Crustal structure of the south-central Andes Cordillera and backarc region from regional waveform modelling

Transcription

1 Geophys. J. Int. (27) 17, doi: /j X x Crustal structure of the south-central Andes Cordillera and backarc region from regional waveform modelling P. Alvarado, 1,2 S. Beck 3 and G. Zandt 3 1 Consejo Nacional de Investigaciones Científicas y Técnicas (CONICET), Argentina. alvarado@unsj.edu.ar 2 Departamento de Geofísica y Astronomía, Facultad de Ciencias Exactas, Físicas y Naturales, Universidad Nacional de San Juan, Meglioli 116 S, San Juan, 54, Argentina 3 Department of Geosciences, University of Arizona, 14 E 4th St., Tucson, AZ 85721, USA GJI Tectonics and geodynamics Accepted 27 March 25. Received 27 January 22; in original form 26 June 26 SUMMARY We investigate the crustal structure in the Andes Cordillera and its backarc region using regional broadband waveforms from crustal earthquakes. We consider seismic waveforms recorded at regional distances by the CHile-ARgentina Geophysical Experiment (CHARGE) during 2 22 and utilize previous seismic moment tensor inversion results. For each single station-earthquake pair, we fixed the source parameters and performed forward waveform modelling using ray paths that sample the crust of the highest elevation Cordillera and the accreted terranes in the backarc region. Our investigation indicates that synthetic seismograms for our earthquake-station geometry are most sensitive to crustal parameters and less sensitive to mantle parameters. We performed a grid search around crustal thickness, P-wave seismic velocity (Vp) and P- to S-wave seismic velocity ratio (Vp/Vs), fixing mantle parameters. We evaluated this waveform analysis by estimating an average correlation coefficient between observed and synthetic data over the three broadband components. We identified all acceptable crustal models that correspond to high correlation coefficients that provide best overall seismogram fits for the data and synthetic waveforms filtered mainly between 1 and 8 s. Our results indicate along strike variations in the crustal structure for the north south high Cordillera with higher P-wave velocity and thickness in the northern segment (north of 33 S), and persistently high Vp/Vs ratio (>1.85) in both segments. This is consistent with a colder mafic composition for the northern segment and a region of crustal thickening above the flat slab region. In contrast, the results for the current volcanic arc (south of 33 S) agree with a warmer crust consistent with partial melt related to Quaternary volcanism presumably of an intermediate to mafic composition. A distinctive feature in the backarc region is the marked contrast between the seismic properties of the Cuyania and Pampia terranes that correlates with their heterogeneous crustal composition. The Cuyania terrane, composed of mafic-ultramafic rocks, exhibits high Vp, high Vp/Vs and a thicker layered crust versus the thinner more quartzrich crust of the eastern Sierras Pampeanas associated with low Vp and low Vp/Vs. These differences may have some effect on the mechanism that unevenly generates crustal seismicity in the upper 3 km in this active compressional region. In particular, the seismic properties of the Cuyania terrane, which shows evidence for a high Vp and high Vp/Vs crust, may be reflecting the complex tectonic evolution history of this terrane including accretion-rifting and re-accretion processes since the Palaeozoic that promote a high level of crustal seismicity in the upper 3 km, enhanced by the flat slab subduction in the segment between 31 S and 32 S. Another possible mechanism could be directly related to the presence of a strong lower crust above the flat slab that efficiently transfer stresses from the slab to the upper crust generating higher seismicity in the Cuyania terrane between 3 S and 34 S. Key words: Andean crust, broadband waveforms, continental seismicity, subduction zone, terranes. Downloaded from by guest on 2 January C 27 The Authors Journal compilation C 27 RAS

2 Crustal structure of the Andes Cordillera and backarc region INTRODUCTION, TECTONIC SETTING AND TERRANE OVERVIEW The highest (>6 m) elevations along the Andean Cordillera lie above the transition in the subduction zone geometry where the Nazca slab changes its dip from near horizontal to 3 (normal) to the east (Barazangi & Isacks 1976; Cahill & Isacks 1992; Anderson et al. 27; Fig. 1 and inset in Fig. 2). Both the convergence between the Nazca and South American plates, at a rate of 6.3 cm yr 1 and azimuth of 8 (Kendrick et al. 23), and the subduction geometry are first-order factors producing north south variations in volcanism and crustal seismicity (Jordan et al. 1983; Gutscher et al. 2a; Kay & Mpodozis 22). The modern volcanic arc runs along the normally dipping slab south of 33 S (see Stern 24 for a summary). North of this latitude, however, volcanism was active until the most pronounced flattening of the subducted plate at 8 1 Ma (Kay et al. 1991). Continental crustal seismic activity is very intense above the flat slab segment (Gutscher et al. 2a), a region characterized by the Cambrian-Ordovician carbonate thinskinned Precordillera fold-thrust belt (Baldis & Bordonaro 1981) and the thick-skinned Sierras Pampeanas basement uplifts, considered a modern example to interpret the Laramide deformation in North America (Jordan & Allmendinger 1986) (Fig. 1). South of 33 S, the frequency of crustal earthquakes in the high Cordillera is also high (Barrientos et al. 24). The south-central Andes backarc region is a mosaic of accreted terranes (Fig. 1) with two main collisions against the Gondwana margin. The Pampia terrane collided at 53 Ma (Rapela 2), and the Cuyania terrane collided at 46 Ma (Ramos 24). The Cuyania terrane is composed of the Precordillera and the westernmost Sierras Pampeanas terranes, neither of which show evidence of Cambrian-Ordovician magmatism related to an active subduction zone by that time (e.g. Ramos 24). In addition, basement rocks of a Grenville (1 12 Ma) age in the Precordillera (Kay et al. 1996) and in this sector of the Western Sierras Pampeanas (WSP) (McDonough et al. 1993) seem to be part of the same basement structure with an intra-grenville suture (Vujovich & Kay 1998; Vujovich et al. 24) (Fig. 1). Yet no consensus exists regarding their time of amalgamation or individual provenances (e.g. Dalla Salda et al. 1992; Aceñolaza et al. 22; Thomas & Astini 23; Finney Downloaded from by guest on 2 January 219 Figure 1. Location map of the region of study showing crustal earthquakes in the Andes Cordillera and its backarc region during the CHARGE experiment, their focal mechanisms in lower hemisphere projection constrained from regional moment tensor inversion by Alvarado et al. (25), the CHARGE seismic network, plate convergence velocity from Kendrick et al. (23), slab contours in km from Cahill & Isacks (1992), sutures, terranes, tectonic provinces, Quaternary volcanism and mineralized areas based on Kay & Mpodozis (21), Ramos et al. (22), Bissig et al. (22), Ramos (24) and Stern (24). C 27 The Authors, GJI, 17, Journal compilation C 27 RAS

