Chapter 5

Determination of the CSZ velocity structure from local earthquake data

Index

5.1 Introduction

To locate an earthquake, one has to determine the optimum values of four earthquake parameters: the latitude, longitude, depth of the hypocentre, and the time of occurrence of the event. The best estimates of these parameters are those that minimize the differences between the observed and the calculated arrival times (i.e., the residuals) for each phase recorded at seismograph stations (Kissling, 1988). Residuals exist due to errors in the observed and the calculated (theoretical) arrival times. For the observed data, the errors come from the inaccuracy of the picked arrival times and of the time control of the recording system. The accuracy of calculated arrival times, on the other hand, depends mostly on how closely the velocity model resembles that of the real Earth, and to a smaller extent, on location errors of recorders, near-surface velocity perturbations and errors in the estimates of the hypocentral parameters.

With reliable observed data and good control on the sources of uncertainties, the residuals can be used to infer the velocity structure of the Earth. For an event, the residuals from all stations can be expressed in a RMS (Root Mean Square) sense as

for I= 1,NS

where the residual is the difference between the observed and the calculated arrival times, the weight is a quality factor assigned to a picked arrival time, NS is the number of stations, and NP is the number of phases used in the solution. The optimum velocity model minimizes the RMS of a series of events.

Extracting velocity information from microearthquakes is called simultaneous inversion of local earthquakes or local earthquake tomography (Thurber, 1993). In contrast with the tomographic inversion of controlled sources and teleseisms, the simultaneous inversion of local earthquakes determines the location and origin time of events, along with the velocity of the material between the source and the recording stations. With sufficient data, the inversion can define a local or a regional velocity model, either one-dimensional (1-D; homogeneous or layered) or three-dimensional (3-D; Kissling, 1988; Kissling et al., 1994).

In order to clarify seismotectonic properties, the main objective of this chapter is to infer the optimum CSZ velocity model contained in the local earthquake data of the period 1977-1997. The chapter is divided into ten sections. In Section 5.2, the previous studies of CSZ velocities are presented, along with the routine procedure for locating earthquakes. Section 5.3 discusses two possible sources of errors of the observed phase data, namely the time control of the seismograph network and the picking precision of Charlevoix phase data. The selected data used in the inversion are presented in Section 5.4. With this data set, optimum velocity models of the CSZ are presented: the average Vp/Vs ratio (Section 5.5); the homogeneous velocity model (Section 5.6); and two layered models (one for the north shore and one for lower velocity Appalachian sediments of the south shore; Section 5.7). The impact of the velocity models on the earthquake locations is discussed in Section 5.8. An inversion of the Laurentides Park velocity model is presented in Section 5.9. Finally, the velocity information is discussed in Section 5.10.

5.2 Previous studies of the regional velocity structure and routine analysis of the CSZ data

Crustal velocity models of the Charlevoix-Saguenay area are shown in Figure 5.1. To the west of the CSZ, across the Grenville Front, P-wave upper crustal velocities in the 6.3-6.5 km/s range were determined, with mid to lower crustal values of about 6.55 km/s (Mereu and Jobidon, 1971; Berry and Fuchs, 1973). Later, these velocities were used to model the strong motion records of the 1988 mbLg 6.5 Saguenay earthquake (Somerville et al., 1990). In the CSZ itself, a seismic refraction-reflection survey showed that the velocity structure differed between the two shores of the St. Lawrence River (Lyons et al., 1980). Logan's Line was modeled as a 20o SE dipping interface separating the Appalachian nappes and the Grenvillian rocks with respective P-velocities of 5.5 and 6.1 km/s. Under the north shore, velocities increase linearly from the surface to 18 km depth, where they jump from 6.6 to 6.8 km/s. Near surface velocities are lower inside the Charlevoix impact crater (6.08 ± 0.04 km/s) than in the north shore (6.27 ± 0.11 km/s), a fact possibly related to the intense fracturing. To the northeast of the crater, velocities progressively move towards the average value of 6.2 km/s. To the west of the crater, in the Laurentians, a higher apparent velocity of 6.48 km/s was found for one station.

