Variations in upper mantle structure under northern Australia

Transcription

1 Geophys. J. hf. (1993) 114,34-31 Variations in upper mantle structure uner northern Australia. C. Dey, B. L. N. Kennett, J. R. Bowman an A. Gooy Research chool of Earth ciences, Australian National University, Canberra, ACT 2, Australia Accepte 1993 January 5. Receivc 1992 Decemhcr 29; in original form 1992 October 21 UMMARY Temporary array eployments of short-perio seismometers in northern Australia have been use to buil up composite recor sections for waves interacting with the upper mantle. table measures of the seismic wavefiel are provie by stacking the complex envelopes of all the seismic waveforms falling in a 1 km istance interval away from the source. Two groups of sources (a) along the Flores Arc, onesia with propagation uner northwestern Australia, an (b) in New Guinea with paths to the NNE of the array, have been use to construct composite recor sections for both P an V waves over the istance range km. The timing an amplitue istributions for P waves from the two regions show noticeable ifferences. Detaile moelling of the recor sections yiels velocity moels with significant variation in velocity for the two sets of propagation paths for which the mipoints are separate by about 1 km. The short-perio V-wave sections inicate efficient propagation of highfrequency waves in a lithosphere extening own to 21 km. Arrivals from the eeper mantle cannot be correlate with confience because of a loss in high-frequency content reveale by broa-ban observations. This requires a significant attenuation zone for beneath 21 km. Key wors: Australia, lateral heterogeneity, P waves, V waves, upper mantle structure. INTRODUCTION The upper part of the mantle is cut by col subucting slabs with high seismic velocities; where hot material is upwelling beneath rifts or in hot spots the seismic velocities are reuce. Even away from these prominent features there is significant heterogeneity in structure. Global tomography stuies (e.g. Woohouse & Dziewonski 1984) have reveale significant variations in the seismic wavespees in the upper mantle or scales of 1 km or more, but currently still have rather limite epth resolution. tuies of refracte waves provie the most irect approach to mapping horizontal variations in upper mantle velocities but require a favourable geometry of sources an receivers. By combining irect mantle observations of H waves with those of the surface multiple at larger istances, Gran & Helmberger (1984) have been able to show significant ifferences in raial structure beneath the shiel an tectonic regions of North America. A strong contrast across a tectonic bounary was also foun in Europe (Rial, Gran & Helmberger 1984). Most of such upper mantle stuies have been acquire using long-perio 34 waveforms because these are more stable an are less influence by small-scale heterogeneity. However, the long perio waves have wavelengths of the orer of 1 km, at a epth of 4 km in the mantle, an this is not sufficient to etermine the etails of the transitions in the mantle. By contrast, short-perio observations have greater potential resolving power but are also more susceptible to the effects of heterogeneity in structure. The influence of such heterogeneity can be reuce by builing composite recor sections for many events across a broa aperture recoring array (Bowman & Kennett 199). The major earthquake belt running through onesia, New Guinea an its offshore islans lies at the appropriate istance range to investigate upper mantle structure using arrays of recoring stations in northern Australia. The activity in this belt is such that a three-month eployment of temporary stations normally allows the preparation of a composite recor section using events with magnitues between 4 an 5. This paper presents the results of two such fiel eployments to contrast the propagation characteristics of paths from the Flores Arc, onesia (previously stuie by

2 Variations in upper mantle structure 35 Bowman & Kennett 199) with paths from New Guinea. The turning points for waves bottoming in the upper mantle from these ifferent groups of events are separate by about lo km. The resulting composite recor sections show clear ifferences in timing an amplitue istribution foi P-wave arrivals, which inicate ifferences in structure within the upper mantle. TEMPORARY ARRAY EXPERIMENT Two separate eployments of temporary arrays of short-perio instruments (LTC, NWB) were installe in the Northern Territory of Australia to recor events in the onesia/new Guinea earthquake belt (Fig. 1). Each experiment ha an aperture of aroun 4 km an laste about three months. The array configurations follow the main highways to provie reasonable access to sites; each consiste of analogue recorers with 3-4 km spacing with a enser igital recorer array with 5-1 km spacing. The igital arrays were linke to the permanent Warramunga array (WRA-a 25 km aperture array of 2 short-perio instruments) which lies about km to the south of the intersection of the two arms of the recorer pattern in Fig. 1. The permanent array provies a useful supplement to the more wiely sprea stations in the temporary eployments. For the LTC experiment from eptember to December 1985 the