Slab1.0: A three-dimensional model of global subduction zone geometries

Authors


Abstract

[1] We describe and present a new model of global subduction zone geometries, called Slab1.0. An extension of previous efforts to constrain the two-dimensional non-planar geometry of subduction zones around the focus of large earthquakes, Slab1.0 describes the detailed, non-planar, three-dimensional geometry of approximately 85% of subduction zones worldwide. While the model focuses on the detailed form of each slab from their trenches through the seismogenic zone, where it combines data sets from active source and passive seismology, it also continues to the limits of their seismic extent in the upper-mid mantle, providing a uniform approach to the definition of the entire seismically active slab geometry. Examples are shown for two well-constrained global locations; models for many other regions are available and can be freely downloaded in several formats from our new Slab1.0 website, http://on.doi.gov/d9ARbS. We describe improvements in our two-dimensional geometry constraint inversion, including the use of ‘average’ active source seismic data profiles in the shallow trench regions where data are otherwise lacking, derived from the interpolation between other active source seismic data along-strike in the same subduction zone. We include several analyses of the uncertainty and robustness of our three-dimensional interpolation methods. In addition, we use the filtered, subduction-related earthquake data sets compiled to build Slab1.0 in a reassessment of previous analyses of the deep limit of the thrust interface seismogenic zone for all subduction zones included in our global model thus far, concluding that the width of these seismogenic zones is on average 30% larger than previous studies have suggested.

1. Introduction

[2] Hayes et al. [2009] combined a collection of complementary data sets to image the two-dimensional structure of subduction zone interfaces in a variety of locations around the world's major convergent margins. Here we follow the same approach as used in their study to produce a suite of 2D profiles for each major global subduction zone, which we subsequently combine to form three-dimensional surfaces of subduction geometry.

[3] For a given location, we compile a dense system of 2D sections evenly spaced along the strike of the margin, and then interpolate between these sections to create a 3D surface of subduction interface geometry, which incorporates the structural variation of the slab in both the downdip and along-strike directions. We show examples of three-dimensional surfaces for several subduction zone segments, including South America and the Kamchatka-Kuril-Japan arc. We explore their data inputs, model sensitivities and uncertainties and show how this approach represents an improvement over existing 3D subduction geometry models.

[4] Our new model (Figure 1), which we call Slab1.0, is unique in Geoscience literature not only because of the systematic and global nature of its application (many other subduction zone models focus on the geometry of a specific, regional interface rather than applying their methods to subduction zones worldwide) but predominantly because of the large number of independent and complimentary data sets it incorporates to solve for non-planar subduction geometry. By combining high-accuracy seismicity (Engdahl et al. [1998], and updates) and moment tensor (gCMT, http://www.globalcmt.org) catalogs, we ensure the seismogenic zone seismicity we fit for geometry occurs on or near the subduction interface. Similarly, by including active source seismic data sets that image the shallow section of subducting slabs, we can constrain the geometry of sections of these interfaces that are otherwise aseismic. Third, the incorporation of deep seismicity (where all earthquakes occur within subducted slabs, regardless of mechanism) allows us to model the true non-planar nature of subduction zone interfaces as slabs turn over and descend into the mantle. Finally, we also accurately image the geometry of the slab at the trench via the inclusion of high resolution bathymetric (from the plate boundary files of Tarr et al. [2010]; and the Marine Geoscience Data System (MGDC) bathymetry database, http://www.marine-geo.org) and sediment thickness (from the National Geophysical Data Center (NGDC) global sediment thickness map; http://www.ngdc.noaa.gov/mgg/sedthick/sedthick.html) data sets. It is the combined use of these data sets, and the attention paid to the detailed geometry over the shallowest portions of each slab (from their trenches through the base of the thrust interface seismogenic zone) that distinguishes Slab1.0 from other major global subduction geometry compilations [e.g., Gudmundsson and Sambridge, 1998; Syracuse and Abers, 2006], which focus on contouring seismicity catalogs alone, at a lower resolution (every 50km in depth) than attempted here, predominantly to image the form of the subduction zone in the upper mantle and deeper.

Figure 1.

Global distribution of three-dimensional subduction zone geometry models included in Slab1.0 to date. Orange shaded areas outlined in red delineate the current extent of each regional model. Red lines describe global plate boundaries, from Tarr et al. [2010]. Base map shaded bathymetry (here and throughout) from Gebco 2008, version 20091120, http://www.gebco.net.

