Seismicity and lithospheric structure in active subduction, relic subduction and intracontinental settings
A THESIS SUBMITTED TO THE FACULTY OF THE GRADUATE SCHOOL OF THE UNIVERSITY OF MINNESOTA BY
Meng Sun
IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY
Maximiliano J. Bezada
September, 2020 © Meng Sun 2020
ALL RIGHTS RESERVED Acknowledgements
I owe many thanks to my advisor, Max Bezada. In the past five years, Max has always been willing to help me with my questions, and provide inspiration and guidance. Especially in the last year of my Ph.D life, Max sacrificed lots of his time to discuss projects with me, and provided very insightful opinions and feedbacks during every meeting. I truly appreciate all the time Max invested in me to help me shape a good understanding of scientific problems and a cautious attitude towards research. I would also like to thank my committee members: Christian Teyssier, Justin Revenaugh, and Ikuko Wada, for their guidance and advice through my entire journey pursuing science at University of Minnesota. I have benefitted from the cooperative atmosphere in the lab. A special thanks to Joseph Byrnes, Joe has always been willing to offer suggestions and encouragements when I found myself stuck at somewhere. I also appreciate the five years’ companion from Hwaju Lee, without each other, the endless Minnesotan winter might have already defeated us. I received a lot of help from fellow graduate student Zhao Zhu and postdoc Aaron Hirsch. I also need to thank my collaborators in one of my research projects: German A. Prieto, Alan Levander, and John Cornthwaite. Last but not least, thanks to my family for the unconditional support in all aspects.
i Abstract Earthquakes occur in a wide range of depths and different tectonic settings. A comprehensive understanding of them can be difficult, this needs acknowledgement from various perspectives, e.g. source properties of earthquakes, such as locations, focal mechanisms, rupture processes; thermal structure, and morphology of seismogenic lithosphere(s); laboratory rock deformation experiments. In this dissertation, three seismogenic regions within distinct tectonic settings are selected, this includes the relic subduction zone in the westernmost Mediterranean, the active northwestern South America subduction zone, and the Reelfoot Rift within the continental interior. In the westernmost Mediterranean, we provide a high-resolution seismic pattern at depths of ~100km using grid-searching and double-difference techniques. A systematic analysis of the clustering seismic pattern indicates this seismicity is associated with active necking and breakoff of the Alboran slab, which is progressing from north to south. For the northwestern South America subduction zone, we provide a well-resolved teleseismic P-wave tomography model and investigate the configuration of this subduction zone utilizing teleseismic tomography. Results reveal NE-SW trending Caribbean plate and Nazca plate, with the subduction of Nazca plate overlapping with CAR between 5°~8°N. This leads to a reinterpretation of the nature of some notable features in the region, namely the Caldas tear and the Bucaramanga nest. In the Reelfoot Rift region, we analyse P-wave attenuation from records of teleseismic events, and provide insights into possible rheology properties of the upper mantle beneath the Reelfoot Rift. Results show relatively low attenuation within the Reelfoot Rift region, indicating the Reelfoot Rift might not be a weak zone. Meanwhile, an association between the attenuation and occurrence of seismicity is revealed, this conveys that the lateral variation of attenuation along the strike of the Reelfoot Rift region might be responsible for the NMSZ. Exploration of seismic patterns and lithosphere structures across these three different regions contributes to our understanding of the occurrence of earthquakes under different tectonic settings. ii Content Table
Acknowledgements ...... i Abstract ......
List of Figures ...... ⅴ
List of Tables ...... ⅵ
1. Introduction ...... 1 1.1 Background ...... 1
1.2 Outline of work presented here ...... 3
2. Seismogenic necking during slab detachment, evidence from relocation of
intermediate depth seismicity in the Alboran slab ...... 5 2.1 Introduction ...... 6 2.2. Data and Method ...... 9 2.2.1 Events recorded by the PICASSO network ...... 9 2.2.2 Events recorded by the Spanish local network ...... 10 2.3. Results ...... 12 2.3.1 Absolute locations of events from PICASSO network ...... 12 2.3.1.1 Relocation results ...... 12 2.3.1.2 Event locations relative to slab structure ...... 13 2.3.1.3 Cross correlation of waveforms ...... 15 2.3.2 DD relocation results of Spanish catalog event ...... 16 2.3.2.1 Seismicity pattern ...... 16 2.3.2.2 Synthetic tests of DD relocation results ...... 21 2.3.2.3 Location of seismicity with respect to the slab ...... 28 2.3.3 Depth-related magnitude-frequency distributions ...... 29 2.4. Discussion ...... 30 iii 2.4.1 Heterogeneous thinning and detachment stages along the strike of Alboran slab. . .30 2.4.2 Mechanism(s) triggering intermediate-depth seismicity ...... 33 2.4.3 Tectonic implications ...... 35 2.5. Conclusion ...... 36
3. Overlapping slabs: untangling subduction in NW South America through finite-frequency teleseismic tomography ...... 39 3.1. Introduction ...... 39 3.2 Background ...... 40 3.3 Data and method ...... 44 3.3.1 Finite-frequency tomography ...... 46 3.3.2 Damping and smoothing regularization ...... 47 3.4 Model ...... 49 3.5. Testing possible slab configurations ...... 54 3.5.1 Architecture of synthetic input models ...... 55 3.5.2 Synthetic results ...... 56 3.6. Discussion ...... 61 3.7. Conclusion ...... 66 4. Is the New Madrid Seismic Zone caused by a weak zone in the upper mantle? Insights from attenuation investigation ...... 68 4.1 Introduction ...... 68 4.2 Data and method ...... 72 4.2.1 Data ...... 72 4.2.2 Inversion scheme ...... 76 4.3 Result ...... 78 4.3.1 Event back azimuth effects...... 78 4.3.2 Network effects ...... 82 4.4. Discussion ...... 85 4.5. Conclusion ...... 87
5. Conclusion ...... 88 Reference ...... 91
Appendix ...... 110
iv List of Figures
Fig. 2.1: Distribution of events and stations ...... 11 Fig. 2.2: Locations changes of 58 PICASSO events before and after relocation ...... 13 Fig. 2.3: Relocated events relative to the slab ...... 14 Fig. 2.4: Seismograms of a pair of repeating earthquakes ...... 15 Fig. 2.5: Distribution of 908 earthquakes before and after double difference relocation...... 18 Fig. 2.6: Relocation results using a subset of events with good station coverage ...... 19 Fig. 2.7: Relocation results and error estimates of the finger-shaped subset using SVD ...... 21 Fig. 2.8: Results after perturbing the catalog locations using a standard deviation of 10km . . . . 23 Fig. 2.9: Results of relocating events before and after perturbing recorded arrival times using standard deviations of 0.6/1.2s and 1.2/2.4s for P and S arrivals, respectively ...... 23 Fig. 2.10: Results using synthetic arrival times of the initial event locations as input ...... 25 Fig. 2.11: Results using synthetic arrival times from events with prescribed scattered locations..26
Fig. 2.12: Results using synthetic arrival times from 3 points with station-event link ...... 28 Fig. 2.13: Depth distribution of M>= 3.5 events (blue) and M<4.0 events (red) ...... 30 Fig. 2.14: Schematic sketch of the necking model of Alboran slab at intermediate depth ...... 32 Fig. 3.1: Study region: northwestern South America ...... 44 Fig. 3.2: L-curve used to tune hyperparameters ...... 46 Fig. 3.3: Tomography result, map view ...... 52 Fig. 3.4: Tomographic cross sections ...... 53 Fig. 3.5: Secondary features to the west of Bucaramanga segment ...... 53 Fig. 3.6: Three sets of synthetic input models ...... 59 Fig. 3.7: Inverted results of synthetic models ...... 60 Fig. 3.8: Hit quality, with slab contours of 2%, 2.5%, 3% anomalies denoted ...... 61 Fig. 4.1. Study region: the Reelfoot Rift ...... 71 Fig. 4.2: Events used in this study ...... 72 v Fig. 4.3: Example comparing between observed and synthetic traces of a same event ...... 75
Fig. 4.4: Distribution of Δt* after inversion using events from different back azimuth ...... 80
Fig. 4.5:. Δt * and earthquake counts along line 1 and line 2 (see Fig. 4.4)...... 81 Fig. 4.6: Δt* results using linear inversion ...... 83 Fig. 4.7: Results using TA, XO, ZL network and dataset from different back azimuth ...... 84 Fig. S1. A zoom in of the distribution of events and some stations ...... 110
Fig. S2. Comparison of relocation results using gridsearch and 3D velocity model vs. hypoDD and 1D velocity model ...... 111
Fig. S3. Relocation results using different velocity models ...... 111
Fig. S4: Station coverage of the event subset ...... 112 Fig. S5: Comparison of relocation results of the finger-shaped subset by using LQSR and SVD respectively ...... 113
Fig. S6: Comparison between relocation errors estimated by SVD for the finger- shaped subset, and the absolute value of the relocation difference using SVD and LSQR for each event in this subset...... 113 Fig. S7: Hypocenters before and after relocation after perturbing catalog hypocenters by standard deviation values of 5km and 20km ...... 113
List of Tables
Table 2.1: PICASSO earthquakes before and after relocation ...... 115
vi Chapter 1
Introduction
1.1 Background
Earthquakes are the expression of rock deformation and the subsequent energy release under different temperatures and pressures. Although they primarily occur along plate boundaries, many occur in the plate interiors and within subducted plates. The occurrence of shallow-depth earthquakes can be explained by the brittle failure of rocks, whereas occurrence of intermediate-depth earthquakes can be complicated to understand. At such depths, being subjected to high pressure and temperature, the deformation of mantle rocks occurs by creep instead of brittle fracture [Kohlstedt et al., 1995]. The stress required to induce brittle fracture cannot be accumulated under these conditions. Nevertheless, approximately 25% of global earthquakes happen at depths greater than 50km [Frohlich, 2006]. Two main hypotheses have been proposed to explain intermediate-depth seismicity: dehydration embrittlement [Hacker et al., 2003; Jung et al., 2004; Jung et al., 2009; Rutter et al., 2009; Okazaki and Hirth, 2016] and adiabatic shear instability [Bercovici, David and Karato, 2002; Green and Marone, 2002]. Consequently, a comprehensive understanding of the mechanisms triggering the occurrence of seismicity can be difficult, this needs acknowledgement of source properties of earthquakes, morphology, thermal structure, and physical properties of the lithosphere etc..
In this dissertation, taking advantage of different seismological tools (e.g. relocation of earthquakes, P wave teleseismic tomography, attenuation estimation), we explore the occurrence of seismicity in three different regions within distinct tectonic settings, and
1 provide insights into their in-situ deformation and long-term tectonic processes. The three regions include: the Alboran slab in the westernmost Mediterranean, the NW South America subduction zone, and the Reelfoot Rift.
Although most intermediate-depth seismicity is associated with active subductions, e.g. the NW of the South America subduction zone in chapter 3, exceptions exist and the Alboran slab is a good example. The Alboran slab is a stagnant relic slab with tens to hundreds of intermediate-depth seismic events occurring every year. Here, we relocate earthquakes that happened between 1997-2017 using grid-searching and double difference methods, and provide a high-resolution seismicity pattern at a depth of ~100km. A careful analysis of the seismicity pattern reveals the heterogeneous thinning and detachment stages along the strike of the Alboran slab.
The NW South America subduction zone is an active subduction zone. It is the conjunction spot of three plates, with the South American plate converging or colliding with both the Caribbean, and Nazca plates. The Bucaramanga nest, one of the world’s most active seismic concentrations within a fairly small volume [Prieto et al., 2012], appears to be located within this region. Debates exist regarding whether it’s of intraslab or interslab origin among various studies, as the most fundamental question regarding the lithosphere morphology and the origin of each segment are still not well-resolved. Here, we approach this problem by investigating the configuration of subduction beneath the northern Andes using teleseismic tomography, providing a well-resolved teleseismic P-wave tomography model, as a necessary guide to understand the tectonic process at play. Our results reveal NE-SW trending Caribbean and Nazca plates, with the subduction of Nazca plate overlapping with CAR between 5°~8°N. This leads us to a reinterpretation of the nature of some notable features in the region, namely the Caldas tear and the Bucaramanga nest.
2 The Reelfoot Rift resides within Mississippi embayment and constitutes the Mississippi Valley Graben [Thomas, 1991; Johnson et al., 1994], is infamous for hosting several M6.8-8 intraplate earthquakes during 1811-1812 and remaining to be seismically active presently [Johnston and Schweig, 1996; Cramer, 2001; Hough and Page, 2011]. Tomography research proposed a low velocity zone in the upper mantle beneath the NMSZ, which was widely interpreted as a weak zone localizing stress. In this dissertation, we analyse P-wave attenuation from records of teleseismic events and provide insights into possible rheology properties of the upper mantle beneath the Reelfoot Rift. Results show relatively low attenuation within the Reelfoot Rift region, indicating the Reelfoot Rift might not be a weak zone. Meanwhile, a consistency between the attenuation gradients and occurrence of seismicity is revealed, this conveys that the lateral variation of attenuation gradient might be accounting for the NMSZ.
1.2 Outline of work presented here
Chapter 2 is a paper that was published in 2020 on Journal of Geophysical Research: Solid Earth with co-author Maximiliano Bezada. This chapter presents the relocation of intermediate-depth clusters within the Alboran slab in the western Mediterranean. Both absolute hypocenters and seismic patterns relative to the subducting slab are revealed using grid-searching and double-difference methods. Well-resolved seismic patterns provide insights into the necking detachment process at depths of ~100km in the relic Alboran slab, which could not be directly constrained by tomography models due to its weak resolvability at this depth.
Chapter 3 is modified from a paper that will be submitted to Earth and Planetary Science Letters, co-authored with Maximiliano J. Bezada, John Cornthwaite, German A. Prieto,
3 Alan Levander, FengLin Niu. The purpose of this chapter is to untangle the subduction zone in western Colombia using teleseismic P wave tomography, providing insights of the tectonic process in the northern Andes and the possible mechanism accounting for the occurrence of the Bucaramanga nest.
Chapter 4 is a study of seismicity in the Reelfoot Rift. In this chapter, a 2D attenuation effect is estimated within the region to investigate whether seismicity in the Reelfoot rift is caused by a weak zone in the upper mantle localizing stress.
Chapter 5 presents a summary of results and a prospect of future directions to pursue based on the results of this work.
4 Chapter 2
Seismogenic necking during slab detachment, evidence from relocation of intermediate depth seismicity in the Alboran slab
An edited version of this paper was published by AGU. Copyright (2020) American Geophysical Union, and is modified and used in this dissertation with permission.
With few exceptions, intermediate-depth seismicity is associated with active subduction. In the Alboran slab that is located just east of the Gibraltar Strait in the westernmost Mediterranean, although evidence suggests subduction is no longer active, tens to hundreds of M<5.0 earthquakes are observed every year at depths of ~80 km. In this paper, we relocate 58 such events recorded by the PICASSO temporary network using a 3D velocity model and implementing a grid-search approach to minimize the normalized misfit between observed and predicted P and S-wave arrival times. Meanwhile we relocate 908 events recorded by the Spanish national network using the double-difference method and give a high-precision picture of seismicity distribution within the slab. Relocation results reveal five clusters in this region separated by small gaps. Jointly considering relocation results and a 3D tomography model indicates one cluster is at a shallower depth above the northern, E-W oriented arm of the Alboran slab, while the other four are aligned from north to south, parallel to the strike of the slab and near its core. Larger events are generally shallower and more scattered, while the deeper, more clustered events tend to have smaller magnitudes. Clusters further to the north show concentrated seismicity on narrow zones that dip ~45° to the south.
5 We suggest this seismicity is associated with active necking and breakoff of the Alboran slab, which is progressing from north to south. Shear instability is likely to be the failure-enabling mechanism for the occurrence of these earthquakes.
2.1 Introduction
Controversies exist on the mechanism(s) accounting for the occurrence of intermediate-depth (50 - 300km) seismicity. At such depths, being subjected to high pressure and temperature, the deformation of mantle rocks occurs by creep instead of brittle fracture [Kohlstedt et al., 1995]. The stress required to induce brittle fracture cannot be accumulated under these conditions. Nevertheless, approximately 25% of global earthquakes happen at depths greater than 50km [Frohlich, 2006]. Two main hypotheses have been proposed to explain intermediate-depth seismicity: dehydration embrittlement [Hacker et al., 2003; Jung et al., 2004; Jung et al., 2009; Okazaki and Hirth, 2016; Rutter et al., 2009] and adiabatic shear instability [Bercovici, David and Karato, 2002; Green and Marone, 2002]. In some cases, elevated concentrations of intermediate-depth seismicity have been associated with lateral tears in subducting slabs [e.g. Clark et al., 2008; Meighan et al., 2013; Sachpazi et al., 2016], although in other cases lateral tears are inferred, yet no concentrations of seismicity are observed [Bezada et al., 2010; Pearce et al., 2012]. Furthermore, the absence of intermediate-depth seismicity has also been proposed as evidence of tearing [Hu and Liu, 2016]. Thus, the relationship between intermediate-depth seismicity and slab deformation is not entirely clear. Enhanced intermediate-depth seismicity rates have also been proposed to be associated with the process of slab detachment and break-off [Martin et al., 2006; Kufner et al., 2017]. The hypothesis of slab detachment after the cessation of subduction was initially proposed to explain magmatism in collisional orogens [Davies and von Blanckenburg, 1995; von Blackenburg and Davies, 1995]. More recent modeling studies have confirmed the feasibility of this process [Boutelier and Cruden, 2017; Duretz et al., 2012], and
6 seismic imaging has been used to propose active slab detachment of stalled slabs in the Mediterranean region [Wortel and Spakman, 2000]. Most conceptual models for slab detachment involve necking of the slab [Boutelier and Cruden, 2017; Duretz et al., 2012; Gerya et al., 2004; Van Hunen and Allen, 2011], which usually progresses from one end of the slab to the other [Boutelier and Cruden, 2017; Van Hunen and Allen, 2011]. This process would likely lead to the concentration of seismicity at the propagating tip of the detachment tear [e.g. Wortel and Spakman, 2000]. The seismicity of the Vrancea nest in Romania has been interpreted in this way [Martin et al., 2006]; yet, in the well-studied case of slab detachment in the Hindu Kush, several clusters of seismicity were observed, rather than one. There, progressive change in the clusters’ characteristics were observed; from very active at the final stages of detachment, to less active where detachment was incipient, inboard of the propagating tip of the tear [Kufner et al., 2017].
In this paper, we examine the intermediate-depth seismicity in the Alboran slab of the westernmost Mediterranean. The most widely accepted hypothesis for the origin of the Alboran slab is initiation of subduction of Alpine Tethys lithosphere near the Gulf of Lyon at ~30 Ma followed by slab fragmentation and rollback. The Alboran slab would represent the westernmost fragment of this subduction which would have rolled back independently toward the west until its arrival at its current position near the Strait of Gibraltar [Faccenna et al., 2004; Lonergan and White, 1997; Royden, 1993; Spakman and Wortel, 2004].
Recent tomography results [Bezada et al., 2013, 2014; Bonnin et al., 2014; Monna et al., 2013; Villaseñor et al., 2015] revealed a vertically continuous slab-shaped high-velocity anomaly hanging beneath the Strait of Gibraltar reaching depths in excess of 600 km, further supporting the hypothesis of slab rollback in the westernmost Mediterranean. Considering this westward rollback evolution and the morphology of the slab, the western edge of the slab represents its pre-subduction bottom, whereas the eastern edge
7 represents its pre-subduction top (Fig. 2.3). At depths of ~100 km, the slab can be described as consisting of a northern, E-W oriented segment, which we will refer to as the northern arm, and a N-S oriented segment which we will refer to as the main arm (Fig. 2.3).
Debates exist as to whether the Alboran slab is still actively retreating [Duarte et al., 2013; Gutscher et al., 2002; Gutscher et al., 2009; Gutscher et al., 2012; Lonergan and White, 1997; Duggen et al., 2004; Duggen et al., 2005]. Nevertheless, some intermediate-depth seismicity with magnitudes varying between ml 0.6 and ml 5 has been consistently observed during the past few decades. Catalog locations from the Spanish local network [González, 2017] indicate these earthquakes occur at depths shallower than 160 km, and concentrate along a narrow N-S trending band at around 4.5°W. Studies on focal mechanisms of several intermediate-depth earthquakes occurring between 2002~2009 show a varied focal mechanism pattern at different locations [Pondrelli et al., 2002; Buforn et al., 2004; Stich et al., 2006; Buforn et al., 2016] and thus indicate spatial variations in the stress pattern within the slab. To the west of 4.5°W, the stress field is dominated by vertical tension; to the east of 4.5°W, the dominant stress changes to vertical compression [Buforn et al., 2016]. Ongoing detachment of the Alboran slab has been proposed based on surface wave tomography [Palomeras et al., 2014] and receiver function imaging [Mancilla et al. 2013; Heit et al., 2017]. Some authors have proposed a relationship between the intermediate-depth seismicity and an ongoing detachment in the Alboran slab, suggesting the earthquakes delineate a vertical tear through a subhorizontal segment of the slab [Heit et al., 2017].
