Bulletin of the Seismological Society of America, Vol. 103, No. 3, pp. 2025–2046, June 2013, doi: 10.1785/0120120328

Tearing and Breaking Off of Subducted Slabs as the Result of Collision of the Panama Arc-Indenter with Northwestern South America by Carlos A. Vargas and Paul Mann

Abstract We present two regional, lithospheric cross sections that illustrate east- ward- and southeastward-dipping, subducted slabs to depths of 315 km beneath the surface of in northwestern South America. These cross-sectional interpre- tations are based on relocated earthquake hypocentral solutions, models supported on gravity and magnetic regional data, and coda-Q (Qc) tomography. The method of tomographic imaging based on spatial inversion of the coda wave has advantages of providing information on the lateral variations of the anelastic properties and ther- mal structure of the lithospheric system. Mapping of earthquake-defined Benioff zones combined with tomographic imaging reveals the presence of an ∼240 km long east–west-striking slab tear, named here the Caldas tear. The proposed Caldas tear separates a zone of shallow, 20°–30°-dipping, southeastward subduction in the area of Colombia adjacent to Panama and the Caribbean Sea, which is not associated with subduction-related volcanism, from an area of steeper, 30°–40°-dipping, slab adjacent to the eastern Pacific Ocean that is associated with an active north–south chain of active arc volcanoes. We propose that the Caldas slab tear separating these two distinct subducted slabs originally formed as the southern boundary of the Panama indenter, an extinct island arc that began subducting beneath northwestern South America about 12 Ma. The area south of the Panama indenter is oceanic crust of the Nazca plate, which subducts eastward beneath northwestern South America at normal angles and melts to form a north–south-trending active volcanic arc. In addition to the for- mation of the Caldas tear, we propose that impedance of the thicker crustal area of the Panama arc-indenter over the past 12 Ma may have led to down-dip break-off of previously subducted oceanic crust that is marked by an extremely concentrated and active earthquake swarm of intermediate-depth earthquakes beneath east-central Colombia.

Introduction and Tectonic Setting Hypocentral solutions recorded by the Colombian Na- Regional compilations of Global Positioning Systems tional Seismological Network (CNSN) show an ∼240 km (GPS) data provide a quantitative tectonic framework for long, right-lateral offset of intermediate to deep events with understanding the widespread crustal effects of the Panama azimuth of 102° (Fig. 1a,b). We infer this discontinuity in arc collision on large areas of northwestern South America earthquakes to be a major slab tear which we have named (Calais and Mann, 2009; Fig. 1a). GPS vectors in western the Caldas tear based on the location in the Colombia show a marked decrease in velocities consistent of Colombia and the alignment of -related surface features with the ongoing collision of the Panama arc with north- (e.g., volcanism, faulting, mineral deposits, geothermal anoma- western South America along a north–south-trending suture lies, etc.). Using the distribution of earthquakes >80 km, zone roughly parallel to the international boundary between Ojeda and Havskov (2001) proposed that the discontinuity Panama and Colombia (Adamek et al., 1988; Trenkamp along the Caldas tear represented a boundary between two sub- et al., 2002; Corredor, 2003; Fig. 1a). The east–west direc- ducted slabs with differing dips and strikes: the northern sub- tion of GPS vectors shows that the effects of east–west duction zone, called the Bucaramanga subduction zone, has shortening and indentation related to the collision of the Pan- a shallower dip (27°) and more northeasterly strike, and the ama arc remains relatively constant over a large, V-shaped, southern, called the Cauca subduction zone, has a steeper fault-bounded area of Colombia due east of the Panama arc- dip (35°–40°) and a more northerly strike (Fig. 1a). indenter (Fig. 1b). GPS vectors on the Maracaibo block of

