research report from Fujita

Fujita Research is a company with a simple mission - to create harmony between people, the environment and technology through sustainable development in construction. We already market several innovative products and, together with industry partners, are working to develop real solutions to the problems of the 21st centruy construction industry.

Fujita Research was founded in 1989 in California and now has branch offices in Bangkok, Tokyo and Brighton, England. We are the primary R&D division of Japan's 7th largest construction company, the Fujita Corporation.

Whilst every effort is made to ensure these reports are accurate, we cannot be held liable for any errors or omissions. You should verify any information contained herein with independent third parties before making any commercial decisions.

Click here to access other reports from Fujita Research
Click to access the Fujita Research homepage


Earthquake Prediction

last updated: 08/97

1. Introduction

Despite the amount of research undertaken over the last century, the prediction of earthquakes still remains as much an art as a science. Reports of animals behaving strangely are still regarded as accurate as more “scientific” techniques. One possible breakthrough in prediction is the VAN technique, derived by Professor Varotsos and co-workers at the University of Athens (1) . But, despite apparent successes, the VAN method still has a large number of critics. Indeed, the debate on VAN was recently the subject of a special issue of Geophysical Research Letters (2).

As staff at Fujita are familiar with VAN, and the work of Professor Varotsos, this report focuses on other, non-VAN, methods of predicting earthquakes. It includes two of the precursors recognized by the IASPEI sub-commission on earthquake prediction: water level changes and increases in concentrations of Radon gas in deep wells. It also examines some phenomena which have not been so recognized: increased low frequency noise, tilt precursors, concentration of anions in ground water, and thermal anomalies.

The quality of the research presented is quite variable. Some is based on many well-documented measurements with a high degree of reproducibility. Some is based on single measurements by a single instrument for a single earthquake. However, with the uncertainties associated with earthquake prediction-work, no plausible technique has been discounted, but is instead presented together with some idea of the data-quality on which the work is based.

2. Ground Water Levels

Changing water levels in deep wells is recognized by the IASPEI as a significant precursor to earthquakes. Perhaps part of the reason for this is explained by the discoveries of German scientists working at “KTB”, which is planned to be the deepest hole ever drilled in the earth’s crust (to 10 km). At a depth of 3 900 m the researchers struck water. A heavy brine with a salinity twice that of sea water, it was at a temperature of 118°C and contained 80% by volume of gases in solution, principally N2(70%) and CH4 (29%). The discovery of this brine led the researchers to postulate a “crustal ocean”, with tides, currents, and flows, all of which could conceivably react to seismic activity.

No simple model exists to connect pre- or co-seismic fluctuation of ground water levels to this crustal ocean. Lomnitz (3) considers that, ultimately, the mechanism will be found to be related to pressure changes, rather than changes in volume in the focal region (as most geophysicists currently believe). Such regional pressure changes, can be detected at deep wells.

The sensitivity of deep wells to seismic activity is remarkably varied. A number of deep wells in China are reported (4) to be extremely sensitive to changes in pressures, and can reliably detect earthquakes occurring halfway round the world. This observed sensitivity is probably due to their being quite protected from surface noise (rainfall, seasonal effects, etc.). As a result, China relies a great deal on deep wells for earthquake prediction. Indeed, over 100 research wells in excess of 1000 m deep have been drilled solely for earthquake prediction purposes. In these wells, water levels are continually monitored to ±0.5 cm and temperatures to ±0.01°C. Japan also relies to some extent on such wells - some 93 wells are monitored for earthquakes. In general a pre-seismic variations at observation wells follows this sequence:

1) A gradual lowering of water levels of a period of months or years
2) An accelerated lowering of water levels (rate often exponential) in the final few months or weeks preceding the earthquake.
3) A “rebound” where water levels begin to increase rapidly in the last few days or hours before the main shock.

In the monitoring of water levels in deep wells, care must be taken to correct the data for “earth tides”. This is due either to volume changes caused in fractured aquifers by tidal strain, or perhaps by changes in gravitational acceleration alone. In either case, it is important that data is corrected for this phenomenon. In addition water extraction from the aquifer must also be considered. In many part’s of the planet the water table is falling due to water abstraction for drinking and irrigation. It is quite possible that such drops could be mistaken for a long-term seismic precursor.

Case Study One: Tangshan Earthquake, China

The Tangshan earthquake occurred in the Hebei Province of China at 0342hrs on 28 July 1976. A magnitude 7.6 quake, it occurred in an area considered to be relatively “non-seismic” and resulted in the deaths of between 240 000 and 650 000 people. A mining area, the subsurface geology of the region was well known, and many records of water levels in wells and pumping rates in mines were available for analysis following the earthquake. Indeed the Tangshan quake provides the largest known dataset for determination of precursor groundwater anomalies.

