Surveillance of Thera Volcano, Greece: Monitoring of the Geomagnetic Field

The total magnetic field was measured at 5 stations on Thera in April 1987 and twice in April 1988. Each time simultaneous data were recorded at one permanent base station and another peripheral one for periods of 10 hours. The main purpose of the initial phases of the experiment was to study the applicability of the volcano magnetic method in the area of Thera for prediction purposes. It is shown that the method should yield useful results; the method of data acquisition, together with an improved version of the standard spatial differencing technique adopted for noise reduction yielded residual uncertainties typically less than 1 nT and sometimes as low as 0.2 nT. Such uncertainties are among the lowest reported. Magnetic field changes prior to volcanic eruptions of 10 nT have been reported elsewhere. If signals of similar magnitude should precede the next eruption of Thera, they could be detected by the method if continuously recording instruments are installed. At some stations a field change of about 3 nT was observed over the period of 1 year. However, without a long and regular observational sequence, no inferences can be made with respect to the cause of the change.

INTRODUCTION

The volcano of Thera is one of the most active in the eastern Mediterranean. A detailed list of its numerous (more than 13) eruptions, including the major one that destroyed the Minoan civilization around 1450 BC, can be found in Papadopoulos (1985) and Xynogalas and Chailas (1987). The last eruption occurred in 1950 and, evidently, this area is characterized by a relatively high volcanic hazard. As a consequence, an efficient assessment of the volcanic hazard is imperative, especially in view of the fact that during summer, the population of the island is more than doubled. For this purpose the Institute of Geology and Mineral Exploration (IGME) together with the Department of Geophysics and Geothermy of Athens University began a multi-disciplinary study on the island to establish a basis for the assessment of its volcanic hazard.

Unlike earthquakes, volcanic eruptions can be successfully predicted today, after careful monitoring of certain precursory phenomena. For example, the eruptions of Kilauea (Hawaii) are predicted accurately on a regular basis (e.g. Klein 1984). The eruptions of this volcano are particularly easy to predict because frequent eruptions give ample opportunity to study geophysical parameters during the pre-eruption phase. In cases where volcanoes are inactive for several decades, the first new eruption can be a surprise, while subsequent ones are easily predicted, as in the case of the Krafla (Iceland) volcanic crisis (e.g. Bjornsson 1985). Some recent predictions of eruptions achieved substantial saving of lives and property, as in the case of Mount St. Augustine, for example, where lives and property were undoubtedly saved by the accurate prediction of Mount St. Helens (Decker 1981). This was achieved despite the fact that this volcano does not give frequent opportunity to study its eruptive pattern: it had not erupted for more than a century (Mullineaux and Crandell 1981; Decker 1981). This successful prediction is particularly encouraging for the case of Thera because there are similarities in the eruptive pattern of Mount St. Helens and that of Thera.

The methods usually applied include various branches of geophysics, geochemistry, geology etc. (e.g. Decker 1982). This paper describes the initial stages and preliminary results of an experimental attempt to monitor the total magnetic field of the earth in the area of Thera, in order to detect possible changes associated with volcanic activity. Such changes and their causative physical processes are known by the generic term volcano magnetic effect.

It is well known that the magnetic susceptibility of ferromagnetic minerals depends on factors like temperature and pressure (stress, e.g. Stacey and Banerjee 1974). Temperature acts as a demagnetization factor, completely demagnetizing a mineral when it rises above its associated Curie point. Pressure will modify the stress regime in which the host rock of the magnetic minerals is embedded. Thus, a stress increase may cause a strengthening of the field through piezomagnetic phenomena, a process that is largely reversible when excessive pressure (stress) is alleviated. Also, dilatancy (the creation of cracks in stressed rocks) may be related to piezomagnetism (see Rossignol 1982). Electrokinetic filtration, which corresponds to water and electrolytes flowing into pores, can be another magnetic field modification factor (Mizutani et al. 1976). For an extensive theoretical treatment of susceptibility to modification through temperature and pressure, the reader is referred to Stacey and Banerjee (1974).

