A physico-mathematical approach for earthquake prediction
A physico-mathematical approach for earthquake prediction
Edelvays Spassov
Geophysical Institute - BAS, 1113 Sofia, Bulgaria
Abstract
The results of several investigations on different subjects brought to the idea of obtaining an objective seismic zonation and then to a possible midterm earthquake prediction. The application of Principal Components Method, Cluster and Factor Analysis and Pattern Recognition by using various geologic input data gives such a zonation. This classification can be improved by changing the initial parameters and checked out by different statistical criteria. The interpretation of the results and Pattern Recognition allow determination of geological and physical peculiarities of seismically active (dangerous) and quiet (safe) regions and thus make probable long or midterm prognosis.
Introduction
The main reasons for unsolved problems in earthquake prediction are basically of physical matter such as unclear processes in earthquake source, the absence of two identical quakes, the unpossibility to "prepare" the real seismic experiment etc. Other problems in this field are the subjectivenes of the estimated observations or difficulties in matching the different in scale and dimensions input parameters while a comprehensive study is necessary for decreasing of subjectiveness. In the same time there are well known mathematical methods making easy the physical understanding of the investigated subject or allowing the joint consideration of quite different parameters. A number of geodynamical and structural investigations ([1], [2], [3], [4]) brought us to a chain of methods supported by a package of computer programs which could be successfully applied in seismic hazard estimation and even in earthquake prediction.
Data and methods
All available geologic parameters as relief denivelation, gravity field, heat flow, photolineaments, fracturing knots, lithosphere and crustal depth etc. are used as input data for the established chain of methods. The investigated territory is usually divided in rectangular or square sells and average value of each parameter is described to the geometrical center of the cell. Thus the initial data matrix has been formed. The details of the data treatment could be found in ([1], [2], [3] and [4]). The main programs used in computational procedure are given in ([5], [6] and [7]). The basic methods in the chain are Principal Components (PC), Cluster (CA) and Factor (FA) Analysis and Pattern Recognition (PR). Only a few things have to be mentioned here. The formation of the group of sells (zones) is done upon the simple condition that each group contains only territorially adjacent sells whose values - of the parameters are adjacent in the space of PC (or factors) too. The objectiveness of the sells classification has been checked out by four different statistical criteria. Some of the initial parameters have been excluded from the computational procedure what gives another zonation but also additional information about the importance of each parameter. After a steady zonation is reached the PR follows. Several sells from well known seismically most active zone and some from most quiet region are used as teaching subjects. Thus a new zonation of the territory is obtained and it is supposed to represent seismic zonation of the region.
Results and discussion
Only results concerning seismic hazard and prediction will be discussed here. They are based on the study of the territory of Czechoslovakia. The zonation achieved by this chain of methods coincides well with the seismic zonation of Czechoslovakia obtained by the classical methods. However, the new classification [8] allows precise determination (in %) of the seismic danger (or probability of an earthquake occurrence with the particular magnitude) for each sell. More important seems to be the result describing the peculiarities of the seismically active (dangerous) regions and those determining quiet (safe) zones. In our case the active areas are characterised by a high denivelation of the relief and heat flow, by thin lithosphere and by low fracturing and number of knots. Seismically quiet zones lie usually over the territories with a low plain relief and vertical movement velocity, thick crust and high fracturing and number of knots. Such results can be easily compared with similar results for other regions. It seems very possible that the including of some of the seismic time sequences in the computations a progress in long and midterm earthquake prognosis could be reached.
Conclusion
Only a short description of some preliminary investigations of a chain of methods supported by a corresponding package of computer programs has been presented here. These results suggest only that some of the physical problems towards the earthquake prediction could be overcame by this way.
References
[1]. Ranguelov et al. 1986 Present geodynamic features of south Bulgaria. Geol. Balc., 16 (3).
[2]. E. Spassov, N. Dotzev, S. Shanov and A. Andreev. 1988. A complex study of the lithosphere beneath territory of Bulgaria. Bulg. Geoph. J., XIV, 1.
[3]. E. Spassov, S. Shanov, A. Andreev and N. Dotzev. 1990. Statistical interrelations of some geologic-geophysical parameters of the lithosphere. Physica Zemly, No 3.
[4]. F. Demirmen. 1969. Multivariate procedure and FORTRAN IV program for evaluation and improvement of classification. Kans. State Geol. Surv. Comp. Contr., 31.
[5]. W. Wahlsted and J. Davis. 1968. FORTRAN IV program for computation and display of principal components. Kans. State Geol. Surv. Comp. contr.., 21
[6]. A. Soloviev. 1988. The package of pattern recognition programs. User's guide. Workshop: Global geophysical informatics with applications to research in earthquake predictions and reduction of seismic risk. ICTR, Trieste, Italy
[7]. E. Spassov, I. Kostadinov and S. Shanov. 1996. Crustal and lithosphere regionalization of Czechoslovakia by statistical geophysics. Studia Geoph. & geodetica (in press).