Of particular interest is the pumping record from the Tangshan mine, which has kept records since 1923, and since that time has shown a generally stable signal with no seasonal fluctuation (probably due to the deep nature of the mine). For several years prior to the 1976 earthquake the required pumping rate dropped, initially very slowly, and then at an exponential rate. At a point between two days and three hours before the main shock a rapid reverse occurred in this trend, with pumping rates increasing from 25 m3sec-1 to 75 m3sec-1 immediately before the shock.

Outside the mines other wells in the region reported similar trends. Over the preceding years water-levels across the region had dropped, with a number of wells drying-up altogether. From 3hrs to 4 min before the shock many wells became artesian. There was also a very large co-seismic rise of up to several meters. Outside the region, an 8 cm increase in water level was observed at least one well (d=125 km).

Following the identification of this trend of a long period of lowering water table followed by a sharp rise, mine records were analyzed to see if similar trends could be observed for other earthquakes. Clear signals were seen for both the 1969 Bohai earthquake (M=7.4 d=200 km) and the 1945 Luan Xian earthquake (M=4.1 d=55 km).

3. Chemical Changes in Ground Water

The changes observed in volume of ground water prior to earthquakes prompted researchers at the University of Tokyo (5) to consider the possibility that the chemical composition of ground water might also be affected by seismic events. In the aftermath of the Kobe earthquake they were fortunate to be able to obtain dated samples of mineral water taken from nearby springs. The water is commercially bottled for sale as drinking water, and thus the researchers were able to assemble a time sequence of water from stocks held at warehouses throughout Japan (total of 59 differently dated bottles and 11 duplicates for the period 5 June 1993 to 13 Jan 1995).

The results of the study showed that the chemical composition of the water changed significantly in the period around the Kobe earthquake. From June 1993 to July 1994 chloride (aq) concentrations were almost constant (13.7-14.1 ppm). Between July 1994 and Jan 1995 a steady rise was observed to 15 ppm. Levels of sulphate also showed a similar rise. However, the rise is not particularly useful from a predictive point of view, as levels of chloride (aq) and sulphate (aq) did not peak until the end of Feb 1995, decreasing throughout March to their former levels.

4. Radon Gas in Ground Water Wells

Increased levels of radon gas (222Rn) in wells is a precursor of earthquakes recognized by the IASPEI. Although radon has a relatively short half life (t1/2=98hrs), and is therefore unlikely to seep to the surface through rocks from the depths at which seismic activity occurs. However, radon is very soluble in water, and can routinely be monitored in wells and springs. Often, radon levels at such springs show reaction to seismic events and, worldwide, many are monitored for earthquake predictions.

Case Study Two: Spring at Bad Brambach (Vogtland, Germany) (6)

Researchers at Bad Bramburgh in Germany have been monitoring the local springs for radon and CO2 levels since 1989, with improved equipment installed in 1992. The region is subject to numerous microearthquakes (M<4.0), which on some occasions occur at such high frequencies as to be considered “swarmquakes”. Over the last five years they have found that radon anomalies are associated with seismic events. Unfortunately from the predictive point of view these anomalies may be noted before, during, or after an earthquake.

Despite not being suitable for prediction of seismic events, the radon anomalies observed at Bad Bramburgh have produced some useful information about the mechanism by which radon levels increase. Measurement of CO2 has shown strong correlation between concentration of the gas and that of Radon. It is suggested that tectonic stress-strain triggers CO2 outgassing which, in turn, acts as a carrier for Radon gas. The d13C values (-3.5 to -3.7‰) of elevated CO2 concentrations confirm that it does indeed come from old groundwater rather than the more surficial groundwater which accounts for 20-40% of spring output.

Case Study Three: Kobe Earthquake, Japan (7)

Over the last twenty years the University of Tokyo and the Geological Society in Japan have monitored radon levels in groundwater in an effort to predict earthquakes in eastern Japan. One such well is located in the southern part of Nishinomiya city, about 30 km NE of the epicenter of the M=7.2 Kobe earthquake of 17 Jan 1995. The well was first monitored between 26 Nov and 02 Dec 1993, with continual monitoring starting on 27 Oct 1994.

