GEOSCOPE Stations Research Articles

Rupture processes of large deep‐focus earthquakes from inversion of moment rate functions

We determine the source time histories of five recent (1994–1996) large, deep‐focus earthquakes using a method that inverts for fault slip from far‐field moment rate functions. The moment rate functions are obtained through the deconvolution of multiple body wave phases using broadband records from Global Seismic Network (GSN) and GEOSCOPE stations. Tests of this method on synthetic data indicate that it is successful in determining the low‐frequency rupture characteristics of deep earthquakes under a variety of complicating conditions. We find that source parameters such as average rupture velocity and stress drop are highly variable among the events studied and that some unusual characteristics exhibited by the June 9, 1994, great Bolivian earthquake are also found for other events. Comparison of the slip distributions with background seismicity and aftershock locations indicates that most of the moment release for large deep‐focus earthquakes is probably occurring within the active slab interior. This provides further evidence that temperature‐controlled mechanisms (such as transformational faulting) play a large role in deep earthquake faulting. Most of the events studied also show a tendency for horizontal rupture propagation, suggesting that isobaric processes may be an important factor in controlling progression of the rupture. Large gaps in the slip distributions point to possible occurrence on multiple fault planes.

Azimuthal anisotropy in the upper mantle from observations ofP-to-Sconverted phases: application to southeast Australia

SUMMARY Converted-waves analysis is used to map the anisotropy as a function of depth in the upper mantle. Stacking of SV -a ndT-component records of several seismic events well distributed in azimuth helps to detect these converted waves.Weighting coe⁄cients are used to extract the anisotropic signal in the records. Synthetic examples show that a harmonic analysis helps to separate azimuthal anisotropy eiects from the eiects of lateral heterogeneity, such as those resultingfrom dipping structures. A good azimuthal coverage of the data set is important in order to separate the diierent eiects in the stack sections obtained for diierent azimuthal orders. The procedure is applied to data observed at GEOSCOPE station CAN and gives a model with two anisotropic layers in the upper mantle.The upper layer, located between the crust and a depth of140 km, has its fast axis in an east^west direction. The lower layer, 40 km thick, has its fast axis in a north^south direction. The fast axis directions are in agreement with results obtained from surface-wave analysis and tectonic information.

Lateral Variations of Lg Coda Q in Southern Mexico

—Broad band digital three-component data recorded at UNM, a GEOSCOPE station, were used to estimate Lg coda Q for 34 medium size (3.9 ≤m b ≤ 6.3) earthquakes with travel paths laying in different geological provinces of southern Mexico in an effort to establish the possible existence of geological structures acting as wave guides and/or travel paths of low attenuation between the Pacific coast and the Valley of Mexico. The stacked spectral ratio method proposed by XIE and NUTTLI (1988) was chosen for computing the coda Q. The variation range of Q 0 (Q at 1 Hz) and the frequency dependence parameter η estimates averaged on the frequency interval of 0.5 to 2 Hz for the regions and the three components considered are: i) Guerrero region 173 ≤Q¯ 0≤ 182 and 0.6 ≤Q¯ 0 ≤ 0.7, ii) Oaxaca region 183 ≤Q¯ 0 ≤ 198 and 0.6 ≤Q¯ 0 ≤ 0.8, iii) Michoacan-Jalisco region 187 ≤Q¯ 0 ≤ 204 and 0.7 ≤Q¯ 0 ≤ 0.8 and iv) eastern portion of the Transmexican Volcanic Belt (TMVB) 313 ≤Q 0≤ 335 and η = 0.9. ¶The results show a very high coda Q for the TMVB as compared to other regions of southern Mexico. This unexpected result is difficult to reconcile with the geophysical characteristics of the TMVB, e.g., low seismicity, high volcanic activity and high heat flow typical of a highly attenuating (low Q) region. Visual inspection of seismograms indicates that for earthquakes with seismic waves traveling along the TMVB, the amplitude decay of Lg coda is anomalously slow as compared to other earthquakes in southern Mexico. Thus, it seems that the high Q value found does not entirely reflect the attenuation characteristics of the TMVB but it is probably contaminated by a wave-guide effect. This phenomenon produces an enhancement in the time duration of the Lg wave trains travelling along this geological structure. This result is important to establish the role played by the transmission medium in the extremely long duration of ground motion observed during the September 19, 1985 Michoacan earthquake. ¶The overall spatial distribution of coda Q values indicates that events with focus in the Michoacan-Jalisco and Oaxaca regions yield slightly higher values than those from Guerrero. This feature is more pronounced for the horizontal component of coda Q. A slight dependence of average coda Q −1 on earthquake focal depth is observed in the frequency range of 0.2 to 1.0 Hz approximately on the horizontal component. Deeper (h > 50 km) events yield lower values of Q −1 than shallower events. For frequencies higher than 1.0 Hz no clear dependence of Q −1 on focal depth is observed. However, due to the estimates uncertainties this result is not clearly established.