Ascending magmas and their associated phenomena can produce all the susceptibility to modification mechanisms mentioned above. As summarized by Rossignol (1982) volcanomagnetic (VM) effects are caused by:

1. Thermal mechanisms, by heat transfer through direct contact, or fumarolic activity and the formation of new fumarolic zones and fields. The demagnetization caused by the difference in temperature is involved in explaining the observed changes in the local magnetic field.
2. Modification of the magnetic properties of rocks from the excessive pressure caused by the volatile constituents of rising magmas. Such a mechanism is analogous to the seismo-magnetic effect.
3. A combination of both processes above.

It becomes apparent that the actual modification mechanism may depend on the type of the volcano and its eruptive process. For hot-spot volcanoes like Mt. Kilauea, Hawaii, rocks are basaltic with high susceptibility, but their eruptive process is not explosive. Therefore, pressure and piezomagnetic effects are not important, although even in this low-stress environment a VM effect was detected (Davis et al. 1979). On the other hand, explosively eruptive volcanoes like Mount St. Helens undergo a considerable pressure build-up prior to eruptions. Although they are poorer in magnetic minerals, the VM precursors can reach several nT (Johnston et al. 1981). This has consequences on the time constants associated with the initiation of detectable VM effects. Obviously, heat transfer through almost insulating rocks is a very slow process, while the propagation of stress fields is almost instantaneous. Therefore, the VM effects due to pressure build-up have a time constant of a few to several days prior to the eruption and are therefore valuable for forecasting (Johnston and Stacey 1969; Davis et al. 1984; Zlotnicki and Le Mouel 1988). The same effect due to heat transfer takes longer and is a more gradual process.

From the above discussion it becomes evident that a careful monitoring of the magnetic fields is an efficient and relatively cheap means of eruption forecasting. Thera is considered to be a Vulcanian type volcano with a history of explosive eruptions the largest of which destroyed a civilization. It also produced moderate lava outpours at other times (Papadopoulos 1985; Xynogalas and Chailas 1987). The former mode of eruption, however, is more worrying, and indicates that pressure build-up may be an important local field modification factor. Given the short time constants of such precursory phenomena, there is strong need for frequent monitoring, as is done in other populated volcanic islands (e.g. Zlotnicki 1986; Zlotnicki et al. 1986).

DATA ACQUISITION

A network of 9 permanent magnetometric stations was installed in 1987; their locations can be seen in Fig. 1. Each station comprises a concrete base, in which a removable, 2 m high aluminum pole can be inserted vertically. The sensor of the magnetometer is fixed on a wooden base, at the top end of the pole. In this manner it is always at a standard height and orientation every time it is installed. Plate 1 shows the above configuration in operation. The station at Pyrgos (PYR) was chosen to be our permanently occupied base, mainly because it is established on the practically non-magnetic Triassic limestones outcropping in the area.

Recording of the earth's magnetic (geomagnetic) field takes place simultaneously in two different stations, one of which is always the base (PYR), on a sunset to sunrise basis, so that the strong daytime variations of the geomagnetic field, as well as intense industrial (anthropogenic) interference, are excluded from our data. The instruments used are automatic, digitally recording, total field proton precession magnetometers, manufactured by Geometrics (type G-856). They are usually set to record at a 60 or 90 second interval. Consequently, about 600 measurements are taken per instrument per night. Only the stations at Pyrgos, Oia (OIA), Fira (AGA), Akrotiri (AKR) and Kolombos cape (KLM) were occupied in the initial stages of the VM experiment. Although better quality results are affordable and more information available if an array of magnetometers and, optionally, a three-component magnetometer is used, limited resources and logistical support in the initial stages of the experiment restricted us to the required minimum of 2 total field magnetometers (also see section on data analysis). Several problems have by now been solved and all future fieldwork will be carried out with a small array of at least four instruments.

During this preliminary phase of our monitoring effort field measurements were made during visits to Thera in April 1987 and in two separate weeks in April 1988.

DATA ANALYSIS - RESULTS

The geomagnetic field at any point on the earth's surface is the sum of (1)the earth's main field; (2) the effects of remanent magnetization; (3) magnetic fields arising directly from the magnetizing susceptibility of rocks and (4) the effects of magnetic fields generated by ionospheric and magnetospheric currents; (5) the effects of magnetic fields generated by subsurface (Telluric) currents, i.e. local electromagnetic induction (EMI) processes due to contrasting rock conductivities; (6) anthropogenic magnetic fields arising from industrial activity. VM signals are expected to be due to changes in contributions (2) and (3) above, produced by the heat released and/or by the pressure exerted by rising magmas and their volatile constituents. All the contributions listed above are unaffected by changes in the stress and/or heat fields and are, therefore, unwanted 'noise' that must be eliminated.