During the 1993 observation period, concentrations of radon were stable at 20Bq/l. By the end of Nov 1994 levels had increased to 60Bq/l. On 7 Jan 1995 a huge increase in radon concentration was observed (to ca. 250Bq/l). These high levels dropped suddenly on 10 Jan, one week before the earthquake. By the time of the earthquake levels had returned to about 30 Bq/l, levels confirmed when the station came back on-line on 22 Jan (monitoring equipment had been damaged by the main shock).
The researchers have examined other possible reasons for the observed increase in Radon levels, but no satisfactory alternate explanation could be found. Over the period of well-monitoring ground water temperature remained almost constant (±0.2°C), and there was no significant rainfall which might have affected the aquifer. Atmospheric pressure is known to have little effect on radon concentrations, so meteorological explanations were also unlikely.

5. Tilt Precusors

In Chile, researchers have reported (8) that, for certain types of earthquakes, tilt may be observed near the epicentral region for some months prior to the main shock. While the use of a tilt precursor in predicting the exact timing of earthquakes is unclear from the case study below, it is clear that as a more generalized signal it is useful in identifying a region in which stress and strain is rapidly accumulating.

Case Study Four: The Rapel Reservoir, Chile

The Rapel Reservoir in Chile is a site at which an extremely good dataset has permitted accurate measurement of tilt prior to a seismic event. Water levels in the reservoir have been measured at two sites (20 km apart) since 1980. About eight months before an earthquake on 3 March 1985 (M=7.9) the levels measured at the two gauges began to show differences due to tilt of the underlying lakebed - differences which increased until 9 months after the shock. The maximum tilt was 13 cm, measured over the 20 km baseline. The shock occurred when the tilt was approximately 0.3 of this maximum value (ca. 4 cm), but no co-seismic perturbation in tilt appeared.

6. VLF Background Noise

The VAN Technique holds that changes in the earth’s electric field prior to an earthquake. A group of researchers in from Russia and Japan (9) believe that similar changes can be noted in the earth’s background noise in the low (LF), very low (VLF), and extremely low (ELF) frequency bands. The details in print are very few; the authors stated in 1991 that they were on the verge of submitting major papers to various journals, but nothing further was published between 1991 and 1996. Nevertheless the possibility that VLF Noise represents a significant precursor to earthquakes was evaluated by the IASPEI in its most recent meeting (the findings of which appeared in Pure and Applied Geophysics, 1997).

As early as 1985 the research group had established a detection network around Tokyo, consisting of six fixed and three mobile stations. Background noise was measured at three frequencies 82kHz (LF), 1525Hz (VLF) and 36Hz (ELF). Between 1986 and 1991 the group claimed to have observed precursors for 20 earthquakes, but no significant details of these were published in the reviewed paper. As a result, the IASPEI reviewed earlier papers by Yoshino’s researchers and also a number by Dr. Gokhberg and his Russian team (10) who reported similar measurements of increased LF, VLF, and ELF noise prior to earthquakes.

The conclusions of the IASPEI were cautious. They were not convinced that a mechanism existed for the transmission of signals from the great depths of the earthquakes (ca. 480 km) which Yoshino claimed generated the signals. Also, there was some confusion about exactly what Yoshino and co-workers were describing as anomalies. Indeed, the panel was more convinced by the observations of Gokhberg and co-workers, who observed three distinct build-ups of noise in the LF (81-82 kHz) prior to three seismic events, after which the observed noise dropped by ca. 10dB. However, even in this case there seem to have been no reported occurrences since late 1981. The panel has not ruled out the possibility that some electromagnetic signals may be noted before earthquakes. Clearly, more research in this field will be necessary in order to assess the value of LF, VLF, and ELF measurements.

7. Thermal Anamoly

One of the most interesting, but least well-documented, proposals put to the recent meeting of the IASPEI sub-commission on Earthquake Prediction, is the idea that anomalies in ground temperature may be connected with seismic activity. The idea that ground temperature anomalies are significant is seductive, as ground-temperature is a parameter which can cheaply and easily monitored over vast areas by use of satellites such as Russia’s Meteosat. However only two papers (11,12) have been published on use of such thermal anomalies, and both deal with only one seismic event - hardly sufficient data on which to test the authors’ claims. Nevertheless, as the method - if it worked - would have so much to recommend it the data from the second paper is summarized here (the first paper was published in Russian and translation was not available).