In this paper, by taking advantage of high-quality records from a dense temporary deployment in Spain and Morocco between 2010~2013, a rich dataset of over 1000 events recorded by the Spanish national network between 1997~2017, and a 3D high-resolution seismic velocity model of the Alboran slab [Bezada et al., 2013;
8 Palomeras et al., 2014] we give a high-precision picture of the seismicity distribution within the Alboran slab. We constrain the absolute locations of 58 events and relative locations among 908 events within/above the slab. Based on the relocation results, we propose a model where the intermediate-depth events are associated with necking and break-off of the slab.
2.2. Data and Method
2.2.1 Events recorded by the PICASSO network
Earthquakes occurring in the westernmost Mediterranean between 2010 and 2013 were recorded with good signal-to-noise ratio by ~70 broadband 3-component stations of the PICASSO network covering a very wide range of back azimuths (Fig. 2.1). We selected 58 events reported by the Instituto Geográfico Nacional (IGN) catalog that occurred during the PICASSO deployment at intermediate depths in the Alboran region, and that were well recorded by at least 14 stations, with a minimum of 10 and 8 records of P and S wave respectively. The selected events have magnitudes varying between ml2.5 and ml4.2, they span depths between 50 and 112 km and appear to concentrate into several clusters (Fig. 2.1). After event-specific bandpass filtering, we visually picked P wave arrival times from vertical components and S wave arrival times from horizontal components. Uncertainties of the picked arrivals were estimated by the length of time between the earliest and latest plausible arrival times, as defined by the analyst. The optimal location for each event was determined using a grid-search through 3D velocity model. Using a 3D model allows us to incorporate travel time effects of the highly laterally heterogeneous velocity structure in this area. We used the P-wave tomography model of Bezada et al. [2014] which was generated by inversion of teleseismic travel time delays, with additional constraints on the shallow (< 80 km) velocity structure coming from the surface wave tomography
9 model of Palomeras et al (2014). An S-wave velocity model is generated from the P- wave model using a constant Vp/Vs ratio of 1.8. Travel times for S and P waves from each station to each point in the 3D velocity model were calculated using a graph-theory based method [Moser 1991, Toomey et al. 1994]. We then found the point that minimized the overall data misfit between observed and calculated P and S wave arrival times, with each travel-time misfit normalized by the estimated picking uncertainty. Details of the methodology are explained in Singh et al. [2016].
2.2.2 Events recorded by the Spanish local network
The PICASSO network provides dense, high-quality data, and allows us to determine absolute earthquake locations. However, given its relatively short deployment period (2010-2013), only a limited amount of events was recorded. Any apparent patterns in the distribution of seismicity stemming from the analysis of these 58 events may not be conclusive given their relatively small number, and the large spatial extent of the study area. For this reason, we retrieved hypocentral parameters and travel-time picks from 1221 intermediate-depth events occurring between 1997 and 2017 from the Spanish national catalog [González, 2017] and utilized the double-difference (DD) earthquake location algorithm hypoDD [Waldhauser and Ellsworth, 2000] to refine their relative locations and obtain more robust constraints on the seismicity distribution within the slab. The DD technique in general, and hypoDD in particular are well-known and widely used methodologies for finding the relative locations of nearby events [Waldhauser and Ellsworth, 2000]. Information reported by the Spanish network includes P and S wave arrival times, origin times and hypocenter locations for all events [González, 2017]. Catalog parameters were used as inputs to the hypoDD relocation program with weights of 1.0 and 0.7 for P and S wave arrival time respectively.
10 Taking into account the event-station distances (Fig. 2.1) and the scale of velocity heterogeneity of the Alboran slab [Bezada et al., 2013], during the first 5 iterations of DD relocation, we searched for event pairs with inter-event distance smaller than 40 km and a minimum of 15 linked observations (including both P and S wave arrivals). During following iterations, we set the distance and residual time cutoff to 15km and 0.5s respectively. A total of 10 iterations were conducted, since variations of locations and time residuals fell into noise levels for subsequent iterations.
Fig. 2.1: Distribution of events and stations. Orange dots are 58 events recorded by the PICASSO network (black triangles, stations mentioned in Fig. 2.4 are outlined using red lines) between 2010 and 2013; blue dots are 1221 events recorded by the Spanish national network (white triangles) between 1997 and 2017. Locations of events shown on this map are all from catalog hypocenters. Topography/bathymetry is shown as shaded relief.
11 2.3. Results
2.3.1 Absolute locations of events from PICASSO network
2.3.1.1 Relocation results
Relocation results of the 58 events recorded by the PICASSO network indicate that hypocenters are more concentrated in space than it would appear from catalog locations, suggesting that events are grouped into distinct clusters. Events tend to shift very slightly southward (average change is -0.0719° along latitude) and eastward (0.0251° along longitude). Although individual events’ depths shift by up to tens of km, the distribution has no clear peaks; therefore no systematic shift in depth is observed (Fig. 2.2). Hypocenters after relocation are distributed within a narrow depth range (80-120 km) and longitude range (-5.28°W to -3.98°W) (Fig. 2.2). Relocated hypocenters are more consistent with the observed travel times. After relocation, the RMS travel-time misfit of P arrivals decreases from 1.4156s to 1.0331s, while for S waves the value decreases from 2.5354s to 1.8436s. The remaining misfit might be due to uncertainties in the velocity model including velocity heterogeneity smaller than model resolution, anisotropy, and spatial variations in the Vp/Vs ratio.
12 Fig. 2.2: Changes in hypocentral parameters of 58 PICASSO events before and after relocation. (a) map view of hypocenter density before relocation; (b) mapview of hypocenter density after relocation; (c)-(e) histograms showing the distribution of change in hypocenters along latitude, longitude and depth respectively. Contours in (a) and (b) show fractional velocity anomalies at 90km depth from the tomography model of Bezada et al., 2014.
2.3.1.2 Event locations relative to slab structure
Considering the relocated hypocenters in the context of the 3D P-wave tomography model of the Alboran slab [Bezada et al., 2014] shows that the seismogenic region is parallel to the strike of the slab. Most events are concentrated in what would have been the middle depths of the lithosphere before subduction, with a few scattered at the pre-subduction bottom (lower lithosphere, western edge) while no seismicity is observed in the pre-subduction top (crust and uppermost mantle lithosphere, eastern edge) of the
13 slab (Fig. 2.3). This distribution of seismicity is not consistent with the general observations in most subduction zones, where intermediate-depth earthquakes often occur within the subducted crust and uppermost mantle and commonly form double seismic zones [Florez and Prieto, 2019; Rietbrock et al., 2004; Brudzinski et al., 2007; Yamasaki et al., 2003] with a second seismic band lying in the subducted lower lithospheric mantle. Events do not seem to be homogeneously distributed along strike within the Alboran slab. There appear to be regions of higher rates of seismicity separated by gaps (Fig. 2.2, 2.3). The error surfaces are determined as isomisfit surfaces that enclose volumes where the misfit is 110% of the minimum normalized misfit for each event, they suggest that the apparent clusters and gaps should be resolvable, i.e. the length scales of clusters and gaps are much larger than the relocation uncertainties (Fig. 2.3).
Fig. 2.3: Locations of relocated events (PICASSO network: black dots; Spanish network: red dots) relative to the slab. Slab surface is defined as the 1.5% iso-anomaly surface from a 3D tomography model (blue
14 volume in (a) & (b)), with uncertainties of PICASSO relocation results outlined by green bubbles. (a). View of the 3D model from the top (b). View of the 3D model from the west (c). Cross section along 36°N (AA’) in Fig. 2.3a, earthquakes within 25km north or south of 36°N are projected.
2.3.1.3 Cross correlation of waveforms
In order to investigate whether there are repeating events within the clusters, we estimated the similarities between same-station records of different events by cross-correlating 60-second windows starting by the picked P wave onsets. In our study, we define as “repeating events” event pairs having cross-correlation coefficients larger than 0.9 at a minimum of 4 stations. Results revealed one pair of repeating events with waveform cross-correlation coefficients larger than 0.95 at 10 common stations (Fig. 2.4). Magnitudes, origin times and hypocentral parameters for these two events are shown in Table 2.1 (occurrence time: 2010-11-22 07:17:14.600; 2011-01-18 04:40:18.200) They are separated by 2 months in time and 6km in space, both lying within cluster B (Fig. 2.5e) and at the interface of the main arm of Alboran slab.
Fig. 2.4: Seismograms of a pair of repeating earthquakes. The plot shows vertical component traces recorded at 10 common stations with cross-correlation values larger than 0.95. Station names are shown on
15 the left of each trace pair, stations are outlined using red lines in Fig. 2.1 and labelled in Fig.S1; values below station names are the cross correlation coefficient of the corresponding trace pair. Each trace is normalized by the maximum amplitude. Red traces are records of the M4 event occurring on 01/18/2011, blue traces are records of the M3 event occurring on 11/22/2010.
2.3.2 DD relocation results of Spanish catalog event
In this section, we show DD relocation results of 1221 events reported by the Spanish national network that confirm the event clustering suggested by the relocation of the 58 PICASSO events, and provide a more detailed picture of the seismicity patterns.
2.3.2.1 Seismicity pattern
80.45% of the 1221 events retrieved from the Spanish catalog were relocated using hypoDD. The mean change of hypocenters is 3.9 km along longitude, 4.1 km along latitude and 5.5 km along depth; after relocation, the RMS time residual decreases by ~30% from 690 ms to 473 ms.
In order to test the impact of different velocity models on relocation results, we repeat the relocation using different velocity models, including a 1D velocity model simplified from the 3D tomography result [Bezada et al., 2013] and the AK135 1D velocity model [Kennett, 1995], while keeping other parameters consistent. No significant difference was observed in the relocation results (Fig. S3).
Those 58 events recorded by PICASSO network were also recorded by the Spanish local network during this time period. 44/58 of them were relocated together with other events recorded by the Spanish network using hypoDD. A comparison of the results show great consistency of event absolute locations determined by different methods and velocity models (Fig. S2), thus in all further discussions, we refer to relocations obtained using the 16 AK135 1D velocity model [Kennett, 1995] for convenience in producing consistent synthetic travel times for test cases in a later section.
DD relocation results confirm event clustering and reveal the presence of five distinct clusters (A-E) in this region (Fig. 5a-5d): cluster A is located to the east of -4.4°W at depths of ~60 km; the other four clusters (B, C, D, E) are deeper, located to the west of -4.4°W and form a N-S trend, dipping slightly eastward. Focusing on cluster B at around 36.6°N, we note a difference in the spatial distribution of events above and below 75 km depth. Above 75km, events are more scattered while those at deeper depths concentrate into a finger-like shape dipping to the southeast (Fig. 5e). During the 20 years between 1997 and 2017, 109 out of 908 relocated Alboran intermediate-depth events were concentrated within this small cylindrical volume with a seismicity density of ~0.03
3 events/km , which is about 5 to 10 times that of the other clusters in this region. Cluster C is located at around 36.4°N and is separated by a few scattered events from Cluster B to the north. This cluster also shows a concentration of events that suggests a linear feature, although the event density is much smaller, and the pattern much less clear than in Cluster B (Fig. 5f). Cluster D is located at around 36.2°N, and is separated from cluster C by a ~20km seismic gap. About 20 km further south, and separated from Cluster D by another seismic gap, we find cluster E. This cluster appears to be more scattered and at a deeper depth relative to the other four, however synthetic tests described in a later section suggest this apparent spatial distribution is an artifact of the event-receiver geometry.
17 Fig. 2.5: Distribution of 908 earthquakes (red dots) before (a, c) and after (b, d) hypoDD relocation, with relocation results 58 PICASSO events also shown (blue dots on b, d). (a) Map view of events distribution before DD relocation. (b) Map view of event distribution after DD relocation. (c) View from the east of event distribution before relocation; (d) View from the east of event distribution after relocation. (e). Zoom-in of cluster B dipping at 42° from vertical (views from the east and south), red circles are relocated events, black squares denote the repeating event pair in Fig. 2.4; (f). Zoom-in of cluster C dipping at 53° from vertical (views from the east and south).
18 Stations involved in our DD relocation are not ideally distributed, with few stations at azimuths between 90° and 270° (Fig. 2.1). During DD relocation, events with poor azimuthal coverage have a negative effect on the relocation results of events that have better station coverage due to the linkage between them [Waldhauser and Ellsworth, 2000]. In order to evaluate this effect, we relocate the subset containing 134 events beneath the Spanish continental lithosphere with latitudes greater than 36.7°N and longitudes greater than -4.3°W, as this subset has better station coverage compared with the rest of the events (Fig. S4). By relocating events in the same subset alone and together with all other catalog events, we found the difference along depth and latitude is negligible with the averaged difference being 0.4 km and -0.3 km respectively. Along longitude, all these events shifted eastwards with an average change of 1.8 km, so the relative pattern between them remains unchanged (Fig. 2.6). We conclude that the less-than ideal azimuthal coverage of the data set taken as a whole does not negatively affect our results.
Fig. 2.6: Relocation results using a subset of events with especially good station coverage alone (red diamond) and together with other events (blue dots), bars connect the same event. (a). Map view (b). Side view.
19 Given the large number of events, we solved the inverse relocation problem using LSQR rather than SVD. Errors estimated by LQSR are usually underestimated [Waldhauser and Ellsworth, 2000], therefore we estimated the errors of our DD relocation results by relocating a subset using SVD. The subset chosen in our study is the finger-shaped concentration in the deeper part of cluster B that contains 102 events and is located around 36.6°N (Fig. 2.5e). For the SVD inversion, we used parameters consistent with those of the LQSR relocation except only 4 iterations are conducted instead of 10.
After 4 iterations using the SVD method, 93% of 102 connected events in the subset were relocated with a 0.29s RMS time residual. The relocated subset using SVD still displays a finger-shaped distribution, with RMS errors of 1.1 km and 1.9 km in the horizontal and vertical directions respectively (Fig. 2.7, Fig.S5). The relocation error estimated using the SVD method is larger than the absolute difference between SVD and LSQR relocation results (Fig.S6), and amount to only one-tenth of the vertical extent of the finger (Fig. 2.7), indicating that the vertical elongation is required to fit the differential travel times and the finger-shaped feature is strongly supported by the data. Beyond this particular feature, the results of this test show that uncertainties in relocation results are much smaller than the seismic gaps we observe, and that the occurrence of seismicity in clusters separated by gaps is a robust result.
20 Fig. 2.7: Relocation results and error estimates of the finger-shaped subset using SVD. Red crosses outline the uncertainty of the relocation results
2.3.2.2 Synthetic tests of DD relocation results
While the relocation errors estimated in the previous section are small, an important question is whether errors in the catalog arrival times and initial hypocenters can affect our results in ways not captured by this test. Considering the small scale of seismic gaps (Fig. 2.5) and sub-optimal station coverage, this is critically important to understand the robustness of our results. To address this issue, we carried out a series of synthetic tests to evaluate whether the overall clustering situation and the particular geometry of each cluster are robust.
21 2.3.2.2.1 Tests with perturbed hypocenters and/or arrival times
To evaluate the sensitivity of our results to the recorded arrival times and/or initial hypocenters, we introduce a series of Gaussian distributed perturbations with different plausible standard deviation (σ) values to either the reported hypocenters, the arrival times, or to both at the same time. We use σ of 5km (see supplement, Fig. S7), 10km (Fig. 2.8) and 20km (see supplement, Fig. S7) to test initial hypocenter effects, and σ of 0.3s/0.6s, 0.6s/1.2s and 1.2s/2.4s to test the effect of P/S wave arrival time uncertainties (Fig. 2.9). The occurrence of clustering and the number of clusters are sustained after relocations with all these perturbations.
Tests show that relocation results are more sensitive to perturbations in arrival times than to perturbations to the initial hypocenters. When introducing normally-distributed noise with a σ of 20km to the recorded hypocenters while leaving the arrival times unperturbed, the finger-shaped geometries of clusters B (Fig. 2.5e) and C (Fig. 2.5f), become ambiguous (Fig.S7); when introducing perturbations to the arrival times with a σ larger than 0.6s/1.2s for P&S waves, these two finger-shaped clusters start to lose their particular well-defined geometries, yet events in the deeper part of these two clusters are still more densely concentrated compared with those in the shallower part (Fig. 2.9).
22 Fig. 2.8: Results of relocating events after perturbing the catalog locations using a standard deviation of 10km. (a). Map view of perturbed data before relocation. (b) Map view of perturbed data after relocation (c) Map view of unperturbed data after relocation. (d) View from the west of perturbed data before relocation. (e) View from the west of perturbed data after relocation. (f) View from the west of unperturbed data after relocation.
Fig. 2.9: Results of relocating events before (c and f) and after perturbing recorded arrival times using standard deviations of 0.6/1.2s (a and d), 1.2/2.4s (b and e) for P and S arrivals, respectively. Panels a-c show Map views of relocated hypocenters. Panels d-f show views from the west of relocated hypocenters.
23 2.3.2.2.2 Tests with synthetic data
Another factor that may artificially collapse earthquakes into clusters is the station links of event pairs selected by hypoDD based on the stations available in the network and the chosen value for relocation parameters. Here we show three tests designed to evaluate the effect of relocation parameters on the results using synthetic data.
(1). For the first test, we assume the catalog locations are correct. We calculate synthetic travel times from these hypocenters and use these synthetic times as input to the DD program while keeping all other parameters the same as in our application to the real data. Under these circumstances the relocation algorithm should not change the location of the events. Any deviation of relocated locations from the initial (correct) ones can shed light on both relative and absolute location uncertainties that are caused by station-event links. Fig. 2.10 shows the comparison between initial and relocated locations. Both the relative and absolute locations of events at latitudes larger than 36.2°N are well relocated with the new locations barely changed from their real ones. Events at latitudes between 36°N and 36.2°N tend to shift to a deeper depth by a similar increment, thus relative locations between them are well resolved. Events at latitudes lower than 36.2°N (the southernmost cluster) systematically shift to deeper depths by different amounts. This artificial deepening trend is more apparent in the south and indicates the deepening trend of clusters’ centroid towards the south in Fig. 2.5 is likely an artifact caused by the station links utilized in our study and the relative locations between events in the cluster E are less well constrained. All these observations are consistent with a comparison between the DD results and the absolute location results. Along a north-south profile, absolute locations of the 58 events recorded by the PICASSO array generally overlap with the DD located catalog events except in the southern end of the line where the DD relocated events are systematically deeper than PICASSO events at the same latitude (Fig. 2.5d).
24 Fig. 2.10: Results of a synthetic test using synthetic arrival times of the initial event locations as input. Blue dots are initial (real) locations, red diamonds are locations after relocation, green bars connect the same event before and after relocation
(2). In order to test whether the relocation procedure is biased towards producing clustered locations, we designed a second synthetic test by introducing a set of σ=20km Gaussian-distributed noise to the initial hypocenters recorded by the Spanish network, and therefore derived 1221 synthetic events that are homogeneously distributed in space, without clustering (Fig. 2.11). We calculated the synthetic travel times of these 1221 scattered synthetic events and used those times as input to the DD algorithm while keeping other parameters such as initial hypocenters, event pair - station linkages, velocity model, distance and time residual cutoff thresholds consistent with our DD relocation of the real data. In principle, results after DD relocation should be able to refute the initial, mildly clustered, catalog hypocenters and reveal the evenly distributed seismicity pattern that is indicated by the synthetic P and S differential times. If relocated events still collapse into clusters, it would imply a bias toward clustering of the event-station link network utilized in our study.
25 After relocation, only 24.4% of these 1221 synthetic events were kept and relocated. The amount of events relocated is constrained by the customized setting of distance and residual time cutoff thresholds. In this test, our setting of the distance (15km) and residual time cutoff (0.5s) thresholds eliminate most of these events since synthetic arrival times of scattered events were adopted as input. The small amount of events relocated indicates a good performance of the system in redefining events’ locations as scattered. From another perspective, events that are kept and relocated by hypoDD show a good agreement with their prescribed locations and do not collapse into clusters, as shown in Fig. 2.11. Based on this, we conclude that the clustering we observe in our results is not an artifact of the relocation procedure.
Fig. 2.11: Results of a test using synthetic arrival times from events with prescribed scattered locations while keeping other parameters including initial hypocenters consistent with the real relocation process. (a),(d). Map view and view from the west of initial locations of earthquakes. (b),(f). Map view and view from the west of prescribed synthetic scattered earthquakes. Grey dots are all prescribed events, green dots are prescribed locations of events that were relocated by DD program. (c),(g). Map view and view from the west of relocated events. Red circles are locations after relocation while green dots are prescribed locations. Black bars show the offset between locations of the same before and after relocation.
26 (3). A complementary test evaluates the procedure’s ability to recover tightly concentrated clusters of events. In this case, instead of using the synthetic travel times of scattered events, we divided events into three groups (Fig. 2.12) and calculated the synthetic travel times from three points, which are roughly the centroids of these three groups (Fig. 2.12), to the stations. We repeat the relocation using this new set of synthetic travel times as input while keeping all other parameters, including the initial hypocenters, consistent with Test 2. In general, results after relocation successfully recover the prescribed locations, which indicates a good sensitivity of the procedure to the concentration(s) of events. In particular, events within the cluster under the northern arm of the Alboran slab collapse exactly into the prescribed overlapped locations; events in the other two clusters generally collapse around the prescribed points but show greater scatter, especially the southernmost cluster. However, this case represents an extreme assumption regarding concentration of seismicity, and not a realistic scenario, so we do not regard scattering in the southernmost cluster as a failure. Absolute location results of 58 events recorded by PICASSO network refute the chance of this extremely concentrated distribution (Fig. 2.3). Based on this, we conclude that DD relocation can robustly resolve the clustering we see, with the caveat that the degree of clustering may be underestimated in the southernmost part of the study area.
