Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
2002
http://www.geosciences.net
_______________________________________________________________________________
TELESEISMIC P-WAVEFORM RECEIVER FUNCTION ANALYSIS AND ITS APPLICATION TO QIONGZHONG STATION (QIZ) OF HAINAN ISLAND, NW SOUTH CHINA SEA
Xuelin QIU 1) , Keith PRIESTLEY 2) and Dan MCKENZIE 2)
1) South China Sea Institute of Oceanology, Chinese Academy of Sciences, Guangzhou 510301, P.R. China, xlqiu@scsio.ac.cn 2) Bullard Laboratories, University of Cambridge, Cambridge CB3 0EZ, UK
Abstract
The teleseismic P-waveform receiver function analysis is a commonly used technique in earthquake seismology to study the crustal structure beneath a broadband seismic station. This technique is applied to digital earthquake data recorded by the Qiongzhong station (QIZ) of Hainan Island, northwestern South China Sea. Both inversion and forward methods are used in modelling the receiver function waveform of QIZ to obtain the S-velocity structure beneath the station. Modelling results show that the crust beneath QIZ is composed of two major layers. The first layer is 8km thick with an S velocity of 3.6km/s. The second layer is 22km thick with an S velocity of 3.8km/s. The Moho is characterized as a gradient transition zone over a depth range of 30-36km. The S velocity increases from 3.8km/s to 4.65km/s. The average Moho depth is 33km. This crustal model is consistent with the results of other studies. Key words: Teleseismic P-waveform / Receiver function / Crustal structure / Qiongzhong station / South China Sea
Introduction
Seismograms of teleseismic P-waveforms contain the combined effects of the earthquake source, the propagation medium and the instrument response. To use these seismograms to determine local Earth structure beneath a broadband seismograph, the effects of local structure, or the site response, must be isolated from other factors. The resulting waveform of the isolated site response is called the receiver function. Many authors have successfully used receiver functions to determine crustal and upper mantle velocity structure (Langston, 1979; 1994; Owens, et al., 1984; Priestley, et al., 1988; Ammon, et al., 1990; Qiu, et al., 1996). A reasonably accurate knowledge of crustal structure is important for regional tectonic and geodynamic studies. Hainan Island is situated at the triple point of the South China Block, Indochina Block and the South China Sea. It is an important area to study the tectonic evolution of Southeast Asia. In this paper, the processing techniques and a time domain inverse modelling procedure of receiver functions are described. These techniques are then applied to broadband data recorded by Qiongzhong station (QIZ), Hainan Island. Finally, the results of modelling are presented and the crustal model is compared with those obtained in previous studies.
D ( ) ER ( ) H R ( ) = R = ER ( ) DV ( ) EV ( ) where D() and E() are the Fourier transforms of D(t) and E(t), respectively. ER() can be approximated by HR() because EV(t) is close to a delta function for teleseisms. In practice, a stable procedure for performing this deconvolution is to divide the Fourier transform of the radial component by that of the vertical component after introducing a water-level spectral fill, a minimum allowable amplitude level, to the amplitude spectrum of the vertical component. This avoids instabilities generated by dividing by a very small number due to troughs in the spectrum. If we smooth the result by a Gaussian function to exclude high-frequency noise, the final expression for the radial receiver function in the frequency domain is then
H R ( ) = Where
* DR ( ) DV ( ) = G ( ) ( )
(5) where dj represents the observed data, m represents the true model, parentheses denote the inner product, m is the model correction vector, and D is a matrix containing the first partial derivatives of the waveform with respect to changes in the layer velocities in the model m0. The operator F represents the nonlinear relationship between the model and the receiver function, which includes the calculation of a plane-wave synthetic seismogram and the source equalization procedure described earlier. Neglecting the higher-order terms, Equation (5) becomes
( D, m ) j F j [ m ] F j [ m0 ] (6) The right-hand side of Equation (6) is the waveform residual vector, and the equation may be solved for m using standard least-squares techniques described by Aki & Richards (1980). A minor modification of Equation (6) allows a solution to be obtained directly for m: adding (D, m0)j to both side of Equation (6) produces ( D, m ) j F j [m ] F j [ m0 ] + ( D, m0 ) j (7) which is a simple matrix equation with m being the only unknown. The importance of solving directly for model m is that constraints, e.g. a smoothness tradeoff parameter, may be applied to the model during the solution of Equation (7). Figure 1 presents a schematic ray diagram for an incident teleseismic plane P-wave together with its near-receiver reverberations and converted phases, which, in this ideal case, dominate the radial ground motion at the receiver. Figure 2 shows vertical and radial synthetic seismograms along with the corresponding radial receiver function for a one-layerover-half-space model. The crustal thickness of the model is 40km and the incident ray parameter corresponds to a surface source at 80 epicentral distance. Measuring times from the arrival of the direct P, the Ps phase arrives at about 5 seconds and the PpPms phase at about 18 seconds. The PpSms and PsPms phases arrive together as a negative pulse about 22.5 seconds. A small negative arrival about 27 seconds after the direct P is identified as the PsSms phase. All these phases are seen on both the radial synthetic seismogram and the receiver function but hardly seen on the vertical synthetic seismogram. The P multiple, PpPmp, arrives about 13 seconds on both vertical and radial synthetic seismograms but is removed from the receiver function by the deconvolution. The relative arrival times of each phase are controlled principally by crustal thickness and velocity, while relative amplitudes are affected by velocity contrasts at the Moho and the ray parameter.