[5] Slab1.0 has applications in a variety of seismological analyses, including seismic hazard studies (e.g., Global Earthquake Model (GEM), http://globalquakemodel.org), tsunami simulations [e.g., Wang and He, 2008], and earthquake source inversions [e.g., Wald et al., 2008; Hayes and Wald, 2009; Moreno et al., 2009]. Indeed, the model has been used in several published studies to date: for numerical modeling of subduction zones [e.g., Mahadevan et al., 2010]; to compare the locations of slow earthquakes to subduction geometries [Beroza and Ide, 2011]; to help constrain geometries for earthquake source inversions [e.g., Newman et al., 2011; Hayes, 2011]; and to help resolve the well-known moment-dip trade-off [Kanamori and Given, 1981] in moment tensor inversions for shallow subduction thrust earthquakes [Duputel et al., 2011]. Having compiled 3D surfaces for major global subduction zone interfaces, and aggregating associated seismic catalogs of well-determined subduction-related events for each, we discuss the potential uses of the model for new seismological studies of subduction zone seismogenesis, including conducting a preliminary analyses of spatial and temporal strain release patterns and thrust interface seismogenic zone limits. We show that by using this new and continually improving global database, we can gain a better understanding of the global subduction zone process.

2. Procedure to Compute Three-Dimensional Interfaces

[6] We follow the approach of Hayes et al. [2009] to constrain subduction geometry in two dimensions using a combined data set consisting of historic earthquake catalogs (global Centroid Moment Tensor (gCMT); National Earthquake Information Center Preliminary Determination of Epicenters (NEIC PDE)), the global relocation catalog of Engdahl et al. [1998] (hereafter referred to as EHB), locations of trench breaks on the seafloor (from the plate boundary files of Tarr et al. [2010] and the Marine Geoscience Data System (MGDC) bathymetry database, http://www.marine-geo.org), trench sediment thicknesses interpreted from the National Geophysical Data Center (NGDC) global sediment thickness map (http://www.ngdc.noaa.gov/mgg/sedthick/sedthick.html), and interpretations of shallow slab interfaces from active source seismic data collected across trenches (see http://earthquake.usgs.gov/research/data/slab for references). Active source data sample the geometry of the downgoing plate both in- and out-board of the trench, to make certain slab geometries are consistent with outer-trench bathymetry. A vertical probability density function (PDF) is associated with each data point used to constrain slab geometry; the width of this PDF is based on the location uncertainty reported in the respective data catalog, or on an assumed uncertainty related to the translation of time to depth for active source seismic data. Data are filtered to include only those events with well-constrained depths and with focal mechanisms indicative of shallow-dip thrust faulting. The strike of each section is computed by averaging the strikes of the shallow-dipping nodal plane of local focal mechanisms that pass filter criteria, and is used to construct a 2D profile that samples the slab over a region ± 100 km in a direction perpendicular to the profile. As in the work of Hayes et al. [2009], intermediate-depth earthquake data are shifted horizontally by 10 km in a direction away from the trench to account for their occurrence within the slab and beneath the interface, rather than on its surface. All filtered data are then fit with a non-planar profile constructed from Hermite Splines, interpolating between (1) a minimum norm polynomial of order 2–3 to shallow (<80 km) data, and (2) a minimum norm polynomial of order 3–4 to intermediate-depth data [Hayes et al., 2009].

[7] To extend this approach to three dimensions, we sweep along the trench of a given subduction zone, sampling two-dimensional geometry every 10 km along strike. We then combine the resulting suite of 2D cross-sections (e.g., Figure 2; sampled into discrete points in the downdip direction) into three dimensions using the Generic Mapping Tools (GMT) blockmean and surface algorithms [e.g., Wessel and Smith, 1991]. These routines smooth our irregularly distributed system of (longitude, latitude, depth) triplets onto a regular grid (with node spacing of 0.5°), and create a continuous surface of these points in three dimensions. We subsequently restrict the dimensions of this grid to the limits of our two-dimensional profiles, such that the grid contains data only where the slab geometry is directly sampled. We have also conducted sensitivity tests of input parameters used in these interpolation algorithms (e.g., internal tension parameter and 2D sample spacing; see auxiliary material) to ascertain their most appropriate values for our purposes.

Figure 2.

Schematic describing the construction of the Tonga-Kermadec slab three-dimensional geometry model from a collection of two-dimensional profiles. Green stars on the top surface represent the beginning and end of each two-dimensional profile, which extend beyond the seismicity sampled within each profile, and thus may extend beyond the slab interpreted from the profile itself. Profiles are taken every 10 km along the strike of the trench (red), in the direction of strike (black arrow). Three representative 2D profiles are shown; red dashed lines are the best fitting 2D non-planar geometries overlain on background seismicity from the EHB catalog (gray circles) within 100 km of each plane.