The 35 microearthquakes recorded during the 1974 Charlevoix experiment were inverted to obtain a crustal velocity model (Leblanc and Buchbinder, 1977). Within the P-velocity range of 5.6 to 6.6 km/s, the 6.2 km/s Pg gave the smallest RMS (with a 3.57 km/s Sg velocity). These velocities were later adopted for Charlevoix routine locations. During the same field survey, surface blasts of the north shore were best located with near surface P-velocities of 6.0 km/s.

In addition to the work of Lyons et al. (1980), the strong velocity contrast between the Grenville rocks and the overlying Appalachian sequence was also modelled with other data sets. Using locally recorded teleseisms, Hearty et al. (1977) modelled the interface as a 19o SE dipping plane separating the 6.1 km/s Grenville from the 5.4 km/s Appalachians. Using some local CSZ earthquakes, Lamontagne (1985) found that a 15o SE dipping boundary between the 6.2 km/s Grenville and the 5.5 km/s Appalachian rocks minimized the RMS. Although these experiments showed that this strong velocity contrast should be considered in CSZ locations, the homogeneous "standard" velocity model continued to be used for routine locations. It was assumed that the shift due to the Appalachian rocks would be approximately the same for all CSZ events, i.e., a few kilometres to the SE (Anglin and Buchbinder, 1980; Lamontagne, 1985).

Over the years, the Geological Survey of Canada (GSC) has used two programs to locate CSZ earthquakes. Between 1978 and 1993, the location program "LOC" (also named "CANCESS") was used. "LOC" uses a nonlinear regression technique, similar to the HYPO-series of programs (Drysdale et al., 1990). Since 1993, Charlevoix earthquakes are routinely located with the GSC program "GRL" which uses a grid search algorithm. The program calculates the RMS for a series of equally-spaced grid points in the origin time, latitude, longitude and depth space. The grid is re-centred on the parameters with the minimum RMS, and the RMS are recomputed using a finer mesh. The grid search stops when the mesh is smaller than a predetermined precision for the origin time, latitude, longitude and depth. The program "GRL" currently accepts a homogeneous or a layered velocity model, with direct or refracted phases. In this thesis, the adopted approach combines the respective advantages of the two location programs. The program "GRL" is less susceptible to local minima due to its grid search approach. "LOC", on the other hand, converges faster to the hypocentre than "GRL" (with a layered velocity model, it could be as much as 30 times faster). Although both "GRL" and "LOC" use layered velocity models, "LOC" can accept two layered velocity models to compute hypocentres. Thus, with "LOC", a velocity structure can be associated with a group of stations. As described in Section 5.7, this property is central to the calculation of the north shore-south shore velocity contrast.

5.3 Sources of uncertainties in the observations

The timing accuracy of the Charlevoix Local Telemetered Network (CLTN) was briefly discussed in Section 1.4. For the period October 1977 to August 1994, time series data for all CLTN stations were time tagged at the concentrator site, providing uniform time across the network. During the period October 1977 to November 1988, two analogue (LMQ and POC) and one digital (LPQ) vertical sensors were in operation at various periods (Figure 1.7). As these stations were vertical component only, their lower precision data were not used to invert the velocity model.

For the post-August 1994 period, Global Positioning System (GPS) timing has been tagged along with the data at each site. This scheme ensures accurate timing at each site, completely independent of the transmission time to the central site. The only possible source of error is the drift of the internal clock that occurs when the signal from a GPS satellite is lost. This, fortunately, rarely occurs and is generally of very short duration (K. Beverly, GSC, pers. comm.). In January 1994, the station LMQ became a digital three-component broadband station, part of the Canadian National Seismograph Network. Accurate timing was provided by a GPS clock, which makes data from this station compatible with the CSZ local network. For station LMQ, however, the 40 Hz sampling rate (0.025 s/sample) makes the phase readings less precise than the 100 Hz (0.01 s/sample) for the local CSZ network.

In terms of station location accuracy, all CSZ stations, including those of the 1996 summer experiment, were located with GPS, accurate to within 25 m for latitude and longitude (G. Girouard, GSC, pers. comm.; Lamontagne et al., 1997). To be consistent with other locations of the network, elevations were scaled on 1/50,000 topographic maps.