2025 2026 C. A. Vargas and P. Mann

Figure 1. (a) Tectonic map of northwestern South America and Panama showing plate boundaries, neotectonic fault systems, and se- lective distribution of hypocentral solutions of ∼30;000 earthquakes extracted from the entire catalog of the CNSN (∼102;000 events) during – 1993 2012 with these criteria: mL ≥ 0:5; GAP ≤ 200; rms ≤ 0:5; error in latitude ≤10:0 km; error in longitude ≤10:0 km; and error in depth ≤10:0 km. Color scale indicates depth of earthquakes. The north and south profiles symbolize the tomographic sections presented in this study. SMM, Santa Marta massif; CB, Choco block; WC, Western Cordillera; CC, Central Cordillera; EC, Eastern Cordillera; PR, Perija Range; GB, Guajira basin; LB, Llanos foreland basin; MMVB, basin; RFZ, Romeral fault zone; SMBF, Santa Marta– Bucaramanga fault; PF, Palestina fault; CF, Cimitarra fault; MGF, Mulato–Getudo fault; HF, Honda fault; SFS, Salinas fault system; GF, Garrapatas fault; LFS, Llanos fault system; IF, Ibague fault; SR, Sandra ridge; BN, Bucaramanga nest; CN, Cauca nest; MN, Murindo nest; PIVC, Paipa–Iza volcanic complex; RSDV, Romeral and San Diego volcanoes. Yellow stars correspond to (1) the Tauramena earthquake (19 January 1995, Mw 6.5); (2) the Armenia earthquake (25 January 1999, Mw 6.2); and (3) the Quetame earthquake (24 May 2008, mL ˆ 5:7). Sections AA0 and BB0 correspond to tomographic profiles presented in Figures 5 and 6. (b) Crustal isochron pattern of the Sandra ridge; pink- colored line, Caldas tear zone; arrows, station velocity GPS vectors relative to stable South America (after Calais and Mann, 2009). CHEP and BOGO are GPS stations used as reference to estimate the onset of the Panama-arc and collision. Other GPS stations in the Panama-arc collision area are MANZ, RION, BUCM, MONT, and CART. Faded blue arrow enclosing 102° azimuth of the approximately 240 km long, right-lateral offset of intermediate to deep events associated with the Caldas tear.

Colombia and Venezuela show a more northerly direction of 20°–30° in the area north of ∼5:6°N (Ojeda and Havskov, plate motions related to northward tectonic escape of the 2001; Vargas et al., 2007). Maracaibo block into the southern Caribbean (Trenkamp Two nests of concentrated intermediate-depth earth- et al., 2002). In contrast to this fairly uniform GPS velocity quakes are present beneath Colombia (Fig. 1a). The Bucara- field of deformed crustal rocks produced by the Panama col- manga earthquake nest (BN) is found at a depth of ∼160 km lision, underlying, eastward-dipping slabs change abruptly on the down-dip extension of the southern (Bucaramanga) across the Caldas tear from dip angles of 30°–40° between subduction zone and has an estimated volume dimension latitudes ∼3:0°–5:6° N in southern Colombia, to dip angles of of ∼13 × 18 × 12 km (Schneider et al., 1987; Frohlich et al., Tearing and Breaking Off of Subducted Slabs as the Result of Collision of the Panama Arc-Indenter 2027

