1 Flat vs. Normal Subduction Zones: A Comparison Based on 3D 2 Regional Traveltime Tomography & Petrological Modeling of 3 Central Chile & Western Argentina (29°-35°S) 4 5 M. Marot(1), T. Monfret(1), M. Gerbault(1), G. Nolet(1), G. Ranalli(2), M. Pardo(3)

6 (1) Géoazur, UNSA, IRD, CNRS, OCA, (UMR 7329), 250 Albert Einstein, 06560 Valbonne, France 7 (2) Department of Earth Sciences, Carleton University, 1125 Colonel By Drive, Ottawa, ON, K1S 5B, Canada 8 (3) Departamento de Geofísica, Universidad de Chile, Blanco Encalada 2002, Santiago, Chile 9

10 Abstract

11 Beneath central Chile and western Argentina, the subducting oceanic Nazca lithosphere 12 drastically changes geometry from dipping at an angle of 27-35° to horizontal, along the inferred 13 subduction path of the Juan Fernandez seamount Ridge (JFR). The aim of our study is to assess the 14 differences in the seismic properties of the overriding lithosphere in these two regions, in order to 15 better understand its deep structure and the links between its surface deformations and the geometry of 16 the slab. In comparison with previous studies, we show the most complete 3D regional seismic 17 tomography images for this region, whereby we use (1) a larger seismic dataset compiled from several 18 short-term seismic catalogs, (2) a denser seismic array enabling us to better resolve the subduction 19 zone from the trench to the backarc and into the upper ~ 30 km of the slab, and (3) a starting 1D 20 background velocity model specifically calculated for this region and refined over the years. We assess 21 and discuss our seismic tomography results with: (i) supporting novel regional seismic attenuation 22 models, and (ii) predicted rock types calculated using the Hacker and Abers (2004) mineral/rock 23 database, and based on a computed thermo-mechanical model of the pressure and temperature 24 conditions at depth. Our results show significant seismic differences between the flat and normal 25 subduction zones: As expected, the lower geotherm of the flat slab region is reflected by faster seismic 26 velocities and increased seismic activity within the slab and overriding lithosphere. We find evidence 27 that the JFR slab dehydrates at the eastern tip of the flat slab segment, indicating that the flat slab 28 segment retains some fluids. The forearc region above the flat slab area is subject to high Vs and very 29 low Vp/Vs ratios (1.69) at two different depth intervals, uncorrelated with typical rock compositions, 30 increased density or reduced temperature; however, possibly linked with the aftershock effects of the 31 Mw 7.1 1997 Punitaqui earthquake, the slab geometry at depth, and/or seismic anisotropy. At the 32 surface, the Andean crust is strongly reduced in seismic velocities along the La Ramada-Aconcagua 33 deformation belt, suggesting a structural damage impact, and the seismic variations correlate with the 34 geological terranes. The Andean lower crust is also impacted by strongly reduced seismic velocities 35 matching hydrous rocks of either mafic and non-eclogitized composition, indicating a possible past 36 delamination event, or a felsic composition, in support for the existence of the Chilenia terrane at these 37 depths. We confirm previous studies that suggest the Cuyania terrane, in the backarc region, is mafic 38 and contains an eclogitized lower crust below 50 km depth. We also suspect major Andean basement 39 detachment faults (or shear zones) to extend towards the plate interface and channelize slab-derived 40 fluids into the continental crust. 41 42 43 Keywords: central Chile; flat subduction; seismic tomography; rock composition; thermo-mechanical 44 modeling; eclogite crust; dehydration/hydration

1

45 1. Introduction

46 Along the South American-Nazca convergence zone, spatial correlations exist between 47 subducting oceanic features (ridges, plateaus, fracture zones), volcanic arc gaps, backarc basement 48 uplift, and past or present occurrences of flat subduction (Skinner and Clayton, 2013). The two most 49 famous present-day flat slabs on Earth are the Peruvian and central Chilean flat slabs, both impacting 50 the Nazca Plate along the South American margin. Nevertheless, the cause for their formation is still 51 subject of debate. 52 The central Chilean “Pampean” flat slab (~ 29°-32.5°S) is perhaps the best documented so far, 53 since the region is advantaged by a high micro-seismic activity, good geological exposure levels, 54 advanced structural evolution, and has been impacting the region for a relatively long time (15-18 Ma; 55 Kay and Mpodozis, 2002; Ramos et al. 2002). It is associated with the subduction of relatively young 56 (35-40 Ma) and hydrated lithosphere (Kopp et al., 2004) and of the Juan Fernandez hotspot seamount 57 Ridge (JFR) beneath the thick continental South American lithosphere. Both the region’s high 58 seismicity and the slab’s rapid geometrical transition (from plunging 30° to 0°), render this location 59 ideal to study the seismic properties and anomalies associated to this flat subduction. 60 In normal subduction circumstances, conventional slab dehydration processes take place due to 61 high geothermal gradients at depth, resulting in arc volcanism, seismicity, and weakening of the 62 continental crust. However, in the case of flat subduction, the asthenosphere is expelled during slab 63 flattening, significantly cooling the subduction system (Kay and Mpodozis, 2002; Ramos et al., 2002; 64 Grevemeyer et al., 2003). In the case of the central Chilean flat slab, it is associated with increases in: 65 (i) plate coupling forces, (ii) compressional stresses farther inland, (iii) topography, (iv) seismicity in 66 the backarc, forearc, plate interface and slab (including a Double Seismic Zone, Marot et al. 2013), 67 and (v) cessation of arc volcanism (Kay and Abbruzzi, 1996; Kay and Mpodozis, 2002; Ramos et al., 68 2002). 69 While slab properties are commonly considered the dominant parameters influencing changes in 70 slab geometry, there is a growing consensus that the upper plate properties also play a non-negligible 71 role in provoking flat subductions (e.g. van Hunen, 2001; 2004; Gerbault et al., 2009; Manea et al., 72 2012). 73 Whereas the processes and timing of slab shallowing are well constrained from the collection of 74 evidences of backarc basement uplifts and eastward volcanic arc migration (Ramos et al., 2002; 75 Alvarado et al., 2007), the factors that triggered flat subduction and maintained it for at least 6 Myr 76 (Kay et al., 1991; Kay and Abbruzzi, 1996), as well as its relationship with the deep composition and 77 deformation of the overriding lithosphere, are ill-constrained and multiple paradoxes exist, such as the 78 obvious influence of the JFR, however, deemed too small to have created by itself the flat slab’s 79 imposing dimension (~ 200-250 km long). 80 For this reason, our study images the deep seismic structure of the continental lithosphere in 81 order to better understand its interactions with the flat slab, its impacts, and what causes and maintains 82 the flat slab. To do so, we performed 3D regional seismic tomography of P- and S-wave residual 83 travel-times using local earthquakes recorded by four temporary seismic campaigns (OVA99, 84 CHARGE, CHARSME, CHASE) to compare the normal (27-30°) subduction region south of 33.5°S 85 (taken to represent conventional subduction conditions) with the flat subduction zone, between 31°S 86 and 32°S. We then compared our absolute seismic velocities with those predicted for subduction-type 87 rocks, using the Hacker and Abers (2004) worksheet, at appropriate pressure and temperature (P-T) 88 conditions obtained from thermo-mechanical modeling. We also analyzed our results with those of 89 Wagner et al. (2005), who performed a similar local experiment for this region, using only the 90 CHARGE database. As an improvement to their work, our study incorporates a much larger 91 earthquake catalog (3770 events) representing different time periods and resulting in greater ray

2

92 coverage, increased resolution, and a more complete temporal view of the seismic properties of the 93 lithospheres. Furthermore, our results include the continental crust and the upper portion of the slab 94 lithosphere, which could not be interpreted in their models due to lack of data and poor resolution. We 95 also supplemented unpublished seismic attenuation models for the region, calculated by Deshayes 96 (2008), using the OVA99 and CHARSME catalogs, to reinforce our interpretations of the nature of 97 our observed velocity perturbations. Our work offers a finer-scale, broader and more complete image 98 of the region’s subduction system. 99 We report seismic differences between the normal and flat slab region, which confirm previous 100 interpretations of (1) a generally cold, dry, and Mg-rich continental mantle, and (2) an eclogitized 101 lower crust in the Cuyania terrane. However, our results also advocate localized mantle hydration, 102 reflecting flat slab dehydration processes, as well as the absence of an eclogitized Andean arc crustal 103 root. The flat slab segment exhibits fast seismic properties which we can only explain with dense 104 eclogite or peridotite for the crust and mantle, respectively, contradicting the flat slab buoyancy 105 effects. We also notice a correlation between geological terrane boundaries at the surface and seismic 106 velocity variations. Regions of localized slow velocity and high-Vp/Vs anomalies are consistent with 107 major basement shear zone domains, previously interpreted to reach the slab interface. Finally, the 108 forearc and Punitaqui aftershock regions, located above the flat slab, show strong increases in Vs and 109 very low Vp/Vs ratios, which we interpret as possibly related to one another,. 110

111 2. Tectonic and Geological Settings

112 The Nazca Plate subducts beneath central Chile and western Argentina (29°-35°S) at a current 113 convergence rate of 6.7 ± 0.2 cm/a in the N78°E direction (Fig. 1) (Kendrick et al., 2003). The region 114 between 30°S and the subducting JFR path (~ 32.5°S at the trench) is defined as the flat slab region 115 (Fig. 1C). The flat slab segment occurs at 100-120 km depth and underplates the continental 116 lithosphere for 200-300 km eastwards, before resuming its descent at 68°W, with a 30°-dip angle (Fig. 117 1D) (e.g. Barazangi and Isacks, 1976; Cahill and Isacks, 1992; Anderson et al., 2007). It is bounded to 118 the north by a gentle (27.5°-29.5°S) along-strike transition back towards a normal dip of 30°, whereas 119 to the south, the transition is more abrupt (32.5°-33.5°S) and occurs over only ~ 100 km width (see 120 slab isocontours in Fig. 1A) (Cahill and Isacks, 1992; Araujo and Suarez, 1994; Giambiagi and 121 Ramos, 2002). The nature of the slab deformation in the southern transition zone is unclear, however, 122 is more often interpreted as a sharp bend rather than a tear (Wagner et al., 2005; Pesicek et al., 2012). 123 About 25 Ma, the Farallon Plate broke-up into the smaller Cocos and Nazca Plates, and around 124 18-15 Ma, the Nazca Plate began shallowing in central Chile from ~ 45° to 30° dip angle (Ramos et 125 al., 2002). Around 12-10 Ma, the southward migrating JFR incepted the central Chilean trench, 126 coinciding with increased slab shallowing and the cessation of arc volcanism 9 Ma (Kay and 127 Mpodozis, 2002). The JFR is currently subducting along the trench at 33°S (Fig. 1A) since 11 Ma 128 (Yañez et al., 2002), and the present flat slab geometry (Fig. 1C) is believed to have been achieved 129 since 6 Ma (Kay et al., 1991; Kay and Abbruzzi, 1996). 130 Since the onset of slab shallowing below the South American plate, eastward volcanic arc 131 migration and increasing compressional stresses have been linked with strong tectonic uplift and 132 shortening ( up to ~ 130 to 150 km, in comparison to less than < 100 km further south, Allmendinger 133 et al., 1997; Kley and Monaldi, 1998; Alvarado et al., 2007). The overriding plate is thus now 134 described by five different morpho-tectono-structural-geological provinces (from west to east) (Fig. 135 1B): the Coastal Cordillera (forearc), Principal and Frontal Cordilleras (present-day active Andean 136 belt), the Precordillera (foothill) and the Sierra Pampeanas (backarc).