In a simultaneous inversion study, the precision of the observed data also depends on the picking of the P and S phases. The picking precision can be evaluated by comparing the picks made by two analysts. To evaluate this precision, phase data from 21 events of the period 1988-1994, originally read by F. Anglin of the GSC, were compared with picks made by the author. The original data were read with the GSC analysis package "SAM" which offers a rotated trace option to ease Sg picking (F. Anglin, pers. comm.). The package "DAN" (Nanometrics Inc) was used for the re-reading. This package is better at enlarging portions of a given trace. The two packages round off the time picks to the nearest hundred of second, implying that the measured time differences between the two picks are given in hundreds of seconds and not in number of samples (0.0125 s/sample for the 1988-1994 CLTN). In general, most phases are impulsive in nature, especially for sites at short epicentral distance or located on the north shore. There were, however, instances where background noise hampered precise phase picking.

The absolute values of the time differences are shown in Figure 5.2. In general, the two sets of P arrivals are more consistent than S arrivals; approximately 80% of phase readings for Pg are within 0.02 seconds whereas about 80% are within 0.05 s for Sg. The larger scatter of Sg picks may be partly due to shear wave splitting which creates smaller amplitude signals prior to the real S arrival (Buchbinder, 1985). Furthermore, the S phase can be emergent rather than impulsive, especially for smaller magnitude events, for larger epicentral distance, and for stations on the south shore which have a higher background noise. In addition, the S onset can be picked as a sharp increase in amplitude or as a weak emergence a few samples before. Compared with three component data, vertical traces do not allow the same precision level, especially for Sg arrivals at larger epicentral distances.

For the post-August 1994 period, the raw data were filtered with a Finite Impulse Response ("FIR") filter which gives rise to unwanted precursory effects prior to the P and S arrivals. In order to eliminate these artifacts, the time series data are routinely band-pass filtered (5-15 Hz pass band) at the analysis stage. This filtering removes the unwanted ringing, while preserving the micro-earthquake signal. With this band-pass filter, the CLTN picks are within 0.01 s of those obtained with the "un-FIR filter" (M. Andrew, GSC, pers. comm.).

5.4 Selection of the earthquake data

The most reliable and consistent hypocentral data for the CSZ were recorded after the local network became a three-component array in November 1988. CSZ events with a minimum magnitude of mN 1.5 were selected to insure conspicuous P and S arrivals on all CSZ stations. To obtain reliable depth control, selected events had to be located within the CLTN 6-station network, and recorded on at least 6 stations with a minimum of 12 seismic phases (P and S). Finally, only events with focal depth of 6 km and deeper were selected, to avoid the poorly converging solutions of some shallow events.

For the period November 1988 to December 1997, 171 events met the selection criteria (Figure 5.3A). Their depth distribution is very similar to that of all CSZ events recorded between 1978 and 1993 (Figure 5.3B). In order to account for the lower precision of S phase picks, P phases were weighted 4 (Quality "A" of the GSC scheme) whereas S phases were weighted 1 (Quality "B"). In the few cases where the residuals were higher than 0.4 s, a weight of 0.25 (Quality "C") was assigned to the Sg pick. With this weighting scheme, the average RMS is 0.084 s, with 90% of RMS less than 0.12 s (Figure 5.3C).

5.5 Vp/Vs Ratio for the CSZ

The traditional method of computing the ratio of P to S wave velocities (Vp/Vs ratio) is the so-called Wadati plot (Wadati, 1933). On this type of graph, where the Sg-Pg travel time is plotted against the Pg travel time, the Vp/Vs ratio is simply equal to the slope of the best fitting line plus one. In the CSZ case, for given hypocentre and station, the Pg travel time is obtained by subtracting the origin time calculated with the "standard" GSC velocity model from the Pg arrival time. The S-P time is computed for each station by subtracting the Pg time from the Sg time. The best fitting line is obtained with a "robust" linear regression method (Huber, 1981). Based on some 1267 P and S phase picks from the data set discussed in Section 5.4, a Vp/Vs ratio of 1.74 is found (Figure 5.4), very close to the assumed value of the GSC "standard" velocity model (1.73, assuming a Poisson's ratio of 0.25). Since the Pg travel time depends on the origin time computed using P and S phases, Wadati plots have some dependence on the S velocity used to derive the hypocentres. In the CSZ, this dependence is small, however, mainly due to the smaller weight assigned to S phases.