Over a time window of a few hours, noise is either frequency independent (e.g. the earth's main field), or frequency dependent (e.g. EMI fields). Anthropogenic noise can fall into both categories depending on its source. The frequency dependent magnetic fields originating in the magnetosphere and ionosphere are slowly varying, long wavelength processes and, over a range of several tens of kilometres, can be very satisfactorily approximated with uniform plane waves. This implies that measurements made up to a frequency of several Hz, in locations separated by a few kilometres, will register these fields, as well as secondary magnetic fields arising from regional (long wavelength) EMI, as equal intensity in-phase processes. It is clear that one straightforward method to remove this kind of noise is to make differential measurements of the magnetic field between two locations on the surface separated by a few kilometres, i.e. calculate

δ B_ij (t) =B_i (t) - B_j (t)

where B_i (t) and B_j (t) are total field measurements at sites i and j respectively. Then the study of the VM effect is reduced to studying the evolution of B_ij with time, along the baseline joining i and j. This technique ensures the total elimination of all uniform in-phase processes that extend over locations i, j and typically reduces the effects of the magnetic disturbances by about 95% (e.g. Rossignol 1982; Davis et al. 1981).

The remaining 5% of the natural magnetic disturbance is due to local EMI which, being dependent on the subsurface conductivity distribution, is generally dissimilar between any two locations and, therefore, out-of-phase. It is thus impossible to remove such effects by simple differential measurements; their intensity, however, may be sufficient to provide large uncertainties and thus mask the VM effect. Moreover, total field magnetometers used in such studies suffer the limitation that they measure only one component of the geomagnetic field, while effective elimination of magnetic disturbances requires the knowledge of all three components (Davis et al. 1981). In general the conductivity structure that generates local EMI fields and, consequently, the EM response (impedance) of the earth at the vicinity of each magnetometer will be invariant over time windows of a few hours duration (unless a major event such as an earthquake occurs). Therefore, any frequency dependence in δB_ij (t), resulting from the out-of-phase part of the EMI fields in the vicinity of magnetometers i and j, will interrelate the EM responses (impedances) of locations i and j. Therefore, one may argue that the frequency dependent part of δB_ij (t) is in itself due to some invariant EM response and thus predictable and removable. This operation will assume some form of predictive deconvolution (e.g. Peacock and Treitel 1969). Poehls and Jackson (1978) predict the total, field at the location of each magnetometer i and j, using a remote reference vector magnetometer k, and the relationship:

^{^}B_i (t) =AX_k (t) + BY_k (t) + CZ_k (t) + e (t)

where X_k (t), Y_k (t) and Z_k (t) are the reference vector field components, A, B, C, are transfer functions interrelating site i with reference site k, ^{^}B_i(t) is the predicted total field value at i and e (t) = B_i (t) - ^{^}B_i (t). Subsequently, they calculate δ^{^}B_ij = ^{^}B_i - ^{^}Bj, thus removing the local EMI fields by common reference to site k. A more efficient method by Davis et al. (1981) predicts the differences δ^{^}Bij (t) between two total field magnetometers from the remaining total field magnetometers of an array, including one of the magnetometers that form the difference and the three components of a remote reference vector magnetometer according to the equation

δ^{^}B_ij  (t) = AX (t) + BY (t) + CZ (t) + D_iB_i (t) + e (t)

where i is an index running over the whole array and e (t) = δ^{^}B_ij  - δB_ij  . Thus, we see that Davis et al. (1981) make use of spatial information from a simultaneous array of magnetometers to remove local EMI effects. The above processing techniques were presented only for the sake of completeness, since we have not been able to operate more than two total field magnetometers simultaneously, or include a reference three-component magnetometer in our instrumentation. Thus, the data presented here have not been processed for local EMI effects. The existing data, however, facilitate an evaluation of the extent and intensity of local EMI effects and their influence on the results of this VM experiment.