27 Fig. 2.12: Results of a test using synthetic arrival times from 3 points (green stars) with station-event links and all other relocation parameters, including initial hypocenters, the same as in the relocation of real data. (a), (c). Map view and view from the west of initial (blue dots) and prescribed (green stars) locations. (b), (d). Map view and view from the west of prescribed (green stars) and relocated (red circles) locations.
2.3.2.3 Location of seismicity with respect to the tomographically defined slab
The error assessments, synthetic tests and constraints from absolute locations of 58 PICASSO events demonstrate that not only the relative locations of DD relocated Spanish network events are well defined, the absolute locations of events at latitudes larger than 36°N (cluster A-D) are also well constrained and thus provide a reliable picture of the seismogenic regions in the Alboran slab. Joint consideration of DD relocation results and a recent 3D tomography model of the region shows that cluster A is located above the Spain-arm of the Alboran slab (Fig. 2.3), this indicates events in cluster
28 A might be caused by the ongoing tearing process of the Alboran slab from Spain’s continental lithosphere due to the gravity and slab drag [Spakman et al., 2018]. The other four clusters (cluster B, C, D, E) are located within or above the main arm of the Alboran slab. They are all within the same pre-subduction depth of the Alboran slab and are aligned parallel to its strike. Events in the shallower part (<75km) of cluster B are scattered above the tomographically defined slab while those at a deeper depth collapse into a finger-shaped concentration within the slab. Cluster D forms a planar feature that is perpendicular to the strike of the Alboran slab. As discussed in 3.2.2.2, uncertainties in the absolute and relative locations of events in cluster E, as well as potential relocation artifacts for this cluster are significant. For these reasons, we will not discuss this cluster any further.
2.3.3 Depth-related magnitude-frequency distributions
Reported magnitudes of events in this region vary between ml0.6 and ml5.0. The magnitude-frequency distribution appears to be non-linear which may indicate spatial variations in the magnitude of completeness (Fig. 2.13). For this reason, we don’t attempt to estimate b-values either for the entire dataset nor for each individual cluster. Nevertheless, taking advantage of the precise relative and absolute relocation results we have obtained, we explore the relative locations between larger and smaller intermediate-depth events in this slab during the past 20 years. We find that the depth distribution of events with M>3.5 is different from that of events with lower magnitudes. Larger events peak at a depth of ~60 km, whereas smaller events peak at a depth of ~90 km (Fig. 2.13). Although we concede the robustness of this result is questionable, it suggests a difference in the processes that are controlling the occurrence of seismicity at different depths in the seismogenic zone.
29 Fig. 2.13: Depth distribution of M>= 3.5 events (blue) and M<4.0 events (red). Note that larger events have a tendency to be shallower.
2.4. Discussion
In previous sections we have documented the heterogeneous distribution of intermediate depth seismicity within the Alboran slab and presented a set of tests that show the robustness of this result. Here, we consider this result in the context of the morphology of the slab as determined by tomographic imaging, the history of the slab, and the most widely accepted mechanisms for enabling intermediate-depth seismicity. Based on these considerations we propose a model of ongoing slab detachment to explain the observations.
2.4.1 Heterogeneous thinning and detachment stages along the strike of Alboran slab
30 Several authors suggest that the Alboran slab is actively detaching due to the buoyancy of the overlying continental crust and the gravitational pull of the hanging oceanic lithosphere [Palomeras et al., 2014; Perouse et al., 2010; Vernant et al., 2010]. Moreover, it has been proposed that the breakoff process has progressed to different degrees at different locations along strike [e.g. Palomeras et al., 2014]. Heit et al. [2017] interpreted intermediate-depth seismicity within the Alboran slab as a manifestation of the tearing process. Yet, in their interpretation, seismicity occurs within a NS-trending planar tear in a sub-horizontal part of the slab. A similar model for the tearing was proposed by Rosell et al. [2011] on the basis of 3-D electrical resistivity images. However, all recent tomography models show a near vertical slab in this depth range [Bezada et al., 2013, 2014; Bonnin et al., 2014; Monna et al., 2013; Villaseñor et al., 2015]. For this reason, we propose a different interpretation that we argue is more consistent with the imaged slab morphology. We propose that in the seismogenic depth range (60-110km) the slab is currently undergoing active necking as part of the break-off process. We speculate that necking may have led to the opening of small-scale gaps laterally along the slab. Therefore, we interpret the clusters of seismicity as zones of high shear strain produced by necking, and the gaps between them as gaps in the subducted lithosphere itself, rather than aseismic sections of the slab, as is shown in Fig. 2.14. A similar scenario was recently proposed in the Andes subduction zone based on geodynamic simulations of slab evolution, where aseismic segments are attributed to gaps in the slab produced by local tearing [Hu and Liu, 2016]. Based on the seismicity distribution, the gaps in the Alboran would have widths on the order of 20 km, and heights of ~40 km which would not be resolvable by teleseismic tomography, even with the recent dense seismic deployments in the area. As a result, the high stress rate accumulated from the slab detachment and necking process is concentrated at the thinned but still continuous parts of the Alboran slab which are separated from each other by ambient mantle, leading to the laterally heterogeneous distribution of regional stress, and triggering earthquakes in these
31 locations. Furthermore, the high concentrations of seismicity along apparent shear zones as they occur at what would be the maximum shear stress orientation produced by the tensional stress caused by the weight of the slab. The fact that this potential shear zone seems to be most highly developed in cluster B, that it is still apparent but less developed in cluster C and absent from clusters D and E further south, suggests that the necking and detachment process is progressing from north to south.
Fig. 2.14: Schematic sketch of the necking model of Alboran slab at intermediate depth. Cluster A-E are annotated.
In the northern part of the slab, whose latitude is larger than 36.6°N (the entire cluster A and shallower part of cluster B), there are scattered events observed above the imaged slab defined by the 1.5% velocity anomaly contour in the Bezada et al. [2016] model (Fig. 2.3). Given the incapability of the convecting mantle to host seismicity, it is reasonable to deduce that at this depth the Spain-arm of the slab is subjected to ongoing delamination and significantly thinned, thus it is not resolvable by tomography imaging. The uplift of the crust in Spain is also consistent with slab detachment in this region [Mancilla et al., 2015; Reinhardt et al, 2007].
In the southernmost part of the Alboran slab beneath the African coast, few earthquakes were observed in our study, and receiver functions suggest that this part of the slab is still
32 attached to the overlying continental crust [Palomeras et al., 2014]. The observation of abundant seismicity within the interior of the detaching part of the slab (cluster B, C, D) and absence/rare occurrences of seismicity within the attached portion of the slab suggests that the detachment process actually leads to the accumulation of stress required to initiate earthquakes. It is also possible however, that this southward reduction in the number of events in the catalog is a result of the decreasing station density of the Spanish national network there.
Our proposed necking model of the Alboran slab at depths between 50-110km is consistent with the absence of earthquakes in what would be the pre-subduction top of the slab. As a result of thinning, this volume would now be occupied by the ambient mantle incapable of hosting earthquakes. However, there are several scattered earthquakes observed near what would be the pre-subduction bottom of the Alboran slab. This seems at odds with the necking model we propose in this paper. However, considering these earthquakes are all at shallower depths relative to the rest of the events and that intermediate depth seismicity has been reported even farther west [Gutscher et al., 2009; Geissler et al., 2010; Monna et al., 2013], these events likely lie in the mantle lithosphere of Atlantic plate.
2.4.2 Mechanism(s) triggering intermediate-depth seismicity
In order to explore whether the seismicity locations we have determined are consistent with dehydration embrittlement, we consider a 3D thermal model of the Alboran slab [Chertova et al., 2018]. We note that the slab morphology in this model does not include the thinning and necking that we propose, and thus the thermal structure is inconsistent with our hypothetical geometry. Nonetheless, we judge it useful to explore this as an alternative scenario. Event locations relative to the 3D thermal [Chertova et al., 2018] and tomography model [Bezada et al., 2016] of the Alboran slab indicates these seismogenic regions are subjected to temperatures of 300~500°C and pressures of 2~3 GPa. Deformation of antigorite is aseismic at this temperature and pressure window, so 33 dehydration embrittlement is not likely to work in an intact slab. However, if the slab is significantly thinned at this depth, as we proposed in this paper, the temperature would be too hot for antigorite to be stable or even dehydrating [Gasc et al, 2017]. In this case, the clustered distribution of seismicity could be interpreted as non-uniform hydration of the slab before subduction. We note a problem with this interpretation: Outer rise faults are likely the main pathways for slab hydration [Hacker et al., 2003]; given the Alboran slab’s history of eastward subduction and westward rollback as well as its current near-vertical orientation [Bezada et al., 2013, 2014; Bonnin et al., 2014; Monna et al., 2013; Villaseñor et al., 2015], outer rise faults preserved in the slab would be oriented NS. Thus one would expect laterally continuous bands of hydrated lithosphere with this orientation. This is inconsistent with our observation of distinct clusters and gaps along strike in the slab. The temperatures for the seismogenic region given above assume a slab of normal thickness; we infer much higher temperatures if the slab has narrowed substantially, as we hypothesize, and additional temperature increases owing to necking-related shear heating should be expected [e.g. Gerya et al., 2004]. If this is the case, temperatures in the seismogenic zones would exceed those at which dehydration embrittlement would be expected to occur.
The mechanism we propose (slab necking) would be more consistent with shear instability and thermal runaway [Bercovici and Karato, 2002; Green and Marone, 2002]. Thus we infer that the necking process localizes shear stress and strain which can reduce grain sizes [Braeck and Podladchikov, 2007; John et al., 2007; Kelemen and Hirth, 2007] and initiate the positive feedback loop that leads to failure by thermal runaway. The observation of zones with high seismicity rates oriented roughly parallel to the expected maximum shear direction in tension (Fig. 2.5e), strongly supports this hypothesis. Additionally, a pair of repeating events with a 2-month interval and 6km shift in space are reported in our study (Fig. 2.4, Fig. 2.5e, Table 1). Shear is thought to lead to the occurrence of repeating earthquakes in the same shear zone [Wiens and Snider, 2001]. Further study of the rupture characteristics of these two events such as rupture geometry
34 and radiation efficiency [Prieto et al., 2013; Poli et al., 2016] will improve our understanding of their enabling mechanism(s).
2.4.3 Tectonic implications
There has been some discussion in the literature in recent years as to whether the subduction of the Alboran slab remains active, with some authors suggesting that the pull of the Alboran slab is initiating incipient subduction of the Atlantic beneath western Europe [Duarte et al., 2013; M. A. Gutscher et al., 2002; M. A. Gutscher et al., 2012; M. A. Gutscher et al., 2009]. Other authors, however, have argued for ongoing detachment of the Alboran slab [Palomeras et al. 2014; Mancilla et al. 2013; Heit et al., 2017]. Detachment of the slab would make it impossible for the weight of the slab to be transferred to the Atlantic oceanic lithosphere, thus eliminating the hypothesized driving force for its incipient subduction. The interpretation of the Alboran intermediate-depth seismicity that we present here strongly supports the detachment hypothesis. We conclude that our results are inconsistent with an intact slab or ongoing subduction. By extension we reject the Alboran slab as possible driving force for subduction initiation in the northeastern Atlantic.
With regards to the dynamics of slab detachment, we note a significant difference between the scenario we propose and current conceptual models for slab breakoff. The prevailing idea is that necking progresses until a tear is formed and this tear most likely propagates from one edge of the slab to the other [Boutelier and Cruden, 2017; Van Hunen and Allen, 2011; Wortel and Spakman, 2000]. While our results are in some ways consistent with this idea, this model predicts a single point of stress concentration at the propagating tip of the tear. What we observe is that there are multiple clusters of seismicity with seismicity rates that increase from the intact slab edge to the actively detaching edge. This is indeed consistent with what has been observed in the well-studied
35 Hindu Kush region, where the seismic expression of the slab-detachment process parallels what we have described here [Kufner et al., 2017].
Nevertheless, an alternative interpretation for these intermediate-depth clusters could also be the subduction of hydrated lineaments observed in the Gulf of Cadiz (Sallares et al., 2011; Gutscher et al., 2012) transporting water into the upper mantle, causing serpentinization and producing zones of rheological weakness, which would then be the first to yield when subjected to high bending stresses in the subducting slab hinge. We do not favor this model for the following two reasons: firstly, we observe a change in the seismicity rate from north to south; it is unclear why the degree of hydration of these lineaments would progressively decrease from north to south along the slab strike; secondly, the distance between hydrated lineaments is on the order of 50 km (Sallares et al., 2010; Gutscher et al., 2012) which is larger than the distance between adjacent clusters in our relocation results.
In addition, focal mechanisms for forty-two intermediate-depth events that were recorded by IberArray and PICASSO network were recently determined [Santos Bueno, et al., - 2019]. Some of these mechanisms show down-dip extension, consistent with expectations if our slab detachment hypothesis is correct. However, the ensemble of focal mechanisms reveals a much more complicated pattern, with stress orientations varying over short distances. Unfortunately, none of 42 events lie within the 45° shear zone in cluster B revealed by hypoDD in our paper, where our hypothesis would predict a specific geometry. More work on focal mechanisms in this region would help test the hypothesis that those earthquake concentrations represent shear zones.
2.5. Conclusion
The Alboran slab in the westernmost Mediterranean is a relic slab that still hosts
36 earthquakes at intermediate depths and provides a rare opportunity to glimpse at the termination stage of subduction. In order to explore the mechanisms accounting for these earthquakes and how they are related to the detachment process, here we use records from the Spanish local network and the temporary PICASSO dense deployment to define earthquake locations in the context of the velocity and thermal structure of the slab. Relocation results revealed distinct seismic clusters within the slab at a depth range of 50-110km. We identify 5 distinct clusters separated by seismic gaps ~20 km wide. The clusters concentrate in the pre-subduction middle levels of the slab, with the orientation of the overall seismogenic plane parallel to the pre-subduction slab surface. There are a few scattered events observed at the pre-subduction bottom surface of Alboran slab, but considering their shallower depth relative to the rest of the events, they may actually lie within the mantle lithosphere of the Atlantic plate. Extensive synthetic testing shows that these results are generally robust, with the constraints being weakest in the southern end of the slab.
The existence of events above the tomographically-resolvable slab surface beneath the Betics and the absence of earthquakes in the pre-subduction top of the slab suggests that, as a result of the detachment process, the Alboran slab is significantly thinned at its shallower depths. We hypothesize the thinning process has advanced to the point of opening small gaps, and that the distribution of seismic clusters and gaps indicates the presence and absence of slab material, respectively. This hypothesis explains why events are concentrated in what would be the pre-subduction middle depths of the slab. The highest concentrations of seismicity occur in what appear to be shear zones oriented at ~45°from the vertical, which corresponds to the maximum shear direction for vertical tension, and is thus consistent with our hypothesis. This, in combination with an inferred temperature that exceeds what would be expected for dehydration reactions suggests that much of the intermediate-depth seismicity in the Alboran slab is enabled by shear instability rather than dehydration embrittlement. Consistently, few events are observed
37 in the southernmost part of Alboran slab, where it is proposed to be still attached to the overlying continental crust by receiver function studies.
Overall, the changes of seismicity rates and clustering concentrations along the strike at intermediate depths of the Alboran slab indicate different stages of development of the apparent shear zones, and thus suggests a detachment process that started in the north and is progressing toward the south. We conclude that the distribution of intermediate depth seismicity provides strong evidence for ongoing slab detachment and is a product of that process. If our interpretation is correct, then as a corollary, this active detachment process precludes the Alboran slab from providing a driving mechanism for initiating Atlantic subduction at the Gulf of Cadiz.
38 Chapter 3
Overlapping slabs: untangling subduction in NW South America through finite-frequency teleseismic tomography
This chapter is modified from a manuscript that will be submitted to Earth and Planetary Science Letters, co-authored with Maximiliano J. Bezada, John Cornthwaite, German A. Prieto, Alan Levander, FengLin Niu.
3.1. Introduction
Although textbook representations of subduction zones depict a pseudo-2D geometry, where along-strike variations are ignored, real subduction zones are much more complicated. Along-strike changes in slab dip, caused by lateral variations in the structure of either the incoming or the overriding plate are not uncommon, and in some instances two different subducting slabs can be in close proximity to each other. Northwestern South America is one such complex setting. There, we find the conjunction of three plates, with the South American plate (SA) converging or colliding with both the Caribbean (CAR), and Nazca (NZ) plates. Matters are further complicated by the subduction of the large, buoyant, Cretaceous Caribbean Large Igneous Province (~ 1.8ｘ1 0⁶ km²) [Burke, 1988; Mora et al., 2017; Kellogg et al., 2019], and the northward migration of the Nazca plate and
39 Nazca-Caribbean-South America triple junction in the past ~70 m.y. [Boschman et al., 2014; Kellogg et al., 2019]. This complex tectonic setting has spawned many conflicting hypothesis for how convergence between the different plates is accommodated by subduction [e.g. Van der Hilst and Mann, 1994; Taboada et al., 2000; Vargas and Mann, 2013; Corte´s and Angelier, 2005].
Although some regional seismic tomography models exist [ van der Hilst and Mann, 1994; Vargas and Mann, 2013; Chiarabba et al., 2015; Syracuse et al., 2016], resolution has been limited by station coverage. Therefore, interpretation of slab(s) structures was still heavily relied on the distribution of the intermediate-depth seismicity in this region, which is scarce to absent deeper than ~200 km and thus does not provide information on the configuration of the deeper slabs. In addition, several recent tomography studies rely on this local seismicity or on surface wave data, leaving the deep structure beneath Colombia unconstrained. In this paper, taking advantage of the Colombian national seismic network and a temporary station deployment, we provide a well-resolved teleseismic P-wave tomography model to investigate the configuration of subduction beneath the northern Andes, as a necessary guide to understand the tectonic process at play. Our results reveal NE-SW trending Caribbean plate and Nazca plate, with the subduction of Nazca plate overlapping with CAR between 5°~8°N. This leads us to a reinterpretation of the nature of some notable features in the region, namely the Caldas tear and the Bucaramanga nest.
3.2 Background
The Northwestern part of SA is underthrust by CAR obliquely at a speed of ~20mm/yr while further to the east the margin transitions to a strike-slip regime [Kellog and Vega, 1995; DeMets et al., 2000; DeMets, 2001; Weber et al., 2001; Trenkamp et al., 2002]. A tear within CAR, is hypothesized to exist and peel the subducted segment off the main 40 body of CAR, accommodating this transition, although its location and geometry are not well constrained [Bezada et al., 2010; Masy et al., 2011; van Benthem and Govers, 2013]. While intermediate depth seismicity is relatively scarce, it has been used for decades to outline the shallow subduction of CAR [e.g. Pennington, 1981; Kellog and Bonini, 1982; Malave and Suarez, 1995; Chiarabba et al., 2016]. The subduction angle is expected to be shallow given that this region of CAR is covered by the buoyant Caribbean Large Igneous Province [e.g. Burke, 1988], which possibly accounts for the absence of arc magmatism associated with this subduction. The morphology and spatial extent of CAR subduction has long been a matter of debate and, although it has become more clear in the last decade, uncertainties about it’s lateral extent persist. van der Hilst and Mann [1994] proposed an incipient subduction whereas Bezada et al., [2010] imaged a ~600km velocity anomaly beneath at latitudes 10°~12°N and therein implied a long-lived subduction. A slab reaching the transition zone was confirmed by Van Benthem et al. [2013] and imaged in detail by Cornthwaite et al. [2020]. Several authors have placed the southern edge of CAR at around ~10°N [van der Hilst and Mann, 1994; Corredor, 2003; Syracuse et al., 2016], whereas others have hypothesized that it extends as far south as ~5°N [Taboada et al., 2000; Kellogg et al., 2019]. Part of the difficulty in discerning the southern boundary of the Caribbean subduction is that although the Pacific coast of Colombia is clearly part of the Nazca-South America boundary today, before collision and accretion of the Panama- Choco block it was the Caribbean that subducted beneath western Colombia [Montes et al., 2012]. In fact, palinspastic reconstructions show a history of ~70 Ma of Caribbean subduction beneath western Colombia, starting in the Late Cretaceous and extending as far south as ~2.3°N in current coordinates. Northeastward motion of the Caribbean relative to South America led to the northward migration of the southern end of this boundary, and the eventual collision of the Choco Block terminated subduction diachronously from south to north [Kellogg and Mora-Páez, 2016] from 15 to 1 Ma.
41 Conversely, Nazca subduction below western Colombia has a diachronous onset from south to north as evidenced by the temporal evolution of volcanism [Wagner et al., 2017; Kellogg et al., 2019], which reaches as far north as 3.6°N by 20 Ma [see Fig. 7 in Kellogg et al., 2019], before eventually shutting down north of ~5.5°N 5 Ma [Wagner et al., 2017; Kellogg et al., 2019]. This leads to the present Nazca-South America plate boundary configuration in the Pacific coast of Colombia. In this modern framework, it is sensible to interpret Wadati-Benioff seismicity, volcanism and any velocity anomalies in the mantle as being related to Nazca subduction. For example, the lateral offset in intermediate depth seismicity at ~5.5°N [Ojeda and Havskov, 2001] is most readily interpreted as a tear in the Nazca plate (the Caldas tear, Vargas and Mann, 2013; Syracuse et al., 2016). This feature coincides with the northern termination of the modern volcanic arc system in south America [Pennington, 1981; Wagner et al., 2017] and is widely thought to separate a northern flat-subducting segment of Nazca (the Bucaramanga segment) from a normally subducting (Cauca) segment to the south (see Fig. 3.1).
