Version classiqueVersion mobile

Risque, nature et société

Lucien Faugères
Christiane Villain-Gandossi

II. Étude de cas. La prévision des séismes (Grèce, Albanie)

Short term earthquake prediction from measurements of the electric field of the earth

Panayiotis Varotsos, Konstantin Alexopoulos et Maria Lazaridou


Depuis 1983, les stations de mesure installées par l'équipe VAN (d'après les initiales des noms de ses trois animateurs) enregistrent de brusques augmentations d'intensité des courants telluriques qui permettent, après interprétation de leurs principales caractéristiques, d'effectuer des prédictions de localisation et de magnitude des séismes dont ils paraissent les précurseurs. Les principes de la méthode sont illustrés à propos de quelques exemples. Les incertitudes de la méthode utilisée et certaines erreurs enregistrées sont analysées en vue de l'amélioration de celle-ci et pour progresser dans la compréhension des mécanismes géophysiques en cause dans l'apparition des signaux électro-sismiques (SES).

Texte intégral


1Since 1983 the electric field of the earth has continuously been monitored at 16 stations within Continental Greece. Transient variations of the field were observed that were followed by earthquakes (EQ), (Varotsos and Alexopoulos 1984a, b). Data obtained from the network of stations allowed in cases of isolated events a one to one correlation between the Seismic Electric Signals (SES) and the ensuing EQ. By studying the signals from each seismic area and their registration at each station a number of empirical rules have emerged that allow the prediction of epicentre and magnitude of the impending EQ. As they have been published in a number of papers (Varotsos and Alexopoulos 1984a, b; 1987, Varotsos et al. 1986; 1988, Varotsos and Lazaridou 1991) we will give here only a brief description of the main points.

2The most important feature for the feasibility of predictions is the fact that EQs of the same magnitude occurring in a given seismic region and measured at a given site always give the same maximum value of the field variation (signal strength). By applying these empirical rules the determination of the epicentre and the magnitude is possible with a satisfactory accuracy.

3The investigation was inspired from theoretical work on the delay that is needed for ionic crystals to reach the final state of polarization when subjected to a change of an electric field. The electric dipoles consisting of point defects in some ionic crystals have a large relaxation time so that they practically never acquire equilibrium polarization. Thermodynamical reasoning, however, predicts a dependence of the relaxation time on the stress (Varotsos and Alexopoulos 1986). If an increasing stress reduces the relaxation time, then, at a certain critical value, equilibrium is achieved at a fast rate. This change of polarization is equivalent to the emission of a current from the surface of the ionic material. A gradually increasing stress during the preparation period is expected. Assuming that the necessary value of the stress is reached before the fracture-stress, one is led to the expectation of premonitory signals.

4Empirical rules resulting from the Greek network of measuring stations have been extensively published. The present paper gives a short description of their main points as well as some more recent aspects. An extensive review of the up to date knowledge have been just published by Varotsos and Lazaridou, 1991 and by Varotsos, Alexopoulos and Lazaridou, 1991.


5The telluric electric field is determined by measuring the potential difference between two electrodes buried in the earth at a depth of 2m. Each pair of electrodes constitutes a dipole and is connected to a differential amplifier. For a continuous operation it was found that on the long run non-polarized electrodes of Pb/PbCl2 give the best results. At each station a number of dipoles are installed with various lengths in the directions North-South and East-West. In a microprocessor the voltage is turned into digital form and transmitted through data-lines of the telecommunication system to the central station near Athens. There, it is transformed back to analog form and registered. The total number of channels in operation amounts approximately to 120. The results are plotted in multipen recorders with a speed of 6 cm per hour. The system operates in the frequency range d.c. to 0.3 Hz. Examples of such registrations have been shown in a large number of publications. Dipoles that were found to give information of minor importance are registered on dot point recorders.

6The importance of a station depends mainly on the number of seismic regions to which it is receptive and to a second degree on its sensitivity (see below).


7The registrations usually show a high background noise which obliges many cross-checks to be made before a deviation is accepted as a SES (see Paper I). We describe the causes of such deceptive deviations:

8Cultural noise. This is produced by electric currents introduced into the earth. Near habitations the earthing of television sets seems to be the worst cause. Currents from trains operating on d.c. disturb up to 20km away (Varotsos and Lazaridou 1991).

9Magnetotelluric effects. They are recognised as such by appearing with the same form on the registrations of all the stations. An automatic neutralisation is shortly envisaged (Hadjiioannou et al. 1991).