The effects of the atmospheric pressure changes on seismic signals or how to improve the quality of a station

Abstract Seismic investigations are mainly limited by seismic noise. Two microbarometers have been installed in the seismic vault of two different GEOSCOPE stations, one at SSB and the other at TAM. All vertical components and most of the horizontal components show a significant correlation with pressure. In order to correct the seismic signals from the atmospheric pressure noise, a transfer function between the pressure data and the seismic data is inverted. Results show that, after correction, the noise levels reached on the horizontal components are similar between the two stations, and the vertical components display noise levels below the low-noise model as defined by Peterson (1993). This technique reduces part of the noise and allows detection of small earthquakes and a better extraction of normal modes. The analysis of the lowest normal modes of the Earth excited by the MS = 8.2 Macquarie Island earthquake is given to illustrate the perspectives of the method.

Mapping S-velocity heterogeneities in the D″ region, from SmKS differential travel times

Seismic velocity heterogeneities in the D″ region are studied using differential travel time residuals between SmKS waves. Previous studies indicate that regional variations in the propagation times of these waves are not due to the propagation through the liquid core. Furthermore, they are only weakly influenced by the topography of the core-mantle boundary (CMB). Thus, the differential travel time anomalies are interpreted as the result of lower-mantle heterogeneity. We compiled all the available published data, and added new observations of travel times from GEOSCOPE stations. We thus obtained 807 S2KS-SKS and 209 S3KS-S2KS differential times. Differential travel time residuals with respect to the mean Earth model PREM (Preliminary Reference Earth Model) are corrected for ellipticity and for mantle heterogeneity down to 2590 km depth (300 km above the CMB) using the 3-dimensional S-wave velocity model SH12-WM13. This model gives a 10% variance reduction of the residuals. The remaining variance of residuals is higher for S2KS-SKS than for S3KS-S2KS, which is consistent with a D″ origin for the heterogeneities. The data are inverted for regional variations using a least-squares analysis, assuming a single-layer D″ structure 300 km thick. The Earth is divided in 150 equal-area sectors, about 115 of which are sampled by four rays or more. We performed inversions using 18 different geographic grids and averaged the results, so as to minimize the artefacts induced by imposing artificial block boundaries. The velocity heterogeneity map, though affected by the poor sampling of the oceans, reveals significant long-wavelength structure, with velocity perturbations exceeding ±3% with respect to a reference velocity, which is slightly lower than the PREM one. This heterogeneity map is consistent with the results from most of previous studies, global as well as regional. The location of some of the high-velocity anomalies coincides with places where subducted slab material may have accumulated at the CMB. These high-velocity heterogeneities are also generally correlated with places where a discontinuous velocity increase of several per cent has been detected at the top of D″.