3 86 P. Alvarado, S. Beck and G. Zandt Figure 2. Earthquake-station paths studied here with forward broadband waveform modelling at near-regional distances (1 35 km). The best layer thickness results for each event-station pair using a layer over half-space model are shown along the ray paths (see Table 1 for other crustal parameters). Tectonic provinces, terranes, sutures and mineralization areas as in Fig. 1. The inset shows elevation contours every 2 km. Note that elevations greater than 5.5 km are in white. Approximate locations of the PANDA seismic experiment (Regnier et al. 1992; Smalley et al. 1993) and CHARGE experiment are also shown. et al. 23, 24; Galindo et al. 24). Additional U-Pb Grenville age observations at 37 S suggest the Cuyania terrane extends to the south (Sato et al. 24; Ramos 24). The most recent accretionary event produced the onset of the allochthonous Chilenia terrane in the west at 4 Ma (Davis et al. 1999). Superimposed rifting events in Late Palaeozoic and Mesozoic caused extensive felsic magmatism and extensional faulting, frequently developing sedimentary backarc basins on the hanging walls of Palaeozoic terrane sutures (Franzese & Spalletti 21; Franzese et al. 23). The Andean compression inverted these previous extensional structures associated with the uplift of the Sierras Pampeanas and caused reactivation of inherited weak fault zones (Ramos 1994; Ramos et al. 22). The contrasting terrane inheritance may be responsible for the different crustal composition and tectonic evolution of the eastern and western terranes. The predominant felsic quartz-rich character of the Eastern Sierras Pampeanas (ESP) is linked to the collision of a microcontinent, the Pampia terrane (Rapela et al. 1998), whereas the dominant mafic-ultramafic composition in the west is associated with oceanic fragments of a crust arc/backarc setting for the WSP (Vujovich & Kay 1998; Ramos et al. 2) and a passive-margin platform for the Precordillera (Thomas & Astini 23). Given the increasing interest in relating provenance and composition of the crust with the evolution of active margins, we have analysed broadband seismic waveforms from regional crustal earthquakes in the central-western part of Argentina using a forward modelling technique to investigate the characteristics and sensitivity of the waveforms to seismic parameters in the crust and uppermost mantle. Investigating the seismic properties along single stationsource pairs in each terrane enable us to compare and contrast the heterogeneous crust of this region and explore the implications for present Andean tectonic deformation. 2 PREVIOUS REGIONAL SEISMIC STUDIES Recent regional studies using broadband data from the CHile ARgentina Geophysical Experiment (CHARGE) have constrained a thick ( 65 km) crust beneath the high Cordillera, which thins slightly to 55 km in the Precordillera and WSP along the northern CHARGE transect (Fromm et al. 24) (Fig. 1). This region has high P-wave velocity (Vp) of 6.4 km s 1 (Alvarado et al. 25) and high P-toS-wave velocity ratio (Vp/Vs) > 1.8 (Alvarado et al. 25; Calkins et al. 26; Gilbert et al. 26). In addition, a complex multilayer structure with probable eclogitization in the lower levels characterizes the western terrane crust, producing a weak teleseismic receiver function Moho converted phase arrival (Gilbert et al. 26). In contrast, the same studies show that the easternmost terranes have a higher-amplitude and single Moho signal defining a much thinner ( 35 km) crust with lower Vp of 6. km s 1 and low Vp/Vs < 1.7. Local studies in using a short-period Downloaded from by guest on 2 January 219 C 27 The Authors, GJI, 17, Journal compilation C 27 RAS

4 Crustal structure of the Andes Cordillera and backarc region 861 seismic network near San Juan (the PANDA experiment; see location in the inset of Fig. 2) had previously suggested a thick (>5 km) crust in the WSP and Precordillera (Regnier et al. 1994). Recent analyses of pmp phases observed at teleseismic distances have constrained crustal thicknesses of 5 59 km beneath the Precordillera and 58 and 61 km in the high Cordillera (McGlashan et al. 26). Seismic reflection studies in the basins around station JUAN in the WSP (Fig. 1) have shown reflectors at 18, 3 and 38 km within the crust (Snyder et al. 199; Comínguez & Ramos 1991; Zapata & Allmendinger 1996) and a transitional acoustic boundary at the base of a 52-km crust (Zapata 1998). At 36 S mid-crustal receiver function arrivals along the southern CHARGE transect are consistent with a thicker crust of 5 km west of 7.5 W related to the active arc, which contrasts with a single Moho interface at 4 km depth in the backarc region (Gilbert et al. 26) (Fig. 1). We explore the crustal structure in between the two CHARGE transects and within the north south trending terranes. Although we selected mainly north south earthquake-receiver paths to conduct a grid-search study, we occasionally considered other geometries to test particular problems. For example, we use the only east west available path in the Río de la Plata craton (Fig. 1) to study the crust in this region. 3 METHODS In this and the following two sections we describe the data used in the regional forward broadband waveform modelling and the grid search technique applied in this study. 3.1 Regional seismic moment tensor inversion data A previous study constrained the seismic moment tensor (SMT) for 27 crustal earthquakes, with magnitudes M w between 3.5 and 5.1, recorded at regional distances out to 7 km on the portable broadband CHARGE stations (Alvarado et al. 25), using an inversion technique from Randall et al. (1995). This technique consists of modelling multiple three-component full-waveforms with good azimuthal coverage, for a fixed hypocentre using an average 1-D crustal model. A series of sensitivity tests including a range of epicentral distances in the inversion were run to evaluate errors related to earthquake mislocations. For each earthquake, a SMT inversion at a series of trial hypocentral crustal depths was performed. In addition, several average crustal models were also tested to observe variations in the SMT inversion results for selected events. Alvarado et al. (25) showed that SMT inversion solutions using an average crustal structure are very robust for the focal mechanism, and the best fit between observed and synthetic waveforms occurs for the best hypocentral depth. Varying the average crustal structure caused no significant variations in the overall focal mechanism and best hypocentral depth results (Alvarado et al. 25). However, the SMT inversion technique has a third-order dependence on the structure and much better synthetic seismogram fits to observed data can be obtained using more appropriate crustal models along each seismic path between the earthquake and the CHARGE station. In this study, we consider those same crustal seismic events (Alvarado et al. 25) and the same frequency bands to filter seismograms. Fixing the previously determined source parameter estimates (SMT for the major dislocation and hypocentral depth already constrained from modelling) for a particular crustal earthquake, we performed a more detailed analysis of the crustal structure between the event and a single CHARGE station. By investigating a number of event-station paths sampling different terranes we develop a regional crustal velocity model for each terrane. 3.2 Regional forward broadband modelling We have used twelve crustal events in the Andean high Cordillera and backarc region from Alvarado et al. (25) ranging in size from M w = 4. to 5.1 and broadband CHARGE data recorded at epicentral distances < 4 km (Figs 1 and 2). We refer to these earthquakes by their year and Julian day of occurrence. These moderate earthquakes have similar epicentral locations with uncertainties < ±6 km determined by the local network of INPRES operating in the region, relocation techniques using the CHARGE data (Anderson et al. 27), and body wave tomographic CHARGE studies (Wagner et al. 25). The events display reverse or strike-slip focal mechanisms and focal depths from 5 to 26 km with maximum errors of 5 km (Table 1 and Fig. 1). The broadband waveforms of these crustal earthquakes display clear direct P wave (Pg) and head P wave (Pn) arrivals, followed by reflections and conversions between the surface and the Moho. The most prominent arrivals in the waveforms are the surface waves. We use the forward modelling technique of Randall (1994) as applied by Swenson et al. (2) for regional-distance waveforms, which consists of calculating the three-component full seismic displacements at regional distances using a layered velocity structure. The most reliable structure between the earthquake and a single station should produce the best correlation of the synthetic to the observed data. As a first step we conducted a series of sensitivity tests to recognize which parameters of the crust and upper mantle have a stronger effect on the waveforms and whether there are any trade offs by using combinations of them. Then, based on the different crustal properties observed in the CHARGE region, we performed a grid search for seismic velocity parameters (average crustal P-wave velocity, P- tos-wave velocity ratio and thickness) that produce the best correlation between synthetic and observed waveforms for the three-component broadband records at regional distances. We use this information to map regional crustal differences. 4 REGIONAL WAVEFORM SENSITIVITY TESTS We have calculated synthetic three-component seismograms for the largest (M w 5.1) crustal earthquake during the CHARGE period, event (Fig. 1), to investigate the sensitivity of the complete waveforms to crustal thickness, average crustal Vp, average crustal Vs, upper-mantle Vp and upper-mantle Vs. Wefixed the seismic source at 5 km depth with a focal mechanism of strike 36, dip 75 and rake 198 (Alvarado et al. 25). We used an epicentral distance and azimuth of 31 km and 192, respectively, to simulate the recording conditions at station MAUL (Figs 1 and 2). We used the TauP-toolkit software (Crotwell et al. 1999) to create our simple seismic models and to predict arrivals and their traveltimes. We observed clear arrivals from the Pn, Pg, the Moho reflection Pmp, Sg as well as the low-frequency regional surface waveforms. In addition, there are several phases between the P- and S-wavetrains, such as the phase PmpSmp with long period energy but smaller amplitudes at regional distances. Lower case p denotes the traditional representation for an upward travelling P wave. Overall, the waveforms in this study are dominated by well-developed regional surface waves which are very sensitive to crustal parameters, as shown below. Downloaded from by guest on 2 January 219 C 27 The Authors, GJI, 17, Journal compilation C 27 RAS