Case Study Five: The Datong Earthquake

On 18 Oct 1989, the Shaxi Province of China was hit by the Datong earthquake (M=6.1 epicenter = 39°57’N, 113°43’E, h=9km). From noon on Oct 15 - 2 am on Oct 16 an area of increasing temperature (ca. 300km long x 20 km wide) was observed running WSW-ENE around Datong. Within this area observed temperatures were 4° higher than in the surrounding mountains. From 2 am on Oct 16 the anomaly increased to a maximum value of 5-6°C higher than the surrounding area, a value reached 22 hours before the earthquake. Following the main shock the temperature anomaly began to decay. The pre-seismic temperature rise was accompanied by cloud formation . In the early phase this was a cloud 350 km x 50 km running generally SW-NE and centering around Datong. In the period of maximum anomaly it was a long thin (1800 x 30 km) cloud, running E-W.

The Chinese researchers assert that this temperature anomaly and cloud formation were the direct result of what they term “earth-degassing”. They believe that the increased levels of CO2, H2 and water vapor (for which they present no data) lead to the creation of a localized greenhouse effect. Clearly the IASPEI panel was more cautious in the their review of the paper. Indeed, one member remarked that the paper as presented would not be accepted by any reputable research journal. Admittedly, the paper has been poorly translated, contains no primary data, and merely asserts that temperature increase is due to seismic activity rather than meteorological causes. Also, if such “degassing” occurs, it should be possible to measure directly, and again no supporting data is presented. However, the panel concludes that the method cannot be discounted without further research, and that, if the method could be proved, it would one of the most useful methods of earthquake prediction - able to cover vast areas and provide real-time data.


In earthquake prediction, there are no simple conclusions which can be drawn. The events on which hypotheses can be tested are often few and far between, and in many cases the recording equipment is affected, or even destroyed by the earthquake. Also, while anomalies in data can often be used to hindcast earthquakes (determine them after the event), changing parameters (e.g. radon levels, chloride concentrations) often continue to rise steadily after the seismic event, rather than showing a dramatic and sudden return to normal levels before the earthquake. In addition the idea of being able to reliably predict earthquakes is in itself so seductive that it is possible to extremely over-optimistic in data analysis.

Nevertheless, this report has examined some techniques which show some success some of the time. The variability of the success of these techniques (especially water level, radon concentration) is probably due to local geological factors as much as the actual existence of the phenomenon. For example, for a well to show changes it probably has to be in contact with an aquifer that is strained by the seismic processes leading to a particular earthquake. This would explain why some wells show no sensitivity despite being close to an epicenter (perhaps analogous to the “directivity effect” of VAN).

Clearly there is much research still to be done before a reliable method of predicting earthquakes is developed - perhaps one never will be. But already it seems to be emerging that a combination of techniques can point to the fact that something is “going on” seismically, that stress is slowly increasing , and that something will give sometime soon. When that prediction becomes reliable then, though the scientific questions may have been answered, the political ones (when to warn citizens, and to evacuate them) will just be beginning.


1. P. Varotsos and K. Alexopoulous, ‘ Physical Properties of the Variations of the Electric Field of the Earth Preecedine Earthquakes’ Tectonophysics v. 110 (1984) pp.73-125.

2. Geophysical Research Letters, v.23 (1996)

3. C. Lomnitz, Fundamentals of Earthquake Prediction (1994)

4. ibid.

5. U.Tsunogai and H.Wakita, ‘Precursory Chemical Changes in Ground Water: Kobe Earthquake, Japan’ Science v.269 (1995)

6. U.Koch & Jens Heinicke, ‘Radon Behaviour in mineral Spring water of Bad Bramburgh (Vogtland, Germany) in the temporal vicinity of the 1992 Rörmond earthquake, the Netherlands’ Geologie en Mijnbouw v.73 (1994) pp399-406

7. G.Igarashi, ‘Ground-Water Radon Anomoly before the Kobe Earthquake in Japan’ Science v.269 (1995) pp. 60-61

8. Barrienntos and Kausel, reported in C. Lomnitz, Fundamentals of Earthquake Prediction (1994)

9. T. Yohino, ‘Increasing VLF Background Noise Level’ Pure and Applied Geophysics 149 (1997) pp.147-157

10. M. Gokhberg et al. , ‘Experimental Measurement of Electromagnetic Emissions Possibly related to Earthquakes in Japan’ Journalof Geophysical Research v.87 (1982) pp.7824-7828

11. Zu-ji Qiang et al. ‘Thermal Infrared Anomoly Precursor of Impending Earthquakes’ Pure and Applied Geophysics 149 (1997) pp.159-171

12. F.Y. Dusmukhamedov, ‘Lito-Atmospheric Re;lations before the Strong Earthquake in Central Asia’ in The Theory, Method and Practice of the Research of Geoindication, abstract, 3rd All-Union Meeting, Kiev, May 16-18 1989, 49-50 (in Russian)

©1997 Fujita Research