5.6 Homogeneous velocity model

The current "standard" homogeneous velocity model was established by Leblanc and Buchbinder (1977) using 35 micro-earthquakes. To test this model with a more complete data set, RMS values were computed for a series of Pg and Sg velocities. Vp was varied between 5.80 and 6.60 by 0.05 km/s increments, while Sg was computed from Vp using a Vp/Vs ratio of between 1.65 and 1.80 by 0.03 km/s increments. Velocities of 6.20 ±0.10 km/s for Pg and 3.58±0.06 km/s for Sg velocities gave the minimum RMS (Figure 5.5). These velocities correspond to the "standard" GSC velocity model, determined by Leblanc and Buchbinder (1977). The Vp/Vs ratio (1.73) is similar to the one determined with the Wadati plot of the previous section.

Although the homogenous velocity model is an acceptable representation, the 0.08s RMS is still higher that the picking precision of P and S phases. For the permanent stations, the Pg residuals, obtained with the "standard" velocity model, show a systematic bias (Figure 5.6). Whereas LMQ, A16 and A11 are more or less centred on 0.00 s, a bias can be observed for the stations A64 (-0.10 s), A61 ( +0.05 s), A54 ( +0.05 s) and A21 ( +0.21 s). Thus, the average velocity to station A64 should be higher than the "standard" velocity model, whereas it should be lower for the three other stations. These results are in line with the seismic refraction results of Lyons et al. (1978), which showed a lower velocity for the south shore (A11 and A21) and for the crater (A54), and higher velocities for the NE portion of the CSZ, outside the crater (A64). To test these regional variations, the data set was subdivided into two groups: one for the southwesternmost stations (the "crater": A11-A16-A61-A54) and one for the northeasternmost stations ("outside the crater": A16-A21-A64-A61). Earthquakes in these two groups were relocated using only the phases from the four stations that defined the polygon, while only the residuals were computed for the two other stations. With this smaller number of stations, the RMS for the two groups were smaller for the "standard" model: 0.046 s for both groups (Figure 5.7). While the "standard" velocity model appears appropriate inside the crater, outside the crater slightly higher velocities provide the lowest RMS (about 6.3 km/s for Pg and 3.64 km/s for Sg). Higher velocities outside the crater were also suggested by Lyons et al. (1980). The Pg residuals for the two groups of stations also reveal a systematic bias: the velocity between the crater events and station A64 should be higher than 6.2 km/s (negative residuals), while velocities to stations A21, A11 and A54, should be somewhat lower (positive residuals; Figure 5.8). To consider this bias, station corrections were defined for the Pg and Sg phases (Table 5.1), bringing the RMS from 0.0838 down to 0.0719 s.

5.7 Calculation of layered velocity models for the north and the south shores

As the minimum RMS with a homogeneous velocity model remains higher than the reading uncertainties, layered velocity models were tested to further reduce the RMS. Since it is known that the largest velocity perturbations arise from the Appalachian rocks of the south shore, it was decided to use two velocity models for the CSZ, one for the north shore stations and one for the south shore stations. To test the validity of this approach, the velocity models of Lyons et al. (1980) were used as a first approximation to calculate travel times to the north shore and south shores (Figure 5.1; Table 5.2). With these models, the RMS was 0.071 s, similar to the best homogeneous model plus the station corrections. On the basis of these positive results, it was decided to further refine these models with the hypocentral data.



Table 5.1: Station corrections to be used with the standard velocity model



Station
Pg

Station Correction

(s)

Sg

Station Correction

(s)

A11 0.00 -0.15
A16 0.00 0.05
A21 -0.10 -0.15
A54 -0.05 0.00
A61 -0.05 0.00
A64 0.10 0.10
LMQ -0.05 -0.05





Table 5.2 Velocities used to model the north and the south shore (based on Lyons et al., 1980)
Depth to top (km) Vp

North Shore

Vp

South Shore

0 6.4 5.5
6 6.5
8 6.5
14 6.6
18 6.8
22 7.0
29 6.9
37 7.1



The best layered velocity models were sought by systematically computing the RMS for a suite of layered velocity models for the north and the south shores. A major constraint of the LOC program is the maximum number of layers (9). In order to prioritize the upper crustal velocity structure, the Appalachian sequence was defined as four 2 km thick layers, while the north shore had only one 8 km thick layer. The north shore and south shore models are the same below 8 km depth. Ranges of velocities and increments were defined for the eight layers of both models (Table 5.3), with the aim of defining the minimum RMS between areas of higher RMS. No velocity reversals with depth are allowed by the program LOC. This restriction is reasonable as a velocity increase with