Noise elimination possibilities are not exhausted with the simple differencing scheme described above. We can introduce a further degree of improvement to our results by minimizing the random fluctuations associated with physical measurements of the magnetic field. Random noise is the additive result of instrumental electronic noise and high frequency (ie. beyond the Nyquist) natural magnetic fields and their associated, highly localized EMI effects. It tends to be amplified upon using any differencing scheme. We reduce their effects by passing both data series that form δB_ij  through an identical rectangular moving average filter, using a time window of 15 min (10 samples), prior to differencing. Thus, every data point is substituted by 15 min averages, using both prior and future information. This is a simple and apparently very effective method for reducing the random noise content of data.

Anthropogenic noise can be the most serious and insurmountable problem in the vicinity of inhabited areas. It can be harmonically dependent or transient depending on the source, or even display dual character. For the long period data we obtain, this type of noise is usually transient and almost invariably, of high intensity that varies with location and time; it is more severe during daytime, when industrial activity is maximal. Because of its irregular characteristics it can only be treated with 'tailor-made' techniques. In order to minimize the probability of registering industrial noise, we have been recording data only during the night. Nonetheless, we have not always avoided industrial interference of the random transient type, that varied from isolated spikes to prolonged intense activity the nature of which is unaccounted for.  Typical examples are presented in Fig. 3. The noise waveforms of Fig. 3 are so intense as to totally corrupt the natural field. No processing method exists powerful enough to separate the geomagnetic field from noise without affecting the weak VM signal. Therefore, data sections thus contaminated can only be omitted from further consideration. On the other hand, there exist several techniques to eliminate isolated spikes contaminating a time series, such as the robust statistical filters of Kleiner et al. (1979) or Martin and Thomson (1982). These filters are very effective interpolators and will be most useful when handling data from magnetometer arrays. In such cases simultaneous data from several magnetometers must be processed at the same time and the existence of spikes or gaps in one data series may be damaging to the rest. For the simple differencing and smoothing scheme we implement here, this type of robust filtering is time-consuming and unnecessary. Isolated spikes and short duration transient noise can be successfully and simply eliminated by screen-editing the digital time series and deleting the corresponding noisy data points from both magnetometers that make up δB_ij .

Fig. 2a, b are examples of the above described procedure, applied to a set of anthropogenic noise-free data. Fig. 3a shows the raw total fields at locations AGA and PYR, and their differences. Fig. 2b is representative of the degree of improvement affordable after smoothing with the rectangular moving average filter described above. The reduction of the random noise content is apparent, as is the reduction of the variance of the δB_ij. This ultimately results in a reduction of the uncertainty associated with the sought mean value of δB_ij and better constraints monitoring of its evolution with time. Fig. 4 shows a few more typical examples of smoothed total field records and their differences, all calculated with respect to our permanent reference station at PYR. The frequency dependence of δB_ij due to local EMI is apparent. However, it is noteworthy that is the examples of Fig. 4, as well as in most data we have studied, the amplitude of residual frequency dependence very rarely exceeds the range ± 5 nT and usually varies within the range ± 3 nT. This yields, typically, a standard deviation of less than 1 nT, for δB_ij which is an encouraging result compared to resolutions reported by other workers (e.g. Zlotnicki 1986; Zlotnicki et al. 1986) and considering the limited range noise elimination techniques applicable to our present data.

DISCUSSION

The main purpose of the present work was to investigate the possibility of conducting a successful VM experiment in Thera. We considered such a preliminary study useful, because Thera has not been electromagnetically investigated in the past; in island areas, local EMI fields can be quite intense, because of the large conductivity contrasts and current channelling effects that may be encountered. We believe we have been successful in showing that a VM experiment in Thera is feasible and will yield useful results. For the simple combination of night-time recording, smoothing and differencing data reduction procedures we have adopted, the standard deviation of the residual field fluctuations is typically less than 1 nT and sometimes as low as 0.30 nT. This is about as good a result as there can be for these procedures (e.g. Rossignol 1982), and facilitates a very detailed, even day to day monitoring of the geomagnetic field. In all future fieldwork, we plan to employ a simultaneous array of at least four magnetometers and implement the more advanced data reduction techniques discussed above. This should lead to further reduction of residual fluctuation in δB_ij and increased confidence in the results.