1995). Previous tectonic interpretations of the origin of the and a high-resolution seismic profile, seeking to define the Bucaramanga nest vary from a zone of two slabs in contact geometry of the Caldas tear and its geotectonic implications (van der Hilst and Mann, 1994), two slabs overlapping in the northwestern corner of South America. (Taboada et al., 2000), or a single slab undergoing extreme bending (Cortés and Angelier, 2005) all occurring in the boundary area of the subducted northern (Bucaramanga) Hypocentral Solutions and Estimation and southern (Cauca) subduction zones (Fig. 1a). The Cauca of the Coda-Wave Attenuation ∼400 intermediate-depth earthquake nest (CN) is located km A catalog has been compiled of ∼102;000 earthquake southwest of the Bucaramanga nest on the trend of our pro- locations calculated by the CNSN during the period 1993– posed Caldas tear and has been previously interpreted 2012 (mL ≤ 6:8). Hypocentral solutions were estimated by by Cortés and Angelier (2005) as a bend in the slab in this using a seismological array of 17 short-period instruments area (Fig. 1a). There is no clear consensus among seismol- (T ˆ 1 s) of the CNSN and complemented by 13 stations ogists for the tectonic interpretation of the two concentrated associated with local volcanic monitoring systems and also Colombian intermediate earthquake nests (Frohlich, 2006; foreign networks (Panama, Ecuador, and Venezuela). Final Zarifi, 2006). solutions were calculated with the HYPOCENTER program The Caldas tear defines the northern limit of the active and the velocity model proposed by Ojeda and Havskov volcanic front of the northern that has formed as a (2001). Then, 9338 waveforms associated with 7645 regional consequence of the steeper subduction of oceanic slab of – earthquakes (3:0 ≤ mL ≤ 6:5; 1993 2012) were selected for normal thickness of the Nazca plate (Fig. 1a). Moreover, estimating the decay rate of the coda amplitudes (Q−1,coda – c associated with active and inactive volcanoes, the east west attenuation). The selected events, on basis of a significant – projected surface trace of the Caldas tear localizes an east number of stations that recorded them (Table 1), have epicen- west alignment of some unusual volcanic rocks including tral distances to stations ranging between 22.6 and 690.0 km adakites (Borrero et al., 2009; Fig. 1a). Other volcanic rocks and depths varying between 0 and 222.0 km. Figure 2 shows in the vicinity of the east–west-trending Caldas tear include −1 all events used for the Qc plotted on a map of northwestern the Plio-Pleistocene Paipa-Iza volcanic complex in the South America along with tectonically significant earth- Eastern Cordillera of Colombia and the Romeral and San quake focal mechanisms including the Quindio, Quetama, Diego volcanoes (Pardo et al., 2005). The presence of these and Tauramena events aligned along the Caldas tear and in- – east west aligned volcanic rocks along with locally elevated termediate-depth focal mechanisms from the Bucaramanga geothermal gradient values (Vargas et al., 2009) suggests that and Cauca nests. the Caldas tear may penetrate the upper crust as a fault zone −1 Estimations of the Qc were done using the Single and provide a conduit for the upward rise of magmas and Backscattering model proposed by Aki and Chouet (1975). hydrothermal fluids produced by melting of the slabs on ei- This model assumes that the coda of a local earthquake is ther side of the Caldas tear (Fig. 2). Furthermore, recent, composed of the sum of secondary S waves produced by shallow-focus, strong motion events such as the Tauramena heterogeneities distributed randomly and uniformly within 25 10 earthquake (19 January 1995; Mw 6.5, h ˆ  km), the lithosphere. The coda is the portion of a seismogram cor- −1 the Quindio earthquake (25 January 1999; Mw 6.2, responding to back-scattered S-waves. The estimation of Q 18 6 c h ˆ : km), and the Quetame earthquake (24 May 2008; used the following equation: mL 5.7; h ˆ superficial) are all in alignment with the surface trace of the Caldas tear. 2 −ωt Previous tomographic studies using both local and 2g θ†jS ω†j Qc P ω;t†ˆ e ; (1) regional earthquakes of varying resolution have produced βt2 differing tectonic interpretations for slabs in this area (van der Hilst and Mann, 1994; Taboada et al., 2000; Vargas where P ω;t† is the time-dependent coda power spectrum, ω et al., 2007). In this paper, we present the results of an in- is the angular frequency, β is the shear-wave velocity, jS ω†j is tegrated geophysics and geologic study that improves the 3D the source spectrum, and g θ† represents the directional scat- imaging of the interactions between the eastward-moving tering coefficient. The g θ† term has been defined as 4π times Panama indenter and its collisional area in northwestern the fractional loss of energy by scattering, per unit travel dis- South America. tance of primary waves, and per unit solid angle of the radi- ation direction θ measured from the direction of primary wave Data and Methods propagation. Using these assumptions, the geometrical spread- ing is assumed to be proportional to r−1, which only applies The following sections describe data and procedures to body waves in a uniform medium. The source factor can be used to estimate hypocentral solutions, the attenuation and its treated as a constant value for single frequency. According to −1 spatial distribution, the simultaneous 2D inversion of gravity equation (1), Qc values can be obtained as the slope of the 2 and magnetic data, and the correlation of these results with least-squares fit of Ln t × P‰ω;tŠ† versus ωt,fort>tβ,where focal mechanisms, geothermal gradients, geological maps, tβ represents the S-wave travel time (Haskov et al., 1989). The 2028 C. A. Vargas and P. Mann

−1 Figure 2. Epicenter projection of events used during the coda-wave-attenuation (Qc ) estimation. Colored circles, earthquakes; blue squares, locations of all seismological stations used in this paper; gray stars are shown with large focal mechanisms, and the most recent and surficial strong-motion events occurring along the Caldas tear are shown by banded-gray polygon. The main focal mechanisms reported by the NEIC-USGS (mb ≥ 4:0) defining the Bucaramanga nest to the northeast and the Cauca nest to the southwest are shown; pink areas identified in the epicentral location of these nests are two main geothermal gradient anomalies reported by Vargas et al. (2009).

time-dependent coda power spectrum was calculated using the mates were performed with short-period records (T ˆ 1 s) at mean squared amplitudes of the coda Aobs ω;t† from band- several frequencies (Table 1). Then we chose estimates in the pass-filtered seismograms around a center frequency. frequency band 1–3(2  1) Hz because of the high availabil- In order to take into account the deep structure using ity and geographical distribution of observations regarding coda waves, ⠈ 4:64 km=s was assumed and calculated as other frequencies; and also best values of correlation coeffi- a weighted average of S-wave speeds in the whole earth vol- cients, the root mean square (rms), and signal-to-noise ratio. ume covered by the scattered waves (Badi et al., 2009). All In addition, it has been reported that the study region presents −1 records were filtered in a chosen frequency band and then Qc values in this frequency band with errors ≤5% (e.g., see −1 used a coda-wave time window (W) of 20 s, starting from Vargas et al. (2004)). In general, errors seen along Qc 2 × tβ s tstart†. The average lapse time, defined as tc ˆ tstart‡ estimations are acceptable, for example, the rms of all esti- W=2 ranges between 11.0 and 384.0 s. These large tc values mations vary between 0.07 and 1.79 (μ ˆ 0:24, σ ˆ 0:07) ensure the sampling of regional structures. Attenuation esti- and the coefficients of correlation are oscillating between Tearing and Breaking Off of Subducted Slabs as the Result of Collision of the Panama Arc-Indenter 2029

