, a combined interpretation of the uncertainty estimates, the number of receiver functions, and the consistency between neighboring stations is a good strategy when interpreting the receiver function results. [27] The H-k stacking results described above provide us with a first look of 1-D estimates of crustal thicknesses and V p /V s ratios in the vicinity of the recording stations, and several interesting observations can be made. At the northern end of the transect, station N00 indicates a crustal thickness greater than 60 km, and the number of good quality seismograms (15) for this station is sufficient to believe that this result is robust. Furthermore, the crustal thickness estimates of 49.5 km and 47 km for the neighboring stations N01 and N02, respectively, are very low compared to the values of station N00. It is also noted that the values of V p /V s ratio and crustal thicknesses of station N01 vary significantly between the Zhu and Kanamori [2000] and the Al-Damegh et al. [2005] approaches (Table 2), which are otherwise consistent. The results of the seismic Deep Probe refraction experiment combined with gravity modeling [Snelson et al., 1998; Gorman et al., 2002; Keller et al., 2005] indicate a major change in crustal thickness and structure in this area. In addition to crustal thickening, an unusually thick high-velocity lower crustal layer was identified in the Deep Probe results [Snelson et al., 1998; Gorman et al., 2002]. Hence while N00 constrains the deepening Moho, it is possible that stations N01 and N02 constrain the top of an underplated layer of mafic material [Snelson et al., 1998; Gorman et al., 2002] rather than the Moho in a region of rapid crustal thinning. [28] Figure 10 shows the results of the CD-ROM northern transect for both crustal thickness (Figure 10, top) and V p /V s ratios (Figure 10, bottom). In Figure 10 (top), the Moho geometry from the controlled source results [e.g., Snelson et al., 2005; Levander et al., 2005] is included (gray line). The crustal thicknesses from the receiver functions are generally lower relative compared to the controlled source results. We attribute this difference to the velocity gradient (7.xx layer) right above the Moho, which produces estimates of crustal 9of17

10 set, and providing a tectonic synthesis of the area. However, to provide an image along the profile, we have chosen common conversion point stacking and migration techniques for further analysis, which will be discussed in further detail below. The northern passive transect is subdivided into a grid, and receiver function binning allows for a two-dimensional view of the transect. 5. Common Conversion Point Imaging (CCP Binning) [30] Common conversion point (CCP) binning is commonly employed in reflection seismology for the velocity analysis of PS data. The main difference between the CCP and the common midpoint (CMP) raypath geometry is the conversion of the incident P wave to an S wave at the reflector (Figure 11). Therefore, the reflection angle changes from the P wave angle of incidence (8) to the reflection angle for the converted S wave (y). Snell s law provides a simple relation between the two angles [Yilmaz, 2001]. In receiver function studies, the waveform is back-projected along a theoretical P wave and S wave raypath (Figure 11b) for a given 1D reference velocity model, and then the amplitude information is stacked into vertical and horizontal CCP bins [Wilson and Aster, 2005]. A constant ray parameter derived from a reference earth model is used to account for the effect of the angle of incidence on the differential arrival time Dt Ps between the P and the converted S wave [Gurrola et al., 1994]. [31] We created a two-dimensional cross section of the CD-ROM northern array in offset and depth space using the CCP method [Wilson and Aster, 2005]. We used bins with a Figure 8. Slant stack result for station N17 using the technique by Al-Damegh et al. [2005]. The weights used for this stack are 40, 30, and 30%, and the average crustal P wave velocity is 6.5 km/s. The best H is 49 ± 0.58 km, and V p /V s is ± The contours are shaded to highlight the maximum as black. (bottom) The weights used for this stack are 50, 40, and 10%, and the average crustal P wave velocity is 6.5 km/s. The best H is 49 ± 0.63 km, and V p /V s is ± thicknesses too low (Figure 7). The V p /V s ratios have a linear trend line added and show a trend of decreasing values from the southern to the northern end of the line that suggest a northward decrease in average mafic content in the Proterozoic crust of northern Colorado and southern Wyoming. The two anomalous