Table 5.3 Range of P-wave velocities used in the 2-D velocity analysis

North Shore South Shore
Depth to top

(km)

Min Vp

(km/s)

Max Vp

(km/s)

Min Vp

(km/s)

Max Vp

(km/s)

0 6.1 6.4 5.5 5.9
2 5.6 5.9
4 5.6 6.0
6 6.0 6.4
8 6.1 6.4 Same as North Shore
10 6.2 6.7
12 6.2 6.7
14 6.2 6.7



depth is generally assumed for the crust. For a given layer, the S velocity was computed from the P velocity using a Vp/Vs ratio of 1.74, as suggested by the Wadati plot (Section 5.5). The RMS calculations were performed on nearly 28000 models which took 18 days to compute on an Axil 320 Sun workstation. Although this systematic search method is more computer intensive than an iterative method that converges to a minimum RMS, it tests all plausible velocity models.

The resulting velocity models for the north and the south shores are presented in Figure 5.9. The figures present the minimum RMS that can be obtained with a velocity value at a given depth, providing a view of the velocity resolution. Using Figure 5.9B as an example, at 1 km depth, the best RMS that can be expected for a 5.9 km/s P-velocity is in the range between 0.061 s and 0.062 s, well above the best RMS value in the 0.059s range. Hence, for the 0 to 2 kilometer layer, the 5.5 to 5.8 km/s values are more likely. The velocity models that minimize the RMS to 0.0594 s are given in Table 5.4. With these velocity models, the P and S residuals become more tightly clustered at around 0.00s (Figure 5.10). The P residuals are clustered near 0.00 s, which implies that most of the homogeneous model P-velocity biases (Figure 5.6), are resolved with the layered velocity models. The S residuals are more dispersed, a combined consequence of their lower weight in the solutions, the lower precision of the observed data, and possibly, variations in the Vp/Vs ratio.

To a certain extent, the details of the resulting velocity models depend on the assumptions made to derive them. The two most important assumptions are the 0-8 km homogeneous velocity layer for the north shore, and the 1-D model under 8 km. Independent models for the north and the south shores could possibly define a layered structure for the north shore. Similarly, below 8 km, there could be lateral variations between the north and the south shore. These questions could be answered with future 3-D inversion.

Considering the picking precision of the S data, resolving the Vp/Vs ratio for each layer is not feasible. An attempt was made, however, to resolve the Vp/Vs ratio for the Appalachian and the Grenvillian sequences taken as two blocks. Using the minimum





Table 5.4 Velocity model that minimizes the RMS
Depth to top (km) Vp North Shore

(km/s)

Vp South Shore

(km/s)

0 6.1 5.5
2 5.6
4 5.6
6 6.1
8 6.4
10 6.4
12 6.6
14 6.6
40 8.0



RMS model of Table 5.4, Vp/Vs ratios for the two blocks were systematically tested between 1.70 to 1.85 by 0.01 increments; one block for the first four layers (0 to 8 km) of the south shore and one block for the remaining layers of both shores. The best Vp/Vs ratios were 1.72 for the 0 to 8 km south shore model, and 1.73 for the north shore model. With these slightly different Vp/Vs ratios, the RMS was lowered to 0.05844 s (from the original 0.05938 s). Hence, the average CSZ Vp/Vs ratio is close to the "standard" value of 1.73.



5.8 Impact on CSZ earthquake locations

Figures 5.11 and 5.12 compare the earthquake locations inferred here with the homogeneous "standard" velocity model. As expected, most earthquakes relocate to the SE due to the lower velocity Appalachian nappes. The average shift is about 1.5 km along an azimuth of 135o. About two thirds of the events are shallower than the original locations, but the average shift for the whole data set is close to 0 km. The epicentral shifts are relatively small (Figure 5.12). The general pattern of the CSZ earthquakes does not change with the improved velocity model.