[8] In extending our 2D approach to the third dimension, one significant challenge faced was dealing with the irregularity of active source data sampling along the strike of a given subduction zone. In some areas, the shallow slab location is well constrained by active source data, while several hundred kilometers along strike where data does not exist, our understanding of the shallow geometry of the same slab relies on interpolations between the trench location and deeper earthquake hypocenters. This variability can therefore result in an interpolated 3D shallow slab that oscillates up and down depending on data coverage. As Hayes et al. [2009] showed, such an oscillation is likely artificial (the result of a lack of data over the shallow portion of the slab in some 2D profiles) thus needs to be removed. To address this issue, we construct regional active source seismic data profiles by averaging existing active source data interpretations along the strike of each subduction zone (auxiliary material). Comparisons of active source seismic data profiles (Figure S3 of the auxiliary material) for a given subduction zone show high correlation along strike, suggesting minimal variability along strike in the depth of this shallow region of the slab; as such, an average profile (applied with a lower weight than ‘real’ active source data) helps guide the form of the shallow 2D polynomial more accurately than the complete absence of active source seismic data. In regions where existing active source data profiles exhibit significant along-strike variability (e.g., Aleutians/Alaska), average profiles are constructed over sub-regions in which data consistency is evident (auxiliary material), and are used to aid geometry constraint only within each specific sub-region.

[9] This procedure results in a three-dimensional geometry for the subduction zone in question. We repeat the approach at each subduction zone well sampled by our data set, giving a global model that represents the three-dimensional geometry of approximately 85% of subduction zones with documented thrust-interface earthquakes worldwide by plate boundary length (Figure 1). For a more detailed discussion of the differences between this model and our previous studies of 2D subduction geometry, see Text S1 of the auxiliary material. The current status of Slab1.0 can be tracked at, and model subsets downloaded from, the URL listed at the end of this manuscript. Here, we provide the model itself in multiple formats (depth, strike and dip grids in ascii and netCDF), contours of each slab surface (ascii, Google Earth KMZ, ArcGIS), text files describing the extent of the model, and a collection of other files and figures related to each subduction zone analyzed.

3. Slab1.0 Examples

[10] We show examples of several subduction zone segments constrained via this approach. Each set of figures shows the model itself colored and contoured according to depth, locations of teleseismically recorded earthquakes, local active source seismic data sets, and the start and end points of each 2D profiles superimposed onto the model. We also show cross-sections through the slab surfaces, and the average differences in depth between the slab and earthquake locations used to calculate the model, a qualitative estimate of model uncertainty and/or slab complexity.

[11] Figure 3 shows our 3D Pacific slab surface model for the Kamchatka/Kuril/Japan subduction zone. This slab reaches depths upwards of 600 km, potentially reaching the base of the transition zone along its entire length from southern Kamchatka to southern Honshu. The interface is very well defined seismically, particularly in the shallow (<80 km) and mid-deep mantle depth intervals, and sinks into the mantle at a relatively shallow angle, such that its deepest extent beneath northern China occurs over 1200 km from the trench.

Figure 3.

Three-dimensional surface for the Kamchatka-Kuril-Japan subduction zone, built via interpolation between approximately 300 individual two-dimensional profiles along the strike of the subduction zone. (a) The 3D surface, colored by depth-to-interface, and contoured at 20 km intervals. Green stars indicate the ends of the 2D profiles used. Red diamonds outlined in white show the locations of active source seismic data incorporated into our model. Grey circles show seismicity from the EHB catalog, sized by magnitude. (b) The same surface, without data overlays. Black dashed lines labeled A-C refer to cross-sections in Figure 5.

[12] On average, subduction-related earthquake depths should be slightly greater than the depth of the slab surface at the same location, as such events occur either on or close to the plate boundary interface itself, or within the subducting plate beneath the interface. Where the Pacific slab is active seismically, our model matches average earthquake depth well (Figure 4). In the intermediate depth levels (100–300 km), where seismicity is sparse, the average misfit increases to between −40 and −80 km. This misfit could indicate that our algorithm underestimates the depth of the slab in these regions, or that the Pacific slab is seismically active throughout its thickness (e.g., evidence for double seismic zones in the Pacific Plate beneath Japan) [Suzuki et al., 1983], and as such deeper zones of earthquakes bias our observed difference. In reality, the cause is probably a result of a combination of these two explanations, as is evident in Figure 5, a selection of cross-sections through the slab taken at several locations along the strike of the trench. We note that in the absence of data, our 3D surface algorithm [Wessel and Smith, 1991] is free from local constraint and interpolates between surrounding data based on an assigned tension value (auxiliary material). A tension value equal to zero, as used in our approach, results in a minimum curvature surface, thus reducing the deviation of this surface from surrounding data values.