stations N01 and N02, which constrain the lower crustal layer, rather than the Moho, are shown in gray. [29] The H-k stacking results described above provide us with a first look of 1D crustal thicknesses and V p /V s ratios in the vicinity of the recording station. Furthermore, the H-k stacking results are used in the companion paper by Rumpfhuber and Keller [2009], which is integrating these results with the outcome of the controlled-source seismic data Figure 9. Stacking results for recording station N17 using a total of 36 receiver functions shown in Figure 5. Different Gaussian width factors (Ga) and weights (W) are used: H-k stacks for (a) Ga =1,W = 50, 40, and 10%; (b) Ga =1,W= 40, 30, and 30%; (c) Ga = 2.5, W = 40, 30, and 30%; (d) Ga = 2.5, W = 20, 40, and 40%; (e) Ga =5,W = 50, 40, and 10%; and (f) Ga =5,W = 40, 30, and 30%. 10 of 17

11 Table 1. Summary of the H-k Stacking Results for Recording Station N17 Shown in Figure 9 a Figure Ga Factor Weights (Ps, PpPs, PsPs+PpSs) (%) H V p /V s 9a 1 50, 40, b 1 40, 30, c , 30, d , 40, e 5 50, 40, f 5 40, 30, a The input parameters used for computing the stacks Ga and weights are shown along with the results for the crustal thickness (H) and V p /V s with the according standard deviations. radius of 25 km and stacked all rays that fell within one bin before plotting them on the 2-D grid. A band-pass filter between 1/3 and 1 s was applied to the traces using a second-order zero-phase Butterworth filter. A 1D velocity model at shot point 7 near Kremmling (CO) was extracted from the controlled source model of Snelson et al. [2005] and used as a reference model. The average V p /V s ratio of derived from the slant stacking approach was used as an estimate for the calculations. [32] The resulting CCP image is shown in Figure 12. Figure 12a shows the CCP image, and has the slant stacking results with the according uncertainties plotted in black. Figure 12b shows the same image with our interpretation, and Figure 12c illustrates the ray coverage. The CCP image shows a prominent Moho with few artifacts. A clear Moho interface is evident up to 42 N latitude, where the interface clearly splits into two branches. However, it is important to compare the resulting CCP image with the associated ray coverage (Figure 12c). The ray coverage shows good ray coverage between 41.5 N and 42.5 N and poor ray coverage between 40 N and 40.5 N. The ray coverage is fair between 40.5 N and 41.5 N latitude, which may be the reason for some smearing of the Moho in the CCP image at this location. [33] This result is not only consistent with results from the independent migration and slant stacking, but it also consistent with previous results in the area from the Deep Probe and CD-ROM [e.g., Snelson et al., 1998, 2005] refraction experiments, which reveal a consistent Moho depth of 50 km up to shot point 10 of the CD-ROM experiment (Figure 2) and a deepening of the Moho north of shot point 43 (Figure 2) of the Deep Probe seismic experiment. In addition, it confirms the existence of the strong lower crustal layer beneath the Wyoming province [Gorman et al., 2002], which is significantly different from the lower crustal layer identified in the CD-ROM experiment. The most significant aspect of this image is the clear southern termination of the strong lower crustal layer of the Wyoming province at 42 N latitude. 6. Kirchhoff Migration [34] The CCP method is easily implemented, and transforms the data into offset and depth space; however, it does not correct for diffracted energy or mismapped multiples, which commonly arise from free surface reverberations. Migration seeks to map seismic energy to its true subsurface origin, and provides for proper correction for wave propagation effects, which can be due to lateral heterogeneity [Wilson and Aster, 2005]. The receiver functions for all stations of the northern CD-ROM transect were bundled and used in the Kirchhoff style migration of Wilson and Aster [2005]. Figure 13a shows the resulting image of the migrated section; Figure 13b shows the resolution or ray coverage with warmer colors indicating a higher resolution or better ray coverage. The selected slant stacking results of section 4 and the according uncertainty estimates are plotted on top of the migrated section in Figure 13. [35] The red/yellow warm colors represent our target interface, which corresponds with the Moho. Along the Table 2. Summary of the Stacking Results for Both the Zhu and Kanamori [2000] and Al-Damegh et al. [2005] Approaches Station Weight Number of RF Zhu and Kanamori [2000] H (km) Al-Damegh et al. [2005] H (km) Zhu and Kanamori [2000] Al-Damegh et al. [2005] Vp/Vs s Vp/Vs s N00 W ± ± ± ± N01 W ± ± ± ± N02 W ± ± ± ± N03 W ± ± ± ± N04 W ± ± ± ± N05 W ± ± ± ± N07 W ± ± ± ± N08 W ± ± ± ± N09 W ± ± ± ± N10 W ± ± ± ± N14A W ± ± ± ± N15A W ± ± ± ± N16 W ± ± ± ± N17 W ± ± ± ± N20 W ± ± ± ± N21 W ± ± ± ± N23 W ± ± ± ± N24 W ± ± ± ± N25 W ± ± ± ± N26 W ± ± ± ± of 17