Between 5°~10°N, the Bucaramanga segment extends sub-horizontally to the east for 400km before transitioning into normal subduction [Chiarabba et al., 2016; Syracuse et al., 2016]. The Bucaramanga nest, one of the world’s most active seismic concentrations within a fairly small volume [Prieto et al., 2012], appears to be located within this segment. In spite of much research on their focal mechanisms and rupture processes [Poli et al., 2016], the most fundamental questions as to which slab(s) hosts it, whether it’s of intraslab or interslab origin are still under debates among various studies. The nest has been variously attributed to the collision between subducted CAR and Nazca plate [van der Hilst and Mann, 1994], inflexion of the subducted paleo-Caribbean plate [Taboada et al., 2000], a margin-parallel slab tear developed within Caribbean plate [Cortes and Angelier, 2005; Vargas and Mann, 2013]. Another dispute is about the location of the Cauca segment's northern edge. Earlier studies of tomographic synthetic tests [Syracuse et al., 2016] and volcanic age 42 compilations in the past 14 m.y [Wagner et al., 2017] argued that Cauca segment subducts normally to the south of the Caldas tear, whereas plate reconstruction based on radiometric ages in the past 100 m.y posit there are overlapping areas between Cauca segment and Bucaramanga segment [Taboada, 2000; Kellogg et al., 2019].
Overall, although the existence of a gap (Calda tear) at latitude ~5°N separating Bucaramang and Cauca segments is evidenced by a variety of studies, e.g. seismic tomography results [Vargas and Mann, 2013], Adakitic samples from Nevado del Ruiz Volcano [Borrero et al., 2009] and SKS measurements [Porritt et al., 2014]. The nature of it, that is whether it stands for the southern edge of CAR [Taboada, 2000; Kellogg et al., 2019] or a slab tear developed within Nazca [Vargas and Mann, 2013; Syracuse et al., 2016; Wagner et al., 2017] due to subduction of Sandra ridge [Lonsdale, 2005; Vargas and Mann, 2013], is still under debate.
43 Fig. 3.1: Study region. Background color stands for topography; dashed lines (AA’--FF’) refer to cross sections in Fig. 3.4; white dash lines (X1X1’--X4X4’) refer to cross sections in Fig. 3.5. GPS data are from [Mora-Páez et al., 2019]. Hypocenters are from Colombian National Seismic Network.
3.3 Data and method
Waveforms from 327 teleseismic events with good azimuthal distribution (Fig. 3.1) were retrieved from 151 broadband stations in the permanent Colombia National Network and the temporary CARibbean-Merida Andes seismic array (CARMArray) deployed from April 2016 to March 2018. P arrival times from events with good signal to noise ratio
44 were initially manually picked. The selected traces then underwent Gaussian bandpass filters centered at 0.3 HZ, 0.5 HZ and 1.0 HZ and, for each frequency band, relative delay times were determined by cross correlation [Vandecar and Crosson, 1990] after removing the travel time predicted by the AK135 model [Kennett et al., 1995]. Eventually, 33871 relative travel time observations were derived.
These measurements are used as input for our velocity tomography procedure. The model is parameterized with variable node spacing to accommodate the loss of resolution with model depth as well as toward the edges of the array. Horizontal mesh node spacing is smallest inside the array footprint, where it is 42 km, and increases stepwise until it reaches 56 km 400 km outside the array. Cell height is also variable from 24 km for the shallowest layer of cells (36–60 km depth) to 60 km for the deepest layer (685–745 km depth). Eventually, the study region with a latitude range of (3.7795°S, 15.9556°N), longitude range of (82.7532° E, 65.6015° E), and depth range of (0km, 750km) is divided into a 40x48x24 dimensional mesh. Given their steep incidence angles, teleseismic rays cannot resolve crustal structure. To avoid mapping delay times caused by lateral variations in crustal velocity into mantle structure, we utilize the crustal Vs model of Poveda et al., [2018]. This shear wave velocity model was obtained through ambient noise tomography using 52 broadband stations in the region [Poveda et al., 2018]. We estimate the corresponding P wave velocities using a constant ratio of 1.75. Model parameters at crustal depths (depth <= 40km) are then derived by calculating the percentage variation from AK135 and used as an a priori constraint on the inversion.
45 Fig. 3.2: L-curve used to tune hyperparameters, hyperparameter choosed in this study is denoted with a star; insert histogram shows the residual between recorded and calculated delay times.
3.3.1 Finite-frequency tomography
The finite-frequency tomography used in our study differs from the classic ray theoretical tomography by taking into account the frequency-dependent sensitivity - and wavefront healing, consequently accounting for shifts in waveform-based delay times caused by velocity perturbations located within the first Fresnel zone and off the geometrical ray path. This approach incorporates a better theoretical representation of wave propagation and imposes physically based smoothness criteria on the inversion, thereby somewhat reducing the reliance on ad hoc regularization of the inverse problem. Research comparing p-value of resulting models using ray-theoretical and finite-frequency approaches demonstrates the advantage of finite frequency theory from a statistical perspective [Larmat et al., 2017]. We use an approximation to the Born theoretical sensitivity kernels commonly known as “banana doughnuts” [Dahlen et al.,
46 2000]. Calculation of sensitivity kernel is restricted to the first Fresnel zone in our study, the approximate radius of which (RF1) is determined as a function of distance along the ray path (DR) for a given frequency band (w) and epicentral distance (D). Relative sensitivity within the first Fresnel zone as a function of ray normal distance (RN) is - given by
where A is a scaling constant introduced to ensure that the integrated sensitivity along the first Fresnel zone volume is equal to that of the full Born kernel [Schmandt and Humphreys, 2010].
In addition to taking into account finite-frequency effects, we incorporate the effect of ray bending by using the hybrid ray-tracing approach of Bezada et al., [2013]. This method combines 1-D ray tracing outside of the model space with a 3-D graph-theory method [Toomey et al., 1994; Hammond and Toomey, 2003] used to calculate ray locations inside the model domain. The sensitivity kernels are then constructed using the equation above where the ray-normal distance is calculated from the 3-D rays. Ray geometries are updated after each iteration and new sensitivity kernels are calculated.
3.3.2 Damping and smoothing regularization
The linear inversion here is an ill-posed problem. Regularization is introduced to overcome the instability of solutions and prevent overfitting. Regularization penalizes model complexity and is constructed based on the prior information of specific problems. In this study, we utilize Tikhonov regularization to penalize the norm of the model
47 vector, and add smoothness regularization to restrict the discretized first order derivatives between adjacent voxels.
Considering the variation of model resolution with depth, global damping and smoothing weights are thereafter adjusted using a linearly interpolated depth-dependent weight vector. For voxels with a low hit quality, additional damping is further applied. The cost function we minimize is then:
J= ‖Gm-d‖²+ λ ‖m‖²+ β ‖Lm‖² where G is the sensitivity matrix that relates the model parameters m (velocity perturbations in each voxel) to the observed delay times d, and L is the first-order difference regularization matrix that refers to discretized smoothness. λ and β are hyperparameters defining the relative global weights of damping and smoothness respectively, and need to be selected before inversion. We use the L-curve method to achieve the trade-off between the fidelity term and penalty term, values of 12 and 12 are selected as hyperparameters for damping and smoothing regularization respectively (Fig. 3.2). We additionally include station and event terms to absorb near-surface structure and variations in the mean travel time anomaly between different events, respectively. After determination of hyperparameters, model parameters that minimize the cost function are optimized using the iterative LSQR method [Paige and Saunders, 1982]. The inversion iteration count plays a similar role as regularization, as too many steps yield an overfitting solution that suffers from a large propagated error due to the error in the data, and too few iterations give an underfitting solution that lacks details that can be of interest. In our research, model iteration is terminated when model updates are no longer significant. The preferred model was reached after 5 iterations and produces a data variance reduction of 73.25% (Fig. 3.2).
48 3.4 Model
In this section, we present the tomographic results by describing the geometry of velocity anomalies in detail, paying special attention to features that are relevant to our tectonic interpretation.
Above 90 km depth, the model shows discontinuous low and high-velocity anomalies dotting the study area (Fig. 3.3). This is most likely caused by a lack of sufficient crossing rays at these depths given the relatively sparse station spacing in the Colombian National Network (Fig. 3.8). Between 90 and 200km, these anomalies coalesce to delineate four distinct high-velocity bodies trending roughly NNE-SSW parallel to the strike of margin which we interpret as subducted lithosphere. The imaged slab segments are well-aligned with Wadati-Benioff seismicity and subduction-related volcanism (Fig. 3.3). For convenience, in the following we refer to these four segments, from north to south, as the Maracaibo, Bucaramanga, Cartago and Pasto segments (see Fig. 3.3, 160km depth slice). We note that previous authors refer to the slab south of 5.5°N as the Cauca segment, here, because we see different characteristics, we divide that segment into Cartago and Pasto north and south of ~3°N respectively.
The Maracaibo segment spans the length of the Caribbean coast of Colombia and has a convex geometry along strike that follows the curvature of the trench. A vertical cross-section through the model shows a normally subducted slab with a dipping angle of ~ 45° above ~350km and ~60° below ~350km (Fig. 3.4B, cross section BB’ in Fig. 3.1). At a depth of 50-70km, this segment seems to be detached ~50km off the trench. Nevertheless, absence of a sub-horizontal architecture here is later demonstrated to be an artifact caused by regularization scheme given poor hit quality at these locations, see synthetic tests in section 4.
49 Shifting to the south, the Bucaramanga segment underthrusts beneath SA with a strike of ~10°NE, and an abrupt change in inclination angle at depth ~150 km (Fig. 3.4C, cross section CC’ in Fig. 3.1). It is revealed to be a 100km-long subhorizontal high velocity architecture at depth above 150km, with synthetic tests (see section 4) indicating its actual spatial extent to be up to ~500km distance from the trench to the Bucaramanga nest location. This speculation is consistent with active occurrence of seismic events at locations where a hypothesized subhorizontal slab is proposed [Chiarabba et al., 2015; Syracuse et al., 2016]. At depth ~150km, the Bucaramanga segment is subjected to a change in slab strike (Fig. 3.3), along with an abrupt change in inclination angle from ~15° to ~75° (Fig. 3.4), which coincides with the seismogenic region hosting Bucaramanga nest.
West of the Bucaramanga segment, between 5°~8°N, additional high velocity anomalies are imaged with amplitudes of ~2% (compared to Bucaramanga ~4%). These anomalies are less continuous and trend roughly NE-SW. They are most clear at 90-230 km depth, but spatially smaller anomalies are found all the way down to the transition zone (Fig. 3.5). We tentatively interpret this as an additional slab segment (segment X) and will test its robustness in the following section.
The southern end of the Bucaramanga segment, coincides with the lateral shift in intermediate depth seismicity that has been dubbed the Caldas Tear at 5.5°N [Vargas and Mann, 2013]. South of there, we image the third clear slab segment (Cartago) that is colocated with the seismicity and thus ~250 km west of the Bucaramanga segment to the north. The offset between these two slab segments appears to decrease with depth, with the tear seeming to close by ~200 km (Fig. 3.3). At a latitude of ~3°N , there exists another gap separating the Cartago and Pasto segments. This gap coincides with an along-strike change in the density of intermediate depth seismicity as well as the spatial frequency of volcanoes and it is evident at all depths between 90 and ~270 km (Fig. 3.3).
50 Below 270 km, the Pasto segment is no longer present in the images.
An intriguing observation is that although above 270km these four slab segments show distinct characteristics including strike, dip and seismicity, the three segments, Maracaibo, Bucaramanga and Cartago appear to merge into a single feature by 270km depth and form a remarkably continuous linear slab by 400km depth (Fig. 3.3). This continuous slab-like feature is consistently observed down to the bottom of the transition zone where it seems to lay flat over the 660 km discontinuity.
51 Maracaibo
Bucaramanga
Cartago
Pasto
Fig. 3.3: Tomography result, map view. Volcanoes are denoted as magenta triangles on the depth slice 125k; seismicity is denoted as red circles at corresponding depths.
52 Fig. 3.4: Tomographic cross sections. (A-F) are lines(AA’-FF’) in Fig. 3.1.
X X X X X X X
Fig. 3.5: Secondary features to the west of Bucaramanga segment, labeled as ''X'. Cross sections from left to right are along X1X1'--X4X4' in Fig. 3.1.
53 3.5. Testing possible slab configurations
In interpreting our tomography model, we are mindful that, like all seismic models, it is an imperfect representation of the subsurface. The true slab structure is to some degree distorted by the imperfect station coverage and ray distribution, in addition to the inherent limitations of travel-time tomography. In this section, we investigate how various subduction configurations would be represented by our tomography procedure, to inform our interpretation. In particular, we mainly seek to answer the following questions: 1) Do the anomalies we have tentatively labeled as “slab segment X” represent subducted lithosphere, or are they likely to be streaking artifacts (incorrect mapping at depth of shallower velocity anomalies)? 2) Do the Bucaramanga and Cartago segments truly merge at depth, or is there a different explanation for the imaged geometry? 3) Are the gaps we image between slab segments real, or are they an artifact of incomplete station coverage.
Questions 1 and 2 above are especially important for a tectonic interpretation. The fact that the Maracaibo, Bucaramanga and Cartago segments appear to merge into a single, continuous feature suggests a common origin for this lithosphere. This would mean they are all fragments of either the Caribbean or Nazca plates. A common Nazca origin can be ruled out because we can follow the Maracaibo segment unequivocally to the Caribbean coast as far north as ~11°N and Nazca subduction has never reached that latitude [Montes et al., 2019; Kellogg et al., 2019]. The Maracaibo segment must be Caribbean. On the other hand, a common Caribbean origin for all three segments can also be ruled out because the Cartago segment is clearly related to recent active volcanism that can only be a product of Nazca subduction [Taboada et al., 2000; Wagner et al., 2017; Mora et al., 2017; Kellogg et al., 2019]. The Pasto segment must belong to
54 Nazca. Meanwhile, the continuity in intermediate-depth seismicity and velocity anomalies between the Maracaibo and Bucaramanga segments strongly suggests a common, Caribbean, origin. Yet, young volcanism occurring as far north as ~8°N (north of the southern end of the Bucaramanga segment) can only be attributed to Nazca subduction, given that it occurs well after the collision and accretion of the Panama-Choco block [Wagner et al., 2017; Mora-Páez et al., 2019; Kellogg et al., 2019]. These apparently conflicting lines of evidence can be resolved if there is overlapping Nazca and Caribbean subducted slabs between ~5.5°N and ~7°N as has been suggested by Kellog et al. [2019] and others. In this scenario Pasto, Cartago and Segment X would all be subducted Nazca lithosphere while the apparent merging of the Cartago and Bucaramanga segments is explained by a change in along-strike width of the Bucaramanga segment, growing further south with depth.
3.5.1 Architecture of synthetic input models
To test the different hypotheses outlined above we design three variations of a synthetic model containing slabs representing Caribbean and Nazca subduction. We note that the goal of these tests is not to exactly replicate what is recovered in the inversion of real data, as the detailed geometry is difficult to reproduce, but rather to investigate the questions posed at the beginning of this section. The synthetic models include different slab segments. Between 5°-12°N, we assume a continuous slab in the depth range of 50-660km, with its southern edge propagating southwards along depth. This segment is common to all three variations and represents a Caribbean slab that includes the Maracaibo and Bucaramanga segments. The slab geometry includes a change in dip from 20° to 70° at a depth of ~120km to emulate the observations of the Bucaramanga segment. At 90-150km depth the synthetic slab coincides with intermediate-depth seismicity, although we don’t include the change in the trend of the seismicity at the Bucaramanga Nest
55 (Fig. 3.6). Other slab segments represent the Nazca subduction and their geometry is different in each variation.
Variation I: A continuous slab extending between 50-300km along depth and 0°-5° N along latitude, following the trend of the intermediate depth seismicity (Fig. 3.6a). In this variation there is no overlap with the Bucaramanga segment.
Variation II: Similar to above but extending from 0° to 8°N, overlapping with the Bucaramanga segment between 5°-8°N (Fig. 3.6b). This would represent a continuous slab including the Pasto and Cartago segments, along with segment X.
Variation III: As in II, the slab extends to 8°N along latitude, but with a gap between 5.5°-6.5°N (Fig. 3.6c). In this configuration there is overlap with the Bucaramanga segment between 6.5°-8°N and a clear separation between Cartago and segment X.
For each of these variations of the input model, synthetic delay times were calculated using the same event-receiver geometry as for the real data. We then invert the synthetic data sets with the same procedure, regularization, and stopping criterion as with the real data. Comparison of the known input and inverted output models provides insights to distinguish robust features from imaging artifacts.
3.5.2 Synthetic results
Firstly, as shown in Fig. 3.6 and Fig. 3.7, comparison of the synthetic input and output models demonstrates that our source-receiver geometry is able to accurately resolve structures deeper than ~200km; as the locations and characteristics of the input velocity
56 anomalies are well recovered by the inversion. We focus this discussion on aspects of the synthetic test results that have a direct bearing on the tectonic interpretation of the inversion of real data. Inspection of the recovered synthetic structures leads to the following conclusions:
(1). Subhorizontal subduction is not imaged as a high-velocity anomaly.
Comparison between synthetic input and output of synthetic model I (Fig. 3.6d vs. Fig. 3.7d) implies weak resolution for subhorizontal slabs at shallow depths as may be found in the Maracaibo and Bucaramanga regions. This is likely caused by the lack of crossing rays at those depths, and corresponding low hit quality (see Fig. 3.8), and is a well-known limitation of teleseismic tomography. Although flat or low angle subduction of the Bucaramanga segment is necessary to explain the location of intermediate- depth seismicity relative to the trench [Vargas and Mann, 2013; Chiarabba et al., 2015; Syracuse et al., 2016; Mora et al., 2017], our test shows that such a feature would be transparent to our tomography. Importantly, this implies that the high-velocity anomalies to the west of the Bucaramanga segment in the inversion of real data (segment X) are not a product of smearing of high-velocity anomalies related to the flat slab.
(2). High velocity anomalies west of the Bucaramanga slab require the presence of the subducted Nazca lithosphere north of 5.5°N.
The secondary feature (segment X) behind the Bucaramanga segment is only successfully reproduced (Fig. 3.7c and Fig. 3.7f) when additional anomalies representing a northward extension of the Nazca slab (overlapping the Caribbean Bucaramanga segment) are introduced in the synthetic input (Fig. 3.6c and Fig. 3.6f); while no artificial features (Fig. 3.7a and Fig. 3.7d) are generated if there is no overlap (Fig. 3.6a and Fig. 3.7d). The short, near-vertically dipping anomaly attached to the surface in Fig. 3.7(f) is consistently recovered along strike where Nazca and Caribbean overlap. This recovered
57 synthetic geometry is mostly similar to the cross section X2X2' in the model derived from real data (Fig. 3.5). As illustrated in section 3, this secondary feature appears to be of varied geometry and depth from north to south along the overlapping region in the real model. However, rather than attempting to constrain the variation of its morphology at different locations, our synthetic test only aims at discerning whether it is likely to be real or caused by streaking effects. Possible mechanisms accounting for the varied morphology of this secondary feature along the strike will be explored in section 5.
(3) Along-strike gaps separating the imaged slab segments are likely real features.
The gap observed between [5.5°~6.5°N, 74°~76°W] in Fig. 3.3 is only reproduced when a gap exists between the Cartago segment and “segment X”, as shown in Fig. 3.6 (a-c) and Fig. 3.7 (a-c), which indicates there likely is a gap at this location, rather than an improperly recovered continuous slab. Further south, at ~3° N the real model shows a gap between what we have called the Cartago and Pasto segments (Fig. 3.3). In contrast, synthetic outputs successfully recover a prescribed slab in this area that is continuous along strike, with no gap artificially generated between 1°~3°N (see Fig. 3.6a-c vs. Fig. 3.7a-c). This suggests that the gap between these two slab segments is also a real feature.
(4). There is likely flat-lying Caribbean slab material atop the 660 km discontinuity.
Tomography results in section 3 show flat-lying slab material at depths of 500-660 km (Fig. 3.3, 3.4d, 3.4e). In the synthetic test, the input models have no flat-lying velocity anomalies at these depths. Accordingly, in the corresponding output velocity models, there are no artificially generated sub-horizontal anomalies at the bottom of the transition zone. This gives us confidence in the robustness of the deep anomalies east of the toe of the slab in the model derived from real data. However, limited resolution
58 at this depth does not allow us to give detailed constraints on its geometry or infer whether or not the slab penetrates across the transition zone.
Fig. 3.6: Three sets of synthetic input models. (a) mapview at depth 90km of model I, (d) and (g) are cross sections along line 1 and line 2 denoted in (a); (b) mapview at depth 90km of model II, (e) and (h) are cross sections along line 1 and line 2 denoted in (b); (c) mapview at depth 90km of model III, (f) and (i) are cross sections along line 1 and line 2 denoted in (c).
59 Fig. 3.7: Inverted results of synthetic models. (a-c): mapview at depth ~300km; (d-i): cross sections of synthetic inverted model along line 1 and line 2 in Fig. 3.6.