3

137 Significant changes occur between the flat and normal subduction zones, in association with the 138 change in slab geometry. To mention only a few changes with respect to the flat slab region (Fig. 1): 139 (i) the strike of the trench rotates from N20°E to N5°E, (ii) the geothermal gradient increases 140 (Miranda, 2001), (iii) the slab becomes younger (Yáñez and Cembrano, 2000), (iv) the arc magmatism 141 resumes, (v) the Central Depression valley re-appears, (vi) the backarc basement no longer uplifts, and 142 the Precordillera and Sierra Pampeanas disappear (Ramos, 2009), (vii) the topography decreases from 143 an average elevation of 4500 m (max. elevation is 6700 m, Mount Aconcagua, at ~ 70°W/32.5°S) to < 144 2000 m at 38°S, (viii) the crustal thickness decreases from 70 to 35 km (Tassara et al., 2006; Alvarado 145 et al., 2007), (ix) the total lithospheric thickness also decreases from 80-100 km (Tassara et al., 2006) 146 to 60 km south of 36°S (compared to > 140 km in Peru and > 160 km in the eastern Brazilian shield, 147 the thickest in South America), and finally, (x) the crustal seismic rates decrease in the slab and 148 backarc regions, however increase along the active volcanic arc. 149 Three major episodes of terrane accretion have occurred against the western proto-Gondwana 150 margin (Rio de la Plata craton) over the past 600 Ma, and have strongly influenced the structure and 151 tectonic evolution of central Chile (Ramos et al., 1986; Ramos et al., 2002; Alvarado and Ramos, 152 2011). These terranes have various provenances, compositions and ages, and comprise of (Fig. 1B): (i) 153 the para-autochthonous Pampia terrane, amalgamated at ~ 530-515 Ma, (ii) the allochthonous Cuyania 154 terrane, accreted at ~ 460 Ma, and (iii) the allochthonous Chilenia terrane (its existence is still 155 debated), added at ~ 420-315 Ma. The Chilenia terrane is believed to compose the Andean basement 156 and part of the forearc (Ramos, 2004). At the surface, suture zones, characterized by narrow ophiolitic 157 belts now representing major shear zones, separate each terrane and control the neo-tectonic style of 158 deformation of the region (Ramos et al., 2002) by influencing the thick-skinned basement uplift of the 159 Sierra Pampeanas region up to ~ 800 km away from the trench (Fig. 1B) (Ramos et al., 2002; 160 Alvarado et al., 2009). However, these terrane boundaries have never been mapped at depth, and the 161 composition of the Andean basement remains enigmatic (Chilenia vs. Cuyania). Such consequent 162 backarc basement uplift over a distance of several hundreds of kilometers, is equally observed above 163 the modern Peruvian flat slab (James and Snoke, 1994) and within the Laramide Orogeny in western 164 USA (DeCelles, 2004), where an ancient flat slab episode of the Farallon Plate is believed to have 165 shaped this uplifted region. Hence, the central Chilean flat subduction is considered and studied as a 166 modern analog of the Laramide Orogeny. 167

168 3. Seismic Data

169 3.1 Seismic Catalog

170 Our seismic tomography inversion is based on passive local earthquakes recorded by four short- 171 term seismic campaigns (Fig. 1A): 172 . OVA99 campaign was deployed in the region of Ovalle (30°-32°S, 72°-70°W) during the period of 173 mid-November 1999 to mid-January 2000. Thirty-seven short-period three-component receivers, 174 about 30 km apart, recorded continuously the local micro-seismicity (0.5 < Ml < 5.5) with a 175 sampling rate of 125 pts/s. 176 . CHARGE (Chile-Argentina Geophysical Experiment, an American project of the University of 177 Arizona; Fromm et al. 2004; Wagner et al. 2005) continuously recorded seismicity from December 178 2000 to May 2002, and consisted of 22 broadband seismometers (10 STS-2, 10 Guralp CMG- 179 3ESPs and 2 Guralp-40T) deployed mainly along two profiles at 30°S and 36 °S, and some in 180 between. Sampling rate was 40 pts/s. 181 . CHARSME (CHile ARgentina Seismological Measurement Experiment) was carried out from mid- 182 November 2002 to March to analyze the regional seismo-tectonic features and map the Nazca slab

4

183 geometry. It comprised 29 portable broadband three-component stations (27 CMG-40T and 2 184 CMG-3T Guralp) that continuously recorded seismicity with a sampling rate of 125 pts/s. 185 . CHASE (CHile-Argentina Seismic Experiment) recorded the seismicity from mid-November 2005 186 to March 2006 included, and was composed of 14 broadband and 12 short-period seismometers. It 187 focused around the area of Santiago city (33.5°-34°S and 70°-71°S) with the main goal of locating 188 shallow seismicity along fault systems and quantifying the seismic hazard of the capital region, and 189 slab events were also recorded. 190 . Events recorded by 15 permanent seismic stations from the Chilean Seismological Service of the 191 University of Chile were added, to increase receptor coverage and improve hypocenter 192 determination near coastal areas. 193 OVA99, CHARSME, and CHASE seismic networks were installed and maintained by a 194 collaborative group of experts from the GéoAzur laboratory in France, the IRD, and the Geophysical 195 Department of the University of Chile, in Santiago. 196

197 3.2 1D Background Model and Event Selection

198 All events were consistently located, using the Hypoinverse program (Klein, 2000) within a 199 ‘minimum’ average 1D velocity model of least-square fit that best describes the region, constructed 200 using passive and active sources from the region, and the Velest program (Kissling et al., 1994) was 201 used for depths greater than 20 km. The chosen average velocity model represents a 17 layer model 202 with average Vp/Vs ratio of 1.76, extrapolated to 215 km depth for the inversion (Fig. 2). It is used by 203 the Chilean Seismological Service of the University of Chile. In comparison with the IASP-91 global 204 average velocity model, which is used in Wagner et al. (2005)’s seismic tomography study, our model 205 produces generally faster seismic velocities and a deeper average continental Moho, which values we 206 will detail below. 207 To ensure stability and reliability of our tomography results, we selected only the highest 208 quality events from our total event catalog, with the following selection criteria for P- and S-waves, 209 respectively: (a) maximum error of manual picking of ± 0.25 s and ± 0.4 s, (b) maximum pick quality 210 index of 2 and 3 (0: excellent, 4: discarded), (c) maximum hypocenter uncertainty of 5 km in all 2 211 directions, (d) maximum RMS misfit of < 0.6 s (Ri /N, where Ri is the time residual at the ith 212 station), and (e) minimum 8 and 4 station observations. Finally, we retain a total of 3770 events for the 213 inversion process, including 55128 and 54889 P- and S-wave arrival times, respectively. 214

215 4. Seismic Tomography: Method

216 4.1 3D Seismic Tomography Inversion

217 We apply the tomography code TLR3 (Latorre et al., 2004; Monteillet et al., 2005) to obtain 3D 218 P- and S-wave velocity models. Source-receiver ray tracing is calculated using the finite-difference 219 algorithm developed by Podvin and Lecomte (1991), within a fine grid model of mesh size 2x2x2 km 220 to enable ray-path smoothness. Our inversion model is discretized into a coarser regular grid of mesh 221 size 40x40x10 km, representing dimensions of 960x840 x220 km, in which body-wave traveltimes are 222 calculated, for computational efficiency purposes. The least-square 3D velocity structure is 223 progressively retrieved through a set of iterations that solve for small velocity perturbations, in order 224 to linearize the problem and reduce computational time constraints caused by the large dataset. 225 Hypocenters are simultaneously relocated at each step. The starting iteration involves the initial 1D 226 velocity model (Fig. 2) and its relocated seismicity. Values for damping and weight ratio (Cp/Cs) of P-

5

227 and S-wave traveltimes were adjusted separately by fixing one parameter at a time and allowing the 228 other to vary, in order to obtain those that minimize the RMS misfit solution. The parameters that led 229 to best results were a damping value of 0.7 and a Cp/Cs value of 0.5. Mesh spacing and final velocity 230 model robustness were examined using the checkerboard and spike tests. The well resolved regions of 231 the model have a spatial resolution representative of the mesh spacing used, and represent areas where 232 the seismicity is densest.

233 4.2 Final Solution Model Quality Assessment

234 A posteriori, we assessed the resolution quality of our final model using the ray density 235 distribution and the harmonic checkerboard (Fig. 3) and spike sensitivity tests (Fig. 4). The latter two 236 examine the possible artifacts introduced in the model space during the inversion process, by assessing 237 the restoration of amplitude, shape and extent of synthetic velocity perturbations. Our checkerboard 238 tests considered three different mesh spacing, 80x80x20, 40x40x10 and 30x30x10 km, with the latter 239 two offering the best resolution solutions. We finally retained a mesh spacing of 40x40x10 km, 240 because 80x80x20 km and 30x30x10 km offer too coarse or too detailed and complex resolutions, 241 respectively. We then constructed a synthetic spike test model that reflects the main seismic 242 characteristics (perturbation sign, location and geometry) obtained in our final velocity model. As 243 shown by the checkerboard and spike tests, the regions closest to the subduction interface, where ray 244 densities are highest, are best resolved. The overriding lithosphere is well resolved above the flat and 245 normal slabs, with a good recovery of both amplitudes and geometry of the synthetic input model. 246 Also, the upper part of the oceanic lithosphere, assumed to be represented by the thick seismogenic 247 zone (~ 20-30 km inside the slab) is well recovered throughout the region. On the other hand, the 248 backarc region over the flat slab (the Cuyania terrane) is less well resolved, with some horizontal 249 smearing and diminished amplitudes, and above the normal slab, it lacks ray paths and is even more 250 poorly resolved. 251

252 5. Seismic Tomography: Results

253 5.1 The Forearc Crust

254 Between 5 and 25 km depth, the forearc crust (Coastal Cordillera) is described throughout the 255 region with fast seismic velocities and low Vp/Vs ratios, limited to the east by the western Andean La 256 Ramada-Aconcagua fold-and-thrust belt, forming the boundary between the Coastal and Principal 257 Cordilleras (Fig. 5A). Between 0 and 10 km depth and at latitudes < 32°S (in the flat slab region), Vs 258 is highest (3.6-3.7 km/s) and results in very low Vp/Vs ratios of 1.69-1.73, peaking in the Punitaqui 259 region (Fig. 5A and 6B). South of 32°S (above the subducting JFR and in the normal slab region), Vs 260 is lower (3.4-3.5 km/s) and Vp/Vs are higher (1.74-1.76). Vp is constant throughout the region (6.2 261 km/s). Directly below 10 km depth, is a 10 km thin area of reduced Vp and Vs, producing higher 262 Vp/Vs ratios (1.78) (Fig. 6A and B), and connecting the plate interface with the Principal Cordillera. 263 Two other areas of distinct reduced seismic velocities and high Vp/Vs ratios occur near the 264 surface, above the normal slab, with unknown origins: (1) northeast of Valparaiso city (70°W/33°S), 265 down to ~ 25 km depth, and which is associated with a cluster of deep seismicity (Fig. 5A); and (2) 266 directly beneath Rancagua city (~ 70.5°W/34°S), down to only ~ 10 km depth, and which displays a 267 Vp/Vs ratio that is by far the highest in the region (1.86). These high Vp/Vs ratios could suggest 268 important fluid (O’Connell and Budiansky, 1974) or (but less likely) melt (Watanabe, 1993) 269 concentrations in the rock, with a possible thermal influence inherited from the relatively recent 270 Andean magmatism (4-9 Ma, i.e. responsible for the famous El Teniente and Rio Blanco-Los Bronces

6

271 porphyry copper deposits) currently active volcanism in the Eastern Principal Cordillera (Farías et al., 272 2010). In both cases, it may be of socials interest to re-evaluate the local seismic risks at these 273 localities, given these new observations.

274 5.2 The Main Andean Crust

275 The entire Andean crust, and particularly the Principal Cordillera, is characterized by 276 significantly reduced seismic velocities and moderately high Vp/Vs ratios (~ 1.78-1.79) (Fig. 5A). 277 Beneath the active and ancient volcanic arcs, down to 25-30 km depth, the Principal Cordillera 278 experiences the slowest seismic velocities together with moderately high Vp/Vs ratios (1.78-1.80) 279 (Fig. 5B). The orientation (NNW-trending) and dimension (~ 100 km wide) of this anomaly mimic 280 those of the La Ramada-Aconcagua deformation belt. Beneath and ~ 50 km west of the active volcanic 281 arc (i.e. in the Principal Cordillera and > 33°S), seismic velocities are more reduced and associated 282 with intense and deep seismicity (Fig. 5A and B), related to the arc volcanism. Alternatively, the 283 Frontal Cordillera shows less reduced seismic velocities, with the exception of the region between 284 33°S and 34°S, bounded to the east by the Cuyania-Chilenia suture zone, where Vp and Vs are as 285 reduced as in the Principal Cordillera and the Vp/Vs ratios are locally higher (Fig. 5A), suggesting a 286 similar impact factor, such as structural control, at least within shallow depths. At deeper crustal 287 depths (> 30 km), this slow velocity anomaly, originating at the surface in the Principal Cordillera, 288 propagates downward and eastward, into the Andean crustal root (beneath the Frontal Cordillera) and 289 into the sublithospheric continental mantle to the slab interface beneath Cuyania (68.5°W) (Fig. 6A 290 and B). The (seismic) continental Moho (Fromm et al., 2004) is well defined by the isocontour lines 291 Vp ~ 8.0 km/s and Vs ~ 4.5 km/s (Fig. 6).