Table 1 −1 Estimated Values of Coda-Wave Attenuation (Qc ) at Various Frequencies

−1 3 Qc × 10 Coefficient of Correlation rms Signal/Noise Waveforms Frequency Analyzed Min Mean±Std Max Min Mean±Std Max Min Mean±Std Max Min Mean±Std Max 2 9338 1.7 7.1±2.7 47.6 −0.97 −0.67±0.11 −0.5 0.07 0.24±0.07 1.79 2.0 16.2±47.5 985.5 8 5421 0.8 1.8±0.7 20.4 −0.96 −0.60±0.08 −0.5 0.18 0.32±0.06 2.67 2.0 11.0±25.6 842.6 12 4441 0.6 1.2±0.4 1.2 −0.94 −0.59±0.08 −0.5 0.15 0.35±0.06 1.89 2.0 9.7±21.1 600.9 16 3741 0.5 0.9±0.3 9.4 −0.95 −0.58±0.07 −0.5 0.17 0.35±0.09 4.13 2.0 9.0±18.1 315.7

−1 Also presented are extreme values, averages, standard deviations, and quality parameters. Qc values at 2 Hz were used to estimate tomograms because of the high availability of observations regarding other frequencies, best values of correlation coefficients, rms, and signal-to-noise ratio. A power law equation −1 −1 3 − 0:970:06† for all Qc observations suggested a high-frequency dependence of the attenuation in this region: Qc f†ˆ 13:2  0:6† × 10 f .

−0:5 and −0:97 (μ ˆ −0:67, σ ˆ 0:11). Table 2 presents a duction processes, are a likely site of large contrasts in statistical summary of the main parameters related with the anelastic attenuation in the subducted lithospheric slabs. −1 −1 estimation of Qc values for 30 seismological stations. Given the ease to estimate Qc , we can use this obser- Figure 3a shows an example of typical waveform used dur- vation for highlighting regional structures related to contrasts ing this analysis, as well the corresponding record filtered for in rigidity (e.g., crust or lithospheric plates). One way to −1 −1 the chosen frequency band. The attenuation factor (Qc )is regionalize Qc is based on the work of Malin (1978) who, suggested as a decay factor for the coda-wave amplitudes. expanding on the work of Aki (1969) and Aki and Chouet −1 Figure 3b presents histograms for Qc values and their cor- (1975), realized that the first-order scatterers responsible for relation coefficients, as well as distributions for the epicentral the generation of coda waves at any given tc can be located distances, focal depths, and local magnitudes of the events on the surface of an ellipsoid with earthquake and station analyzed. Figures 2 and 3b emphasize the presence of attenu- locations as foci (Singh and Herrmann, 1983). In the ellip- ation contrasts in the region and at least two sources of soidal volume sampled by coda waves at any time t, Pulli events, one of them surficial and dispersed, and the other (1984) defined the large semi-axis as a1 ˆ βt=2, and defined 2 2 1=2 located at an intermediate depth (linked to the nests of Buca- the small semi-axis as a2 ˆ a3 ˆ a1 − r =4† , where r is ramanga and Cauca). the source–receiver distance of the ellipsoid. The horizontal projection of this volume is coincident with the elliptical en- Tomographic Imaging Using Coda-Wave Attenuation velope proposed by several authors as the area occupied by the scattered energy of the coda-wave record (Mitchell et al., Mukhopadhyay and Sharma (2010) have proposed that −1 1997; Mitchell and Cong, 1998; Xie, 2002; Vargas et al., the variation of Qc with tc shows a direct relationship with −1 2004). Following these observations and knowing the values depth. These authors interpreted that Qc values related to of t , W, and β, it is possible to deduce the volumes of the 200 c scattering processes that penetrate > km depth are con- ellipsoidal shells where the seismic energy is scattered. trolled by a crust and a relatively more transparent mantle. −1 −1 Hence Qc values estimated with large tc correspond to large These results support the idea that Qc estimated with a large sampled volumes, and vice versa. Based on these hypotheses tc is representative of a large sampled volume and large we can perform a generalized inversion for regionalizing sampled depths. A corollary of this hypothesis is that the −1 −1 −1 Qc . For the purpose of the inversion, we define a geo- Qc value must be near to the intrinsic absorption (Qi ) con- graphic grid around the seismic station that also encloses the trolled mainly by the mantle. Following these ideas, Vargas −1 hypocenter. We recognize that each measured Qc is an aver- et al. (2004) developed a regional tomographic study using −1 −1 age estimate Qav (or Qapparent) for the volume as sampled by stations of the CNSN with relative large tc, (up to 180 s) and −1 −1 the ellipsoidal shell given by found that the Qc values are near to the Qi values for almost all stations, meaning that a large portion of the upper VTOTAL VBlock-j mantle is being sampled. Other studies have suggested a ˆ ; (2) Q Q direct relation between the thermal field and anelastic attenu- av Xj j ation (Faul and Jackson, 2005; Priestley and McKenzie, 2006; Yang et al., 2007). The physical meaning of this rela- where VBlock-j is the fraction of volume (block) sampled by −1 tionship is not been completely understood, but Karato and the ellipsoidal shell with the true attenuation coefficient Qj −1 Jung (1998) proposed that the higher water content in the (or Qtrue). Assuming a constant S-wave velocity of propaga- asthenosphere significantly reduces the seismic-wave veloc- tion, the volume traveled by a ray that leaves the hypocenter ities through anelastic relaxation and increasing temperature. moves outward to the ellipsoidal shell as defined by the Convergent margins such as Colombia, which involve large observation time of the coda and is scattered to the receiver, amounts of sediments and water mobilized during the sub- can be determined. Equation (2) can be written as 2030 C. A. Vargas and P. Mann

