3D TOMOGRAPHIC STRUCTURE OF THE UPPER MANTLE BENEATH THE CENTRAL PART OF THE EUROASIAN CONTINENT



Ivan Kulakov
United Institute of Geology, Geophysics and Mineralogy,
Siberian Branch of Russian Academy of Sciences.
3, University Ave., Novosibirsk, 630090, Russia.
E-mail: ivank@uiggm.nsc.ru
Phones: (3832) 351444, 355832 (home)



Introduction



Seismic tomography is the most powerful instrument for determination of 3D structure of deep Earth's interiors. Tomographic models obtained at the global and regional scales serve as the basis for determination of geodynamical state of the Earth. The global tomographic models (Dziewonski, 1984, Woodhouse and Dziewonski, 1984, Su et al., 1994) have evident correlation with other geophysical and geological characteristics (geoid, thickness of the lithosphere, relief) that proves some theoretical geodynamical speculations based on numerical and physical modelling (Hager and Richards, 1989). The same situation takes place in the regional modelling. For example, tomographic model of Europe obtained by Spakman et al, 1993, clearly shows all perculiarities predicted by geologists (existence of slabs, root of mountains, spreding) and also gives fundamentally new possibilities for investigation of the actual geodynamical state of the region and to trace the history of its development (De Jonge et al., 1994).

Unfortunately there are only few regions where the regional tomographic models of the upper mantle were obtained. Mostly these regions are coinciding with centers of technical civilization and, at the same time, of high seismological activity: Europe (Spakman et al., 1993), west of the North America (Humphreys and Dueker, 1993), Asia Minor (Hearn and Ni, 1993), Eastern Pacific (Puspito et al., 1994) etc. But most of the Earth's territory is not yet studied satisfactorily. The center of the Euroasian continent is very interesting region from geological point of view. On the one hand it is the center of the largest continent and such location should definitely determine the structure of flows in the mantle, on the other hand a great collizion zone connected with closure of Paloasian ocean (Sengor et al., 1995) is associated with this area. Manifestations of recent tectonic activities, like as Baikal rift zone, orogenesis, high seismisity, take place within this inner plate region.
The attemptions to obtain an information about mantle structure under this territory was mounted by several authors since 60th. First quantitative tomographic model was obtained by Alekseev et al., 1971 The seismic structure of the upper mantle beneath a broad area located in the center of Euroasian continent is described in the paper. This area includes such tectonicly active regions as: Altai and Sayan orogenic areas, Baikal rift zone, western part of Mongoly and North-Western part of China (Fig.1). A part of this area (Altai-Sayan region) was studied in our recent investigation (Kulakov et al., 1995). A large negative seismic anomalies was obtained under the regional depression of Ubsu-Nur lake and under Hubsugul lake. These anomalies were interpreted in the paper as hot, ascending flows On the contrary, the mountain areas of Altai and Sayan was associated with positive anomalies in the upper mantle interpreted as descending cold flows. However this tomographic model was obtained in condition of very poor data, when the level of noise was higher than the relevant residuals. Therefore the problem of reliability of these results was very sharp. In spite of several tests made in the work, destined to prove the validity of the model, there was many doubts with regards to fairness of the conclusions proposed. We confessed in the paper that only a verification using independent data could prove more or less definitely the reliability of the obtained configuration of anomalies. The main aim of the actual research is verification of the recent tomographic model (Kulakov et al., 1995), and, in the case of success, expanding of the research to the adjacent areas. In so doing, the key problem was the following: which data we can use?

Problem of the Data


More than 20 permanent, stationary seismic stations registering the information both on local and telesesmical events are situated on the territory of Altay-Sayan region. However, application of the information about far sources registered at these stations is connected with many difficulties which made our recent research (Kulakov et al., 1995) problematic.

First, these data are stored "on the paper". Their selection is hard manual work expanding the time of realisation of research to the years. In addition, when the adjacent territories were studied, the information is often stored in different towns and it is practically impossible to get them into computer readable form.