292 5.3 The Backarc Crust

293 Our resolution of the backarc region is limited but is best within the flat slab region (< 32.5°S) 294 (Fig. 3 and 4), representing the Cuyania terrane (Precordillera and western Sierra Pampeanas), due to 295 the higher ray coverage produced by the flat slab’s seismicity. The Cuyania lithospheric column, down 296 to 75-80 km depth, is described by slightly higher Vp and Vs values and lower Vp/Vs ratios (1.75- 297 1.76) (Fig. 5, 6A and B) than the environing lithosphere. Inconsistently, previous studies, based on 1D 298 velocity forward modeling, characterized it rather with high Vp/Vs values (> 1.80) (Gilbert et al., 299 2006; Alvarado et al., 2007). Finally, as mentioned in Section 4.2, our resolution of the backarc above 300 the normal slab (> 33°S) is inadequate for discussion.

301 5.4 The Continental Mantle

302 Throughout the region, between 50 and 60 km depth, the continental mantle exhibits reduced 303 Vp and Vs with higher Vp/Vs ratios (1.77-1.78) than the surrounding mantle (Fig. 6A-C). Above the 304 flat slab segment (> 70 km depth), the continental mantle is characterized by relatively high Vp (8.0 to 305 8.5 km/s), high Vs (4.5 to 4.8 km/s), and relatively low Vp/Vs ratios (1.75-1.77), with localized 306 patches of higher Vp/Vs values (1.79) (Fig. 6A and B). Beneath the Frontal Cordillera, the mantle is 307 reduced in seismic velocities, a feature which appears to extend deeper eastward to the flat slab’s 308 interface at 68.5°W, directly beneath the Cuyania terrane. This 10-20 km thin area of slow velocities, 309 located beneath the Cuyania terrane, is located above the eastern tip of the flat slab segment, before re- 310 subduction occurs, and represents locally increased Vp/Vs ratios (1.78) (Fig. 6D), possibly indicating 311 hydrous mantle conditions. 312 The only marked difference in the seismic velocities of the continental mantle between the flat 313 and normal regions is observed between 35 and 50 km depth. Here, at latitudes < 32°S (in the flat slab

7

314 region), Vs increases (4.0-4.4 km/s) substantially compared to above the normal slab (3.9-4.1 km/s), 315 and Vp remains constant (6.9-7.7 km/s) everywhere, resulting in very low Vp/Vs ratios (1.70-1.74) 316 (Fig. 5B), analogous to the shallow forearc directly above it.

317 5.5 The Slab

318 In the flat slab region (< 32.5°S), the slab down to ~ 50 km depth is highly seismogenic, 319 strongly reduced in seismic velocities and contains high Vp/Vs ratios (1.80-1.82), suggesting the 320 presence of fluids (Fig. 6A and B). Between 50 and 75 km depth, both the Vp/Vs ratios and the DSZ’s 321 upper plane seismic rate decrease (Fig. 6A). At 31°S, 71°W and 50-80 km depth, a local anomaly of 322 decreased Vp (8.0-8.1 km/s), strongly increased Vs (4.5-4.7 km/s) and very low Vp/Vs ratios (1.70- 323 1.73), defines the 1997 Punitaqui (Mw 7.1) aftershock region (shown by the blue star in Fig. 6B). Its 324 peak low Vp/Vs value is located in the interplanar region of the DSZ. Along the flat slab segment (> 325 100 km depth), the western part of the slab, between 69°W and 70.5°W, is expressed with higher Vp 326 (8.5-8.6 km/s) and Vs (4.8-4.9 km/s) than the slightly slower eastern part (8.4-8.5 and 4.7-4.8 km/s, 327 respectively), a difference which is particularly pronounced along the JFR axis (Fig. 6A and B). 328 Nevertheless, the Vp/Vs ratios remain constant and relatively low (1.75-1.77) throughout the flat slab 329 segment. 330 In comparison, the normal slab (> 33°S) down to 50 km depth is less seismogenic and less 331 pronounced in slow seismic velocities, with slightly lower Vp/Vs ratios (1.76-1.78) (Fig. 6C). 332 Between 50 and 75 km depth, the seismic activity increases and Vp decreases, resulting in even lower 333 Vp/Vs ratios (1.74-1.76). Below 125-150 km depth, the seismic activity becomes rare. 334

335 6. Petrological Analysis: Method

336 6.1 Thermo-Mechanical Estimation of P-T Conditions

337 In order to correlate our obtained seismic velocities with those predicted for rock types likely to 338 constitute the lithospheres, we need to estimate the appropriate pressure (P) and temperature (T) 339 conditions at depth. A conventional approach consists in defining kinematical models of subduction 340 that calculate a geothermal gradient equilibrated with the velocity of subduction (Springer, 1999; 341 Gutscher et al., 2000) and assume a pressure simply equal to the lithostatic pressure. Here instead, we 342 chose to estimate the P-T field along two vertical cross-sections representing the central Chilean flat 343 (31.5°S) and normal (33.5°S) subduction, with a thermo-mechanical method that can account for some 344 tectonic overpressure as well as shear heating along the plate interface. The plane-strain finite 345 differences code Parovoz (Poliakov and Podladchikov, 1992) was used, which is based on the FLAC 346 method (Fast Lagrangian Analysis of Continuum, Cundall and Board, 1988). This code has been used 347 in a variety of geodynamic contexts (e.g. Gerbault et al. 2009, and references therein), and resolves the 348 equations of motion and heat transfer explicitly in a time-marching scheme, with self-consistent 349 elasto-visco-plastic T-dependent rheologies (details in Gerbault et al. 2009). 350 The modeled domain is first sub-divided into several rheological units describing the different 351 portions of the continental and oceanic lithospheres at present-day. The geometries of our two starting 352 modeled profiles are constructed based on our catalog hypocenter distribution at depth, and 353 complemented by seismic reflection and wide-angle data for depths less than 10 km (Flueh et al., 354 1998). The initial geotherms are evaluated according to the method proposed by Burov and Diament 355 (1996) assuming: i) effective thermal ages of the oceanic and continental lithospheres of 35 Ma and 356 200 Ma, respectively (similar to estimates by Tassara et al. 2006), ii) radiogenic heating in the 35 km 357 thick reference continental crust (8.10-10 W/kg with an exponential decay length set to 10 km), and iii)

8

358 best fitting surface heat flow data (Hamza and Muñoz, 1996). The thickness of the lithosphere (LAB) 359 is determined by the 1350°C isotherm, set at 100 km and 150 km depths above the flat and normal 360 slabs, respectively (according to values determined by Tassara et al. 2006). Plate convergence is 361 applied at 7.5 cm/yr, as proposed by Somoza and Ghidella (2005) and O'Neill (2005). Other thermo- 362 mechanical parameters can be found in Gerbault et al. (2009), as it is not the aim of the present paper 363 to discuss them. 364 Our aim here is not to reproduce the progressive formation (or not) of a flat subduction, for 365 which mechanical causes are still subject of debate (for a recent discussion, see e.g. Cerpa et al., 366 submitted). Instead, we determine thermo-mechanical parameters for which the present day geometry 367 is stable over several millions of years, a timing a priori consistent with the geological record (e.g. 368 Haschke et al. 2006) and appropriate to generate the observed seismic velocities. Therefore, the 369 computational time is chosen to span ~ 5 Ma, during which the geometrical setup evolves relatively 370 little (yet external changes in kinematical and thermal initial conditions can be considered negligible at 371 this time-scale as well as internal phase changes, e.g. see discussion in Gerbault et al., 2009). As a 372 result, two synthetic P-T models are obtained, for the normal and flat dipping slab sections, shown in 373 Fig. 7. These P-T conditions can now be used as input data for the petrological analysis, with a rather 374 permissive range of uncertainty, as detailed below.

375 6.2 Petrological Modeling

376 For each cell and for several discrete sites in our 2D representative cross-sections of the central 377 Chilean flat (31.5°S) and normal (33.5°S) subduction zones (Fig. 8), we analyzed which rock types 378 matched best our calculated seismic velocities for (i) the forearc crust, (ii) the continental lower crust, 379 (iii) the continental mantle, and (iii) the resolved upper slab regions. We based our analysis on 380 experimentally measured seismic isotropic properties of mafic and ultramafic rock compositions, 381 representative of subduction zones (Hacker et al., 2003). 382 The Hacker and Abers (2004) rock and mineral database provides compositional details for 383 many common subduction rock types, including 25 MORB-type rocks, 19 hydrous peridotites (10 384 harzburgites, 9 lherzolites) and 21 anhydrous peridotites (10 lherzolites, 7 harzburgites, 1 dunite, 385 wherlite, olivine clinopyroxenite, and pyrolite). We tested for all rock types and accounted for a 386 permissive uncertainty range of ± 0.1 km/s in our Vp and Vs, ± 100°C in T and ± 0.5 GPa in P. We did 387 not compare Vp/Vs ratios, since small variations within the uncertainty constraint of either Vp or Vs, 388 or both, resulted in substantial misleading differences in the Vp/Vs values. Hence, we only 389 acknowledged absolute velocities. 390 The aim of our exercise is not to find specific rock compositions that best match our seismic 391 wave field at depth, but instead to observe whether trends exists, such as recurring rock facies (e.g. 392 eclogites vs. blueschists) or in the case of mantle rocks, hydrated vs. non-hydrated, garnet vs. 393 plagioclase peridotites. In order to do so, we calculated the total average volume percentage of water ∑푁(푉표푙%) 394 and other minerals (vol%) contained in the rock solutions describing each cell ( 푖 푖, where i is 푁 395 each rock solution and N is the total number of rock solutions per cell). In such a statistical analysis, 396 the reliability of our results is quantified by the number of rock solutions in each cell, shown in Fig. 397 8B, which in turn is dependent on the total number of rocks present in our database, considered to be 398 fairly high. 399 We have not used more complete or self-consistent mineral databases, such as Perplex 400 (Connolly, 2005) and Theriak (Capitano and Petrakakis, 2010), since these methods generally require 401 that detailed rock and mineral compositions be known (obtained from field sampling and laboratory 402 analyses), which is not our case here. Furthermore, the resolutions associated to our thermo- 403 mechanical and tomographic data are too large to justify the use of more complex and precise

9

404 methods, which furthermore, are best applied to crustal domains than to mantle domains. The Hacker 405 and Abers (2004) database, in turn, has been quite well developed for mantle domains, which is our 406 principal interest in this study. 407 Figure 8B shows that the continental and oceanic mantle domains have a large set of rock 408 solutions matching our seismic velocities. Because there is little seismic distinction between 409 lherzolites and harzburgites of the same type (e.g. garnet) at similar P-T conditions, the number of 410 rock solutions for the mantle domain are much high compared to the continent crustal domain, which 411 contains a larger panoply of rock compositions, describing a larger range of seismic velocities for 412 which our velocities could match only fewer. Nevertheless, our results are considered reliable for both 413 domains, since our seismic velocities in each cell generally fitted rocks with similar compositions in 414 major elements (e.g. garnet peridotites) or trends (e.g. hydrous vs. anhydrous), whereby in the latter 415 case we calculated total averages of volume percentages (e.g. average vol% H2O). 416

417 7. Discussion and Interpretation

418 7.1 The Forearc Crust

419 The forearc crust in the flat slab region (< 32°S) displays several unusual seismic characteristics 420 which are difficultly explained, and possibly linked to one another and/or with the flat slab geometry. 421 Here, the elevated Vs and low Vp/Vs ratios, with respect to the normal slab region, suggest the 422 presence of denser and/or colder rocks, with negligible water content. Indeed, a compositional change 423 was proposed by Sallarès and Ranero (2005), who also noted similar results of faster Vs (> 4.0 km/s) 424 north of 33°S than to the south (Vs ~ 3.5 km/s), possibly attributed to the higher degree of plutonism 425 and metamorphism in the forearc basement rocks (Ramos et al., 1986). 426 Comparing our seismic velocities (Vp: 6.2 km/s, Vs: 3.6 km/s) with those predicted for typical 427 mafic rocks at 5 km depth using the Hacker and Abers (2004) database, we find that normal mafic 428 rocks should have much faster seismic velocities at such depths (average ∆Vp: +1.25 km/s, ∆Vs: 429 +0.66 km/s), suggesting either abnormal/deformed mafic rocks or prevailing felsic compositions. We 430 could not test for typical felsic rock compositions using the Hacker and Abers (2004) rock database 431 since these are not included however, we could examine the influence of increasing the silica content 432 on the calculated seismic velocities. We observe that increasing the silica content of a rock decreases 433 Vp significantly more than Vs, rather than increasing the Vs. Furthermore, we calculated seismic 434 velocities for an average granite composition (74.5% SiO2, 14% Al2O3, 9.5% Na2O and K2O, 2% 435 oxides (Fe, Mn, Mg, Ca) at 5 km depth (P: 0.5 GPa, T: 100°C) using PerpleX (Connolly, 2005) and 436 found much slower seismic velocities (Vp: 5.9 km/s, Vs: 2.5 km/s) than our observed values, in 437 particular fir Vs. Since our Vp is constant throughout the region, we consider that our Vs must be 438 abnormally high in the flat slab region, contradicting the effects of increased silica content. 439 If a compositional change poorly justifies the seismic differences between the northern and 440 southern part of the forearc crust, a compositional change between the forearc (Coastal Cordillera) and 441 the Andean crust (Principal Cordillera) seems to more easily explain its eastern boundary at or near the 442 western La Ramada-Aconcagua fold-and-thrust-belt (Principal Cordillera), since such compositional 443 contrasts are often associated with major shear zones (Twiss and Moores, 1992). This correlation was 444 reported in northern Chile, between 15°S and 23°S, where a sharp density contrast exists between the 445 forearc and the Altiplano-Puna crusts (Tassara, 2005), associated with high gravity, high resistivity 446 and low P-wave attenuation values, and interpreted as cold and rigid material acting as an indenter into 447 the weaker Altiplano-Puna lithosphere (Tassara et al., 2006; Gerbault et al., 2009). In comparison with 448 our case study, the forearc crust in the flat slab region also reflects high gravity anomalies (Andrès 449 Tassara, personal communication), but does not promote any seismic attenuation pattern, as shown in