60 Fig. 3.8: Hit quality, with slab contours with 2%, 2.5%, 3% anomalies denoted by red, blue and green lines respectively.
3.6. Discussion
Resolving the debate about the current configuration of subduction beneath the northern Andes has been hindered by lack of constraints in the deeper subsurface structure. Here, taking advantage of the Colombian National Seismic Network and a dense temporary deployment, we provide a well resolved tomography velocity model with uncertainties illustrated by synthetic tests in western Colombia and western Venezuela.
Considering the results of the tests described in section 5, our tomography model is most consistent with the scenario described by Kellog et al. [2019], and hypothesized by others [e.g. Cortes and Angelier, 2005; Taboada et al., 2000]. Namely, there is about 200 km of overlap between the Caribbean and Nazca subductions between ~5.5° and ~7.5°N. The
61 subducted Caribbean slab extends from ~11°N to ~5.5°N at ~100 km depth, with its southern end extending further south with increasing depth, and includes the Bucaramanga and Maracaibo segments. The offset of intermediate-depth seismicity at the “Caldas tear”, marks the southern end of Caribbean subduction instead of offset between two Nazca segments, in accordance with Kellog et al. [2019]. Nazca subduction does seem to be fragmented in our study area, with three distinct segments, from north to south: “segment X” north of the “Caldas tear” is the one that overlaps with Caribbean subduction and shows no volcanism or intermediate-depth seismicity (Fig. 3.4 and Fig. 3.5). The Cartago segment, directly south of the “Caldas tear” and north of ~3°N shows relatively abundant intermediate-depth seismicity and is co-located with active volcanism (Fig. 3.4). Finally, the Pasto segment south of 3°N is related to volcanism in the surface and shows a modest amount of intermediate depth seismicity. The southern end of the Pasto segment lies outside our study area but inspection of the intermediate-depth seismicity hypocenters suggests its southern terminus is ~1.3°S, given that south of that latitude the density of intermediate-depth seismicity increases again.
The Bucaramanga slab segment has been variably attributed over the decades to Caribbean [Taboada et al., 2000; Cortes and Angelier, 2005; Kellogg et al., 2019] or Nazca [Syracuse et al., 2016; Wagner et al., 2017] lithosphere. Hypotheses arguing for a Caribbean origin are mainly based on the continuous distribution of intermediate-depth seismicity from ~11°N to ~5.5°N, across the Bucaramanga and Maracaibo segments, at a depth range of 90~150km [e.g. Kellogg et al, 2019. Fig. 3.1, Fig. 3.3]; as well as by palinspastic reconstructions [e.g. Kellogg et al, 2019; Montes et al., 2012; Cortes and Angelier, 2005]. In contrast, a Nazca origin is suggested not only by the current position and direction of the Nazca plate with respect to South America, but also from compilations of volcanic ages in the past 14Ma [Wagner et al., 2017]. Wagner et al. [2017] point out that arc volcanism associated with Nazca subduction extended along Colombia’s entire Pacific margin in the mid-Miocene. They interpret a shut-off of
62 volcanic activity north of ~ 3°N at ~6 Ma as the onset of flat-slab Nazca subduction, with arc volcanism resuming only south of 5.5°N after 4 Ma. This pattern was interpreted by Wagner et al. [2017] as showing Nazca slab flattening at 6 Ma and subsequent tearing at the Caldas tear, resulting in a flat slab segment to the north (Bucaramanga) and a re-steepened segment to the south (Cauca, which we separate into Cartago and Pasto).
Our tomography results show a laterally continuous slab-shaped positive velocity anomaly spanning all the northern Andes (3°N-10.5°N) at depths deeper than 300 km (Fig. 3.3). This is strong evidence supporting a common origin for the Maracaibo and Bucaramanga segments, that can only be associated with the Caribbean as discussed in section 5. Distinct morphologies of these two segments at shallower depths (90-300 km), including different strike and dip (Fig. 3.3, Fig. 3.4), are likely to be an expression of the heterogeneity of the incoming Caribbean plate along strike and the diachronous collision of the Panama-Choco block at the trailing end of the plate [e.g. Kellogg et al. 2019, Montes et al. 2019]. Furthermore, the spatial extent of the subducted Caribbean slab as constrained by our tomographic results is highly consistent with estimates of the length of the trench in the geologic past as presented in several recent palinspastic reconstructions [e.g. Kellogg et al., 2019; Montes et al., 2019].
Additional high-velocity anomalies are observed to the west of the Bucaramanga segment between 5°~8°N and synthetic tests rule out the possibility that these are merely imaging artifacts. Instead, we interpret these anomalies as the northernmost expression of Nazca subduction, where it subducts beneath the Choco block, once a part of the Caribbean, but currently accreted to South America [Kellog et al., 2019, Montes et al., 2019]. This interpretation is consistent with plate reconstruction studies [Kellogg et al., 2019, Montes et al. 2019] that argue for exactly this sort of overlap between Nazca and CAR based on the volcanism and deformation in the past ~100 m.y., as well as with
63 current GPS observations [Kellog and Vega, 1995; DeMets et al., 2000; DeMets, 2001; Weber et al., 2001; Trenkamp et al., 2002]. The imaged slab geometry is consistent with the volcanic history described by Wagner et al. [2017]. As Kellog et al. [2019] point out, the spike in volcanic activity in the Choco terrane and just east of the terrane from 15 to 5 m.y was indeed related to Nazca subduction, but the overriding plate was the Caribbean. In their interpretation, the abrupt cessation of volcanic activity to the north of 5.5°N ~6 Ma can be explained by thermal isolation caused by the Nazca slab coming up to the subducted Caribbean slab. An outstanding question is why volcanism ceased as far south as ~3°N if the southern limit of slab overlap (and the hypothesized thermal isolation) is 5.5°N. We propose as a working hypothesis that as the Nazca trench approached the Caribbean slab trench suction may have indeed produced some flattening of the northern end of Nazca, as far south as ~3°N, where we place the limit between the Cartago and Paso segments. Subsequently, the Cartago segment (3°-5.5°N) would have resteepened and resumed volcanism, while segment X remained avolcanic given its overlap with Caribbean lithosphere as described by Kellog et al. [2019]. This scenario would provide a mechanism for slab flattening of northernmost Nazca, which remained unidentified in Wagner et al. [2017]. Under this hypothesis, there would indeed be a tear in the Nazca plate at 5.5°N but it would be different in nature from what was originally posited by Vargas and Mann [2013].
One important argument against our interpretation is the lack of intermediate-depth seismicity associated with ‘segment X’. We note firstly that our imaging and synthetic tests argue strongly for the presence of subducted lithosphere west of the Bucaramanga segment between 5.5° and 8°N. Subduction without intermediate depth seismicity is also found elsewhere. The subduction of the Juan de Fuca plate (another remnant of Farallon, e.g. Atwater, 1970) is bracketed to the north and south by the subduction of the Explorer and Gorda microplates, respectively (e.g. Stock and Lee, 1994). Neither of these show seismicity below ~60 km (e.g. McCrory et al., 2012), in spite of appearing as
64 high-velocity anomalies at those depths in tomography models [e.g. Bodmer et al. 2018]. Although we have referred to the Cartago segment, and segment X as Nazca subduction (in accordance with most authors), in northernmost Nazca 2 microplates have been identified: the Malpelo and Coiba microplates [e.g. Adamek et al. 1988, Zhang et al. 2017]. What we have called “segment X” may be more accurately called subducted Coiba lithosphere. Coiba lithosphere is young (<10 Ma, e.g. Müller et al, 2008) and may be too warm to host intermediate-depth seismicity. This scenario would be somewhat analogous to Gorda or Explorer subduction, with the added complication of the overlap with the Caribbean slab.
The formation of the Nazca plate took place after the fragmentation of the Farallon plate during the Late Oligocene–Early Miocene [Hoernle et al., 2002; Lonsdale, 2005; Meschede and Barckhausen, 2000]. The real-data model in our study indicates the Nazca plate behind the Caribbean plate might be subjected to detachment or fragmentation. Several events of seafloor spreading have been reported in this plate over the last 20 m.y., which means very young ages of the lithosphere distributed in an asymmetric pattern [Müller et al., 2008]. Specifically, the gap sitting between 5.5°~6.5° N (Fig. 3) is collinear with the W-E Sandra ridge, with such a ridge subducted, the fragmented Nazca plate would be easily drifted into different directions beneath SA. The low-angle subducted Caribbean above and to the east of it might play a role of thermal blanket. These effects might explain the absence of a second Wadati-Benioff zone to the west of Caribbean plate.
Lastly, our interpretation has implications for the origin of the Bucaramanga Nest. This volume of very frequent intermediate depth seismicity has been interpreted as occurring within the Nazca plate [e.g. Chiarabba et al. 2016], within the Caribbean plate [Taboada et al. 2000, Cortes and Angelier, 2005, Prieto et al., 2012], and at the
65 boundary of the two [Zarifi, et al. 2007, Syracuse et al, 2016]. The unusually high concentration of seismicity has been attributed to the tip of propagating slab tears [e.g. Vargas and Mann 2013, Cortes and Angelier, 2005], and to the subduction of extraordinarily hydrated lithosphere [e.g. Chiarabba et al. 2016]. According to our interpretation of the tomography model presented here, the Bucaramanga nest is located in the interior of the subducted Caribbean plate. Furthermore, the nest coincides with elevated curvature in the slab surface; it is located where there is a large increase in dip and also a change in the strike of the slab (Fig. 3.3, Fig. 3.4). Modeling has shown that slab bending and associated larger strain rates produce zones of elevated stress within subducted slabs [Billen, 2020]. Although we cannot rule out a contribution by locally enhanced hydration of the slab, we suggest that elevated stresses related to the high curvature of the slab in both a vertical and horizontal plane play an important role in the concentration of seismicity at the Bucaramanga nest. The likely complicated stress distribution at a doubly bending point in the slab may help explain the diversity of focal mechanisms found in the Bucaramanga nest [e.g. Prieto et al., 2012].
3.7. Conclusion
In this paper, we provide a-well resolved P wave velocity model of the NW South America subduction zones. Some features proposed by previous work are further confirmed and several new features with significant implications are observed for the first time. To summarize: • The Maracaibo and Bucaramanga segments are of Caribbean origin, with the ‘Caldas tear’ being the southern edge of the subducted Caribbean plate. The southern edge of this slab propagates southward with depth. Differences at shallow depths
66 between these two segments is likely an expression of heterogeneity in the incoming Caribbean plate and the diachronous collision of the Panama-Choco block. • The Maracaibo segment transitions from subhorizontal subduction at shallow depth into normal subduction (dip angle ~40°) at depth deeper than 100km;
• The Bucaramanga segment is subjected to subhorizontal subduction and extends ~600 km away from trench, and transitions into normal subduction (dip angle: ~75°) at depth deeper than 150km. The Bucaramanga nest is hosted within the subducted Caribbean plate where there is an abrupt change in both dip and strike at depths of ~150km. Therein highly accumulated stress should be available resulting from the bending in two different planes, a corollary is the high seismicity rate may be caused by the enhanced stress and not by excess dehydration;
• The Nazca plate is observed to overlap with Caribbean plate between 5.5°~8°N, in what we are calling “segment X”. South of ~5.5°N, we identify the Cartago and Pasto segments (Cauca segment for previous authors). These are normally-subducted to the south of the ‘Caldas tear’. Subduction of EW-oriented weaknesses related to relic spreading centers might account for the fragmentation of the Nazca plate and drifting of the subducted fragments into different directions.
67 Chapter 4
Is the New Madrid Seismic Zone caused by a weak zone in the upper mantle? Insights from attenuation investigation
4.1 Introduction
Although intracontinental earthquakes are less common than interplate earthquakes, they can be more devastating as their vicinity to vulnerable populations [Bilham, 2014]. Hypotheses emerged proposing the causality between intracontinental seismicity and abandoned rifts. As a consequence of the central United States being subjected to a long tectonic history of continental collision and rifting, a complex of rift systems are hosted within the interior of the North America continent. Among these, the Reelfoot Rift, which resides within Mississippi embayment (UME) and constitutes the Mississippi Valley Graben [Thomas, 1991; Johnson et al., 1994], is infamous for hosting several M6.8-8 intraplate earthquakes during 1811-1812 and remaining to be seismic active nowadays [Johnston and Schweig, 1996; Cramer, 2001; Hough and Page, 2011] (Fig. 4.1). The Mississippi embayment (UME) underwent initial extensive rifting in the Late Paleozoic, uplifting and subsequent erosion processes associated with the passage of the Bermuda hotspot in mid-Cretaceous [Cox and Van Arsdale, 1997; Cox and Van Arsdale, 2002]. A second phase of subsidence was proposed to occur in the Late Cretaceous and ended by mid-Eocene [Cox and Van Arsdale, 1997], and no consensus has been achieved over the mechanism causing it [Ervin and McGinnis, 1975; Kane et al., 1981; Braile et al., 1986; Cox and Van Arsdale, 1997; Cox and Van Arsdale, 2002;Wolaver et al., 2012] .
68 As indicated by the slow strain rate measured by Global Positioning System (GPS) [Calais and Stein, 2009; Frankel et al., 2012], long term tectonic stress loaded and transferred from plate boundaries are unlikely to be responsible for the frequent occurrence of seismicity within the New Madrid Seismic Zone (NMSZ). Alternatively, hypotheses with local stress origin include isostatic adjustment caused by river incision since last Ice age [Calais et al., 2010], stalling of Farallon slab in the lower mantle [Forte et al., 2007], and high density lower crust elevating gravity derived stress at seismogenic - depths [Levandowski et al., 2016]. A hypothesis that a weak zone within the upper mantle localizing differential stress was also proposed [Pollitz and Mooney, 2014; Chen et al., 2015; Nyamwandha et al., 2016; Zhan et al., 2016]. An existence of a weak zone should display geophysical signatures such as low velocity zone and high attenuation, thus understanding the seismic structure beneath this area is important for testing this hypothesis. Recently a variety of tomography studies revealed a low velocity zone within the upper mantle at depths approximately 100~200km under the Reelfoot Rift [Chen et al., 2014; Pollitz and Mooney, 2014; Nyamwandha et al., 2015; Wang et al., 2019]. However, the location, scale and geometry of this proposed LVZ varies significantly across different studies. Zhang et al. [2009] delineated a narrow LVZ that is restricted within and parallel to the Reelfoot Rift. Chen et al. [2014] delineated a LVZ that spans a much larger volume and, at depths between 80-120km, the western edge of the Reelfoot Rift overlaps with the boundary between the LVZ and a high velocity anomaly to the east. With increasing depth, the LVZ shifts ~1° to the west of the Reelfoot Rift, and a high velocity anomaly lies at depths deeper than 250km. Nyamwandha et al. [2015] shows an LVZ that is of a smaller scale than the model proposed by Chen et al. [2015], but unlike the model proposed by Zhang et al. [2009], it is not restricted within the Reelfoot Rift. At a depth of ~200km, the LVZ shifts ~20km to the west and becomes parallel to the strike of Reelfoot Rift. High velocity anomalies were also reported in their research, but was inconsistent with the high velocity revealed by Chen at al., [2014]. Chen et al. [2014]’s result reported that one large scale of high velocity lying to the west
69 of the LVZ at depth between 80-200km, and another high velocity anomaly beneath the LVZ at depth deeper than 250km. However, high velocity anomalies demonstrated in Nyamwandha et al. [2015]’s study distributed above and along the side of the LVZ at depth between 80-200km, they were interpreted as the remnants of the depleted cratonic lithosphere. Wang et al. [2019] explored this problem by simultaneously inverting the P wave azimuth anisotropy and P wave velocity, and revealed a LVZ between 80-200km that does not have a consistent geometry and locations with increasing depth, a high velocity anomaly spreads a large volume and lies at depths deeper than 200km.
Finite element numerical experiments [Zhan et al., 2016] based on Pollitz et al. [2015] model illustrated that if the low velocity can be fully attributed to a thermal origin [Liu and Zoback, 1997], the extensive mantle LVZ would have a significant impact on the differential stress localization beneath New Madrid Seismic Zone, this stress can be further transfer to the base of preexisting rift faults and lead to their reactivation. But McKenna et al. [2007] argued that according to the surface heat flow data, no significant positive thermal anomaly exists beneath the Reelfoot Rift. Therefore, thermal origin can not, at least not be fully responsible for the LVZ. Alternatively, Pollitz and Mooney [2015] speculated this LVZ resulted from mantle refertilization or emplacement that was associated with the passage of the Bermuda hotspot at ~90 Ma [Chu et al., 2013]. Chen et al. [2014] and Nyamwandha et al. [2015] proposed the LVZ is caused by the addition of water that dehydrated from the stalled Farallon slab within the lower mantle.
To summarize, although various studies have been conducted, a comprehensive understanding regarding which mechanism(s), or a qualitative contribution from various mechanisms to the occurrence of seismicity within the Reelfoot Rift have not been achieved. Among these, although hypotheses resorting to the LVZ have not agreed upon the mechanism accounting for the low velocity, it is widely speculated that this LVZ stands for a weak zone localizing stress. Previous studies [Bezada, 2017; Bezada and
70 Smale, 2019] have found a quantitative and quantitative relationship between seismic attenuation and the distribution of intraplate seismicity, therefore attention estimation is another tool for us to gain further insights of the physical structure. In this paper, we analyse P-wave attenuation from records of teleseismic events and provide insights into possible rheology properties of the upper mantle beneath the Reelfoot Rift, and whether the upper mantle LVZ could be a possible mechanism triggering earthquakes in the NMSZ. In our study, unexpectedly, results show relatively low attenuation within the Reelfoot Rift region, indicating the Reelfoot Rift might not be a weak zone. Meanwhile, an association between the attenuation and occurrence of seismicity is revealed, this conveys that the lateral variation of attenuation along the strike of the Reelfoot Rift region might be responsible for the NMSZ.
Fig. 4.1: Study region. Earthquakes are denoted by red dots; Black, white and yellow triangles represent stations from XO, ZL, TA networks respectively. Blue dashed lines outline of the Mississippi Embayment.
71 4.2 Data and method
4.2.1 Data
Attenuation is the loss of energy as the wavefield propagates through an anelastic medium and is often measured with the parameter t*, which is defined as the integrated attenuation effect along the ray path and determined by the following equation.
Q(r), V(r) are the quality factor (inverse of fractional energy loss per cycle) and wave propagation velocity at the corresponding distance r along the ray path from source to receiver. As demonstrated by the equation above, in our study, attenuation evaluation is restricted to two dimensions, with the measured Δt* representing an integrated attenuation effect along the ray path in reference to the least attenuated records; the mean is removed afterwards, leaving us differential attenuation relative to the average attenuation in the study area.
The widely used spectral ratio method [Teng et al., 1968] measures Δt* between stations in the frequency domain by finding the slope of the best linear fit to the logarithm of the ratio of the amplitude spectra of two seismograms. Results measured by this method can be affected by extrinsic factors such as the frequency range to fit the slope, the magnitude of events, directionality of source time functions, and scattering and focusing effects [Adams and Humphreys, 2010; Cafferky and Schmandt, 2015; Bezada et al., 2019; Byrnes and Bezada, 2020]. To alleviate these problems and achieve more reliable measurements, firstly, we resort to a time domain waveform matching method [Adams and Humphreys, 2010; Bezada, 2017] that considers both amplitude and dispersion
72 effects of attenuation to measure the differential attenuations. Secondly, only teleseismic events with hypocentral depths greater than 250km, magnitudes larger than M5.5 are selected. Given lower mantle is of high Q and attenuation mostly originates from asthenosphere layers, selecting teleseismic events with deep hypocentral depths provides us an opportunity to analyse the physical rheology condition of the asthenosphere beneath our target study area.
Records were retrieved from the Ozark Illinois Indiana Kentucky Flexible Array Experiment (XO), Northern Embayment Lithospheric Experiment (NL), USArray Transportable Array (TA) between 2010-2015 (Fig. 4.1), with their waveforms filtered between 0.02~3 Hz. XO network is a flexible Array experiment centered on the Illinois Basin [Pavlis and Gilbert, 2011] and covers the northernmost area of our study region (Fig. 4.7). NL network is a passage of the EarthScope TA stations across the Mississippi Embayment [Langston et al., 2011], and covers the middle part of our study region (Fig. 4.7). After visual quality control [Bezada, 2017], we selected 19 teleseismic events (10 events from Andes regions, 8 events from Kuril Islands and 1 from the Aleutians, see Fig. 4.2) recorded by these networks. For each event, we make an estimate of the unattenuated source-time function, and match attenuated versions of this estimate to the first arriving P wave recordings using a grid search over Δt*p [Bezada, 2017]. A total of 1444 measurements of Δt*p were used in this study.
73 Fig. 4.2: Events used in this study. Events are denoted as red stars; the study region is outlined by green lines.
74 Fig. 4.3: Example comparing between observed (balck) and synthetic (red) traces of a same event. (A). Distribution of attenuation estimation for each station; (B). Observed and synthetic waveforms at four stations, estimated Δt* are denoted in each subplot, station names are denoted in (A). Q42 and LD14 are examples of low attenuation, X47A and MF08 are examples of high attenuation.