5 862 P. Alvarado, S. Beck and G. Zandt Table 1. Grid search results for different earthquake-single CHARGE-station pairs used in this study (Figs 1 and 2). Event/station M w /depth (km) Distance (km) Periods (s) Correlation coefficient Thickness (km) Vp (km s 1 ) Vp/Vs 1-127/PACH 4.1/ (35 43) 6.4 ( ) 1.85 ( ) /HUER 4.1/ (2 35) 6. (5.9 6.) 1.7 ( ) 1-138/LLAN 4.3/ (35 4) 6.4 ( ) 1.65 ( ) 1-285/NEGR 5.1/ (5 64) 6.6 ( ) 1.85 ( ) 1-285/ELBO (45 52) /MAUL (45 5) 5.8 (5.8 6.) 1.85 ( ) 1-285/LENA (42 48) 5.8 ( ) 1.8 ( ) 1-285/PENA (43 46) 6.4 ( ) 1.8 ( ) /BARD (4 47) 6.2 ( ) /NIEB (38 47) 6.2 ( ) /PACH (45 5) 6.6 ( ) 1.9 ( ) (55 6) /PENA 4.1/ (32 37) 6. ( ) 1.8 ( ) 1-348/RINC( ) 4.4/ (23 37) 6.4 ( ) 1.85 ( ) (42 44) 6.2 ( ) 1.85 ( ) 1-349/LLAN 4./ (27 33) /PICH (27 33) ( ) 1-352/LLAN 4.5/ (33 37) 6. (6. 6.2) 1.7 ( ) 1-352/PICH (4 45) 5.8 ( ) /HEDI (35 48) 6.2 ( ) /JUAN (25 3) (55 62) /USPA (25 35) 6.2 ( ) /USPA 4./ (55 63) /JUAN (55 6) 6.2 (6. 6.3) /JUAN 4.7/ (3 37) 6.2 (6. 6.2) 1.85 ( ) 2-5/PENA (4 5) 6.2 ( ) 1.8 ( ) 2-5/HEDI (27 35) 6. ( ) /BARD (3 37) 6.2 ( ) 1.8 ( ) /NIEB (3 45) 6.2 ( ) /PICH 4./ (3 35) 6.2 ( ) /JUAN 4.2/ (35 4) 6.2 (6. 6.2) 1.8 ( ) 2-17/USPA (38 53) ( ) 2-117/USPA 5.1/ (45 5) ( ) /HEDI (35 39) 6. ( ) 1.85 Notes: The source parameters in the first column include year-julian day of occurrence, station name, magnitude (M w ) and focal depth from a previous regional moment tensor inversion analysis (Alvarado et al. 25). Distance refers to the respective epicentral distance; Periods, to the corner periods used in the bandpass filtering; Correlation coefficient, to the maximum average correlation between observed and synthetic data among the three-component waveforms for the best model; Thickness, Vp and Vp/Vs, to the best layer over half-space model that produces the best waveform fits and hence the maximum correlation. The last three columns also show a range of crustal parameters in parentheses for a 5 per cent of less correlated data around the best model estimated simultaneously in the three seismogram components. A star after the station name indicates that only two seismic components were used. 4.1 Sensitivity to crustal thickness We started testing sensitivity of the waveforms to crustal thickness using a crustal layer over mantle half-space model. We fixed the crustal Vp at 6.4 km s 1, mantle Vp at 8.15 km s 1 and Vp/Vs at 1.8, and calculated three-component full waveforms for crustal thicknesses varying from 3 to 75 km in steps of 5 km. The fixed crustal parameters are consistent with previous observations in this region (Fromm et al. 24; Alvarado et al. 25; Gilbert et al. 26). Fig. 3 shows the results of this sensitivity analysis for five different thicknesses on the radial and tangential seismic displacement components. Radial components are affected by both P- and S waves (Figs 3a and c) while S-wave phases are more clearly observed in tangential components (Figs 3b and d). In this and following waveform figures, all seismograms are aligned on the first P-wave arrival. We observe a variation in the amplitudes of the P waves and the surface waves, which show an increase in amplitude for an increase in the crustal thickness on all three components. There are also variations in timing and amplitude of the phases arriving between the P and S waves. 4.2 Sensitivity to average crustal P-wave velocity We explore the sensitivity of the waveforms to variations in crustal Vp. Wefixed the crustal thickness at 5 km, used mantle parameters as before and varied the average crustal Vp from 5.6 to 6.6 km s 1 in steps of.2 km s 1 (Fig. 4). Since this will also introduce Downloaded from by guest on 2 January 219 C 27 The Authors, GJI, 17, Journal compilation C 27 RAS