Attenuation in the uppermost inner core from broad-band GEOSCOPEPKPdata

The attenuation in the uppermost 500 km of the inner core is investigated at the global scale from the comparison of PKP(DF) and PKP(BC) phases recorded at broad-band GEOSCOPE stations. From the amplitude ratio DFIBC, we show that for most paths the attenuation is not frequency-dependent in the range 0.2–2 Hz. This result follows from the linearity of ln(DF/BC) with frequency. A notable exception occurs for paths from the south-west Pacific to Europe, as analysed in previous studies which infer a frequency dependence of inner core attenuation. The DF/BC amplitude ratios are determined at 3 s periods for the different paths. The data include the diffracted branch of BC at long distances, providing a deeper sampling of the inner core by the DF phase. The variations of the DF/BC amplitude ratio with epicentral distance exhibit a rather large scatter around a mean curve. This mean curve is compared to the theoretical ratios obtained for several earth models using synthetic seismograms. A trade-off exists between the velocity at the base of the liquid core and the attenuation in the uppermost inner core. Robust features include a velocity decrease in the lowermost liquid core, inferred from previous studies of the PKP(BC) propagation times, and a Q decrease at the top of the inner core, required by the DF/BC amplitude ratios. The preferred model exhibits a constant velocity gradient in the lowermost 150 km of the liquid core, and Q = 200 in the uppermost 100 km of the inner core, Q= 440 below. With the adopted velocity model, an increase of Q with depth is required by the data. The amplitude ratio DF/BC exhibits regional variations, which cannot be attributed with certainty to regional variations of Q inside the inner core, due to the large Fresnel zone at the inner core and the possibility of wave focusing and defocusing in the mantle. On the other hand, with the available data, we failed to detect an anisotropy in attenuation inside the inner core.

On the possibility of anisotropy in the D″ layer as inferred from the polarization of diffracted S waves

S waves diffracted along the core-mantle boundary (Sd) show in a few cases anomalously large SV components. To test if azimuthal anisotropy in D″ can be a cause for these anomalies, a method based on the Langer approximation and on perturbation theory is developed to model Sd waves in the presence of anisotropy in D″. By comparing the synthetics in azimuthally anisotropic models with those in isotropic or transversely isotropic models, we search for the characteristic features in the data which would discriminate between the different models. The SVd/SHd amplitude ratio turns out not to be a discriminating feature. A ratio of 30% at an epicentral distance of 115° can, for example, be explained by 1% of azimuthal anisotropy at the core-mantle boundary (CMB), decreasing linearly to zero 150 km above, by a mean of 0.5% azimuthal anisotropy through D″, by a realistic negative velocity gradient in an isotropic model of D″, or by transverse isotropy in D″. The phase difference between the SH and SV components and their frequency content are not discriminating either. Thus, with a single recording, azimuthal anisotropy cannot be discriminated from isotropy or transverse isotropy in D″. The only effect characteristic of azimuthal anisotropy is that the horizontal particle motion of Sd hardly depends on the focal mechanism, as opposed to the linear dependence which exists in isotropic or transversely isotropic models. Also, it becomes invariant with epicentral distance beyond 115°. To discriminate between different kinds of models, the particle motion of Sd waves from several events with known focal mechanisms therefore needs to be analysed. To test the hypothesis of Vinnik et al. (Geophys. Res. Lett., 16: 519–522, 1989) concerning the origin of the anomalous SVd they observed, we analyse three of their data. These are Fiji events recorded at the North American Geoscope station WFM and at the North Atlantic World Wide Standardized Seismograph Network station BEC, all showing large SVd waves. The Sd waveforms of the two events recorded at the smallest epicentral distance cannot be explained simultaneously by a laterally homogeneous isotropic or a transversely isotropic model for D″. An azimuthally anisotropic model with 1% azimuthal anisotropy at the CMB explains them better. However, neither an isotropic nor an anisotropic model could be found which explains the waveforms observed at the furthest station. The hypothesis of azimuthal anisotropy in D″ under the Pacific Ocean to explain these clearly anomalous waveforms is not confirmed. A much larger data set should, however, be analysed before drawing a conclusion for that region. The example of the data of Vinnik et al. shows that the clear predictions which are made here as to which waveforms are expected if D″ is anisotropic allow discriminating studies of D″.