(2)
(3) and
G ( ) = exp(
2
4a 2
(4)
where * indicates complex conjugation; c is the minimum allowable spectral amplitude of the vertical component, the water level; ' covers the range of all possible frequencies; the constant normalizes the Gaussian filter to unit amplitude in the time domain; and a is the Gaussian factor which controls the width of the Gaussian pulse. Transformation of HR() back into the time domain produces HR(t), an estimate of ER(t). In fact, as the plane P-wave approaches vertical incidence, the receiver function HR(t) becomes nothing more than a scaled version of the radial component of the site response, ER(t), with the P multiples removed. The tangential receiver function, HT(t), is produced by the same procedure. For a laterally homogeneous structure, HT(t) should theoretically be zero. This provides a way to judge the degree of lateral variation in Earth structure beneath a station (e.g. the presence of dipping velocity interfaces). The inversion procedure and notation summarized below is presented in detail by Ammon et al. (1990). The nonlinear relationship between velocity structure and waveform is approximated by a Taylor series expansion about an initial model, m0, such that
http://www.geosciences.net
_______________________________________________________________________________
Fig.1 Simplified ray diagram showing the major P-to-S converted phases that comprise the radial receiver function for a single layer over a half-space
In the phase nomenclature, the first letter P denotes a plane P-wave coming up from the mantle; the other letters denote travel paths in the crust: upper case letters denote downgoing paths, lower case letters denote upgoing paths, and m indicates reflection from the Moho interface
Fig.2 Vertical and radial synthetic seismograms and the corresponding receiver function calculated for a single-layer-over-half-space model
The receiver function is annotated with P-to-S converted phases corresponding to those in Fig.1. Note the removal of the P multiple, PpPmp, from the receiver function
middle of Hainan Island with an elevation of 230 m (Fig.3). Topography of the island shows that the northern part of the island is a flat plain (<200 m) and
Fig.3 Topography of Hainan Island superimposed with simplified geological map (after Li et al., 2000)
1 Quaternary; 2 Cenozoic volcanic rocks; 3 Mesozoic basins; 4 Paleozoic; 5 Precambrian basement; 6 granites
QIZ is one of the China Digital Seismic Network (CDSN) stations. It was upgraded in 1992 with broadband instruments of the Global Digital Seismic Network (GDSN). The broadband channels are digitized at 20 sps and have a flat velocity response between 0.05 and 5.0 Hz. The USGS PDE catalogue was searched for events of 5.8 and = 30-90 during the period 1992-2000. An appropriate suite of teleseismic earthquake seismograms was obtained from the IRIS DMC. Broadband P-waves were windowed, rotated to the theoretical back azimuth to the event, and then deconvolved to form radial and tangential receiver functions. As stated above, examination of the tangential component of motion is useful for determining the degree of lateral velocity heterogeneity in the structure. Only data with the highest signal-to-noise ratios and whose structural signals could be clearly seen were chosen. Receiver functions for-events with similar ray parameters and
Eight events from the Tonga Trench (azimuth 117 and distance 81) were chosen for receiver function analysis, based on a good signal-to-noise ratio and the robust features in the resulting receiver functions. The
http://www.geosciences.net
_______________________________________________________________________________
wave velocity was the free parameter in the inversion, the P-wave velocity was set assuming a Poisson's ratio of 0.25 (i.e. Vp = 3 Vs), the density was set by assuming the relationship =0.32Vp+0.77 (Priestley et al., 1988), and the layer thicknesses were fixed at 2km. In this study, a Gaussian parameter of 1.0, which low-passes the waveforms at ~0.4Hz, and a water-level of 0.001 are used for the deconvolutions, and a smoothness trade-off parameter of 0.5 is used for the inversion modelling. A summary of the inversion results is shown in Figure 6. Although the starting models are spread over a large space, the inversion solution models converge into a relative narrow range. The stacked receiver function and its 1 standard deviation () bounds are shown in Figure 6(b). All synthetic receiver functions of the inversion solution models well fit the observed receiver function.