(a)(a)

(b)(b) σ µ=-0.67 σ=0.11 µ=155.0 σ=96.0 4000 1600 4500 3500 1400 4000 3000 1200 3500 3000 2500 1000 2500 2000 800 2000 Frequency 1500 Frequency 600 Frequency 1500 1000 400 1000 500 200 500 0 0 0 0 10 20 30 40 50 -1 -0.9 -0.8 -0.7 -0.6 -0.5 0 100 200 300 400 500 600 700 800 900 Coefficient of correlation Epicentral distance µ=87.7 σ=65.6 µ=3.1 σ=0.7 4000 3500 3500 3000 3000 2500 2500 2000 2000 Frequency Frequency 1500 1500 1000 1000 500 500 0 0 0 50 100 150 200 250 0 1 2 3 4 5 6 7 Depth (km)

Figure 3. (a) Example of waveform used for estimating the Qc values. Upper trace represents the original record of an earthquake recorded by a short-period seismological station of the CNSN. Middle trace represents the filtered record in frequency band 1–3 (2  1) Hz. Lower trace represents the decay envelope of the coda wave in a window of 20 s, starting from 2 × tβ s. Qc value was obtained 2 as the slope of the least-squares fit of Ln‰t × P ω;t†Š versus ωt (dashed line with arrowheads), for t>tβ, where tβ represents the S-wave −1 travel time (Haskov et al., 1989). (b) Histograms for Qc values and their correlation coefficients, as well as distributions for the epicentral distances, focal depths, and local magnitudes of all events analyzed.

1 1 V 1 1 V 1 where ˆ Block- ‡‡ Block- ‡ Q Q1 V Q1 V av TOTAL TOTAL 1 1 V 1 Block-i VBlock-n y ˆ xi ˆ ai ˆ : ‡ ; (3) Qav Qi VTOTAL Qn VTOTAL Then, a least-squares estimation of the xi is given by the where the ratio VBlock-j=VTOTAL is the volume fraction compact matrix equation AX ˆ Y where A is a (k × n) associated with the total scattered-wave travel path spent coefficient matrix, X is a (n × 1) vector, Y is a (k × 1) vector, in the jth block. If the process is repeated for each station– and k is the number of station–hypocenter pairs. A linear hypocentral pair, the entire region is sampled. Equation (3) is inversion of the matrix equation was formulated as an iter- of the form atively damped least-squares technique (Levenberg, 1944; Marquardt, 1963). The damping factor (σ), which adds a1x1 ‡‡aixi ‡‡anxn ˆ y; (4) to the diagonal parameters of the matrix, was computed ern n raigOfo udce lb steRsl fCliino h aaaArc-Indenter Panama the of Collision of Result the as Slabs Subducted of Off Breaking and Tearing Table 2 Seismological Stations of the CNSN Used in this Study