ata comes primarily from the analogue recorers (the igital recorers i not perform well in the high temperatures of the tropical summer). For the NWB experiment from July to eptember 1986 much of the ata use comes from the igital array an has been supplemente with analogue recorings to give a larger aperture. The earthquakes for which ata were extracte are isplaye in Fig. 1. The events marke with a triangle were use by Bowman & Kennett (199) to generate the P-velocity moel NWB-1 for paths from the Flores Arc to northern Australia. The iamons inicate aitional events use in the construction of recor sections for propagation paths to the LTC an NWB arrays. The events chosen have NEIC boy wave magnitues between 4 an 5 with the object of trying to secure simple waveforms. However, a isavantage of working with events of this size is that locations from global observations are of limite reliability. The uncertainties introuce by possible mislocation can be reuce by combining ata from many events. The earthquakes have been chosen to lie above 4 km epth so that the propagation paths woul have the smallest possible interaction with the subuction-zone structures. the subsequent isplays we present composite recor sections for two major classes of events base on the work of Dey (1989). The first consists of events along the front of the Flores Arc from 11"-132"E, 6"-123 an the secon coniprises events in New Guinea from 134"-153"E, "-9". Both these groups of events cover a moerate span in azimuths. A thir group with a smaller azimuthal ispersal lies through the Bana Arc towars the Philippines but this is not iscusse here, because of the likely contamination of the propagation characteristics through the upper mantle by the complex pattern of subuction zones surrouning the Bana ea. The mipoints for propagation from the selecte events lie beneath the Australian continent or continental shelf. For the Flores events the turning points in the mantle lie beneath the Kimberley ranges in northwestern Australia. The turning points for the events in New Guinea lie mostly beneath the Gulf of Carpentaria. There is a horizontal LTC/N W B -1-2 Figure 1. Earthquake istribution in the onesia-new Guinea region uring the eployment of the portable arrays use in this stuy. The soli triangles enote the events use by Bowman & Kennett (19%)) an the soli iamons the events ae for this stuy. The open triangles inicate the recorer locations in the LTC array an the pentagons the recorer locations for the NWB array.

3 36. C. Dey et al. separation between the two clusters of turning points of the orer of 1 km. orer to gain the maximum coverage of the upper mantle we nee to combine ata from many events, an to o this we nee to bring each event to a common epth. We have chosen to correct all events to surface focus base on an estimate of the mean slowness across the array of stations. This proceure is more successful for P waves than for the later waves which suggests that there are resiual errors in the epth an origin times for the events. Once each event has been correcte to surface focus, we have use stacking of all the arrivals in a narrow istance range to enhance the features of the composite recor sections, particularly the later phases associate with mantle transitions. A very stable measure of the seismic wavefiel is provie by a stack of the complex envelopes of those waveforms lying within 1km istance bins (Bowman & Kennett 199). This proceure reuces the influence of variations in source waveform an also counteracts the effects of small-scale heterogeneity. The stacking proceure has the consequence of lowering the apparent frequency of the ata an imposing a slight elay on the location of amplitue maxima compare with unstacke ata. The stacking proceure reuces the clutter in the recor sections - Y ai \ W cn r I F an allows a clear picture of the arrivals to been seen (Fig. 2). Whilst the main emphasis here is on the raial structure in the upper mantle, we note that Kennett & Bowman (199) have use the variation in amplitue istribution for iniviual events across arrays of ifferent aperture an spacing to infer the presence of pervasive P-wave heterogeneity, of the orer of 1 per cent with horizontal scale lengths of 2-3 km, superimpose on the main raial variation in velocity. P-WAVE RECORD ECTION Composite recor sections for P-wave arrivals from the events along the Flores Arc, onesia an in New Guinea are compare in Fig. 2. Each trace is compose of a stack of the complex envelopes of all traces falling in a 1 km bin an is separately normalize to the maximum amplitue on the trace. Before stacking, all traces were filtere with a Hz ban pass. The application of epth corrections for each event prior to stacking has the effect of enhancing the arrivals on the later branches e.g. those associate with the 4 km iscontinuity. Also, the epth corrections ten to suppress the influence of the surface-reflecte phases pp an n Y cr; \ W cn r l- I t ~ Figure 2. Composite recor sections for P waves generate by stacking the complex envelopes of seismograms from many events recore at the arrays of portable stations in northern Australia with reuction velocity 9km s-', (a) events in the Flores Arc, onesia, (b) events in New Guinea.