75 4.2.2 Inversion scheme
Measurements of Δt* are normally very noisy and the raw measurements are hard to interpret without some manner of smoothing [e.g. Cafferky and Schmandt, 2015]. One difficulty is that there is no clear way to objectively choose a smoothing length. To address this problem, a two-dimensional (map view) model of attenuation across the study area was inverted using a reversible jump Markov chain Monte Carlo (rj-MCMC) algorithm [Bodin et al., 2012a, 2012b; Eilon et al., 2018; Olugboji et al., 2017; Byrnes et al., 2019]. Optimization schemes that aim at minimizing the misfits require some parameters to be set beforehand, e.g. the number of free parameters, penalty terms, relative weights between multiple data sets etc.. However, station distribution tends to be uneven across the space and thus resolvable model complexity at different locations ought to vary. A global regularization term in optimization schemes is likely to under or over damp different sub-regions of the spatial domain, that is to say, the data information content is not being fully utilized. In contrast, the rj-MCMC is a self-adapted sampling method in Bayesian scheme that allows simultaneous inference on both model and parameter space. It utilizes the prior knowledge of model distribution p(m), effectively transforms prior into posterior by likelihood function. The likelihood function p(d|m) represents the probability of obtaining the observations d given the model m, and measures the fit between observations and calculations. The level of data uncertainty is introduced as a free parameter, the width of probability distribution in the likelihood function. When multiple types of data sets are involved, the estimated data uncertainties play the role of natural relative weights. Estimated data uncertainties tend to converge towards the true uncertainty level when maximizing likelihood.
In this study, we set the prior of the model to be a uniform distribution bounded by [-0.5s,
76 0.5s], which is a physically reasonable limit for attenuation. For each Markov chain, a Voronoi model is initialized with number of cells, locations, and Δt* values of each cell to be randomly drawn from the prior uniform distribution. A nearest neighbor interpolation is applied in this study to obtain Δt* at an evenly spaced grid covering the study area.
Along each Markov chain, after a random walk during the burn-in period, the model converges and stabilizes to a projection of the posterior probability distribution. Each step of the Markov chain is updated iteratively following two steps: a proposal and an acceptance of the proposal. A change is proposed with an equal probability of five possibilities (1). Change of Δt* value of a random Voronoi cell according to Guassian distribution centered at the current Δt* value; (2) Perturbation the location of a random Voronoi node according to 2-D Gaussian distribution centered at the current location value; (3) birth of a new node, (4) death of a random node, (5) change of noise parameter.
The burn-in and convergence rate is sensitive to the starting point, chains with randomly initialized starting points that are closer to the posterior tend to converge faster. In this study, a burn-in value of 2.5ｘ10⁵ is selected and post burn-in results are recorded every other 2.5ｘ10³ iterations. 50 Markov chains with different starting points that are randomly drawn from the prior are parallized. Eventually, we derive an ensemble of 5050 Voronoi models with variable parameterizations and noise estimations, distributed according to the posterior PDF. These models are then averaged to get one single smooth model with varied spatial complexity that is directly determined by the data. Model uncertainty is defined as dividing the standard deviation of the PDFs by two, which gives twice the standard error (i.e., the 95% confidence interval). More details of this method has been addressed by Byrnes et al. [2019].
77 4.3 Result
An ensemble of 5050 post burn-in models from 50 Markov chains are sampled and recorded. We therefore derive our final model by taking the mean value of PDF sets at each point in a regularly spaced grid. Model uncertainty is estimated by dividing the standard deviation of the PDFs by two, which gives twice the standard error [Byrnes et al., 2019]. A large model error can be caused by multiple factors such as large divergence of starting points from posterior distribution, small sampler width, small values for burn-in and iterations etc.. Thus can be used as an indicator as for model convergence and provide confidence level for derived values.
As shown in Fig. 4.4, evaluated Δt* distributes between [-0.15s, 0.15s]. Contrary to what might be expected, the NMSZ occurs within a low attenuation area. This N-S trending low attenuated block spans between 34°-36.5°N with width tapering from north to south. It is overlaid by the Reelfoot Rift system, and the northern and southern ends of NMSZ coincide with the large gradients in Δt* of this low attenuated anomaly. To the west of the NMSZ, there lies a NE-SW trended high attenuation belt wrapping around the western and northwestern edge of the N-S low attenuation anomaly.
4.3.1 Event back azimuth effects
In this study, in order to check the back-azimuth dependence for the inversion results, we re-conducted the inversions using events from the Andes and the Kuril Islands separately (Fig. 4.4). If we assume that most of the attenuation occurs in the relatively shallow, low-Q, asthenosphere [Bezada, 2017], and given the steep incidence angle of teleseismic ray paths, we would expect events from different azimuths sample similar structure on the receiver side. Under these assumptions, no significant variations 78 of inversion results are expected by using event subsets from different back azimuths separately, which is found to be the case in Iberia and Morocco [Bezada, 2017]. However, in our study, results using event sets of different back azimuth are only partially consistent with each other. As shown in Fig. 4.4b and 4.4c, the existence of the NE-SW trend high attenuation belt lying to the west of the Reelfoot Rift holds robust between these two inversions; the region hosting the NMSZ are of low attenuation in both inversions. But Δt* estimations within the Reelfoot Rift using Andes events is ~0.05 s less than it is when using Kuril Islands events. The area between -90°~-87°W is revealed to be of low attenuation when only Andes event dataset is utilized (Fig. 4.4B), whereas when using events from Kuril Islands (Fig. 4.4C), a high attenuated anomaly is revealed to be between -88°~-87°W and a low attenuation block is between -90°~-88°W. Moreover, unlike the inversion result by using Kuril Islands events that the NMSZ occurs within the lowest attenuated anomaly (Fig. 4.4B), results using Andes events suggest that the NMSZ lies within a low attenuation area that is between the NE-SW trended high attenuation belt to the west and the lowest attenuated block to the east (Fig. 4.4C).
We took two profiles across the Reelfoot Rift to further investigate the association between attenuation and the occurrence of seismicity. Line 1 is perpendicular to the strike of the Reelfoot Rift, and line 2 is along the strike of the Reelfoot Rift (see Fig. 4.4). Profiles along line 1 demonstrate the same phenomena as we mentioned previously, namely that the NMSZ occurs within a low attenuation block and near the boundary with an adjacent high-attenuation block (see Fig. 4.5A, C, E). Interestingly, in spite of the small lateral variation along the profile that is within the Reelfoot Rift, profiles along line 2 (Fig. 4.5B, D, F) convey a good consistency between Δt* and the occurrence of seismicity. This is most apparent in the result derived from Andes events (Fig. 4.5D). Despite the NMSZ lying within a low-attenuation area, earthquakes tend to occur at local highs in attenuation and are scarce in regions that are of relatively lower attenuation. However this association is slightly different in the results using events from Kurils Islands, although between 92°~90.5°W, the relative Δt* is consistent with the occurrence
79 of earthquakes. The spike of earthquake occurrence observed between 90°~89°W coincides with a gradient in attenuation, rather than a local high.
Fig. 4.4: (A-C) are distribution of Δt* after inversion by using records from TA, XO and ZL networks together; (D-F) are model uncertainty as defined in text. Red dashed lines (line 1, line 2) refers to the profiles in Fig. 4.5.
80 Fig. 4.5:. Δt * and earthquake counts along line 1 and line 2 (see Fig. 4.4).
81 4.3.2 Network effects
Networks employed in this region include TA, XO and ZL. In order to investigate whether different station installation protocols play a role in the inconsistency of attenuation estimations when different back azimuth event sets are utilized separately, we subset our data according to network and event back azimuth, invert each subset and compare the attenuation results. As shown in Fig. 4.6, inversion results using MCMC and linear inversion are very similar in our study, therefore, in this section, we explore the network effects using linear inversion for the sake of computation efficiency (Fig. 4.7).
As shown in Fig. 4.7A-C, the TA network covers the entire region. In the case that events from all back azimuths are utilized, the attenuation result is generally consistent with the inversion of the entire data set (Fig. 4.4, Fig. 4.6), though absolute Δt* values of the low attenuation block at the NMSZ location are slightly higher, and the northern and southern ends of NMSZ are constrained by lower attenuation materials. The difference of results using two different event sets holds to be similar with the variations when using all networks, as shown in Fig. 4.4 and Fig. 4.7.
The XO network covers the northern part of this region (Fig. 4.7D-F) and can only constrain the northern part of Reelfoot Rift. The general patterns are mostly consistent among results using different event sets recorded by XO network, with high attenuated areas between 91°W~89°W and being surrounded by low attenuated regions. The north of NMSZ is characterized as a high-attenuation area from results inverted using Kuril Islands events, whereas it is shown to be the boundary between a high attenuated area to the west and a low attenuated area to the east when Kuril Islands events are utilized. But this small scale difference could also be caused by smoothing settings in the linear
82 inversion. Nevertheless, at locations between 92°~90°W along longitude and 36.5°~38°N along latitude, results using XO network show high attenuation, which differs from results using TA network that indicate low attenuation within this region.
ZL network covers the northern and central of the west part of this region (Fig. 4.7G-I). Results using two different event sets diverge a lot from each other. Andes results show the NMSZ lies at the boundary between a NE-SW trend low attenuated to the west and W-E high attenuated materials to the east. In contrast, results using Kuril Islands events show the NMSZ lying within and is restricted by the lowest attenuation block in this area, with a belt of high attenuation surrounding it to the west. This feature constrained by the ZL network is supposed to contribute the most to the feature at this location by using all these three networks, as shown in Fig. 4.3 and Fig. 4.5.
In this part, we explored whether network deployments play in a role in attenuation results, which was reported in an experiment using different installation procedures [Bezada, Byrnes, and Z. Eilon, 2019], results in this section demonstrates that the difference between networks is smaller than the difference between azimuths. Consequently, we suspect event azimuth is the main driver leading to the different lateral variation patterns when event datasets are utilized separately.
Fig. 4.6: Δt* results using linear inversion
83 Fig. 4.7: Results using TA, XO, ZL network and event dataset from different back azimuth separately. (A-C) are results using TA network; (D-F) are results using XO network; (G-I) are results using ZL network.
84 4.4. Discussion
Previous tomography studies indicated a low velocity zone in the upper mantle beneath NMSZ. Although morphology of the LVZ is inconsistent across studies, the existence of it beneath or nearby the Reelfoot Rift has been widely believed. Hypotheses therein emerged proposing that the LVZ is a weak zone localizing stress and triggering seismic events within NMSZ. Addition of water into the upper mantle from the stalled Farallon slab [Chen et al., 2014; Nyamwandha et al., 2015], modification or emplacement of fertile mantle [Pollitz and Mooney, 2014] have been proposed to explain the observed LVZ. Regardless of the factor responsible for the LVZ, it is expected to be a weak zone that has low viscosity and thus high attenuation effects. Interestingly, direct attenuation analysis in our study demonstrates low attenuation beneath NMSZ, which contradicts the expectation that the LVZ beneath Reelfoot Rift stands for a weak zone. In this study, we show that using event datasets from two different back azimuth leads us different lateral variations patterns of the estimated Δt*. Results using events from Andes indicate NMSZ lies at transition locations between high attenuation to the west and low attenuation to the east of it, as shown in Fig. 4.4. This is consistent with observations in the study of Bezada et al. [2017]. However, using events from Kurils Islands revealed a different pattern, with the NMSZ occurring at locations that have the lowest attenuation. Additionally, the region to the east of the NMSZ is revealed to be of low attenuation when Andes events are utilized, whereas it is of high attenuation when only Kurils Islands events are used. This is at odds with our expectations, as teleseismic waves from all back azimuths should be sampling the same structures and thus should reveal a similar lateral variation, which is the case in a similar study in Morocco and Spain [Bezada et al., 2017]. In section 3, we further explored whether the deployment of networks plays a role in affecting the results when different event datasets are involved, and it turns out that the effect of networks is smaller than the event back azimuth. Although at this moment, the difference between
85 attenuation estimations using different event datasets is not yet fully understood, a deeper structure locating at depths that coincides with the divergence of ray paths originating from different back azimuth, or the propagation of waves along the slab at the Andes side, might be responsible for the results.
Another observation needs to be pointed out is, a high attenuated belt lies ~60m to the west of NMSZ and this feature is robust regardless of event sets and networks involved in inversions, as shown in Fig. 4.4 & Fig. 4.7. The proposed LVZ in some tomographic models [Chen et al., 2014; Nyamwandha et al., 2015] extends to this location at depths deeper than 150km, and may fulfil the weak zone hypothesis at this particular location. However, the question as to why this high attenuation band is to the west of the NMSZ and so much less concentrated than in the NMSZ needs to be explored. Given morphology of this proposed LVZ and its surrounding structures in the upper mantle were described significantly distinctly along depth across different studies, it hinders us from grasping a comprehensive sense of upper mantle structure and exploring the mechanism accounting for the lateral physical rheology variations within the Reelfoot Rift region. Additionally, recent SKS study [Nyamwandha and Powell, 2016] indicated a certain amount of anisotropy beneath the Mississippi embayment, tomography study [Wang et al., 2019] considering P wave azimuthal anisotropy revealed a much smaller scale of LVZ compared with isotropic tomography [Chen et al., 2014]. But the anisotropy effects on the velocity structure beneath and surrounding the Reelfoot Rift have yet not systematically discussed and addressed. Before diving into the mechanism of LVZ localizing stress and triggering earthquakes, a systematic tomography study exploring anisotropy contribution is highly encouraged. It will help us investigate whether the LVZ is robust and better constrain velocity structures in the upper mantle beneath Mississippi embayment. At this moment, we can not exclude the possibility that the imaged LVZ is artificially caused, if not all of it, by anisotropy at these regions.
86 Attenuation studies in other regions proposed a spatial correlation between great attenuation gradients and zones of intraplate deformation and seismicity [Bezada, 2017;Bezada and Smale, 2019]. Result in our study does not exactly follow this pattern. NMSZ occurs within a low attenuation zone and possibly being constrained by the largest gradients of the attenuation on the north and south ends. Along the strike of the NMSZ and within the Reelfoot Rift, results demonstrate the association between Δt* and occurrence of seismicity. In spite of the small lateral variation along the profile that is within the Reelfoot Rift, it conveys a very good consistency between Δt* and the occurrence of seismicity. An alternative possibility is the NMSZ was initiated at northernmost and southernmost locations of NMSZ where greatest attenuation gradients occur, which led to stress perturbations from earthquakes transmitting along fault networks and triggered the reactivation of the preexisting faults, which has been suggested before in other studies [e.g. Li, Liu, and Seth, 2009, 2016]. At this moment, we are not able to rule out other possible mechanisms such as river incision since last Ice age [Calais et al., 2010], stallation of Farallon slab in the lower mantle [Forte et al., 2007].
4.5. Conclusion
In this study, we investigated the lateral differential variations of Δt* in the Reelfoot rift region and its surroundings, provided insights into the physical rheology properties of the asthenosphere of this region. Δt* estimation of this region indicates the NMSZ occurs within a low attenuation region, this contradicts with our expectation if a low velocity zone lies beneath this region. Using event datasets from two different back azimuth leads to different lateral variation estimations, which is at odds with a previous similar study in Morocco and Spain [Bezada et al., 2017]. We further explored the contribution of event back azimuth and different network deployments, results demonstrated that the
87 difference between networks is smaller than event back azimuth. Although the mechanism accounting for the difference when using events from different back azimuth is not fully understood yet, a deeper anomaly coinciding with the divergence of ray paths, or angle between the ray path and slab strike, might be accountable.
In addition, morphology of the asthenosphere here is not shown to be consistent among different studies. Further study about tomography, especially by taking off anisotropy, are encouraged to be conducted to test the robustness of the LVZ and its morphology.
Chapter 5
Conclusion
In this thesis, we presented three projects exploring unsolved problems related to the occurrence of seismicity in three regions using different seismological methods. We relocated seismicity in the Alboran slab, constructed a tomographic velocity model in the northwestern South America subduction zone, and estimated the attenuation effects in the Reelfoot Rift.
Chapter 2: In this study, using grid-searching and the hypoDD program, data from the Spanish local network and the temporary PICASSO network, we provided high-resolution earthquake locations in the context of the velocity and thermal structure of the slab. Relocation results revealed distinct seismic clusters within the slab at a depth
88 range of 50-110km. The clusters concentrate in the pre-subduction middle levels of the slab, with the orientation of the overall seismogenic plane parallel to the pre-subduction slab surface. The existence of events above the tomographically-resolvable slab surface beneath the Betics and the absence of earthquakes in the pre-subduction top of the slab suggests that, as a result of the detachment process, the Alboran slab is significantly thinned at its shallower depths. We hypothesize the thinning process has advanced to the point of opening small gaps, and that the distribution of seismic clusters and gaps indicates the presence and absence of slab material, respectively.
Changes in seismicity rates and clustering concentrations along the strike at intermediate depths of the Alboran slab indicate different stages of development of the apparent shear zones, and thus suggests a detachment process that started in the north and is progressing toward the south. We conclude that the distribution of intermediate depth seismicity provides strong evidence for ongoing slab detachment and is a product of that process. If our interpretation is correct, then as a corollary, this active detachment process precludes the Alboran slab from providing a driving mechanism for initiating Atlantic subduction at the Gulf of Cadiz.
Chapter 3: In this study, we provide a well-resolved P wave velocity model in the NW South America subduction zone. The Maracaibo segment transitions from subhorizontal subduction at shallow depth into normal subduction (dip angle ~40°) at depths deeper than 100km; the Bucaramanga segment is subjected to subhorizontal subduction and extends ~600 km away from trench, and transitions into normal subduction (dip angle: ~75°) at depths deeper than 150 km. The continuous integrated feature of the Maracaibo and Bucaramanga segments at depths greater than 350km indicated they are both of Caribbean origin, with the ‘Caldas tear’ representing the southern edge of the subducted Caribbean plate, and this edge propagating southwards with depth. The Bucaramanga nest is hosted within the subducted Caribbean plate where there is an abrupt change in both dip and strike
89 at depths of ~150km. Therein highly accumulated stress should be available resulting from the bending in two different planes, a corollary is the high seismicity rate may be caused by the enhanced stress and not by excess dehydration. The Nazca plate is observed to overlap with the Caribbean plate between 5°~8°N. The Cartago and Pasto segments are normally subducted to the south of the ‘Caldas tear’. Subduction of EW- oriented weaknesses related to relic spreading centers might account for the fragmentation of the Nazca plate and drifting of the subducted fragments in different directions.
Chapter 4: In this study, we investigated the lateral differential variations of attenuation (Δt*) in the Reelfoot rift region and its surroundings, shedding light on the physical rheology properties of the asthenosphere of this region. Δt* estimation of this region indicates the NMSZ occurs within a low attenuation region, which contradicts our expectation if a low velocity zone lies beneath this region. An association between the attenuation and occurrence of seismicity is revealed; this conveys that the lateral variation of attenuation along the strike of the Reelfoot Rift region might be responsible for the NMSZ. Additionally, using event datasets from two different back azimuth leads to different lateral variation estimations, which is at odds with a previous similar study in Morocco and Spain [Bezada et al., 2017]. We further explored the contribution of event back azimuth and different network deployments, with results demonstrating that the difference between networks is smaller than event back azimuth. Although the mechanism accounting for the difference when using events from different back azimuth is not fully understood yet, a deeper anomaly coinciding with the divergence of ray paths, or angle between the ray path and slab strike, might be accountable. In addition, the morphology of the asthenosphere here is not shown to be consistent among different studies. Further tomographic imaging, especially by taking into anisotropy, are encouraged to be conducted to test the robustness of the LVZ and its morphology.
90 Reference
Adamek, S., Frohlich, C., & Pennington, W. D. (1988). Seismicity of the Caribbean-Nazca Boundary: Constraints on microplate tectonics of the Panama region. Journal of Geophysical Research: Solid Earth, 93(B3), 2053–2075. https://doi.org/10.1029/JB093iB03p02053
Aki, Keiiti. "Maximum likelihood estimate of b in the formula log N= a-bM and its confidence limits." Bull. Earthq. Res. Inst., Tokyo Univ. 43 (1965): 237-239.
Atwater, T. (1970). Implications of Plate Tectonics for the Cenozoic Tectonic Evolution of Western North America. GSA Bulletin, 81(12), 3513–3536. https://doi.org/10.1130/0016-7606(1970)81[3513:IOPTFT]2.0.CO;2
Bercovici, David, and Shun-ichiro Karato. "Theoretical analysis of shear localization in the lithosphere." Reviews in mineralogy and geochemistry 51.1 (2002): 387-420.
Bezada, M. J., Humphreys, E. D., Toomey, D. R., Harnafi, M., Dávila, J. M., & Gallart, J. (2013). Evidence for slab rollback in westernmost Mediterranean from improved upper mantle imaging. Earth and Planetary Science Letters, 368, 51–60.
Bezada, M. J., Humphreys, E. D., Davila, J. m., Carbonell, R., Harnafi, M., Palomeras, I., & Levander, A. (2014). Piecewise delamination of Moroccan lithosphere from beneath the Atlas Mountains. Geochemistry, Geophysics, Geosystems, n/a-n/a.
Bezada, M. J., M. Faccenda, and D. R. Toomey. "Representing anisotropic subduction zones with isotropic velocity models: A characterization of the problem and some steps on a possible path forward." Geochemistry, Geophysics, Geosystems 17.8 (2016): 3164-3189.
Bezada, M. J., A. Levander, and B. Schmandt. "Subduction in the southern Caribbean: Images from finite frequency P wave tomography." Journal of Geophysical - Research: Solid Earth 115.B12 (2010).
