SEARCH

SEARCH BY CITATION

Keywords:

  • TAIGER;
  • ambient seismic noises;
  • gravity anomaly;
  • multi-scale tomography;
  • surface waves

Abstract

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Data and Data processing
  5. 3. Wavelet-based multi-scale Inversion
  6. 4. Results and Discussions
  7. Acknowledgments
  8. References
  9. Supporting Information

[1] We construct the first broad-band surface wave group velocity dispersion maps for the entire island of Taiwan using the ambient seismic noise tomography. Continuous data from three island-wide broad-band networks are used. In particular, taking advantage of the temporary arrays deployed by the TAiwan Integrated GEodynamics Research project (TAIGER), we have collected an unprecedented data amount for the noise tomography in Taiwan. We construct 2D group velocity maps for Rayleigh waves from 4 to 20 seconds using a wavelet-based multi-scale inversion technique. Patterns of lateral variations of our shorter period (<10 seconds) model demonstrate very good correlation with the surficial geology, whereas the overall structure, albeit with much better resolution in the shallow depth, is generally consistent with previously established body wave models. Besides seismic structure, our model also provides vital constraint on resolving the long-lasting controversy about most prominent Bouguer gravity anomaly in central Taiwan, implying that it is likely caused by a deeper mountain root. With regard to various scenarios of the tectonic evolution of Taiwan, our results seem to favor the lithospheric collision model that invokes significant crustal thickening during the collisional orogeny.

1. Introduction

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Data and Data processing
  5. 3. Wavelet-based multi-scale Inversion
  6. 4. Results and Discussions
  7. Acknowledgments
  8. References
  9. Supporting Information

[2] Sitting along the active arc-continent collision boundary between the Philippine Sea Plate and the Eurasia Plate, Taiwan is known for its complicated tectonics. Various scientific efforts have been made to unwarp its geological complexity, and seismic tomography is one of the most powerful tools to explore the subsurface structure as benefited from the high regional seismicity. In the last decade, high resolution models using vast body wave picks have been published [e.g.,Kim et al., 2005; Wu et al., 2007]. Despite general agreements among these models, several fundamental and interrelated issues remain to be clarified, such as the mechanisms of collision and orogeny, the Moho morphology, and the source of the most prominent Bouguer anomaly in central Taiwan [Yen and Hsieh, 2010]. The ambiguities are largely due to the fact that resolutions of these seismic models are relatively poor in both the deeper (>∼40 km) and the near-surface portions (<∼5 km) because of the intrinsic limitations of body wave tomography of local scale. In particular, adequate resolution of the near-surface structure is essential to decipher the nature of gravity anomalies.

[3] It is well-known that sensitivities of surface waves and body waves are complementary and short-period surface waves may provide greater constraints to the shallow crust. However, in constrast to the body wave tomography, the progress of the traditional surface wave tomography in Taiwan has been limited by (1) insufficient robust measurement of short period surface waves, as they are usually obscured by body waves at short distance and fade away at larger distance due to anelastic attenuation, and (2) nonuniform distribution of earthquakes of moderate size inside or near the island; surface wave energy excited by small earthquakes is weak, and large earthquakes are avoided for their source complexity. These problems have been resolved with the observation that the cross-correlation function (CCF) of continuous records of ambient noise at two seismic stations resembles the impulse response of elastic waves between these two stations [e.g.,Weaver and Lobkis, 2001, 2002; Shapiro and Campillo, 2004]. The noise-derived empirical Green's functions (EGF) are dominated by fundamental mode surface waves [e.g.,Shapiro et al., 2005] and they are ideal data for the crust and uppermost mantle tomography [e.g., Brenguier et al., 2007; Yang et al., 2007; Nishida et al., 2008; Liang and Langston, 2008; Saygin and Kennett, 2010; Porritt et al., 2011].