Antarctica II: Upper-mantle structure from velocities and anisotropy

Abstract To improve the lateral resolution of three-dimensional seismic wave velocity models in Antarctica and the surrounding oceans, we have analysed direct earthquake-to-station Rayleigh-wave data observed on the vertical high-gain long-period and the very long period components of seven GEOSCOPE stations located in the southern hemisphere and three other stations at equatorial latitudes. The phase velocities of Rayleigh waves along 400 well-distributed paths are obtained in the period range 60–300 s, by fitting the data with synthetic seismograms computed with known source parameters in a reference earth model represented by the Preliminary Reference Earth Model (PREM). Corrections for shallow layers have been carefully applied to the observed phase velocities. The geographical distributions of phase velocities and azimuthal anisotropy are then computed with the tomographic method without any a priori regionalization developed by Montagner (Ann. Geophys., 4(B3): 283–294, 1986). The results show some new and important features of Antarctica and the southern hemisphere. The locations of velocity anomalies are well resolved. The eastern part of Antarctica corresponds to a craton-like structure down to depths of about 250 km, and the highest velocities are observed in Enderby Land, where some of the oldest rocks in the world have been sampled. The low velocities are located along the ridges encircling the Antarctic continent. The lowest velocities appear in some areas corresponding to hotspots (Crozet, Kerguelen, Macquarie and Balleny Islands). Also, an elongated low velocity is found on the western flank of the Transantarctic Mountains, which might be related to the existence of a rift zone similar to the African rift. The Australia-Antarctica Discordance (AAD) presents slow velocities near the surface but fast velocities below the lithosphere. These main features are discussed in the framework of the Gondwana hypothesis and the earlier supercontinent. The first azimuthal anisotropy results are also discussed. Anisotropy values are smaller within the Antarctic continent than in the surrounding oceans. They are also small in the AAD but particularly large in the areas around it, suggesting an active tectonic process characterized by a downward flow at depth, a good candidate for a cold spot or a new subduction zone.

The French Pilot Experiment OFM-SISMOBS: first scientific results on noise level and event detection

Abstract The seismic data recorded during the French Pilot Experiment OFM-SISMOBS (Observatoire Fond de Mer) were carefully analyzed. This experiment was successfully conducted between 28 April and 11 May 1992. Notwithstanding the technical goal of the experiment, which was to show the feasibility of installing and recovering two sets of three-component broadband seismometers (one inside an Ocean Drilling Program (ODP) borehole and another inside an ocean-bottom seismograph (OBS) sphere in the vicinity of the hole), the second goal of the experiment was to obtain for the first time the seismic noise level in the broadband range 0.5–3600 s, to conduct a comparative study of broadband noise on the sea-floor, downhole and on a continent, and to determine the detection threshold of seismic events. After the installation of both sets of seismometers, seismic signals were recorded continuously for 10 days. The analysis of these signals shows that the seismic noise is smaller in the period range 4–30 s for both sea-floor seismometers (OFM; Observatoire Fond de Mer) and downhole seismometers (OFP; Observatoire Fond de Puits) than at a typical broadband continental station such as GEOSCOPE Station SSB. The noise is smaller than that at SSB up to 600 s for OFM. The noise on vertical components is much smaller than on the horizontal ones. This difference might be explained by instrument settling. It was also observed that the noise level tends to decrease with time for both OFM and OFP, which means that the equilibrium stage was not attained by the end of the experiment. The patterns of microseismic noise in oceanic and continental areas are completely different. The background microseismic noise is shifted towards shorter periods for OFM and OFP compared with a continental station. This might be related to the difference in the crustal structure between oceans and continents. The low level of seismic noise implies that the detection threshold of earthquakes is very low, and it has been possible to record correctly teleseismic earthquakes of magnitude as small as 5.3. It was also possible to extract the Earth tide oceanic signal. Therefore, the experiment was a complete technical and scientific success, and demonstrated that the installation of a permanent broadband seismic and geophysical observatory on the sea-floor is now possible and can provide the scientific community with high-quality seismic data.

Origin of precursors to teleseismic S waves

Abstract In the literature, there are descriptions of precursors to teleseismic S waves that are polarized as teleseimic P. They are commonly regarded as phases converted from SV to P underneath the seismographic stations. We present observations of precursors at the broadband digital GEOSCOPE stations in the distance range from 45° to 95° and at periods around 10 sec. These precursors are polarized as the teleseismic P but cannot be interpreted in terms of conversion underneath the station. We propose that they are formed by conversion and scattering from S to P at the free surface and scattering of the resulting P in the lithosphere of the region between the source and the receiver. The apparent velocity of S in the region of scattering is usually around 7 km/sec, which implies a long wavepath of the converted P in the continental crust. Apparently, S-to-P and P-to-P scattering play an important role in forming teleseismic wave fields even in the relatively long-period band around 10 sec. A possibility of observing true S-P converted phases originating in the lithosphere underneath the station is certainly not ruled out. However, our analysis is a warning that scattered waves sometimes can be erroneously taken for converted phases.