4 ~ 28. Variations in upper mantle structure 37 P which can be tracke across the recors of a number of the iniviual events. The ata set for the Flores Arc presente in Fig. 2 is augmente from that use by Bowman & Kennett (199) by the inclusion of some extra events. Much of the ata use for the New Guinea section came from the LTC experiment because there were relatively few suitable events along the northern margin of New Guinea uring the NWB recoring perio. Fig. 3, we present synthetic seismograms calculate for the nip reference moel (Fig. 6) using the WKBJ technique (Chapman 1978) in orer to provie a key to the main sets of arrivals in the composite sections in Fig. 2. We inicate the branches associate with each of the mantle iscontinuities by a combination of the epth an suffix (R for the waves reflecte back from the transition, r for waves refracte below the iscontinuity). The reference moel is chosen to give a general match to the character of both of the recor sections, but is not intene to be a specific interpretation of either section. Comparison of the P-recor sections (Fig. 2) with the synthetics for the reference moel (Fig. 3) shows significant ifferences in the timing an amplitue istribution between the sections for the sources in the Ffores an New Guinea regions. A striking feature of both observe sections is the prominence of arrivals associate with the 41 km transition, but there is notable ifference in the amplitue patterns in the two cases, an also the timing relative to the first arrivals. For the sources in the Flores Arc, with propagation through the zone to the northwest of the arrays, the refracte branch from below the 41 iscontinuity (41r) is prominent out to 2 km. For the sources in New Guinea to the NNE of the arrays, the retrograe reflecte branch is larger in this istance interval; the refraction is more prominent beyon 21km as a first arrival. The larger reflecte arrivals suggest that the contrast at the 41 km iscontinuity may be generally larger for paths from New Guinea. The balance of the amplitue of ifferent phase branches can be change by the nature of the source mechanism but the two branches 41r, 41R have similar take-off angles. The consistency of amplitue patterns over several hunre kilometres helps with the association of the character of the recor sections with structural features. However, - it shoul be note that Bowman & Kennett (199) present recors from two events in the Flores Arc with exceptionally clear arrivals from the 41 km transition but the majority of events have no such ramatic features. At larger istances the ifferences in the sections arise in part from the event istribution uring the recoring perios. For the Flores Arc, most of the recors beyon 22 km come from just two events with prominent arrivals that can be associate with the '66 km' iscontinuity. The branches for the reflecte an refracte arrivals are rather close together in this range but the timing of the branches favour a iscontinuity shallower than 66 km. For the New Guinea region, events in the New Britain subuction zone provie a much enser coverage at greater istances an there is more inication of the reflecte branch (66R) than the corresponing refraction beneath the 66 km transition. The emergence of the 66r branch as a first arrival occurs just outsie the istance range isplaye in Fig. 2. At short ranges there is much better ata coverage from the sources in the Flores Arc an in consequence better evience for the presence of a 21km iscontinuity to explain the separation between the onsets an the prominent energy 2s or more later over the interval km. ince we recognize that the locations use in preparing the composite recor sections may well have significant errors, we nee to know how these may affect the character of the composite sections. To simulate the effects of the errors we have prepare sections with the location of the sources ranomly perturbe by up to.3" from the original location. The character of the composite sections is preserve but there can be bulk time offsets of about a secon. The moification of the locations is not sufficient to isturb the prominent features of the section which are controlle by the triplications in the traveltime curve Figure 3. Theoretical seismogram section for P waves for the reference moel nip calculate using WKBJ technique for a surface source. The traveltime branches are enote by the iscontinuity with which they are associate (r-refraction below, R-retrograe reflection).