5.9 Velocity model between the CSZ and station DAQ

One interesting question is how different the CSZ velocity structure is compared with the mostly aseismic Grenville Province. The seismograph station Lac Daran, Québec (DAQ), located some 100 km west of the CSZ (Figure 1.11), has recorded numerous CSZ events, as small as mN 1.0. These phase data can be used to infer a 1-D velocity structure between the CSZ and station DAQ (thereafter named the Laurentides Park region). In addition to providing comparisons with the CSZ, the regional crustal velocity model could be used in future studies of the Saguenay earthquake zone.

A total of 96 events with mN 2.0 were chosen from the data set used in the CSZ inversion (Figure 5.13A). The hypocentral information (latitude, longitude, depth, origin time) was pegged at the values computed with the CSZ stations and the "standard" homogeneous velocity model. This data set has a RMS of 0.7883 s, due to rather high residuals for Pg (centred near -0.7 s) and Sg (centred near -0.3 s), which indicate higher velocities than the "standard" velocity model (Figure 5.13B). DAQ phase readings are used to compute the RMS of the events, with a weighting scheme of 4, 1 and 0.25 for P, S and the absolute value of the residuals 1.0 s respectively. The Wadati plot indicates a Vp/Vs ratio of 1.786, with some scatter in the phase information (Figure 5.14. The best homogeneous velocity model gives a minimum RMS of 0.439, with Pg and Sg velocities of 6.45 and 3.69 km/s respectively (Vp/Vs ratio of 1.75; Figure 5.15). The best layered model gives a minimum RMS of 0.430 s, assuming a Vp/Vs ratio of 1.786 (Figure 5.16; Table 5.5). With this velocity model, Pg and Sg residuals are closer to 0.0 s, with some remaining events with residuals exceeding ±0.5 s (Figure 5.17). Some events have anomalous P and S residuals, due possibly to differences in timing between the CLTN network (prior to August 1994) and station DAQ of the ECTN network. Two events, recognized as earthquake multiplets, had very different relative arrival times on DAQ. The timing difference between the CLTN and the ECTN cannot be measured directly.

To clarify whether the high P and S residuals could be due to timing, the layered velocity model was recomputed with events recorded after November 1995, i.e., after both the CLTN and station DAQ relied on GPS clocks. Events were selected with the same criteria as before, with additional events recorded between January and March 1998. The 44 event data set had a RMS of 0.644 s, with a large scatter of P and S residuals (Figure 5.18). The Wadati plot is less scattered and gives a Vp/Vs ratio of 1.89 (Figure 5.19). The homogeneous velocity model minimizes the RMS (0.239) with Pg and Sg velocities of 6.45 km/s and 3.57 km/s respectively (Vp/Vs ratio of 1.81; Figure 5.20). The best layered model has P-velocity of 6.0 km/s down to 6 km, and 6.6 km/s beneath 6 km (assuming



Table 5.5 Velocity model for the Laurentides Park region

Depth to top of layer

(km)

Range of

Vp values

(km/s)

Optimum value

(km/s)

Minimum Maximum
0 6.0 6.6 5.9
2 6.0 6.6 6.1
4 6.0 6.6 6.2
6 6.2 6.7 6.3
8 6.2 6.7 6.4
10 6.3 6.7 6.7
14 6.3 6.7 6.7
18 6.3 6.8 6.7
40 8



Vp/Vs of 1.81; Figure 5.21). The RMS is reduced to 0.238 s, slightly less than the homogeneous model. Although most of the P and S residuals are recentred near 0.00 s (Figure 5.22), high anomalous residuals remain, especially for the S phase. Except for one cluster of five events with high S residuals, most events with high P and/or S residuals are not clustered in one specific part of the CSZ. Since the timing uncertainty is not the source of the problem, variations in the Vp/Vs ratio must exist in the Laurentides Park region. The positive S residuals suggest a lower-than-average S velocity, implying a Vp/Vs ratio > 1.81.

The anomalous S residuals imply that the misfits of the 1988-1997 data set were not entirely due to bad time control of the ECTN and the CLTN. One reason for these anomalies can be that the geology between the CSZ and station DAQ is not homogeneous, but made up of various lithologies. For a given ray path, S residuals will vary according to the Vp/Vs ratio of the geological unit. Another possibility is that the P and the S phases are using different ray paths, possibly not the direct ones, due to the complex geology of the area. This could lead to anomalous S residuals.