Q−1 × 103 t Coefficients of Correlation Epicentral Distances Longitude Latitude Altitude Waveforms c c Station (°) (°) (masl) Analyzed Min Mean±Std Max Min Mean±Std Max Min Mean±Std Max Min Mean±Std Max ANIL −75.40 4.49 2300 248 2.9 6.8±1.6 17.2 12.1 89.3 ± 51.5 231 −0.94 −0.67±0.11 −0.50 26.7 166.3±90.2 413.6 BAR −73.18 6.58 1864 754 1.7 6.1±2.5 32.3 11.8 71.2±19.2 227 −0.94 0.67±0.11 −0.50 26.6 142.0±33.7 414.6 BCIP −79.84 9.17 61 5 5.5 6.9±2.0 12.2 104.7 126.4±15.7 146 −0.86 −0.69±0.10 −0.59 200.7 238.7±27.5 272.8 BET −75.44 2.68 540 66 2.9 6.6±3.0 14.9 16.7 68.1±56.3 246 −0.91 −0.68±0.11 −0.50 46.7 136.8±98.5 448.0 BRI −72.79 7.72 1427 46 2.9 6.0±3.5 47.6 13.1 81.6±36.7 150 −0.97 −0.66±0.13 −0.50 24.7 159.6±65.8 280.2 CHI −73.73 4.63 3140 724 2.7 6.8±3.0 20.0 11 89.4±53.3 263 −0.97 −0.69±0.11 −0.50 24.5 173.5±93.8 477.6 CLIM −77.89 0.94 4232 17 2.5 7.0±5.0 12.0 31.7 78.9±35.5 132 −0.91 −0.69±0.11 −0.53 73.0 155.5±62.1 247.6 COD −73.44 9.94 108 104 3.4 7.3±3.1 18.2 19.3 47.3±48.7 180 −0.95 −0.70±0.12 −0.50 33.8 100.2±85.3 332.2 CPAS −77.25 1.22 2620 8 5.1 8.8±4.6 15.9 16.6 14.1±10.2 38.4 −0.91 −0.72±0.12 −0.56 29.1 42.2±17.8 84.7 CRU −76.95 1.57 2761 330 2.5 6.7±2.9 16.1 17.6 79.2±60.2 339 −0.95 −0.69±0.11 −0.50 30.8 156.0±105.3 610.4 CTAB −74.2 5.01 3500 2 5.6 8.1±8.2 14.5 132.4 133±0.85 134 −0.86 −0.70±0.22 −0.55 249.2 250.2±1.48 251.3 CTAU −74.04 5.20 3868 5 4.3 8.0±6.7 28.6 52.8 96.2±31.2 119.1 −0.9 −0.73±0.12 −0.6 109.9 185.9±54.5 225.9 CUM −77.83 0.94 3420 234 3.0 7.0±3.2 22.7 12.9 82.7±57.9 384 −0.97 −0.69±0.12 −0.50 22.6 162.3±101.3 690.0 GCAL −77.42 1.21 2353 8 7.9 10.3±1.8 13.5 14.6 22.8±21.12 69.4 −0.91 −0.83±0.05 −0.74 25.6 57.4±37.0 139.0 GCUF −77.35 1.23 3800 56 3.7 7.4±2.7 18.2 22.0 52.3±40.1 152 −0.93 −0.70±0.11 −0.50 38.5 108.1±70.5 283.5 GUA −72.63 2.54 217 11 3.6 6.2±2.9 10.2 20.3 174.4±85.5 227 −0.86 −0.67±0.11 −0.51 53.0 322.7±149.7 415.3 HEL −75.53 6.19 2815 216 3.0 6.1±2.1 19.6 21.0 110.2±40.9 190 −0.96 −0.65±0.10 −0.50 36.8 210.0±71.5 350.2 MAL −77.34 4.01 75 391 2.4 6.2±0.0 15.9 14.9 65.1±37.5 323 −0.95 −0.67±0.11 −0.50 43.6 131.4±65.7 582.4 MARA −75.95 2.84 2207 78 3.0 5.9±2.4 18.2 17.9 87.6±65.2 259 −0.92 −0.65±0.11 −0.51 31.3 170.8±114.1 469.9 NOR −74.87 5.57 536 596 2.5 5.9±2.1 20.4 16.3 113.4±35.5 297 −0.94 −0.66±0.11 −0.50 28.5 215.8±62.1 537.4 OCA −73.32 8.24 1264 3304 2.0 6.0±2.3 21.7 16.4 102.8±23.1 287 −0.96 −0.66±0.11 −0.50 28.7 197.4±40.4 520.3 OTAV −78.45 0.24 3492 26 4.3 7.6±2.1 13.9 23.6 70.2±25.7 147 −0.88 −0.73±0.12 −0.53 58.8 140.3±44.9 274.1 PCON −76.4 2.33 4294 120 2.0 6.6±3.2 16.9 12.5 72.4±52.8 289 −0.96 −0.67±0.11 −0.50 39.4 144.1±92.4 522.6 PRA −74.89 3.71 468 313 2.5 6.1±2.7 19.6 17.6 101.1±61.3 259 −0.95 −0.67±0.11 −0.50 30.8 194.5±107.4 470.1 ROSC −74.33 4.86 3020 149 2.9 5.9±2.4 33.3 17.9 88.2±47.6 204 −0.95 −0.66±0.12 −0.50 31.3 171.8±83.3 373.6 RREF −75.35 4.9 4743 205 2.5 6.1±2.3 12.7 113.7 118.1±48.8 220 −0.92 −0.65±0.10 −0.50 41.5 224.1±85.4 402.7 RUS −73.08 5.89 3697 150 2.6 6.0±2.7 20.4 17.5 72.6±39.9 240 −0.94 −0.66±0.11 −0.50 30.6 144.5±69.7 437.9 SDV −70.63 8.88 1620 64 2.8 5.8±2.0 12.5 114.5 176.3±23.7 262 −0.90 −0.65±0.11 −0.50 217.9 326.0±41.4 476.5 SOL −77.41 6.23 38 850 3.1 7.3±3.1 25.6 17.2 64.6±39.8 305 −0.97 −0.71±0.11 −0.50 30.1 130.6±69.6 551.1 TOL −75.32 4.59 2577 258 2.5 6.3±2.6 20.0 19.8 100.4±56.1 248 −0.95 −0.67±0.11 −0.50 34.7 193.1±98.2 451.9