5 38. C. Dey et al. Although we have presente theoretical seismograms for a perfectly elastic meium for comparison with the recor sections of Fig. 3, the interpretations leaing to the velocity moels (fls-flores section, ngr-new Guinea section) shown in Fig. 6 were base in part on reflectivity calculations incluing attenuation. There can be a significant trae-off between the etails of velocity structure an attenuation (see e.g. Kennett 1975), but broa-ban observations at the Warramunga array (Gooy 1991) o not suggest the nee for any significant zones of increase attenuation for P waves. V-WAVE RECORD ECTION Using a similar set of proceures to those use to construct the composite P-wave sections (Fig. 2) we have attempte to generate composite recor sections for mantle arrivals from both the Flores an New Guinea source regions. The complex envelope stacks of the -wave arrivals from the vertical component seismometers in the temporary arrays are isplaye in Fig. 4. Three-component ata was only available from the permanent Warramunga array an was not sufficient to create recor-sections from the horizontal components. Although the V-wave sections of Fig. 4 have nothing like the clarity of the P-wave sections, they o show some consistent features. We note that the waves from the sources in New Guinea appear to have slightly lower frequency than those from the sources in the Flores Arc. There is moerately coherent high-frequency propagation for out to about 22km, thereafter it is ifficult to correlate arrivals. There are hints of later branches, but on these short-perio recors it is ifficult to have confience in any particular choices. Part of the problem arises because trace alignment before stacking has not been as effective as for P, the corrections are larger an therefore errors in location parameters have greater effect. However, the main problem arises from the lower-frequency content of the waves returne from the eeper-mantle iscontinuities. A broa-ban seismometer (Guralp CMG-3) has been operate at the Warramunga array since 1988 an there is now sufficient ata to buil up three-component recor sections (Gooy 1991). High-frequency waves are recore as the onset of the waveform from both V an H waves out to 22 km. At larger istances the onset of has much lower-frequency content ( Hz) an such intermeiate perio arrivals characterize the waves returne from the 41 an 66 km iscontinuities. n v) U \ aj, r t- I n v) U \ aj, r t- I Figure 4. Composite recor sections for waves generate by stacking the complex envelopes of seismograms from many events recore at the arrays of portable stations in northern Australia with reuction velocity 5 km s (a) events in the Flores Arc, onesia, (b) events in New Guinea.

6 Variations in upper mantle structure 39 VI v; 5. \ aj Figure 5. Theoretical seismogram section for waves for the reference moel njp calculate using WKBJ technique for a surface source. The traveltime branches are enote by the iscontinuity with which they are associatc (r-refraction below, R-retrograe reflection). WKBJ seismograms for the reference moel njp (base on the broa-ban results) are isplaye in Fig. 5 to provie a guie to the expecte locations of arrivals from eeper iscontinuities. No attenuation has been inclue in the synthetic seismogram calculations in Fig. 5. The ramatic loss of high frequencies for the 41 an 66km arrivals requires the presence of significant attenuation in the zone below 21 km. The times for the 21 km arrivals give a goo general match to the observations, occasional highfrequency arrivals beyon 22 km are consistent with surface multiples trappe in a high Q lithosphere extening own to 21 km. With the insight provie by the constraints from the broa-ban ata it is possible to fin reasonable correlations of features in the -wave sections with the theoretical times. For example, the 41r branch between 18 an 21 km for reuce times aroun 55 s may be evient in the ata of Fig. 4 but it woul have been ifficult to sustain this correlation without aitional information. DICUION the two previous sections we have isplaye composite recor sections for P- an -wave propagation through the upper mantle from sources in the Flores Arc, onesia an New Guinea to arrays of stations in northern Australia. The recor sections inicate noticeable ifferences for the propagation characteristics from the two sets of sources both in the relative timing of phase branches an the amplitue istribution between branches. Detaile moelling of the P-wave sections using synthetic-seismogram matching of the times an amplitue characteristics. following the proceures iscusse by Bowman & Kennett (199O), has le to the construction of two istinct moels for the Flores an New Guinea profiles. The P-wave velocity moels 3s-Flores, ngr-new Guinea are isplaye in Fig. 6 together with the reference moel njp use for the preparation of the theoretical seismogram sections in Figs 3 an 5. 1 ook D.a km/s Figure 6. Comparison of velocity moels for the northern Australian region. The P-wave moels fis, ngr euce by moelling the features of the composite recor sections in Fig. 3 are compare with the reference moel njp use to prepare the theoretical seismogram sections in Figs 3 an 5. The features of the -wavc velocity istribution below 21 km are erive from moelling of broa-ban observations. The P -wave moels, erive from the etaile analysis of the recor sections, sustain the ifferences in structure suggeste by the earlier work of Bowman & Kennett (199) for sources in the Flores Arc, an the slowness an traveltime stuies of impson, Mereu & King (1974) for sources in New Guinea observe at the Warramunga array. The P moels are well controlle to a epth of about 55 km, since the available aperture is not sufficient to give a goo resolution of the structure at the 66 km iscontinuity. The ifferences in upper mantle structure below 21 km summarize by the ifferences between the fls an ngr moels occur over a horizontal istance of about 1 km (the separation between the clusters of mipoints for the two profiles). The ifferences in the moels imply significant meium-scale variability within the upper mantle.