10

450 Fig. 9A (Deshayes, 2008). Nevertheless, a compositional change associated to increases in density 451 should induce simultaneous increases in Vp and Vs, not simply Vs (as it seems to be our case). In 452 addition, a decrease in rock temperature, reflecting the flat slab’s lower geotherm, can also be 453 discarded since it should produce an equal effect as increased density. To resume, normal isotropic 454 mafic rock compositions, higher rock density, and lower rock temperatures, do not convincingly 455 explain the shallow (< 10 km depth) forearc seismic anomaly. 456 Deeper, between 35 and 50 km depth, an identical anomaly of high-Vs and very-low-Vp/Vs also 457 locates in the forearc crust at similar latitudes (< 32°S) as its shallower counterpart, suggesting a 458 relationship between one another. Understanding this deeper forearc anomaly might shed light onto 459 the shallower anomaly. 460 The deeper forearc anomaly lies within the uncertainty range of the poorly defined seismic 461 Moho (Fromm et al., 2004), and hence, it could represent crust or mantle material. Our petrological 462 analysis shows that mafic rocks, rather than mantle rocks, match far better the seismic velocities at this 463 depth range and at appropriate P-T conditions (n°8 in Fig. 10A); whereas, mantle rocks explain better 464 the deeper seismic velocities (n°9 in Fig. 10A), indicating the lower limit of this anomaly probably 465 represents the boundary between the crust and mantle, which is supported by the sudden decrease in 466 the slab’s seismic activity (Fig. 6A and B) and the change to seismic creep at this depth (50 km) 467 reported by Gardi et al. (2006) based on Coulomb stress transfer analyses of major interface events 468 that occurred in the region. However, no rocks match the very low Vp/Vs values of the shallow and 469 deep forearc crust (1.69-1.73). 470 A study by Ramachandran and Hyndman (2012) reported even lower Vp/Vs values (1.65) for 471 the forearc lower crust in the Cascadia subduction zone, located between 25 and 35 km depth, equally 472 associated with elevated Vs (4.1-4.2 km/s), however also by decreased Vp (6.6-6.8 km/s), unlike us. 473 Comparing absolute seismic velocities, our Vs is similar, however our Vp is significantly higher (7.0- 474 7.1 km/s). Takei (2002) indicated that adding only 5% of silica to a rock decreases significantly its 475 Vp/Vs ratio, and Ramachandran and Hyndman (2012) explained their seismic anomaly with 20% 476 added silica, derived from consequent amounts of subducted eroded continental material precipitated 477 from slab fluids. Nevertheless, as mentioned earlier, increasing the silica content of a rock decreases 478 Vp substantially more than Vs, contradicting our observations. 479 In addition, a link with the intraslab Punitaqui earthquake aftershock region (1997, Mw 7.1), 480 located at 31°S and 70 km depth (Fig. 6B) cannot be excluded, considering that they all reflect 481 identical seismic behaviors. Another correlation can also be made with the flat slab geometry at depth. 482 To resume, within the flat slab region, two very-low-Vp/Vs anomalies occur inside the forearc 483 crust and another in the slab, for which we can find no reasonable explanation for their seismic 484 behaviors nor a link with the flat slab geometry underneath. Simple mechanisms, such as normal mafic 485 and ultramafic rock compositions, increased silica concentrations, decreased temperature or increased 486 rock density, are discarded. However, one might consider seismic anisotropy effects resulting from 487 convergence deformation, since MacDougall et al. (2012) reported strong seismic anisotropy in the 488 mantle wedge below the forearc and above the flat slab. However, this interpretation is beyond the 489 reach of our data. 490 Finally, we superposed the basement detachment fault geometry of Farías et al. (2010)’s 491 tectonic model for the Andes at 34°S (La Ramada-Aconcagua fold-and-thrust belt) onto our 492 tomography models (Fig. 6) and noticed a good correlation between the fault’s westward extension 493 and the high-Vp/Vs region separating both very-low-Vp/Vs anomalies, located at 30 km depth in the 494 forearc crust (Fig. 6B). Thus, we support Farías et al. (2010)’s proposition that a crustal shear zone 495 crosses the forearc at mid-crustal depths, reaching out for the subduction interface and possibly 496 channelizing slab fluids into the continental crust, where high-Vp/Vs ratios dominate.

11

497 7.2 The Main Andean Crust

498 The NS-trending slow velocity and high-Vp/Vs anomaly affects the upper 30 km of the 499 Principal Cordillera crust and mimics well the location, geometry and dimension of the La Ramada- 500 Aconcagua fold-and-thrust belt (Fig. 1B, 6A and B) (~ 100 km wide and ~ 20-30 km deep; Ramos et 501 al., 2002; Giambiagi et al., 2003; Farías et al., 2010). Therefore, this slow anomaly is interpreted as the 502 consequence of structural damage, in addition to the regular presence of meta-sedimentary rocks and 503 fluids (reflected by the higher Vp/Vs ratios). In section 7.1, we interpreted the basement detachment 504 fault of the La Ramada-Aconcagua belt reaching the subduction interface across the forearc’s mid- 505 crust. 506 The Frontal Cordillera is less affected by such strong velocity reductions, indicating stronger 507 and less fractured crust (Ramos et al., 2002), in agreement with its lower seismic rate. The only 508 exception lies in the Cordon del Plata, west of Mendoza city (33°-34°S), displaying similar significant 509 decreases in seismic velocities as in the Principal Cordillera, down to 35 km depth (Fig. 5A). Although 510 this region lies outside the limits of the Aconcagua fold-and-thrust belt at this latitude, it is also 511 interpreted to experience structural damage (Alvarado et al., 2009), in addition to being at proximity to 512 the active volcanic arc, thus creating a higher geothermal gradient that could explain slow seismic 513 velocities. 514 At depth, little is known about the composition and origin of the Andean basement rocks, as 515 there is little evidence left for the existence of the Chilenia terrane, obliterated by abundant old 516 intrusions and covered by volcanism and sedimentation. However, the continental affinity of the oldest 517 rocks found in the Frontal Cordillera (Caminos et al., 1982) supports its existence. Also, there is no 518 reconnaissance for the location of the Chilenia-Cuyania suture zone at depth (Ramos et al., 1986), 519 which would answer the questions about the Andean basement’s composition. Our results show that 520 the lower Andean crust is strongly reduced in seismic velocities, with high Vp/Vs ratios (1.77-1.79), 521 relative to the Andean mid- and Cuyania crusts (Fig. 6A and B). Our seismic velocities for the Andean 522 root, between 50 and 60 km depth, can be explained by lawsonite amphibole eclogite (57% H2O), as 523 well as jadeite epidote blueschist (74% H2O) and garnet granulite (0% H2O) assemblages. And 524 between 60 and 70 km depth (within the uncertainty range for the seismic Moho), the seismic 525 velocities match either hydrated eclogites (zoisite and zoisite-amphibole eclogites, 32 and 10% H2O, 526 respectively) or may represent mantle rocks hydrated with an average of 10-30% H2O (n°10, Fig. 527 10A). Therefore, the higher Vp/Vs ratios and the mafic rock solutions for the Andean lower crust 528 suggest it is likely hydrated and principally non-eclogitized (at least until 60 depths), contrary to our 529 expectations, since the crustal root extends well into the eclogite stability field, normally starting 530 around 50 km depth (e.g. Hacker et al., 2003; Hacker and Abers 2004). 531 In general, the absence of an eclogitized crustal root can be explained by the following: (1) the 532 rock composition is too felsic to transform to eclogite. This is highly likely, since felsic rocks have 533 slower seismic velocities than mafic rocks and the Andean crust is suspected to encompass the 534 continental Chilenia terrane; (2) the lower crust temperature is too low for eclogite transformation to 535 occur efficiently, even in the presence of catalyzing fluids (Artemieva and Meissner, 2012); and (3) 536 crustal delamination may have removed the previously dense eclogitized lower crust, due to overcome 537 gravitional instability (DeCelles et al., 2009). Crustal detachment was proposed by Kay and Gordillo 538 (1994) to explain the deep and old crustal signatures in the Pocho lava rocks above the re-subducting 539 slab (65°W, 32°S). 540 Therefore, the Andean slow velocity lower crust can be explained by a felsic or a mafic non- 541 eclogitized crustal root, with the likely presence of fluids. In the case of a felsic root, this would attest 542 for the presence of Chilenia and provide insight to the location of the suture zone and distribution of 543 the Chilenia and Cuyania terranes at depth.

12

544 7.3 The Backarc Crust

545 The Grenvillian Cuyania terrane (Precordillera and western Sierra Pampeanas) stands out as a 546 fast seismic velocity anomaly down to 80 km depth, with relatively low Vp/Vs ratios (Fig 6A-B). 547 Consistent with fast seismic velocities, Cuyania is also mapped as a high gravity anomaly, overprinted 548 by a low geothermal gradient, until 37°S (Tassara et al., 2006). These observations reflect its 549 mafic/ultramafic, and cold and old character (Ramos et al., 1986; 2002; Ramos, 2004). However, we 550 can only observe its high Vp until ~ 33°S and high Vs until ~ 32°S, rather than 37°S (Fig. 5A). The 551 peak of intensity is situated more-or-less beneath San Juan city, which is directly above the inferred 552 subducting JFR track. 553 Multiple seismic and gravity studies measured a crustal thickness of 60 km beneath the 554 Precordillera and 50-55 km beneath the western Sierra Pampeanas (Regnier et al., 1994; Fromm et al., 555 2004; Alvarado et al., 2005; 2007; Gilbert et al., 2006, Calkins et al., 2006; Corona, 2007; Alvarado et 556 al., 2009; Castro de Machuca et al., 2012). These all proposed the existence of a dense eclogitized 557 lower crust to explain seismic velocity jumps and their unexpected high gravity values relative to their 558 average low elevations (~ 2 and 1 km, respectively) (Miranda, 2001; Alvarado et al., 2005; 2007; 559 Gilbert et al., 2006). Indeed, our obtained high seismic velocities for the Cuyania lower crust (around 560 68.75°W) confirm the presence of eclogites below 50 km depth, matching amphibole (26% H2O) and 561 zoisite (10% H2O) eclogites (~ 3.50 g/cm3). Between 40 and 50 km depth, lawsonite amphibole 562 eclogite (57% H2O) could occur, amongst jadeite epidote blueschist (74% H2O) and garnet granulite 563 (0% H2O). Since eclogite is not seismically distinguishable from mantle rocks, we are unable to 564 provide information about the maximum depth extent of this eclogite layer. 565 The presence of eclogite supports a mafic composition, indicating that Cuyania (and not 566 Chilenia) forms the backarc basement, constraining further the suture zone location between the 567 terranes.