10Lightning. It produces a signal like a step-function. Thunderbolts have destroyed the preamplifier although protection was embodied.

11Rain. It is most bothersome. The resulting change of the electrochemical potential can be recognized by the deviation not starting on all parallel dipoles at the same time; this is due to the water not reaching all the electrodes at the same moment.

12In order for a deviation to be accepted as an SES it has to correspond to the same current density on all parallel dipoles after suitable correction factors. These factors can be determined only after a long series of experimentation; they are due to inhomogeneities in the conductivity of the earth under the station (Varotsos and Lazaridou 1991).

13Each SES is characterised by its polarity which is defined as increase or decrease of the electric field component under consideration.

14Figure 1 shows a series of SES registered on a multipen recorder. Each of the signals in this example has a duration of around a few minutes. The depicted signals appeared on 4 channels of one station, in this case IOA. They preceded the destructive EQ sequence of Killini (September 22 and October 16, 1988) on the west coast of Greece (240 km to the west of Athens).


15We will briefly describe certain empirical rules that have been established concerning the SES. An extensive description can be found elsewhere (Varotsos and Lazaridou 1991).

16Form of signals. The duration of an SES lies between half a minute and several hours. It appears in different forms starting either progressively or abruptly (by the latter we mean faster than the movement of the recording pens). The ends can have both forms but an "abrupt end" has never been observed for signals that started progressively. No correlation between duration and site of the epicentre with the magnitude could be established.

17Polarity. It has been found that SESs from a given seismic region and registered at a given station always have the same polarity.

18Signal strength. The experimentally measured quantity is the potential difference V between the two electrodes at a distance L, so that V/L describes the field strength in the direction of the dipole. The maximum transient deviation AV/L (=E) will be defined as the strength of the signal. We will describe the empirically found properties for various arrays of dipoles.

19a) One dipole of a given station. The strength of premonitory signals from a given seismic area are found to increase with the magnitude of the EQ according to: log E = O.35 M + c (1)

20The field strength is connected to the current density j by: E = j ρ(2)

21where ρ is the resistivity of the earth. Because of inhomogeneities it was found that p depends on the dipole considered even on a local scale i.e. for parallel dipoles of a station. The resistivity cannot be calculated because j cannot be directly measured. As mentioned signals for EQ of the same magnitude and the same seismic area always have the same strength for a given dipole.

22b) Perpendicular dipoles of a given station. When the measurements are made on two perpendicular dipoles, e.g. in the directions EW and NS, the signal-strengths for EQs of equal magnitude and for the same seismic area do not have the same value. This is equivalent to stating Ens=Eew. In the plot therefore that connects the logarithm of the field at a given station and a given seismic area to the magnitude two lines have to be drawn. They are found to be parallel which gives a constant ratio Ens=Eew for all magnitudes from the corresponding area (Varotsos and Lazaridou 1991). This fact helps in predicting the epicentre of an EQ.

23c) Parallel dipoles of different stations. In order to compare simultaneous signals (if they are observed) at different stations we have to revert to relative values of ρns and ρew by comparing them to the ρvalues of a station that we choose as a "base station". As a suitable station we chose PIR (Varotsos and Alexopoulos 1984a,b) for which the two resistivities ρew, pir and ρns, pir are set equal to 1.

24Equation (2) can now be transformed into

25ENS = JNSρNS = JNSρNS ρNS’NS’PIR = JNSρNS’rel ρNS’PIR = JNSρNS,rel where ρns,rel is defined by ρns, rel = ρnsns,pir

26A similar relation holds for ρew,rel.

27Thus a working formula for each dipole of each station has been derived. Solving equation (2) we get a quantity jNS which now gives an indication about the current density in relative units. The total current density is: j=(j2 + ns + j2ew) 1/2= (Ens/ρrel.NS + E/ρrel,ew) 1/2(3)

28Stations with large values of ρrel,ns and prel,ew have a large sensitivity as they give strong signal strengths for a given current density.

291/r - law (r= distance between seism focus and station). By comparing the values of j calculated from Eq.3 for a given station for a large number of EQs of the same magnitude it was found that j is proportional to 1/r so that j.r depends only on the magnitude. This gives a general equation: log (j.r) = 0.35 + c

30A plot of an example is given in Fig. 2 where the points illustrate individual measurements (Varotsos and Alexopoulos 1984a,b); they lie on both sides of a straight line with deviations around 50% in the value of log (j.r). The error in deriving M from the plot is around 0.5 R.