Figure 4.

A comparison between the depth of our modeled slab interface for the Kamchatka-Kuril-Japan subduction zone, and the average depth of earthquakes within each grid cell used to invert for the geometry of the slab, contoured at 20 km intervals. Negative values (blue-yellow colors) indicate regions where the slab is on average shallower than seismicity; positive values (red-orange colors) where the slab is deeper. In Figure 4a, the difference grid is overlain with the earthquake hypocenters used to invert for the slab geometry.

Figure 5.

Representative two-dimensional cross-sections of the inverted three-dimensional geometry of the Kamchatka-Kuril-Japan subduction zone. Locations A-C are shown in Figure 3. In each figure, the red dashed lines represent the best fitting non-planar geometries, overlain on background seismicity from the EHB catalog (gray circles) within 100 km of each line, and shallow subduction-related thrust earthquakes (yellow CMT solutions) used to constrain the slab geometry. All earthquakes deeper than 80 km are used to help constrain the deep form of the slab interface. Red diamonds in the shallowest portion of the subduction zone represent active source seismic data used in the inversion process.

[13] Figure 6 shows our 3D Nazca slab surface for the South American subduction zone. In southern Chile, the extent of the slab is not well imaged, due to a lack of significant subduction-related seismicity; as a result, the slab can only be traced to depths of ∼40 km adjacent to the Chile triple junction using locally collected active source seismic profiles [Krawczyk et al., 2006; Loreto et al., 2007; Rubio et al., 2000; Scherwath et al., 2009; von Huene et al., 1997]. Moving northward, the slab is imaged to increasingly greater depths as it becomes more active seismically. In northern Chile, the flat-slab region between 30°S and 34°S is imaged well, as demonstrated in cross-sectional view in Figure 7, as is the subsequent rollover to transitional zone depths north of 30°S beneath Argentina and Bolivia. We note that by using our relatively automated approach and our combined, filtered catalog, we reproduce the Benioff zone contours of the South American slab inferred from other detailed studies conducted in this region extremely well [e.g., Bevis and Isacks, 1984; Pardo et al., 2004; Tassara et al., 2006; Anderson et al., 2007].

Figure 6.

Three-dimensional surface for the South America subduction zone, built via interpolation between approximately 600 individual two-dimensional profiles along the strike of the subduction zone. (a) The 3D surface, colored by depth-to-interface, and contoured at 20 km intervals. Green stars indicate the ends of the 2D profiles used. Red diamonds outlined in white show the locations of active source seismic data profiles incorporated into our model. Grey circles show seismicity from the EHB catalog, sized by magnitude. (b) The same surface, without data overlays. Black dashed lines labeled A-C refer to cross-sections in Figure 7.

Figure 7.

Representative two-dimensional cross-sections of the inverted three-dimensional geometry of the South America subduction zone. Locations A-C are shown in Figure 6. In each figure, the red dashed lines represent the best fitting non-planar geometries, overlain on background seismicity from the EHB catalog (gray circles) within 100 km of each line, and shallow subduction-related thrust earthquakes (yellow CMT solutions) used to constrain the slab geometry. All earthquakes deeper than 80 km are used to help constrain the deep form of the slab interface. Red diamonds in the shallowest portion of the subduction zone represent active source seismic data used in the inversion process.

[14] Significant slab flattening is also imaged adjacent to the Nazca Ridge (Figures 6 and 7), extending over a broad region to the north and south of the downdip projection of the ridges' bathymetric expression. In this section of the subduction zone, and in particular over the region north of the Nazca Ridge, the short-wavelength features seen in our 3D slab surface and in our slab-earthquake misfit plots (Figure 8) indicate difficulty fitting the true subduction interface as a result of insufficient seismicity in a region of complex geometry. Short wavelength complexity may also reflect artifacts resulting from the interpolation algorithm itself, and how this approach models regions of sparse or irregularly sampled data.

Figure 8.

A comparison between the depth of our modeled slab interface for the South America subduction zone, and the average depth of earthquakes within each grid cell used to invert for the geometry of the slab, contoured at 20 km intervals. Negative values (blue-yellow colors) indicate regions where the slab is on average shallower than seismicity; positive values (red-orange colors) where the slab is deeper. In Figure 8a, the difference grid is overlain with the earthquake hypocenters used to invert for the slab geometry.

[15] Further north between northern Peru and Colombia, the Nazca slab remains fairly shallow (imaged to depths of ∼100–200 km), though it may sink deeper, as evidenced by a series of deep earthquakes beneath Colombia near 1.5°S, −72.5°W at the northern extent of our model.