6 Crustal structure of the Andes Cordillera and backarc region 863 a) 35 km Crustal b) 1 Thickness displacement (x 1 m) km 55 km 65 km 75 km Radial component 5 1 displacement (x 1 m) km 45 km 55 km 65 km 75 km Tangential component 5 1 Crustal Thickness c) displacement (x 1 m) Bandpass 1-8 s 5 1 time (s) d) displacement (x 1 m) km 45 km 55 km 65 km 75 km Bandpass 1-8 s 5 1 time (s) Figure 3. Sensitivity of synthetic broadband waveforms to variations in crustal thickness using a single layer over half-space model. We show broadband seismic displacements computed for event at a distance and azimuth of 31 km and 192, respectively. Fixed seismic parameters are crustal Vp (6.4 km s 1 ), crustal Vp/Vs (1.8), mantle Vp (8.15 km s 1 ) and mantle Vp/Vs (1.8). (a) Radial component broadband seismograms. (b) Tangential component broadband seismograms. (c) Radial component filtered using a bandpass between 1 and 8 s. (d) Same as (c) for the tangential component. variations in Vp/Vs for the crust, we fixed Vp/Vs at 1.8 calculating the associated S-wave velocity in each model (Vs varying from 3.11 to 3.67 km s 1 ). The most direct impact on waveforms of variations in crustal Vp is a variation in timing and amplitude of the Rayleigh waves and the intermediate arrivals between P and S waves (Fig. 4). Increasing the crustal Vp decreases the traveltime and decreases the amplitude between peaks and troughs for the waves. This is to be expected, because the contrast between crustal and mantle velocities is diminishing. 4.3 Sensitivity to crustal P- to S-wave velocity ratio We analyse the sensitivity of waveforms to the crustal Vp/Vs ratio by initially fixing Vp at 6.4 km s 1 and allowing Vp/Vs to vary from 1.6 to 1.9. We also fixed the crustal thickness at 5 km and mantle parameters as before. We repeated this test for different fixed crustal Vp calculating each time the corresponding Vs to produce the same Vp/Vs range explored. As expected, increasing Vp/Vs produces a significant delay in the S waves and surface waves (Fig. 5). On the filtered records the large amplitude arrival increases for a greater Vp/Vs on the tangential component. The effect is also more visible on the radial than on the vertical component. This makes sense, as the tangential and radial components are dominated by S waves (Fig. 5). 4.4 Sensitivity to mantle parameters We ran two different tests allowing first variations in the mantle P-wave velocity and later in the mantle Vp/Vs ratio. We started fixing the average crustal Vp at 6.4 km s 1, crustal and mantle Vp/Vs at 1.8 and crustal thickness at 5 km and computed synthetic waveforms for a series of mantle Vp varying between 7.6 and 8.4 km s 1 with an increment of.2 km s 1. The second test consisted of fixing the crustal parameters as described above, fixing the mantle Vp at 8.15 km s 1 and calculating synthetic seismograms for Vp/Vs in the mantle varying from 1.65 to 1.9 with an increment of.5. This range in Vp/Vs corresponds to variations in the mantle S-wave velocity from 4.23 to 4.94 km s 1. Neither variations in Vp nor Vp/Vs for the mantle significantly modify the regional calculated waveforms (Figs 6 and 7). This result is very useful in our grid search approach. Because waveforms are much less sensitive to the mantle parameters, these parameters can be fixed. 5 GRID-SEARCH FORWARD MODELLING TO DETERMINE CRUSTAL STRUCTURE The sensitivity test results indicate that regional broadband waveforms (at 3 km distance) from crustal earthquakes are much more sensitive to parameters in the crust than in the upper mantle. However, there is a trade off among the crustal parameters themselves (thickness, Vp and Vs) (compare Figs 3 5). Since variations in the mantle parameters do not produce significant variations in the waveforms, we fixed Vp at 8.15 km s 1 and Vp/Vs at 1.8 in the upper mantle for all tested models, in agreement with average continental upper-mantle velocities (Kennett & Engdahl 1991). In order to refine the grid search around the crustal parameters we simultaneously tested variations of the crustal thickness, Vp and Vp/Vs using sensitivity analysis to identify the appropriate crustal parameters producing the best matches between synthetic and observed seismograms. For every single path, we tested more than 5 crustal models consisting of a layer over half-space with fixed Downloaded from by guest on 2 January 219 C 27 The Authors, GJI, 17, Journal compilation C 27 RAS

7 864 P. Alvarado, S. Beck and G. Zandt a) displacement (x 1 m) km/s 5.8 km/s 6. km/s 6.2 km/s 6.4 km/s 6.6 km/s 5 1 Crustal Vp a) displacement (x 1 m) Crustal Vp/Vs b) displacement (x 1 m) time (s) Figure 4. Sensitivity of the broadband waveforms to average crustal P-wave velocity (Vp) calculated at the same distance and azimuth as in Fig. 3. For simplicity we show the radial component in this and the following waveform sensitivity figures. We assumed a 5-km thick crustal layer of Vp/Vs of 1.8. Mantle-half-space parameters fixed as in Fig. 3. (a) Broadband radialcomponent seismograms. (b) Synthetic waveforms using a 1 8 s bandpass filter. (mantle) half-space parameters. Our method investigated possible crustal thicknesses of 2 75 km with a grid spacing of 5 km, crustal Vp velocities of km s 1 with a grid spacing of.2 km s 1, and crustal Vs velocities that produce Vp/Vs of with a grid spacing of.5. In order to avoid circular reasoning, we varied the crustal thickness and Vp within the specified ranges using a fixed Vp/Vs. We then repeated the same process for a different Vp/Vs, and so on, until completing the test for all Vp/Vs values. For each combination of crustal parameters, we calculated synthetic broadband full waveforms at a given regional epicentral distance and azimuth using the reflectivity code of Randall (1994). To avoid errors in timing and mislocation, we chose the time window for waveform modelling aligned at the P-wave first arrival and of record length successful in including all phases with good signal-to-noise levels. A shorter time window is usually needed to analyse the full seismogram for seismic stations closer to the source. In contrast a larger time window is needed when the seismic station is further away from the event. The observed and synthetic data were then filtered with a Butterworth bandpass filter from intermediate to long periods. In general, we applied a bandpass filter of 1 8 s for earthquakes with M w > 5, 1 5 s for 4 < M w < 5 and 1 35 s for M w 4. We did not test smaller period wavelengths because they become more sensitive to small-scale lateral heterogeneities, which are not resolvable by our method, and are often contaminated with noise. After filtering, we compared every (vertical, radial and tangential) synthetic seismogram with the corresponding observed seismic com- b) displacement (x 1 m) time (s) Figure 5. Sensitivity of the broadband waveforms to crustal S-wave velocity (Vs) calculated at the same distance and azimuth as in Fig. 3. We used a 5 km thick crust of Vp = 6.4 km s 1. Mantle parameters fixed as in Fig. 3. (a) Broadband radial-component seismograms. (b) Synthetic waveforms using a1 8 s bandpass filter. ponent by estimating their cross-correlation coefficient and taking an average over the three components. The correlation coefficient varies between 1 for inversely correlated data to 1 for perfectly correlated data. One single map of averaged correlation coefficients for crustal thickness as a function of Vp was constructed for each (fixed) Vp/Vs tested. We determined the maximum in each map and carefully examined its corresponding seismogram matches. We discuss in detail the results for the pairs 2-5/BARD, 1-352/PICH, 1-285/NEGR and 1-285/LENA to illustrate the grid search analysis. Fig. 8 shows an example of the grid-search results around crustal thickness and Vp for event 2-5 and station BARD (see the corresponding seismic path in Fig. 2). Each point in the map is the correlation coefficient between synthetic and observed data, averaged over the three-component waveforms, which were bandpass filtered between 1 and 5 s and compared over 12 s of record length. In this example, we fixed Vp/Vs at 1.8 in the crust and mantle parameters remained fixed as before. The maximum correlation (.93) over this grid is considered to be the best one layer model, with a crustal Vp of 6.2 km s 1 and thickness of 35 km (red dot in Fig. 8). Table 1 compiles these data, together with the results for a range of correlation coefficients corresponding to waveform matches that differ by less than 5 per cent of the maximum correlation coefficient obtained for each component (Fig. 8b). The inspection of the waveform matches around the maximum averaged correlation coefficient enables us to discriminate high quality results (shown in Table 1) from estimations of less correlated data (compare waveform matches in Figs 8b d). The acceptable waveform matches Downloaded from by guest on 2 January 219 C 27 The Authors, GJI, 17, Journal compilation C 27 RAS