91 Bezada, M. J., and J. Smale. "Lateral variations in lithospheric mantle structure control the location of intracontinental seismicity in Australia." Geophysical Research Letters 46.22 (2019): 12862-12869.
Bezada, M. J. (2017). Insights into the lithospheric architecture of Iberia and Morocco from teleseismic body-wave attenuation. Earth and Planetary Science Letters, 478, 14-26. Billen, M. I. (2020). Deep slab seismicity limited by rate of deformation in the transition zone. Science Advances, 6(22), eaaz7692. https://doi.org/10.1126/sciadv.aaz7692 Bodmer, M., Toomey, D. R., Hooft, E. E. E., & Schmandt, B. (2018). Buoyant Asthenosphere Beneath Cascadia Influences Megathrust Segmentation. Geophysical Research Letters, 45(14), 6954–6962. https://doi.org/10.1029/2018GL078700 Bonnin, M., Nolet, G., Villaseñor, A., Gallart, J., & Thomas, C. (2014).
Multiple-frequency tomography of the upper mantle beneath the African/Iberian collision zone. Geophysical Journal International, 198(3), 1458–1473.
Boschman, L.M., van Hinsbergen, D.J.J., Torsvik, T.H., Spakman, W., Pindell, J.L., 2014. Kinematic reconstruction of the Caribbean region since the Early Jurassic. Earth Sci. Rev. 138, 102–136. https://doi.org/10.1016/j.earscirev.2014.08.007. Boutelier, D., and A. R. Cruden. "Slab breakoff: Insights from 3D thermo-mechanical analogue modelling experiments." Tectonophysics 694 (2017): 197-213.
Braeck, S., and Y. Y. Podladchikov. "Spontaneous thermal runaway as an ultimate failure mechanism of materials." Physical Review Letters 98.9 (2007): 095504.
Braile, L. W., W. J. Hinze, G. R. Keller, E. G. Lidiak, and J. L. Sexton (1986), Tectonic development of the New Madrid rift complex, Mississippi Embayment, North America, Tectonophysics, 131(1), 1–21. Brudzinski, Michael R., et al. "Global prevalence of double Benioff zones." Science 316.5830 (2007): 1472-1474. 92 Buforn, Elisa, et al. "Seismic sources on the Iberia-African plate boundary and their tectonic implications." Pure and Applied Geophysics 161.3 (2004): 623-646.
Buforn, E., A. Udías, and C. Pro. "Source mechanism studies of earthquakes in the Ibero-Maghrebian region and their tectonic implications." Journal of Seismology 20.4 (2016): 1075-1088.
Burke, K., 1988. Tectonic evolution of the Caribbean. Annu. Rev. Earth Planet. Sci. 16, 201–230. (A88-52285 22- 46). Annual Reviews, Inc., Palo Alto, CA.
Borrero, C., et al. "Geochemistry and tectonic controls of the effusive activity related with the ancestral Nevado del Ruiz volcano, Colombia." Geofísica internacional 48.1 (2009): 149-169.
Byrnes, J. S., Bezada, M., Long, M. D., & Benoit, M. H. (2019). Thin lithosphere beneath the central Appalachian Mountains: Constraints from seismic attenuation beneath the MAGIC array. Earth and Planetary Science Letters, 519, 297–307. https://doi.org/10.1016/j.epsl.2019.04.045
Bilham, R., 2014. Aggravated earthquake risk in south Asia: engineering versus human
nature. In: Wyss, M. (Ed.), Earthquake Hazard, Risk, and Disasters. Elsevier,
Amsterdam, pp. 103–141.
Bodin, T., Sambridge, M., Rawlinson, N., Arroucau, P., 2012a. Transdimensional to- mography with unknown data noise. Geophys. J. Int. 189, 1536–1556. https:// doi .org /10 .1111 /j .1365 -246X .2012 .05414 .x.
Bodin, T., Sambridge, M., Tkalˇci ´c, H., Arroucau, P., Gallagher, K., Rawlinson, N., 2012b. Transdimensional inversion of receiver functions and surface wave dispersion. J. Geophys.
Cafferky, S., Schmandt, B., 2015. Teleseismic P wave spectra from USArray and implications for upper mantle attenuation and scattering. Geochem. Geophys. Geosyst. 16, 3343–3361. http://dx.doi.org/10.1002/2015GC005993 93 Calais, Eric, et al. "Triggering of New Madrid seismicity by late-Pleistocene erosion." Nature 466.7306 (2010): 608-611.
Calais, Eric, and Seth Stein. "Time-variable deformation in the New Madrid seismic zone." Science 323.5920 (2009): 1442-1442.
Cramer, Chris H. "A seismic hazard uncertainty analysis for the New Madrid seismic zone." Engineering Geology 62.1-3 (2001): 251-266.
Corredor, Freddy. "Seismic strain rates and distributed continental deformation in the northern Andes and three-dimensional seismotectonics of northwestern South America." Tectonophysics 372.3-4 (2003): 147-166.
Cox, Randel T., and Roy B. Van Arsdale. "Hotspot origin of the Mississippi embayment and its possible impact on contemporary seismicity." Engineering Geology 46.3-4 (1997): 201-216.
Cox, Randel Tom, and Roy B. Van Arsdale. "The Mississippi Embayment, North America: a first order continental structure generated by the Cretaceous superplume mantle event." Journal of Geodynamics 34.2 (2002): 163-176.
Charles Langston, H. D. S. (2011). Northern Embayment Lithospheric Experiment. International Federation of Digital Seismograph Networks. https://doi.org/10.7914/SN/ZL_2011
Chen, C., Zhao, D., & Wu, S. (2014). Crust and upper mantle structure of the New Madrid Seismic Zone: Insight into intraplate earthquakes. Physics of the Earth and Planetary Interiors, 230, 1–14. https://doi.org/10.1016/j.pepi.2014.01.016 Chiarabba, Claudio, et al. "Subduction system and flat slab beneath the Eastern Cordillera of Colombia." Geochemistry, Geophysics, Geosystems 17.1 (2016): 16-27.
94 Chu, R., W. Leng, D. V. Helmberger, and M. Gurnis (2013), Hidden hotspot track beneath the eastern United States, Nat. Geosci., 6(11), 963–966.
Chiarabba, Claudio, et al. "Subduction system and flat slab beneath the E astern
Cordillera of Colombia." Geochemistry, Geophysics, Geosystems 17.1 (2016): 16-27.
Cortés, Martín, and Jacques Angelier. "Current states of stress in the northern Andes as indicated by focal mechanisms of earthquakes." Tectonophysics 403.1-4 (2005): 29-58.
Chertova, Maria V., Wim Spakman, and Bernhard Steinberger. "Mantle flow influence
on subduction evolution." Earth and Planetary Science Letters 489 (2018): 258-266.
Clark, S. A., et al. "Identification and tectonic implications of a tear in the South American plate at the southern end of the Lesser Antilles." Geochemistry, Geophysics, Geosystems 9.11 (2008).
DeMets, C. (2001), A new estimate for present-day Cocos-Caribbean plate motion: Implications for slip along the Central American volcanic arc, Geophys. Res. Lett., 28, 4043–4046.
DeMets, C., P. E. Jansma, G. S. Mattioli, T. H. Dixon, F. Farina, R. Bilham, E. Calais, and P. Mann (2000), GPS geodetic constraints on CaribbeanNorth America plate motion, Geophys. Res. Lett., 27, 437–440.
Domingues, A., Custódio, S., & Cesca, S. (2013). Waveform inversion of
small-to-moderate earthquakes located offshore southwest Iberia. Geophysical Journal International, 192(1), 248–259.
95 Duarte, J. C., Rosas, F. M., Terrinha, P., Schellart, W. P., Boutelier, D., Gutscher, M.-A., & Ribeiro, A. (2013). Are subduction zones invading the Atlantic? Evidence from the southwest Iberia margin. Geology, 41(8), 839–842.
Duggen, Svend, et al. "Magmatic evolution of the Alboran region: The role of subduction in forming the western Mediterranean and causing the Messinian Salinity Crisis." Earth and Planetary Science Letters 218.1-2 (2004): 91-108.
Duggen, Svend, et al. "Post-collisional transition from subduction-to intraplate-type magmatism in the westernmost Mediterranean: evidence for continental-edge delamination of subcontinental lithosphere." Journal of Petrology 46.6 (2005): 1155-1201.
Duretz, T., S. M. Schmalholz, and T. V. Gerya. "Dynamics of slab detachment." Geochemistry, Geophysics, Geosystems 13.3 (2012).
Davies, J. Huw, and Friedhelm von Blanckenburg. "Slab breakoff: a model of lithosphere detachment and its test in the magmatism and deformation of collisional orogens." Earth and Planetary Science Letters 129.1-4 (1995): 85-102.
Duarte, J. C., Rosas, F. M., Terrinha, P., Schellart, W. P., Boutelier, D., Gutscher, M.-A., & Ribeiro, A. (2013). Are subduction zones invading the Atlantic? Evidence from the southwest Iberia margin. Geology, 41(8), 839–842.
Dahlen, F.A., Hung, S.H., Nolet, G., 2000. Frechet kernels for finite-frequency traveltimes; I, Theory. Geophys. J. Int. 141, 157–174.
Eilon, Z., Fischer, K.M., Dalton, C.A., 2018. An adaptive Bayesian inversion for upper- mantle structure using surface waves and scattered body waves. Geophys. J. Int. 214, 232–253. https://doi .org /10 .1093 /gji /ggy137.
Ervin, C. Patrick, and L. D. McGinnis. "Reelfoot rift: Reactivated precursor to the Mississippi embayment." Geological Society of America Bulletin 86.9 (1975): 1287-1295.
96 Frankel, Arthur, Robert Smalley, and J. Paul. "Significant motions between GPS sites in the New Madrid region: Implications for seismic hazard." Bulletin of the Seismological Society of America 102.2 (2012): 479-489.
Forte, A. M., J. X. Mitrovica, R. Moucha, N. A. Simmons, and P. S. Grand (2007), Descent of the ancient Farallon slab drives localized mantle flow below the New Madrid Seismic Zone, Geophys. Res. Lett., 34, L04308, doi:10.1029/2006GL027895.
Faccenna, Claudio, et al. "Lateral slab deformation and the origin of the western Mediterranean arcs." Tectonics 23.1 (2004).
Florez, M. A., and G. A. Prieto. "Controlling Factors of Seismicity and Geometry in Double Seismic Zones." Geophysical Research Letters 46.8 (2019): 4174-4181.
Frohlich, Cliff. Deep earthquakes. Cambridge University Press, 2006.
Geissler, W. H., Matias, L., Stich, D., Carrilho, F., Jokat, W., Monna, S., et al. (2010). Focal mechanisms for sub-crustal earthquakes in the Gulf of Cadiz from a dense OBS deployment. Geophysical Research Letters, 37(18).
Gerya, Taras V., David A. Yuen, and Walter V. Maresch. "Thermomechanical modelling of slab detachment." Earth and Planetary Science Letters 226.1-2 (2004): 101-116.
González, Á. (2017). The Spanish National Earthquake Catalogue: Evolution, precision and completeness. Journal of Seismology, 21(3), 435–471.
Green, Harry W., and Chris Marone. "Instability of deformation." Reviews in mineralogy and geochemistry 51.1 (2002): 181-199.
Grevemeyer, Ingo, Luis Matias, and Sonia Silva. "Mantle earthquakes beneath the South Iberia continental margin and Gulf of Cadiz–constraints from an onshore-offshore seismological network." Journal of Geodynamics 99 (2016): 39-50.
97 Gutscher, Marc-André, Dominguez, S., Westbrook, G. K., & Leroy, P. (2009). Deep structure, recent deformation and analog modeling of the Gulf of Cadiz accretionary wedge: Implications for the 1755 Lisbon earthquake. Tectonophysics, 475(1), 85–97.
Gutscher, M. A., Malod, J., Rehault, J. P., Contrucci, I., Klingelhoefer, F., Mendes-Victor, L., & Spakman, W. (2002). Evidence for active subduction beneath Gibraltar. Geology, 30(12), 1071–1074.
Gutscher, M-A., et al. "The Gibraltar subduction: A decade of new geophysical data." Tectonophysics 574 (2012): 72-91.
Hacker, Bradley R., et al. "Subduction factory 2. Are intermediate depth earthquakes in - subducting slabs linked to metamorphic dehydration reactions?." Journal of Geophysical Research: Solid Earth 108.B1 (2003).
Hough, Susan E., and Morgan Page. "Toward a consistent model for strain accrual and release for the New Madrid seismic zone, central United States." Journal of Geophysical Research: Solid Earth 116.B3 (2011).
Hammond, W.C., Toomey, D.R., 2003. Seismic velocity anisotropy and heterogeneity beneath the Mantle Electromagnetic and Tomography Experiment (MELT) region of the East Pacific Rise from analysis of P and S body waves. J. Geophys. Res.-Solid Earth 108.
Hilst, Rob van der, and Paul Mann. "Tectonic implications of tomographic images of subducted lithosphere beneath northwestern South America." Geology 22.5 (1994): 451-454.
Hoernle, K., et al., 2002. Missing history (16–71 Ma) of the Galápagos hotspot: implications for the tectonic and biological evolution of the Americas. Geology 30 (9), 795–798. Hacker, Bradley R., et al. "Subduction factory 2. Are intermediate depth earthquakes in - 98 subducting slabs linked to metamorphic dehydration reactions?". Journal of Geophysical Research: Solid Earth 108.B1 (2003).
Heit, Benjamin, et al. "Tearing of the mantle lithosphere along the intermediate depth - seismicity zone beneath the Gibraltar Arc: The onset of lithospheric delamination." Geophysical Research Letters 44.9 (2017): 4027-4035. Hirata, Takayuki. "A correlation between the b value and the fractal dimension of earthquakes." Journal of Geophysical Research: Solid Earth 94.B6 (1989): 7507-7514.
Hu, Jiashun, and Lijun Liu. "Abnormal seismological and magmatic processes controlled by the tearing South American flat slabs." Earth and Planetary Science Letters 450 (2016): 40-51. John, T., et al. "Spontaneous thermal runaway as an earthquake mechanism at elevated pressure: insights from petrological and numerical studies." AGU Fall Meeting Abstracts. Vol. 1. 2007. Jung, Haemyeong, Harry W. Green Ii, and Larissa F. Dobrzhinetskaya. "Intermediate-depth earthquake faulting by dehydration embrittlement with negative volume change." Nature 428.6982 (2004): 545-549.
Jung, Haemyeong, et al. "Frictional sliding in serpentine at very high pressure." Earth and Planetary Science Letters 277.1 (2009): 273-279.
Kane, M. F., T. G. Hildenbrand, and J. D. Hendricks (1981), Model for the tectonic evolution of the Mississippi Embayment and its contemporary seismicity, Geology, 9(12), 563–568. Kelemen, Peter B., and Greg Hirth. "A periodic shear-heating mechanism for intermediate-depth earthquakes in the mantle." Nature 446.7137 (2007): 787-790.
Kellog, J. N., and V. Vega (1995), Tectonic development of Panama, Costa Rica, and the Colombian Andes: Constraints from global positioning system geodetic studies
99 and gravity, in Geologic and Tectonic Development of the Caribbean Plate Boundary in Southern Central America, edited by P. Mann, pp. 75–90, Geol. Soc. of Am., Boulder, Colo.
Kellogg, James N., and William E. Bonini. "Subduction of the Caribbean plate and basement uplifts in the overriding South American plate." Tectonics 1.3 (1982): 251-276.
Kellogg, J., Mora-Páez, H., 2016. Panama Arc-North Andes Collision: “Broken Indenter” Model From New GPS Velocity Field, IASPEI—Regional Assembly. Latin-American and Caribbean Seismological Commission-LACSC, San José.
Kellogg, J. N., Camelio, G. B. F., & Mora-Páez, H. (2019). Cenozoic tectonic evolution of the North Andes with constraints from volcanic ages, seismic reflection, and satellite geodesy. Andean Tectonics. https://doi.org/10.1016/b978-0-12-816009-1.00006-x
Kennett, Brian LN, E. R. Engdahl, and R. Buland. "Constraints on seismic velocities in the Earth from traveltimes." Geophysical Journal International 122.1 (1995): 108-124.
Kohlstedt, D. L., Brian Evans, and S. J. Mackwell. "Strength of the lithosphere: Constraints imposed by laboratory experiments." Journal of Geophysical Research: Solid Earth100.B9 (1995): 17587-17602.
Kufner, Sofia-Katerina, et al. "Zooming into the Hindu Kush slab break-off: A rare glimpse on the terminal stage of subduction." Earth and Planetary Science Letters 461 (2017): 127-140.
Johnson, P. R., I. Zietz, and W. A. Thomas (1994), Possible Neo-Proterozoic-Early Paleozoic grabens in Mississippi, Alabama and Tennessee, Geology, 22, 11–14.
Johnston, Arch C., and Eugene S. Schweig. "The enigma of the New Madrid earthquakes of 1811–1812." Annual Review of Earth and Planetary Sciences 24.1 (1996): 339-384. 100 Levandowski, W., Boyd, O. S., & Ramirez Guzmán, L. (2016). Dense lower crust - elevates long term earthquake rates in the New Madrid seismic zone. Geophysical - Research Letters, 43(16), 8499-8510.
Larmat, C., Maceira, M., Higdon, D. M., & Anderson, D. N. (2017). Joining statistics and geophysics for assessment and uncertainty quantification of three-dimensional seismic Earth models. Statistical Analysis and Data Mining, 10(5), 277–289. https://doi.org/10.1002/sam.11353
Lonsdale, Peter. "Creation of the Cocos and Nazca plates by fission of the Farallon plate." Tectonophysics 404.3-4 (2005): 237-264.
Mancilla, F. de L., Stich, D., Berrocoso, M., Martín, R., Morales, J., Fernandez-Ros, A., et al. (2013). Delamination in the Betic Range: Deep structure, seismicity, and GPS motion. Geology, 41(3), 307–310.
Malavé, Gustavo, and Gerardo Suárez. "Intermediate depth seismicity in northern - Colombia and western Venezuela and its relationship to Caribbean plate subduction." Tectonics 14.3 (1995): 617-628.
Mancilla, Flor de Lis, and Jordi Diaz. "High resolution Moho topography map beneath Iberia and Northern Morocco from receiver function analysis." Tectonophysics 663 (2015): 203-211.
Masy, J., A. Levander, and F. Niu. "Receiver Functions Imaging of the Moho and LAB in the Southern Caribbean plate boundary and Venezuela." AGUFM 2011 (2011): T11B-2320.
Martin, M., Wenzel, F. and CALIXTO Working Group, 2006. High-resolution teleseismic body wave tomography beneath SE-Romania-II. Imaging of a slab detachment scenario. Geophysical Journal International, 164(3), pp.579-595.
McCrory, P. A., Blair, J. L., Waldhauser, F., & Oppenheimer, D. H. (2012). Juan de Fuca slab geometry and its relation to Wadati-Benioff zone seismicity. Journal of
101 Geophysical Research: Solid Earth, 117(B9), B09306. https://doi.org/10.1029/2012JB009407
Meschede, M., Barckhausen, U., 2000. Plate tectonic evolution of the Cocos–Nazca spreading center. In: Silver, E.A., Kimura, G., Shipley, T.H. (Eds.), Proceedings of the Ocean Drilling Program, Scientific Results vol. 170. Federal Institute for Geosciences and Natural Resources, Hannover, Germany, pp. 1–10.
Mora, J. A., Oncken, O., Le Breton, E., Ibánez-Mejia, M., Faccenna, C., Veloza, G., … Mesa, A. (2017). Linking Late Cretaceous to Eocene Tectonostratigraphy of the San Jacinto Fold Belt of NW Colombia With Caribbean Plateau Collision and Flat Subduction. Tectonics, 36(11), 2599–2629. https://doi.org/10.1002/2017TC004612
McKenna, J., Stein, S., & Stein, C. A. (2007). Is the New Madrid seismic zone hotter and weaker than its surroundings? Special Paper of the Geological Society of
America, 425(12), 167–175. https://doi.org/10.1130/2007.2425(12) Montes, C., Bayona, G., Cardona, A., Buchs, D.M., Silva, C.A., Morón, S., Hoyos, N., Ramírez, D.A., Jaramillo, C.A., Valencia, V., 2012. Arc-continent collision and orocline formation: closing of the Central American seaway. J. Geophys. Res. 117, B04105. https://doi.org/10.1029/2011JB008959.
Mora-Paez, H., Kellogg, J., Freymueller, J., Mencin, D., Fernandes, R.M., Diederix, H., LaFemina, P., Cardona-Piedrahita, L., Lizarazo, S., Pelaez-Gaviria, J.R., Díaz-Mila, F., Bohorquez-Orozco, O.P., Giraldo-Londoño, L., Corchuelo-Cuervo, Y., 2019. Crustal deformation in the northern Andes – Space geodesy velocity field. J. South Am. Earth Sci. 89, 76–91.
Müller, R. D., Sdrolias, M., Gaina, C., & Roest, W. R. (2008). Age, spreading rates, and spreading asymmetry of the world’s ocean crust. Geochemistry, Geophysics, Geosystems, 9(4). https://doi.org/10.1029/2007GC001743.
