July/August 1997


When people at a party find out I'm a seismologist, their first question is usually, "When will the next earthquake occur?" This suggests (1) that most people are fascinated and frightened by earthquakes, and (2) they think seismologists can predict them.

Unsuccessful earthquake prediction research has been going on for over one hundred years, as has criticism of that research. A paper by de Montalk in 1934 (BSSA, 1934, v. 24, 100-108) has the marvelously descriptive title "Earthquakes: The futility of predicting them." The criticisms by Macelwane in 1946 (BSSA, 1946, v. 36, 1-4) apply equally today: "Is it possible, in the present state of scientific knowledge, to predict earthquakes ...? All reputable seismologists agree that we have no means at the present time of arriving at a reliable forecast of any earthquake anywhere. The problem of earthquake forecasting has been under intensive investigation in California and elsewhere for some forty years, and we seem to be no nearer a solution of the problem than we were in the beginning. In fact the outlook is much less hopeful." Richter in 1977 (BSSA, 1977, v. 67, 1244-1247) commented: "Since my first attachment to seismology, I have had a horror of predictions and of predictors. Journalists and the general public rush to any suggestion of earthquake prediction like hogs toward a full trough. ... [Prediction] provides a happy hunting ground for amateurs, cranks, and outright publicity-seeking fakers. The vaporings of such people are from time to time seized upon by the news media, who then encroach on the time of men who are occupied in serious research." Developments in the last 20 years do not require any modification of the above views.

The popular belief that earthquakes are predictable is at odds with current research. At the recent RAS-JAG (Royal Astronomical Society-Joint Association for Geophysics) discussion meeting, "Assessment of schemes for earthquake prediction" (London, 7-8 November 1996; see meeting reports by Main in Nature, 1997, v. 385, 19-20 and Geller, Eos, 1997, v. 78, 63-67), the clear consensus was that individual earthquakes are inherently unpredictable because of the chaotic, highly nonlinear nature of the source process. The Earth appears to be in a state of "self-organized criticality"--always teetering on the edge of instability. Exactly when and where earthquakes occur, and how large they will grow after they start, depend on a myriad of fine and unmeasurable details of the physical state of the Earth over a large volume, not just in the immediate vicinity of the fault. The idea that the Earth telegraphs its punches, i.e., that large earthquakes are preceded by observable and identifiable precursors, like many such "obvious" ideas, isn't backed up by the facts. For further discussion of this point see my recent "Perspectives" article with Jackson, Kagan, and Mulargia in Science (1997, v. 275, 1,616-1,617). The popular idea of an "earthquake prediction" is a warning of an imminent damaging earthquake, given with sufficient accuracy and reliability to permit closing schools, factories, and businesses, mobilizing civil defense forces, enforcing full or partial evacuation, etc. The costs of such measures would be enormous, so extremely high reliability and accuracy would be required. Pulling some numbers out of a hat, let's say we would need a 50% chance of being right, and accuracy of one day in time, about 50 km in space, and 0.5 in magnitude. While there's room for disagreement, something of this order is required before prediction would pay off. But there are no good prospects for prediction at all at present, let alone with this accuracy or reliability.

Even if the prediction of individual large earthquakes were a goal that could be realized, it would still be of questionable utility. People would be far better off living and working in buildings that were designed to withstand earthquakes when they did occur. The design of earthquake-resistant structures is primarily a job for civil engineers, but seismologists play an important supporting role by providing information on the expected seismic hazard (e.g., Main, BSSA, 1995, v. 85, 1,299-1,308 and Jackson, BSSA, 1995, v. 85, 379-439) and the expected levels of strong ground motion (e.g., Anderson and Yu, BSSA, 1996, v. 86, S100-S114 and Kawase, SRL, 1996, v. 67, no. 5, 25-34). There are complex trade-offs here--it is economically impossible to design all structures to withstand all conceivable future earthquakes--but this is one area where seismologists can make major contributions to public safety.

Seismologists can also help to mitigate the effects of a quake by quickly determining source parameters (location, depth, origin time, energy, etc.) and getting this information to government leaders and the public as soon as possible after a large quake hits. The importance of this task, which is sometimes called "real-time seismology", is underscored by the experience of the 1995 Kobe earthquake. The quake struck at 5:46 a.m. (local time), but due to poor communication the Cabinet did not appreciate the extent of the disaster until that afternoon. Most of the dead were crushed in collapsed structures; perhaps a quarter to a third of them could have been saved if relief forces had been mobilized promptly. Another big contribution to public safety by seismologists is issuing warnings of tsunamis.

Unfortunately, the media do not consider seismic hazard mitigation to be a sexy topic, while earthquake prediction claims are avidly publicized. For example, the claims of J. Zschau of the Geo-Research Center in Potsdam, Germany were announced in a newspaper story (see sidebar). Zschau's claims have also been extensively reported in the German popular science magazines (Bild der Wissenschaft, December, 1993; Geo, March, 1996), but he has not yet published his work in a major refereed journal (cited in Georefs, 1991-1996). Nevertheless, the Ministry of Education and Research of the German government granted him approximately $2 million in funding (for one year) for a project on "earthquake prediction research", apparently without peer review.