Fig.6 Summary of the inversion results of the QIZ stacked receiver function
Left panel: the range of starting model velocities (black lines), the inversion solution models (green lines) and the average inversion model (red line). Right panel: the stacked receiver function (red line) and its 1 bounds (black lines), and the synthetic receiver functions computed from each inversion solution model (green lines)
We simplified the velocity model by first averaging the acceptable inversion models and then collecting adjacent layers of similar velocities into single layers. Finally, we used forward modelling to test these simplified models and to assess how well individual model features were constrained by arrivals in the data. In the forward modelling we required the simplified models to contain the general features of
the inversion models but with a smaller number of parameters. Large tangential arrivals in the receiver functions indicate laterally heterogeneous structure (Langston 1977; Cassidy 1992); where these occurred we attempted to match only the largest-amplitude features of the radial receiver function so as to minimise the effect of varying lateral structure on the 1-D earth model.
Fig.7 A comparison of the stacked receiver function (red line) with synthetic receiver functions for forward test models (Fig.8)
Black line for the average inversion model, green line for the two-layered model and the blue line for the final model
Fig.8 The average inversion model (red lines), the two-layered forward test model (green lines) and the final model (blue lines) of QIZ. In the same model, Vp= 3 Vs and =0.32Vp+0.77 are assumed.
http://www.geosciences.net
_______________________________________________________________________________
which is comparable with our results. Moho depths from gravity modeling vary from 35km (Wang et al., 1982; Wei et al., 2001) to 30km (Jiang et al., 2001; Song et al., 2002), which are also in broad agreement with our model. Although no refraction profiles exist from the central part of Hainan Island near the QIZ station, some controlled-source seismic investigations have been carried out around the island. Results from a refraction profile (Lin et al., 1988) across the Qiongzhou Strait suggest a Moho depth of 28km in northern Hainan Island. OBH experiments in the Yinggehai Basin (Xia et al., 1998) and across the Xisha Trough (Qiu et al., 2000; Qiu et al., 2001) show that compressional velocity at the top of upper crust is 5.8-5.9km/s in the offshore areas southwest and southeast of Hainan Island. Our model of QIZ has the Moho a few kilometres deeper and the top crustal velocity a few percent faster. However, these differences are reasonable, since the central and southern parts of Hainan Island are mountain regions and the northern and offshore areas are closer to centres of extension.
3. Discussion
Results from our receiver function analysis indicate that an upper crustal thickness of 8km and a lower crustal thickness of 22km. The upper crust corresponds to a granitic layer, consistent with granitic outcrop around the QIZ station. Chen Guoneng et al. (2001) suggested that these granites were generated by in-situ melting. The lower crust is inferred to correspond to a basaltic layer and the proposed gradational Moho suggests that the crustmantle boundary is most probably a transitional one in this area. The seismic velocity in the uppermost mantle is not well constrained, and the mantle has been modelled as a half-space with a shear velocity of 4.5-4.7km/s. The receiver function method is not sensitive to the absolute velocity of the half-space, and therefore does not provide a reliable measurement of the upper mantle velocity below the station. Liu et al. (1997) analysed early CDSN data in order to obtain the receiver structure of the Chinese stations, using a similar technique to the one described in this paper. For the QIZ station, only three events from the same distance of 34 were used together with a different azimuth for the stacked receiver function. Liu et al. (1997) obtained a Moho depth of 34km,
5. Acknowledgments
We would like to thank IRIS DMC for providing the digital seismic data and James McKenzie for assistance in computer system and software installations. This work is jointly supported by the Royal Society, the Chinese Academy of Sciences (KZCX2-209, KZ951-B1-406), and the Chinese Ministry of Science and Technology (G2000-046701).