Source parameters of the 1989 Loma Prieta Earthquake determined from long‐period Rayleigh waves

The source parameters of the Loma Prieta earthquake are determined using long‐period Rayleigh waves recorded by USGS/ERIS, IDA/IRIS, and GEOSCOPE stations. The source mechanism is well‐constrained by the Rayleigh wave radiation pattern, with a dip = 70 (±5)°, strike = 130 (±5)°, rake = 135 (±5)°, and moment = 3.4 (±0.5) × 1019 Nm (Mw = 7.0). This mechanism is generally consistent with independent body wave determinations. The most stable long‐period waves, with periods from 200 to 275 s, indicate that the source process has a centroid time of about 10 s, somewhat longer than that indicated by body waves (about 5–6 s). This discrepancy cannot be uniquely attributed to source effects because of uncertainties in the propagation corrections. The importance of using surface waves with short propagation paths for analysis of moderate size earthquakes such as the Loma Prieta event is demonstrated by the unreasonably long source durations inferred from R3 arrivals.

The Romanian earthquake of August 30, 1986: A study based on GEOSCOPE very long-period and broadband data

The Romanian earthquake of August 30, 1986 is the second largest intermediate depth event in this area since the worldwide deployment of digital instrumentation, and the first one since the installation of GEOSCOPE network. It offers the unique opportunity to document this well-known but poorly understood zone of deep continental seismicity using high quality teleseismic data in different frequency bands. The source is well constrained both from very-long period surface wave data observed on GEOSCOPE stations and, independently, from body wave modelling at various worldwide stations. The depth obtained is approximately 140 kilometers, the seismic moment, 0.8 1027 dyne-cm and the mechanism from both data sets is very similar to that of the previous 1940 and 1977 Vancrea events, indicating that these events, although having occurred at noticeably different depths, are expressions of the same tectonic process. However, from the detailed study of the source using broadband data, it can be inferred that the source presents much less complexity than the 1977 event.

Macquarie earthquake of May 23, 1989

We read with great interest the news report in Eos (June 16, 1989) on the very large Macquarie earthquake of May 23, 1989, which left U.S. scientists divided as to whether this earthquake was of strike slip or thrust type or a mixture of both.Indeed, we think we have the solution to this dilemma, since thanks to teletransmitted data from 11 Geoscope stations recording both very long‐period and broad band data, we were able to obtain a reliable preliminary mechanism less than 48 hours after this earthquake occurred. Figure 1 shows this mechanism, which is a practically pure right‐lateral strike slip, consistent with the general trend in the Macquarie ridge area as indicated from the CMT catalog for the past 10 years [Dziewonski et al., 1989], and in particular, very similar to the mechanisms of the two largest events which occurred in this area in this time period (May 25, 1981 and July 7, 1982). By moment tensor inversion of very long‐period Rayleigh waves, we obtained a moment of 2.2×1028 dyne‐cm and a centroid time of 28 s, which yields a rating for this earthquake as “average‐to‐fast” with respect to the moment/duration relation. So far, we haven't found any conclusive evidence for strong horizontal directivity, while it seems that the rupture may have extended down to (or up from) at least 40 km. These features of the solution offer a good explanation as to why no tsunami was observed for this earthquake. The confidence in our “very‐long period” solution comes from its very good compatibility with broad band data from the three teletransmitted Geoscope stations that lie within the epicentral distance range 30°− 90° from the epicenter of the Macquarie event, both in terms of mechanism as well as source duration. For shallow earthquakes the earthquake depth and the vertical dip‐slip components of faulting are poorly constrained by the very‐long period data. The addition of only a few broad band seismograms in the analysis can remove this uncertainty. Figure 2 shows two examples of fits for SH waves obtained in a combined analysis of long‐period surface waves and broad band body waves [Ekstrom, 1989]. The teleseismic SH arrival at Papeete had an amplitude of 0.9 mm. The Macquarie region lies so far south in the southern hemisphere that very few records from existing broad band digital stations appropriate for mantle P and S wave analysis will ever be available. For Geoscope (which presently counts 20 operational broad band digital stations), besides PPT (Tahiti), KIP (Hawaii) and RER (Reunion Island), which were teletransmitted (CAN, Australia,is too close), we still expect data in this distance range from stations NOU (New Caledonia), DRV (Dumont d'Urville), PAF (Kerguelen) and CRZF (Crozet Island), the last three of which will take several months to arrive because they will travel part of the way by ship. Table 1 lists the coordinates and position with respect to the epicenter of the 11 stations whose data were teletransmitted.