[4] In Taiwan, ambient noise tomography was first implemented by Huang et al. [2010], who inverted for maps of very short period (0.5–3.0 seconds) surface waves of Taipei Basin using data recorded by a dense local network. You et al. [2010]used data from 34 short-period seismometers to construct maps of short period surface waves (2–5 seconds) of northern Taiwan, which indicate consistent correlation to the lateral surface geological variations. While these studies have clearly demonstrated the strength of surface wave tomography in imaging the shallow crust, the potential of ambient noise tomography in Taiwan was still highly limited by the instrument response of the seismometers and/or the spatial extent in these earlier works, and no surface wave models of the entire island of Taiwan have been developed yet.

[5] In this study, we apply seismic noise tomography for Rayleigh waves using continuous data recorded by three island-wide broadband networks. With this unprecedented data set, we are able to derive high resolution broad-band models (4–20 seconds) of Rayleigh waves for the island of Taiwan. The models provide thorough constraint on the crust structure from the surface down to the depth of about 30 km. More importantly, with their exceptional near-surface resolving power, the results shed light on the celebrated debate about the most prominent gravity anomaly in central Taiwan and provide important implications to the orogeny of Taiwan.

2. Data and Data processing

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Data and Data processing
  5. 3. Wavelet-based multi-scale Inversion
  6. 4. Results and Discussions
  7. Acknowledgments
  8. References
  9. Supporting Information

[6] Besides two major permanent seismic networks, the Central Weather Bureau Broad-Band network (CWBBB), operated by the Central Weather Bureau, Taiwan, and the Broadband Array in Taiwan for Seismology (BATS), operated by the Institute of Earth Science, Academia Sinica, we have incorporated data from three east-west linear arrays from a temporary broadband network deployed by the TAIGER project [Okaya et al., 2009]. In total, we have collected continuous data from 85 stations, including 25 CWBBB stations, 14 BATS stations and the other 46 from TAIGER arrays. Because continuous records of a common time period are required for the operation of cross-correlation between station pairs, we have selected data from the year 2007 to cope with the operation period of TAIGER project. The distribution of stations is shown inFigure 1. As the number of cross-correlation paris isn(n-1)/2, where n is the number of stations, the additional 46 TAIGER stations bring in about four times more available paths than those provided solely by the permanent CWBBB and BATS networks. This significant increase in data amount could largely improve the path coverage for the tomography.

image

Figure 1. Distribution of broad-band seismic stations used in this study and the colored topography of Taiwan. Stations of three networks are represented by different red symbols as indicated in the figure. The dark lines are boundaries of major tectonic units, and the indices shown in each geological unit are as follows: (I) Coastal Plain, which includes Ilan Plain (Ia), Western Coastal Plain (1b) and Pintung Plain (1c), (II) Western Foothills, (III) Hsueshan Range, (IV) Central Range, (V) Costal Range, (VI) Longitudinal Valley and (VII) Hengchun Peninsula.

Download figure to PowerPoint

[7] The vertical component of the data is used in this study, and thus the major signals of derived EGFs mainly contain fundamental mode Rayleigh waves. We basically follow the widely used procedure of Bensen et al. [2008]to derive EGFs from noise cross-correlation. Examples of the final CCF stacks at two period bands are presented inFigure 2.

image

Figure 2. Final stacks of CCFs at two different period bands with corner frequencies at (a) 2.5 and 8 seconds, and (b) 8 and 25 seconds, respectively. The traces are sorted by distance, and at most 5 representative traces with higher SNR are shown for each 5 kilometer bin.

Download figure to PowerPoint

[8] The dispersion of group velocity from 4 to 20 seconds for the noise-derived EGFs is measured using the frequency-time analysis [e.g.,Levshin et al., 1989]. Limited by the spatial extent of the island and the requirement of three-wavelength inter-station distance, longer period surface wave dispersion measurements are not available for most station pairs. Careful examination is employed in the stage of raw EGFs and the dispersion measurements, and about 60% of all available data at each period band are retained for the tomography. The numbers of CCFs used for tomography are 2026 for the shortest period (4 seconds) and 862 for the longest period (20 seconds). More details on data processing, data selection and types of seismographs used in this study are given in theSupplementary material.