568 7.4 The Continental Mantle

569 Whether the flat slab retains or releases fluids is key information to better constrain the reasons 570 for its flat geometry and the fate of deformation in the overriding lithosphere, since fluids decrease the 571 bulk rock density, increase buoyancy, and reduce rock strength (e.g. Reynard et al., 2013). 572 Above both the flat and normal slab regions, and between 50 and 60 km depth, our results show 573 that the continental mantle’s seismic velocities have a 70-80% chance of matching serpentinized 574 mantle rocks (i.e. 6 out of 9, and 5 out of 7 rocks, respectively, contain hydrous phases) with an 575 average of 30 % hydration, represented by the minerals amphibole (avg. 15%), talc (avg. < 5%), 576 chlorite and Å-phase (avg. < 10%). This is in agreement with the locally higher Vp/Vs ratios in the 577 mantle at this depth range, and the reduced seismic activity in the Wadati-Benioff zone of the adjacent 578 slab (Fig. 6A-B). The mantle in the flat slab region, down to 80 km depth, exhibits a pronounced low 579 P-wave attenuation anomaly (Fig. 9A-B) (Deshayes, 2008). However, the effects of saturated versus 580 under-saturated fluid conditions on seismic attenuation behaviors are complex and difficult to assess 581 (Toksöz et al., 1979). 582 The very small variations in the seismic velocities of the mantle above both normal and flat 583 slabs (Fig. 10A-B) imply similar degrees of mantle hydration; yet, the mantle above the normal slab 584 strongly attenuates P- and S-waves (Fig. 9C) (Deshayes, 2008) and appears able to produce sufficient 585 partial melts to feed the overlying volcanoes. Hence, we wonder what is the role of temperature in 586 causing both the flat and normal slabs to dehydrate similarly, and its influence in triggering the mantle 587 wedge’s partial melting. We also question whether the mantle wedge adjacent to the normal slab is

13

588 unusually cold and dry, or if instead the mantle above the flat slab is abnormally hydrated, given the 589 lower temperature circumstances and our expectation of differences in seismic velocities. 590 The erupted lavas of the Tupungatito, San José and Maipo volcanoes, situated between 33.5°S 591 and 35°S, reflect higher degrees of fractionation and crustal contamination than the rest of the 592 Southern Volcanic Zone (SVZ), due to longer magma ascent and residence times in the crust (Michael 593 Dungan, personal communication). This can be explained by (1) smaller magma production, relative 594 to the rest of the SVZ, reflecting poor slab dehydration processes or low mantle temperatures, or by 595 (2) high crustal compressional stresses associated to the flat subduction and consequent impeding 596 magma ascent. In both cases, it appears that the flat slab conditions to the north may affect the normal 597 slab conditions directly to its south over a few hundred kilometers, at least between ~ 32.5°-34.5°S. 598 Based on geological evidences, Folguera and Ramos (2011) proposed that current Pampean flat 599 subduction once extended further south until 38°S at 17 Ma, then its southern segment progressively 600 steepened to its present angle of subduction (30°) at 5 Ma, resulting in the SVZ. This hypothesis 601 explain the similar seismic structure of the mantle wedges above both normal and flat slabs in the 602 region, and also renders our assumed “normal” subduction conditions at 33.5°S closer to a transitional 603 state, with closer to normal conditions expected to be found further south along the SVZ (> 38°S). 604 Therefore, although the tectonics near the surface changes rapidly as a response to the dramatic shift in 605 the slab geometry at depth, on the other hand, the continental mantle appears more homogeneous 606 (seismically) and less abruptly affected by these sudden changes, perhaps owing to an ambient low 607 viscosity flow that homogenizes the mantle properties. 608 The continental mantle above the flat slab segment contains faster Vp (8.0 to 8.4 km/s) and Vs 609 (4.5 to 4.8 km/s) values, and Vp/Vs ratios mainly around 1.75-1.77, in agreement with previous 610 interpretations for a cold and dry mantle (n°12, 14, 22 and 23 in Fig. 10A). We find that mostly Opx- 611 and Mg-rich garnet peridotites (lherzolites and harzburgites) match these seismic velocities, a 612 conclusion also settled by Wagner et al. (2005; 2006; 2008), based on another processing method of 613 the CHARGE seismic data. The slower continental mantle directly below the Frontal Cordillera (n°13) 614 is significantly attenuated in P-waves (Fig. 9A) (Deshayes, 2008) and, along with the mantle directly 615 beneath the Cuyania crust (n°21), matches partially hydrated (avg. 10-20 vol% H2O) mantle rocks. 616 However, the proximity of this material to the ill-defined Moho could also indicate that it represents 617 lower crust material instead (hydrous eclogites, see section 7.3). 618 We present here additional evidence that the mantle is probably locally hydrated directly above 619 the flat slab’s eastern extremity, between 68°W and 69°W, shown by Vp and Vs reductions and 620 increased Vp/Vs ratios over a small volume of 100x100x20 km3 (Fig. 5C, 6A-B and D). This is 621 supported by other geophysical investigations which highlighted this region as strongly reduced in Vp 622 (Bianchi et al., 2013), highly attenuated in S-waves (Fig. 9A) (Deshayes, 2008) and significantly 623 conductive (Booker et al., 2004; Orozco et al., 2013), all partisan of fluids effects. Also, its 624 geographical position coincides with the subducting JFR track (Fig. 5C, 6A-B and D), indicating that 625 the oceanic lithosphere along the seamount ridge may dehydrate into the overlying continental mantle. 626 On the other hand, though Wagner et al. (2005) also found strong Vp reductions (< 7.4 km/s) 627 associated to this area, their Vs remains high (4.6 km/s), resulting in very low Vp/Vs values (~ 1.65). 628 In comparison, our model indicates a similar Vs value, but a much higher Vp (8.2 km/s) and Vp/Vs 629 ratio (1.78) (Fig. 6D). By analyzing mantle rock seismic properties over a range of possible P-T 630 conditions appropriate for here, we find no correspondence at all with Wagner et al. (2005)’s seismic 631 velocities (n°23w in Fig. 10A), suggesting that their Vp might be too low because of their use of a 632 global background model. As an interpretation of our own seismic velocities, only dry garnet 633 peridotites (n°23) can explain our results (Fig. 10A), creating a paradox for mantle hydration. 634 To remedy this paradox of too fast seismic velocities, we tested the influence on our results of a 635 slower background 1D velocity model for mantle depths (Fig. 11A). The resulting effects on the final

14

636 velocities at the flat slab’s depth are a reduction in Vp of up to 0.1 km/s, and in Vs of up to 0.3 km/s. 637 The region exhibiting mantle hydration signatures (n°23) is now described by a Vp of 8.0-8.1 km/s 638 and a Vs of 4.5-4.6 km/s (Fig. 11B), corresponding to dry or slightly serpentinized peridotites (avg. 639 10-20 vol% H2O) (Fig. 12A). 640 Therefore, we hypothesize that the flat slab segment slowly dehydrates along the track of the 641 JFR prior to its re-subduction into the deeper mantle, as a consequence of its proximity to the hot 642 asthenosphere directly east. We expect those released fluids to be stored in the adjacent continental 643 mantle where the slow velocity anomaly can be observed (n°23), and possibly to migrate upwards into 644 the Andean lower crust via crustal shear zones that cross-cut the mantle, explaining the locally higher 645 Vp/Vs ratios and the seismic attenuations. Crustal faults are believed to penetrate into the continental 646 mantle (e.g. Fromm et al., 2004; Gilbert et al., 2006), despite the region’s low geothermal gradient and 647 probably higher rock rigidity.

648 7.5 The Oceanic Lithosphere

649 The vertical resolution of our model (10 km) limits us to studying the oceanic mantle rather than 650 the oceanic crust, considering that the crust there has a normal thickness of less than 10 km, as 651 indicated by the seismic surveys offshore (~ 8 km thick along the JFR; Kopp et al. 2004). 652 In the flat slab region, the oceanic lithosphere is characterized by slow seismic velocities and 653 high Vp/Vs ratios (1.80) down to about 60-70 km depth, below which the Vp/Vs ratio diminishes to 654 1.75-1.77 (Fig. 6A-B). At 25 km depth, we match these velocities with fully serpentinized mantle 655 rocks (n°1), and with 60-70% (n°24), 50-60% (n°25) and 0-20% serpentinization at 40 km, 50 km and 656 70 km depths, respectively (Fig. 10A). Similar results are obtained for the normal slab region, with a 657 slab appearing to be 50-60% serpentinized at 50 km depth (n°9-10), and hardly (n° 17-18) or totally 658 dry (n° 14-13) below 70 km depth (Fig. 10B). Ray coverage is poor above 50 km depth, but tends to 659 assimilate to highly serpentinized rocks. Furthermore, the plate interface at 25 km depth in the flat slab 660 region is likely a zone of strong fluid concentration (n°1 in Fig. 10A), shown by the high Vp/Vs ratios 661 (1.82) and decreased Vs (Fig. 6A-B), coherent with the effects of fluid-filled pore compaction and 662 primary rock dehydration reactions causing expulsion of slab fluids at these depths (Kirby et al., 1996; 663 Peacock, 2001; Hacker et al., 2003). 664 Although the slight velocity variations (~ 0.05 km/s) between the normal and flat slab regions 665 may result in no difference in the predicted rock compositions, they nevertheless bring non-negligible 666 discrepancies in the Vp/Vs ratios. Therefore, the higher Vp/Vs ratios in the flat slab region attest in 667 favor of higher plate hydration along the subducting JFR. Indeed, there is intense lithospheric 668 hydration offshore, observed by seismic and gravity investigations and identified as the Challenger 669 Fracture zone which the JFR is part of (Yáñez et al., 2002; Grevemeyer et al., 2003; 2005; Kopp et al., 670 2004; Ranero et al., 2005; Clouard et al., 2007; Contreras-Reyes et al., 2010), whereby intense 671 fracturing and faulting allow particularly deep seawater penetration to occur, at least down to 20 km 672 depth (Clouard et al., 2007). 673 Along the flat slab segment, Gans et al. (2011) used receiver function analyses to locate the 674 slab’s surface and Moho, and advocate for a thicker oceanic crust (~ 15 km) than measured offshore (~ 675 8 km). They also interpreted a vertical offset along an inherited fault (shown in Fig. 6A-B), and found 676 that the distribution of this overthickened crust spans a larger area (~ 300 km) than the flat slab’s 677 width estimated by the dense seismicity and the JFR width offshore (both 100 km; shown by the white 678 line and band in Fig. 5; Kopp et al. 2004). In the case of a 15 km-thick oceanic crust, our maximum 679 vertical resolution of 10 km should enable us to detect the oceanic crust, which should be noticeable as 680 a slow velocity anomaly especially if it is non-eclogitized. Non- or partially-eclogitized oceanic crust 681 is suggested as a possible factor inducing flat subduction in this region (Martinod et al., 2005; 2010;

15

682 Gans et al., 2011), since basaltic crust is on average 0.5 g/cm3 lighter than eclogite (3.5 g/cm3). Its 683 effects would evermore be enhanced in the case of an overthickened crust (Gans et al., 2011). 684 At flat slab P-T conditions (400-800°C at 2-4 GPa, van Hunen et al., 2002), normal average 685 basaltic crust has Vp values between 7.2 and 7.6 km/s (Hacker et al., 2003). Not only do our results 686 not show a slow velocity anomaly along the flat slab segment possibly indicating non-eclogitized 687 conditions or enabling us to confirm Gans et al. (2011)’s observation, but our fast seismic velocities 688 (Vp ~ 8.5-8.6 km/s, Vs ~ 4.8-4.9 km/s) also correspond to only dense and anhydrous eclogites (3.5-3.6 689 g/cm3). When we test our results using a slower background velocity model (Fig. 11A), the resulting 690 smaller seismic velocities (Fig. 11B) now match hydrous amphibole eclogite (20% H2O); however, 691 the paradox for a buoyant slab is maintained, since its bulk rock density is as high as for dry eclogites 692 (3.5 g/cm3), representing a positive density anomaly of 0.2-0.3 g/cm3 relative to normal mantle (3.3 693 g/cm3). In the case where our seismic velocities, from both background models, rather illuminate 694 oceanic mantle (due to the lack of model resolution for the oceanic crust), the corresponding rock 695 types are dry and dense peridotites (n°15-17, 27 in Fig. 10A and 12A). 696 To resume, the flat slab segment is composed of dry oceanic mantle with either (1) a very dense 697 (3.5-3.6 g/cm3), thick (> 10 km) and fully eclogitized oceanic crust, or (2) a normally thick oceanic 698 crust (7-8 km) which we cannot observe in our model due to lack of resolution. This observation 699 challenges the concept of buoyancy, posing a paradox of how can flat subduction occur when the 700 lithospheric density is so high? Then one should seek for another mechanism to explain flat 701 subduction (e.g. see alternative mechanisms in Skinner and Clayton, 2013; Cerpa et al., submitted). In 702 addition, a fully transformed eclogitic crust should no longer be seismically active, contradicting the 703 very high seismic activity defining the flat slab geometry (Fig. 1C) and the subducting JFR path. 704 Furthermore, in agreement with other studies (see Section 7.4), our results show that the JFR may be 705 dehydrating at the eastern tip of the flat slab, indicating that fluids are retained in the slab until this 706 point and that eclogite transformation is incomplete. This interpretation is backed by the high mantle 707 conductivity values adjacent to the re-subducting slab (Booker et al., 2004) and by the recent (1.9 Ma, 708 Kay and Abbruzzi, 1996) volcanism above it, in San Luis located in the Sierras Cordoba province 709 (65°W). This hypothesis of a partially eclogitized and hydrous oceanic crust with thickness less than 710 10 km, explains well the flat slab’s high seismic activity and positive buoyancy. 711

712 Conclusions

713 In conclusion, our results show that there are significant seismic differences between the 714 continental lithospheres above the flat and normal slabs, in central Chile and western Argentina: 715 . The forearc crust above the flat slab (< 32°S) exhibits high Vs and very low Vp/Vs ratios at two 716 distinct depth ranges, 0-10 and 35-50 km. No normal rock type was found to correlate with these 717 seismic velocities. However, we suggest possible links with transient short time-scale variations 718 similar to the Punitaqui aftershock region that portrays identical seismic behaviors, the flat slab 719 geometry, and/or seismic anisotropy. 720 . The Andean crust throughout the region is characterized by significantly slow seismic velocities 721 and relatively high Vp/Vs ratios, especially at depth ranges of 0-30 km and 40-70 km. The shallow 722 anomaly is explained by intense structural damage correlated with the La Ramada-Aconcagua fold- 723 and-thrust belt. And the deeper crustal root anomaly is interpreted as a fluid-bearing non- 724 eclogitized mafic crust possibly indicating a past delamination event (Kay and Gordillo, 1994), or a 725 felsic crust, coinciding with a deep extent of the Chilenia terrane. 726 . The backarc crust is described with fast seismic velocities outlining the Cuyania terrane, which are 727 coherent with eclogitic compositions below 50 km depth, supporting many previous interpretations.