4. Previous Studies of Global Subduction Zone Geometry

[16] As discussed by Hayes and Wald [2009], and Hayes et al. [2009], a variety of previous studies have attempted to model Wadati-Benioff Zone (WBZ) geometry in subduction zones; we focus on comparisons to other global, publically available models here [Gudmundsson and Sambridge, 1998; Syracuse and Abers, 2006]. Other global or partially global models focused on deeper slab structure, and not available electronically, include Bevis and Isacks [1984], and Rhodes and Davies [2004].

[17] Gudmundsson and Sambridge [1998] (hereafter GS98) produce three-dimensional surfaces of global subduction zones by contouring earthquake catalog data. Slab contours are produced every 50 km in depth, and used to estimate slab volume in the mantle, which can be subsequently used to calculate the effects of slabs on earthquake travel times, particularly useful for global tomography studies. Syracuse and Abers [2006] (hereafter SA06) hand-contoured seismicity from the EHB catalog for slabs with volcanic arcs to explore the relationship between the depth to the top of the slab beneath each arc, other subduction parameters such as slab dip and plate age, and the processes that drive arc magmatism.

[18] The major distinction between these models and Slab1.0 is the focus in both GS98 and SA06 on mantle geometries, and their lack of definition in shallow subduction zones (no resolution between 0 and 50 km for GS98; seafloor and 50 km for SA06), versus the detailed shallow geometries available in Slab1.0. As this shallow environment hosts the greatest hazard from large subduction zone earthquakes, this makes Slab1.0 more applicable for hazard-related studies. Furthermore, in Slab1.0 we filter earthquake data sets to remove both poorly located events and those earthquakes without thrust mechanisms, thereby assuring we are focusing on seismicity (and thus structure) associated with the subduction process. No other global subduction zone geometry model includes this step.

[19] In Figures S4a-S4c of the auxiliary material, we show a detailed comparison of these three models by computing the slab-normal distance between subduction-related earthquakes (from the filtered catalog compiled to produce Slab1.0), and the surface of the slab in each model. In theory, as shallow subduction-related earthquakes occur on or near the plate interface, these distances should cluster around zero in the seismogenic zone. For deeper regions, where seismicity occurs within the subducting plate, these distances should be close to zero, but biased toward negative values (in a positive-up reference frame relative to the slab surface). How well each model matches surrounding seismicity can thus be evaluated by comparing the mean and standard deviation of these offsets. Figures S4a-S4c of the auxiliary material shows that, for the majority of regions and data sets (separated into shallow earthquakes, all earthquakes, and all data), Slab1.0 either performs better than or comparable to GS98 and AS06. The poorest data match in Slab1.0 is for the Ryukyu slab beneath southwest Japan. Here, sparse seismicity and complex subduction zone structure [e.g., Ide et al., 2010] make the slab geometry difficult to model with an automated approach. In all models except for Ryukyu, Slab1.0 matches shallow earthquake data better than GS98 and AS06.

5. Modeling Slabs Routinely and Globally

[20] Another distinct advantage of Slab1.0 is its automated nature, allowing the approach to be applied anywhere globally that subduction zones are seismically well defined. In its current form, Slab1.0 does not attempt to incorporate what are often referred to as segment boundaries (generally bathymetric features of the oceanic plate proposed to arrest ruptures of great earthquakes) because these should be reflected in the geometry of existing seismicity if they are physiographically prominent. Similarly, Slab1.0 does not currently incorporate segmentation or tears in slabs [e.g., Chen and Brudzinski, 2001; Miller et al., 2006; Ide et al., 2010], under a similar assumption that abrupt changes in slab geometry should be reflected in the geometry of local seismicity, and thus captured by our approach. However, it remains to be seen how much of a damping effect the width and along-strike sampling of each 2D cross-section used in our approach plays on the representation of such features. These regions are thus ideal locations to test the variation of these parameters in the future. For the Tonga-Kermadec slab in particular, shown by Chen and Brudzinski [2001] to be detached at depth into at least two fragments spatially distinct from the main slab, a more complicated approach specific to this region would be necessary to capture the true complexity of the slab at depth.