3. Wavelet-based multi-scale Inversion

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Data and Data processing
  5. 3. Wavelet-based multi-scale Inversion
  6. 4. Results and Discussions
  7. Acknowledgments
  8. References
  9. Supporting Information

[9] Instead of the commonly used pixel parameterization in local tomography, we implement a wavelet-based multi-scale inversion, by which both the spatial localization and non-stationary model smoothing are assured. More specifically, since spatially and spectrally compromised localization of the wavelet representation accommodates non-uniform distribution of available data constraints, the regularization acts to sort through successive scales depending on the local data constraints and automatically achieves data-adaptive, spatially varying optimal resolution. This technique has been applied to tomography in global [Chiao and Kuo, 2001; Chiao et al., 2006], regional [Gung et al., 2009] and local scales [You et al., 2010; Chiao et al., 2010]. In this study, we follow the wavelet based local parameterization described by You et al. [2010]. More details of this technique, inversion schemes and the evaluation of model reliability are given in the supplementary material.

4. Results and Discussions

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Data and Data processing
  5. 3. Wavelet-based multi-scale Inversion
  6. 4. Results and Discussions
  7. Acknowledgments
  8. References
  9. Supporting Information

[10] We present results of inverted maps of Rayleigh wave group velocities at 5 selected periods, 4, 8.3, 12.5, 16.7 and 20 seconds (Figure 3). To illustrate the relationship between these results and surficial geology, the map of topography and major geological units are also shown in Figure 1.

image

Figure 3. (a)-(e) Maps of group velocity of fundamental mode Rayleigh waves at 5 selected periods, (a) 4, (b) 8.3, (c) 12.5, (d) 16.7 and (e) 20 seconds. In each panel, the boundaries of major tectonic units are shown in dark lines, and the topographic shading is also shown as background. The locations of Taichung Basin (TB), Peikang Basement High (PBH) and Pingtung Basin (PB) are labeled on the panel (b). In each panel of the velocity maps, the average velocity of the final model (Vavg), number of station pairs used in the inversion, and the achieved variance reduction (VR) are shown the upper left corner. The variance reduction of each final model is evaluated w.r.t. the corresponding reference model, in which the constant group velocity Vavg is assigned everywhere. (f) Bouguer Gravity anomaly of Taiwan. Modified fromYen and Hsieh [2010], where two major negative gravity anomalies Taichung-Puli low (TPL) and Pingtung low (PL) are labeled on the map.

Download figure to PowerPoint

[11] At shorter periods, the lateral variations of velocity correlate well with the surficial geology. Lower velocity regions are distributed in lowlands with recent sediments, such as the Western Coastal Plain, Pingtung Plain, Longitudinal Valley and Hengchun Peninsula, whereas higher velocity is generally associated with regions in mountain ranges with either aged sedimentary or metamorphic units. Since the geological and elastic characteristics of the shallow crust is closely related to the topography in Taiwan, a very strong correlation between wave velocities and topography is also observed in shorter period results, where higher/lower velocities are situated in topographic highs/lows, and the topographic transitional zone, the Western Foothills, is marked by intermediate velocities.

[12] As expected, the lack of focused depth sensitivities of surface waves results in a slow transition in lateral variations with the increasing periods (Figure 3 (b)–(e)), and the extent of surficial geology in depth is generally revealed by the persistence of the short period signatures on maps of longer periods. For instance, the three major geological units of the Western Coastal Plain, Taichung Basin (TB), Peikang Basement High (PBH) and Pingtung Basin (PB), are all characterized by lower velocity at the period 4.0 seconds. The PBH low velocity signature is getting weaker at the map of 8.3 seconds (Figure 3 (b)), and the TB low velocity fades away at 12.5 seconds (Figure 3(c)). In contrast, the low velocity anomaly in PB persists all the way to the map of 20 seconds (Figure 3 (e)). The results can be explained by different thicknesses of younger sedimentary sequences in these regions, which are substantiated by the well log data in the Western Coastal Plain [e.g., Chou, 1980; Shaw, 1996; Mouthereau et al., 2002].