−1 The stations detected 7645 earthquakes (1993–2012) that were used for estimating 9338 Qc values in frequency band 1–3(2  1) Hz and coda-wave time window (W)of20s.tc values and relatively large epicentral distances allowed us to estimate the coda-wave tomography regionally. 2031 2032 C. A. Vargas and P. Mann automatically for each iteration (Hoerl and Kennard, 1970; For the 3D inverse problem, we estimated the fraction −1 Hoerl et al., 1975). According to this technique, the solution of volumes associated with each Qav in order to establish and resolution matrixes can be found for the following equation (3). Using all foci related to the events selected in equations: this study, we assembled the compact matrix of equation (4) −1 T 2 −1 T and then we inverted the Qav values using equation (5). X ˆ‰A A ‡ σ IŠ A Y; (5) −1 Finally, a spatial interpolation of the Qav values was done and based on the Kriging method (Oliver and Webster, 1990) and presented on Mercator projection. Figure 4d shows T 2 −1 T R ˆ‰A A ‡ σ IŠ A A: (6) the results of the inversion for the synthetic experiment based on two domains of contrasting attenuation (Fig. 4b). This −1 Similar procedures for the Qc imaging have been used experiment is comparable to a slab-tearing model for which in previous works in order to establish a deterministic char- two zones with different angles of subduction, are related to acterization of the heterogeneity in the lithosphere as an different attenuations. This hypothetical model linked a flat alternative technique for traditional tomographic measure- subduction zone in the north (lower attenuation) and a nor- ments (O’Doherty et al., 1997; Lacruz et al., 2009). mal subduction zone in the south (higher attenuation). In general, the available data may allow detection of large struc- Resolution and Reliability tures with significant contrasts of attenuation as much as ∼270 km depth. On the other hand, Figure 4e presents 32 32 8 A spatial inversion of attenuation of × × the inversion for a chessboard based on two areas of contrast- ∼60 blocks with block dimensions of km latitude†× ing attenuation (Fig. 4c). This experiment suggests that ∼50 ∼40 km longitude† × km thickness† was designed in the available data may allow detection of smaller bodies order to detect relevant structures in the region. We qualified (e.g., 100 km × 100 km × 60 km) with significant contrasts the tomographic inversion by means of three approaches: of attenuation, mainly in Colombia, and as much as ∼180 km (a) hit count of ellipsoidal shells; (b) solving controlled tests; depth. and (c) mapping the diagonal elements of the resolution ma- After several trials of accurate resolution and sta- trix (RDE) by using equation (6). The hit count is a very bility, the spatial inversion of attenuation with real data rough quality estimation that only tells about summing up was performed with the same grid (32 × 32 × 8 blocks). the number of ellipsoids that contribute to the solution of Figure 4f,g shows results of the tomographic estimation a block. Based on this discretized volume, we mapped the and their maps of the RDE at different depths. Because each hit count with the available data in eight layers (0, 45, 90, RDE shows the amount of independence in the solution of 135, 180, 225, 270, and 315 km; Fig. 4a). Although a large one model parameter (RDE oscillates between 0 and 1), the part of northwestern South America (including Colombia, larger value of the RDE for one model parameter suggests a western Venezuela, eastern Panama, and northern Ecuador) is covered by ellipsoidal shells (over 500 crossings), it is in more independent solution for this parameter. 3D inversion presents higher RDE values (e.g., >0:4) limited by the avail- northern Colombia and northwestern Venezuela (71° W to −1 76° W; 5° N to 10° N; 0–180 km depth) where the largest ability and geographical concentration of Qc values, indi- number of shells run through the blocks (based on more than cating that the method is useful for areas for which a large 5000 hits per block). This approach emphasizes the impor- stacking of attenuation observations is present. From the tance of the Bucaramanga nest data in the estimation of to- available earthquake data, the tomographic solution of the mographic images. attenuation efficiently images large areas of the crust and To incorporate the second approach, we evaluated the upper mantle of northwestern South America including efficiency of the method described above solving the 3D di- Colombia, eastern Panama, and western Venezuela with rect problem. The ellipsoidal shell volume associated with all sampling depths reaching >315 km (Fig. 4f). Because the foci (pairs of earthquake and station) were used to relate the ellipsoids related to deeper hypocentral solutions can sample −1 profounder volumes, the 3D inversion may detect the thermal Qc values in two controlled test boards. The first one was for appraising large domains of attenuation, for example, a influx from the mantle adequately. zone with flat subduction in the north, and the other zone In order to infer the geometry of the Caldas tear and its with normal subduction in the south (Fig. 4b). The second relationships with the adjacent Nazca and Caribbean plates, test board evaluated the ability of the method to detect small we made two regional cross sections: (1) a northern sec- anomalies by use of a typical chessboard (Fig. 4c). In tests, tion (AA0, Fig. 1a) from the to the Llanos −1 we assigned two values of Qtrue that represent attenuation foreland basin of eastern Colombia and crossing the inter- contrasts (1=70 and 1=200) into the 3D grid. Then we esti- mediate-depth earthquakes of the Bucaramanga nest; (2) a −1 −1 mated the Qav values (or theoretical Qc values) for all el- southern section designed for imaging the corridor between lipsoids (each one related to foci [earthquake–station]) by the Nazca plate and the Llanos basin and crossing the inter- −1 0 estimating the weighted average of Qtrue involved in the vol- mediate-depth earthquakes of the Cauca nest (BB , Fig. 1a). ume of each ellipsoid. As discussed subsequently, it is essential to incorporate all Tearing and Breaking Off of Subducted Slabs as the Result of Collision of the Panama Arc-Indenter 2033 available geophysical data for proper interpretation of the longitude ≤10:0 km; error in depth ≤5:0 km) were plotted on tomograms along these sections. the tomographic profiles along two 60 km wide corridors (Figs. 5 and 6). Because seismicity in a corridor parallel to the northern section is sparse, we have included an interpreta- Integrating Earthquake Data with Regional tion of the Trans-Andean megaregional seismic-reflection line Seismic-Reflection Lines that extends from the Caribbean coast to the Eastern Cordil- Hypocentral solutions of the CNSN (rms < 0:3 s; lera of Colombia (Vargas et al., 2010) and to the northern to- GAP < 200; stations ≥6; error in latitude ≤10:0 km; error in mographic section. This 383 km long reflection line is a 20 s