P‐coda at distances 80° – 120° and structure beneath subduction zones: Example of Chile

Strong phases are sometimes observed between P and PP for earthquakes at epicentral distances 80° – 120°. The present study analyzes as an example broadband data from a Chilean earthquake recorded in Europe by the NARS network, and by worldwide distributed GEOSCOPE stations. Clear phases are observed in the P‐coda only for the European stations.Underside reflections at upper mantle discontinuities beneath the PP bounce point at the Mid‐Atlantic Ridge are unable to explain these phases. A discontinuity in the lowermost mantle at the top of D″ is also discarded as a possible explanation.Reflections and conversions at a strong reflector about 200 km deep beneath Chile provide a satisfying explanation of the observations. The P‐coda is dominated by waves leaving the focus as S‐waves. The focal solution and the resulting radiation pattern explain the variations with azimuth and epicentral distance of the phases observed in the P‐coda.The reflector at 200 km beneath Chile may be the upper boundary of a piece of slab detached from the Nazca plate. Alternatively, it may result from the chemical differenciation of the slab during its subduction. Modelling P‐coda generated in other subduction zones may help to specify this differenciation process.

Geographical Distribution of Surface-Wave Velocities and 3-D Upper-Mantle Structure In Africa

SUMMARY Long-period records of surface waves from WWSSN, GDSN, IDA and GEOSCOPE stations are analysed to retrieve purely continental dispersion of the African continent in the period range of 20-200 s for Rayleigh waves and 20-125 s for Love waves. The phase and group velocities are obtained by two-station and single-station methods, with earthquakes within or near the continent. A regionalization without any a priori constraints is performed over the whole African continent to retrieve local phase and group velocities. Phase- and group-velocity geographical distributions show in the northeast of Africa, near the Red Sea, a region with lower velocities extending to the west, and high velocities in the northwestern and southern parts of Africa. At long periods, there is a general agreement with global models, but the resolution is improved in this regional study. In the resulting 3-D structure, cratonic areas show a fast lithosphere from the NW to the SE. Whereas the high SV velocities of the NW area extend deeper than 200 km, lower velocities appear below this depth in the centre. In the Red Sea region low velocities are locally deep-rooted, with a slow lithospheric layer 150 km thick extending to the west. In this region polarization anisotropy is evidenced by Love-Rayleigh incompatibility. Short-period lateral variations are used to model the crustal thickness.

Evidence for anisotropy in north east Africa, from geographical and azimuthal distribution of Rayleigh wave velocities, and average upper mantle structure

Long period surface waves from WWSSN, GDSN, SRO and GEOSCOPE stations are analysed to obtain the dispersion of surface waves in the North‐East of Africa. Lateral variations of Rayleigh wave velocities are retrieved simultaneously with the average azimuthal anisotropy. As a first result, the low velocity anomaly under the Red Sea is apparent at all periods. A second result is a nearly North‐South direction of maximum group and phase velocities. This azimuthal anisotropy direction is in agreement with the direction of absolute plate motion in this region.Anisotropy is also evidenced by simultaneous inversion of Love and Rayleigh wave dispersion along the path TAM‐AGD. In the resulting models, isotropic and anisotropic, very low mantle SV‐wave velocities extend up to the Moho. For SH waves, we find a strong discrepancy with SV waves, SH‐wave velocities being higher by 5% down to 200 km under the crust. This is characteristic of lithospheric polarization anisotropy.The S‐wave anisotropy (polarization and azimuthal) found in this region, including the Red Sea, is compatible with the absolute motion of the African plate: it can be interpreted, considering the low lithospheric S‐velocities, as a consequence of plastic flow in the weak lithosphere of an anomalous mantle. This anisotropy, evidenced here on S waves, may be due, as Pn and lithospheric S‐wave anisotropy found in other continental areas, to dynamical processes related to plate motion.