16

728 . We interpret localized regions of higher Vp/Vs ratios in the forearc crust and mantle wedge as 729 hydrated zones representing basement detachment shear zones reaching out for the subduction 730 interface and channelizing slab-derived fluids into the Andean crust. 731 . The continental mantle above the flat slab portion is expressed by fast seismic velocities and 732 relatively low Vp/Vs ratios, probably enhanced by the region’s low geotherm, and matching dry 733 Mg-rich peridotite seismic properties at such depths, as previously suggested. This coincides well 734 with the more intense deformation style at the surface, and the cessation of arc volcanism at the 735 surface. 736 . We show evidence that the subducting Juan Fernandez Ridge locally dehydrates at the eastern tip 737 of the flat slab (68.5°W/ 31.5°S), before re-subduction occurs, suggesting that some fluids are 738 retained within the flat slab, explaining part of its buoyancy. In contradiction with a previous study, 739 we interpret the oceanic crust along the flat slab segment as thinner than our model’s vertical 740 resolution of 10 km, since our seismicity velocities here are high and can only reflect very dense 741 eclogites. An eclogitic oceanic crust strongly disrupts the role of the slab’s buoyancy in 742 maintaining flat subduction, thus calling for an alternative causative mechanism (which is not our 743 aim to discuss here). In addition to advocating for non-overthickened flat slab oceanic crust, we 744 propose that it is instead non- or only very partially-eclogitized, which is consistent with its 745 ongoing dehydration process along the JFR, the observed dense seismic activity, as well as further 746 east, the higher mantle conductivity values (Booker et al., 2004) and recent volcanism (Kay and 747 Gordillo, 1994). 748 Nevertheless, there appears to be little seismic variations in the mantle wedges above both flat 749 and normally dipping slabs, describing similar levels and depths of hydration (until 70 km depth), and 750 indicating that this similitude is caused by the influence of, and proximity with, the flat slab impacting 751 on the region directly to its south. 752

753 Acknowledgements

754 Local data were obtained thanks to projects FONDECYT 1020972-1050758 and IRD- 755 GéoAzur. Marianne Marot and Guust Nolet are supported by the GlobalSeis project (ERC 226837). 756 Giorgio Ranalli’s participation in this project is supported by a grant by NSERC (Natural Sciences and 757 Engineering Research Council of Canada). We are very grateful to all the members who have actively 758 discussed and debated our subject with us, bringing us new insights and challenges to our topic. These 759 include, Andres Tassara, Eric Ferre, Lara Wagner, Stephan Rondenay and Manuele Faccenda. We also 760 would like to thank Perrine Deshayes who shared with us her seismic attenuation tomography results 761 for this region, which she performed during her PhD thesis project. 762 763

17

1 References

2 3 Allmendinger, R. W., Jordan, T. E., Kay, S. M., and Isacks, B. L. (1997). The evolution of the 4 Altiplano-Puna plateau of the Central Andes. Annual Review of Earth and Planetary Sciences, 5 25(1), 139-174. Doi: 10.1146/annurev.earth.25.1.139

6 Alvarado, P., Beck, S., Zandt, G., Araujo, M., and Triep, E. (2005). Crustal deformation in the 7 south‐central Andes backarc terranes as viewed from regional broad‐band seismic waveform 8 modeling. Geophysical Journal International, 163(2), 580-598. doi:10.1111/j.1365- 9 246X.2005.02759.x.

10 Alvarado, P., Beck, S., and Zandt, G. (2007). Crustal structure of the south‐central Andes Cordillera 11 and backarc region from regional waveform modeling. Geophysical Journal International, 12 170(2), 858-875. doi :10.1111/j.1365-246X.2007.03452.x

13 Alvarado, P., Pardo, M., Gilbert, H., Miranda, S., Anderson, M., Saez, M., and Beck, S. (2009). Flat- 14 slab subduction and crustal models for the seismically active Sierras Pampeanas region of 15 Argentina. MWR204: Backbone of the Americas: Shallow subduction, plateau uplift, and ridge 16 and terrane collision: Boulder, Colorado, Geological Society of America, 261-278. James, D. 17 E., and Snoke, J. A. (1994). Structure and tectonics in the region of flat subduction beneath 18 central Peru: Crust and uppermost mantle. Journal of Geophysical Research: Solid Earth 19 (1978–2012), 99(B4), 6899-6912. doi:10.1130/2009.1204(12)

20 Alvarado, P., and Ramos, V. A. (2011). Earthquake deformation in the northwestern Sierras 21 Pampeanas of Argentina based on seismic waveform modeling. Journal of Geodynamics, 51(4), 22 205-218.

23 Anderson, M., Alvarado, P., Zandt, G., and Beck, S. (2007). Geometry and brittle deformation of the 24 subducting Nazca Plate, Central Chile and Argentina. Geophysical Journal International, 25 171(1), 419-434. doi: 10.1111/j.1365-246X.2007.03483.x

26 Araujo, M., and Suarez, G., (1994). Geometry and state of stress of the subducted Nazca Plate beneath 27 central Chile and Argentina: Evidence from teleseismic data. Geophysical Journal 28 International, 116, 283-303. doi: 10.1111/j.1365-246X.1994.tb01799.x

29 Artemieva, I. M., and Meissner, R. (2012). Crustal thickness controlled by plate tectonics : A review 30 of crust-mantle interaction processes illustrated by European examples. Tectonophysics, 530, 31 18-49.

32 Barazangi, M., and Isacks, B. L. (1976). Spatial distribution of earthquakes and subduction of the 33 Nazca plate beneath South America. Geology, 4(11), 686-692. doi: 10.1130/0091-7613(1976)

34 Booker, J. R., Favetto, A., and Pomposiello, M. C. (2004). Low electrical resistivity associated with 35 plunging of the Nazca flat slab beneath Argentina. Nature, 429(6990), 399-403. 36 doi:10.1038/nature02565.

37 Burov, E., and Diament, M. (1996). Isostasy, equivalent elastic thickness, and inelastic rheology of 38 continents and oceans. Geology, 24(5), 419-422. doi:10.1130/ 0091- 39 7613(1996)024<0419:IEETAI>2.3.CO;2. 40 Cahill, T., and Isacks, B. L. (1992). Seismicity and shape of the subducted Nazca plate. Journal of 41 Geophysical Research, 97(B12), 17503-17. doi:10.1029/92JB00493

42 Calkins, J. A., Zandt, G., Gilbert, H. J., and Beck, S. L. (2006). Crustal images from San Juan, 43 Argentina, obtained using high frequency local event receiver functions. Geophysical Research 44 Letters, 33(7). doi:10.1029/2005GL025516.

45 Caminos, R., Cingolani, C. A., Hervé, F., and Linares, E. (1982). Geochronology of the pre-Andean 46 metamorphism and magmatism in the Andean Cordillera between latitudes 30 and 36 S. Earth- 47 Science Reviews, 18(3), 333-352. http://dx.doi.org/10.1016/0012-8252(82)90043-5

48 Capitano, C., de, and Petrakakis, K. (2010). The computation of equilibrium assemblage diagrams 49 with the Theriak/Domino software. America Mineralogist, 95, 7, 1,006-1,016. 50 doi:10.2138/am.2010.3354

51 Castro de Machuca, B., Perarnau, M., Alvarado, P., López, G., and Saez, M. (2012). Modelo cortical 52 sismológico y petrológico para el sudoeste de la sierra de Pie de Palo, provincia de San Juan. 53 Revista de la Asociación Geológica Argentina, 69(2), 179-186.

54 Cerpa Gilvonio, N., Hassani, R., Gerbault, M., Prévost, J. H. A fictitious domain method for 55 lithosphere/asthenosphere interaction : application to periodic slab folding in the upper mantle, 56 Geochemistry, Geophysics, Geosystems, in submission.

57 Clouard, V., Campos, J., Lemoine, A., Perez, A., and Kausel, E. (2007). Outer rise stress changes 58 related to the subduction of the Juan Fernandez Ridge, central Chile. Journal of Geophysical 59 Research: Solid Earth (1978–2012), 112(B5). doi:10.1029/2005JB003999.

60 Connolly, J. A. D. (2005). Computation of phase equilibria by linear programming: a tool for 61 geodynamic modeling and its application to subduction zone decarbonation. Earth and 62 Planetary Science Letters, 236(1), 524-541.

63 Contreras‐Reyes, E., and Osses, A. (2010a). Lithospheric flexure modelling seaward of the Chile 64 trench: implications for oceanic plate weakening in the Trench Outer Rise region. Geophysical 65 Journal International, 182(1), 97-112. doi:10.1111/j.1365-246X.2010.04629.x

66 Corona, G., (2007). Estructura Litosférica del Sistema Andes Sierras Pampeanas en la Banda 30°S– 67 31°S a Partir de Datos de Gravedad y Sísmicos [Undergraduate fi nal work thesis]: San Juan, 68 Argentina, Facultad de Ciencias (in Spanish)

69 Cundall, P. A., and Board, M. (1988). A microcomputer program for modeling large strain plasticity 70 problems, Numerical Methods in Geomechanics G. Swoboda, Rotterdam, Netherlands. 6, 2101– 71 2108.

72 DeCelles, P. G. (2004). Late Jurassic to Eocene evolution of the Cordilleran thrust belt and foreland 73 basin system, western USA. American Journal of Science, 304(2), 105-168.

74 Deshayes, P. (2008). Tomographie en vitesse et en atténuation de la zone de subduction au Chili 75 central-ouest de l'Argentine (29° S-34° S) à partir de données sismologiques locales: apport à 76 l'étude de la composition minéralogique (Doctoral dissertation, Université de Nice Sophia- 77 Antipolis) (in French) 78 Farías, M., Comte, D., Charrier, R., Martinod, J., Tassara, A., and Fock, A. (2010). Crustal-scale 79 structural architecture of the Central Chile Andes based on 3D seismic tomography, seismicity, 80 and surface geology: Implications for mountain building in subduction zones. Tectonics, 29. 81 doi:10.1029/2009TC002480

82 Flueh, E. R., Vidal, N., Ranero, C. R., Hojka, A., Von Huene, R., Bialas, J., and Zelt, C. (1998). 83 Seismic investigation of the continental margin off-and onshore Valparaiso, Chile. 84 Tectonophysics, 288(1), 251-263.

85 Fromm, R., Zandt, G., and Beck, S. L. (2004). Crustal thickness beneath the Andes and Sierras 86 Pampeanas at 30 S inferred from Pn apparent phase velocities. Geophysical Research Letters, 87 31(6). doi:10.1029/2003GL019231.

88 Gans, C. R., Beck, S. L., Zandt, G., Gilbert, H., Alvarado, P., Anderson, M., and Linkimer, L. (2011). 89 Continental and oceanic crustal structure of the Pampean flat slab region, western Argentina, 90 using receiver function analysis: new high‐resolution results. Geophysical Journal 91 International, 186(1), 45-58.

92 Gardi, A., Lemoine, A., Madariaga, R., and Campos, J. (2006). Modeling of stress transfer in the 93 Coquimbo region of central Chile. Journal of Geophysical Research: Solid Earth (1978–2012), 94 111(B4). doi:10.1029/2004JB003440

95 Gerbault, M., Cembrano, J., Mpodozis, C., Farias, M., and Pardo, M. (2009). Continental margin 96 deformation along the Andean subduction zone: Thermo-mechanical models. Physics of the 97 Earth and Planetary Interiors, 177(3), 180-205.