7 31. C. Dey et al. The 17s moel is very close to the NWB-1 moel of Bowman & Kennett (199) but the shallow structure which is not constraine by the available ata has been moifie slightly. The inclusion of an aitional event at larger range has also le to a slightly eeper 66O km iscontinuity, though still somewhat shallower than for ngr. A prominent 21 km iscontinuity is neee for paths thro!igh northwestern Australia (Hales, Muirhea & Rynn 198; Leven 1985) but only a slight transition is require to match the character of the New Guinea profile. The large ifferences in amplitue between the 21 r branch an the arrivals from the 41 km iscontinuity require a zone of reuce velocity graients below 21 km. The ngr moel is generally similar to MAK-1 (impson et al. 1974) but has sharper iscontinuities, which have been moelle as graient zones over a 1 km interval. A lower velocity than for fis is require at 21km to match the apparent velocity of the first arrivals between 18 an 22 km range. The graient above the 41 km iscontinuity is quite low an a large contrast at the iscontinuity than for fls is inclue to match the enhance amplitues on the 41 R branch in Fig. 3. Below the 41 km iscontinuity the velocity is slightly higher at first in the fis but the ngr moel is faster at epth. The ifferences in the size of the feature in the P-wave-spee istributions for fls an ngr near 21 km epth may well be associate with anisotropy at the base of the lithosphere as suggeste by Leven, Jackson & Ringwoo (1981). The presence of such anisotropy is supporte by consistent observations of about 1 s shear wave splitting between V an H arrivals observe on broa-ban recors for the 21R branch (Gooy 1991). uch splitting is not observe consistently from eeper structure. For -wave propagation we have been able to enhance the interpretation of the short-perio ata by incorporating aitional information from three-component broa-ban recorings. The general character of the sections is similar, although the arrivals from New Guinea ten to be of slightly lower frequency. There is efficient high-frequency propagation of V in the lithosphere (own to 21km) an the comparative absence of eeper mantle arrivals can be explaine by a frequency shift to intermeiate perios ue to attenuation below 21 km. Unfortunately we o not have enough information at this stage to tell whether the -wave structures at epth mirror those of the P-wave velocity istribution. REFERENCE Bowman, J. R. & Kennett, B. L. N., 199. An investigation of the upper mantle beneath NW Australia using a hybri seismograph array, Geophys. 1. t.. 11, Chapman, C. H., A new metho for computing synthetic seismograms. Geophys. J. R. astr. oc., 54, Dey,. C., Lateral variations in the upper mantle velocity structure uner northern Australia, PhD thesis, Australia National University. Gooy, A., 1991, Broa-ban stuies of the upper mantle beneath northern Australia, Honours thesis, Australian National University. Gran,. & Helmberger, D. V., Upper mantle shear structure of North America, Geophys. J. R. astr. oc., 76, Hales, A. L., Muirhea, K. J. & Rynn, J. W., 198. A compressional velocity istribution for the upper mantle, Tectonophysics, 63, Kennett, B. L. N., The effects of attenuation on seismograms, Bull. seism. oc. Am., 65, Kennett, B. L. N. & Bowman, J. R., The velocity structure an heterogeneity of the mantle, Phys. Earth planet. ter., 59, Leven, J. H., The application of synthctic seismograms in the interpretation of the uppcr mantle P-wave velocity structure in northern Australia, Phys. Earth. planet. ter., 38, Leven, J. H., Jackson, I. & Ringwoo, A. E., Upper mantle seismic anisotropy an lithospheric ecoupling, Nature, 239, Rial, J. A., Gran,. & Helmberger, D. V., A note on lateral variation in upper mantle shear-wave velocity across the Alpine front, Geophys. J. R. am. oc., 77, impson, D. W., Mereu, R. F. & King, D. W., An array stuy of P wave velocities in the upper mantle transition zone beneath northeastern Australia, Bull. seism. oc. Am., 64, Woohouse, J. H. & Dziewonski, A. M., Mapping the upper mantle, three imensional moelling of Earth structure by inversion of seismic waveforms, J. geophys. Res., 89,