A possible local anomaly in the 670-km discontinuity beneath Kerguelen

The mapping of the 670-km discontinuity in regions of ascending or descending flows is of major interest for a better understanding of the nature of this discontinuity. In this paper, an attempt is made to investigate the properties of deep mantle discontinuities beneath the Kerguelen hot spot. Anomalous P to S converted waves are observed for some earthquakes recorded at the Kerguelen long-period GEOSCOPE station; the characteristics of these phases are specified owing to a three-dimensional polarization analysis of the signals. A tilted discontinuity is able to explain the data; this discontinuity dips toward the northeast with at 20°. The depth at which the conversion takes place is ∼ 880 km, assuming standard velocities in the upper mantle beneath Kerguelen, or 810 km if a hot upper mantle with a Poisson's ratio of 0.30–0.31 is present. However, the constraints given by the data are not strong enough to specify unambiguously this depth, a conversion at 600 km (or 550 km assuming a hot upper mantle) may alternatively be proposed. A strong velocity increase at the discontinuity is in each case necessary to explain the amplitude of the converted waves. Geophysical arguments show that this tilted discontinuity must be a local feature; possible other mechanisms, such as wave focusing due to undulations of the discontinuity, may also be proposed.

Mantle wave deviations from ‘pure-path’ propagation on aspherical models of the Earth by Gaussian beam waveform synthesis

Departures from the classical approximations of geometrical optics and great circle propagation are investigated by surface wave ray-tracing on laterally heterogeneous models and computation of synthetic seismograms by the Gaussian beam method. Higher order deviations, such as time differences between odd and even orbits, residuals from great-circle average times, and fluctuations in the succession of eigenfrequencies in the spectra are sought for and compared with long-period great-circle observations. In the case of the Akita-Oki (Honshu) earthquake (1983) at the Geoscope station PAF, the deviations are of the order of the first order tomographic effect. They are interpreted by a transverse velocity gradient shifting the orbits. In other examples where great-circle propagation is nearly satisfied, geometrical effects of curvature are displayed. Frequency fluctuations due to lateral heterogeneity appear in the synthetic spectrum.

Phase velocity distribution in the Indian Ocean and Indonesian region inferred from GEOSCOPE records

To improve the lateral resolution of 3 dimensional models of seismic wave velocities in the Indian Ocean and surrounding regions we analyzed direct earthquake to station Rayleigh waves observed on the vertical long‐period components of four GEOSCOPE stations [TAM (Algeria), PCR (La Réunion), PAF (Kerguelen) and AGR (Djibouti)].The phase velocities along 51 well‐distributed paths are obtained in the period range 40‐350 s after corrections for the phase pattern at the source and for the instrumental response. A tomographic method is then applied (Montagner, 1986a) to compute the geographical distribution of phase velocities, without any a priori regionalization.The location of velocity anomalies is quite stable and well resolved. The phase velocity distributions are compared with global results from Nakanishi and Anderson (1984b), Woodhouse and Dziewonski (1984), Tanimoto (1985), Nataf et al. (1986) and with regional studies using group velocities for the same region (Montagner, 1986a,b,c).

First two‐station results for long‐period surface waves velocity from the Geoscope Stations in Africa

Rayleigh and Love phase and group velocity dispersion measurements are performed along a purely african path between the Geoscope stations TAM and AGR, for the Leewards Isl. earthquake of March 16, 1985. This regional study is important as it brings deep‐seated structure information in a seismologically blank region of Africa. It enables to improve the spatial resolution of larger scale models. For periods longer than 160 s. and up to 250 s., the fundamental Rayleigh wave dispersion is in agreement with prediction by global models. For shorter periods, the results fall below this prediction, which is an indication of lithospheric S‐wave velocities lower than those of the models. The results are also compared with previous regionalizations. The Rayleigh second higher mode is extracted at AGR, and the phase velocity dispersion obtained in the period range 50‐140 s is in agreement with the theoretical dispersion for PREM.