5.10 Impact of the velocity models on the understanding of the CSZ

From the local hypocentre data, it has been possible to infer velocity information for the CSZ and the Laurentides Park region and to reduce significantly the RMS of the earthquake solutions. The current CSZ network is sufficient to invert a "pseudo-2-D" P-velocity model, i.e., two 1-D velocity models, one for the Appalachian sequence under the south shore, and one for the Grenville. For the south shore model, the 6 km thick Appalachian rocks have P-velocities in the 5.5-5.6 km/s range. For the Grenville, transitions in the P-wave velocities occur at 8 and 12 kilometres, with a passage from an average 6.2 km/s upper crustal to > 6.5 km/s mid-crustal velocities. Interestingly, this velocity transition also corresponds to the peak in the earthquake occurrence distribution, possibly related to a lithological change. In general, the velocity results are similar to those obtained in seismic refraction surveys (Figure 5.23). The layered velocity models could be used as the starting model in future 3-D hypocentre-velocity inversion. Appendix 3 discusses how the current CSZ network could be upgraded to allow 3-D velocity determination.

Using the best layered velocity model, earthquake locations shift by an average 1.5 km to the southeast, with most hypocentral depths remaining within 2 km of the original positions. Although individual hypocentres may shift with the new velocity model, the hypocentre distribution at a whole may not be very different. The uncertainty in absolute hypocentre positions must be kept in mind while correlating earthquakes with lineaments or structures at depth.

In the Laurentides Park region, the P-velocity structure differs from the CSZ, especially in terms of the Vp/Vs ratio (1.81 and 1.73, respectively). Variations in Vp/Vs ratios are related to difference in pore fluid pressure, fabric and composition (Musacchio et al., 1997). High Vp/Vs ratios can correspond to high pore pressure, foliation parallel to the ray path or low quartz content. With high pore-fluid pressure, it is a decrease in S velocity, rather than an increase in P velocity, that increases the Vp/Vs ratio. Based on the homogeneous model for the Laurentides Park (Figure 5.20) however, we know that Vp is higher and Vs is similar to the CSZ. The second possible explanation, the parallel foliation, is also unlikely: constant foliation attitude is not realistic over some 100 km distance in the very geologically complex Laurentides Park region. The probable explanation lies in a difference in the mid crustal lithology. The Laurentides Park contains typical Grenvillian rocks, such as mangerite, charnockite and garnet gneiss. Similar high Vp/Vs ratios and high P-wave velocities are found in Grenvillian crust further south, in the Southern Ontario-Northern New York region (Musacchio et al., 1997). Consequently, the Laurentides Park region appears to be a "standard" Grenvillian crust, with some spatial Vp/Vs variations as suggested by the misfit S residuals.

The CSZ, on the other hand, appears to differ from the Grenvillian crust found in the Southern Ontario-Northern New York region (average crustal Vp/Vs ratios: 1.73 for the Appalachians; 1.81 for the Grenville; Musacchio et al., 1997). While the CSZ Appalachian ratio is similar (1.72), the CSZ Grenville ratio is much lower than elsewhere (1.73). The reason for the CSZ average Vp/Vs ratio is not obvious. High pore fluid pressure can explain fault reactivation (Chapter 2), but its relationship with low Vp/Vs in unclear. High pore fluid pressure can be highly localized along fault zones. The Vp/Vs ratio, on the other hand, results from averaging over a whole region, including some inactive, and possibly dry, blocks. Unless pore fluids are pervasive, their presence may not be revealed by the current data. In some experiments with high pore pressure, the reduced effective pressure brings about an increase in the Vp/Vs ratio (increase in Poisson's ratio; Spencer and Nur, 1976). The interrelations between Vp/Vs, lithology, pore geometry and pressure are not sufficiently well understood to be definite (Wenzel and Sandmeier, 1992). Lithologically, an average Vp/Vs ratio corresponds to granite at mid-crustal depth (Musacchio et al., 1997). This is incompatible with the outcropping CSZ lithologies (mangerites, charnockites and garnet gneisses) that have Vp/Vs ratios in the 1.78 to 1.81 range. The most likely explanation is that, with the CSZ data set, the more felsic upper crust (Vp/Vs 1.73) is better sampled than the more basic lower crust. According to Figure 5.9A, velocities in the 6.5-6.6 km/s range, generally associated with more basic rocks, are found below 12 km depth. Hence, most of the travel path, on which Vp/Vs is based, goes through more felsic rocks. In the Laurentides Park case, on the other hand, the basic rocks are shallower (below 6 km depth), and most of the travel paths take place within them. In the future, independent calculations of Vp and Vs at depth may validate this interpretation.