8 Crustal structure of the Andes Cordillera and backarc region 865 a) displacement (x 1 m) b) displacement (x 1 m) km/s 7.8 km/s 8. km/s 8.2 km/s 8.4 km/s time (s) Mantle Vp Figure 6. Sensitivity of the broadband waveforms to mantle Vp calculated at the same distance and azimuth as in Fig. 3. We fixed crustal parameters using a layer of 5 km thickness, Vp = 6.4 km s 1 and Vp/Vs = 1.8 over a mantle-half-space of constant Vp/Vs = 1.8. (a) Broadband radialcomponent seismograms. (b) Synthetic waveforms using a 1 8 s bandpass filter. outline the broadness of the best crustal model solution and gives some estimate of uncertainty (Table 1). In a few cases the map of correlation coefficients has a tendency to define a secondary maximum peak in the contouring of averaged correlation coefficients for the same event-station path, which we are also reporting in Table 1. For example, Fig. 8a shows the occurrence of a secondary peak with average correlation coefficient of.77 around thicknesses of km and Vp of approximately km s 1. Although this coefficient may not be related to acceptable fits we still include it in Table 1 to show the behaviour around the maximum average correlation coefficient. This enables us to better identify regions of single well-defined maximum peaks. Another example is shown in Fig. 9 for event and station PICH located at an epicentral distance of 28 km (Fig. 2). A set of forward synthetic three-component waveforms is calculated at a series of crustal Vp/Vs, varying crustal Vp and crustal thickness. Mantle parameters remained fixed as before. We use a bandpass filter between 1 and 5 s and a time window of 72 s after the first P-wave arrival to estimate the cross-correlation between synthetic and observed waveforms. Shown in Fig. 9a is the map of correlation coefficients between forward modelled synthetic waveforms, using a fixed crustal Vp/Vs of 1.7, and observed waveforms, averaged among the three components. A similar map was created for each of the six different crustal Vp/Vs values explored for this seismic path. A Vp/Vs ratio of 1.7 produces the overall maximum correlation coefficient in the entire grid-search process. The corresponding best layer over half-space model occurs for a crustal thickness of 45 km a) displacement (x 1 m) b) displacement (x 1 m) time (s) Mantle Vp/Vs Figure 7. Sensitivity of the broadband waveforms to mantle Vp/Vs calculated at the same distance and azimuth as in Fig. 3. We used the same crustal layer parameters as in Fig. 6. Mantle-half-space Vp was fixed at 8.15 km s 1. (a) Broadband radial-component seismograms. (b) Synthetic waveforms usinga1 8 s bandpass filter. and a crustal Vp of 5.8 km s 1, with an average correlation coefficient of.94 (red dot in Fig. 9). For comparison, forward-modelled waveforms corresponding to relatively smaller maximum average correlation coefficients, determined for the six different crustal Vp/Vs explored in the grid-search, are also shown (Fig. 9b). Although all models (last column in Fig. 9b) define a maximum correlation between synthetic and observed seismograms around thicknesses of 3 45 km, the best overall matches indicate a thickness of 45 km, a low crustal Vp of 5.8 km s 1 andalowvp/vs of 1.7 (Fig. 9 and Table 1). The last example (Fig. 1) shows the grid search results for two different regions using the same event (1-285) but different stations (NEGR and LENA) (Fig. 2). The three-component synthetic waveforms were calculated for the pairs 1-285/NEGR and 1-285/LENA (Fig. 2) separately, for combinations of crustal parameters (thickness, Vp and Vp/Vs) varying within the range of all reasonable estimates and maintaining mantle parameters fixed as before. For each thickness-vp-vp/vs model, the synthetic waveforms were compared to observed data using a bandpass filter of 1 8 s and a time record window of 14 s for the pair 1-285/NEGR. The same grid search was carried out for the pair 1-285/LENA using a time record window of 5 s (Figs 1c and d). We calculated the correlation coefficient for the respective displacement components and took the average among the three components. The best singlelayer model for the northern seismic path using station NEGR has an average correlation coefficient of.93, with a crustal thickness of 6 km, a crustal Vp of 6.6 km/s and a Vp/Vs of 1.85 (Figs 1a and b, Downloaded from by guest on 2 January 219 C 27 The Authors, GJI, 17, Journal compilation C 27 RAS

9 866 P. Alvarado, S. Beck and G. Zandt a) /BARD a) /PICH crustal Vp (km/s) vertical radial tangential crustal thickness (km) b) c) d) observed synthetic s Figure 8. (a) Example of a map of the average correlation coefficient over the three-component waveforms between synthetic and observed data for the event/station pair 2-5/BARD (see Fig. 2 for location). Sixty combinations of crustal thickness and Vp were tested with forward waveform modelling for seven different crustal Vp/Vs ratios. In addition mantle parameters Vp = 8.15 km s 1 and Vp/Vs = 1.8 remained fixed during the entire search. In this case, a specific grid search using a fixed crustal Vp/Vs at 1.8 is shown. A bandpass filter between 1 and 5 s was used. The red dot indicates the maximum correlation for the best layer-model consisting of a thickness of 35 km, a Vp of 6.2 km s 1 and a Vp/Vs of 1.8. (b) Synthetic (red line) and observed (black line) seismic displacements for the best layer-model in figure (a) corresponding to a maximum average correlation coefficient of.93. The number near the waveforms indicates the correlation percentage for each seismic component. (c) Waveform matches for a layer-model with thickness of 55 km, Vp of 6.2 km s 1 and Vp/Vs of 1.8, corresponding to an average correlation coefficient of.75 (a star). (d) Waveform matches for a layer-model with thickness of 45 km, Vp of 6.4 km s 1 and Vp/Vs of 1.8 corresponding to an average correlation coefficient of.44 (a square). Table 1). The best results for the pair 1-285/LENA occur for an average correlation coefficient of.91 with a crustal thickness of 45 km, a Vp of 5.8 km s 1 and a Vp/Vs of 1.8 (Figs 1c and d, Table 1). Other possible combinations of crustal parameters around crustal Vp (km/s) crustal thickness (km) b) vertical radial tangential (Corr. ) Model (74) 3_6.6_ s observed (82) 35_6.6_1.85 (85) 4_6.4_1.8 (83) 45_6.2_1.75 (87) 4_5.8_1.65 (94) 45_5.8_1.7 synthetic Figure 9. (a) Example map of the maximum correlation coefficient over the three component waveforms for event (M w 4.5) and station PICH (Fig. 2). We used a bandpass filter between 1 and 5 s. The best layer over half-space model (red dot) occurs for a Vp of 5.8 km s 1, a thickness of 45 km and Vp/Vs of 1.7, with a maximum correlation coefficient of.94. (b) Synthetic three-component forward-modelled waveforms for the maximum correlation results of each grid-search analysis around crustal thickness (in km) and crustal Vp (in km s 1 ) using different crustal Vp/Vs. The average correlation coefficients for each best layer over half-space model resulting from the respective grid search are indicated to the right. Numbers above and on the left of each trace indicate the percentage of correlation between the corresponding forward synthetic seismogram (red) and the observed data (black). Note that the best overall results occur for a model with crustal thickness of 45 km, Vp of 5.8 km s 1 and crustal Vp/Vs of 1.7. the best ones are also acceptable for both examples along the Cordillera, as reported in Table 1. Using the same analysis as used for the single station-event pairs 2-5/BARD, 1-352/PICH, 1-285/NEGR and 1-285/LENA we constrained the best layer-overhalf-space model for a total of 32 different paths (Figs 2, 11a c). In some cases one of the observed seismic components was noisy but we were able to use the remaining components for the grid search study (see pairs with a star in Table 1). Downloaded from by guest on 2 January 219 C 27 The Authors, GJI, 17, Journal compilation C 27 RAS