12 Figure 10. Selected results for (top) crustal thickness and (bottom) V p /V s ratios for the CD-ROM northern transect. The uncertainties estimated by the bootstrap technique are shown by the error bars. The gray line shows (top) the crustal thicknesses of the CD-ROM controlled source results and (bottom) a linear trend of the V p /V s results. Stations N01 and N02 are attributed to the top of a lower crustal layer rather than the Moho and hence are illustrated in gray. southern portion of the profile, this boundary occurs at 50 km depth. The crust appears to thin in the vicinity of station N09. Between stations N8 and N05, the Moho is a strong interface before it splits into two separate branches at station N04 or 42 N latitude. These observations are consistent with the CCP image and the slant stacking results, which also show a deep Moho at station N00 and shallower estimates of 49.5 and 47.5 km at stations N01 and N02, respectively. Figure 13b shows information about the resolution of the migrated image, and indicates that there is low ray coverage at the edges of the migrated section as expected, but higher resolution is obtained in the center and in particular at 42 N latitude, which also shows the important southern termination of the lower crustal layer. Hence, the migrated image contributes to a comprehensive picture of the crust-mantle interface of the northern CD- ROM transect. 7. Discussion [36] We have used three difference receiver function techniques (stacking, CCP, and migration), and evaluated the results to provide a new interpretation of crustal structure along the northern CD-ROM transect. We first independently applied two stacking techniques, which are based on the same approach to estimate crustal thickness H and V p /V s on a station-by-station basis. However, the associated error analyses were conducted differently and, on average, varied by over a factor of two. On the basis of many iterations in which key parameters were varied, we found that the values for H and V p /V s should be considered 12 of 17

13 Figure 11. (a) Common midpoint (CMP) and common conversion point (CCP) geometries of an incident P wave. The P-to-P and P-to-S reflections at a reflecting interface are shown; the arrows represent the particle motion (modified from Yilmaz [2001]). (b) The CCP binning equivalent for receiver function studies. The source is a teleseismic earthquake. estimates that are dependent on careful data selection and parameters such as the weights for the crustal phases, the Gaussian width (Ga) factor, and the average crustal P wave velocity assumed. Thus, it is important to assess the results from these techniques with caution, since there are many potential trade-offs that produce uncertainty that may be hard to quantify. Both the quality of the receiver functions and the number of receiver functions used in the bootstrap error analysis come into play since a smaller number of receiver functions available leads to a smaller number of statistical samples. In addition, the uncertainties represented are solely based on statistical methods, and thus, the actual error in determining the Moho depth may exceed these values. This is true for both stacking techniques used in this study, and the bootstrap results show relatively high uncertainties with an average of ±6 km for the crustal thickness and ±0.09 for V p /V s. Therefore, this analysis points out the need for integration with other data and other receiver function methods of analysis. In addition, we created CCP and migrated images from our receiver functions. The three techniques applied in this study delivered consistent results, and only the combined interpretation produced a complete picture of crustal structure variations along the transect. [37] The slant stack results indicated crustal thickness estimates at the northern end of the profile varied substantially, with a crustal thickness >60 km at the very northern end and values of >50 km just to the south at neighboring stations. After a combined analysis with the images resulting from migration and CCP stacking, it became clear that these values are attributed to two different crustal disconti- Figure 12. Image resulting from CCP binning. (a) Band-pass-filtered CCP image, with the slant stacking results and according uncertainties plotted on top. (b) The same image, however, with our interpretation plotted on top. (c) Ray coverage for the CCP images. 13 of 17