[21] We note that, as with any model, Slab1.0 is not perfect. The model performs poorly fitting overturned slabs (i.e., where slabs turn back toward the trench with increasing depth; e.g., central Izu-Bonin Arc), an issue we hope to correct in the future by investigating more complex interpolating algorithms [e.g., Hale, 2009]. This model also does not (and cannot, in its current form) define slab geometries beyond their seismic definition in the deep mantle. As a result, Slab1.0 does not include descriptions of slabs penetrating the mantle transition zone, such as those proposed from tomographic studies beneath the Indonesian Islands [e.g., Richards et al., 2007]. Furthermore, this means that the edges of each slab model are governed by the extent of each slabs' seismicity, and may not reflect the true spatial dimensions of the slab in the mantle. Slab1.0s lack of resolution where slabs are aseismic also means models are undefined for several key subduction zones such as Cascadia, the Caribbean, the Hellenic Arc, and the Pacific Plate beneath the North Island of New Zealand. In areas such as these, and poorly fit regions such as the Ryukyu slab under southwestern Japan discussed in section 4, the addition of regional, high-resolution seismicity catalogs and other data sets (e.g., receiver functions) would likely significantly improve Slab1.0. The incorporation of double-difference relocated seismicity beneath Alaska from Fuis et al. [2008] enabled the extension of the Aleutian slab model approximately 450 km further north and east beneath the North American continent, making it much more useful for modeling the hazard of subduction zone earthquakes to Alaska.

6. The Seismogenic Width of Subduction Zone Thrusts

[22] Because of its combination and use of multiple independent data sets, and through the systematic filtering of those data to focus on subduction-related seismicity, Slab1.0 is of particular relevance to studies of the thrust interface seismogenic zone, and facilitates improved modeling of subduction zone seismogenesis via the reduction of uncertainties associated with inaccurate assumptions of planar geometries [e.g., Moreno et al., 2009]. Slab1.0 can also help better define the limits of the seismogenic zone itself, as shown in Figures 9 and 10. Pacheco et al. [1993] (hereafter referred to as PSS93) define the deep limit of the seismogenic zone as the 95th percentile of a Gaussian or double-normal distribution fit to the number of thrust earthquakes in moderate-to-highly active subduction zones worldwide, using earthquake locations and mechanisms from the Harvard moment tensor catalog and their own body wave moment tensor inversions. Their results suggest that this seismic-aseismic transition varies from around 30 km in Mexico, to approximately 70 km in Java. Here, we employ the combined and filtered catalog of seismicity used to define each subduction zone interface, and follow the same procedure as PSS93, constraining the deep limit of the seismogenic zone using histograms of earthquake frequency with depth. Like PSS93, we limit the catalog used to earthquakes with magnitudes of M < 7.5, avoiding complications resulting from large earthquakes that may rupture beyond the entire width of the seismogenic zone. Depth is also often assigned to such large earthquakes in modern catalogs as a result of their difficulty to model as simple sources. We explore constraining the upper transition (sd) using the 5% percentile of these double-normal distributions; however results show sd is not well-characterized using this approach (Figure 9). The PSS93 study estimates the shallow aseismic-seismic transition as ds = 10 km for all subduction zones studied, so we adopt the same assumption here.

Figure 9.

Earthquake depth distributions showing the normalized number of earthquakes as a function of depth (red lines) for each of the subduction zones included in Slab1.0. Histograms are fit with a double Gaussian distribution (dashed blue line). Black dashed line marks the interpreted maximum depth of the seismogenic zone, the 95th percentile of the best fit distribution.

Figure 10.

Earthquake depth distributions showing the normalized number of earthquakes as a function of depth (red lines), for a selection of sub-regions of Slab1.0 models that were studied by Pacheco et al. [1993]. Histograms are fit with a double Gaussian distribution (dashed blue line). Black dashed line marks the interpreted maximum depth of the seismogenic zone, the 95th percentile of the best fit distribution.

[23] Hueret et al. [2011] recently published a similar analysis, characterizing seismogenic widths of subduction zones using thrust earthquakes from a combined gCMT and Centennial Catalog [Engdahl and Villaseñor, 2002]. In addition to the vertical histograms used here and by PSS, Hueret et al. [2011] analyze histograms of earthquakes with distance from the trench horizontally, and combine the 5% and 95% quartiles of both analyses to estimate a linear interface dip and width. We include results from their study below, for completeness.

[24] Results indicate that the deep limit of the thrust interface seismogenic zone varies from approximately 34 km in Mexico and northern Peru, to approximately 60 km in New Britain. The deep transitions for all subduction zones studied are listed in Table 1, and are compared to the results of PSS93. The fits to all regional subduction zones (entire arcs) are shown in Figure 9, while in Figure 10 we shown fits within sub-regions that may form distinct segments of each subduction zone, and that were used in PSS93. In each location, we also show the average distance of earthquakes to the slab interface model with depth, which shows that in general the deep limit of the seismogenic zone is coincident with a transition toward earthquakes occurring within the subducting slab (i.e., negative dZ in these panels; Figures 9 and 10). This is more apparent for the sub-region plots (Figure 10), suggesting that fits to entire arcs (Figure 9) mask variations in the deep limits of some subduction zones in the along strike direction.