10 Crustal structure of the Andes Cordillera and backarc region 867 a) /NEGR c) /LENA crustal Vp (km/s) crustal Vp (km/s) crustal thickness (km) crustal thickness (km) b) vertical radial tangential Model d) vertical radial tangential Model Av. Corr..93 thickness 6 km Vp 6.6 km/s Av. Corr..91 thickness 45 km Vp 5.8 km/s 14 s Vp/Vs 1.85 Vp/Vs s observed synthetic observed synthetic Figure 1. Comparison of the grid search results along the high Cordillera. (a) Map of the maximum correlation coefficient over the three component waveforms for event (M w 5.1) and station NEGR (Fig. 2). We used a bandpass filter between 1 and 8 s. The best layer over half-space model (red dot) occurs for a Vp of 6.6 km s 1, a thickness of 6 km and a Vp/Vs of 1.85, with a maximum correlation coefficient of.93. (b) Synthetic (red) three-component forwardmodelled waveforms for the best combination of crustal parameters [red dot in (a)] compared to observed seismic displacements (black). The percentage of correlation for each component between synthetic and observed waveforms is shown above and to the left of each trace. (c) Map of the maximum correlation coefficient over the three component waveforms for the same event using station LENA (Fig. 2). We used a bandpass filter between 15 and 5 s. The best layer over half-space model (blue triangle) occurs for a Vp of 5.8 km s 1, a thickness of 45 km and a Vp/Vs of 1.8, with a maximum correlation coefficient of.91. (d) Synthetic (cyan) three-component forward-modelled waveforms for the best combination of crustal parameters [blue triangle in (c)] compared to observed seismic displacements (black). The percentage of correlation for each component between synthetic and observed waveforms is shown above and to the left of each trace. 6 RESULTS AND DISCUSSION Our results for Vp and Vp/Vs in the crust are summarized in Figs 11a and b. Both maps show an interpolation of the results listed in Table 1. We assigned two equally spaced data-points along each studied seismic path (Fig. 2). For example, an estimation along a seismic path of 3 km epicentral distance is represented by two intermediate Vp (or Vp/Vs) data points with a spacing of 1 km. Our estimates and previous studies (e.g. Gilbert et al. 26; McGlashan et al. 26), show that the listed thicknesses in Table 1 do not necessarily imply Moho depths. This is the case for seismic paths in the WSP and the Precordillera as we discuss in Section 6.3. In other regions, if more than one thickness is listed in Table 1 for a particular earthquake-station path (e.g /JUAN), we consider the Vp and Vp/Vs estimations for the best layer model associated with the overall maximum averaged correlation coefficient, which is listed first. The map in Fig. 11c shows our crustal thickness results using data points as in Fig. 11a. In addition, this map includes previous regional crustal thickness results compiled by Gilbert et al. (26) and described in Section 2, as well as recent local determinations beneath station JUAN by Calkins et al. (26) and teleseismic observations of pmp phases between 3 S and 33 S beneath the Precordillera and Cordillera by McGlashan et al. (26). We discuss the results shown in Fig. 11 for crustal Vp, Vp/Vs and thickness, which are later compared to crustal seismicity in the region (Fig. 12). These results are mainly based on limited observations along few seismic paths (Fig. 2 and Table 1). Thus the crustal models in Fig. 11 are averaged over a large area. Many more single earthquake-receiver paths are required to accurately map the region and investigate the structure in more detail. A common feature observed in the region is the expected larger thickness beneath the high Cordillera and its decrease in the eastern terranes that support lower elevations (Fig. 11c and inset in Fig. 2). To the north of 33 S we found a greater thickness beneath the high Cordillera than to the south (Figs 1 and 11c). This thick crust is also observed along paths in the Precordillera fold-thrust belt (Fig. 2 and Table 1). This is consistent with an overthickened crust due to tectonic shortening, where the slab subducts horizontally as suggested by Allmendinger et al. (199). Another feature observed from the numerous single event-station paths is the different character of the best layer over half-space models between different regions. The eastern terranes and high Cordillera show one single crustal layer, perhaps in isostatic compensation with their elevations (inset in Fig. 2 and Fig. 11c). If velocity discontinuities exist within the crust of these regions, they are relatively minor in comparison with the crust-mantle boundary for our surface wave modelling. The only results available for a comparison in both regions are from receiver function analysis, which show a very well defined Moho signal for the ESP (beneath stations LLAN and PICH, Fig. 2) (Gilbert et al. 26). In contrast, the WSP and Precordillera terranes exhibit a more complex structure that makes it difficult to identify one single crustal layer model (e.g /PACH). We have compared our results in Fig. 11 with empirical data from a compilation of a wide variety of lithologies in active continental Downloaded from by guest on 2 January 219 C 27 The Authors, GJI, 17, Journal compilation C 27 RAS

11 868 P. Alvarado, S. Beck and G. Zandt Figure 11. (a) Map of P-wave velocity results from the best-correlated waveforms in the grid search analysis (Table 1). Vp-contours represent interpolated Vp-data obtained in this study assigning two equally spaced data-points along each studied seismic path (Fig. 2). By comparing with Fig. 2, note the sparse data used to obtain this and the following crustal models (b) and (c), especially in the ESP. Slab contours are from Anderson et al. (27). All other symbols as in Fig. 1. (b) Same as (a) for Vp/Vs ratio. This contour map also indicates variations of Vs determined from the best layer model along each studied path in Fig. 2. (c) Contour map for crustal thickness results from the best-correlated waveforms in the grid search analysis (Table 1), using the same criteria to assign thickness-data points as in (a) and (b), and other estimations by Regnier et al. (1994), Fromm et al. (24), Gilbert et al. (26), Calkins et al. (26) and McGlashan et al. (26). All other symbols as in Fig. 1. (d) GPS velocity measurements and their corresponding 95 per cent confidence ellipses relative to stations in the eastern part of South America according to Brooks et al. (23). The white bar represents the Nazca-South American plate convergence at a rate of 63 mm yr 1 (Kendrick et al. 23). Terranes boundaries as in Fig. 1. margins by Brocher (25) (Fig. 13). This compilation of Vp and Vp/Vs, based on independent estimations of Vp, Vs and density corresponds to specific lithologies from laboratory and field measurements, borehole, seismic vertical profiles and seismic tomography. Our constraints are shown by rectangles in Fig. 13. We note that Vs determinations are more important than Vp to place our constraints within mafic/felsic limits and, some of the expected variation of the elastic Lamé s constants with depth within the crust. We discuss each region below. 6.1 High Cordillera and active arc North of 33 S, the best observed-synthetic waveform matches were achieved using a crustal layer of 5 6 km thickness with average Vp of km s 1 and Vp/Vs of 1.85 (Figs 1a and b, 11a c). These results are consistent with an intermediate to mafic crustal composition (Christensen & Mooney 1995; Brocher 25; Fig. 13). To the south, results along the active volcanic arc indicate also a high Vp/Vs ratio (>1.8), with a much lower Vp ( km s 1 ) (Figs 1c and d, 11a c). Our grid search results suggest a crustal layer model of 45 km along the active arc consistent with receiver functions determinations by Gilbert et al. (26) and pmp estimations from McGlashan et al. (26). The magmatic fluid source above the normal subduction segment south of 33 S could decrease the P- and S-wave velocities. In contrast, the results in the currently amagmatic northern segment are less straightforward and require more discussion. Comparing this region with its adjacent southern segment along the active Downloaded from by guest on 2 January 219 C 27 The Authors, GJI, 17, Journal compilation C 27 RAS