98 Giambiagi, L. B., and Ramos, V. A. (2002). Structural evolution of the Andes in a transitional zone 99 between flat and normal subduction (33.30°–33.45° S), Argentina and Chile. Journal of South 100 American Earth Sciences, 15(1), 101-116.

101 Giambiagi, L. B., Ramos, V. A., Godoy, E., Alvarez, P. P., and Orts, S. (2003). Cenozoic deformation 102 and tectonic style of the Andes, between 33 and 34 south latitude. Tectonics, 22(4), 1041.

103 Gilbert, H., Beck, S., and Zandt, G. (2006). Lithospheric and upper mantle structure of central Chile 104 and Argentina. Geophysical Journal International, 165, 383–398. doi:10.1111/j.1365- 105 246X.2006.02867.x.

106 Grevemeyer, I., Diaz-Naveas, J. L., Ranero, C. R., and Villinger, H. W. (2003). Heat flow over the 107 descending Nazca plate in central Chile, 32 S to 41 S: Observations from ODP Leg 202 and the 108 occurrence of natural gas hydrates. Earth and Planetary Science Letters, 213(3), 285-298. 109 doi.org/10.1016/S0012-821X(03)00303-0

110 Grevemeyer, I., Kaul, N., Diaz-Naveas, J. L., Villinger, H. W., Ranero, C. R., and Reichert, C. (2005). 111 Heat flow and bending-related faulting at subduction trenches: case studies offshore of 112 Nicaragua and Central Chile. Earth and Planetary Science Letters, 236(1), 238-248. doi: 113 10.1016/j.epsl.2005.04.048

114 Gutscher, M. A., Spakman, W., Bijwaard, H., and Engdahl, E. R. (2000). Geodynamics of flat 115 subduction: Seismicity and tomographic constraints from the Andean margin. Tectonics, 19(5), 116 814-833. 117 Hacker, B. R., G. A. Abers, and S. M. Peacock (2003). Subduction factory 1. Theoretical mineralogy, 118 densities, seismic wave speeds, and H2O contents. Journal Geophysical Research, 108(B1), 119 2029. doi:10.1029/2001JB001127.

120 Hacker, B. R., and Abers, G. A. (2004). Subduction Factory 3: An Excel worksheet and macro for 121 calculating the densities, seismic wave speeds, and H2O contents of minerals and rocks at 122 pressure and temperature. Geochemistry, Geophysics, Geosystems, 5(1). 123 doi:10.1029/2003GC000614

124 Hamza, V. and Munoz, M. (1996). Heat flow map of South America. Geothermics, 25, 48, 599-656

125 Haschke, M., Günther, A., Melnick, D., Echtler, H., Reutter, K. J., Scheuber, E., and Oncken, O. 126 (2006). Central and southern Andean tectonic evolution inferred from arc magmatism. In The 127 Andes (pp. 337-353). Springer Berlin Heidelberg. 128 129 James, D. E., and Snoke, J. A. (1994). Structure and tectonics in the region of flat subduction beneath 130 central Peru: Crust and uppermost mantle. Journal of Geophysical Research: Solid Earth 131 (1978–2012), 99(B4), 6899-6912.

132 Kay, S. M., Mpodozis, C., Ramos, V. A., and Munizaga, F. (1991). Magma source variations for mid– 133 late Tertiary magmatic rocks associated with a shallowing subduction zone and a thickening 134 crust in the central Andes (28 to 33 S). Geological Society of America Special Papers, 265, 113- 135 138.

136 Kay, S. M., and Gordillo, C. E. (1994). Pocho volcanic rocks and the melting of depleted continental 137 lithosphere above a shallowly dipping subduction zone in the Central Andes. Contributions to 138 Mineralogy and Petrology, 117(1), 25-44.

139 Kay, S. M., and Abbruzzi, J. M. (1996). Magmatic evidence for Neogene lithospheric evolution of the 140 central Andean “flat-slab” between 30 S and 32 S. Tectonophysics, 259(1), 15-28.

141 Kay, S.M. and Mpodozis, C. (2002). Magmatism as a probe to the Neogene shallowing of the Nazca 142 plate beneath the modern Chilean flat- slab. Journal of South American Earth Sciences, 15, 39- 143 57.

144 Kendrick, E., Bevis, M., Smalley Jr, R., Brooks, B., Vargas, R. B., Laurıa, E., and Fortes, L. P. S. 145 (2003). The Nazca–South America Euler vector and its rate of change. Journal of South 146 American Earth Sciences, 16(2), 125-131. doi:10.1016/S0895-9811(03)00028-2

147 Kirby, S., Engdahl, R. E., and Denlinger, R. (1996). Intermediate-depth intraslab earthquakes and arc 148 volcanism as physical expressions of crustal and uppermost mantle metamorphism in 149 subducting slabs. Geophysical Monograph Series, 96, 195-214. doi:10.1029/GM096p0195

150 Kissling, E., Ellsworth, W. L., Eberhart‐Phillips, D., and Kradolfer, U. (1994). Initial reference models 151 in local earthquake tomography. Journal of Geophysical Research: Solid Earth (1978–2012), 152 99(B10), 19635-19646. doi :10.1029/93JB03138

153 Klein, F. W. (2000). User's Guide to HYPOINVERSE-2000 a Fortran Program to Solve for 154 Earthquake Locations and Magnitudes, USGS Open File Report. Draft 5/19. 155 Kley, J., and Monaldi, C. R. (1998). Tectonic shortening and crustal thickness in the Central Andes: 156 How good is the correlation?. Geology, 26(8), 723-726. doi: 10.1130/0091-7613(1998) 157 026<0723:TSACTI>2.3.CO;2

158 Kopp, H., Flueh, E. R., Papenberg, C., and Klaeschen, D. (2004). Seismic investigations of the 159 O'Higgins Seamount Group and Juan Fernández Ridge: Aseismic ridge emplacement and 160 lithosphere hydration. Tectonics, 23(2). doi:10.1029/2003TC001590.

161 Latorre, D., Virieux, J., Monfret, T., Monteiller, V., Vanorio, T., Got, J. L., and Lyon‐Caen, H. (2004). 162 A new seismic tomography of Aigion area (Gulf of Corinth, Greece) from the 1991 data set. 163 Geophysical Journal International, 159(3), 1013-1031. doi:10.1111/j.1365-246X.2004.02412.x

164 MacDougall, J. G., Fischer, K. M., and Anderson, M. L. (2012). Seismic anisotropy above and below 165 the subducting Nazca lithosphere in southern South America. Journal of Geophysical Research: 166 Solid Earth (1978–2012), 117(B12). doi:10.1029/2012JB009538

167 Manea, V. C., Pérez-Gussinyé, M., and Manea, M. (2012). Chilean flat slab subduction controlled by 168 overriding plate thickness and trench rollback. Geology, 40(1), 35-38.

169 Marot, M., Monfret, T., Pardo, M., Ranalli, G., and Nolet, G. (2012). An intermediate-depth tensional 170 earthquake (Mw 5.7) and its aftershocks within the Nazca slab, central Chile: A reactivated 171 outer rise fault?. Earth and Planetary Science Letters, 327, 9-16. doi: 172 10.1016/j.epsl.2012.02.003.

173 Martinod, J., Funiciello, F., Faccenna, C., Labanieh, S., and Regard, V. (2005). Dynamical effects of 174 subducting ridges: insights from 3‐D laboratory models. Geophysical Journal International, 175 163(3), 1137-1150. doi:10.1111/j.1365246X.2005.02797.x.

176 Martinod, J., Husson, L., Roperch, P., Guillaume, B., and Espurt, N. (2010). Horizontal subduction 177 zones, convergence velocity and the building of the Andes. Earth and Planetary Science 178 Letters, 299(3), 299-309. doi:10.1016/j.epsl.2010.09.010.

179 Miranda, S. (2001). Gravity crustal model for an Andean section at 32°S considering upper mantle 180 lateral density variation. Brazilian Journal of Geophysics, 18(2)

181 Monteiller V., Got J.-L., Virieux J. and P.G. Okubo, (2005). An efficient algorithm for double- 182 difference tomography and location in heterogeneous media, with an application to Kilauea 183 volcano, Hawaii, Journal Geophysical Research, 110, B12306, doi:10.1029/2004BJ003466

184 O’Neill, C., Muller, R.D., and Steinberger, B., (2005). On the uncertainties in hot spot reconstructions 185 and the significance of moving hotspot reference frames. Geochemistry, Geophysics, 186 Geosystems, 6(4). doi:10.1029/2004GC000784.

187 O’Connell, R. J., and Budiansky, B. (1974). Seismic velocities in dry and saturated cracked solids. 188 Journal of Geophysical Research, 79(35), 5412-5426. doi: 10.1029/JB079i035p05412

189 Orozco L. A., Favetto A., Pomposiello C., Rossello E., Booker J. (2013). Crustal deformation of the 190 Andean foreland at 31° 30′S (Argentina) constrained by magnetotelluric survey. 191 Tectonophysics, 582, 126–139. doi:10.1145/355993.356000. 192 Peacock, S. M. (2001). Are the lower planes of double seismic zones caused by serpentine dehydration 193 in subducting oceanic mantle?. Geology, 29(4), 299-302. doi:10.1130/0091- 194 7613(2001)029<0299:ATLPOD>2.0.CO;2.

195 Pesicek, J. D., Engdahl, E. R., Thurber, C. H., DeShon, H. R., and Lange, D. (2012). Mantle 196 subducting slab structure in the region of the 2010 M8. 8 Maule earthquake (30–40° S), Chile. 197 Geophysical Journal International, 191(1), 317-324.

198 Podvin P. and Lecompte I. (1991). Finite difference computation of traveltimes in very contrasted 199 velocity models: a massively parallel approach and its associated tools. Geophysical Journal 200 International, 105(1): 271-284

201 Poliakov A. and Podladchikov Y. (1992). Diapirism and topography. Geophysical Journal 202 International, 105(1), 271-284

203 Ramachandran, K., and Hyndman, R. D. (2012). The fate of fluids released from subducting slab in 204 northern Cascadia. Solid Earth, 3, 121-129.

205 Ramos, V. A., Jordan, T. E., Allmendinger, R. W., Mpodozis, C., Kay, S. M., Cortés, J. M., and 206 Palma, M. (1986). Paleozoic terranes of the central Argentine-Chilean Andes. Tectonics, 5(6), 207 855-880. doi:10.1029/TC5i006p855

208 Ramos, V. A., Cristallini, E. O., and Pérez, D. J. (2002). The Pampean flat-slab of the Central Andes. 209 Journal of South American Earth Sciences, 15(1), 59-78. doi:10.1016/S0895-9811(02)00006-8

210 Ramos, V. A. (2004). Cuyania, an exotic block to Gondwana: review of a historical success and the 211 present problems. Gondwana Research, 7(4), 1009-1026.

212 Ramos V. A. (2009) Anatomy and global context of the Andes: Main geologic features and the 213 Andean Orogenic cycle, in Kay S. M., Ramos V. A., and Dickinson W. R., eds, Backbone of the 214 Americas: shallow subduction, plateau uplift, and ridge and terrane collision. Society of America 215 Memoir 204, p. 31-65 doi: 10.1130/2009

216 Ramos, V. A., and Folguera, A. (2011). Payenia volcanic province in the Southern Andes: An 217 appraisal of an exceptional Quaternary tectonic setting. Journal of Volcanology and geothermal 218 Research, 201(1), 53-64. http://dx.doi.org/10.1016/j.jvolgeores.2010.09.008

219 Ranero, C. R., Villasenor, A., Morgan, J. P., Weinrebe, W., (2005). Relationship between bend- 220 faulting at trenches and intermediate-depth seismicity. Geochemistry Geophysics Geosystems, 6, 221 12, Q12002, doi:10.1029/2005GC000997

222 Regnier, M., Chiu, J.M., Smalley, R., Jr., Isacks, B.L., and Araujo, M., (1994). Crustal thickness 223 variation in the Andean foreland, Argentina, from converted waves. Bulletin of the 224 Seismological Society of America, 84,1097–1111.

225 Reynard, B. (2013). Serpentine in active subduction zones. Lithos. 178, 171-15. 226 doi.org/10.1016/j.lithos.2012.10.012 227 Sallarès, V., and Ranero, C. R. (2005). Structure and tectonics of the erosional convergent margin off 228 Antofagasta, north Chile (23°-30°S). Journal of Geophysical Research: Solid Earth (1978– 229 2012), 110(B6). doi:10.1029/2004JB003418.