Table 1. Thrust Interface Seismogenic Zone Parameters From Slab1.0a
Subduction Zoneequation image (deg)dd (km)SW (km)
  • a

    Here, θ is the average dip of the interface, dd the deep transition between seismic and aseismic interface regions, and SW the downdip seismogenic zone width. Comparisons to the same parameters from Pacheco et al. [1993] are given in regular parentheses; italicized values are from Tichelaar and Ruff [1993], and bold values are from Hueret et al. [2011], for reference. Note that PSS93 estimates the shallow aseismic-seismic transition as ds = 10 km for all subduction zones studied, which we adopt here. Asterisks denote poorly constrained sub-regions, due to a lack of significant seismicity over the time period studied.

Aleutians1753146
Rat Islands19 (22) {31}52 (30) {48}108 (53) {72}
Central Aleutians19 (25) {35}55 (50) (35–41) {56}139 (94) {75}
Eastern Aleutians15 (21) {33}43 (50) {50}130 (113) {72}
Alaska*8 {19}31 (37–41) {55}151 {135}
Central America223875
Mexico19 (21) {24}34 (40) (20–30) {41}73 (89) {74}
Central America24 (24) {28}44 (47) {63}83 (93) {103}
South America1648137
Central Chile16 (23)51 (50) (44–49)146 (107)
Southern Chile*14 {14}35 (≥ 41) {50}105 {190}
Northern Chile16 {22}50 (36–53) {51}141 {105}
Southern Peru17 {24}47 {43}133 {79}
Northern Peru11 {17}34 {44}128 {118}
Scotia (South Sandwich)17 (28) {22}35 (35) {60}68 (53) {131}
Kermadec-Tonga2047109
Kermadec22 (28) {26}55 (55) {58}119 (97) {114}
Tonga17 (26) {26}39 (40) {44}101 (68) {89}
Vanuatu25 (32) {29}42 (45) {53}70 (66) {85}
Solomon Islands315683
Solomon's30 (32) {24}46 (65) {46}62 (107) {87}
New Britain29 (25) {22}60 (64) {63}101 (134) {121}
Sumatra-Java1551154
Java*17 (27) {13}61 (70) {57}173 (138) {188}
Sumatra (Sunda)16 (17) {12}53 (50) {53}159 (150) {174}
Izu-Bonin (Marianas)16 (24) {20}34 (40) {45}85 (74) {104}
Philippines26 {30}46 {55}79 {81}
Ryukyu18 {24}46 {52}116 {98}
Japan-Kuril Islands1755153
Japan16 (22) {18}53 (50) (37–55) {60}162 (109) {161}
South Kuril18 (22) {22}55 (40) (37–43) {54}148 (82) {102}
North Kuril18 (32) {26}53 (50) {51}137 (75) {95}
Kamchatka17 (28) {27}55 (55) (38–40) {61}150 (87) {110}

[25] The most significant difference between our results and those of PSS93 are our estimates of thrust interface seismogenic width, calculated as the average downdip length of non-planar subduction zone interface between the shallow and deep seismogenic zone limits. Our estimates of this parameter tend to be larger than those of PSS93, by an average of approximately 30%. The difference between the two studies is greater than would be expected purely from the differences in our estimates of the deep limit of seismogenesis, implying that the PSS93 study also overestimated the average dip angle of the subduction interface through the seismogenic zone, or that their assumption of a linear geometry is inaccurate. We favor the former alternative, as a width estimate calculated from an average linear dip of the seismogenic zone agrees well with the seismogenic width calculated from the 3D model. Hayes and Wald [2009] and Hayes et al. [2009] both identified a difference between the dips of the slab interface derived from our technique and the dips of the best fitting fault planes of the gCMT mechanisms of events occurring on or close to those interfaces, biased on average ∼7–11° toward oversteepened gCMT dips for all subduction zones analyzed. This discrepancy could lead to the overestimation of the dip of the seismogenic zone in PSS93, as those authors calculate dip by averaging the dips of shallow fault planes of local thrusting CMT solutions. As our model accurately represents the trend of subduction zone seismicity through the seismogenic zone, and the non-planar nature of those interfaces, our estimates of seismogenic zone width are likely more accurate than those of Pacheco et al. [1993].

[26] Similarly, although our estimates of the deep limit of thrust interface seismogenesis compare well with those from the Hueret et al. [2011] study, seismogenic width assessments often differ considerably. We ascribe these differences to the linear geometry assumed by Hueret et al. [2011], as with the PSS93 study, misrepresentation of the dip of the subducting interface leads to errors in resulting estimates of seismogenic width.