31Lead time. By studying signals isolated in time and space so as to be sure that they correspond to the ensuing EQ it was found that they occur with a lead time between 7 hours and 11 days. In an approximative way one can state that signals corresponding to aftershocks i.e. belonging to activated areas, have usually the shorter values e.g. of the order of several hours. We stress that for cases of electric activities (i.e. series of signals) the time-window of 11 days holds only between the initiation of the electrical activity and the initiation of the seismic activity; the time-lag between the largest SES and the strongest EQ may be appreciably higher i.e. a few weeks (for details see Varotsos and Lazaridou 1991).

32General remarks. All these rules have been found by studying signals for magnitudes larger than 4-units. Not all such events are suitable for the proof of a one to one correlation because during periods of high activity the time between consecutive EQs is of the same order of magnitude as the lead time. Therefore when extending the study to such weak EQs only events clearly isolated in time and site were considered.


33It is an empirical fact that a station does not register signals from all seismic areas for which magnitude and epicentral distance would warrant a measurable signal. Each seismic area is therefore characterised by the stations that can respond to the SES that it emits. This selectivity effect is not a directivity effect in the sense that it does not depend on the direction in which the station lies in relation to the signal-source. Assume that sites A,B and C lie in a straight line and that a signal is emitted from site A. Station B might not be receptive to the emitted SES although the latter was registered at C which lies at a larger distance. Selectivity is also not reversible in the sense that in the above example signals from area A could well be registered at C but not vice-versa.

34It is admitted that directivity of emission might contribute to minor extent to the selectivity observation although we think that the latter depends mainly on the path followed by the current between the seismic source and the registering station. The 1/r-law suggests that the signal strength does not dissipate as expected for a dipole source which would give a 1/r3 - law. It rather indicates that the current chooses to some extent channels of higher conductivity.


35The prediction of the epicentre is rather straight forward when the SES is simultaneously registered at many stations. As the current density decreases with 1/r, after determining its value from Eq.3 of paragraph 4 from the measured components of E one can draw around each pair of stations 1 and 2 an apollonian circle for the geometrical locus of points that fulfil the condition rl/r2 = j2/j1. By drawing the same circles for all pairs of stations one finds at their intersection the desired epicentre.

36When the SES is registered only at one station A (which is the most usual case), the epicentre can still be determined step by step by iterated exclusion of seismic regions. The first step derives from the selectivity effect which excludes all seismic areas to which the station A is non-receptive. The second step is based on the polarity of the signal. As mentioned in paragraph 4, signals from a given seismic area always appear on each dipole of a given station with the same polarity. Because of this fact all seismic areas which would be acceptable from the point of view of selectivity but gave different polarity have to be excluded. The third step pinpoints to the seismic areas that have given in the past the same value for Ens/Eew as the one occurring in the SES under examination. In paragraph 4,b we have already stressed that this ratio has a characteristic value for each pair "station-seismic area".

37One sees that the prediction from a registration at a single station A leads to a result only if the behaviour of all seismic areas in respect to station A is already known. The number of seismic areas that can be excluded obviously increases with the total number of stations of the network because of the enhanced probability of a receptive station existing beyond station A. When the observed value of Ens/Eew happens to be near the value for other areas an estimation of the epicentre can be made by intercalation.

38Once the epicentre has been determined the magnitude can be found on hand of the log AV/L versus M plot that is valid for a given station and a given seismic area. After a number of signals have been registered from various seismic sources such plots can be drawn for each station. The results can be depicted for each station by a selectivity map. This is a survey of all regions that are known to give signals that can be registered at that station. Of course regions outside the selectivity areas do not have obligatorily to be regions for which we are sure that signals cannot be registered; they simply are regions for which the situation is unknown. On the other hand one can also draw for each station a "non-selectivity map" in which regions are indicated for which it is empirically known that their signals are not collected at station A. Details on the construction of a selectivity map can be found elsewhere (Varotsos, Alexopoulos and Lazaridou 1991).


39The long period of experimentation has given the possibility to check the reliability of the predictions that have been made. The predictions are filed in the form of telegrams that state the epicentral vector in regards to Athens and the magnitude. They are only expedited when the expected magnitude is around 5-units or larger. In cases where several signals appear within a short time the telegram mentions "electrical activity". As the exact value of the lead time cannot be stated, it is implicitly understood that the ensuing EQ is predicted to occur within the time window mentioned above.