14 Figure 13. (a) Kirchhoff-style migration on all receiver functions of the CD-ROM northern array after Wilson and Aster [2005]. The stacking results with the according error bars are plotted as white circles. (b) Ray coverage or resolution according to the migration of Figure 13a. The map shows the limited coverage at the edges of the northern CD-ROM transect but good coverage in the center of the profile. This map is an important feature for the interpretation of the migrated section. 14 of 17

15 Figure 14. Profile showing the shared shot point at the northern end of the CD-ROM (SP 10) and the southern end of the Deep Probe (SP 43) experiments. The dashed line indicates typical ray coverage for a Pn refracted wave. The vertical lines indicate the CD-ROM passive stations, including the Moho depths. nuities. The depth values of stations N01 and N02 can be related to the top of an anomalous lower crustal layer, whereas at station N00 and all other stations depth values can be attributed the Moho. This structural complexity occurs near the ends of the CD-ROM and Deep Probe controlled source experiments where they share a common shot point in central Wyoming (Figure 2). Thus, this complexity could not be resolved from these experiments alone owing to ray coverage considerations (Figure 14). However, through our receiver function analysis, we were able to establish an independent seismic tie between the CD-ROM and Deep Probe refraction/wide-angle seismic experiments. [38] The Cheyenne belt area along the Colorado/Wyoming border (Figure 2) was a major target of the CR-ROM project, and Levander et al. [2005] point out that although the Cheyenne belt is a major geologic boundary, it does not stand out as a major seismic velocity boundary in the refraction models. Our results show that major change in crustal structure occurs at 42 N latitude, whereas the Cheyenne belt is located at 41 N latitude. However, the image derived from the CCP stacking shows a zone of crustal thinning just north of the Cheyenne belt that was suggested by Johnson et al. [1984] and Snelson et al. [1998]. This thinning is most likely to be a remnant of the rifting (2.0 Ga) that formed the passive margin that preceded the Cheyenne belt collision. [39] South of the Cheyenne belt (Colorado/Wyoming border), the refraction results [Snelson et al., 2005; Levander et al., 2005] and receiver functions show that the Moho attains a depth of 55 km, and south of the passive array, the refraction data shows that the crust begins to thin in central Colorado. Similarly, Crosswhite and Humphreys [2003] show thick crust in the vicinity of the Cheyenne belt on the basis of receiver function analysis. These results suggest that there is long-lived signature of the 1.8 Ga collision that created the Cheyenne belt although younger events have reworked the region to some extent. Thus, these results also indicate that the reflection tentatively identified as the Moho on the CD-ROM reflection profile [Morozova et al., 2005] is from the top of the lower crust. [40] We provide a summary of the main results of our analysis in Figure 15. We modified gravity model of Snelson et al. [2005] that integrated the previous results from the CD-ROM and Deep Probe seismic experiments using gravity modeling to provide a continuous structural picture across the Cheyenne belt suture area due to the lack of ray coverage under the Day Loma shot point. The dotted line represents the results of this study, and locates the termination of the thick lower crustal layer associated with the Wyoming craton north of the Cheyenne belt suture [Gorman et al., 2002]. Hence, this study provides a strong seismic tie between the CD-ROM and Deep Probe seismic experiments and completes a north-south transect across the Figure 15. The results of this study viewed in the context of previous results and key geologic features. The image shows the gravity model of Snelson et al. [2005] as gray lines, which tied the previous results from the CD-ROM and Deep Probe seismic experiments together. The dashed lines are interfaces from the Deep Probe model of Gorman et al. [2002]. The dotted lines depict the Moho and lower crustal interfaces detected in this study. The gray layers are the strongly mafic lower crustal layers delineated jointly interpreted from the CD-ROM and Deep Probe results and this study. This geometry emphasizes the profound change in crustal structure across the Cheyenne belt suture zone. 15 of 17