102 Nyamwandha, C. A., and C. A. Powell (2016), Seismic anisotropy beneath the Mississippi Embayment and the New Madrid Seismic Zone: A study of shear wave splitting, J. Geophys. Res. Solid Earth, 121, 8239–8253, doi:10.1002/2015JB012729.
Nyamwandha, Cecilia A., Christine A. Powell, and Charles A. Langston. "A joint local and teleseismic tomography study of the Mississippi Embayment and New Madrid Seismic Zone." Journal of Geophysical Research: Solid Earth 121.5 (2016): 3570-3585.
Olugboji, T.M., Lekic, V., McDonough, W., 2017. A statistical assessment of seismic models of the U.S. continental crust using Bayesian inversion of ambient noise surface wave dispersion data. Tectonics 36, 1232–1253. https://doi .org /10 .1002 / 2017TC004468.
Okazaki, Keishi, and Greg Hirth. "Dehydration of lawsonite could directly trigger earthquakes in subducting oceanic crust." Nature 530.7588 (2016): 81-84.
Ogawa, Masaki. "Shear instability in a viscoelastic material as the cause of deep focus earthquakes." Journal of Geophysical Research: Solid Earth 92.B13 (1987): 13801-13810.
Paige, Christopher C., and Michael A. Saunders. "LSQR: An algorithm for sparse linear equations and sparse least squares." ACM Transactions on Mathematical Software (TOMS) 8.1 (1982): 43-71.
Palomeras, I., et al. "Finite frequency Rayleigh wave tomography of the western - Mediterranean: Mapping its lithospheric structure." Geochemistry, Geophysics, Geosystems 15.1 (2014): 140-160.
Pearce, F.D., Rondenay, S., Sachpazi, M., Charalampakis, M. and Royden, L.H., 2012. Seismic investigation of the transition from continental to oceanic subduction along the western Hellenic Subduction Zone. Journal of Geophysical Research: Solid Earth, 117(B7).
103 Pérouse, Eugénie, et al. "Active surface deformation and sub-lithospheric processes in the western Mediterranean constrained by numerical models." Geology 38.9 (2010): 823-826.
Poli, Piero, et al. "Earthquakes initiation and thermal shear instability in the Hindu Kush intermediate depth nest." Geophysical Research Letters 43.4 (2016): 1537-1542.
Poli, P., et al. "Complex rupture of the M 6.3 2015 March 10 Bucaramanga earthquake: Evidence of strong weakening process." Geophysical Journal International 205.2 (2016): 988-994.
Pollitz, Fred F., and Walter D. Mooney. "Seismic structure of the Central US crust and shallow upper mantle: Uniqueness of the Reelfoot Rift." Earth and Planetary Science Letters 402 (2014): 157-166.
Pondrelli, S., et al. "European–Mediterranean regional centroid-moment tensors: 1997–2000." Physics of the Earth and Planetary Interiors 130.1-2 (2002): 71-101.
Porritt, Robert W., Thorsten W. Becker, and Gaspar Monsalve. "Seismic anisotropy and slab dynamics from SKS splitting recorded in Colombia." Geophysical Research Letters 41.24 (2014): 8775-8783.
Prieto, Germán A., et al. "Seismic evidence for thermal runaway during intermediate depth earthquake rupture." Geophysical Research Letters 40.23 - (2013): 6064-6068.
Prieto, Germán A., et al. "Earthquake nests as natural laboratories for the study of intermediate-depth earthquake mechanics." Tectonophysics 570 (2012): 42-56.
Reinhardt, L. J., et al. "Tectonic denudation and topographic development in the Spanish Sierra Nevada." Tectonics 26.3 (2007).
Rietbrock, Andreas, and Felix Waldhauser. "A narrowly spaced double seismic zone in - the subducting Nazca plate." Geophysical Research Letters 31.10 (2004).
104 Rosell, O., Marti, A., Marcuello, A., Ledo, J., Queralt, P., Roca, E., & Campanya, J. (2011). Deep electrical resistivity structure of the northern Gibraltar Arc (western Mediterranean): evidence of lithospheric slab break-off. Terra Nova, 23(3), 179–186.
Royden, Leigh H. "Evolution of retreating subduction boundaries formed during continental collision." Tectonics 12.3 (1993): 629-638.
Rutter, E. H., S. Llana-Funez, and K. H. Brodie. "Dehydration and deformation of intact cylinders of serpentinite." Journal of Structural Geology 31.1 (2009): 29-43.
Sachpazi, M., Laigle, M., Charalampakis, M., Diaz, J., Kissling, E., Gesret, A., Becel, A., Flueh, E., Miles, P. and Hirn, A., 2016. Segmented Hellenic slab rollback driving Aegean deformation and seismicity. Geophysical Research Letters, 43(2), pp.651-658.
Sallarès, Valentí, et al. "Seismic evidence for the presence of Jurassic oceanic crust in the central Gulf of Cadiz (SW Iberian margin)." Earth and Planetary Science Letters 311.1-2 (2011): 112-123.
Schmandt, Brandon, and Eugene Humphreys. "Complex subduction and small-scale convection revealed by body-wave tomography of the western United States upper mantle." Earth and Planetary Science Letters 297.3-4 (2010): 435-445.
Schorlemmer, Danijel, Stefan Wiemer, and Max Wyss. "Variations in earthquake-size distribution across different stress regimes." Nature 437.7058 (2005): 539.
Singh, S. K., Bezada, M. J., Elouai, D., & Harnafi, M. (2016). Apparently-deep events in the Middle Atlas resolved to be shallow: Implications for lithospheric deformation. Tectonophysics, 691, Part B, 263–270.
Spakman, Wim, et al. "Puzzling features of western Mediterranean tectonics explained by slab dragging." Nature Geoscience 11.3 (2018): 211.
105 Spakman, Wim, and Rinus Wortel. "A tomographic view on western Mediterranean geodynamics." The TRANSMED atlas. The Mediterranean region from crust to mantle. Springer, Berlin, Heidelberg, 2004. 31-52.
Stich, Daniel, et al. "Kinematics of the Iberia–Maghreb plate contact from seismic moment tensors and GPS observations." Tectonophysics 426.3-4 (2006): 295-317.
Stock, J. M., & Lee, J. (1994). Do microplates in subduction zones leave a geological record? Tectonics, 13(6), 1472–1487. https://doi.org/10.1029/94TC01808 Syracuse, Ellen M., et al. "Multiple plates subducting beneath Colombia, as illuminated by seismicity and velocity from the joint inversion of seismic and gravity data." Earth and Planetary Science Letters 444 (2016): 139-149.
Li, Q., Liu, M., & Stein, S. (2009). Spatiotemporal Complexity of Continental Intraplate Seismicity: Insights from Geodynamic Modeling and Implications for Seismic Hazard EstimationSpatiotemporal Complexity of Continental Intraplate Seismicity. Bulletin of the Seismological Society of America, 99(1), 52–60. https://doi.org/10.1785/0120080005
Liu, Mian, and Seth Stein. "Mid-continental earthquakes: Spatiotemporal occurrences, causes, and hazards." Earth-Science Reviews 162 (2016): 364-386.
Lonergan, Lidia, and Nicky White. "Origin of the Betic Rif mountain belt." Tectonics - 16.3 (1997): 504-522.
Meighan, Hallie E., Uri Brink, and Jay Pulliam. "Slab tears and intermediate depth - seismicity." Geophysical Research Letters 40.16 (2013): 4244-4248.
Monna, S., Cimini, G. B., Montuori, C., Matias, L., Geissler, W. H., & Favali, P. (2013). New insights from seismic tomography on the complex geodynamic evolution of two adjacent domains: Gulf of Cadiz and Alboran Sea. Journal of Geophysical Research: Solid Earth, 118(4), 1587–1601. Moser, T. J. "Shortest path calculation of seismic rays." Geophysics 56.1 (1991): 59-67.
106 Taboada, Alfredo, et al. "Geodynamics of the northern Andes: Subductions and intracontinental deformation (Colombia)." Tectonics 19.5 (2000): 787-813.
Thomas, W. A. (1991), The Appalachian-Ouachita rifted margin of southeastern North America, Geol. Soc. Am. Bull., 103, 415–431.
Toomey, D., Solomon, S., Purdy, G., 1994. Tomographic imaging of the shallow crustal structure of the East Pacific rise at 9-Degrees-30′n. J. Geophys. Res.-Solid Earth 99, 24135–24157.
Toomey, Douglas R., Sean C. Solomon, and G. M. Purdy. "Tomographic imaging of the shallow crustal structure of the East Pacific Rise at 9° 30′ N." Journal of Geophysical Research: Solid Earth 99.B12 (1994): 24135-24157.
Trenkamp, R., J. N. Kellogg, J. T. Freymueller, and H. P. Mora (2002), Wide plate margin deformation, southern Central America and northwestern South America, CASA GPS observations, J. South Am. Earth Sci., 15(2), 157–171.
Utsu, T. "A method for determining the b-value in a formula logN= a-bm showing magnitude frequency relation for earthquakes, Hokkaido Univ. Geophys. Bull., 13, 99–103." Japanese with English abstract. 1965. van Benthem, Steven, et al. "Tectonic evolution and mantle structure of the Caribbean." Journal of Geophysical Research: Solid Earth 118.6 (2013): 3019-3036.
VanDecar, J. C., and R. S. Crosson. "Determination of teleseismic relative phase arrival times using multi-channel cross-correlation and least squares." Bulletin of the Seismological Society of America 80.1 (1990): 150-169. Vargas, Carlos A., and Paul Mann. "Tearing and breaking off of subducted slabs as the result of collision of the Panama Arc Indenter with northwestern South America." - Bulletin of the Seismological Society of America 103.3 (2013): 2025-2046.
Vernant, Philippe, et al. "Geodetic constraints on active tectonics of the Western
107 Mediterranean: Implications for the kinematics and dynamics of the Nubia-Eurasia plate boundary zone." Journal of Geodynamics 49.3-4 (2010): 123-129.
Villaseñor, A., Chevrot, S., Harnafi, M., Gallart, J., Pazos, A., Serrano, I., et al. (2015). Subduction and volcanism in the Iberia–North Africa collision zone from tomographic images of the upper mantle. Tectonophysics, 663, 238–249. von Blanckenburg, Friedhelm, and J. Huw Davies. "Slab breakoff: a model for syn collisional magmatism and tectonics in the Alps." Tectonics 14.1 (1995): 120-131.
Wagner, L. S., et al. "Transient slab flattening beneath Colombia." Geophysical Research Letters 44.13 (2017): 6616-6623.
Waldhauser, Felix, and William L. Ellsworth. "A double-difference earthquake location algorithm: Method and application to the northern Hayward fault, California." Bulletin of the Seismological Society of America 90.6 (2000): 1353-1368.
Weber, J. C., T. H. Dixon, C. DeMets, W. B. Ambeh, P. Jansma, G. Mattioli, J. Saleh, G. Sella, R. Bilham, and O. Perez (2001), GPS estimate of relative motion between the Caribbean and South American plates, and geologic implications for Trinidad and Venezuela, Geology, 29(1), 75–78.
Wiens, Douglas A., and Nathaniel O. Snider. "Repeating deep earthquakes: evidence for fault reactivation at great depth." Science 293.5534 (2001): 1463-1466.
Wortel, M. J. R., and Wim Spakman. "Subduction and slab detachment in the Mediterranean-Carpathian region." Science 290.5498 (2000): 1910-1917.
Wolaver, B. D., L. J. Crossey, K. E. Karlstrom, J. L. Banner, M. B. Cardenas, C. G. Ojeda, and J. M. Sharp (2012), Identifying origins of and pathways for spring waters in a semiarid basin using He, Sr, and C isotopes: Cuatrociénegas Basin, Mexico, Geosphere GES00849.
108 Yamasaki, Tadashi, and Tetsuzo Seno. "Double seismic zone and dehydration embrittlement of the subducting slab." Journal of Geophysical Research: Solid Earth 108.B4 (2003).
Zhang, T., Gordon, R. G., Mishra, J. K., & Wang, C. (2017). The Malpelo Plate Hypothesis and implications for nonclosure of the Cocos-Nazca-Pacific plate motion circuit. Geophysical Research Letters, 44(16), 8213–8218. https://doi.org/10.1002/2017GL073704.
109 Appendix
Fig. S1: A zoom in of the distribution of events and some stations. Orange dots are 58 events recorded by the PICASSO network (black triangles, stations mentioned in Figure.4 are outlines using red lines) between 2015 and 2017; blue dots are 1221 events recorded by the Spanish national network (white triangles) between 1997 and 2017.
110 Fig. S2: (a). Comparison of relocation results using gridsearch and 3D velocity model (red dots) vs. hypoDD and 1D velocity model (blue dots); (b). difference of longitudes (gridsearch result – hypoDD result) of same events; (c). difference of latitudes; (d). differences of depth
111 Fig. S3: Relocation results using different velocity models while keeping other parameters consistent. (a) Map view. (b) View from the west. Blue dots are relocation results using a 1D velocity model simplified from the 3D tomography model; red dots are relocation results using the AK135 velocity model.
Fig. S4: Station coverage of the subset extracted to test the effect of station coverage effect on relocation results (Fig. 2.6).
112 Fig. S5: Comparison of relocation results of the finger-shaped subset by using LQSR (blue dots) and SVD (red diamonds) respectively. (a) Map view. (b) View from the west
Fig.S6S6 : Comparison between relocation errors estimated by SVD for the finger- shaped subset (blue crosses), and the absolute value of the relocation difference using SVD and LSQR (red crosses) for each event in this subset. (a) Along longitude. (b)Along latitude. (c) Along depth
113 Fig.S7: Hypocenters before and after relocation after perturbing catalog hypocenters using standard deviation values of 5km (a-d), and 20km (e-h). Map views before (a, e) and after (b, f) relocation, as well as views from the west before (c, g) and after (d, h) relocation.
114 Table 2.1 PICASSO earthquakes before and after relocation
Catalog occurrence time (Y-M-D h-m-s) Catalog magnitude Catalog location(lon/°, lat/°, dep/km) Relocated location(lon/°, lat/°, dep/km) 2010-03-07 16:35:12.240 3.6 (-4.9779, 35.2030, -73.8) (-4.9714, 35.1761, -67.9958) 2010-05-02 16:12:50.200 3.1 (-4.7072, 35.7973, -68.9) (-4.6184, 35.6759, -84.6853) 2010-05-09 16:56:20.450 4.1 (-4.6811, 36.5506, -63.5) (-4.5638, 36.4223, -50.864) 2010-05-22 08:04:42.400 3.3 (-4.6164, 36.4745, -69.4) (-4.4976, 36.2373, -74.759) 2010-06-14 12:23:05.300 3.0 (-4.65, 35.9723, -80) (-4.5667, 35.8171, -83.8921) 2010-08-08 02:54:17.100 3.1 (-4.5761, 36.6602, -74.3) (-4.5322, 36.4603, -84.6853) 2010-11-05 09:43:47.380 3.6 (-5.1409, 36.1900, -64.3) (-5.117, 36.1455, -55.451) 2010-11-22 07:17:14.600 3.3 (-4.6651, 36.7300, -60) (-4.6508, 36.5566, -71.1785) 2010-12-30 00:33:35.730 3.6 (-4.4527, 36.6760, -97) (-4.4347, 36.6265, -82.7023) 2011-01-01 20:48:22 3.0 (-4.4816, 36.7987, -57) (-4.4248, 36.7768, -61.8263) 2011-01-06 23:54:11 2.7 (-4.0550, 36.8269, -66) (-3.9874, 36.8158, -67.2001) 2011-01-17 03:20:44 2.7 (-4.6608, 35.9024, -76) (-4.6088, 35.8607, -87.065) 2011-01-18 04:40:18.200 3.1 (-4.6752, 36.7628, -60) (-4.5676, 36.5636, -71.1785) 2011-02-09 11:37:18.540 3.6 (-4.5819, 35.6634, -100.8) (-4.5052, 35.6647, -91.8243) 2011-02-27 07:10:30.100 3.2 (-4.5138, 36.6142, -80) (-4.4582, 36.4672, -82.9001) 2011-03-01 10:26:24.270 3.7 (-4.7261, 35.1406, -97.9) (-4.5985, 35.1285, -85.082) 2011-03-12 12:48:48 3.4 (-4.7137, 35.7051, -65) (-4.7272, 35.6834, -79.3251) 2011-03-16 15:03:38.400 3.2 (-4.7869, 35.8392, -70) (-4.5318, 35.6756, -85.8752)
115 2011-03-26 11:58:04 2.6 (-4.5184, 36.3028, -72) (-4.4883, 36.3007, -76.5445) 2011-04-06 03:43:55.960 3.6 (-4.4858, 35.7093, -126.1) (-4.3792, 35.5753, -112.0074) 2011-04-27 03:22:05 2.7 (-4.5959, 35.8393, -83) (-4.5312, 35.7825, -80.914) 2011-06-15 01:31:50 2.7 (-4.4696, 36.5141, -82) (-4.4447, 36.4744, -84.6853)
2011-06-18 19:37:10 2.7 (-4.5897, 35.3352, -82) (-4.516, 35.3095, -88.2548) 2011-06-22 01:42:14.800 3.1 (-4.5992, 36.5772, -80) (-4.5277, 36.4639, -81.7085) 2011-07-02 09:52:50.100 3.0 (-4.5640, 36.5340, -71) (-4.5684, 36.4078, -84.0918) 2011-08-27 07:54:48.210 3.7 (-4.1800, 36.8303, -75.3) (-4.1724, 36.7882, -65.2109)
2011-09-22 08:42:11 2.6 (-4.6560, 36.7095, -65) (-4.6101, 36.6163, -78.5307) 2011-09-30 23:10:04 3.1 (-4.7645, 35.8036, -50) (-4.7179, 35.8247, -67.9958) 2011-10-21 05:39:09 3.1 (-4.6596, 36.3662, -83) (-4.56, 36.2755, -90.6345) 2011-10-28 19:37:10 3.4 (-4.6011, 36.5668, -50) (-4.5725, 36.4857, -59.834) 2011-11-02 23:47:38 3.0 (-4.6294, 35.6321, -70) (-4.723, 35.5982, -68.7914)
2011-11-07 21:20:08 2.6 (-5.0207, 35.2420, -86) (-5.0442, 35.196, -68.3936) 2012-01-10 06:38:28 2.7 (-4.5503, 36.3154 , -74) (-4.5064, 36.2645, -69.1893) 2012-02-03 17:46:20 2.6 (-4.4377, 36.6350, -68) (-4.5137, 36.5779, -71.9741) 2012-03-09 12:22:28 2.7 (-4.2561, 36.7809, -66) (-4.2179, 36.7432, -52.4603) 2012-04-02 22:59:33.900 3.0 (-4.6557, 36.3192, -64.9) (-4.5565, 36.0907, -70.7806) 2012-04-10 19:22:58.300 3.2 (-4.7179, 36.0006, -88.4) (-4.5581, 35.7428, -98.36)
2012-04-12 19:41:30.200 3.4 (-4.6358, 35.9932, -80) (-4.5093, 35.7281, -93.0142) 2012-05-16 11:19:57 3.1 (-4.6033, 36.0516, -81) (-4.539, 36.0091, -89.0481)
116 2012-06-21 16:54:10 2.5 (-4.6560, 35.7656, -92) (-4.6405, 35.7086, -97.568) 2012-06-22 12:10:13 3.1 (-4.5927, 35.7105, -81) (-4.5453, 35.6467, -83.6946) 2012-07-20 05:32:35 3.0 (-4.2115, 36.6966, -58) (-4.1729, 36.7356, -59.4356) 2012-07-21 01:06:02 3.3 (-4.6653, 35.8937, -65) (-4.6446, 35.8064, -78.3396) 2012-07-26 03:10:41 2.8 (-4.6487, 36.0408, -72) (-4.5749, 35.9712, -92.2209) 2012-07-28 16:49:41 3.2 (-4.0593, 36.8239, -63) (-3.9774, 36.8972, -61.8263) 2012-08-08 02:36:34 2.6 (-4.5955, 36.4821, -56) (-4.5495, 36.5998, -68.3936) 2012-08-13 13:13:10 3.4 (-4.5088, 35.0235, -68) (-4.491, 35.0248, -80.914) 2012-08-17 23:42:03 2.6 (-5.1059, 35.1693, -78) (-5.0753, 35.2811, -59.0371) 2012-09-04 00:29:56 2.5 (-4.4533, 36.5849, -74) (-4.4126, 36.5703, -73.764) 2012-09-20 04:04:33 2.8 (-4.5646, 36.6759, -78) (-4.5538, 36.6451, -76.9418) 2012-09-30 12:13:42 2.5 (-4.5234, 36.8383, -58) (-4.4988, 36.826, -55.451) 2012-10-28 20:43:44 3.3 (-4.4558, 36.5609, -87) (-4.6107, 36.4822, -64.813) 2012-11-02 21:13:40 2.6 (-4.4817, 35.0845, -71) (-4.4401, 35.0645, -97.568) 2013-01-14 02:00:55 2.8 (-4.5719, 36.3583, -88) (-4.9584, 36.2127, -69.9849) 2013-01-16 02:32:45 3.3 (-4.7780, 36.0761, -83) (-5.0809, 35.939, -52.4603) 2013-01-26 11:36:43 4.2 (-4.7192, 35.1209, -52) (-4.9272, 35.0655, -71.3806) 2013-02-28 17:45:59 4.2 (-5.0207, 35.8507, -62) (-5.2892, 35.6722, -73.1677)
117