40In 1985 an Earthquake Prediction Council was established in Greece in which the content of the telegrams was discussed before the EQoccurrence (Varotsos et al. 1988). Since 1986 the procedure agreed upon with the Greek Government is to announce the predictions by telegrams forwarded to the Interministerial Council. From May 15, 1988 on telegrams have also been sent to a restricted number of Seismological Institutes in France and Japan. The comparison of predictions with ensuing EQ allow a quantitative reliability evaluation to be made. Two probability values are of interest. The probability of a correct prediction P1 is the ratio of correct predictions to the total number of predictions; it serves a measure of the extent to which one can rely on a prediction that has been made to actually occur, of course, under the prescribed error limits of time epicentre and magnitude. A second type of probability is the probability P2 of an EQ that has occurred to have been correctly predicted (i.e. the ratio of the correct predictions over the total number of EQ above a certain magnitude threshold). Its value is of interest for cases where warning could be envisaged.

41The value of P1 and P2 depend on the conditions which must be fulfilled in order for the prediction to be considered as correct. Beyond the demand that the EQ should occur within the time-window the errors Ar and AM in the site of the epicentre and the magnitude should not exceed an appropriate limit. These can be arbitrarily chosen and strongly influence the calculated result. A further limitation has to be made as to the extent of the area in which an EQ that occurred should be considered as having the possibility of being connected to the prediction. We propose the following error limits: 100 km for Ar which is the mean distance between the stations of the network; 0.7 magnitude units for AM which is about double the magnitude deviations declared in the Bulletin by the various seismological centres. The most arbitrary point in calculating P2 is the distance from the predicted epicentre within which EQs that occurred within the lead time should be considered as connected to the prediction. One simple method is to extract from the Bulletin the EQ that occurred within or not far from the perimeter of the network of stations. As the network has a length of 500 km in the NS direction such a decision is not very meaningful as a 5 M EQ is of importance for a circle of less than 100 km. A more realistic assumption, maybe, is to consider only EQs that occurred within a circle the radius of which is chosen in each case depending on the actual magnitude. Such a criterion although quite as arbitrary as the first one, gives a better view for the reliability of warning.

42To the number of incorrect predictions one must add the cases where an EQ occurred under the conditions that would warrant the appearance of a signal but not prediction had been issued. In some of these missed cases the study of the registrations in retrospect showed that a signal had actually appeared but had been misregarded in the noise.

43In reliability calculations an estimation of the value of the probability for achieving a correct prediction by chance should be made; such a calculation demands the separate study in each case of the time elapsed since an EQ of similar magnitude had occurred in the region. Of course for stronger EQs the occurrence of an event within the 100 km which is one of the criteria for a correct prediction is so rare so that it is highly improbable to happen within the lead time except for cases of aftershock series that correspond to main events with magnitudes 6.0-units or larger.


44Tables of predictions and ensuing EQs have been published in various papers (Varotsos and Alexopoulos 1984a, Varotsos et al. 1986; 1988, Varotsos and Lazaridou 1991) so that a complete and uninterrupted list is available. In the present paper we give two time charts: the first time chart (Fig. 3) is given for the period from May 15, 1988 to Nov. 30, 1989. On the upper side all expedited predictions are marked. As already mentioned they refer (all except one) to signals that should be precursors to magnitudes equal or above 5 R. On the lower side all EQs stronger than (or equal to) the indicated threshold magnitude of 5-units are shown. Those EQs marked with a dashed bar are weaker than the threshold but always within the error limit of 0.7 R. In a second time chart (Fig. 4) we give all EQs with Ms>, 5.5 along with their predictions for a three year period i.e. January 1, 1987 to November 30, 1989 (see also Table 1). The detailed data can be found in the paper by Varotsos and Lazaridou (1991) whereas copies of all telegrams are published by Dologlou (1990).

45By disregarding the case of the EQ on March 19, 1989 (when two of the authors were absent in France) and inspection of Table 1 indicates the following: Twelve EQs with Ms > 5.5 have occurred during a three year period. For ten of these 12 EQs predictions were issued well in advance. Among these 10 predictions, only two have relatively large deviations i.e. the inaccuracy in the epicentre determination was larger than 100 km. If we restrict ourselves to EQs with Ms > 5.8, 4 events have occurred (i.e. Febr. 27, 1987; May 18, 1988; October 16, 1988; August 20, 1989); for all these 4 EQs predictions were issued, three of which are considered as successful because they achieved Ar < 30 km and AM < 0.7 mag-units.