12 Crustal structure of the Andes Cordillera and backarc region Seismic energy (x 1 J) 1.9/4.5 JFR 8 6 Chilenia km 5 1 Valle Fertil fault Cuyania Pampia Fig. 15 Rio de La Plata? Magnitude >4.5 Depth (km) < / / /5.1./.8 5.3/ /1. ( ) ( ) 23.6/8.1 Figure 12. Moderate crustal seismicity of depths <5 km and magnitude >3.5 from 1986 to 26, reported by the local networks in Argentina (INPRES) and Chile (SSN), as indicated in the NEIC (USGS) and ISC catalogs. White stars are large (M 7.) historical crustal earthquakes from INPRES (27). The histogram shows the seismic energy (in 1 6 J) released in the upper 7 km of the high Cordillera and backarc region from (dark green) and (light green), from Gutscher et al. (2a). Terranes boundaries from Ramos et al. (22) and slab contours from Anderson et al. (27), labelled in km. Brackets show the approximate location of the cross-section in Fig. 15. Vp/Vs (4) (1) (4) (3) (4) A E (2) (5) W H (4) (5) line, and Calcium-rich Eqn. rocks 6, Regression fit (6) Brocher (25) empirical Vp (km/s) Vp/Vs Figure 13. Simplified diagram based on independent observations of Vp, Vs and density for common lithologies from Brocher (25). The mafic (thick black) line delineates calcium-rich rocks, gabbros, and mafic rocks with Vs.2.3 km s 1 below the general trend of Brocher s empirical fit (thin black line). Regions of differing Lamé s elastic constants determined from independent observations of Vp, Vs and density are also shown by a dashed blue line according to Brocher (25). Ellipses encircle data provided by single references in Brocher (25). Numbers in parentheses link specific lithologies: (1): serpentinite; (2): sedimentary rocks; (3): granite; (4): metagraywackes and mafic rocks; (5): metamorphic rocks and (6): mafic rocks. Rectangles represent each region in this study: E: Pampia (Eastern Sierras Pampeanas); W: Cuyania (Western Sierras Pampeanas and Precordillera); A: Active arc and H: High Cordillera north of 33 S. See Brocher (25) for complete references. arc, the high Vp/Vs ratio could be in part explained by the greater Vp (> 6.4 km s 1 ) values but also by the low Vs. Magmatism in this northern part of the high Cordillera has been interpreted to cease at 6 5 Ma as the subducted slab flattened to the east (Kay et al. 1991). Significant periods of andesitic and dacitic volcanism have occurred between 27 and 8 Ma (Kay et al. 1999; Bissig et al. 23), and silicic volcanism at 5 Ma (Ramos et al. 1989) with mineralization intruding granitiod basement rocks. Moreover, studies of rhyolitic rocks showing negative Europium geochemical anomalies in the northern segment have revealed that localized crustal melting continued until 2 Ma (Bissig et al. 22). Thus, recent partial melting is associated with the economic mineralization shown in Figs 1 and 2 in the northern segment of the high Cordillera. Different tectonicmagmatic mechanisms have been proposed to explain these melts, such as flat-slab derived melting (Gutscher et al. 2b), secondary melting within the crust (Kay & Mpodozis 22), or high-pressure melting and assimilation in the lower crust (Bissig et al. 23). The remarkable contrast in Vp and thickness between the northern and active arc segments is consistent with a much cooler northern region, although some possible localized recent melt in the lower crust cannot be ruled out, as shown by the low S-wave regional velocities. Although different average Vp estimations characterize the northern high Cordillera and the active arc, it is likely that they both have similar mafic compositions (Fig. 13). According to Brocher (25), the mafic line in Fig. 13 represents mafic and calcium-rich rocks, which have a Vs.2.3 km s 1 below general empirical Vp versus Vp/Vs trends. These results taken together indicate a relatively larger percentage of partial melt along the active arc. As laboratory measurements show, the effect of partial melt decreases Vp, and still more dramatically, Vs, increasing the Vp/Vs ratio, independently of the melt composition or confining pressure (Makovsky & Klemperer 1999). The region immediately east of the active arc shows an increase in both Vp and Vs which might be related to a region of relatively cooler backarc volcanism. 6.2 Eastern Sierras Pampeanas The average crustal model in the Pampia terrane basement indicates low values of Vp of km s 1, a layer of km thickness and low Vp/Vs of (Figs 9 and 11a c). Both the Pampia Downloaded from by guest on 2 January 219 C 27 The Authors, GJI, 17, Journal compilation C 27 RAS