Why did the German government allocate so much money for Zschau's unproven and apparently ineffective system, which is in any case a variation on an old theme? Two factors seem to play a part here. First, the responsible officials were apparently unaware of the long history of failed research on earthquake prediction. Second, the potential payoff is enticing: a system that could save large number of lives by predicting damaging earthquakes.

In recent years public officials in many nations seem to have decided that "curiosity-driven research" is bad and "strategic research" is good. Zschau's project shows what is wrong with this idea: The bureaucrats who dole out the funds seem to base their evaluation entirely on the stated goals of the project, without realistically assessing the possibility of achieving them. "Strategic research", at least in this cartoon-like form, seems doomed for the same reasons as the Marxist command economy: Government planners simply lack the knowledge to make sensible decisions. The stakes get much higher, and the media get even more frenzied, when scientists, either soi-disant or recognized, issue public predictions. For example, in May 1980, Brian Brady of the U.S. Bureau of Mines said that earthquakes with magnitude Mw 9.8 and Mw 8.8 would occur at the subduction zone off the coast of Peru in August 1981 and May 1982, respectively (see Kerr, Science, 1981, v. 211, 808-809; 1981, v. 213, 527; and Olson, Podesta, and Nigg, The Politics of Earthquake Prediction, Princeton University Press, 1989). A foreshock with Mw 7.5-8 was forecast for June 1981. When the foreshock failed to occur, Brady retracted his prediction. It had no sound basis, but much time and effort had to be wasted in evaluation. The prediction caused substantial unrest; the head of the Office of Earthquake Studies of the U.S. Geological Survey had to travel to Lima in June 1981 to assure the Peruvian people there was no cause for alarm.

An even more spectacular debacle involved Iben Browning, a business consultant with a Ph.D. in biology but no background in Earth science. Browning predicted that an earthquake with magnitude between 6.5 and 7.5 would strike New Madrid, Missouri (about 250 km southeast of St. Louis) between 1 and 5 December 1990 (Kerr, Science, 1990, v. 250, 511; 1991, v. 253, 622-623; Gori, BSSA, 1993, v. 83, 963-980; Spence, Herrmann, Johnston, and Reagor, USGS Circular 1083, 1993). The prediction had no scientific basis, but Browning's promotional efforts generated a flood of publicity and panic; public schools were even closed on December 3. Browning profited by selling explanatory videotapes at $99 a copy. Spence et al. reproduce over one hundred newspaper and magazine articles on Browning's prediction and provide a wealth of other information.

Another example comes from P. Varotsos of the University of Athens, Greece, who has been claiming for fifteen years to be able to predict earthquakes on the basis of geo-electrical data. During this period the details of Varotsos' claims have varied greatly. In 1981 he claimed to be able to detect precursory electrical signals a few minutes before every earthquake of M ≥ 2.6 within 80 km of his observatory, and before some large events at distances of hundreds of kilometers. He now claims only to be able to predict larger quakes (roughly M ≥ 5) one or two months in advance (see Varotsos, Lazaridou, Eftaxias, Antonopoulos, Makris, and Kopanas; and Geller in A Critical Review of VAN, World Scientific, Singapore, 1996), and the special issue "Debate on 'VAN'", GRL, 1996, v. 23, 1,291-1,452, for details and further references).

Varotsos has been remarkably successful in attracting funds, supporters, and publicity. At the RAS-JAG Meeting, S. Stiros reported that, for the past ten years, Varotsos and his colleagues have obtained, without peer review, about 40% of the funds allocated to the antiseismic organization in Greece. Several Japanese scientists have become strong supporters. After the Kobe earthquake they launched an all-out publicity campaign claiming that if only Varotsos' methods had been used in Japan the Kobe earthquake could have been predicted and thousands of lives saved.

▲ Artist's rendition of P. Varotsos. From the title page of Part II of Earthquakes Can Be Predicted! (from the comic book Shorten Sunday, 28 June 1995)

Tried and Tested?

Zschau's prediction techniques were announced in the following story in the December 16-22, 1994 issue of The European newspaper.

Researchers close to quake warning system
Till Mundzeck, London

German and Turkish scientists believe they are on the brink of a breakthrough in forecasting earthquakes. It could save lives and reduce the huge cost of damage--more than $100 billion in the past ten years. "On average a major earthquake happens every third day," said Professor Jochen Zschau of the Geo-Research Center in Potsdam, Germany. "The devastating effects could be considerably reduced if the approach of the quake was predictable." He spoke on 14 December as two earthquakes within an hour shook ski resorts in the French Alps.

An earthquake analysis system called Seismolap, developed as a joint venture between the Germans and Turkish scientists in northwest Anatolia, could provide the key.