[13] Compared to the topographic lows, the subsurface elastic properties associated with high mountain ranges seem to go deeper. This is clearly shown that the major high velocity spots along the crests of the Central Range (CR) and Hsueshan Range (HR) extend for a wide spectral range in the associated surface wave velocities to about 12.5 seconds.

[14] The lateral variations at longer periods (Figure 3 (d) and (e)) are relatively homogeneous and show little connection with the surficial geology. Note that, other than the PB low velocity anomaly mentioned earlier, the high velocity beneath PBH is another noticeable anomaly at longer periods, indicating that the PBH structure may extend deep.

[15] To further demonstrate the strength of the resolving power of the short-period surface waves at shallow crust, we compare our results with those predicted by two recent 3D high resolution body wave models byKim et al. [2005] and Wu et al. [2007]. VP and VS in both models are used to compute group velocity maps of fundamental mode Rayleigh waves at the five selected periods (Figure 3). The results for periods 4 seconds and 16.7 seconds along with their corresponding average velocities and the fits to the EGFs in terms of variance reductions are shown here (Figure 4), whereas maps for the other periods are presented in the supplementary material.

image

Figure 4. Group velocity maps of fundamental-mode Rayleigh waves calculated from two representative body-wave models: (a) map of 4-second period derived from Vp and Vs models ofWu et al. [2007]; (b) map of 4-second period derived fromKim et al. [2005]; (c) same as (a), but for 16.7-second period; and (d) same as (b), but for 16.7-second period.

Download figure to PowerPoint

[16] For the results of period 4 seconds, although the general patterns in both models of Figure 4 are similar with marginal correlation to the surficial geology, the average velocities are too high and the velocity perturbations are too weak, i.e. the velocity is overestimated in the Coastal Plain and underestimated in mountain Ranges. This agrees well with the conclusions in Lin et al. [2011], who assess the 3D body wave models by comparing the first-arrival times from the active sources conducted by TAIGER project with those computed from the 3D models using numerical methods.

[17] The weaker constraint on shallow structure in body wave tomography is not surprising and is mainly caused by the combination of two factors: (1) most rays arrive at the stations in quasi-vertical direction, thus, the inter-station area in the near surface is inherently much less sampled; and (2) to compromise the noises from the inevitable source uncertainties, regularization invoked in the inversion exercise over-damping especially for the shallow structure.

[18] On the other hand, for results of longer periods derived from body wave models (Figure 4 (c), (d) and Figure S2 in the supplementary material), the average velocities and data fits are much closer to results from our ambient noise tomography, suggesting that the body wave models do have a fair overall constraint for the deeper structure.

[19] Since density anomaly is one dominant factor responsible for the corresponding seismic velocity anomaly and the fact that the gravity inversion is highly non-unique, seismic models are commonly used as an important calibration reference for the gravity interpretation. In Taiwan, detailed field measurements of gravity anomaly have been conducted over the last two decades [e.g.,Yen et al., 1990; Hwang et al., 2007; Yen and Hsieh, 2010]. There are three major anomalies easily identifiable on the map of Bouguer gravity anomaly (Figure 3(f)). This is, the positive anomalous belt along the south-east coast, the negative anomaly in the southern Taiwan, the Pingtung low (PL), and the negative anomaly in the central Taiwan, the Taichung-Puli low (TPL).

[20] It is well known that the positive anomaly along the south-east coast is induced by the adjacent oceanic crust [e.g.,Yen et al., 1998]. For the PL anomaly, it is commonly accepted as a consequence of the accreted thick sediments [e.g., Hsieh, 1970; Wu et al., 2007] and the associated low-velocity anomaly is also resolved in our broad-band surface wave maps. It is worth mentioning that the positive gravity anomaly in PBH is also in good agreement with high velocity anomaly shown in our results.