[27] Our updated results show extremely broad thrust interface seismogenic zones in the Sumatra-Java (159 km and 173 km, respectively) and Japan-Kuril-Kamchatka (162–150 km from south to north) arcs. Of interest in these figures is the fact that, prior to 2011, only two segments of these long zones (Sumatra and Kamchatka) were known to have hosted M9+ earthquakes, and the broadest seismogenic zone in central-northern Japan had no documented event greater than ∼Mw8.2 in the past several hundred years or more [e.g., Ruff and Kanamori, 1980; Stein and Okal, 2007]. This changed on March 11 2011, when the Mw 9.0 Tohoku earthquake struck offshore of northern Honshu. This earthquake was considered to be perhaps unexpected by many [e.g., Geller, 2011]; indeed, this portion of the Japan Trench subduction zone had not been considered for an event even close to this size in the latest Japanese seismic hazard maps (Headquarters for Earthquake Research Promotion, National Seismic Hazard Maps for Japan, 2005; http://bit.ly/eBzkfD). Our results are summarized further in Figure 11, and show that, until the seismology community can definitively identify which properties play a dominant role in determining the maximum potential earthquake magnitude for a given subduction zone (such as sediment input and oceanic plate roughness) [e.g., Tanioka et al., 1997], all broad thrust interface seismogenic zones should be considered capable of such hazard [e.g., McCaffrey, 2008].

Figure 11.

Inferred thrust interface seismogenic zone width (SW, Table 1) versus maximum earthquake magnitude for subduction zone segments included in (a) this study, and (b) Pacheco et al. [1993] Maximum (moment) magnitudes are taken from Stein and Okal [2007], updated to include the 2010 Mw 8.8 Central Chile and 2011 Mw 9.0 Japan earthquakes. Though data from the PSS study (Figure 11b) shows weak correlation (correlation coefficient of r = 0.50), data from this study (Figure 11a) shows a strong positive correlation of r = 0.87. When poorly constrained data sets from southern Chile, Alaska and Java (shown transparent) are included, r = 0.61.

7. Conclusions

[28] Slab1.0 is the first three-dimensional model of global subduction zones to describe their detailed geometries from the trench, through the shallow seismogenic zone and into the mid-mantle. As such, the model has a variety of applications to geophysical analyses that rely on fault and/or subduction geometries as known a priori: for example, models of mantle convection [e.g., van Keken et al., 2002], tsunami modeling and propagation [e.g., Wang and He, 2008], and seismic hazard (e.g., GEM, http://www.globalquakemodel.org/).

[29] Slab1.0 will be explicitly incorporated into the forthcoming Global Active Fault and Seismic Source Database, under development through the GEM initiative. To this end, future versions of Slab1.0 will aim to supplement the existing model (incorporating data of slab depth, strike and dip at each grid node) with information related to rake, slip rate, seismogenic width, and coupling coefficients. Rake angle is a natural addition to Slab1.0, and can be derived from averaging moment tensors of subduction related earthquakes in a similar manner to the calculation of strike angles in the current model. As discussed above, and shown in Figures 9 and 10, the calculation of seismogenic width is also a viable product of the current model. Slip rates and coupling ratios are explicitly linked (given by plate motion velocities and the percentage of aseismic slip for a given location), and are somewhat complicated in subduction zone environments because most are not 100% seismic (i.e., they do not slip at the full plate motion rate over their seismogenic area), and because we have not witnessed and documented a full seismogenic cycle for any subduction zone (thus we cannot accurately calculate what percentage of accumulated moment was released seismically over that time period). Their incorporation into Slab1.0 will involve the use of geodetic models, which measure the current distribution of slipping slab surfaces, such as those that currently exist for Sumatra [e.g., Chilieh et al., 2008] and the Caribbean [e.g., Manaker et al., 2008].

[30] Slab1.0 currently represents the three-dimensional geometry of approximately 85% of subduction zones worldwide. The model is freely available online (see URL at the end of this manuscript); updates to the existing model, and new regional models, will be posted to this site. These models represent an evolving compilation; while in their current form they reflect an accurate representation of three-dimensional subduction zone geometries worldwide, future improvements are inevitable as new data become available, and as we supplement existing teleseismic data sets with regional, higher resolution studies.

Acknowledgments

[31] We thank two anonymous reviewers and the Associate Editor for their helpful comments. We also thank Daniel Garcia and James Dewey for comments and discussions during the preparation of this manuscript. Many of the figures were made using GMT, and we thank their developers and the many users of their support list. Finally, we thank Bob Engdahl and the Global Centroid Moment Tensor Group for the maintenance of and open access to their respective earthquake catalogs. The Slab1.0 model is available for download at http://earthquake.usgs.gov/research/data/slab.

Ancillary

Advertisement