The scientists noted that before major seismic activity in an earthquake-prone area there occurred a pause in the constant but normally imperceptible micro-earthquakes. "This pause, or hiatus, may start years prior to the actual quake, depending on its magnitude," Zschau said. "Once this quiet period passed, we saw the microquakes increasing again and clustering round the epicenter of the earthquake." According to Zschau this usually happened a couple of days before the quake. The hiatus-cluster combination has shown up in all the earthquake data from the eastern Mediterranean that the scientists have analyzed. So far the results are retrospective but the team is optimistic that, in conjunction with its new satellite monitoring program, Readiness, a viable warning system will emerge.

Comment on newspaper quote:
Zschau's Seismolap system is not an earthquake prediction system, because it does not forecast the time, location, and magnitude of an earthquake within narrow and explicitly stated limits. And claims of precursory variations in seismicity are an old story. Richter (JGR, 1964, 69, 3025) commented as follows: "Short periods of exceptional quiet in the area of an approaching earthquake are often reported. ... There is difficulty, and some arbitrariness ... in selecting the area which is to be included in each individual study. Working after the occurrence, it will almost always be possible to select the area so as to give the expected result; it has to be shown that this selection then has a real tectonic significance."

Reasenberg (Eos, 1996 Fall Meeting Supplement) applied the Seismolap algorithm to seismicity in California, with the following conclusion. "Seismolap appears to be unstable both in these forecasting tests and in the behavior of the underlying algorithm, whose output is sensitive to the values selected for the free parameters. Seismolap calculations are often based on very few earthquakes, adding to its instability. And a nonlinear filter in the algorithm apparently amplifies these instabilities. The result is that Seismolap tends to generate a large number of extrema and forecasts, some of which, apparently by chance, then become associated with some large earthquakes."

Eschewing refereed journals, they carried out this campaign in newspapers and magazines, on TV, and even in comic books. The title page of an 80-page, two-part series "Earthquakes can be predicted!" in the comic book (manga) called Shonen Sunday lists T. Nagao, a geophysicist now at Tokai University, as an editorial consultant. The comic was prepared with the full cooperation of Varotsos.

The cumulative effect of these and other specious claims, extensively reported by the media, has been to give ordinary citizens and government officials the incorrect impression that earthquakes can be predicted. This not only leads to wasting funds on pointless prediction research, it also leads to neglect of practical precautions that could save lives and reduce property damage when a quake comes.

The scientific community has buried its head in the sand when it comes to dealing with prediction claims. The basic problem is that most of these claims--Varotsos' are a classic example--are stated in such a vague and ambiguous way that objective testing is impossible. First, we have to waste enormous amounts of time just sorting out whether or not the "predictions" were "successful." But this is an exercise in futility because of the ambiguity of the "prediction" announcements. And after that we're faced with the even more daunting task of evaluating whether or not the "predictions" have outperformed the null hypothesis.

The predictors outflank the peer review process by taking their case directly to the media. If they would instead submit their results for publication in scientific journals, the burden of proof would be on them to show that their predictions were formulated unambiguously. They would also have to show the statistical significance of the results--just like all other researchers. And their claims would be published in a way that permitted verification by independent workers.

The long and tragic history of earthquake prediction research exposes a dark side of our science. Work in this field has consisted of incorrectly attributing significance to signals at or below the noise level. While there might have been some isolated instances of intentional improper scientific conduct, in general workers in this field have systematically deceived themselves. Noisy data have been treated as signals, studies using posterior adjustment of parameters were evaluated using statistical tests for hypotheses with a priori fixed parameters, and alternative hypotheses were systematically excluded from consideration. The situation is like unfortunate examples in other fields ("pathological science", Langmuir, Phys. Today, 1989, v. 42(10), 36-48, or "cargo cult science", R. P. Feynman in Surely You're Joking, Mr. Feynman, Norton, 1985).

The peer review process isn't just an arbitrary social convention--it's a system that has been developed over the past three hundred years for sorting out scientific controversies. Prediction proponents should publish in refereed scientific journals, not newspapers or comic books.

Incidentally, just in case you're wondering, I've published many papers in BSSA and other refereed journals, but haven't yet published in a comic book.

Robert J. Geller
Department of Earth and Planetary Physics
Faculty of Science
Tokyo University, Yayoi 2-11-16
Bunkyo-ku, Tokyo 113 JAPAN
e-mail: bob@eps.s.u-tokyo.ac.jp

This article was first printed in the February/March 1997 issue of Astronomy & Geophysics, the journal of the Royal Astronomical Society. It is reprinted here with the permission of the Royal Astronomical Society.

To send a letter to the editor regarding this opinion or to write your own opinion, you may contact the SRL editor by sending e-mail to <srled [at] seismosoc [dot] org>.

Posted: 22 January 1999
E-mail updated: 27 September 2005
Updated (to include cartoon): 7 October 2011