Most reported VM studies deal with the problem of volcanic eruption prediction and therefore seek to detect short time constant changes (Rossignol 1982). Our measurements, sparse in time, forbid the observation of any short term changes associated with volcanic activity, which, in any case, has not been significant over the past two years. However, we have been able to observe a significant change in δB_ij between 1987 and 1988. This is tabulated in Table 1 for the baselines PYR-AGA, PYR-KLM, and PYR-OIA; a similar change was not observable in the southern base line PYR-AKR. As can be seen, the change is consistently positive in all three baselines, indicating a decrease of field intensity at the peripheral stations of the northern part of the island, or an increase at PYR, or a combination of the above effects. The results from two field campaigns in 1988, help to establish the consistency of the observations of this year, as well as the validity of the observed change in δB_ij.

Due to the absence of regular data between 1987 and 1988, we cannot study the evolution of δB_ij  along the baselines and, therefore, we cannot make inferences about the causes of the changes. However, the existing literature provides a theoretical basis for a few comments on the observations. Short time constant modifications in the local magnetic field have been presented by several workers (Johnston and Stacey 1969; Pozzi et al. 1979; Zlotnicki and Le Mouel 1988) who show impressive plots of the evolution of δB_ij within periods of a few to several days before volcanic eruptions and intrusions. In these cases there has been an increase in the local magnetic field of 4 to 10 nT, indicating piezomagnetic rather than thermal causes of the modification; δB_ij   tends to recover its 'quiet times' value after the crisis is over.

Other results obtained by Hurst and Christoffel (1973) in New Zealand associate changes in the magnetization of rocks with thermal processes, while Norris (1978) shows the relation between hot vents and demagnetization (decrease in the local field) in Papua, New Guinea. In both cases the changes were linked to increased volcanic activity. On the other hand, long time constant anomalies of the secular variation of the magnetic field have also been reported, although the literature is rather poor. In the GDR, Mundt (1978) reports secular anomalies of the order of 3-4 nT/yr, most of them over regions where recent vertical and horizontal movements have taken place and possibly correlated with positive heat flow. Tazima et al. (1976), in Japan, report anomalies of the order of 2 nT/yr, apparently correlated and explainable with thermal modifications (volcanism) and/or earthquakes (i.e. linked to changes in the regional stress field). Also, aseismic modification of the regional stress field in the Urals may be the cause of the secular anomalies observed by Shapiro et al. (1978).

In the case of Thera, the apparent field weakening between 1987 and 1988 would seem to imply apparent demagnetization in the northern part of the island and therefore possible increase in heat flow. A possible source of the excess heat could be the area of the Kolombos submarine volcano, an underwater crater off the north-east coast of the island. According to Xynogalas and Chailas (1987) most micro-earthquake activity in the immediate vicinity of Thera is concentrated around the Kolombos submarine volcano, and that near Kameni island does not appear to be associated with significant seismic activity. This would appear to imply that the activity of the volcano has at present shifted to the area of Kolombos.

Unfortunately, the operation of the seismic network installed as part of the seismological leg of the volcanic hazard project has been discontinuous, like our own magnetometric field campaigns, and we did not always overlap. Therefore, we cannot gain insight into the actual processes involved by studying the parallel evolution of earthquake activity. In the absence of recent heat flow measurements and other apparent precursory phenomena, we cannot regard the observed changes as a pre-indication of volcanic activity. On the other hand, the lack of a long observational data base does not allow us to confidently attribute the observed change to some kind of anomalous secular variation. Thus, at present, we can only speculate on the causes of the observed differences. It is clear, however, that the region of Thera is alive and active. Although our studies do not indicate any cause for immediate concern, they do point out that Thera is still potentially hazardous and a continuous monitoring of its activities is imperative.

---------------------------------------

--------------------------------------------

Source: "Thera and the Aegean World III, Vol 2" (pp. 207 - 214)

Authors: E. Lagios, A. Tzanis, S. Chailas,  - Department of Geophysics and Geothermy, University of Athens and M. Wyss - University of Colorado at Boulder nnn