Second, record of the seismograms at the stations is occured by old, analogue equipement, interpretation of the seismograms is manual, that cause significant errors of phase determination and travel time calculation. Third, small number of the stations and their uneven distribution in the region make the study of the upper 100-200 km impossible, since the rays from differend stations do not form there sufficient system of intersections. There was attempts to apply the data on the travel timef from local events registered by regional seismological network. But, as was apparent, the Altai- Sayan Seismological Expedition, which is responsible for processing of all seismological information in Altai-Sayan region, takes into accont only, so called, direct waves, passing in the crust, that is convinient for study of hypocentral parameters of a source, but does not give absolutely an information about subcrustal structure.

All these reasons have sent us to break with application of the regional seismological network data and to invoke methods applying the data of world seismological network allowing extraction of a relevant information on interiors of the region under study. One of the such methods is:

Inverse Teleseismic Scheme (ITS)



An essence of ITS is based on utilization of travel times of the rays from events located within a region under study registered at the stations of world seismological network. All this information is stored at the servers of ISC and can be easily taken by everyone for any region. In particular, for Altay-Sayan region we have about 200 events registered by more then 1000 station. High number of stations registering one event and high number of events registered by one station allow a precize calculation of the source and receiver corrections. The stations of world seismological network equiped by instruments of high quality, and it allows the calculation of travel times of necessary phases to a higher precision than that at regional stations.

A dense distribution of sources within the study region allows the itersection of rays from different ray tubes at the shallower depths in comparison with traditional teleseismic scheme, that permits us to obtain a relevant information above the 100 km in depth.

ITS allows the study of any tectonicly active region, beyond the dependence on technical state of the regional seismological service (and even in the case of its absence) avoiding permissions of local authorities. It has been this scheme which has served as a basis of current research.

Data processing



Preliminary processing of the data has to do with all sources within the study area (Fig. 1). The number of sources within the area used in the research is about 500. About 2000 stations of the world seiemological network take part into their registration. The total number of the rays available is about 36 000. The processing of initial data begins with determination of an average travel-time curve for the total set of data. A problem of determination of a reference model for the study region corresponding to this travel-time curve did not occupy our attention, since, as the experience of teleseismic research implies (Kulakov et al., 1995), variation of a reference model, especially in its upper part, has not a marked effect on the result of teleseismic inversion. The reference travel time curve was approximated by a polynom of L-th order. In the region from 20* to 93*, corresponding to the rays passing through the lower mantle and not entering to the core, the travel-time curve of refracted P waves, which is quite smooth, is well described by the order L=6. The calculation of these decomposition parameters is made by the Least Squares

Method.



Here, as in the traditional teleseismic scheme, the relevant residuals are presented as:

where is real travel time
is travel time according the reference curve
is receiver (station) correction,
is source correction. Source and receiver corrections are determined from a minimum condition of residual norm: . The station correction includes all error factors superimposed on the travel time of a ray between the station and bottom border of the study volume: anomalies in the mantle and the lithosphere and relief beneath the station, errors of reference model determination etc. The source corrections includes errors connected with the sources and the uppermost region under them: error of hypocenter depth determination, anomalies in the crust etc.

Preliminary estimation of these correction is made by simple averaging of travel times relatively refenence travel-time curve:

The further, more accurate determination of the source and receiver corrections is occured on the assumption of munimum of norm , that is realized by SVD decomposition.

When processing the data and searching of the source and receivers corrections, the selection was made only over the sources registered by no less than N stations, and receivers registering no less than M sources. In our case the value of N and M was correspondingly 25 and 10.

The next processing was made over separated fragments of the region, but the data obtained for the whole area were used as the residuals.

Percuilarities of inversion technics.



Here, as in the vast majority of practical tomographic investigations, the inversion was based on the matrix representation:

where is the data vector, is the matrix of the first derivatives , is finit set of unknown parameters describing the velocity distribution in the study volume. As in our previous research, the parameterization of the model (method of representation of velocity distribution within the study area by finit number of parameters) was made by Vertex Method described in (Kulakov et al., 1995). Here this method is realized by subdivision of the study volume by tetrahedral sells with constant gradient of a velocity parameter (velocity, or slowness, or squared slowness). Such approximation allows continuous velocity distribution within the study volume. The values of velocity variations at the vertices of these tetrahedral blocks play the role of unknown parameters. The direct problem of ray tracing in such media is described in (Cerveny, 1991). An important perculiarity of our method of parametrization consists of coordination of the density of input and output information which is realized by determination of a sell size according to the density of the rays. Automatic algorithms allowing such construction were developed in the actual research. The coordination of input and output information allows the uniform filling of rows of A-matrix by nonzero elements, that significantly improves its quality and, correspondingly, increases stability of inversion toward the random noise. The advantages of such method of parameterisation are described in (Kulakov et al., 1995). In particular, it was noted that such way allows stable reconstruction of anomaly even if the noise in the data is two time higher than the relevant signal. The tests showing the advantages of such coordination in comparison with uniform grid are shown in the chapter "Verification of the results and tests". The inversion of the A-matrix was made by the Singular Value Decomposition (SVD) method. The criteria for determination of optimal quantity of singular numbers, which play the role of regularization parameter, were developed in the paper (Kulakov and Druzhinina, 1996).

Results



The calculation of velocity structure in the upper mantle was made within separated regions covering the study territory and overlaping partially one another. That allowed us, on the one hand, to operate with fairly small matrix (300x3000 elements) which can be inverted by the SVD method, and, on the other hand, it give us another way of verification of the obtained results by comparison of anomaly configuration in the overlaping ares. ITS permits a movement from one fragment to another along all tectonically areas with sufficient number of earthquakes registered at the world seismological network. The obtained velocity model is given in Figures as set of vertical sections, whose location is shown in the firure . Dark areas mean the negative seismic anomalies.

Comparison with configuration of anomalies obtained in (Kulakov et al., 1995) shows fairly good correlation of these two models, obtained by different approaches with absolutely independent sets of data. In the both models large negative anomalies are associated with a large depression in the area of Ubsu- Nur lake and with Hubsugul lake. In the contrary, large positive anomalies are observed beneath the mountain areas of Altai and Sayan. Such correlation proves evidently the general conclusions made in our previous paper.

Verification of the results and tests



Good correlation of the obtained models in the overlaping areas between different fragments as well as satisfactory agreement with the tomografic model obtained latter attests the reliability of the obtained results, but to verify them and to test the algorithm we used several tests.

Reconstruction with different grids. To study the influence of a parameterization upon the result of velocity calculation different grids where applied for the same region. In particular, an effect given by the coordination of imput and output information was investigated by inversion with an uniform grid. In the current research a triangulate grid with constant size of sells was applied for the real data. The analysis of the matrix structure shows that the condition numver in such case (a ratio between maximum and minimum eigenvalues reflecting the quality of the matrix) was about 14000, while in the case with coordination it was about 16. Correspondingly a more or less stable solution in the first case was achieved when only 60-70 singular numbers were used, while in the second case such stability was acheved at 110-130 singular numbers. Results of inversion in both case (fig. ) evidently shows the effectivity of our way of parameterization.

The parametrization by rectangular sells of uniform size were analized and compared with triangulate grid with coordination at syntetic models in the paper (Druzhinina and Kulakov, 1996). This research showed significant efficiency of the vertix parameterization in comparison with traditional rectangular parameterization.

A reconstruction of the same anomaly by use of the same type of grids, but with different nunber of parameters is very important to show how the increasing of parameter number can improve the resolution of a model. This question was also studied in the paper (Druzhinina and Kulakov, 1996), and it was shown that in the case of significant noise in the data, an increase of the number of parameters does not improve the quality of the reconstruction, but even gives significant nonstability and causes manifestation of falsh details. Reconstruction with independent sets of data. In order to estimate the influence of random noise in the data upon the final reconstruction, an independent inversion of two independent groups of data was made. All events was separated by a random criterium (with odd and even numbers) and the all processing was made independently. The result of inversion (fig. ) shows a significantly better correlation that in the same test made for model KTK95. That means that the quality of data in this research is significanly higher. Reconstruction with noised data. In order to show abilities of the algorithm used in the actual research to give the useful information in condition of high level of noise, an inversion both for syntethic (Druzhinina and Kulakov, 1996) and real anomalies were made in the case of high level of noise. A configuration of noise distribution was taken from (Spakman et al., 1993). Such reconstruction was made for different types of anomalies and the results was generally satisfactory.
</