> > > To Chapter 6


Figure Captions

Figure 5.1 P-wave velocity models of the Charlevoix and the Saguenay regions. The CSZ North and South shores are the velocity models obtained from seismic refection-refraction surveys (Lyons et al., 1980). The "standard" velocity model is the model used for routine earthquake locations in eastern Canada, including the CSZ. The "standard" model assumes a Vp/Vs ratio of 1.73. The Saguenay model was defined by Somerville et al. (1990) on the seismic-refraction survey described in Berry and Fuchs (1973).

Figure 5.2 Absolute time differences between the Pg and Sg arrivals picked by F. Anglin of the GSC and the author for some 21 events. About 90 % of the Pg picks and 80% of the Sg picks match within 0.05 seconds.

Figure 5.3 Characteristics of the events selected for the velocity inversion: (A) Epicentral map with black lines representing the faults of the Precambrian shield; (B) Depth distribution of the events by 1 km classes in absolute numbers (left) and in cumulative numbers (right). (C) RMS distribution of the events selected, calculated from the "standard" velocity model.

Figure 5.4 Wadati plot for the events selected.

Figure 5.5 RMS values for various combinations of Vp and Vs (scale in seconds). The white areas represent Vp and Vs combinations that were not used.

Figure 5.6 Residuals for the various CSZ stations grouped by 0.05 s classes; (A) For P arrivals; (B) For S arrivals. As in similar graphs to follow, the value plotted represents the number of events that is less or equal to the X value. For example, the 80 values at 0.00 s for A16, means that 80 values are in the -0.04 to 0.00 s range.

Figure 5.7 RMS values, as in Figure 5.5: (A) Using events within the Charlevoix crater, located only with data from stations A11, A16, A61 and A54. (B) Using events outside the crater, located only with data from stations A16, A21, A64 and A61.

Figure 5.8 P-wave residuals for events located inside (A64 and A21) and outside (A54 and A11) the crater.

Figure 5.9 Best RMS obtained as a function of P-wave velocity (scale in seconds) and depth of layers: (A) North Shore; (B) South Shore.

Figure 5.10 Number of residuals grouped in 0.05 s classes for the CSZ stations: (A) P wave; (B) S wave.

Figure 5.11 Shifts between original event locations (X = Y = 0) and locations computed with the best layered model: (A) Epicentres; (B) Hypocentres along E-W section. Blue crosses are shallower relocated hypocentres and red diamonds deeper ones.

Figure 5.12 Epicentral map showing shift between original (yellow) and the relocated (red) epicentres.

Figure 5.13 CSZ events recorded at station DAQ; (A) Hypocentres; (B) Distribution of P and S residuals.

Figure 5.14 Wadati plot for the CSZ events recorded on DAQ.

Figure 5.15 Vp-Vs tests for the DAQ data with RMS values in seconds.

Figure 5.16 Best RMS for DAQ data obtained for given P-wave velocities and depth of layers.

Figure 5.17 Number of DAQ residuals for P and S waves grouped by 0.05 s classes.

Figure 5.18 Distribution of the P and S residuals grouped in 0.05 s classes for the CSZ earthquake data recorded on DAQ between November 1994 and March 1998.

Figure 5.19 Wadati plot for CSZ events recorded on DAQ between November 1995 and March 1998.

Figure 5.20 RMS calculated for crustal P and S velocities with CSZ events recorded on DAQ between November 1995 and March 1998.

Figure 5.21 Best RMS values at given depth and P velocities for a layered velocity model, using CSZ events recorded on DAQ between November 1995 and March 1998.

Figure 5.22 P and S residuals computed with the best layered velocity model, using CSZ events recorded on DAQ between November 1995 and March 1998.

Figure 5.23 Comparison of the computed 1-D velocity models with other models of the Charlevoix and Saguenay regions.