13 87 P. Alvarado, S. Beck and G. Zandt terrane and Río de la Plata craton have a more felsic quartz-rich composition (Rapela et al. 1998) compared to the western basement terranes. Our results are consistent with some granite-gneiss crystalline lithologies (rectangle E in Fig. 13) in good agreement with the exposed rocks in this region (Rapela et al. 1998). These results show a contrast with the properties of the western terranes as we discuss in next section. Recent volcanism in this region at 2 Ma has been reported as a consequence of its eastward migration related to the flattening of the subducted plate (e.g. Ramos et al. 22). Although averaged crustal seismic properties, our results do not show evidence for extensive currently active volcanism. However, localized partial melts cannot be ruled out. 6.3 Precordillera and Western Sierras Pampeanas Our results for the Precordillera and WSP region (Cuyania terrane) using a simple one crustal layer indicate high Vp (>6.4 km s 1 ), high Vp/Vs ( ) and thicknesses of 3 6 km (Figs 2 and 11a c). Comparing the Cuyania terrane with the Cordillera region above the flat slab segment, we observe a slightly lower Vp and similar Vp/Vs along the Cuyania terrane (Figs 11a and b). Our determinations for the Cuyania terrane lie in the region of metagraywackes and mafic rocks, yet are close to the mafic-line lithologies (rectangle W in Fig. 13). We note that good fits are achieved using a very simple seismic velocity structure with a crustal thickness of 35 4 km and a half-space (with mantle parameters) immediately below (e.g. results for 2-117/USPA or 2-17/JUAN pairs). However, there are many ray paths that also predict a thicker crust (e.g /USPA; 1-352/JUAN). In addition, receiver function, Pn and multiple broad-band waveform inversions that also used the CHARGE data (Fromm et al. 24; Alvarado et al. 25; Calkins et al. 26; Gilbert et al. 26), local earthquake converted phases at the Moho (Regnier et al. 1994) and teleseismic observations of pmp phases beneath the Precordillera (McGlashan et al. 26), provide significant evidence for a Moho depth of 5 55 km in this region (Fig. 11c). For this reason, a densification in the lower crust has been suggested by Snyder et al. (199), Gilbert et al. (26) and Calkins et al. (26), which would also reconcile seismic observations and gravimetric data (e.g. Smalley & Introcaso 23). Therefore, this region is likely characterized by crustal models consisting of more than one single layer. In light of previous CHARGE studies suggesting a gradational increase of Vp and Vs between 4 and 55 km depth in the lower crust for the Cuyania terrane (Calkins et al. 26), we have tested for the presence of a higher velocity layer in the lower crust. We used 11 different paths (Fig. 14a) and a crustal model consisting of two layers of 4 ± 5 and 15 ± 5 km thickness, respectively, corresponding to a total thickness between 55 and 6 km (Fig. 14b). We tested a full combination of P-wave velocities from 5.7 to 6.6 km s 1 in the upper crust and 6.6 to 7.6 km s 1 in the lower crust, varying crustal Vp/Vs from 1.65 to 1.85 and fixing the density at 2.85 g cm 3 in the crust, 3.33 g cm 3 in the mantle, and mantle velocities as before. Our analyses suggest that these waveforms are sensitive to details in the crustal structure. A velocity of km s 1 increasing to km s 1 at a depth of 35 4 km with a total thickness of 55 km and crustal Vp/Vs of 1.8 produce the best fits (Figs 14b and c). In some cases we obtained better fits modifying slightly Vp/Vs in the lower crust for the pairs 1-127/PACH (1.85) and 1-352/USPA (1.7). We have observed that waveform modelling is much less sensitive to variations in density. Using an increased density of 3. g cm 3 in the lower crust produces similar results to those shown in Fig. 14c. These results provide the first evidence from regional modelling for a thick crust with mafic material or partial eclogite in the lower crust, which has seismic parameters intermediate between the crust and upper mantle. This confirms previous results in the region from receiver functions (e.g. Gilbert et al. 26). We note that the receiver function study is only sensitive to discontinuities and gradients in the velocity model but not to absolute velocities. We did not include any upper-crust low velocity layers to account for the basins surrounding the WSP ranges. Thus our estimates represent minimum high velocity and Vp/Vs values. Modelling higher frequencies, which are more sensitive to crustal parameters, might be useful to explore details like thin uppermost crustal layers. However, for the magnitudes of the earthquakes and epicentral distances used in this study, we were not able to use higher frequency data due to the high noise content. 7 COMPARISON OF CRUSTAL STRUCTURE WITH REGIONAL CRUSTAL SEISMICITY We observe a correlation between a higher rate of seismicity and moment release in the Cuyania terrane, and its relatively high Vp, high Vp/Vs ratio, thick crust and evidence for a distinct two layer crust with a possible partially ecologitized lower crust. Crustal earthquakes are also frequent along the active arc south of 33 S. In this section we explore the relationship between crustal structure and seismicity including in our discussion quantifiable factors such as plate convergence rate, heat flow, down-going plate geometry, composition, and increased strain (as shown by the GPS measurements) for comparison across terranes. 7.1 Observations Continental seismicity over the last 2 yr in the region of study is not biased by any large (magnitude > 6.) earthquake. The region has good station coverage due to the Chile and Argentina seismic networks, so it is unlikely that earthquakes of magnitude above 4.5 go undetected. This seismicity with maximum location uncertainty of 2 km shows that the Andean crustal earthquake distribution is highly dense in the backarc region near San Juan and Mendoza (Figs 1, 12 and 15). The region above the flat slab generates more earthquakes and of a higher magnitude compared to adjacent segments, where the slab starts dipping more steeply (Gutscher et al. 2a). Historically, three large damaging crustal earthquakes of magnitudes M w 7., 6.8 and 7.5 have occurred in the Precordillera and the WSP in 1944, 1952 and 1977, respectively (Kadinsky-Cade 1985; Langer & Hartzell 1996; Alvarado & Beck 26) (Figs 12 and 15). Local and regional studies in this region show a predominance of thrust focal mechanisms with focal depths between 1 and 35 km for moderate seismicity (Regnier et al. 1992; Smalley et al. 1993; Alvarado et al. 25; Salazar 25). This high seismicity contrasts with that of the region located immediately to the east of the active Sierra Valle Fértil Fault, interpreted as the suture between the Cuyania and Pampia terranes (Ramos 1994; Figs 1 and 12). The occurrence of crustal earthquakes in the Pampia terrane with magnitudes over 4.5 are rare, although historical records indicate one M 6. event in the south, near San Luis (Fig. 1) (Costa & Vita-Finzi 1996; INPRES 27). Seismicity in the Río de la Plata craton is also lower than the crustal Downloaded from by guest on 2 January 219 C 27 The Authors, GJI, 17, Journal compilation C 27 RAS

14 Crustal structure of the Andes Cordillera and backarc region a) 9o 8o -3 PACH HEDI o -32 USPA o o 9 8 o 7 o velocitiy (km/s) depth (km) b) / PACH / USPA 92 Model_1 Model_2 Model_ / HEDI 91 s_(89) Model_1 86 s_(78) Model_ s_(87) 75 Model_ s_(91) Model_ s_(84) Model_1 2-5/ JUAN / USPA / JUAN / USPA / HEDI s_(88) Model_ s_(94) Model_ s_(9) Model_ s_(97) Model_ s_(91) Model_3 observed synthetic Figure 14. Testing for a velocity increase in the lower crust of the Cuyania terrane. (a) Single earthquake-receiver paths used in this analysis. (b) Best crustal models for the maximum average correlation results between synthetic and observed waveforms shown in (c) using a density of 2.85 g cm 3 in the crust and 3.33 g cm 3 in the mantle. We used a Vp/Vs ratio of 1.8 except for the lower crust layer associated with the pairs 1-127/PACH (1.85) and 1-352/USPA (1.7). (c) Forward synthetic waveforms calculated for the event/station pairs in (a) using the crustal models in (b) compared with observed seismograms. We used a bandpass filter between 15 and 5 s. Numbers above and to the left of each seismic component are the percentage of cross-correlation between synthetic and observed data. The time record length after the first arrival and averaged correlation coefficient (in parentheses) over the three seismic components using each crustal model are indicated in the last column. Principal and Frontal Precordillera Cordilleras 72oW Sierras Pampeanas Chilenia Probable interface in the upper crust 5 1 high Vp; Cuyania high Vp/Vs Nazca slab with the JF R low Vp; low Vp/Vs Pampia Moho Partially eclogitized lower crust Río de la Plata km 5 Probably cold, dry, Mg-rich depleted lithosphere 15 2 km Figure 15. Schematic lithospheric scale cross-section across the flat-slab region (see location in Fig. 12) showing main crustal structure results in this study, different terranes (Ramos et al. 22), seismogenic crustal zones, Moho depths (Fromm et al. 24; Alvarado et al. 25; Gilbert et al. 26), probable interface at 2 km depth (Gilbert et al. 26; Calkins et al. 26) and dry-cold upper mantle (Wagner et al. 25) relative to the top of the subducted Nazca plate possibly associated with the buoyant JFR (Anderson et al. 27). Focal mechanisms with dark compressional quadrants in vertical projection from synthetic modelling studies for historical large earthquakes (Langer & Hartzell 1996; Alvarado & Beck 26) and moderate (3.5 < M w < 5.1) CHARGE events (Alvarado et al. 25). Figs 12 and 15 show that major changes in crustal structure correlate with different rates of seismicity and seismic moment release. C 27 The Authors, GJI, 17, C 27 RAS Journal compilation Downloaded from by guest on 2 January o? tangential / JUAN JUAN o radial / USPA -356 o vertical c) 7o 871