[21] On the other hand, the interpretation for the largest and the most prominent TPL remains debatable. A shallow sedimentary structure is the conventionally favored source by previous gravity modeling [Yen and Hsieh, 2010] and the mechanical modeling [Upton et al., 2008]. If this is the case, it is then reasonable to presume that such a large scale shallow density anomaly should be accompanied by a near surface velocity anomaly of the comparable scale. However, this has not been (in)validated by body wave models for their poor resolution at shallow depths.

[22] Our short period surface wave maps derived from seismic ambient noises, for the first time, provide robust constraint on the near surface structure of Taiwan, in which no low velocity anomaly beneath TPL is observed in the map of short period Rayleigh waves. Based on our result, the possibility of a shallow source responsible for TPL is thus confidently excluded. While we could not pinpoint the source depth of TPL, it has to lie below the resolvable depth range of our tomography. The depth sensitivity of 20 seconds Rayleigh goes to about 30 km, thus, our results suggest that a thicker mountain root is very likely to be held responsible for the low density anomaly.

[23] Thicker crust beneath TPL is also supported by recent studies using seismic receiver functions [Wang et al., 2009; Wang et al., 2010], in which a deep Moho discontinuity up to ∼50 km beneath this area is reported.

[24] In summary, with EGFs derived from seismic ambient noises, we have constructed the first broad-band Rayleigh wave models of Taiwan using a wavelet-based multi-scale inversion technique. Reliability of our model is demonstrated by consistent correlation between lateral variations of shorter period (<10 seconds) surface waves and surficial geology, and by comparing with two most recent body wave models. Besides seismic structures, our models also provide important constraint on the origin of the local Bouguer gravity anomalies, indicating that sources for the PL and the positive anomaly around PBH are within the shallow crust, while the largest and most prominent TPL Bouguer gravity low is likely caused by a deeper mountain root beneath the Taichung-Puli area. With regard to various scenarios of the tectonic evolution of Taiwan [Suppe, 1987; Wu et al., 1997; Lin, 2000], the implication for a deeper mountain root seems to be more consistent with the lithospheric collision model [Wu et al., 1997] that invokes significant crustal thickening during the collisional orogeny.

Acknowledgments

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Data and Data processing
  5. 3. Wavelet-based multi-scale Inversion
  6. 4. Results and Discussions
  7. Acknowledgments
  8. References
  9. Supporting Information

[25] We wish to thank the operators of the CWB, BATs, and TAIGER project for providing high-quality seismic data. Comments from two anonymous reviewers and discussions with Ban-Yuan Kuo, Horng-Yuan Yen and John Suppe have improved this work considerably. This research is supported by the National Science Council of Taiwan under the grants NSC 100-2116-M-002-025 and NSC 100-2119-M-001-020.

[26] The Editor thanks the two anonymous reviewers for their assistance in evaluating this paper.

References

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Data and Data processing
  5. 3. Wavelet-based multi-scale Inversion
  6. 4. Results and Discussions
  7. Acknowledgments
  8. References
  9. Supporting Information

Supporting Information

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Data and Data processing
  5. 3. Wavelet-based multi-scale Inversion
  6. 4. Results and Discussions
  7. Acknowledgments
  8. References
  9. Supporting Information

Auxiliary material for this article contains a text file.

Auxiliary material files may require downloading to a local drive depending on platform, browser, configuration, and size. To open auxiliary materials in a browser, click on the label. To download, Right-click and select “Save Target As…” (PC) or CTRL-click and select “Download Link to Disk” (Mac).

Additional file information is provided in the readme.txt.

FilenameFormatSizeDescription
grl28955-sup-0001-readme.txtplain text document1Kreadme.txt
grl28955-sup-0002-txts01.docWord document1124KText S1. Details on data processing and data selection, seismographs, wavelet-based multi-scale inversion, group velocity models calculated from body wave models, and evaluation of model reliability.
grl28955-sup-0003-txts01.pdfPDF document1030KText S1. Details on data processing and data selection, seismographs, wavelet-based multi-scale inversion, group velocity models calculated from body wave models, and evaluation of model reliability.

Please note: Wiley Blackwell is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.