230 Skinner, S. M., and Clayton, R. W. (2013). The lack of correlation between flat slabs and bathymetric 231 impactors in South America. Earth and Planetary Science Letters, 371, 1-5. 232 http://dx.doi.org/10.1016/j.epsl.2013.04.013

233 Somoza, R., and Ghidella, M.E., (2005). Convergencia en el margen occidental de America del sur 234 durante el Cenozoico: Subduccion de Nazca. Farallon y Aluk. Revista de la Asociacion 235 Geologica Argentina 60, 797–809.

236 Springer, M. (1999). Interpretation of heat-flow density in the Central Andes. Tectonophysics, 306(3), 237 377-395. http://dx.doi.org/10.1016/S0040-1951(99)00067-0

238 Takei, Y. (2002). Effect of pore geometry on Vp/Vs: From equilibrium geometry to crack. Journal of 239 Geophysical Research: Solid Earth (1978–2012), 107(B2), ECV-6.

240 Tassara, A. (2005). Interaction between the Nazca and South American plates and formation of the 241 Altiplano–Puna plateau: Review of a flexural analysis along the Andean margin (15–34 S). 242 Tectonophysics, 399(1), 39-57.

243 Tassara, A., Götze, H. J., Schmidt, S., and Hackney, R. (2006). Three‐dimensional density model of 244 the Nazca plate and the Andean continental margin. Journal of Geophysical Research: Solid 245 Earth (1978–2012), 111(B9). doi:10.1029/2005JB003976.

246 Toksöz, M. N., Johnston, D. H., and Timur, A. (1979). Attenuation of seismic waves in dry and 247 saturated rocks: I. Laboratory measurements. Geophysics, 44(4), 681-690. doi: 248 10.1190/1.1440969

249 Twiss, R. J., and Moore, E. M., (1992). Normal faults, in: Structural Geology, Chap.5, Publisher: 250 Freeman, H. W., first edition, p.74-95

251 van Hunen, J., van den Berg, A. P., and Vlaar, N. J. (2001). Latent heat effects of the major mantle 252 phase transitions on low-angle subduction. Earth and Planetary Science Letters, 190(3), 125- 253 135.

254 van Hunen, J., van den Berg, A. P., and Vlaar, N. J. (2002). On the role of subducting oceanic plateaus 255 in the development of shallow flat subduction. Tectonophysics, 352(3), 317-333.

256 van Hunen, J., van den Berg, A.P., and Vlaar, N.J., (2004). Various mechanisms to induce present-day 257 shallow flat subduction and implications for the younger Earth: a numerical parameter study. 258 Physics of the Earth and Planetary Interiors, 146, 179–194.

259 Wagner, L. S., Beck, S., and Zandt, G. (2005). Upper mantle structure in the south central Chilean 260 subduction zone (30 to 36 S). Journal of Geophysical Research, 110(B1), B01308. 261 doi:10.1029/2004JB003238 262 Wagner, L. S., Beck, S., Zandt, G., and Ducea, M. N. (2006). Depleted lithosphere, cold, trapped 263 asthenosphere, and frozen melt puddles above the flat slab in central Chile and Argentina. Earth 264 and Planetary Science Letters, 245(1), 289-301. doi:10.1016/j.epsl.2006.02.014.

265 Wagner, L. S., Anderson, M. L., Jackson, J. M., Beck, S. L., and Zandt, G. (2008). Seismic evidence 266 for orthopyroxene enrichment in the continental lithosphere. Geology, 36(12), 935-938.

267 Watanabe, T. (1993). Effects of water and melt on seismic velocities and their application to 268 characterization of seismic reflectors. Geophysical Research Letters, 20(24), 2933-2936. 269 Doi: 10.1029/93GL03170

270 Yáñez, G., and Cembrano, J. (2000). Tectonic models for ridge collision at a continental plate 271 boundary: The case of Juan Fernández ridge and Chile rise, Preliminary Results. In Congreso 272 Geológico Chileno (No. 9, pp. 649-654).

273 Yáñez, G., Cembrano, J., Pardo, M., Ranero, C., and Selles, D. (2002). The Challenger–Juan 274 Fernández–Maipo major tectonic transition of the Nazca–Andean subduction system at 33–34 275 S: geodynamic evidence and implications. Journal of South American Earth Sciences, 15(1), 276 23-38. doi:10.1016/S0895-9811(02)00004-4 1 Figure Captions

2 Figure 1: Tectono-seismo-structural-geological context of central Chile and western Argentina. (A) 3 Seismological context of the seismic networks (inverted triangles) and its recorded seismicity (small 4 circles). Shown are: active volcanoes (red triangles), main cities (white circles, capital city Santiago 5 with a star), slab contours from Anderson et al. (2007), the political border between Chile and 6 Argentina (white line), the inferred Juan Fernandez Ridge subduction path and width (semi- 7 transparent white line and band, respectively). B) Tectono-structural-geological context, showing the 8 accreted terranes, major suture zones (thick black lines), geological provinces and their uplifted 9 outcrops (dotted lines), the approximate location of the La Ramada and Aconcagua fold-and-thrust 10 belts (lines with triangles), and main cities (white circles and circled labeled in (A). (C) and (D) The 11 recorded seismicity along two vertical E-W cross-sections along the flat and normal subduction 12 regions, respectively, positioned in (A).

13 Figure 2 : 1D velocity models determined for the region of Central Chile and used as initial model for 14 our tomography study.

15 Figure 3: Checkerboard test results for our final velocity models of P- and S-waves in: (A) plan view 16 at 5, 35 and 95 km depths; (B) and (C) vertical E-W cross-sections along the flat slab at 31.5°S and the 17 normal slab at 33.5°S; and (D) vertical N-S cross-section along the backarc region at 68.5°W. The 18 initial synthetic velocity perturbations for P- and S-waves are 500 and 300 m/s, respectively. The 19 horizontal and vertical node spacing distances are chosen to be 40 and 10 km, respectively. Grey 20 shaded cells in vertical cross-sections reflect areas of poor ray density, determined by fixing a certain 21 threshold value for the total ray-length crossed in each cell. Also shown are the topography and 22 geological and tectonic terrane boundaries (thick dotted and solid lines, respectively), the slab 23 geometry (isocontours in thin dotted lines, from Anderson et al. 2007)the location of the active 24 volcanoes (red triangles) and of the seismic stations used in this study (inverted blue triangles), the 25 inferred slab interface based on our relocated seismicity (dark gray line) and the expected location of 26 the subducting Juan Fernandez Ridge material (thick brow line segment), the continental Moho 27 calculated from seismic (Fromm et al., 2004) and gravity studies (Tassara et al., 2006).

28 Figure 4: Spike test results for our final velocity models of P- and S-waves, shown in: (A) plan view at 29 depth 5, 35 and 95 km depth; (B) and (C) Vertical E-W cross-sections through the flat slab at 31.5°S 30 and normal slab at 33.5°S; and (D) Vertical N-S cross-section along the backarc region at 68.5°W. The 31 initial velocity perturbations imposed on our final velocity inversion model are shown by blue 32 (positive) and red (negative) outlined boxes of amplitude ±500 m/s. We notice that the region closest 33 to the subduction front (i.e. the forearc) is well resolved with almost not distortions, however, the 34 backarc region is less well constrained by the ray coverage, leading to some horizontal smearing, 35 mostly in the continental mantle below Cuyania.

36 Figure 5: Plan view of our final 3D velocity model perturbations for P- and S-waves, relative to our 37 background model, and deduced Vp/Vs ratios, in at depths (A) 5 km, (B) 35 km, and (C) 95 km, cross- 38 cutting the shallow forearc, deeper forearc and continental mantle seismic anomalies (details in text). 39 Contour lines are the absolute P- and S-wave velocities. Red dots are the seismicity at these depth 40 slices. Small empty triangle at the center indicates the location of the Aconcagua highest peak of the 41 region (6500 m). Figure legend same as Figure 3.

42 Figure 6: Vertical cross-sections of our final 3D velocity model perturbations for P- and S-waves, 43 relative to our background model, and deduced Vp/Vs ratios, along latitudes (A) 31.5°S, (B) 31°S, and 44 (C) 33.5°S, describing the flat slab segment where the JFR subducts, the Punitaqui aftershock region; 45 and the normal slab region, respectively; and along the longitude (D) 68.5°W, describing the Cuyania 46 lithosphere and its possible hydrated continental mantle. Shown are: absolute seismic velocities 47 (contour lines); the selected seismicity for our tomography along each profile(red dots, see Section 3.2 48 for details); the tectonic models for the Andes (from Farías et al. 2010) and for the backarc region 49 (from Ramos et al. 2002) (black lines) and our inferred extensions of major basement detachment 50 faults (purely hypothetical and based on our tomography results, dotted line); and another model for 51 the flat slab’s interface (dark red dotted line), oceanic Moho (pink dotted line) and speculative vertical 52 offset (black line) based on receiver function analysis by Gans et al. (2011). Figure legend same as 53 Figure 3.

54 Figure 7: Our final synthetic P-T models computed to represent the central Chilean (A) flat and (B) 55 normal subduction zones, and used for our petrological analysis. See Section 6.1 for details on the 56 computation method. Variations in temperature (T) and pressure (P) are delineated by isocontour lines. 57 The uncertainty of T and P are of the order of ±100°C and ±0.5 GPa, respectively. Figure legend same 58 as Figure 3.

59 Figure 8: Areas sampled for our petrological analysis, whereby our absolute seismic velocities 60 obtained by tomography are compared with predicted rock seismic velocities computed from the 61 Hacker and Abers (2004) worksheet. (A) Discrete and precise locations of interest. Numbers to make 62 reference to Figure 10. (B) Within in cell (40 and 10 km in length and width respectively). Numbers 63 and color code refer to the number of rock solutions found matching our seismic velocities observed in 64 each cell. The left and right columns represent the flat and normal subduction zones at 31.5°S and 65 33.5°S, respectively. Figure legend same as Figure 3.

66 Figure 9: Novel seismic attenuation models for P- and S-waves calculated by Deshayes (2008) for the 67 region using events from the OVA99 and CHARM campaigns, along the (A) flat slab at 31.5°S, (B) 68 transition zone at 32.5°S, and (C) normal slab at 33.5°S. Note that the model is limited at 72°W. The 69 color scale describes a Q-value variation from 400 to 1600, representing strongly attenuating and non- 70 attenuating regions, respectively.

71 Figure 10: Petrological analysis. Comparison of our seismic tomography velocities with predicted 72 rock seismic behaviors (Vp vs. Vs) for the continental and oceanic mantle, at varying pressures (1-3 73 GPa), temperatures (400-600°C) and fluid content (0-100 vol% H2O) along (A) the flat and (B) 74 normal slabs, using the Hacker and Abers (2004) worksheet and listed ultramafic rock compositions. 75 Our seismic tomography velocities are based on the initial velocity model shown in Figure 2 and 76 sampled at locations shown in Figure 8A. We have calculated the predicted rock seismic properties at 77 varying pressures and temperatures, to account for the uncertainty range in our petrological model, 78 shown in Figure 7. Each rock type (Harzburgites, Lherzolites and others) is represented by a 79 geometrical shape and color coded according to its volume proportion of hydrous minerals (also 80 representing its percent hydration). Increasing the rock’s temperature and fluid content, decreases its 81 seismic velocities, and increasing its pressure, increases its seismic velocities. Rocks with the highest 82 seismic velocities are rich in magnesium (Mg) rather than iron (Fe). The continental mantle above the 83 flat slab segment is described with a Mg-rich and dry composition.

84 Figure 11: A comparison of our seismic tomography results when using (A) a slower background 85 velocity model (grey line) at depths of the continental mantle, with respect to the faster one initially 86 used (black line) and shown in Figure 2. The new seismic tomography results obtained using this 87 slower background model are shown along the (B) flat and (C) normal subduction zones, and describe 88 the same velocity trends as those obtained using the faster model in Figure 2, with however only a 89 slight reduction in Vp and Vs of up to 0.1 and 0.3 km/s, respectively, at flat slab depths, sufficient to 90 describe the continental and oceanic mantle with seismic velocities now closer to hydrous conditions, 91 as shown in Figure 12.

92 Figure 12: New comparison between predicted rock seismic behaviors (Vp vs. Vs) using the Hacker 93 and Abers (2004) worksheet for the continental and oceanic mantle and our seismic tomography 94 velocities obtained using a slower background velocity model shown in figure 11A. Numbers refer to 95 the locations where our seismic velocities were sampled, shown in Figure 8A. The previously 96 described as dry continental mantle above the flat slab segment is now described as lying along the 97 borderline between dry and slight hydrated.

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Figure 9

Figure 10

Figure 11

Figure 12