Figure 4. Resolution, reliability, and results of the spatial inversion of attenuation based on a geometry of 32 × 32 × 8 blocks with ∼ ∼ ∼ −1 dimensions of 60 km latitude† × 50 km longitude† × 40 km thickness†. Coda-wave tomograms were estimated with 9338 Qc ob- – – servations associated with 7645 regional earthquakes (mL ≤ 6:7; 1993 2012) in the frequency band 1 3(2  1) Hz. (a) Hit count of ellip- −1 soidal shells, suggesting that the 3D inversion of Qc may solve large areas of Colombia, eastern Panama, western Venezuela, and northern Ecuador. (b) Synthetic model that represents two large domains of attenuation (e.g., a zone with flat subduction in the north and normal subduction in the south, limited by a slab tear). The contrasts of attenuation incorporated into the model to evaluate the effectiveness of the −1 −1 method were Qc ˆ 1=200 and Qc ˆ 1=70. (c) Chessboard with smaller and regular distribution of attenuation contrasts. As in the previous −1 −1 case, the contrasts of attenuation incorporated into the model were Qc ˆ 1=200 and Qc ˆ 1=70. (d) 3D inversion of the synthetic model presented in (b) suggesting that the distribution of the available data may allow detection of large structures with significant contrasts of attenuation as much as ∼270 km depth. (e) 3D inversion of the chessboard model presented in (c) suggesting that the distribution of the available data may permit detection of smaller bodies (e.g., ∼100 km × 100 km × 60 km) with significant contrasts of attenuation, mainly in Colombia, and as much as ∼180 km depth. (f) Results of the tomographic inversion with the available data. (g) Maps of the RDE at different depths. Higher RDE values (e.g., ≥0:5) indicate zones efficiently solved. However, these higher RDE values were limited by the geographical −1 concentration of Qc values, indicating that the method is useful for areas where a large stacking of attenuation observations is present. Tomographic solution of the attenuation efficiently images large areas of the crust and upper mantle of northwestern South America including Colombia, eastern Panama, and western Venezuela with sampling depths reaching >315 km. High-attenuation anomalies suggest that Buca- ramanga and Cauca seismic nests may be related to asthenospheric emplacements. (Continued) 2034 C. A. Vargas and P. Mann

Figure 4. Continued.