Table 1. All earthquakes with MS > 5.5 and their corresponding predictions. Area (41.00-36.00)N (25.00-19.00)E. Time period: 1987-1989

Table 1. All earthquakes with MS &gt; 5.5 and their corresponding predictions. Area (41.00-36.00)N (25.00-19.00)E. Time period: 1987-1989

*the first earthquake in the destructive sequence of Killini (Main shock on OCT. 16, 1988, MS=6.0)



Fig. 1. Sequence of 4 SES that were collected at the station IOA (Northwestern Greece) before the Killini-Vartholomio destructive seismic activity. In view of this SES-activity a public announcement was made by Prof. Haroun Tazieff (for details see Varotsos and Lazaridou 1991).

Fig. 2. The amplitude of the SES after considering the relative effective resistivities in the two directions (EW and NS) scales with the magnitude. The data of this plot can be found in the papers by Varotsos and Alexopoulos 1984a, b.

Fig. 3. Time chart of all the telegrams issued along with all earthquakes with Ms > 5.0-units. The dashed bars correspond to EQs with Ms < 5.0 that followed some telegrams. For the data see Dologlou (1990) and Varotsos and Lazaridou (1991).

Fig. 4. Time chart of all EQ with Ms >5.5 during the period January 1, 1987 to November 30, 1989. In this figure we have also inserted the corresponding predictions. For the data see Dologlou (1990) and Varotsos and Lazaridou (1991); the symbol x denotes a period in which the authors were absent in France and hence a prediction could not be issued.



1. E. Dologlou. "Earthquake Prediction in Greece by means of Seismic Electric Signals". Paper presented at the International Conference on Measurements and Theoretical models of the Earth's Electric Field Variations Related to Earthquakes, Athens, February 6-8, 1990.

2. D. Hadjioannou, F. Vallianatos, K. Eftaxias, V. Hadjicontis, and K. Nomicos. 1991. "Magnetotelluric noise subtraction from the VAN measurements". Tectonophysics (to be published).

3. P. Varotsos, and K. Alexopoulos. 1984a. "Physical properties of the variations of the electric field of the earth preceding earthquakes", I. Tectonophysics, 110: 73-98.

4. P. Varotsos, and K. Alexopoulos,. 1984b. "Physical properties of the electric field of the earth preceding earthquakes, II. Determination of epicenter and magnitude". Tectonophysics, 110: 99-125.

5. P. Varotsos and K. Alexopoulos. 1986."Stimulated current emission in the earth: piezostimulated currents and related geophysical aspects”. In: S. Amelinckx, R. Gevers and J. Nihoul (Editors). Thermodynamics of Point Defects and their Relation with Bulk Properties. North-Holland, Amsterdam, pp. 136-142, 403-406, 410-412, 417-420.

5. P. Varotsos and K. Alexopoulos. 1987. "Physical properties of the variations of the electric field of the earth preceding earthquakes". III. Tectonophysics, 136: 335-339.

6. P. Varotsos, K. Alexopoulos, K. Nomicos and M. Lazaridou. 1986. "Earthquake prediction and electric signals". Nature, 322: 120.

7. P. Varotsos, K. Alexopoulos, K. Nomicos, and M. Lazaridou, M. 1988. "Official earthquake prediction procedure in Greece". In: O. Kulhanek (Editor), Seismic Source Physics and Earthquake Prediction Research. Tectonophysics 152: 193-196.

8. P. Varotsos and M. Lazaridou. 1991."Latest aspects on earthquake prediction in Greece based on seismic electric signals I". Tectonophysics 186 (in press).

9. P. Varotsos, K. Alexopoulos and M. Lazaridou. "Latest aspects on earthquake prediction in Greece based on seismic electric signals II". Tectonophysics (to be published).


Département de Physique, Université d'Athènes.

Académie des Sciences d'Athènes.

Laboratoire VAN, Département de Physique, Université d'Athènes.

© Éditions de la Sorbonne, 1996

Conditions d’utilisation :

Cette publication numérique est issue d’un traitement automatique par reconnaissance optique de caractères.
Rechercher dans OpenEdition Search

Vous allez être redirigé vers OpenEdition Search